xfs: set up scrub cross-referencing helpers
authorDarrick J. Wong <darrick.wong@oracle.com>
Wed, 17 Jan 2018 02:52:14 +0000 (18:52 -0800)
committerDarrick J. Wong <darrick.wong@oracle.com>
Thu, 18 Jan 2018 05:00:44 +0000 (21:00 -0800)
Create some helper functions that we'll use later to deal with problems
we might encounter while cross referencing metadata with other metadata.

Signed-off-by: Darrick J. Wong <darrick.wong@oracle.com>
Reviewed-by: Dave Chinner <dchinner@redhat.com>
fs/xfs/scrub/btree.c
fs/xfs/scrub/btree.h
fs/xfs/scrub/common.c
fs/xfs/scrub/common.h
fs/xfs/scrub/scrub.c
fs/xfs/scrub/trace.h

index df0766132acea690e0d883e87a02365539810168..4751ed03b9fba7c78e6c82f2796e6c8aa0fc3b0b 100644 (file)
  * Check for btree operation errors.  See the section about handling
  * operational errors in common.c.
  */
-bool
-xfs_scrub_btree_process_error(
+static bool
+__xfs_scrub_btree_process_error(
        struct xfs_scrub_context        *sc,
        struct xfs_btree_cur            *cur,
        int                             level,
-       int                             *error)
+       int                             *error,
+       __u32                           errflag,
+       void                            *ret_ip)
 {
        if (*error == 0)
                return true;
@@ -60,36 +62,80 @@ xfs_scrub_btree_process_error(
        case -EFSBADCRC:
        case -EFSCORRUPTED:
                /* Note the badness but don't abort. */
-               sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
+               sc->sm->sm_flags |= errflag;
                *error = 0;
                /* fall through */
        default:
                if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
                        trace_xfs_scrub_ifork_btree_op_error(sc, cur, level,
-                                       *error, __return_address);
+                                       *error, ret_ip);
                else
                        trace_xfs_scrub_btree_op_error(sc, cur, level,
-                                       *error, __return_address);
+                                       *error, ret_ip);
                break;
        }
        return false;
 }
 
+bool
+xfs_scrub_btree_process_error(
+       struct xfs_scrub_context        *sc,
+       struct xfs_btree_cur            *cur,
+       int                             level,
+       int                             *error)
+{
+       return __xfs_scrub_btree_process_error(sc, cur, level, error,
+                       XFS_SCRUB_OFLAG_CORRUPT, __return_address);
+}
+
+bool
+xfs_scrub_btree_xref_process_error(
+       struct xfs_scrub_context        *sc,
+       struct xfs_btree_cur            *cur,
+       int                             level,
+       int                             *error)
+{
+       return __xfs_scrub_btree_process_error(sc, cur, level, error,
+                       XFS_SCRUB_OFLAG_XFAIL, __return_address);
+}
+
 /* Record btree block corruption. */
-void
-xfs_scrub_btree_set_corrupt(
+static void
+__xfs_scrub_btree_set_corrupt(
        struct xfs_scrub_context        *sc,
        struct xfs_btree_cur            *cur,
-       int                             level)
+       int                             level,
+       __u32                           errflag,
+       void                            *ret_ip)
 {
-       sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
+       sc->sm->sm_flags |= errflag;
 
        if (cur->bc_flags & XFS_BTREE_ROOT_IN_INODE)
                trace_xfs_scrub_ifork_btree_error(sc, cur, level,
-                               __return_address);
+                               ret_ip);
        else
                trace_xfs_scrub_btree_error(sc, cur, level,
-                               __return_address);
+                               ret_ip);
+}
+
+void
+xfs_scrub_btree_set_corrupt(
+       struct xfs_scrub_context        *sc,
+       struct xfs_btree_cur            *cur,
+       int                             level)
+{
+       __xfs_scrub_btree_set_corrupt(sc, cur, level, XFS_SCRUB_OFLAG_CORRUPT,
+                       __return_address);
+}
+
+void
+xfs_scrub_btree_xref_set_corrupt(
+       struct xfs_scrub_context        *sc,
+       struct xfs_btree_cur            *cur,
+       int                             level)
+{
+       __xfs_scrub_btree_set_corrupt(sc, cur, level, XFS_SCRUB_OFLAG_XCORRUPT,
+                       __return_address);
 }
 
