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>

Following symptoms were observed:
1. After raid0->raid10 takeover operation we have array with 2
missing disks.
When we add disk for rebuild, recovery process starts as expected
but it does not finish- it stops at about 90%, md126_resync process
hangs in "D" state.
2. Similar behavior is when we have mounted raid0 array and we
execute takeover to raid10. After this when we try to unmount array-
it causes process umount hangs in "D"

In scenarios above processes hang at the same function- wait_barrier
in raid10.c.
Process waits in macro "wait_event_lock_irq" until the
"!conf->barrier" condition will be true.
In scenarios above it never happens.

Reason was that at the end of level_store, after calling pers->run,
we call mddev_resume. This calls pers->quiesce(mddev, 0) with
RAID10, that calls lower_barrier.
However raise_barrier hadn't been called on that 'conf' yet,
so conf->barrier becomes negative, which is bad.

This patch introduces setting conf->barrier=1 after takeover
operation. It prevents to become barrier negative after call
lower_barrier().
Signed-off-by: NKrzysztof Wojcik <krzysztof.wojcik@intel.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Add new parameter to 'sync_page_io'.

The new parameter allows us to distinguish between metadata and data
operations.  This becomes important later when we add the ability to
use separate devices for data and metadata.
Signed-off-by: NJonathan Brassow <jbrassow@redhat.com>

Noticed-by: NRussell King <linux@arm.linux.org.uk>
Signed-off-by: NJoe Perches <joe@perches.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

When we fail to start a raid10 for some reason, we call
md_unregister_thread to kill the thread that was created.

Unfortunately md_thread() will then make one call into the handler
(raid10d) even though md_wakeup_thread has not been called.  This is
not safe and as md_unregister_thread is called after mddev->private
has been set to NULL, it will definitely cause a NULL dereference.

So fix this at both ends:
 - md_thread should only call the handler if THREAD_WAKEUP has been
   set.
 - raid10 should call md_unregister_thread before setting things
   to NULL just like all the other raid modules do.

This is applicable to 2.6.35 and later.

Cc: stable@kernel.org
Reported-by: N"Citizen" <citizen_lee@thecus.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

bio_clone and bio_alloc allocate from a common bio pool.
If an md device is stacked with other devices that use this pool, or under
something like swap which uses the pool, then the multiple calls on
the pool can cause deadlocks.

So allocate a local bio pool for each md array and use that rather
than the common pool.

This pool is used both for regular IO and metadata updates.
Signed-off-by: NNeilBrown <neilb@suse.de>

Currently sync_page_io takes a 'bdev'.
Every caller passes 'rdev->bdev'.
We will soon want another field out of the rdev in sync_page_io,
So just pass the rdev instead of the bdev out of it.
Signed-off-by: NNeilBrown <neilb@suse.de>

bio_alloc can never fail (as it uses a mempool) but an block
indefinitely, especially if the caller is holding a reference to a
previously allocated bio.

So these to places which both handle failure and hold multiple bios
should not use bio_alloc, they should use bio_kmalloc.
Signed-off-by: NNeilBrown <neilb@suse.de>

It is not safe to allocate from a mempool while holding an item
previously allocated from that mempool as that can deadlock when the
mempool is close to exhaustion.

So don't use a bio list to collect the bios to write to multiple
devices in raid1 and raid10.
Instead queue each bio as it becomes available so an unplug will
activate all previously allocated bios and so a new bio has a chance
of being allocated.

This means we must set the 'remaining' count to '1' before submitting
any requests, then when all are submitted, decrement 'remaining' and
possible handle the write completion at that point.
Reported-by: NTorsten Kaiser <just.for.lkml@googlemail.com>
Tested-by: NTorsten Kaiser <just.for.lkml@googlemail.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

bitmap_get_counter returns the number of sectors covered
by the counter in a pass-by-reference variable.
In some cases this can be very large, so make it a sector_t
for safety.
Signed-off-by: NNeilBrown <neilb@suse.de>

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>

commit 7b6d91da changed the behaviour
of a few variables in raid1 and raid10 from flags to bit-sets, but
left them as type 'bool' so they did not work.

Change them (back) to unsigned long.
(historical note: see 1ef04fef)
Signed-off-by: NNeilBrown <neilb@suse.de>
Reported-by: Jiri Slaby <jslaby@suse.cz> and many others

md_check_recovery expects ->spare_active to return 'true' if any
spares were activated, but none of them do, so the consequent change
in 'degraded' is not notified through sysfs.

