Pull core block updates from Jens Axboe:
"This first core part of the block IO changes contains:
- Cleanup of the bio IO error signaling from Christoph. We used to
rely on the uptodate bit and passing around of an error, now we
store the error in the bio itself.
- Improvement of the above from myself, by shrinking the bio size
down again to fit in two cachelines on x86-64.
- Revert of the max_hw_sectors cap removal from a revision again,
from Jeff Moyer. This caused performance regressions in various
tests. Reinstate the limit, bump it to a more reasonable size
instead.
- Make /sys/block/<dev>/queue/discard_max_bytes writeable, by me.
Most devices have huge trim limits, which can cause nasty latencies
when deleting files. Enable the admin to configure the size down.
We will look into having a more sane default instead of UINT_MAX
sectors.
- Improvement of the SGP gaps logic from Keith Busch.
- Enable the block core to handle arbitrarily sized bios, which
enables a nice simplification of bio_add_page() (which is an IO hot
path). From Kent.
- Improvements to the partition io stats accounting, making it
faster. From Ming Lei.
- Also from Ming Lei, a basic fixup for overflow of the sysfs pending
file in blk-mq, as well as a fix for a blk-mq timeout race
condition.
- Ming Lin has been carrying Kents above mentioned patches forward
for a while, and testing them. Ming also did a few fixes around
that.
- Sasha Levin found and fixed a use-after-free problem introduced by
the bio->bi_error changes from Christoph.
- Small blk cgroup cleanup from Viresh Kumar"
* 'for-4.3/core' of git://git.kernel.dk/linux-block: (26 commits)
blk: Fix bio_io_vec index when checking bvec gaps
block: Replace SG_GAPS with new queue limits mask
block: bump BLK_DEF_MAX_SECTORS to 2560
Revert "block: remove artifical max_hw_sectors cap"
blk-mq: fix race between timeout and freeing request
blk-mq: fix buffer overflow when reading sysfs file of 'pending'
Documentation: update notes in biovecs about arbitrarily sized bios
block: remove bio_get_nr_vecs()
fs: use helper bio_add_page() instead of open coding on bi_io_vec
block: kill merge_bvec_fn() completely
md/raid5: get rid of bio_fits_rdev()
md/raid5: split bio for chunk_aligned_read
block: remove split code in blkdev_issue_{discard,write_same}
btrfs: remove bio splitting and merge_bvec_fn() calls
bcache: remove driver private bio splitting code
block: simplify bio_add_page()
block: make generic_make_request handle arbitrarily sized bios
blk-cgroup: Drop unlikely before IS_ERR(_OR_NULL)
block: don't access bio->bi_error after bio_put()
block: shrink struct bio down to 2 cache lines again
...
When a write to one of the devices of a RAID5/6 fails, the failure is
recorded in the metadata of the other devices so that after a restart
the data on the failed drive wont be trusted even if that drive seems
to be working again (maybe a cable was unplugged).
Similarly when we record a bad-block in response to a write failure,
we must not let the write complete until the bad-block update is safe.
Currently there is no interlock between the write request completing
and the metadata update. So it is possible that the write will
complete, the app will confirm success in some way, and then the
machine will crash before the metadata update completes.
This is an extremely small hole for a racy to fit in, but it is
theoretically possible and so should be closed.
So:
- set MD_CHANGE_PENDING when requesting a metadata update for a
failed device, so we can know with certainty when it completes
- queue requests that completed when MD_CHANGE_PENDING is set to
only be processed after the metadata update completes
- call raid_end_bio_io() on bios in that queue when the time comes.
Signed-off-by: NeilBrown <neilb@suse.com>
It is possible (though unlikely) for a reshape to be
interrupted between the time that end_reshape is called
and the time when raid5_finish_reshape is called.
This can leave conf->reshape_progress set to MaxSector,
but mddev->reshape_position not.
This combination confused reshape_request() when ->reshape_backwards.
As conf->reshape_progress is so high, it seems the reshape hasn't
really begun. But assuming MaxSector is a valid address only
leads to sorrow.
So ensure reshape_position and reshape_progress both agree,
and add an extra check in reshape_request() just in case they don't.
Signed-off-by: NeilBrown <neilb@suse.com>
While it generally shouldn't happen, it is not impossible for
curr_resync_completed to exceed resync_max.
This can particularly happen when reshaping RAID5 - the current
status isn't copied to curr_resync_completed promptly, so when it
is, it can exceed resync_max.
This happens when the reshape is 'frozen', resync_max is set low,
and reshape is re-enabled.
