mirror_zfs/cmd/zdb/zdb.c

4487 lines
114 KiB
C
Raw Normal View History

2008-11-20 23:01:55 +03:00
/*
* CDDL HEADER START
*
* The contents of this file are subject to the terms of the
* Common Development and Distribution License (the "License").
* You may not use this file except in compliance with the License.
*
* You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
* or http://www.opensolaris.org/os/licensing.
* See the License for the specific language governing permissions
* and limitations under the License.
*
* When distributing Covered Code, include this CDDL HEADER in each
* file and include the License file at usr/src/OPENSOLARIS.LICENSE.
* If applicable, add the following below this CDDL HEADER, with the
* fields enclosed by brackets "[]" replaced with your own identifying
* information: Portions Copyright [yyyy] [name of copyright owner]
*
* CDDL HEADER END
*/
2008-11-20 23:01:55 +03:00
/*
* Copyright (c) 2005, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2011, 2016 by Delphix. All rights reserved.
* Copyright (c) 2014 Integros [integros.com]
* Copyright 2016 Nexenta Systems, Inc.
* Copyright (c) 2017 Lawrence Livermore National Security, LLC.
* Copyright (c) 2015, 2017, Intel Corporation.
2008-11-20 23:01:55 +03:00
*/
#include <stdio.h>
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
#include <unistd.h>
2008-11-20 23:01:55 +03:00
#include <stdio_ext.h>
#include <stdlib.h>
#include <ctype.h>
#include <sys/zfs_context.h>
#include <sys/spa.h>
#include <sys/spa_impl.h>
#include <sys/dmu.h>
#include <sys/zap.h>
#include <sys/fs/zfs.h>
#include <sys/zfs_znode.h>
#include <sys/zfs_sa.h>
#include <sys/sa.h>
#include <sys/sa_impl.h>
2008-11-20 23:01:55 +03:00
#include <sys/vdev.h>
#include <sys/vdev_impl.h>
#include <sys/metaslab_impl.h>
#include <sys/dmu_objset.h>
#include <sys/dsl_dir.h>
#include <sys/dsl_dataset.h>
#include <sys/dsl_pool.h>
#include <sys/dbuf.h>
#include <sys/zil.h>
#include <sys/zil_impl.h>
#include <sys/stat.h>
#include <sys/resource.h>
#include <sys/dmu_traverse.h>
#include <sys/zio_checksum.h>
#include <sys/zio_compress.h>
#include <sys/zfs_fuid.h>
#include <sys/arc.h>
#include <sys/ddt.h>
#include <sys/zfeature.h>
#include <sys/abd.h>
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 15:45:09 +04:00
#include <zfs_comutil.h>
2008-11-20 23:01:55 +03:00
#include <libzfs.h>
#define ZDB_COMPRESS_NAME(idx) ((idx) < ZIO_COMPRESS_FUNCTIONS ? \
zio_compress_table[(idx)].ci_name : "UNKNOWN")
#define ZDB_CHECKSUM_NAME(idx) ((idx) < ZIO_CHECKSUM_FUNCTIONS ? \
zio_checksum_table[(idx)].ci_name : "UNKNOWN")
#define ZDB_OT_TYPE(idx) ((idx) < DMU_OT_NUMTYPES ? (idx) : \
(((idx) == DMU_OTN_ZAP_DATA || (idx) == DMU_OTN_ZAP_METADATA) ? \
DMU_OT_ZAP_OTHER : DMU_OT_NUMTYPES))
static char *
zdb_ot_name(dmu_object_type_t type)
{
if (type < DMU_OT_NUMTYPES)
return (dmu_ot[type].ot_name);
else if ((type & DMU_OT_NEWTYPE) &&
((type & DMU_OT_BYTESWAP_MASK) < DMU_BSWAP_NUMFUNCS))
return (dmu_ot_byteswap[type & DMU_OT_BYTESWAP_MASK].ob_name);
else
return ("UNKNOWN");
}
extern int reference_tracking_enable;
extern int zfs_recover;
extern uint64_t zfs_arc_max, zfs_arc_meta_limit;
extern int zfs_vdev_async_read_max_active;
2008-11-20 23:01:55 +03:00
const char cmdname[] = "zdb";
uint8_t dump_opt[256];
typedef void object_viewer_t(objset_t *, uint64_t, void *data, size_t size);
extern void dump_intent_log(zilog_t *);
uint64_t *zopt_object = NULL;
int zopt_objects = 0;
libzfs_handle_t *g_zfs;
uint64_t max_inflight = 1000;
2008-11-20 23:01:55 +03:00
static void snprintf_blkptr_compact(char *, size_t, const blkptr_t *);
2008-11-20 23:01:55 +03:00
/*
* These libumem hooks provide a reasonable set of defaults for the allocator's
* debugging facilities.
*/
const char *
_umem_debug_init(void)
2008-11-20 23:01:55 +03:00
{
return ("default,verbose"); /* $UMEM_DEBUG setting */
}
const char *
_umem_logging_init(void)
{
return ("fail,contents"); /* $UMEM_LOGGING setting */
}
static void
usage(void)
{
(void) fprintf(stderr,
"Usage:\t%s [-AbcdDFGhiLMPsvX] [-e [-V] [-p <path> ...]] "
"[-I <inflight I/Os>]\n"
"\t\t[-o <var>=<value>]... [-t <txg>] [-U <cache>] [-x <dumpdir>]\n"
"\t\t[<poolname> [<object> ...]]\n"
"\t%s [-AdiPv] [-e [-V] [-p <path> ...]] [-U <cache>] <dataset> "
"[<object> ...]\n"
"\t%s -C [-A] [-U <cache>]\n"
"\t%s -l [-Aqu] <device>\n"
"\t%s -m [-AFLPX] [-e [-V] [-p <path> ...]] [-t <txg>] "
"[-U <cache>]\n\t\t<poolname> [<vdev> [<metaslab> ...]]\n"
"\t%s -O <dataset> <path>\n"
"\t%s -R [-A] [-e [-V] [-p <path> ...]] [-U <cache>]\n"
"\t\t<poolname> <vdev>:<offset>:<size>[:<flags>]\n"
"\t%s -S [-AP] [-e [-V] [-p <path> ...]] [-U <cache>] "
"<poolname>\n\n",
cmdname, cmdname, cmdname, cmdname, cmdname, cmdname, cmdname,
cmdname);
(void) fprintf(stderr, " Dataset name must include at least one "
"separator character '/' or '@'\n");
(void) fprintf(stderr, " If dataset name is specified, only that "
"dataset is dumped\n");
(void) fprintf(stderr, " If object numbers are specified, only "
"those objects are dumped\n\n");
(void) fprintf(stderr, " Options to control amount of output:\n");
(void) fprintf(stderr, " -b block statistics\n");
(void) fprintf(stderr, " -c checksum all metadata (twice for "
2009-07-03 02:44:48 +04:00
"all data) blocks\n");
(void) fprintf(stderr, " -C config (or cachefile if alone)\n");
(void) fprintf(stderr, " -d dataset(s)\n");
(void) fprintf(stderr, " -D dedup statistics\n");
(void) fprintf(stderr, " -h pool history\n");
(void) fprintf(stderr, " -i intent logs\n");
(void) fprintf(stderr, " -l read label contents\n");
2009-01-16 00:59:39 +03:00
(void) fprintf(stderr, " -L disable leak tracking (do not "
"load spacemaps)\n");
(void) fprintf(stderr, " -m metaslabs\n");
(void) fprintf(stderr, " -M metaslab groups\n");
(void) fprintf(stderr, " -O perform object lookups by path\n");
2008-11-20 23:01:55 +03:00
(void) fprintf(stderr, " -R read and display block from a "
"device\n");
(void) fprintf(stderr, " -s report stats on zdb's I/O\n");
(void) fprintf(stderr, " -S simulate dedup to measure effect\n");
(void) fprintf(stderr, " -v verbose (applies to all "
"others)\n\n");
(void) fprintf(stderr, " Below options are intended for use "
"with other options:\n");
(void) fprintf(stderr, " -A ignore assertions (-A), enable "
"panic recovery (-AA) or both (-AAA)\n");
(void) fprintf(stderr, " -e pool is exported/destroyed/"
"has altroot/not in a cachefile\n");
(void) fprintf(stderr, " -F attempt automatic rewind within "
"safe range of transaction groups\n");
(void) fprintf(stderr, " -G dump zfs_dbgmsg buffer before "
"exiting\n");
(void) fprintf(stderr, " -I <number of inflight I/Os> -- "
"specify the maximum number of\n "
"checksumming I/Os [default is 200]\n");
(void) fprintf(stderr, " -o <variable>=<value> set global "
"variable to an unsigned 32-bit integer\n");
(void) fprintf(stderr, " -p <path> -- use one or more with "
"-e to specify path to vdev dir\n");
(void) fprintf(stderr, " -P print numbers in parseable form\n");
(void) fprintf(stderr, " -q don't print label contents\n");
(void) fprintf(stderr, " -t <txg> -- highest txg to use when "
"searching for uberblocks\n");
(void) fprintf(stderr, " -u uberblock\n");
(void) fprintf(stderr, " -U <cachefile_path> -- use alternate "
"cachefile\n");
(void) fprintf(stderr, " -V do verbatim import\n");
(void) fprintf(stderr, " -x <dumpdir> -- "
"dump all read blocks into specified directory\n");
(void) fprintf(stderr, " -X attempt extreme rewind (does not "
"work with dataset)\n");
2008-11-20 23:01:55 +03:00
(void) fprintf(stderr, "Specify an option more than once (e.g. -bb) "
"to make only that option verbose\n");
(void) fprintf(stderr, "Default is to dump everything non-verbosely\n");
exit(1);
}
static void
dump_debug_buffer(void)
{
if (dump_opt['G']) {
(void) printf("\n");
zfs_dbgmsg_print("zdb");
}
}
2009-07-03 02:44:48 +04:00
/*
* Called for usage errors that are discovered after a call to spa_open(),
* dmu_bonus_hold(), or pool_match(). abort() is called for other errors.
*/
2008-11-20 23:01:55 +03:00
static void
fatal(const char *fmt, ...)
{
va_list ap;
va_start(ap, fmt);
(void) fprintf(stderr, "%s: ", cmdname);
(void) vfprintf(stderr, fmt, ap);
va_end(ap);
(void) fprintf(stderr, "\n");
dump_debug_buffer();
2009-07-03 02:44:48 +04:00
exit(1);
2008-11-20 23:01:55 +03:00
}
/* ARGSUSED */
static void
dump_packed_nvlist(objset_t *os, uint64_t object, void *data, size_t size)
{
nvlist_t *nv;
size_t nvsize = *(uint64_t *)data;
char *packed = umem_alloc(nvsize, UMEM_NOFAIL);
2009-07-03 02:44:48 +04:00
VERIFY(0 == dmu_read(os, object, 0, nvsize, packed, DMU_READ_PREFETCH));
2008-11-20 23:01:55 +03:00
VERIFY(nvlist_unpack(packed, nvsize, &nv, 0) == 0);
umem_free(packed, nvsize);
dump_nvlist(nv, 8);
nvlist_free(nv);
}
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 15:45:09 +04:00
/* ARGSUSED */
static void
dump_history_offsets(objset_t *os, uint64_t object, void *data, size_t size)
{
spa_history_phys_t *shp = data;
if (shp == NULL)
return;
(void) printf("\t\tpool_create_len = %llu\n",
(u_longlong_t)shp->sh_pool_create_len);
(void) printf("\t\tphys_max_off = %llu\n",
(u_longlong_t)shp->sh_phys_max_off);
(void) printf("\t\tbof = %llu\n",
(u_longlong_t)shp->sh_bof);
(void) printf("\t\teof = %llu\n",
(u_longlong_t)shp->sh_eof);
(void) printf("\t\trecords_lost = %llu\n",
(u_longlong_t)shp->sh_records_lost);
}
static void
zdb_nicenum(uint64_t num, char *buf)
{
if (dump_opt['P'])
(void) sprintf(buf, "%llu", (longlong_t)num);
else
nicenum(num, buf);
}
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
const char histo_stars[] = "****************************************";
const int histo_width = sizeof (histo_stars) - 1;
2008-11-20 23:01:55 +03:00
static void
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
dump_histogram(const uint64_t *histo, int size, int offset)
2008-11-20 23:01:55 +03:00
{
int i;
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
int minidx = size - 1;
2008-11-20 23:01:55 +03:00
int maxidx = 0;
uint64_t max = 0;
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
for (i = 0; i < size; i++) {
2008-11-20 23:01:55 +03:00
if (histo[i] > max)
max = histo[i];
if (histo[i] > 0 && i > maxidx)
maxidx = i;
if (histo[i] > 0 && i < minidx)
minidx = i;
}
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
if (max < histo_width)
max = histo_width;
2008-11-20 23:01:55 +03:00
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
for (i = minidx; i <= maxidx; i++) {
(void) printf("\t\t\t%3u: %6llu %s\n",
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
i + offset, (u_longlong_t)histo[i],
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
&histo_stars[(max - histo[i]) * histo_width / max]);
}
2008-11-20 23:01:55 +03:00
}
static void
dump_zap_stats(objset_t *os, uint64_t object)
{
int error;
zap_stats_t zs;
error = zap_get_stats(os, object, &zs);
if (error)
return;
if (zs.zs_ptrtbl_len == 0) {
ASSERT(zs.zs_num_blocks == 1);
(void) printf("\tmicrozap: %llu bytes, %llu entries\n",
(u_longlong_t)zs.zs_blocksize,
(u_longlong_t)zs.zs_num_entries);
return;
}
(void) printf("\tFat ZAP stats:\n");
(void) printf("\t\tPointer table:\n");
(void) printf("\t\t\t%llu elements\n",
(u_longlong_t)zs.zs_ptrtbl_len);
(void) printf("\t\t\tzt_blk: %llu\n",
(u_longlong_t)zs.zs_ptrtbl_zt_blk);
(void) printf("\t\t\tzt_numblks: %llu\n",
(u_longlong_t)zs.zs_ptrtbl_zt_numblks);
(void) printf("\t\t\tzt_shift: %llu\n",
(u_longlong_t)zs.zs_ptrtbl_zt_shift);
(void) printf("\t\t\tzt_blks_copied: %llu\n",
(u_longlong_t)zs.zs_ptrtbl_blks_copied);
(void) printf("\t\t\tzt_nextblk: %llu\n",
(u_longlong_t)zs.zs_ptrtbl_nextblk);
(void) printf("\t\tZAP entries: %llu\n",
(u_longlong_t)zs.zs_num_entries);
(void) printf("\t\tLeaf blocks: %llu\n",
(u_longlong_t)zs.zs_num_leafs);
(void) printf("\t\tTotal blocks: %llu\n",
(u_longlong_t)zs.zs_num_blocks);
(void) printf("\t\tzap_block_type: 0x%llx\n",
(u_longlong_t)zs.zs_block_type);
(void) printf("\t\tzap_magic: 0x%llx\n",
(u_longlong_t)zs.zs_magic);
(void) printf("\t\tzap_salt: 0x%llx\n",
(u_longlong_t)zs.zs_salt);
(void) printf("\t\tLeafs with 2^n pointers:\n");
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
dump_histogram(zs.zs_leafs_with_2n_pointers, ZAP_HISTOGRAM_SIZE, 0);
2008-11-20 23:01:55 +03:00
(void) printf("\t\tBlocks with n*5 entries:\n");
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
dump_histogram(zs.zs_blocks_with_n5_entries, ZAP_HISTOGRAM_SIZE, 0);
2008-11-20 23:01:55 +03:00
(void) printf("\t\tBlocks n/10 full:\n");
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
dump_histogram(zs.zs_blocks_n_tenths_full, ZAP_HISTOGRAM_SIZE, 0);
2008-11-20 23:01:55 +03:00
(void) printf("\t\tEntries with n chunks:\n");
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
dump_histogram(zs.zs_entries_using_n_chunks, ZAP_HISTOGRAM_SIZE, 0);
2008-11-20 23:01:55 +03:00
(void) printf("\t\tBuckets with n entries:\n");
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
dump_histogram(zs.zs_buckets_with_n_entries, ZAP_HISTOGRAM_SIZE, 0);
2008-11-20 23:01:55 +03:00
}
/*ARGSUSED*/
static void
dump_none(objset_t *os, uint64_t object, void *data, size_t size)
{
}
/*ARGSUSED*/
static void
dump_unknown(objset_t *os, uint64_t object, void *data, size_t size)
{
(void) printf("\tUNKNOWN OBJECT TYPE\n");
}
2008-11-20 23:01:55 +03:00
/*ARGSUSED*/
void
dump_uint8(objset_t *os, uint64_t object, void *data, size_t size)
{
}
/*ARGSUSED*/
static void
dump_uint64(objset_t *os, uint64_t object, void *data, size_t size)
{
}
/*ARGSUSED*/
static void
dump_zap(objset_t *os, uint64_t object, void *data, size_t size)
{
zap_cursor_t zc;
zap_attribute_t attr;
void *prop;
int i;
dump_zap_stats(os, object);
(void) printf("\n");
for (zap_cursor_init(&zc, os, object);
zap_cursor_retrieve(&zc, &attr) == 0;
zap_cursor_advance(&zc)) {
(void) printf("\t\t%s = ", attr.za_name);
if (attr.za_num_integers == 0) {
(void) printf("\n");
continue;
}
prop = umem_zalloc(attr.za_num_integers *
attr.za_integer_length, UMEM_NOFAIL);
(void) zap_lookup(os, object, attr.za_name,
attr.za_integer_length, attr.za_num_integers, prop);
if (attr.za_integer_length == 1) {
(void) printf("%s", (char *)prop);
} else {
for (i = 0; i < attr.za_num_integers; i++) {
switch (attr.za_integer_length) {
case 2:
(void) printf("%u ",
((uint16_t *)prop)[i]);
break;
case 4:
(void) printf("%u ",
((uint32_t *)prop)[i]);
break;
case 8:
(void) printf("%lld ",
(u_longlong_t)((int64_t *)prop)[i]);
break;
}
}
}
(void) printf("\n");
umem_free(prop, attr.za_num_integers * attr.za_integer_length);
}
zap_cursor_fini(&zc);
}
static void
dump_bpobj(objset_t *os, uint64_t object, void *data, size_t size)
{
bpobj_phys_t *bpop = data;
uint64_t i;
char bytes[32], comp[32], uncomp[32];
if (bpop == NULL)
return;
zdb_nicenum(bpop->bpo_bytes, bytes);
zdb_nicenum(bpop->bpo_comp, comp);
zdb_nicenum(bpop->bpo_uncomp, uncomp);
(void) printf("\t\tnum_blkptrs = %llu\n",
(u_longlong_t)bpop->bpo_num_blkptrs);
(void) printf("\t\tbytes = %s\n", bytes);
if (size >= BPOBJ_SIZE_V1) {
(void) printf("\t\tcomp = %s\n", comp);
(void) printf("\t\tuncomp = %s\n", uncomp);
}
if (size >= sizeof (*bpop)) {
(void) printf("\t\tsubobjs = %llu\n",
(u_longlong_t)bpop->bpo_subobjs);
(void) printf("\t\tnum_subobjs = %llu\n",
(u_longlong_t)bpop->bpo_num_subobjs);
}
if (dump_opt['d'] < 5)
return;
for (i = 0; i < bpop->bpo_num_blkptrs; i++) {
char blkbuf[BP_SPRINTF_LEN];
blkptr_t bp;
int err = dmu_read(os, object,
i * sizeof (bp), sizeof (bp), &bp, 0);
if (err != 0) {
(void) printf("got error %u from dmu_read\n", err);
break;
}
snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), &bp);
(void) printf("\t%s\n", blkbuf);
}
}
/* ARGSUSED */
static void
dump_bpobj_subobjs(objset_t *os, uint64_t object, void *data, size_t size)
{
dmu_object_info_t doi;
int64_t i;
VERIFY0(dmu_object_info(os, object, &doi));
uint64_t *subobjs = kmem_alloc(doi.doi_max_offset, KM_SLEEP);
int err = dmu_read(os, object, 0, doi.doi_max_offset, subobjs, 0);
if (err != 0) {
(void) printf("got error %u from dmu_read\n", err);
kmem_free(subobjs, doi.doi_max_offset);
return;
}
int64_t last_nonzero = -1;
for (i = 0; i < doi.doi_max_offset / 8; i++) {
if (subobjs[i] != 0)
last_nonzero = i;
}
for (i = 0; i <= last_nonzero; i++) {
(void) printf("\t%llu\n", (u_longlong_t)subobjs[i]);
}
kmem_free(subobjs, doi.doi_max_offset);
}
/*ARGSUSED*/
static void
dump_ddt_zap(objset_t *os, uint64_t object, void *data, size_t size)
{
dump_zap_stats(os, object);
/* contents are printed elsewhere, properly decoded */
}
/*ARGSUSED*/
static void
dump_sa_attrs(objset_t *os, uint64_t object, void *data, size_t size)
{
zap_cursor_t zc;
zap_attribute_t attr;
dump_zap_stats(os, object);
(void) printf("\n");
for (zap_cursor_init(&zc, os, object);
zap_cursor_retrieve(&zc, &attr) == 0;
zap_cursor_advance(&zc)) {
(void) printf("\t\t%s = ", attr.za_name);
if (attr.za_num_integers == 0) {
(void) printf("\n");
continue;
}
(void) printf(" %llx : [%d:%d:%d]\n",
(u_longlong_t)attr.za_first_integer,
(int)ATTR_LENGTH(attr.za_first_integer),
(int)ATTR_BSWAP(attr.za_first_integer),
(int)ATTR_NUM(attr.za_first_integer));
}
zap_cursor_fini(&zc);
}
/*ARGSUSED*/
static void
dump_sa_layouts(objset_t *os, uint64_t object, void *data, size_t size)
{
zap_cursor_t zc;
zap_attribute_t attr;
uint16_t *layout_attrs;
int i;
dump_zap_stats(os, object);
(void) printf("\n");
for (zap_cursor_init(&zc, os, object);
zap_cursor_retrieve(&zc, &attr) == 0;
zap_cursor_advance(&zc)) {
(void) printf("\t\t%s = [", attr.za_name);
if (attr.za_num_integers == 0) {
(void) printf("\n");
continue;
}
VERIFY(attr.za_integer_length == 2);
layout_attrs = umem_zalloc(attr.za_num_integers *
attr.za_integer_length, UMEM_NOFAIL);
VERIFY(zap_lookup(os, object, attr.za_name,
attr.za_integer_length,
attr.za_num_integers, layout_attrs) == 0);
for (i = 0; i != attr.za_num_integers; i++)
(void) printf(" %d ", (int)layout_attrs[i]);
(void) printf("]\n");
umem_free(layout_attrs,
attr.za_num_integers * attr.za_integer_length);
}
zap_cursor_fini(&zc);
}
2008-11-20 23:01:55 +03:00
/*ARGSUSED*/
static void
dump_zpldir(objset_t *os, uint64_t object, void *data, size_t size)
{
zap_cursor_t zc;
zap_attribute_t attr;
const char *typenames[] = {
/* 0 */ "not specified",
/* 1 */ "FIFO",
/* 2 */ "Character Device",
/* 3 */ "3 (invalid)",
/* 4 */ "Directory",
/* 5 */ "5 (invalid)",
/* 6 */ "Block Device",
/* 7 */ "7 (invalid)",
/* 8 */ "Regular File",
/* 9 */ "9 (invalid)",
/* 10 */ "Symbolic Link",
/* 11 */ "11 (invalid)",
/* 12 */ "Socket",
/* 13 */ "Door",
/* 14 */ "Event Port",
/* 15 */ "15 (invalid)",
};
dump_zap_stats(os, object);
(void) printf("\n");
for (zap_cursor_init(&zc, os, object);
zap_cursor_retrieve(&zc, &attr) == 0;
zap_cursor_advance(&zc)) {
(void) printf("\t\t%s = %lld (type: %s)\n",
attr.za_name, ZFS_DIRENT_OBJ(attr.za_first_integer),
typenames[ZFS_DIRENT_TYPE(attr.za_first_integer)]);
}
zap_cursor_fini(&zc);
}
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
int
get_dtl_refcount(vdev_t *vd)
{
int refcount = 0;
int c;
if (vd->vdev_ops->vdev_op_leaf) {
space_map_t *sm = vd->vdev_dtl_sm;
if (sm != NULL &&
sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
return (1);
return (0);
}
for (c = 0; c < vd->vdev_children; c++)
refcount += get_dtl_refcount(vd->vdev_child[c]);
return (refcount);
}
int
get_metaslab_refcount(vdev_t *vd)
{
int refcount = 0;
int c, m;
Illumos 4976-4984 - metaslab improvements 4976 zfs should only avoid writing to a failing non-redundant top-level vdev 4978 ztest fails in get_metaslab_refcount() 4979 extend free space histogram to device and pool 4980 metaslabs should have a fragmentation metric 4981 remove fragmented ops vector from block allocator 4982 space_map object should proactively upgrade when feature is enabled 4983 need to collect metaslab information via mdb 4984 device selection should use fragmentation metric Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <adam.leventhal@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/4976 https://www.illumos.org/issues/4978 https://www.illumos.org/issues/4979 https://www.illumos.org/issues/4980 https://www.illumos.org/issues/4981 https://www.illumos.org/issues/4982 https://www.illumos.org/issues/4983 https://www.illumos.org/issues/4984 https://github.com/illumos/illumos-gate/commit/2e4c998 Notes: The "zdb -M" option has been re-tasked to display the new metaslab fragmentation metric and the new "zdb -I" option is used to control the maximum number of in-flight I/Os. The new fragmentation metric is derived from the space map histogram which has been rolled up to the vdev and pool level and is presented to the user via "zpool list". Add a number of module parameters related to the new metaslab weighting logic. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2595
2014-07-20 00:19:24 +04:00
if (vd->vdev_top == vd && !vd->vdev_removing) {
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
for (m = 0; m < vd->vdev_ms_count; m++) {
space_map_t *sm = vd->vdev_ms[m]->ms_sm;
if (sm != NULL &&
sm->sm_dbuf->db_size == sizeof (space_map_phys_t))
refcount++;
}
}
for (c = 0; c < vd->vdev_children; c++)
refcount += get_metaslab_refcount(vd->vdev_child[c]);
return (refcount);
}
static int
verify_spacemap_refcounts(spa_t *spa)
{
uint64_t expected_refcount = 0;
uint64_t actual_refcount;
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
(void) feature_get_refcount(spa,
&spa_feature_table[SPA_FEATURE_SPACEMAP_HISTOGRAM],
&expected_refcount);
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
actual_refcount = get_dtl_refcount(spa->spa_root_vdev);
actual_refcount += get_metaslab_refcount(spa->spa_root_vdev);
if (expected_refcount != actual_refcount) {
(void) printf("space map refcount mismatch: expected %lld != "
"actual %lld\n",
(longlong_t)expected_refcount,
(longlong_t)actual_refcount);
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
return (2);
}
return (0);
}
2008-11-20 23:01:55 +03:00
static void
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
dump_spacemap(objset_t *os, space_map_t *sm)
2008-11-20 23:01:55 +03:00
{
uint64_t alloc, offset, entry;
char *ddata[] = { "ALLOC", "FREE", "CONDENSE", "INVALID",
"INVALID", "INVALID", "INVALID", "INVALID" };
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
if (sm == NULL)
2008-11-20 23:01:55 +03:00
return;
/*
* Print out the freelist entries in both encoded and decoded form.
