My editor shows much of this is RED.
Signed-off-by: NNeilBrown <neilb@suse.de>

The stripe cache has two goals:
1. cache data, so next time if data can be found in stripe cache, disk access
can be avoided.
2. stable data. data is copied from bio to stripe cache and calculated parity.
data written to disk is from stripe cache, so if upper layer changes bio data,
data written to disk isn't impacted.

In my environment, I can guarantee 2 will not happen. And BDI_CAP_STABLE_WRITES
can guarantee 2 too. For 1, it's not common too. block plug mechanism will
dispatch a bunch of sequentail small requests together. And since I'm using
SSD, I'm using small chunk size. It's rare case stripe cache is really useful.

So I'd like to avoid the copy from bio to stripe cache and it's very helpful
for performance. In my 1M randwrite tests, avoid the copy can increase the
performance more than 30%.

Of course, this shouldn't be enabled by default. It's reported enabling
BDI_CAP_STABLE_WRITES can harm some workloads before, so I added an option to
control it.

Neilb:
  changed BUG_ON to WARN_ON
  Removed some assignments from raid5_build_block which are now not needed.
Signed-off-by: NShaohua Li <shli@fusionio.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

track empty inactive list count, so md_raid5_congested() can use it to make
decision.
Signed-off-by: NShaohua Li <shli@fusionio.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

get_active_stripe() is the last place we have lock contention. It has two
paths. One is stripe isn't found and new stripe is allocated, the other is
stripe is found.

The first path basically calls __find_stripe and init_stripe. It accesses
conf->generation, conf->previous_raid_disks, conf->raid_disks,
conf->prev_chunk_sectors, conf->chunk_sectors, conf->max_degraded,
conf->prev_algo, conf->algorithm, the stripe_hashtbl and inactive_list. Except
stripe_hashtbl and inactive_list, other fields are changed very rarely.

With this patch, we split inactive_list and add new hash locks. Each free
stripe belongs to a specific inactive list. Which inactive list is determined
by stripe's lock_hash. Note, even a stripe hasn't a sector assigned, it has a
lock_hash assigned. Stripe's inactive list is protected by a hash lock, which
is determined by it's lock_hash too. The lock_hash is derivied from current
stripe_hashtbl hash, which guarantees any stripe_hashtbl list will be assigned
to a specific lock_hash, so we can use new hash lock to protect stripe_hashtbl
list too. The goal of the new hash locks introduced is we can only use the new
locks in the first path of get_active_stripe(). Since we have several hash
locks, lock contention is relieved significantly.

The first path of get_active_stripe() accesses other fields, since they are
changed rarely, changing them now need take conf->device_lock and all hash
locks. For a slow path, this isn't a problem.

If we need lock device_lock and hash lock, we always lock hash lock first. The
tricky part is release_stripe and friends. We need take device_lock first.
Neil's suggestion is we put inactive stripes to a temporary list and readd it
to inactive_list after device_lock is released. In this way, we add stripes to
temporary list with device_lock hold and remove stripes from the list with hash
lock hold. So we don't allow concurrent access to the temporary list, which
means we need allocate temporary list for all participants of release_stripe.

One downside is free stripes are maintained in their inactive list, they can't
across between the lists. By default, we have total 256 stripes and 8 lists, so
each list will have 32 stripes. It's possible one list has free stripe but
other list hasn't. The chance should be rare because stripes allocation are
even distributed. And we can always allocate more stripes for cache, several
mega bytes memory isn't a big deal.

This completely removes the lock contention of the first path of
get_active_stripe(). It slows down the second code path a little bit though
because we now need takes two locks, but since the hash lock isn't contended,
the overhead should be quite small (several atomic instructions). The second
path of get_active_stripe() (basically sequential write or big request size
randwrite) still has lock contentions.
Signed-off-by: NShaohua Li <shli@fusionio.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Signed-off-by: NMichael Opdenacker <michael.opdenacker@free-electrons.com>
Signed-off-by: NJiri Kosina <jkosina@suse.cz>

If there are not enough stripes to handle, we'd better not always
queue all available work_structs. If one worker can only handle small
or even none stripes, it will impact request merge and create lock
contention.

