Move stats related fields - stamp, in_flight, dkstats - from disk to
part0 and unify stat handling such that...

* part_stat_*() now updates part0 together if the specified partition
  is not part0.  ie. part_stat_*() are now essentially all_stat_*().

* {disk|all}_stat_*() are gone.

* part_round_stats() is updated similary.  It handles part0 stats
  automatically and disk_round_stats() is killed.

* part_{inc|dec}_in_fligh() is implemented which automatically updates
  part0 stats for parts other than part0.

* disk_map_sector_rcu() is updated to return part0 if no part matches.
  Combined with the above changes, this makes NULL special case
  handling in callers unnecessary.

* Separate stats show code paths for disk are collapsed into part
  stats show code paths.

* Rename disk_stat_lock/unlock() to part_stat_lock/unlock()

While at it, reposition stat handling macros a bit and add missing
parentheses around macro parameters.
Signed-off-by: NTejun Heo <tj@kernel.org>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

There are two variants of stat functions - ones prefixed with double
underbars which don't care about preemption and ones without which
disable preemption before manipulating per-cpu counters.  It's unclear
whether the underbarred ones assume that preemtion is disabled on
entry as some callers don't do that.

This patch unifies diskstats access by implementing disk_stat_lock()
and disk_stat_unlock() which take care of both RCU (for partition
access) and preemption (for per-cpu counter access).  diskstats access
should always be enclosed between the two functions.  As such, there's
no need for the versions which disables preemption.  They're removed
and double underbars ones are renamed to drop the underbars.  As an
extra argument is added, there's no danger of using the old version
unconverted.

disk_stat_lock() uses get_cpu() and returns the cpu index and all
diskstat functions which access per-cpu counters now has @cpu
argument to help RT.

This change adds RCU or preemption operations at some places but also
collapses several preemption ops into one at others.  Overall, the
performance difference should be negligible as all involved ops are
very lightweight per-cpu ones.
Signed-off-by: NTejun Heo <tj@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

raid5 can overflow with more than 255 stripes, and we can increase it
to an int for free on both 32 and 64-bit archs due to the padding.
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

When we have externally managed metadata, we need to mark a failed
device as 'Blocked' and not allow any writes until that device
have been marked as faulty in the metadata and the Blocked flag has
been removed.

However it is perfectly OK to allow read requests when there is a
Blocked device, and with a readonly array, there may not be any
metadata-handler watching for blocked devices.

So in raid5/raid6 only allow a Blocked device to interfere with
Write request or resync.  Read requests go through untouched.

raid1 and raid10 already differentiate between read and write
properly.
Signed-off-by: NNeilBrown <neilb@suse.de>

We cannot currently change the size of a write-intent bitmap.
So if we change the size of an array which has such a bitmap, it
tries to set bits beyond the end of the bitmap.

For now, simply reject any request to change the size of an array
which has a bitmap.  mdadm can remove the bitmap and add a new one
after the array has changed size.
Signed-off-by: NNeilBrown <neilb@suse.de>

handle_stripe will take no action on a stripe when waiting for userspace
to unblock the array, so do not report completed sectors.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

The original STRIPE_OP_IO removal patch had the following hunk:

-               for (i = conf->raid_disks; i--; ) {
+               for (i = conf->raid_disks; i--; )
                        set_bit(R5_Wantwrite, &sh->dev[i].flags);
-                       if (!test_and_set_bit(STRIPE_OP_IO, &sh->ops.pending))
-                               sh->ops.count++;
-               }

However it appears the hunk became broken after merging:
-               for (i = conf->raid_disks; i--; ) {
+               for (i = conf->raid_disks; i--; )
                        set_bit(R5_Wantwrite, &sh->dev[i].flags);
                        set_bit(R5_LOCKED, &dev->flags);
                        s.locked++;
-                       if (!test_and_set_bit(STRIPE_OP_IO, &sh->ops.pending))
-                               sh->ops.count++;
-               }
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

Some dma drivers need to call spin_lock_bh in their device_issue_pending
routines.  This change avoids:

WARNING: at kernel/softirq.c:136 local_bh_enable_ip+0x3a/0x85()
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

This patch renames the array_size field of struct mddev_s to array_sectors
and converts all instances to use units of 512 byte sectors instead of 1k
blocks.
Signed-off-by: NAndre Noll <maan@systemlinux.org>
Signed-off-by: NNeilBrown <neilb@suse.de>

Remove the dubious attempt to prefer 'compute' over 'read'.  Not only is it
wrong given commit c337869d (md: do not compute parity unless it is on a failed
drive), but it can trigger a BUG_ON in handle_parity_checks5().

Cc: <stable@kernel.org>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

When devices are stacked, one device's merge_bvec_fn may need to perform
the mapping and then call one or more functions for its underlying devices.

The following bio fields are used:
  bio->bi_sector
  bio->bi_bdev
  bio->bi_size
  bio->bi_rw  using bio_data_dir()

This patch creates a new struct bvec_merge_data holding a copy of those
fields to avoid having to change them directly in the struct bio when
going down the stack only to have to change them back again on the way
back up.  (And then when the bio gets mapped for real, the whole
exercise gets repeated, but that's a problem for another day...)
Signed-off-by: NAlasdair G Kergon <agk@redhat.com>
Cc: Neil Brown <neilb@suse.de>
Cc: Milan Broz <mbroz@redhat.com>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

md_allow_write() marks the metadata dirty while holding mddev->lock and then
waits for the write to complete.  For externally managed metadata this causes a
deadlock as userspace needs to take the lock to communicate that the metadata
update has completed.

Change md_allow_write() in the 'external' case to start the 'mark active'
operation and then return -EAGAIN.  The expected side effects while waiting for
userspace to write 'active' to 'array_state' are holding off reshape (code
currently handles -ENOMEM), cause some 'stripe_cache_size' change requests to
fail, cause some GET_BITMAP_FILE ioctl requests to fall back to GFP_NOIO, and
cause updates to 'raid_disks' to fail.  Except for 'stripe_cache_size' changes
these failures can be mitigated by coordinating with mdmon.

md_write_start() still prevents writes from occurring until the metadata
handler has had a chance to take action as it unconditionally waits for
MD_CHANGE_CLEAN to be cleared.

[neilb@suse.de: return -EAGAIN, try GFP_NOIO]
Signed-off-by: NDan Williams <dan.j.williams@intel.com>

From: Dan Williams <dan.j.williams@intel.com>

Commit a4456856 refactored some of the deep code paths in raid5.c into separate
functions.  The names chosen at the time do not consistently indicate what is
going to happen to the stripe.  So, update the names, and since a stripe is a
cache element use cache semantics like fill, dirty, and clean.

(also, fix up the indentation in fetch_block5)
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

From: Dan Williams <dan.j.williams@intel.com>

Neil said:
> At the end of ops_run_compute5 you have:
>         /* ack now if postxor is not set to be run */
>         if (tx && !test_bit(STRIPE_OP_POSTXOR, &s->ops_run))
>                 async_tx_ack(tx);
>
> It looks odd having that test there.  Would it fit in raid5_run_ops
> better?

The intended global interpretation is that raid5_run_ops can build a chain
of xor and memcpy operations.  When MD registers the compute-xor it tells
async_tx to keep the operation handle around so that another item in the
dependency chain can be submitted. If we are just computing a block to
satisfy a read then we can terminate the chain immediately.  raid5_run_ops
gives a better context for this test since it cares about the entire chain.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

From: Dan Williams <dan.j.williams@intel.com>

Currently ops_run_biodrain and other locations have extra logic to determine
which blocks are processed in the prexor and non-prexor cases.  This can be
eliminated if handle_write_operations5 flags the blocks to be processed in all
cases via R5_Wantdrain.  The presence of the prexor operation is tracked in
sh->reconstruct_state.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

From: Dan Williams <dan.j.williams@intel.com>

Track the state of reconstruct operations (recalculating the parity block
usually due to incoming writes, or as part of array expansion)  Reduces the
scope of the STRIPE_OP_{BIODRAIN,PREXOR,POSTXOR} flags to only tracking whether
a reconstruct operation has been requested via the ops_request field of struct
stripe_head_state.