*/
alloc = 0;
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
for (offset = 0; offset < space_map_length(sm);
offset += sizeof (entry)) {
uint8_t mapshift = sm->sm_shift;
VERIFY0(dmu_read(os, space_map_object(sm), offset,
2009-07-03 02:44:48 +04:00
sizeof (entry), &entry, DMU_READ_PREFETCH));
2008-11-20 23:01:55 +03:00
if (SM_DEBUG_DECODE(entry)) {
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
(void) printf("\t [%6llu] %s: txg %llu, pass %llu\n",
2008-11-20 23:01:55 +03:00
(u_longlong_t)(offset / sizeof (entry)),
ddata[SM_DEBUG_ACTION_DECODE(entry)],
(u_longlong_t)SM_DEBUG_TXG_DECODE(entry),
(u_longlong_t)SM_DEBUG_SYNCPASS_DECODE(entry));
} else {
(void) printf("\t [%6llu] %c range:"
" %010llx-%010llx size: %06llx\n",
2008-11-20 23:01:55 +03:00
(u_longlong_t)(offset / sizeof (entry)),
SM_TYPE_DECODE(entry) == SM_ALLOC ? 'A' : 'F',
(u_longlong_t)((SM_OFFSET_DECODE(entry) <<
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
mapshift) + sm->sm_start),
2008-11-20 23:01:55 +03:00
(u_longlong_t)((SM_OFFSET_DECODE(entry) <<
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
mapshift) + sm->sm_start +
(SM_RUN_DECODE(entry) << mapshift)),
2008-11-20 23:01:55 +03:00
(u_longlong_t)(SM_RUN_DECODE(entry) << mapshift));
if (SM_TYPE_DECODE(entry) == SM_ALLOC)
alloc += SM_RUN_DECODE(entry) << mapshift;
else
alloc -= SM_RUN_DECODE(entry) << mapshift;
}
}
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
if (alloc != space_map_allocated(sm)) {
2008-11-20 23:01:55 +03:00
(void) printf("space_map_object alloc (%llu) INCONSISTENT "
"with space map summary (%llu)\n",
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
(u_longlong_t)space_map_allocated(sm), (u_longlong_t)alloc);
2008-11-20 23:01:55 +03:00
}
}
2009-07-03 02:44:48 +04:00
static void
dump_metaslab_stats(metaslab_t *msp)
{
char maxbuf[32];
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
range_tree_t *rt = msp->ms_tree;
avl_tree_t *t = &msp->ms_size_tree;
int free_pct = range_tree_space(rt) * 100 / msp->ms_size;
2009-07-03 02:44:48 +04:00
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
zdb_nicenum(metaslab_block_maxsize(msp), maxbuf);
2009-07-03 02:44:48 +04:00
(void) printf("\t %25s %10lu %7s %6s %4s %4d%%\n",
2009-07-03 02:44:48 +04:00
"segments", avl_numnodes(t), "maxsize", maxbuf,
"freepct", free_pct);
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
(void) printf("\tIn-memory histogram:\n");
dump_histogram(rt->rt_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
2009-07-03 02:44:48 +04:00
}
2008-11-20 23:01:55 +03:00
static void
dump_metaslab(metaslab_t *msp)
{
vdev_t *vd = msp->ms_group->mg_vd;
spa_t *spa = vd->vdev_spa;
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
space_map_t *sm = msp->ms_sm;
char freebuf[32];
2008-11-20 23:01:55 +03:00
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
zdb_nicenum(msp->ms_size - space_map_allocated(sm), freebuf);
2008-11-20 23:01:55 +03:00
(void) printf(
"\tmetaslab %6llu offset %12llx spacemap %6llu free %5s\n",
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
(u_longlong_t)msp->ms_id, (u_longlong_t)msp->ms_start,
(u_longlong_t)space_map_object(sm), freebuf);
2008-11-20 23:01:55 +03:00
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
if (dump_opt['m'] > 2 && !dump_opt['L']) {
2009-07-03 02:44:48 +04:00
mutex_enter(&msp->ms_lock);
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
metaslab_load_wait(msp);
if (!msp->ms_loaded) {
VERIFY0(metaslab_load(msp));
range_tree_stat_verify(msp->ms_tree);
}
2009-07-03 02:44:48 +04:00
dump_metaslab_stats(msp);
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
metaslab_unload(msp);
2009-07-03 02:44:48 +04:00
mutex_exit(&msp->ms_lock);
}
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
if (dump_opt['m'] > 1 && sm != NULL &&
spa_feature_is_active(spa, SPA_FEATURE_SPACEMAP_HISTOGRAM)) {
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
/*
* The space map histogram represents free space in chunks
* of sm_shift (i.e. bucket 0 refers to 2^sm_shift).
*/
Illumos 4976-4984 - metaslab improvements 4976 zfs should only avoid writing to a failing non-redundant top-level vdev 4978 ztest fails in get_metaslab_refcount() 4979 extend free space histogram to device and pool 4980 metaslabs should have a fragmentation metric 4981 remove fragmented ops vector from block allocator 4982 space_map object should proactively upgrade when feature is enabled 4983 need to collect metaslab information via mdb 4984 device selection should use fragmentation metric Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <adam.leventhal@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/4976 https://www.illumos.org/issues/4978 https://www.illumos.org/issues/4979 https://www.illumos.org/issues/4980 https://www.illumos.org/issues/4981 https://www.illumos.org/issues/4982 https://www.illumos.org/issues/4983 https://www.illumos.org/issues/4984 https://github.com/illumos/illumos-gate/commit/2e4c998 Notes: The "zdb -M" option has been re-tasked to display the new metaslab fragmentation metric and the new "zdb -I" option is used to control the maximum number of in-flight I/Os. The new fragmentation metric is derived from the space map histogram which has been rolled up to the vdev and pool level and is presented to the user via "zpool list". Add a number of module parameters related to the new metaslab weighting logic. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2595
2014-07-20 00:19:24 +04:00
(void) printf("\tOn-disk histogram:\t\tfragmentation %llu\n",
(u_longlong_t)msp->ms_fragmentation);
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
dump_histogram(sm->sm_phys->smp_histogram,
Illumos 4976-4984 - metaslab improvements 4976 zfs should only avoid writing to a failing non-redundant top-level vdev 4978 ztest fails in get_metaslab_refcount() 4979 extend free space histogram to device and pool 4980 metaslabs should have a fragmentation metric 4981 remove fragmented ops vector from block allocator 4982 space_map object should proactively upgrade when feature is enabled 4983 need to collect metaslab information via mdb 4984 device selection should use fragmentation metric Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <adam.leventhal@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/4976 https://www.illumos.org/issues/4978 https://www.illumos.org/issues/4979 https://www.illumos.org/issues/4980 https://www.illumos.org/issues/4981 https://www.illumos.org/issues/4982 https://www.illumos.org/issues/4983 https://www.illumos.org/issues/4984 https://github.com/illumos/illumos-gate/commit/2e4c998 Notes: The "zdb -M" option has been re-tasked to display the new metaslab fragmentation metric and the new "zdb -I" option is used to control the maximum number of in-flight I/Os. The new fragmentation metric is derived from the space map histogram which has been rolled up to the vdev and pool level and is presented to the user via "zpool list". Add a number of module parameters related to the new metaslab weighting logic. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2595
2014-07-20 00:19:24 +04:00
SPACE_MAP_HISTOGRAM_SIZE, sm->sm_shift);
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
}
if (dump_opt['d'] > 5 || dump_opt['m'] > 3) {
ASSERT(msp->ms_size == (1ULL << vd->vdev_ms_shift));
2009-07-03 02:44:48 +04:00
mutex_enter(&msp->ms_lock);
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
dump_spacemap(spa->spa_meta_objset, msp->ms_sm);
2009-07-03 02:44:48 +04:00
mutex_exit(&msp->ms_lock);
}
}
2008-11-20 23:01:55 +03:00
static void
print_vdev_metaslab_header(vdev_t *vd)
{
(void) printf("\tvdev %10llu\n\t%-10s%5llu %-19s %-15s %-10s\n",
(u_longlong_t)vd->vdev_id,
"metaslabs", (u_longlong_t)vd->vdev_ms_count,
"offset", "spacemap", "free");
(void) printf("\t%15s %19s %15s %10s\n",
"---------------", "-------------------",
"---------------", "-------------");
2008-11-20 23:01:55 +03:00
}
Illumos 4976-4984 - metaslab improvements 4976 zfs should only avoid writing to a failing non-redundant top-level vdev 4978 ztest fails in get_metaslab_refcount() 4979 extend free space histogram to device and pool 4980 metaslabs should have a fragmentation metric 4981 remove fragmented ops vector from block allocator 4982 space_map object should proactively upgrade when feature is enabled 4983 need to collect metaslab information via mdb 4984 device selection should use fragmentation metric Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <adam.leventhal@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/4976 https://www.illumos.org/issues/4978 https://www.illumos.org/issues/4979 https://www.illumos.org/issues/4980 https://www.illumos.org/issues/4981 https://www.illumos.org/issues/4982 https://www.illumos.org/issues/4983 https://www.illumos.org/issues/4984 https://github.com/illumos/illumos-gate/commit/2e4c998 Notes: The "zdb -M" option has been re-tasked to display the new metaslab fragmentation metric and the new "zdb -I" option is used to control the maximum number of in-flight I/Os. The new fragmentation metric is derived from the space map histogram which has been rolled up to the vdev and pool level and is presented to the user via "zpool list". Add a number of module parameters related to the new metaslab weighting logic. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2595
2014-07-20 00:19:24 +04:00
static void
dump_metaslab_groups(spa_t *spa)
{
vdev_t *rvd = spa->spa_root_vdev;
metaslab_class_t *mc = spa_normal_class(spa);
uint64_t fragmentation;
int c;
metaslab_class_histogram_verify(mc);
for (c = 0; c < rvd->vdev_children; c++) {
vdev_t *tvd = rvd->vdev_child[c];
metaslab_group_t *mg = tvd->vdev_mg;
if (mg->mg_class != mc)
continue;
metaslab_group_histogram_verify(mg);
mg->mg_fragmentation = metaslab_group_fragmentation(mg);
(void) printf("\tvdev %10llu\t\tmetaslabs%5llu\t\t"
"fragmentation",
(u_longlong_t)tvd->vdev_id,
(u_longlong_t)tvd->vdev_ms_count);
if (mg->mg_fragmentation == ZFS_FRAG_INVALID) {
(void) printf("%3s\n", "-");
} else {
(void) printf("%3llu%%\n",
(u_longlong_t)mg->mg_fragmentation);
}
dump_histogram(mg->mg_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
}
(void) printf("\tpool %s\tfragmentation", spa_name(spa));
fragmentation = metaslab_class_fragmentation(mc);
if (fragmentation == ZFS_FRAG_INVALID)
(void) printf("\t%3s\n", "-");
else
(void) printf("\t%3llu%%\n", (u_longlong_t)fragmentation);
dump_histogram(mc->mc_histogram, RANGE_TREE_HISTOGRAM_SIZE, 0);
}
2008-11-20 23:01:55 +03:00
static void
dump_metaslabs(spa_t *spa)
{
vdev_t *vd, *rvd = spa->spa_root_vdev;
uint64_t m, c = 0, children = rvd->vdev_children;
2008-11-20 23:01:55 +03:00
(void) printf("\nMetaslabs:\n");
if (!dump_opt['d'] && zopt_objects > 0) {
c = zopt_object[0];
if (c >= children)
(void) fatal("bad vdev id: %llu", (u_longlong_t)c);
2008-11-20 23:01:55 +03:00
if (zopt_objects > 1) {
vd = rvd->vdev_child[c];
print_vdev_metaslab_header(vd);
for (m = 1; m < zopt_objects; m++) {
if (zopt_object[m] < vd->vdev_ms_count)
dump_metaslab(
vd->vdev_ms[zopt_object[m]]);
else
(void) fprintf(stderr, "bad metaslab "
"number %llu\n",
(u_longlong_t)zopt_object[m]);
}
(void) printf("\n");
return;
}
children = c + 1;
}
for (; c < children; c++) {
vd = rvd->vdev_child[c];
print_vdev_metaslab_header(vd);
2008-11-20 23:01:55 +03:00
for (m = 0; m < vd->vdev_ms_count; m++)
dump_metaslab(vd->vdev_ms[m]);
(void) printf("\n");
}
}
static void
dump_dde(const ddt_t *ddt, const ddt_entry_t *dde, uint64_t index)
{
const ddt_phys_t *ddp = dde->dde_phys;
const ddt_key_t *ddk = &dde->dde_key;
char *types[4] = { "ditto", "single", "double", "triple" };
char blkbuf[BP_SPRINTF_LEN];
blkptr_t blk;
int p;
for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
if (ddp->ddp_phys_birth == 0)
continue;
ddt_bp_create(ddt->ddt_checksum, ddk, ddp, &blk);
snprintf_blkptr(blkbuf, sizeof (blkbuf), &blk);
(void) printf("index %llx refcnt %llu %s %s\n",
(u_longlong_t)index, (u_longlong_t)ddp->ddp_refcnt,
types[p], blkbuf);
}
}
static void
dump_dedup_ratio(const ddt_stat_t *dds)
{
double rL, rP, rD, D, dedup, compress, copies;
if (dds->dds_blocks == 0)
return;
rL = (double)dds->dds_ref_lsize;
rP = (double)dds->dds_ref_psize;
rD = (double)dds->dds_ref_dsize;
D = (double)dds->dds_dsize;
dedup = rD / D;
compress = rL / rP;
copies = rD / rP;
(void) printf("dedup = %.2f, compress = %.2f, copies = %.2f, "
"dedup * compress / copies = %.2f\n\n",
dedup, compress, copies, dedup * compress / copies);
}
static void
dump_ddt(ddt_t *ddt, enum ddt_type type, enum ddt_class class)
{
char name[DDT_NAMELEN];
ddt_entry_t dde;
uint64_t walk = 0;
dmu_object_info_t doi;
uint64_t count, dspace, mspace;
int error;
error = ddt_object_info(ddt, type, class, &doi);
if (error == ENOENT)
return;
ASSERT(error == 0);
error = ddt_object_count(ddt, type, class, &count);
ASSERT(error == 0);
if (count == 0)
return;
dspace = doi.doi_physical_blocks_512 << 9;
mspace = doi.doi_fill_count * doi.doi_data_block_size;
ddt_object_name(ddt, type, class, name);
(void) printf("%s: %llu entries, size %llu on disk, %llu in core\n",
name,
(u_longlong_t)count,
(u_longlong_t)(dspace / count),
(u_longlong_t)(mspace / count));
if (dump_opt['D'] < 3)
return;
zpool_dump_ddt(NULL, &ddt->ddt_histogram[type][class]);
if (dump_opt['D'] < 4)
return;
if (dump_opt['D'] < 5 && class == DDT_CLASS_UNIQUE)
return;
(void) printf("%s contents:\n\n", name);
while ((error = ddt_object_walk(ddt, type, class, &walk, &dde)) == 0)
dump_dde(ddt, &dde, walk);
ASSERT(error == ENOENT);
(void) printf("\n");
}
static void
dump_all_ddts(spa_t *spa)
{
ddt_histogram_t ddh_total;
ddt_stat_t dds_total;
enum zio_checksum c;
enum ddt_type type;
enum ddt_class class;
bzero(&ddh_total, sizeof (ddt_histogram_t));
bzero(&dds_total, sizeof (ddt_stat_t));
for (c = 0; c < ZIO_CHECKSUM_FUNCTIONS; c++) {
ddt_t *ddt = spa->spa_ddt[c];
for (type = 0; type < DDT_TYPES; type++) {
for (class = 0; class < DDT_CLASSES;
class++) {
dump_ddt(ddt, type, class);
}
}
}
ddt_get_dedup_stats(spa, &dds_total);
if (dds_total.dds_blocks == 0) {
(void) printf("All DDTs are empty\n");
return;
}
(void) printf("\n");
if (dump_opt['D'] > 1) {
(void) printf("DDT histogram (aggregated over all DDTs):\n");
ddt_get_dedup_histogram(spa, &ddh_total);
zpool_dump_ddt(&dds_total, &ddh_total);
}
dump_dedup_ratio(&dds_total);
}
2009-01-16 00:59:39 +03:00
static void
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
dump_dtl_seg(void *arg, uint64_t start, uint64_t size)
2009-01-16 00:59:39 +03:00
{
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
char *prefix = arg;
2009-01-16 00:59:39 +03:00
(void) printf("%s [%llu,%llu) length %llu\n",
prefix,
(u_longlong_t)start,
(u_longlong_t)(start + size),
(u_longlong_t)(size));
}
2008-11-20 23:01:55 +03:00
static void
dump_dtl(vdev_t *vd, int indent)
{
2009-01-16 00:59:39 +03:00
spa_t *spa = vd->vdev_spa;
boolean_t required;
char *name[DTL_TYPES] = { "missing", "partial", "scrub", "outage" };
char prefix[256];
int c, t;
2009-01-16 00:59:39 +03:00
spa_vdev_state_enter(spa, SCL_NONE);
2009-01-16 00:59:39 +03:00
required = vdev_dtl_required(vd);
(void) spa_vdev_state_exit(spa, NULL, 0);
2008-11-20 23:01:55 +03:00
if (indent == 0)
(void) printf("\nDirty time logs:\n\n");
2009-01-16 00:59:39 +03:00
(void) printf("\t%*s%s [%s]\n", indent, "",
vd->vdev_path ? vd->vdev_path :
2009-01-16 00:59:39 +03:00
vd->vdev_parent ? vd->vdev_ops->vdev_op_type : spa_name(spa),
required ? "DTL-required" : "DTL-expendable");
2008-11-20 23:01:55 +03:00
for (t = 0; t < DTL_TYPES; t++) {
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
range_tree_t *rt = vd->vdev_dtl[t];
if (range_tree_space(rt) == 0)
2009-01-16 00:59:39 +03:00
continue;
(void) snprintf(prefix, sizeof (prefix), "\t%*s%s",
indent + 2, "", name[t]);
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
mutex_enter(rt->rt_lock);
range_tree_walk(rt, dump_dtl_seg, prefix);
mutex_exit(rt->rt_lock);
2009-01-16 00:59:39 +03:00
if (dump_opt['d'] > 5 && vd->vdev_children == 0)
dump_spacemap(spa->spa_meta_objset,
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
vd->vdev_dtl_sm);
2008-11-20 23:01:55 +03:00
}
for (c = 0; c < vd->vdev_children; c++)
2008-11-20 23:01:55 +03:00
dump_dtl(vd->vdev_child[c], indent + 4);
}
static void
dump_history(spa_t *spa)
{
nvlist_t **events = NULL;
char *buf;
uint64_t resid, len, off = 0;
uint_t num = 0;
int error;
time_t tsec;
struct tm t;
char tbuf[30];
char internalstr[MAXPATHLEN];
int i;
if ((buf = malloc(SPA_OLD_MAXBLOCKSIZE)) == NULL) {
(void) fprintf(stderr, "%s: unable to allocate I/O buffer\n",
__func__);
return;
}
do {
len = SPA_OLD_MAXBLOCKSIZE;
if ((error = spa_history_get(spa, &off, &len, buf)) != 0) {
(void) fprintf(stderr, "Unable to read history: "
"error %d\n", error);
free(buf);
return;
}
if (zpool_history_unpack(buf, len, &resid, &events, &num) != 0)
break;
off -= resid;
} while (len != 0);
(void) printf("\nHistory:\n");
for (i = 0; i < num; i++) {
uint64_t time, txg, ievent;
char *cmd, *intstr;
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 15:45:09 +04:00
boolean_t printed = B_FALSE;
if (nvlist_lookup_uint64(events[i], ZPOOL_HIST_TIME,
&time) != 0)
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 15:45:09 +04:00
goto next;
if (nvlist_lookup_string(events[i], ZPOOL_HIST_CMD,
&cmd) != 0) {
if (nvlist_lookup_uint64(events[i],
ZPOOL_HIST_INT_EVENT, &ievent) != 0)
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 15:45:09 +04:00
goto next;
verify(nvlist_lookup_uint64(events[i],
ZPOOL_HIST_TXG, &txg) == 0);
verify(nvlist_lookup_string(events[i],
ZPOOL_HIST_INT_STR, &intstr) == 0);
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 15:45:09 +04:00
if (ievent >= ZFS_NUM_LEGACY_HISTORY_EVENTS)
goto next;
(void) snprintf(internalstr,
sizeof (internalstr),
"[internal %s txg:%lld] %s",
zfs_history_event_names[ievent],
(longlong_t)txg, intstr);
cmd = internalstr;
}
tsec = time;
(void) localtime_r(&tsec, &t);
(void) strftime(tbuf, sizeof (tbuf), "%F.%T", &t);
(void) printf("%s %s\n", tbuf, cmd);
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 15:45:09 +04:00
printed = B_TRUE;
next:
if (dump_opt['h'] > 1) {
if (!printed)
(void) printf("unrecognized record:\n");
dump_nvlist(events[i], 2);
}
}
free(buf);
}
2008-11-20 23:01:55 +03:00
/*ARGSUSED*/
static void
dump_dnode(objset_t *os, uint64_t object, void *data, size_t size)
{
}
static uint64_t
blkid2offset(const dnode_phys_t *dnp, const blkptr_t *bp,
const zbookmark_phys_t *zb)
2008-11-20 23:01:55 +03:00
{
if (dnp == NULL) {
ASSERT(zb->zb_level < 0);
if (zb->zb_object == 0)
return (zb->zb_blkid);
return (zb->zb_blkid * BP_GET_LSIZE(bp));
}
ASSERT(zb->zb_level >= 0);
2008-11-20 23:01:55 +03:00
return ((zb->zb_blkid <<
(zb->zb_level * (dnp->dn_indblkshift - SPA_BLKPTRSHIFT))) *
2008-11-20 23:01:55 +03:00
dnp->dn_datablkszsec << SPA_MINBLOCKSHIFT);
}
static void
snprintf_blkptr_compact(char *blkbuf, size_t buflen, const blkptr_t *bp)
2008-11-20 23:01:55 +03:00
{
const dva_t *dva = bp->blk_dva;
int ndvas = dump_opt['d'] > 5 ? BP_GET_NDVAS(bp) : 1;
int i;
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
if (dump_opt['b'] >= 6) {
snprintf_blkptr(blkbuf, buflen, bp);
return;
}
2008-11-20 23:01:55 +03:00
if (BP_IS_EMBEDDED(bp)) {
(void) sprintf(blkbuf,
"EMBEDDED et=%u %llxL/%llxP B=%llu",
(int)BPE_GET_ETYPE(bp),
(u_longlong_t)BPE_GET_LSIZE(bp),
(u_longlong_t)BPE_GET_PSIZE(bp),
(u_longlong_t)bp->blk_birth);
return;
}
2008-11-20 23:01:55 +03:00
blkbuf[0] = '\0';
for (i = 0; i < ndvas; i++)
(void) snprintf(blkbuf + strlen(blkbuf),
buflen - strlen(blkbuf), "%llu:%llx:%llx ",
2008-11-20 23:01:55 +03:00
(u_longlong_t)DVA_GET_VDEV(&dva[i]),
(u_longlong_t)DVA_GET_OFFSET(&dva[i]),
(u_longlong_t)DVA_GET_ASIZE(&dva[i]));
if (BP_IS_HOLE(bp)) {
(void) snprintf(blkbuf + strlen(blkbuf),
buflen - strlen(blkbuf),
"%llxL B=%llu",
(u_longlong_t)BP_GET_LSIZE(bp),
(u_longlong_t)bp->blk_birth);
} else {
(void) snprintf(blkbuf + strlen(blkbuf),
buflen - strlen(blkbuf),
"%llxL/%llxP F=%llu B=%llu/%llu",
(u_longlong_t)BP_GET_LSIZE(bp),
(u_longlong_t)BP_GET_PSIZE(bp),
(u_longlong_t)BP_GET_FILL(bp),
(u_longlong_t)bp->blk_birth,
(u_longlong_t)BP_PHYSICAL_BIRTH(bp));
}
2008-11-20 23:01:55 +03:00
}
static void
print_indirect(blkptr_t *bp, const zbookmark_phys_t *zb,
const dnode_phys_t *dnp)
2008-11-20 23:01:55 +03:00
{
char blkbuf[BP_SPRINTF_LEN];
2008-11-20 23:01:55 +03:00
int l;
if (!BP_IS_EMBEDDED(bp)) {
ASSERT3U(BP_GET_TYPE(bp), ==, dnp->dn_type);
ASSERT3U(BP_GET_LEVEL(bp), ==, zb->zb_level);
}
2008-11-20 23:01:55 +03:00
(void) printf("%16llx ", (u_longlong_t)blkid2offset(dnp, bp, zb));
2008-11-20 23:01:55 +03:00
ASSERT(zb->zb_level >= 0);
for (l = dnp->dn_nlevels - 1; l >= -1; l--) {
if (l == zb->zb_level) {
(void) printf("L%llx", (u_longlong_t)zb->zb_level);
2008-11-20 23:01:55 +03:00
} else {
(void) printf(" ");
2008-11-20 23:01:55 +03:00
}
}
snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp);
(void) printf("%s\n", blkbuf);
}
static int
visit_indirect(spa_t *spa, const dnode_phys_t *dnp,
blkptr_t *bp, const zbookmark_phys_t *zb)
{
int err = 0;
if (bp->blk_birth == 0)
return (0);
print_indirect(bp, zb, dnp);
if (BP_GET_LEVEL(bp) > 0 && !BP_IS_HOLE(bp)) {
arc_flags_t flags = ARC_FLAG_WAIT;
int i;
blkptr_t *cbp;
int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
arc_buf_t *buf;
uint64_t fill = 0;
err = arc_read(NULL, spa, bp, arc_getbuf_func, &buf,
ZIO_PRIORITY_ASYNC_READ, ZIO_FLAG_CANFAIL, &flags, zb);
if (err)
return (err);
ASSERT(buf->b_data);
/* recursively visit blocks below this */