Taking a difference between two unsigned numbers is always dangerous
anyway, so add a test to behave correctly if
curr_resync_completed > resync_max
Signed-off-by: NeilBrown <neilb@suse.com>
This code is calculating:
writepos, which is the furthest along address (device-space) that we
*will* be writing to
readpos, which is the earliest address that we *could* possible read
from, and
safepos, which is the earliest address in the 'old' section that we
might read from after a crash when the reshape position is
recovered from metadata.
The first is a precise calculation, so clipping at zero doesn't
make sense. As the reshape position is now guaranteed to always be
a multiple of reshape_sectors and as we already BUG_ON when
reshape_progress is zero, there is no point in this min_t() call.
The readpos and safepos are worst case - actual value depends on
precise geometry. That worst case could be negative, which is only
a problem because we are storing the value in an unsigned.
So leave the min_t() for those.
Signed-off-by: NeilBrown <neilb@suse.com>
When reshaping, we work in units of the largest chunk size.
If changing from a larger to a smaller chunk size, that means we
reshape more than one stripe at a time. So the required alignment
of reshape_position needs to take into account both the old
and new chunk size.
This means that both 'here_new' and 'here_old' are calculated with
respect to the same (maximum) chunk size, so testing if they are the
same when delta_disks is zero becomes pointless.
Signed-off-by: NeilBrown <neilb@suse.com>
The chunk_sectors and new_chunk_sectors fields of mddev can be changed
any time (via sysfs) that the reconfig mutex can be taken. So raid5
keeps internal copies in 'conf' which are stable except for a short
locked moment when reshape stops/starts.
So any access that does not hold reconfig_mutex should use the 'conf'
values, not the 'mddev' values.
Several don't.
This could result in corruption if new values were written at awkward
times.
Also use min() or max() rather than open-coding.
Signed-off-by: NeilBrown <neilb@suse.com>
These aren't really needed when no reshape is happening,
but it is safer to have them always set to a meaningful value.
The next patch will use ->prev_chunk_sectors without checking
if a reshape is happening (because that makes the code simpler),
and this patch makes that safe.
Signed-off-by: NeilBrown <neilb@suse.com>
md/raid5 only updates ->reshape_position (which is stored in
metadata and is authoritative) occasionally, but particularly
when getting closed to ->resync_max as it must be correct
when ->resync_max is reached.
When mdadm tries to stop an array which is reshaping it will:
- freeze the reshape,
- set resync_max to where the reshape has reached.
- unfreeze the reshape.
When this happens, the reshape is aborted and then restarted.
The restart doesn't check that resync_max is close, and so doesn't
update ->reshape_position like it should.
This results in the reshape stopping, but ->reshape_position being
incorrect.
So on that first call to reshape_request, make sure ->reshape_position
is updated if needed.
Signed-off-by: NeilBrown <neilb@suse.com>
As generic_make_request() is now able to handle arbitrarily sized bios,
it's no longer necessary for each individual block driver to define its
own ->merge_bvec_fn() callback. Remove every invocation completely.
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Lars Ellenberg <drbd-dev@lists.linbit.com>
Cc: drbd-user@lists.linbit.com
Cc: Jiri Kosina <jkosina@suse.cz>
Cc: Yehuda Sadeh <yehuda@inktank.com>
Cc: Sage Weil <sage@inktank.com>
Cc: Alex Elder <elder@kernel.org>
Cc: ceph-devel@vger.kernel.org
Cc: Alasdair Kergon <agk@redhat.com>
Cc: Mike Snitzer <snitzer@redhat.com>
Cc: dm-devel@redhat.com
Cc: Neil Brown <neilb@suse.de>
Cc: linux-raid@vger.kernel.org
Cc: Christoph Hellwig <hch@infradead.org>
Cc: "Martin K. Petersen" <martin.petersen@oracle.com>
Acked-by: NeilBrown <neilb@suse.de> (for the 'md' bits)
Acked-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
[dpark: also remove ->merge_bvec_fn() in dm-thin as well as
dm-era-target, and resolve merge conflicts]
Signed-off-by: Dongsu Park <dpark@posteo.net>
Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
Remove bio_fits_rdev() as sufficient merge_bvec_fn() handling is now
performed by blk_queue_split() in md_make_request().
