Commit Graph

2540 Commits

Author SHA1 Message Date
cao
e2c7d3785a Remove unused sa_update_from_cb()
It looks like this was functionality which was added in the
original SA implementation and then never needed.  It can
be safely removed now and easily added back if we find a
use for it.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: cao.xuewen <cao.xuewen@zte.com.cn>
Closes #5440
2016-12-01 16:39:06 -07:00
cao
6a8fd57fa7 Compile zio.h and zio_impl.h mutual include
zio.h includes zio_impl.h but zio_impl.h also includes zio.h, so the
header files to contain each other.  Get rid of the zio_impl.h include
in zio.h and update zio_inject.c to include zio.h instead of zio_impl.h.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: cao.xuewen <cao.xuewen@zte.com.cn>
Closes #5439
2016-12-01 16:36:25 -07:00
Håkan Johansson
5f20c145df Do not force VDEV_NAME_TYPE_ID in max_width()
Do not force VDEV_NAME_TYPE_ID in max_width(), instead add it
in the relevant calls to max_width().

The first location of max_width() where VDEV_NAME_TYPE_ID is
now added in show_import() is followed by print_import_config() and
print_logs().  Both these print children vdev names that have been
retrieved using an explicit VDEV_NAME_TYPE_ID added.

The second location is in status_callback().  This is followed by
print_status_config(), print_logs(), print_l2cache(), and
print_spares(). For l2cache and spares it should not matter as there
are no mirror-X or raidz-X involved.  print_status_config() as above
retrieves the name using explicit VDEV_NAME_TYPE_ID before
calling itself to print children.

The call of max_width() in get_namewidth() is not changed, as this is
used by zpool_do_iostat(), followed by print_iostat(), which does not
add VDEV_NAME_TYPE_ID.

Overall, we should consider adding VDEV_NAME_TYPE_ID to the
relevant name_flags / cb_name_flags fields, and remove the explicit
adding in called routines.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Haakan T Johansson <f96hajo@chalmers.se>
Closes #5401
2016-11-30 17:46:16 -07:00
Brian Behlendorf
a3fd9d9e15 Convert zio_buf_alloc() consumers
In multiple cases zio_buf_alloc() was used instead of kmem_alloc()
or vmem_alloc().  This was often done because the allocations
could be large and it was easy to use zfs_buf_alloc() for them.

But this isn't ideal for allocations which are small or short
lived.  In these cases it is better to use kmem_alloc() or
vmem_alloc().  If possible we want to avoid the case where
we have slabs allocated for kmem caches which are rarely used.

Note for small allocations vmem_alloc() will be internally
converted to kmem_alloc().  Therefore as long as large
allocations are infrequent and short lived the penalty for
using vmem_alloc() is small.

Reviewed-by: Chunwei Chen <david.chen@osnexus.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #5409
2016-11-30 16:18:20 -07:00
Brian Behlendorf
7657defc48 Introduce ARC Buffer Data (ABD)
ZFS currently uses ARC buffers which are backed by virtual memory.
While functional, there are some major problems with this approach
which can be observed on all OpenZFS platforms.  ABD was designed
to address these issues and includes contributions from OpenZFS
developers from multiple platforms.

While all OpenZFS platforms will benefit from ABD this functionality
is critical for Linux.  Unlike the other OpenZFS platforms the Linux
kernel discourages extensive use of virtual memory.  The provided
interfaces are not optimized for frequent allocations from the virtual
address space.  To maintain good performance a kmem cache is
used which contains relatively long lived slabs backed by virtual
memory.  The downside to the approach is that those slabs can
become highly fragmented resulting in an inefficient use of memory.

Another issue is that on 32-bit systems the available virtual
address space in the kernel is only a small fraction of total
system memory.  This means the ARC size is highly constrained
which hurts performance and make allocating memory difficult
and OOMs more likely.

ABD is designed to address these issues by using scatter lists
of pages for data buffers.  This removes the need for slabs
which resolves the fragmentation issue.  It also allows high
memory pages to be allocated which alleviates the virtual
address space pressure on 32-bit systems.

