Skip to content

K
Kernel

openeuler / Kernel
1 年多 前同步成功

通知 8
Star 0
Fork 0
K
Kernel
切换分支/标签
  1. 30 5月, 2015 10 次提交
  3. 29 5月, 2015 1 次提交
    • M
      dm: fix false warning in free_rq_clone() for unmapped requests · e5d8de32
      由 Mike Snitzer 提交于 
      When stacking request-based dm device on non blk-mq device and
device-mapper target could not map the request (error target is used,
multipath target with all paths down, etc), the WARN_ON_ONCE() in
free_rq_clone() will trigger when it shouldn't.

The warning was added by commit aa6df8dd ("dm: fix free_rq_clone() NULL
pointer when requeueing unmapped request").  But free_rq_clone() with
clone->q == NULL is valid usage for the case where
dm_kill_unmapped_request() initiates request cleanup.

Fix this false warning by just removing the WARN_ON -- it only generated
false positives and was never useful in catching the intended case
(completing clone request not being mapped e.g. clone->q being NULL).

Fixes: aa6df8dd ("dm: fix free_rq_clone() NULL pointer when requeueing unmapped request")
Reported-by: NBart Van Assche <bart.vanassche@sandisk.com>
Reported-by: NJun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
      e5d8de32
  5. 28 5月, 2015 2 次提交
    • M
      dm: requeue from blk-mq dm_mq_queue_rq() using BLK_MQ_RQ_QUEUE_BUSY · 45714fbe
      由 Mike Snitzer 提交于 
      Use BLK_MQ_RQ_QUEUE_BUSY to requeue a blk-mq request directly from the
DM blk-mq device's .queue_rq.  This cleans up the previous convoluted
handling of request requeueing that would return BLK_MQ_RQ_QUEUE_OK
(even though it wasn't) and then run blk_mq_requeue_request() followed
by blk_mq_kick_requeue_list().

Also, document that DM blk-mq ontop of old request_fn devices cannot
fail in clone_rq() since the clone request is preallocated as part of
the pdu.
Reported-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
      45714fbe
    • M
      dm mpath: fix leak of dm_mpath_io structure in blk-mq .queue_rq error path · 4c6dd53d
      由 Mike Snitzer 提交于 
      Otherwise kmemleak reported:

unreferenced object 0xffff88009b14e2b0 (size 16):
  comm "fio", pid 4274, jiffies 4294978034 (age 1253.210s)
  hex dump (first 16 bytes):
    40 12 f3 99 01 88 ff ff 00 10 00 00 00 00 00 00  @...............
  backtrace:
    [<ffffffff81600029>] kmemleak_alloc+0x49/0xb0
    [<ffffffff811679a8>] kmem_cache_alloc+0xf8/0x160
    [<ffffffff8111c950>] mempool_alloc_slab+0x10/0x20
    [<ffffffff8111cb37>] mempool_alloc+0x57/0x150
    [<ffffffffa04d2b61>] __multipath_map.isra.17+0xe1/0x220 [dm_multipath]
    [<ffffffffa04d2cb5>] multipath_clone_and_map+0x15/0x20 [dm_multipath]
    [<ffffffffa02889b5>] map_request.isra.39+0xd5/0x220 [dm_mod]
    [<ffffffffa028b0e4>] dm_mq_queue_rq+0x134/0x240 [dm_mod]
    [<ffffffff812cccb5>] __blk_mq_run_hw_queue+0x1d5/0x380
    [<ffffffff812ccaa5>] blk_mq_run_hw_queue+0xc5/0x100
    [<ffffffff812ce350>] blk_sq_make_request+0x240/0x300
    [<ffffffff812c0f30>] generic_make_request+0xc0/0x110
    [<ffffffff812c0ff2>] submit_bio+0x72/0x150
    [<ffffffff811c07cb>] do_blockdev_direct_IO+0x1f3b/0x2da0
    [<ffffffff811c166e>] __blockdev_direct_IO+0x3e/0x40
    [<ffffffff8120aa1a>] ext4_direct_IO+0x1aa/0x390

Fixes: e5863d9a ("dm: allocate requests in target when stacking on blk-mq devices")
Reported-by: NBart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org # 4.0+
      4c6dd53d
  7. 27 5月, 2015 1 次提交
    • J
      dm: fix NULL pointer when clone_and_map_rq returns !DM_MAPIO_REMAPPED · 3a140755
      由 Junichi Nomura 提交于 
      When stacking request-based DM on blk_mq device, request cloning and
remapping are done in a single call to target's clone_and_map_rq().
The clone is allocated and valid only if clone_and_map_rq() returns
DM_MAPIO_REMAPPED.

