xfs: remove filestream item xfs_inode reference

The filestreams allocator stores an xfs_fstrm_item structure in the MRU to
cache inode number to agno mappings for a particular length of time.  Each
xfs_fstrm_item contains the internal MRU structure, an inode pointer and
agno value.

The inode pointer stored in the xfs_fstrm_item is not referenced, however,
which means the inode itself can be removed and reclaimed before the MRU
item is freed. If this occurs, xfs_fstrm_free_func() can access freed or
unrelated memory through xfs_fstrm_item->ip and crash.

The obvious solution is to grab an inode reference for xfs_fstrm_item.
The filestream mechanism only actually uses the inode pointer as a means
to access the xfs_mount, however.  Rather than add unnecessary
complexity, simplify the implementation to store an xfs_mount pointer in
struct xfs_mru_cache, and pass it to the free callback.  This also
requires updates to the tracepoint class to provide the associated data
via parameters rather than the inode and a minor hack to peek at the MRU
key to establish the inode number at free time.

Based on debugging work and an earlier patch from Brian Foster, who
also wrote most of this changelog.

Reported-by: Brian Foster <bfoster@redhat.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>

1 files changed, 9 insertions, 10 deletions

diff --git a/fs/xfs/xfs_filestream.c b/fs/xfs/xfs_filestream.c
index 043ca3808ea2..5131a6e25fc9 100644
--- a/fs/xfs/xfs_filestream.c
+++ b/fs/xfs/xfs_filestream.c
@@ -34,7 +34,6 @@
 
 struct xfs_fstrm_item {
 	struct xfs_mru_cache_elem	mru;
-	struct xfs_inode		*ip;
 	xfs_agnumber_t			ag; /* AG in use for this directory */
 };
 
@@ -122,14 +121,15 @@ xfs_filestream_put_ag(
 
 static void
 xfs_fstrm_free_func(
+	void			*data,
 	struct xfs_mru_cache_elem *mru)
 {
+	struct xfs_mount	*mp = data;
 	struct xfs_fstrm_item	*item =
 		container_of(mru, struct xfs_fstrm_item, mru);
 
-	xfs_filestream_put_ag(item->ip->i_mount, item->ag);
-
-	trace_xfs_filestream_free(item->ip, item->ag);
+	xfs_filestream_put_ag(mp, item->ag);
+	trace_xfs_filestream_free(mp, mru->key, item->ag);
 
 	kmem_free(item);
 }
@@ -165,7 +165,7 @@ xfs_filestream_pick_ag(
 	trylock = XFS_ALLOC_FLAG_TRYLOCK;
 
 	for (nscan = 0; 1; nscan++) {
-		trace_xfs_filestream_scan(ip, ag);
+		trace_xfs_filestream_scan(mp, ip->i_ino, ag);
 
 		pag = xfs_perag_get(mp, ag);
 
@@ -265,7 +265,6 @@ next_ag:
 		goto out_put_ag;
 
 	item->ag = *agp;
-	item->ip = ip;
 
 	err = xfs_mru_cache_insert(mp->m_filestream, ip->i_ino, &item->mru);
 	if (err) {
@@ -333,7 +332,7 @@ xfs_filestream_lookup_ag(
 		ag = container_of(mru, struct xfs_fstrm_item, mru)->ag;
 		xfs_mru_cache_done(mp->m_filestream);
 
-		trace_xfs_filestream_lookup(ip, ag);
+		trace_xfs_filestream_lookup(mp, ip->i_ino, ag);
 		goto out;
 	}
 
@@ -399,7 +398,7 @@ xfs_filestream_new_ag(
 	 * Only free the item here so we skip over the old AG earlier.
 	 */
 	if (mru)
-		xfs_fstrm_free_func(mru);
+		xfs_fstrm_free_func(mp, mru);
 
 	IRELE(pip);
 exit:
@@ -426,8 +425,8 @@ xfs_filestream_mount(
 	 * timer tunable to within about 10 percent.  This requires at least 10
 	 * groups.
 	 */
-	return xfs_mru_cache_create(&mp->m_filestream, xfs_fstrm_centisecs * 10,
-				    10, xfs_fstrm_free_func);
+	return xfs_mru_cache_create(&mp->m_filestream, mp,
+			xfs_fstrm_centisecs * 10, 10, xfs_fstrm_free_func);
 }
 
 void