 /*
index 4de825a626d19404122ec9e527aef91a06a98560..e2b868ede70beb9174bafe9d2d6bf4874d47c5f3 100644 (file)
 bool xfs_scrub_btree_process_error(struct xfs_scrub_context *sc,
                struct xfs_btree_cur *cur, int level, int *error);
 
+/* Check for btree xref operation errors. */
+bool xfs_scrub_btree_xref_process_error(struct xfs_scrub_context *sc,
+                               struct xfs_btree_cur *cur, int level,
+                               int *error);
+
 /* Check for btree corruption. */
 void xfs_scrub_btree_set_corrupt(struct xfs_scrub_context *sc,
                struct xfs_btree_cur *cur, int level);
 
+/* Check for btree xref discrepancies. */
+void xfs_scrub_btree_xref_set_corrupt(struct xfs_scrub_context *sc,
+               struct xfs_btree_cur *cur, int level);
+
 struct xfs_scrub_btree;
 typedef int (*xfs_scrub_btree_rec_fn)(
        struct xfs_scrub_btree  *bs,
index d5c37d8d2fe66a84f6316b94fe5612e6f08e1b28..68fea09cd673a0a4d8931abd3498b546674a3d12 100644 (file)
  */
 
 /* Check for operational errors. */
-bool
-xfs_scrub_process_error(
+static bool
+__xfs_scrub_process_error(
        struct xfs_scrub_context        *sc,
        xfs_agnumber_t                  agno,
        xfs_agblock_t                   bno,
-       int                             *error)
+       int                             *error,
+       __u32                           errflag,
+       void                            *ret_ip)
 {
        switch (*error) {
        case 0:
@@ -95,24 +97,48 @@ xfs_scrub_process_error(
        case -EFSBADCRC:
        case -EFSCORRUPTED:
                /* Note the badness but don't abort. */
-               sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
+               sc->sm->sm_flags |= errflag;
                *error = 0;
                /* fall through */
        default:
                trace_xfs_scrub_op_error(sc, agno, bno, *error,
-                               __return_address);
+                               ret_ip);
                break;
        }
        return false;
 }
 
-/* Check for operational errors for a file offset. */
 bool
-xfs_scrub_fblock_process_error(
+xfs_scrub_process_error(
+       struct xfs_scrub_context        *sc,
+       xfs_agnumber_t                  agno,
+       xfs_agblock_t                   bno,
+       int                             *error)
+{
+       return __xfs_scrub_process_error(sc, agno, bno, error,
+                       XFS_SCRUB_OFLAG_CORRUPT, __return_address);
+}
+
+bool
+xfs_scrub_xref_process_error(
+       struct xfs_scrub_context        *sc,
+       xfs_agnumber_t                  agno,
+       xfs_agblock_t                   bno,
+       int                             *error)
+{
+       return __xfs_scrub_process_error(sc, agno, bno, error,
+                       XFS_SCRUB_OFLAG_XFAIL, __return_address);
+}
+
+/* Check for operational errors for a file offset. */
+static bool
+__xfs_scrub_fblock_process_error(
        struct xfs_scrub_context        *sc,
        int                             whichfork,
        xfs_fileoff_t                   offset,
-       int                             *error)
+       int                             *error,
+       __u32                           errflag,
+       void                            *ret_ip)
 {
        switch (*error) {
        case 0:
@@ -124,17 +150,39 @@ xfs_scrub_fblock_process_error(
        case -EFSBADCRC:
        case -EFSCORRUPTED:
                /* Note the badness but don't abort. */
-               sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
+               sc->sm->sm_flags |= errflag;
                *error = 0;
                /* fall through */
        default:
                trace_xfs_scrub_file_op_error(sc, whichfork, offset, *error,
-                               __return_address);
+                               ret_ip);
                break;
        }
        return false;
 }
 
+bool
+xfs_scrub_fblock_process_error(
+       struct xfs_scrub_context        *sc,
+       int                             whichfork,
+       xfs_fileoff_t                   offset,
+       int                             *error)
+{
+       return __xfs_scrub_fblock_process_error(sc, whichfork, offset, error,
+                       XFS_SCRUB_OFLAG_CORRUPT, __return_address);
+}
+
+bool
+xfs_scrub_fblock_xref_process_error(
+       struct xfs_scrub_context        *sc,
+       int                             whichfork,
+       xfs_fileoff_t                   offset,
+       int                             *error)
+{
+       return __xfs_scrub_fblock_process_error(sc, whichfork, offset, error,
+                       XFS_SCRUB_OFLAG_XFAIL, __return_address);
+}
+
 /*
  * Handling scrub corruption/optimization/warning checks.
  *
@@ -183,6 +231,16 @@ xfs_scrub_block_set_corrupt(
        trace_xfs_scrub_block_error(sc, bp->b_bn, __return_address);
 }
 
