1) Make mmap flushes synchronous. Linux may skip flushing dirty pages
already in writeback unless data-integrity sync is requested.
2) Change zfs_putpage to use TXG_WAIT. Otherwise dirty pages may be
skipped due to DMU pushing back on TX assign.
3) Add missing mmap flush when doing block cloning.
4) While here, pass errors from putpage to writepage/writepages.
This change fixes corruption edge cases, but unfortunately adds
synchronous ZIL flushes for dirty mmap pages to llseek and bclone
operations. It may be possible to avoid these sync writes later
but would need more tricky refactoring of the writeback code.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Robert Evans <evansr@google.com>
Closes#15933Closes#16019
The regular ABD iterators yield data buffers, so they have to map and
unmap pages into kernel memory. If the caller only wants to count
chunks, or can use page pointers directly, then the map/unmap is just
unnecessary overhead.
This adds adb_iterate_page_func, which yields unmapped struct page
instead.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Closes#15533Closes#15588
(cherry picked from commit 390b448726)
When ZFS overwrites a whole block, it does not bother to read the
old content from disk. It is a good optimization, but if the buffer
fill fails due to page fault or something else, the buffer ends up
corrupted, neither keeping old content, nor getting the new one.
On FreeBSD this is additionally complicated by page faults being
blocked by VFS layer, always returning EFAULT on attempt to write
from mmap()'ed but not yet cached address range. Normally it is
not a big problem, since after original failure VFS will retry the
write after reading the required data. The problem becomes worse
in specific case when somebody tries to write into a file its own
mmap()'ed content from the same location. In that situation the
only copy of the data is getting corrupted on the page fault and
the following retries only fixate the status quo. Block cloning
makes this issue easier to reproduce, since it does not read the
old data, unlike traditional file copy, that may work by chance.
This patch provides the fill status to dmu_buf_fill_done(), that
in case of error can destroy the corrupted buffer as if no write
happened. One more complication in case of block cloning is that
if error is possible during fill, dmu_buf_will_fill() must read
the data via fall-back to dmu_buf_will_dirty(). It is required
to allow in case of error restoring the buffer to a state after
the cloning, not not before it, that would happen if we just call
dbuf_undirty().
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Rob Norris <robn@despairlabs.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15665
Similar to deduplication, the size of data duplicated by block cloning
should not be included in the slop space calculation.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Yuxin Wang <yuxinwang9999@gmail.com>
Closes#15874
On Linux the ioctl_ficlonerange() and ioctl_ficlone() system calls
are expected to either fully clone the specified range or return an
error. The range may be for an entire file. While internally ZFS
supports cloning partial ranges there's no way to return the length
cloned to the caller so we need to make this all or nothing.
As part of this change support for the REMAP_FILE_CAN_SHORTEN flag
has been added. When REMAP_FILE_CAN_SHORTEN is set zfs_clone_range()
will return a shortened range when encountering pending dirty records.
When it's clear zfs_clone_range() will block and wait for the records
to be written out allowing the blocks to be cloned.
Furthermore, the file range lock is held over the region being cloned
to prevent it from being modified while cloning. This doesn't quite
provide an atomic semantics since if an error is encountered only a
portion of the range may be cloned. This will be converted to an
error if REMAP_FILE_CAN_SHORTEN was not provided and returned to the
caller. However, the destination file range is left in an undefined
state.
A test case has been added which exercises this functionality by
verifying that `cp --reflink=never|auto|always` works correctly.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Closes#15728Closes#15842
During device removal stress tests, we noticed that we were tripping
the assertion that mg_initialized was true. After investigation, it was
determined that the mg in question was the embedded log metaslab
group for a newly added vdev; the normal mg had been initialized (by
metaslab_sync_reassess, via vdev_sync_done). However, because the spa
config alloc lock is not held as writer across both calls to
metaslab_sync_reassess, it is possible for an allocation to happen
between the two metaslab_groups being initialized. Because the metaslab
code doesn't check the group in question, just the vdev's main mg, it
is possible to get past the initial check in vdev_allocatable and
later fail due to the assertion.
