c/mirror_zfs
1
0
Fork 0
mirror of https://git.proxmox.com/git/mirror_zfs.git synced 2024-11-17 18:11:00 +03:00
Code Issues Packages Projects Releases Wiki Activity
e8b96c6007
mirror_zfs/module/zfs/dsl_scan.c
Matthew Ahrens e8b96c6007 Illumos #4045 write throttle & i/o scheduler performance work 
4045 zfs write throttle & i/o scheduler performance work

1. The ZFS i/o scheduler (vdev_queue.c) now divides i/os into 5 classes: sync
read, sync write, async read, async write, and scrub/resilver.  The scheduler
issues a number of concurrent i/os from each class to the device.  Once a class
has been selected, an i/o is selected from this class using either an elevator
algorithem (async, scrub classes) or FIFO (sync classes).  The number of
concurrent async write i/os is tuned dynamically based on i/o load, to achieve
good sync i/o latency when there is not a high load of writes, and good write
throughput when there is.  See the block comment in vdev_queue.c (reproduced
below) for more details.

2. The write throttle (dsl_pool_tempreserve_space() and
txg_constrain_throughput()) is rewritten to produce much more consistent delays
when under constant load.  The new write throttle is based on the amount of
dirty data, rather than guesses about future performance of the system.  When
there is a lot of dirty data, each transaction (e.g. write() syscall) will be
delayed by the same small amount.  This eliminates the "brick wall of wait"
that the old write throttle could hit, causing all transactions to wait several
seconds until the next txg opens.  One of the keys to the new write throttle is
decrementing the amount of dirty data as i/o completes, rather than at the end
of spa_sync().  Note that the write throttle is only applied once the i/o
scheduler is issuing the maximum number of outstanding async writes.  See the
block comments in dsl_pool.c and above dmu_tx_delay() (reproduced below) for
more details.

This diff has several other effects, including:

 * the commonly-tuned global variable zfs_vdev_max_pending has been removed;
use per-class zfs_vdev_*_max_active values or zfs_vdev_max_active instead.

 * the size of each txg (meaning the amount of dirty data written, and thus the
time it takes to write out) is now controlled differently.  There is no longer
an explicit time goal; the primary determinant is amount of dirty data.
Systems that are under light or medium load will now often see that a txg is
always syncing, but the impact to performance (e.g. read latency) is minimal.
Tune zfs_dirty_data_max and zfs_dirty_data_sync to control this.

 * zio_taskq_batch_pct = 75 -- Only use 75% of all CPUs for compression,
checksum, etc.  This improves latency by not allowing these CPU-intensive tasks
to consume all CPU (on machines with at least 4 CPU's; the percentage is
rounded up).

--matt

APPENDIX: problems with the current i/o scheduler

The current ZFS i/o scheduler (vdev_queue.c) is deadline based.  The problem
with this is that if there are always i/os pending, then certain classes of
i/os can see very long delays.

For example, if there are always synchronous reads outstanding, then no async
writes will be serviced until they become "past due".  One symptom of this
situation is that each pass of the txg sync takes at least several seconds
(typically 3 seconds).

If many i/os become "past due" (their deadline is in the past), then we must
service all of these overdue i/os before any new i/os.  This happens when we
enqueue a batch of async writes for the txg sync, with deadlines 2.5 seconds in
the future.  If we can't complete all the i/os in 2.5 seconds (e.g. because
there were always reads pending), then these i/os will become past due.  Now we
must service all the "async" writes (which could be hundreds of megabytes)
before we service any reads, introducing considerable latency to synchronous
i/os (reads or ZIL writes).

Notes on porting to ZFS on Linux:

- zio_t gained new members io_physdone and io_phys_children.  Because
  object caches in the Linux port call the constructor only once at
  allocation time, objects may contain residual data when retrieved
  from the cache. Therefore zio_create() was updated to zero out the two
  new fields.

- vdev_mirror_pending() relied on the depth of the per-vdev pending queue
  (vq->vq_pending_tree) to select the least-busy leaf vdev to read from.
  This tree has been replaced by vq->vq_active_tree which is now used
  for the same purpose.

- vdev_queue_init() used the value of zfs_vdev_max_pending to determine
  the number of vdev I/O buffers to pre-allocate.  That global no longer
  exists, so we instead use the sum of the *_max_active values for each of
  the five I/O classes described above.

- The Illumos implementation of dmu_tx_delay() delays a transaction by
  sleeping in condition variable embedded in the thread
  (curthread->t_delay_cv).  We do not have an equivalent CV to use in
  Linux, so this change replaced the delay logic with a wrapper called
  zfs_sleep_until(). This wrapper could be adopted upstream and in other
  downstream ports to abstract away operating system-specific delay logic.

- These tunables are added as module parameters, and descriptions added
  to the zfs-module-parameters.5 man page.

  spa_asize_inflation
  zfs_deadman_synctime_ms
  zfs_vdev_max_active
  zfs_vdev_async_write_active_min_dirty_percent
  zfs_vdev_async_write_active_max_dirty_percent
  zfs_vdev_async_read_max_active
  zfs_vdev_async_read_min_active
  zfs_vdev_async_write_max_active
  zfs_vdev_async_write_min_active
  zfs_vdev_scrub_max_active
  zfs_vdev_scrub_min_active
  zfs_vdev_sync_read_max_active
  zfs_vdev_sync_read_min_active
  zfs_vdev_sync_write_max_active
  zfs_vdev_sync_write_min_active
  zfs_dirty_data_max_percent
  zfs_delay_min_dirty_percent
  zfs_dirty_data_max_max_percent
  zfs_dirty_data_max
  zfs_dirty_data_max_max
  zfs_dirty_data_sync
  zfs_delay_scale

  The latter four have type unsigned long, whereas they are uint64_t in
  Illumos.  This accommodates Linux's module_param() supported types, but
  means they may overflow on 32-bit architectures.

  The values zfs_dirty_data_max and zfs_dirty_data_max_max are the most
  likely to overflow on 32-bit systems, since they express physical RAM
  sizes in bytes.  In fact, Illumos initializes zfs_dirty_data_max_max to
  2^32 which does overflow. To resolve that, this port instead initializes
  it in arc_init() to 25% of physical RAM, and adds the tunable
  zfs_dirty_data_max_max_percent to override that percentage.  While this
  solution doesn't completely avoid the overflow issue, it should be a
  reasonable default for most systems, and the minority of affected
  systems can work around the issue by overriding the defaults.

- Fixed reversed logic in comment above zfs_delay_scale declaration.

- Clarified comments in vdev_queue.c regarding when per-queue minimums take
  effect.

- Replaced dmu_tx_write_limit in the dmu_tx kstat file
  with dmu_tx_dirty_delay and dmu_tx_dirty_over_max.  The first counts
  how many times a transaction has been delayed because the pool dirty
  data has exceeded zfs_delay_min_dirty_percent.  The latter counts how
  many times the pool dirty data has exceeded zfs_dirty_data_max (which
  we expect to never happen).

- The original patch would have regressed the bug fixed in
  zfsonlinux/zfs@c418410, which prevented users from setting the
  zfs_vdev_aggregation_limit tuning larger than SPA_MAXBLOCKSIZE.
  A similar fix is added to vdev_queue_aggregate().

- In vdev_queue_io_to_issue(), dynamically allocate 'zio_t search' on the
  heap instead of the stack.  In Linux we can't afford such large
  structures on the stack.

Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed by: Adam Leventhal <ahl@delphix.com>
Reviewed by: Christopher Siden <christopher.siden@delphix.com>
Reviewed by: Ned Bass <bass6@llnl.gov>
Reviewed by: Brendan Gregg <brendan.gregg@joyent.com>
Approved by: Robert Mustacchi <rm@joyent.com>

References:
  http://www.illumos.org/issues/4045
  illumos/illumos-gate@69962b5647

Ported-by: Ned Bass <bass6@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #1913
2013-12-06 09:32:43 -08:00

