pandora-kernel.git
15 years agoBtrfs: fix RHEL test for ClearPageFsMisc
Eric Sandeen [Thu, 7 Aug 2008 15:19:42 +0000 (11:19 -0400)]
Btrfs: fix RHEL test for ClearPageFsMisc

Newer RHEL5 kernels define both ClearPageFSMisc and
ClearPageChecked, so test for both before redefining.

Signed-off-by: Eric Sandeen <sandeen@redhat.com>
---

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Update version.sh to v0.16
Chris Mason [Mon, 4 Aug 2008 19:45:30 +0000 (15:45 -0400)]
Btrfs: Update version.sh to v0.16

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Avoid calling into the FS for the final iput on fake root inodes
Chris Mason [Tue, 5 Aug 2008 17:30:48 +0000 (13:30 -0400)]
Btrfs: Avoid calling into the FS for the final iput on fake root inodes

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix nodatacow for the new data=ordered mode
Yan Zheng [Tue, 5 Aug 2008 17:05:02 +0000 (13:05 -0400)]
Btrfs: Fix nodatacow for the new data=ordered mode

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoGet rid of BTRFS_I(inode)->index and use local vars instead
Chris Mason [Tue, 5 Aug 2008 15:18:09 +0000 (11:18 -0400)]
Get rid of BTRFS_I(inode)->index and use local vars instead

rename and link don't always have a lock on the source inode, and
our use of a per-inode index variable was racy.  This changes things to
store the index in a local variable instead.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix the multi-bio code to save the original bio for completion
Chris Mason [Tue, 5 Aug 2008 14:13:57 +0000 (10:13 -0400)]
Btrfs: Fix the multi-bio code to save the original bio for completion

The multi-bio code is responsible for duplicating blocks in raid1 and
single spindle duplication.  It has counters to make sure all of
the locations for a given extent are properly written before io completion
is returned to the higher layers.

But, it didn't always complete the same bio it was given, sometimes a
clone was completed instead.  This lead to problems with the async
work queues because they saved a pointer to the bio in a struct off
bi_private.

The fix is to remember the original bio and only complete that one.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Update clone file ioctl
Yan Zheng [Tue, 5 Aug 2008 03:23:47 +0000 (23:23 -0400)]
Btrfs: Update clone file ioctl

This patch updates the file clone ioctl for the tree locking and new
data ordered code.

---

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Various small fixes.
Yan Zheng [Tue, 5 Aug 2008 03:23:47 +0000 (23:23 -0400)]
Btrfs: Various small fixes.

This trivial patch contains two locking fixes and a off by one fix.

---

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agobtrfs_lookup_bio_sums seems broken, go back to the readpage_io_hook for now
Chris Mason [Tue, 5 Aug 2008 03:17:27 +0000 (23:17 -0400)]
btrfs_lookup_bio_sums seems broken, go back to the readpage_io_hook for now

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Maintain a list of inodes that are delalloc and a way to wait on them
Chris Mason [Tue, 5 Aug 2008 03:17:27 +0000 (23:17 -0400)]
Btrfs: Maintain a list of inodes that are delalloc and a way to wait on them

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Don't corrupt ram in shrink_extent_tree, leak it instead
Chris Mason [Tue, 5 Aug 2008 03:17:26 +0000 (23:17 -0400)]
Btrfs: Don't corrupt ram in shrink_extent_tree, leak it instead

Far from the perfect fix, but these structs are small.  TODO for the
next release.  The block group cache structs are referenced in many
different places, and it isn't safe to just free them while resizing.

A real fix will be a larger change to the allocator so that it doesn't
have to carry about the block group cache structs to find good places
to search for free blocks.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: fix ioctl-initiated transactions vs wait_current_trans()
Sage Weil [Mon, 4 Aug 2008 14:41:27 +0000 (10:41 -0400)]
Btrfs: fix ioctl-initiated transactions vs wait_current_trans()

Commit 597:466b27332893 (btrfs_start_transaction: wait for commits in
progress) breaks the transaction start/stop ioctls by making
btrfs_start_transaction conditionally wait for the next transaction to
start.  If an application artificially is holding a transaction open,
things deadlock.

This workaround maintains a count of open ioctl-initiated transactions in
fs_info, and avoids wait_current_trans() if any are currently open (in
start_transaction() and btrfs_throttle()).  The start transaction ioctl
uses a new btrfs_start_ioctl_transaction() that _does_ call
wait_current_trans(), effectively pushing the join/wait decision to the
outer ioctl-initiated transaction.

This more or less neuters btrfs_throttle() when ioctl-initiated
transactions are in use, but that seems like a pretty fundamental
consequence of wrapping lots of write()'s in a transaction.  Btrfs has no
way to tell if the application considers a given operation as part of it's
transaction.

Obviously, if the transaction start/stop ioctls aren't being used, there
is no effect on current behavior.

