openwrt/staging/blogic.git
13 years agodm ioctl: fill in device parameters in more ioctls
Mikulas Patocka [Tue, 2 Aug 2011 11:32:06 +0000 (12:32 +0100)]
dm ioctl: fill in device parameters in more ioctls

Move parameter filling from find_device to __find_device_hash_cell.

This patch causes ioctls using __find_device_hash_cell
(DM_DEV_REMOVE_CMD, DM_DEV_SUSPEND_CMD - resume, DM_TABLE_CLEAR_CMD)
to return device parameters, bringing them into line with the other
ioctls.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm flakey: add corrupt_bio_byte feature
Mike Snitzer [Tue, 2 Aug 2011 11:32:06 +0000 (12:32 +0100)]
dm flakey: add corrupt_bio_byte feature

Add corrupt_bio_byte feature to simulate corruption by overwriting a byte at a
specified position with a specified value during intervals when the device is
"down".

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm flakey: add drop_writes
Mike Snitzer [Tue, 2 Aug 2011 11:32:05 +0000 (12:32 +0100)]
dm flakey: add drop_writes

Add 'drop_writes' option to drop writes silently while the
device is 'down'.  Reads are not touched.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm flakey: support feature args
Mike Snitzer [Tue, 2 Aug 2011 11:32:05 +0000 (12:32 +0100)]
dm flakey: support feature args

Add the ability to specify arbitrary feature flags when creating a
flakey target.  This code uses the same target argument helpers that
the multipath target does.

Also remove the superfluous 'dm-flakey' prefixes from the error messages,
as they already contain the prefix 'flakey'.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm flakey: use dm_target_offset and support discards
Mike Snitzer [Tue, 2 Aug 2011 11:32:05 +0000 (12:32 +0100)]
dm flakey: use dm_target_offset and support discards

Use dm_target_offset() and support discards.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm table: share target argument parsing functions
Mike Snitzer [Tue, 2 Aug 2011 11:32:04 +0000 (12:32 +0100)]
dm table: share target argument parsing functions

Move multipath target argument parsing code into dm-table so other
targets can share it.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm snapshot: skip reading origin when overwriting complete chunk
Mikulas Patocka [Tue, 2 Aug 2011 11:32:04 +0000 (12:32 +0100)]
dm snapshot: skip reading origin when overwriting complete chunk

If we write a full chunk in the snapshot, skip reading the origin device
because the whole chunk will be overwritten anyway.

This patch changes the snapshot write logic when a full chunk is written.
In this case:
  1. allocate the exception
  2. dispatch the bio (but don't report the bio completion to device mapper)
  3. write the exception record
  4. report bio completed

Callbacks must be done through the kcopyd thread, because callbacks must not
race with each other.  So we create two new functions:

  dm_kcopyd_prepare_callback: allocate a job structure and prepare the callback.
  (This function must not be called from interrupt context.)

  dm_kcopyd_do_callback: submit callback.
  (This function may be called from interrupt context.)

Performance test (on snapshots with 4k chunk size):
  without the patch:
    non-direct-io sequential write (dd):    17.7MB/s
    direct-io sequential write (dd):        20.9MB/s
    non-direct-io random write (mkfs.ext2): 0.44s

  with the patch:
    non-direct-io sequential write (dd):    26.5MB/s
    direct-io sequential write (dd):        33.2MB/s
    non-direct-io random write (mkfs.ext2): 0.27s

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm: ignore merge_bvec for snapshots when safe
Mikulas Patocka [Tue, 2 Aug 2011 11:32:04 +0000 (12:32 +0100)]
dm: ignore merge_bvec for snapshots when safe

Add a new flag DMF_MERGE_IS_OPTIONAL to struct mapped_device to indicate
whether the device can accept bios larger than the size its merge
function returns.  When set, use this to send large bios to snapshots
which can split them if necessary.  Snapshot I/O may be significantly
fragmented and this approach seems to improve peformance.

Before the patch, dm_set_device_limits restricted bio size to page size
if the underlying device had a merge function and the target didn't
provide a merge function.  After the patch, dm_set_device_limits
restricts bio size to page size if the underlying device has a merge
function, doesn't have DMF_MERGE_IS_OPTIONAL flag and the target doesn't
provide a merge function.

The snapshot target can't provide a merge function because when the merge
function is called, it is impossible to determine where the bio will be
remapped.  Previously this led us to impose a 4k limit, which we can
now remove if the snapshot store is located on a device without a merge
function.  Together with another patch for optimizing full chunk writes,
it improves performance from 29MB/s to 40MB/s when writing to the
filesystem on snapshot store.

If the snapshot store is placed on a non-dm device with a merge function
(such as md-raid), device mapper still limits all bios to page size.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm table: clean dm_get_device and move exports
Mike Snitzer [Tue, 2 Aug 2011 11:32:04 +0000 (12:32 +0100)]
dm table: clean dm_get_device and move exports

There is no need for __table_get_device to be factored out.
Also move the exports to the end of their respective functions.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm raid: tidy includes
Alasdair G Kergon [Tue, 2 Aug 2011 11:32:03 +0000 (12:32 +0100)]
dm raid: tidy includes

A dm target only needs to use include/linux dm headers.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm ioctl: prevent empty message
Alasdair G Kergon [Tue, 2 Aug 2011 11:32:03 +0000 (12:32 +0100)]
dm ioctl: prevent empty message

Detect invalid empty messages in core dm instead of requiring every target to
check this.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm raid: cleanup parameter handling
Jonathan Brassow [Tue, 2 Aug 2011 11:32:03 +0000 (12:32 +0100)]
dm raid: cleanup parameter handling

Re-order the parameters so they are handled consistently in the same order
where defined, parsed and output.

Only include rebuild parameters in the STATUSTYPE_TABLE output if they were
supplied in the original table line.

Correct the parameter count when outputting rebuild: there are two words,
not one.

Use case-independent checks for keywords (as in other device-mapper targets).

Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm snapshot: style cleanups
Jonathan Brassow [Tue, 2 Aug 2011 11:32:03 +0000 (12:32 +0100)]
dm snapshot: style cleanups

Coding style cleanups.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Signed-off-by: Jonathan Brassow <jbrassow@redhat.com>
13 years agodm snapshot: remove unused definitions
Mikulas Patocka [Tue, 2 Aug 2011 11:32:03 +0000 (12:32 +0100)]
dm snapshot: remove unused definitions

Remove a couple of unused #defines.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm kcopyd: remove nr_pages field from job structure
Mikulas Patocka [Tue, 2 Aug 2011 11:32:02 +0000 (12:32 +0100)]
dm kcopyd: remove nr_pages field from job structure

The nr_pages field in struct kcopyd_job is only used temporarily in
run_pages_job() to count the number of required pages.
We can use a local variable instead.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm kcopyd: remove offset field from job structure
Mikulas Patocka [Tue, 2 Aug 2011 11:32:02 +0000 (12:32 +0100)]
dm kcopyd: remove offset field from job structure

The offset field in struct kcopyd_job is always zero so remove it.

Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm: use vzalloc
Joe Perches [Tue, 2 Aug 2011 11:32:02 +0000 (12:32 +0100)]
dm: use vzalloc

Use vzalloc() instead of vmalloc()+memset().

Signed-off-by: Joe Perches <joe@perches.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm log: userspace use list_move
Kirill A. Shutemov [Tue, 2 Aug 2011 11:32:02 +0000 (12:32 +0100)]
dm log: userspace use list_move

Replace list_del() followed by list_add() with list_move().

