| Age | Commit message (Collapse) | Author |
|---|
|
Currently this function handles both the READ and WRITE dio cases. This
is facilitated by a bunch of 'if' statements, a goto short-circuit
statement and a very perverse aliasing of "!created"(READ) case
by setting lockstart = lockend and checking for lockstart < lockend for
detecting the write. Let's simplify this mess by extracting the
READ-only code into a separate __btrfs_get_block_direct_read function.
This is only the first step, the next one will be to factor out the
write side as well. The end goal will be to have the common locking/
unlocking code in btrfs_get_blocks_direct and then it will call either
the read|write subvariants. No functional changes.
Signed-off-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
tg in throtl_select_dispatch is used first and then do check. Since tg
may be NULL, it has potential NULL pointer dereference risk. So fix
it.
Signed-off-by: Joseph Qi <joseph.qi@linux.alibaba.com>
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
Currently, kyber is very unfriendly with merging. kyber depends
on ctx rq_list to do merging, however, most of time, it will not
leave any requests in ctx rq_list. This is because even if tokens
of one domain is used up, kyber will try to dispatch requests
from other domain and flush the rq_list there.
To improve this, we setup kyber_ctx_queue (kcq) which is similar
with ctx, but it has rq_lists for different domain and build same
mapping between kcq and khd as the ctx & hctx. Then we could merge,
insert and dispatch for different domains separately. At the same
time, only flush the rq_list of kcq when get domain token successfully.
Then if one domain token is used up, the requests could be left in
the rq_list of that domain and maybe merged with following io.
Following is my test result on machine with 8 cores and NVMe card
INTEL SSDPEKKR128G7
fio size=256m ioengine=libaio iodepth=64 direct=1 numjobs=8
seq/random
+------+---------------------------------------------------------------+
|patch?| bw(MB/s) | iops | slat(usec) | clat(usec) | merge |
+----------------------------------------------------------------------+
| w/o | 606/612 | 151k/153k | 6.89/7.03 | 3349.21/3305.40 | 0/0 |
+----------------------------------------------------------------------+
| w/ | 1083/616 | 277k/154k | 4.93/6.95 | 1830.62/3279.95 | 223k/3k |
+----------------------------------------------------------------------+
When set numjobs to 16, the bw and iops could reach 1662MB/s and 425k
on my platform.
Signed-off-by: Jianchao Wang <jianchao.w.wang@oracle.com>
Tested-by: Holger Hoffstätte <holger@applied-asynchrony.com>
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
|
|
No functional changes in this patch, just a prep patch for utilizing
this in an IO scheduler.
Signed-off-by: Jens Axboe <axboe@kernel.dk>
Reviewed-by: Omar Sandoval <osandov@fb.com>
|
|
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Atul Gupta <atul.gupta@chelsio.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
- unindented continue
- check for null page
- signed return
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Atul Gupta <atul.gupta@chelsio.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
skb dereferenced before check in sendpage
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Atul Gupta <atul.gupta@chelsio.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
address suspicious code <gustavo@embeddedor.com>
1210 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1211 }
The issue is that in the code above, set_bit is never reached
due to the 'continue' statement at line 1208.
Also reported by bug report:<dan.carpenter@oracle.com>
1210 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Not reachable.
Its required to wait for buffer in the send path and takes care of
unaddress and un-handled SOCK_NOSPACE.
v2: use csk_mem_free where appropriate
proper indent of goto do_nonblock
replace out with do_rm_wq
Reported-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Atul Gupta <atul.gupta@chelsio.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
corrected the key length to copy 128b key. Removed 192b and 256b
key as user input supports key of size 128b in gcm_ctx
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Atul Gupta <atul.gupta@chelsio.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
This reverts commit eb772f37ae8163a89e28a435f6a18742ae06653b, as now the
x86 Salsa20 implementation has been removed and the generic helpers are
no longer needed outside of salsa20_generic.c.
We could keep this just in case someone else wants to add a new
optimized Salsa20 implementation. But given that we have ChaCha20 now
too, I think it's unlikely. And this can always be reverted back.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
The x86 assembly implementations of Salsa20 use the frame base pointer
register (%ebp or %rbp), which breaks frame pointer convention and
breaks stack traces when unwinding from an interrupt in the crypto code.
Recent (v4.10+) kernels will warn about this, e.g.
WARNING: kernel stack regs at 00000000a8291e69 in syzkaller047086:4677 has bad 'bp' value 000000001077994c
[...]
