xfs: online scrub support

index 3422711..43005fb 100644
@@ -610,7 +610,7 @@ xfs_iget(
 	} else {
 		rcu_read_unlock();
 		if (flags & XFS_IGET_INCORE) {
-			error = -ENOENT;
+			error = -ENODATA;
 			goto out_error_or_again;
 		}
 		XFS_STATS_INC(mp, xs_ig_missed);

index 744dcae..bd3a943 100644
@@ -2923,3 +2923,52 @@ xfs_alloc_query_all(
 	query.fn = fn;
 	return xfs_btree_query_all(cur, xfs_alloc_query_range_helper, &query);
 }
+
+/* Find the size of the AG, in blocks. */
+xfs_agblock_t
+xfs_ag_block_count(
+	struct xfs_mount	*mp,
+	xfs_agnumber_t		agno)
+{
+	ASSERT(agno < mp->m_sb.sb_agcount);
+
+	if (agno < mp->m_sb.sb_agcount - 1)
+		return mp->m_sb.sb_agblocks;
+	return mp->m_sb.sb_dblocks - (agno * mp->m_sb.sb_agblocks);
+}
+
+/*
+ * Verify that an AG block number pointer neither points outside the AG
+ * nor points at static metadata.
+ */
+bool
+xfs_verify_agbno_ptr(
+	struct xfs_mount	*mp,
+	xfs_agnumber_t		agno,
+	xfs_agblock_t		agbno)
+{
+	xfs_agblock_t		eoag;
+
+	eoag = xfs_ag_block_count(mp, agno);
+	if (agbno >= eoag)
+		return false;
+	if (agbno <= XFS_AGFL_BLOCK(mp))
+		return false;
+	return true;
+}
+
+/*
+ * Verify that an FS block number pointer neither points outside the
+ * filesystem nor points at static AG metadata.
+ */
+bool
+xfs_verify_fsbno_ptr(
+	struct xfs_mount	*mp,
+	xfs_fsblock_t		fsbno)
+{
+	xfs_agnumber_t		agno = XFS_FSB_TO_AGNO(mp, fsbno);
+
+	if (agno >= mp->m_sb.sb_agcount)
+		return false;
+	return xfs_verify_agbno_ptr(mp, agno, XFS_FSB_TO_AGBNO(mp, fsbno));
+}

index ef26edc..3185807 100644
@@ -232,5 +232,9 @@ int xfs_alloc_query_range(struct xfs_btree_cur *cur,
 		xfs_alloc_query_range_fn fn, void *priv);
 int xfs_alloc_query_all(struct xfs_btree_cur *cur, xfs_alloc_query_range_fn fn,
 		void *priv);
+xfs_agblock_t xfs_ag_block_count(struct xfs_mount *mp, xfs_agnumber_t agno);
+bool xfs_verify_agbno_ptr(struct xfs_mount *mp, xfs_agnumber_t agno,
+		xfs_agblock_t agbno);
+bool xfs_verify_fsbno_ptr(struct xfs_mount *mp, xfs_fsblock_t fsbno);
 
 #endif	/* __XFS_ALLOC_H__ */

index 5d4e43e..0a49348 100644
@@ -1086,3 +1086,15 @@ xfs_rtalloc_query_all(
 
 	return xfs_rtalloc_query_range(tp, &keys[0], &keys[1], fn, priv);
 }
+
+/*
+ * Verify that an realtime block number pointer doesn't point off the
+ * end of the realtime device.
+ */
+bool
+xfs_verify_rtbno_ptr(
+	struct xfs_mount	*mp,
+	xfs_rtblock_t		rtbno)
+{
+	return rtbno < mp->m_sb.sb_rblocks;
+}

index 79defa7..11b8554 100644
@@ -138,6 +138,7 @@ int xfs_rtalloc_query_range(struct xfs_trans *tp,
 int xfs_rtalloc_query_all(struct xfs_trans *tp,
 			  xfs_rtalloc_query_range_fn fn,
 			  void *priv);
+bool xfs_verify_rtbno_ptr(struct xfs_mount *mp, xfs_rtblock_t rtbno);
 #else
 # define xfs_rtallocate_extent(t,b,min,max,l,f,p,rb)    (ENOSYS)
 # define xfs_rtfree_extent(t,b,l)                       (ENOSYS)

@@ -146,6 +147,7 @@ int xfs_rtalloc_query_all(struct xfs_trans *tp,
 # define xfs_rtalloc_query_range(t,l,h,f,p)             (ENOSYS)
 # define xfs_rtalloc_query_all(t,f,p)                   (ENOSYS)
 # define xfs_rtbuf_get(m,t,b,i,p)                       (ENOSYS)
+# define xfs_verify_rtbno_ptr(m, r)			(false)
 static inline int		/* error */
 xfs_rtmount_init(
 	xfs_mount_t	*mp)	/* file system mount structure */

index 044a363..b4cbd1a 100644
@@ -657,8 +657,8 @@ xfs_bmap_btree_to_extents(
 	cbno = be64_to_cpu(*pp);
 	*logflagsp = 0;
 #ifdef DEBUG
-	if ((error = xfs_btree_check_lptr(cur, cbno, 1)))
-		return error;
+	XFS_WANT_CORRUPTED_RETURN(cur->bc_mp,
+			xfs_btree_check_lptr(cur, cbno, 1));
 #endif
 	error = xfs_btree_read_bufl(mp, tp, cbno, 0, &cbp, XFS_BMAP_BTREE_REF,
 				&xfs_bmbt_buf_ops);