This is the final step in the removal of ops.{pending,ack,complete,count}, i.e.
the STRIPE_OP_{BIODRAIN,PREXOR,POSTXOR} flags only request an operation and do
not track the state of the operation.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

From: Dan Williams <dan.j.williams@intel.com>

Track the state of compute operations (recalculating a block from all the other
blocks in a stripe) with a state flag.  Reduces the scope of the
STRIPE_OP_COMPUTE_BLK flag to only tracking whether a compute operation has
been requested via the ops_request field of struct stripe_head_state.

Note, the compute operation that is performed in the course of doing a 'repair'
operation (check the parity block, recalculate it and write it back if the
check result is not zero) is tracked separately with the 'check_state'
variable.  Compute operations are held off while a 'check' is in progress, and
moving this check out to handle_issuing_new_read_requests5 the helper routine
__handle_issuing_new_read_requests5 can be simplified.

This is another step towards the removal of ops.{pending,ack,complete,count},
i.e. STRIPE_OP_COMPUTE_BLK only requests an operation and does not track the
state of the operation.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

From: Dan Williams <dan.j.williams@intel.com>

Track the state of read operations (copying data from the stripe cache to bio
buffers outside the lock) with a state flag.  Reduce the scope of the
STRIPE_OP_BIOFILL flag to only tracking whether a biofill operation has been
requested via the ops_request field of struct stripe_head_state.

This is another step towards the removal of ops.{pending,ack,complete,count},
i.e. STRIPE_OP_BIOFILL only requests an operation and does not track the state
of the operation.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

From: Dan Williams <dan.j.williams@intel.com>

The STRIPE_OP_* flags record the state of stripe operations which are
performed outside the stripe lock.  Their use in indicating which
operations need to be run is straightforward; however, interpolating what
the next state of the stripe should be based on a given combination of
these flags is not straightforward, and has led to bugs.  An easier to read
implementation with minimal degrees of freedom is needed.

Towards this goal, this patch introduces explicit states to replace what was
previously interpolated from the STRIPE_OP_* flags.  For now this only converts
the handle_parity_checks5 path, removing a user of the
ops.{pending,ack,complete,count} fields of struct stripe_operations.

This conversion also found a remaining issue with the current code.  There is
a small window for a drive to fail between when we schedule a repair and when
the parity calculation for that repair completes.  When this happens we will
writeback to 'failed_num' when we really want to write back to 'pd_idx'.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

From: Dan Williams <dan.j.williams@intel.com>

Let the raid6 path call ops_run_io to get pending i/o submitted.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

From: Dan Williams <dan.j.williams@intel.com>

In handle_stripe after taking sh->lock we sample some bits into 's' (struct
stripe_head_state):

	s.syncing = test_bit(STRIPE_SYNCING, &sh->state);
	s.expanding = test_bit(STRIPE_EXPAND_SOURCE, &sh->state);
	s.expanded = test_bit(STRIPE_EXPAND_READY, &sh->state);

Use these values from 's' in ops_run_io() rather than re-sampling the bits.
This ensures a consistent snapshot (as seen under sh->lock) is used.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

From: Dan Williams <dan.j.williams@intel.com>

The R5_Want{Read,Write} flags already gate i/o.  So, this flag is
superfluous and we can unconditionally call ops_run_io().
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

From: Dan Williams <dan.j.williams@intel.com>

This micro-optimization allowed the raid code to skip a re-read of the
parity block after checking parity.  It took advantage of the fact that
xor-offload-engines have their own internal result buffer and can check
parity without writing to memory.  Remove it for the following reasons:

1/ It is a layering violation for MD to need to manage the DMA and
   non-DMA paths within async_xor_zero_sum
2/ Bad precedent to toggle the 'ops' flags outside the lock
3/ Hard to realize a performance gain as reads will not need an updated
   parity block and writes will dirty it anyways.
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

For all array types but linear, ->hot_add_disk returns 1 on
success, 0 on failure.
For linear, it returns 0 on success and -errno on failure.