The "IS_ERR(clone)" check in map_request() does not cover all the
!DM_MAPIO_REMAPPED cases that are possible (E.g. if underlying devices
are not ready or unavailable, clone_and_map_rq() may return
DM_MAPIO_REQUEUE without ever having established an ERR_PTR).  Fix this
by explicitly checking for a return that is not DM_MAPIO_REMAPPED in
map_request().

Without this fix, DM core may call setup_clone() for a NULL clone
and oops like this:

   BUG: unable to handle kernel NULL pointer dereference at 0000000000000068
   IP: [<ffffffff81227525>] blk_rq_prep_clone+0x7d/0x137
   ...
   CPU: 2 PID: 5793 Comm: kdmwork-253:3 Not tainted 4.0.0-nm #1
   ...
   Call Trace:
    [<ffffffffa01d1c09>] map_tio_request+0xa9/0x258 [dm_mod]
    [<ffffffff81071de9>] kthread_worker_fn+0xfd/0x150
    [<ffffffff81071cec>] ? kthread_parkme+0x24/0x24
    [<ffffffff81071cec>] ? kthread_parkme+0x24/0x24
    [<ffffffff81071fdd>] kthread+0xe6/0xee
    [<ffffffff81093a59>] ? put_lock_stats+0xe/0x20
    [<ffffffff81071ef7>] ? __init_kthread_worker+0x5b/0x5b
    [<ffffffff814c2d98>] ret_from_fork+0x58/0x90
    [<ffffffff81071ef7>] ? __init_kthread_worker+0x5b/0x5b

Fixes: e5863d9a ("dm: allocate requests in target when stacking on blk-mq devices")
Reported-by: NBart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: NJun'ichi Nomura <j-nomura@ce.jp.nec.com>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Cc: stable@vger.kernel.org # 4.0+
      3a140755
  9. 26 5月, 2015 1 次提交
  11. 22 5月, 2015 2 次提交
    • C
      block, dm: don't copy bios for request clones · 5f1b670d
      由 Christoph Hellwig 提交于 
      Currently dm-multipath has to clone the bios for every request sent
to the lower devices, which wastes cpu cycles and ties down memory.

This patch instead adds a new REQ_CLONE flag that instructs req_bio_endio
to not complete bios attached to a request, which we set on clone
requests similar to bios in a flush sequence.  With this change I/O
errors on a path failure only get propagated to dm-multipath, which
can then either resubmit the I/O or complete the bios on the original
request.

I've done some basic testing of this on a Linux target with ALUA support,
and it survives path failures during I/O nicely.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NJens Axboe <axboe@fb.com>
      5f1b670d
    • M
      block: remove management of bi_remaining when restoring original bi_end_io · 326e1dbb
      由 Mike Snitzer 提交于 
      Commit c4cf5261 ("bio: skip atomic inc/dec of ->bi_remaining for
non-chains") regressed all existing callers that followed this pattern:
 1) saving a bio's original bi_end_io
 2) wiring up an intermediate bi_end_io
 3) restoring the original bi_end_io from intermediate bi_end_io
 4) calling bio_endio() to execute the restored original bi_end_io

The regression was due to BIO_CHAIN only ever getting set if
bio_inc_remaining() is called.  For the above pattern it isn't set until
step 3 above (step 2 would've needed to establish BIO_CHAIN).  As such
the first bio_endio(), in step 2 above, never decremented __bi_remaining
before calling the intermediate bi_end_io -- leaving __bi_remaining with
the value 1 instead of 0.  When bio_inc_remaining() occurred during step
3 it brought it to a value of 2.  When the second bio_endio() was
called, in step 4 above, it should've called the original bi_end_io but
it didn't because there was an extra reference that wasn't dropped (due
to atomic operations being optimized away since BIO_CHAIN wasn't set
upfront).

Fix this issue by removing the __bi_remaining management complexity for
all callers that use the above pattern -- bio_chain() is the only
interface that _needs_ to be concerned with __bi_remaining.  For the
above pattern callers just expect the bi_end_io they set to get called!
Remove bio_endio_nodec() and also remove all bio_inc_remaining() calls
that aren't associated with the bio_chain() interface.

Also, the bio_inc_remaining() interface has been moved local to bio.c.