For metadata buffers, which are small, linear ABDs are allocated
from the slab.  This is preferable because there are many places
in the code which expect to be able to read from a given offset
in the buffer.  Using linear ABDs means none of that code needs
to be modified.  The majority of these buffers are allocated with
kmalloc so there's minimal impact of the virtual address space.

Tested-by: Kash Pande <kash@tripleback.net>
Tested-by: kernelOfTruth <kerneloftruth@gmail.com>
Tested-by: RageLtMan <rageltman@sempervictus>
Tested-by: DHE <git@dehacked.net>
Reviewed-by: Chunwei Chen <david.chen@osnexus.com>
Reviewed-by: Dan Kimmel <dan.kimmel@delphix.com>
Reviewed-by: David Quigley <david.quigley@intel.com>
Reviewed-by: Gvozden Neskovic <neskovic@gmail.com>
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: Isaac Huang <he.huang@intel.com>
Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #3441 
Closes #5135
2016-11-30 14:48:16 -07:00
ChaoyuZhang
ce43e88dd6 Enable ro_props_001_pos
This script was disabled as the avail/used space changed slightly.
Add sync_pool() and a short delay after snapshots are created to
ensure everything in flight has been written.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: ChaoyuZhang <zhang.chaoyu@zte.com.cn>
Closes #5201 
Closes #5419
2016-11-30 11:27:04 -07:00
luozhengzheng
2d44b04b79 Fix coverity defects: CID 154591
CID 154591: Incorrect expression (SIZEOF_MISMATCH)

Reviewed-by: Gvozden Neskovic <neskovic@gmail.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: luozhengzheng <luo.zhengzheng@zte.com.cn>
Closes #5435
2016-11-30 10:48:01 -07:00
Chunwei Chen
9829574834 ABD optimized page allocation code
* Convert ABD to use the Linux Kernel scatterlist implementation
  instead of the hand rolled one from illumos.

* Scatter ABDs are preferentially populated with higher order
  compound pages from a single zone.  Allocation size is
  progressively decreased until it can be satisfied without
  performing reclaim or compaction.

* An alternate page allocator is provided for kernels older
  than 3.6 and for CONFIG_HIGHMEM systems.  This allocator
  is designed as a fallback for maximum compatibility.

* Extended abdstats to provide visibility in the the allocator.

* Add cached value for PAGESIZE in userspace.

Contributions-by:
Chunwei Chen <david.chen@osnexus.com>
Gvozden Neskovic <neskovic@gmail.com>
Jinshan Xiong <jinshan.xiong@intel.com>
Isaac Huang <he.huang@intel.com>
David Quigley <david.quigley@intel.com>
Brian Behlendorf <behlendorf1@llnl.gov>
2016-11-29 14:34:33 -08:00
Chunwei Chen
4f60152910 ABD kmap to kmap_atomic
Convert usage of kmap to kmap_atomic while correctly saving off
irq state.
2016-11-29 14:34:33 -08:00
Romain Dolbeau
88cc2352ea ABD raidz NEON support
Port NEON implementation of RAID-Z functions to ABD.

Signed-off-by: Roomain Dolbeau <romain.dolbeau@atos.net>
2016-11-29 14:34:33 -08:00
Gvozden Neskovic
65d71d4212 ABD raidz avx512f support
Implement shift based multiplication for 512f. Higher IPC over lookup based
methods yields up to 40% better performance on the current hardware.