With this patch, the number of work_struct running will depend on
pending stripes number. Note: some statistics info used in the patch
are accessed without locking protection. This should doesn't matter,
we just try best to avoid queue unnecessary work_struct.
Signed-off-by: NShaohua Li <shli@fusionio.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

make_request() access various shape parameters (raid_disks, chunk_size
etc) which might be changed by raid5_start_reshape().

If the later is called at and awkward time during the form, the wrong
stripe_head might be used.

So introduce a 'seqcount' and after finding a stripe_head make sure
there is no reason to expect that we got the wrong one.
Signed-off-by: NNeilBrown <neilb@suse.de>

This is another attempt to create multiple threads to handle raid5 stripes.
This time I use workqueue.

raid5 handles request (especially write) in stripe unit. A stripe is page size
aligned/long and acrosses all disks. Writing to any disk sector, raid5 runs a
state machine for the corresponding stripe, which includes reading some disks
of the stripe, calculating parity, and writing some disks of the stripe. The
state machine is running in raid5d thread currently. Since there is only one
thread, it doesn't scale well for high speed storage. An obvious solution is
multi-threading.

To get better performance, we have some requirements:
a. locality. stripe corresponding to request submitted from one cpu is better
handled in thread in local cpu or local node. local cpu is preferred but some
times could be a bottleneck, for example, parity calculation is too heavy.
local node running has wide adaptability.
b. configurablity. Different setup of raid5 array might need diffent
configuration. Especially the thread number. More threads don't always mean
better performance because of lock contentions.

My original implementation is creating some kernel threads. There are
interfaces to control which cpu's stripe each thread should handle. And
userspace can set affinity of the threads. This provides biggest flexibility
and configurability. But it's hard to use and apparently a new thread pool
implementation is disfavor.

Recent workqueue improvement is quite promising. unbound workqueue will be
bound to numa node. If WQ_SYSFS is set in workqueue, there are sysfs option to
do affinity setting. For example, we can only include one HT sibling in
affinity. Since work is non-reentrant by default, and we can control running
thread number by limiting dispatched work_struct number.

In this patch, I created several stripe worker group. A group is a numa node.
stripes from cpus of one node will be added to a group list. Workqueue thread
of one node will only handle stripes of worker group of the node. In this way,
stripe handling has numa node locality. And as I said, we can control thread
number by limiting dispatched work_struct number.

The work_struct callback function handles several stripes in one run. A typical
work queue usage is to run one unit in each work_struct. In raid5 case, the
unit is a stripe. But we can't do that:
a. Though handling a stripe doesn't need lock because of reference accounting
and stripe isn't in any list, queuing a work_struct for each stripe will make
workqueue lock contended very heavily.
b. blk_start_plug()/blk_finish_plug() should surround stripe handle, as we
might dispatch request. If each work_struct only handles one stripe, such block
plug is meaningless.

This implementation can't do very fine grained configuration. But the numa
binding is most popular usage model, should be enough for most workloads.

Note: since we have only one stripe queue, switching to multi-thread might
decrease request size dispatching down to low level layer. The impact depends
on thread number, raid configuration and workload. So multi-thread raid5 might
not be proper for all setups.

Changes V1 -> V2:
1. remove WQ_NON_REENTRANT
2. disabling multi-threading by default
3. Add more descriptions in changelog
Signed-off-by: NShaohua Li <shli@fusionio.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

release_stripe still has big lock contention. We just add the stripe to a llist
without taking device_lock. We let the raid5d thread to do the real stripe
release, which must hold device_lock anyway. In this way, release_stripe
doesn't hold any locks.

The side effect is the released stripes order is changed. But sounds not a big
deal, stripes are never handled in order. And I thought block layer can already
do nice request merge, which means order isn't that important.