Signed-off-by: Sage Weil <sage@newdream.net>
---
 ctree.h       |    1 +
 ioctl.c       |   12 +++++++++++-
 transaction.c |   18 +++++++++++++-----
 transaction.h |    2 ++
 4 files changed, 27 insertions(+), 6 deletions(-)

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add support for HW assisted crc32c
Chris Mason [Mon, 4 Aug 2008 15:10:20 +0000 (11:10 -0400)]
Btrfs: Add support for HW assisted crc32c

Intel doesn't yet ship hardware to the public with this enabled, but when they
do, they will be ready.  Original code from:

Austin Zhang <austin_zhang@linux.intel.com>

It is currently disabled, but edit crc32c.h to turn it on.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Hold csum mutex while reading in sums during readpages
Chris Mason [Mon, 4 Aug 2008 12:35:53 +0000 (08:35 -0400)]
Btrfs: Hold csum mutex while reading in sums during readpages

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: More throttle tuning
Chris Mason [Mon, 4 Aug 2008 12:20:15 +0000 (08:20 -0400)]
Btrfs: More throttle tuning

* Make walk_down_tree wake up throttled tasks more often
* Make walk_down_tree call cond_resched during long loops
* As the size of the ref cache grows, wait longer in throttle
* Get rid of the reada code in walk_down_tree, the leaves don't get
  read anymore, thanks to the ref cache.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agobtrfs_search_slot: reduce lock contention by cowing in two stages
Chris Mason [Fri, 1 Aug 2008 19:11:20 +0000 (15:11 -0400)]
btrfs_search_slot: reduce lock contention by cowing in two stages

A btree block cow has two parts, the first is to allocate a destination
block and the second is to copy the old bock over.

The first part needs locks in the extent allocation tree, and may need to
do IO.  This changeset splits that into a separate function that can be
called without any tree locks held.

btrfs_search_slot is changed to drop its path and start over if it has
to COW a contended block.  This often means that many writers will
pre-alloc a new destination for a the same contended block, but they
cache their prealloc for later use on lower levels in the tree.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Throttle less often waiting for snapshots to delete
Chris Mason [Fri, 1 Aug 2008 17:11:41 +0000 (13:11 -0400)]
Btrfs: Throttle less often waiting for snapshots to delete

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Improve and cleanup locking done by walk_down_tree
Chris Mason [Fri, 1 Aug 2008 15:27:23 +0000 (11:27 -0400)]
Btrfs: Improve and cleanup locking done by walk_down_tree

While dropping snapshots, walk_down_tree does most of the work of checking
reference counts and limiting tree traversal to just the blocks that
we are freeing.

It dropped and held the allocation mutex in strange and confusing ways,
this commit changes it to only hold the mutex while actually freeing a block.

The rest of the checks around reference counts should be safe without the lock
because we only allow one process in btrfs_drop_snapshot at a time.  Other
processes dropping reference counts should not drop it to 1 because
their tree roots already have an extra ref on the block.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Hold a reference on bios during submit_bio, add some extra bio checks
Chris Mason [Thu, 31 Jul 2008 20:29:02 +0000 (16:29 -0400)]
Btrfs: Hold a reference on bios during submit_bio, add some extra bio checks

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Drop some debugging around the extent_map pinned flag
Chris Mason [Thu, 31 Jul 2008 19:42:54 +0000 (15:42 -0400)]
Btrfs: Drop some debugging around the extent_map pinned flag

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix streaming read performance with checksumming on
Chris Mason [Thu, 31 Jul 2008 19:42:53 +0000 (15:42 -0400)]
Btrfs: Fix streaming read performance with checksumming on

Large streaming reads make for large bios, which means each entry on the
list async work queues represents a large amount of data.  IO
congestion throttling on the device was kicking in before the async
worker threads decided a single thread was busy and needed some help.

The end result was that a streaming read would result in a single CPU
running at 100% instead of balancing the work off to other CPUs.

This patch also changes the pre-IO checksum lookup done by reads to
work on a per-bio basis instead of a per-page.  This results in many
extra btree lookups on large streaming reads.  Doing the checksum lookup
right before bio submit allows us to reuse searches while processing
adjacent offsets.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Throttle tuning
Chris Mason [Thu, 31 Jul 2008 14:48:37 +0000 (10:48 -0400)]
Btrfs: Throttle tuning

This avoids waiting for transactions with pages locked by breaking out
the code to wait for the current transaction to close into a function
called by btrfs_throttle.

It also lowers the limits for where we start throttling.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add missing hunk from Yan Zheng's cache reclaim patch
Chris Mason [Thu, 31 Jul 2008 13:46:18 +0000 (09:46 -0400)]
Btrfs: Add missing hunk from Yan Zheng's cache reclaim patch

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add compatibility for kernels >= 2.6.27-rc1
Sven Wegener [Wed, 30 Jul 2008 20:54:26 +0000 (16:54 -0400)]
Btrfs: Add compatibility for kernels >= 2.6.27-rc1

Add a couple of #if's to follow API changes.