cbp = buf->b_data;
for (i = 0; i < epb; i++, cbp++) {
zbookmark_phys_t czb;
SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
zb->zb_level - 1,
zb->zb_blkid * epb + i);
err = visit_indirect(spa, dnp, cbp, &czb);
if (err)
break;
fill += BP_GET_FILL(cbp);
}
2009-01-16 00:59:39 +03:00
if (!err)
ASSERT3U(fill, ==, BP_GET_FILL(bp));
OpenZFS 6950 - ARC should cache compressed data Authored by: George Wilson <george.wilson@delphix.com> Reviewed by: Prakash Surya <prakash.surya@delphix.com> Reviewed by: Dan Kimmel <dan.kimmel@delphix.com> Reviewed by: Matt Ahrens <mahrens@delphix.com> Reviewed by: Paul Dagnelie <pcd@delphix.com> Reviewed by: Tom Caputi <tcaputi@datto.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Ported by: David Quigley <david.quigley@intel.com> This review covers the reading and writing of compressed arc headers, sharing data between the arc_hdr_t and the arc_buf_t, and the implementation of a new dbuf cache to keep frequently access data uncompressed. I've added a new member to l1 arc hdr called b_pdata. The b_pdata always hangs off the arc_buf_hdr_t (if an L1 hdr is in use) and points to the physical block for that DVA. The physical block may or may not be compressed. If compressed arc is enabled and the block on-disk is compressed, then the b_pdata will match the block on-disk and remain compressed in memory. If the block on disk is not compressed, then neither will the b_pdata. Lastly, if compressed arc is disabled, then b_pdata will always be an uncompressed version of the on-disk block. Typically the arc will cache only the arc_buf_hdr_t and will aggressively evict any arc_buf_t's that are no longer referenced. This means that the arc will primarily have compressed blocks as the arc_buf_t's are considered overhead and are always uncompressed. When a consumer reads a block we first look to see if the arc_buf_hdr_t is cached. If the hdr is cached then we allocate a new arc_buf_t and decompress the b_pdata contents into the arc_buf_t's b_data. If the hdr already has a arc_buf_t, then we will allocate an additional arc_buf_t and bcopy the uncompressed contents from the first arc_buf_t to the new one. Writing to the compressed arc requires that we first discard the b_pdata since the physical block is about to be rewritten. The new data contents will be passed in via an arc_buf_t (uncompressed) and during the I/O pipeline stages we will copy the physical block contents to a newly allocated b_pdata. When an l2arc is inuse it will also take advantage of the b_pdata. Now the l2arc will always write the contents of b_pdata to the l2arc. This means that when compressed arc is enabled that the l2arc blocks are identical to those stored in the main data pool. This provides a significant advantage since we can leverage the bp's checksum when reading from the l2arc to determine if the contents are valid. If the compressed arc is disabled, then we must first transform the read block to look like the physical block in the main data pool before comparing the checksum and determining it's valid. OpenZFS-issue: https://www.illumos.org/issues/6950 OpenZFS-commit: https://github.com/openzfs/openzfs/commit/7fc10f0 Issue #5078
2016-06-02 07:04:53 +03:00
arc_buf_destroy(buf, &buf);
2008-11-20 23:01:55 +03:00
}
return (err);
2008-11-20 23:01:55 +03:00
}
/*ARGSUSED*/
static void
dump_indirect(dnode_t *dn)
2008-11-20 23:01:55 +03:00
{
dnode_phys_t *dnp = dn->dn_phys;
int j;
zbookmark_phys_t czb;
2008-11-20 23:01:55 +03:00
(void) printf("Indirect blocks:\n");
SET_BOOKMARK(&czb, dmu_objset_id(dn->dn_objset),
dn->dn_object, dnp->dn_nlevels - 1, 0);
for (j = 0; j < dnp->dn_nblkptr; j++) {
czb.zb_blkid = j;
(void) visit_indirect(dmu_objset_spa(dn->dn_objset), dnp,
&dnp->dn_blkptr[j], &czb);
}
2008-11-20 23:01:55 +03:00
(void) printf("\n");
}
/*ARGSUSED*/
static void
dump_dsl_dir(objset_t *os, uint64_t object, void *data, size_t size)
{
dsl_dir_phys_t *dd = data;
time_t crtime;
char nice[32];
2008-11-20 23:01:55 +03:00
if (dd == NULL)
return;
ASSERT3U(size, >=, sizeof (dsl_dir_phys_t));
crtime = dd->dd_creation_time;
(void) printf("\t\tcreation_time = %s", ctime(&crtime));
(void) printf("\t\thead_dataset_obj = %llu\n",
(u_longlong_t)dd->dd_head_dataset_obj);
(void) printf("\t\tparent_dir_obj = %llu\n",
(u_longlong_t)dd->dd_parent_obj);
(void) printf("\t\torigin_obj = %llu\n",
(u_longlong_t)dd->dd_origin_obj);
(void) printf("\t\tchild_dir_zapobj = %llu\n",
(u_longlong_t)dd->dd_child_dir_zapobj);
zdb_nicenum(dd->dd_used_bytes, nice);
(void) printf("\t\tused_bytes = %s\n", nice);
zdb_nicenum(dd->dd_compressed_bytes, nice);
(void) printf("\t\tcompressed_bytes = %s\n", nice);
zdb_nicenum(dd->dd_uncompressed_bytes, nice);
(void) printf("\t\tuncompressed_bytes = %s\n", nice);
zdb_nicenum(dd->dd_quota, nice);
(void) printf("\t\tquota = %s\n", nice);
zdb_nicenum(dd->dd_reserved, nice);
(void) printf("\t\treserved = %s\n", nice);
2008-11-20 23:01:55 +03:00
(void) printf("\t\tprops_zapobj = %llu\n",
(u_longlong_t)dd->dd_props_zapobj);
(void) printf("\t\tdeleg_zapobj = %llu\n",
(u_longlong_t)dd->dd_deleg_zapobj);
(void) printf("\t\tflags = %llx\n",
(u_longlong_t)dd->dd_flags);
#define DO(which) \
zdb_nicenum(dd->dd_used_breakdown[DD_USED_ ## which], nice); \
(void) printf("\t\tused_breakdown[" #which "] = %s\n", nice)
DO(HEAD);
DO(SNAP);
DO(CHILD);
DO(CHILD_RSRV);
DO(REFRSRV);
#undef DO
2008-11-20 23:01:55 +03:00
}
/*ARGSUSED*/
static void
dump_dsl_dataset(objset_t *os, uint64_t object, void *data, size_t size)
{
dsl_dataset_phys_t *ds = data;
time_t crtime;
char used[32], compressed[32], uncompressed[32], unique[32];
2008-11-20 23:01:55 +03:00
char blkbuf[BP_SPRINTF_LEN];
if (ds == NULL)
return;
ASSERT(size == sizeof (*ds));
crtime = ds->ds_creation_time;
zdb_nicenum(ds->ds_referenced_bytes, used);
zdb_nicenum(ds->ds_compressed_bytes, compressed);
zdb_nicenum(ds->ds_uncompressed_bytes, uncompressed);
zdb_nicenum(ds->ds_unique_bytes, unique);
snprintf_blkptr(blkbuf, sizeof (blkbuf), &ds->ds_bp);
2008-11-20 23:01:55 +03:00
(void) printf("\t\tdir_obj = %llu\n",
2008-11-20 23:01:55 +03:00
(u_longlong_t)ds->ds_dir_obj);
(void) printf("\t\tprev_snap_obj = %llu\n",
(u_longlong_t)ds->ds_prev_snap_obj);
(void) printf("\t\tprev_snap_txg = %llu\n",
(u_longlong_t)ds->ds_prev_snap_txg);
(void) printf("\t\tnext_snap_obj = %llu\n",
(u_longlong_t)ds->ds_next_snap_obj);
(void) printf("\t\tsnapnames_zapobj = %llu\n",
(u_longlong_t)ds->ds_snapnames_zapobj);
(void) printf("\t\tnum_children = %llu\n",
(u_longlong_t)ds->ds_num_children);
2009-08-18 22:43:27 +04:00
(void) printf("\t\tuserrefs_obj = %llu\n",
(u_longlong_t)ds->ds_userrefs_obj);
2008-11-20 23:01:55 +03:00
(void) printf("\t\tcreation_time = %s", ctime(&crtime));
(void) printf("\t\tcreation_txg = %llu\n",
(u_longlong_t)ds->ds_creation_txg);
(void) printf("\t\tdeadlist_obj = %llu\n",
(u_longlong_t)ds->ds_deadlist_obj);
(void) printf("\t\tused_bytes = %s\n", used);
(void) printf("\t\tcompressed_bytes = %s\n", compressed);
(void) printf("\t\tuncompressed_bytes = %s\n", uncompressed);
(void) printf("\t\tunique = %s\n", unique);
(void) printf("\t\tfsid_guid = %llu\n",
(u_longlong_t)ds->ds_fsid_guid);
(void) printf("\t\tguid = %llu\n",
(u_longlong_t)ds->ds_guid);
(void) printf("\t\tflags = %llx\n",
(u_longlong_t)ds->ds_flags);
(void) printf("\t\tnext_clones_obj = %llu\n",
(u_longlong_t)ds->ds_next_clones_obj);
(void) printf("\t\tprops_obj = %llu\n",
(u_longlong_t)ds->ds_props_obj);
2008-11-20 23:01:55 +03:00
(void) printf("\t\tbp = %s\n", blkbuf);
}
/* ARGSUSED */
static int
dump_bptree_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
{
char blkbuf[BP_SPRINTF_LEN];
if (bp->blk_birth != 0) {
snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
(void) printf("\t%s\n", blkbuf);
}
return (0);
}
static void
dump_bptree(objset_t *os, uint64_t obj, char *name)
{
char bytes[32];
bptree_phys_t *bt;
dmu_buf_t *db;
if (dump_opt['d'] < 3)
return;
VERIFY3U(0, ==, dmu_bonus_hold(os, obj, FTAG, &db));
bt = db->db_data;
zdb_nicenum(bt->bt_bytes, bytes);
(void) printf("\n %s: %llu datasets, %s\n",
name, (unsigned long long)(bt->bt_end - bt->bt_begin), bytes);
dmu_buf_rele(db, FTAG);
if (dump_opt['d'] < 5)
return;
(void) printf("\n");
(void) bptree_iterate(os, obj, B_FALSE, dump_bptree_cb, NULL, NULL);
}
/* ARGSUSED */
static int
dump_bpobj_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
{
char blkbuf[BP_SPRINTF_LEN];
ASSERT(bp->blk_birth != 0);
snprintf_blkptr_compact(blkbuf, sizeof (blkbuf), bp);
(void) printf("\t%s\n", blkbuf);
return (0);
}
2008-11-20 23:01:55 +03:00
static void
dump_full_bpobj(bpobj_t *bpo, char *name, int indent)
2008-11-20 23:01:55 +03:00
{
char bytes[32];
char comp[32];
char uncomp[32];
uint64_t i;
2008-11-20 23:01:55 +03:00
if (dump_opt['d'] < 3)
return;
zdb_nicenum(bpo->bpo_phys->bpo_bytes, bytes);
if (bpo->bpo_havesubobj && bpo->bpo_phys->bpo_subobjs != 0) {
zdb_nicenum(bpo->bpo_phys->bpo_comp, comp);
zdb_nicenum(bpo->bpo_phys->bpo_uncomp, uncomp);
(void) printf(" %*s: object %llu, %llu local blkptrs, "
"%llu subobjs in object, %llu, %s (%s/%s comp)\n",
indent * 8, name,
(u_longlong_t)bpo->bpo_object,
(u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
(u_longlong_t)bpo->bpo_phys->bpo_num_subobjs,
(u_longlong_t)bpo->bpo_phys->bpo_subobjs,
2008-11-20 23:01:55 +03:00
bytes, comp, uncomp);
for (i = 0; i < bpo->bpo_phys->bpo_num_subobjs; i++) {
uint64_t subobj;
bpobj_t subbpo;
int error;
VERIFY0(dmu_read(bpo->bpo_os,
bpo->bpo_phys->bpo_subobjs,
i * sizeof (subobj), sizeof (subobj), &subobj, 0));
error = bpobj_open(&subbpo, bpo->bpo_os, subobj);
if (error != 0) {
(void) printf("ERROR %u while trying to open "
"subobj id %llu\n",
error, (u_longlong_t)subobj);
continue;
}
dump_full_bpobj(&subbpo, "subobj", indent + 1);
bpobj_close(&subbpo);
}
2008-11-20 23:01:55 +03:00
} else {
(void) printf(" %*s: object %llu, %llu blkptrs, %s\n",
indent * 8, name,
(u_longlong_t)bpo->bpo_object,
(u_longlong_t)bpo->bpo_phys->bpo_num_blkptrs,
bytes);
2008-11-20 23:01:55 +03:00
}
if (dump_opt['d'] < 5)
2008-11-20 23:01:55 +03:00
return;
if (indent == 0) {
(void) bpobj_iterate_nofree(bpo, dump_bpobj_cb, NULL, NULL);
(void) printf("\n");
}
}
2008-11-20 23:01:55 +03:00
static void
dump_deadlist(dsl_deadlist_t *dl)
{
dsl_deadlist_entry_t *dle;
uint64_t unused;
char bytes[32];
char comp[32];
char uncomp[32];
if (dump_opt['d'] < 3)
return;
2008-11-20 23:01:55 +03:00
if (dl->dl_oldfmt) {
dump_full_bpobj(&dl->dl_bpobj, "old-format deadlist", 0);
return;
}
zdb_nicenum(dl->dl_phys->dl_used, bytes);
zdb_nicenum(dl->dl_phys->dl_comp, comp);
zdb_nicenum(dl->dl_phys->dl_uncomp, uncomp);
(void) printf("\n Deadlist: %s (%s/%s comp)\n",
bytes, comp, uncomp);
if (dump_opt['d'] < 4)
return;
(void) printf("\n");
/* force the tree to be loaded */
dsl_deadlist_space_range(dl, 0, UINT64_MAX, &unused, &unused, &unused);
for (dle = avl_first(&dl->dl_tree); dle;
dle = AVL_NEXT(&dl->dl_tree, dle)) {
if (dump_opt['d'] >= 5) {
char buf[128];
(void) snprintf(buf, sizeof (buf),
"mintxg %llu -> obj %llu",
(longlong_t)dle->dle_mintxg,
(longlong_t)dle->dle_bpobj.bpo_object);
dump_full_bpobj(&dle->dle_bpobj, buf, 0);
} else {
(void) printf("mintxg %llu -> obj %llu\n",
(longlong_t)dle->dle_mintxg,
(longlong_t)dle->dle_bpobj.bpo_object);
}
}
2008-11-20 23:01:55 +03:00
}
static avl_tree_t idx_tree;
static avl_tree_t domain_tree;
static boolean_t fuid_table_loaded;
static objset_t *sa_os = NULL;
static sa_attr_type_t *sa_attr_table = NULL;
static int
open_objset(const char *path, dmu_objset_type_t type, void *tag, objset_t **osp)
{
int err;
uint64_t sa_attrs = 0;
uint64_t version = 0;
VERIFY3P(sa_os, ==, NULL);
err = dmu_objset_own(path, type, B_TRUE, tag, osp);
if (err != 0) {
(void) fprintf(stderr, "failed to own dataset '%s': %s\n", path,
strerror(err));
return (err);
}
if (dmu_objset_type(*osp) == DMU_OST_ZFS) {
(void) zap_lookup(*osp, MASTER_NODE_OBJ, ZPL_VERSION_STR,
8, 1, &version);
if (version >= ZPL_VERSION_SA) {
(void) zap_lookup(*osp, MASTER_NODE_OBJ, ZFS_SA_ATTRS,
8, 1, &sa_attrs);
}
err = sa_setup(*osp, sa_attrs, zfs_attr_table, ZPL_END,
&sa_attr_table);
if (err != 0) {
(void) fprintf(stderr, "sa_setup failed: %s\n",
strerror(err));
dmu_objset_disown(*osp, tag);
*osp = NULL;
}
}
sa_os = *osp;
return (0);
}
static void
close_objset(objset_t *os, void *tag)
{
VERIFY3P(os, ==, sa_os);
if (os->os_sa != NULL)
sa_tear_down(os);
dmu_objset_disown(os, tag);
sa_attr_table = NULL;
sa_os = NULL;
}
2008-11-20 23:01:55 +03:00
static void
fuid_table_destroy(void)
2008-11-20 23:01:55 +03:00
{
if (fuid_table_loaded) {
zfs_fuid_table_destroy(&idx_tree, &domain_tree);
fuid_table_loaded = B_FALSE;
}
}
/*
* print uid or gid information.
* For normal POSIX id just the id is printed in decimal format.
* For CIFS files with FUID the fuid is printed in hex followed by
* the domain-rid string.
2008-11-20 23:01:55 +03:00
*/
static void
print_idstr(uint64_t id, const char *id_type)
{
if (FUID_INDEX(id)) {
char *domain;
domain = zfs_fuid_idx_domain(&idx_tree, FUID_INDEX(id));
(void) printf("\t%s %llx [%s-%d]\n", id_type,
(u_longlong_t)id, domain, (int)FUID_RID(id));
} else {
(void) printf("\t%s %llu\n", id_type, (u_longlong_t)id);
}
}
static void
dump_uidgid(objset_t *os, uint64_t uid, uint64_t gid)
2008-11-20 23:01:55 +03:00
{
uint32_t uid_idx, gid_idx;
uid_idx = FUID_INDEX(uid);
gid_idx = FUID_INDEX(gid);
2008-11-20 23:01:55 +03:00
/* Load domain table, if not already loaded */
if (!fuid_table_loaded && (uid_idx || gid_idx)) {
uint64_t fuid_obj;
/* first find the fuid object. It lives in the master node */
VERIFY(zap_lookup(os, MASTER_NODE_OBJ, ZFS_FUID_TABLES,
8, 1, &fuid_obj) == 0);
2009-07-03 02:44:48 +04:00
zfs_fuid_avl_tree_create(&idx_tree, &domain_tree);
2008-11-20 23:01:55 +03:00
(void) zfs_fuid_table_load(os, fuid_obj,
&idx_tree, &domain_tree);
fuid_table_loaded = B_TRUE;
}
print_idstr(uid, "uid");
print_idstr(gid, "gid");
2008-11-20 23:01:55 +03:00
}
static void
dump_znode_sa_xattr(sa_handle_t *hdl)
{
nvlist_t *sa_xattr;
nvpair_t *elem = NULL;
int sa_xattr_size = 0;
int sa_xattr_entries = 0;
int error;
char *sa_xattr_packed;
error = sa_size(hdl, sa_attr_table[ZPL_DXATTR], &sa_xattr_size);
if (error || sa_xattr_size == 0)
return;
sa_xattr_packed = malloc(sa_xattr_size);
if (sa_xattr_packed == NULL)
return;
error = sa_lookup(hdl, sa_attr_table[ZPL_DXATTR],
sa_xattr_packed, sa_xattr_size);
if (error) {
free(sa_xattr_packed);
return;
}
error = nvlist_unpack(sa_xattr_packed, sa_xattr_size, &sa_xattr, 0);
if (error) {
free(sa_xattr_packed);
return;
}
while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL)
sa_xattr_entries++;
(void) printf("\tSA xattrs: %d bytes, %d entries\n\n",
sa_xattr_size, sa_xattr_entries);
while ((elem = nvlist_next_nvpair(sa_xattr, elem)) != NULL) {
uchar_t *value;
uint_t cnt, idx;
(void) printf("\t\t%s = ", nvpair_name(elem));
nvpair_value_byte_array(elem, &value, &cnt);
for (idx = 0; idx < cnt; ++idx) {
if (isprint(value[idx]))
(void) putchar(value[idx]);
else
(void) printf("\\%3.3o", value[idx]);
}
(void) putchar('\n');
}
nvlist_free(sa_xattr);
free(sa_xattr_packed);
}
2008-11-20 23:01:55 +03:00
/*ARGSUSED*/
static void
dump_znode(objset_t *os, uint64_t object, void *data, size_t size)
{
char path[MAXPATHLEN * 2]; /* allow for xattr and failure prefix */
sa_handle_t *hdl;
uint64_t xattr, rdev, gen;
uint64_t uid, gid, mode, fsize, parent, links;
uint64_t pflags;
uint64_t acctm[2], modtm[2], chgtm[2], crtm[2];
time_t z_crtime, z_atime, z_mtime, z_ctime;
sa_bulk_attr_t bulk[12];
int idx = 0;
2008-11-20 23:01:55 +03:00
int error;
VERIFY3P(os, ==, sa_os);
if (sa_handle_get(os, object, NULL, SA_HDL_PRIVATE, &hdl)) {
(void) printf("Failed to get handle for SA znode\n");
return;
}
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_UID], NULL, &uid, 8);
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GID], NULL, &gid, 8);
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_LINKS], NULL,
&links, 8);
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_GEN], NULL, &gen, 8);
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MODE], NULL,
&mode, 8);
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_PARENT],
NULL, &parent, 8);
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_SIZE], NULL,
&fsize, 8);
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_ATIME], NULL,
acctm, 16);
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_MTIME], NULL,
modtm, 16);
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CRTIME], NULL,
crtm, 16);
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_CTIME], NULL,
chgtm, 16);
SA_ADD_BULK_ATTR(bulk, idx, sa_attr_table[ZPL_FLAGS], NULL,
&pflags, 8);
if (sa_bulk_lookup(hdl, bulk, idx)) {
(void) sa_handle_destroy(hdl);
return;
}
2008-11-20 23:01:55 +03:00
z_crtime = (time_t)crtm[0];
z_atime = (time_t)acctm[0];
z_mtime = (time_t)modtm[0];
z_ctime = (time_t)chgtm[0];
2008-11-20 23:01:55 +03:00
if (dump_opt['d'] > 4) {
error = zfs_obj_to_path(os, object, path, sizeof (path));
if (error != 0) {
(void) snprintf(path, sizeof (path),
"\?\?\?<object#%llu>", (u_longlong_t)object);
}
(void) printf("\tpath %s\n", path);
}
dump_uidgid(os, uid, gid);
2008-11-20 23:01:55 +03:00
(void) printf("\tatime %s", ctime(&z_atime));
(void) printf("\tmtime %s", ctime(&z_mtime));
(void) printf("\tctime %s", ctime(&z_ctime));
(void) printf("\tcrtime %s", ctime(&z_crtime));
(void) printf("\tgen %llu\n", (u_longlong_t)gen);
(void) printf("\tmode %llo\n", (u_longlong_t)mode);
(void) printf("\tsize %llu\n", (u_longlong_t)fsize);
(void) printf("\tparent %llu\n", (u_longlong_t)parent);
(void) printf("\tlinks %llu\n", (u_longlong_t)links);
(void) printf("\tpflags %llx\n", (u_longlong_t)pflags);
if (sa_lookup(hdl, sa_attr_table[ZPL_XATTR], &xattr,
sizeof (uint64_t)) == 0)
(void) printf("\txattr %llu\n", (u_longlong_t)xattr);
if (sa_lookup(hdl, sa_attr_table[ZPL_RDEV], &rdev,
sizeof (uint64_t)) == 0)
(void) printf("\trdev 0x%016llx\n", (u_longlong_t)rdev);
dump_znode_sa_xattr(hdl);
sa_handle_destroy(hdl);
2008-11-20 23:01:55 +03:00
}
/*ARGSUSED*/
static void
dump_acl(objset_t *os, uint64_t object, void *data, size_t size)
{
}
/*ARGSUSED*/
static void
dump_dmu_objset(objset_t *os, uint64_t object, void *data, size_t size)
{
}
static object_viewer_t *object_viewer[DMU_OT_NUMTYPES + 1] = {
2008-11-20 23:01:55 +03:00
dump_none, /* unallocated */
dump_zap, /* object directory */
dump_uint64, /* object array */
dump_none, /* packed nvlist */
dump_packed_nvlist, /* packed nvlist size */
dump_none, /* bpobj */
dump_bpobj, /* bpobj header */
2008-11-20 23:01:55 +03:00
dump_none, /* SPA space map header */
dump_none, /* SPA space map */
dump_none, /* ZIL intent log */
dump_dnode, /* DMU dnode */
dump_dmu_objset, /* DMU objset */
dump_dsl_dir, /* DSL directory */
dump_zap, /* DSL directory child map */
dump_zap, /* DSL dataset snap map */
dump_zap, /* DSL props */
dump_dsl_dataset, /* DSL dataset */
dump_znode, /* ZFS znode */
dump_acl, /* ZFS V0 ACL */
dump_uint8, /* ZFS plain file */
dump_zpldir, /* ZFS directory */
dump_zap, /* ZFS master node */
dump_zap, /* ZFS delete queue */
dump_uint8, /* zvol object */
dump_zap, /* zvol prop */
dump_uint8, /* other uint8[] */
dump_uint64, /* other uint64[] */
dump_zap, /* other ZAP */
dump_zap, /* persistent error log */
dump_uint8, /* SPA history */
Illumos #2882, #2883, #2900 2882 implement libzfs_core 2883 changing "canmount" property to "on" should not always remount dataset 2900 "zfs snapshot" should be able to create multiple, arbitrary snapshots at once Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Chris Siden <christopher.siden@delphix.com> Reviewed by: Garrett D'Amore <garrett@damore.org> Reviewed by: Bill Pijewski <wdp@joyent.com> Reviewed by: Dan Kruchinin <dan.kruchinin@gmail.com> Approved by: Eric Schrock <Eric.Schrock@delphix.com> References: https://www.illumos.org/issues/2882 https://www.illumos.org/issues/2883 https://www.illumos.org/issues/2900 illumos/illumos-gate@4445fffbbb1ea25fd0e9ea68b9380dd7a6709025 Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1293 Porting notes: WARNING: This patch changes the user/kernel ABI. That means that the zfs/zpool utilities built from master are NOT compatible with the 0.6.2 kernel modules. Ensure you load the matching kernel modules from master after updating the utilities. Otherwise the zfs/zpool commands will be unable to interact with your pool and you will see errors similar to the following: $ zpool list failed to read pool configuration: bad address no pools available $ zfs list no datasets available Add zvol minor device creation to the new zfs_snapshot_nvl function. Remove the logging of the "release" operation in dsl_dataset_user_release_sync(). The logging caused a null dereference because ds->ds_dir is zeroed in dsl_dataset_destroy_sync() and the logging functions try to get the ds name via the dsl_dataset_name() function. I've got no idea why this particular code would have worked in Illumos. This code has subsequently been completely reworked in Illumos commit 3b2aab1 (3464 zfs synctask code needs restructuring). Squash some "may be used uninitialized" warning/erorrs. Fix some printf format warnings for %lld and %llu. Apply a few spa_writeable() changes that were made to Illumos in illumos/illumos-gate.git@cd1c8b8 as part of the 3112, 3113, 3114 and 3115 fixes. Add a missing call to fnvlist_free(nvl) in log_internal() that was added in Illumos to fix issue 3085 but couldn't be ported to ZoL at the time (zfsonlinux/zfs@9e11c73) because it depended on future work.