+/* Record a corruption while cross-referencing. */
+void
+xfs_scrub_block_xref_set_corrupt(
+       struct xfs_scrub_context        *sc,
+       struct xfs_buf                  *bp)
+{
+       sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
+       trace_xfs_scrub_block_error(sc, bp->b_bn, __return_address);
+}
+
 /*
  * Record a corrupt inode.  The trace data will include the block given
  * by bp if bp is given; otherwise it will use the block location of the
@@ -198,6 +256,17 @@ xfs_scrub_ino_set_corrupt(
        trace_xfs_scrub_ino_error(sc, ino, bp ? bp->b_bn : 0, __return_address);
 }
 
+/* Record a corruption while cross-referencing with an inode. */
+void
+xfs_scrub_ino_xref_set_corrupt(
+       struct xfs_scrub_context        *sc,
+       xfs_ino_t                       ino,
+       struct xfs_buf                  *bp)
+{
+       sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
+       trace_xfs_scrub_ino_error(sc, ino, bp ? bp->b_bn : 0, __return_address);
+}
+
 /* Record corruption in a block indexed by a file fork. */
 void
 xfs_scrub_fblock_set_corrupt(
@@ -209,6 +278,17 @@ xfs_scrub_fblock_set_corrupt(
        trace_xfs_scrub_fblock_error(sc, whichfork, offset, __return_address);
 }
 
+/* Record a corruption while cross-referencing a fork block. */
+void
+xfs_scrub_fblock_xref_set_corrupt(
+       struct xfs_scrub_context        *sc,
+       int                             whichfork,
+       xfs_fileoff_t                   offset)
+{
+       sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XCORRUPT;
+       trace_xfs_scrub_fblock_error(sc, whichfork, offset, __return_address);
+}
+
 /*
  * Warn about inodes that need administrative review but is not
  * incorrect.
@@ -588,3 +668,38 @@ out:
        /* scrub teardown will unlock and release the inode for us */
        return error;
 }
+
+/*
+ * Predicate that decides if we need to evaluate the cross-reference check.
+ * If there was an error accessing the cross-reference btree, just delete
+ * the cursor and skip the check.
+ */
+bool
+xfs_scrub_should_check_xref(
+       struct xfs_scrub_context        *sc,
+       int                             *error,
+       struct xfs_btree_cur            **curpp)
+{
+       if (*error == 0)
+               return true;
+
+       if (curpp) {
+               /* If we've already given up on xref, just bail out. */
+               if (!*curpp)
+                       return false;
+
+               /* xref error, delete cursor and bail out. */
+               xfs_btree_del_cursor(*curpp, XFS_BTREE_ERROR);
+               *curpp = NULL;
+       }
+
+       sc->sm->sm_flags |= XFS_SCRUB_OFLAG_XFAIL;
+       trace_xfs_scrub_xref_error(sc, *error, __return_address);
+
+       /*
+        * Errors encountered during cross-referencing with another
+        * data structure should not cause this scrubber to abort.
+        */
+       *error = 0;
+       return false;
+}
index fe12053aa0e7740a5886e592076dadb8fef333dd..84c302f1d63498fcab8618cbe061e4e7a5a244f7 100644 (file)
@@ -56,6 +56,11 @@ bool xfs_scrub_process_error(struct xfs_scrub_context *sc, xfs_agnumber_t agno,
 bool xfs_scrub_fblock_process_error(struct xfs_scrub_context *sc, int whichfork,
                xfs_fileoff_t offset, int *error);
 
