xfs: _{attr,data}_map_shared should take ILOCK_EXCL until iread_extents is completely...

While fuzzing the data fork extent count on a btree-format directory
with xfs/375, I observed the following (excerpted) splat:

XFS: Assertion failed: xfs_isilocked(ip, XFS_ILOCK_EXCL), file: fs/xfs/libxfs/xfs_bmap.c, line: 1208
------------[ cut here ]------------
WARNING: CPU: 0 PID: 43192 at fs/xfs/xfs_message.c:104 assfail+0x46/0x4a [xfs]
Call Trace:
 <TASK>
 xfs_iread_extents+0x1af/0x210 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xchk_dir_walk+0xb8/0x190 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xchk_parent_count_parent_dentries+0x41/0x80 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xchk_parent_validate+0x199/0x2e0 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xchk_parent+0xdf/0x130 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xfs_scrub_metadata+0x2b8/0x730 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xfs_scrubv_metadata+0x38b/0x4d0 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xfs_ioc_scrubv_metadata+0x111/0x160 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 xfs_file_ioctl+0x367/0xf50 [xfs 09f66509ece4938760fac7de64732a0cbd3e39cd]
 __x64_sys_ioctl+0x82/0xa0
 do_syscall_64+0x2b/0x80
 entry_SYSCALL_64_after_hwframe+0x46/0xb0

The cause of this is a race condition in xfs_ilock_data_map_shared,
which performs an unlocked access to the data fork to guess which lock
mode it needs:

Thread 0                          Thread 1

xfs_need_iread_extents
<observe no iext tree>
xfs_ilock(..., ILOCK_EXCL)
xfs_iread_extents
<observe no iext tree>
<check ILOCK_EXCL>
<load bmbt extents into iext>
<notice iext size doesn't
 match nextents>
                                  xfs_need_iread_extents
                                  <observe iext tree>
                                  xfs_ilock(..., ILOCK_SHARED)
<tear down iext tree>
xfs_iunlock(..., ILOCK_EXCL)
                                  xfs_iread_extents
                                  <observe no iext tree>
                                  <check ILOCK_EXCL>
                                  *BOOM*

Fix this race by adding a flag to the xfs_ifork structure to indicate
that we have not yet read in the extent records and changing the
predicate to look at the flag state, not if_height.  The memory barrier
ensures that the flag will not be set until the very end of the
function.

Signed-off-by: Darrick J. Wong <djwong@kernel.org>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
Signed-off-by: Dave Chinner <david@fromorbit.com>

diff --git a/fs/xfs/libxfs/xfs_bmap.c b/fs/xfs/libxfs/xfs_bmap.c
index 34de6e6..f11ef33 100644 (file)
--- a/fs/xfs/libxfs/xfs_bmap.c
+++ b/fs/xfs/libxfs/xfs_bmap.c
@@ -1171,6 +1171,12 @@ xfs_iread_extents(
                goto out;
        }
        ASSERT(ir.loaded == xfs_iext_count(ifp));
+       /*
+        * Use release semantics so that we can use acquire semantics in
+        * xfs_need_iread_extents and be guaranteed to see a valid mapping tree
+        * after that load.
+        */
+       smp_store_release(&ifp->if_needextents, 0);
        return 0;
 out:
        xfs_iext_destroy(ifp);

diff --git a/fs/xfs/libxfs/xfs_inode_fork.c b/fs/xfs/libxfs/xfs_inode_fork.c
index 6b21760..1bbe5ea 100644 (file)
--- a/fs/xfs/libxfs/xfs_inode_fork.c
+++ b/fs/xfs/libxfs/xfs_inode_fork.c
@@ -226,10 +226,15 @@ xfs_iformat_data_fork(
 
        /*
         * Initialize the extent count early, as the per-format routines may
-        * depend on it.
+        * depend on it.  Use release semantics to set needextents /after/ we
+        * set the format. This ensures that we can use acquire semantics on
+        * needextents in xfs_need_iread_extents() and be guaranteed to see a
+        * valid format value after that load.
         */
        ip->i_df.if_format = dip->di_format;
        ip->i_df.if_nextents = xfs_dfork_data_extents(dip);
+       smp_store_release(&ip->i_df.if_needextents,
+                          ip->i_df.if_format == XFS_DINODE_FMT_BTREE ? 1 : 0);
 
        switch (inode->i_mode & S_IFMT) {
        case S_IFIFO:
@@ -282,8 +287,17 @@ xfs_ifork_init_attr(
        enum xfs_dinode_fmt     format,
        xfs_extnum_t            nextents)
 {
+       /*
+        * Initialize the extent count early, as the per-format routines may
+        * depend on it.  Use release semantics to set needextents /after/ we
+        * set the format. This ensures that we can use acquire semantics on
+        * needextents in xfs_need_iread_extents() and be guaranteed to see a
+        * valid format value after that load.
+        */
        ip->i_af.if_format = format;
        ip->i_af.if_nextents = nextents;
+       smp_store_release(&ip->i_af.if_needextents,
+                          ip->i_af.if_format == XFS_DINODE_FMT_BTREE ? 1 : 0);
 }
 
 void

diff --git a/fs/xfs/libxfs/xfs_inode_fork.h b/fs/xfs/libxfs/xfs_inode_fork.h
index d3943d6..96d3077 100644 (file)
--- a/fs/xfs/libxfs/xfs_inode_fork.h
+++ b/fs/xfs/libxfs/xfs_inode_fork.h
@@ -24,6 +24,7 @@ struct xfs_ifork {
        xfs_extnum_t            if_nextents;    /* # of extents in this fork */
        short                   if_broot_bytes; /* bytes allocated for root */
        int8_t                  if_format;      /* format of this fork */
+       uint8_t                 if_needextents; /* extents have not been read */
 };
 
 /*
@@ -260,9 +261,10 @@ int xfs_iext_count_upgrade(struct xfs_trans *tp, struct xfs_inode *ip,
                uint nr_to_add);
 
 /* returns true if the fork has extents but they are not read in yet. */
-static inline bool xfs_need_iread_extents(struct xfs_ifork *ifp)
+static inline bool xfs_need_iread_extents(const struct xfs_ifork *ifp)
 {
-       return ifp->if_format == XFS_DINODE_FMT_BTREE && ifp->if_height == 0;
+       /* see xfs_iformat_{data,attr}_fork() for needextents semantics */
+       return smp_load_acquire(&ifp->if_needextents) != 0;
 }
 
 #endif /* __XFS_INODE_FORK_H__ */