The number of IO and checksum events should match the number of errors
seen in zpool status. Previously there was a mismatch between the
two counts because zpool status would only count unrecovered errors,
while zpool events would get an event for *all* errors (recovered or
not). This lead to situations where disks could be faulted for
"too many errors", while at the same time showing zero errors in zpool
status.
This fixes the zpool status error counters to increment at the same
times we post the error events.
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Olaf Faaland <faaland1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes#4851Closes#7817
This patch simply fixes some small memory leaks that can happen
during error handling in zfsvfs_create_impl(). If the function
fails, it frees all the memory / references it created.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8490
This patch attempts to address some user concerns that have arisen
since errata 4 was introduced.
* The errata warning has been made less scary for users without
any encrypted datasets.
* The errata warning now clears itself without a pool reimport if
the bookmark_v2 feature is enabled and no encrypted datasets
exist.
* It is no longer possible to create new encrypted datasets without
enabling the bookmark_v2 feature, thus helping to ensure that the
errata is resolved.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Issue ##8308
Closes#8504
Before sequential scrub patches ZFS never aggregated I/Os above 128KB.
Sequential scrub bumped that to 1MB, supposedly to reduce number of
head seeks for spinning disks. But for SSDs it makes little to no
sense, especially on FreeBSD, where due to MAXPHYS limitation device
will likely still see bunch of 128KB I/Os instead of one large.
Having more strict aggregation limit for SSDs allows to avoid
allocation of large memory buffer and copy to/from it, that is a
serious problem when throughput reaches gigabytes per second.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Richard Elling <Richard.Elling@RichardElling.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Closes#8494
Currently, there is an issue in the raw receive code where
raw receives are allowed to happen on top of previously
non-raw received datasets. This is a problem because the
source-side dataset doesn't know about how the blocks on
the destination were encrypted. As a result, any MAC in
the objset's checksum-of-MACs tree that is a parent of both
blocks encrypted on the source and blocks encrypted by the
destination will be incorrect. This will result in
authentication errors when we decrypt the dataset.
This patch fixes this issue by adding a new check to the
raw receive code. The code now maintains an "IVset guid",
which acts as an identifier for the set of IVs used to
encrypt a given snapshot. When a snapshot is raw received,
the destination snapshot will take this value from the
DRR_BEGIN payload. Non-raw receives and normal "zfs snap"
operations will cause ZFS to generate a new IVset guid.
When a raw incremental stream is received, ZFS will check
that the "from" IVset guid in the stream matches that of
the "from" destination snapshot. If they do not match, the
code will error out the receive, preventing the problem.
This patch requires an on-disk format change to add the
IVset guids to snapshots and bookmarks. As a result, this
patch has errata handling and a tunable to help affected
users resolve the issue with as little interruption as
possible.
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8308
This patch adds the bookmark v2 feature to the on-disk format. This
feature will be needed for the upcoming redacted sends and for an
upcoming fix that for raw receives. The feature is not currently
used by any code and thus this change is a no-op, aside from the
fact that the user can now enable the feature.
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Issue #8308
Currently, the receive code can create an unreadable dataset from
a correct raw send stream. This is because it is currently
impossible to set maxblkid to a lower value without freeing the
associated object. This means truncating files on the send side
to a non-0 size could result in corruption. This patch solves this
issue by adding a new 'force' flag to dnode_new_blkid() which will
allow the raw receive code to force the DMU to accept the provided
maxblkid even if it is a lower value than the existing one.
For testing purposes the send_encrypted_files.ksh test has been
extended to include a variety of truncated files and multiple
snapshots. It also now leverages the xattrtest command to help
ensure raw receives correctly handle xattrs.
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8168Closes#8487
Most of the zfs_arc_* module parameters do not have their values used by
the ARC code directly. Instead, there is a function, arc_tuning_update,
which is called during module initialization and periodically
thereafter, whose job is to fetch the module parameter values, clamp/
limit them appropriately, and then assign those values to a separate set
of internal variables that are actually referenced by the ARC code.
Commit 3ec34e55 featured an overhaul of arc_reclaim_thread, which is the
former location where the post-init-time calls to arc_tuning_update
would occur. The rework split the work previously done by the
arc_reclaim_thread into a pair of replacement threads; and
unfortunately, the call to arc_tuning_update fell through the cracks and
was lost in the reorganization.
This meant that changing almost any ARC-related zfs module parameter via
/sys/module/zfs/parameters/ would result in the module parameter value
itself appearing to change; however the modification would not actually
propagate to the ARC code and have any real effect.
This commit reinstates the post-init-time call to arc_tuning_update. It
is now called during arc_adjust_cb_check; this should be equivalent to
its former call location in arc_reclaim_thread.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Justin Gottula <justin@jgottula.com>
Closes#8405Closes#8463
This patch modifies the zfs_ioc_snapshot_list_next() ioctl to enable it
to take input parameters that alter the way looping through the list of
snapshots is performed. The idea here is to restrict functions that
throw away some of the snapshots returned by the ioctl to a range of
snapshots that these functions actually use. This improves efficiency
and execution speed for some rollback and send operations.
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Signed-off-by: Alek Pinchuk <apinchuk@datto.com>
Closes#8077
Resolve a vdev_initialize crash uncovered by ztest. Similar
to when starting a new initialization verify that a removal
is not in progress. Additionally, do not restart when the
thread already exists. This check is now congruent with the
POOL_INITIALIZE_DO handling in spa_vdev_initialize_impl().
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8477
Instead of choosing a leaf vdev quasi-randomly, by starting at the root
vdev and randomly choosing children, rotate over leaves to issue MMP
writes. This fixes an issue in a pool whose top-level vdevs have
different numbers of leaves.
The issue is that the frequency at which individual leaves are chosen
for MMP writes is based not on the total number of leaves but based on
how many siblings the leaves have.
For example, in a pool like this:
root-vdev
+------+---------------+
vdev1 vdev2
| |
| +------+-----+-----+----+
disk1 disk2 disk3 disk4 disk5 disk6
vdev1 and vdev2 will each be chosen 50% of the time. Every time vdev1
is chosen, disk1 will be chosen. However, every time vdev2 is chosen,
disk2 is chosen 20% of the time. As a result, disk1 will be sent 5x as
many MMP writes as disk2.
This may create wear issues in the case of SSDs. It also reduces the
effectiveness of MMP as it depends on the writes being evenly
distributed for the case where some devices fail or are partitioned.
The new code maintains a list of leaf vdevs in the pool. MMP records
the last leaf used for an MMP write in mmp->mmp_last_leaf. To choose
the next leaf, MMP starts at mmp->mmp_last_leaf and traverses the list,
continuing from the head if the tail is reached. It stops when a
suitable leaf is found or all leaves have been examined.
Added a test to verify MMP write distribution is even.
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: Kash Pande <kash@tripleback.net>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Signed-off-by: Olaf Faaland <faaland1@llnl.gov>
Closes#7953
The linux kernel's nfsd implementation use RWF_SYNC to determine if the
write is synchronous or not. This flag is used to set the kernel's I/O
control block flags. Unfortunately, ZFS was not updated to inspect these
flags so NFS sync writes were not being honored.