2013-08-28 15:45:09 +04:00
dump_history_offsets, /* SPA history offsets */
2008-11-20 23:01:55 +03:00
dump_zap, /* Pool properties */
dump_zap, /* DSL permissions */
dump_acl, /* ZFS ACL */
dump_uint8, /* ZFS SYSACL */
dump_none, /* FUID nvlist */
dump_packed_nvlist, /* FUID nvlist size */
dump_zap, /* DSL dataset next clones */
dump_zap, /* DSL scrub queue */
2009-07-03 02:44:48 +04:00
dump_zap, /* ZFS user/group used */
dump_zap, /* ZFS user/group quota */
2009-08-18 22:43:27 +04:00
dump_zap, /* snapshot refcount tags */
dump_ddt_zap, /* DDT ZAP object */
dump_zap, /* DDT statistics */
dump_znode, /* SA object */
dump_zap, /* SA Master Node */
dump_sa_attrs, /* SA attribute registration */
dump_sa_layouts, /* SA attribute layouts */
dump_zap, /* DSL scrub translations */
dump_none, /* fake dedup BP */
dump_zap, /* deadlist */
dump_none, /* deadlist hdr */
dump_zap, /* dsl clones */
dump_bpobj_subobjs, /* bpobj subobjs */
dump_unknown, /* Unknown type, must be last */
2008-11-20 23:01:55 +03:00
};
static void
dump_object(objset_t *os, uint64_t object, int verbosity, int *print_header)
{
dmu_buf_t *db = NULL;
dmu_object_info_t doi;
dnode_t *dn;
void *bonus = NULL;
size_t bsize = 0;
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 04:25:34 +03:00
char iblk[32], dblk[32], lsize[32], asize[32], fill[32], dnsize[32];
char bonus_size[32];
2008-11-20 23:01:55 +03:00
char aux[50];
int error;
if (*print_header) {
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 04:25:34 +03:00
(void) printf("\n%10s %3s %5s %5s %5s %6s %5s %6s %s\n",
"Object", "lvl", "iblk", "dblk", "dsize", "dnsize",
"lsize", "%full", "type");
2008-11-20 23:01:55 +03:00
*print_header = 0;
}
if (object == 0) {
dn = DMU_META_DNODE(os);
2008-11-20 23:01:55 +03:00
} else {
error = dmu_bonus_hold(os, object, FTAG, &db);
if (error)
fatal("dmu_bonus_hold(%llu) failed, errno %u",
object, error);
bonus = db->db_data;
bsize = db->db_size;
dn = DB_DNODE((dmu_buf_impl_t *)db);
2008-11-20 23:01:55 +03:00
}
dmu_object_info_from_dnode(dn, &doi);
zdb_nicenum(doi.doi_metadata_block_size, iblk);
zdb_nicenum(doi.doi_data_block_size, dblk);
zdb_nicenum(doi.doi_max_offset, lsize);
zdb_nicenum(doi.doi_physical_blocks_512 << 9, asize);
zdb_nicenum(doi.doi_bonus_size, bonus_size);
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 04:25:34 +03:00
zdb_nicenum(doi.doi_dnodesize, dnsize);
(void) sprintf(fill, "%6.2f", 100.0 * doi.doi_fill_count *
doi.doi_data_block_size / (object == 0 ? DNODES_PER_BLOCK : 1) /
doi.doi_max_offset);
2008-11-20 23:01:55 +03:00
aux[0] = '\0';
if (doi.doi_checksum != ZIO_CHECKSUM_INHERIT || verbosity >= 6) {
(void) snprintf(aux + strlen(aux), sizeof (aux), " (K=%s)",
ZDB_CHECKSUM_NAME(doi.doi_checksum));
2008-11-20 23:01:55 +03:00
}
if (doi.doi_compress != ZIO_COMPRESS_INHERIT || verbosity >= 6) {
(void) snprintf(aux + strlen(aux), sizeof (aux), " (Z=%s)",
ZDB_COMPRESS_NAME(doi.doi_compress));
2008-11-20 23:01:55 +03:00
}
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 04:25:34 +03:00
(void) printf("%10lld %3u %5s %5s %5s %6s %5s %6s %s%s\n",
(u_longlong_t)object, doi.doi_indirection, iblk, dblk,
asize, dnsize, lsize, fill, zdb_ot_name(doi.doi_type), aux);
2008-11-20 23:01:55 +03:00
if (doi.doi_bonus_type != DMU_OT_NONE && verbosity > 3) {
Implement large_dnode pool feature Justification ------------- This feature adds support for variable length dnodes. Our motivation is to eliminate the overhead associated with using spill blocks. Spill blocks are used to store system attribute data (i.e. file metadata) that does not fit in the dnode's bonus buffer. By allowing a larger bonus buffer area the use of a spill block can be avoided. Spill blocks potentially incur an additional read I/O for every dnode in a dnode block. As a worst case example, reading 32 dnodes from a 16k dnode block and all of the spill blocks could issue 33 separate reads. Now suppose those dnodes have size 1024 and therefore don't need spill blocks. Then the worst case number of blocks read is reduced to from 33 to two--one per dnode block. In practice spill blocks may tend to be co-located on disk with the dnode blocks so the reduction in I/O would not be this drastic. In a badly fragmented pool, however, the improvement could be significant. ZFS-on-Linux systems that make heavy use of extended attributes would benefit from this feature. In particular, ZFS-on-Linux supports the xattr=sa dataset property which allows file extended attribute data to be stored in the dnode bonus buffer as an alternative to the traditional directory-based format. Workloads such as SELinux and the Lustre distributed filesystem often store enough xattr data to force spill bocks when xattr=sa is in effect. Large dnodes may therefore provide a performance benefit to such systems. Other use cases that may benefit from this feature include files with large ACLs and symbolic links with long target names. Furthermore, this feature may be desirable on other platforms in case future applications or features are developed that could make use of a larger bonus buffer area. Implementation -------------- The size of a dnode may be a multiple of 512 bytes up to the size of a dnode block (currently 16384 bytes). A dn_extra_slots field was added to the current on-disk dnode_phys_t structure to describe the size of the physical dnode on disk. The 8 bits for this field were taken from the zero filled dn_pad2 field. The field represents how many "extra" dnode_phys_t slots a dnode consumes in its dnode block. This convention results in a value of 0 for 512 byte dnodes which preserves on-disk format compatibility with older software. Similarly, the in-memory dnode_t structure has a new dn_num_slots field to represent the total number of dnode_phys_t slots consumed on disk. Thus dn->dn_num_slots is 1 greater than the corresponding dnp->dn_extra_slots. This difference in convention was adopted because, unlike on-disk structures, backward compatibility is not a concern for in-memory objects, so we used a more natural way to represent size for a dnode_t. The default size for newly created dnodes is determined by the value of a new "dnodesize" dataset property. By default the property is set to "legacy" which is compatible with older software. Setting the property to "auto" will allow the filesystem to choose the most suitable dnode size. Currently this just sets the default dnode size to 1k, but future code improvements could dynamically choose a size based on observed workload patterns. Dnodes of varying sizes can coexist within the same dataset and even within the same dnode block. For example, to enable automatically-sized dnodes, run # zfs set dnodesize=auto tank/fish The user can also specify literal values for the dnodesize property. These are currently limited to powers of two from 1k to 16k. The power-of-2 limitation is only for simplicity of the user interface. Internally the implementation can handle any multiple of 512 up to 16k, and consumers of the DMU API can specify any legal dnode value. The size of a new dnode is determined at object allocation time and stored as a new field in the znode in-memory structure. New DMU interfaces are added to allow the consumer to specify the dnode size that a newly allocated object should use. Existing interfaces are unchanged to avoid having to update every call site and to preserve compatibility with external consumers such as Lustre. The new interfaces names are given below. The versions of these functions that don't take a dnodesize parameter now just call the _dnsize() versions with a dnodesize of 0, which means use the legacy dnode size. New DMU interfaces: dmu_object_alloc_dnsize() dmu_object_claim_dnsize() dmu_object_reclaim_dnsize() New ZAP interfaces: zap_create_dnsize() zap_create_norm_dnsize() zap_create_flags_dnsize() zap_create_claim_norm_dnsize() zap_create_link_dnsize() The constant DN_MAX_BONUSLEN is renamed to DN_OLD_MAX_BONUSLEN. The spa_maxdnodesize() function should be used to determine the maximum bonus length for a pool. These are a few noteworthy changes to key functions: * The prototype for dnode_hold_impl() now takes a "slots" parameter. When the DNODE_MUST_BE_FREE flag is set, this parameter is used to ensure the hole at the specified object offset is large enough to hold the dnode being created. The slots parameter is also used to ensure a dnode does not span multiple dnode blocks. In both of these cases, if a failure occurs, ENOSPC is returned. Keep in mind, these failure cases are only possible when using DNODE_MUST_BE_FREE. If the DNODE_MUST_BE_ALLOCATED flag is set, "slots" must be 0. dnode_hold_impl() will check if the requested dnode is already consumed as an extra dnode slot by an large dnode, in which case it returns ENOENT. * The function dmu_object_alloc() advances to the next dnode block if dnode_hold_impl() returns an error for a requested object. This is because the beginning of the next dnode block is the only location it can safely assume to either be a hole or a valid starting point for a dnode. * dnode_next_offset_level() and other functions that iterate through dnode blocks may no longer use a simple array indexing scheme. These now use the current dnode's dn_num_slots field to advance to the next dnode in the block. This is to ensure we properly skip the current dnode's bonus area and don't interpret it as a valid dnode. zdb --- The zdb command was updated to display a dnode's size under the "dnsize" column when the object is dumped. For ZIL create log records, zdb will now display the slot count for the object. ztest ----- Ztest chooses a random dnodesize for every newly created object. The random distribution is more heavily weighted toward small dnodes to better simulate real-world datasets. Unused bonus buffer space is filled with non-zero values computed from the object number, dataset id, offset, and generation number. This helps ensure that the dnode traversal code properly skips the interior regions of large dnodes, and that these interior regions are not overwritten by data belonging to other dnodes. A new test visits each object in a dataset. It verifies that the actual dnode size matches what was stored in the ztest block tag when it was created. It also verifies that the unused bonus buffer space is filled with the expected data patterns. ZFS Test Suite -------------- Added six new large dnode-specific tests, and integrated the dnodesize property into existing tests for zfs allow and send/recv. Send/Receive ------------ ZFS send streams for datasets containing large dnodes cannot be received on pools that don't support the large_dnode feature. A send stream with large dnodes sets a DMU_BACKUP_FEATURE_LARGE_DNODE flag which will be unrecognized by an incompatible receiving pool so that the zfs receive will fail gracefully. While not implemented here, it may be possible to generate a backward-compatible send stream from a dataset containing large dnodes. The implementation may be tricky, however, because the send object record for a large dnode would need to be resized to a 512 byte dnode, possibly kicking in a spill block in the process. This means we would need to construct a new SA layout and possibly register it in the SA layout object. The SA layout is normally just sent as an ordinary object record. But if we are constructing new layouts while generating the send stream we'd have to build the SA layout object dynamically and send it at the end of the stream. For sending and receiving between pools that do support large dnodes, the drr_object send record type is extended with a new field to store the dnode slot count. This field was repurposed from unused padding in the structure. ZIL Replay ---------- The dnode slot count is stored in the uppermost 8 bits of the lr_foid field. The bits were unused as the object id is currently capped at 48 bits. Resizing Dnodes --------------- It should be possible to resize a dnode when it is dirtied if the current dnodesize dataset property differs from the dnode's size, but this functionality is not currently implemented. Clearly a dnode can only grow if there are sufficient contiguous unused slots in the dnode block, but it should always be possible to shrink a dnode. Growing dnodes may be useful to reduce fragmentation in a pool with many spill blocks in use. Shrinking dnodes may be useful to allow sending a dataset to a pool that doesn't support the large_dnode feature. Feature Reference Counting -------------------------- The reference count for the large_dnode pool feature tracks the number of datasets that have ever contained a dnode of size larger than 512 bytes. The first time a large dnode is created in a dataset the dataset is converted to an extensible dataset. This is a one-way operation and the only way to decrement the feature count is to destroy the dataset, even if the dataset no longer contains any large dnodes. The complexity of reference counting on a per-dnode basis was too high, so we chose to track it on a per-dataset basis similarly to the large_block feature. Signed-off-by: Ned Bass <bass6@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #3542
2016-03-17 04:25:34 +03:00
(void) printf("%10s %3s %5s %5s %5s %5s %5s %6s %s\n",
"", "", "", "", "", "", bonus_size, "bonus",
zdb_ot_name(doi.doi_bonus_type));
2008-11-20 23:01:55 +03:00
}
if (verbosity >= 4) {
(void) printf("\tdnode flags: %s%s%s%s\n",
2009-07-03 02:44:48 +04:00
(dn->dn_phys->dn_flags & DNODE_FLAG_USED_BYTES) ?
"USED_BYTES " : "",
(dn->dn_phys->dn_flags & DNODE_FLAG_USERUSED_ACCOUNTED) ?
"USERUSED_ACCOUNTED " : "",
(dn->dn_phys->dn_flags & DNODE_FLAG_USEROBJUSED_ACCOUNTED) ?
"USEROBJUSED_ACCOUNTED " : "",
(dn->dn_phys->dn_flags & DNODE_FLAG_SPILL_BLKPTR) ?
"SPILL_BLKPTR" : "");
2009-07-03 02:44:48 +04:00
(void) printf("\tdnode maxblkid: %llu\n",
(longlong_t)dn->dn_phys->dn_maxblkid);
object_viewer[ZDB_OT_TYPE(doi.doi_bonus_type)](os, object,
bonus, bsize);
object_viewer[ZDB_OT_TYPE(doi.doi_type)](os, object, NULL, 0);
2008-11-20 23:01:55 +03:00
*print_header = 1;
}
if (verbosity >= 5)
dump_indirect(dn);
2008-11-20 23:01:55 +03:00
if (verbosity >= 5) {
/*
* Report the list of segments that comprise the object.
*/
uint64_t start = 0;
uint64_t end;
uint64_t blkfill = 1;
int minlvl = 1;
if (dn->dn_type == DMU_OT_DNODE) {
minlvl = 0;
blkfill = DNODES_PER_BLOCK;
}
for (;;) {
char segsize[32];
error = dnode_next_offset(dn,
0, &start, minlvl, blkfill, 0);
2008-11-20 23:01:55 +03:00
if (error)
break;
end = start;
error = dnode_next_offset(dn,
DNODE_FIND_HOLE, &end, minlvl, blkfill, 0);
zdb_nicenum(end - start, segsize);
2008-11-20 23:01:55 +03:00
(void) printf("\t\tsegment [%016llx, %016llx)"
" size %5s\n", (u_longlong_t)start,
(u_longlong_t)end, segsize);
if (error)
break;
start = end;
}
}
if (db != NULL)
dmu_buf_rele(db, FTAG);
}
static char *objset_types[DMU_OST_NUMTYPES] = {
"NONE", "META", "ZPL", "ZVOL", "OTHER", "ANY" };
static void
dump_dir(objset_t *os)
{
dmu_objset_stats_t dds;
uint64_t object, object_count;
uint64_t refdbytes, usedobjs, scratch;
char numbuf[32];
2009-07-03 02:44:48 +04:00
char blkbuf[BP_SPRINTF_LEN + 20];
char osname[ZFS_MAX_DATASET_NAME_LEN];
2008-11-20 23:01:55 +03:00
char *type = "UNKNOWN";
int verbosity = dump_opt['d'];
int print_header = 1;
int i, error;
dsl_pool_config_enter(dmu_objset_pool(os), FTAG);
2008-11-20 23:01:55 +03:00
dmu_objset_fast_stat(os, &dds);
dsl_pool_config_exit(dmu_objset_pool(os), FTAG);
2008-11-20 23:01:55 +03:00
if (dds.dds_type < DMU_OST_NUMTYPES)
type = objset_types[dds.dds_type];
if (dds.dds_type == DMU_OST_META) {
dds.dds_creation_txg = TXG_INITIAL;
usedobjs = BP_GET_FILL(os->os_rootbp);
refdbytes = dsl_dir_phys(os->os_spa->spa_dsl_pool->dp_mos_dir)->
dd_used_bytes;
2008-11-20 23:01:55 +03:00
} else {
dmu_objset_space(os, &refdbytes, &scratch, &usedobjs, &scratch);
}
ASSERT3U(usedobjs, ==, BP_GET_FILL(os->os_rootbp));
2008-11-20 23:01:55 +03:00
zdb_nicenum(refdbytes, numbuf);
2008-11-20 23:01:55 +03:00
if (verbosity >= 4) {
(void) snprintf(blkbuf, sizeof (blkbuf), ", rootbp ");
(void) snprintf_blkptr(blkbuf + strlen(blkbuf),
sizeof (blkbuf) - strlen(blkbuf), os->os_rootbp);
2008-11-20 23:01:55 +03:00
} else {
blkbuf[0] = '\0';
}
dmu_objset_name(os, osname);
(void) printf("Dataset %s [%s], ID %llu, cr_txg %llu, "
"%s, %llu objects%s\n",
osname, type, (u_longlong_t)dmu_objset_id(os),
(u_longlong_t)dds.dds_creation_txg,
numbuf, (u_longlong_t)usedobjs, blkbuf);
if (zopt_objects != 0) {
for (i = 0; i < zopt_objects; i++)
dump_object(os, zopt_object[i], verbosity,
&print_header);
(void) printf("\n");
return;
}
if (dump_opt['i'] != 0 || verbosity >= 2)
dump_intent_log(dmu_objset_zil(os));
if (dmu_objset_ds(os) != NULL)
dump_deadlist(&dmu_objset_ds(os)->ds_deadlist);
if (verbosity < 2)
return;
if (BP_IS_HOLE(os->os_rootbp))
return;
2008-11-20 23:01:55 +03:00
dump_object(os, 0, verbosity, &print_header);
2009-07-03 02:44:48 +04:00
object_count = 0;
if (DMU_USERUSED_DNODE(os) != NULL &&
DMU_USERUSED_DNODE(os)->dn_type != 0) {
2009-07-03 02:44:48 +04:00
dump_object(os, DMU_USERUSED_OBJECT, verbosity, &print_header);
dump_object(os, DMU_GROUPUSED_OBJECT, verbosity, &print_header);
}
2008-11-20 23:01:55 +03:00
object = 0;
while ((error = dmu_object_next(os, &object, B_FALSE, 0)) == 0) {
dump_object(os, object, verbosity, &print_header);
object_count++;
}
ASSERT3U(object_count, ==, usedobjs);
(void) printf("\n");
2009-07-03 02:44:48 +04:00
if (error != ESRCH) {
(void) fprintf(stderr, "dmu_object_next() = %d\n", error);
abort();
}
2008-11-20 23:01:55 +03:00
}
static void
dump_uberblock(uberblock_t *ub, const char *header, const char *footer)
2008-11-20 23:01:55 +03:00
{
time_t timestamp = ub->ub_timestamp;
(void) printf("%s", header ? header : "");
2008-11-20 23:01:55 +03:00
(void) printf("\tmagic = %016llx\n", (u_longlong_t)ub->ub_magic);
(void) printf("\tversion = %llu\n", (u_longlong_t)ub->ub_version);
(void) printf("\ttxg = %llu\n", (u_longlong_t)ub->ub_txg);
(void) printf("\tguid_sum = %llu\n", (u_longlong_t)ub->ub_guid_sum);
(void) printf("\ttimestamp = %llu UTC = %s",
(u_longlong_t)ub->ub_timestamp, asctime(localtime(&timestamp)));
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
if (dump_opt['u'] >= 4) {
2008-11-20 23:01:55 +03:00
char blkbuf[BP_SPRINTF_LEN];
snprintf_blkptr(blkbuf, sizeof (blkbuf), &ub->ub_rootbp);
2008-11-20 23:01:55 +03:00
(void) printf("\trootbp = %s\n", blkbuf);
}
(void) printf("%s", footer ? footer : "");
2008-11-20 23:01:55 +03:00
}
static void
dump_config(spa_t *spa)
2008-11-20 23:01:55 +03:00
{
dmu_buf_t *db;
size_t nvsize = 0;
int error = 0;
2008-11-20 23:01:55 +03:00
error = dmu_bonus_hold(spa->spa_meta_objset,
spa->spa_config_object, FTAG, &db);
if (error == 0) {
nvsize = *(uint64_t *)db->db_data;
dmu_buf_rele(db, FTAG);
(void) printf("\nMOS Configuration:\n");
dump_packed_nvlist(spa->spa_meta_objset,
spa->spa_config_object, (void *)&nvsize, 1);
} else {
(void) fprintf(stderr, "dmu_bonus_hold(%llu) failed, errno %d",
(u_longlong_t)spa->spa_config_object, error);
}
}
static void
dump_cachefile(const char *cachefile)
{
int fd;
struct stat64 statbuf;
char *buf;
nvlist_t *config;
if ((fd = open64(cachefile, O_RDONLY)) < 0) {
(void) printf("cannot open '%s': %s\n", cachefile,
strerror(errno));
exit(1);
}
if (fstat64(fd, &statbuf) != 0) {
(void) printf("failed to stat '%s': %s\n", cachefile,
strerror(errno));
exit(1);
}
if ((buf = malloc(statbuf.st_size)) == NULL) {
(void) fprintf(stderr, "failed to allocate %llu bytes\n",
(u_longlong_t)statbuf.st_size);
exit(1);
}
if (read(fd, buf, statbuf.st_size) != statbuf.st_size) {
(void) fprintf(stderr, "failed to read %llu bytes\n",
(u_longlong_t)statbuf.st_size);
exit(1);
}
(void) close(fd);
if (nvlist_unpack(buf, statbuf.st_size, &config, 0) != 0) {
(void) fprintf(stderr, "failed to unpack nvlist\n");
exit(1);
}
free(buf);
dump_nvlist(config, 0);
nvlist_free(config);
}
/*
* ZFS label nvlist stats
*/
typedef struct zdb_nvl_stats {
int zns_list_count;
int zns_leaf_count;
size_t zns_leaf_largest;
size_t zns_leaf_total;
nvlist_t *zns_string;
nvlist_t *zns_uint64;
nvlist_t *zns_boolean;
} zdb_nvl_stats_t;
static void
collect_nvlist_stats(nvlist_t *nvl, zdb_nvl_stats_t *stats)
{
nvlist_t *list, **array;
nvpair_t *nvp = NULL;
char *name;
uint_t i, items;
stats->zns_list_count++;
while ((nvp = nvlist_next_nvpair(nvl, nvp)) != NULL) {
name = nvpair_name(nvp);
switch (nvpair_type(nvp)) {
case DATA_TYPE_STRING:
fnvlist_add_string(stats->zns_string, name,
fnvpair_value_string(nvp));
break;
case DATA_TYPE_UINT64:
fnvlist_add_uint64(stats->zns_uint64, name,
fnvpair_value_uint64(nvp));
break;
case DATA_TYPE_BOOLEAN:
fnvlist_add_boolean(stats->zns_boolean, name);
break;
case DATA_TYPE_NVLIST:
if (nvpair_value_nvlist(nvp, &list) == 0)
collect_nvlist_stats(list, stats);
break;
case DATA_TYPE_NVLIST_ARRAY:
if (nvpair_value_nvlist_array(nvp, &array, &items) != 0)
break;
for (i = 0; i < items; i++) {
collect_nvlist_stats(array[i], stats);
/* collect stats on leaf vdev */
if (strcmp(name, "children") == 0) {
size_t size;
(void) nvlist_size(array[i], &size,
NV_ENCODE_XDR);
stats->zns_leaf_total += size;
if (size > stats->zns_leaf_largest)
stats->zns_leaf_largest = size;
stats->zns_leaf_count++;
}
}
break;
default:
(void) printf("skip type %d!\n", (int)nvpair_type(nvp));
}
}
}
static void
dump_nvlist_stats(nvlist_t *nvl, size_t cap)
{
zdb_nvl_stats_t stats = { 0 };
size_t size, sum = 0, total;
size_t noise;
/* requires nvlist with non-unique names for stat collection */
VERIFY0(nvlist_alloc(&stats.zns_string, 0, 0));
VERIFY0(nvlist_alloc(&stats.zns_uint64, 0, 0));
VERIFY0(nvlist_alloc(&stats.zns_boolean, 0, 0));
VERIFY0(nvlist_size(stats.zns_boolean, &noise, NV_ENCODE_XDR));
(void) printf("\n\nZFS Label NVList Config Stats:\n");
VERIFY0(nvlist_size(nvl, &total, NV_ENCODE_XDR));
(void) printf(" %d bytes used, %d bytes free (using %4.1f%%)\n\n",
(int)total, (int)(cap - total), 100.0 * total / cap);
collect_nvlist_stats(nvl, &stats);
VERIFY0(nvlist_size(stats.zns_uint64, &size, NV_ENCODE_XDR));
size -= noise;
sum += size;
(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "integers:",
(int)fnvlist_num_pairs(stats.zns_uint64),
(int)size, 100.0 * size / total);
VERIFY0(nvlist_size(stats.zns_string, &size, NV_ENCODE_XDR));
size -= noise;
sum += size;
(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "strings:",
(int)fnvlist_num_pairs(stats.zns_string),
(int)size, 100.0 * size / total);
VERIFY0(nvlist_size(stats.zns_boolean, &size, NV_ENCODE_XDR));
size -= noise;
sum += size;
(void) printf("%12s %4d %6d bytes (%5.2f%%)\n", "booleans:",
(int)fnvlist_num_pairs(stats.zns_boolean),
(int)size, 100.0 * size / total);
size = total - sum; /* treat remainder as nvlist overhead */
(void) printf("%12s %4d %6d bytes (%5.2f%%)\n\n", "nvlists:",
stats.zns_list_count, (int)size, 100.0 * size / total);
if (stats.zns_leaf_count > 0) {
size_t average = stats.zns_leaf_total / stats.zns_leaf_count;
(void) printf("%12s %4d %6d bytes average\n", "leaf vdevs:",
stats.zns_leaf_count, (int)average);
(void) printf("%24d bytes largest\n",
(int)stats.zns_leaf_largest);
if (dump_opt['l'] >= 3 && average > 0)
(void) printf(" space for %d additional leaf vdevs\n",
(int)((cap - total) / average));
}
(void) printf("\n");
nvlist_free(stats.zns_string);
nvlist_free(stats.zns_uint64);
nvlist_free(stats.zns_boolean);
}
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
typedef struct cksum_record {
zio_cksum_t cksum;
boolean_t labels[VDEV_LABELS];
avl_node_t link;
} cksum_record_t;
static int
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
cksum_record_compare(const void *x1, const void *x2)
2008-11-20 23:01:55 +03:00
{
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
const cksum_record_t *l = (cksum_record_t *)x1;
const cksum_record_t *r = (cksum_record_t *)x2;
int arraysize = ARRAY_SIZE(l->cksum.zc_word);
int difference;
for (int i = 0; i < arraysize; i++) {
difference = AVL_CMP(l->cksum.zc_word[i], r->cksum.zc_word[i]);
if (difference)
break;
}
return (difference);
}
static cksum_record_t *
cksum_record_alloc(zio_cksum_t *cksum, int l)
{
cksum_record_t *rec;
rec = umem_zalloc(sizeof (*rec), UMEM_NOFAIL);
rec->cksum = *cksum;
rec->labels[l] = B_TRUE;
return (rec);
}
static cksum_record_t *
cksum_record_lookup(avl_tree_t *tree, zio_cksum_t *cksum)
{
cksum_record_t lookup = { .cksum = *cksum };
avl_index_t where;
return (avl_find(tree, &lookup, &where));
}
static cksum_record_t *
cksum_record_insert(avl_tree_t *tree, zio_cksum_t *cksum, int l)
{
cksum_record_t *rec;
rec = cksum_record_lookup(tree, cksum);
if (rec) {
rec->labels[l] = B_TRUE;
} else {
rec = cksum_record_alloc(cksum, l);
avl_add(tree, rec);
}
return (rec);
}
static int
first_label(cksum_record_t *rec)
{
for (int i = 0; i < VDEV_LABELS; i++)
if (rec->labels[i])
return (i);
return (-1);
}
static void
print_label_numbers(char *prefix, cksum_record_t *rec)
{
printf("%s", prefix);
for (int i = 0; i < VDEV_LABELS; i++)
if (rec->labels[i] == B_TRUE)
printf("%d ", i);
printf("\n");
}
#define MAX_UBERBLOCK_COUNT (VDEV_UBERBLOCK_RING >> UBERBLOCK_SHIFT)
typedef struct label {
2008-11-20 23:01:55 +03:00
vdev_label_t label;
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
nvlist_t *config_nv;
cksum_record_t *config;
cksum_record_t *uberblocks[MAX_UBERBLOCK_COUNT];
boolean_t header_printed;
boolean_t read_failed;
} label_t;
static void
print_label_header(label_t *label, int l)
{
if (dump_opt['q'])
return;
if (label->header_printed == B_TRUE)
return;
(void) printf("------------------------------------\n");
(void) printf("LABEL %d\n", l);
(void) printf("------------------------------------\n");
label->header_printed = B_TRUE;
}
static void
dump_config_from_label(label_t *label, size_t buflen, int l)
{
if (dump_opt['q'])
return;
if ((dump_opt['l'] < 3) && (first_label(label->config) != l))
return;
print_label_header(label, l);
dump_nvlist(label->config_nv, 4);
print_label_numbers(" labels = ", label->config);
if (dump_opt['l'] >= 2)
dump_nvlist_stats(label->config_nv, buflen);
}
#define ZDB_MAX_UB_HEADER_SIZE 32
static void
dump_label_uberblocks(label_t *label, uint64_t ashift, int label_num)
{
vdev_t vd;
char header[ZDB_MAX_UB_HEADER_SIZE];
vd.vdev_ashift = ashift;
vd.vdev_top = &vd;
for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
uberblock_t *ub = (void *)((char *)&label->label + uoff);
cksum_record_t *rec = label->uberblocks[i];
if (rec == NULL) {
if (dump_opt['u'] >= 2) {
print_label_header(label, label_num);
(void) printf(" Uberblock[%d] invalid\n", i);
}
continue;
}
if ((dump_opt['u'] < 3) && (first_label(rec) != label_num))
continue;
print_label_header(label, label_num);
(void) snprintf(header, ZDB_MAX_UB_HEADER_SIZE,
" Uberblock[%d]\n", i);
dump_uberblock(ub, header, "");
print_label_numbers(" labels = ", rec);
}
}
static char curpath[PATH_MAX];
/*
* Iterate through the path components, recursively passing
* current one's obj and remaining path until we find the obj
* for the last one.
