This patch adds a helper function 'update_pinned_extents' to
extent-tree.c. The usage of the helper function is similar to
'update_block_group',  the last parameter of the function indicates
pin vs unpin.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

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

It could return the bit as set when there was actually a hole at the
very end of the range.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

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

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

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

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

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

When 'item_end' is equal to 'inode->i_size',   'found_type' is updated
and current item is skipped. This behavior is correct for extent item,
but incorrect for csum item. For example, there is a csum item with
'offset == 0'. When deleting the inode,  'inode->i_size' is set to 0,
so the csum item isn't deleted.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

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

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

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

Don't set hint_byte to EXTENT_MAP_INLINE when 'end == extent_end' or
'start == key.offset' . The inline extent will be truncated in these
cases.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

When calculating the size of inline extent,  inode->i_size should also
be take into consideration, otherwise sys_write may drop some data
silently.  You can test this bug by:

#dd if=/dev/zero bs=4k count=1 of=test_file
#dd if=/dev/zero bs=2k count=1 of=test_file conv=notrunc
Signed-off-by: Chris Mason <chris.mason@oracle.com>

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

When pin_down_bytes decides not to pin a block because it was from the
current transaction, make sure the in memory cache of free extents is updated
Signed-off-by: Chris Mason <chris.mason@oracle.com>

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

There is a 'finish_wait', but no 'prepare_to_wait' . So I think that
the 'prepare_to_wait' is missing. The second change is  according to
the name of variable.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

This patch fixes a typo in update_block_group and memory leak in
btrfs_free_block_groups.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

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: Chris Mason <chris.mason@oracle.com>

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

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

Execution should goto label 'insert' when 'btrfs_next_leaf' return a
non-zero value, otherwise the parameter 'slot' for
'btrfs_item_key_to_cpu' may be out of bounds. The original codes jump
to  label 'insert' only when 'btrfs_next_leaf' return a negative
value.
Signed-off-by: Chris 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: Chris Mason <chris.mason@oracle.com>

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

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

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

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

1) Forced defrag wasn't working properly (btrfsctl -d) because some
cache only checks were incorrect.

2) Defrag only the leaves unless in forced defrag mode.

3) Don't use complex logic to figure out if a leaf is needs defrag
Signed-off-by: Chris Mason <chris.mason@oracle.com>

This modifies inline extent size calculation, so that
insert_inline_extent can handle the case that parameter 'offset' is
not zero; it also a few codes to zero uninitialized area in inline
extent.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

When making room for a new item, it is ok to create an empty leaf, but
when making room to extend an item, split_leaf needs to make sure it
keeps the item we're extending in the path and make sure we don't end up
with an empty leaf.
Signed-off-by: Chris Mason <chris.mason@oracle.com>

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

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

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

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

This avoids recursive use of KM_USER0 during btrfs_file_write
Signed-off-by: Chris Mason <chris.mason@oracle.com>

Just use kobject_set_name(), that works in all kernels (I think...).
Kernels newer than 2.6.23 currently fail with:

/home/axboe/git/btrfs/btrfs-unstable/sysfs.c:188: error: unknown field
'name' specified in initializer
Signed-off-by: Chris 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: Chris Mason <chris.mason@oracle.com>

It now returns void and it is never called for partial completions, so
the bio->bi_size check must go.
Signed-off-by: Chris Mason <chris.mason@oracle.com>