1785 lines
51 KiB
C
Raw Blame History

/*
* 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
*/
/*
* Copyright (c) 2008, 2010, Oracle and/or its affiliates. All rights reserved.
* Copyright (c) 2013 by Delphix. All rights reserved.
*/
#include <sys/dsl_scan.h>
#include <sys/dsl_pool.h>
#include <sys/dsl_dataset.h>
#include <sys/dsl_prop.h>
#include <sys/dsl_dir.h>
#include <sys/dsl_synctask.h>
#include <sys/dnode.h>
#include <sys/dmu_tx.h>
#include <sys/dmu_objset.h>
#include <sys/arc.h>
#include <sys/zap.h>
#include <sys/zio.h>
#include <sys/zfs_context.h>
#include <sys/fs/zfs.h>
#include <sys/zfs_znode.h>
#include <sys/spa_impl.h>
#include <sys/vdev_impl.h>
#include <sys/zil_impl.h>
#include <sys/zio_checksum.h>
#include <sys/ddt.h>
#include <sys/sa.h>
#include <sys/sa_impl.h>
#include <sys/zfeature.h>
#ifdef _KERNEL
#include <sys/zfs_vfsops.h>
#endif
typedef int (scan_cb_t)(dsl_pool_t *, const blkptr_t *, const zbookmark_t *);
static scan_cb_t dsl_scan_scrub_cb;
static void dsl_scan_cancel_sync(void *, dmu_tx_t *);
static void dsl_scan_sync_state(dsl_scan_t *, dmu_tx_t *tx);
int zfs_top_maxinflight = 32; /* maximum I/Os per top-level */
int zfs_resilver_delay = 2; /* number of ticks to delay resilver */
int zfs_scrub_delay = 4; /* number of ticks to delay scrub */
int zfs_scan_idle = 50; /* idle window in clock ticks */
int zfs_scan_min_time_ms = 1000; /* min millisecs to scrub per txg */
int zfs_free_min_time_ms = 1000; /* min millisecs to free per txg */
int zfs_resilver_min_time_ms = 3000; /* min millisecs to resilver per txg */
int zfs_no_scrub_io = B_FALSE; /* set to disable scrub i/o */
int zfs_no_scrub_prefetch = B_FALSE; /* set to disable srub prefetching */
enum ddt_class zfs_scrub_ddt_class_max = DDT_CLASS_DUPLICATE;
int dsl_scan_delay_completion = B_FALSE; /* set to delay scan completion */
#define DSL_SCAN_IS_SCRUB_RESILVER(scn) \
((scn)->scn_phys.scn_func == POOL_SCAN_SCRUB || \
(scn)->scn_phys.scn_func == POOL_SCAN_RESILVER)
/* the order has to match pool_scan_type */
static scan_cb_t *scan_funcs[POOL_SCAN_FUNCS] = {
NULL,
dsl_scan_scrub_cb, /* POOL_SCAN_SCRUB */
dsl_scan_scrub_cb, /* POOL_SCAN_RESILVER */
};
int
dsl_scan_init(dsl_pool_t *dp, uint64_t txg)
{
int err;
dsl_scan_t *scn;
spa_t *spa = dp->dp_spa;
uint64_t f;
scn = dp->dp_scan = kmem_zalloc(sizeof (dsl_scan_t), KM_SLEEP);
scn->scn_dp = dp;
/*
* It's possible that we're resuming a scan after a reboot so
* make sure that the scan_async_destroying flag is initialized
* appropriately.
*/
ASSERT(!scn->scn_async_destroying);
scn->scn_async_destroying = spa_feature_is_active(dp->dp_spa,
&spa_feature_table[SPA_FEATURE_ASYNC_DESTROY]);
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
"scrub_func", sizeof (uint64_t), 1, &f);
if (err == 0) {
/*
* There was an old-style scrub in progress. Restart a
* new-style scrub from the beginning.
*/
scn->scn_restart_txg = txg;
zfs_dbgmsg("old-style scrub was in progress; "
"restarting new-style scrub in txg %llu",
scn->scn_restart_txg);
/*
* Load the queue obj from the old location so that it
* can be freed by dsl_scan_done().
*/
(void) zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
"scrub_queue", sizeof (uint64_t), 1,
&scn->scn_phys.scn_queue_obj);
} else {
err = zap_lookup(dp->dp_meta_objset, DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
&scn->scn_phys);
if (err == ENOENT)
return (0);
else if (err)
return (err);
if (scn->scn_phys.scn_state == DSS_SCANNING &&
spa_prev_software_version(dp->dp_spa) < SPA_VERSION_SCAN) {
/*
* A new-type scrub was in progress on an old
* pool, and the pool was accessed by old
* software. Restart from the beginning, since
* the old software may have changed the pool in
* the meantime.
*/
scn->scn_restart_txg = txg;
zfs_dbgmsg("new-style scrub was modified "
"by old software; restarting in txg %llu",
scn->scn_restart_txg);
}
}
spa_scan_stat_init(spa);
return (0);
}
void
dsl_scan_fini(dsl_pool_t *dp)
{
if (dp->dp_scan) {
kmem_free(dp->dp_scan, sizeof (dsl_scan_t));
dp->dp_scan = NULL;
}
}
/* ARGSUSED */
static int
dsl_scan_setup_check(void *arg, dmu_tx_t *tx)
{
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
if (scn->scn_phys.scn_state == DSS_SCANNING)
return (SET_ERROR(EBUSY));
return (0);
}
static void
dsl_scan_setup_sync(void *arg, dmu_tx_t *tx)
{
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
pool_scan_func_t *funcp = arg;
dmu_object_type_t ot = 0;
dsl_pool_t *dp = scn->scn_dp;
spa_t *spa = dp->dp_spa;
ASSERT(scn->scn_phys.scn_state != DSS_SCANNING);
ASSERT(*funcp > POOL_SCAN_NONE && *funcp < POOL_SCAN_FUNCS);
bzero(&scn->scn_phys, sizeof (scn->scn_phys));
scn->scn_phys.scn_func = *funcp;
scn->scn_phys.scn_state = DSS_SCANNING;
scn->scn_phys.scn_min_txg = 0;
scn->scn_phys.scn_max_txg = tx->tx_txg;
scn->scn_phys.scn_ddt_class_max = DDT_CLASSES - 1; /* the entire DDT */
scn->scn_phys.scn_start_time = gethrestime_sec();
scn->scn_phys.scn_errors = 0;
scn->scn_phys.scn_to_examine = spa->spa_root_vdev->vdev_stat.vs_alloc;
scn->scn_restart_txg = 0;
scn->scn_done_txg = 0;
spa_scan_stat_init(spa);
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
scn->scn_phys.scn_ddt_class_max = zfs_scrub_ddt_class_max;
/* rewrite all disk labels */
vdev_config_dirty(spa->spa_root_vdev);
if (vdev_resilver_needed(spa->spa_root_vdev,
&scn->scn_phys.scn_min_txg, &scn->scn_phys.scn_max_txg)) {
spa_event_notify(spa, NULL, FM_EREPORT_ZFS_RESILVER_START);
} else {
spa_event_notify(spa, NULL, FM_EREPORT_ZFS_SCRUB_START);
}
spa->spa_scrub_started = B_TRUE;
/*
* If this is an incremental scrub, limit the DDT scrub phase
* to just the auto-ditto class (for correctness); the rest
* of the scrub should go faster using top-down pruning.
*/
if (scn->scn_phys.scn_min_txg > TXG_INITIAL)
scn->scn_phys.scn_ddt_class_max = DDT_CLASS_DITTO;
}
/* back to the generic stuff */
if (dp->dp_blkstats == NULL) {
dp->dp_blkstats = kmem_alloc(sizeof (zfs_all_blkstats_t),
KM_PUSHPAGE | KM_NODEBUG);
}
bzero(dp->dp_blkstats, sizeof (zfs_all_blkstats_t));
if (spa_version(spa) < SPA_VERSION_DSL_SCRUB)
ot = DMU_OT_ZAP_OTHER;
scn->scn_phys.scn_queue_obj = zap_create(dp->dp_meta_objset,
ot ? ot : DMU_OT_SCAN_QUEUE, DMU_OT_NONE, 0, tx);
dsl_scan_sync_state(scn, tx);
spa_history_log_internal(spa, "scan setup", tx,
"func=%u mintxg=%llu maxtxg=%llu",
*funcp, scn->scn_phys.scn_min_txg, scn->scn_phys.scn_max_txg);
}
/* ARGSUSED */
static void
dsl_scan_done(dsl_scan_t *scn, boolean_t complete, dmu_tx_t *tx)
{
static const char *old_names[] = {
"scrub_bookmark",
"scrub_ddt_bookmark",
"scrub_ddt_class_max",
"scrub_queue",
"scrub_min_txg",
"scrub_max_txg",
"scrub_func",
"scrub_errors",
NULL
};
dsl_pool_t *dp = scn->scn_dp;
spa_t *spa = dp->dp_spa;
int i;
/* Remove any remnants of an old-style scrub. */
for (i = 0; old_names[i]; i++) {