91 lines not on this page · raw message

index f2a88c3..8f52eda 100644
@@ -269,10 +269,19 @@ xfs_btree_check_block(
 /*
  * Check that (long) pointer is ok.
  */
-int					/* error (0 or EFSCORRUPTED) */
+bool					/* error (0 or EFSCORRUPTED) */
 xfs_btree_check_lptr(
 	struct xfs_btree_cur	*cur,	/* btree cursor */
-	xfs_fsblock_t		ptr,	/* btree block disk address */
+	xfs_fsblock_t		fsbno,	/* btree block disk address */
+	int			level);	/* btree block level */
+
+/*
+ * Check that (short) pointer is ok.
+ */
+bool					/* error (0 or EFSCORRUPTED) */
+xfs_btree_check_sptr(
+	struct xfs_btree_cur	*cur,	/* btree cursor */
+	xfs_agblock_t		agbno,	/* btree block disk address */
 	int			level);	/* btree block level */
 
 /*

index e7e033a..2266a5a 100644
@@ -63,44 +63,61 @@ xfs_btree_magic(
 	return magic;
 }
 
-STATIC int				/* error (0 or EFSCORRUPTED) */
-xfs_btree_check_lblock(
-	struct xfs_btree_cur	*cur,	/* btree cursor */
-	struct xfs_btree_block	*block,	/* btree long form block pointer */
-	int			level,	/* level of the btree block */
-	struct xfs_buf		*bp)	/* buffer for block, if any */
+/*
+ * Check a long btree block header.  Return the address of the failing check,
+ * or NULL if everything is ok.
+ */
+void *
+__xfs_btree_check_lblock(
+	struct xfs_btree_cur	*cur,
+	struct xfs_btree_block	*block,
+	int			level,
+	struct xfs_buf		*bp)
 {
-	int			lblock_ok = 1; /* block passes checks */
-	struct xfs_mount	*mp;	/* file system mount point */
+	struct xfs_mount	*mp = cur->bc_mp;
 	xfs_btnum_t		btnum = cur->bc_btnum;
-	int			crc;
-
-	mp = cur->bc_mp;
-	crc = xfs_sb_version_hascrc(&mp->m_sb);
+	int			crc = xfs_sb_version_hascrc(&mp->m_sb);
 
 	if (crc) {
-		lblock_ok = lblock_ok &&
-			uuid_equal(&block->bb_u.l.bb_uuid,
-				   &mp->m_sb.sb_meta_uuid) &&
-			block->bb_u.l.bb_blkno == cpu_to_be64(
-				bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
+		if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
+			return __this_address;
+		if (block->bb_u.l.bb_blkno !=
+		    cpu_to_be64(bp ? bp->b_bn : XFS_BUF_DADDR_NULL))
+			return __this_address;
 	}
 
-	lblock_ok = lblock_ok &&
-		be32_to_cpu(block->bb_magic) == xfs_btree_magic(crc, btnum) &&
-		be16_to_cpu(block->bb_level) == level &&
-		be16_to_cpu(block->bb_numrecs) <=
-			cur->bc_ops->get_maxrecs(cur, level) &&
-		block->bb_u.l.bb_leftsib &&
-		(block->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK) ||
-		 XFS_FSB_SANITY_CHECK(mp,
-			be64_to_cpu(block->bb_u.l.bb_leftsib))) &&
-		block->bb_u.l.bb_rightsib &&
-		(block->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK) ||
-		 XFS_FSB_SANITY_CHECK(mp,
-			be64_to_cpu(block->bb_u.l.bb_rightsib)));
-
-	if (unlikely(XFS_TEST_ERROR(!lblock_ok, mp,
+	if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum))
+		return __this_address;
+	if (be16_to_cpu(block->bb_level) != level)
+		return __this_address;
+	if (be16_to_cpu(block->bb_numrecs) >
+	    cur->bc_ops->get_maxrecs(cur, level))
+		return __this_address;
+	if (block->bb_u.l.bb_leftsib != cpu_to_be64(NULLFSBLOCK) &&
+	    !xfs_btree_check_lptr(cur, be64_to_cpu(block->bb_u.l.bb_leftsib),
+			level + 1))
+		return __this_address;
+	if (block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK) &&
+	    !xfs_btree_check_lptr(cur, be64_to_cpu(block->bb_u.l.bb_rightsib),
+			level + 1))
+		return __this_address;
+
+	return NULL;
+}
+
+/* Check a long btree block header. */
+int
+xfs_btree_check_lblock(
+	struct xfs_btree_cur	*cur,
+	struct xfs_btree_block	*block,
+	int			level,
+	struct xfs_buf		*bp)
+{
+	struct xfs_mount	*mp = cur->bc_mp;
+	void			*failed_at;
+
+	failed_at = __xfs_btree_check_lblock(cur, block, level, bp);
+	if (unlikely(XFS_TEST_ERROR(failed_at != NULL, mp,
 			XFS_ERRTAG_BTREE_CHECK_LBLOCK))) {
 		if (bp)
 			trace_xfs_btree_corrupt(bp, _RET_IP_);

