When ZIL allocates space for new LWBs without knowing how much it
will require, it can use new metaslab_alloc_range() function to
allocate slightly more or less than it predicted. It allows to
improve space efficiency by allocating bigger LWBs on RAIDZ/dRAID
instead of padding and possibly packing more ZIL records there.
It may also allow to reduce ganging in some cases by allowing to
allocate smaller LWBs when we are not sure we'll need bigger.
On the opposite side, when we allocate space for already closed
LWBs, when we precisely know how much space we need, we may just
allocate what we need instead of relying on writing less than
allocated, that does not work for RAIDZ.
Space for LWBs in open state (still being filled) is allocated
same as before.
Reviewed-by: Rob Norris <robn@despairlabs.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <alexander.motin@TrueNAS.com>
Closes#17613
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#17622
Make sure we properly inform the nolwb waiters of the error, and don't
keep trying.
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#17622
Just making it easier to not get the locking and broadcast wrong.
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#17622
If the ZIL crashed, any outstanding LWBs are no longer interesting, so
if they return, we need to just clean them up and return, not try to do
any work on them. This is true even if they return success, as that may
be long after the pool suspended and resumed, depending on when/if the
kernel decides to return the IO to us. In particular, we must not try to
get the "next" LWB from zl_lwb_list, since they're no longer on that
list.
So, we put a flag on in-flight LWBs in zil_crash() when we move them
from zl_lwb_list to zl_lwb_crash_list, so we know what's going on when
they return.
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#17622
I'm soon about to need another LWB flag, and boolean_t is just so big
for only storing a single bit. Changing to a bitfield is far less
wasteful.
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#17622
The vast majority of calls to zil_commit() follow VFS ops, and should
honour the failmode= setting - either wait for sync, or return error.
Some calls however are part of a larger syncing op, and shouldn't ever
block if something goes wrong.
To allow this, we introduce zil_commit_flags(), with a flag
ZIL_COMMIT_FAILMODE to indicate whether or not the pool failmode should
be honoured. zil_commit() is now a wrapper that always sets this flag,
but any caller wanting a different behaviour can request ZIL_COMMIT_NOW
instead to have the call return failure if the pool suspends, regardless
of the failmode= setting.
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#17398
If the ZIL runs into trouble, it calls txg_wait_synced(), which blocks
on suspend. We want it to not block on suspend, instead returning an
error. On the surface, this is simple: change all calls to
txg_wait_synced_flags(TXG_WAIT_SUSPEND), and then thread the error
return back to the zil_commit() caller.
Handling suspension means returning an error to all commit waiters. This
is relatively straightforward, as zil_commit_waiter_t already has
zcw_zio_error to hold the write IO error, which signals a fallback to
txg_wait_synced_flags(TXG_WAIT_SUSPEND), which will fail, and so the
waiter can now return an error from zil_commit().
However, commit waiters are normally signalled when their associated
write (LWB) completes. If the pool has suspended, those IOs may not
return for some time, or maybe not at all. We still want to signal those
waiters so they can return from zil_commit(). We have a list of those
in-flight LWBs on zl_lwb_list, so we can run through those, detach them
and signal them. The LWB itself is still in-flight, but no longer has
attached waiters, so when it returns there will be nothing to do.
(As an aside, ITXs can also supply completion callbacks, which are
called when they are destroyed. These are directly connected to LWBs
though, so are passed the error code and destroyed there too).
At this point, all ZIL waiters have been ejected, so we only have to
consider the internal state. We potentially still have ITXs that have
not been committed, LWBs still open, and LWBs in-flight. The on-disk ZIL
is in an unknown state; some writes may have been written but not
returned to us. We really can't rely on any of it; the best thing to do
is abandon it entirely and start over when the pool returns to service.
But, since we may have IO out that won't return until the pool resumes,
we need something for it to return to.
The simplest solution I could find, implemented here, is to "crash" the
ZIL: accept no new ITXs, make no further updates, and let it empty out
on its normal schedule, that is, as txgs complete and zil_sync() and
zil_clean() are called. We set a "restart txg" to three txgs in the
future (syncing + TXG_CONCURRENT_STATES), at which point all the
internal state will have been cleared out, and the ZIL can resume
operation (handled at the top of zil_clean()).
This commit adds zil_crash(), which handles all of the above:
- sets the restart txg
- capture and signal all waiters
- zero the header
zil_crash() is called when txg_wait_synced_flags(TXG_WAIT_SUSPEND)
returns because the pool suspended (ESHUTDOWN).