(void) zap_remove(dp->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT, old_names[i], tx);
}
if (scn->scn_phys.scn_queue_obj != 0) {
VERIFY(0 == dmu_object_free(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, tx));
scn->scn_phys.scn_queue_obj = 0;
}
/*
* If we were "restarted" from a stopped state, don't bother
* with anything else.
*/
if (scn->scn_phys.scn_state != DSS_SCANNING)
return;
if (complete)
scn->scn_phys.scn_state = DSS_FINISHED;
else
scn->scn_phys.scn_state = DSS_CANCELED;
spa_history_log_internal(spa, "scan done", tx,
"complete=%u", complete);
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
mutex_enter(&spa->spa_scrub_lock);
while (spa->spa_scrub_inflight > 0) {
cv_wait(&spa->spa_scrub_io_cv,
&spa->spa_scrub_lock);
}
mutex_exit(&spa->spa_scrub_lock);
spa->spa_scrub_started = B_FALSE;
spa->spa_scrub_active = B_FALSE;
/*
* If the scrub/resilver completed, update all DTLs to
* reflect this. Whether it succeeded or not, vacate
* all temporary scrub DTLs.
*/
vdev_dtl_reassess(spa->spa_root_vdev, tx->tx_txg,
complete ? scn->scn_phys.scn_max_txg : 0, B_TRUE);
if (complete) {
spa_event_notify(spa, NULL, scn->scn_phys.scn_min_txg ?
FM_EREPORT_ZFS_RESILVER_FINISH :
FM_EREPORT_ZFS_SCRUB_FINISH);
}
spa_errlog_rotate(spa);
/*
* We may have finished replacing a device.
* Let the async thread assess this and handle the detach.
*/
spa_async_request(spa, SPA_ASYNC_RESILVER_DONE);
}
scn->scn_phys.scn_end_time = gethrestime_sec();
}
/* ARGSUSED */
static int
dsl_scan_cancel_check(void *arg, dmu_tx_t *tx)
{
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
if (scn->scn_phys.scn_state != DSS_SCANNING)
return (SET_ERROR(ENOENT));
return (0);
}
/* ARGSUSED */
static void
dsl_scan_cancel_sync(void *arg, dmu_tx_t *tx)
{
dsl_scan_t *scn = dmu_tx_pool(tx)->dp_scan;
dsl_scan_done(scn, B_FALSE, tx);
dsl_scan_sync_state(scn, tx);
}
int
dsl_scan_cancel(dsl_pool_t *dp)
{
return (dsl_sync_task(spa_name(dp->dp_spa), dsl_scan_cancel_check,
dsl_scan_cancel_sync, NULL, 3));
}
static void dsl_scan_visitbp(blkptr_t *bp,
const zbookmark_t *zb, dnode_phys_t *dnp, arc_buf_t *pbuf,
dsl_dataset_t *ds, dsl_scan_t *scn, dmu_objset_type_t ostype,
dmu_tx_t *tx);
inline __attribute__((always_inline)) static void dsl_scan_visitdnode(
dsl_scan_t *, dsl_dataset_t *ds, dmu_objset_type_t ostype,
dnode_phys_t *dnp, arc_buf_t *buf, uint64_t object, dmu_tx_t *tx);
void
dsl_free(dsl_pool_t *dp, uint64_t txg, const blkptr_t *bp)
{
zio_free(dp->dp_spa, txg, bp);
}
void
dsl_free_sync(zio_t *pio, dsl_pool_t *dp, uint64_t txg, const blkptr_t *bpp)
{
ASSERT(dsl_pool_sync_context(dp));
zio_nowait(zio_free_sync(pio, dp->dp_spa, txg, bpp, pio->io_flags));
}
static uint64_t
dsl_scan_ds_maxtxg(dsl_dataset_t *ds)
{
uint64_t smt = ds->ds_dir->dd_pool->dp_scan->scn_phys.scn_max_txg;
if (dsl_dataset_is_snapshot(ds))
return (MIN(smt, ds->ds_phys->ds_creation_txg));
return (smt);
}
static void
dsl_scan_sync_state(dsl_scan_t *scn, dmu_tx_t *tx)
{
VERIFY0(zap_update(scn->scn_dp->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_SCAN, sizeof (uint64_t), SCAN_PHYS_NUMINTS,
&scn->scn_phys, tx));
}
static boolean_t
dsl_scan_check_pause(dsl_scan_t *scn, const zbookmark_t *zb)
{
uint64_t elapsed_nanosecs;
int mintime;
/* we never skip user/group accounting objects */
if (zb && (int64_t)zb->zb_object < 0)
return (B_FALSE);
if (scn->scn_pausing)
return (B_TRUE); /* we're already pausing */
if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark))
return (B_FALSE); /* we're resuming */
/* We only know how to resume from level-0 blocks. */
if (zb && zb->zb_level != 0)
return (B_FALSE);
mintime = (scn->scn_phys.scn_func == POOL_SCAN_RESILVER) ?
zfs_resilver_min_time_ms : zfs_scan_min_time_ms;
elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
if (elapsed_nanosecs / NANOSEC > zfs_txg_timeout ||
(NSEC2MSEC(elapsed_nanosecs) > mintime &&
txg_sync_waiting(scn->scn_dp)) ||
spa_shutting_down(scn->scn_dp->dp_spa)) {
if (zb) {
dprintf("pausing at bookmark %llx/%llx/%llx/%llx\n",
(longlong_t)zb->zb_objset,
(longlong_t)zb->zb_object,
(longlong_t)zb->zb_level,
(longlong_t)zb->zb_blkid);
scn->scn_phys.scn_bookmark = *zb;
}
dprintf("pausing at DDT bookmark %llx/%llx/%llx/%llx\n",
(longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class,
(longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type,
(longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum,
(longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor);
scn->scn_pausing = B_TRUE;
return (B_TRUE);
}
return (B_FALSE);
}
typedef struct zil_scan_arg {
dsl_pool_t *zsa_dp;
zil_header_t *zsa_zh;
} zil_scan_arg_t;
/* ARGSUSED */
static int
dsl_scan_zil_block(zilog_t *zilog, blkptr_t *bp, void *arg, uint64_t claim_txg)
{
zil_scan_arg_t *zsa = arg;
dsl_pool_t *dp = zsa->zsa_dp;
dsl_scan_t *scn = dp->dp_scan;
zil_header_t *zh = zsa->zsa_zh;
zbookmark_t zb;
if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
return (0);
/*
* One block ("stubby") can be allocated a long time ago; we
* want to visit that one because it has been allocated
* (on-disk) even if it hasn't been claimed (even though for
* scrub there's nothing to do to it).
*/
if (claim_txg == 0 && bp->blk_birth >= spa_first_txg(dp->dp_spa))
return (0);
SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET],
ZB_ZIL_OBJECT, ZB_ZIL_LEVEL, bp->blk_cksum.zc_word[ZIL_ZC_SEQ]);
VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb));
return (0);
}
/* ARGSUSED */
static int
dsl_scan_zil_record(zilog_t *zilog, lr_t *lrc, void *arg, uint64_t claim_txg)
{
if (lrc->lrc_txtype == TX_WRITE) {
zil_scan_arg_t *zsa = arg;
dsl_pool_t *dp = zsa->zsa_dp;
dsl_scan_t *scn = dp->dp_scan;
zil_header_t *zh = zsa->zsa_zh;
lr_write_t *lr = (lr_write_t *)lrc;
blkptr_t *bp = &lr->lr_blkptr;
zbookmark_t zb;
if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
return (0);
/*
* birth can be < claim_txg if this record's txg is
* already txg sync'ed (but this log block contains
* other records that are not synced)
*/
if (claim_txg == 0 || bp->blk_birth < claim_txg)
return (0);
SET_BOOKMARK(&zb, zh->zh_log.blk_cksum.zc_word[ZIL_ZC_OBJSET],
lr->lr_foid, ZB_ZIL_LEVEL,
lr->lr_offset / BP_GET_LSIZE(bp));
VERIFY(0 == scan_funcs[scn->scn_phys.scn_func](dp, bp, &zb));
}
return (0);
}
static void
dsl_scan_zil(dsl_pool_t *dp, zil_header_t *zh)
{
uint64_t claim_txg = zh->zh_claim_txg;
zil_scan_arg_t zsa = { dp, zh };
zilog_t *zilog;
/*
* We only want to visit blocks that have been claimed but not yet
* replayed (or, in read-only mode, blocks that *would* be claimed).
*/
if (claim_txg == 0 && spa_writeable(dp->dp_spa))
return;
zilog = zil_alloc(dp->dp_meta_objset, zh);
(void) zil_parse(zilog, dsl_scan_zil_block, dsl_scan_zil_record, &zsa,
claim_txg);
zil_free(zilog);
}
/* ARGSUSED */
static void
dsl_scan_prefetch(dsl_scan_t *scn, arc_buf_t *buf, blkptr_t *bp,
uint64_t objset, uint64_t object, uint64_t blkid)