This doesn't cause a functional problem because the ->hot_add_disk
function of linear is used quite differently to the others.
However it is confusing.

So convert all to return 0 for success or -errno on failure
and fix call sites to match.
Signed-off-by: NNeil Brown <neilb@suse.de>

i.e. extend the 'md/dev-XXX/slot' attribute so that you can
tell a device to fill an vacant slot in an and md array.
Signed-off-by: NNeil Brown <neilb@suse.de>

Turn calls to bi->bi_end_io() into bio_endio(). Apparently bio_endio does
exactly the same error processing as is hardcoded at these places.

bio_endio() avoids recursion (or will soon), so it should be used.
Signed-off-by: NMikulas Patocka <mpatocka@redhat.com>
Signed-off-by: NNeil Brown <neilb@suse.de>

We shouldn't acknowledge that a stripe has been expanded (When
reshaping a raid5 by adding a device) until the moved data has
actually been written out.  However we are currently
acknowledging (by calling md_done_sync) when the POST_XOR
is complete and before the write.

So track in s.locked whether there are pending writes, and don't
call md_done_sync yet if there are.

Note: we all set R5_LOCKED on devices which are are about to
read from.  This probably isn't technically necessary, but is
usually done when writing a block, and justifies the use of
s.locked here.

This bug can lead to a crash if an array is stopped while an reshape
is in progress.

Cc: <stable@kernel.org>
Signed-off-by: NNeil Brown <neilb@suse.de>

If, while assembling an array, we find a device which is not fully
in-sync with the array, it is important to set the "fullsync" flags.
This is an exact analog to the setting of this flag in hot_add_disk
methods.

Currently, only v1.x metadata supports having devices in an array
which are not fully in-sync (it keep track of how in sync they are).
The 'fullsync' flag only makes a difference when a write-intent bitmap
is being used.  In this case it tells recovery to ignore the bitmap
and recovery all blocks.

This fix is already in place for raid1, but not raid5/6 or raid10.

So without this fix, a raid1 ir raid4/5/6 array with version 1.x
metadata and a write intent bitmaps, that is stopped in the middle
of a recovery, will appear to complete the recovery instantly
after it is reassembled, but the recovery will not be correct.

If you might have an array like that, issueing
   echo repair > /sys/block/mdXX/md/sync_action

will make sure recovery completes properly.

Cc: <stable@kernel.org>
Signed-off-by: NNeil Brown <neilb@suse.de>

If a block is computed (rather than read) then a check/repair operation
may be lead to believe that the data on disk is correct, when infact it
isn't.  So only compute blocks for failed devices.

This issue has been around since at least 2.6.12, but has become harder to
hit in recent kernels since most reads bypass the cache.

echo repair > /sys/block/mdN/md/sync_action will set the parity blocks to the
correct state.

Cc: <stable@kernel.org>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

During the initial array synchronization process there is a window between
when a prexor operation is scheduled to a specific stripe and when it
completes for a sync_request to be scheduled to the same stripe.  When
this happens the prexor completes and the stripe is unconditionally marked
"insync", effectively canceling the sync_request for the stripe.  Prior to
2.6.23 this was not a problem because the prexor operation was done under
sh->lock.  The effect in older kernels being that the prexor would still
erroneously mark the stripe "insync", but sync_request would be held off
and re-mark the stripe as "!in_sync".

Change the write completion logic to not mark the stripe "in_sync" if a
prexor was performed.  The effect of the change is to sometimes not set
STRIPE_INSYNC.  The worst this can do is cause the resync to stall waiting
for STRIPE_INSYNC to be set.  If this were happening, then STRIPE_SYNCING
would be set and handle_issuing_new_read_requests would cause all
available blocks to eventually be read, at which point prexor would never
be used on that stripe any more and STRIPE_INSYNC would eventually be set.

echo repair > /sys/block/mdN/md/sync_action will correct arrays that may
have lost this race.