The rest of the commit is just threading the errors through, and related
housekeeping.
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#17398
ITX callbacks are used to signal that something can be cleaned up after
a itx is committed. Presently that's only used when syncing out mapped
pages (msync()) to mark dirty pages clean.
This extends the callback interface so it can be passed an error, and
take a different cleanup action if necessary.
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#17398
This changes zil_commit() to have an int return, and updates all callers
to check it. There are no corresponding internal changes yet; it will
always return 0.
Since zil_commit() is an indication that the caller _really_ wants the
associated data to be durability stored, I've annotated it with the
__warn_unused_result__ compiler attribute (via __must_check), to emit a
warning if it's ever ussd without doing something with the return code.
I hope this will mean we never misuse it in the future.
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#17398
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/Closes#17591
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/Closes#17591
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <alexander.motin@TrueNAS.com>
Signed-off-by: Rob Norris <robn@despairlabs.com>
Sponsored-by: https://despairlabs.com/sponsor/Closes#17591
During regular block writes ZFS sets both logical and physical
birth times equal to the current TXG. During dedup and block
cloning logical birth time is still set to the current TXG, but
physical may be copied from the original block that was used.
This represents the fact that logically user data has changed,
but the physically it is the same old block.
But block rewrite introduces a new situation, when block is not
changed logically, but stored in a different place of the pool.
From ARC, scrub and some other perspectives this is a new block,
but for example for user applications or incremental replication
it is not. Somewhat similar thing happen during remap phase of
device removal, but in that case space blocks are still acounted
as allocated at their logical birth times.
This patch introduces a new "rewrite" flag in the block pointer
structure, allowing to differentiate physical rewrite (when the
block is actually reallocated at the physical birth time) from
the device reval case (when the logical birth time is used).
The new functionality is not used at this point, and the only
expected change is that error log is now kept in terms of physical
physical birth times, rather than logical, since if a block with
logged error was somehow rewritten, then the previous error does
not matter any more.
This change also introduces a new TRAVERSE_LOGICAL flag to the
traverse code, allowing zfs send, redact and diff to work in
context of logical birth times, ignoring physical-only rewrites.
It also changes nothing at this point due to lack of those writes,
but they will come in a following patch.
Reviewed-by: Rob Norris <robn@despairlabs.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <alexander.motin@TrueNAS.com>
Closes#17565
Before this change ZIL blocks were allocated only from normal or
SLOG vdevs. In typical situation when special vdevs are SSDs and
normal are HDDs it could cause weird inversions when data blocks
are written to SSDs, but ZIL referencing them to HDDs.
This change assumes that special vdevs typically have much better
(or at least not worse) latency than normal, and so in absence of
SLOGs should store ZIL blocks. It means similar to normal vdevs
introduction of special embedded log allocation class and updating
the allocation fallback order to: SLOG -> special embedded log ->
special -> normal embedded log -> normal.
The code tries to guess whether data block is going to be written
to normal or special vdev (it can not be done precisely before
compression) and prefer indirect writes for blocks written to a
special vdev to avoid double-write. For blocks that are going to
be written to normal vdev, special vdev by default plays as SLOG,
reducing write latency by the cost of higher special vdev wear,
but it is tunable via module parameter.
This should allow HDD pools with decent SSD as special vdev to
work under synchronous workloads without requiring additional
SLOG SSD, impractical in many scenarios.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Rob Norris <rob.norris@klarasystems.com>
Reviewed-by: Paul Dagnelie <paul.dagnelie@klarasystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#17505
Under parallel workloads ZIL may delay writes of open LWBs that
are not full enough. On suspend we do not expect anything new to
appear since zil_get_commit_list() will not let it pass, only
returning TXG number to wait for. But I suspect that waiting for
the TXG commit without having the last LWB issued may not wait for
its completion, resulting in panic described in #17509.
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Rob Norris <rob.norris@klarasystems.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#17521
ZIL introduced dependencies between its write ZIOs to permit flush
defer, when we flush vdev caches only once all the write ZIOs has
completed. But it was recently spotted that it serializes not only
ZIO completions handling, but also their ready stage. It means ZIO
pipeline can't calculate checksums for the following ZIOs until all
the previous are checksumed, even though it is not required. On a
systems where memory throughput of a single CPU core is limited,
it creates single-core CPU bottleneck, which is difficult to see
due to ZIO pipeline design with many taskqueue threads.