Results on Xeon Phi(TM) CPU 7210:
implementation   gen_p           gen_pq          gen_pqr         rec_p           rec_q           rec_r           rec_pq          rec_pr          rec_qr          rec_pqr
original         142232671       24411492        12948205        283053705       22348167        4215911         9171609         2265548         2378370         1648495
scalar           295711162       49851491        33253815        293198109       88179448        61866752        27941684        25764416        17384442        12138153
sse2             410055998       199642658       117973654       406240463       152688682       121092250       84968180        79291076        47473657        20779719
ssse3            411641595       199669571       117937647       406211024       137638508       117050346       81263322        76120405        46281559        32696722
avx2             616485806       311515332       188595628       605455115       260602390       230554476       148198817       138800254       92273356        62937819
avx512f          832191523       408509425       253599522       810094481       404325734       317590971       218235687       197204920       133101937       94001219
fastest          avx512f         avx512f         avx512f         avx512f         avx512f         avx512f         avx512f         avx512f         avx512f         avx512f

Signed-off-by: Gvozden Neskovic <neskovic@gmail.com>
2016-11-29 14:34:33 -08:00
Gvozden Neskovic
cbf484f8ad ABD Vectorized raidz
Enable vectorized raidz code on ABD buffers.  The avx512f,
avx512bw, neon and aarch64_neonx2 are disabled in this commit.
With the exception of avx512bw these implementations are
updated for ABD in the subsequent commits.

Signed-off-by: Gvozden Neskovic <neskovic@gmail.com>
2016-11-29 14:34:33 -08:00
Gvozden Neskovic
a206522c4f ABD changes for vectorized RAIDZ
* userspace: aligned buffers. Minimum of 32B alignment is
  needed for AVX2. Kernel buffers are aligned 512B or more.
* add abd_get_offset_size() interface
* abd_iter_map(): fix calculation of iter_mapsize
* add abd_raidz_gen_iterate() and abd_raidz_rec_iterate()

Signed-off-by: Gvozden Neskovic <neskovic@gmail.com>
2016-11-29 14:34:33 -08:00
Isaac Huang
b0be93e81a ABD page support to vdev_disk.c
Signed-off-by: Isaac Huang <he.huang@intel.com>
2016-11-29 14:34:32 -08:00
David Quigley
a6255b7fce DLPX-44812 integrate EP-220 large memory scalability 2016-11-29 14:34:27 -08:00
Tim Chase
616fa7c02b zstreamdump needs to initialize fletcher 4 support
Otherwise, the checksum function pointer isn't initialized.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tim Chase <tim@chase2k.com>
Closes #5411
2016-11-29 14:47:05 -07:00
Tony Hutter
8720e9e748 Add -c to zpool iostat & status to run command
This patch adds a command (-c) option to zpool status and zpool iostat.  The
-c option allows you to run an arbitrary command on each vdev and display
the first line of output in zpool status/iostat.  The environment vars
VDEV_PATH and VDEV_UPATH are set to the vdev's path and "underlying path"
before running the command.  For device mapper, multipath, or partitioned
vdevs, VDEV_UPATH is the actual underlying /dev/sd* disk.  This can be useful
if the command you're running requires a /dev/sd* device.

The patch also uses /sys/block/<dev>/slaves/ to lookup the underlying device
instead of using libdevmapper.  This not only removes the libdevmapper
requirement at build time, but also allows you to resolve device mapper
devices without being root.  This means that UDEV_UPATH get set correctly
when running zpool status/iostat as an unprivileged user.

Example:

$ zpool status -c 'echo I am $VDEV_PATH, $VDEV_UPATH'

NAME        STATE     READ WRITE CKSUM
mypool      ONLINE       0     0     0
  mirror-0  ONLINE       0     0     0
    mpatha  ONLINE       0     0     0  I am /dev/mapper/mpatha, /dev/sdc
    sdb     ONLINE       0     0     0  I am /dev/sdb1, /dev/sdb

Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes #5368
2016-11-29 14:45:38 -07:00
LOLi
2f71caf2d9 Allow zfs unshare <protocol> -a
Allow `zfs unshare <protocol> -a` command to share or unshare all datasets
of a given protocol, nfs or smb.

Additionally, enable most of ZFS Test Suite zfs_share/zfs_unshare test cases.
To work around some Illumos-specific functionalities ($SHARE/$UNSHARE) some
function wrappers were added around them.