Signed-off-by: Kirill A. Shutemov <kirill@shutemov.name>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm log: clean up bit little endian bitops
Akinobu Mita [Tue, 2 Aug 2011 11:32:01 +0000 (12:32 +0100)]
dm log: clean up bit little endian bitops

Using __test_and_{set,clear}_bit_le() with ignoring its return value
can be replaced with __{set,clear}_bit_le().

This also removes unnecessary casts.

Signed-off-by: Akinobu Mita <akinobu.mita@gmail.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm table: fix discard support
Mike Snitzer [Tue, 2 Aug 2011 11:32:01 +0000 (12:32 +0100)]
dm table: fix discard support

Remove 'discards_supported' from the dm_table structure.  The same
information can be easily discovered from the table's target(s) in
dm_table_supports_discards().

Before this fix dm_table_supports_discards() would skip checking the
individual targets' 'discards_supported' flag if any one target in the
table didn't set num_discard_requests > 0.  Now the per-target
'discards_supported' flag is effective at insuring the final DM device
advertises discard support.  But, to be clear, targets that don't
support discards (!num_discard_requests) will not receive discard
requests.

Also DMWARN if a target sets 'discards_supported' override but forgets
to set 'num_discard_requests'.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm: suppress endian warnings
Alasdair G Kergon [Tue, 2 Aug 2011 11:32:01 +0000 (12:32 +0100)]
dm: suppress endian warnings

Suppress sparse warnings about cpu_to_le32() by using __le32 types for
on-disk data etc.

Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm: fix idr leak on module removal
Alasdair G Kergon [Tue, 2 Aug 2011 11:32:01 +0000 (12:32 +0100)]
dm: fix idr leak on module removal

Destroy _minor_idr when unloading the core dm module.  (Found by kmemleak.)

Cc: stable@kernel.org
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm io: flush cpu cache with vmapped io
Mikulas Patocka [Tue, 2 Aug 2011 11:32:01 +0000 (12:32 +0100)]
dm io: flush cpu cache with vmapped io

For normal kernel pages, CPU cache is synchronized by the dma layer.
However, this is not done for pages allocated with vmalloc. If we do I/O
to/from vmallocated pages, we must synchronize CPU cache explicitly.

Prior to doing I/O on vmallocated page we must call
flush_kernel_vmap_range to flush dirty cache on the virtual address.
After finished read we must call invalidate_kernel_vmap_range to
invalidate cache on the virtual address, so that accesses to the virtual
address return newly read data and not stale data from CPU cache.

This patch fixes metadata corruption on dm-snapshots on PA-RISC and
possibly other architectures with caches indexed by virtual address.

Cc: stable <stable@kernel.org>
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agodm mpath: fix potential NULL pointer in feature arg processing
Mike Snitzer [Tue, 2 Aug 2011 11:32:00 +0000 (12:32 +0100)]
dm mpath: fix potential NULL pointer in feature arg processing

Avoid dereferencing a NULL pointer if the number of feature arguments
supplied is fewer than indicated.

Signed-off-by: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
Cc: stable@kernel.org
13 years agodm snapshot: flush disk cache when merging
Mikulas Patocka [Tue, 2 Aug 2011 11:32:00 +0000 (12:32 +0100)]
dm snapshot: flush disk cache when merging

This patch makes dm-snapshot flush disk cache when writing metadata for
merging snapshot.

Without cache flushing the disk may reorder metadata write and other
data writes and there is a possibility of data corruption in case of
power fault.

Cc: stable@kernel.org
Signed-off-by: Mikulas Patocka <mpatocka@redhat.com>
Signed-off-by: Alasdair G Kergon <agk@redhat.com>
13 years agoMerge branch 'for-linus' of git://neil.brown.name/md
Linus Torvalds [Thu, 28 Jul 2011 12:50:27 +0000 (05:50 -0700)]
Merge branch 'for-linus' of git://neil.brown.name/md

* 'for-linus' of git://neil.brown.name/md: (75 commits)
  md/raid10: handle further errors during fix_read_error better.
  md/raid10: Handle read errors during recovery better.
  md/raid10: simplify read error handling during recovery.
  md/raid10: record bad blocks due to write errors during resync/recovery.
  md/raid10:  attempt to fix read errors during resync/check
  md/raid10:  Handle write errors by updating badblock log.
  md/raid10: clear bad-block record when write succeeds.
  md/raid10: avoid writing to known bad blocks on known bad drives.
  md/raid10 record bad blocks as needed during recovery.
  md/raid10: avoid reading known bad blocks during resync/recovery.
  md/raid10 - avoid reading from known bad blocks - part 3
  md/raid10: avoid reading from known bad blocks - part 2
  md/raid10: avoid reading from known bad blocks - part 1
  md/raid10: Split handle_read_error out from raid10d.
  md/raid10: simplify/reindent some loops.
  md/raid5: Clear bad blocks on successful write.
  md/raid5.  Don't write to known bad block on doubtful devices.
  md/raid5: write errors should be recorded as bad blocks if possible.
  md/raid5: use bad-block log to improve handling of uncorrectable read errors.
  md/raid5: avoid reading from known bad blocks.
  ...

13 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6
Linus Torvalds [Thu, 28 Jul 2011 12:49:31 +0000 (05:49 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/tiwai/sound-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6:
  sound: oss: rename local change_bits to avoid powerpc bitsops.h definition
  ALSA: hda - Fix duplicated DAC assignments for Realtek
  ALSA: asihpi - off by one in asihpi_hpi_ioctl()
  ALSA: hda - Fix Oops with Realtek quirks with NULL adc_nids
  ALSA: asihpi - bug fix pa use before init.
  ALSA: hda - Add support for vref-out based mute LED control on IDT codecs

13 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris...
Linus Torvalds [Thu, 28 Jul 2011 02:26:38 +0000 (19:26 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/jmorris/security-testing-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/jmorris/security-testing-2.6: (54 commits)
  tpm_nsc: Fix bug when loading multiple TPM drivers
  tpm: Move tpm_tis_reenable_interrupts out of CONFIG_PNP block
  tpm: Fix compilation warning when CONFIG_PNP is not defined
  TOMOYO: Update kernel-doc.
  tpm: Fix a typo
  tpm_tis: Probing function for Intel iTPM bug
  tpm_tis: Fix the probing for interrupts
  tpm_tis: Delay ACPI S3 suspend while the TPM is busy
  tpm_tis: Re-enable interrupts upon (S3) resume
  tpm: Fix display of data in pubek sysfs entry
  tpm_tis: Add timeouts sysfs entry
  tpm: Adjust interface timeouts if they are too small
  tpm: Use interface timeouts returned from the TPM
  tpm_tis: Introduce durations sysfs entry
  tpm: Adjust the durations if they are too small
  tpm: Use durations returned from TPM
  TOMOYO: Enable conditional ACL.
  TOMOYO: Allow using argv[]/envp[] of execve() as conditions.
  TOMOYO: Allow using executable's realpath and symlink's target as conditions.
  TOMOYO: Allow using owner/group etc. of file objects as conditions.
  ...

Fix up trivial conflict in security/tomoyo/realpath.c

13 years agomd/raid10: handle further errors during fix_read_error better.
NeilBrown [Thu, 28 Jul 2011 01:39:25 +0000 (11:39 +1000)]
md/raid10: handle further errors during fix_read_error better.

If we find more read/write errors we should record a bad block before
failing the device.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10: Handle read errors during recovery better.
NeilBrown [Thu, 28 Jul 2011 01:39:25 +0000 (11:39 +1000)]
md/raid10: Handle read errors during recovery better.

Currently when we get a read error during recovery, we simply abort
the recovery.

Instead, repeat the read in page-sized blocks.
On successful reads, write to the target.
On read errors, record a bad block on the destination,
and only if that fails do we abort the recovery.