While it would be great to bypass the ready stage waits, it would
require changes to ZIO code, and I haven't found a clean way to do
it. But I've noticed that we don't need any dependency between
the write ZIOs if the previous one has some waiters, which means
it won't defer any flushes and work as a barrier for the earlier
ones.
Bypassing it won't help large single-thread writes, since all the
write ZIOs except the last in that case won't have waiters, and
so will be dependent. But in that case the ZIO processing might
not be a bottleneck, since there will be only one thread populating
the write buffers, that will likely be the bottleneck.
But bypassing the ZIO dependency on multi-threaded write workloads
really allows them to scale beyond the checksuming throughput of
one CPU core.
My tests with writing 12 files on a same dataset on a pool with
4 striped NVMes as SLOGs from 12 threads with 1MB blocks on a
system with Xeon Silver 4114 CPU show total throughput increase
from 4.3GB/s to 8.5GB/s, increasing the SLOGs busy from ~30% to
~70%.
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#17458
This is the cheap way to keep non-user functions working after
break-on-suspend becomes default.
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: Paul Dagnelie <paul.dagnelie@klarasystems.com>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#17355
This helps to avoids confusion with the similarly-named
txg_wait_synced().
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Mariusz Zaborski <mariusz.zaborski@klarasystems.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Sponsored-by: https://despairlabs.com/sponsor/
Signed-off-by: Rob Norris <robn@despairlabs.com>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Make sure log record don't stray beyond valid memory region.
There is a lack of verification of the space occupied by fixed members
of lr_t in the zil_parse.
We can create a crafted image to trigger an out of bounds read by
following these steps:
1) Do some file operations and reboot to simulate abnormal exit
without umount
2) zil_chain.zc_nused: 0x1000
3) First lr_t
lr_t.lrc_txtype: 0x0
lr_t.lrc_reclen: 0x1000-0xb8-0x1
lr_t.lrc_txg: 0x0
lr_t.lrc_seq: 0x1
4) Update checksum in zil_chain.zc_eck
Fix:
Add some checks to make sure the remaining bytes are large enough to
hold an log record.
Signed-off-by: XDTG <click1799@163.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
There's no good way to tell when a ZIL commit fails and falls back to a
transaction sync, other than perhaps a throughput drop. This adds
counters so we can see when it happens and why.
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Tony Hutter <hutter2@llnl.gov>
There's no other options, so we can just always assume its a flush.
Includes some light refactoring where a switch statement was doing
control flow that no longer works.
Sponsored-by: Klara, Inc.
Sponsored-by: Wasabi Technology, Inc.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Signed-off-by: Rob Norris <rob.norris@klarasystems.com>
Closes#16064
There exist a couple of macros that are used to update the blkptr birth
times but they can often be confusing. For example, the
BP_PHYSICAL_BIRTH() macro will provide either the physical birth time
if it is set or else return back the logical birth time. The
complement to this macro is BP_SET_BIRTH() which will set the logical
birth time and set the physical birth time if they are not the same.
Consumers may get confused when they are trying to get the physical
birth time and use the BP_PHYSICAL_BIRTH() macro only to find out that
the logical birth time is what is actually returned.
This change cleans up these macros and makes them symmetrical. The same
functionally is preserved but the name is changed. Instead of calling
BP_PHYSICAL_BIRTH(), consumer can now call BP_GET_BIRTH(). In
additional to cleaning up this naming conventions, two new sets of
macros are introduced -- BP_[SET|GET]_LOGICAL_BIRTH() and
BP_[SET|GET]_PHYSICAL_BIRTH. These new macros allow the consumer to
get and set the specific birth time.
As part of the cleanup, the unused GRID macros have been removed and
that portion of the blkptr are currently unused.
Reviewed-by: Matthew Ahrens <mahrens@delphix.com>
Reviewed-by: Alexander Motin <mav@FreeBSD.org>
Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Signed-off-by: George Wilson <gwilson@delphix.com>
Closes#15962
Track history in context of bursts, not individual log blocks. It
allows to not blow away all the history by single large burst of
many block, and same time allows optimizations covering multiple
blocks in a burst and even predicted following burst. For each
burst account its optimal block size and minimal first block size.
Use that statistics from the last 8 bursts to predict first block
size of the next burst.
Remove predefined set of block sizes. Allocate any size we see fit,
multiple of 4KB, as required by ZIL now. With compression enabled
by default, ZFS already writes pretty random block sizes, so this
should not surprise space allocator any more.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15635
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
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
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
- Generalize vdev_nowritecache handling by traversing through the
VDEV tree and skipping children ZIOs where not supported.