Cc: Neil Brown <neilb@suse.de>
Cc: linux-raid@vger.kernel.org
Acked-by: NeilBrown <neilb@suse.de>
Signed-off-by: Kent Overstreet <kent.overstreet@gmail.com>
[dpark: add more description in commit message]
Signed-off-by: Dongsu Park <dpark@posteo.net>
Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
If a read request fits entirely in a chunk, it will be passed directly to the
underlying device (providing it hasn't failed of course). If it doesn't fit,
the slightly less efficient path that uses the stripe_cache is used.
Requests that get to the stripe cache are always completely split up as
necessary.
So with RAID5, ripping out the merge_bvec_fn doesn't cause it to stop work,
but could cause it to take the less efficient path more often.
All that is needed to manage this is for 'chunk_aligned_read' do some bio
splitting, much like the RAID0 code does.
Cc: Neil Brown <neilb@suse.de>
Cc: linux-raid@vger.kernel.org
Acked-by: NeilBrown <neilb@suse.de>
Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
I have a report of drop_one_stripe() called from
raid5_cache_scan() apparently finding ->max_nr_stripes == 0.
This should not be allowed.
So add a test to keep max_nr_stripes above min_nr_stripes.
Also use a 'mask' rather than a 'mod' in drop_one_stripe
to ensure 'hash' is valid even if max_nr_stripes does reach zero.
Fixes: edbe83ab4c ("md/raid5: allow the stripe_cache to grow and shrink.")
Cc: stable@vger.kernel.org (4.1 - please release with 2d5b569b66)
Reported-by: Tomas Papan <tomas.papan@gmail.com>
Signed-off-by: NeilBrown <neilb@suse.com>
Some places use helpers now, others don't. We only have the 'is set'
helper, add helpers for setting and clearing flags too.
It was a bit of a mess of atomic vs non-atomic access. With
BIO_UPTODATE gone, we don't have any risk of concurrent access to the
flags. So relax the restriction and don't make any of them atomic. The
flags that do have serialization issues (reffed and chained), we
already handle those separately.
Signed-off-by: Jens Axboe <axboe@fb.com>
Currently we have two different ways to signal an I/O error on a BIO:
(1) by clearing the BIO_UPTODATE flag
(2) by returning a Linux errno value to the bi_end_io callback
The first one has the drawback of only communicating a single possible
error (-EIO), and the second one has the drawback of not beeing persistent
when bios are queued up, and are not passed along from child to parent
bio in the ever more popular chaining scenario. Having both mechanisms
available has the additional drawback of utterly confusing driver authors
and introducing bugs where various I/O submitters only deal with one of
them, and the others have to add boilerplate code to deal with both kinds
of error returns.
So add a new bi_error field to store an errno value directly in struct
bio and remove the existing mechanisms to clean all this up.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: NeilBrown <neilb@suse.com>
Signed-off-by: Jens Axboe <axboe@fb.com>
This flag is currently never cleared, which can in rare cases
trigger a warn-on if it is still set but the block isn't
InSync.
So clear it when it isn't need, which includes if the replacement
device has failed.
Signed-off-by: NeilBrown <neilb@suse.com>
Cache size can grow or shrink due to various pressures at
any time. So when we resize the cache as part of a 'grow'
operation (i.e. change the size to allow more devices) we need
to blocks that automatic growing/shrinking.
So introduce a mutex. auto grow/shrink uses mutex_trylock()
and just doesn't bother if there is a blockage.
Resizing the whole cache holds the mutex to ensure that
the correct number of new stripes is allocated.
This bug can result in some stripes not being freed when an
array is stopped. This leads to the kmem_cache not being
freed and a subsequent array can try to use the same kmem_cache
and get confused.
Fixes: edbe83ab4c ("md/raid5: allow the stripe_cache to grow and shrink.")
Cc: stable@vger.kernel.org (4.1 - please delay until 2 weeks after release of 4.2)
Signed-off-by: NeilBrown <neilb@suse.com>
conf->released_stripes list isn't always related to where there are
free stripes pending. Active stripes can be in the list too.
And even free stripes were active very recently.
Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: NeilBrown <neilb@suse.de>
I noticed heavy spin lock contention at get_active_stripe() with fsmark
multiple thread write workloads.
Here is how this hot contention comes from. We have limited stripes, and
it's a multiple thread write workload. Hence, those stripes will be taken
soon, which puts later processes to sleep for waiting free stripes. When
enough stripes(>= 1/4 total stripes) are released, all process are woken,
trying to get the lock. But there is one only being able to get this lock
for each hash lock, making other processes spinning out there for acquiring
the lock.
Thus, it's effectiveless to wakeup all processes and let them battle for
a lock that permits one to access only each time. Instead, we could make
it be a exclusive wake up: wake up one process only. That avoids the heavy
spin lock contention naturally.