We simply remove the assertions. We could also consider locking the
ALLOC lock around the reassess calls in vdev_sync_done, but that risks
deadlocks. We could check the actual target mg in vdev_allocatable,
but that risks racing with a passivation that comes in after that
check but before the assertion. We still won't be able to actually
allocate from the metaslab group if no metaslabs are ready, so this
change shouldn't break anything.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Wilson <george.wilson@delphix.com>
Signed-off-by: Paul Dagnelie <pcd@delphix.com>
Closes#15818
If devid or physpath for a vdev changes between imports, ensure it is
updated to the new value.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Ameer Hamza <ahamza@ixsystems.com>
Closes#15816
Pool import logic uses vdev paths, so it makes sense to add path
information on AUX vdev as well.
Reviewed-by: Umer Saleem <usaleem@ixsystems.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Ameer Hamza <ahamza@ixsystems.com>
Closes#15737
When spare or l2cache (aux) vdev is added during pool creation,
spa->spa_uberblock is not dumped until that point. Subsequently,
the aux label is never synchronized after its initial creation,
resulting in the uberblock label remaining undumped. The uberblock
is crucial for lib_blkid in identifying the ZFS partition type. To
address this issue, we now ensure sync of the uberblock label once
if it's not dumped initially.
Reviewed-by: Umer Saleem <usaleem@ixsystems.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Ameer Hamza <ahamza@ixsystems.com>
Closes#15737
In general, VOPs must not load the "z_log" field until having called
zfs_enter_verify_zp().
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Mark Johnston <markj@FreeBSD.org>
Closes#15752
If the destination file is mmaped and the mmaped region was already
read, so it is cached, we need to update mmaped pages after successful
clone using update_pages().
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Pointed out by: Ka Ho Ng <khng@freebsd.org>
Signed-off-by: Pawel Jakub Dawidek <pawel@dawidek.net>
Closes#15772
Switch from cv_wait() to cv_wait_idle() in vdev_autotrim_wait_kick(),
which should mitigate the high load average while waiting.
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: jxdking <lostking2008@hotmail.com>
Closes#15781
For FreeBSD sysctls, we don't want the extra newline, since the
sysctl(8) utility will format strings appropriately.
Reviewed-by: Rob Norris <robn@despairlabs.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reported-by: Peter Holm <pho@FreeBSD.org>
Signed-off-by: Mark Johnston <markj@FreeBSD.org>
Closes#15719
sbuf_cpy() resets the sbuf state, which is wrong for sbufs allocated by
sbuf_new_for_sysctl(). In particular, this code triggers an assertion
failure in sbuf_clear().
Simplify by just using sysctl_handle_string() for both reading and
setting the tunable.
Fixes: 6930ecbb7 ("spa: make read/write queues configurable")
Reviewed-by: Rob Norris <robn@despairlabs.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reported-by: Peter Holm <pho@FreeBSD.org>
Signed-off-by: Mark Johnston <markj@FreeBSD.org>
Closes#15719
Two block pointers in livelist pointing to the same location may
be caused not only by dedup, but also by block cloning. We should
not assert D bit set in them.
Two block pointers in livelist pointing to the same location may
have different logical birth time in case of dedup or cloning. We
should assert identical physical birth time instead.
Assert identical physical block size between pointers in addition
to checksum, since that is what checksums are calculated on.
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15732
- Fail if source block is smaller than destination. We can only
grow blocks, not shrink them.
- Fail if we do not have full znode range lock. In that case grow
is not even called. We should improve zfs_rangelock_cb() somehow
to know when cloning needs to grow the block size unlike write.
- Fail of we tried to resize, but failed. There are many reasons
for it to fail that we can not predict at this level, so be ready
for them. Unlike write, that may proceed after growth failure,
block cloning can't and must return error.
This fixes assertion inside dmu_brt_clone() when it sees different
number of blocks held in destination than it got block pointers.