As we now retry reads we need to know where we read from.  This was in
bi_sector but that can be changed during a read attempt.
So store the correct from_addr and to_addr in the r10_bio for later
access.

Signed-off-by: NeilBrown<neilb@suse.de>
13 years agomd/raid10: simplify read error handling during recovery.
NeilBrown [Thu, 28 Jul 2011 01:39:25 +0000 (11:39 +1000)]
md/raid10: simplify read error handling during recovery.

If a read error is detected during recovery the code currently
fails the read device.
This isn't really necessary.  recovery_request_write will signal
a write error to end_sync_write and it will record a write
error on the destination device which will record a bad block
there or kick it from the array.

So just remove this call to do md_error.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10: record bad blocks due to write errors during resync/recovery.
NeilBrown [Thu, 28 Jul 2011 01:39:25 +0000 (11:39 +1000)]
md/raid10: record bad blocks due to write errors during resync/recovery.

If we get a write error during resync/recovery don't fail the device
but instead record a bad block.  If that fails we can then fail the
device.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10: attempt to fix read errors during resync/check
NeilBrown [Thu, 28 Jul 2011 01:39:25 +0000 (11:39 +1000)]
md/raid10:  attempt to fix read errors during resync/check

We already attempt to fix read errors found during normal IO
and a 'repair' process.
It is best to try to repair them at any time they are found,
so move a test so that during sync and check a read error will
be corrected by over-writing with good data.

If both (all) devices have known bad blocks in the sync section we
won't try to fix even though the bad blocks might not overlap.  That
should be considered later.

Also if we hit a read error during recovery we don't try to fix it.
It would only be possible to fix if there were at least three copies
of data, which is not very common with RAID10.  But it should still
be considered later.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10: Handle write errors by updating badblock log.
NeilBrown [Thu, 28 Jul 2011 01:39:24 +0000 (11:39 +1000)]
md/raid10:  Handle write errors by updating badblock log.

When we get a write error (in the data area, not in metadata),
update the badblock log rather than failing the whole device.

As the write may well be many blocks, we trying writing each
block individually and only log the ones which fail.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10: clear bad-block record when write succeeds.
NeilBrown [Thu, 28 Jul 2011 01:39:24 +0000 (11:39 +1000)]
md/raid10: clear bad-block record when write succeeds.

If we succeed in writing to a block that was recorded as
being bad, we clear the bad-block record.

This requires some delayed handling as the bad-block-list update has
to happen in process-context.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10: avoid writing to known bad blocks on known bad drives.
NeilBrown [Thu, 28 Jul 2011 01:39:24 +0000 (11:39 +1000)]
md/raid10: avoid writing to known bad blocks on known bad drives.

Writing to known bad blocks on drives that have seen a write error
is asking for trouble.  So try to avoid these blocks.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10 record bad blocks as needed during recovery.
NeilBrown [Thu, 28 Jul 2011 01:39:24 +0000 (11:39 +1000)]
md/raid10 record bad blocks as needed during recovery.

When recovering one or more devices, if all the good devices have
bad blocks we should record a bad block on the device being rebuilt.

If this fails, we need to abort the recovery.

To ensure we don't think that we aborted later than we actually did,
we need to move the check for MD_RECOVERY_INTR earlier in md_do_sync,
in particular before mddev->curr_resync is updated.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10: avoid reading known bad blocks during resync/recovery.
NeilBrown [Thu, 28 Jul 2011 01:39:24 +0000 (11:39 +1000)]
md/raid10: avoid reading known bad blocks during resync/recovery.

During resync/recovery limit the size of the request to avoid
reading into a bad block that does not start at-or-before the current
read address.

Similarly if there is a bad block at this address, don't allow the
current request to extend beyond the end of that bad block.

Now that we don't ever read from known bad blocks, it is safe to allow
devices with those blocks into the array.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10 - avoid reading from known bad blocks - part 3
NeilBrown [Thu, 28 Jul 2011 01:39:24 +0000 (11:39 +1000)]
md/raid10 - avoid reading from known bad blocks - part 3

When attempting to repair a read error, don't read from
devices with a known bad block.

As we are only reading PAGE_SIZE blocks, we don't try to
narrow down to smaller regions in the hope that only part of this
page is bad - it isn't worth the effort.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10: avoid reading from known bad blocks - part 2
NeilBrown [Thu, 28 Jul 2011 01:39:23 +0000 (11:39 +1000)]
md/raid10: avoid reading from known bad blocks - part 2

When redirecting a read error to a different device, we must
again avoid bad blocks and possibly split the request.

Spin_lock typo fixed thanks to Dan Carpenter <error27@gmail.com>

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10: avoid reading from known bad blocks - part 1
NeilBrown [Thu, 28 Jul 2011 01:39:23 +0000 (11:39 +1000)]
md/raid10: avoid reading from known bad blocks - part 1

This patch just covers the basic read path:
 1/ read_balance needs to check for badblocks, and return not only
    the chosen slot, but also how many good blocks are available
    there.
 2/ read submission must be ready to issue multiple reads to
    different devices as different bad blocks on different devices
    could mean that a single large read cannot be served by any one
    device, but can still be served by the array.
    This requires keeping count of the number of outstanding requests
    per bio.  This count is stored in 'bi_phys_segments'

On read error we currently just fail the request if another target
cannot handle the whole request.  Next patch refines that a bit.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10: Split handle_read_error out from raid10d.
NeilBrown [Thu, 28 Jul 2011 01:39:23 +0000 (11:39 +1000)]
md/raid10: Split handle_read_error out from raid10d.

raid10d() is too big and is about to get bigger, so split
handle_read_error() out as a separate function.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid10: simplify/reindent some loops.
NeilBrown [Thu, 28 Jul 2011 01:39:23 +0000 (11:39 +1000)]
md/raid10: simplify/reindent some loops.

When a loop ends with a large if, it can be neater to change the
if to invert the condition and just 'continue'.
Then the body of the if can be indented to a lower level.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid5: Clear bad blocks on successful write.
NeilBrown [Thu, 28 Jul 2011 01:39:23 +0000 (11:39 +1000)]
md/raid5: Clear bad blocks on successful write.

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: NeilBrown <neilb@suse.de>
13 years agomd/raid5. Don't write to known bad block on doubtful devices.
NeilBrown [Thu, 28 Jul 2011 01:39:22 +0000 (11:39 +1000)]
md/raid5.  Don't write to known bad block on doubtful devices.

If a device has seen write errors, don't write to any known
bad blocks on that device.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid5: write errors should be recorded as bad blocks if possible.
NeilBrown [Thu, 28 Jul 2011 01:39:22 +0000 (11:39 +1000)]
md/raid5: write errors should be recorded as bad blocks if possible.

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: NeilBrown <neilb@suse.de>
13 years agomd/raid5: use bad-block log to improve handling of uncorrectable read errors.
NeilBrown [Thu, 28 Jul 2011 01:39:22 +0000 (11:39 +1000)]
md/raid5: use bad-block log to improve handling of uncorrectable read errors.

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: NeilBrown <neilb@suse.de>
13 years agomd/raid5: avoid reading from known bad blocks.
NeilBrown [Thu, 28 Jul 2011 01:39:22 +0000 (11:39 +1000)]
md/raid5: avoid reading from known bad blocks.

There are two times that we might read in raid5:
1/ when a read request fits within a chunk on a single
   working device.
   In this case, if there is any bad block in the range of
   the read, we simply fail the cache-bypass read and
   perform the read though the stripe cache.