{
zbookmark_t czb;
uint32_t flags = ARC_NOWAIT | ARC_PREFETCH;
if (zfs_no_scrub_prefetch)
return;
if (BP_IS_HOLE(bp) || bp->blk_birth <= scn->scn_phys.scn_min_txg ||
(BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_DNODE))
return;
SET_BOOKMARK(&czb, objset, object, BP_GET_LEVEL(bp), blkid);
(void) arc_read(scn->scn_zio_root, scn->scn_dp->dp_spa, bp,
NULL, NULL, ZIO_PRIORITY_ASYNC_READ,
ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD, &flags, &czb);
}
static boolean_t
dsl_scan_check_resume(dsl_scan_t *scn, const dnode_phys_t *dnp,
const zbookmark_t *zb)
{
/*
* We never skip over user/group accounting objects (obj<0)
*/
if (!ZB_IS_ZERO(&scn->scn_phys.scn_bookmark) &&
(int64_t)zb->zb_object >= 0) {
/*
* If we already visited this bp & everything below (in
* a prior txg sync), don't bother doing it again.
*/
if (zbookmark_is_before(dnp, zb, &scn->scn_phys.scn_bookmark))
return (B_TRUE);
/*
* If we found the block we're trying to resume from, or
* we went past it to a different object, zero it out to
* indicate that it's OK to start checking for pausing
* again.
*/
if (bcmp(zb, &scn->scn_phys.scn_bookmark, sizeof (*zb)) == 0 ||
zb->zb_object > scn->scn_phys.scn_bookmark.zb_object) {
dprintf("resuming at %llx/%llx/%llx/%llx\n",
(longlong_t)zb->zb_objset,
(longlong_t)zb->zb_object,
(longlong_t)zb->zb_level,
(longlong_t)zb->zb_blkid);
bzero(&scn->scn_phys.scn_bookmark, sizeof (*zb));
}
}
return (B_FALSE);
}
/*
* Return nonzero on i/o error.
* Return new buf to write out in *bufp.
*/
inline __attribute__((always_inline)) static int
dsl_scan_recurse(dsl_scan_t *scn, dsl_dataset_t *ds, dmu_objset_type_t ostype,
dnode_phys_t *dnp, const blkptr_t *bp,
const zbookmark_t *zb, dmu_tx_t *tx, arc_buf_t **bufp)
{
dsl_pool_t *dp = scn->scn_dp;
int zio_flags = ZIO_FLAG_CANFAIL | ZIO_FLAG_SCAN_THREAD;
int err;
if (BP_GET_LEVEL(bp) > 0) {
uint32_t flags = ARC_WAIT;
int i;
blkptr_t *cbp;
int epb = BP_GET_LSIZE(bp) >> SPA_BLKPTRSHIFT;
err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp,
ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
return (err);
}
for (i = 0, cbp = (*bufp)->b_data; i < epb; i++, cbp++) {
dsl_scan_prefetch(scn, *bufp, cbp, zb->zb_objset,
zb->zb_object, zb->zb_blkid * epb + i);
}
for (i = 0, cbp = (*bufp)->b_data; i < epb; i++, cbp++) {
zbookmark_t czb;
SET_BOOKMARK(&czb, zb->zb_objset, zb->zb_object,
zb->zb_level - 1,
zb->zb_blkid * epb + i);
dsl_scan_visitbp(cbp, &czb, dnp,
*bufp, ds, scn, ostype, tx);
}
} else if (BP_GET_TYPE(bp) == DMU_OT_USERGROUP_USED) {
uint32_t flags = ARC_WAIT;
err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp,
ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
return (err);
}
} else if (BP_GET_TYPE(bp) == DMU_OT_DNODE) {
uint32_t flags = ARC_WAIT;
dnode_phys_t *cdnp;
int i, j;
int epb = BP_GET_LSIZE(bp) >> DNODE_SHIFT;
err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp,
ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
return (err);
}
for (i = 0, cdnp = (*bufp)->b_data; i < epb; i++, cdnp++) {
for (j = 0; j < cdnp->dn_nblkptr; j++) {
blkptr_t *cbp = &cdnp->dn_blkptr[j];
dsl_scan_prefetch(scn, *bufp, cbp,
zb->zb_objset, zb->zb_blkid * epb + i, j);
}
}
for (i = 0, cdnp = (*bufp)->b_data; i < epb; i++, cdnp++) {
dsl_scan_visitdnode(scn, ds, ostype,
cdnp, *bufp, zb->zb_blkid * epb + i, tx);
}
} else if (BP_GET_TYPE(bp) == DMU_OT_OBJSET) {
uint32_t flags = ARC_WAIT;
objset_phys_t *osp;
err = arc_read(NULL, dp->dp_spa, bp, arc_getbuf_func, bufp,
ZIO_PRIORITY_ASYNC_READ, zio_flags, &flags, zb);
if (err) {
scn->scn_phys.scn_errors++;
return (err);
}
osp = (*bufp)->b_data;
dsl_scan_visitdnode(scn, ds, osp->os_type,
&osp->os_meta_dnode, *bufp, DMU_META_DNODE_OBJECT, tx);
if (OBJSET_BUF_HAS_USERUSED(*bufp)) {
/*
* We also always visit user/group accounting
* objects, and never skip them, even if we are
* pausing. This is necessary so that the space
* deltas from this txg get integrated.
*/
dsl_scan_visitdnode(scn, ds, osp->os_type,
&osp->os_groupused_dnode, *bufp,
DMU_GROUPUSED_OBJECT, tx);
dsl_scan_visitdnode(scn, ds, osp->os_type,
&osp->os_userused_dnode, *bufp,
DMU_USERUSED_OBJECT, tx);
}
}
return (0);
}
inline __attribute__((always_inline)) static void
dsl_scan_visitdnode(dsl_scan_t *scn, dsl_dataset_t *ds,
dmu_objset_type_t ostype, dnode_phys_t *dnp, arc_buf_t *buf,
uint64_t object, dmu_tx_t *tx)
{
int j;
for (j = 0; j < dnp->dn_nblkptr; j++) {
zbookmark_t czb;
SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object,
dnp->dn_nlevels - 1, j);
dsl_scan_visitbp(&dnp->dn_blkptr[j],
&czb, dnp, buf, ds, scn, ostype, tx);
}
if (dnp->dn_flags & DNODE_FLAG_SPILL_BLKPTR) {
zbookmark_t czb;
SET_BOOKMARK(&czb, ds ? ds->ds_object : 0, object,
0, DMU_SPILL_BLKID);
dsl_scan_visitbp(&dnp->dn_spill,
&czb, dnp, buf, ds, scn, ostype, tx);
}
}
/*
* The arguments are in this order because mdb can only print the
* first 5; we want them to be useful.
*/
static void
dsl_scan_visitbp(blkptr_t *bp, const zbookmark_t *zb,
dnode_phys_t *dnp, arc_buf_t *pbuf,
dsl_dataset_t *ds, dsl_scan_t *scn, dmu_objset_type_t ostype,
dmu_tx_t *tx)
{
dsl_pool_t *dp = scn->scn_dp;
arc_buf_t *buf = NULL;
blkptr_t *bp_toread;
bp_toread = kmem_alloc(sizeof (blkptr_t), KM_PUSHPAGE);
*bp_toread = *bp;
/* ASSERT(pbuf == NULL || arc_released(pbuf)); */
if (dsl_scan_check_pause(scn, zb))
goto out;
if (dsl_scan_check_resume(scn, dnp, zb))
goto out;
if (bp->blk_birth == 0)
goto out;
scn->scn_visited_this_txg++;
/*
* This debugging is commented out to conserve stack space. This
* function is called recursively and the debugging addes several
* bytes to the stack for each call. It can be commented back in
* if required to debug an issue in dsl_scan_visitbp().
*
* dprintf_bp(bp,
* "visiting ds=%p/%llu zb=%llx/%llx/%llx/%llx buf=%p bp=%p",
* ds, ds ? ds->ds_object : 0,
* zb->zb_objset, zb->zb_object, zb->zb_level, zb->zb_blkid,
* pbuf, bp);
*/
if (bp->blk_birth <= scn->scn_phys.scn_cur_min_txg)
goto out;
if (dsl_scan_recurse(scn, ds, ostype, dnp, bp_toread, zb, tx,
&buf) != 0)
goto out;
/*
* If dsl_scan_ddt() has aready visited this block, it will have
* already done any translations or scrubbing, so don't call the
* callback again.
*/
if (ddt_class_contains(dp->dp_spa,
scn->scn_phys.scn_ddt_class_max, bp)) {
ASSERT(buf == NULL);
goto out;
}
/*
* If this block is from the future (after cur_max_txg), then we
* are doing this on behalf of a deleted snapshot, and we will
* revisit the future block on the next pass of this dataset.
* Don't scan it now unless we need to because something
* under it was modified.
*/
if (BP_PHYSICAL_BIRTH(bp) <= scn->scn_phys.scn_cur_max_txg) {
scan_funcs[scn->scn_phys.scn_func](dp, bp, zb);
}
if (buf)
(void) arc_buf_remove_ref(buf, &buf);
out:
kmem_free(bp_toread, sizeof(blkptr_t));
}
static void
dsl_scan_visit_rootbp(dsl_scan_t *scn, dsl_dataset_t *ds, blkptr_t *bp,