Builds without ZFS_DEBUG returned EXDEV, so are not affected much.
Reviewed-by: Pawel Jakub Dawidek <pawel@dawidek.net>
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15724Closes#15735
While 763ca47 closes the situation of block cloning creating
unencrypted records in encrypted datasets, existing data still causes
panic on read. Setting zfs_recover bypasses this but at the cost of
potentially ignoring more serious issues.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Chris Peredun <chris.peredun@ixsystems.com>
Closes#15677
Block cloning normally creates dirty record without dr_data. But if
the block is read after cloning, it is moved into DB_CACHED state and
receives the data buffer. If after that we call dbuf_unoverride()
to convert the dirty record into normal write, we should give it the
data buffer from dbuf and release one.
Reviewed-by: Kay Pedersen <mail@mkwg.de>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15654Closes#15656
In some cases dbuf_assign_arcbuf() may be called on a block that
was recently cloned. If it happened in current TXG we must undo
the block cloning first, since the only one dirty record per TXG
can't and shouldn't mean both cloning and overwrite same time.
Reviewed-by: Kay Pedersen <mail@mkwg.de>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15653
dmu_assign_arcbuf_by_dnode() should drop dn_struct_rwlock lock in
case dbuf_hold() failed. I don't have reproduction for this, but
it looks inconsistent with dmu_buf_hold_noread_by_dnode() and co.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15644
Without this patch on pool of 60 vdevs with ZFS_DEBUG enabled clone
takes much more time than copy, while heavily trashing dbgmsg for
no good reason, repeatedly dumping all vdevs BRTs again and again,
even unmodified ones.
I am generally not sure this dumping is not excessive, but decided
to keep it for now, just restricting its scope to more reasonable.
Reviewed-by: Kay Pedersen <mail@mkwg.de>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15625
To improve 128KB block write performance in case of multiple VDEVs
ZIL used to spit those writes into two 64KB ones. Unfortunately it
was found to cause LWB buffer overflow, trying to write maximum-
sizes 128KB TX_CLONE_RANGE record with 1022 block pointers into
68KB buffer, since unlike TX_WRITE ZIL code can't split it.
This is a minimally-invasive temporary block cloning fix until the
following more invasive prediction code refactoring.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Ameer Hamza <ahamza@ixsystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15634
When two datasets share the same master encryption key, it is safe
to clone encrypted blocks. Currently only snapshots and clones
of a dataset share with it the same encryption key.
Added a test for:
- Clone from encrypted sibling to encrypted sibling with
non encrypted parent
- Clone from encrypted parent to inherited encrypted child
- Clone from child to sibling with encrypted parent
- Clone from snapshot to the original datasets
- Clone from foreign snapshot to a foreign dataset
- Cloning from non-encrypted to encrypted datasets
- Cloning from encrypted to non-encrypted datasets
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Original-patch-by: Pawel Jakub Dawidek <pawel@dawidek.net>
Signed-off-by: Kay Pedersen <mail@mkwg.de>
Closes#15544
ZIL claim can not handle block pointers cloned from the future,
since they are not yet allocated at that point. It may happen
either if the block was just written when it was cloned, or if
the pool was frozen or somehow else rewound on import.
Handle it from two sides: prevent cloning of blocks with physical
birth time from not yet synced or frozen TXG, and abort ZIL claim
if we still detect such blocks due to rewind or something else.
While there, assert that any cloned blocks we claim are really
allocated by calling metaslab_check_free().
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15617
zil_claim_clone_range() takes references on cloned blocks before ZIL
replay. Later zil_free_clone_range() drops them after replay or on
dataset destroy. The total balance is neutral. It means we do not
need to do anything (drop the references) for not implemented yet
TX_CLONE_RANGE replay for ZVOLs.
This is a logical follow up to #15603.