@@ -110,48 +127,61 @@ xfs_btree_check_lblock(
 	return 0;
 }
 
-STATIC int				/* error (0 or EFSCORRUPTED) */
-xfs_btree_check_sblock(
-	struct xfs_btree_cur	*cur,	/* btree cursor */
-	struct xfs_btree_block	*block,	/* btree short form block pointer */
-	int			level,	/* level of the btree block */
-	struct xfs_buf		*bp)	/* buffer containing block */
+/*
+ * Check a short btree block header.  Return the address of the failing check,
+ * or NULL if everything is ok.
+ */
+void *
+__xfs_btree_check_sblock(
+	struct xfs_btree_cur	*cur,
+	struct xfs_btree_block	*block,
+	int			level,
+	struct xfs_buf		*bp)
 {
-	struct xfs_mount	*mp;	/* file system mount point */
-	struct xfs_buf		*agbp;	/* buffer for ag. freespace struct */
-	struct xfs_agf		*agf;	/* ag. freespace structure */
-	xfs_agblock_t		agflen;	/* native ag. freespace length */
-	int			sblock_ok = 1; /* block passes checks */
+	struct xfs_mount	*mp = cur->bc_mp;
 	xfs_btnum_t		btnum = cur->bc_btnum;
-	int			crc;
-
-	mp = cur->bc_mp;
-	crc = xfs_sb_version_hascrc(&mp->m_sb);
-	agbp = cur->bc_private.a.agbp;
-	agf = XFS_BUF_TO_AGF(agbp);
-	agflen = be32_to_cpu(agf->agf_length);
+	int			crc = xfs_sb_version_hascrc(&mp->m_sb);
 
 	if (crc) {
-		sblock_ok = sblock_ok &&
-			uuid_equal(&block->bb_u.s.bb_uuid,
-				   &mp->m_sb.sb_meta_uuid) &&
-			block->bb_u.s.bb_blkno == cpu_to_be64(
-				bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
+		if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
+			return __this_address;
+		if (block->bb_u.s.bb_blkno !=
+		    cpu_to_be64(bp ? bp->b_bn : XFS_BUF_DADDR_NULL))
+			return __this_address;
 	}
 
-	sblock_ok = sblock_ok &&
-		be32_to_cpu(block->bb_magic) == xfs_btree_magic(crc, btnum) &&
-		be16_to_cpu(block->bb_level) == level &&
-		be16_to_cpu(block->bb_numrecs) <=
-			cur->bc_ops->get_maxrecs(cur, level) &&
-		(block->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) ||
-		 be32_to_cpu(block->bb_u.s.bb_leftsib) < agflen) &&
-		block->bb_u.s.bb_leftsib &&
-		(block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK) ||
-		 be32_to_cpu(block->bb_u.s.bb_rightsib) < agflen) &&
-		block->bb_u.s.bb_rightsib;
-
-	if (unlikely(XFS_TEST_ERROR(!sblock_ok, mp,
+	if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum))
+		return __this_address;
+	if (be16_to_cpu(block->bb_level) != level)
+		return __this_address;
+	if (be16_to_cpu(block->bb_numrecs) >
+	    cur->bc_ops->get_maxrecs(cur, level))
+		return __this_address;
+	if (block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK) &&
+	    !xfs_btree_check_sptr(cur, be32_to_cpu(block->bb_u.s.bb_leftsib),
+			level + 1))
+		return __this_address;
+	if (block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK) &&
+	    !xfs_btree_check_sptr(cur, be32_to_cpu(block->bb_u.s.bb_rightsib),
+			level + 1))
+		return __this_address;
+
+	return NULL;
+}
+
+/* Check a short btree block header. */
+STATIC int
+xfs_btree_check_sblock(
+	struct xfs_btree_cur	*cur,
+	struct xfs_btree_block	*block,
+	int			level,
+	struct xfs_buf		*bp)
+{
+	struct xfs_mount	*mp = cur->bc_mp;
+	void			*failed_at;
+
+	failed_at = __xfs_btree_check_sblock(cur, block, level, bp);
+	if (unlikely(XFS_TEST_ERROR(failed_at != NULL, mp,
 			XFS_ERRTAG_BTREE_CHECK_SBLOCK))) {
 		if (bp)
 			trace_xfs_btree_corrupt(bp, _RET_IP_);

index 8f52eda..baf7064 100644
@@ -255,6 +255,11 @@ typedef struct xfs_btree_cur
  */
 #define	XFS_BUF_TO_BLOCK(bp)	((struct xfs_btree_block *)((bp)->b_addr))
 
+/* Internal long and short btree block checks. */
+void *__xfs_btree_check_lblock(struct xfs_btree_cur *cur,
+		struct xfs_btree_block *block, int level, struct xfs_buf *bp);
+void *__xfs_btree_check_sblock(struct xfs_btree_cur *cur,
+		struct xfs_btree_block *block, int level, struct xfs_buf *bp);
 
 /*
  * Check that block header is ok.

index dcd1292..b825953 100644
@@ -142,6 +142,13 @@ typedef __u32			xfs_nlink_t;
 #define SYNCHRONIZE()	barrier()
 #define __return_address __builtin_return_address(0)
 