Signed-off-by: Sven Wegener <sven.wegener@stealer.net>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: implement memory reclaim for leaf reference cache
Yan [Wed, 30 Jul 2008 20:29:20 +0000 (16:29 -0400)]
Btrfs: implement memory reclaim for leaf reference cache

The memory reclaiming issue happens when snapshot exists. In that
case, some cache entries may not be used during old snapshot dropping,
so they will remain in the cache until umount.

The patch adds a field to struct btrfs_leaf_ref to record create time. Besides,
the patch makes all dead roots of a given snapshot linked together in order of
create time. After a old snapshot was completely dropped, we check the dead
root list and remove all cache entries created before the oldest dead root in
the list.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix verify_parent_transid
Chris Mason [Wed, 30 Jul 2008 14:29:12 +0000 (10:29 -0400)]
Btrfs: Fix verify_parent_transid

It was incorrectly clearing the up to date flag on the buffer even
when the buffer properly verified.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Update and fix mount -o nodatacow
Yan Zheng [Wed, 30 Jul 2008 13:26:11 +0000 (09:26 -0400)]
Btrfs: Update and fix mount -o nodatacow

To check whether a given file extent is referenced by multiple snapshots, the
checker walks down the fs tree through dead root and checks all tree blocks in
the path.

We can easily detect whether a given tree block is directly referenced by other
snapshot. We can also detect any indirect reference from other snapshot by
checking reference's generation. The checker can always detect multiple
references, but can't reliably detect cases of single reference. So btrfs may
do file data cow even there is only one reference.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: async-thread: fix possible memory leak
Li Zefan [Wed, 30 Jul 2008 13:24:37 +0000 (09:24 -0400)]
Btrfs: async-thread: fix possible memory leak

When kthread_run() returns failure, this worker hasn't been
added to the list, so btrfs_stop_workers() won't free it.

Signed-off-by: Li Zefan <lizf@cn.fujitsu.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Throttle operations if the reference cache gets too large
Chris Mason [Tue, 29 Jul 2008 20:15:18 +0000 (16:15 -0400)]
Btrfs: Throttle operations if the reference cache gets too large

A large reference cache is directly related to a lot of work pending
for the cleaner thread.  This throttles back new operations based on
the size of the reference cache so the cleaner thread will be able to keep
up.

Overall, this actually makes the FS faster because the cleaner thread will
be more likely to find things in cache.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix version.sh when used outside of an hg repo
Chris Mason [Tue, 29 Jul 2008 20:11:35 +0000 (16:11 -0400)]
Btrfs: Fix version.sh when used outside of an hg repo

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Leaf reference cache update
Chris Mason [Mon, 28 Jul 2008 19:32:51 +0000 (15:32 -0400)]
Btrfs: Leaf reference cache update

This changes the reference cache to make a single cache per root
instead of one cache per transaction, and to key by the byte number
of the disk block instead of the keys inside.

This makes it much less likely to have cache misses if a snapshot
or something has an extra reference on a higher node or a leaf while
the first transaction that added the leaf into the cache is dropping.

Some throttling is added to functions that free blocks heavily so they
wait for old transactions to drop.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add a leaf reference cache
Yan Zheng [Mon, 28 Jul 2008 19:32:19 +0000 (15:32 -0400)]
Btrfs: Add a leaf reference cache

Much of the IO done while dropping snapshots is done looking up
leaves in the filesystem trees to see if they point to any extents and
to drop the references on any extents found.

This creates a cache so that IO isn't required.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Rev the disk format magic
Chris Mason [Thu, 24 Jul 2008 16:25:50 +0000 (12:25 -0400)]
Btrfs: Rev the disk format magic

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Null terminate strings passed in from userspace
Mark Fasheh [Thu, 24 Jul 2008 16:20:14 +0000 (12:20 -0400)]
Btrfs: Null terminate strings passed in from userspace

The 'char name[BTRFS_PATH_NAME_MAX]' member of struct btrfs_ioctl_vol_args
is passed directly to strlen() after being copied from user. I haven't
verified this, but in theory a userspace program could pass in an
unterminated string and cause a kernel crash as strlen walks off the end of
the array.

This patch terminates the ->name string in all btrfs ioctl functions which
currently use a 'struct btrfs_ioctl_vol_args'. Since the string is now
properly terminated, it's length will never be longer than
BTRFS_PATH_NAME_MAX so that error check has been removed.

By the way, it might be better overall to just have the ioctl pass an
unterminated string + length structure but I didn't bother with that since
it'd change the kernel/user interface.

Signed-off-by: Mark Fasheh <mfasheh@suse.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoFix path slots selection in btrfs_search_forward
Yan [Thu, 24 Jul 2008 16:19:49 +0000 (12:19 -0400)]
Fix path slots selection in btrfs_search_forward

We should decrease the found slot by one as btrfs_search_slot does
when bin_search return 1 and node level > 0.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix .. lookup corner case
Yan [Thu, 24 Jul 2008 16:19:32 +0000 (12:19 -0400)]
Btrfs: Fix .. lookup corner case

Inode ref item can be in the next leaf when we find "path->slots[0] ==
btrfs_header_nritems(...)".