dmu_tx_t *tx)
{
zbookmark_t zb;
SET_BOOKMARK(&zb, ds ? ds->ds_object : DMU_META_OBJSET,
ZB_ROOT_OBJECT, ZB_ROOT_LEVEL, ZB_ROOT_BLKID);
dsl_scan_visitbp(bp, &zb, NULL, NULL,
ds, scn, DMU_OST_NONE, tx);
dprintf_ds(ds, "finished scan%s", "");
}
void
dsl_scan_ds_destroyed(dsl_dataset_t *ds, dmu_tx_t *tx)
{
dsl_pool_t *dp = ds->ds_dir->dd_pool;
dsl_scan_t *scn = dp->dp_scan;
uint64_t mintxg;
if (scn->scn_phys.scn_state != DSS_SCANNING)
return;
if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) {
if (dsl_dataset_is_snapshot(ds)) {
/* Note, scn_cur_{min,max}_txg stays the same. */
scn->scn_phys.scn_bookmark.zb_objset =
ds->ds_phys->ds_next_snap_obj;
zfs_dbgmsg("destroying ds %llu; currently traversing; "
"reset zb_objset to %llu",
(u_longlong_t)ds->ds_object,
(u_longlong_t)ds->ds_phys->ds_next_snap_obj);
scn->scn_phys.scn_flags |= DSF_VISIT_DS_AGAIN;
} else {
SET_BOOKMARK(&scn->scn_phys.scn_bookmark,
ZB_DESTROYED_OBJSET, 0, 0, 0);
zfs_dbgmsg("destroying ds %llu; currently traversing; "
"reset bookmark to -1,0,0,0",
(u_longlong_t)ds->ds_object);
}
} else if (zap_lookup_int_key(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) {
ASSERT3U(ds->ds_phys->ds_num_children, <=, 1);
VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds->ds_object, tx));
if (dsl_dataset_is_snapshot(ds)) {
/*
* We keep the same mintxg; it could be >
* ds_creation_txg if the previous snapshot was
* deleted too.
*/
VERIFY(zap_add_int_key(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj,
ds->ds_phys->ds_next_snap_obj, mintxg, tx) == 0);
zfs_dbgmsg("destroying ds %llu; in queue; "
"replacing with %llu",
(u_longlong_t)ds->ds_object,
(u_longlong_t)ds->ds_phys->ds_next_snap_obj);
} else {
zfs_dbgmsg("destroying ds %llu; in queue; removing",
(u_longlong_t)ds->ds_object);
}
} else {
zfs_dbgmsg("destroying ds %llu; ignoring",
(u_longlong_t)ds->ds_object);
}
/*
* dsl_scan_sync() should be called after this, and should sync
* out our changed state, but just to be safe, do it here.
*/
dsl_scan_sync_state(scn, tx);
}
void
dsl_scan_ds_snapshotted(dsl_dataset_t *ds, dmu_tx_t *tx)
{
dsl_pool_t *dp = ds->ds_dir->dd_pool;
dsl_scan_t *scn = dp->dp_scan;
uint64_t mintxg;
if (scn->scn_phys.scn_state != DSS_SCANNING)
return;
ASSERT(ds->ds_phys->ds_prev_snap_obj != 0);
if (scn->scn_phys.scn_bookmark.zb_objset == ds->ds_object) {
scn->scn_phys.scn_bookmark.zb_objset =
ds->ds_phys->ds_prev_snap_obj;
zfs_dbgmsg("snapshotting ds %llu; currently traversing; "
"reset zb_objset to %llu",
(u_longlong_t)ds->ds_object,
(u_longlong_t)ds->ds_phys->ds_prev_snap_obj);
} else if (zap_lookup_int_key(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds->ds_object, &mintxg) == 0) {
VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds->ds_object, tx));
VERIFY(zap_add_int_key(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj,
ds->ds_phys->ds_prev_snap_obj, mintxg, tx) == 0);
zfs_dbgmsg("snapshotting ds %llu; in queue; "
"replacing with %llu",
(u_longlong_t)ds->ds_object,
(u_longlong_t)ds->ds_phys->ds_prev_snap_obj);
}
dsl_scan_sync_state(scn, tx);
}
void
dsl_scan_ds_clone_swapped(dsl_dataset_t *ds1, dsl_dataset_t *ds2, dmu_tx_t *tx)
{
dsl_pool_t *dp = ds1->ds_dir->dd_pool;
dsl_scan_t *scn = dp->dp_scan;
uint64_t mintxg;
if (scn->scn_phys.scn_state != DSS_SCANNING)
return;
if (scn->scn_phys.scn_bookmark.zb_objset == ds1->ds_object) {
scn->scn_phys.scn_bookmark.zb_objset = ds2->ds_object;
zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
"reset zb_objset to %llu",
(u_longlong_t)ds1->ds_object,
(u_longlong_t)ds2->ds_object);
} else if (scn->scn_phys.scn_bookmark.zb_objset == ds2->ds_object) {
scn->scn_phys.scn_bookmark.zb_objset = ds1->ds_object;
zfs_dbgmsg("clone_swap ds %llu; currently traversing; "
"reset zb_objset to %llu",
(u_longlong_t)ds2->ds_object,
(u_longlong_t)ds1->ds_object);
}
if (zap_lookup_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj,
ds1->ds_object, &mintxg) == 0) {
int err;
ASSERT3U(mintxg, ==, ds1->ds_phys->ds_prev_snap_txg);
ASSERT3U(mintxg, ==, ds2->ds_phys->ds_prev_snap_txg);
VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds1->ds_object, tx));
err = zap_add_int_key(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds2->ds_object, mintxg, tx);
VERIFY(err == 0 || err == EEXIST);
if (err == EEXIST) {
/* Both were there to begin with */
VERIFY(0 == zap_add_int_key(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj,
ds1->ds_object, mintxg, tx));
}
zfs_dbgmsg("clone_swap ds %llu; in queue; "
"replacing with %llu",
(u_longlong_t)ds1->ds_object,
(u_longlong_t)ds2->ds_object);
} else if (zap_lookup_int_key(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds2->ds_object, &mintxg) == 0) {
ASSERT3U(mintxg, ==, ds1->ds_phys->ds_prev_snap_txg);
ASSERT3U(mintxg, ==, ds2->ds_phys->ds_prev_snap_txg);
VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds2->ds_object, tx));
VERIFY(0 == zap_add_int_key(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds1->ds_object, mintxg, tx));
zfs_dbgmsg("clone_swap ds %llu; in queue; "
"replacing with %llu",
(u_longlong_t)ds2->ds_object,
(u_longlong_t)ds1->ds_object);
}
dsl_scan_sync_state(scn, tx);
}
struct enqueue_clones_arg {
dmu_tx_t *tx;
uint64_t originobj;
};
/* ARGSUSED */
static int
enqueue_clones_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
{
struct enqueue_clones_arg *eca = arg;
dsl_dataset_t *ds;
int err;
dsl_scan_t *scn = dp->dp_scan;
if (hds->ds_dir->dd_phys->dd_origin_obj != eca->originobj)
return (0);
err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
if (err)
return (err);
while (ds->ds_phys->ds_prev_snap_obj != eca->originobj) {
dsl_dataset_t *prev;
err = dsl_dataset_hold_obj(dp,
ds->ds_phys->ds_prev_snap_obj, FTAG, &prev);
dsl_dataset_rele(ds, FTAG);
if (err)
return (err);
ds = prev;
}
VERIFY(zap_add_int_key(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds->ds_object,
ds->ds_phys->ds_prev_snap_txg, eca->tx) == 0);
dsl_dataset_rele(ds, FTAG);
return (0);
}
static void
dsl_scan_visitds(dsl_scan_t *scn, uint64_t dsobj, dmu_tx_t *tx)
{
dsl_pool_t *dp = scn->scn_dp;
dsl_dataset_t *ds;
objset_t *os;
char *dsname;
VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
if (dmu_objset_from_ds(ds, &os))
goto out;
/*
* Only the ZIL in the head (non-snapshot) is valid. Even though
* snapshots can have ZIL block pointers (which may be the same
* BP as in the head), they must be ignored. So we traverse the
* ZIL here, rather than in scan_recurse(), because the regular
* snapshot block-sharing rules don't apply to it.
*/
if (DSL_SCAN_IS_SCRUB_RESILVER(scn) && !dsl_dataset_is_snapshot(ds))
dsl_scan_zil(dp, &os->os_zil_header);
/*
* Iterate over the bps in this ds.
*/
dmu_buf_will_dirty(ds->ds_dbuf, tx);
dsl_scan_visit_rootbp(scn, ds, &ds->ds_phys->ds_bp, tx);
dsname = kmem_alloc(ZFS_MAXNAMELEN, KM_PUSHPAGE);