Reviewed-by: Kay Pedersen <mail@mkwg.de>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15612
Since we use a limited set of kmem caches, quite often we have unused
memory after the end of the buffer. Put there up to a 512-byte canary
when built with debug to detect buffer overflows at the free time.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15553
This should make sure we have log written without overflows.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15517
PR #15457 exposed weird logic in L2ARC write sizing. If it appeared
bigger than device size, instead of liming write it reset all the
system-wide tunables to their default. Aside of being excessive,
it did not actually help with the problem, still allowing infinite
loop to happen.
This patch removes the tunables reverting logic, but instead limits
L2ARC writes (or at least eviction/trim) to 1/4 of the capacity.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Amanakis <gamanakis@gmail.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15519
Add a dataset_kstats_rename function, and call it when renaming
a zvol on FreeBSD and Linux.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alan Somers <asomers@gmail.com>
Sponsored-by: Axcient
Closes#15482Closes#15486
Once we verified the ABDs and asserted the sizes we should never
see premature ABDs ends. Assert that and remove extra branches
from production builds.
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15428
We are finding that as customers get larger and faster machines
(hundreds of cores, large NVMe-backed pools) they keep hitting
relatively low performance ceilings. Our profiling work almost always
finds that they're running into bottlenecks on the SPA IO taskqs.
Unfortunately there's often little we can advise at that point, because
there's very few ways to change behaviour without patching.
This commit adds two load-time parameters `zio_taskq_read` and
`zio_taskq_write` that can configure the READ and WRITE IO taskqs
directly.
This achieves two goals: it gives operators (and those that support
them) a way to tune things without requiring a custom build of OpenZFS,
which is often not possible, and it lets us easily try different config
variations in a variety of environments to inform the development of
better defaults for these kind of systems.
Because tuning the IO taskqs really requires a fairly deep understanding
of how IO in ZFS works, and generally isn't needed without a pretty
serious workload and an ability to identify bottlenecks, only minimal
documentation is provided. Its expected that anyone using this is going
to have the source code there as well.
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
zil_claim_clone_range() takes references on cloned blocks before ZIL
replay. Later zil_free_clone_range() drops them after replay or on
dataset destroy. The total balance is neutral. It means on actual
replay we must take additional references, which would stay in BRT.
Without this blocks could be freed prematurely when either original
file or its clone are destroyed. I've observed BRT being emptied
and the feature being deactivated after ZIL replay completion, which
should not have happened. With the patch I see expected stats.
Reviewed-by: Kay Pedersen <mail@mkwg.de>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Rob Norris <robn@despairlabs.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15603
Previously, dmu_buf_will_clone() would roll back any dirty record, but
would not clean out the modified data nor reset the state before
releasing the lock. That leaves the last-written data in db_data, but
the dbuf in the wrong state.
This is eventually corrected when the dbuf state is made NOFILL, and
dbuf_noread() called (which clears out the old data), but at this point
its too late, because the lock was already dropped with that invalid
state.
Any caller acquiring the lock before the call into
dmu_buf_will_not_fill() can find what appears to be a clean, readable
buffer, and would take the wrong state from it: it should be getting the
data from the cloned block, not from earlier (unwritten) dirty data.
Even after the state was switched to NOFILL, the old data was still not
cleaned out until dbuf_noread(), which is another gap for a caller to
take the lock and read the wrong data.
This commit fixes all this by properly cleaning up the previous state
and then setting the new state before dropping the lock. The
DBUF_VERIFY() calls confirm that the dbuf is in a valid state when the
lock is down.
Sponsored-by: Klara, Inc.
Sponsored-By: OpenDrives Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Pawel Jakub Dawidek <pawel@dawidek.net>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#15566Closes#15526
So that zdb (and others!) can get at the BRT on-disk structures.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Kay Pedersen <mail@mkwg.de>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Closes#15541
Over its history this the dirty dnode test has been changed between
checking for a dnodes being on `os_dirty_dnodes` (`dn_dirty_link`) and
`dn_dirty_record`.
de198f2d9 Fix lseek(SEEK_DATA/SEEK_HOLE) mmap consistency
2531ce372 Revert "Report holes when there are only metadata changes"
ec4f9b8f3 Report holes when there are only metadata changes
454365bba Fix dirty check in dmu_offset_next()
66aca2473 SEEK_HOLE should not block on txg_wait_synced()
Also illumos/illumos-gate@c543ec060dillumos/illumos-gate@2bcf0248e9
It turns out both are actually required.
