xfs: More robust inode extent count validation
authorDave Chinner <dchinner@redhat.com>
Fri, 22 Jun 2018 06:25:57 +0000 (23:25 -0700)
committerDarrick J. Wong <darrick.wong@oracle.com>
Fri, 22 Jun 2018 06:25:57 +0000 (23:25 -0700)
When the inode is in extent format, it can't have more extents that
fit in the inode fork. We don't currenty check this, and so this
corruption goes unnoticed by the inode verifiers. This can lead to
crashes operating on invalid in-memory structures.

Attempts to access such a inode will now error out in the verifier
rather than allowing modification operations to proceed.

Reported-by: Wen Xu <wen.xu@gatech.edu>
Signed-off-by: Dave Chinner <dchinner@redhat.com>
Reviewed-by: Darrick J. Wong <darrick.wong@oracle.com>
[darrick: fix a typedef, add some braces and breaks to shut up compiler warnings]
Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
fs/xfs/libxfs/xfs_format.h
fs/xfs/libxfs/xfs_inode_buf.c

index 1c5a8aa..7b4a43d 100644 (file)
@@ -962,6 +962,9 @@ typedef enum xfs_dinode_fmt {
                XFS_DFORK_DSIZE(dip, mp) : \
                XFS_DFORK_ASIZE(dip, mp))
 
+#define XFS_DFORK_MAXEXT(dip, mp, w) \
+       (XFS_DFORK_SIZE(dip, mp, w) / sizeof(struct xfs_bmbt_rec))
+
 /*
  * Return pointers to the data or attribute forks.
  */
index d38d724..33dc346 100644 (file)
@@ -374,6 +374,47 @@ xfs_log_dinode_to_disk(
        }
 }
 
+static xfs_failaddr_t
+xfs_dinode_verify_fork(
+       struct xfs_dinode       *dip,
+       struct xfs_mount        *mp,
+       int                     whichfork)
+{
+       uint32_t                di_nextents = XFS_DFORK_NEXTENTS(dip, whichfork);
+
+       switch (XFS_DFORK_FORMAT(dip, whichfork)) {
+       case XFS_DINODE_FMT_LOCAL:
+               /*
+                * no local regular files yet
+                */
+               if (whichfork == XFS_DATA_FORK) {
+                       if (S_ISREG(be16_to_cpu(dip->di_mode)))
+                               return __this_address;
+                       if (be64_to_cpu(dip->di_size) >
+                                       XFS_DFORK_SIZE(dip, mp, whichfork))
+                               return __this_address;
+               }
+               if (di_nextents)
+                       return __this_address;
+               break;
+       case XFS_DINODE_FMT_EXTENTS:
+               if (di_nextents > XFS_DFORK_MAXEXT(dip, mp, whichfork))
+                       return __this_address;
+               break;
+       case XFS_DINODE_FMT_BTREE:
+               if (whichfork == XFS_ATTR_FORK) {
+                       if (di_nextents > MAXAEXTNUM)
+                               return __this_address;
+               } else if (di_nextents > MAXEXTNUM) {
+                       return __this_address;
+               }
+               break;
+       default:
+               return __this_address;
+       }
+       return NULL;
+}
+
 xfs_failaddr_t
 xfs_dinode_verify(
        struct xfs_mount        *mp,
@@ -441,24 +482,9 @@ xfs_dinode_verify(
        case S_IFREG:
        case S_IFLNK:
        case S_IFDIR:
-               switch (dip->di_format) {
-               case XFS_DINODE_FMT_LOCAL:
-                       /*
-                        * no local regular files yet
-                        */
-                       if (S_ISREG(mode))
-                               return __this_address;
-                       if (di_size > XFS_DFORK_DSIZE(dip, mp))
-                               return __this_address;
-                       if (dip->di_nextents)
-                               return __this_address;
-                       /* fall through */
-               case XFS_DINODE_FMT_EXTENTS:
-               case XFS_DINODE_FMT_BTREE:
-                       break;
-               default:
-                       return __this_address;
-               }
+               fa = xfs_dinode_verify_fork(dip, mp, XFS_DATA_FORK);
+               if (fa)
+                       return fa;
                break;
        case 0:
                /* Uninitialized inode ok. */
@@ -468,17 +494,9 @@ xfs_dinode_verify(
        }
 
        if (XFS_DFORK_Q(dip)) {
-               switch (dip->di_aformat) {
-               case XFS_DINODE_FMT_LOCAL:
-                       if (dip->di_anextents)
-                               return __this_address;
-               /* fall through */
-               case XFS_DINODE_FMT_EXTENTS:
-               case XFS_DINODE_FMT_BTREE:
-                       break;
-               default:
-                       return __this_address;
-               }
+               fa = xfs_dinode_verify_fork(dip, mp, XFS_ATTR_FORK);
+               if (fa)
+                       return fa;
        } else {
                /*
                 * If there is no fork offset, this may be a freshly-made inode