I kept the unplug release batch, which is unnecessary with this patch from lock
contention avoid point of view, and actually if we delete it, the stripe_head
release_list and lru can share storage. But the unplug release batch is also
helpful for request merge. We probably can delay wakeup raid5d till unplug, but
I'm still afraid of the case which raid5d is running.
Signed-off-by: NShaohua Li <shli@fusionio.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

If a device in a RAID4/5/6 is being replaced while another is being
recovered, then the writes to the replacement device currently don't
happen, resulting in corruption when the replacement completes and the
new drive takes over.

This is because the replacement writes are only triggered when
's.replacing' is set and not when the similar 's.sync' is set (which
is the case during resync and recovery - it means all devices need to
be read).

So schedule those writes when s.replacing is set as well.

In this case we cannot use "STRIPE_INSYNC" to record that the
replacement has happened as that is needed for recording that any
parity calculation is complete.  So introduce STRIPE_REPLACED to
record if the replacement has happened.

For safety we should also check that STRIPE_COMPUTE_RUN is not set.
This has a similar effect to the "s.locked == 0" test.  The latter
ensure that now IO has been flagged but not started.  The former
checks if any parity calculation has been flagged by not started.
We must wait for both of these to complete before triggering the
'replace'.

Add a similar test to the subsequent check for "are we finished yet".
This possibly isn't needed (is subsumed in the STRIPE_INSYNC test),
but it makes it more obvious that the REPLACE will happen before we
think we are finished.

Finally if a NeedReplace device is not UPTODATE then that is an
error.  We really must trigger a warning.

This bug was introduced in commit 9a3e1101
(md/raid5:  detect and handle replacements during recovery.)
which introduced replacement for raid5.
That was in 3.3-rc3, so any stable kernel since then would benefit
from this fix.

Cc: stable@vger.kernel.org (3.3+)
Reported-by: Nqindehua <13691222965@163.com>
Tested-by: Nqindehua <qindehua@163.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Once instance of this Kconfig macro remained after commit
51acbcec ("md: remove
CONFIG_MULTICORE_RAID456"). Remove that one too. And, while we're at it,
also remove it from the defconfig files that carry it.
Signed-off-by: NPaul Bolle <pebolle@tiscali.nl>
Signed-off-by: NNeilBrown <neilb@suse.de>

A number of problems can occur due to races between
resync/recovery and discard.

- if sync_request calls handle_stripe() while a discard is
  happening on the stripe, it might call handle_stripe_clean_event
  before all of the individual discard requests have completed
  (so some devices are still locked, but not all).
  Since commit ca64cae9
     md/raid5: Make sure we clear R5_Discard when discard is finished.
  this will cause R5_Discard to be cleared for the parity device,
  so handle_stripe_clean_event() will not be called when the other
  devices do become unlocked, so their ->written will not be cleared.
  This ultimately leads to a WARN_ON in init_stripe and a lock-up.

- If handle_stripe_clean_event() does clear R5_UPTODATE at an awkward
  time for resync, it can lead to s->uptodate being less than disks
  in handle_parity_checks5(), which triggers a BUG (because it is
  one).

So:
 - keep R5_Discard on the parity device until all other devices have
   completed their discard request
 - make sure we don't try to have a 'discard' and a 'sync' action at
   the same time.
   This involves a new stripe flag to we know when a 'discard' is
   happening, and the use of R5_Overlap on the parity disk so when a
   discard is wanted while a sync is active, so we know to wake up
   the discard at the appropriate time.

Discard support for RAID5 was added in 3.7, so this is suitable for
any -stable kernel since 3.7.

Cc: stable@vger.kernel.org (v3.7+)
Reported-by: NJes Sorensen <Jes.Sorensen@redhat.com>
Tested-by: NJes Sorensen <Jes.Sorensen@redhat.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Discard for raid4/5/6 has limitation. If discard request size is
small, we do discard for one disk, but we need calculate parity and
write parity disk.  To correctly calculate parity, zero_after_discard
must be guaranteed. Even it's true, we need do discard for one disk
but write another disks, which makes the parity disks wear out
fast. This doesn't make sense. So an efficient discard for raid4/5/6
should discard all data disks and parity disks, which requires the
write pattern to be (A, A+chunk_size, A+chunk_size*2...). If A's size
is smaller than chunk_size, such pattern is almost impossible in
practice. So in this patch, I only handle the case that A's size
equals to chunk_size. That is discard request should be aligned to
stripe size and its size is multiple of stripe size.

