This patch moves the caching of the block group off to a kthread in order to
allow people to allocate sooner.  Instead of blocking up behind the caching
mutex, we instead kick of the caching kthread, and then attempt to make an
allocation.  If we cannot, we wait on the block groups caching waitqueue, which
the caching kthread will wake the waiting threads up everytime it finds 2 meg
worth of space, and then again when its finished caching.  This is how I tested
the speedup from this

mkfs the disk
mount the disk
fill the disk up with fs_mark
unmount the disk
mount the disk
time touch /mnt/foo

Without my changes this took 11 seconds on my box, with these changes it now
takes 1 second.

Another change thats been put in place is we lock the super mirror's in the
pinned extent map in order to keep us from adding that stuff as free space when
caching the block group.  This doesn't really change anything else as far as the
pinned extent map is concerned, since for actual pinned extents we use
EXTENT_DIRTY, but it does mean that when we unmount we have to go in and unlock
those extents to keep from leaking memory.

I've also added a check where when we are reading block groups from disk, if the
amount of space used == the size of the block group, we go ahead and mark the
block group as cached.  This drastically reduces the amount of time it takes to
cache the block groups.  Using the same test as above, except doing a dd to a
file and then unmounting, it used to take 33 seconds to umount, now it takes 3
seconds.

This version uses the commit_root in the caching kthread, and then keeps track
of how many async caching threads are running at any given time so if one of the
async threads is still running as we cross transactions we can wait until its
finished before handling the pinned extents.  Thank you,
Signed-off-by: Josef Bacik <jbacik@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

Currently btrfs has a problem where it can use a ridiculous amount of RAM simply
tracking free space.  As free space gets fragmented, we end up with thousands of
entries on an rb-tree per block group, which usually spans 1 gig of area.  Since
we currently don't ever flush free space cache back to disk this gets to be a
bit unweildly on large fs's with lots of fragmentation.

This patch solves this problem by using PAGE_SIZE bitmaps for parts of the free
space cache.  Initially we calculate a threshold of extent entries we can
handle, which is however many extent entries we can cram into 16k of ram.  The
maximum amount of RAM that should ever be used to track 1 gigabyte of diskspace
will be 32k of RAM, which scales much better than we did before.

Once we pass the extent threshold, we start adding bitmaps and using those
instead for tracking the free space.  This patch also makes it so that any free
space thats less than 4 * sectorsize we go ahead and put into a bitmap.  This is
nice since we try and allocate out of the front of a block group, so if the
front of a block group is heavily fragmented and then has a huge chunk of free
space at the end, we go ahead and add the fragmented areas to bitmaps and use a
normal extent entry to track the big chunk at the back of the block group.

I've also taken the opportunity to revamp how we search for free space.
Previously we indexed free space via an offset indexed rb tree and a bytes
indexed rb tree.  I've dropped the bytes indexed rb tree and use only the offset
indexed rb tree.  This cuts the number of tree operations we were doing
previously down by half, and gives us a little bit of a better allocation
pattern since we will always start from a specific offset and search forward
from there, instead of searching for the size we need and try and get it as
close as possible to the offset we want.

I've given this a healthy amount of testing pre-new format stuff, as well as
post-new format stuff.  I've booted up my fedora box which is installed on btrfs
with this patch and ran with it for a few days without issues.  I've not seen
any performance regressions in any of my tests.

Since the last patch Yan Zheng fixed a problem where we could have overlapping
entries, so updating their offset inline would cause problems.  Thanks,
Signed-off-by: Josef Bacik <jbacik@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

If the tree roots hit read errors during mount, btrfs is not properly
erroring out.  We need to check the uptodate bits after
reading in the tree root node.
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

This removes the continues call's of btrfs_header_level. One call of
btrfs_header_level(c) its enough.

Signed-off-by Daniel Cadete <danielncadete10@gmail.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

Move the call to BUG_ON to before the dereference of the tested value.
Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

It was never actually doing anything anyway (see the loop condition),
and it would be difficult to make it work for RAID[56].

Even if it was actually working, it's checking for the wrong thing
anyway. Instead of checking whether we list a block which _doesn't_ land
at the relevant physical location, it should be checking that we _have_
listed all the logical blocks which refer to the required physical
location on all devices.

This function is only called from remove_sb_from_cache() to ensure that
we reserve the logical blocks which would reside at the same physical
location as the superblock copies. So listing more blocks than we need
is actually OK.

With RAID[56] we're going to throw away an entire stripe for each block
we have to ignore, so we _are_ going to list blocks other than the
ones which actually contain the superblock.
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

If spin_lock_irqsave is called twice in a row with the same second
argument, the interrupt state at the point of the second call overwrites
the value saved by the first call.  Indeed, the second call does not need
to save the interrupt state, so it is changed to a simple spin_lock.
Signed-off-by: Julia Lawall <julia@diku.dk>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

Write dirty block groups may allocate new block, and so may add new delayed
back ref. btrfs_run_delayed_refs may make some block groups dirty.

commit_cowonly_roots does not handle the recursion properly, and some dirty
blocks can be left unwritten at commit time. This patch moves
btrfs_run_delayed_refs into the loop that writes dirty block groups, and makes
the code not break out of the loop until there are no dirty block groups or
delayed back refs.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

When walking up the tree, btrfs_find_next_key assumes the upper level tree
block is properly locked. This isn't always true even path->keep_locks is 1.
This is because btrfs_find_next_key may advance path->slots[] several times
instead of only once.