dsl_dataset_name(ds, dsname);
zfs_dbgmsg("scanned dataset %llu (%s) with min=%llu max=%llu; "
"pausing=%u",
(longlong_t)dsobj, dsname,
(longlong_t)scn->scn_phys.scn_cur_min_txg,
(longlong_t)scn->scn_phys.scn_cur_max_txg,
(int)scn->scn_pausing);
kmem_free(dsname, ZFS_MAXNAMELEN);
if (scn->scn_pausing)
goto out;
/*
* We've finished this pass over this dataset.
*/
/*
* If we did not completely visit this dataset, do another pass.
*/
if (scn->scn_phys.scn_flags & DSF_VISIT_DS_AGAIN) {
zfs_dbgmsg("incomplete pass; visiting again");
scn->scn_phys.scn_flags &= ~DSF_VISIT_DS_AGAIN;
VERIFY(zap_add_int_key(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds->ds_object,
scn->scn_phys.scn_cur_max_txg, tx) == 0);
goto out;
}
/*
* Add descendent datasets to work queue.
*/
if (ds->ds_phys->ds_next_snap_obj != 0) {
VERIFY(zap_add_int_key(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, ds->ds_phys->ds_next_snap_obj,
ds->ds_phys->ds_creation_txg, tx) == 0);
}
if (ds->ds_phys->ds_num_children > 1) {
boolean_t usenext = B_FALSE;
if (ds->ds_phys->ds_next_clones_obj != 0) {
uint64_t count;
/*
* A bug in a previous version of the code could
* cause upgrade_clones_cb() to not set
* ds_next_snap_obj when it should, leading to a
* missing entry. Therefore we can only use the
* next_clones_obj when its count is correct.
*/
int err = zap_count(dp->dp_meta_objset,
ds->ds_phys->ds_next_clones_obj, &count);
if (err == 0 &&
count == ds->ds_phys->ds_num_children - 1)
usenext = B_TRUE;
}
if (usenext) {
VERIFY0(zap_join_key(dp->dp_meta_objset,
ds->ds_phys->ds_next_clones_obj,
scn->scn_phys.scn_queue_obj,
ds->ds_phys->ds_creation_txg, tx));
} else {
struct enqueue_clones_arg eca;
eca.tx = tx;
eca.originobj = ds->ds_object;
VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
enqueue_clones_cb, &eca, DS_FIND_CHILDREN));
}
}
out:
dsl_dataset_rele(ds, FTAG);
}
/* ARGSUSED */
static int
enqueue_cb(dsl_pool_t *dp, dsl_dataset_t *hds, void *arg)
{
dmu_tx_t *tx = arg;
dsl_dataset_t *ds;
int err;
dsl_scan_t *scn = dp->dp_scan;
err = dsl_dataset_hold_obj(dp, hds->ds_object, FTAG, &ds);
if (err)
return (err);
while (ds->ds_phys->ds_prev_snap_obj != 0) {
dsl_dataset_t *prev;
err = dsl_dataset_hold_obj(dp, ds->ds_phys->ds_prev_snap_obj,
FTAG, &prev);
if (err) {
dsl_dataset_rele(ds, FTAG);
return (err);
}
/*
* If this is a clone, we don't need to worry about it for now.
*/
if (prev->ds_phys->ds_next_snap_obj != ds->ds_object) {
dsl_dataset_rele(ds, FTAG);
dsl_dataset_rele(prev, FTAG);
return (0);
}
dsl_dataset_rele(ds, FTAG);
ds = prev;
}
VERIFY(zap_add_int_key(dp->dp_meta_objset, scn->scn_phys.scn_queue_obj,
ds->ds_object, ds->ds_phys->ds_prev_snap_txg, tx) == 0);
dsl_dataset_rele(ds, FTAG);
return (0);
}
/*
* Scrub/dedup interaction.
*
* If there are N references to a deduped block, we don't want to scrub it
* N times -- ideally, we should scrub it exactly once.
*
* We leverage the fact that the dde's replication class (enum ddt_class)
* is ordered from highest replication class (DDT_CLASS_DITTO) to lowest
* (DDT_CLASS_UNIQUE) so that we may walk the DDT in that order.
*
* To prevent excess scrubbing, the scrub begins by walking the DDT
* to find all blocks with refcnt > 1, and scrubs each of these once.
* Since there are two replication classes which contain blocks with
* refcnt > 1, we scrub the highest replication class (DDT_CLASS_DITTO) first.
* Finally the top-down scrub begins, only visiting blocks with refcnt == 1.
*
* There would be nothing more to say if a block's refcnt couldn't change
* during a scrub, but of course it can so we must account for changes
* in a block's replication class.
*
* Here's an example of what can occur:
*
* If a block has refcnt > 1 during the DDT scrub phase, but has refcnt == 1
* when visited during the top-down scrub phase, it will be scrubbed twice.
* This negates our scrub optimization, but is otherwise harmless.
*
* If a block has refcnt == 1 during the DDT scrub phase, but has refcnt > 1
* on each visit during the top-down scrub phase, it will never be scrubbed.
* To catch this, ddt_sync_entry() notifies the scrub code whenever a block's
* reference class transitions to a higher level (i.e DDT_CLASS_UNIQUE to
* DDT_CLASS_DUPLICATE); if it transitions from refcnt == 1 to refcnt > 1
* while a scrub is in progress, it scrubs the block right then.
*/
static void
dsl_scan_ddt(dsl_scan_t *scn, dmu_tx_t *tx)
{
ddt_bookmark_t *ddb = &scn->scn_phys.scn_ddt_bookmark;
ddt_entry_t dde;
int error;
uint64_t n = 0;
bzero(&dde, sizeof (ddt_entry_t));
while ((error = ddt_walk(scn->scn_dp->dp_spa, ddb, &dde)) == 0) {
ddt_t *ddt;
if (ddb->ddb_class > scn->scn_phys.scn_ddt_class_max)
break;
dprintf("visiting ddb=%llu/%llu/%llu/%llx\n",
(longlong_t)ddb->ddb_class,
(longlong_t)ddb->ddb_type,
(longlong_t)ddb->ddb_checksum,
(longlong_t)ddb->ddb_cursor);
/* There should be no pending changes to the dedup table */
ddt = scn->scn_dp->dp_spa->spa_ddt[ddb->ddb_checksum];
ASSERT(avl_first(&ddt->ddt_tree) == NULL);
dsl_scan_ddt_entry(scn, ddb->ddb_checksum, &dde, tx);
n++;
if (dsl_scan_check_pause(scn, NULL))
break;
}
zfs_dbgmsg("scanned %llu ddt entries with class_max = %u; pausing=%u",
(longlong_t)n, (int)scn->scn_phys.scn_ddt_class_max,
(int)scn->scn_pausing);
ASSERT(error == 0 || error == ENOENT);
ASSERT(error != ENOENT ||
ddb->ddb_class > scn->scn_phys.scn_ddt_class_max);
}
/* ARGSUSED */
void
dsl_scan_ddt_entry(dsl_scan_t *scn, enum zio_checksum checksum,
ddt_entry_t *dde, dmu_tx_t *tx)
{
const ddt_key_t *ddk = &dde->dde_key;
ddt_phys_t *ddp = dde->dde_phys;
blkptr_t bp;
zbookmark_t zb = { 0 };
int p;
if (scn->scn_phys.scn_state != DSS_SCANNING)
return;
for (p = 0; p < DDT_PHYS_TYPES; p++, ddp++) {
if (ddp->ddp_phys_birth == 0 ||
ddp->ddp_phys_birth > scn->scn_phys.scn_max_txg)
continue;
ddt_bp_create(checksum, ddk, ddp, &bp);
scn->scn_visited_this_txg++;
scan_funcs[scn->scn_phys.scn_func](scn->scn_dp, &bp, &zb);
}
}
static void
dsl_scan_visit(dsl_scan_t *scn, dmu_tx_t *tx)
{
dsl_pool_t *dp = scn->scn_dp;
zap_cursor_t *zc;
zap_attribute_t *za;
if (scn->scn_phys.scn_ddt_bookmark.ddb_class <=
scn->scn_phys.scn_ddt_class_max) {
scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg;
scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg;
dsl_scan_ddt(scn, tx);
if (scn->scn_pausing)
return;
}
if (scn->scn_phys.scn_bookmark.zb_objset == DMU_META_OBJSET) {
/* First do the MOS & ORIGIN */
scn->scn_phys.scn_cur_min_txg = scn->scn_phys.scn_min_txg;
scn->scn_phys.scn_cur_max_txg = scn->scn_phys.scn_max_txg;
dsl_scan_visit_rootbp(scn, NULL,
&dp->dp_meta_rootbp, tx);
spa_set_rootblkptr(dp->dp_spa, &dp->dp_meta_rootbp);
if (scn->scn_pausing)
return;
if (spa_version(dp->dp_spa) < SPA_VERSION_DSL_SCRUB) {
VERIFY0(dmu_objset_find_dp(dp, dp->dp_root_dir_obj,
enqueue_cb, tx, DS_FIND_CHILDREN));
} else {
dsl_scan_visitds(scn,
