Re: [PATCH v3 2/7] btrfs: check for priority ticket granting before flushing
From: Nikolay Borisov <hidden>
Date: 2021-11-11 13:14:28
On 9.11.21 г. 17:12, Josef Bacik wrote:
Since we're dropping locks before we enter the priority flushing loops we could have had our ticket granted before we got the space_info->lock. So add this check to avoid doing some extra flushing in the priority flushing cases. Signed-off-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Nikolay Borisov <redacted>
quoted hunk ↗ jump to hunk
---> fs/btrfs/space-info.c | 9 ++++++++- 1 file changed, 8 insertions(+), 1 deletion(-)diff --git a/fs/btrfs/space-info.c b/fs/btrfs/space-info.c index 9d6048f54097..9a362f3a6df4 100644 --- a/fs/btrfs/space-info.c +++ b/fs/btrfs/space-info.c@@ -1264,7 +1264,7 @@ static void priority_reclaim_metadata_space(struct btrfs_fs_info *fs_info, spin_lock(&space_info->lock); to_reclaim = btrfs_calc_reclaim_metadata_size(fs_info, space_info); - if (!to_reclaim) { + if (!to_reclaim || ticket->bytes == 0) {
nit: This is purely an optimization, handling the case where a prio ticket N is being added to the list, but at the same time we might have had ticket N-1 just satisfied (or failed) and having called try_granting_ticket might have satisfied concurrently added ticket N, right? And this is a completely independent change of the other cleanups being done here?
quoted hunk ↗ jump to hunk
spin_unlock(&space_info->lock); return; }@@ -1297,6 +1297,13 @@ static void priority_reclaim_data_space(struct btrfs_fs_info *fs_info, struct reserve_ticket *ticket) { spin_lock(&space_info->lock); + + /* We could have been granted before we got here. */ + if (ticket->bytes == 0) { + spin_unlock(&space_info->lock); + return; + } + while (!space_info->full) { spin_unlock(&space_info->lock); flush_space(fs_info, space_info, U64_MAX, ALLOC_CHUNK_FORCE, false);