When 'path->slots[level] >= btrfs_header_nritems(path->nodes[level])' is found,
we can't guarantee the original value of 'path->slots[level]' is
'btrfs_header_nritems(path->nodes[level]) - 1'. If it's not, the tree block at
'level + 1' isn't locked.

This patch fixes the issue by explicitly checking the locking state,
re-searching the tree if it's not locked.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

if 1 is returned by btrfs_search_slot, the path already points to the
first item with 'key > searching key'. So increasing path->slots[0] by
one is superfluous in that case.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

Change 'goto done' to 'break' for the case of all device extents have
been freed, so that the code updates space information will be execute.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

use __le64 instead of u64 in on-disk structure definition.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

Make an error msg look nicer by inserting a space between number and word.
Signed-off-by: Hu Tao <hu.taoo@gmail.com>
Signed-off-by: Jiri Kosina <jkosina@suse.cz>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

worker memory is already freed on one fail path in btrfs_start_workers,
but is still dereferenced. Switch the dereference and kfree.
Signed-off-by: Jiri Slaby <jirislaby@gmail.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

The btrfs attr patches unconditionally inherited the inode flags field
without honoring nodatacow and nodatasum.  This fix makes sure
we properly record the nodatacow/sum mount options in new inodes.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

The new backref format has restriction on type of backref item.  If a tree
block isn't referenced by its owner tree, full backrefs must be used for the
pointers in it. When a tree block loses its owner tree's reference, backrefs
for the pointers in it should be updated to full backrefs. Current
btrfs_drop_snapshot misses the code that updates backrefs, so it's unsafe for
general use.

This patch adds backrefs update code to btrfs_drop_snapshot.  It isn't a
problem in the restricted form btrfs_drop_snapshot is used today, but for
general snapshot deletion this update is required.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

Using Eric Sandeen's xfstest for fallocate, you can easily trigger a ENOSPC
panic on btrfs.  This is because we do not account for data we may use when
doing the fallocate.  This patch fixes the problem by properly reserving space,
and then just freeing it when we are done.  The reservation stuff was made with
delalloc in mind, so its a little crude for this case, but it keeps the box
from panicing.
Signed-off-by: Josef Bacik <jbacik@redhat.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

commit_fs_roots skips updating root items for fs trees that aren't modified.
This is unsafe now that relocation code modifies root item's last_snapshot
field without modifying corresponding fs tree.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

During tree log replay, we read in the tree log roots,
process them and then free them.  A recent change
takes an extra reference on the root node of the tree
when the root is read in, and stores that reference
in root->commit_root.

This reference was not being freed, leaving us with
one buffer pinned in ram for each subvol with
a tree log root after a crash.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

This happens during subvol creation.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

It was printing nodatacsum, which was not the correct option name.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

lookup_inline_extent_backref only checks for duplicate backref for data
extents. It assumes backrefs for tree block never conflict.

This patch makes lookup_inline_extent_backref check for duplicate backrefs
for both data and tree block, so that we can detect potential bug earlier.
This is a safety check, strictly speaking it is not required.
Signed-off-by: Yan Zheng <zheng.yan@oracle.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

This patch fixes a bug which may result race condition
between btrfs_start_workers() and worker_loop().

btrfs_start_workers() executed in a parent thread writes
on workers->worker and worker_loop() in a child thread
reads workers->worker. However, there is no synchronization
enforcing the order of two operations.

This patch makes btrfs_start_workers() fill workers->worker
before it starts a child thread with worker_loop()
Signed-off-by: Chris Mason <chris.mason@oracle.com>

write_dev_supers is called in sequence.  First is it called with wait == 0,
which starts IO on all of the super blocks for a given device.  Then it is
called with wait == 1 to make sure they all reach the disk.

It doesn't currently pin the buffers between the two calls, and it also
assumes the buffers won't go away between the two calls, leading to
an oops if the VM manages to free the buffers in the middle of the sync.

This fixes that assumption and updates the code to return an error if things
are not up to date when the wait == 1 run is done.
Signed-off-by: Hisashi Hifumi <hifumi.hisashi@oss.ntt.co.jp>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

On multi-device filesystems, btrfs writes supers to all of the devices
before considering a sync complete.  There wasn't any additional
locking between super writeout and the device list management code
because device management was done inside a transaction and
super writeout only happened  with no transation writers running.

With the btrfs fsync log and other async transaction updates, this
has been racey for some time.  This adds a mutex to protect
the device list.  The existing volume mutex could not be reused due to
transaction lock ordering requirements.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

... otherwise generic_permission() will allow *anything* for all
files you don't own and that have some group permissions.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

In btrfs, fdatasync and fsync are identical, but
fdatasync should skip committing transaction when
inode->i_state is set just I_DIRTY_SYNC and this indicates
only atime or/and mtime updates.
Following patch improves fdatasync throughput.