In the case of appending data to a newly created file, the dnode proper
is dirtied (at least to change the blocksize) and dirty records are
added. Thus, a single logical operation is represented by separate
dirty indicators, and must not be separated.
The incorrect dirty check becomes a problem when the first block of a
file is being appended to while another process is calling lseek to skip
holes. There is a small window where the dnode part is undirtied while
there are still dirty records. In this case, `lseek(fd, 0, SEEK_DATA)`
would not know that the file is dirty, and would go to
`dnode_next_offset()`. Since the object has no data blocks yet, it
returns `ESRCH`, indicating no data found, which results in `ENXIO`
being returned to `lseek()`'s caller.
Since coreutils 9.2, `cp` performs sparse copies by default, that is, it
uses `SEEK_DATA` and `SEEK_HOLE` against the source file and attempts to
replicate the holes in the target. When it hits the bug, its initial
search for data fails, and it goes on to call `fallocate()` to create a
hole over the entire destination file.
This has come up more recently as users upgrade their systems, getting
OpenZFS 2.2 as well as a newer coreutils. However, this problem has been
reproduced against 2.1, as well as on FreeBSD 13 and 14.
This change simply updates the dirty check to check both types of dirty.
If there's anything dirty at all, we immediately go to the "wait for
sync" stage, It doesn't really matter after that; both changes are on
disk, so the dirty fields should be correct.
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Rich Ercolani <rincebrain@gmail.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#15571Closes#15526
This reverts commit bd7a02c251 which
can trigger an unlikely existing bio alignment issue on Linux.
This change is good, but the underlying issue it exposes needs to
be resolved before this can be re-applied.
Signed-off-by: Brian Behlendorf <behlendorf1@llnl.gov>
Issue #15533
Currently vdev_queue_class_length is responsible for checking how long
the queue length is, however, it doesn't check the length when a list
is used, rather it just returns whether it is empty or not. To fix this
I added a counter variable to vdev_queue_class to keep track of the sync
IO ops, and changed vdev_queue_class_length to reference this variable
instead.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: MigeljanImeri <ImeriMigel@gmail.com>
Closes#15478
There is no sense to have separate implementations for FreeBSD and
Linux. Make Linux code shared as more functional and just register
FreeBSD-specific prune callback with arc_add_prune_callback() API.
Aside of code cleanup this should fix excessive pruning on FreeBSD:
https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=274698
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Mark Johnston <markj@FreeBSD.org>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15456
We should not always use PAGESIZE alignment for caches bigger than
it and SPA_MINBLOCKSIZE otherwise. Doing that caches for 5, 6, 7,
10 and 14KB rounded up to 8, 12 and 16KB respectively make no sense.
Instead specify as alignment the biggest power-of-2 divisor. This
way 2KB and 6KB caches are both aligned to 2KB, while 4KB and 8KB
are aligned to 4KB.
Reduce number of caches to half-power of 2 instead of quarter-power
of 2. This removes caches difficult for underlying allocators to
fit into page-granular slabs, such as: 2.5, 3.5, 5, 7, 10KB, etc.
Since these caches are mostly used for transient allocations like
ZIOs and small DBUF cache it does not worth being too aggressive.
Due to the above alignment issue some of those caches were not
working properly any way. 6KB cache now finally has a chance to
work right, placing 2 buffers into 3 pages, that makes sense.
Remove explicit alignment in Linux user-space case. I don't think
it should be needed any more with the above fixes.