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Properly release lock in pin_down_bytes
Yan [Thu, 24 Jul 2008 16:18:16 +0000 (12:18 -0400)]
Btrfs: Properly release lock in pin_down_bytes

When buffer isn't uptodate, pin_down_bytes may leave the tree locked
after it returns.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Remove unused variable in fixup_tree_root_location
Balaji Rao [Thu, 24 Jul 2008 16:17:15 +0000 (12:17 -0400)]
Btrfs: Remove unused variable in fixup_tree_root_location

Remove a unused variable 'path' in fixup_tree_root_location.

Signed-off-by: Balaji Rao <balajirrao@gmail.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix a few functions that exit without stopping their transaction
Josef Bacik [Thu, 24 Jul 2008 16:17:14 +0000 (12:17 -0400)]
Btrfs: Fix a few functions that exit without stopping their transaction

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Create orphan inode records to prevent lost files after a crash
Josef Bacik [Thu, 24 Jul 2008 16:17:14 +0000 (12:17 -0400)]
Btrfs: Create orphan inode records to prevent lost files after a crash

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add ACL support
Josef Bacik [Thu, 24 Jul 2008 16:16:36 +0000 (12:16 -0400)]
Btrfs: Add ACL support

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Remove unused xattr code
Josef Bacik [Thu, 24 Jul 2008 16:16:03 +0000 (12:16 -0400)]
Btrfs: Remove unused xattr code

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Implement new dir index format
Josef Bacik [Thu, 24 Jul 2008 16:12:38 +0000 (12:12 -0400)]
Btrfs: Implement new dir index format

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix the defragmention code and the block relocation code for data=ordered
Chris Mason [Thu, 24 Jul 2008 15:57:52 +0000 (11:57 -0400)]
Btrfs: Fix the defragmention code and the block relocation code for data=ordered

Before setting an extent to delalloc, the code needs to wait for
pending ordered extents.

Also, the relocation code needs to wait for ordered IO before scanning
the block group again.  This is because the extents are not removed
until the IO for the new extents is finished

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Use assert_spin_locked instead of spin_trylock
David Woodhouse [Thu, 24 Jul 2008 14:09:43 +0000 (10:09 -0400)]
Btrfs: Use assert_spin_locked instead of spin_trylock

On UP systems spin_trylock always succeeds

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add version strings on module load
Chris Mason [Wed, 23 Jul 2008 16:12:13 +0000 (12:12 -0400)]
Btrfs: Add version strings on module load

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix some build problems on 2.6.18 based enterprise kernels
Chris Mason [Thu, 24 Jul 2008 13:51:08 +0000 (09:51 -0400)]
Btrfs: Fix some build problems on 2.6.18 based enterprise kernels

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Search data ordered extents first for checksums on read
Chris Mason [Thu, 24 Jul 2008 13:41:53 +0000 (09:41 -0400)]
Btrfs: Search data ordered extents first for checksums on read

Checksum items are not inserted into the tree until all of the io from a
given extent is complete.  This means one dirty page from an extent may
be written, freed, and then read again before the entire extent is on disk
and the checksum item is inserted.

The checksums themselves are stored in the ordered extent so they can
be inserted in bulk when IO is complete.  On read, if a checksum item isn't
found, the ordered extents were being searched for a checksum record.

This all worked most of the time, but the checksum insertion code tries
to reduce the number of tree operations by pre-inserting checksum items
based on i_size and a few other factors.  This means the read code might
find a checksum item that hasn't yet really been filled in.

This commit changes things to check the ordered extents first and only
dive into the btree if nothing was found.  This removes the need for
extra locking and is more reliable.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix 32 bit compiles by using an unsigned long byte count in the ordered extent
Chris Mason [Wed, 23 Jul 2008 13:26:26 +0000 (09:26 -0400)]
Btrfs: Fix 32 bit compiles by using an unsigned long byte count in the ordered extent

The ordered extents have to fit in memory, so an unsigned long is sufficient.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Take the csum mutex while reading checksums
Chris Mason [Wed, 23 Jul 2008 03:06:42 +0000 (23:06 -0400)]
Btrfs: Take the csum mutex while reading checksums

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: alloc_mutex latency reduction
Chris Mason [Wed, 23 Jul 2008 03:06:41 +0000 (23:06 -0400)]
Btrfs: alloc_mutex latency reduction

This releases the alloc_mutex in a few places that hold it for over long
operations.  btrfs_lookup_block_group is changed so that it doesn't need
the mutex at all.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add some conditional schedules near the alloc_mutex
Chris Mason [Tue, 22 Jul 2008 16:08:37 +0000 (12:08 -0400)]
Btrfs: Add some conditional schedules near the alloc_mutex

This helps prevent stalls, especially while the snapshot cleaner is
running hard

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Use mutex_lock_nested for tree locking
Chris Mason [Tue, 22 Jul 2008 15:18:09 +0000 (11:18 -0400)]
Btrfs: Use mutex_lock_nested for tree locking

Lockdep has the notion of locking subclasses so that you can identify
locks you expect to be taken after other locks of the same class.  This
changes the per-extent buffer btree locking routines to use a subclass based
on the level in the tree.