2/ when reading into the stripe cache.  In this case we
   mark as failed any device which has a bad block in that
   strip (1 page wide).
   Note that we will both avoid reading and avoid writing.
   This is correct (as we will never read from the block, there
   is no point writing), but not optimal (as writing could 'fix'
   the error) - that will be addressed later.

If we have not seen any write errors on the device yet, we treat a bad
block like a recent read error.  This will encourage an attempt to fix
the read error which will either generate a write error, or will
ensure good data is stored there.  We don't yet forget the bad block
in that case.  That comes later.

Now that we honour bad blocks when reading we can allow devices with
bad blocks into the array.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid1: factor several functions out or raid1d()
NeilBrown [Thu, 28 Jul 2011 01:38:13 +0000 (11:38 +1000)]
md/raid1: factor several functions out or raid1d()

raid1d is too big with several deep branches.
So separate them out into their own functions.

Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Namhyung Kim <namhyung@gmail.com>
13 years agomd/raid1: improve handling of read failure during recovery.
NeilBrown [Thu, 28 Jul 2011 01:33:42 +0000 (11:33 +1000)]
md/raid1: improve handling of read failure during recovery.

If we cannot read a block from anywhere during recovery, there is
now a better approach than just giving up.
We can record a bad block on each device and keep going - being
careful not to clear the bad block when a write succeeds as it might -
it will be a write of incorrect data.

We have now reached the state where - for raid1 - we only call
md_error if md_set_badblocks has failed.

Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Namhyung Kim <namhyung@gmail.com>
13 years agomd/raid1: record badblocks found during resync etc.
NeilBrown [Thu, 28 Jul 2011 01:33:00 +0000 (11:33 +1000)]
md/raid1: record badblocks found during resync etc.

If we find a bad block while writing as part of resync/recovery we
need to report that back to raid1d which must record the bad block,
or fail the device.

Similarly when fixing a read error, a further error should just
record a bad block if possible rather than failing the device.

Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Namhyung Kim <namhyung@gmail.com>
13 years agomd/raid1: Handle write errors by updating badblock log.
NeilBrown [Thu, 28 Jul 2011 01:32:41 +0000 (11:32 +1000)]
md/raid1:  Handle write errors by updating badblock log.

When we get a write error (in the data area, not in metadata),
update the badblock log rather than failing the whole device.

As the write may well be many blocks, we trying writing each
block individually and only log the ones which fail.

Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Namhyung Kim <namhyung@gmail.com>
13 years agomd/raid1: store behind-write pages in bi_vecs.
NeilBrown [Thu, 28 Jul 2011 01:32:10 +0000 (11:32 +1000)]
md/raid1: store behind-write pages in bi_vecs.

When performing write-behind we allocate pages to store the data
during write.
Previously we just keep a list of pages.  Now we keep a list of
bi_vec which includes offset and size.
This means that the r1bio has complete information to create a new
bio which will be needed for retrying after write errors.

Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Namhyung Kim <namhyung@gmail.com>
13 years agomd/raid1: clear bad-block record when write succeeds.
NeilBrown [Thu, 28 Jul 2011 01:31:49 +0000 (11:31 +1000)]
md/raid1: clear bad-block record when write succeeds.

If we succeed in writing to a block that was recorded as
being bad, we clear the bad-block record.

This requires some delayed handling as the bad-block-list update has
to happen in process-context.

Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Namhyung Kim <namhyung@gmail.com>
13 years agomd/raid1: avoid writing to known-bad blocks on known-bad drives.
NeilBrown [Thu, 28 Jul 2011 01:31:48 +0000 (11:31 +1000)]
md/raid1: avoid writing to known-bad blocks on known-bad drives.

If we have seen any write error on a drive, then don't write to
any known-bad blocks on that drive.
If necessary, we divide the write request up into pieces just
like we do for reads, so each piece is either all written or
all not written to any given drive.

Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Namhyung Kim <namhyung@gmail.com>
13 years agomd: update documentation for md/rdev/state sysfs interface
Namhyung Kim [Thu, 28 Jul 2011 01:31:48 +0000 (11:31 +1000)]
md: update documentation for md/rdev/state sysfs interface

Previous patches in the bad block series extended behavior of
rdev's 'state' interface but lacked documentation update.
Fix it.

Signed-off-by: Namhyung Kim <namhyung@gmail.com>
Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd: make it easier to wait for bad blocks to be acknowledged.
NeilBrown [Thu, 28 Jul 2011 01:31:48 +0000 (11:31 +1000)]
md: make it easier to wait for bad blocks to be acknowledged.

It is only safe to choose not to write to a bad block if that bad
block is safely recorded in metadata - i.e. if it has been
'acknowledged'.

If it hasn't we need to wait for the acknowledgement.

We support that using rdev->blocked wait and
md_wait_for_blocked_rdev by introducing a new device flag
'BlockedBadBlock'.

This flag is only advisory.
It is cleared whenever we acknowledge a bad block, so that a waiter
can re-check the particular bad blocks that it is interested it.

It should be set by a caller when they find they need to wait.
This (set after test) is inherently racy, but as
md_wait_for_blocked_rdev already has a timeout, losing the race will
have minimal impact.

When we clear "Blocked" was also clear "BlockedBadBlocks" incase it
was set incorrectly (see above race).

We also modify the way we manage 'Blocked' to fit better with the new
handling of 'BlockedBadBlocks' and to make it consistent between
externally managed and internally managed metadata.   This requires
that each raidXd loop checks if the metadata needs to be written and
triggers a write (md_check_recovery) if needed.  Otherwise a queued
write request might cause raidXd to wait for the metadata to write,
and only that thread can write it.

Before writing metadata, we set FaultRecorded for all devices that
are Faulty, then after writing the metadata we clear Blocked for any
device for which the Fault was certainly Recorded.

The 'faulty' device flag now appears in sysfs if the device is faulty
*or* it has unacknowledged bad blocks.  So user-space which does not
understand bad blocks can continue to function correctly.
User space which does, should not assume a device is faulty until it
sees the 'faulty' flag, and then sees the list of unacknowledged bad
blocks is empty.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd: add 'write_error' flag to component devices.
NeilBrown [Thu, 28 Jul 2011 01:31:48 +0000 (11:31 +1000)]
md: add 'write_error' flag to component devices.

If a device has ever seen a write error, we will want to handle
known-bad-blocks differently.
So create an appropriate state flag and export it via sysfs.

Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Namhyung Kim <namhyung@gmail.com>
13 years agomd/raid1: avoid reading known bad blocks during resync
NeilBrown [Thu, 28 Jul 2011 01:31:48 +0000 (11:31 +1000)]
md/raid1: avoid reading known bad blocks during resync

When performing resync/etc, keep the size of the request
small enough that it doesn't overlap any known bad blocks.
Devices with badblocks at the start of the request are completely
excluded.
If there is nowhere to read from due to bad blocks, record
a bad block on each target device.

Now that we never read from known-bad-blocks we can allow devices with
known-bad-blocks into a RAID1.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/raid1: avoid reading from known bad blocks.
NeilBrown [Thu, 28 Jul 2011 01:31:48 +0000 (11:31 +1000)]
md/raid1: avoid reading from known bad blocks.

Now that we have a bad block list, we should not read from those
blocks.
There are several main parts to this:
  1/ read_balance needs to check for bad blocks, and return not only
     the chosen device, but also how many good blocks are available
     there.
  2/ fix_read_error needs to avoid trying to read from bad blocks.
  3/ read submission must be ready to issue multiple reads to
     different devices as different bad blocks on different devices
     could mean that a single large read cannot be served by any one
     device, but can still be served by the array.
     This requires keeping count of the number of outstanding requests
     per bio.  This count is stored in 'bi_phys_segments'
  4/ retrying a read needs to also be ready to submit a smaller read
     and queue another request for the rest.