As result on FreeBSD instead of such numbers of pages per slab:
vm.uma.zio_buf_comb_16384.keg.ppera: 4
vm.uma.zio_buf_comb_14336.keg.ppera: 4
vm.uma.zio_buf_comb_12288.keg.ppera: 3
vm.uma.zio_buf_comb_10240.keg.ppera: 3
vm.uma.zio_buf_comb_8192.keg.ppera: 2
vm.uma.zio_buf_comb_7168.keg.ppera: 2
vm.uma.zio_buf_comb_6144.keg.ppera: 2 <= Broken
vm.uma.zio_buf_comb_5120.keg.ppera: 2
vm.uma.zio_buf_comb_4096.keg.ppera: 1
vm.uma.zio_buf_comb_3584.keg.ppera: 7 <= Hard to free
vm.uma.zio_buf_comb_3072.keg.ppera: 3
vm.uma.zio_buf_comb_2560.keg.ppera: 2
vm.uma.zio_buf_comb_2048.keg.ppera: 1
vm.uma.zio_buf_comb_1536.keg.ppera: 2
vm.uma.zio_buf_comb_1024.keg.ppera: 1
vm.uma.zio_buf_comb_512.keg.ppera: 1
I am now getting such:
vm.uma.zio_buf_comb_16384.keg.ppera: 4
vm.uma.zio_buf_comb_12288.keg.ppera: 3
vm.uma.zio_buf_comb_8192.keg.ppera: 2
vm.uma.zio_buf_comb_6144.keg.ppera: 3 <= Fixed, 2 in 3 pages
vm.uma.zio_buf_comb_4096.keg.ppera: 1
vm.uma.zio_buf_comb_3072.keg.ppera: 3
vm.uma.zio_buf_comb_2048.keg.ppera: 1
vm.uma.zio_buf_comb_1536.keg.ppera: 2
vm.uma.zio_buf_comb_1024.keg.ppera: 1
vm.uma.zio_buf_comb_512.keg.ppera: 1
Reviewed-by: Allan Jude <allan@klarasystems.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15452
dmu_tx_check_ioerr() pre-reads blocks that are going to be dirtied
as part of transaction to both prefetch them and check for errors.
But it makes no sense to do it for holes, since there are no disk
reads to prefetch and there can be no errors. On the other side
those blocks are anonymous, and they are freed immediately by the
dbuf_rele() without even being put into dbuf cache, so we just
burn CPU time on decompression and overheads and get absolutely
no result at the end.
Use of dbuf_hold_impl() with fail_sparse parameter allows to skip
the extra work, and on my tests with sequential 8KB writes to empty
ZVOL with 32KB blocks shows throughput increase from 1.7 to 2GB/s.
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15371
Prefetched buffers are currently read from L2ARC if, and only if,
l2arc_noprefetch is set to non-default value of 0. This means that
a streaming read which can be served from L2ARC will instead engage
the main pool.
For example, consider what happens when a file is sequentially read:
- application requests contiguous data, engaging the prefetcher;
- ARC buffers are initially marked as prefetched but, as the calling
application consumes data, the prefetch tag is cleared;
- these "normal" buffers become eligible for L2ARC and are copied to it;
- re-reading the same file will *not* engage L2ARC even if it contains
the required buffers;
- main pool has to suffer another sequential read load, which (due to
most NCQ-enabled HDDs preferring sequential loads) can dramatically
increase latency for uncached random reads.
In other words, current behavior is to write data to L2ARC (wearing it)
without using the very same cache when reading back the same data. This
was probably useful many years ago to preserve L2ARC read bandwidth but,
with current SSD speed/size/price, it is vastly sub-optimal.
Setting l2arc_noprefetch=1, while enabling L2ARC to serve these reads,
means that even prefetched but unused buffers will be copied into L2ARC,
further increasing wear and load for potentially not-useful data.
This patch enable prefetched buffer to be read from L2ARC even when
l2arc_noprefetch=1 (default), increasing sequential read speed and
reducing load on the main pool without polluting L2ARC with not-useful
(ie: unused) prefetched data. Moreover, it clear users confusion about
L2ARC size increasing but not serving any IO when doing sequential
reads.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Gionatan Danti <g.danti@assyoma.it>
Closes#15451
Many long-running ZFS ioctls lock the spa_namespace_lock, forcing
concurrent ioctls to sleep for the mutex. Previously, the only
option is to call mutex_enter() which sleeps uninterruptibly. This
is a usability issue for sysadmins, for example, if the admin runs
`zpool status` while a slow `zpool import` is ongoing, the admin's
shell will be locked in uninterruptible sleep for a long time.
