There isn't locking setting STRIPE_DELAYED and STRIPE_PREREAD_ACTIVE bits, but
the two bits have relationship. A delayed stripe can be moved to hold list only
when preread active stripe count is below IO_THRESHOLD. If a stripe has both
the bits set, such stripe will be in delayed list and preread count not 0,
which will make such stripe never leave delayed list.
Signed-off-by: NShaohua Li <shli@fusionio.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

We may not be able to fix a bad block if:
 - the array is degraded
 - the over-write fails.

In these cases we currently eject the device, but we should
record a bad block if possible.
Signed-off-by: Nmajianpeng <majianpeng@gmail.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Having the 'name' arg optional and defaulting to the current
personality name is no necessary and leads to errors, as when
changing the level of an array we can end up using the
name of the old level instead of the new one.

So make it non-optional and always explicitly pass the name
of the level that the array will be.
Reported-by: Nmajianpeng <majianpeng@gmail.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

commit 58c54fcc
     md/raid10: handle further errors during fix_read_error better.

in 3.1 added "r10_sync_page_io" which takes an IO size in sectors.
But we were passing the IO size in bytes!!!
This resulting in bio_add_page failing, and empty request being sent
down, and a consequent BUG_ON in scsi_lib.

[fix missing space in error message at same time]

This fix is suitable for 3.1.y and later.

Cc: stable@vger.kernel.org
Reported-by: NChristian Balzer <chibi@gol.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

commit 43220aa0
    md/raid5: fix a hang on device failure.

fixed a hang, but introduced a refcounting in-balance so
that if the presence of bad-blocks ever caused an rdev to
be 'blocked' we would increment the refcount on the rdev and
never decrement it.

So added the needed rdev_dec_pending when md_wait_for_blocked_rdev
is not called.
Reported-by: Nmajianpeng <majianpeng@gmail.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Add blk_plug in sync_thread will increase the performance of sync.
Because sync_thread did not blk_plug,so when raid sync, the bio merge
not well.

Testing environment:
SATA controller: Intel Corporation 82801JI (ICH10 Family) SATA AHCI
Controller.
OS:Linux xxx 3.5.0-rc2+ #340 SMP Tue Jun 12 09:00:25 CST 2012
x86_64 x86_64 x86_64 GNU/Linux.
RAID5: four ST31000524NS disk.

Without blk_plug:recovery speed about 63M/Sec;
Add blk_plug:recovery speed about 120M/Sec.

Using blktrace:
blktrace -d /dev/sdb -w 60  -o -|blkparse -i -

without blk_plug:
Total (8,16):
 Reads Queued:      309811,     1239MiB	 Writes Queued:           0,        0KiB
 Read Dispatches:   283583,     1189MiB	 Write Dispatches:        0,        0KiB
 Reads Requeued:         0		 Writes Requeued:         0
 Reads Completed:   273351,     1149MiB	 Writes Completed:        0,        0KiB
 Read Merges:        23533,    94132KiB	 Write Merges:            0,        0KiB
 IO unplugs:             0        	 Timer unplugs:           0

add blk_plug:
Total (8,16):
 Reads Queued:      428697,     1714MiB	 Writes Queued:           0,        0KiB
 Read Dispatches:     3954,     1714MiB	 Write Dispatches:        0,        0KiB
 Reads Requeued:         0		 Writes Requeued:         0
 Reads Completed:     3956,     1715MiB	 Writes Completed:        0,        0KiB
 Read Merges:       424743,     1698MiB	 Write Merges:            0,        0KiB
 IO unplugs:             0        	 Timer unplugs:        3384

The ratio of merge will be markedly increased.
Signed-off-by: Nmajianpeng <majianpeng@gmail.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

In ops_run_io(), the call to md_wait_for_blocked_rdev will decrement
nr_pending so we lose the reference we hold on the rdev.
So atomic_inc it first to maintain the reference.

This bug was introduced by commit  73e92e51
    md/raid5.  Don't write to known bad block on doubtful devices.

which appeared in 3.0, so patch is suitable for stable kernels since
then.

Cc: stable@vger.kernel.org
Signed-off-by: Nmajianpeng <majianpeng@gmail.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

In chunk_aligned_read() we are adding data_offset before calling
is_badblock.  But is_badblock also adds data_offset, so that is bad.

So move the addition of data_offset to after the call to
is_badblock.

