| Message ID | 155259761740.31886.12775146464846669280.stgit@magnolia (mailing list archive) |
|---|---|
| State | Superseded, archived |
| Headers | show |
| Series | xfsprogs-5.0: fix various problems | expand |
On 3/14/19 4:06 PM, Darrick J. Wong wrote: > From: Darrick J. Wong <darrick.wong@oracle.com> > > Retain the ifork ops used to validate the inode so that we can use the > same one to iflush it. xfs_repair phase 6 can use multiple transactions > to fix various inode problems, which means that the inode might not be > fully fixed when each transaction commits. > > This can be a particular problem if there's a shortform directory with > both invalid directory entries and incorrect i8count. Phase 3 will set > the parent inode to "0" to signal to phase 6 that it needs to reset the > parent and i8count, but phase 6 starts a transaction to junk the bad > entries which fail to commit because the parent is invalid: > > fixing i8count in inode 69022994673 > Invalid inode number 0x0 > xfs_dir_ino_validate: XFS_ERROR_REPORT > Metadata corruption detected at 0x464eb0, inode 0x10121750f1 data fork > xfs_repair: warning - iflush_int failed (-117) > > And thus the inode fixes never get written out. This feels like it would be better to use consistently by dropping the ops arg to libxfs_inode_verify_forks, using ->i_fork_ops inside it, and setting it prior to the call. Er, same for libxfs_iget? A mishmash of grabbing what was last set in the inode vs. explicitly pointing at a global ops structure seems a bit ad hoc. Thoughts? > > Reported-by: Arkadiusz Miskiewicz <arekm@maven.pl> > Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com> > --- > include/xfs_inode.h | 1 + > libxfs/rdwr.c | 1 + > libxfs/util.c | 2 +- > 3 files changed, 3 insertions(+), 1 deletion(-) > > > diff --git a/include/xfs_inode.h b/include/xfs_inode.h > index 79ec3a2d..e1e8b430 100644 > --- a/include/xfs_inode.h > +++ b/include/xfs_inode.h > @@ -51,6 +51,7 @@ typedef struct xfs_inode { > > xfs_fsize_t i_size; /* in-memory size */ > const struct xfs_dir_ops *d_ops; /* directory ops vector */ > + struct xfs_ifork_ops *i_fork_ops; /* fork verifiers */ > struct inode i_vnode; > } xfs_inode_t; > > diff --git a/libxfs/rdwr.c b/libxfs/rdwr.c > index a00360e7..69d5abb2 100644 > --- a/libxfs/rdwr.c > +++ b/libxfs/rdwr.c > @@ -1391,6 +1391,7 @@ libxfs_iget( > return error; > } > > + ip->i_fork_ops = ifork_ops; > if (!libxfs_inode_verify_forks(ip, ifork_ops)) { > libxfs_irele(ip); > return -EFSCORRUPTED; > diff --git a/libxfs/util.c b/libxfs/util.c > index bd414043..0799f965 100644 > --- a/libxfs/util.c > +++ b/libxfs/util.c > @@ -420,7 +420,7 @@ libxfs_iflush_int(xfs_inode_t *ip, xfs_buf_t *bp) > VFS_I(ip)->i_version++; > > /* Check the inline fork data before we write out. */ > - if (!libxfs_inode_verify_forks(ip, &xfs_default_ifork_ops)) > + if (!libxfs_inode_verify_forks(ip, ip->i_fork_ops)) > return -EFSCORRUPTED; > > /* >
On 4/5/19 1:17 PM, Eric Sandeen wrote: > On 3/14/19 4:06 PM, Darrick J. Wong wrote: >> From: Darrick J. Wong <darrick.wong@oracle.com> >> >> Retain the ifork ops used to validate the inode so that we can use the >> same one to iflush it. xfs_repair phase 6 can use multiple transactions >> to fix various inode problems, which means that the inode might not be >> fully fixed when each transaction commits. >> >> This can be a particular problem if there's a shortform directory with >> both invalid directory entries and incorrect i8count. Phase 3 will set >> the parent inode to "0" to signal to phase 6 that it needs to reset the >> parent and i8count, but phase 6 starts a transaction to junk the bad >> entries which fail to commit because the parent is invalid: >> >> fixing i8count in inode 69022994673 >> Invalid inode number 0x0 >> xfs_dir_ino_validate: XFS_ERROR_REPORT >> Metadata corruption detected at 0x464eb0, inode 0x10121750f1 data fork >> xfs_repair: warning - iflush_int failed (-117) >> >> And thus the inode fixes never get written out. > > This feels like it would be better to use consistently by dropping the ops > arg to libxfs_inode_verify_forks, using ->i_fork_ops inside it, and setting > it prior to the call. Er, same for libxfs_iget? er, no not for libxfs_iget, sorry. But libxfs_iget can set the ops on the inode if found and from then on stuff like libxfs_inode_verify_forks uses what's currently set in the inode ....) > A mishmash of grabbing what was last set in the inode vs. explicitly pointing > at a global ops structure seems a bit ad hoc. Thoughts?