So count the number of spares activated, subtract it from 'degraded'
just once, and return it.
Reported-by: NAdrian Drzewiecki <adriand@vmware.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

When RAID1 is done syncing disks, it'll update the state
of synced rdevs to In_sync. But it neglected to notify
sysfs that the attribute changed. So any programs that
are waiting for an rdev's state to change will not be
woken.

(raid5/raid10 added by neilb)
Signed-off-by: NAdrian Drzewiecki <adriand@vmware.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Remove the current bio flags and reuse the request flags for the bio, too.
This allows to more easily trace the type of I/O from the filesystem
down to the block driver.  There were two flags in the bio that were
missing in the requests:  BIO_RW_UNPLUG and BIO_RW_AHEAD.  Also I've
renamed two request flags that had a superflous RW in them.

Note that the flags are in bio.h despite having the REQ_ name - as
blkdev.h includes bio.h that is the only way to go for now.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NJens Axboe <jaxboe@fusionio.com>

If the 'bio_split' path in raid10-read is used while
resync/recovery is happening it is possible to deadlock.
Fix this be elevating ->nr_waiting for the duration of both
parts of the split request.

This fixes a bug that has been present since 2.6.22
but has only started manifesting recently for unknown reasons.
It is suitable for and -stable since then.
Reported-by: NJustin Bronder <jsbronder@gentoo.org>
Tested-by: NJustin Bronder <jsbronder@gentoo.org>
Signed-off-by: NNeilBrown <neilb@suse.de>
Cc: stable@kernel.org

Use mddev->new_layout in setup_conf.
Also use new_chunk, and don't set ->degraded in takeover().  That
gets set in run()
Signed-off-by: NMaciej Trela <maciej.trela@intel.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Most array level changes leave the list of devices largely unchanged,
possibly causing one at the end to become redundant.
However conversions between RAID0 and RAID10 need to renumber
all devices (except 0).

This renumbering is currently being done in the ->run method when the
new personality takes over.  However this is too late as the common
code in md.c might already have invalidated some of the devices if
they had a ->raid_disk number that appeared to high.

Moving it into the ->takeover method is too early as the array is
still active at that time and wrong ->raid_disk numbers could cause
confusion.

So add a ->new_raid_disk field to mdk_rdev_s and use it to communicate
the new raid_disk number.
Now the common code knows exactly which devices need to be renumbered,
and which can be invalidated, and can do it all at a convenient time
when the array is suspend.
It can also update some symlinks in sysfs which previously were not be
updated correctly.
Reported-by: NMaciej Trela <maciej.trela@intel.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Such NULL pointer dereference can occur when the driver was fixing the
read errors/bad blocks and the disk was physically removed
causing a system crash. This patch check if the
rcu_dereference() returns valid rdev before accessing it in fix_read_error().

Cc: stable@kernel.org
Signed-off-by: NPrasanna S. Panchamukhi <prasanna.panchamukhi@riverbed.com>
Signed-off-by: NRob Becker <rbecker@riverbed.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

read_balance uses a "unsigned long" for a sector number which
will get truncated beyond 2TB.
This will cause read-balancing to be non-optimal, and can cause
data to be read from the 'wrong' branch during a resync.  This has a
very small chance of returning wrong data.
Reported-by: NJordan Russell <jr-list-2010@quo.to>
Cc: stable@kernel.org
Signed-off-by: NNeilBrown <neilb@suse.de>

All raid10 printk messages now start
   md/raid10:md-device-name:
Signed-off-by: NNeilBrown <neilb@suse.de>

We used to pass the personality make_request function direct
to the block layer so the first argument had to be a queue.
But now we have the intermediary md_make_request so it makes
at lot more sense to pass a struct mddev_s.
It makes it possible to have an mddev without its own queue too.
Signed-off-by: NNeilBrown <neilb@suse.de>

While I generally prefer letting personalities do as much as possible,
given that we have a central md_make_request anyway we may as well use
it to simplify code.
Also this centralises knowledge of ->gendisk which will help later.
Signed-off-by: NNeilBrown <neilb@suse.de>

Signed-off-by: NMaciej Trela <maciej.trela@intel.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

In a subsequent patch we will make it possible to change
mddev->raid_disks while a RAID0 or RAID10 array is active.  This is
part of the process of reshaping such an array.