This bug was introduced by commit 31c176ec
     md/raid5: avoid reading from known bad blocks.
which first appeared in 3.0.  So that patch is suitable for any
-stable kernel from 3.0.y onwards.  However it will need minor
revision for most of those (as the comment didn't appear until
recently).

Cc: stable@vger.kernel.org
Signed-off-by: Nmajianpeng <majianpeng@gmail.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

If a RAID5 has both a failed device and a device marked as
'WantReplacement', then we should preferentially replace the failed
device.
However the current code replaces whichever is found first.
So split into 2 loops, check fail failed/missing first, and only check
for WantReplacement if nothing is failed or missing.
Reported-by: Nmajianpeng <majianpeng@gmail.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

If a RAID10 has an odd number of chunks - as might happen when there
are an odd number of devices - the last chunk has no pair and so is
not mirrored.  We don't store data there, but when recovering the last
device in an array we retry to recover that last chunk from a
non-existent location.  This results in an error, and the recovery
aborts.

When we get to that last chunk we should just stop - there is nothing
more to do anyway.

This bug has been present since the introduction of RAID10, so the
patch is appropriate for any -stable kernel.

Cc: stable@vger.kernel.org
Reported-by: NChristian Balzer <chibi@gol.com>
Tested-by: NChristian Balzer <chibi@gol.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

This patch implements two new messages that can be sent to the thin
pool target allowing it to take a snapshot of the _metadata_.  This,
read-only snapshot can be accessed by userland, concurrently with the
live target.

Only one metadata snapshot can be held at a time.  The pool's status
line will give the block location for the current msnap.

Since version 0.1.5 of the userland thin provisioning tools, the
thin_dump program displays the msnap as follows:

    thin_dump -m <msnap root> <metadata dev>

Available here: https://github.com/jthornber/thin-provisioning-tools

Now that userland can access the metadata we can do various things
that have traditionally been kernel side tasks:

     i) Incremental backups.

     By using metadata snapshots we can work out what blocks have
     changed over time.  Combined with data snapshots we can ensure
     the data doesn't change while we back it up.

     A short proof of concept script can be found here:

     https://github.com/jthornber/thinp-test-suite/blob/master/incremental_backup_example.rb

     ii) Migration of thin devices from one pool to another.

     iii) Merging snapshots back into an external origin.

     iv) Asyncronous replication.
Signed-off-by: NJoe Thornber <ejt@redhat.com>
Signed-off-by: NAlasdair G Kergon <agk@redhat.com>

Use dedicated caches prefixed with a "dm_" name rather than relying on
kmalloc mempools backed by generic slab caches so the memory usage of
thin provisioning (and any leaks) can be accounted for independently.
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NAlasdair G Kergon <agk@redhat.com>

After the failure of a group of paths, any alternative paths that
need initialising do not become available until further I/O is sent to
the device.  Until this has happened, ioctls return -EAGAIN.

With this patch, new paths are made available in response to an ioctl
too.  The processing of the ioctl gets delayed until this has happened.

Instead of returning an error, we submit a work item to kmultipathd
(that will potentially activate the new path) and retry in ten
milliseconds.

Note that the patch doesn't retry an ioctl if the ioctl itself fails due
to a path failure.  Such retries should be handled intelligently by the
code that generated the ioctl in the first place, noting that some SCSI
commands should not be retried because they are not idempotent (XOR write
commands).  For commands that could be retried, there is a danger that
if the device rejected the SCSI command, the path could be errorneously
marked as failed, and the request would be retried on another path which
might fail too.  It can be determined if the failure happens on the
device or on the SCSI controller, but there is no guarantee that all
SCSI drivers set these flags correctly.
Signed-off-by: NMikulas Patocka <mpatocka@redhat.com>
Signed-off-by: NAlasdair G Kergon <agk@redhat.com>

If I/O needs retrying and only bypassed priority groups are available,
set the pg_init_delay_retry flag to wait before retrying.

If, for example, the reason for the bypass is that the controller is
getting reset or there is a firmware upgrade happening, retrying right
away would cause a flood of log messages and retries for what could be a
few seconds or even several minutes.
Signed-off-by: NMike Christie <michaelc@cs.wisc.edu>
Acked-by: NMike Snitzer <snitzer@redhat.com>
Signed-off-by: NAlasdair G Kergon <agk@redhat.com>

Move multipath structure's 'lock' and 'queue_size' members to eliminate
two 4-byte holes.  Also use a bit within a single unsigned int for each
existing flag (saves 8-bytes).  This allows future flags to be added
without each consuming an unsigned int.
Signed-off-by: NMike Snitzer <snitzer@redhat.com>
Acked-by: NHannes Reinecke <hare@suse.de>
Signed-off-by: NAlasdair G Kergon <agk@redhat.com>

The new merge_bvec_fn which calls the corresponding function
in subsidiary devices requires that mddev->merge_check_needed
be set if any child has a merge_bvec_fn.