Cc: <stable@kernel.org>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

When we get any IO error during a recovery (rebuilding a spare), we abort
the recovery and restart it.

For RAID6 (and multi-drive RAID1) it may not be best to restart at the
beginning: when multiple failures can be tolerated, the recovery may be
able to continue and re-doing all that has already been done doesn't make
sense.

We already have the infrastructure to record where a recovery is up to
and restart from there, but it is not being used properly.
This is because:
  - We sometimes abort with MD_RECOVERY_ERR rather than just MD_RECOVERY_INTR,
    which causes the recovery not be be checkpointed.
  - We remove spares and then re-added them which loses important state
    information.

The distinction between MD_RECOVERY_ERR and MD_RECOVERY_INTR really isn't
needed.  If there is an error, the relevant drive will be marked as
Faulty, and that is enough to ensure correct handling of the error.  So we
first remove MD_RECOVERY_ERR, changing some of the uses of it to
MD_RECOVERY_INTR.

Then we cause the attempt to remove a non-faulty device from an array to
fail (unless recovery is impossible as the array is too degraded).  Then
when remove_and_add_spares attempts to remove the devices on which
recovery can continue, it will fail, they will remain in place, and
recovery will continue on them as desired.

Issue:  If we are halfway through rebuilding a spare and another drive
fails, and a new spare is immediately available,  do we want to:
 1/ complete the current rebuild, then go back and rebuild the new spare or
 2/ restart the rebuild from the start and rebuild both devices in
    parallel.

Both options can be argued for.  The code currently takes option 2 as
  a/ this requires least code change
  b/ this results in a minimally-degraded array in minimal time.

Cc: "Eivind Sarto" <ivan@kasenna.com>
Signed-off-by: NNeil Brown <neilb@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Last night we had scsi problems and a hardware raid unit was offlined
during heavy i/o.  While this happened we got for about 3 minutes a huge
number messages like these

Apr 12 03:36:07 pfs1n14 kernel: [197510.696595] raid5:md7: read error not correctable (sector 2993096568 on sdj2).

I guess the high error rate is responsible for not scheduling other events
- during this time the system was not pingable and in the end also other
devices run into scsi command timeouts causing problems on these unrelated
devices as well.
Signed-off-by: NBernd Schubert <bernd-schubert@gmx.de>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

As setting and clearing queue flags now requires that we hold a spinlock
on the queue, and as blk_queue_stack_limits is called without that lock,
get the lock inside blk_queue_stack_limits.

For blk_queue_stack_limits to be able to find the right lock, each md
personality needs to set q->queue_lock to point to the appropriate lock.
Those personalities which didn't previously use a spin_lock, us
q->__queue_lock.  So always initialise that lock when allocated.

With this in place, setting/clearing of the QUEUE_FLAG_PLUGGED bit will no
longer cause warnings as it will be clear that the proper lock is held.

Thanks to Dan Williams for review and fixing the silly bugs.
Signed-off-by: NNeilBrown <neilb@suse.de>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Jens Axboe <jens.axboe@oracle.com>
Cc: Alistair John Strachan <alistair@devzero.co.uk>
Cc: Nick Piggin <npiggin@suse.de>
Cc: "Rafael J. Wysocki" <rjw@sisk.pl>
Cc: Jacek Luczak <difrost.kernel@gmail.com>
Cc: Prakash Punnoor <prakash@punnoor.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

commit bd2ab670 "md: close a livelock window
in handle_parity_checks5" introduced a bug in handling 'repair' operations.
After a repair operation completes we clear the state bits tracking this
operation.  However, they are cleared too early and this results in the code
deciding to re-run the parity check operation.  Since we have done the repair
in memory the second check does not find a mismatch and thus does not do a
writeback.