This patch resolves this admin usability issue by introducing
mutex_enter_interruptible() which sleeps interruptibly while waiting
to acquire a lock. It is implemented for both Linux and FreeBSD.
The ZFS_IOC_POOL_CONFIGS ioctl, used by `zpool status`, is changed to
use this new macro so that the command can be interrupted if it is
issued during a concurrent `zpool import` (or other long-running
operation).
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Thomas Bertschinger <bertschinger@lanl.gov>
Closes#15360
In my understanding ARC_BUF_SHARED() and arc_buf_is_shared() should
return identical results, except the second also asserts it deeper.
The first is much cheaper though, saving few pointer dereferences.
Replace production arc_buf_is_shared() calls with ARC_BUF_SHARED(),
and call arc_buf_is_shared() in random assertions, while making it
even more strict.
On my tests this in half reduces arc_buf_destroy_impl() time, that
noticeably reduces hash_lock congestion under heavy dbuf eviction.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: George Wilson <george.wilson@delphix.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15397
Torn reads/writes of dp_dirty_total are unlikely: on 64-bit systems
due to register size, while on 32-bit due to memory constraints.
And even if we hit some race, the code implementing the delay takes
the lock any way.
Removal of the poll-wide lock acquisition saves ~1% of CPU time on
8-thread 8KB write workload.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15390
When a vdev is to be expanded -- either via `zpool online -e` or via
the autoexpand option -- a SPA_ASYNC_CONFIG_UPDATE request is queued
to be handled via an asynchronous worker thread (spa_async_thread).
This normally happens almost immediately; but will be delayed up to
zfs_ccw_retry_interval seconds (default 5 minutes) if an attempt to
write the zpool configuration cache failed.
When FreeBSD boots ZFS-root VM images generated using `makefs -t zfs`,
the zpoolupgrade rc.d script runs `zpool upgrade`, which modifies the
pool configuration and triggers an attempt to write to the cache file.
This attempted write fails because the filesystem is still mounted
read-only at this point in the boot process, triggering a 5-minute
cooldown before SPA_ASYNC_CONFIG_UPDATE requests will be handled by
the asynchronous worker thread.
When expanding a vdev, reset the "when did a configuration cache
write last fail" value so that the SPA_ASYNC_CONFIG_UPDATE request
will be handled promptly. A cleaner but more intrusive option would
be to use separate SPA_ASYNC_ flags for "configuration changed" and
"try writing the configuration cache again", but with FreeBSD 14.0
coming very soon I'd prefer to leave such refactoring for a later
date.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Colin Percival <cperciva@FreeBSD.org>
Closes#15405
For synchronous write workloads with large IO sizes, a pool configured
with a slog performs worse than one with an embedded zil:
sequential_writes 1m sync ios, 16 threads
Write IOPS: 1292 438 -66.10%
Write Bandwidth: 1323570 448910 -66.08%
Write Latency: 12128400 36330970 3.0x
sequential_writes 1m sync ios, 32 threads
Write IOPS: 1293 430 -66.74%
Write Bandwidth: 1324184 441188 -66.68%
Write Latency: 24486278 74028536 3.0x
The reason is the `zil_slog_bulk` variable. In `zil_lwb_write_open`,
if a zil block is greater than 768K, the priority of the write is
downgraded from sync to async. Increasing the value allows greater
throughput. To select a value for this PR, I ran an fio workload with
the following values for `zil_slog_bulk`:
zil_slog_bulk KiB/s
1048576 422132
2097152 478935
4194304 533645
8388608 623031
12582912 827158
16777216 1038359
25165824 1142210
33554432 1211472
50331648 1292847
67108864 1308506
100663296 1306821
134217728 1304998
At 64M, the results with a slog are now improved to parity with an
embedded zil:
sequential_writes 1m sync ios, 16 threads
Write IOPS: 438 1288 2.9x
Write Bandwidth: 448910 1319062 2.9x
Write Latency: 36330970 12163408 -66.52%
sequential_writes 1m sync ios, 32 threads
Write IOPS: 430 1290 3.0x