+/*
+ * Return the address of a label.  Use asm volatile so that the optimizer
+ * won't try anything stupid like refactoring the error jumpouts into a
+ * single return, which throws off the reported address.
+ */
+#define __this_address  ({ __label__ __here; __here: asm volatile(""); &&__here; })
+
 #define XFS_PROJID_DEFAULT	0
 
 #define MIN(a,b)	(min(a,b))

index ae19f24..67a5b0f 100644
@@ -63,44 +63,61 @@ xfs_btree_magic(
 	return magic;
 }
 
-STATIC int				/* error (0 or EFSCORRUPTED) */
-xfs_btree_check_lblock(
-	struct xfs_btree_cur	*cur,	/* btree cursor */
-	struct xfs_btree_block	*block,	/* btree long form block pointer */
-	int			level,	/* level of the btree block */
-	struct xfs_buf		*bp)	/* buffer for block, if any */
+/*
+ * Check a long btree block header.  Return the address of the failing check,
+ * or NULL if everything is ok.
+ */
+void *
+__xfs_btree_check_lblock(
+	struct xfs_btree_cur	*cur,
+	struct xfs_btree_block	*block,
+	int			level,
+	struct xfs_buf		*bp)
 {
-	int			lblock_ok = 1; /* block passes checks */
-	struct xfs_mount	*mp;	/* file system mount point */
+	struct xfs_mount	*mp = cur->bc_mp;
 	xfs_btnum_t		btnum = cur->bc_btnum;
-	int			crc;
-
-	mp = cur->bc_mp;
-	crc = xfs_sb_version_hascrc(&mp->m_sb);
+	int			crc = xfs_sb_version_hascrc(&mp->m_sb);
 
 	if (crc) {
-		lblock_ok = lblock_ok &&
-			uuid_equal(&block->bb_u.l.bb_uuid,
-				   &mp->m_sb.sb_meta_uuid) &&
-			block->bb_u.l.bb_blkno == cpu_to_be64(
-				bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
+		if (!uuid_equal(&block->bb_u.l.bb_uuid, &mp->m_sb.sb_meta_uuid))
+			return __this_address;
+		if (block->bb_u.l.bb_blkno !=
+		    cpu_to_be64(bp ? bp->b_bn : XFS_BUF_DADDR_NULL))
+			return __this_address;
 	}
 
-	lblock_ok = lblock_ok &&
-		be32_to_cpu(block->bb_magic) == xfs_btree_magic(crc, btnum) &&
-		be16_to_cpu(block->bb_level) == level &&
-		be16_to_cpu(block->bb_numrecs) <=
-			cur->bc_ops->get_maxrecs(cur, level) &&
-		block->bb_u.l.bb_leftsib &&
-		(block->bb_u.l.bb_leftsib == cpu_to_be64(NULLFSBLOCK) ||
-		 XFS_FSB_SANITY_CHECK(mp,
-			be64_to_cpu(block->bb_u.l.bb_leftsib))) &&
-		block->bb_u.l.bb_rightsib &&
-		(block->bb_u.l.bb_rightsib == cpu_to_be64(NULLFSBLOCK) ||
-		 XFS_FSB_SANITY_CHECK(mp,
-			be64_to_cpu(block->bb_u.l.bb_rightsib)));
-
-	if (unlikely(XFS_TEST_ERROR(!lblock_ok, mp,
+	if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum))
+		return __this_address;
+	if (be16_to_cpu(block->bb_level) != level)
+		return __this_address;
+	if (be16_to_cpu(block->bb_numrecs) >
+	    cur->bc_ops->get_maxrecs(cur, level))
+		return __this_address;
+	if (block->bb_u.l.bb_leftsib != cpu_to_be64(NULLFSBLOCK) &&
+	    !xfs_btree_check_lptr(cur, be64_to_cpu(block->bb_u.l.bb_leftsib),
+			level + 1))
+		return __this_address;
+	if (block->bb_u.l.bb_rightsib != cpu_to_be64(NULLFSBLOCK) &&
+	    !xfs_btree_check_lptr(cur, be64_to_cpu(block->bb_u.l.bb_rightsib),
+			level + 1))
+		return __this_address;
+
+	return NULL;
+}
+
+/* Check a long btree block header. */
+int
+xfs_btree_check_lblock(
+	struct xfs_btree_cur	*cur,
+	struct xfs_btree_block	*block,
+	int			level,
+	struct xfs_buf		*bp)
+{
+	struct xfs_mount	*mp = cur->bc_mp;
+	void			*failed_at;
+
+	failed_at = __xfs_btree_check_lblock(cur, block, level, bp);
+	if (unlikely(XFS_TEST_ERROR(failed_at != NULL, mp,
 			XFS_ERRTAG_BTREE_CHECK_LBLOCK))) {
 		if (bp)
 			trace_xfs_btree_corrupt(bp, _RET_IP_);