But after looking into it, I believe there's very little reason to still
retain the x86 Salsa20 code. First, these are *not* vectorized
(SSE2/SSSE3/AVX2) implementations, which would be needed to get anywhere
close to the best Salsa20 performance on any remotely modern x86
processor; they're just regular x86 assembly. Second, it's still
unclear that anyone is actually using the kernel's Salsa20 at all,
especially given that now ChaCha20 is supported too, and with much more
efficient SSSE3 and AVX2 implementations. Finally, in benchmarks I did
on both Intel and AMD processors with both gcc 8.1.0 and gcc 4.9.4, the
x86_64 salsa20-asm is actually slightly *slower* than salsa20-generic
(~3% slower on Skylake, ~10% slower on Zen), while the i686 salsa20-asm
is only slightly faster than salsa20-generic (~15% faster on Skylake,
~20% faster on Zen). The gcc version made little difference.
So, the x86_64 salsa20-asm is pretty clearly useless. That leaves just
the i686 salsa20-asm, which based on my tests provides a 15-20% speed
boost. But that's without updating the code to not use %ebp. And given
the maintenance cost, the small speed difference vs. salsa20-generic,
the fact that few people still use i686 kernels, the doubt that anyone
is even using the kernel's Salsa20 at all, and the fact that a SSE2
implementation would almost certainly be much faster on any remotely
modern x86 processor yet no one has cared enough to add one yet, I don't
think it's worthwhile to keep.
Thus, just remove both the x86_64 and i686 salsa20-asm implementations.
Reported-by: syzbot+ffa3a158337bbc01ff09@syzkaller.appspotmail.com
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
The GET_ID command, added as of SEV API v0.16, allows the SEV firmware
to be queried about a unique CPU ID. This unique ID can then be used
to obtain the public certificate containing the Chip Endorsement Key
(CEK) public key signed by the AMD SEV Signing Key (ASK).
For more information please refer to "Section 5.12 GET_ID" of
https://support.amd.com/TechDocs/55766_SEV-KM%20API_Specification.pdf
Signed-off-by: Janakarajan Natarajan <Janakarajan.Natarajan@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
The DOWNLOAD_FIRMWARE command, added as of SEV API v0.15, allows the OS
to install SEV firmware newer than the currently active SEV firmware.
For the new SEV firmware to be applied it must:
* Pass the validation test performed by the existing firmware.
* Be of the same build or a newer build compared to the existing firmware.
For more information please refer to "Section 5.11 DOWNLOAD_FIRMWARE" of
https://support.amd.com/TechDocs/55766_SEV-KM%20API_Specification.pdf
Signed-off-by: Janakarajan Natarajan <Janakarajan.Natarajan@amd.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
Signed-off-by: Conor McLoughlin <conor.mcloughlin@intel.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
The cache parameter register configuration was being too verbose.
Use dev_dbg() to only provide the information if needed.
Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
Use managed clock handling, differentiate between no clock (possibly OK)
and clock init failure (never OK) and correctly handle clock detection
being deferred.
Suggested-by: Geert Uytterhoeven <geert@linux-m68k.org>
Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
The product signature and HW revision register have different offset on the
older HW revisions.
This fixes the problem of the driver failing sanity check on silicon
despite working on the FPGA emulation systems.
Fixes: 27b3b22dd98c ("crypto: ccree - add support for older HW revs")
Cc: stable@vger.kernel.org
Signed-off-by: Gilad Ben-Yossef <gilad@benyossef.com>
Reviewed-by: Simon Horman <horms+renesas@verge.net.au>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
Extends memory required for IV to include B0 Block and DMA map in
single operation.
Signed-off-by: Harsh Jain <harsh@chelsio.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
Send IV in WR as immediate instead of dma mapped entry for cipher.
Signed-off-by: Harsh Jain <harsh@chelsio.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
Change the return type based on following patch
https://www.mail-archive.com/linux-crypto@vger.kernel.org/msg28552.html
Signed-off-by: Harsh Jain <harsh@chelsio.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
Coverity warns about an
"Unintentional integer overflow (OVERFLOW_BEFORE_WIDEN)"
when computing the congestion threshold value.
Even though it is highly unlikely for an overflow to happen,
use this as an opportunity to simplify the code.
Signed-off-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
In some cases the CCB DMA-based internal transfer started by the MOVE
command (src=M3 register, dst=descriptor buffer) does not finish
in time and DECO executes the unpatched descriptor.
This leads eventually to a DECO Watchdog Timer timeout error.
To make sure the transfer ends, change the MOVE command to be blocking.
Signed-off-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
Management Complex (MC) f/w detection is based on CTPR_MS[DPAA2] bit.