Since we can only handle request with specific alignment and size (or
part of the request fitting stripes), we can't guarantee
zero_after_discard even zero_after_discard is true in low level
drives.

The block layer doesn't send down correctly aligned requests even
correct discard alignment is set, so I must filter out.

For raid4/5/6 parity calculation, if data is 0, parity is 0. So if
zero_after_discard is true for all disks, data is consistent after
discard.  Otherwise, data might be lost. Let's consider a scenario:
discard a stripe, write data to one disk and write parity disk. The
stripe could be still inconsistent till then depending on using data
from other data disks or parity disks to calculate new parity. If the
disk is broken, we can't restore it. So in this patch, we only enable
discard support if all disks have zero_after_discard.

If discard fails in one disk, we face the similar inconsistent issue
above. The patch will make discard follow the same path as normal
write request. If discard fails, a resync will be scheduled to make
the data consistent. This isn't good to have extra writes, but data
consistency is important.

If a subsequent read/write request hits raid5 cache of a discarded
stripe, the discarded dev page should have zero filled, so the data is
consistent. This patch will always zero dev page for discarded request
stripe. This isn't optimal because discard request doesn't need such
payload. Next patch will avoid it.
Signed-off-by: NShaohua Li <shli@fusionio.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

make_request() does stripe release for every stripe and the stripe usually has
count 1, which makes previous release_stripe() optimization not work. In my
test, this release_stripe() becomes the heaviest pleace to take
conf->device_lock after previous patches applied.

Below patch makes stripe release batch. All the stripes will be released in
unplug. The STRIPE_ON_UNPLUG_LIST bit is to protect concurrent access stripe
lru.
Signed-off-by: NShaohua Li <shli@fusionio.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Because bios will merge at block-layer,so bios-error may caused by other
bio which be merged into to the same request.
Using this flag,it will find exactly error-sector and not do redundant
operation like re-write and re-read.

V0->V1:Using REQ_FLUSH instead REQ_NOMERGE avoid bio merging at block
layer.
Signed-off-by: NJianpeng Ma <majianpeng@gmail.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Add a per-stripe lock to protect stripe specific data. The purpose is to reduce
lock contention of conf->device_lock.

stripe ->toread, ->towrite are protected by per-stripe lock.  Accessing bio
list of the stripe is always serialized by this lock, so adding bio to the
lists (add_stripe_bio()) and removing bio from the lists (like
ops_run_biofill()) not race.

If bio in ->read, ->written ... list are not shared by multiple stripes, we
don't need any lock to protect ->read, ->written, because STRIPE_ACTIVE will
protect them. If the bio are shared,  there are two protections:
1. bi_phys_segments acts as a reference count
2. traverse the list uses r5_next_bio, which makes traverse never access bio
not belonging to the stripe

Let's have an example:
|  stripe1 |  stripe2    |  stripe3  |
...bio1......|bio2|bio3|....bio4.....

stripe2 has 4 bios, when it's finished, it will decrement bi_phys_segments for
all bios, but only end_bio for bio2 and bio3. bio1->bi_next still points to
bio2, but this doesn't matter. When stripe1 is finished, it will not touch bio2
because of r5_next_bio check. Next time stripe1 will end_bio for bio1 and
stripe3 will end_bio bio4.

before add_stripe_bio() addes a bio to a stripe, we already increament the bio
bi_phys_segments, so don't worry other stripes release the bio.
Signed-off-by: NShaohua Li <shli@fusionio.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

REQ_SYNC is ignored in current raid5 code. Block layer does use it to do
policy,
for example ioscheduler. This patch adds it.
Signed-off-by: NShaohua Li <shli@fusionio.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

The important issue here is incorporating the different in data_offset
into calculations concerning when we might need to over-write data
that is still thought to be valid.