@@ -110,48 +127,61 @@ xfs_btree_check_lblock(
 	return 0;
 }
 
-STATIC int				/* error (0 or EFSCORRUPTED) */
-xfs_btree_check_sblock(
-	struct xfs_btree_cur	*cur,	/* btree cursor */
-	struct xfs_btree_block	*block,	/* btree short form block pointer */
-	int			level,	/* level of the btree block */
-	struct xfs_buf		*bp)	/* buffer containing block */
+/*
+ * Check a short btree block header.  Return the address of the failing check,
+ * or NULL if everything is ok.
+ */
+void *
+__xfs_btree_check_sblock(
+	struct xfs_btree_cur	*cur,
+	struct xfs_btree_block	*block,
+	int			level,
+	struct xfs_buf		*bp)
 {
-	struct xfs_mount	*mp;	/* file system mount point */
-	struct xfs_buf		*agbp;	/* buffer for ag. freespace struct */
-	struct xfs_agf		*agf;	/* ag. freespace structure */
-	xfs_agblock_t		agflen;	/* native ag. freespace length */
-	int			sblock_ok = 1; /* block passes checks */
+	struct xfs_mount	*mp = cur->bc_mp;
 	xfs_btnum_t		btnum = cur->bc_btnum;
-	int			crc;
-
-	mp = cur->bc_mp;
-	crc = xfs_sb_version_hascrc(&mp->m_sb);
-	agbp = cur->bc_private.a.agbp;
-	agf = XFS_BUF_TO_AGF(agbp);
-	agflen = be32_to_cpu(agf->agf_length);
+	int			crc = xfs_sb_version_hascrc(&mp->m_sb);
 
 	if (crc) {
-		sblock_ok = sblock_ok &&
-			uuid_equal(&block->bb_u.s.bb_uuid,
-				   &mp->m_sb.sb_meta_uuid) &&
-			block->bb_u.s.bb_blkno == cpu_to_be64(
-				bp ? bp->b_bn : XFS_BUF_DADDR_NULL);
+		if (!uuid_equal(&block->bb_u.s.bb_uuid, &mp->m_sb.sb_meta_uuid))
+			return __this_address;
+		if (block->bb_u.s.bb_blkno !=
+		    cpu_to_be64(bp ? bp->b_bn : XFS_BUF_DADDR_NULL))
+			return __this_address;
 	}
 
-	sblock_ok = sblock_ok &&
-		be32_to_cpu(block->bb_magic) == xfs_btree_magic(crc, btnum) &&
-		be16_to_cpu(block->bb_level) == level &&
-		be16_to_cpu(block->bb_numrecs) <=
-			cur->bc_ops->get_maxrecs(cur, level) &&
-		(block->bb_u.s.bb_leftsib == cpu_to_be32(NULLAGBLOCK) ||
-		 be32_to_cpu(block->bb_u.s.bb_leftsib) < agflen) &&
-		block->bb_u.s.bb_leftsib &&
-		(block->bb_u.s.bb_rightsib == cpu_to_be32(NULLAGBLOCK) ||
-		 be32_to_cpu(block->bb_u.s.bb_rightsib) < agflen) &&
-		block->bb_u.s.bb_rightsib;
-
-	if (unlikely(XFS_TEST_ERROR(!sblock_ok, mp,
+	if (be32_to_cpu(block->bb_magic) != xfs_btree_magic(crc, btnum))
+		return __this_address;
+	if (be16_to_cpu(block->bb_level) != level)
+		return __this_address;
+	if (be16_to_cpu(block->bb_numrecs) >
+	    cur->bc_ops->get_maxrecs(cur, level))
+		return __this_address;
+	if (block->bb_u.s.bb_leftsib != cpu_to_be32(NULLAGBLOCK) &&
+	    !xfs_btree_check_sptr(cur, be32_to_cpu(block->bb_u.s.bb_leftsib),
+			level + 1))
+		return __this_address;
+	if (block->bb_u.s.bb_rightsib != cpu_to_be32(NULLAGBLOCK) &&
+	    !xfs_btree_check_sptr(cur, be32_to_cpu(block->bb_u.s.bb_rightsib),
+			level + 1))
+		return __this_address;
+
+	return NULL;
+}
+
+/* Check a short btree block header. */
+STATIC int
+xfs_btree_check_sblock(
+	struct xfs_btree_cur	*cur,
+	struct xfs_btree_block	*block,
+	int			level,
+	struct xfs_buf		*bp)
+{
+	struct xfs_mount	*mp = cur->bc_mp;
+	void			*failed_at;
+
+	failed_at = __xfs_btree_check_sblock(cur, block, level, bp);
+	if (unlikely(XFS_TEST_ERROR(failed_at != NULL, mp,
 			XFS_ERRTAG_BTREE_CHECK_SBLOCK))) {
 		if (bp)
 			trace_xfs_btree_corrupt(bp, _RET_IP_);

index 8f52eda..43694e4 100644
@@ -255,6 +255,14 @@ typedef struct xfs_btree_cur
  */
 #define	XFS_BUF_TO_BLOCK(bp)	((struct xfs_btree_block *)((bp)->b_addr))
 
+/*
+ * Internal long and short btree block checks.  They return NULL if the
+ * block is ok or the address of the failed check otherwise.
+ */
+void *__xfs_btree_check_lblock(struct xfs_btree_cur *cur,
+		struct xfs_btree_block *block, int level, struct xfs_buf *bp);
+void *__xfs_btree_check_sblock(struct xfs_btree_cur *cur,
+		struct xfs_btree_block *block, int level, struct xfs_buf *bp);
 
 /*
  * Check that block header is ok.