This is incorrect since:
-the bit is set for all CAAM blocks integrated in SoCs with a certain
Layerscape Chassis
-some SoCs with LS Chassis don't have an MC block (thus no MC f/w)
To fix this, MC f/w detection will be based on the presence of
"fsl,qoriq-mc" compatible string in the device tree.
Fixes: 297b9cebd2fc0 ("crypto: caam/jr - add support for DPAA2 parts")
Signed-off-by: Horia Geantă <horia.geanta@nxp.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
Several source files have been taken from OpenSSL. In some of them a
comment that "permission to use under GPL terms is granted" was
included below a contradictory license statement. In several cases,
there was no indication that the license of the code was compatible
with the GPLv2.
This change clarifies the licensing for all of these files. I've
confirmed with the author (Andy Polyakov) that a) he has licensed the
files with the GPLv2 comment under that license and b) that he's also
happy to license the other files under GPLv2 too. In one case, the
file is already contained in his CRYPTOGAMS bundle, which has a GPLv2
option, and so no special measures are needed.
In all cases, the license status of code has been clarified by making
the GPLv2 license prominent.
The .S files have been regenerated from the updated .pl files.
This is a comment-only change. No code is changed.
Signed-off-by: Adam Langley <agl@chromium.org>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
Commit 56e8e57fc3a7 ("crypto: morus - Add common SIMD glue code for
MORUS") accidetally consiedered the glue code to be usable by different
architectures, but it seems to be only usable on x86.
This patch moves it under arch/x86/crypto and adds 'depends on X86' to
the Kconfig options and also removes the prompt to hide these internal
options from the user.
Reported-by: kbuild test robot <lkp@intel.com>
Signed-off-by: Ondrej Mosnacek <omosnacek@gmail.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
Currently testmgr has separate encryption and decryption test vectors
for symmetric ciphers. That's massively redundant, since with few
exceptions (mostly mistakes, apparently), all decryption tests are
identical to the encryption tests, just with the input/result flipped.
Therefore, eliminate the redundancy by removing the decryption test
vectors and updating testmgr to test both encryption and decryption
using what used to be the encryption test vectors. Naming is adjusted
accordingly: each cipher_testvec now has a 'ptext' (plaintext), 'ctext'
(ciphertext), and 'len' instead of an 'input', 'result', 'ilen', and
'rlen'. Note that it was always the case that 'ilen == rlen'.
AES keywrap ("kw(aes)") is special because its IV is generated by the
encryption. Previously this was handled by specifying 'iv_out' for
encryption and 'iv' for decryption. To make it work cleanly with only
one set of test vectors, put the IV in 'iv', remove 'iv_out', and add a
boolean that indicates that the IV is generated by the encryption.
In total, this removes over 10000 lines from testmgr.h, with no
reduction in test coverage since prior patches already copied the few
unique decryption test vectors into the encryption test vectors.
This covers all algorithms that used 'struct cipher_testvec', e.g. any
block cipher in the ECB, CBC, CTR, XTS, LRW, CTS-CBC, PCBC, OFB, or
keywrap modes, and Salsa20 and ChaCha20. No change is made to AEAD
tests, though we probably can eliminate a similar redundancy there too.
The testmgr.h portion of this patch was automatically generated using
the following awk script, with some slight manual fixups on top (updated
'struct cipher_testvec' definition, updated a few comments, and fixed up
the AES keywrap test vectors):
BEGIN { OTHER = 0; ENCVEC = 1; DECVEC = 2; DECVEC_TAIL = 3; mode = OTHER }
/^static const struct cipher_testvec.*_enc_/ { sub("_enc", ""); mode = ENCVEC }
/^static const struct cipher_testvec.*_dec_/ { mode = DECVEC }
mode == ENCVEC && !/\.ilen[[:space:]]*=/ {
sub(/\.input[[:space:]]*=$/, ".ptext =")
sub(/\.input[[:space:]]*=/, ".ptext\t=")
sub(/\.result[[:space:]]*=$/, ".ctext =")
sub(/\.result[[:space:]]*=/, ".ctext\t=")
sub(/\.rlen[[:space:]]*=/, ".len\t=")
print
}
mode == DECVEC_TAIL && /[^[:space:]]/ { mode = OTHER }
mode == OTHER { print }
mode == ENCVEC && /^};/ { mode = OTHER }
mode == DECVEC && /^};/ { mode = DECVEC_TAIL }
Note that git's default diff algorithm gets confused by the testmgr.h
portion of this patch, and reports too many lines added and removed.
It's better viewed with 'git diff --minimal' (or 'git show --minimal'),
which reports "2 files changed, 919 insertions(+), 11723 deletions(-)".