- Remove intermediate zio_null() in case of several VDEV children.
- Remove children handling from zio_ioctl(). There are no other
use cases for this code beside DKIOCFLUSHWRITECACHED, and would there
be, I doubt they would so straightforward apply to all VDEV children.
Comparing to removed previous optimization this should improve cases
of redundant ZILs/SLOGs.
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#15515
zio_root() has no arguments for ready callback or parent ZIO. Except
one recent case in ZIL code if root ZIOs ever have a parent it is
also a root ZIO. It means we do not need READY pipeline stage for
them, which takes some time to process, but even more time to wait
for the children and be woken by them, and both for no good reason.
The most visible effect of this change is that it avoids one taskq
wakeup per ZIL block written, previously used to run zio_ready()
for lwb_root_zio and skipped now.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15398
... by checking that previous block is fully written and flushed.
It allows to skip commit delays since we can give up on aggregation
in that case. This removes zil_min_commit_timeout parameter, since
for single-threaded workloads it is not needed at all, while on very
fast devices even some multi-threaded workloads may get detected as
single-threaded and still bypass the wait. To give multi-threaded
workloads more aggregation chances increase zfs_commit_timeout_pct
from 5 to 10%, as they should suffer less from additional latency.
Also single-threaded workloads detection allows in perspective better
prediction of the next block size.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Prakash Surya <prakash.surya@delphix.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15381
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
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
zil_lwb_set_zio_dependency() can not set write ZIO dependency on
previous LWB's write ZIO if one is already in done handler and set
state to LWB_STATE_WRITE_DONE. So theoretically done handler of
next LWB's write ZIO may run before done handler of previous LWB
write ZIO completes. In such case we can not defer flushes, since
the flush issue process is not locked.
This may fix some reported assertions of lwb_vdev_tree not being
empty inside zil_free_lwb().
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15278
In zil_lwb_write_issue(), after issuing lwb_root_zio/lwb_write_zio,
we have no right to access lwb->lwb_child_zio. If it was not there,
the first two ZIOs may have already completed and freed the lwb.
ZIOs issue in opposite order from children to parent should keep
the lwb valid till the end, since the lwb can be freed only after
lwb_root_zio completion callback.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15233
While I have no reports of it, I suspect possible use-after-free
scenario when zil_commit_waiter() tries to dereference zcw_lwb
for lwb already freed by zil_sync(), while zcw_done is not set.
Extension of zl_lock scope as it was originally should block
zil_sync() from freeing the lwb, closing this race.
This reverts #14959 and couple chunks of #14841.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15228
In zil_free_lwb() we should first assert lwb_state or the rest of
assertions can be misleading if it is false.
Add lwb_state assertions in zil_lwb_add_block() to make sure we are
not trying to add elements to lwb_vdev_tree after it was processed.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15227
The previous patch #14841 appeared to have significant flaw, causing
deadlocks if zl_get_data callback got blocked waiting for TXG sync. I
already handled some of such cases in the original patch, but issue
#14982 shown cases that were impossible to solve in that design.
This patch fixes the problem by postponing log blocks allocation till
the very end, just before the zios issue, leaving nothing blocking after
that point to cause deadlocks. Before that point though any sleeps are
now allowed, not causing sync thread blockage. This require slightly
more complicated lwb state machine to allocate blocks and issue zios
in proper order. But with removal of special early issue workarounds
the new code is much cleaner now, and should even be more efficient.
Since this patch uses null zios between write, I've found that null
zios do not wait for logical children ready status in zio_ready(),
that makes parent write to proceed prematurely, producing incorrect
log blocks. Added ZIO_CHILD_LOGICAL_BIT to zio_wait_for_children()
fixes it.
Reviewed-by: Rob Norris <rob.norris@klarasystems.com>
Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Reviewed-by: George Wilson <george.wilson@delphix.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15122
If we get next block allocation error during log write, we trigger
transaction commit. But the block we have just completed is still
written and transactions it covers will be acknowledged normally.
If after that we ignore the block during replay just because it is
the last in the chain, we may not replay some transactions that we
have acknowledged as synced, that is not right.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15132
Fastwrite was introduced many years ago to improve ZIL writes spread
between multiple top-level vdevs by tracking number of allocated but
not written blocks and choosing vdev with smaller count. It suposed
to reduce ZIL knowledge about allocation, but actually made ZIL to
even more actively report allocation code about the allocations,
complicating both ZIL and metaslabs code.
