diff options
author | 2023-04-11 19:00:26 -0700 | |
---|---|---|
committer | 2023-04-11 19:00:26 -0700 | |
commit | 1e59fdb7d6157ff685a250e0873a015a2b16a4f2 (patch) | |
tree | 551c11e030cb8d29a8a07f4b1aa6cf34ba9c7f59 /fs/xfs/scrub | |
parent | xfs: split the xchk_bmap_check_rmaps into a predicate (diff) | |
download | linux-1e59fdb7d6157ff685a250e0873a015a2b16a4f2.tar.gz linux-1e59fdb7d6157ff685a250e0873a015a2b16a4f2.tar.bz2 linux-1e59fdb7d6157ff685a250e0873a015a2b16a4f2.zip |
xfs: don't call xchk_bmap_check_rmaps for btree-format file forks
The logic at the end of xchk_bmap_want_check_rmaps tries to detect a
file fork that has been zapped by what will become the online inode
repair code. Zapped forks are in FMT_EXTENTS with zero extents, and
some sort of hint that there's supposed to be data somewhere in the
filesystem.
Unfortunately, the inverted logic here is confusing and has the effect
that we always call xchk_bmap_check_rmaps for FMT_BTREE forks. This is
horribly inefficient and unnecessary, so invert the logic to get rid of
this performance problem. This has caused 8h delays in generic/333 and
generic/334.
Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Diffstat (limited to 'fs/xfs/scrub')
-rw-r--r-- | fs/xfs/scrub/bmap.c | 30 |
1 files changed, 14 insertions, 16 deletions
diff --git a/fs/xfs/scrub/bmap.c b/fs/xfs/scrub/bmap.c index 64ed5b6585d5..87ab9f95a487 100644 --- a/fs/xfs/scrub/bmap.c +++ b/fs/xfs/scrub/bmap.c @@ -645,7 +645,6 @@ xchk_bmap_want_check_rmaps( { struct xfs_scrub *sc = info->sc; struct xfs_ifork *ifp; - bool zero_size; if (!xfs_has_rmapbt(sc->mp)) return false; @@ -659,24 +658,23 @@ xchk_bmap_want_check_rmaps( return false; /* - * Only do this for complex maps that are in btree format, or for - * situations where we would seem to have a size but zero extents. - * The inode repair code can zap broken iforks, which means we have - * to flag this bmap as corrupt if there are rmaps that need to be - * reattached. + * The inode repair code zaps broken inode forks by resetting them back + * to EXTENTS format and zero extent records. If we encounter a fork + * in this state along with evidence that the fork isn't supposed to be + * empty, we need to scan the reverse mappings to decide if we're going + * to rebuild the fork. Data forks with nonzero file size are scanned. + * xattr forks are never empty of content, so they are always scanned. */ - - if (info->whichfork == XFS_DATA_FORK) - zero_size = i_size_read(VFS_I(sc->ip)) == 0; - else - zero_size = false; - ifp = xfs_ifork_ptr(sc->ip, info->whichfork); - if (ifp->if_format != XFS_DINODE_FMT_BTREE && - (zero_size || ifp->if_nextents > 0)) - return false; + if (ifp->if_format == XFS_DINODE_FMT_EXTENTS && ifp->if_nextents == 0) { + if (info->whichfork == XFS_DATA_FORK && + i_size_read(VFS_I(sc->ip)) == 0) + return false; - return true; + return true; + } + + return false; } /* Make sure each rmap has a corresponding bmbt entry. */ |