This means that we cannot use this value while processes requests
(it is OK to use it during initialisation as we are locked against
changes then).
Both RAID0 and RAID10 have the same value stored in the private data
structure, so use that value instead.
Signed-off-by: NNeilBrown <neilb@suse.de>

void pointers do not need to be cast to other pointer types.
Signed-off-by: NH Hartley Sweeten <hsweeten@visionengravers.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

include cleanup: Update gfp.h and slab.h includes to prepare for breaking implicit slab.h inclusion from percpu.h

percpu.h is included by sched.h and module.h and thus ends up being
included when building most .c files.  percpu.h includes slab.h which
in turn includes gfp.h making everything defined by the two files
universally available and complicating inclusion dependencies.

percpu.h -> slab.h dependency is about to be removed.  Prepare for
this change by updating users of gfp and slab facilities include those
headers directly instead of assuming availability.  As this conversion
needs to touch large number of source files, the following script is
used as the basis of conversion.

  http://userweb.kernel.org/~tj/misc/slabh-sweep.py

The script does the followings.

* Scan files for gfp and slab usages and update includes such that
  only the necessary includes are there.  ie. if only gfp is used,
  gfp.h, if slab is used, slab.h.

* When the script inserts a new include, it looks at the include
  blocks and try to put the new include such that its order conforms
  to its surrounding.  It's put in the include block which contains
  core kernel includes, in the same order that the rest are ordered -
  alphabetical, Christmas tree, rev-Xmas-tree or at the end if there
  doesn't seem to be any matching order.

* If the script can't find a place to put a new include (mostly
  because the file doesn't have fitting include block), it prints out
  an error message indicating which .h file needs to be added to the
  file.

The conversion was done in the following steps.

1. The initial automatic conversion of all .c files updated slightly
   over 4000 files, deleting around 700 includes and adding ~480 gfp.h
   and ~3000 slab.h inclusions.  The script emitted errors for ~400
   files.

2. Each error was manually checked.  Some didn't need the inclusion,
   some needed manual addition while adding it to implementation .h or
   embedding .c file was more appropriate for others.  This step added
   inclusions to around 150 files.

3. The script was run again and the output was compared to the edits
   from #2 to make sure no file was left behind.

4. Several build tests were done and a couple of problems were fixed.
   e.g. lib/decompress_*.c used malloc/free() wrappers around slab
   APIs requiring slab.h to be added manually.

5. The script was run on all .h files but without automatically
   editing them as sprinkling gfp.h and slab.h inclusions around .h
   files could easily lead to inclusion dependency hell.  Most gfp.h
   inclusion directives were ignored as stuff from gfp.h was usually
   wildly available and often used in preprocessor macros.  Each
   slab.h inclusion directive was examined and added manually as
   necessary.

6. percpu.h was updated not to include slab.h.

7. Build test were done on the following configurations and failures
   were fixed.  CONFIG_GCOV_KERNEL was turned off for all tests (as my
   distributed build env didn't work with gcov compiles) and a few
   more options had to be turned off depending on archs to make things
   build (like ipr on powerpc/64 which failed due to missing writeq).

   * x86 and x86_64 UP and SMP allmodconfig and a custom test config.
   * powerpc and powerpc64 SMP allmodconfig
   * sparc and sparc64 SMP allmodconfig
   * ia64 SMP allmodconfig
   * s390 SMP allmodconfig
   * alpha SMP allmodconfig
   * um on x86_64 SMP allmodconfig

8. percpu.h modifications were reverted so that it could be applied as
   a separate patch and serve as bisection point.

Given the fact that I had only a couple of failures from tests on step
6, I'm fairly confident about the coverage of this conversion patch.
If there is a breakage, it's likely to be something in one of the arch
headers which should be easily discoverable easily on most builds of
the specific arch.
Signed-off-by: NTejun Heo <tj@kernel.org>
Guess-its-ok-by: NChristoph Lameter <cl@linux-foundation.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>

If a component device has a merge_bvec_fn then as we never call it
we must ensure we never need to.  Currently this is done by setting
max_sector to 1 PAGE, however this does not stop a bio being created
with several sub-page iovecs that would violate the merge_bvec_fn.

So instead set max_segments to 1 and set the segment boundary to the
same as a page boundary to ensure there is only ever one single-page
segment of IO requested at a time.

This can particularly be an issue when 'xen' is used as it is
known to submit multiple small buffers in a single bio.
Signed-off-by: NNeilBrown <neilb@suse.de>
Cc: stable@kernel.org

The block layer calling convention is blk_queue_<limit name>.
blk_queue_max_sectors predates this practice, leading to some confusion.
Rename the function to appropriately reflect that its intended use is to
set max_hw_sectors.