dp->dp_origin_snap->ds_object, tx);
}
ASSERT(!scn->scn_pausing);
} else if (scn->scn_phys.scn_bookmark.zb_objset !=
ZB_DESTROYED_OBJSET) {
/*
* If we were paused, continue from here. Note if the
* ds we were paused on was deleted, the zb_objset may
* be -1, so we will skip this and find a new objset
* below.
*/
dsl_scan_visitds(scn, scn->scn_phys.scn_bookmark.zb_objset, tx);
if (scn->scn_pausing)
return;
}
/*
* In case we were paused right at the end of the ds, zero the
* bookmark so we don't think that we're still trying to resume.
*/
bzero(&scn->scn_phys.scn_bookmark, sizeof (zbookmark_t));
zc = kmem_alloc(sizeof(zap_cursor_t), KM_PUSHPAGE);
za = kmem_alloc(sizeof(zap_attribute_t), KM_PUSHPAGE);
/* keep pulling things out of the zap-object-as-queue */
while (zap_cursor_init(zc, dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj),
zap_cursor_retrieve(zc, za) == 0) {
dsl_dataset_t *ds;
uint64_t dsobj;
dsobj = strtonum(za->za_name, NULL);
VERIFY3U(0, ==, zap_remove_int(dp->dp_meta_objset,
scn->scn_phys.scn_queue_obj, dsobj, tx));
/* Set up min/max txg */
VERIFY3U(0, ==, dsl_dataset_hold_obj(dp, dsobj, FTAG, &ds));
if (za->za_first_integer != 0) {
scn->scn_phys.scn_cur_min_txg =
MAX(scn->scn_phys.scn_min_txg,
za->za_first_integer);
} else {
scn->scn_phys.scn_cur_min_txg =
MAX(scn->scn_phys.scn_min_txg,
ds->ds_phys->ds_prev_snap_txg);
}
scn->scn_phys.scn_cur_max_txg = dsl_scan_ds_maxtxg(ds);
dsl_dataset_rele(ds, FTAG);
dsl_scan_visitds(scn, dsobj, tx);
zap_cursor_fini(zc);
if (scn->scn_pausing)
goto out;
}
zap_cursor_fini(zc);
out:
kmem_free(za, sizeof(zap_attribute_t));
kmem_free(zc, sizeof(zap_cursor_t));
}
static boolean_t
dsl_scan_free_should_pause(dsl_scan_t *scn)
{
uint64_t elapsed_nanosecs;
if (zfs_recover)
return (B_FALSE);
elapsed_nanosecs = gethrtime() - scn->scn_sync_start_time;
return (elapsed_nanosecs / NANOSEC > zfs_txg_timeout ||
(NSEC2MSEC(elapsed_nanosecs) > zfs_free_min_time_ms &&
txg_sync_waiting(scn->scn_dp)) ||
spa_shutting_down(scn->scn_dp->dp_spa));
}
static int
dsl_scan_free_block_cb(void *arg, const blkptr_t *bp, dmu_tx_t *tx)
{
dsl_scan_t *scn = arg;
if (!scn->scn_is_bptree ||
(BP_GET_LEVEL(bp) == 0 && BP_GET_TYPE(bp) != DMU_OT_OBJSET)) {
if (dsl_scan_free_should_pause(scn))
return (SET_ERROR(ERESTART));
}
zio_nowait(zio_free_sync(scn->scn_zio_root, scn->scn_dp->dp_spa,
dmu_tx_get_txg(tx), bp, 0));
dsl_dir_diduse_space(tx->tx_pool->dp_free_dir, DD_USED_HEAD,
-bp_get_dsize_sync(scn->scn_dp->dp_spa, bp),
-BP_GET_PSIZE(bp), -BP_GET_UCSIZE(bp), tx);
scn->scn_visited_this_txg++;
return (0);
}
boolean_t
dsl_scan_active(dsl_scan_t *scn)
{
spa_t *spa = scn->scn_dp->dp_spa;
uint64_t used = 0, comp, uncomp;
if (spa->spa_load_state != SPA_LOAD_NONE)
return (B_FALSE);
if (spa_shutting_down(spa))
return (B_FALSE);
if (scn->scn_phys.scn_state == DSS_SCANNING ||
scn->scn_async_destroying)
return (B_TRUE);
if (spa_version(scn->scn_dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
(void) bpobj_space(&scn->scn_dp->dp_free_bpobj,
&used, &comp, &uncomp);
}
return (used != 0);
}
void
dsl_scan_sync(dsl_pool_t *dp, dmu_tx_t *tx)
{
dsl_scan_t *scn = dp->dp_scan;
spa_t *spa = dp->dp_spa;
int err;
/*
* Check for scn_restart_txg before checking spa_load_state, so
* that we can restart an old-style scan while the pool is being
* imported (see dsl_scan_init).
*/
if (scn->scn_restart_txg != 0 &&
scn->scn_restart_txg <= tx->tx_txg) {
pool_scan_func_t func = POOL_SCAN_SCRUB;
dsl_scan_done(scn, B_FALSE, tx);
if (vdev_resilver_needed(spa->spa_root_vdev, NULL, NULL))
func = POOL_SCAN_RESILVER;
zfs_dbgmsg("restarting scan func=%u txg=%llu",
func, tx->tx_txg);
dsl_scan_setup_sync(&func, tx);
}
if (!dsl_scan_active(scn) ||
spa_sync_pass(dp->dp_spa) > 1)
return;
scn->scn_visited_this_txg = 0;
scn->scn_pausing = B_FALSE;
scn->scn_sync_start_time = gethrtime();
spa->spa_scrub_active = B_TRUE;
/*
* First process the free list. If we pause the free, don't do
* any scanning. This ensures that there is no free list when
* we are scanning, so the scan code doesn't have to worry about
* traversing it.
*/
if (spa_version(dp->dp_spa) >= SPA_VERSION_DEADLISTS) {
scn->scn_is_bptree = B_FALSE;
scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
NULL, ZIO_FLAG_MUSTSUCCEED);
err = bpobj_iterate(&dp->dp_free_bpobj,
dsl_scan_free_block_cb, scn, tx);
VERIFY3U(0, ==, zio_wait(scn->scn_zio_root));
if (err == 0 && spa_feature_is_active(spa,
&spa_feature_table[SPA_FEATURE_ASYNC_DESTROY])) {
ASSERT(scn->scn_async_destroying);
scn->scn_is_bptree = B_TRUE;
scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
NULL, ZIO_FLAG_MUSTSUCCEED);
err = bptree_iterate(dp->dp_meta_objset,
dp->dp_bptree_obj, B_TRUE, dsl_scan_free_block_cb,
scn, tx);
VERIFY0(zio_wait(scn->scn_zio_root));
if (err == 0) {
zfeature_info_t *feat = &spa_feature_table
[SPA_FEATURE_ASYNC_DESTROY];
/* finished; deactivate async destroy feature */
spa_feature_decr(spa, feat, tx);
ASSERT(!spa_feature_is_active(spa, feat));
VERIFY0(zap_remove(dp->dp_meta_objset,
DMU_POOL_DIRECTORY_OBJECT,
DMU_POOL_BPTREE_OBJ, tx));
VERIFY0(bptree_free(dp->dp_meta_objset,
dp->dp_bptree_obj, tx));
dp->dp_bptree_obj = 0;
scn->scn_async_destroying = B_FALSE;
}
}
if (scn->scn_visited_this_txg) {
zfs_dbgmsg("freed %llu blocks in %llums from "
"free_bpobj/bptree txg %llu",
(longlong_t)scn->scn_visited_this_txg,
(longlong_t)
NSEC2MSEC(gethrtime() - scn->scn_sync_start_time),
(longlong_t)tx->tx_txg);
scn->scn_visited_this_txg = 0;
/*
* Re-sync the ddt so that we can further modify
* it when doing bprewrite.
*/
ddt_sync(spa, tx->tx_txg);
}
if (err == ERESTART)
return;
}
if (scn->scn_phys.scn_state != DSS_SCANNING)
return;
if (scn->scn_done_txg == tx->tx_txg) {
ASSERT(!scn->scn_pausing);
/* finished with scan. */
zfs_dbgmsg("txg %llu scan complete", tx->tx_txg);
dsl_scan_done(scn, B_TRUE, tx);
ASSERT3U(spa->spa_scrub_inflight, ==, 0);
dsl_scan_sync_state(scn, tx);
return;
}
if (scn->scn_phys.scn_ddt_bookmark.ddb_class <=
scn->scn_phys.scn_ddt_class_max) {
zfs_dbgmsg("doing scan sync txg %llu; "
"ddt bm=%llu/%llu/%llu/%llx",
(longlong_t)tx->tx_txg,
(longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_class,
(longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_type,
(longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_checksum,
(longlong_t)scn->scn_phys.scn_ddt_bookmark.ddb_cursor);
ASSERT(scn->scn_phys.scn_bookmark.zb_objset == 0);
ASSERT(scn->scn_phys.scn_bookmark.zb_object == 0);
ASSERT(scn->scn_phys.scn_bookmark.zb_level == 0);
ASSERT(scn->scn_phys.scn_bookmark.zb_blkid == 0);
} else {
zfs_dbgmsg("doing scan sync txg %llu; bm=%llu/%llu/%llu/%llu",
(longlong_t)tx->tx_txg,
(longlong_t)scn->scn_phys.scn_bookmark.zb_objset,
(longlong_t)scn->scn_phys.scn_bookmark.zb_object,
(longlong_t)scn->scn_phys.scn_bookmark.zb_level,
(longlong_t)scn->scn_phys.scn_bookmark.zb_blkid);