Unfortunately, lockdep can only handle 8 total subclasses, and the btrfs
max level is also 8.  So when lockdep is on, use a lower max level.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix some data=ordered related data corruptions
Chris Mason [Tue, 22 Jul 2008 15:18:09 +0000 (11:18 -0400)]
Btrfs: Fix some data=ordered related data corruptions

Stress testing was showing data checksum errors, most of which were caused
by a lookup bug in the extent_map tree.  The tree was caching the last
pointer returned, and searches would check the last pointer first.

But, search callers also expect the search to return the very first
matching extent in the range, which wasn't always true with the last
pointer usage.

For now, the code to cache the last return value is just removed.  It is
easy to fix, but I think lookups are rare enough that it isn't required anymore.

This commit also replaces do_sync_mapping_range with a local copy of the
related functions.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Use a mutex in the extent buffer for tree block locking
Chris Mason [Tue, 22 Jul 2008 15:18:08 +0000 (11:18 -0400)]
Btrfs: Use a mutex in the extent buffer for tree block locking

This replaces the use of the page cache lock bit for locking, which wasn't
suitable for block size < page size and couldn't be used recursively.

The mutexes alone don't fix either problem, but they are the first step.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Index extent buffers in an rbtree
Chris Mason [Tue, 22 Jul 2008 15:18:07 +0000 (11:18 -0400)]
Btrfs: Index extent buffers in an rbtree

Before, extent buffers were a temporary object, meant to map a number of pages
at once and collect operations on them.

But, a few extra fields have crept in, and they are also the best place to
store a per-tree block lock field as well.  This commit puts the extent
buffers into an rbtree, and ensures a single extent buffer for each
tree block.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Data ordered fixes
Chris Mason [Mon, 21 Jul 2008 14:29:44 +0000 (10:29 -0400)]
Btrfs: Data ordered fixes

* In btrfs_delete_inode, wait for ordered extents after calling
truncate_inode_pages.  This is much faster, and more correct

* Properly clear our the PageChecked bit everywhere we redirty the page.

* Change the writepage fixup handler to lock the page range and check to
see if an ordered extent had been inserted since the improperly dirtied
page was discovered

* Wait for ordered extents outside the transaction.  This isn't required
for locking rules but does improve transaction latencies

* Reduce contention on the alloc_mutex by dropping it while incrementing
refs on a node/leaf and while dropping refs on a leaf.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoFix btrfs_wait_ordered_extent_range to properly wait
Chris Mason [Sat, 19 Jul 2008 00:42:20 +0000 (20:42 -0400)]
Fix btrfs_wait_ordered_extent_range to properly wait

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Keep extent mappings in ram until pending ordered extents are done
Chris Mason [Fri, 18 Jul 2008 16:01:11 +0000 (12:01 -0400)]
Btrfs: Keep extent mappings in ram until pending ordered extents are done

It was possible for stale mappings from disk to be used instead of the
new pending ordered extent.  This adds a flag to the extent map struct
to keep it pinned until the pending ordered extent is actually on disk.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Don't allow releasepage to succeed if EXTENT_ORDERED is set
Chris Mason [Fri, 18 Jul 2008 15:56:15 +0000 (11:56 -0400)]
Btrfs: Don't allow releasepage to succeed if EXTENT_ORDERED is set

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Handle data checksumming on bios that span multiple ordered extents
Chris Mason [Fri, 18 Jul 2008 10:17:13 +0000 (06:17 -0400)]
Btrfs: Handle data checksumming on bios that span multiple ordered extents

Data checksumming is done right before the bio is sent down the IO stack,
which means a single bio might span more than one ordered extent.  In
this case, the checksumming data is split between two ordered extents.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Cleanup and comment ordered-data.c
Chris Mason [Thu, 17 Jul 2008 17:53:27 +0000 (13:53 -0400)]
Btrfs: Cleanup and comment ordered-data.c

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Force caching of metadata block groups on mount to avoid deadlock
Chris Mason [Thu, 17 Jul 2008 16:54:48 +0000 (12:54 -0400)]
Btrfs: Force caching of metadata block groups on mount to avoid deadlock

This is a temporary change to avoid deadlocks until the extent tree locking
is fixed up.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agobtrfs_next_leaf: do readahead when skip_locking is turned on
Chris Mason [Thu, 17 Jul 2008 16:54:43 +0000 (12:54 -0400)]
btrfs_next_leaf: do readahead when skip_locking is turned on

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoAdd a per-inode lock around btrfs_drop_extents
Chris Mason [Thu, 17 Jul 2008 16:54:40 +0000 (12:54 -0400)]
Add a per-inode lock around btrfs_drop_extents

btrfs_drop_extents is always called with a range lock held on the inode.
But, it may operate on extents outside that range as it drops and splits
them.