Test results:
$ echo repair > /sys/block/md0/md/sync_action
$ cat /sys/block/md0/md/mismatch_cnt
51072
$ echo repair > /sys/block/md0/md/sync_action
$ cat /sys/block/md0/md/mismatch_cnt
0

(also fix incorrect indentation)

Cc: <stable@kernel.org>
Tested-by: NGeorge Spelvin <linux@horizon.com>
Acked-by: NNeilBrown <neilb@suse.de>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Allows a userspace metadata handler to take action upon detecting a device
failure.

Based on an original patch by Neil Brown.

Changes:
-added blocked_wait waitqueue to rdev
-don't qualify Blocked with Faulty always let userspace block writes
-added md_wait_for_blocked_rdev to wait for the block device to be clear, if
 userspace misses the notification another one is sent every 5 seconds
-set MD_RECOVERY_NEEDED after clearing "blocked"
-kill DoBlock flag, just test mddev->external
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

MD drivers use one printk() call to print 2 log messages and the second line
may be prefixed by a TAB character.  It may also output a trailing space
before newline.  klogd (I think) turns the TAB character into the 2 characters
'^I' when logging to a file.  This looks ugly.

Instead of a leading TAB to indicate continuation, prefix both output lines
with 'raid:' or similar.  Also remove any trailing space in the vicinity of
the affected code and consistently end the sentences with a period.
Signed-off-by: NNick Andrew <nick@nick-andrew.net>
Cc: Neil Brown <neilb@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

strict_strtoul handles the open-coded sanity checks in
raid5_store_stripe_cache_size and raid5_store_preread_threshold
Acked-by: NNeilBrown <neilb@suse.de>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Improve write performance by preventing the delayed_list from dumping all its
stripes onto the handle_list in one shot.  Delayed stripes are now further
delayed by being held on the 'hold_list'.  The 'hold_list' is bypassed when:

  * a STRIPE_IO_STARTED stripe is found at the head of 'handle_list'
  * 'handle_list' is empty and i/o is being done to satisfy full stripe-width
    write requests
  * 'bypass_count' is less than 'bypass_threshold'.  By default the threshold
    is 1, i.e. every other stripe handled is a preread stripe provided the
    top two conditions are false.

Benchmark data:
System: 2x Xeon 5150, 4x SATA, mem=1GB
Baseline: 2.6.24-rc7
Configuration: mdadm --create /dev/md0 /dev/sd[b-e] -n 4 -l 5 --assume-clean
Test1: dd if=/dev/zero of=/dev/md0 bs=1024k count=2048
  * patched:  +33% (stripe_cache_size = 256), +25% (stripe_cache_size = 512)

Test2: tiobench --size 2048 --numruns 5 --block 4096 --block 131072 (XFS)
  * patched: +13%
  * patched + preread_bypass_threshold = 0: +37%

Changes since v1:
* reduce bypass_threshold from (chunk_size / sectors_per_chunk) to (1) and
  make it configurable.  This defaults to fairness and modest performance
  gains out of the box.
Changes since v2:
* [neilb@suse.de]: kill STRIPE_PRIO_HI and preread_needed as they are not
  necessary, the important change was clearing STRIPE_DELAYED in
  add_stripe_bio and this has been moved out to make_request for the hang
  fix.
* [neilb@suse.de]: simplify get_priority_stripe
* [dan.j.williams@intel.com]: reset the bypass_count when ->hold_list is
  sampled empty (+11%)
* [dan.j.williams@intel.com]: decrement the bypass_count at the detection
  of stripes being naturally promoted off of hold_list +2%.  Note, resetting
  bypass_count instead of decrementing on these events yields +4% but that is
  probably too aggressive.
Changes since v3:
* cosmetic fixups
Tested-by: NJames W. Laferriere <babydr@baby-dragons.com>
Signed-off-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeil Brown <neilb@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

__FUNCTION__ is gcc-specific, use __func__
Signed-off-by: NHarvey Harrison <harvey.harrison@gmail.com>
Cc: Neil Brown <neilb@suse.de>
Signed-off-by: NAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>