| Message ID | 20201217145334.2512475-1-bfoster@redhat.com (mailing list archive) |
|---|---|
| State | Superseded |
| Headers | show |
| Series | xfs: sync lazy sb accounting on quiesce of read-only mounts | expand |
On Thu, Dec 17, 2020 at 09:53:34AM -0500, Brian Foster wrote: > xfs_log_sbcount() syncs the superblock specifically to accumulate > the in-core percpu superblock counters and commit them to disk. This > is required to maintain filesystem consistency across quiesce > (freeze, read-only mount/remount) or unmount when lazy superblock > accounting is enabled because individual transactions do not update > the superblock directly. > > This mechanism works as expected for writable mounts, but > xfs_log_sbcount() skips the update for read-only mounts. Read-only > mounts otherwise still allow log recovery and write out an unmount > record during log quiesce. If a read-only mount performs log > recovery, it can modify the in-core superblock counters and write an > unmount record when the filesystem unmounts without ever syncing the > in-core counters. This leaves the filesystem with a clean log but in > an inconsistent state with regard to lazy sb counters. > > Update xfs_log_sbcount() to use the same logic > xfs_log_unmount_write() uses to determine when to write an unmount > record. We can drop the freeze state check because the update is > already allowed during the freezing process and no context calls > this function on an already frozen fs. This ensures that lazy > accounting is always synced before the log is cleaned. Refactor this > logic into a new helper to distinguish between a writable filesystem > and a writable log. Specifically, the log is writable unless the > filesystem is mounted with the norecovery mount option, the > underlying log device is read-only, or the filesystem is shutdown. > > Signed-off-by: Brian Foster <bfoster@redhat.com> It seems I get the point... Although a minor thing is that I think if xfs_log_writable() in xfs_log_sbcount() is false, there is no need to xfs_log_quiesce() entirely then... But that may be another rework story. So overall, it looks good to me: Reviewed-by: Gao Xiang <hsiangkao@redhat.com> Thanks, Gao Xiang > --- > > This is something I noticed when reworking the log quiesce code to reuse > log covering. It seems like a bug worth an independent fix, so I peeled > it off into a standalone patch. Note that the broader rework currently > removes both xfs_log_sbcount() and xfs_quiesce_attr(), so this is > intended to be an isolated/backportable bug fix. The problem is easily > reproducible with a small tweak to generic/388 that I'll post shortly... > > Brian > > fs/xfs/xfs_log.c | 28 ++++++++++++++++++++-------- > fs/xfs/xfs_log.h | 1 + > fs/xfs/xfs_mount.c | 3 +-- > 3 files changed, 22 insertions(+), 10 deletions(-) > > diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c > index fa2d05e65ff1..b445e63cbc3c 100644 > --- a/fs/xfs/xfs_log.c > +++ b/fs/xfs/xfs_log.c > @@ -347,6 +347,25 @@ xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type) > tic->t_res_num++; > } > > +bool > +xfs_log_writable( > + struct xfs_mount *mp) > +{ > + /* > + * Never write to the log on norecovery mounts, if the block device is > + * read-only, or if the filesystem is shutdown. Read-only mounts still > + * allow internal writes for log recovery and unmount purposes, so don't > + * restrict that case here. > + */ > + if (mp->m_flags & XFS_MOUNT_NORECOVERY) > + return false; > + if (xfs_readonly_buftarg(mp->m_log->l_targ)) > + return false; > + if (XFS_FORCED_SHUTDOWN(mp)) > + return false; > + return true; > +} > + > /* > * Replenish the byte reservation required by moving the grant write head. > */ > @@ -886,15 +905,8 @@ xfs_log_unmount_write( > { > struct xlog *log = mp->m_log; > > - /* > - * Don't write out unmount record on norecovery mounts or ro devices. > - * Or, if we are doing a forced umount (typically because of IO errors). > - */ > - if (mp->m_flags & XFS_MOUNT_NORECOVERY || > - xfs_readonly_buftarg(log->l_targ)) { > - ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); > + if (!xfs_log_writable(mp)) > return; > - } > > xfs_log_force(mp, XFS_LOG_SYNC); > > diff --git a/fs/xfs/xfs_log.h b/fs/xfs/xfs_log.h > index 58c3fcbec94a..98c913da7587 100644 > --- a/fs/xfs/xfs_log.h > +++ b/fs/xfs/xfs_log.h > @@ -127,6 +127,7 @@ int xfs_log_reserve(struct xfs_mount *mp, > int xfs_log_regrant(struct xfs_mount *mp, struct xlog_ticket *tic); > void xfs_log_unmount(struct xfs_mount *mp); > int xfs_log_force_umount(struct xfs_mount *mp, int logerror); > +bool xfs_log_writable(struct xfs_mount *mp); > > struct xlog_ticket *xfs_log_ticket_get(struct xlog_ticket *ticket); > void xfs_log_ticket_put(struct xlog_ticket *ticket); > diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c > index 7110507a2b6b..a62b8a574409 100644 > --- a/fs/xfs/xfs_mount.c > +++ b/fs/xfs/xfs_mount.c > @@ -1176,8 +1176,7 @@ xfs_fs_writable( > int > xfs_log_sbcount(xfs_mount_t *mp) > { > - /* allow this to proceed during the freeze sequence... */ > - if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE)) > + if (!xfs_log_writable(mp)) > return 0; > > /* > -- > 2.26.2 >
diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c index fa2d05e65ff1..b445e63cbc3c 100644 --- a/fs/xfs/xfs_log.c +++ b/fs/xfs/xfs_log.c @@ -347,6 +347,25 @@ xlog_tic_add_region(xlog_ticket_t *tic, uint len, uint type) tic->t_res_num++; } +bool +xfs_log_writable( + struct xfs_mount *mp) +{ + /* + * Never write to the log on norecovery mounts, if the block device is + * read-only, or if the filesystem is shutdown. Read-only mounts still + * allow internal writes for log recovery and unmount purposes, so don't + * restrict that case here. + */ + if (mp->m_flags & XFS_MOUNT_NORECOVERY) + return false; + if (xfs_readonly_buftarg(mp->m_log->l_targ)) + return false; + if (XFS_FORCED_SHUTDOWN(mp)) + return false; + return true; +} + /* * Replenish the byte reservation required by moving the grant write head. */ @@ -886,15 +905,8 @@ xfs_log_unmount_write( { struct xlog *log = mp->m_log; - /* - * Don't write out unmount record on norecovery mounts or ro devices. - * Or, if we are doing a forced umount (typically because of IO errors). - */ - if (mp->m_flags & XFS_MOUNT_NORECOVERY || - xfs_readonly_buftarg(log->l_targ)) { - ASSERT(mp->m_flags & XFS_MOUNT_RDONLY); + if (!xfs_log_writable(mp)) return; - } xfs_log_force(mp, XFS_LOG_SYNC); diff --git a/fs/xfs/xfs_log.h b/fs/xfs/xfs_log.h index 58c3fcbec94a..98c913da7587 100644 --- a/fs/xfs/xfs_log.h +++ b/fs/xfs/xfs_log.h @@ -127,6 +127,7 @@ int xfs_log_reserve(struct xfs_mount *mp, int xfs_log_regrant(struct xfs_mount *mp, struct xlog_ticket *tic); void xfs_log_unmount(struct xfs_mount *mp); int xfs_log_force_umount(struct xfs_mount *mp, int logerror); +bool xfs_log_writable(struct xfs_mount *mp); struct xlog_ticket *xfs_log_ticket_get(struct xlog_ticket *ticket); void xfs_log_ticket_put(struct xlog_ticket *ticket); diff --git a/fs/xfs/xfs_mount.c b/fs/xfs/xfs_mount.c index 7110507a2b6b..a62b8a574409 100644 --- a/fs/xfs/xfs_mount.c +++ b/fs/xfs/xfs_mount.c @@ -1176,8 +1176,7 @@ xfs_fs_writable( int xfs_log_sbcount(xfs_mount_t *mp) { - /* allow this to proceed during the freeze sequence... */ - if (!xfs_fs_writable(mp, SB_FREEZE_COMPLETE)) + if (!xfs_log_writable(mp)) return 0; /*
xfs_log_sbcount() syncs the superblock specifically to accumulate the in-core percpu superblock counters and commit them to disk. This is required to maintain filesystem consistency across quiesce (freeze, read-only mount/remount) or unmount when lazy superblock accounting is enabled because individual transactions do not update the superblock directly. This mechanism works as expected for writable mounts, but xfs_log_sbcount() skips the update for read-only mounts. Read-only mounts otherwise still allow log recovery and write out an unmount record during log quiesce. If a read-only mount performs log recovery, it can modify the in-core superblock counters and write an unmount record when the filesystem unmounts without ever syncing the in-core counters. This leaves the filesystem with a clean log but in an inconsistent state with regard to lazy sb counters. Update xfs_log_sbcount() to use the same logic xfs_log_unmount_write() uses to determine when to write an unmount record. We can drop the freeze state check because the update is already allowed during the freezing process and no context calls this function on an already frozen fs. This ensures that lazy accounting is always synced before the log is cleaned. Refactor this logic into a new helper to distinguish between a writable filesystem and a writable log. Specifically, the log is writable unless the filesystem is mounted with the norecovery mount option, the underlying log device is read-only, or the filesystem is shutdown. Signed-off-by: Brian Foster <bfoster@redhat.com> --- This is something I noticed when reworking the log quiesce code to reuse log covering. It seems like a bug worth an independent fix, so I peeled it off into a standalone patch. Note that the broader rework currently removes both xfs_log_sbcount() and xfs_quiesce_attr(), so this is intended to be an isolated/backportable bug fix. The problem is easily reproducible with a small tweak to generic/388 that I'll post shortly... Brian fs/xfs/xfs_log.c | 28 ++++++++++++++++++++-------- fs/xfs/xfs_log.h | 1 + fs/xfs/xfs_mount.c | 3 +-- 3 files changed, 22 insertions(+), 10 deletions(-)