This patch adds a per-inode mutex that is held while calling
btrfs_drop_extents and while inserting new extents into the tree.  It
prevents races from two procs working against adjacent ranges in the tree.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Don't pin pages in ram until the entire ordered extent is on disk.
Chris Mason [Thu, 17 Jul 2008 16:54:15 +0000 (12:54 -0400)]
Btrfs: Don't pin pages in ram until the entire ordered extent is on disk.

Checksum items are not inserted until the entire ordered extent is on disk,
but individual pages might be clean and available for reclaim long before
the whole extent is on disk.

In order to allow those pages to be freed, we need to be able to search
the list of ordered extents to find the checksum that is going to be inserted
in the tree.  This way if the page needs to be read back in before
the checksums are in the btree, we'll be able to verify the checksum on
the page.

This commit adds the ability to search the pending ordered extents for
a given offset in the file, and changes btrfs_releasepage to allow
ordered pages to be freed.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agobtrfs_start_transaction: wait for commits in progress to finish
Chris Mason [Thu, 17 Jul 2008 16:54:14 +0000 (12:54 -0400)]
btrfs_start_transaction: wait for commits in progress to finish

btrfs_commit_transaction has to loop waiting for any writers in the
transaction to finish before it can proceed.  btrfs_start_transaction
should be polite and not join a transaction that is in the process
of being finished off.

There are a few places that can't wait, basically the ones doing IO that
might be needed to finish the transaction.  For them, btrfs_join_transaction
is added.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Update on disk i_size only after pending ordered extents are done
Chris Mason [Thu, 17 Jul 2008 16:54:05 +0000 (12:54 -0400)]
Btrfs: Update on disk i_size only after pending ordered extents are done

This changes the ordered data code to update i_size after the extent
is on disk.  An on disk i_size is maintained in the in-memory btrfs inode
structures, and this is updated as extents finish.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Use async helpers to deal with pages that have been improperly dirtied
Chris Mason [Thu, 17 Jul 2008 16:53:51 +0000 (12:53 -0400)]
Btrfs: Use async helpers to deal with pages that have been improperly dirtied

Higher layers sometimes call set_page_dirty without asking the filesystem
to help.  This causes many problems for the data=ordered and cow code.
This commit detects pages that haven't been properly setup for IO and
kicks off an async helper to deal with them.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: New data=ordered implementation
Chris Mason [Thu, 17 Jul 2008 16:53:50 +0000 (12:53 -0400)]
Btrfs: New data=ordered implementation

The old data=ordered code would force commit to wait until
all the data extents from the transaction were fully on disk.  This
introduced large latencies into the commit and stalled new writers
in the transaction for a long time.

The new code changes the way data allocations and extents work:

* When delayed allocation is filled, data extents are reserved, and
  the extent bit EXTENT_ORDERED is set on the entire range of the extent.
  A struct btrfs_ordered_extent is allocated an inserted into a per-inode
  rbtree to track the pending extents.

* As each page is written EXTENT_ORDERED is cleared on the bytes corresponding
  to that page.

* When all of the bytes corresponding to a single struct btrfs_ordered_extent
  are written, The previously reserved extent is inserted into the FS
  btree and into the extent allocation trees.  The checksums for the file
  data are also updated.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Drop some verbose printks
Chris Mason [Tue, 8 Jul 2008 18:32:12 +0000 (14:32 -0400)]
Btrfs: Drop some verbose printks

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add locking around volume management (device add/remove/balance)
Chris Mason [Tue, 8 Jul 2008 18:19:17 +0000 (14:19 -0400)]
Btrfs: Add locking around volume management (device add/remove/balance)

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix deadlock while searching for dead roots on mount
Chris Mason [Thu, 26 Jun 2008 14:34:20 +0000 (10:34 -0400)]
Btrfs: Fix deadlock while searching for dead roots on mount

btrfs_find_dead_roots called btrfs_read_fs_root_no_radix, which
means we end up calling btrfs_search_slot with a path already held.

The fix is to remember the key inside btrfs_find_dead_roots and drop
the path.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Reduce contention on the root node
Chris Mason [Wed, 25 Jun 2008 20:14:04 +0000 (16:14 -0400)]
Btrfs: Reduce contention on the root node

This calls unlock_up sooner in btrfs_search_slot in order to decrease the
amount of work done with the higher level tree locks held.

Also, it changes btrfs_tree_lock to spin for a big against the page lock
before scheduling.  This makes a big difference in context switch rate under
highly contended workloads.