This change maps the IOCB_* flags to the ZFS equivalent.
Reviewed-by: Don Brady <don.brady@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: George Wilson <george.wilson@delphix.com>
Closes#8474Closes#8452Closes#8486
Commit torvalds/linux@736706bee has removed the get_fs() function
as a bit of cleanup. It has been defined as KERNEL_DS on all
architectures for all supported kernels. Replace get_fs() with
KERNEL_DS as was done in the kernel.
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8479
The function bpobj_iterate_impl overflows the stack when bpobjs
are deeply nested. Rewrite the function to eliminate the recursion.
Reviewed-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Paul Zuchowski <pzuchowski@datto.com>
Closes#7674Closes#7675Closes#7908
Before allowing new allocations to the metaslab we need to ensure
that any issued initializing writes have been synced. Otherwise,
it's possible for metaslab_block_alloc() to allocate a range which
is about to be overwritten by an initializing IO.
Serapheim Dimitropoulos <serapheim@delphix.com>
Reviewed-by: Richard Elling <Richard.Elling@RichardElling.com>
Reviewed-by: Tim Chase <tim@chase2k.com>
Reviewed-by: George Wilson <george.wilson@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8461
Fix indentation of code in ifdef's.
Remove obsolete comment.
Make if/else statements more readable by adding braces.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Matthew Ahrens <mahrens@delphix.com>
Closes#8459
When multihost is enabled, and a pool is suspended, return
EINVAL in response to "zpool clear <pool>". The pool
may have been imported on another host while I/O was suspended.
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Olaf Faaland <faaland1@llnl.gov>
Closes#6933Closes#8460
abd_alloc() normally does scatter allocations, thus solving the problem
that ABD originally set out to: the bulk of ZFS's allocations are single
pages, which are faster to allocate and free, and don't suffer from
internal fragmentation (and the inability to reclaim memory because some
buffers in the slab are still allocated).
However, the current code does linear allocations for 4KB and smaller
allocations, defeating the purpose of ABD.
Scatter ABD's use at least one page each, so sub-page allocations waste
some space when allocated as scatter (e.g. 2KB scatter allocation wastes
half of each page). Using linear ABD's for small allocations means that
they will be put on slabs which contain many allocations. This can
improve memory efficiency, but it also makes it much harder for ARC
evictions to actually free pages, because all the buffers on one slab
need to be freed in order for the slab (and underlying pages) to be
freed. Typically, 512B and 1KB kmem caches have 16 buffers per slab, so
it's possible for them to actually waste more memory than scatter (one
page per buf = wasting 3/4 or 7/8th; one buf per slab = wasting
15/16th).
Spill blocks are typically 512B and are heavily used on systems running
selinux with the default dnode size and the `xattr=sa` property set.
By default we will use linear allocations for 512B and 1KB, and scatter
allocations for larger (1.5KB and up).
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: DHE <git@dehacked.net>
Reviewed-by: Chunwei Chen <tuxoko@gmail.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Don Brady <don.brady@delphix.com>
Signed-off-by: Matthew Ahrens <mahrens@delphix.com>
Closes#8455
The spa_txg_history_init_io() and spa_txg_history_fini_io() were
mistakenly taking SCL_ALL when only SCL_CONFIG is required to
access the vdev stats. This could result in a deadlock which
was observed when running ztest.
Reviewed-by: Olaf Faaland <faaland1@llnl.gov>
Reviewed-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8445
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <guss80@gmail.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#8444
The issue is caused by a small discrepancy in how userland creates the
partition layout and the kernel estimates available space:
* zpool command: subtract 9M from the usable device size, then align
to 1M boundary. 9M is the sum of 1M "start" partition alignment + 8M
EFI "reserved" partition.
* kernel module: subtract 10M from the device size. 10M is the sum of
1M "start" partition alignment + 1m "end" partition alignment + 8M
EFI "reserved" partition.
For devices where the number of sectors is not a multiple of the
alignment size the zpool command will create a partition layout which
reserves less than 1M after the 8M EFI "reserved" partition:
Disk /dev/sda: 1024 MiB, 1073739776 bytes, 2097148 sectors
Units: sectors of 1 * 512 = 512 bytes
Sector size (logical/physical): 512 bytes / 512 bytes
I/O size (minimum/optimal): 512 bytes / 512 bytes
Disklabel type: gpt
Disk identifier: 49811D40-16F4-4E41-84A9-387703950D7F
Device Start End Sectors Size Type
/dev/sda1 2048 2078719 2076672 1014M Solaris /usr & Apple ZFS
/dev/sda9 2078720 2095103 16384 8M Solaris reserved 1
When the kernel module vdev_open() the device its max_asize ends up
being slightly smaller than asize: this results in a huge number (16E)
reported by metaslab_class_expandable_space().
This change prevents bdev_max_capacity() from returing a size smaller
than bdev_capacity().
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Wilson <george.wilson@delphix.com>
Reviewed by: Sara Hartse <sara.hartse@delphix.com>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes#1468Closes#8391
Soft lockups could happen when multiple threads trying
to get zrl on the same dnode handle in order to allocate
and initialize the dnode marked as DN_SLOT_ALLOCATED.
Don't loop from beginning when we can't get zrl, otherwise
we would increase the zrl refcount and nobody can actually
lock it.
Reviewed by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Li Dongyang <dongyangli@ddn.com>
Closes#8433
The SCST driver (SCSI target driver implementation) and possibly
others may issue read bio's with a length of zero bytes. Although
this is unusual, such bio's issued under certain condition can cause
kernel oops, due to how rangelock is implemented.
rangelock_add_reader() is not made to handle overlap of two (or more)
ranges from read bio's with the same offset when one of them has size
of 0, even though they conceptually overlap. Allowing them to enter
rangelock results in kernel oops by dereferencing invalid pointer,
or assertion failure on AVL tree manipulation with debug enabled
kernel module.
For example, this happens when read bio whose (offset, size) is
(0, 0) enters rangelock followed by another read bio with (0, 4096)
when (0, 0) rangelock is still locked, when there are no pending
write bio's. It can also happen with reverse order, which is (0, N)
followed by (0, 0) when (0, N) is still locked. More details
mentioned in #8379.
Kernel Oops on ->make_request_fn() of ZFS volume
https://github.com/zfsonlinux/zfs/issues/8379
Prevent this by returning bio with size 0 as success without entering
rangelock. This has been done for write bio after checking flusher
bio case (though not for the same reason), but not for read bio.
Reviewed-by: Alek Pinchuk <apinchuk@datto.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tomohiro Kusumi <kusumi.tomohiro@osnexus.com>
Closes#8379Closes#8401
This patch introduces 3 new histograms per metaslab. These
histograms track segments that have made it to the metaslab's
space map histogram (and are part of the spacemap) but have
not yet reached the ms_allocatable tree on loaded metaslab's
because these metaslab's are currently syncing and haven't
gone through metaslab_sync_done() yet.