To do the exclusive wake up, we've to split wait_for_stripe into multiple
wait queues, to make it per hash value, just like the hash lock.
Here are some test results I have got with this patch applied(all test run
3 times):
`fsmark.files_per_sec'
=====================
next-20150317 this patch
------------------------- -------------------------
metric_value ±stddev metric_value ±stddev change testbox/benchmark/testcase-params
------------------------- ------------------------- -------- ------------------------------
25.600 ±0.0 92.700 ±2.5 262.1% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-btrfs-4M-30G-fsyncBeforeClose
25.600 ±0.0 77.800 ±0.6 203.9% ivb44/fsmark/1x-64t-9BRD_6G-RAID5-btrfs-4M-30G-fsyncBeforeClose
32.000 ±0.0 93.800 ±1.7 193.1% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-ext4-4M-30G-fsyncBeforeClose
32.000 ±0.0 81.233 ±1.7 153.9% ivb44/fsmark/1x-64t-9BRD_6G-RAID5-ext4-4M-30G-fsyncBeforeClose
48.800 ±14.5 99.667 ±2.0 104.2% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-xfs-4M-30G-fsyncBeforeClose
6.400 ±0.0 12.800 ±0.0 100.0% ivb44/fsmark/1x-64t-3HDD-RAID5-btrfs-4M-40G-fsyncBeforeClose
63.133 ±8.2 82.800 ±0.7 31.2% ivb44/fsmark/1x-64t-9BRD_6G-RAID5-xfs-4M-30G-fsyncBeforeClose
245.067 ±0.7 306.567 ±7.9 25.1% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-f2fs-4M-30G-fsyncBeforeClose
17.533 ±0.3 21.000 ±0.8 19.8% ivb44/fsmark/1x-1t-3HDD-RAID5-xfs-4M-40G-fsyncBeforeClose
188.167 ±1.9 215.033 ±3.1 14.3% ivb44/fsmark/1x-1t-4BRD_12G-RAID5-btrfs-4M-30G-NoSync
254.500 ±1.8 290.733 ±2.4 14.2% ivb44/fsmark/1x-1t-9BRD_6G-RAID5-btrfs-4M-30G-NoSync
`time.system_time'
=====================
next-20150317 this patch
------------------------- -------------------------
metric_value ±stddev metric_value ±stddev change testbox/benchmark/testcase-params
------------------------- ------------------------- -------- ------------------------------
7235.603 ±1.2 185.163 ±1.9 -97.4% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-btrfs-4M-30G-fsyncBeforeClose
7666.883 ±2.9 202.750 ±1.0 -97.4% ivb44/fsmark/1x-64t-9BRD_6G-RAID5-btrfs-4M-30G-fsyncBeforeClose
14567.893 ±0.7 421.230 ±0.4 -97.1% ivb44/fsmark/1x-64t-3HDD-RAID5-btrfs-4M-40G-fsyncBeforeClose
3697.667 ±14.0 148.190 ±1.7 -96.0% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-xfs-4M-30G-fsyncBeforeClose
5572.867 ±3.8 310.717 ±1.4 -94.4% ivb44/fsmark/1x-64t-9BRD_6G-RAID5-ext4-4M-30G-fsyncBeforeClose
5565.050 ±0.5 313.277 ±1.5 -94.4% ivb44/fsmark/1x-64t-4BRD_12G-RAID5-ext4-4M-30G-fsyncBeforeClose
2420.707 ±17.1 171.043 ±2.7 -92.9% ivb44/fsmark/1x-64t-9BRD_6G-RAID5-xfs-4M-30G-fsyncBeforeClose
3743.300 ±4.6 379.827 ±3.5 -89.9% ivb44/fsmark/1x-64t-3HDD-RAID5-ext4-4M-40G-fsyncBeforeClose
3308.687 ±6.3 363.050 ±2.0 -89.0% ivb44/fsmark/1x-64t-3HDD-RAID5-xfs-4M-40G-fsyncBeforeClose
Where,
1x: where 'x' means iterations or loop, corresponding to the 'L' option of fsmark
1t, 64t: where 't' means thread
4M: means the single file size, corresponding to the '-s' option of fsmark
40G, 30G, 120G: means the total test size
4BRD_12G: BRD is the ramdisk, where '4' means 4 ramdisk, and where '12G' means
the size of one ramdisk. So, it would be 48G in total. And we made a
raid on those ramdisk
As you can see, though there are no much performance gain for hard disk
workload, the system time is dropped heavily, up to 97%. And as expected,
the performance increased a lot, up to 260%, for fast device(ram disk).
v2: use bits instead of array to note down wait queue need to wake up.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Signed-off-by: NeilBrown <neilb@suse.de>
I noticed heavy spin lock contention at get_active_stripe(), introduced
at being wake up stage, where a bunch of processes try to re-hold the
spin lock again.