index dcd1292..00a5efe 100644
@@ -142,6 +142,13 @@ typedef __u32			xfs_nlink_t;
 #define SYNCHRONIZE()	barrier()
 #define __return_address __builtin_return_address(0)
 
+/*
+ * Return the address of a label.  Use barrier() so that the optimizer
+ * won't reorder code to refactor the error jumpouts into a single
+ * return, which throws off the reported address.
+ */
+#define __this_address	({ __label__ __here; __here: barrier(); &&__here; })
+
 #define XFS_PROJID_DEFAULT	0
 
 #define MIN(a,b)	(min(a,b))

index ccf9783..ee4d2a3 100644
@@ -30,6 +30,7 @@
 #include "xfs_bmap.h"
 #include "xfs_dir2.h"
 #include "xfs_dir2_priv.h"
+#include "xfs_ialloc.h"
 #include "xfs_error.h"
 #include "xfs_trace.h"
 

@@ -202,22 +203,8 @@ xfs_dir_ino_validate(
 	xfs_mount_t	*mp,
 	xfs_ino_t	ino)
 {
-	xfs_agblock_t	agblkno;
-	xfs_agino_t	agino;
-	xfs_agnumber_t	agno;
-	int		ino_ok;
-	int		ioff;
-
-	agno = XFS_INO_TO_AGNO(mp, ino);
-	agblkno = XFS_INO_TO_AGBNO(mp, ino);
-	ioff = XFS_INO_TO_OFFSET(mp, ino);
-	agino = XFS_OFFBNO_TO_AGINO(mp, agblkno, ioff);
-	ino_ok =
-		agno < mp->m_sb.sb_agcount &&
-		agblkno < mp->m_sb.sb_agblocks &&
-		agblkno != 0 &&
-		ioff < (1 << mp->m_sb.sb_inopblog) &&
-		XFS_AGINO_TO_INO(mp, agno, agino) == ino;
+	bool		ino_ok = xfs_verify_dir_ino_ptr(mp, ino);
+
 	if (unlikely(XFS_TEST_ERROR(!ino_ok, mp, XFS_ERRTAG_DIR_INO_VALIDATE))) {
 		xfs_warn(mp, "Invalid inode number 0x%Lx",
 				(unsigned long long) ino);

84 lines not on this page · raw message

index b32cfb5..904d69a 100644
@@ -173,5 +173,12 @@ void xfs_inobt_btrec_to_irec(struct xfs_mount *mp, union xfs_btree_rec *rec,
 		struct xfs_inobt_rec_incore *irec);
 
 int xfs_ialloc_cluster_alignment(struct xfs_mount *mp);
+void xfs_ialloc_aginode_range(struct xfs_mount *mp, xfs_agnumber_t agno,
+		xfs_agino_t *first, xfs_agino_t *last);
+bool xfs_verify_agino_ptr(struct xfs_mount *mp, xfs_agnumber_t agno,
+		xfs_agino_t agino);
+bool xfs_verify_ino_ptr(struct xfs_mount *mp, xfs_ino_t ino);
+bool xfs_internal_inum(struct xfs_mount *mp, xfs_ino_t ino);
+bool xfs_verify_dir_ino_ptr(struct xfs_mount *mp, xfs_ino_t ino);
 
 #endif	/* __XFS_IALLOC_H__ */

index c393a2f..0172d0b 100644
@@ -31,16 +31,6 @@
 #include "xfs_trace.h"
 #include "xfs_icache.h"
 
-int
-xfs_internal_inum(
-	xfs_mount_t	*mp,
-	xfs_ino_t	ino)
-{
-	return (ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
-		(xfs_sb_version_hasquota(&mp->m_sb) &&
-		 xfs_is_quota_inode(&mp->m_sb, ino)));
-}
-
 /*
  * Return stat information for one inode.
  * Return 0 if ok, else errno.

index 17e86e0..6ea8b39 100644
@@ -96,6 +96,4 @@ xfs_inumbers(
 	void			__user *buffer, /* buffer with inode info */
 	inumbers_fmt_pf		formatter);
 
-int xfs_internal_inum(struct xfs_mount *mp, xfs_ino_t ino);
-
 #endif	/* __XFS_ITABLE_H__ */

index ccf9783..ee5e916 100644
@@ -30,6 +30,7 @@
 #include "xfs_bmap.h"
 #include "xfs_dir2.h"
 #include "xfs_dir2_priv.h"
+#include "xfs_ialloc.h"
 #include "xfs_error.h"
 #include "xfs_trace.h"
 

@@ -202,22 +203,8 @@ xfs_dir_ino_validate(
 	xfs_mount_t	*mp,
 	xfs_ino_t	ino)
 {
-	xfs_agblock_t	agblkno;
-	xfs_agino_t	agino;
-	xfs_agnumber_t	agno;
-	int		ino_ok;
-	int		ioff;
-
-	agno = XFS_INO_TO_AGNO(mp, ino);
-	agblkno = XFS_INO_TO_AGBNO(mp, ino);
-	ioff = XFS_INO_TO_OFFSET(mp, ino);
-	agino = XFS_OFFBNO_TO_AGINO(mp, agblkno, ioff);
-	ino_ok =
-		agno < mp->m_sb.sb_agcount &&
-		agblkno < mp->m_sb.sb_agblocks &&
-		agblkno != 0 &&
-		ioff < (1 << mp->m_sb.sb_inopblog) &&
-		XFS_AGINO_TO_INO(mp, agno, agino) == ino;
+	bool		ino_ok = xfs_verify_dir_ino(mp, ino);
+
 	if (unlikely(XFS_TEST_ERROR(!ino_ok, mp, XFS_ERRTAG_DIR_INO_VALIDATE))) {
 		xfs_warn(mp, "Invalid inode number 0x%Lx",
 				(unsigned long long) ino);