The histograms help when we decide whether to load an unloaded
metaslab in-order to allocate from it. When calculating the
weight of an unloaded metaslab traditionally, we look at the
highest bucket of its spacemap's histogram. The problem is
that we are not guaranteed to be able to allocated that
segment when we load the metaslab because it may still be at
the freeing, freed, or defer trees. The new histograms are
used when we try to calculate an unloaded metaslab's weight
to deal with this issue by removing segments that have would
not be in the allocatable tree at runtime. Note, that this
method of dealing with this is not completely accurate as
adjacent segments are not always consolidated in the space
map histogram of a metaslab.
In addition and to make things deterministic, we always reset
the weight of unloaded metaslabs based on their space map
weight (instead of doing that on a need basis). Thus, every
time a metaslab is loaded and its weight is reset again (from
the weight based on its space map to the one based on its
allocatable range tree) we expect (and assert) that this
change in weight can only get better if it doesn't stay the
same.
Reviewed by: Paul Dagnelie <pcd@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#8358
Trying to mount a dataset from a readonly pool could inadvertently start
the user accounting upgrade task, leading to the following failure:
VERIFY3(tx->tx_threads == 2) failed (0 == 2)
PANIC at txg.c:680:txg_wait_synced()
Showing stack for process 2541
CPU: 2 PID: 2541 Comm: z_upgrade Tainted: P O 3.16.0-4-amd64 #1 Debian 3.16.51-3
Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
Call Trace:
[<0>] ? dump_stack+0x5d/0x78
[<0>] ? spl_panic+0xc9/0x110 [spl]
[<0>] ? dnode_next_offset+0x1d4/0x2c0 [zfs]
[<0>] ? dmu_object_next+0x77/0x130 [zfs]
[<0>] ? dnode_rele_and_unlock+0x4d/0x120 [zfs]
[<0>] ? txg_wait_synced+0x91/0x220 [zfs]
[<0>] ? dmu_objset_id_quota_upgrade_cb+0x10f/0x140 [zfs]
[<0>] ? dmu_objset_upgrade_task_cb+0xe3/0x170 [zfs]
[<0>] ? taskq_thread+0x2cc/0x5d0 [spl]
[<0>] ? wake_up_state+0x10/0x10
[<0>] ? taskq_thread_should_stop.part.3+0x70/0x70 [spl]
[<0>] ? kthread+0xbd/0xe0
[<0>] ? kthread_create_on_node+0x180/0x180
[<0>] ? ret_from_fork+0x58/0x90
[<0>] ? kthread_create_on_node+0x180/0x180
This patch updates both functions responsible for checking if we can
perform user accounting to verify the pool is not readonly.
Reviewed-by: Alek Pinchuk <apinchuk@datto.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes#8424
If we hit the (NSEC_TO_TICK(diff) == 0) condition in
zio_delay_interrupt, zio_interrupt is never called and the
zio does not progress.
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: sara hartse <sara.hartse@delphix.com>
Closes#8404
Add the zio_deadman_log_all tunable to print all zios in
zio_deadman_impl(). Also, in all cases, display the depth of the
zio relative to the original parent zio. This is meant to be used by
developers to gain diagnostic information for hangs which don't involve
fully set-up zio trees or are otherwise stuck or hung in an early stage.
Reviewed-by: Olaf Faaland <faaland1@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Signed-off-by: Tim Chase <tim@chase2k.com>
Closes#8362
Add -h switch to zfs send command to send dataset holds. If
holds are present in the stream, zfs receive will create them
on the target dataset, unless the zfs receive -h option is used
to skip receive of holds.
Reviewed-by: Alek Pinchuk <apinchuk@datto.com>
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed by: Paul Dagnelie <pcd@delphix.com>
Signed-off-by: Paul Zuchowski <pzuchowski@datto.com>
Closes#7513
5d43cc9a59 renamed it to rangelock_enter().
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tomohiro Kusumi <kusumi.tomohiro@osnexus.com>
Closes#8408
Deletion throttle currently does not account for holes in a file.
This means that it can activate when it shouldn't.
To fix it we switch the throttle to be based on the number of
L1 blocks we will have to dirty when freeing
Reviewed by: Tom Caputi <tcaputi@datto.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alek Pinchuk <apinchuk@datto.com>
Closes#7725Closes#7888
This patch is an async implementation of the existing sync
zfs_unlinked_drain() function. This function is called at mount time and
is responsible for freeing znodes that we didn't get to freeing before.
We don't have to hold mounting of the dataset until the unlinked list is
fully drained as is done now. Since we can process the unlinked set
asynchronously this results in a better user experience when mounting a
dataset with entries in the unlinked set.
Reviewed by: Jorgen Lundman <lundman@lundman.net>
Reviewed by: Tom Caputi <tcaputi@datto.com>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: Paul Dagnelie <pcd@delphix.com>
Signed-off-by: Alek Pinchuk <apinchuk@datto.com>
Closes#8142
Initially, metaslabs and space maps used to be the same thing
in ZFS. Later, we started differentiating them by referring
to the space map as the on-disk state of the metaslab, making
the metaslab a higher-level concept that is metadata that deals
with space accounting. Today we've managed to split that code
furthermore, with the space map being its own on-disk data
structure used in areas of ZFS besides metaslabs (e.g. the
vdev-wide space maps used for zpool checkpoint or vdev removal
features).
This patch refactors the space map code to further split the
space map code from the metaslab code. It does so by getting
rid of the idea that the space map can have a different in-core
and on-disk length (sm_length vs smp_length) which is something
that is only used for the metaslab code, and other consumers
of space maps just have to deal with. Instead, this patch
introduces changes that move the old in-core length of the
metaslab's space map to the metaslab structure itself (see
ms_synced_length field) while making the space map code only
care about the actual space map's length on-disk.
The result of this is that space map consumers no longer have
to deal with syncing two different lengths for the same
structure (e.g. space_map_update() goes away) while metaslab
specific behavior stays within the metaslab code. Specifically,
the ms_synced_length field keeps track of the amount of data
metaslab_load() can read from the metaslab's space map while
working concurrently with metaslab_sync() that may be
appending to that same space map.
As a side note, the patch also adds a few comments around
the metaslab code documenting some assumptions and expected
behavior.
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#8328
zfs create, receive and rename can bypass this hierarchy rule. Update
both userland and kernel module to prevent this issue and use pyzfs
unit tests to exercise the ioctls directly.
Note: this commit slightly changes zfs_ioc_create() ABI. This allow to
differentiate a generic error (EINVAL) from the specific case where we
tried to create a dataset below a ZVOL (ZFS_ERR_WRONG_PARENT).
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Due to an off-by-one condition in spa_preferred_class() we are picking
the "normal" allocation class instead of the "special" one for file
blocks with size equal to the special_small_blocks property value.
This change fix the small code issue, update the ZFS Test Suite and the
zfs(8) man page.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Don Brady <don.brady@delphix.com>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes#8351Closes#8361
Re-factor arc_read() to better account for embedded data blkptrs.
Previously, reading the payload from an embedded blkptr would cause
arcstats such as demand_metadata_misses to be bumped when there was
actually no cache "miss" because the data are already available in
the blkptr.
The following test procedure was used to demonstrate the problem:
zpool create tank ...
zfs create -o compression=lz4 tank/fs
echo blah > /tank/fs/blah
stat /tank/fs/blah
grep 'meta.*mis' /proc/spl/kstat/zfs/arcstats
and repeating the last two steps to watch the metadata miss counter
increment. This can also be demonstrated via the zfs_arc_miss DTRACE4
probe in arc_read().