This does not yet handle bad blocks when reading to perform resync,
recovery, or check.

'md_trim_bio' will also be used for RAID10, so put it in md.c and
export it.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd: Disable bad blocks and v0.90 metadata.
NeilBrown [Thu, 28 Jul 2011 01:31:47 +0000 (11:31 +1000)]
md: Disable bad blocks and v0.90 metadata.

v0.90 metadata cannot record bad blocks, so when loading metadata
for such a device, set shift to -1.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd: load/store badblock list from v1.x metadata
NeilBrown [Thu, 28 Jul 2011 01:31:47 +0000 (11:31 +1000)]
md: load/store badblock list from v1.x metadata

Space must have been allocated when array was created.
A feature flag is set when the badblock list is non-empty, to
ensure old kernels don't load and trust the whole device.

We only update the on-disk badblocklist when it has changed.
If the badblocklist (or other metadata) is stored on a bad block, we
don't cope very well.

If metadata has no room for bad block, flag bad-blocks as disabled,
and do the same for 0.90 metadata.

Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd: don't allow arrays to contain devices with bad blocks.
NeilBrown [Thu, 28 Jul 2011 01:31:47 +0000 (11:31 +1000)]
md: don't allow arrays to contain devices with bad blocks.

As no personality understand bad block lists yet, we must
reject any device that is known to contain bad blocks.
As the personalities get taught, these tests can be removed.

This only applies to raid1/raid5/raid10.
For linear/raid0/multipath/faulty the whole concept of bad blocks
doesn't mean anything so there is no point adding the checks.

Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Namhyung Kim <namhyung@gmail.com>
13 years agomd: add documentation for bad block log
Namhyung Kim [Thu, 28 Jul 2011 01:31:47 +0000 (11:31 +1000)]
md: add documentation for bad block log

Previous patch in the bad block series added new sysfs interfaces
([unacknowledged_]bad_blocks) for each rdev without documentation.
Add it.

Signed-off-by: Namhyung Kim <namhyung@gmail.com>
Signed-off-by: NeilBrown <neilb@suse.de>
13 years agomd/bad-block-log: add sysfs interface for accessing bad-block-log.
NeilBrown [Thu, 28 Jul 2011 01:31:47 +0000 (11:31 +1000)]
md/bad-block-log: add sysfs interface for accessing bad-block-log.

This can show the log (providing it fits in one page) and
allows bad blocks to be 'acknowledged' meaning that they
have safely been recorded in metadata.

Clearing bad blocks is not allowed via sysfs (except for
code testing).  A bad block can only be cleared when
a write to the block succeeds.

Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Namhyung Kim <namhyung@gmail.com>
13 years agomd: beginnings of bad block management.
NeilBrown [Thu, 28 Jul 2011 01:31:46 +0000 (11:31 +1000)]
md: beginnings of bad block management.

This the first step in allowing md to track bad-blocks per-device so
that we can fail individual blocks rather than the whole device.

This patch just adds a data structure for recording bad blocks, with
routines to add, remove, search the list.

Signed-off-by: NeilBrown <neilb@suse.de>
Reviewed-by: Namhyung Kim <namhyung@gmail.com>
13 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs...
Linus Torvalds [Wed, 27 Jul 2011 23:43:52 +0000 (16:43 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/mason/btrfs-unstable

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mason/btrfs-unstable:
  Btrfs: make sure reserve_metadata_bytes doesn't leak out strange errors
  Btrfs: use the commit_root for reading free_space_inode crcs
  Btrfs: reduce extent_state lock contention for metadata
  Btrfs: remove lockdep magic from btrfs_next_leaf
  Btrfs: make a lockdep class for each root
  Btrfs: switch the btrfs tree locks to reader/writer
  Btrfs: fix deadlock when throttling transactions
  Btrfs: stop using highmem for extent_buffers
  Btrfs: fix BUG_ON() caused by ENOSPC when relocating space
  Btrfs: tag pages for writeback in sync
  Btrfs: fix enospc problems with delalloc
  Btrfs: don't flush delalloc arbitrarily
  Btrfs: use find_or_create_page instead of grab_cache_page
  Btrfs: use a worker thread to do caching
  Btrfs: fix how we merge extent states and deal with cached states
  Btrfs: use the normal checksumming infrastructure for free space cache
  Btrfs: serialize flushers in reserve_metadata_bytes
  Btrfs: do transaction space reservation before joining the transaction
  Btrfs: try to only do one btrfs_search_slot in do_setxattr

13 years agomd: remove suspicious size_of()
NeilBrown [Wed, 27 Jul 2011 21:56:24 +0000 (07:56 +1000)]
md: remove suspicious size_of()

When calling bioset_create we pass the size of the front_pad as
   sizeof(mddev)
which looks suspicious as mddev is a pointer and so it looks like a
common mistake where
   sizeof(*mddev)
was intended.
The size is actually correct as we want to store a pointer in the
front padding of the bios created by the bioset, so make the intent
more explicit by using
   sizeof(mddev_t *)

Reported-by: Zdenek Kabelac <zdenek.kabelac@gmail.com>
Signed-off-by: NeilBrown <neilb@suse.de>
13 years agoMerge branch 'for-linus' of git://oss.sgi.com/xfs/xfs
Linus Torvalds [Wed, 27 Jul 2011 20:41:51 +0000 (13:41 -0700)]
Merge branch 'for-linus' of git://oss.sgi.com/xfs/xfs

* 'for-linus' of git://oss.sgi.com/xfs/xfs:
  xfs: optimize the negative xattr caching
  xfs: prevent against ioend livelocks in xfs_file_fsync
  xfs: flag all buffers as metadata
  xfs: encapsulate a block of debug code

13 years agoMerge branch 'nfs-for-3.1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs
Linus Torvalds [Wed, 27 Jul 2011 20:23:02 +0000 (13:23 -0700)]
Merge branch 'nfs-for-3.1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs

* 'nfs-for-3.1' of git://git.linux-nfs.org/projects/trondmy/linux-nfs: (44 commits)
  NFSv4: Don't use the delegation->inode in nfs_mark_return_delegation()
  nfs: don't use d_move in nfs_async_rename_done
  RDMA: Increasing RPCRDMA_MAX_DATA_SEGS
  SUNRPC: Replace xprt->resend and xprt->sending with a priority queue
  SUNRPC: Allow caller of rpc_sleep_on() to select priority levels
  SUNRPC: Support dynamic slot allocation for TCP connections
  SUNRPC: Clean up the slot table allocation
  SUNRPC: Initalise the struct xprt upon allocation
  SUNRPC: Ensure that we grab the XPRT_LOCK before calling xprt_alloc_slot
  pnfs: simplify pnfs files module autoloading
  nfs: document nfsv4 sillyrename issues
  NFS: Convert nfs4_set_ds_client to EXPORT_SYMBOL_GPL
  SUNRPC: Convert the backchannel exports to EXPORT_SYMBOL_GPL
  SUNRPC: sunrpc should not explicitly depend on NFS config options
  NFS: Clean up - simplify the switch to read/write-through-MDS
  NFS: Move the pnfs write code into pnfs.c
  NFS: Move the pnfs read code into pnfs.c
  NFS: Allow the nfs_pageio_descriptor to signal that a re-coalesce is needed
  NFS: Use the nfs_pageio_descriptor->pg_bsize in the read/write request
  NFS: Cache rpc_ops in struct nfs_pageio_descriptor
  ...