After giving some thoughts on this issue, I found the lock could be
relieved(and even avoided) if we turn the wait_for_stripe to per
waitqueue for each lock hash and make the wake up exclusive: wake up
one process each time, which avoids the lock contention naturally.
Before go hacking with wait_for_stripe, I found it actually has 2
usages: for the array to enter or leave the quiescent state, and also
to wait for an available stripe in each of the hash lists.
So this patch splits the first usage off into a separate wait_queue,
wait_for_quiescent, and the next patch will turn the second usage into
one waitqueue for each hash value, and make it exclusive, to relieve
the lock contention.
v2: wake_up(wait_for_quiescent) when (active_stripes == 0)
Commit log refactor suggestion from Neil.
Signed-off-by: Yuanhan Liu <yuanhan.liu@linux.intel.com>
Signed-off-by: NeilBrown <neilb@suse.de>
MD_RECOVERY_DONE is normally cleared by md_check_recovery after a
resync etc finished. However it is possible for raid5_start_reshape
to race and start a reshape before MD_RECOVERY_DONE is cleared. This
can lean to multiple reshapes running at the same time, which isn't
good.
To make sure it is cleared before starting a reshape, and also clear
it when reaping a thread, just to be safe.
Signed-off-by: NeilBrown <neilb@suse.de>
Now that the code in break_stripe_batch_list() is nearly identical
to the end of handle_stripe_clean_event, replace the later
with a function call.
The only remaining difference of any interest is the masking that is
applieds to dev[i].flags copied from head_sh.
R5_WriteError certainly isn't wanted as it is set per-stripe, not
per-patch. R5_Overlap isn't wanted as it is explicitly handled.
Signed-off-by: NeilBrown <neilb@suse.de>
When a batch of stripes is broken up, we keep some of the flags
that were per-stripe, and copy other flags from the head to all
others.
This only happens while a stripe is being handled, so many of the
flags are irrelevant.
The "SYNC_FLAGS" (which I've renamed to make it clear there are
several) and STRIPE_DEGRADED are set per-stripe and so need to be
preserved. STRIPE_INSYNC is the only flag that is set on the head
that needs to be propagated to all others.
For safety, add a WARN_ON if others are set, except:
STRIPE_HANDLE - this is safe and per-stripe and we are going to set
in several cases anyway
STRIPE_INSYNC
STRIPE_IO_STARTED - this is just a hint and doesn't hurt.
STRIPE_ON_PLUG_LIST
STRIPE_ON_RELEASE_LIST - It is a point pointless for a batched
stripe to be on one of these lists, but it can happen
as can be safely ignored.
Signed-off-by: NeilBrown <neilb@suse.de>
When we break a stripe_batch_list we sometimes want to set
STRIPE_HANDLE on the individual stripes, and sometimes not.
So pass a 'handle_flags' arg. If it is zero, always set STRIPE_HANDLE
(on non-head stripes). If not zero, only set it if any of the given
flags are present.
Signed-off-by: NeilBrown <neilb@suse.de>
break_stripe_batch list didn't clear head_sh->batch_head.
This was probably a bug.
Also clear all R5_Overlap flags and if any were cleared, wake up
'wait_for_overlap'.
This isn't always necessary but the worst effect is a little
extra checking for code that is waiting on wait_for_overlap.
Also, don't use wake_up_nr() because that does the wrong thing
if 'nr' is zero, and it number of flags cleared doesn't
strongly correlate with the number of threads to wake.
Signed-off-by: NeilBrown <neilb@suse.de>
handle_stripe_clean_event() contains a chunk of code very
similar to check_break_stripe_batch_list().
If we make the latter more like the former, we can end up
with just one copy of this code.
This first step removed the condition (and the 'check_') part
of the name. This has the added advantage of making it clear
what check is being performed at the point where the function is
called.
Signed-off-by: NeilBrown <neilb@suse.de>
If a stripe is a member of a batch, but not the head, it must
not be handled separately from the rest of the batch.
'clear_batch_ready()' handles this requirement to some
extent but not completely. If a member is passed to handle_stripe()
a second time it returns '0' indicating the stripe can be handled,
which is wrong.