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Reviewed-by: George Wilson <george.wilson@delphix.com>
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Tim Chase <tim@chase2k.com>
Closes#8319
Get rid of the majority metaslab metadata when removing log vdevs
in spa_vdev_remove_log() with a call to metaslab_fini() instead
of duplicating a lot of that in vdev_remove_empty_log().
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#8347
The current L2 ARC device code consistently uses psize to
increment vs_alloc but varies between psize and lsize when
decrementing it. The result of this behavior is that
vs_alloc can be decremented more that it is incremented
and underflow. This patch changes the code so asize is
used anywhere.
In addition, it ensures that vs_alloc gets incremented by
the L2 ARC device code as buffers are written and not at
the end of the l2arc_write_buffers() routine. The latter
(and old) way would temporarily underflow vs_alloc as
buffers that were just written, would be destroyed while
l2arc_write_buffers() was still looping.
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#8298
Address a deadlock caused by simultaneous wakeup and cancel on a zthr
by remove the hold of zthr_request_lock from zthr_wakeup. This
allows thr_wakeup to not block a thread that is in the process of
being cancelled.
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Signed-off-by: Sara Hartse <sara.hartse@delphix.com>
Closes#8333
The Linux 5.0 kernel updated the bio_set_dev() macro so it calls the
GPL-only bio_associate_blkg() symbol thus inadvertently converting
the entire macro. Provide a minimal version which always assigns the
request queue's root_blkg to the bio.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8287
SUBDIRs has been deprecated for a long time, and was finally removed in
the 5.0 kernel. Use "M=" instead.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes#8257
In the 5.0 kernel, only the mount namespace code should use the MS_*
macos. Filesystems should use the SB_* ones.
https://patchwork.kernel.org/patch/10552493/
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes#8264
totalram_pages() was converted to an atomic variable in 5.0:
https://patchwork.kernel.org/patch/10652795/
Its value should now be read though the totalram_pages() helper
function.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes#8263
access_ok no longer needs a 'type' parameter in the 5.0 kernel.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes#8261
= Old behavior
For vdev sizes 100GB to 50TB we keep ~200 metaslabs per
vdev and the metaslab size grows from 512MB to 256GB.
For vdev's bigger than that we start increasing the
number of metaslabs until we hit the 128K limit.
= New Behavior
For vdev sizes 100GB to 3TB we keep ~200 metaslabs per
vdev and the metaslab size grows from 512MB to 16GB.
For vdev's bigger than that we start increasing the
number of metaslabs until we hit the 128K limit.
= Reasoning
The old behavior makes metaslabs grow in size when
the vdev range is between 3TB (ms_size 16GB) and
32PB (ms_size 256GB). Even though keeping the number
of metaslabs is good in terms of potential number of
I/Os per TXG, these bigger metaslabs take longer
to be loaded and after they are loaded they can
take up a lot of memory because of their range trees.
This change tries to put a boundary in memory and
loading time for the specific range of vdev sizes.
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Don Brady <don.brady@delphix.com>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#8324
The range_tree_verify function looks for a segment in a
range tree and panics if the segment is present on the
tree. This patch gives the function a more descriptive
name.
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#8327
This allows the spa config refcounts to use tracking in debug builds
without triggering the "No such hold %p on refcount" panic.
Reviewed-by: Olaf Faaland <faaland1@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tim Chase <tim@chase2k.com>
Closes#8326
Currently, zvol_rename_minors_impl() calls kmem_asprintf()
to allocate and initialize a string. This function is a thin
wrapper around the kernel's kvasprintf() and does not call
into the SPL's kmem tracking code when it is enabled. However,
this function frees the string with the tracked kmem_free()
instead of the untracked strfree(), which causes the SPL
kmem tracking code to believe that the function is attempting
to free memory it never allocated, triggering an ASSERT. This
patch simply corrects this issue.
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8307
Since d8fdfc2 was integrated dsl_pool_create() does not call
dmu_objset_create_impl() for the root dataset when running in
userland (ztest): this creates a pool with a partially initialized
root dataset. Trying to import and use this pool results in both
zpool and zfs executables dumping core.
Fix this by adopting an alternative change suggested in OpenZFS 8607
code review.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed by: Tom Caputi <tcaputi@datto.com>
Original-patch-by: Robert Mustacchi <rm@joyent.com>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes#8277
This check provides no real additional protection and unnecessarily
introduces a dependency on the "oops_in_progress" kernel symbol.
Remove the check, it there are special circumstances on other
platforms which make this a requirement it can be reintroduced
for all relevant call paths in a more portable comprehensive manor.
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: George Melikov <mail@gmelikov.ru>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8297
Most callers that need to operate on a loaded metaslab, always
call metaslab_load_wait() before loading the metaslab just in
case someone else is already doing the work.
Factoring metaslab_load_wait() within metaslab_load() makes the
later more robust, as callers won't have to do the load-wait
check explicitly every time they need to load a metaslab.
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#8290
Currently, when a DRR_OBJECT record is read into memory in
receive_read_record(), memory is allocated for the bonus buffer.
However, if the object doesn't have a bonus buffer the code will
still "allocate" the zero bytes, but the memory will not be passed
to the processing thread for cleanup later. This causes the spl
kmem tracking code to report a leak. This patch simply changes the
code so that it only allocates this memory if it has a non-zero
length.
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8266
The point of this refactoring is to break the high-level conceptual
steps of spa_sync() to their own helper functions. In general large
functions can enhance readability if structured well, but in this
case the amount of conceptual steps taken could use the help of
helper functions.
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#8293
Currently, the functions dbuf_prefetch_indirect_done() and
dmu_assign_arcbuf_by_dnode() assume that dbuf_hold_level() cannot
fail. In the event of an error the former will cause a NULL pointer
dereference and the later will trigger a VERIFY. This patch adds
error handling to these functions and their callers where necessary.
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8291
The following fields from the vdev_t struct are not used anywhere.
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#8285
The ztest_ddt_repair() test is designed inflict damage to the
ddt which can be repairable by a scrub. Unfortunately, this
repair logic was broken at some point and it went undetected.
This issue is not specific to ztest, but thankfully this extra
redundancy is rarely enabled and even more rarely needed.
The root cause was identified to be the ddt_bp_create()
function called by dsl_scan_ddt_entry() which did not set the
dedup bit of the generated block pointer.
The consequence of this was that the ZIO_DDT_READ_PIPELINE was
never enabled for the block pointer during the scrub, and the
dedup ditto repair logic was never run. Note that for demand
reads which don't rely on ddt_bp_create() the required pipeline
stages would be enabled and the repair performed.
This was resolved by unconditionally setting the dedup bit in
ddt_bp_create(). This way all codes paths which may need to
perform a repair from a block pointer generated from the dtt
entry will be able too. The only exception is that the dedup
bit is cleared in ddt_phys_free() which is required to avoid
leaking space.