13 years agoMerge branch 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target...
Linus Torvalds [Wed, 27 Jul 2011 20:21:40 +0000 (13:21 -0700)]
Merge branch 'for-next' of git://git./linux/kernel/git/nab/target-pending

* 'for-next' of git://git.kernel.org/pub/scm/linux/kernel/git/nab/target-pending:
  target: Convert to DIV_ROUND_UP_SECTOR_T usage for sectors / dev_max_sectors
  kernel.h: Add DIV_ROUND_UP_ULL and DIV_ROUND_UP_SECTOR_T macro usage
  iscsi-target: Add iSCSI fabric support for target v4.1
  iscsi: Add Serial Number Arithmetic LT and GT into iscsi_proto.h
  iscsi: Use struct scsi_lun in iscsi structs instead of u8[8]
  iscsi: Resolve iscsi_proto.h naming conflicts with drivers/target/iscsi

13 years agoMerge branch 'integration' into for-linus
Chris Mason [Wed, 27 Jul 2011 20:13:10 +0000 (16:13 -0400)]
Merge branch 'integration' into for-linus

13 years agoBtrfs: make sure reserve_metadata_bytes doesn't leak out strange errors
Chris Mason [Wed, 27 Jul 2011 19:57:44 +0000 (15:57 -0400)]
Btrfs: make sure reserve_metadata_bytes doesn't leak out strange errors

The btrfs transaction code will return any errors that come from
reserve_metadata_bytes.  We need to make sure we don't return funny
things like 1 or EAGAIN.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
13 years agosignals: sys_ssetmask/sys_rt_sigsuspend should use set_current_blocked()
Oleg Nesterov [Wed, 27 Jul 2011 19:49:44 +0000 (12:49 -0700)]
signals: sys_ssetmask/sys_rt_sigsuspend should use set_current_blocked()

sys_ssetmask(), sys_rt_sigsuspend() and compat_sys_rt_sigsuspend()
change ->blocked directly.  This is not correct, see the changelog in
e6fa16ab "signal: sigprocmask() should do retarget_shared_pending()"

Change them to use set_current_blocked().

Another change is that now we are doing ->saved_sigmask = ->blocked
lockless, it doesn't make any sense to do this under ->siglock.

Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Matt Fleming <matt.fleming@linux.intel.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agosparc: rename atomic_add_unless
Stephen Rothwell [Wed, 27 Jul 2011 19:49:44 +0000 (12:49 -0700)]
sparc: rename atomic_add_unless

Should have been done in commit 1af08a1407f4 ("This is in preparation
for more generic atomic").

Signed-off-by: Stephen Rothwell <sfr@canb.auug.org.au>
Cc: Arun Sharma <asharma@fb.com>
Cc: David Miller <davem@davemloft.net>
Cc: "Hans-Christian Egtvedt" <hans-christian.egtvedt@atmel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoproc: make struct proc_dir_entry::name a terminal array rather than a pointer
David Howells [Wed, 27 Jul 2011 18:47:03 +0000 (21:47 +0300)]
proc: make struct proc_dir_entry::name a terminal array rather than a pointer

Since __proc_create() appends the name it is given to the end of the PDE
structure that it allocates, there isn't a need to store a name pointer.
Instead we can just replace the name pointer with a terminal char array of
_unspecified_ length.  The compiler will simply append the string to statically
defined variables of PDE type overlapping any hole at the end of the structure
and, unlike specifying an explicitly _zero_ length array, won't give a warning
if you try to statically initialise it with a string of more than zero length.

Also, whilst we're at it:

 (1) Move namelen to end just prior to name and reduce it to a single byte
     (name shouldn't be longer than NAME_MAX).

 (2) Move pde_unload_lock two places further on so that if it's four bytes in
     size on a 64-bit machine, it won't cause an unused hole in the PDE struct.

Signed-off-by: David Howells <dhowells@redhat.com>
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agosound: oss: rename local change_bits to avoid powerpc bitsops.h definition
Andy Whitcroft [Wed, 27 Jul 2011 16:48:41 +0000 (17:48 +0100)]
sound: oss: rename local change_bits to avoid powerpc bitsops.h definition

This collides with powerpc exported functions from bitops.h.  Rename the
local copy in the oss soundblaster mixer and ad1848 driver.

Signed-off-by: Andy Whitcroft <apw@canonical.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
13 years agoBtrfs: use the commit_root for reading free_space_inode crcs
Chris Mason [Tue, 26 Jul 2011 19:35:09 +0000 (15:35 -0400)]
Btrfs: use the commit_root for reading free_space_inode crcs

Now that we are using regular file crcs for the free space cache,
we can deadlock if we try to read the free_space_inode while we are
updating the crc tree.

This commit fixes things by using the commit_root to read the crcs.  This is
safe because we the free space cache file would already be loaded if
that block group had been changed in the current transaction.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
13 years agoBtrfs: reduce extent_state lock contention for metadata
Chris Mason [Mon, 25 Jul 2011 10:50:50 +0000 (06:50 -0400)]
Btrfs: reduce extent_state lock contention for metadata

For metadata buffers that don't straddle pages (all of them), btrfs
can safely use the page uptodate bits and extent_buffer uptodate bit
instead of needing to use the extent_state tree.

This greatly reduces contention on the state tree lock.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
13 years agoBtrfs: remove lockdep magic from btrfs_next_leaf
Chris Mason [Tue, 26 Jul 2011 20:01:59 +0000 (16:01 -0400)]
Btrfs: remove lockdep magic from btrfs_next_leaf

Before the reader/writer locks, btrfs_next_leaf needed to keep
the path blocking to avoid making lockdep upset.

Now that btrfs_next_leaf only takes read locks, this isn't required.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
13 years agoBtrfs: make a lockdep class for each root
Chris Mason [Tue, 26 Jul 2011 20:11:19 +0000 (16:11 -0400)]
Btrfs: make a lockdep class for each root

This patch was originally from Tejun Heo.  lockdep complains about the btrfs
locking because we sometimes take btree locks from two different trees at the
same time.  The current classes are based only on level in the btree, which
isn't enough information for lockdep to figure out if the lock is safe.

This patch makes a class for each type of tree, and lumps all the FS trees that
actually have files and directories into the same class.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
13 years agoBtrfs: switch the btrfs tree locks to reader/writer
Chris Mason [Sat, 16 Jul 2011 19:23:14 +0000 (15:23 -0400)]
Btrfs: switch the btrfs tree locks to reader/writer

The btrfs metadata btree is the source of significant
lock contention, especially in the root node.   This
commit changes our locking to use a reader/writer
lock.

The lock is built on top of rw spinlocks, and it
extends the lock tracking to remember if we have a
read lock or a write lock when we go to blocking.  Atomics
count the number of blocking readers or writers at any
given time.

It removes all of the adaptive spinning from the old code
and uses only the spinning/blocking hints inside of btrfs
to decide when it should continue spinning.

In read heavy workloads this is dramatically faster.  In write
heavy workloads we're still faster because of less contention
on the root node lock.

We suffer slightly in dbench because we schedule more often
during write locks, but all other benchmarks so far are improved.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
13 years agoBtrfs: fix deadlock when throttling transactions
Josef Bacik [Sun, 24 Jul 2011 19:45:34 +0000 (15:45 -0400)]
Btrfs: fix deadlock when throttling transactions

Hit this nice little deadlock.  What happens is this

__btrfs_end_transaction with throttle set, --use_count so it equals 0
  btrfs_commit_transaction
    <somebody else actually manages to start the commit>
    btrfs_end_transaction --use_count so now its -1 <== BAD
      we just return and wait on the transaction

This is bad because we just return after our use_count is -1 and don't let go
of our num_writer count on the transaction, so the guy committing the
transaction just sits there forever.  Fix this by inc'ing our use_count if we're
going to call commit_transaction so that if we call btrfs_end_transaction it's
valid.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
13 years agoBtrfs: stop using highmem for extent_buffers
Chris Mason [Tue, 19 Jul 2011 16:04:14 +0000 (12:04 -0400)]
Btrfs: stop using highmem for extent_buffers

The extent_buffers have a very complex interface where
we use HIGHMEM for metadata and try to cache a kmap mapping
to access the memory.