Longer term, a better locking structure is needed than the page lock.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Online btree defragmentation fixes
Chris Mason [Wed, 25 Jun 2008 20:01:31 +0000 (16:01 -0400)]
Btrfs: Online btree defragmentation fixes

The btree defragger wasn't making forward progress because the new key wasn't
being saved by the btrfs_search_forward function.

This also disables the automatic btree defrag, it wasn't scaling well to
huge filesystems.  The auto-defrag needs to be done differently.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add a per-inode csum mutex to avoid races creating csum items
Chris Mason [Wed, 25 Jun 2008 20:01:31 +0000 (16:01 -0400)]
Btrfs: Add a per-inode csum mutex to avoid races creating csum items

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Change find_extent_buffer to use TestSetPageLocked
Chris Mason [Wed, 25 Jun 2008 20:01:31 +0000 (16:01 -0400)]
Btrfs: Change find_extent_buffer to use TestSetPageLocked

This makes it possible for callers to check for extent_buffers in cache
without deadlocking against any btree locks held.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add btree locking to the tree defragmentation code
Chris Mason [Wed, 25 Jun 2008 20:01:31 +0000 (16:01 -0400)]
Btrfs: Add btree locking to the tree defragmentation code

The online btree defragger is simplified and rewritten to use
standard btree searches instead of a walk up / down mechanism.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Replace the transaction work queue with kthreads
Chris Mason [Wed, 25 Jun 2008 20:01:31 +0000 (16:01 -0400)]
Btrfs: Replace the transaction work queue with kthreads

This creates one kthread for commits and one kthread for
deleting old snapshots.  All the work queues are removed.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoAdd btrfs_end_transaction_throttle to force writers to wait for pending commits
Chris Mason [Wed, 25 Jun 2008 20:01:31 +0000 (16:01 -0400)]
Add btrfs_end_transaction_throttle to force writers to wait for pending commits

The existing throttle mechanism was often not sufficient to prevent
new writers from coming in and making a given transaction run forever.
This adds an explicit wait at the end of most operations so they will
allow the current transaction to close.

There is no wait inside file_write, inode updates, or cow filling, all which
have different deadlock possibilities.

This is a temporary measure until better asynchronous commit support is
added.  This code leads to stalls as it waits for data=ordered
writeback, and it really needs to be fixed.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix snapshot deletion to release the alloc_mutex much more often.
Chris Mason [Wed, 25 Jun 2008 20:01:30 +0000 (16:01 -0400)]
Btrfs: Fix snapshot deletion to release the alloc_mutex much more often.

This lowers the impact of snapshot deletion on the rest of the FS.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add a skip_locking parameter to struct path, and make various funcs honor it
Chris Mason [Wed, 25 Jun 2008 20:01:30 +0000 (16:01 -0400)]
Btrfs: Add a skip_locking parameter to struct path, and make various funcs honor it

Allocations may need to read in block groups from the extent allocation tree,
which will require a tree search and take locks on the extent allocation
tree.  But, those locks might already be held in other places, leading
to deadlocks.

Since the alloc_mutex serializes everything right now, it is safe to
skip the btree locking while caching block groups.  A better fix will be
to either create a recursive lock or find a way to back off existing
locks while caching block groups.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoFix btrfs_next_leaf to check for new items after dropping locks
Chris Mason [Wed, 25 Jun 2008 20:01:30 +0000 (16:01 -0400)]
Fix btrfs_next_leaf to check for new items after dropping locks

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoFix btrfs_del_ordered_inode to allow forcing the drop during unlinks
Chris Mason [Wed, 25 Jun 2008 20:01:30 +0000 (16:01 -0400)]
Fix btrfs_del_ordered_inode to allow forcing the drop during unlinks

This allows us to delete an unlinked inode with dirty pages from the list
instead of forcing commit to write these out before deleting the inode.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoDrop locks in btrfs_search_slot when reading a tree block.
Chris Mason [Wed, 25 Jun 2008 20:01:30 +0000 (16:01 -0400)]
Drop locks in btrfs_search_slot when reading a tree block.

One lock per btree block can make for significant congestion if everyone
has to wait for IO at the high levels of the btree.  This drops
locks held by a path when doing reads during a tree search.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Replace the big fs_mutex with a collection of other locks
Chris Mason [Wed, 25 Jun 2008 20:01:30 +0000 (16:01 -0400)]
Btrfs: Replace the big fs_mutex with a collection of other locks

Extent alloctions are still protected by a large alloc_mutex.
Objectid allocations are covered by a objectid mutex
Other btree operations are protected by a lock on individual btree nodes

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Start btree concurrency work.
Chris Mason [Wed, 25 Jun 2008 20:01:30 +0000 (16:01 -0400)]
Btrfs: Start btree concurrency work.

The allocation trees and the chunk trees are serialized via their own
dedicated mutexes.  This means allocation location is still not very
fine grained.

The main FS btree is protected by locks on each block in the btree.  Locks
are taken top / down, and as processing finishes on a given level of the
tree, the lock is released after locking the lower level.