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: Tom Caputi <tcaputi@datto.com>
Reviewed by: Serapheim Dimitropoulos <serapheim@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8270
Since the new spacemap encoding was ported to ZoL that's no longer
a limitation. This patch updates vdev_is_spacemap_addressable()
that was performing that check.
It also updates the appropriate test to ensure that the same
functionality is tested. The test does so by creating pools that
don't have the new spacemap encoding enabled - just the checkpoint
feature. This patch also reorganizes that same tests in order to
cut in half its memory consumption.
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#8286
Increase the default allowed number of reconstruction attempts.
There's not an exact right number for this setting. It needs
to be set large enough to cover any realistic failure scenarios
and small enough to avoid stalling the IO pipeline and invoking
the dead man detection.
The current value of 256 was empirically determined to be too
low based on multi-day runs of ztest. The fault injection code
would inject more damage than could be reconstructed given the
relatively small number of attempts. However, in all observed
cases the block could be reconstructed using a slightly higher
limit.
Based on local testing increasing the default value to 4096 was
determined to strike the best balance. Checking all combinations
takes less than 10s in the worst case, and has so far eliminated
the vast majority of false positives detected by ztest. This
delay is roughly on par with how long retries may be performed
to a misbehaving HDD and was deemed to be reasonable. Better to
err on the side of a brief delay rather than fail to reconstruct
the data.
Lastly, the -Y flag has been added to zdb to make it easy to try all
possible combinations when performing split block reconstruction.
For badly damaged blocks with 18 splits, they can be fully enumerated
within a few minutes. This has been done to ensure permanent errors
are never incorrectly reported when ztest verifies the pool with zdb.
Reviewed by: Tom Caputi <tcaputi@datto.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: Serapheim Dimitropoulos <serapheim@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8271
Currently, dbuf_read() may decide to create a zio_root which is
used as a parent for any child zios created in dbuf_read_impl().
However, if there is an error in dbuf_read_impl(), this zio is
never executed and ends up leaked. This patch simply ensures
that we always execute the root zio, even i it has no real work
to do.
Reviewed-by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8267
Some minor spelling mistakes and typos. No functional changes.
Reviewed-by: Neal Gompa <ngompa@datto.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: Giuseppe Di Natale <guss80@gmail.com>
Reviewed-by: bunder2015 <omfgbunder@gmail.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8272
Adds a new lock for serializing operations on zthrs.
The commit also includes some code cleanup and
refactoring.
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Closes#8229
On full pool when pool root filesystem references very few bytes,
the f_blocks returned to statvfs is 0 but should be at least 1.
Reviewed by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Paul Zuchowski <pzuchowski@datto.com>
Closes#8253Closes#8254
Object allocation performance can be improved for complex operations
by providing an interface which returns the newly allocated dnode.
This allows the caller to immediately use the dnode without incurring
the expense of looking up the dnode by object number.
The functions dmu_object_alloc_hold(), zap_create_hold(), and
dmu_bonus_hold_by_dnode() were added for this purpose.
The zap_create_* functions have been updated to take advantage of
this new functionality. The dmu_bonus_hold_impl() function should
really have never been included in sys/dmu.h and was removed.
It's sole caller was converted to use dmu_bonus_hold_by_dnode().
The new symbols have been exported for use by Lustre.
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8015
This patch simply fixes a small bug where dnode_hold_impl() could
attempt to allocate a dnode that was in the process of being freed,
but which still had active references. This patch simply adds the
required check.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8249
This commit fixes a small issue which causes both zfs receive and
rollback operations to incorrectly increase the "filesystem_count"
property value.
This change also adds a new test group "limits" to the ZFS Test Suite
to exercise both filesystem_count/limit and snapshot_count/limit
functionality.
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes#8232
Scrubbing is supposed to detect and repair all errors in the pool.
However, it wrongly ignores active spare devices. The problem can
easily be reproduced in OpenZFS at git rev 0ef125d with these
commands:
truncate -s 64m /tmp/a /tmp/b /tmp/c
sudo zpool create testpool mirror /tmp/a /tmp/b spare /tmp/c
sudo zpool replace testpool /tmp/a /tmp/c
/bin/dd if=/dev/zero bs=1024k count=63 oseek=1 conv=notrunc of=/tmp/c
sync
sudo zpool scrub testpool
zpool status testpool # Will show 0 errors, which is wrong
sudo zpool offline testpool /tmp/a
sudo zpool scrub testpool
zpool status testpool # Will show errors on /tmp/c,
# which should've already been fixed
FreeBSD head is partially affected: the first scrub will detect
some errors, but the second scrub will detect more. This same
test was run on Linux before applying the fix and the FreeBSD
head behavior was observed.
Authored by: asomers <asomers@FreeBSD.org>
Reviewed by: Andy Stormont <astormont@racktopsystems.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: George Wilson <george.wilson@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Approved by: Richard Lowe <richlowe@richlowe.net>
Ported-by: Brian Behlendorf <behlendorf1@llnl.gov>
Sponsored by: Spectra Logic Corp
OpenZFS-issue: https://www.illumos.org/issues/8473
FreeBSD-commit: https://github.com/freebsd/freebsd/commit/e20ec8879
OpenZFS-commit: https://github.com/illumos/illumos-gate/commit/554675eeCloses#8251
PROBLEM
========
When invoking "zpool initialize" on a pool the command will
create a thread to initialize each disk. Unfortunately, it does
this serially across many transaction groups which can result
in commands taking a long time to return to the user and may
appear hung. The same thing is true when trying to suspend/cancel
the operation.
SOLUTION
=========
This change refactors the way we invoke the initialize interface
to ensure we can start or stop the intialization in just a few
transaction groups.
When stopping or cancelling a vdev initialization perform it
in two phases. First signal each vdev initialization thread
that it should exit, then after all threads have been signaled
wait for them to exit.
On a pool with 40 leaf vdevs this reduces the vdev initialize
stop/cancel time from ~10 minutes to under a second. The reason
for this is spa_vdev_initialize() no longer needs to wait on
multiple full TXGs per leaf vdev being stopped.
This commit additionally adds some missing checks for the passed
"initialize_vdevs" input nvlist. The contents of the user provided
input "initialize_vdevs" nvlist must be validated to ensure all
values are uint64s. This is done in zfs_ioc_pool_initialize() in
order to keep all of these checks in a single location.
Updated the innvl and outnvl comments to match the formatting used
for all other new sytle ioctls.
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed-by: loli10K <ezomori.nozomu@gmail.com>
Reviewed-by: Tim Chase <tim@chase2k.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: George Wilson <george.wilson@delphix.com>
Closes#8230
PROBLEM
========
The first access to a block incurs a performance penalty on some platforms
(e.g. AWS's EBS, VMware VMDKs). Therefore we recommend that volumes are
"thick provisioned", where supported by the platform (VMware). This can
create a large delay in getting a new virtual machines up and running (or
adding storage to an existing Engine). If the thick provision step is
omitted, write performance will be suboptimal until all blocks on the LUN
have been written.
SOLUTION
=========
This feature introduces a way to 'initialize' the disks at install or in the
background to make sure we don't incur this first read penalty.