The next commit adds reader/writer locks, and concurrent use
of this kmap cache would make it even more complex.

This commit drops the ability to use HIGHMEM with extent buffers,
and rips out all of the related code.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
13 years agoBtrfs: fix BUG_ON() caused by ENOSPC when relocating space
Miao Xie [Fri, 15 Jul 2011 10:34:36 +0000 (10:34 +0000)]
Btrfs: fix BUG_ON() caused by ENOSPC when relocating space

When we balanced the chunks across the devices, BUG_ON() in
__finish_chunk_alloc() was triggered.

------------[ cut here ]------------
kernel BUG at fs/btrfs/volumes.c:2568!
[SNIP]
Call Trace:
 [<ffffffffa049525e>] btrfs_alloc_chunk+0x8e/0xa0 [btrfs]
 [<ffffffffa04546b0>] do_chunk_alloc+0x330/0x3a0 [btrfs]
 [<ffffffffa045c654>] btrfs_reserve_extent+0xb4/0x1f0 [btrfs]
 [<ffffffffa045c86b>] btrfs_alloc_free_block+0xdb/0x350 [btrfs]
 [<ffffffffa048a8d8>] ? read_extent_buffer+0xd8/0x1d0 [btrfs]
 [<ffffffffa04476fd>] __btrfs_cow_block+0x14d/0x5e0 [btrfs]
 [<ffffffffa044660d>] ? read_block_for_search+0x14d/0x4d0 [btrfs]
 [<ffffffffa0447c9b>] btrfs_cow_block+0x10b/0x240 [btrfs]
 [<ffffffffa044dd5e>] btrfs_search_slot+0x49e/0x7a0 [btrfs]
 [<ffffffffa044f07d>] btrfs_insert_empty_items+0x8d/0xf0 [btrfs]
 [<ffffffffa045e973>] insert_with_overflow+0x43/0x110 [btrfs]
 [<ffffffffa045eb0d>] btrfs_insert_dir_item+0xcd/0x1f0 [btrfs]
 [<ffffffffa0489bd0>] ? map_extent_buffer+0xb0/0xc0 [btrfs]
 [<ffffffff812276ad>] ? rb_insert_color+0x9d/0x160
 [<ffffffffa046cc40>] ? inode_tree_add+0xf0/0x150 [btrfs]
 [<ffffffffa0474801>] btrfs_add_link+0xc1/0x1c0 [btrfs]
 [<ffffffff811dacac>] ? security_inode_init_security+0x1c/0x30
 [<ffffffffa04a28aa>] ? btrfs_init_acl+0x4a/0x180 [btrfs]
 [<ffffffffa047492f>] btrfs_add_nondir+0x2f/0x70 [btrfs]
 [<ffffffffa046af16>] ? btrfs_init_inode_security+0x46/0x60 [btrfs]
 [<ffffffffa0474ac0>] btrfs_create+0x150/0x1d0 [btrfs]
 [<ffffffff81159c63>] ? generic_permission+0x23/0xb0
 [<ffffffff8115b415>] vfs_create+0xa5/0xc0
 [<ffffffff8115ce6e>] do_last+0x5fe/0x880
 [<ffffffff8115dc0d>] path_openat+0xcd/0x3d0
 [<ffffffff8115e029>] do_filp_open+0x49/0xa0
 [<ffffffff8116a965>] ? alloc_fd+0x95/0x160
 [<ffffffff8114f0c7>] do_sys_open+0x107/0x1e0
 [<ffffffff810bcc3f>] ? audit_syscall_entry+0x1bf/0x1f0
 [<ffffffff8114f1e0>] sys_open+0x20/0x30
 [<ffffffff81484ec2>] system_call_fastpath+0x16/0x1b
[SNIP]
RIP  [<ffffffffa049444a>] __finish_chunk_alloc+0x20a/0x220 [btrfs]

The reason is:
Task1 Space balance task
do_chunk_alloc()
  __finish_chunk_alloc()
    update device info
    in the chunk tree
      alloc system metadata block
relocate system metadata block group
  set system metadata block group
  readonly, This block group is the
  only one that can allocate space. So
  there is no free space that can be
  allocated now.
        find no space and don't try
        to alloc new chunk, and then
        return ENOSPC
  BUG_ON() in __finish_chunk_alloc()
  was triggered.

Fix this bug by allocating a new system metadata chunk before relocating the
old one if we find there is no free space which can be allocated after setting
the old block group to be read-only.

Reported-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Miao Xie <miaox@cn.fujitsu.com>
Tested-by: Tsutomu Itoh <t-itoh@jp.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
13 years agoBtrfs: tag pages for writeback in sync
Josef Bacik [Fri, 15 Jul 2011 21:26:38 +0000 (21:26 +0000)]
Btrfs: tag pages for writeback in sync

Everybody else does this, we need to do it too.  If we're syncing, we need to
tag the pages we're going to write for writeback so we don't end up writing the
same stuff over and over again if somebody is constantly redirtying our file.
This will keep us from having latencies with heavy sync workloads.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
13 years agoBtrfs: fix enospc problems with delalloc
Josef Bacik [Fri, 15 Jul 2011 15:16:44 +0000 (15:16 +0000)]
Btrfs: fix enospc problems with delalloc

So I had this brilliant idea to use atomic counters for outstanding and reserved
extents, but this turned out to be a bad idea.  Consider this where we have 1
outstanding extent and 1 reserved extent

Reserver Releaser
atomic_dec(outstanding) now 0
atomic_read(outstanding)+1 get 1
atomic_read(reserved) get 1
don't actually reserve anything because
they are the same
atomic_cmpxchg(reserved, 1, 0)
atomic_inc(outstanding)
atomic_add(0, reserved)
free reserved space for 1 extent

Then the reserver now has no actual space reserved for it, and when it goes to
finish the ordered IO it won't have enough space to do it's allocation and you
get those lovely warnings.

Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
13 years agoBtrfs: don't flush delalloc arbitrarily
Josef Bacik [Fri, 15 Jul 2011 16:01:03 +0000 (16:01 +0000)]
Btrfs: don't flush delalloc arbitrarily

Kill the check to see if we have 512mb of reserved space in delalloc and
shrink_delalloc if we do.  This causes unexpected latencies and we have other
logic to see if we need to throttle.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
13 years agoBtrfs: use find_or_create_page instead of grab_cache_page
Josef Bacik [Mon, 11 Jul 2011 14:47:06 +0000 (10:47 -0400)]
Btrfs: use find_or_create_page instead of grab_cache_page

grab_cache_page will use mapping_gfp_mask(), which for all inodes is set to
GFP_HIGHUSER_MOVABLE.  So instead use find_or_create_page in all cases where we
need GFP_NOFS so we don't deadlock.  Thanks,

Signed-off-by: Josef Bacik <josef@redhat.com>
13 years agoBtrfs: use a worker thread to do caching
Josef Bacik [Thu, 30 Jun 2011 18:42:28 +0000 (14:42 -0400)]
Btrfs: use a worker thread to do caching

