On Mon, May 04, 2020 at 06:13:53PM -0700, Darrick J. Wong wrote: > From: Darrick J. Wong <darrick.wong@xxxxxxxxxx> > > In xfs_bui_item_recover, there exists a use-after-free bug with regards > to the inode that is involved in the bmap replay operation. If the > mapping operation does not complete, we call xfs_bmap_unmap_extent to > create a deferred op to finish the unmapping work, and we retain a > pointer to the incore inode. > > Unfortunately, the very next thing we do is commit the transaction and > drop the inode. If reclaim tears down the inode before we try to finish > the defer ops, we dereference garbage and blow up. Therefore, create a > way to join inodes to the defer ops freezer so that we can maintain the > xfs_inode reference until we're done with the inode. > > Note: This imposes the requirement that there be enough memory to keep > every incore inode in memory throughout recovery. > > Signed-off-by: Darrick J. Wong <darrick.wong@xxxxxxxxxx> > --- Maybe I'm missing something, but I thought the discussion on the previous version[1] landed on an approach where the intent would hold a reference to the inode. Wouldn't that break the dependency on the dfops freeze/thaw mechanism? Brian [1] https://lore.kernel.org/linux-xfs/20200429235818.GX6742@magnolia/ > fs/xfs/libxfs/xfs_defer.c | 50 +++++++++++++++++++++++++++++++++++++++++++++ > fs/xfs/libxfs/xfs_defer.h | 10 +++++++++ > fs/xfs/xfs_bmap_item.c | 7 ++++-- > fs/xfs/xfs_icache.c | 19 +++++++++++++++++ > 4 files changed, 83 insertions(+), 3 deletions(-) > > > diff --git a/fs/xfs/libxfs/xfs_defer.c b/fs/xfs/libxfs/xfs_defer.c > index ea4d28851bbd..72933fdafcb2 100644 > --- a/fs/xfs/libxfs/xfs_defer.c > +++ b/fs/xfs/libxfs/xfs_defer.c > @@ -16,6 +16,7 @@ > #include "xfs_inode.h" > #include "xfs_inode_item.h" > #include "xfs_trace.h" > +#include "xfs_icache.h" > > /* > * Deferred Operations in XFS > @@ -583,8 +584,19 @@ xfs_defer_thaw( > struct xfs_defer_freezer *dff, > struct xfs_trans *tp) > { > + int i; > + > ASSERT(tp->t_flags & XFS_TRANS_PERM_LOG_RES); > > + /* Re-acquire the inode locks. */ > + for (i = 0; i < XFS_DEFER_FREEZER_INODES; i++) { > + if (!dff->dff_inodes[i]) > + break; > + > + dff->dff_ilocks[i] = XFS_ILOCK_EXCL; > + xfs_ilock(dff->dff_inodes[i], dff->dff_ilocks[i]); > + } > + > /* Add the dfops items to the transaction. */ > list_splice_init(&dff->dff_dfops, &tp->t_dfops); > tp->t_flags |= dff->dff_tpflags; > @@ -597,5 +609,43 @@ xfs_defer_freeezer_finish( > struct xfs_defer_freezer *dff) > { > xfs_defer_cancel_list(mp, &dff->dff_dfops); > + xfs_defer_freezer_irele(dff); > kmem_free(dff); > } > + > +/* > + * Attach an inode to this deferred ops freezer. Callers must hold ILOCK_EXCL, > + * which will be dropped and reacquired when we're ready to thaw the frozen > + * deferred ops. > + */ > +int > +xfs_defer_freezer_ijoin( > + struct xfs_defer_freezer *dff, > + struct xfs_inode *ip) > +{ > + unsigned int i; > + > + ASSERT(xfs_isilocked(ip, XFS_ILOCK_EXCL)); > + > + for (i = 0; i < XFS_DEFER_FREEZER_INODES; i++) { > + if (dff->dff_inodes[i] == ip) > + goto out; > + if (dff->dff_inodes[i] == NULL) > + break; > + } > + > + if (i == XFS_DEFER_FREEZER_INODES) { > + ASSERT(0); > + return -EFSCORRUPTED; > + } > + > + /* > + * Attach this inode to the freezer and drop its ILOCK because we > + * assume the caller will need to allocate a transaction. > + */ > + dff->dff_inodes[i] = ip; > + dff->dff_ilocks[i] = 0; > +out: > + xfs_iunlock(ip, XFS_ILOCK_EXCL); > + return 0; > +} > diff --git a/fs/xfs/libxfs/xfs_defer.h b/fs/xfs/libxfs/xfs_defer.h > index 7ae05e10d750..0052a0313283 100644 > --- a/fs/xfs/libxfs/xfs_defer.h > +++ b/fs/xfs/libxfs/xfs_defer.h > @@ -76,6 +76,11 @@ struct xfs_defer_freezer { > /* Deferred ops state saved from the transaction. */ > struct list_head dff_dfops; > unsigned int dff_tpflags; > + > + /* Inodes to hold when we want to finish the deferred work items. */ > +#define XFS_DEFER_FREEZER_INODES 2 > + unsigned int dff_ilocks[XFS_DEFER_FREEZER_INODES]; > + struct xfs_inode *dff_inodes[XFS_DEFER_FREEZER_INODES]; > }; > > /* Functions to freeze a chain of deferred operations for later. */ > @@ -83,5 +88,10 @@ int xfs_defer_freeze(struct xfs_trans *tp, struct xfs_defer_freezer **dffp); > void xfs_defer_thaw(struct xfs_defer_freezer *dff, struct xfs_trans *tp); > void xfs_defer_freeezer_finish(struct xfs_mount *mp, > struct xfs_defer_freezer *dff); > +int xfs_defer_freezer_ijoin(struct xfs_defer_freezer *dff, > + struct xfs_inode *ip); > + > +/* These functions must be provided by the xfs implementation. */ > +void xfs_defer_freezer_irele(struct xfs_defer_freezer *dff); > > #endif /* __XFS_DEFER_H__ */ > diff --git a/fs/xfs/xfs_bmap_item.c b/fs/xfs/xfs_bmap_item.c > index c733bdeeeb9b..bbce191d8fcd 100644 > --- a/fs/xfs/xfs_bmap_item.c > +++ b/fs/xfs/xfs_bmap_item.c > @@ -530,12 +530,13 @@ xfs_bui_item_recover( > } > > error = xlog_recover_trans_commit(tp, dffp); > - xfs_iunlock(ip, XFS_ILOCK_EXCL); > - xfs_irele(ip); > - return error; > + if (error) > + goto err_rele; > + return xfs_defer_freezer_ijoin(*dffp, ip); > > err_inode: > xfs_trans_cancel(tp); > +err_rele: > if (ip) { > xfs_iunlock(ip, XFS_ILOCK_EXCL); > xfs_irele(ip); > diff --git a/fs/xfs/xfs_icache.c b/fs/xfs/xfs_icache.c > index 17a0b86fe701..b96ddf5ff334 100644 > --- a/fs/xfs/xfs_icache.c > +++ b/fs/xfs/xfs_icache.c > @@ -12,6 +12,7 @@ > #include "xfs_sb.h" > #include "xfs_mount.h" > #include "xfs_inode.h" > +#include "xfs_defer.h" > #include "xfs_trans.h" > #include "xfs_trans_priv.h" > #include "xfs_inode_item.h" > @@ -1847,3 +1848,21 @@ xfs_start_block_reaping( > xfs_queue_eofblocks(mp); > xfs_queue_cowblocks(mp); > } > + > +/* Release all the inode resources attached to this freezer. */ > +void > +xfs_defer_freezer_irele( > + struct xfs_defer_freezer *dff) > +{ > + unsigned int i; > + > + for (i = 0; i < XFS_DEFER_FREEZER_INODES; i++) { > + if (!dff->dff_inodes[i]) > + break; > + > + if (dff->dff_ilocks[i]) > + xfs_iunlock(dff->dff_inodes[i], dff->dff_ilocks[i]); > + xfs_irele(dff->dff_inodes[i]); > + dff->dff_inodes[i] = NULL; > + } > +} >