}
scn->scn_zio_root = zio_root(dp->dp_spa, NULL,
NULL, ZIO_FLAG_CANFAIL);
dsl_pool_config_enter(dp, FTAG);
dsl_scan_visit(scn, tx);
dsl_pool_config_exit(dp, FTAG);
(void) zio_wait(scn->scn_zio_root);
scn->scn_zio_root = NULL;
zfs_dbgmsg("visited %llu blocks in %llums",
(longlong_t)scn->scn_visited_this_txg,
(longlong_t)NSEC2MSEC(gethrtime() - scn->scn_sync_start_time));
if (!scn->scn_pausing) {
scn->scn_done_txg = tx->tx_txg + 1;
zfs_dbgmsg("txg %llu traversal complete, waiting till txg %llu",
tx->tx_txg, scn->scn_done_txg);
}
if (DSL_SCAN_IS_SCRUB_RESILVER(scn)) {
mutex_enter(&spa->spa_scrub_lock);
while (spa->spa_scrub_inflight > 0) {
cv_wait(&spa->spa_scrub_io_cv,
&spa->spa_scrub_lock);
}
mutex_exit(&spa->spa_scrub_lock);
}
dsl_scan_sync_state(scn, tx);
}
/*
* This will start a new scan, or restart an existing one.
*/
void
dsl_resilver_restart(dsl_pool_t *dp, uint64_t txg)
{
if (txg == 0) {
dmu_tx_t *tx;
tx = dmu_tx_create_dd(dp->dp_mos_dir);
VERIFY(0 == dmu_tx_assign(tx, TXG_WAIT));
txg = dmu_tx_get_txg(tx);
dp->dp_scan->scn_restart_txg = txg;
dmu_tx_commit(tx);
} else {
dp->dp_scan->scn_restart_txg = txg;
}
zfs_dbgmsg("restarting resilver txg=%llu", txg);
}
boolean_t
dsl_scan_resilvering(dsl_pool_t *dp)
{
return (dp->dp_scan->scn_phys.scn_state == DSS_SCANNING &&
dp->dp_scan->scn_phys.scn_func == POOL_SCAN_RESILVER);
}
/*
* scrub consumers
*/
static void
count_block(zfs_all_blkstats_t *zab, const blkptr_t *bp)
{
int i;
/*
* If we resume after a reboot, zab will be NULL; don't record
* incomplete stats in that case.
*/
if (zab == NULL)
return;
for (i = 0; i < 4; i++) {
int l = (i < 2) ? BP_GET_LEVEL(bp) : DN_MAX_LEVELS;
int t = (i & 1) ? BP_GET_TYPE(bp) : DMU_OT_TOTAL;
int equal;
zfs_blkstat_t *zb;
if (t & DMU_OT_NEWTYPE)
t = DMU_OT_OTHER;
zb = &zab->zab_type[l][t];
zb->zb_count++;
zb->zb_asize += BP_GET_ASIZE(bp);
zb->zb_lsize += BP_GET_LSIZE(bp);
zb->zb_psize += BP_GET_PSIZE(bp);
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_2_of_2_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 == 1)
zb->zb_ditto_2_of_3_samevdev++;
else if (equal == 3)
zb->zb_ditto_3_of_3_samevdev++;
break;
}
}
}
static void
dsl_scan_scrub_done(zio_t *zio)
{
spa_t *spa = zio->io_spa;
zio_data_buf_free(zio->io_data, zio->io_size);
mutex_enter(&spa->spa_scrub_lock);
spa->spa_scrub_inflight--;
cv_broadcast(&spa->spa_scrub_io_cv);
if (zio->io_error && (zio->io_error != ECKSUM ||
!(zio->io_flags & ZIO_FLAG_SPECULATIVE))) {
spa->spa_dsl_pool->dp_scan->scn_phys.scn_errors++;
}
mutex_exit(&spa->spa_scrub_lock);
}
static int
dsl_scan_scrub_cb(dsl_pool_t *dp,
const blkptr_t *bp, const zbookmark_t *zb)
{
dsl_scan_t *scn = dp->dp_scan;
size_t size = BP_GET_PSIZE(bp);
spa_t *spa = dp->dp_spa;
uint64_t phys_birth = BP_PHYSICAL_BIRTH(bp);
boolean_t needs_io = B_FALSE;
int zio_flags = ZIO_FLAG_SCAN_THREAD | ZIO_FLAG_RAW | ZIO_FLAG_CANFAIL;
int scan_delay = 0;
int d;
if (phys_birth <= scn->scn_phys.scn_min_txg ||
phys_birth >= scn->scn_phys.scn_max_txg)
return (0);
count_block(dp->dp_blkstats, bp);
ASSERT(DSL_SCAN_IS_SCRUB_RESILVER(scn));
if (scn->scn_phys.scn_func == POOL_SCAN_SCRUB) {
zio_flags |= ZIO_FLAG_SCRUB;
needs_io = B_TRUE;
scan_delay = zfs_scrub_delay;
} else {
ASSERT3U(scn->scn_phys.scn_func, ==, POOL_SCAN_RESILVER);
zio_flags |= ZIO_FLAG_RESILVER;
needs_io = B_FALSE;
scan_delay = zfs_resilver_delay;
}
/* If it's an intent log block, failure is expected. */
if (zb->zb_level == ZB_ZIL_LEVEL)
zio_flags |= ZIO_FLAG_SPECULATIVE;
for (d = 0; d < BP_GET_NDVAS(bp); d++) {
vdev_t *vd = vdev_lookup_top(spa,
DVA_GET_VDEV(&bp->blk_dva[d]));
/*
* Keep track of how much data we've examined so that
* zpool(1M) status can make useful progress reports.
*/
scn->scn_phys.scn_examined += DVA_GET_ASIZE(&bp->blk_dva[d]);
spa->spa_scan_pass_exam += DVA_GET_ASIZE(&bp->blk_dva[d]);
/* if it's a resilver, this may not be in the target range */
if (!needs_io) {
if (DVA_GET_GANG(&bp->blk_dva[d])) {
/*
* Gang members may be spread across multiple
* vdevs, so the best estimate we have is the
* scrub range, which has already been checked.
* XXX -- it would be better to change our
* allocation policy to ensure that all
* gang members reside on the same vdev.
*/
needs_io = B_TRUE;
} else {
needs_io = vdev_dtl_contains(vd, DTL_PARTIAL,
phys_birth, 1);
}
}
}
if (needs_io && !zfs_no_scrub_io) {
vdev_t *rvd = spa->spa_root_vdev;
uint64_t maxinflight = rvd->vdev_children * zfs_top_maxinflight;
void *data = zio_data_buf_alloc(size);
mutex_enter(&spa->spa_scrub_lock);
while (spa->spa_scrub_inflight >= maxinflight)
cv_wait(&spa->spa_scrub_io_cv, &spa->spa_scrub_lock);
spa->spa_scrub_inflight++;
mutex_exit(&spa->spa_scrub_lock);
/*
* If we're seeing recent (zfs_scan_idle) "important" I/Os
* then throttle our workload to limit the impact of a scan.
*/
if (ddi_get_lbolt64() - spa->spa_last_io <= zfs_scan_idle)
delay(scan_delay);
zio_nowait(zio_read(NULL, spa, bp, data, size,
dsl_scan_scrub_done, NULL, ZIO_PRIORITY_SCRUB,
zio_flags, zb));
}
/* do not relocate this block */
return (0);
}
int
dsl_scan(dsl_pool_t *dp, pool_scan_func_t func)
{
spa_t *spa = dp->dp_spa;
/*
* Purge all vdev caches and probe all devices. We do this here
* rather than in sync context because this requires a writer lock
* on the spa_config lock, which we can't do from sync context. The
* spa_scrub_reopen flag indicates that vdev_open() should not
* attempt to start another scrub.
*/
spa_vdev_state_enter(spa, SCL_NONE);
spa->spa_scrub_reopen = B_TRUE;
vdev_reopen(spa->spa_root_vdev);
spa->spa_scrub_reopen = B_FALSE;
(void) spa_vdev_state_exit(spa, NULL, 0);
return (dsl_sync_task(spa_name(spa), dsl_scan_setup_check,
dsl_scan_setup_sync, &func, 0));
}
#if defined(_KERNEL) && defined(HAVE_SPL)
module_param(zfs_top_maxinflight, int, 0644);
MODULE_PARM_DESC(zfs_top_maxinflight, "Max I/Os per top-level");
module_param(zfs_resilver_delay, int, 0644);
MODULE_PARM_DESC(zfs_resilver_delay, "Number of ticks to delay resilver");
module_param(zfs_scrub_delay, int, 0644);
MODULE_PARM_DESC(zfs_scrub_delay, "Number of ticks to delay scrub");
module_param(zfs_scan_idle, int, 0644);
MODULE_PARM_DESC(zfs_scan_idle, "Idle window in clock ticks");
module_param(zfs_scan_min_time_ms, int, 0644);
MODULE_PARM_DESC(zfs_scan_min_time_ms, "Min millisecs to scrub per txg");
module_param(zfs_free_min_time_ms, int, 0644);
MODULE_PARM_DESC(zfs_free_min_time_ms, "Min millisecs to free per txg");
module_param(zfs_resilver_min_time_ms, int, 0644);
MODULE_PARM_DESC(zfs_resilver_min_time_ms, "Min millisecs to resilver per txg");
module_param(zfs_no_scrub_io, int, 0644);
MODULE_PARM_DESC(zfs_no_scrub_io, "Set to disable scrub I/O");
module_param(zfs_no_scrub_prefetch, int, 0644);
MODULE_PARM_DESC(zfs_no_scrub_prefetch, "Set to disable scrub prefetching");
#endif
Reference in New Issue View Git Blame Copy Permalink