Write Bandwidth: 441188 1321693 3.0x
Write Latency: 74028536 24519698 -66.88%
None of the other tests in the performance suite (run with a zil or
slog) had a significant change, including the random_write_zil tests,
which use multiple datasets.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Tony Nguyen <tony.nguyen@delphix.com>
Signed-off-by: John Wren Kennedy <john.kennedy@delphix.com>
Closes#14378
Block cloning from an encrypted dataset into an unencrypted dataset
and vice versa is not possible. The current code did allow cloning
unencrypted files into an encrypted dataset causing a panic when
these were accessed. Block cloning between encrypted and encrypted
is currently supported on the same filesystem only.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Kay Pedersen <mail@mkwg.de>
Reviewed-by: Rob N <robn@despairlabs.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Martin Matuska <mm@FreeBSD.org>
Closes#15464Closes#15465
When doing a manual TRIM on a zpool, the metaslab being TRIMmed is
potentially re-enabled before all queued TRIM zios for that metaslab
have completed. Since TRIM zios have the lowest priority, it is
possible to get into a situation where allocations occur from the
just re-enabled metaslab and cut ahead of queued TRIMs to the same
metaslab. If the ranges overlap, this will cause corruption.
We were able to trigger this pretty consistently with a small single
top-level vdev zpool (i.e. small number of metaslabs) with heavy
parallel write activity while performing a manual TRIM against a
somewhat 'slow' device (so TRIMs took a bit of time to complete).
With the patch, we've not been able to recreate it since. It was on
illumos, but inspection of the OpenZFS trim code looks like the
relevant pieces are largely unchanged and so it appears it would be
vulnerable to the same issue.
Reviewed-by: Igor Kozhukhov <igor@dilos.org>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Jason King <jking@racktopsystems.com>
Illumos-issue: https://www.illumos.org/issues/15939Closes#15395
Benchmarks show that at certain write sizes range lock/unlock take
not so much time as extra memory copy. The exact threshold is not
obvious due to other overheads, but it is definitely lower than
~63KB used before. Make it configurable, defaulting at 7.5KB,
that is 8KB of nearest malloc() size minus itx and lr structs.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15353
Previously, if a cachefile is passed to zpool import, the cached config
is mostly offered as-is to ZFS_IOC_POOL_TRYIMPORT->spa_tryimport(), and
the results are taken as the canonical pool config and handed back to
ZFS_IOC_POOL_IMPORT.
In the course of its operation, spa_load() will inspect the pool and
build a new config from what it finds on disk. However, it then
regenerates a new config ready to import, and so rightly sets the hostid
and hostname for the local host in the config it returns.
Because of this, the "require force" checks always decide the pool is
exported and last touched by the local host, even if this is not true,
which is possible in a HA environment when MMP is not enabled. The pool
may be imported on another head, but the import checks still pass here,
so the pool ends up imported on both.
(This doesn't happen when a cachefile isn't used, because the pool
config is discovered in userspace in zpool_find_import(), and that does
find the on-disk hostid and hostname correctly).
Since the systemd zfs-import-cache.service unit uses cachefile imports,
this can lead to a system returning after a crash with a "valid"
cachefile on disk and automatically, quietly, importing a pool that has
already been taken up by a secondary head.
This commit causes the on-disk hostid and hostname to be included in the
ZPOOL_CONFIG_LOAD_INFO item in the returned config, and then changes the
"force" checks for zpool import to use them if present.
This method should give no change in behaviour for old userspace on new
kernels (they won't know to look for the new config items) and for new
userspace on old kernels (the won't find the new config items).
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Closes#15290