So add an extra test.
Signed-off-by: NeilBrown <neilb@suse.de>
When we add a write to a stripe we need to make sure the bitmap
bit is set. While doing that the stripe is not locked so it could
be added to a batch after which further changes to STRIPE_BIT_DELAY
and ->bm_seq are ineffective.
So we need to hold off adding to a stripe until bitmap_startwrite has
completed at least once, and we need to avoid further changes to
STRIPE_BIT_DELAY once the stripe has been added to a batch.
If a bitmap_startwrite() completes after the stripe was added to a
batch, it will not have set the bit, only incremented a counter, so no
extra delay of the stripe is needed.
Reported-by: Shaohua Li <shli@kernel.org>
Signed-off-by: NeilBrown <neilb@suse.de>
When we add a stripe to a batch, we need to be sure that
head stripe will wait for the bitmap update required for the new
stripe.
Signed-off-by: NeilBrown <neilb@suse.de>
ops_run_reconstruct6() doesn't correctly chain asyn operations. The tx returned
by async_gen_syndrome should be added as the dependent tx of next stripe.
The issue is introduced by commit 59fc630b8b
RAID5: batch adjacent full stripe write
Reported-and-tested-by: Maxime Ripard <maxime.ripard@free-electrons.com>
Signed-off-by: Shaohua Li <shli@fb.com>
Signed-off-by: NeilBrown <neilb@suse.de>
There is no need for special handling of stripe-batches when the array
is degraded.
There may be if there is a failure in the batch, but STRIPE_DEGRADED
does not imply an error.
So don't set STRIPE_BATCH_ERR in ops_run_io just because the array is
degraded.
This actually causes a bug: the STRIPE_DEGRADED flag gets cleared in
check_break_stripe_batch_list() and so the bitmap bit gets cleared
when it shouldn't.
So in check_break_stripe_batch_list(), split the batch up completely -
again STRIPE_DEGRADED isn't meaningful.
Also don't set STRIPE_BATCH_ERR when there is a write error to a
replacement device. This simply removes the replacement device and
requires no extra handling.
Signed-off-by: NeilBrown <neilb@suse.de>
As the new 'scribble' array is sized based on chunk size,
we need to make sure the size matches the largest of 'old'
and 'new' chunk sizes when the array is undergoing reshape.
We also potentially need to resize it even when not resizing
the stripe cache, as chunk size can change without changing
number of devices.
So move the 'resize' code into a separate function, and
consider old and new sizes when allocating.
Signed-off-by: NeilBrown <neilb@suse.de>
Fixes: 46d5b78562 ("raid5: use flex_array for scribble data")
If any memory allocation in resize_stripes fails we will return
-ENOMEM, but in some cases we update conf->pool_size anyway.
This means that if we try again, the allocations will be assumed
to be larger than they are, and badness results.
So only update pool_size if there is no error.
This bug was introduced in 2.6.17 and the patch is suitable for
-stable.
Fixes: ad01c9e375 ("[PATCH] md: Allow stripes to be expanded in preparation for expanding an array")
Cc: stable@vger.kernel.org (v2.6.17+)
Signed-off-by: NeilBrown <neilb@suse.de>
When performing a reconstruct write, we need to read all blocks
that are not being over-written .. except the parity (P and Q) blocks.
The code currently reads these (as they are not being over-written!)
unnecessarily.
Signed-off-by: NeilBrown <neilb@suse.de>
Fixes: ea664c8245 ("md/raid5: need_this_block: tidy/fix last condition.")
It is not incorrect to call handle_stripe_fill() when
a batch of full-stripe writes is active.
It is, however, a BUG if fetch_block() then decides
it needs to actually fetch anything.
So move the 'BUG_ON' to where it belongs.
Signed-off-by: NeilBrown <neilb@suse.de>
Fixes: 59fc630b8b ("RAID5: batch adjacent full stripe write")
The new batch_lock and batch_list fields are being initialized in
grow_one_stripe() but not in resize_stripes(). This causes a crash
on resize.
So separate the core initialization into a new function and call it
from both allocation sites.
Signed-off-by: NeilBrown <neilb@suse.de>
Fixes: 59fc630b8b ("RAID5: batch adjacent full stripe write")
When array is degraded, read data landed on failed drives will result in
reading rest of data in a stripe. So a single sequential read would
result in same data being read twice.
This patch is to avoid chunk aligned read for degraded array. The
downside is to involve stripe cache which means associated CPU overhead
and extra memory copy.