To this end we find the minimum offset difference across all devices
and add that where appropriate.
Signed-off-by: NNeilBrown <neilb@suse.de>

During recovery we want to write to the replacement but not
the original.  So we have two new flags
 - R5_NeedReplace if this stripe has a replacement that needs to
   be written at some stage
 - R5_WantReplace if NeedReplace, and the data is available, and
   a 'sync' has been requested on this stripe.

We also distinguish between 'sync and replace' which need to read
all other devices, and 'replace' which only needs to read the
devices being replaced.

Note that during resync we always write to any replacement device.
It might not need to be written to, but as we don't read to compare,
we have to write to be sure.
Signed-off-by: NNeilBrown <neilb@suse.de>

When writing, we need to submit two writes, one to the original, and
one to the replacement - if there is a replacement.

If the write to the replacement results in a write error, we just fail
the device.  We only try to record write errors to the original.

When writing for recovery, we shouldn't write to the original.  This
will be addressed in a subsequent patch that generally addresses
recovery.
Reviewed-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Remove some #defines that are no longer used, and replace some
others with an enum.
And remove an unused field.
Reviewed-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Just enhance data structures to record a second device per slot to be
used as a 'replacement' device, replacing the original.
We also have a second bio in each slot in each stripe_head.  This will
only be used when writing to the array - we need to write to both the
original and the replacement at the same time, so will need two bios.

For now, only try using the replacement drive for aligned-reads.
In this case, we prefer the replacement if it has been recovered far
enough, otherwise use the original.

This includes a small enhancement.  Previously we would only do
aligned reads if the target device was fully recovered.  Now we also
do them if it has recovered far enough.
Reviewed-by: NDan Williams <dan.j.williams@intel.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Signed-off-by: NNeilBrown <neilb@suse.de>

Signed-off-by: NNeilBrown <neilb@suse.de>

Having mddev_t and 'struct mddev_s' is ugly and not preferred
Signed-off-by: NNeilBrown <neilb@suse.de>

The typedefs are just annoying. 'mdk' probably refers to 'md_k.h'
which used to be an include file that defined this thing.
Signed-off-by: NNeilBrown <neilb@suse.de>

On a successful write to a known bad block, flag the sh
so that raid5d can remove the known bad block from the list.
Signed-off-by: NNeilBrown <neilb@suse.de>

When a write error is detected, don't mark the device as failed
immediately but rather record the fact for handle_stripe to deal with.

Handle_stripe then attempts to record a bad block.  Only if that fails
does the device get marked as faulty.
Signed-off-by: NNeilBrown <neilb@suse.de>

If we get an uncorrectable read error - record a bad block rather than
failing the device.
And if these errors (which may be due to known bad blocks) cause
recovery to be impossible, record a bad block on the recovering
devices, or abort the recovery.

As we might abort a recovery without failing a device we need to teach
RAID5 about recovery_disabled handling.
Signed-off-by: NNeilBrown <neilb@suse.de>

Adding these three fields will allow more common code to be moved
to handle_stripe()

struct field rearrangement by Namhyung Kim.
Signed-off-by: NNeilBrown <neilb@suse.de>
Reviewed-by: NNamhyung Kim <namhyung@gmail.com>

'struct stripe_head_state' stores state about the 'current' stripe
that is passed around while handling the stripe.
For RAID6 there is an extension structure: r6_state, which is also
passed around.
There is no value in keeping these separate, so move the fields from
the latter into the former.

This means that all code now needs to treat s->failed_num as an small
array, but this is a small cost.
Signed-off-by: NNeilBrown <neilb@suse.de>
Reviewed-by: NNamhyung Kim <namhyung@gmail.com>

sh->lock is now mainly used to ensure that two threads aren't running
in the locked part of handle_stripe[56] at the same time.

That can more neatly be achieved with an 'active' flag which we set
while running handle_stripe.  If we find the flag is set, we simply
requeue the stripe for later by setting STRIPE_HANDLE.

For safety we take ->device_lock while examining the state of the
stripe and creating a summary in 'stripe_head_state / r6_state'.
This possibly isn't needed but as shared fields like ->toread,
->towrite are checked it is safer for now at least.