However were were only setting that when a device was hot-added,
not when a device was present from the start.

This bug was introduced in 3.4 so patch is suitable for 3.4.y
kernels.  However that are conflicts in raid10.c so a separate
patch will be needed for 3.4.y.

Cc: stable@vger.kernel.org
Reported-by: NSebastian Riemer <sebastian.riemer@profitbricks.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Now that bitmaps can grow and shrink it is best if we record
how much space is available.  This means that when
we reduce the size of the bitmap we won't "lose" the space
for late when we might want to increase the size of the bitmap
again.
Signed-off-by: NNeilBrown <neilb@suse.de>

When a reshape which reduced the number of devices finishes
we must remove the extra devices.

So ensure  that raid10_remove_disk won't try to keep them, and
have raid10_finish_reshape clear the 'in_sync' flag.  Then
remove_and_add_spares will be able to remove them.
Reported-by: NHannes Reinecke <hare@suse.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

After a reshape which reduced the number of devices we need
to disconnect the extra devices.
The code for this doesn't currently handle 'replacement' devices.
It is very unlikely that such devices will be present, but it is
safest to handle them anyway.

So simplify the handling.  Just clear In_sync and leave it
to remove_and_add_spaces (which will be called soon) to do
the real works.
Signed-off-by: NNeilBrown <neilb@suse.de>

Check the return of mddev_find(), since it may fail due to out of
memeory or out of usable minor number.

The reason I chose -ENODEV instead of -ENOMEM or something else is
md_alloc() function chose that ;)
Signed-off-by: NYuanhan Liu <yuanhan.liu@linux.intel.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

A RAID1 device does not necessarily need a fullsync if the bitmap can be used instead.

Similar to commit d6b212f4 in raid5.c, if a raid1
device can be brought back (i.e. from a transient failure) it shouldn't need a
complete resync.  Provided the bitmap is not to old, it will have recorded the areas
of the disk that need recovery.
Signed-off-by: NJonathan Brassow <jbrassow@redhat.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

When encountering an error while reading the superblock, call md_error.

We are currently setting the 'Faulty' bit on one of the array devices when an
error is encountered while reading the superblock of a dm-raid array.  We should
be calling md_error(), as it handles the error more completely.
Signed-off-by: NJonathan Brassow <jbrassow@redhat.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Missing dm-raid devices should be recorded in the superblock

When specifying the devices that compose a DM RAID array, it is possible to denote
failed or missing devices with '-'s.  When this occurs, we must record this in the
superblock.  We do this by checking if the array position's data device is missing
and then forcing MD to record the superblock by setting 'MD_CHANGE_DEVS' in
'raid_resume'.  If we do not cause the superblock to be rewritten by the resume
function, it is possible for a stale superblock to be written by an out-going
in-active table (during 'raid_dtr').
Signed-off-by: NJonathan Brassow <jbrassow@redhat.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

Properly initialize MD recovery flags when resuming device-mapper devices.

When a device-mapper device is suspended, all I/O must stop.  This is done by
calling 'md_stop_writes' and 'mddev_suspend'.  These calls in-turn manipulate
the recovery flags - including setting 'MD_RECOVERY_FROZEN'.  The DM device
may have been suspended while recovery was not yet complete, so the process
needs to pick-up where it left off.  Since 'mddev_resume' does not unset
'MD_RECOVERY_FROZEN' and set 'MD_RECOVERY_NEEDED', we must do it ourselves.
'MD_RECOVERY_NEEDED' can safely be set in 'mddev_resume', but 'MD_RECOVERY_FROZEN'
must be set outside of 'mddev_resume' due to how MD handles RAID reshaping.
(e.g.  It is possible for a user to delay reshaping a RAID5->RAID6 by purposefully
setting 'MD_RECOVERY_FROZEN'.  Clearing it in 'mddev_resume' would override the
desired behavior.)

Because 'mddev_resume' already unconditionally calls 'md_wakeup_thread(mddev->thread)'
there is no need to make this call from 'raid_resume' since it calls 'mddev_resume'.