Finally, fix and issue in smb_is_share_active() that would leave SMB shares
exported when invoking 'zfs unshare -a'

Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Turbo Fredriksson <turbo@bayour.com>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes #3238 
Closes #5367
2016-11-29 12:22:38 -07:00
Giuseppe Di Natale
251cb8dfac Ensure that perf regression tests cleanup properly
Each test in the performance regression test suite
creates a pool and a dataset for use. Unfortunately,
these tests do not cleanup the pool and dataset
correctly once they complete. Each test now kills
fio and iostat, destroys the dataset, and finally
destroys the pool. Each test also now traps the
SIGTERM signal to handle cases where test-runner
kills a test.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Requires-builders: all
Closes #5407
2016-11-28 17:24:47 -07:00
ChaoyuZhang
ce4197c1ca Enable user_property_002_pos
The user_property_002_pos passes as expected.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: ChaoyuZhang <zhang.chaoyu@zte.com.cn>
Closes #5406
2016-11-18 16:25:06 -08:00
DeHackEd
7ca25051b6 Kernel 4.9 compat: file_operations->aio_fsync removal
Linux kernel commit 723c038475b78 removed this field.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: DHE <git@dehacked.net>
Closes #5393
2016-11-15 09:20:46 -08:00
DeHackEd
6146e17ebe Fix man page formatting in zfs-module-parameters
Bold and Normal codes were mixed up in a few places resulting in
bad highlighting.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: DHE <git@dehacked.net>
Closes #5397
2016-11-14 17:03:57 -08:00
Håkan Johansson
1c91161258 Repair indent of zpool.8 man page
Repair indent of zpool.8 man page, just before zpool labelclear
details.  Accidentally introduced by 193a37cb2 (git bisect).

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Haakan T Johansson <f96hajo@chalmers.se>
Closes #5394
2016-11-14 09:47:49 -08:00
Brian Behlendorf
c30d8ded0c Fix 'zpool import' detection issue
Before adding the entry to the configuration verify that the
device can be opened exclusively.  This ensures that as long
as multipathd is running the underlying multipath devices, which
otherwise appear identical to their /dev/mapper counterpart,
are pruned from the configuration.

Failure to do so can result in a result in the vdev appearing
as UNAVAIL when the vdev path provided to the kernel can't be
opened exclusively.

This check would normally be performed in zpool_open_func()
but placing it there would result in false positives because
it is called concurrently for many devices.


Reviewed-by: Olaf Faaland <faaland1@llnl.gov>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #5387
2016-11-14 09:40:18 -08:00
Don Brady
0df15db98f Add a statechange notify zedlet
Now that ZED has internal fault diagnosis and the statechange event
is generated for faulted states, we can replace the io-notify and
checksum-notify zedlets with one based on statechange.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Don Brady <don.brady@intel.com>
Closes #5383
2016-11-10 13:52:59 -08:00
luozhengzheng
32dec7bd1a Fix coverity defects: CID 147503
CID 147503: Dereference after null check (FORWARD_NULL)

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: luozhengzheng <luo.zhengzheng@zte.com.cn>
Closes #5326
2016-11-10 08:50:32 -08:00
cao
3bfd95d589 Fix coverity defects: CID 147540, 147542
CID 147540: unsigned_compare
- Cast nsec to a int32_t to properly detect the expected overflow.
CID 147542: unsigned_compare
- intval can never be less than ZIO_FAILURE_MODE_WAIT which is
  defined to be zero.  Remove this useless check.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: cao.xuewen <cao.xuewen@zte.com.cn>
Closes #5379
2016-11-09 17:35:26 -08:00
Gvozden Neskovic
7e5ea7be7f Fix ZFS_AC_KERNEL_SET_CACHED_ACL_USABLE check
Pass `ACL_TYPE_ACCESS` for type parameter of `set_cached_acl()` and
`forget_cached_acl()` to avoid removal of dead code after BUG() in
compile time. Tested on 3.2.0 kernel.