The end result of a search is now a path where only the lowest level
is locked.  Releasing or freeing the path drops any locks held.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add a thread pool just for submit_bio
Chris Mason [Thu, 12 Jun 2008 18:46:17 +0000 (14:46 -0400)]
Btrfs: Add a thread pool just for submit_bio

If a bio submission is after a lock holder waiting for the bio
on the work queue, it is possible to deadlock.  Move the bios
into their own pool.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBTRFS_IOC_TRANS_START should be privilegued
Christoph Hellwig [Thu, 12 Jun 2008 01:53:58 +0000 (21:53 -0400)]
BTRFS_IOC_TRANS_START should be privilegued

As mentioned in the comment next to it btrfs_ioctl_trans_start can
do bad damage to filesystems and thus should be limited to privilegued
users.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: split out ioctl.c
Christoph Hellwig [Thu, 12 Jun 2008 01:53:53 +0000 (21:53 -0400)]
Btrfs: split out ioctl.c

Split the ioctl handling out of inode.c into a file of it's own.
Also fix up checkpatch.pl warnings for the moved code.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: kerneldoc comments for extent_map.c
Christoph Hellwig [Thu, 12 Jun 2008 01:52:17 +0000 (21:52 -0400)]
Btrfs: kerneldoc comments for extent_map.c

Add kerneldoc comments for all exported functions.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add a mount option to control worker thread pool size
Chris Mason [Thu, 12 Jun 2008 01:47:56 +0000 (21:47 -0400)]
Btrfs: Add a mount option to control worker thread pool size

mount -o thread_pool_size changes the default, which is
min(num_cpus + 2, 8).  Larger thread pools would make more sense on
very large disk arrays.

This mount option controls the max size of each thread pool.  There
are multiple thread pools, so the total worker count will be larger
than the mount option.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Worker thread optimizations
Chris Mason [Thu, 12 Jun 2008 00:21:24 +0000 (20:21 -0400)]
Btrfs: Worker thread optimizations

This changes the worker thread pool to maintain a list of idle threads,
avoiding a complex search for a good thread to wake up.

Threads have two states:

idle - we try to reuse the last thread used in hopes of improving the batching
ratios

busy - each time a new work item is added to a busy task, the task is
rotated to the end of the line.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add backport for the kthread work on kernels older than 2.6.20
Chris Mason [Wed, 11 Jun 2008 21:09:48 +0000 (17:09 -0400)]
Btrfs: Add backport for the kthread work on kernels older than 2.6.20

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Fix mount -o max_inline=0
Chris Mason [Wed, 11 Jun 2008 20:51:38 +0000 (16:51 -0400)]
Btrfs: Fix mount -o max_inline=0

max_inline=0 used to force the max_inline size to one sector instead.  Now
it properly disables inline data items, while still being able to read
any that happen to exist on disk.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agoBtrfs: Add async worker threads for pre and post IO checksumming
Chris Mason [Wed, 11 Jun 2008 20:50:36 +0000 (16:50 -0400)]
Btrfs: Add async worker threads for pre and post IO checksumming

Btrfs has been using workqueues to spread the checksumming load across
other CPUs in the system.  But, workqueues only schedule work on the
same CPU that queued the work, giving them a limited benefit for systems with
higher CPU counts.

This code adds a generic facility to schedule work with pools of kthreads,
and changes the bio submission code to queue bios up.  The queueing is
important to make sure large numbers of procs on the system don't
turn streaming workloads into random workloads by sending IO down
concurrently.

The end result of all of this is much higher performance (and CPU usage) when
doing checksumming on large machines.  Two worker pools are created,
one for writes and one for endio processing.  The two could deadlock if
we tried to service both from a single pool.

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agobtrfs: allow scanning multiple devices during mount
Christoph Hellwig [Tue, 10 Jun 2008 14:40:46 +0000 (10:40 -0400)]
btrfs: allow scanning multiple devices during mount

Allows to specify one or multiple device=/dev/foo options during mount
so that ioctls on the control device can be avoided.  Especially useful
when trying to mount a multi-device setup as root.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agobtrfs: sanity mount option parsing and early mount code
Christoph Hellwig [Tue, 10 Jun 2008 14:40:29 +0000 (10:40 -0400)]
btrfs: sanity mount option parsing and early mount code

Also adds lots of comments to describe what's going on here.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agobtrfs: fix strange indentation in lookup_extent_mapping
Christoph Hellwig [Tue, 10 Jun 2008 14:21:04 +0000 (10:21 -0400)]
btrfs: fix strange indentation in lookup_extent_mapping

Signed-off-by: Chris Mason <chris.mason@oracle.com>
15 years agobtrfs: tiny makefile cleanup
Christoph Hellwig [Tue, 10 Jun 2008 14:20:57 +0000 (10:20 -0400)]
btrfs: tiny makefile cleanup

use normal kbuild syntax to build acl.o conditinally and remove comment
out lines.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Chris Mason <chris.mason@oracle.com>