93 lines not on this page · raw message

index b32cfb5..d2bdcd5 100644
@@ -173,5 +173,12 @@ void xfs_inobt_btrec_to_irec(struct xfs_mount *mp, union xfs_btree_rec *rec,
 		struct xfs_inobt_rec_incore *irec);
 
 int xfs_ialloc_cluster_alignment(struct xfs_mount *mp);
+void xfs_ialloc_agino_range(struct xfs_mount *mp, xfs_agnumber_t agno,
+		xfs_agino_t *first, xfs_agino_t *last);
+bool xfs_verify_agino(struct xfs_mount *mp, xfs_agnumber_t agno,
+		xfs_agino_t agino);
+bool xfs_verify_ino(struct xfs_mount *mp, xfs_ino_t ino);
+bool xfs_internal_inum(struct xfs_mount *mp, xfs_ino_t ino);
+bool xfs_verify_dir_ino(struct xfs_mount *mp, xfs_ino_t ino);
 
 #endif	/* __XFS_IALLOC_H__ */

index c393a2f..0172d0b 100644
@@ -31,16 +31,6 @@
 #include "xfs_trace.h"
 #include "xfs_icache.h"
 
-int
-xfs_internal_inum(
-	xfs_mount_t	*mp,
-	xfs_ino_t	ino)
-{
-	return (ino == mp->m_sb.sb_rbmino || ino == mp->m_sb.sb_rsumino ||
-		(xfs_sb_version_hasquota(&mp->m_sb) &&
-		 xfs_is_quota_inode(&mp->m_sb, ino)));
-}
-
 /*
  * Return stat information for one inode.
  * Return 0 if ok, else errno.

index 17e86e0..6ea8b39 100644
@@ -96,6 +96,4 @@ xfs_inumbers(
 	void			__user *buffer, /* buffer with inode info */
 	inumbers_fmt_pf		formatter);
 
-int xfs_internal_inum(struct xfs_mount *mp, xfs_ino_t ino);
-
 #endif	/* __XFS_ITABLE_H__ */

65 lines not on this page · raw message

54 lines not on this page · raw message

new file mode 100644
index 0000000..eb1cd9d
@@ -0,0 +1,25 @@
+/*
+ * Copyright (C) 2017 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#ifndef __XFS_SCRUB_SCRUB_H__
+#define __XFS_SCRUB_SCRUB_H__
+
+/* Metadata scrubbers */
+
+#endif	/* __XFS_SCRUB_SCRUB_H__ */

41 lines not on this page · raw message

new file mode 100644
index 0000000..a95a7c8
@@ -0,0 +1,33 @@
+/*
+ * Copyright (C) 2017 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#undef TRACE_SYSTEM
+#define TRACE_SYSTEM xfs_scrub
+
+#if !defined(_TRACE_XFS_SCRUB_TRACE_H) || defined(TRACE_HEADER_MULTI_READ)
+#define _TRACE_XFS_SCRUB_TRACE_H
+
+#include <linux/tracepoint.h>
+
+#endif /* _TRACE_XFS_SCRUB_TRACE_H */
+
+#undef TRACE_INCLUDE_PATH
+#define TRACE_INCLUDE_PATH .
+#define TRACE_INCLUDE_FILE scrub/trace
+#include <trace/define_trace.h>

new file mode 100644
index 0000000..e00e0ea
@@ -0,0 +1,29 @@
+/*
+ * Copyright (C) 2017 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#ifndef __XFS_SCRUB_H__
+#define __XFS_SCRUB_H__
+
+#ifndef CONFIG_XFS_ONLINE_SCRUB
+# define xfs_scrub_metadata(ip, sm)	(-ENOTTY)
+#else
+int xfs_scrub_metadata(struct xfs_inode *ip, struct xfs_scrub_metadata *sm);
+#endif /* CONFIG_XFS_ONLINE_SCRUB */
+
+#endif	/* __XFS_SCRUB_H__ */

index aa75389..6ff012f 100644
@@ -44,6 +44,7 @@
 #include "xfs_btree.h"
 #include <linux/fsmap.h>
 #include "xfs_fsmap.h"
+#include "scrub/xfs_scrub.h"
 
 #include <linux/capability.h>
 #include <linux/cred.h>

@@ -1703,6 +1704,30 @@ xfs_ioc_getfsmap(
 	return 0;
 }
 
+STATIC int
+xfs_ioc_scrub_metadata(
+	struct xfs_inode		*ip,
+	void				__user *arg)
+{
+	struct xfs_scrub_metadata	scrub;
+	int				error;
+
+	if (!capable(CAP_SYS_ADMIN))
+		return -EPERM;
+
+	if (copy_from_user(&scrub, arg, sizeof(scrub)))
+		return -EFAULT;
+
+	error = xfs_scrub_metadata(ip, &scrub);
+	if (error)
+		return error;
+
+	if (copy_to_user(arg, &scrub, sizeof(scrub)))
+		return -EFAULT;
+
+	return 0;
+}
+
 int
 xfs_ioc_swapext(
 	xfs_swapext_t	*sxp)