Introduced in 3779913

Reviewed-by: Massimo Maggi <me@massimo-maggi.eu>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Chunwei Chen <david.chen@osnexus.com>
Signed-off-by: Gvozden Neskovic <neskovic@gmail.com>
Closes #5378
2016-11-09 13:53:13 -08:00
jxiong
126ae9f4e9 Export symbol dmu_objset_userobjspace_upgradable
It's used by Lustre to determine if the objset can be upgraded.
The inline version doesn't work because dmu_objset_is_snapshot()
is not exported.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Jinshan Xiong <jinshan.xiong@intel.com>
Closes #5385
2016-11-09 13:51:12 -08:00
tuxoko
0420c126ce Linux 3.14 compat: assign inode->set_acl
Linux 3.14 introduces inode->set_acl(). Normally, acl modification will come
from setxattr, which will handle by the acl xattr_handler, and we already
handles that well. However, nfsd will directly calls inode->set_acl or
return error if it doesn't exists.

Reviewed-by: Tim Chase <tim@chase2k.com>
Reviewed-by: Massimo Maggi <me@massimo-maggi.eu>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Chunwei Chen <david.chen@osnexus.com>
Closes #5371 
Closes #5375
2016-11-09 10:37:17 -08:00
Olaf Faaland
66f801f00a Fix symlinks for {vdev_clear,statechange}-led.sh
These were named in the zed/Makefile.am as vdev_clear-blinkled.sh
and statechange-blinkled.sh causing bad symlinks to be created.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Olaf Faaland <faaland1@llnl.gov>
Closes #5384
2016-11-09 10:19:43 -08:00
cao
010782be91 Fix coverity defects: CID 147586
CID 147586: function:allow_usage Type:out-of-bounds read

Reviewed-by: Chunwei Chen <david.chen@osnexus.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: cao.xuewen <cao.xuewen@zte.com.cn>
Closes #5364
2016-11-08 17:33:23 -08:00
cao
acc6d0e223 Fix coverity defects: CID 147629
CID 147629: Type:Dereference before null check

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov
Signed-off-by: <cao.xuewen cao.xuewen@zte.com.cn>
Closes #5376
2016-11-08 16:41:31 -08:00
luozhengzheng
70df4c21c0 Fix coverity defects: 154021
CID 154021: Null pointer dereference

Reviewed-by: Giuseppe Di Natale <dinatale2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: luozhengzheng <luo.zhengzheng@zte.com.cn>
Closes #5380
2016-11-08 14:34:52 -08:00
cao
a36cc8d242 Fix coverity defects: CID 147626, 147628
CID 147626: Type:Dereference before null check
CID 147628: Type:Dereference before null check

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov
Reviewed-by: Chunwei Chen <david.chen@osnexus.com>
Signed-off-by: cao.xuewen <cao.xuewen@zte.com.cn>
Closes #5304
2016-11-08 14:28:17 -08:00
Brian Behlendorf
ffb78320c8 Skip test suites on 32-bit TEST builders
The ztest, filebench, xfstests, and zfsstress test suites should
be skipped when testing on 32-bit platforms until they pass
reliably.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #5381
2016-11-08 13:57:17 -08:00
Don Brady
976246fadd Add illumos FMD ZFS logic to ZED -- phase 2
The phase 2 work primarily entails the Diagnosis Engine and
the Retire Agent modules. It also includes infrastructure
to support a crude FMD environment to host these modules.

The Diagnosis Engine consumes I/O and checksum ereports and
feeds them into a SERD engine which will generate a corres-
ponding fault diagnosis when the SERD engine fires. All the
diagnosis state data is collected into cases, one case per
vdev being tracked.

The Retire Agent responds to diagnosed faults by isolating
the faulty VDEV. It will notify the ZFS kernel module of
the new VDEV state (degraded or faulted). This agent is
also responsible for managing hot spares across pools.
When it encounters a device fault or a device removal it
replaces the device with an appropriate spare if available.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Don Brady <don.brady@intel.com>
Closes #5343
2016-11-07 15:01:38 -08:00
cao
f4bae2ed63 Fix coverity defects: CID 147575, 147577, 147578, 147579
CID 147575, Type:Unintentional integer overflow
CID 147577, Type:Unintentional integer overflow
CID 147578, Type:Unintentional integer overflow
CID 147579, Type:Unintentional integer overflow

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: cao.xuewen <cao.xuewen@zte.com.cn>
Closes #5365
2016-11-07 14:54:32 -08:00
Chunwei Chen
3779913b35 Use set_cached_acl and forget_cached_acl when possible
Originally, these two function are inline, so their usability is tied to
posix_acl_release. However, since Linux 3.14, they became EXPORT_SYMBOL, so we
can always use them. In this patch, we create an independent test for these
two functions so we can use them when possible.