On Fri, Apr 05, 2019 at 01:19:52PM -0500, Eric Sandeen wrote: > > > On 4/5/19 1:17 PM, Eric Sandeen wrote: > > On 3/14/19 4:06 PM, Darrick J. Wong wrote: > >> From: Darrick J. Wong <darrick.wong@oracle.com> > >> > >> Retain the ifork ops used to validate the inode so that we can use the > >> same one to iflush it. xfs_repair phase 6 can use multiple transactions > >> to fix various inode problems, which means that the inode might not be > >> fully fixed when each transaction commits. > >> > >> This can be a particular problem if there's a shortform directory with > >> both invalid directory entries and incorrect i8count. Phase 3 will set > >> the parent inode to "0" to signal to phase 6 that it needs to reset the > >> parent and i8count, but phase 6 starts a transaction to junk the bad > >> entries which fail to commit because the parent is invalid: > >> > >> fixing i8count in inode 69022994673 > >> Invalid inode number 0x0 > >> xfs_dir_ino_validate: XFS_ERROR_REPORT > >> Metadata corruption detected at 0x464eb0, inode 0x10121750f1 data fork > >> xfs_repair: warning - iflush_int failed (-117) > >> > >> And thus the inode fixes never get written out. > > > > This feels like it would be better to use consistently by dropping the ops > > arg to libxfs_inode_verify_forks, using ->i_fork_ops inside it, and setting > > it prior to the call. Er, same for libxfs_iget? > > er, no not for libxfs_iget, sorry. But libxfs_iget can set the ops on > the inode if found and from then on stuff like libxfs_inode_verify_forks > uses what's currently set in the inode ....) > > > A mishmash of grabbing what was last set in the inode vs. explicitly pointing > > at a global ops structure seems a bit ad hoc. Thoughts? Yeah, there's no reason to have the second parameter when we've already got it in @ip, so I'll make a new patch to refactor it out. --D
diff --git a/include/xfs_inode.h b/include/xfs_inode.h index 79ec3a2d..e1e8b430 100644 --- a/include/xfs_inode.h +++ b/include/xfs_inode.h @@ -51,6 +51,7 @@ typedef struct xfs_inode { xfs_fsize_t i_size; /* in-memory size */ const struct xfs_dir_ops *d_ops; /* directory ops vector */ + struct xfs_ifork_ops *i_fork_ops; /* fork verifiers */ struct inode i_vnode; } xfs_inode_t; diff --git a/libxfs/rdwr.c b/libxfs/rdwr.c index a00360e7..69d5abb2 100644 --- a/libxfs/rdwr.c +++ b/libxfs/rdwr.c @@ -1391,6 +1391,7 @@ libxfs_iget( return error; } + ip->i_fork_ops = ifork_ops; if (!libxfs_inode_verify_forks(ip, ifork_ops)) { libxfs_irele(ip); return -EFSCORRUPTED; diff --git a/libxfs/util.c b/libxfs/util.c index bd414043..0799f965 100644 --- a/libxfs/util.c +++ b/libxfs/util.c @@ -420,7 +420,7 @@ libxfs_iflush_int(xfs_inode_t *ip, xfs_buf_t *bp) VFS_I(ip)->i_version++; /* Check the inline fork data before we write out. */ - if (!libxfs_inode_verify_forks(ip, &xfs_default_ifork_ops)) + if (!libxfs_inode_verify_forks(ip, ip->i_fork_ops)) return -EFSCORRUPTED; /*