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
One "kw(aes)" decryption test vector doesn't exactly match an encryption
test vector with input and result swapped. In preparation for removing
the decryption test vectors, add this test vector to the encryption test
vectors, so we don't lose any test coverage.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
None of the four "ecb(tnepres)" decryption test vectors exactly match an
encryption test vector with input and result swapped. In preparation
for removing the decryption test vectors, add these to the encryption
test vectors, so we don't lose any test coverage.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
One "cbc(des)" decryption test vector doesn't exactly match an
encryption test vector with input and result swapped. It's *almost* the
same as one, but the decryption version is "chunked" while the
encryption version is "unchunked". In preparation for removing the
decryption test vectors, make the encryption one both chunked and
unchunked, so we don't lose any test coverage.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
Two "ecb(des)" decryption test vectors don't exactly match any of the
encryption test vectors with input and result swapped. In preparation
for removing the decryption test vectors, add these to the encryption
test vectors, so we don't lose any test coverage.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Herbert Xu <herbert@gondor.apana.org.au>
|
|
When the call to hw_params for a component fails, the error code is held
by the variable '__ret' but the error message displays the value held by
the variable 'ret'. Fix the return code shown when hw_params fails for
a component.
Fixes: b8135864d4d3 ("ASoC: snd_soc_component_driver has snd_pcm_ops")
Signed-off-by: Jon Hunter <jonathanh@nvidia.com>
Signed-off-by: Mark Brown <broonie@kernel.org>
|
|
The error code does not match the reason of failure and may confuse the
callers.
Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux
Pull s390 fixes from Martin Schwidefsky:
- a missing -msoft-float for the compile of the kexec purgatory
- a fix for the dasd driver to avoid the double use of a field in the
'struct request'
[ That latter one is being discussed, and Christoph asked for something
cleaner, but for now it's a fix ]
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
s390/dasd: use blk_mq_rq_from_pdu for per request data
s390/purgatory: Fix endless interrupt loop
|
|
In the quest to remove all stack VLA usage from the kernel[1], this
allocates the working buffers during regular init, instead of using stack
space. This refactors the allocation code a bit to make it easier
to review.
[1] https://lkml.kernel.org/r/CA+55aFzCG-zNmZwX4A2FQpadafLfEzK6CC=qPXydAacU1RqZWA@mail.gmail.com
Signed-off-by: Kees Cook <keescook@chromium.org>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
If one of the backup superblocks is found to differ seriously from
superblock 0, write out a fresh copy from the in-core sb.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
|
|
Add a helper routine to attach quota information to inodes that are
about to undergo repair. If that fails, we need to schedule a
quotacheck for the next mount but allow the corrupted metadata repair to
continue.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
|
|
Add a helper function to help us recover btree roots from the rmap data.
Callers pass in a list of rmap owner codes, buffer ops, and magic
numbers. We iterate the rmap records looking for owner matches, and
then read the matching blocks to see if the magic number & uuid match.
If so, we then read-verify the block, and if that passes then we retain
a pointer to the block with the highest level, assuming that by the end
of the call we will have found the root. This will be used to reset the
AGF/AGI btree root fields during their rebuild procedures.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
|
|
Now that we've plumbed in the ability to construct a list of dead btree
blocks following a repair, add more helpers to dispose of them. This is
done by examining the rmapbt -- if the btree was the only owner we can
free the block, otherwise it's crosslinked and we can only remove the
rmapbt record.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
|
|
Add some helpers to assemble a list of fs block extents. Generally,
repair functions will iterate the rmapbt to make a list (1) of all
extents owned by the nominal owner of the metadata structure; then they
will iterate all other structures with the same rmap owner to make a
list (2) of active blocks; and finally we have a subtraction function to
subtract all the blocks in (2) from (1), with the result that (1) is now
a list of blocks that were owned by the old btree and must be disposed.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
|
|
Add a pair of helper functions to allocate and initialize fresh btree
roots. The repair functions will use these as part of recreating
corrupted metadata.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
|
|
For repairs, we need to reserve at least as many blocks as we think
we're going to need to rebuild the data structure, and we're going to
need some helpers to roll transactions while maintaining locks on the AG
headers so that other threads cannot wander into the middle of a repair.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Allison Henderson <allison.henderson@oracle.com>
|
|
Grab and hold the per-AG data across a scrub run whenever relevant.
This helps us avoid repeated trips through rcu and the radix tree
in the repair code.
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
|
|
In cow_file_range(), create_io_em() may fail, but its return value is
not recorded. Then return value may be 0 even it failed which is a
wrong behavior.
Let cow_file_range() return PTR_ERR(em) if create_io_em() failed.