When an entire LUN is added to ZFS, we make all space available immediately,
and allow ZFS to find unallocated space and zero it out. This works with
concurrent writes to arbitrary offsets, ensuring that we don't zero out
something that has been (or is in the middle of being) written. This scheme
can also be applied to existing pools (affecting only free regions on the
vdev). Detailed design:
- new subcommand:zpool initialize [-cs] <pool> [<vdev> ...]
- start, suspend, or cancel initialization
- Creates new open-context thread for each vdev
- Thread iterates through all metaslabs in this vdev
- Each metaslab:
- select a metaslab
- load the metaslab
- mark the metaslab as being zeroed
- walk all free ranges within that metaslab and translate
them to ranges on the leaf vdev
- issue a "zeroing" I/O on the leaf vdev that corresponds to
a free range on the metaslab we're working on
- continue until all free ranges for this metaslab have been
"zeroed"
- reset/unmark the metaslab being zeroed
- if more metaslabs exist, then repeat above tasks.
- if no more metaslabs, then we're done.
- progress for the initialization is stored on-disk in the vdev’s
leaf zap object. The following information is stored:
- the last offset that has been initialized
- the state of the initialization process (i.e. active,
suspended, or canceled)
- the start time for the initialization
- progress is reported via the zpool status command and shows
information for each of the vdevs that are initializing
Porting notes:
- Added zfs_initialize_value module parameter to set the pattern
written by "zpool initialize".
- Added zfs_vdev_{initializing,removal}_{min,max}_active module options.
Authored by: George Wilson <george.wilson@delphix.com>
Reviewed by: John Wren Kennedy <john.kennedy@delphix.com>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com>
Reviewed by: Prakash Surya <prakash.surya@delphix.com>
Reviewed by: loli10K <ezomori.nozomu@gmail.com>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Approved by: Richard Lowe <richlowe@richlowe.net>
Signed-off-by: Tim Chase <tim@chase2k.com>
Ported-by: Tim Chase <tim@chase2k.com>
OpenZFS-issue: https://www.illumos.org/issues/9102
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/c3963210ebCloses#8230
This change is required to ease the transition when upgrading
from 0.7.x to 0.8.x. It allows 0.8.x user space utilities to
remain compatible with 0.7.x and older kernel modules.
Reviewed-by: Don Brady <don.brady@delphix.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8231
The dmu_objset_remap_indirects_impl() logic depends on dnode_hold()
returning ENOENT for dnodes which will be freed and should be skipped.
This behavior can only be relied upon when taking a new hold and
while the caller has an open transaction. This ensures that the
open txg cannot advance and that a concurrent free will end up
in the same txg (which is critical). Relying on an existing hold
will not prevent dnode_free() from succeeding.
The solution is to take an additional dnode_hold() after assigning
the transaction. This ensures the remap will never dirty the dnode
if it was freed while we were waiting in dmu_tx_assign(, TXG_WAIT).
Randomly set zfs_object_remap_one_indirect_delay_ms in ztest. This
increases the likelihood of an operation racing with the remap.
Converted from ticks to milliseconds.
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: Tom Caputi <tcaputi@datto.com>
Reviewed by: Igor Kozhukhov <igor@dilos.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8215
Following the fix for 9018 (Replace kmem_cache_reap_now() with
kmem_cache_reap_soon), the arc_reclaim_thread() no longer blocks
while reaping. However, the code is still confusing and error-prone,
because this thread has two responsibilities. We should instead
separate this into two threads each with their own responsibility:
1. keep `arc_size` under `arc_c`, by calling `arc_adjust()`, which
improves `arc_is_overflowing()`
2. keep enough free memory in the system, by calling
`arc_kmem_reap_now()` plus `arc_shrink()`, which improves
`arc_available_memory()`.
Furthermore, we can use the zthr infrastructure to separate the
"should we do something" from "do it" parts of the logic, and
normalize the start up / shut down of the threads.
Authored by: Brad Lewis <brad.lewis@delphix.com>
Reviewed by: Matt Ahrens <mahrens@delphix.com>
Reviewed by: Serapheim Dimitropoulos <serapheim@delphix.com>
Reviewed by: Pavel Zakharov <pavel.zakharov@delphix.com>
Reviewed by: Dan Kimmel <dan.kimmel@delphix.com>
Reviewed by: Paul Dagnelie <pcd@delphix.com>
Reviewed by: Dan McDonald <danmcd@joyent.com>
Reviewed by: Tim Kordas <tim.kordas@joyent.com>
Reviewed by: Tim Chase <tim@chase2k.com>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Ported-by: Brad Lewis <brad.lewis@delphix.com>
Signed-off-by: Brad Lewis <brad.lewis@delphix.com>
OpenZFS-issue: https://www.illumos.org/issues/9284
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/de753e34f9Closes#8165
In dfbe2675 zfs_dirty_data_sync was changed to a new tunable named
zfs_dirty_data_sync_percent. Unfortunately, the module parameter
documentation is the code was not updated accordingly. This patch
simply corrects that.
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8212
This patch simply removes an invalid assert from the zap_update()
function. The ASSERT is invalid because it does not hold the zap
lock from the time it fetches the old value to the time it confirms
that it is what it should be.
Reviewed by: Matt Ahrens <matt@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8209
Porting Notes:
* Additional changes to recv_rename_impl() were required due to
encryption code not being merged in OpenZFS yet.
* libzfs_core python bindings (pyzfs) were updated to fully support
both lzc_rename() and lzc_destroy()
Authored by: Andriy Gapon <avg@FreeBSD.org>
Reviewed by: Andy Stormont <astormont@racktopsystems.com>
Reviewed by: Matt Ahrens <matt@delphix.com>
Reviewed by: Serapheim Dimitropoulos <serapheim.dimitro@delphix.com>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Approved by: Dan McDonald <danmcd@joyent.com>
Ported-by: loli10K <ezomori.nozomu@gmail.com>
OpenZFS-issue: https://www.illumos.org/issues/9630
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/049ba63Closes#8207
This patch addresses an issue found in ztest where resilver
write zios that were passed to an indirect vdev would end up
being handled as though they were resilver read zios. This
caused issues where the zio->io_abd would be both read to
and written from at the same time, causing asserts to fail.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed by: Matt Ahrens <matt@delphix.com>
Reviewed-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8193
Linux kstat IO and TIMER printed values as signed. However the counters
only increment. Thus humans looking at the data can be confused when
the counters roll over.
Note: The recommended use of these values is to monitor the derivative,
which don't really care about the sign. See explanations related to
non-negative derivatives in the various time-series databases.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <guss80@gmail.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Richard Elling <Richard.Elling@RichardElling.com>
Closes#8131Closes#8198
Macro ZFS_MINOR, introduced in commit a6cc9756 to record the chosen
static minor number for /dev/zfs, conflicts with an existing macro
in Lustre. The lustre macro (along with _MAJOR, _PATCH, _FIX) is
used to record the zfsonlinux version Lustre is being built against.
Since the Lustre macro came first, and is used in past versions of
lustre at least going back to 2.10, it makes sense to rename the
macro in ZFS instead of doing so in Lustre which would require
backporting the patch.
Reviewed-by: Giuseppe Di Natale <guss80@gmail.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Olaf Faaland <faaland1@llnl.gov>
Closes#8195
As a result of the changes made in 8585, it's possible for an excessive
amount of vdev flush commands to be issued under some workloads.
