| Message ID | 9a8e2f9639330dd5e82db11a49f84fa17cb9988d.1701464169.git.boris@bur.io (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | btrfs: qgroups rsv fixes | expand |
On 2023/12/2 07:30, Boris Burkov wrote: > The EXTENT_QGROUP_RESERVED bit is used to "lock" regions of the file for > duplicate reservations. That is two writes to that range in one > transaction shouldn't create two reservations, as the reservation will > only be freed once when the write finally goes down. Therefore, it is > never OK to clear that bit without freeing the associated qgroup rsv. At > this point, we don't want to be freeing the rsv, so mask off the bit. Did you have any dmesg output for this case? I believe this would lead to a underflow. > > Signed-off-by: Boris Burkov <boris@bur.io> Reviewed-by: Qu Wenruo <wqu@suse.com> Thanks, Qu > --- > fs/btrfs/extent_io.c | 3 ++- > 1 file changed, 2 insertions(+), 1 deletion(-) > > diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c > index 0143bf63044c..87283087c669 100644 > --- a/fs/btrfs/extent_io.c > +++ b/fs/btrfs/extent_io.c > @@ -2310,7 +2310,8 @@ static int try_release_extent_state(struct extent_io_tree *tree, > ret = 0; > } else { > u32 clear_bits = ~(EXTENT_LOCKED | EXTENT_NODATASUM | > - EXTENT_DELALLOC_NEW | EXTENT_CTLBITS); > + EXTENT_DELALLOC_NEW | EXTENT_CTLBITS > + | EXTENT_QGROUP_RESERVED); > > /* > * At this point we can safely clear everything except the
diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index 0143bf63044c..87283087c669 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -2310,7 +2310,8 @@ static int try_release_extent_state(struct extent_io_tree *tree, ret = 0; } else { u32 clear_bits = ~(EXTENT_LOCKED | EXTENT_NODATASUM | - EXTENT_DELALLOC_NEW | EXTENT_CTLBITS); + EXTENT_DELALLOC_NEW | EXTENT_CTLBITS + | EXTENT_QGROUP_RESERVED); /* * At this point we can safely clear everything except the
The EXTENT_QGROUP_RESERVED bit is used to "lock" regions of the file for duplicate reservations. That is two writes to that range in one transaction shouldn't create two reservations, as the reservation will only be freed once when the write finally goes down. Therefore, it is never OK to clear that bit without freeing the associated qgroup rsv. At this point, we don't want to be freeing the rsv, so mask off the bit. Signed-off-by: Boris Burkov <boris@bur.io> --- fs/btrfs/extent_io.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-)