Fixes: 6f9994dbabe5 ("Btrfs: create a helper to create em for IO")
CC: stable@vger.kernel.org # 4.11+
Signed-off-by: Su Yue <suy.fnst@cn.fujitsu.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
Since there is no more use of qgroup_reserved member in struct
btrfs_pending_snapshot, remove it.
Signed-off-by: Gu JinXiang <gujx@cn.fujitsu.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
Since commit 7775c8184ec0 ("btrfs: remove unused parameter from
btrfs_subvolume_release_metadata") parameter qgroup_reserved is not used
by caller of function btrfs_subvolume_reserve_metadata. So remove it.
Signed-off-by: Gu JinXiang <gujx@cn.fujitsu.com>
Reviewed-by: David Sterba <dsterba@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
[Problem description and how we fix it]
We should balance dirty metadata pages at the end of
btrfs_finish_ordered_io, since a small, unmergeable random write can
potentially produce dirty metadata which is multiple times larger than
the data itself. For example, a small, unmergeable 4KiB write may
produce:
16KiB dirty leaf (and possibly 16KiB dirty node) in subvolume tree
16KiB dirty leaf (and possibly 16KiB dirty node) in checksum tree
16KiB dirty leaf (and possibly 16KiB dirty node) in extent tree
Although we do call balance dirty pages in write side, but in the
buffered write path, most metadata are dirtied only after we reach the
dirty background limit (which by far only counts dirty data pages) and
wakeup the flusher thread. If there are many small, unmergeable random
writes spread in a large btree, we'll find a burst of dirty pages
exceeds the dirty_bytes limit after we wakeup the flusher thread - which
is not what we expect. In our machine, it caused out-of-memory problem
since a page cannot be dropped if it is marked dirty.
Someone may worry about we may sleep in btrfs_btree_balance_dirty_nodelay,
but since we do btrfs_finish_ordered_io in a separate worker, it will not
stop the flusher consuming dirty pages. Also, we use different worker for
metadata writeback endio, sleep in btrfs_finish_ordered_io help us throttle
the size of dirty metadata pages.
[Reproduce steps]
To reproduce the problem, we need to do 4KiB write randomly spread in a
large btree. In our 2GiB RAM machine:
1) Create 4 subvolumes.
2) Run fio on each subvolume:
[global]
direct=0
rw=randwrite
ioengine=libaio
bs=4k
iodepth=16
numjobs=1
group_reporting
size=128G
runtime=1800
norandommap
time_based
randrepeat=0
3) Take snapshot on each subvolume and repeat fio on existing files.
4) Repeat step (3) until we get large btrees.
In our case, by observing btrfs_root_item->bytes_used, we have 2GiB of
metadata in each subvolume tree and 12GiB of metadata in extent tree.
5) Stop all fio, take snapshot again, and wait until all delayed work is
completed.
6) Start all fio. Few seconds later we hit OOM when the flusher starts
to work.
It can be reproduced even when using nocow write.
Signed-off-by: Ethan Lien <ethanlien@synology.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ add comment ]
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
In nocow path, we check if the extent is snapshotted in
btrfs_cross_ref_exist(). We can do the similar check earlier and avoid
unnecessary search into extent tree.
A fio test on a Intel D-1531, 16GB RAM, SSD RAID-5 machine as follows:
[global]
group_reporting
time_based
thread=1
ioengine=libaio
bs=4k
iodepth=32
size=64G
runtime=180
numjobs=8
rw=randwrite
[file1]
filename=/mnt/nocow/testfile
IOPS result: unpatched patched
1 fio round: 46670 46958
snapshot
2 fio round: 51826 54498
3 fio round: 59767 61289
After snapshot, the first fio get about 5% performance gain. As we
continually write to the same file, all writes will resume to nocow mode
and eventually we have no performance gain.
Signed-off-by: Ethan Lien <ethanlien@synology.com>
Reviewed-by: David Sterba <dsterba@suse.com>
[ update comments ]
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
This function always takes a transaction handle which contains a
reference to the fs_info. Use that and remove the extra argument.
Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
[ rename the function ]
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
This function always takes a transaction handle which contains a
reference to the fs_info. Use that and remove the extra argument.
Signed-off-by: Lu Fengqi <lufq.fnst@cn.fujitsu.com>
Reviewed-by: Nikolay Borisov <nborisov@suse.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
|
The check is superfluous since all callers who set search_for_commit
also have skip_locking set.
ASSERT() is put in place to ensure skip_locking is set by new callers.
Reviewed-by: Qu Wenruo <wqu@suse.com>
Signed-off-by: Liu Bo <bo.liu@linux.alibaba.com>
Signed-off-by: David Sterba <dsterba@suse.com>
|