Also clean up where  level_store calls mddev_resume() - it current
duplicates some of the funcitons of that call. - NB
Signed-off-by: NJonathan Brassow <jbrassow@redhat.com>
Signed-off-by: NNeilBrown <neilb@suse.de>

We always should have allowed this.  A raid5 reshape doesn't change
the size of the bitmap, so not need to restrict it.

Also add a test to make sure we don't try to start a reshape on a
failed array.
Signed-off-by: NNeilBrown <neilb@suse.de>

If a reshape changes the size of the array, then we can now
update the bitmap to suit - so do so.
Signed-off-by: NNeilBrown <neilb@suse.de>

Now that bitmaps can be resized, we can allow an array to be resized
while the bitmap is present.

This only covers resizing that involves changing the effective size
of member devices, not resizing that changes the number of devices.
Signed-off-by: NNeilBrown <neilb@suse.de>

As a reshape may change the sync_size and/or chunk_size, we need
to update these whenever we write out the bitmap superblock.
Signed-off-by: NNeilBrown <neilb@suse.de>

This function will allocate the new data structures and copy
bits across from old to new, allowing for the possibility that the
chunksize has changed.

Use the same function for performing the initial allocation
of the structures.  This improves test coverage.

When bitmap_resize is used to resize an existing bitmap, it
only copies '1' bits in, not '0' bits.
So when allocating the bitmap, ensure everything is initialised
to ZERO.
Signed-off-by: NNeilBrown <neilb@suse.de>

Also take the opportunity to simplify CHUNK_BLOCK_RATIO.
Signed-off-by: NNeilBrown <neilb@suse.de>

The new "struct bitmap_counts" contains all the fields that are
related to counting the number of active writes in each bitmap chunk.

Having this separate will make it easier to change the chunksize
or overall size of a bitmap atomically.
Signed-off-by: NNeilBrown <neilb@suse.de>

This allows us to remove spinlock protection which is
more heavy-weight than simple atomics.
Signed-off-by: NNeilBrown <neilb@suse.de>

Using e.g. set_bit instead of __set_bit and using test_and_clear_bit
allow us to remove some locking and contract other locked ranges.

It is rare that we set or clear a lot of these bits, so gain should
outweigh any cost.
Signed-off-by: NNeilBrown <neilb@suse.de>

There functions really do one thing together: release the
'bitmap_storage'.  So make them just one function.

Since we removed the locking (previous patch), we don't need to zero
any fields before freeing them, so it all becomes a bit simpler.
Signed-off-by: NNeilBrown <neilb@suse.de>

There is no real value in freeing things the moment there is an error.
It is just as good to free the bitmap file and pages when the bitmap
is explicitly removed (and replaced?) or at shutdown.

With this gone, the bitmap will only disappear when the array is
quiescent, so we can remove some locking.

As the 'filemap' doesn't disappear now, include extra checks before
trying to write any of it out.
Also remove the check for "has it disappeared" in
bitmap_daemon_write().
Signed-off-by: NNeilBrown <neilb@suse.de>

All of these sites can only be called from process context with
irqs enabled, so using irqsave/irqrestore just adds noise.
Remove it.
Signed-off-by: NNeilBrown <neilb@suse.de>

We currently use '&' and '|' which isn't the norm in the kernel
and doesn't allow easy atomicity.
So change to bit numbers and {set,clear,test}_bit.
This allows us to remove a spinlock/unlock (which was dubious anyway)
and some other simplifications.
Signed-off-by: NNeilBrown <neilb@suse.de>

Just do single-bit manipulations on bitmap->flags and copy whole
value between that and sb->state.

This will allow next patch which changes how bit manipulations are
performed on bitmap->flags.

This does result in BITMAP_STALE not being set in sb by
bitmap_read_sb, however as the setting is determined by other
information in the 'sb' we do not lose information this way.
Normally, bitmap_load will be called shortly which will clear
BITMAP_STALE anyway.
Signed-off-by: NNeilBrown <neilb@suse.de>

This function isn't really needed.  It sets or clears a flag in both
bitmap->flags and sb->state.
However both times it is called, bitmap_update_sb is called soon
afterwards which copies bitmap->flags to sb->state.
So just make changes to bitmap->flags, and open-code those rather than
hiding in a function.
Signed-off-by: NNeilBrown <neilb@suse.de>

We should allocate memory for the storage-bitmap at create-time, not
load time.
Signed-off-by: NNeilBrown <neilb@suse.de>