We leave the label after the old 'unlock' called "unlock" because it
will disappear in a few patches, so renaming seems pointless.

This leaves the stripe 'locked' for longer as we clear STRIPE_ACTIVE
later, but that is not a problem.
Signed-off-by: NNeilBrown <neilb@suse.de>
Reviewed-by: NNamhyung Kim <namhyung@gmail.com>

This is the start of a series of patches to remove sh->lock.

sync_request takes sh->lock before setting STRIPE_SYNCING to ensure
there is no race with testing it in handle_stripe[56].

Instead, use a new flag STRIPE_SYNC_REQUESTED and test it early
in handle_stripe[56] (after getting the same lock) and perform the
same set/clear operations if it was set.
Signed-off-by: NNeilBrown <neilb@suse.de>
Reviewed-by: NNamhyung Kim <namhyung@gmail.com>

md has some plugging infrastructure for RAID5 to use because the
normal plugging infrastructure required a 'request_queue', and when
called from dm, RAID5 doesn't have one of those available.

This relied on the ->unplug_fn callback which doesn't exist any more.

So remove all of that code, both in md and raid5.  Subsequent patches
with restore the plugging functionality.
Signed-off-by: NNeilBrown <neilb@suse.de>

Code has been converted over to the new explicit on-stack plugging,
and delay users have been converted to use the new API for that.
So lets kill off the old plugging along with aops->sync_page().
Signed-off-by: NJens Axboe <jaxboe@fusionio.com>

This patch converts md to support REQ_FLUSH/FUA instead of now
deprecated REQ_HARDBARRIER.  In the core part (md.c), the following
changes are notable.

* Unlike REQ_HARDBARRIER, REQ_FLUSH/FUA don't interfere with
  processing of other requests and thus there is no reason to mark the
  queue congested while FLUSH/FUA is in progress.

* REQ_FLUSH/FUA failures are final and its users don't need retry
  logic.  Retry logic is removed.

* Preflush needs to be issued to all member devices but FUA writes can
  be handled the same way as other writes - their processing can be
  deferred to request_queue of member devices.  md_barrier_request()
  is renamed to md_flush_request() and simplified accordingly.

For linear, raid0 and multipath, the core changes are enough.  raid1,
5 and 10 need the following conversions.

* raid1: Handling of FLUSH/FUA bio's can simply be deferred to
  request_queues of member devices.  Barrier related logic removed.

* raid5: Queue draining logic dropped.  FUA bit is propagated through
  biodrain and stripe resconstruction such that all the updated parts
  of the stripe are written out with FUA writes if any of the dirtying
  writes was FUA.  preread_active_stripes handling in make_request()
  is updated as suggested by Neil Brown.

* raid10: FUA bit needs to be propagated to write clones.

linear, raid0, 1, 5 and 10 tested.
Signed-off-by: NTejun Heo <tj@kernel.org>
Reviewed-by: NNeil Brown <neilb@suse.de>
Signed-off-by: NJens Axboe <jaxboe@fusionio.com>

Also remove remaining accesses to ->queue and ->gendisk when ->queue
is NULL (As it is in a DM target).
Signed-off-by: NNeilBrown <neilb@suse.de>

md/raid5 uses the plugging infrastructure provided by the block layer
and 'struct request_queue'.  However when we plug raid5 under dm there
is no request queue so we cannot use that.

So create a similar infrastructure that is much lighter weight and use
it for raid5.
Signed-off-by: NNeilBrown <neilb@suse.de>

the dm module will need this for dm-raid45.

Also only access ->queue->backing_dev_info->congested_fn
if ->queue actually exists.  It won't in a dm target.
Signed-off-by: NNeilBrown <neilb@suse.de>

We will shortly allow md devices with no gendisk (they are attached to
a dm-target instead).  That will cause mdname() to return 'mdX'.
There is one place where mdname really needs to be unique: when
creating the name for a slab cache.
So in that case, if there is no gendisk, you the address of the mddev
formatted in HEX to provide a unique name.
Signed-off-by: NNeilBrown <neilb@suse.de>