Fixes: c4cf5261 ("bio: skip atomic inc/dec of ->bi_remaining for non-chains")
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Reviewed-by: NJan Kara <jack@suse.cz>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NJens Axboe <axboe@fb.com>
      326e1dbb
  13. 21 5月, 2015 3 次提交
  15. 08 5月, 2015 7 次提交
    • N
      md/raid5: fix handling of degraded stripes in batches. · bb27051f
      由 NeilBrown 提交于 
      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: NNeilBrown <neilb@suse.de>
      bb27051f
    • N
      md/raid5: fix allocation of 'scribble' array. · 738a2738
      由 NeilBrown 提交于 
      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: NNeilBrown <neilb@suse.de>
Fixes: 46d5b785 ("raid5: use flex_array for scribble data")
      738a2738
    • N
      md/raid5: don't record new size if resize_stripes fails. · 6e9eac2d
      由 NeilBrown 提交于 
      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: ad01c9e3 ("[PATCH] md: Allow stripes to be expanded in preparation for expanding an array")
Cc: stable@vger.kernel.org (v2.6.17+)
Signed-off-by: NNeilBrown <neilb@suse.de>
      6e9eac2d
    • N
      md/raid5: avoid reading parity blocks for full-stripe write to degraded array · 10d82c5f
      由 NeilBrown 提交于 
      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: NNeilBrown <neilb@suse.de>
Fixes: ea664c82 ("md/raid5: need_this_block: tidy/fix last condition.")
      10d82c5f
    • N
      md/raid5: more incorrect BUG_ON in handle_stripe_fill. · b0c783b3
      由 NeilBrown 提交于 
      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: NNeilBrown  <neilb@suse.de>
Fixes: 59fc630b ("RAID5: batch adjacent full stripe write")
      b0c783b3
    • N
      md/raid5: new alloc_stripe() to allocate an initialize a stripe. · f18c1a35
      由 NeilBrown 提交于 
      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: NNeilBrown <neilb@suse.de>
Fixes: 59fc630b ("RAID5: batch adjacent full stripe write")
      f18c1a35
    • H
      md-raid0: conditional mddev->queue access to suit dm-raid · b6538fe3
      由 Heinz Mauelshagen 提交于 
      This patch is a prerequisite for dm-raid "raid0" support to allow
dm-raid to access the MD RAID0 personality doing unconditional
accesses to mddev->queue, which is NULL in case of dm-raid stacked on
top of MD.

Most of the conditional mddev->queue accesses made it to upstream but
this missing one, which prohibits md raid0 to set disk stack limits
(being done in dm core in case of md underneath dm).
Signed-off-by: NHeinz Mauelshagen <heinzm@redhat.com>
Tested-by: NHeinz Mauelshagen <heinzm@redhat.com>
Signed-off-by: NNeilBrown <neilb@suse.de>
      b6538fe3
  17. 06 5月, 2015 3 次提交
    • J
      bio: skip atomic inc/dec of ->bi_cnt for most use cases · dac56212
      由 Jens Axboe 提交于 
      Struct bio has a reference count that controls when it can be freed.
Most uses cases is allocating the bio, which then returns with a
single reference to it, doing IO, and then dropping that single
reference. We can remove this atomic_dec_and_test() in the completion
path, if nobody else is holding a reference to the bio.

If someone does call bio_get() on the bio, then we flag the bio as
now having valid count and that we must properly honor the reference
count when it's being put.
Tested-by: NRobert Elliott <elliott@hp.com>
Signed-off-by: NJens Axboe <axboe@fb.com>
      dac56212
    • J
      bio: skip atomic inc/dec of ->bi_remaining for non-chains · c4cf5261
      由 Jens Axboe 提交于 
      Struct bio has an atomic ref count for chained bio's, and we use this
to know when to end IO on the bio. However, most bio's are not chained,
so we don't need to always introduce this atomic operation as part of
ending IO.

Add a helper to elevate the bi_remaining count, and flag the bio as
now actually needing the decrement at end_io time. Rename the field
to __bi_remaining to catch any current users of this doing the
incrementing manually.

For high IOPS workloads, this reduces the overhead of bio_endio()
substantially.
Tested-by: NRobert Elliott <elliott@hp.com>
Acked-by: NKent Overstreet <kent.overstreet@gmail.com>
Reviewed-by: NJan Kara <jack@suse.cz>
Signed-off-by: NJens Axboe <axboe@fb.com>
      c4cf5261
    • R
      Revert "dm crypt: fix deadlock when async crypto algorithm returns -EBUSY" · c0403ec0
      由 Rabin Vincent 提交于 
      This reverts Linux 4.1-rc1 commit 0618764c.

