Just as Filipe pointed out, the most time consuming parts of qgroup are
btrfs_qgroup_account_extents() and
btrfs_qgroup_prepare_account_extents().
Which both call btrfs_find_all_roots() to get old_roots and new_roots
ulist.

What makes things worse is, we're calling that expensive
btrfs_find_all_roots() at transaction committing time with
TRANS_STATE_COMMIT_DOING, which will blocks all incoming transaction.

Such behavior is necessary for @new_roots search as current
btrfs_find_all_roots() can't do it correctly so we do call it just
before switch commit roots.

However for @old_roots search, it's not necessary as such search is
based on commit_root, so it will always be correct and we can move it
out of transaction committing.

This patch moves the @old_roots search part out of
commit_transaction(), so in theory we can half the time qgroup time
consumption at commit_transaction().

But please note that, this won't speedup qgroup overall, the total time
consumption is still the same, just reduce the performance stall.

Cc: Filipe Manana <fdmanana@suse.com>
Signed-off-by: NQu Wenruo <quwenruo@cn.fujitsu.com>
Reviewed-by: NFilipe Manana <fdmanana@suse.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Never used.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Both unused after the call to update_cache_item has been moved to
__btrfs_wait_cache_io.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

bitmap_list is unused since the io_ctl framework.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Unused since the helper has been split, eb used in the caller.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Never used.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

After the page locking has been reworked, we get all pages prepared via
cmp_pages.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Never used.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Never used.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The name parameters have never been used, as the name is passed via the
dentry.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The 'device' used to be added in that function, but now it's done by the
caller.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We grab fs_info from other parameters.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Never used for anything meaningful since we have our own superblock
filler.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The 'tree' was used to call locking hook that does not exist anymore.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Never used.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The logic has been updated in "Btrfs: make mapping->writeback_index
point to the last written page" (a9132667) and page is not
needed anymore.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

This used to hold number of maximum pages to allocate, but this is now
limited by BIO_MAX_PAGES. The local are now unused and removed as well.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Not needed.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Never needed.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

None of the checks need to know the ro/rw status as they're all not
changing the superblock. Moreover, we can access the sb flags directly
if we'd need to decide by the ro/rw status.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Added but never used.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Unused since qgroup refactoring that split data and metadata accounting,
the btrfs_qgroup_free helper.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Unused since long ago.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Unused since long ago.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

write_all_supers and write_ctree_super are almost equal, the parameter
'trans' is unused so we can drop it and have just one helper.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The barriers are handled by the caller.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Never used.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Added but never needed.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Added but never used.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Never used in that function.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Change the name so it matches the naming we already use eg. for
btrfs_path.
Suggested-by: NQu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

There was never need for RCU protection around reading nodesize or other
fairly constant filesystem data.
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The helper name is not too helpful and is just wrapping a simple call.
Signed-off-by: NDavid Sterba <dsterba@suse.com>

The quota status used to be tracked as a variable, so the mutex was
needed (until "Btrfs: add a flags field to btrfs_fs_info" afcdd129).
Since the status is a bit modified atomically and we don't hold the
mutex beyond the check, we can drop it.
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Status of quotas should be the first check in
btrfs_qgroup_account_extent and we can return immediatelly, no need to
do no-op ulist frees.
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We can embed range_changed to the extent changeset to address following
problems:

- no need to allocate ulist dynamically, we also get rid of the GFP_NOFS
  for free
- fix lack of allocation failure checking in btrfs_qgroup_reserve_data

The stack consuption where extent_changeset is used slightly increases:

before: 16
after: 16 - 8 (for pointer) + 32 (sizeof ulist) = 40

Which is bearable.
Reviewed-by: NQu Wenruo <quwenruo@cn.fujitsu.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Make ulist_fini externally visible so the ulist API is complete.
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Internal helper.
Signed-off-by: NDavid Sterba <dsterba@suse.com>

We do a readahead of the free space cache inode to speed things up but
the failure is not fatal, like in other readahead cases. Proper reads
would need to happen anyway and any errors would be caught there.
Reviewed-by: NLiu Bo <bo.li.liu@oracle.com>
Signed-off-by: NDavid Sterba <dsterba@suse.com>

Qgroup relations are added/deleted from ioctl, we hold the high level
qgroup lock, no deadlocks or recursion from the allocation possible
here.
Signed-off-by: NDavid Sterba <dsterba@suse.com>