summaryrefslogtreecommitdiffstats
path: root/fs/xfs/xfs_trans.c
diff options
context:
space:
mode:
authorDarrick J. Wong <djwong@kernel.org>2022-04-11 22:49:42 +0200
committerDave Chinner <david@fromorbit.com>2022-04-11 22:49:42 +0200
commit2229276c5283264b8c2241c1ed972bbb136cab22 (patch)
tree2b4a00c1b7be5c610b250ff7349fae960a17e955 /fs/xfs/xfs_trans.c
parentxfs: recalculate free rt extents after log recovery (diff)
downloadlinux-2229276c5283264b8c2241c1ed972bbb136cab22.tar.xz
linux-2229276c5283264b8c2241c1ed972bbb136cab22.zip
xfs: use a separate frextents counter for rt extent reservations
As mentioned in the previous commit, the kernel misuses sb_frextents in the incore mount to reflect both incore reservations made by running transactions as well as the actual count of free rt extents on disk. This results in the superblock being written to the log with an underestimate of the number of rt extents that are marked free in the rtbitmap. Teaching XFS to recompute frextents after log recovery avoids operational problems in the current mount, but it doesn't solve the problem of us writing undercounted frextents which are then recovered by an older kernel that doesn't have that fix. Create an incore percpu counter to mirror the ondisk frextents. This new counter will track transaction reservations and the only time we will touch the incore super counter (i.e the one that gets logged) is when those transactions commit updates to the rt bitmap. This is in contrast to the lazysbcount counters (e.g. fdblocks), where we know that log recovery will always fix any incorrect counter that we log. As a bonus, we only take m_sb_lock at transaction commit time. Signed-off-by: Darrick J. Wong <djwong@kernel.org> Reviewed-by: Dave Chinner <dchinner@redhat.com> Signed-off-by: Dave Chinner <david@fromorbit.com>
Diffstat (limited to 'fs/xfs/xfs_trans.c')
-rw-r--r--fs/xfs/xfs_trans.c43
1 files changed, 37 insertions, 6 deletions
diff --git a/fs/xfs/xfs_trans.c b/fs/xfs/xfs_trans.c
index 0ac717aad380..6d9df2e9b267 100644
--- a/fs/xfs/xfs_trans.c
+++ b/fs/xfs/xfs_trans.c
@@ -498,10 +498,31 @@ xfs_trans_apply_sb_deltas(
be64_add_cpu(&sbp->sb_fdblocks, tp->t_res_fdblocks_delta);
}
- if (tp->t_frextents_delta)
- be64_add_cpu(&sbp->sb_frextents, tp->t_frextents_delta);
- if (tp->t_res_frextents_delta)
- be64_add_cpu(&sbp->sb_frextents, tp->t_res_frextents_delta);
+ /*
+ * Updating frextents requires careful handling because it does not
+ * behave like the lazysb counters because we cannot rely on log
+ * recovery in older kenels to recompute the value from the rtbitmap.
+ * This means that the ondisk frextents must be consistent with the
+ * rtbitmap.
+ *
+ * Therefore, log the frextents change to the ondisk superblock and
+ * update the incore superblock so that future calls to xfs_log_sb
+ * write the correct value ondisk.
+ *
+ * Don't touch m_frextents because it includes incore reservations,
+ * and those are handled by the unreserve function.
+ */
+ if (tp->t_frextents_delta || tp->t_res_frextents_delta) {
+ struct xfs_mount *mp = tp->t_mountp;
+ int64_t rtxdelta;
+
+ rtxdelta = tp->t_frextents_delta + tp->t_res_frextents_delta;
+
+ spin_lock(&mp->m_sb_lock);
+ be64_add_cpu(&sbp->sb_frextents, rtxdelta);
+ mp->m_sb.sb_frextents += rtxdelta;
+ spin_unlock(&mp->m_sb_lock);
+ }
if (tp->t_dblocks_delta) {
be64_add_cpu(&sbp->sb_dblocks, tp->t_dblocks_delta);
@@ -614,7 +635,12 @@ xfs_trans_unreserve_and_mod_sb(
if (ifreedelta)
percpu_counter_add(&mp->m_ifree, ifreedelta);
- if (rtxdelta == 0 && !(tp->t_flags & XFS_TRANS_SB_DIRTY))
+ if (rtxdelta) {
+ error = xfs_mod_frextents(mp, rtxdelta);
+ ASSERT(!error);
+ }
+
+ if (!(tp->t_flags & XFS_TRANS_SB_DIRTY))
return;
/* apply remaining deltas */
@@ -622,7 +648,12 @@ xfs_trans_unreserve_and_mod_sb(
mp->m_sb.sb_fdblocks += tp->t_fdblocks_delta + tp->t_res_fdblocks_delta;
mp->m_sb.sb_icount += idelta;
mp->m_sb.sb_ifree += ifreedelta;
- mp->m_sb.sb_frextents += rtxdelta;
+ /*
+ * Do not touch sb_frextents here because we are dealing with incore
+ * reservation. sb_frextents is not part of the lazy sb counters so it
+ * must be consistent with the ondisk rtbitmap and must never include
+ * incore reservations.
+ */
mp->m_sb.sb_dblocks += tp->t_dblocks_delta;
mp->m_sb.sb_agcount += tp->t_agcount_delta;
mp->m_sb.sb_imax_pct += tp->t_imaxpct_delta;