The problem which that commit attempts to fix actually lies in the
Freescale CAAM crypto driver not dm-crypt.

dm-crypt uses CRYPTO_TFM_REQ_MAY_BACKLOG.  This means the the crypto
driver should internally backlog requests which arrive when the queue is
full and process them later.  Until the crypto hw's queue becomes full,
the driver returns -EINPROGRESS.  When the crypto hw's queue if full,
the driver returns -EBUSY, and if CRYPTO_TFM_REQ_MAY_BACKLOG is set, is
expected to backlog the request and process it when the hardware has
queue space.  At the point when the driver takes the request from the
backlog and starts processing it, it calls the completion function with
a status of -EINPROGRESS.  The completion function is called (for a
second time, in the case of backlogged requests) with a status/err of 0
when a request is done.

Crypto drivers for hardware without hardware queueing use the helpers,
crypto_init_queue(), crypto_enqueue_request(), crypto_dequeue_request()
and crypto_get_backlog() helpers to implement this behaviour correctly,
while others implement this behaviour without these helpers (ccp, for
example).

dm-crypt (before the patch that needs reverting) uses this API
correctly.  It queues up as many requests as the hw queues will allow
(i.e. as long as it gets back -EINPROGRESS from the request function).
Then, when it sees at least one backlogged request (gets -EBUSY), it
waits till that backlogged request is handled (completion gets called
with -EINPROGRESS), and then continues.  The references to
af_alg_wait_for_completion() and af_alg_complete() in that commit's
commit message are irrelevant because those functions only handle one
request at a time, unlink dm-crypt.

The problem is that the Freescale CAAM driver, which that commit
describes as having being tested with, fails to implement the
backlogging behaviour correctly.  In cam_jr_enqueue(), if the hardware
queue is full, it simply returns -EBUSY without backlogging the request.
What the observed deadlock was is not described in the commit message
but it is obviously the wait_for_completion() in crypto_convert() where
dm-crypto would wait for the completion being called with -EINPROGRESS
in the case of backlogged requests.  This completion will never be
completed due to the bug in the CAAM driver.

Commit 0618764c incorrectly made dm-crypt wait for every request,
even when the driver/hardware queues are not full, which means that
dm-crypt will never see -EBUSY.  This means that that commit will cause
a performance regression on all crypto drivers which implement the API
correctly.

Revert it.  Correct backlog handling should be implemented in the CAAM
driver instead.

Cc'ing stable purely because commit 0618764c did.  If for some reason
a stable@ kernel did pick up commit 0618764c it should get reverted.
Signed-off-by: NRabin Vincent <rabin.vincent@axis.com>
Reviewed-by: NHoria Geanta <horia.geanta@freescale.com>
Cc: stable@vger.kernel.org
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
      c0403ec0
  19. 30 4月, 2015 2 次提交
    • M
      dm: fix free_rq_clone() NULL pointer when requeueing unmapped request · aa6df8dd
      由 Mike Snitzer 提交于 
      Commit 02233342 ("dm: optimize dm_mq_queue_rq to _not_ use kthread if
using pure blk-mq") mistakenly removed free_rq_clone()'s clone->q check
before testing clone->q->mq_ops.  It was an oversight to discontinue
that check for 1 of the 2 use-cases for free_rq_clone():
1) free_rq_clone() called when an unmapped original request is requeued
2) free_rq_clone() called in the request-based IO completion path

The clone->q check made sense for case #1 but not for #2.  However, we
cannot just reinstate the check as it'd mask a serious bug in the IO
completion case #2 -- no in-flight request should have an uninitialized
request_queue (basic block layer refcounting _should_ ensure this).

The NULL pointer seen for case #1 is detailed here:
https://www.redhat.com/archives/dm-devel/2015-April/msg00160.html

Fix this free_rq_clone() NULL pointer by simply checking if the
mapped_device's type is DM_TYPE_MQ_REQUEST_BASED (clone's queue is
blk-mq) rather than checking clone->q->mq_ops.  This avoids the need to
dereference clone->q, but a WARN_ON_ONCE is added to let us know if an
uninitialized clone request is being completed.
Reported-by: NBart Van Assche <bart.vanassche@sandisk.com>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
      aa6df8dd
    • C
      dm: only initialize the request_queue once · 3e6180f0
      由 Christoph Hellwig 提交于 
      Commit bfebd1cd ("dm: add full blk-mq support to request-based DM")
didn't properly account for the need to short-circuit re-initializing
DM's blk-mq request_queue if it was already initialized.