*/
static int
dump_path_impl(objset_t *os, uint64_t obj, char *name)
{
int err;
int header = 1;
uint64_t child_obj;
char *s;
dmu_buf_t *db;
dmu_object_info_t doi;
if ((s = strchr(name, '/')) != NULL)
*s = '\0';
err = zap_lookup(os, obj, name, 8, 1, &child_obj);
(void) strlcat(curpath, name, sizeof (curpath));
if (err != 0) {
(void) fprintf(stderr, "failed to lookup %s: %s\n",
curpath, strerror(err));
return (err);
}
child_obj = ZFS_DIRENT_OBJ(child_obj);
err = sa_buf_hold(os, child_obj, FTAG, &db);
if (err != 0) {
(void) fprintf(stderr,
"failed to get SA dbuf for obj %llu: %s\n",
(u_longlong_t)child_obj, strerror(err));
return (EINVAL);
}
dmu_object_info_from_db(db, &doi);
sa_buf_rele(db, FTAG);
if (doi.doi_bonus_type != DMU_OT_SA &&
doi.doi_bonus_type != DMU_OT_ZNODE) {
(void) fprintf(stderr, "invalid bonus type %d for obj %llu\n",
doi.doi_bonus_type, (u_longlong_t)child_obj);
return (EINVAL);
}
if (dump_opt['v'] > 6) {
(void) printf("obj=%llu %s type=%d bonustype=%d\n",
(u_longlong_t)child_obj, curpath, doi.doi_type,
doi.doi_bonus_type);
}
(void) strlcat(curpath, "/", sizeof (curpath));
switch (doi.doi_type) {
case DMU_OT_DIRECTORY_CONTENTS:
if (s != NULL && *(s + 1) != '\0')
return (dump_path_impl(os, child_obj, s + 1));
/*FALLTHROUGH*/
case DMU_OT_PLAIN_FILE_CONTENTS:
dump_object(os, child_obj, dump_opt['v'], &header);
return (0);
default:
(void) fprintf(stderr, "object %llu has non-file/directory "
"type %d\n", (u_longlong_t)obj, doi.doi_type);
break;
}
return (EINVAL);
}
/*
* Dump the blocks for the object specified by path inside the dataset.
*/
static int
dump_path(char *ds, char *path)
{
int err;
objset_t *os;
uint64_t root_obj;
err = open_objset(ds, DMU_OST_ZFS, FTAG, &os);
if (err != 0)
return (err);
err = zap_lookup(os, MASTER_NODE_OBJ, ZFS_ROOT_OBJ, 8, 1, &root_obj);
if (err != 0) {
(void) fprintf(stderr, "can't lookup root znode: %s\n",
strerror(err));
dmu_objset_disown(os, FTAG);
return (EINVAL);
}
(void) snprintf(curpath, sizeof (curpath), "dataset=%s path=/", ds);
err = dump_path_impl(os, root_obj, path);
close_objset(os, FTAG);
return (err);
}
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
static int
dump_label(const char *dev)
{
char path[MAXPATHLEN];
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
label_t labels[VDEV_LABELS];
uint64_t psize, ashift;
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
struct stat64 statbuf;
boolean_t config_found = B_FALSE;
boolean_t error = B_FALSE;
avl_tree_t config_tree;
avl_tree_t uberblock_tree;
void *node, *cookie;
int fd;
bzero(labels, sizeof (labels));
2008-11-20 23:01:55 +03:00
/*
* Check if we were given absolute path and use it as is.
* Otherwise if the provided vdev name doesn't point to a file,
* try prepending expected disk paths and partition numbers.
*/
(void) strlcpy(path, dev, sizeof (path));
if (dev[0] != '/' && stat64(path, &statbuf) != 0) {
int error;
error = zfs_resolve_shortname(dev, path, MAXPATHLEN);
if (error == 0 && zfs_dev_is_whole_disk(path)) {
if (zfs_append_partition(path, MAXPATHLEN) == -1)
error = ENOENT;
}
if (error || (stat64(path, &statbuf) != 0)) {
(void) printf("failed to find device %s, try "
"specifying absolute path instead\n", dev);
return (1);
}
}
if ((fd = open64(path, O_RDONLY)) < 0) {
(void) printf("cannot open '%s': %s\n", path, strerror(errno));
2008-11-20 23:01:55 +03:00
exit(1);
}
if (ioctl(fd, BLKFLSBUF) != 0)
(void) printf("failed to invalidate cache '%s' : %s\n", path,
strerror(errno));
if (fstat64_blk(fd, &statbuf) != 0) {
(void) printf("failed to stat '%s': %s\n", path,
2008-11-20 23:01:55 +03:00
strerror(errno));
(void) close(fd);
exit(1);
}
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
avl_create(&config_tree, cksum_record_compare,
sizeof (cksum_record_t), offsetof(cksum_record_t, link));
avl_create(&uberblock_tree, cksum_record_compare,
sizeof (cksum_record_t), offsetof(cksum_record_t, link));
2008-11-20 23:01:55 +03:00
psize = statbuf.st_size;
psize = P2ALIGN(psize, (uint64_t)sizeof (vdev_label_t));
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
ashift = SPA_MINBLOCKSHIFT;
2008-11-20 23:01:55 +03:00
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
/*
* 1. Read the label from disk
* 2. Unpack the configuration and insert in config tree.
* 3. Traverse all uberblocks and insert in uberblock tree.
*/
for (int l = 0; l < VDEV_LABELS; l++) {
label_t *label = &labels[l];
char *buf = label->label.vl_vdev_phys.vp_nvlist;
size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
nvlist_t *config;
cksum_record_t *rec;
zio_cksum_t cksum;
vdev_t vd;
if (pread64(fd, &label->label, sizeof (label->label),
vdev_label_offset(psize, l, 0)) != sizeof (label->label)) {
if (!dump_opt['q'])
(void) printf("failed to read label %d\n", l);
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
label->read_failed = B_TRUE;
error = B_TRUE;
2008-11-20 23:01:55 +03:00
continue;
}
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
label->read_failed = B_FALSE;
if (nvlist_unpack(buf, buflen, &config, 0) == 0) {
nvlist_t *vdev_tree = NULL;
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
size_t size;
if ((nvlist_lookup_nvlist(config,
ZPOOL_CONFIG_VDEV_TREE, &vdev_tree) != 0) ||
(nvlist_lookup_uint64(vdev_tree,
ZPOOL_CONFIG_ASHIFT, &ashift) != 0))
ashift = SPA_MINBLOCKSHIFT;
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
if (nvlist_size(config, &size, NV_ENCODE_XDR) != 0)
size = buflen;
fletcher_4_native_varsize(buf, size, &cksum);
rec = cksum_record_insert(&config_tree, &cksum, l);
label->config = rec;
label->config_nv = config;
config_found = B_TRUE;
} else {
error = B_TRUE;
2008-11-20 23:01:55 +03:00
}
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
vd.vdev_ashift = ashift;
vd.vdev_top = &vd;
for (int i = 0; i < VDEV_UBERBLOCK_COUNT(&vd); i++) {
uint64_t uoff = VDEV_UBERBLOCK_OFFSET(&vd, i);
uberblock_t *ub = (void *)((char *)label + uoff);
if (uberblock_verify(ub))
continue;
fletcher_4_native_varsize(ub, sizeof (*ub), &cksum);
rec = cksum_record_insert(&uberblock_tree, &cksum, l);
label->uberblocks[i] = rec;
}
}
/*
* Dump the label and uberblocks.
*/
for (int l = 0; l < VDEV_LABELS; l++) {
label_t *label = &labels[l];
size_t buflen = sizeof (label->label.vl_vdev_phys.vp_nvlist);
if (label->read_failed == B_TRUE)
continue;
if (label->config_nv) {
dump_config_from_label(label, buflen, l);
} else {
if (!dump_opt['q'])
(void) printf("failed to unpack label %d\n", l);
}
if (dump_opt['u'])
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
dump_label_uberblocks(label, ashift, l);
nvlist_free(label->config_nv);
2008-11-20 23:01:55 +03:00
}
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
cookie = NULL;
while ((node = avl_destroy_nodes(&config_tree, &cookie)) != NULL)
umem_free(node, sizeof (cksum_record_t));
cookie = NULL;
while ((node = avl_destroy_nodes(&uberblock_tree, &cookie)) != NULL)
umem_free(node, sizeof (cksum_record_t));
avl_destroy(&config_tree);
avl_destroy(&uberblock_tree);
(void) close(fd);
Dump unique configurations and Uberblocks in zdb -lu For zdb -l, detect when the configuration nvlist in some label l (l>0) is the same as a configuration already dumped. If so, do not dump it. Make a similar check when dumping Uberblocks for zdb -lu. Check whether a label already dumped contains an identical Uberblock. If so, do not dump the Uberblock. When dumping a configuration or Uberblock, state which labels it is found in (0-3), for example: labels = 1 2 3 Detecting redundant uberblocks or configurations is accomplished by calculating checksums of the uberblocks and the packed nvlists containing the configuration. If there is nothing unique to be dumped for a label (ie the configuration and uberblocks have checksums matching those already dumped) print nothing for that label. With additional l's or u's, increase verbosity as follows: -l Dump each unique configuration only once. Indicate which labels it appears in. -ll In addition, dump label space usage stats. -lll Dump every configuration, unique or not. -u Dump each unique, valid, uberblock only once. Indicate which labels it appears in. -uu In addition, state which slots are invalid. -uuu Dump every uberblock, unique or not. -uuuu Dump the uberblock blockpointer (used to be -uuu) Make exit values conform to the manual page. Failing to unpack a configuration nvlist is considered an error, as well as failing to open or read from the device. Add three tests, zdb_00{3,4,5}_pos to verify the above functionality. An example of the output: ------------------------------------ LABEL 0 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 880 < ... redacted ... > features_for_read: com.delphix:hole_birth com.delphix:embedded_data labels = 0 Uberblock[0] magic = 0000000000bab10c version = 5000 txg = 0 guid_sum = 3038694082047428541 timestamp = 1487715500 UTC = Tue Feb 21 14:18:20 2017 labels = 0 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 772 guid_sum = 9045970794941528051 timestamp = 1487727291 UTC = Tue Feb 21 17:34:51 2017 labels = 0 < ... redacted ... > ------------------------------------ LABEL 1 ------------------------------------ version: 5000 name: 'pool' state: 1 txg: 14 < ... redacted ... > com.delphix:embedded_data labels = 1 2 3 Uberblock[4] magic = 0000000000bab10c version = 5000 txg = 4 guid_sum = 7793930272573252584 timestamp = 1487727521 UTC = Tue Feb 21 17:38:41 2017 labels = 1 2 3 < ... redacted ... > Reviewed-by: Tim Chase <tim@chase2k.com> Reviewed-by: Don Brady <don.brady@intel.com> Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Olaf Faaland <faaland1@llnl.gov> Closes #5738
2017-03-07 03:01:45 +03:00
return (config_found == B_FALSE ? 2 :
(error == B_TRUE ? 1 : 0));
2008-11-20 23:01:55 +03:00
}
static uint64_t dataset_feature_count[SPA_FEATURES];
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 23:15:08 +03:00
2008-11-20 23:01:55 +03:00
/*ARGSUSED*/
static int
dump_one_dir(const char *dsname, void *arg)
2008-11-20 23:01:55 +03:00
{
int error;
objset_t *os;
spa_feature_t f;
2008-11-20 23:01:55 +03:00
error = open_objset(dsname, DMU_OST_ANY, FTAG, &os);
if (error != 0)
2008-11-20 23:01:55 +03:00
return (0);
for (f = 0; f < SPA_FEATURES; f++) {
if (!dmu_objset_ds(os)->ds_feature_inuse[f])
continue;
ASSERT(spa_feature_table[f].fi_flags &
ZFEATURE_FLAG_PER_DATASET);
dataset_feature_count[f]++;
}
2008-11-20 23:01:55 +03:00
dump_dir(os);
close_objset(os, FTAG);
2008-11-20 23:01:55 +03:00
fuid_table_destroy();
return (0);
}
/*
* Block statistics.
2008-11-20 23:01:55 +03:00
*/
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 23:15:08 +03:00
#define PSIZE_HISTO_SIZE (SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 2)
2008-11-20 23:01:55 +03:00
typedef struct zdb_blkstats {
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
uint64_t zb_asize;
uint64_t zb_lsize;
uint64_t zb_psize;
uint64_t zb_count;
uint64_t zb_gangs;
uint64_t zb_ditto_samevdev;
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
uint64_t zb_psize_histogram[PSIZE_HISTO_SIZE];
2008-11-20 23:01:55 +03:00
} zdb_blkstats_t;
/*
* Extended object types to report deferred frees and dedup auto-ditto blocks.
*/
#define ZDB_OT_DEFERRED (DMU_OT_NUMTYPES + 0)
#define ZDB_OT_DITTO (DMU_OT_NUMTYPES + 1)
#define ZDB_OT_OTHER (DMU_OT_NUMTYPES + 2)
#define ZDB_OT_TOTAL (DMU_OT_NUMTYPES + 3)
static char *zdb_ot_extname[] = {
"deferred free",
"dedup ditto",
"other",
"Total",
};
2008-11-20 23:01:55 +03:00
#define ZB_TOTAL DN_MAX_LEVELS
2008-11-20 23:01:55 +03:00
typedef struct zdb_cb {
zdb_blkstats_t zcb_type[ZB_TOTAL + 1][ZDB_OT_TOTAL + 1];
uint64_t zcb_dedup_asize;
uint64_t zcb_dedup_blocks;
uint64_t zcb_embedded_blocks[NUM_BP_EMBEDDED_TYPES];
uint64_t zcb_embedded_histogram[NUM_BP_EMBEDDED_TYPES]
[BPE_PAYLOAD_SIZE + 1];
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
uint64_t zcb_start;
uint64_t zcb_lastprint;
uint64_t zcb_totalasize;
2008-11-20 23:01:55 +03:00
uint64_t zcb_errors[256];
int zcb_readfails;
int zcb_haderrors;
spa_t *zcb_spa;
2008-11-20 23:01:55 +03:00
} zdb_cb_t;
static void
zdb_count_block(zdb_cb_t *zcb, zilog_t *zilog, const blkptr_t *bp,
dmu_object_type_t type)
2008-11-20 23:01:55 +03:00
{
uint64_t refcnt = 0;
int i;
ASSERT(type < ZDB_OT_TOTAL);
if (zilog && zil_bp_tree_add(zilog, bp) != 0)
return;
for (i = 0; i < 4; i++) {
2008-11-20 23:01:55 +03:00
int l = (i < 2) ? BP_GET_LEVEL(bp) : ZB_TOTAL;
int t = (i & 1) ? type : ZDB_OT_TOTAL;
int equal;
2008-11-20 23:01:55 +03:00
zdb_blkstats_t *zb = &zcb->zcb_type[l][t];
zb->zb_asize += BP_GET_ASIZE(bp);
zb->zb_lsize += BP_GET_LSIZE(bp);
zb->zb_psize += BP_GET_PSIZE(bp);
zb->zb_count++;
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 23:15:08 +03:00
/*
* The histogram is only big enough to record blocks up to
* SPA_OLD_MAXBLOCKSIZE; larger blocks go into the last,
* "other", bucket.