On top of that, it seems ZIO_FLAG_FASTWRITE setting in dmu_sync()
was lost many years ago, that was one of the declared benefits. Plus
introduction of embedded log metaslab class solved another problem
with allocation rotor accounting both normal and log allocations,
since in most cases those are now in different metaslab classes.
After all that, I'd prefer to simplify already too complicated ZIL,
ZIO and metaslab code if the benefit of complexity is not obvious.
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#15107
This locking was recently added as part of #14979. But appears it
is illegal to take zl_issuer_lock while holding dp_config_rwlock,
taken by dsl_pool_hold(). It causes deadlock with sync thread in
spa_sync_upgrades(). On a second thought, we should not
need this locking, since zil_commit_impl() we call below takes
zl_issuer_lock, that should sufficiently protect zl_suspend reads,
combined with other logic from #14979.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15103
When we have some LWBs closed and their ZIOs ready to be issued, we
can not afford sleeping on config lock if somebody else try to lock
it as writer, or it will cause a deadlock.
To solve it, move spa_config_enter() from zil_lwb_write_issue() to
zil_lwb_write_close() under zl_issuer_lock to enforce lock ordering
with other threads. Now if we can't immediately lock config, issue
all previously closed LWBs so that they could drop their config
locks after completion, and only then allow sleeping on our lock.
Reviewed-by: Mark Maybee <mark.maybee@delphix.com>
Reviewed-by: Prakash Surya <prakash.surya@delphix.com>
Reviewed-by: George Wilson <george.wilson@delphix.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#15078Closes#15080
With zl_suspend read in zil_commit() not protected by any locks it
is possible for new ZIL writes to be in progress while zil_destroy()
called by zil_suspend() freeing them. This patch closes the race
by taking zl_issuer_lock in zil_suspend() and adding the second
zl_suspend check to zil_get_commit_list(), protected by the lock.
It allows all already queued transactions to be logged normally,
while blocks any new ones, calling txg_wait_synced() for the TXGs.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#14979
lwb->lwb_issued_txg can not be accessed after lwb_state is set to
LWB_STATE_FLUSH_DONE and zl_lock is dropped, since the lwb may be
freed by zil_sync(). We must save the txg number before that.
This is similar to the 55b1842f92, but as I see the bug is not new.
It existed for quite a while, just was not triggered due to smaller
race window.
Reviewed-by: Allan Jude <allan@klarasystems.com>
Reviewed-by: Brian Atkinson <batkinson@lanl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#14988Closes#14999
482da24e2 missed arc_buf_destroy() calls on log parse errors, possibly
leaking up to 128KB of memory per dataset during ZIL replay.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#14987
We are not allowed to access lwb after setting LWB_STATE_FLUSH_DONE
state and dropping zl_lock, since it may be freed by zil_sync().
To free itxs and waiters after dropping the lock we need to move
lwb_itxs and lwb_waiters lists elements to local storage.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#14957Closes#14959
There seems to be no reason for ZIL blocks to be limited by 128KB
other than replay code is written in such a way. This change does
not increase the limit yet, just removes the artificial limitation.
Avoided extra memcpy() may save us a second during replay.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Prakash Surya <prakash.surya@delphix.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#14910
In addition to a number of actual log bytes written, account also a
total written bytes including padding and total allocated bytes (bytes
<= write <= alloc). It should allow to monitor zil traffic and space
efficiency.
Add dtrace probe for zil block size selection.
Make zilstat report more information and fit it into less width.
Reviewed-by: Ameer Hamza <ahamza@ixsystems.com>
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#14863
Before this change ZIL copied all log data while holding the lock.
It caused huge lock contention on workloads with many big parallel
writes. This change splits the process into two parts: first,
zil_lwb_assign() estimates the log space needed for all transactions,
and zil_lwb_write_close() allocates blocks and zios while holding the
lock, then, after the lock in dropped, zil_lwb_commit() copies the
data, and zil_lwb_write_issue() issues the I/Os.
Also while there slightly reduce scope of zl_lock.
Reviewed-by: Paul Dagnelie <pcd@delphix.com>
Reviewed-by: Prakash Surya <prakash.surya@delphix.com>
Reviewed-by: Richard Yao <richard.yao@alumni.stonybrook.edu>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#14841
There is no sense to keep that memory allocated during the flush.
Reviewed-by: Brian Behlendorf <behlendorf1@llnl.gov>
Reviewed-by: Prakash Surya <prakash.surya@delphix.com>
Signed-off-by: Alexander Motin <mav@FreeBSD.org>
Sponsored by: iXsystems, Inc.
Closes#14855