Otherwise, reloading a blk-mq request-based DM table (either manually
or via multipathd) resulted in errors, see:
 https://www.redhat.com/archives/dm-devel/2015-April/msg00132.html

Fix is to only initialize the request_queue on the initial table load
(when the mapped_device type is assigned).

This is better than having dm_init_request_based_blk_mq_queue() return
early if the queue was already initialized because it elevates the
constraint to a more meaningful location in DM core.  As such the
pre-existing early return in dm_init_request_based_queue() can now be
removed.

Fixes: bfebd1cd ("dm: add full blk-mq support to request-based DM")
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
      3e6180f0
  21. 28 4月, 2015 1 次提交
    • N
      block: destroy bdi before blockdev is unregistered. · 6cd18e71
      由 NeilBrown 提交于 
      Because of the peculiar way that md devices are created (automatically
when the device node is opened), a new device can be created and
registered immediately after the
	blk_unregister_region(disk_devt(disk), disk->minors);
call in del_gendisk().

Therefore it is important that all visible artifacts of the previous
device are removed before this call.  In particular, the 'bdi'.

Since:
commit c4db59d3
Author: Christoph Hellwig <hch@lst.de>
    fs: don't reassign dirty inodes to default_backing_dev_info

moved the
   device_unregister(bdi->dev);
call from bdi_unregister() to bdi_destroy() it has been quite easy to
lose a race and have a new (e.g.) "md127" be created after the
blk_unregister_region() call and before bdi_destroy() is ultimately
called by the final 'put_disk', which must come after del_gendisk().

The new device finds that the bdi name is already registered in sysfs
and complains

> [ 9627.630029] WARNING: CPU: 18 PID: 3330 at fs/sysfs/dir.c:31 sysfs_warn_dup+0x5a/0x70()
> [ 9627.630032] sysfs: cannot create duplicate filename '/devices/virtual/bdi/9:127'

We can fix this by moving the bdi_destroy() call out of
blk_release_queue() (which can happen very late when a refcount
reaches zero) and into blk_cleanup_queue() - which happens exactly when the md
device driver calls it.

Then it is only necessary for md to call blk_cleanup_queue() before
del_gendisk().  As loop.c devices are also created on demand by
opening the device node, we make the same change there.

Fixes: c4db59d3Reported-by: NAzat Khuzhin <a3at.mail@gmail.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: stable@vger.kernel.org (v4.0)
Signed-off-by: NNeilBrown <neilb@suse.de>
Reviewed-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <axboe@fb.com>
      6cd18e71
  23. 22 4月, 2015 7 次提交
    • E
      md/raid5: don't do chunk aligned read on degraded array. · 9ffc8f7c
      由 Eric Mei 提交于 
      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: NEric Mei <eric.mei@seagate.com>
Signed-off-by: NNeilBrown <neilb@suse.de>
      9ffc8f7c
    • N
      md/raid5: allow the stripe_cache to grow and shrink. · edbe83ab
      由 NeilBrown 提交于 
      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: NNeilBrown <neilb@suse.de>
      edbe83ab
    • N
      md/raid5: change ->inactive_blocked to a bit-flag. · 5423399a
      由 NeilBrown 提交于 
      This allows us to easily add more (atomic) flags.
Signed-off-by: NNeilBrown <neilb@suse.de>
      5423399a
    • N
      md/raid5: move max_nr_stripes management into grow_one_stripe and drop_one_stripe · 486f0644
      由 NeilBrown 提交于 
      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: NNeilBrown <neilb@suse.de>
      486f0644
    • N
      md/raid5: pass gfp_t arg to grow_one_stripe() · a9683a79
      由 NeilBrown 提交于 
      This is needed for future improvement to stripe cache management.
Signed-off-by: NNeilBrown <neilb@suse.de>
      a9683a79
    • M
      md/raid5: introduce configuration option rmw_level · d06f191f
      由 Markus Stockhausen 提交于 
      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: NMarkus Stockhausen <stockhausen@collogia.de>
Signed-off-by: NNeilBrown <neilb@suse.de>
      d06f191f
    • M
      md/raid5: activate raid6 rmw feature · 584acdd4
      由 Markus Stockhausen 提交于 
      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: NMarkus Stockhausen <stockhausen@collogia.de>
Signed-off-by: NNeilBrown <neilb@suse.de>
      584acdd4