Signed-off-by: Chunwei Chen <david.chen@osnexus.com>
2016-11-07 11:04:44 -08:00
Chunwei Chen
8e71ab99dc Batch free zpl_posix_acl_release
Currently every calls to zpl_posix_acl_release will schedule a delayed task,
and each delayed task will add a timer. This used to be fine except for
possibly bad performance impact.

However, in Linux 4.8, a new timer wheel implementation[1] is introduced. In
this new implementation, the larger the delay, the less accuracy the timer is.
So when we have a flood of timer from zpl_posix_acl_release, they will expire
at the same time. Couple with the fact that task_expire will do linear search
with lock held. This causes an extreme amount of contention inside interrupt
and would actually lockup the system.

We fix this by doing batch free to prevent a flood of delayed task. Every call
to zpl_posix_acl_release will put the posix_acl to be freed on a lockless
list. Every batch window, 1 sec, the zpl_posix_acl_free will fire up and free
every posix_acl that passed the grace period on the list. This way, we only
have one delayed task every second.

[1] https://lwn.net/Articles/646950/

Signed-off-by: Chunwei Chen <david.chen@osnexus.com>
2016-11-07 11:04:44 -08:00
Brian Behlendorf
83bf769d50 Fix 'zpool import' detection issues
This patch addresses multiple 'zpool import' block device
indentification problems which are most likely to occur on a
system configured to use blkid, by_vdev paths, multipath and
failover.  The symptom most commonly observed is the import
uses different path names to import the pool than would
normally be expected.

* When using blkid to identify vdevs the listed devices may
be added to the cache in any order.  In order to apply the
preferred search order heuristic a zfs_path_order() function
was added to calculate the order given full path names.

* Since it's possible to have multiple block devices with
different vdev guids which refer to the same ZPOOL_CONFIG_PATH
the slice cache must be indexed by guid and name.  By avoiding
collisions the preferred ordering can be maintaining even
when multiple block devices claim the same ZPOOL_CONFIG_PATH.
The preferred sorting by partition was never benefitial for
a Linux system and was removed as part of this change.

* When adding entries to the blkid cache avl_find/avl_insert
are used instead of avl_add because collisions are possible
and must be handled gracefully.

* For pools using multipath devices there are, at a minimum,
three devices where a vdev label may be read.  They are the
dm-* device and each underlying /dev/sd* device.  Due to the
way the block cache is implemented each of these devices may
have a different cached copy of the vdev label.  This can
result in "ghost pools" which appear to persist even after
a 'zpool labelclear' has been done to the dm-* device.  In
order to prevent this the vdev label is read with O_DIRECT
in order to bypass any caching to get the on-disk version.

* When opening a block device verify that vdev guid read from
the disk matches the expected vdev guid.  This allows for bad
labels to be filtered out.

Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes #5359
2016-11-07 10:28:57 -08:00
Brian Behlendorf
34328f3cf8 Allow 16M zio buffers in user space
Only restrict the maximum zio alloc size to 32-bit kernel space.
The same virtual address space limitations don't apply to user
space.  This resolves a memory allocation failure in raidz_test
where it expects to be able to exercises all valid zio sizes.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2016-11-07 10:26:17 -08:00
Brian Behlendorf
61f9b2cd12 Replace ISAINFO with is_32bit function
The isainfo(1) utility was used by the ZFS Test Suite to determine
when running on a 32-bit platform.  This non-portable check has been
replaced with an is_32bit helper function which uses getconf(1).
The getconf(1) utility is available for Linux, FreeBSD, and Illumos.

Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
2016-11-07 10:26:17 -08:00
Tony Hutter
1ad9de6d08 Allow autoreplace even when enclosure LED sysfs entries don't exist
The previous autoreplace code assumed that if you were using autoreplace, then
you also had the enclosure SES driver loaded.  This could lead to autoreplace
not working if the SES driver wasn't loaded, or if it wasn't creating the
proper enclosure_device symlinks (which has happened).  This patch removes
that assumption.

Reviewed by: Don Brady <don.brady@intel.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes #5363
2016-11-04 13:34:13 -07:00
Romain Dolbeau
7f3194932d Add superscalar fletcher4
This is the Fletcher4 algorithm implemented in pure C, but using
multiple counters using algorithms identical to those used for
SSE/NEON and AVX2.

This allows for faster execution on core with strong superscalar
capabilities but weak SIMD capabilities.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Romain Dolbeau <romain.dolbeau@atos.net>
Closes #5317
2016-11-04 10:53:03 -07:00
Chunwei Chen
ace1eae84c Add support for O_TMPFILE
Linux 3.11 add O_TMPFILE to open(2), which allow creating an unlinked file on
supported filesystem. It's basically doing open(2) and unlink(2) atomically.

The filesystem support is added through i_op->tmpfile. We basically copy the
create operation except we get rid of the link and name related stuff and add
the new node to unlinked set.

We also add support for linkat(2) to link tmpfile. However, since all previous
file operation will skip ZIL, we force a txg_wait_synced to make sure we are
sync safe.

Signed-off-by: Chunwei Chen <david.chen@osnexus.com>
2016-11-04 10:46:40 -07:00
Chunwei Chen
987014903f Fix unlinked file cannot do xattr operations
Currently, doing things like fsetxattr(2) on an unlinked file will result in
ENODATA. There's two places that cause this: zfs_dirent_lock and zfs_zget.

The fix in zfs_dirent_lock is pretty straightforward. In zfs_zget though, we
need it to not return error when the zp is unlinked. This is a pretty big
change in behavior, but skimming through all the callers, I don't think this
change would cause any problem. Also there's nothing preventing z_unlinked
from being set after the z_lock mutex is dropped before but before zfs_zget
returns anyway.

The rest of the stuff is to make sure we don't log xattr stuff when owner is
unlinked.

Signed-off-by: Chunwei Chen <david.chen@osnexus.com>
2016-11-04 10:46:40 -07:00
Romain Dolbeau
7f547f85fe Add parity generation/rebuild using AVX-512 for x86-64
avx512f should work on all AVX512 hardware, since it only uses
Foundation instructions.

avx512bw should be faster on hardware supporting the AVW512BW
extension. We can use full-width pshufb (instead of relying on the 256
bits AVX2 pshufb). As a side-effect, the code is also unrolled more.

Reviewed-by: Richard Laager <rlaager@wiktel.com>
Reviewed-by: Gvozden Neskovic <neskovic@gmail.com>
Reviewed-by: Jinshan Xiong <jinshan.xiong@intel.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Romain Dolbeau <romain.github@dolbeau.name>
Closes #5219
2016-11-02 12:40:23 -07:00
BearBabyLiu
6d4210052b Fix dsl_prop_get_all_dsl() memory leak
On error dsl_prop_get_all_ds() does not free the nvlist it allocates.
This behavior may have been intentional when originally written
but is atypical and often confusing.  Since no callers rely on this
behavior the function has been updated to always free the nvlist
on error.

Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: BearBabyLiu <liu.huang@zte.com.cn>
Closes #5320
2016-11-02 12:34:10 -07:00
Brian Behlendorf
e676a19624 Skip async_destroy_001_pos on 32-bit systems
The async_destroy_001_pos test case currently hangs when testing on
a 32-bit system.  Conditionally skip this test case on 32-bit
systems until the root cause is identified and resolved.

Reviewed-by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #5352
Issue #5347
2016-11-02 12:14:45 -07:00