Specifically, when the workload consists of mostly async write activity,
interspersed with some sync write and/or fsync activity, we can end up
issuing more flush commands to the underlying storage than is actually
necessary. As a result of these flush commands, the write latency and
overall throughput of the pool can be poorly impacted (latency
increases, throughput decreases).
Currently, any time an lwb completes, the vdev(s) written to as a result
of that lwb will be issued a flush command. The intenion is so the data
written to that vdev is on stable storage, prior to communicating to any
waiting threads that their data is safe on disk.
The problem with this scheme, is that sometimes an lwb will not have any
threads waiting for it to complete. This can occur when there's async
activity that gets "converted" to sync requests, as a result of calling
the zil_async_to_sync() function via zil_commit_impl(). When this
occurs, the current code may issue many lwbs that don't have waiters
associated with them, resulting in many flush commands, potentially to
the same vdev(s).
For example, given a pool with a single vdev, and a single fsync() call
that results in 10 lwbs being written out (e.g. due to other async
writes), that will result in 10 flush commands to that single vdev (a
flush issued after each lwb write completes). Ideally, we'd only issue a
single flush command to that vdev, after all 10 lwb writes completed.
Further, and most important as it pertains to this change, since the
flush commands are often very impactful to the performance of the pool's
underlying storage, unnecessarily issuing these flush commands can
poorly impact the performance of the lwb writes themselves. Thus, we
need to avoid issuing flush commands when possible, in order to acheive
the best possible performance out of the pool's underlying storage.
This change attempts to address this problem by changing the ZIL's logic
to only issue a vdev flush command when it detects an lwb that has a
thread waiting for it to complete. When an lwb does not have threads
waiting for it, the responsibility of issuing the flush command to the
vdevs involved with that lwb's write is passed on to the "next" lwb.
It's only once a write for an lwb with waiters completes, do we issue
the vdev flush command(s). As a result, now when we issue the flush(s),
we will issue them to the vdevs involved with that specific lwb's write,
but potentially also to vdevs involved with "previous" lwb writes (i.e.
if the previous lwbs did not have waiters associated with them).
Thus, in our prior example with 10 lwbs, it's only once the last lwb
completes (which will be the lwb containing the waiter for the thread
that called fsync) will we issue the vdev flush command; all of the
other lwbs will find they have no waiters, so they'll pass the
responsibility of the flush to the "next" lwb (until reaching the last
lwb that has the waiter).
Porting Notes:
* Reconciled conflicts with the fastwrite feature.
Authored by: Prakash Surya <prakash.surya@delphix.com>
Reviewed by: Matt Ahrens <matt@delphix.com>
Reviewed by: Brad Lewis <brad.lewis@delphix.com>
Reviewed by: Patrick Mooney <patrick.mooney@joyent.com>
Reviewed by: Jerry Jelinek <jerry.jelinek@joyent.com>
Approved by: Joshua M. Clulow <josh@sysmgr.org>
Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
OpenZFS-issue: https://www.illumos.org/issues/9962
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/545190c6Closes#8188
Porting Notes:
* Add options to zfs-module-parameters(5) man page.
* zfs_nocacheflush move to vdev.c instead of vdev_disk.c, since
the latter doesn't get built for user space.
Authored by: Prakash Surya <prakash.surya@delphix.com>
Reviewed by: Matt Ahrens <matt@delphix.com>
Reviewed by: Brad Lewis <brad.lewis@delphix.com>
Reviewed by: Patrick Mooney <patrick.mooney@joyent.com>
Reviewed by: Tom Caputi <tcaputi@datto.com>
Reviewed by: George Melikov <mail@gmelikov.ru>
Approved by: Dan McDonald <danmcd@joyent.com>
Ported-by: Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
OpenZFS-issue: https://www.illumos.org/issues/9963
OpenZFS-commit: https://github.com/openzfs/openzfs/commit/f8fdf68125Closes#8186
This patch simply ensures that scn->scn_prefetch_queue is emptied
before the kernel module is unloaded and when scanning completes.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alek Pinchuk <apinchuk@datto.com>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8178
Commit 4c5b89f59 refactored dnode_hold() and in the process
accidentally introduced a slight change in behavior which was
not intended. The required behavior is that once the ZPL,
or other consumer, declares its intent to free a dnode then
dnode_hold() should immediately start failing. This updated
code wouldn't return the failure until after it was freed.
When DNODE_MUST_BE_ALLOCATED is set it must return ENOENT, and
when DNODE_MUST_BE_FREE is set it must return EEXIST;
This issue was uncovered by ztest_remap() which attempted
to remap a freeing object which should have been skipped as
described by the comment in dmu_objset_remap_indirects_impl().
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: Olaf Faaland <faaland1@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#8172
The verbose output of 'zpool list' was not correctly aligned due
to differences in the vdev name lengths. Minimally update the
code the correct the alignment using the same strategy employed
by 'zpool status'.
Missing dashes were added for the empty defaults columns, and
the vdev state is now printed for all vdevs.
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#7308Closes#8147
This patch corrects an issue where spa_vdev_remove() would
call spa_history_log_internal() while holding the spa config
lock. This function may decide to block until the next txg if
the current one seems too full. However, since the thread is
holding the config log, the txg sync thread cannot progress
and the system ends up deadlocked. This patch simply moves
all calls to spa_history_log_internal() outside of the config
lock.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8162
* Detect IO errors during device removal
While device removal cannot verify the checksums of individual
blocks during device removal, it can reasonably detect hard IO
errors from the leaf vdevs. Failure to perform this error
checking can result in device removal completing successfully,
but moving no data which will permanently corrupt the pool.
Situation 1: faulted/degraded vdevs
In the configuration shown below, the removal of mirror-0 will
permanently corrupt the pool. Device removal will preferentially
copy data from 'vdev1 -> vdev3' and from 'vdev2 -> vdev4'. Which
in this case will result in nothing being copied since one vdev
in each of those groups in unavailable. However, device removal
will complete successfully since all IO errors are ignored.
tank DEGRADED 0 0 0
mirror-0 DEGRADED 0 0 0
/var/tmp/vdev1 FAULTED 0 0 0 external fault
/var/tmp/vdev2 ONLINE 0 0 0
mirror-1 DEGRADED 0 0 0
/var/tmp/vdev3 ONLINE 0 0 0
/var/tmp/vdev4 FAULTED 0 0 0 external fault
This issue is resolved by updating the source child selection
logic to exclude unreadable leaf vdevs. Additionally, unwritable
destination child vdevs which can never succeed are skipped to
prevent generating a large number of write IO errors.
Situation 2: individual hard IO errors
During removal if an unexpected hard IO error is encountered when
either reading or writing the child vdev the entire removal
operation is cancelled. While it may be possible to reconstruct
the data after removal that cannot be guaranteed. The only
strictly safe thing to do is to cancel the removal.
As a future improvement we may want to instead suspend the removal
process and allow the damaged region to be retried. But that work
is left for another time, hard IO errors during the removal process
are expected to be exceptionally rare.