Test Results:
Following test are done on a enterprise storage node with Seagate 6T SAS
drives and Xeon E5-2648L CPU (10 cores, 1.9Ghz), 10 disks MD RAID6 8+2,
chunk size 128 KiB.
I use FIO, using direct-io with various bs size, enough queue depth,
tested sequential and 100% random read against 3 array config:
1) optimal, as baseline;
2) degraded;
3) degraded with this patch.
Kernel version is 4.0-rc3.
Each individual test I only did once so there might be some variations,
but we just focus on big trend.
Sequential Read:
bs=(KiB) optimal(MiB/s) degraded(MiB/s) degraded-with-patch (MiB/s)
1024 1608 656 995
512 1624 710 956
256 1635 728 980
128 1636 771 983
64 1612 1119 1000
32 1580 1420 1004
16 1368 688 986
8 768 647 953
4 411 413 850
Random Read:
bs=(KiB) optimal(IOPS) degraded(IOPS) degraded-with-patch (IOPS)
1024 163 160 156
512 274 273 272
256 426 428 424
128 576 592 591
64 726 724 726
32 849 848 837
16 900 970 971
8 927 940 929
4 948 940 955
Some notes:
* In sequential + optimal, as bs size getting smaller, the FIO thread
become CPU bound.
* In sequential + degraded, there's big increase when bs is 64K and
32K, I don't have explanation.
* In sequential + degraded-with-patch, the MD thread mostly become CPU
bound.
If you want to we can discuss specific data point in those data. But in
general it seems with this patch, we have more predictable and in most
cases significant better sequential read performance when array is
degraded, and almost no noticeable impact on random read.
Performance is a complicated thing, the patch works well for this
particular configuration, but may not be universal. For example I
imagine testing on all SSD array may have very different result. But I
personally think in most cases IO bandwidth is more scarce resource than
CPU.
Signed-off-by: Eric Mei <eric.mei@seagate.com>
Signed-off-by: NeilBrown <neilb@suse.de>
The default setting of 256 stripe_heads is probably
much too small for many configurations. So it is best to make it
auto-configure.
Shrinking the cache under memory pressure is easy. The only
interesting part here is that we put a fairly high cost
('seeks') on shrinking the cache as the cost is greater than
just having to read more data, it reduces parallelism.
Growing the cache on demand needs to be done carefully. If we allow
fast growth, that can upset memory balance as lots of dirty memory can
quickly turn into lots of memory queued in the stripe_cache.
It is important for the raid5 block device to appear congested to
allow write-throttling to work.
So we only add stripes slowly. We set a flag when an allocation
fails because all stripes are in use, allocate at a convenient
time when that flag is set, and don't allow it to be set again
until at least one stripe_head has been released for re-use.
This means that a spurt of requests will only cause one stripe_head
to be allocated, but a steady stream of requests will slowly
increase the cache size - until memory pressure puts it back again.
It could take hours to reach a steady state.
The value written to, and displayed in, stripe_cache_size is
used as a minimum. The cache can grow above this and shrink back
down to it. The actual size is not directly visible, though it can
be deduced to some extent by watching stripe_cache_active.
Signed-off-by: NeilBrown <neilb@suse.de>
Rather than adjusting max_nr_stripes whenever {grow,drop}_one_stripe()
succeeds, do it inside the functions.
Also choose the correct hash to handle next inside the functions.
This removes duplication and will help with future new uses of
{grow,drop}_one_stripe.
This also fixes a minor bug where the "md/raid:%md: allocate XXkB"
message always said "0kB".
Signed-off-by: NeilBrown <neilb@suse.de>
Depending on the available coding we allow optimized rmw logic for write
operations. To support easier testing this patch allows manual control
of the rmw/rcw descision through the interface /sys/block/mdX/md/rmw_level.
The configuration can handle three levels of control.
rmw_level=0: Disable rmw for all RAID types. Hardware assisted P/Q
calculation has no implementation path yet to factor in/out chunks of
a syndrome. Enforcing this level can be benefical for slow CPUs with
hardware syndrome support and fast SSDs.
rmw_level=1: Estimate rmw IOs and rcw IOs. Execute rmw only if we will
save IOs. This equals the "old" unpatched behaviour and will be the
default.
rmw_level=2: Execute rmw even if calculated IOs for rmw and rcw are
equal. We might have higher CPU consumption because of calculating the
parity twice but it can be benefical otherwise. E.g. RAID4 with fast
dedicated parity disk/SSD. The option is implemented just to be
forward-looking and will ONLY work with this patch!