--file-block-size=4K --file-total-size=16G --file-test-mode=rndwr
--file-fsync-mode=fdatasync run

Results:
-2.6.30-rc8
Test execution summary:
    total time:                          1980.6540s
    total number of events:              10001
    total time taken by event execution: 1192.9804
    per-request statistics:
         min:                            0.0000s
         avg:                            0.1193s
         max:                            15.3720s
         approx.  95 percentile:         0.7257s

Threads fairness:
    events (avg/stddev):           625.0625/151.32
    execution time (avg/stddev):   74.5613/9.46

-2.6.30-rc8-patched
Test execution summary:
    total time:                          1695.9118s
    total number of events:              10000
    total time taken by event execution: 871.3214
    per-request statistics:
         min:                            0.0000s
         avg:                            0.0871s
         max:                            10.4644s
         approx.  95 percentile:         0.4787s

Threads fairness:
    events (avg/stddev):           625.0000/131.86
    execution time (avg/stddev):   54.4576/8.98
Signed-off-by: Hisashi Hifumi <hifumi.hisashi@oss.ntt.co.jp>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

There's no need to preserve this abstraction; it used to let us use
hardware crc32c support directly, but libcrc32c is already doing that for us
through the crypto API -- so we're already using the Intel crc32c
acceleration where appropriate.
Signed-off-by: David Woodhouse <David.Woodhouse@intel.com>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

Add support for the standard attributes set via chattr and read via
lsattr.  Currently we store the attributes in the flags value in
the btrfs inode, but I wonder whether we should split it into two so
that we don't have to keep converting between the two formats.

Remove the btrfs_clear_flag/btrfs_set_flag/btrfs_test_flag macros
as they were confusing the existing code and got in the way of the
new additions.

Also add the FS_IOC_GETVERSION ioctl for getting i_generation as it's
trivial.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Chris Mason <chris.mason@oracle.com>

During mount, btrfs will check the queue nonrot flag
for all the devices found in the FS.  If they are all
non-rotating, SSD mode is enabled by default.

If the FS was mounted with -o nossd, the non-rotating
flag is ignored.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

Some SSDs perform best when reusing block numbers often, while
others perform much better when clustering strictly allocates
big chunks of unused space.

The default mount -o ssd will find rough groupings of blocks
where there are a bunch of free blocks that might have some
allocated blocks mixed in.

mount -o ssd_spread will make sure there are no allocated blocks
mixed in.  It should perform better on lower end SSDs.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

In SSD mode for data, and all the time for metadata the allocator
will try to find a cluster of nearby blocks for allocations.  This
commit adds extra checks to make sure that each free block in the
cluster is close to the last one.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

This allows you to turn off the ssd mode via remount.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

The btrfs IO submission threads try to service a bunch of devices with a small
number of threads.  They do a congestion check to try and avoid waiting
on requests for a busy device.

The checks make sure we've sent a few requests down to a given device just so
that we aren't bouncing between busy devices without actually sending down
any IO.  The counter used to decide if we can switch to the next device
is somewhat overloaded.  It is also being used to decide if we've done
a good batch of requests between the WRITE_SYNC or regular priority lists.
It may get reset to zero often, leaving us hammering on a busy device
instead of moving on to another disk.

This commit adds a new counter for the number of bios sent while
servicing a device.  It doesn't get reset or fiddled with.  On
multi-device filesystems, this fixes IO stalls in streaming
write workloads.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

Btrfs uses dedicated threads to submit bios when checksumming is on,
which allows us to make sure the threads dedicated to checksumming don't get
stuck waiting for requests.  For each btrfs device, there are
two lists of bios.  One list is for WRITE_SYNC bios and the other
is for regular priority bios.

The IO submission threads used to process all of the WRITE_SYNC bios first and
then switch to the regular bios.  This commit makes sure we don't completely
starve the regular bios by rotating between the two lists.

WRITE_SYNC bios are still favored 2:1 over the regular bios, and this tries
to run in batches to avoid seeking.  Benchmarking shows this eliminates
stalls during streaming buffered writes on both multi-device and
single device filesystems.

If the regular bios starve, the system can end up with a large amount of ram
pinned down in writeback pages.  If we are a little more fair between the two
classes, we're able to keep throughput up and make progress on the bulk of
our dirty ram.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

Once a metadata block has been written, it must be recowed, so the
btrfs dirty balancing call has a check to make sure a fair amount of metadata
was actually dirty before it started writing it back to disk.

A previous commit had changed the dirty tracking for metadata without
updating the btrfs dirty balancing checks.  This commit switches it
to use the correct counter.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

The block allocator in SSD mode will try to find groups of free blocks
that are close together.  This commit makes it loop less on a given
group size before bumping it.

The end result is that we are less likely to fill small holes in the
available free space, but we don't waste as much CPU building the
large cluster used by ssd mode.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

With the new back reference code, the cost of a balance has gone down
in terms of the number of back reference updates done.  This commit
makes us more aggressively balance leaves and nodes as they become
less full.
Signed-off-by: Chris Mason <chris.mason@oracle.com>