*/
int idx = BP_GET_PSIZE(bp) >> SPA_MINBLOCKSHIFT;
idx = MIN(idx, SPA_OLD_MAXBLOCKSIZE / SPA_MINBLOCKSIZE + 1);
zb->zb_psize_histogram[idx]++;
zb->zb_gangs += BP_COUNT_GANG(bp);
switch (BP_GET_NDVAS(bp)) {
case 2:
if (DVA_GET_VDEV(&bp->blk_dva[0]) ==
DVA_GET_VDEV(&bp->blk_dva[1]))
zb->zb_ditto_samevdev++;
break;
case 3:
equal = (DVA_GET_VDEV(&bp->blk_dva[0]) ==
DVA_GET_VDEV(&bp->blk_dva[1])) +
(DVA_GET_VDEV(&bp->blk_dva[0]) ==
DVA_GET_VDEV(&bp->blk_dva[2])) +
(DVA_GET_VDEV(&bp->blk_dva[1]) ==
DVA_GET_VDEV(&bp->blk_dva[2]));
if (equal != 0)
zb->zb_ditto_samevdev++;
break;
}
2008-11-20 23:01:55 +03:00
}
if (BP_IS_EMBEDDED(bp)) {
zcb->zcb_embedded_blocks[BPE_GET_ETYPE(bp)]++;
zcb->zcb_embedded_histogram[BPE_GET_ETYPE(bp)]
[BPE_GET_PSIZE(bp)]++;
return;
}
if (dump_opt['L'])
return;
if (BP_GET_DEDUP(bp)) {
ddt_t *ddt;
ddt_entry_t *dde;
ddt = ddt_select(zcb->zcb_spa, bp);
ddt_enter(ddt);
dde = ddt_lookup(ddt, bp, B_FALSE);
if (dde == NULL) {
refcnt = 0;
} else {
ddt_phys_t *ddp = ddt_phys_select(dde, bp);
ddt_phys_decref(ddp);
refcnt = ddp->ddp_refcnt;
if (ddt_phys_total_refcnt(dde) == 0)
ddt_remove(ddt, dde);
2008-11-20 23:01:55 +03:00
}
ddt_exit(ddt);
2008-11-20 23:01:55 +03:00
}
VERIFY3U(zio_wait(zio_claim(NULL, zcb->zcb_spa,
refcnt ? 0 : spa_first_txg(zcb->zcb_spa),
bp, NULL, NULL, ZIO_FLAG_CANFAIL)), ==, 0);
2008-11-20 23:01:55 +03:00
}
Illumos #3306, #3321 3306 zdb should be able to issue reads in parallel 3321 'zpool reopen' command should be documented in the man page and help Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com> Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: illumos/illumos-gate@31d7e8fa33fae995f558673adb22641b5aa8b6e1 https://www.illumos.org/issues/3306 https://www.illumos.org/issues/3321 The vdev_file.c implementation in this patch diverges significantly from the upstream version. For consistenty with the vdev_disk.c code the upstream version leverages the Illumos bio interfaces. This makes sense for Illumos but not for ZoL for two reasons. 1) The vdev_disk.c code in ZoL has been rewritten to use the Linux block device interfaces which differ significantly from those in Illumos. Therefore, updating the vdev_file.c to use the Illumos interfaces doesn't get you consistency with vdev_disk.c. 2) Using the upstream patch as is would requiring implementing compatibility code for those Solaris block device interfaces in user and kernel space. That additional complexity could lead to confusion and doesn't buy us anything. For these reasons I've opted to simply move the existing vn_rdwr() as is in to the taskq function. This has the advantage of being low risk and easy to understand. Moving the vn_rdwr() function in to its own taskq thread also neatly avoids the possibility of a stack overflow. Finally, because of the additional work which is being handled by the free taskq the number of threads has been increased. The thread count under Illumos defaults to 100 but was decreased to 2 in commit 08d08e due to contention. We increase it to 8 until the contention can be address by porting Illumos #3581. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1354
2013-05-03 03:36:32 +04:00
static void
zdb_blkptr_done(zio_t *zio)
{
spa_t *spa = zio->io_spa;
blkptr_t *bp = zio->io_bp;
int ioerr = zio->io_error;
zdb_cb_t *zcb = zio->io_private;
zbookmark_phys_t *zb = &zio->io_bookmark;
Illumos #3306, #3321 3306 zdb should be able to issue reads in parallel 3321 'zpool reopen' command should be documented in the man page and help Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com> Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: illumos/illumos-gate@31d7e8fa33fae995f558673adb22641b5aa8b6e1 https://www.illumos.org/issues/3306 https://www.illumos.org/issues/3321 The vdev_file.c implementation in this patch diverges significantly from the upstream version. For consistenty with the vdev_disk.c code the upstream version leverages the Illumos bio interfaces. This makes sense for Illumos but not for ZoL for two reasons. 1) The vdev_disk.c code in ZoL has been rewritten to use the Linux block device interfaces which differ significantly from those in Illumos. Therefore, updating the vdev_file.c to use the Illumos interfaces doesn't get you consistency with vdev_disk.c. 2) Using the upstream patch as is would requiring implementing compatibility code for those Solaris block device interfaces in user and kernel space. That additional complexity could lead to confusion and doesn't buy us anything. For these reasons I've opted to simply move the existing vn_rdwr() as is in to the taskq function. This has the advantage of being low risk and easy to understand. Moving the vn_rdwr() function in to its own taskq thread also neatly avoids the possibility of a stack overflow. Finally, because of the additional work which is being handled by the free taskq the number of threads has been increased. The thread count under Illumos defaults to 100 but was decreased to 2 in commit 08d08e due to contention. We increase it to 8 until the contention can be address by porting Illumos #3581. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1354
2013-05-03 03:36:32 +04:00
abd_free(zio->io_abd);
Illumos #3306, #3321 3306 zdb should be able to issue reads in parallel 3321 'zpool reopen' command should be documented in the man page and help Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com> Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: illumos/illumos-gate@31d7e8fa33fae995f558673adb22641b5aa8b6e1 https://www.illumos.org/issues/3306 https://www.illumos.org/issues/3321 The vdev_file.c implementation in this patch diverges significantly from the upstream version. For consistenty with the vdev_disk.c code the upstream version leverages the Illumos bio interfaces. This makes sense for Illumos but not for ZoL for two reasons. 1) The vdev_disk.c code in ZoL has been rewritten to use the Linux block device interfaces which differ significantly from those in Illumos. Therefore, updating the vdev_file.c to use the Illumos interfaces doesn't get you consistency with vdev_disk.c. 2) Using the upstream patch as is would requiring implementing compatibility code for those Solaris block device interfaces in user and kernel space. That additional complexity could lead to confusion and doesn't buy us anything. For these reasons I've opted to simply move the existing vn_rdwr() as is in to the taskq function. This has the advantage of being low risk and easy to understand. Moving the vn_rdwr() function in to its own taskq thread also neatly avoids the possibility of a stack overflow. Finally, because of the additional work which is being handled by the free taskq the number of threads has been increased. The thread count under Illumos defaults to 100 but was decreased to 2 in commit 08d08e due to contention. We increase it to 8 until the contention can be address by porting Illumos #3581. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1354
2013-05-03 03:36:32 +04:00
mutex_enter(&spa->spa_scrub_lock);
spa->spa_scrub_inflight--;
cv_broadcast(&spa->spa_scrub_io_cv);
if (ioerr && !(zio->io_flags & ZIO_FLAG_SPECULATIVE)) {
char blkbuf[BP_SPRINTF_LEN];
zcb->zcb_haderrors = 1;
zcb->zcb_errors[ioerr]++;
if (dump_opt['b'] >= 2)
snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
Illumos #3306, #3321 3306 zdb should be able to issue reads in parallel 3321 'zpool reopen' command should be documented in the man page and help Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com> Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: illumos/illumos-gate@31d7e8fa33fae995f558673adb22641b5aa8b6e1 https://www.illumos.org/issues/3306 https://www.illumos.org/issues/3321 The vdev_file.c implementation in this patch diverges significantly from the upstream version. For consistenty with the vdev_disk.c code the upstream version leverages the Illumos bio interfaces. This makes sense for Illumos but not for ZoL for two reasons. 1) The vdev_disk.c code in ZoL has been rewritten to use the Linux block device interfaces which differ significantly from those in Illumos. Therefore, updating the vdev_file.c to use the Illumos interfaces doesn't get you consistency with vdev_disk.c. 2) Using the upstream patch as is would requiring implementing compatibility code for those Solaris block device interfaces in user and kernel space. That additional complexity could lead to confusion and doesn't buy us anything. For these reasons I've opted to simply move the existing vn_rdwr() as is in to the taskq function. This has the advantage of being low risk and easy to understand. Moving the vn_rdwr() function in to its own taskq thread also neatly avoids the possibility of a stack overflow. Finally, because of the additional work which is being handled by the free taskq the number of threads has been increased. The thread count under Illumos defaults to 100 but was decreased to 2 in commit 08d08e due to contention. We increase it to 8 until the contention can be address by porting Illumos #3581. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1354
2013-05-03 03:36:32 +04:00
else
blkbuf[0] = '\0';
(void) printf("zdb_blkptr_cb: "
"Got error %d reading "
"<%llu, %llu, %lld, %llx> %s -- skipping\n",
ioerr,
(u_longlong_t)zb->zb_objset,
(u_longlong_t)zb->zb_object,
(u_longlong_t)zb->zb_level,
(u_longlong_t)zb->zb_blkid,
blkbuf);
}
mutex_exit(&spa->spa_scrub_lock);
}
2008-11-20 23:01:55 +03:00
static int
zdb_blkptr_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
2008-11-20 23:01:55 +03:00
{
zdb_cb_t *zcb = arg;
2009-07-03 02:44:48 +04:00
dmu_object_type_t type;
boolean_t is_metadata;
2008-11-20 23:01:55 +03:00
if (bp == NULL)
return (0);
if (dump_opt['b'] >= 5 && bp->blk_birth > 0) {
char blkbuf[BP_SPRINTF_LEN];
snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
(void) printf("objset %llu object %llu "
"level %lld offset 0x%llx %s\n",
(u_longlong_t)zb->zb_objset,
(u_longlong_t)zb->zb_object,
(longlong_t)zb->zb_level,
(u_longlong_t)blkid2offset(dnp, bp, zb),
blkbuf);
}
if (BP_IS_HOLE(bp))
return (0);
2009-07-03 02:44:48 +04:00
type = BP_GET_TYPE(bp);
zdb_count_block(zcb, zilog, bp,
(type & DMU_OT_NEWTYPE) ? ZDB_OT_OTHER : type);
is_metadata = (BP_GET_LEVEL(bp) != 0 || DMU_OT_IS_METADATA(type));
if (!BP_IS_EMBEDDED(bp) &&
(dump_opt['c'] > 1 || (dump_opt['c'] && is_metadata))) {
size_t size = BP_GET_PSIZE(bp);
abd_t *abd = abd_alloc(size, B_FALSE);
int flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCRUB | ZIO_FLAG_RAW;
2009-07-03 02:44:48 +04:00
/* If it's an intent log block, failure is expected. */
if (zb->zb_level == ZB_ZIL_LEVEL)
flags |= ZIO_FLAG_SPECULATIVE;
Illumos #3306, #3321 3306 zdb should be able to issue reads in parallel 3321 'zpool reopen' command should be documented in the man page and help Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com> Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: illumos/illumos-gate@31d7e8fa33fae995f558673adb22641b5aa8b6e1 https://www.illumos.org/issues/3306 https://www.illumos.org/issues/3321 The vdev_file.c implementation in this patch diverges significantly from the upstream version. For consistenty with the vdev_disk.c code the upstream version leverages the Illumos bio interfaces. This makes sense for Illumos but not for ZoL for two reasons. 1) The vdev_disk.c code in ZoL has been rewritten to use the Linux block device interfaces which differ significantly from those in Illumos. Therefore, updating the vdev_file.c to use the Illumos interfaces doesn't get you consistency with vdev_disk.c. 2) Using the upstream patch as is would requiring implementing compatibility code for those Solaris block device interfaces in user and kernel space. That additional complexity could lead to confusion and doesn't buy us anything. For these reasons I've opted to simply move the existing vn_rdwr() as is in to the taskq function. This has the advantage of being low risk and easy to understand. Moving the vn_rdwr() function in to its own taskq thread also neatly avoids the possibility of a stack overflow. Finally, because of the additional work which is being handled by the free taskq the number of threads has been increased. The thread count under Illumos defaults to 100 but was decreased to 2 in commit 08d08e due to contention. We increase it to 8 until the contention can be address by porting Illumos #3581. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1354
2013-05-03 03:36:32 +04:00
mutex_enter(&spa->spa_scrub_lock);
while (spa->spa_scrub_inflight > max_inflight)
cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
spa->spa_scrub_inflight++;
mutex_exit(&spa->spa_scrub_lock);
zio_nowait(zio_read(NULL, spa, bp, abd, size,
Illumos #3306, #3321 3306 zdb should be able to issue reads in parallel 3321 'zpool reopen' command should be documented in the man page and help Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com> Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: illumos/illumos-gate@31d7e8fa33fae995f558673adb22641b5aa8b6e1 https://www.illumos.org/issues/3306 https://www.illumos.org/issues/3321 The vdev_file.c implementation in this patch diverges significantly from the upstream version. For consistenty with the vdev_disk.c code the upstream version leverages the Illumos bio interfaces. This makes sense for Illumos but not for ZoL for two reasons. 1) The vdev_disk.c code in ZoL has been rewritten to use the Linux block device interfaces which differ significantly from those in Illumos. Therefore, updating the vdev_file.c to use the Illumos interfaces doesn't get you consistency with vdev_disk.c. 2) Using the upstream patch as is would requiring implementing compatibility code for those Solaris block device interfaces in user and kernel space. That additional complexity could lead to confusion and doesn't buy us anything. For these reasons I've opted to simply move the existing vn_rdwr() as is in to the taskq function. This has the advantage of being low risk and easy to understand. Moving the vn_rdwr() function in to its own taskq thread also neatly avoids the possibility of a stack overflow. Finally, because of the additional work which is being handled by the free taskq the number of threads has been increased. The thread count under Illumos defaults to 100 but was decreased to 2 in commit 08d08e due to contention. We increase it to 8 until the contention can be address by porting Illumos #3581. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1354
2013-05-03 03:36:32 +04:00
zdb_blkptr_done, zcb, ZIO_PRIORITY_ASYNC_READ, flags, zb));
2008-11-20 23:01:55 +03:00
}
zcb->zcb_readfails = 0;
/* only call gethrtime() every 100 blocks */
static int iters;
if (++iters > 100)
iters = 0;
else
return (0);
if (dump_opt['b'] < 5 && gethrtime() > zcb->zcb_lastprint + NANOSEC) {
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
uint64_t now = gethrtime();
char buf[10];
uint64_t bytes = zcb->zcb_type[ZB_TOTAL][ZDB_OT_TOTAL].zb_asize;
int kb_per_sec =
1 + bytes / (1 + ((now - zcb->zcb_start) / 1000 / 1000));
int sec_remaining =
(zcb->zcb_totalasize - bytes) / 1024 / kb_per_sec;
zfs_nicenum(bytes, buf, sizeof (buf));
(void) fprintf(stderr,
"\r%5s completed (%4dMB/s) "
"estimated time remaining: %uhr %02umin %02usec ",
buf, kb_per_sec / 1024,
sec_remaining / 60 / 60,
sec_remaining / 60 % 60,
sec_remaining % 60);
zcb->zcb_lastprint = now;
}
2008-11-20 23:01:55 +03:00
return (0);
}
static void
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
zdb_leak(void *arg, uint64_t start, uint64_t size)
{
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
vdev_t *vd = arg;
(void) printf("leaked space: vdev %llu, offset 0x%llx, size %llu\n",
(u_longlong_t)vd->vdev_id, (u_longlong_t)start, (u_longlong_t)size);
}
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
static metaslab_ops_t zdb_metaslab_ops = {
Illumos 4976-4984 - metaslab improvements 4976 zfs should only avoid writing to a failing non-redundant top-level vdev 4978 ztest fails in get_metaslab_refcount() 4979 extend free space histogram to device and pool 4980 metaslabs should have a fragmentation metric 4981 remove fragmented ops vector from block allocator 4982 space_map object should proactively upgrade when feature is enabled 4983 need to collect metaslab information via mdb 4984 device selection should use fragmentation metric Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <adam.leventhal@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/4976 https://www.illumos.org/issues/4978 https://www.illumos.org/issues/4979 https://www.illumos.org/issues/4980 https://www.illumos.org/issues/4981 https://www.illumos.org/issues/4982 https://www.illumos.org/issues/4983 https://www.illumos.org/issues/4984 https://github.com/illumos/illumos-gate/commit/2e4c998 Notes: The "zdb -M" option has been re-tasked to display the new metaslab fragmentation metric and the new "zdb -I" option is used to control the maximum number of in-flight I/Os. The new fragmentation metric is derived from the space map histogram which has been rolled up to the vdev and pool level and is presented to the user via "zpool list". Add a number of module parameters related to the new metaslab weighting logic. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2595
2014-07-20 00:19:24 +04:00
NULL /* alloc */
};
static void
zdb_ddt_leak_init(spa_t *spa, zdb_cb_t *zcb)
{
ddt_bookmark_t ddb = { 0 };
ddt_entry_t dde;
int error;
int p;
while ((error = ddt_walk(spa, &ddb, &dde)) == 0) {
blkptr_t blk;
ddt_phys_t *ddp = dde.dde_phys;
if (ddb.ddb_class == DDT_CLASS_UNIQUE)
return;
ASSERT(ddt_phys_total_refcnt(&dde) > 1);
for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
if (ddp->ddp_phys_birth == 0)
continue;
ddt_bp_create(ddb.ddb_checksum,
&dde.dde_key, ddp, &blk);
if (p == DDT_PHYS_DITTO) {
zdb_count_block(zcb, NULL, &blk, ZDB_OT_DITTO);
} else {
zcb->zcb_dedup_asize +=
BP_GET_ASIZE(&blk) * (ddp->ddp_refcnt - 1);
zcb->zcb_dedup_blocks++;
}
}
if (!dump_opt['L']) {
ddt_t *ddt = spa->spa_ddt[ddb.ddb_checksum];
ddt_enter(ddt);
VERIFY(ddt_lookup(ddt, &blk, B_TRUE) != NULL);
ddt_exit(ddt);
}
}
ASSERT(error == ENOENT);
}
static void
zdb_leak_init(spa_t *spa, zdb_cb_t *zcb)
{
zcb->zcb_spa = spa;
uint64_t c, m;
if (!dump_opt['L']) {
vdev_t *rvd = spa->spa_root_vdev;
/*
* We are going to be changing the meaning of the metaslab's
* ms_tree. Ensure that the allocator doesn't try to
* use the tree.
*/
spa->spa_normal_class->mc_ops = &zdb_metaslab_ops;
spa->spa_log_class->mc_ops = &zdb_metaslab_ops;
for (c = 0; c < rvd->vdev_children; c++) {
vdev_t *vd = rvd->vdev_child[c];
ASSERTV(metaslab_group_t *mg = vd->vdev_mg);
for (m = 0; m < vd->vdev_ms_count; m++) {
metaslab_t *msp = vd->vdev_ms[m];
ASSERT3P(msp->ms_group, ==, mg);
mutex_enter(&msp->ms_lock);
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
metaslab_unload(msp);
/*
* For leak detection, we overload the metaslab
* ms_tree to contain allocated segments
* instead of free segments. As a result,
* we can't use the normal metaslab_load/unload
* interfaces.
*/
if (msp->ms_sm != NULL) {
(void) fprintf(stderr,
"\rloading space map for "
"vdev %llu of %llu, "
"metaslab %llu of %llu ...",
(longlong_t)c,
(longlong_t)rvd->vdev_children,
(longlong_t)m,
(longlong_t)vd->vdev_ms_count);
/*
* We don't want to spend the CPU
* manipulating the size-ordered
* tree, so clear the range_tree
* ops.
*/
msp->ms_tree->rt_ops = NULL;
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
VERIFY0(space_map_load(msp->ms_sm,
msp->ms_tree, SM_ALLOC));
if (!msp->ms_loaded)
msp->ms_loaded = B_TRUE;
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
}
mutex_exit(&msp->ms_lock);
}
}
(void) fprintf(stderr, "\n");
}
spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
zdb_ddt_leak_init(spa, zcb);
spa_config_exit(spa, SCL_CONFIG, FTAG);
}
static void
zdb_leak_fini(spa_t *spa)
{
int c, m;
if (!dump_opt['L']) {
vdev_t *rvd = spa->spa_root_vdev;
for (c = 0; c < rvd->vdev_children; c++) {
vdev_t *vd = rvd->vdev_child[c];
ASSERTV(metaslab_group_t *mg = vd->vdev_mg);
for (m = 0; m < vd->vdev_ms_count; m++) {
metaslab_t *msp = vd->vdev_ms[m];
ASSERT3P(mg, ==, msp->ms_group);
mutex_enter(&msp->ms_lock);
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
/*
* The ms_tree has been overloaded to
* contain allocated segments. Now that we
* finished traversing all blocks, any
* block that remains in the ms_tree
* represents an allocated block that we
* did not claim during the traversal.
* Claimed blocks would have been removed
* from the ms_tree.
*/
range_tree_vacate(msp->ms_tree, zdb_leak, vd);
if (msp->ms_loaded)
msp->ms_loaded = B_FALSE;
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
mutex_exit(&msp->ms_lock);
}
}
}
}
/* ARGSUSED */
static int
count_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
{
zdb_cb_t *zcb = arg;
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
if (dump_opt['b'] >= 5) {
char blkbuf[BP_SPRINTF_LEN];
snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
(void) printf("[%s] %s\n",
"deferred free", blkbuf);
}
zdb_count_block(zcb, NULL, bp, ZDB_OT_DEFERRED);
return (0);
}
2008-11-20 23:01:55 +03:00
static int
dump_block_stats(spa_t *spa)
{
zdb_cb_t zcb;
2008-11-20 23:01:55 +03:00
zdb_blkstats_t *zb, *tzb;
uint64_t norm_alloc, norm_space, total_alloc, total_found;
int flags = TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA | TRAVERSE_HARD;
boolean_t leaks = B_FALSE;
int e, c;
bp_embedded_type_t i;
2008-11-20 23:01:55 +03:00
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
(void) printf("\nTraversing all blocks %s%s%s%s%s...\n\n",
(dump_opt['c'] || !dump_opt['L']) ? "to verify " : "",
(dump_opt['c'] == 1) ? "metadata " : "",
dump_opt['c'] ? "checksums " : "",
(dump_opt['c'] && !dump_opt['L']) ? "and verify " : "",
!dump_opt['L'] ? "nothing leaked " : "");
2008-11-20 23:01:55 +03:00
/*
* Load all space maps as SM_ALLOC maps, then traverse the pool
* claiming each block we discover. If the pool is perfectly
* consistent, the space maps will be empty when we're done.
* Anything left over is a leak; any block we can't claim (because
* it's not part of any space map) is a double allocation,
* reference to a freed block, or an unclaimed log block.
2008-11-20 23:01:55 +03:00
*/
bzero(&zcb, sizeof (zdb_cb_t));
zdb_leak_init(spa, &zcb);
2008-11-20 23:01:55 +03:00
/*
* If there's a deferred-free bplist, process that first.
*/
(void) bpobj_iterate_nofree(&spa->spa_deferred_bpobj,
count_block_cb, &zcb, NULL);
if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
(void) bpobj_iterate_nofree(&spa->spa_dsl_pool->dp_free_bpobj,
count_block_cb, &zcb, NULL);
}
if (spa_feature_is_active(spa, SPA_FEATURE_ASYNC_DESTROY)) {
VERIFY3U(0, ==, bptree_iterate(spa->spa_meta_objset,
spa->spa_dsl_pool->dp_bptree_obj, B_FALSE, count_block_cb,
&zcb, NULL));
}
2008-11-20 23:01:55 +03:00
if (dump_opt['c'] > 1)
flags |= TRAVERSE_PREFETCH_DATA;
2008-11-20 23:01:55 +03:00
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
zcb.zcb_totalasize = metaslab_class_get_alloc(spa_normal_class(spa));
zcb.zcb_start = zcb.zcb_lastprint = gethrtime();
zcb.zcb_haderrors |= traverse_pool(spa, 0, flags, zdb_blkptr_cb, &zcb);
2008-11-20 23:01:55 +03:00
Illumos #3306, #3321 3306 zdb should be able to issue reads in parallel 3321 'zpool reopen' command should be documented in the man page and help Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com> Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: illumos/illumos-gate@31d7e8fa33fae995f558673adb22641b5aa8b6e1 https://www.illumos.org/issues/3306 https://www.illumos.org/issues/3321 The vdev_file.c implementation in this patch diverges significantly from the upstream version. For consistenty with the vdev_disk.c code the upstream version leverages the Illumos bio interfaces. This makes sense for Illumos but not for ZoL for two reasons. 1) The vdev_disk.c code in ZoL has been rewritten to use the Linux block device interfaces which differ significantly from those in Illumos. Therefore, updating the vdev_file.c to use the Illumos interfaces doesn't get you consistency with vdev_disk.c. 2) Using the upstream patch as is would requiring implementing compatibility code for those Solaris block device interfaces in user and kernel space. That additional complexity could lead to confusion and doesn't buy us anything. For these reasons I've opted to simply move the existing vn_rdwr() as is in to the taskq function. This has the advantage of being low risk and easy to understand. Moving the vn_rdwr() function in to its own taskq thread also neatly avoids the possibility of a stack overflow. Finally, because of the additional work which is being handled by the free taskq the number of threads has been increased. The thread count under Illumos defaults to 100 but was decreased to 2 in commit 08d08e due to contention. We increase it to 8 until the contention can be address by porting Illumos #3581. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1354
2013-05-03 03:36:32 +04:00
/*
* If we've traversed the data blocks then we need to wait for those
* I/Os to complete. We leverage "The Godfather" zio to wait on
* all async I/Os to complete.
*/
if (dump_opt['c']) {
for (c = 0; c < max_ncpus; c++) {
(void) zio_wait(spa->spa_async_zio_root[c]);
spa->spa_async_zio_root[c] = zio_root(spa, NULL, NULL,
ZIO_FLAG_CANFAIL | ZIO_FLAG_SPECULATIVE |
ZIO_FLAG_GODFATHER);
}
Illumos #3306, #3321 3306 zdb should be able to issue reads in parallel 3321 'zpool reopen' command should be documented in the man page and help Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com> Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: illumos/illumos-gate@31d7e8fa33fae995f558673adb22641b5aa8b6e1 https://www.illumos.org/issues/3306 https://www.illumos.org/issues/3321 The vdev_file.c implementation in this patch diverges significantly from the upstream version. For consistenty with the vdev_disk.c code the upstream version leverages the Illumos bio interfaces. This makes sense for Illumos but not for ZoL for two reasons. 1) The vdev_disk.c code in ZoL has been rewritten to use the Linux block device interfaces which differ significantly from those in Illumos. Therefore, updating the vdev_file.c to use the Illumos interfaces doesn't get you consistency with vdev_disk.c. 2) Using the upstream patch as is would requiring implementing compatibility code for those Solaris block device interfaces in user and kernel space. That additional complexity could lead to confusion and doesn't buy us anything. For these reasons I've opted to simply move the existing vn_rdwr() as is in to the taskq function. This has the advantage of being low risk and easy to understand. Moving the vn_rdwr() function in to its own taskq thread also neatly avoids the possibility of a stack overflow. Finally, because of the additional work which is being handled by the free taskq the number of threads has been increased. The thread count under Illumos defaults to 100 but was decreased to 2 in commit 08d08e due to contention. We increase it to 8 until the contention can be address by porting Illumos #3581. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1354
2013-05-03 03:36:32 +04:00
}
if (zcb.zcb_haderrors) {
2008-11-20 23:01:55 +03:00
(void) printf("\nError counts:\n\n");
(void) printf("\t%5s %s\n", "errno", "count");
for (e = 0; e < 256; e++) {
2008-11-20 23:01:55 +03:00
if (zcb.zcb_errors[e] != 0) {
(void) printf("\t%5d %llu\n",
e, (u_longlong_t)zcb.zcb_errors[e]);
}
}
}
/*
* Report any leaked segments.