Reviewed-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Tom Caputi <tcaputi@datto.com>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #6900Closes#8161
ztest currently uses the boolean flag ztest_device_removal_active
to protect some tests that may not run successfully if they occur
at the same time as ztest_device_removal(). Unfortunately, in the
event that ztest is in the middle of a device removal when it
decides to issue a SIGKILL, the device removal will be
automatically restarted (without setting the flag) when the pool
is re-imported on the next run. This patch corrects this by
ensuring that any in-progress removals are completed before running
further tests after the re-import.
This patch also makes a few small changes to prevent race conditions
involving the creation and destruction of spa->spa_vdev_removal,
since this field is not protected by any locks. Some checks that
may run concurrently with setting / unsetting this field have been
updated to check spa->spa_removing_phys.sr_state instead. The most
significant change here is that spa_removal_get_stats() no longer
accounts for in-flight work done, since that could result in a NULL
pointer dereference.
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Serapheim Dimitropoulos <serapheim@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8105
This commit reverts to using printk() instead of zfs_dbgmsg() to log
messages in vdev_disk_error(): this is necessary because the latter can
be called from interrupt context where we are not allowed to sleep.
Unfortunately zfs_dbgmsg() performs its allocations calling kmalloc()
with the KM_SLEEP flag which may result in the following oops:
BUG: scheduling while atomic: swapper/4/0/0x10000100
Call Trace:
<IRQ> [<0>] dump_stack+0x19/0x1b
...
[<0>] spl_kmem_alloc+0xdf/0x140 [spl] <-- kmem_alloc(size, KM_SLEEP)
[<0>] __dprintf+0x69/0x150 [zfs]
[<0>] ? kmem_cache_free+0x1e2/0x200
[<0>] vdev_disk_error.part.15+0x5f/0x70 [zfs]
[<0>] vdev_disk_io_flush_completion+0x48/0x70 [zfs]
[<0>] bio_endio+0x67/0xb0
[<0>] blk_update_request+0x90/0x360
...
[<0>] scsi_finish_command+0xdc/0x140
[<0>] scsi_softirq_done+0x132/0x160
[<0>] blk_done_softirq+0x96/0xc0
[<0>] __do_softirq+0xf5/0x280
[<0>] call_softirq+0x1c/0x30
[<0>] do_softirq+0x65/0xa0
[<0>] irq_exit+0x105/0x110
[<0>] do_IRQ+0x56/0xf0
[<0>] common_interrupt+0x162/0x162
<EOI> [<0>] ? cpuidle_enter_state+0x54/0xd0
[<0>] cpuidle_idle_call+0xde/0x230
[<0>] arch_cpu_idle+0xe/0xb0
[<0>] cpu_startup_entry+0x14a/0x1e0
[<0>] start_secondary+0x1f7/0x270
[<0>] start_cpu+0x5/0x14
Reviewed-by: Olaf Faaland <faaland1@llnl.gov>
Reviewed by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes#8137Closes#8150
Currently, several tests in the ZFS Test Suite that attempt to
test scrub and resilver behavior occasionally fail. A big reason
for this is that these tests use a combination of zinject and
zfs_scan_vdev_limit to attempt to slow these operations enough
to verify their test commands. This method works most of the time,
but provides no guarantees and leads to flaky behavior. This patch
adds a new tunable, zfs_scan_suspend_progress, that ensures that
scans make no progress, guaranteeing that tests can be run without
racing.
This patch also changes zfs_remove_max_bytes_pause to match this
new tunable. This provides some consistency between these two
similar tunables and ensures that the tunable will not misbehave
on 32-bit systems.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <guss80@gmail.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8111
CID 184285: Read from pointer after free (USE_AFTER_FREE)
This patch fixes an use-after-free in vdev_config_generate_stats()
moving the kmem_free() call at the end of the function.
Reviewed-by: George Melikov <mail@gmelikov.ru>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Giuseppe Di Natale <guss80@gmail.com>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes#8120
This commit adds a new test case to the ZFS Test Suite to verify ZED
can detect when a device is physically removed from a running system:
the device will be offlined if a spare is not available in the pool.
We implement this by using the existing libudev functionality and
without relying solely on the FM kernel module capabilities which have
been observed to be unreliable with some kernels.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Don Brady <don.brady@delphix.com>
Signed-off-by: loli10K <ezomori.nozomu@gmail.com>
Closes#1537Closes#7926
This patch adds a new slow I/Os (-s) column to zpool status to show the
number of VDEV slow I/Os. This is the number of I/Os that didn't
complete in zio_slow_io_ms milliseconds. It also adds a new parsable
(-p) flag to display exact values.
NAME STATE READ WRITE CKSUM SLOW
testpool ONLINE 0 0 0 -
mirror-0 ONLINE 0 0 0 -
loop0 ONLINE 0 0 0 20
loop1 ONLINE 0 0 0 0
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed by: Matthew Ahrens <mahrens@delphix.com>
Signed-off-by: Tony Hutter <hutter2@llnl.gov>
Closes#7756Closes#6885
It's disabled by default, update code and tests to reflect
the documentation.
Minor cleanup in delegate_common.kshlib.
Reviewed-by: Gregor Kopka <gregor@kopka.net>
Reviewed-by: John Kennedy <john.kennedy@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: George Melikov <mail@gmelikov.ru>
Closes#7835Closes#8045
This patch simply ensures that vdev_indirect_splits_damage()
cannot hit a divide by zero exception if a split has no
children with valid data. The normal reconstruction code
path in vdev_indirect_reconstruct_io_done() already has this
check.
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8086
This patch simply corrects an issue where vdev_dtl_reassess()
could attempt to dirty the vdev config even when the spa was
not elligable for writing.
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8085
This patch ensures that logs are replayed on all datasets prior
to starting ztest workers. This ensures that the call to
vdev_offline() a log device in ztest_fault_inject() will not fail
due to the log device being required for replay.
This patch also fixes a small issue found during testing where
spa_keystore_load_wkey() does not check that the dataset specified
is an encryption root. This check was present in libzfs, however.
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Closes#8084
This patch fixes a race condition where the end of
vdev_remove_replace_with_indirect(), which holds
svr_lock, would race against spa_vdev_removal_destroy(),
which destroys the same lock and is called asynchronously
via dsl_sync_task_nowait().
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Issue #6900Closes#8083
vdev_clear() can call vdev_set_deferred_resilver() with a
non-leaf vdev to setup a deferred resilver. However, this
function is currently written to only handle leaf vdevs.
This bug was introduced with deferred resilvers in 80a91e74.
This patch makes this function recursive so that it can find
appropriate vdevs to resilver and set vdev_resilver_deferred
on them.
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Tom Caputi <tcaputi@datto.com>
Issue #7732Closes#8082
Tony Hutter
Tom Caputi
Alexander Motin
Andrew Stormont
Justin Gottula
Alek P
Brian Behlendorf
Olaf Faaland
George Wilson
mzhivich
Paul Zuchowski
Matthew Ahrens
Serapheim Dimitropoulos
loli10K
lidongyang
Tomohiro Kusumi
Sara Hartse
Tim Chase
asomers
George Wilson
Brad Lewis
Andriy Gapon
Richard Elling
Prakash Surya
loli10K
George Melikov