Signed-off-by: Markus Stockhausen <stockhausen@collogia.de>
Signed-off-by: NeilBrown <neilb@suse.de>
Glue it altogehter. The raid6 rmw path should work the same as the
already existing raid5 logic. So emulate the prexor handling/flags
and split functions as needed.
1) Enable xor_syndrome() in the async layer.
2) Split ops_run_prexor() into RAID4/5 and RAID6 logic. Xor the syndrome
at the start of a rmw run as we did it before for the single parity.
3) Take care of rmw run in ops_run_reconstruct6(). Again process only
the changed pages to get syndrome back into sync.
4) Enhance set_syndrome_sources() to fill NULL pages if we are in a rmw
run. The lower layers will calculate start & end pages from that and
call the xor_syndrome() correspondingly.
5) Adapt the several places where we ignored Q handling up to now.
Performance numbers for a single E5630 system with a mix of 10 7200k
desktop/server disks. 300 seconds random write with 8 threads onto a
3,2TB (10*400GB) RAID6 64K chunk without spare (group_thread_cnt=4)
bsize rmw_level=1 rmw_level=0 rmw_level=1 rmw_level=0
skip_copy=1 skip_copy=1 skip_copy=0 skip_copy=0
4K 115 KB/s 141 KB/s 165 KB/s 140 KB/s
8K 225 KB/s 275 KB/s 324 KB/s 274 KB/s
16K 434 KB/s 536 KB/s 640 KB/s 534 KB/s
32K 751 KB/s 1,051 KB/s 1,234 KB/s 1,045 KB/s
64K 1,339 KB/s 1,958 KB/s 2,282 KB/s 1,962 KB/s
128K 2,673 KB/s 3,862 KB/s 4,113 KB/s 3,898 KB/s
256K 7,685 KB/s 7,539 KB/s 7,557 KB/s 7,638 KB/s
512K 19,556 KB/s 19,558 KB/s 19,652 KB/s 19,688 Kb/s
Signed-off-by: Markus Stockhausen <stockhausen@collogia.de>
Signed-off-by: NeilBrown <neilb@suse.de>
expansion/resync can grab a stripe when the stripe is in batch list. Since all
stripes in batch list must be in the same state, we can't allow some stripes
run into expansion/resync. So we delay expansion/resync for stripe in batch
list.
Signed-off-by: Shaohua Li <shli@fusionio.com>
Signed-off-by: NeilBrown <neilb@suse.de>
If io error happens in any stripe of a batch list, the batch list will be
split, then normal process will run for the stripes in the list.
Signed-off-by: Shaohua Li <shli@fusionio.com>
Signed-off-by: NeilBrown <neilb@suse.de>
stripe cache is 4k size. Even adjacent full stripe writes are handled in 4k
unit. Idealy we should use big size for adjacent full stripe writes. Bigger
stripe cache size means less stripes runing in the state machine so can reduce
cpu overhead. And also bigger size can cause bigger IO size dispatched to under
layer disks.
With below patch, we will automatically batch adjacent full stripe write
together. Such stripes will be added to the batch list. Only the first stripe
of the list will be put to handle_list and so run handle_stripe(). Some steps
of handle_stripe() are extended to cover all stripes of the list, including
ops_run_io, ops_run_biodrain and so on. With this patch, we have less stripes
running in handle_stripe() and we send IO of whole stripe list together to
increase IO size.
Stripes added to a batch list have some limitations. A batch list can only
include full stripe write and can't cross chunk boundary to make sure stripes
have the same parity disks. Stripes in a batch list must be in the same state
(no written, toread and so on). If a stripe is in a batch list, all new
read/write to add_stripe_bio will be blocked to overlap conflict till the batch
list is handled. The limitations will make sure stripes in a batch list be in
exactly the same state in the life circly.
I did test running 160k randwrite in a RAID5 array with 32k chunk size and 6
PCIe SSD. This patch improves around 30% performance and IO size to under layer
disk is exactly 32k. I also run a 4k randwrite test in the same array to make
sure the performance isn't changed with the patch.
Signed-off-by: Shaohua Li <shli@fusionio.com>
Signed-off-by: NeilBrown <neilb@suse.de>
NeilBrown
Linus Torvalds
Kent Overstreet
Ming Lin
Sasha Levin
Jens Axboe
Christoph Hellwig
Shaohua Li
Yuanhan Liu
NeilBrown
Shaohua Li
Eric Mei
Markus Stockhausen