*/
zdb_leak_fini(spa);
2008-11-20 23:01:55 +03:00
tzb = &zcb.zcb_type[ZB_TOTAL][ZDB_OT_TOTAL];
2008-11-20 23:01:55 +03:00
norm_alloc = metaslab_class_get_alloc(spa_normal_class(spa));
norm_space = metaslab_class_get_space(spa_normal_class(spa));
2008-11-20 23:01:55 +03:00
total_alloc = norm_alloc + metaslab_class_get_alloc(spa_log_class(spa));
total_found = tzb->zb_asize - zcb.zcb_dedup_asize;
2008-11-20 23:01:55 +03:00
if (total_found == total_alloc) {
2009-01-16 00:59:39 +03:00
if (!dump_opt['L'])
(void) printf("\n\tNo leaks (block sum matches space"
" maps exactly)\n");
2008-11-20 23:01:55 +03:00
} else {
(void) printf("block traversal size %llu != alloc %llu "
2009-01-16 00:59:39 +03:00
"(%s %lld)\n",
(u_longlong_t)total_found,
(u_longlong_t)total_alloc,
2009-01-16 00:59:39 +03:00
(dump_opt['L']) ? "unreachable" : "leaked",
(longlong_t)(total_alloc - total_found));
leaks = B_TRUE;
2008-11-20 23:01:55 +03:00
}
if (tzb->zb_count == 0)
return (2);
(void) printf("\n");
(void) printf("\tbp count: %10llu\n",
(u_longlong_t)tzb->zb_count);
(void) printf("\tganged count: %10llu\n",
(longlong_t)tzb->zb_gangs);
(void) printf("\tbp logical: %10llu avg: %6llu\n",
2008-11-20 23:01:55 +03:00
(u_longlong_t)tzb->zb_lsize,
(u_longlong_t)(tzb->zb_lsize / tzb->zb_count));
(void) printf("\tbp physical: %10llu avg:"
" %6llu compression: %6.2f\n",
2008-11-20 23:01:55 +03:00
(u_longlong_t)tzb->zb_psize,
(u_longlong_t)(tzb->zb_psize / tzb->zb_count),
(double)tzb->zb_lsize / tzb->zb_psize);
(void) printf("\tbp allocated: %10llu avg:"
" %6llu compression: %6.2f\n",
2008-11-20 23:01:55 +03:00
(u_longlong_t)tzb->zb_asize,
(u_longlong_t)(tzb->zb_asize / tzb->zb_count),
(double)tzb->zb_lsize / tzb->zb_asize);
(void) printf("\tbp deduped: %10llu ref>1:"
" %6llu deduplication: %6.2f\n",
(u_longlong_t)zcb.zcb_dedup_asize,
(u_longlong_t)zcb.zcb_dedup_blocks,
(double)zcb.zcb_dedup_asize / tzb->zb_asize + 1.0);
(void) printf("\tSPA allocated: %10llu used: %5.2f%%\n",
(u_longlong_t)norm_alloc, 100.0 * norm_alloc / norm_space);
2008-11-20 23:01:55 +03:00
for (i = 0; i < NUM_BP_EMBEDDED_TYPES; i++) {
if (zcb.zcb_embedded_blocks[i] == 0)
continue;
(void) printf("\n");
(void) printf("\tadditional, non-pointer bps of type %u: "
"%10llu\n",
i, (u_longlong_t)zcb.zcb_embedded_blocks[i]);
if (dump_opt['b'] >= 3) {
(void) printf("\t number of (compressed) bytes: "
"number of bps\n");
dump_histogram(zcb.zcb_embedded_histogram[i],
sizeof (zcb.zcb_embedded_histogram[i]) /
sizeof (zcb.zcb_embedded_histogram[i][0]), 0);
}
}
if (tzb->zb_ditto_samevdev != 0) {
(void) printf("\tDittoed blocks on same vdev: %llu\n",
(longlong_t)tzb->zb_ditto_samevdev);
}
2008-11-20 23:01:55 +03:00
if (dump_opt['b'] >= 2) {
int l, t, level;
(void) printf("\nBlocks\tLSIZE\tPSIZE\tASIZE"
"\t avg\t comp\t%%Total\tType\n");
for (t = 0; t <= ZDB_OT_TOTAL; t++) {
char csize[32], lsize[32], psize[32], asize[32];
char avg[32], gang[32];
2008-11-20 23:01:55 +03:00
char *typename;
if (t < DMU_OT_NUMTYPES)
typename = dmu_ot[t].ot_name;
else
typename = zdb_ot_extname[t - DMU_OT_NUMTYPES];
2008-11-20 23:01:55 +03:00
if (zcb.zcb_type[ZB_TOTAL][t].zb_asize == 0) {
(void) printf("%6s\t%5s\t%5s\t%5s"
"\t%5s\t%5s\t%6s\t%s\n",
"-",
"-",
"-",
"-",
"-",
"-",
"-",
typename);
continue;
}
for (l = ZB_TOTAL - 1; l >= -1; l--) {
level = (l == -1 ? ZB_TOTAL : l);
zb = &zcb.zcb_type[level][t];
if (zb->zb_asize == 0)
continue;
if (dump_opt['b'] < 3 && level != ZB_TOTAL)
continue;
if (level == 0 && zb->zb_asize ==
zcb.zcb_type[ZB_TOTAL][t].zb_asize)
continue;
zdb_nicenum(zb->zb_count, csize);
zdb_nicenum(zb->zb_lsize, lsize);
zdb_nicenum(zb->zb_psize, psize);
zdb_nicenum(zb->zb_asize, asize);
zdb_nicenum(zb->zb_asize / zb->zb_count, avg);
zdb_nicenum(zb->zb_gangs, gang);
2008-11-20 23:01:55 +03:00
(void) printf("%6s\t%5s\t%5s\t%5s\t%5s"
"\t%5.2f\t%6.2f\t",
csize, lsize, psize, asize, avg,
(double)zb->zb_lsize / zb->zb_psize,
100.0 * zb->zb_asize / tzb->zb_asize);
if (level == ZB_TOTAL)
(void) printf("%s\n", typename);
else
(void) printf(" L%d %s\n",
level, typename);
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
if (dump_opt['b'] >= 3 && zb->zb_gangs > 0) {
(void) printf("\t number of ganged "
"blocks: %s\n", gang);
}
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
if (dump_opt['b'] >= 4) {
(void) printf("psize "
"(in 512-byte sectors): "
"number of blocks\n");
dump_histogram(zb->zb_psize_histogram,
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
PSIZE_HISTO_SIZE, 0);
Illumos #3641 compressed block histograms with zdb This patch is a zdb extension of the '-b' option, producing a histogram of the physical compressed block sizes per DMU object type on disk. The '-bbbb' option to zdb will uncover this new feature; here's an example usage on a new pool and snippet of the output it generates: # zpool create tank /dev/vd{b,c,d} # dd bs=1k if=/dev/urandom of=/tank/1kfile count=1 # dd bs=3k if=/dev/urandom of=/tank/3kfile count=1 # dd bs=64k if=/dev/urandom of=/tank/64kfile count=1 # zdb -bbbb tank ... 3 68.0K 68.0K 68.0K 22.7K 1.00 34.26 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... The blocks are also broken down by their indirection level. Expanding on the above example: # zfs set recordsize=1k tank # dd bs=1k if=/dev/urandom of=/tank/2x1kfile count=2 # zdb -bbbb tank ... 1 16K 1K 2K 2K 16.00 1.02 L1 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 1 * 5 70.0K 70.0K 70.0K 14.0K 1.00 35.71 L0 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 3 *** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * 6 86.0K 71.0K 72.0K 12.0K 1.21 36.73 ZFS plain file psize (in 512-byte sectors): number of blocks 2: 4 **** 3: 0 4: 0 5: 0 6: 1 * 7: 0 ... 127: 0 128: 1 * ... There's now a single 1K L1 block which is the indirect block needed for the '2x1kfile' file just created, as well as two more 1K L0 blocks from the same file. This can be used to get a distribution of the block sizes used within the pool, on a per object type basis. References: https://illumos.org/issues/3641 https://github.com/illumos/illumos-gate/commit/490d05b Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Signed-off-by: Boris Protopopov <boris.protopopov@me.com> Closes #2456
2013-03-25 01:24:51 +04:00
}
2008-11-20 23:01:55 +03:00
}
}
}
(void) printf("\n");
if (leaks)
return (2);
if (zcb.zcb_haderrors)
return (3);
return (0);
}
typedef struct zdb_ddt_entry {
ddt_key_t zdde_key;
uint64_t zdde_ref_blocks;
uint64_t zdde_ref_lsize;
uint64_t zdde_ref_psize;
uint64_t zdde_ref_dsize;
avl_node_t zdde_node;
} zdb_ddt_entry_t;
/* ARGSUSED */
static int
zdb_ddt_add_cb(spa_t *spa, zilog_t *zilog, const blkptr_t *bp,
const zbookmark_phys_t *zb, const dnode_phys_t *dnp, void *arg)
{
avl_tree_t *t = arg;
avl_index_t where;
zdb_ddt_entry_t *zdde, zdde_search;
if (bp == NULL || BP_IS_HOLE(bp) || BP_IS_EMBEDDED(bp))
return (0);
if (dump_opt['S'] > 1 && zb->zb_level == ZB_ROOT_LEVEL) {
(void) printf("traversing objset %llu, %llu objects, "
"%lu blocks so far\n",
(u_longlong_t)zb->zb_objset,
(u_longlong_t)BP_GET_FILL(bp),
avl_numnodes(t));
}
if (BP_IS_HOLE(bp) || BP_GET_CHECKSUM(bp) == ZIO_CHECKSUM_OFF ||
BP_GET_LEVEL(bp) > 0 || DMU_OT_IS_METADATA(BP_GET_TYPE(bp)))
return (0);
ddt_key_fill(&zdde_search.zdde_key, bp);
zdde = avl_find(t, &zdde_search, &where);
if (zdde == NULL) {
zdde = umem_zalloc(sizeof (*zdde), UMEM_NOFAIL);
zdde->zdde_key = zdde_search.zdde_key;
avl_insert(t, zdde, where);
}
zdde->zdde_ref_blocks += 1;
zdde->zdde_ref_lsize += BP_GET_LSIZE(bp);
zdde->zdde_ref_psize += BP_GET_PSIZE(bp);
zdde->zdde_ref_dsize += bp_get_dsize_sync(spa, bp);
return (0);
}
static void
dump_simulated_ddt(spa_t *spa)
{
avl_tree_t t;
void *cookie = NULL;
zdb_ddt_entry_t *zdde;
ddt_histogram_t ddh_total;
ddt_stat_t dds_total;
bzero(&ddh_total, sizeof (ddt_histogram_t));
bzero(&dds_total, sizeof (ddt_stat_t));
avl_create(&t, ddt_entry_compare,
sizeof (zdb_ddt_entry_t), offsetof(zdb_ddt_entry_t, zdde_node));
spa_config_enter(spa, SCL_CONFIG, FTAG, RW_READER);
(void) traverse_pool(spa, 0, TRAVERSE_PRE | TRAVERSE_PREFETCH_METADATA,
zdb_ddt_add_cb, &t);
spa_config_exit(spa, SCL_CONFIG, FTAG);
while ((zdde = avl_destroy_nodes(&t, &cookie)) != NULL) {
ddt_stat_t dds;
uint64_t refcnt = zdde->zdde_ref_blocks;
ASSERT(refcnt != 0);
dds.dds_blocks = zdde->zdde_ref_blocks / refcnt;
dds.dds_lsize = zdde->zdde_ref_lsize / refcnt;
dds.dds_psize = zdde->zdde_ref_psize / refcnt;
dds.dds_dsize = zdde->zdde_ref_dsize / refcnt;
dds.dds_ref_blocks = zdde->zdde_ref_blocks;
dds.dds_ref_lsize = zdde->zdde_ref_lsize;
dds.dds_ref_psize = zdde->zdde_ref_psize;
dds.dds_ref_dsize = zdde->zdde_ref_dsize;
ddt_stat_add(&ddh_total.ddh_stat[highbit64(refcnt) - 1],
&dds, 0);
umem_free(zdde, sizeof (*zdde));
}
avl_destroy(&t);
ddt_histogram_stat(&dds_total, &ddh_total);
(void) printf("Simulated DDT histogram:\n");
zpool_dump_ddt(&dds_total, &ddh_total);
dump_dedup_ratio(&dds_total);
}
2008-11-20 23:01:55 +03:00
static void
dump_zpool(spa_t *spa)
{
dsl_pool_t *dp = spa_get_dsl(spa);
int rc = 0;
if (dump_opt['S']) {
dump_simulated_ddt(spa);
return;
}
if (!dump_opt['e'] && dump_opt['C'] > 1) {
(void) printf("\nCached configuration:\n");
dump_nvlist(spa->spa_config, 8);
}
if (dump_opt['C'])
dump_config(spa);
2008-11-20 23:01:55 +03:00
if (dump_opt['u'])
dump_uberblock(&spa->spa_uberblock, "\nUberblock:\n", "\n");
2008-11-20 23:01:55 +03:00
if (dump_opt['D'])
dump_all_ddts(spa);
if (dump_opt['d'] > 2 || dump_opt['m'])
dump_metaslabs(spa);
Illumos 4976-4984 - metaslab improvements 4976 zfs should only avoid writing to a failing non-redundant top-level vdev 4978 ztest fails in get_metaslab_refcount() 4979 extend free space histogram to device and pool 4980 metaslabs should have a fragmentation metric 4981 remove fragmented ops vector from block allocator 4982 space_map object should proactively upgrade when feature is enabled 4983 need to collect metaslab information via mdb 4984 device selection should use fragmentation metric Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <adam.leventhal@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/4976 https://www.illumos.org/issues/4978 https://www.illumos.org/issues/4979 https://www.illumos.org/issues/4980 https://www.illumos.org/issues/4981 https://www.illumos.org/issues/4982 https://www.illumos.org/issues/4983 https://www.illumos.org/issues/4984 https://github.com/illumos/illumos-gate/commit/2e4c998 Notes: The "zdb -M" option has been re-tasked to display the new metaslab fragmentation metric and the new "zdb -I" option is used to control the maximum number of in-flight I/Os. The new fragmentation metric is derived from the space map histogram which has been rolled up to the vdev and pool level and is presented to the user via "zpool list". Add a number of module parameters related to the new metaslab weighting logic. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2595
2014-07-20 00:19:24 +04:00
if (dump_opt['M'])
dump_metaslab_groups(spa);
if (dump_opt['d'] || dump_opt['i']) {
spa_feature_t f;
2008-11-20 23:01:55 +03:00
dump_dir(dp->dp_meta_objset);
if (dump_opt['d'] >= 3) {
dump_full_bpobj(&spa->spa_deferred_bpobj,
"Deferred frees", 0);
if (spa_version(spa) >= SPA_VERSION_DEADLISTS) {
dump_full_bpobj(
&spa->spa_dsl_pool->dp_free_bpobj,
"Pool snapshot frees", 0);
}
if (spa_feature_is_active(spa,
SPA_FEATURE_ASYNC_DESTROY)) {
dump_bptree(spa->spa_meta_objset,
spa->spa_dsl_pool->dp_bptree_obj,
"Pool dataset frees");
}
2008-11-20 23:01:55 +03:00
dump_dtl(spa->spa_root_vdev, 0);
}
(void) dmu_objset_find(spa_name(spa), dump_one_dir,
NULL, DS_FIND_SNAPSHOTS | DS_FIND_CHILDREN);
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 23:15:08 +03:00
for (f = 0; f < SPA_FEATURES; f++) {
uint64_t refcount;
if (!(spa_feature_table[f].fi_flags &
ZFEATURE_FLAG_PER_DATASET) ||
!spa_feature_is_enabled(spa, f)) {
ASSERT0(dataset_feature_count[f]);
continue;
}
if (feature_get_refcount(spa, &spa_feature_table[f],
&refcount) == ENOTSUP)
continue;
if (dataset_feature_count[f] != refcount) {
(void) printf("%s feature refcount mismatch: "
"%lld datasets != %lld refcount\n",
spa_feature_table[f].fi_uname,
(longlong_t)dataset_feature_count[f],
(longlong_t)refcount);
rc = 2;
} else {
(void) printf("Verified %s feature refcount "
"of %llu is correct\n",
spa_feature_table[f].fi_uname,
(longlong_t)refcount);
}
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 23:15:08 +03:00
}
2008-11-20 23:01:55 +03:00
}
Illumos 5027 - zfs large block support 5027 zfs large block support Reviewed by: Alek Pinchuk <pinchuk.alek@gmail.com> Reviewed by: George Wilson <george.wilson@delphix.com> Reviewed by: Josef 'Jeff' Sipek <josef.sipek@nexenta.com> Reviewed by: Richard Elling <richard.elling@richardelling.com> Reviewed by: Saso Kiselkov <skiselkov.ml@gmail.com> Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov> Approved by: Dan McDonald <danmcd@omniti.com> References: https://www.illumos.org/issues/5027 https://github.com/illumos/illumos-gate/commit/b515258 Porting Notes: * Included in this patch is a tiny ISP2() cleanup in zio_init() from Illumos 5255. * Unlike the upstream Illumos commit this patch does not impose an arbitrary 128K block size limit on volumes. Volumes, like filesystems, are limited by the zfs_max_recordsize=1M module option. * By default the maximum record size is limited to 1M by the module option zfs_max_recordsize. This value may be safely increased up to 16M which is the largest block size supported by the on-disk format. At the moment, 1M blocks clearly offer a significant performance improvement but the benefits of going beyond this for the majority of workloads are less clear. * The illumos version of this patch increased DMU_MAX_ACCESS to 32M. This was determined not to be large enough when using 16M blocks because the zfs_make_xattrdir() function will fail (EFBIG) when assigning a TX. This was immediately observed under Linux because all newly created files must have a security xattr created and that was failing. Therefore, we've set DMU_MAX_ACCESS to 64M. * On 32-bit platforms a hard limit of 1M is set for blocks due to the limited virtual address space. We should be able to relax this one the ABD patches are merged. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #354
2014-11-03 23:15:08 +03:00
if (rc == 0 && (dump_opt['b'] || dump_opt['c']))
2008-11-20 23:01:55 +03:00
rc = dump_block_stats(spa);
Illumos #4101, #4102, #4103, #4105, #4106 4101 metaslab_debug should allow for fine-grained control 4102 space_maps should store more information about themselves 4103 space map object blocksize should be increased 4105 removing a mirrored log device results in a leaked object 4106 asynchronously load metaslab Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Sebastien Roy <seb@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> Prior to this patch, space_maps were preferred solely based on the amount of free space left in each. Unfortunately, this heuristic didn't contain any information about the make-up of that free space, which meant we could keep preferring and loading a highly fragmented space map that wouldn't actually have enough contiguous space to satisfy the allocation; then unloading that space_map and repeating the process. This change modifies the space_map's to store additional information about the contiguous space in the space_map, so that we can use this information to make a better decision about which space_map to load. This requires reallocating all space_map objects to increase their bonus buffer size sizes enough to fit the new metadata. The above feature can be enabled via a new feature flag introduced by this change: com.delphix:spacemap_histogram In addition to the above, this patch allows the space_map block size to be increase. Currently the block size is set to be 4K in size, which has certain implications including the following: * 4K sector devices will not see any compression benefit * large space_maps require more metadata on-disk * large space_maps require more time to load (typically random reads) Now the space_map block size can adjust as needed up to the maximum size set via the space_map_max_blksz variable. A bug was fixed which resulted in potentially leaking an object when removing a mirrored log device. The previous logic for vdev_remove() did not deal with removing top-level vdevs that are interior vdevs (i.e. mirror) correctly. The problem would occur when removing a mirrored log device, and result in the DTL space map object being leaked; because top-level vdevs don't have DTL space map objects associated with them. References: https://www.illumos.org/issues/4101 https://www.illumos.org/issues/4102 https://www.illumos.org/issues/4103 https://www.illumos.org/issues/4105 https://www.illumos.org/issues/4106 https://github.com/illumos/illumos-gate/commit/0713e23 Porting notes: A handful of kmem_alloc() calls were converted to kmem_zalloc(). Also, the KM_PUSHPAGE and TQ_PUSHPAGE flags were used as necessary. Ported-by: Tim Chase <tim@chase2k.com> Signed-off-by: Prakash Surya <surya1@llnl.gov> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2488
2013-10-02 01:25:53 +04:00
if (rc == 0)
rc = verify_spacemap_refcounts(spa);
2008-11-20 23:01:55 +03:00
if (dump_opt['s'])
show_pool_stats(spa);
if (dump_opt['h'])
dump_history(spa);
if (rc != 0) {
dump_debug_buffer();
2008-11-20 23:01:55 +03:00
exit(rc);
}
2008-11-20 23:01:55 +03:00
}
#define ZDB_FLAG_CHECKSUM 0x0001
#define ZDB_FLAG_DECOMPRESS 0x0002
#define ZDB_FLAG_BSWAP 0x0004
#define ZDB_FLAG_GBH 0x0008
#define ZDB_FLAG_INDIRECT 0x0010
#define ZDB_FLAG_PHYS 0x0020
#define ZDB_FLAG_RAW 0x0040
#define ZDB_FLAG_PRINT_BLKPTR 0x0080
int flagbits[256];
static void
zdb_print_blkptr(blkptr_t *bp, int flags)
{
char blkbuf[BP_SPRINTF_LEN];
2008-11-20 23:01:55 +03:00
if (flags & ZDB_FLAG_BSWAP)
byteswap_uint64_array((void *)bp, sizeof (blkptr_t));
snprintf_blkptr(blkbuf, sizeof (blkbuf), bp);
(void) printf("%s\n", blkbuf);
2008-11-20 23:01:55 +03:00
}
static void
zdb_dump_indirect(blkptr_t *bp, int nbps, int flags)
{
int i;
for (i = 0; i < nbps; i++)
zdb_print_blkptr(&bp[i], flags);
}
static void
zdb_dump_gbh(void *buf, int flags)
{
zdb_dump_indirect((blkptr_t *)buf, SPA_GBH_NBLKPTRS, flags);
}
static void
zdb_dump_block_raw(void *buf, uint64_t size, int flags)
{
if (flags & ZDB_FLAG_BSWAP)
byteswap_uint64_array(buf, size);
VERIFY(write(fileno(stdout), buf, size) == size);
2008-11-20 23:01:55 +03:00
}
static void
zdb_dump_block(char *label, void *buf, uint64_t size, int flags)
{
uint64_t *d = (uint64_t *)buf;
int nwords = size / sizeof (uint64_t);
int do_bswap = !!(flags & ZDB_FLAG_BSWAP);
int i, j;
char *hdr, *c;
if (do_bswap)
hdr = " 7 6 5 4 3 2 1 0 f e d c b a 9 8";
else
hdr = " 0 1 2 3 4 5 6 7 8 9 a b c d e f";
(void) printf("\n%s\n%6s %s 0123456789abcdef\n", label, "", hdr);
#ifdef _LITTLE_ENDIAN
/* correct the endianness */
do_bswap = !do_bswap;
#endif
2008-11-20 23:01:55 +03:00
for (i = 0; i < nwords; i += 2) {
(void) printf("%06llx: %016llx %016llx ",
(u_longlong_t)(i * sizeof (uint64_t)),
(u_longlong_t)(do_bswap ? BSWAP_64(d[i]) : d[i]),
(u_longlong_t)(do_bswap ? BSWAP_64(d[i + 1]) : d[i + 1]));
c = (char *)&d[i];
for (j = 0; j < 2 * sizeof (uint64_t); j++)
(void) printf("%c", isprint(c[j]) ? c[j] : '.');
(void) printf("\n");
}
}
/*
* There are two acceptable formats:
* leaf_name - For example: c1t0d0 or /tmp/ztest.0a
* child[.child]* - For example: 0.1.1
*
* The second form can be used to specify arbitrary vdevs anywhere
* in the hierarchy. For example, in a pool with a mirror of
2008-11-20 23:01:55 +03:00
* RAID-Zs, you can specify either RAID-Z vdev with 0.0 or 0.1 .
