The fixes do a number of things:

1) Most btrfs_drop_extent callers will try to leave the inline extents in
place.  It can truncate bytes off the beginning of the inline extent if
required.

2) writepage can now update the inline extent, allowing mmap writes to
go directly into the inline extent.

3) btrfs_truncate_in_transaction truncates inline extents

4) extent_map.c fixed to not merge inline extent mappings and hole
mappings together
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

1.  Reorder kmap and the test for  'page != NULL'
2.  Zero-fill rest area of a block when inline extent isn't big enough.
3.  Do not insert extent_map into the map tree when page == NULL.
(If insert the extent_map into the map tree,  subsequent read requests
will find it in the map tree directly and the corresponding inline
extent data aren't copied into page by the the get_extent function.
extent_read_full_page can't handle that case)
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

The ENOTEMPTY check in btrfs_rmdir isn't reliable. It's possible that
the backward search finds . or .. at first, then some other directory
entry. In that case,  btrfs_rmdir delete . or .. improperly.  The
patch also fixes a fs_mutex unlock issue in  btrfs_rmdir.

--
Signed-off-by: NChris Mason <chris.mason@oracle.com>

When inode is found, the return value is from the uninitialized
variable 'ret'.

--
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

This reduces the number of calls to btrfs_extend_item and greatly lowers
the cpu usage while writing large files.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

endio handling is typically called with interrupts disabled, but can
also be called with it enabled. So save interrupts before using KM_IRQ0
to be completely safe.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

An assorted set of casts to get rid of the warnings on 32-bit archs.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

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

The first change removes  potential double free, the second fix a off
by one error.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

I think check whether extent is a hole before update 'inode->i_blocks'
is unconditional required. (original codes check it only when
del_item isn't equal to 0)
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

found_type has already been decreased by codes above the change,  I
think decrease it by one again doesn't make sense.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

btrfs_btree_balance_dirty is changed to pass the number of pages dirtied
for more accurate dirty throttling.  This lets the VM make better decisions
about when to force some writeback.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Add a helper per ioctl function to make the code more readable.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

No reason to grab the BKL before calling into the btrfs ioctl code.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

generic_bmap is completely trivial, while the extent to bh mapping in
btrfs is rather complex.  So provide a extent_bmap instead that takes
a get_extent callback and can be used by filesystem using the extent_map
code.
Signed-off-by: NChristoph Hellwig <hch@lst.de>
Signed-off-by: NChris Mason <chris.mason@oracle.com>

File data checksums are only done during writepage, so we have to make sure
all pages are written when the snapshot is taken.  This also adds some
locking so that new writes don't race in and add new dirty pages.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

defrag.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

instead of buffer heads.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>

This adds two types of btree defrag, a run time form that tries to
defrag recently allocated blocks in the btree when they are still in ram,
and an ioctl that forces defrag of all btree blocks.

File data blocks are not defragged yet, but this can make a huge difference
in sequential btree reads.
Signed-off-by: NChris Mason <chris.mason@oracle.com>

Signed-off-by: NChris Mason <chris.mason@oracle.com>