A user reported a deadlock when copying a bunch of files.  This is because they
were low on memory and kthreadd got hung up trying to migrate pages for an
allocation when starting the caching kthread.  The page was locked by the person
starting the caching kthread.  To fix this we just need to use the async thread
stuff so that the threads are already created and we don't have to worry about
deadlocks.  Thanks,

Reported-by: Roman Mamedov <rm@romanrm.ru>
Signed-off-by: Josef Bacik <josef@redhat.com>
13 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/shaggy...
Linus Torvalds [Wed, 27 Jul 2011 16:26:39 +0000 (09:26 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/shaggy/jfs-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/shaggy/jfs-2.6:
  jfs: clean up some compiler warnings

13 years agoMerge git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-fixes
Linus Torvalds [Wed, 27 Jul 2011 16:26:22 +0000 (09:26 -0700)]
Merge git://git./linux/kernel/git/steve/gfs2-2.6-fixes

* git://git.kernel.org/pub/scm/linux/kernel/git/steve/gfs2-2.6-fixes:
  GFS2: Fix mount hang caused by certain access pattern to sysfs files

13 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6
Linus Torvalds [Wed, 27 Jul 2011 16:25:15 +0000 (09:25 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/tiwai/sound-2.6

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/tiwai/sound-2.6: (22 commits)
  ALSA: hda - Cirrus Logic CS421x support
  ALSA: Make pcm.h self-contained
  ALSA: hda - Allow codec-specific set_power_state ops
  ALSA: hda - Add post_suspend patch ops
  ALSA: hda - Make CONFIG_SND_HDA_POWER_SAVE depending on CONFIG_PM
  ALSA: hda - Make sure mute led reflects master mute state
  ALSA: hda - Fix invalid mute led state on resume of IDT codecs
  ASoC: Revert "ASoC: SAMSUNG: Add I2S0 internal dma driver"
  ALSA: hda - Add support of the 4 internal speakers on certain HP laptops
  ALSA: Make snd_pcm_debug_name usable outside pcm_lib
  ALSA: hda - Fix DAC filling for multi-connection pins in Realtek parser
  ASoC: dapm - Add methods to retrieve snd_card and soc_card from dapm context.
  ASoC: SAMSUNG: Add I2S0 internal dma driver
  ASoC: SAMSUNG: Modify I2S driver to support idma
  ASoC: davinci: add missing break statement
  ASoC: davinci: fix codec start and stop functions
  ASoC: dapm - add DAPM macro for external enum widgets
  ASoC: Acknowledge WM8962 interrupts before acting on them
  ASoC: sgtl5000: guide user when regulator support is needed
  ASoC: sgtl5000: refactor registering internal ldo
  ...

13 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
Linus Torvalds [Wed, 27 Jul 2011 16:24:56 +0000 (09:24 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/dtor/input

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input: (53 commits)
  Input: synaptics - fix reporting of min coordinates
  Input: tegra-kbc - enable key autorepeat
  Input: kxtj9 - fix locking typo in kxtj9_set_poll()
  Input: kxtj9 - fix bug in probe()
  Input: intel-mid-touch - remove pointless checking for variable 'found'
  Input: hp_sdc - staticize hp_sdc_kicker()
  Input: pmic8xxx-keypad - fix a leak of the IRQ during init failure
  Input: cy8ctmg110_ts - set reset_pin and irq_pin from platform data
  Input: cy8ctmg110_ts - constify i2c_device_id table
  Input: cy8ctmg110_ts - fix checking return value of i2c_master_send
  Input: lifebook - make dmi callback functions return 1
  Input: atkbd - make dmi callback functions return 1
  Input: gpio_keys - switch to using SIMPLE_DEV_PM_OPS
  Input: gpio_keys - add support for device-tree platform data
  Input: aiptek - remove double define
  Input: synaptics - set minimum coordinates as reported by firmware
  Input: synaptics - process button bits in AGM packets
  Input: synaptics - rename set_slot to be more descriptive
  Input: synaptics - fuzz position for touchpad with reduced filtering
  Input: synaptics - set resolution for MT_POSITION_X/Y axes
  ...

13 years agoMerge branch 'next' of git://git.monstr.eu/linux-2.6-microblaze
Linus Torvalds [Wed, 27 Jul 2011 16:24:20 +0000 (09:24 -0700)]
Merge branch 'next' of git://git.monstr.eu/linux-2.6-microblaze

* 'next' of git://git.monstr.eu/linux-2.6-microblaze:
  microblaze: Do not show error message for 32 interrupt lines
  Revert "microblaze: PCI fix typo fault in of_node pointer moving into pci_bus"
  microblaze: PCI fix typo fault in of_node pointer moving into pci_bus
  microblaze: Add support for early console on mdm
  microblaze: Simplify early console binding from DT
  microblaze: Get early printk console earlier
  microblaze: Standardise cpuinfo output for cache policy
  microblaze: Unprivileged stream instruction awareness
  microblaze: trivial: Fix typo fault
  microblaze: exec: Remove redundant set_fs(USER_DS)
  microblaze: Remove duplicated prototype of start_thread()
  microblaze: Fix unaligned value saving to the stack for system with MMU
  microblaze/irqs: Do not trace arch_local_{*,irq_*} functions

13 years agostaging: brcm80211: Fix double include introduced by bad merge
Daniel Morsing [Wed, 27 Jul 2011 11:53:28 +0000 (13:53 +0200)]
staging: brcm80211: Fix double include introduced by bad merge

A merge with Linus' tree added a double include of linux/interrupt.h.
Fix by removing one of the includes.

Signed-off-by: Daniel Morsing <daniel.morsing@gmail.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
13 years agoALSA: hda - Fix duplicated DAC assignments for Realtek
Takashi Iwai [Wed, 27 Jul 2011 14:41:57 +0000 (16:41 +0200)]
ALSA: hda - Fix duplicated DAC assignments for Realtek

Copying hp_pins and speaker_pins from line_out_pins may confuse the
parser, and it can lead to duplicated initializations for the same pin
with a wrong DAC assignment.  The problem appears in 3.0 kernel code.

Cc: <stable@kernel.org> (for 3.0)
Signed-off-by: Takashi Iwai <tiwai@suse.de>
13 years agoALSA: asihpi - off by one in asihpi_hpi_ioctl()
Dan Carpenter [Wed, 27 Jul 2011 12:02:26 +0000 (15:02 +0300)]
ALSA: asihpi - off by one in asihpi_hpi_ioctl()

"adapter" is used as an array index in the adapters[] array so
the off by one would make us read past the end.

1c073b67979 "ALSA: asihpi - Remove spurious adapter index check"
reverted Dan Rosenberg's check that would have prevented the
overflow here.

Signed-off-by: Dan Carpenter <error27@gmail.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
13 years agoALSA: hda - Fix Oops with Realtek quirks with NULL adc_nids
Takashi Iwai [Wed, 27 Jul 2011 12:01:24 +0000 (14:01 +0200)]
ALSA: hda - Fix Oops with Realtek quirks with NULL adc_nids

Somce quirk models don't set adc_nids but let the parser filling it.
But the recent code has unnecessary NULL-checks of spec->input_mux,
and it resulted in NULL dereferences.
This patch fixes that regression.

Reported-and-tested-by: Oliver Neukum <oneukum@suse.de>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
13 years agomicroblaze: Do not show error message for 32 interrupt lines
Michal Simek [Wed, 27 Jul 2011 08:45:32 +0000 (10:45 +0200)]
microblaze: Do not show error message for 32 interrupt lines

When interrupt controller uses 32 interrupts lines the kernel
show error message about mismatch in kind-of-intr parameter
because it exceeds u32. Recast fixs this issue.

Signed-off-by: Michal Simek <monstr@monstr.eu>