Also introduce a temporary wrapper for backwards compability.  This can
be removed after the merge window is closed.
Signed-off-by: NMartin K. Petersen <martin.petersen@oracle.com>
Signed-off-by: NJens Axboe <jens.axboe@oracle.com>

Suggested by  Oren Held <orenhe@il.ibm.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

We've noticed severe lasting performance degradation of our raid
arrays when we have drives that yield large amounts of media errors.
The raid10 module will queue each failed read for retry, and also
will attempt call fix_read_error() to perform the read recovery.
Read recovery is performed while the array is frozen, so repeated
recovery attempts can degrade the performance of the array for
extended periods of time.

With this patch I propose adding a per md device max number of
corrected read attempts.  Each rdev will maintain a count of
read correction attempts in the rdev->read_errors field (not
used currently for raid10). When we enter fix_read_error()
we'll check to see when the last read error occurred, and
divide the read error count by 2 for every hour since the
last read error. If at that point our read error count
exceeds the read error threshold, we'll fail the raid device.

In addition in this patch I add sysfs nodes (get/set) for
the per md max_read_errors attribute, the rdev->read_errors
attribute, and added some printk's to indicate when
fix_read_error fails to repair an rdev.

For testing I used debugfs->fail_make_request to inject
IO errors to the rdev while doing IO to the raid array.
Signed-off-by: NRobert Becker <Rob.Becker@riverbed.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

When we get a read error on a device in a RAID10, and attempting to
repair the error fails, print more useful messages about why it
failed.
Signed-off-by: NRobert Becker <Rob.Becker@riverbed.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

As bitmap_create and bitmap_destroy already set thread->timeout
as appropriate, there is no need to do it in raid10_quiesce.
There is a possible need to wake the thread after the timeout
has been set low, but it is better to do that where the timeout
is actually set low, in bitmap_create.
Signed-off-by: NNeilBrown <neilb@suse.de>

This removes a lot of multiplications by HZ.
Signed-off-by: NNeilBrown <neilb@suse.de>

... and into bitmap_info.  These are all configuration parameters
that need to be set before the bitmap is created.
Signed-off-by: NNeilBrown <neilb@suse.de>

Previously barriers were only supported on RAID1.  This is because
other levels requires synchronisation across all devices and so needed
a different approach.
Here is that approach.

When a barrier arrives, we send a zero-length barrier to every active
device.  When that completes - and if the original request was not
empty -  we submit the barrier request itself (with the barrier flag
cleared) and then submit a fresh load of zero length barriers.

The barrier request itself is asynchronous, but any subsequent
request will block until the barrier completes.

The reason for clearing the barrier flag is that a barrier request is
allowed to fail.  If we pass a non-empty barrier through a striping
raid level it is conceivable that part of it could succeed and part
could fail.  That would be way too hard to deal with.
So if the first run of zero length barriers succeed, we assume all is
sufficiently well that we send the request and ignore errors in the
second run of barriers.

RAID5 needs extra care as write requests may not have been submitted
to the underlying devices yet.  So we flush the stripe cache before
proceeding with the barrier.

Note that the second set of zero-length barriers are submitted
immediately after the original request is submitted.  Thus when
a personality finds mddev->barrier to be set during make_request,
it should not return from make_request until the corresponding
per-device request(s) have been queued.

That will be done in later patches.
Signed-off-by: NNeilBrown <neilb@suse.de>
Reviewed-by: NAndre Noll <maan@systemlinux.org>

Both raid1 and raid10 create a mempool during startup.
If the 'alloc' function for this mempool fails, unplug_slaves
is called.
If that happens when the pool is being initialised, unplug_slaves
will try to use the 'conf' structure that isn't filled in yet, and
badness will happen.

So ensure that unplug_slaves doesn't get called unless we know
that the conf structure if fully initialised.
Signed-off-by: NNeilBrown <neilb@suse.de>

During 'check' of a raid1 or raid10 it is possible for the management
thread to spend a lot of time running 'memcmp' on blocks from
different devices, so make sure the thread has a chance to schedule.
raid5d already has a cond_resched (in process_stripe).
Reported-By: NLee Howard <faxguy@howardsilvan.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Recently Jens has changed bio_rw_flagged() logic by following
commit 1f98a13f. Now it returns
bool instead of int. This broke raid1/raid10 RW bits manipulation logic.
One of visible result is BUG_ON triggering due to empty barrier
here scsi_lib.c:1108 scsi_setup_fs_cmnd()
Signed-off-by: NDmitry Monakhov <dmonakhov@openvz.org>
Signed-off-by: NNeilBrown <neilb@suse.de>

This should writeback from coming when the device is temporarily
suspended.
Signed-off-by: NNeilBrown <neilb@suse.de>