+bool xfs_scrub_xref_process_error(struct xfs_scrub_context *sc,
+               xfs_agnumber_t agno, xfs_agblock_t bno, int *error);
+bool xfs_scrub_fblock_xref_process_error(struct xfs_scrub_context *sc,
+               int whichfork, xfs_fileoff_t offset, int *error);
+
 void xfs_scrub_block_set_preen(struct xfs_scrub_context *sc,
                struct xfs_buf *bp);
 void xfs_scrub_ino_set_preen(struct xfs_scrub_context *sc, xfs_ino_t ino,
@@ -68,6 +73,13 @@ void xfs_scrub_ino_set_corrupt(struct xfs_scrub_context *sc, xfs_ino_t ino,
 void xfs_scrub_fblock_set_corrupt(struct xfs_scrub_context *sc, int whichfork,
                xfs_fileoff_t offset);
 
+void xfs_scrub_block_xref_set_corrupt(struct xfs_scrub_context *sc,
+               struct xfs_buf *bp);
+void xfs_scrub_ino_xref_set_corrupt(struct xfs_scrub_context *sc, xfs_ino_t ino,
+               struct xfs_buf *bp);
+void xfs_scrub_fblock_xref_set_corrupt(struct xfs_scrub_context *sc,
+               int whichfork, xfs_fileoff_t offset);
+
 void xfs_scrub_ino_set_warning(struct xfs_scrub_context *sc, xfs_ino_t ino,
                struct xfs_buf *bp);
 void xfs_scrub_fblock_set_warning(struct xfs_scrub_context *sc, int whichfork,
@@ -76,6 +88,10 @@ void xfs_scrub_fblock_set_warning(struct xfs_scrub_context *sc, int whichfork,
 void xfs_scrub_set_incomplete(struct xfs_scrub_context *sc);
 int xfs_scrub_checkpoint_log(struct xfs_mount *mp);
 
+/* Are we set up for a cross-referencing check? */
+bool xfs_scrub_should_check_xref(struct xfs_scrub_context *sc, int *error,
+                          struct xfs_btree_cur **curpp);
+
 /* Setup functions */
 int xfs_scrub_setup_fs(struct xfs_scrub_context *sc, struct xfs_inode *ip);
 int xfs_scrub_setup_ag_allocbt(struct xfs_scrub_context *sc,
index cd4607782a19bbd155bb1b0763dcc9da71eb6530..0ed2a123cbb855ce1aed1dcdd153453ed01abea2 100644 (file)
  * structure itself is corrupt, the CORRUPT flag will be set.  If
  * the metadata is correct but otherwise suboptimal, the PREEN flag
  * will be set.
+ *
+ * We perform secondary validation of filesystem metadata by
+ * cross-referencing every record with all other available metadata.
+ * For example, for block mapping extents, we verify that there are no
+ * records in the free space and inode btrees corresponding to that
+ * space extent and that there is a corresponding entry in the reverse
+ * mapping btree.  Inconsistent metadata is noted by setting the
+ * XCORRUPT flag; btree query function errors are noted by setting the
+ * XFAIL flag and deleting the cursor to prevent further attempts to
+ * cross-reference with a defective btree.
  */
 
 /*
index ffa4a70ebdcc9e07aad7d5d2f915fbd3e0780ecd..a0a6d3cd131a48fed3366a3ee65828d78f32d178 100644 (file)
@@ -491,6 +491,28 @@ DEFINE_EVENT(xfs_scrub_sbtree_class, name, \
 DEFINE_SCRUB_SBTREE_EVENT(xfs_scrub_btree_rec);
 DEFINE_SCRUB_SBTREE_EVENT(xfs_scrub_btree_key);
 
+TRACE_EVENT(xfs_scrub_xref_error,
+       TP_PROTO(struct xfs_scrub_context *sc, int error, void *ret_ip),
+       TP_ARGS(sc, error, ret_ip),
+       TP_STRUCT__entry(
+               __field(dev_t, dev)
+               __field(int, type)
+               __field(int, error)
+               __field(void *, ret_ip)
+       ),
+       TP_fast_assign(
+               __entry->dev = sc->mp->m_super->s_dev;
+               __entry->type = sc->sm->sm_type;
+               __entry->error = error;
+               __entry->ret_ip = ret_ip;
+       ),
+       TP_printk("dev %d:%d type %u xref error %d ret_ip %pF",
+                 MAJOR(__entry->dev), MINOR(__entry->dev),
+                 __entry->type,
+                 __entry->error,
+                 __entry->ret_ip)
+);
+
 #endif /* _TRACE_XFS_SCRUB_TRACE_H */
 
 #undef TRACE_INCLUDE_PATH