*/
static vdev_t *
zdb_vdev_lookup(vdev_t *vdev, char *path)
{
char *s, *p, *q;
int i;
if (vdev == NULL)
return (NULL);
/* First, assume the x.x.x.x format */
i = (int)strtoul(path, &s, 10);
if (s == path || (s && *s != '.' && *s != '\0'))
goto name;
if (i < 0 || i >= vdev->vdev_children)
return (NULL);
vdev = vdev->vdev_child[i];
if (s && *s == '\0')
2008-11-20 23:01:55 +03:00
return (vdev);
return (zdb_vdev_lookup(vdev, s+1));
name:
for (i = 0; i < vdev->vdev_children; i++) {
vdev_t *vc = vdev->vdev_child[i];
if (vc->vdev_path == NULL) {
vc = zdb_vdev_lookup(vc, path);
if (vc == NULL)
continue;
else
return (vc);
}
p = strrchr(vc->vdev_path, '/');
p = p ? p + 1 : vc->vdev_path;
q = &vc->vdev_path[strlen(vc->vdev_path) - 2];
if (strcmp(vc->vdev_path, path) == 0)
return (vc);
if (strcmp(p, path) == 0)
return (vc);
if (strcmp(q, "s0") == 0 && strncmp(p, path, q - p) == 0)
return (vc);
}
return (NULL);
}
/* ARGSUSED */
static int
random_get_pseudo_bytes_cb(void *buf, size_t len, void *unused)
{
return (random_get_pseudo_bytes(buf, len));
}
2008-11-20 23:01:55 +03:00
/*
* Read a block from a pool and print it out. The syntax of the
* block descriptor is:
*
* pool:vdev_specifier:offset:size[:flags]
*
* pool - The name of the pool you wish to read from
* vdev_specifier - Which vdev (see comment for zdb_vdev_lookup)
* offset - offset, in hex, in bytes
* size - Amount of data to read, in hex, in bytes
* flags - A string of characters specifying options
* b: Decode a blkptr at given offset within block
* *c: Calculate and display checksums
* d: Decompress data before dumping
2008-11-20 23:01:55 +03:00
* e: Byteswap data before dumping
* g: Display data as a gang block header
* i: Display as an indirect block
2008-11-20 23:01:55 +03:00
* p: Do I/O to physical offset
* r: Dump raw data to stdout
*
* * = not yet implemented
*/
static void
zdb_read_block(char *thing, spa_t *spa)
2008-11-20 23:01:55 +03:00
{
blkptr_t blk, *bp = &blk;
dva_t *dva = bp->blk_dva;
2008-11-20 23:01:55 +03:00
int flags = 0;
uint64_t offset = 0, size = 0, psize = 0, lsize = 0, blkptr_offset = 0;
2008-11-20 23:01:55 +03:00
zio_t *zio;
vdev_t *vd;
abd_t *pabd;
void *lbuf, *buf;
char *s, *p, *dup, *vdev, *flagstr;
int i, error;
boolean_t borrowed = B_FALSE;
2008-11-20 23:01:55 +03:00
dup = strdup(thing);
s = strtok(dup, ":");
vdev = s ? s : "";
s = strtok(NULL, ":");
offset = strtoull(s ? s : "", NULL, 16);
s = strtok(NULL, ":");
size = strtoull(s ? s : "", NULL, 16);
s = strtok(NULL, ":");
flagstr = s ? s : "";
s = NULL;
if (size == 0)
s = "size must not be zero";
if (!IS_P2ALIGNED(size, DEV_BSIZE))
s = "size must be a multiple of sector size";
if (!IS_P2ALIGNED(offset, DEV_BSIZE))
s = "offset must be a multiple of sector size";
if (s) {
(void) printf("Invalid block specifier: %s - %s\n", thing, s);
free(dup);
return;
}
for (s = strtok(flagstr, ":"); s; s = strtok(NULL, ":")) {
for (i = 0; flagstr[i]; i++) {
int bit = flagbits[(uchar_t)flagstr[i]];
if (bit == 0) {
(void) printf("***Invalid flag: %c\n",
flagstr[i]);
continue;
}
flags |= bit;
/* If it's not something with an argument, keep going */
if ((bit & (ZDB_FLAG_CHECKSUM |
2008-11-20 23:01:55 +03:00
ZDB_FLAG_PRINT_BLKPTR)) == 0)
continue;
p = &flagstr[i + 1];
if (bit == ZDB_FLAG_PRINT_BLKPTR) {
2008-11-20 23:01:55 +03:00
blkptr_offset = strtoull(p, &p, 16);
i = p - &flagstr[i + 1];
}
2008-11-20 23:01:55 +03:00
if (*p != ':' && *p != '\0') {
(void) printf("***Invalid flag arg: '%s'\n", s);
free(dup);
return;
}
}
}
vd = zdb_vdev_lookup(spa->spa_root_vdev, vdev);
if (vd == NULL) {
(void) printf("***Invalid vdev: %s\n", vdev);
free(dup);
return;
} else {
if (vd->vdev_path)
(void) fprintf(stderr, "Found vdev: %s\n",
vd->vdev_path);
2008-11-20 23:01:55 +03:00
else
(void) fprintf(stderr, "Found vdev type: %s\n",
2008-11-20 23:01:55 +03:00
vd->vdev_ops->vdev_op_type);
}
psize = size;
lsize = size;
pabd = abd_alloc_for_io(SPA_MAXBLOCKSIZE, B_FALSE);
lbuf = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
BP_ZERO(bp);
2008-11-20 23:01:55 +03:00
DVA_SET_VDEV(&dva[0], vd->vdev_id);
DVA_SET_OFFSET(&dva[0], offset);
DVA_SET_GANG(&dva[0], !!(flags & ZDB_FLAG_GBH));
DVA_SET_ASIZE(&dva[0], vdev_psize_to_asize(vd, psize));
BP_SET_BIRTH(bp, TXG_INITIAL, TXG_INITIAL);
BP_SET_LSIZE(bp, lsize);
BP_SET_PSIZE(bp, psize);
BP_SET_COMPRESS(bp, ZIO_COMPRESS_OFF);
BP_SET_CHECKSUM(bp, ZIO_CHECKSUM_OFF);
BP_SET_TYPE(bp, DMU_OT_NONE);
BP_SET_LEVEL(bp, 0);
BP_SET_DEDUP(bp, 0);
BP_SET_BYTEORDER(bp, ZFS_HOST_BYTEORDER);
2008-11-20 23:01:55 +03:00
spa_config_enter(spa, SCL_STATE, FTAG, RW_READER);
2008-11-20 23:01:55 +03:00
zio = zio_root(spa, NULL, NULL, 0);
if (vd == vd->vdev_top) {
/*
* Treat this as a normal block read.
*/
zio_nowait(zio_read(zio, spa, bp, pabd, psize, NULL, NULL,
ZIO_PRIORITY_SYNC_READ,
ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW, NULL));
} else {
/*
* Treat this as a vdev child I/O.
*/
zio_nowait(zio_vdev_child_io(zio, bp, vd, offset, pabd,
psize, ZIO_TYPE_READ, ZIO_PRIORITY_SYNC_READ,
ZIO_FLAG_DONT_CACHE | ZIO_FLAG_DONT_QUEUE |
ZIO_FLAG_DONT_PROPAGATE | ZIO_FLAG_DONT_RETRY |
ZIO_FLAG_CANFAIL | ZIO_FLAG_RAW, NULL, NULL));
}
2008-11-20 23:01:55 +03:00
error = zio_wait(zio);
spa_config_exit(spa, SCL_STATE, FTAG);
2008-11-20 23:01:55 +03:00
if (error) {
(void) printf("Read of %s failed, error: %d\n", thing, error);
goto out;
}
if (flags & ZDB_FLAG_DECOMPRESS) {
/*
* We don't know how the data was compressed, so just try
* every decompress function at every inflated blocksize.
*/
enum zio_compress c;
void *pbuf2 = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
void *lbuf2 = umem_alloc(SPA_MAXBLOCKSIZE, UMEM_NOFAIL);
abd_copy_to_buf(pbuf2, pabd, psize);
VERIFY0(abd_iterate_func(pabd, psize, SPA_MAXBLOCKSIZE - psize,
random_get_pseudo_bytes_cb, NULL));
VERIFY0(random_get_pseudo_bytes((uint8_t *)pbuf2 + psize,
SPA_MAXBLOCKSIZE - psize));
/*
* XXX - On the one hand, with SPA_MAXBLOCKSIZE at 16MB,
* this could take a while and we should let the user know
* we are not stuck. On the other hand, printing progress
* info gets old after a while. What to do?
*/
for (lsize = psize + SPA_MINBLOCKSIZE;
lsize <= SPA_MAXBLOCKSIZE; lsize += SPA_MINBLOCKSIZE) {
for (c = 0; c < ZIO_COMPRESS_FUNCTIONS; c++) {
(void) fprintf(stderr,
"Trying %05llx -> %05llx (%s)\n",
(u_longlong_t)psize, (u_longlong_t)lsize,
zio_compress_table[c].ci_name);
if (zio_decompress_data(c, pabd,
lbuf, psize, lsize) == 0 &&
zio_decompress_data_buf(c, pbuf2,
lbuf2, psize, lsize) == 0 &&
bcmp(lbuf, lbuf2, lsize) == 0)
break;
}
if (c != ZIO_COMPRESS_FUNCTIONS)
break;
}
umem_free(pbuf2, SPA_MAXBLOCKSIZE);
umem_free(lbuf2, SPA_MAXBLOCKSIZE);
if (lsize <= psize) {
(void) printf("Decompress of %s failed\n", thing);
goto out;
}
buf = lbuf;
size = lsize;
} else {
size = psize;
buf = abd_borrow_buf_copy(pabd, size);
borrowed = B_TRUE;
}
2008-11-20 23:01:55 +03:00
if (flags & ZDB_FLAG_PRINT_BLKPTR)
zdb_print_blkptr((blkptr_t *)(void *)
((uintptr_t)buf + (uintptr_t)blkptr_offset), flags);
else if (flags & ZDB_FLAG_RAW)
zdb_dump_block_raw(buf, size, flags);
else if (flags & ZDB_FLAG_INDIRECT)
zdb_dump_indirect((blkptr_t *)buf, size / sizeof (blkptr_t),
flags);
else if (flags & ZDB_FLAG_GBH)
zdb_dump_gbh(buf, flags);
else
zdb_dump_block(thing, buf, size, flags);
if (borrowed)
abd_return_buf_copy(pabd, buf, size);
2008-11-20 23:01:55 +03:00
out:
abd_free(pabd);
umem_free(lbuf, SPA_MAXBLOCKSIZE);
2008-11-20 23:01:55 +03:00
free(dup);
}
static boolean_t
pool_match(nvlist_t *cfg, char *tgt)
2008-11-20 23:01:55 +03:00
{
uint64_t v, guid = strtoull(tgt, NULL, 0);
2008-11-20 23:01:55 +03:00
char *s;
if (guid != 0) {
if (nvlist_lookup_uint64(cfg, ZPOOL_CONFIG_POOL_GUID, &v) == 0)
return (v == guid);
2008-11-20 23:01:55 +03:00
} else {
if (nvlist_lookup_string(cfg, ZPOOL_CONFIG_POOL_NAME, &s) == 0)
return (strcmp(s, tgt) == 0);
2008-11-20 23:01:55 +03:00
}
return (B_FALSE);
2008-11-20 23:01:55 +03:00
}
static char *
find_zpool(char **target, nvlist_t **configp, int dirc, char **dirv)
2008-11-20 23:01:55 +03:00
{
nvlist_t *pools;
nvlist_t *match = NULL;
char *name = NULL;
char *sepp = NULL;
char sep = '\0';
int count = 0;
importargs_t args = { 0 };
2008-11-20 23:01:55 +03:00
args.paths = dirc;
args.path = dirv;
args.can_be_active = B_TRUE;
if ((sepp = strpbrk(*target, "/@")) != NULL) {
sep = *sepp;
*sepp = '\0';
}
pools = zpool_search_import(g_zfs, &args);
2008-11-20 23:01:55 +03:00
if (pools != NULL) {
nvpair_t *elem = NULL;
while ((elem = nvlist_next_nvpair(pools, elem)) != NULL) {
verify(nvpair_value_nvlist(elem, configp) == 0);
if (pool_match(*configp, *target)) {
count++;
2008-11-20 23:01:55 +03:00
if (match != NULL) {
/* print previously found config */
if (name != NULL) {
(void) printf("%s\n", name);
dump_nvlist(match, 8);
name = NULL;
}
(void) printf("%s\n",
nvpair_name(elem));
dump_nvlist(*configp, 8);
2008-11-20 23:01:55 +03:00
} else {
match = *configp;
name = nvpair_name(elem);
2008-11-20 23:01:55 +03:00
}
}
}
}
if (count > 1)
(void) fatal("\tMatched %d pools - use pool GUID "
"instead of pool name or \n"
"\tpool name part of a dataset name to select pool", count);
if (sepp)
*sepp = sep;
/*
* If pool GUID was specified for pool id, replace it with pool name
*/
if (name && (strstr(*target, name) != *target)) {
int sz = 1 + strlen(name) + ((sepp) ? strlen(sepp) : 0);
*target = umem_alloc(sz, UMEM_NOFAIL);
(void) snprintf(*target, sz, "%s%s", name, sepp ? sepp : "");
}
2008-11-20 23:01:55 +03:00
*configp = name ? match : NULL;
2008-11-20 23:01:55 +03:00
return (name);
2008-11-20 23:01:55 +03:00
}
int
main(int argc, char **argv)
{
int i, c;
struct rlimit rl = { 1024, 1024 };
spa_t *spa = NULL;
2008-11-20 23:01:55 +03:00
objset_t *os = NULL;
int dump_all = 1;
int verbose = 0;
int error = 0;
char **searchdirs = NULL;
int nsearch = 0;
char *target;
nvlist_t *policy = NULL;
uint64_t max_txg = UINT64_MAX;
int flags = ZFS_IMPORT_MISSING_LOG;
int rewind = ZPOOL_NEVER_REWIND;
char *spa_config_path_env;
boolean_t target_is_spa = B_TRUE;
2008-11-20 23:01:55 +03:00
(void) setrlimit(RLIMIT_NOFILE, &rl);
(void) enable_extended_FILE_stdio(-1, -1);
dprintf_setup(&argc, argv);
/*
* If there is an environment variable SPA_CONFIG_PATH it overrides
* default spa_config_path setting. If -U flag is specified it will
* override this environment variable settings once again.
*/
spa_config_path_env = getenv("SPA_CONFIG_PATH");
if (spa_config_path_env != NULL)
spa_config_path = spa_config_path_env;
while ((c = getopt(argc, argv,
"AbcCdDeFGhiI:lLmMo:Op:PqRsSt:uU:vVx:X")) != -1) {
2008-11-20 23:01:55 +03:00
switch (c) {
case 'b':
case 'c':
case 'C':
case 'd':
case 'D':
case 'G':
case 'h':
case 'i':
case 'l':
2009-07-03 02:44:48 +04:00
case 'm':
Illumos 4976-4984 - metaslab improvements 4976 zfs should only avoid writing to a failing non-redundant top-level vdev 4978 ztest fails in get_metaslab_refcount() 4979 extend free space histogram to device and pool 4980 metaslabs should have a fragmentation metric 4981 remove fragmented ops vector from block allocator 4982 space_map object should proactively upgrade when feature is enabled 4983 need to collect metaslab information via mdb 4984 device selection should use fragmentation metric Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <adam.leventhal@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/4976 https://www.illumos.org/issues/4978 https://www.illumos.org/issues/4979 https://www.illumos.org/issues/4980 https://www.illumos.org/issues/4981 https://www.illumos.org/issues/4982 https://www.illumos.org/issues/4983 https://www.illumos.org/issues/4984 https://github.com/illumos/illumos-gate/commit/2e4c998 Notes: The "zdb -M" option has been re-tasked to display the new metaslab fragmentation metric and the new "zdb -I" option is used to control the maximum number of in-flight I/Os. The new fragmentation metric is derived from the space map histogram which has been rolled up to the vdev and pool level and is presented to the user via "zpool list". Add a number of module parameters related to the new metaslab weighting logic. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2595
2014-07-20 00:19:24 +04:00
case 'M':
case 'O':
2008-11-20 23:01:55 +03:00
case 'R':
case 's':
case 'S':
case 'u':
2008-11-20 23:01:55 +03:00
dump_opt[c]++;
dump_all = 0;
break;
case 'A':
case 'e':
case 'F':
2009-01-16 00:59:39 +03:00
case 'L':
case 'P':
case 'q':
case 'X':
2009-01-16 00:59:39 +03:00
dump_opt[c]++;
break;
/* NB: Sort single match options below. */
Illumos 4976-4984 - metaslab improvements 4976 zfs should only avoid writing to a failing non-redundant top-level vdev 4978 ztest fails in get_metaslab_refcount() 4979 extend free space histogram to device and pool 4980 metaslabs should have a fragmentation metric 4981 remove fragmented ops vector from block allocator 4982 space_map object should proactively upgrade when feature is enabled 4983 need to collect metaslab information via mdb 4984 device selection should use fragmentation metric Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <adam.leventhal@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/4976 https://www.illumos.org/issues/4978 https://www.illumos.org/issues/4979 https://www.illumos.org/issues/4980 https://www.illumos.org/issues/4981 https://www.illumos.org/issues/4982 https://www.illumos.org/issues/4983 https://www.illumos.org/issues/4984 https://github.com/illumos/illumos-gate/commit/2e4c998 Notes: The "zdb -M" option has been re-tasked to display the new metaslab fragmentation metric and the new "zdb -I" option is used to control the maximum number of in-flight I/Os. The new fragmentation metric is derived from the space map histogram which has been rolled up to the vdev and pool level and is presented to the user via "zpool list". Add a number of module parameters related to the new metaslab weighting logic. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2595
2014-07-20 00:19:24 +04:00
case 'I':
Illumos #3306, #3321 3306 zdb should be able to issue reads in parallel 3321 'zpool reopen' command should be documented in the man page and help Reviewed by: Adam Leventhal <ahl@delphix.com> Reviewed by: Matt Ahrens <matthew.ahrens@delphix.com> Reviewed by: Christopher Siden <chris.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: illumos/illumos-gate@31d7e8fa33fae995f558673adb22641b5aa8b6e1 https://www.illumos.org/issues/3306 https://www.illumos.org/issues/3321 The vdev_file.c implementation in this patch diverges significantly from the upstream version. For consistenty with the vdev_disk.c code the upstream version leverages the Illumos bio interfaces. This makes sense for Illumos but not for ZoL for two reasons. 1) The vdev_disk.c code in ZoL has been rewritten to use the Linux block device interfaces which differ significantly from those in Illumos. Therefore, updating the vdev_file.c to use the Illumos interfaces doesn't get you consistency with vdev_disk.c. 2) Using the upstream patch as is would requiring implementing compatibility code for those Solaris block device interfaces in user and kernel space. That additional complexity could lead to confusion and doesn't buy us anything. For these reasons I've opted to simply move the existing vn_rdwr() as is in to the taskq function. This has the advantage of being low risk and easy to understand. Moving the vn_rdwr() function in to its own taskq thread also neatly avoids the possibility of a stack overflow. Finally, because of the additional work which is being handled by the free taskq the number of threads has been increased. The thread count under Illumos defaults to 100 but was decreased to 2 in commit 08d08e due to contention. We increase it to 8 until the contention can be address by porting Illumos #3581. Ported-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #1354
2013-05-03 03:36:32 +04:00
max_inflight = strtoull(optarg, NULL, 0);
if (max_inflight == 0) {
(void) fprintf(stderr, "maximum number "
"of inflight I/Os must be greater "
"than 0\n");
usage();
}
break;
case 'o':
error = set_global_var(optarg);
if (error != 0)
usage();
break;
2008-11-20 23:01:55 +03:00
case 'p':
if (searchdirs == NULL) {
searchdirs = umem_alloc(sizeof (char *),
UMEM_NOFAIL);
} else {
char **tmp = umem_alloc((nsearch + 1) *
sizeof (char *), UMEM_NOFAIL);
bcopy(searchdirs, tmp, nsearch *
sizeof (char *));
umem_free(searchdirs,
nsearch * sizeof (char *));
searchdirs = tmp;
}
searchdirs[nsearch++] = optarg;
2008-11-20 23:01:55 +03:00
break;
2009-01-16 00:59:39 +03:00
case 't':
max_txg = strtoull(optarg, NULL, 0);
if (max_txg < TXG_INITIAL) {
2009-01-16 00:59:39 +03:00
(void) fprintf(stderr, "incorrect txg "
"specified: %s\n", optarg);
usage();
}
break;
case 'U':
spa_config_path = optarg;
break;
Illumos 4976-4984 - metaslab improvements 4976 zfs should only avoid writing to a failing non-redundant top-level vdev 4978 ztest fails in get_metaslab_refcount() 4979 extend free space histogram to device and pool 4980 metaslabs should have a fragmentation metric 4981 remove fragmented ops vector from block allocator 4982 space_map object should proactively upgrade when feature is enabled 4983 need to collect metaslab information via mdb 4984 device selection should use fragmentation metric Reviewed by: Matthew Ahrens <mahrens@delphix.com> Reviewed by: Adam Leventhal <adam.leventhal@delphix.com> Reviewed by: Christopher Siden <christopher.siden@delphix.com> Approved by: Garrett D'Amore <garrett@damore.org> References: https://www.illumos.org/issues/4976 https://www.illumos.org/issues/4978 https://www.illumos.org/issues/4979 https://www.illumos.org/issues/4980 https://www.illumos.org/issues/4981 https://www.illumos.org/issues/4982 https://www.illumos.org/issues/4983 https://www.illumos.org/issues/4984 https://github.com/illumos/illumos-gate/commit/2e4c998 Notes: The "zdb -M" option has been re-tasked to display the new metaslab fragmentation metric and the new "zdb -I" option is used to control the maximum number of in-flight I/Os. The new fragmentation metric is derived from the space map histogram which has been rolled up to the vdev and pool level and is presented to the user via "zpool list". Add a number of module parameters related to the new metaslab weighting logic. Ported by: Tim Chase <tim@chase2k.com> Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov> Closes #2595
2014-07-20 00:19:24 +04:00
case 'v':
verbose++;
break;
case 'V':
flags = ZFS_IMPORT_VERBATIM;
break;
case 'x':
vn_dumpdir = optarg;
break;
2008-11-20 23:01:55 +03:00
default:
usage();
break;
}
}
if (!dump_opt['e'] && searchdirs != NULL) {
(void) fprintf(stderr, "-p option requires use of -e\n");
usage();
}
2008-11-20 23:01:55 +03:00
#if defined(_LP64)
/*
* ZDB does not typically re-read blocks; therefore limit the ARC
* to 256 MB, which can be used entirely for metadata.
*/
zfs_arc_max = zfs_arc_meta_limit = 256 * 1024 * 1024;
#endif
/*
* "zdb -c" uses checksum-verifying scrub i/os which are async reads.
* "zdb -b" uses traversal prefetch which uses async reads.
* For good performance, let several of them be active at once.
*/
zfs_vdev_async_read_max_active = 10;
/*
* Disable reference tracking for better performance.
*/
reference_tracking_enable = B_FALSE;
2008-11-20 23:01:55 +03:00
kernel_init(FREAD);
if ((g_zfs = libzfs_init()) == NULL) {
(void) fprintf(stderr, "%s", libzfs_error_init(errno));
return (1);
}
2008-11-20 23:01:55 +03:00
if (dump_all)
verbose = MAX(verbose, 1);
2008-11-20 23:01:55 +03:00
for (c = 0; c < 256; c++) {
if (dump_all && strchr("AeFlLOPRSX", c) == NULL)
2008-11-20 23:01:55 +03:00
dump_opt[c] = 1;
if (dump_opt[c])
dump_opt[c] += verbose;
}
aok = (dump_opt['A'] == 1) || (dump_opt['A'] > 2);
zfs_recover = (dump_opt['A'] > 1);
2008-11-20 23:01:55 +03:00
argc -= optind;
argv += optind;
if (argc < 2 && dump_opt['R'])
usage();
2008-11-20 23:01:55 +03:00
if (argc < 1) {
if (!dump_opt['e'] && dump_opt['C']) {
dump_cachefile(spa_config_path);
2008-11-20 23:01:55 +03:00
return (0);
}
usage();
}
if (dump_opt['l'])
return (dump_label(argv[0]));
2008-11-20 23:01:55 +03:00
if (dump_opt['O']) {
if (argc != 2)
usage();
dump_opt['v'] = verbose + 3;
return (dump_path(argv[0], argv[1]));
}
if (dump_opt['X'] || dump_opt['F'])
rewind = ZPOOL_DO_REWIND |
(dump_opt['X'] ? ZPOOL_EXTREME_REWIND : 0);
2008-11-20 23:01:55 +03:00
if (nvlist_alloc(&policy, NV_UNIQUE_NAME_TYPE, 0) != 0 ||
nvlist_add_uint64(policy, ZPOOL_REWIND_REQUEST_TXG, max_txg) != 0 ||
nvlist_add_uint32(policy, ZPOOL_REWIND_REQUEST, rewind) != 0)
fatal("internal error: %s", strerror(ENOMEM));
2008-11-20 23:01:55 +03:00
error = 0;
target = argv[0];
2008-11-20 23:01:55 +03:00
if (dump_opt['e']) {
nvlist_t *cfg = NULL;
char *name = find_zpool(&target, &cfg, nsearch, searchdirs);
2008-11-20 23:01:55 +03:00
error = ENOENT;
if (name) {
if (dump_opt['C'] > 1) {
(void) printf("\nConfiguration for import:\n");
dump_nvlist(cfg, 8);
}
if (nvlist_add_nvlist(cfg,
ZPOOL_REWIND_POLICY, policy) != 0) {
fatal("can't open '%s': %s",
target, strerror(ENOMEM));
}
error = spa_import(name, cfg, NULL, flags);
2008-11-20 23:01:55 +03:00
}
}
if (strpbrk(target, "/@") != NULL) {
size_t targetlen;
target_is_spa = B_FALSE;
targetlen = strlen(target);
if (targetlen && target[targetlen - 1] == '/')
target[targetlen - 1] = '\0';
}
if (error == 0) {
if (target_is_spa || dump_opt['R']) {
error = spa_open_rewind(target, &spa, FTAG, policy,
NULL);
if (error) {
/*
* If we're missing the log device then
* try opening the pool after clearing the
* log state.
*/
mutex_enter(&spa_namespace_lock);
if ((spa = spa_lookup(target)) != NULL &&
spa->spa_log_state == SPA_LOG_MISSING) {
spa->spa_log_state = SPA_LOG_CLEAR;
error = 0;
}
mutex_exit(&spa_namespace_lock);
if (!error) {
error = spa_open_rewind(target, &spa,
FTAG, policy, NULL);
}
}
} else {
error = open_objset(target, DMU_OST_ANY, FTAG, &os);
}
2008-11-20 23:01:55 +03:00
}
nvlist_free(policy);
2008-11-20 23:01:55 +03:00
if (error)
fatal("can't open '%s': %s", target, strerror(error));
2008-11-20 23:01:55 +03:00
argv++;
argc--;
if (!dump_opt['R']) {
if (argc > 0) {
zopt_objects = argc;
zopt_object = calloc(zopt_objects, sizeof (uint64_t));
for (i = 0; i < zopt_objects; i++) {
errno = 0;
zopt_object[i] = strtoull(argv[i], NULL, 0);
if (zopt_object[i] == 0 && errno != 0)
fatal("bad number %s: %s",
argv[i], strerror(errno));
}
2008-11-20 23:01:55 +03:00
}
if (os != NULL) {
dump_dir(os);
} else if (zopt_objects > 0 && !dump_opt['m']) {
dump_dir(spa->spa_meta_objset);
} else {
dump_zpool(spa);
}
2008-11-20 23:01:55 +03:00
} else {
flagbits['b'] = ZDB_FLAG_PRINT_BLKPTR;
flagbits['c'] = ZDB_FLAG_CHECKSUM;
flagbits['d'] = ZDB_FLAG_DECOMPRESS;
flagbits['e'] = ZDB_FLAG_BSWAP;
flagbits['g'] = ZDB_FLAG_GBH;
flagbits['i'] = ZDB_FLAG_INDIRECT;
flagbits['p'] = ZDB_FLAG_PHYS;
flagbits['r'] = ZDB_FLAG_RAW;
for (i = 0; i < argc; i++)
zdb_read_block(argv[i], spa);
2008-11-20 23:01:55 +03:00
}
if (os != NULL)
close_objset(os, FTAG);
else
spa_close(spa, FTAG);
2008-11-20 23:01:55 +03:00
fuid_table_destroy();
dump_debug_buffer();
2008-11-20 23:01:55 +03:00
libzfs_fini(g_zfs);
kernel_fini();
return (0);
}