On Thu, Dec 09, 2021 at 04:41:49PM +1100, Dave Chinner wrote:
> On Wed, Dec 08, 2021 at 03:15:16PM -0800, Darrick J. Wong wrote:
> > From: Darrick J. Wong <djwong@xxxxxxxxxx>
> >
> > As part of multiple customer escalations due to file data corruption
> > after copy on write operations, I wrote some fstests that use fsstress
> > to hammer on COW to shake things loose. Regrettably, I caught some
> > filesystem shutdowns due to incorrect rmap operations with the following
> > loop:
> >
> > mount <filesystem> # (0)
> > fsstress <run only readonly ops> & # (1)
> > while true; do
> > fsstress <run all ops>
> > mount -o remount,ro # (2)
> > fsstress <run only readonly ops>
> > mount -o remount,rw # (3)
> > done
> >
> > When (2) happens, notice that (1) is still running. xfs_remount_ro will
> > call xfs_blockgc_stop to walk the inode cache to free all the COW
> > extents, but the blockgc mechanism races with (1)'s reader threads to
> > take IOLOCKs and loses, which means that it doesn't clean them all out.
> > Call such a file (A).
> >
> > When (3) happens, xfs_remount_rw calls xfs_reflink_recover_cow, which
> > walks the ondisk refcount btree and frees any COW extent that it finds.
> > This function does not check the inode cache, which means that incore
> > COW forks of inode (A) is now inconsistent with the ondisk metadata. If
> > one of those former COW extents are allocated and mapped into another
> > file (B) and someone triggers a COW to the stale reservation in (A), A's
> > dirty data will be written into (B) and once that's done, those blocks
> > will be transferred to (A)'s data fork without bumping the refcount.
> >
> > The results are catastrophic -- file (B) and the refcount btree are now
> > corrupt. In the first patch, we fixed the race condition in (2) so that
> > (A) will always flush the COW fork. In this second patch, we move the
> > _recover_cow call to the initial mount call in (0) for safety.
> >
> > As mentioned previously, xfs_reflink_recover_cow walks the refcount
> > btree looking for COW staging extents, and frees them. This was
> > intended to be run at mount time (when we know there are no live inodes)
> > to clean up any leftover staging events that may have been left behind
> > during an unclean shutdown. As a time "optimization" for readonly
> > mounts, we deferred this to the ro->rw transition, not realizing that
> > any failure to clean all COW forks during a rw->ro transition would
> > result in catastrophic corruption.
> >
> > Therefore, remove this optimization and only run the recovery routine
> > when we're guaranteed not to have any COW staging extents anywhere,
> > which means we always run this at mount time. While we're at it, move
> > the callsite to xfs_log_mount_finish because any refcount btree
> > expansion (however unlikely given that we're removing records from the
> > right side of the index) must be fed by a per-AG reservation, which
> > doesn't exist in its current location.
> >
> > Fixes: 174edb0e46e5 ("xfs: store in-progress CoW allocations in the refcount btree")
> > Signed-off-by: Darrick J. Wong <djwong@xxxxxxxxxx>
> > ---
> > fs/xfs/xfs_log.c | 23 ++++++++++++++++++++++-
> > fs/xfs/xfs_mount.c | 10 ----------
> > fs/xfs/xfs_reflink.c | 5 ++++-
> > fs/xfs/xfs_super.c | 9 ---------
> > 4 files changed, 26 insertions(+), 21 deletions(-)
> >
> >
> > diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c
> > index 89fec9a18c34..c17344fc1275 100644
> > --- a/fs/xfs/xfs_log.c
> > +++ b/fs/xfs/xfs_log.c
> > @@ -10,6 +10,7 @@
> > #include "xfs_log_format.h"
> > #include "xfs_trans_resv.h"
> > #include "xfs_mount.h"
> > +#include "xfs_inode.h"
> > #include "xfs_errortag.h"
> > #include "xfs_error.h"
> > #include "xfs_trans.h"
> > @@ -20,6 +21,7 @@
> > #include "xfs_sysfs.h"
> > #include "xfs_sb.h"
> > #include "xfs_health.h"
> > +#include "xfs_reflink.h"
> >
> > struct kmem_cache *xfs_log_ticket_cache;
> >
> > @@ -847,9 +849,28 @@ xfs_log_mount_finish(
> > /* Make sure the log is dead if we're returning failure. */
> > ASSERT(!error || xlog_is_shutdown(log));
> >
> > - return error;
> > + if (error)
> > + return error;
> > +
> > + /*
> > + * Recover any CoW staging blocks that are still referenced by the
> > + * ondisk refcount metadata. During mount there cannot be any live
> > + * staging extents as we have not permitted any user modifications.
> > + * Therefore, it is safe to free them all right now, even on a
> > + * read-only mount.
> > + */
> > + error = xfs_reflink_recover_cow(mp);
> > + if (error) {
> > + xfs_err(mp, "Error %d recovering leftover CoW allocations.",
> > + error);
> > + xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
> > + return error;
> > + }
> > +
> > + return 0;
> > }
>
> THis is after we've emitted an "Ending recovery ...." log message.
> I kinda expected you to move this up to just after the
> evict_inodes() call before we force the log, push the AIL, drain
> the buffers used during recovery and potentially turn the
> "filesystem is read-only" flag back on.
>
> i.e. if this is a recovery operation, it should be done before we
> finish recovery....
>
> Other than that, the change is fine.
It's a recovery function, albeit one that we always run during mount,
even if the log didn't require recovery. That's a behavior change,
which is beyond the scope of a fix patch. Not to mention it causes a
hang in xfs/434, which means now I have to withdraw this until I sort
out why we stall forever in xfs_buftarg_drain, and ftrace isn't helpful
in this regard.
--D
>
> Cheers,
>
> Dave.
> --
> Dave Chinner
> david@xxxxxxxxxxxxx