@@ -1886,6 +1911,9 @@ xfs_file_ioctl(
 	case FS_IOC_GETFSMAP:
 		return xfs_ioc_getfsmap(ip, arg);
 
+	case XFS_IOC_SCRUB_METADATA:
+		return xfs_ioc_scrub_metadata(ip, arg);
+
 	case XFS_IOC_FD_TO_HANDLE:
 	case XFS_IOC_PATH_TO_HANDLE:
 	case XFS_IOC_PATH_TO_FSHANDLE: {

index fa0bc4d..35c79e2 100644
@@ -556,6 +556,7 @@ xfs_file_compat_ioctl(
 	case XFS_IOC_ERROR_INJECTION:
 	case XFS_IOC_ERROR_CLEARALL:
 	case FS_IOC_GETFSMAP:
+	case XFS_IOC_SCRUB_METADATA:
 		return xfs_file_ioctl(filp, cmd, p);
 #ifndef BROKEN_X86_ALIGNMENT
 	/* These are handled fine if no alignment issues */

index 5db2a6f..75c318b 100644
@@ -44,11 +44,201 @@
 #include "scrub/scrub.h"
 #include "scrub/trace.h"
 
+/*
+ * Online Scrub and Repair
+ *
+ * Traditionally, XFS (the kernel driver) did not know how to check or
+ * repair on-disk data structures.  That task was left to the xfs_check
+ * and xfs_repair tools, both of which require taking the filesystem
+ * offline for a thorough but time consuming examination.  Online
+ * scrub & repair, on the other hand, enables us to check the metadata
+ * for obvious errors while carefully stepping around the filesystem's
+ * ongoing operations, locking rules, etc.
+ *
+ * Given that most XFS metadata consist of records stored in a btree,
+ * most of the checking functions iterate the btree blocks themselves
+ * looking for irregularities.  When a record block is encountered, each
+ * record can be checked for obviously bad values.  Record values can
+ * also be cross-referenced against other btrees to look for potential
+ * misunderstandings between pieces of metadata.
+ *
+ * It is expected that the checkers responsible for per-AG metadata
+ * structures will lock the AG headers (AGI, AGF, AGFL), iterate the
+ * metadata structure, and perform any relevant cross-referencing before
+ * unlocking the AG and returning the results to userspace.  These
+ * scrubbers must not keep an AG locked for too long to avoid tying up
+ * the block and inode allocators.
+ *
+ * Block maps and b-trees rooted in an inode present a special challenge
+ * because they can involve extents from any AG.  The general scrubber
+ * structure of lock -> check -> xref -> unlock still holds, but AG
+ * locking order rules /must/ be obeyed to avoid deadlocks.  The
+ * ordering rule, of course, is that we must lock in increasing AG
+ * order.  Helper functions are provided to track which AG headers we've
+ * already locked.  If we detect an imminent locking order violation, we
+ * can signal a potential deadlock, in which case the scrubber can jump
+ * out to the top level, lock all the AGs in order, and retry the scrub.
+ *
+ * For file data (directories, extended attributes, symlinks) scrub, we
+ * can simply lock the inode and walk the data.  For btree data
+ * (directories and attributes) we follow the same btree-scrubbing
+ * strategy outlined previously to check the records.
+ *
+ * We use a bit of trickery with transactions to avoid buffer deadlocks
+ * if there is a cycle in the metadata.  The basic problem is that
+ * travelling down a btree involves locking the current buffer at each
+ * tree level.  If a pointer should somehow point back to a buffer that
+ * we've already examined, we will deadlock due to the second buffer
+ * locking attempt.  Note however that grabbing a buffer in transaction
+ * context links the locked buffer to the transaction.  If we try to
+ * re-grab the buffer in the context of the same transaction, we avoid
+ * the second lock attempt and continue.  Between the verifier and the
+ * scrubber, something will notice that something is amiss and report
+ * the corruption.  Therefore, each scrubber will allocate an empty
+ * transaction, attach buffers to it, and cancel the transaction at the
+ * end of the scrub run.  Cancelling a non-dirty transaction simply
+ * unlocks the buffers.
+ *
+ * There are four pieces of data that scrub can communicate to
+ * userspace.  The first is the error code (errno), which can be used to
+ * communicate operational errors in performing the scrub.  There are
+ * also three flags that can be set in the scrub context.  If the data
+ * structure itself is corrupt, the CORRUPT flag will be set.  If
+ * the metadata is correct but otherwise suboptimal, the PREEN flag
+ * will be set.
+ */
+
+/* Scrub setup and teardown */
+
+/* Free all the resources and finish the transactions. */
+STATIC int
+xfs_scrub_teardown(
+	struct xfs_scrub_context	*sc,
+	int				error)
+{
+	if (sc->tp) {
+		xfs_trans_cancel(sc->tp);
+		sc->tp = NULL;
+	}
+	return error;
+}
+
+/* Scrubbing dispatch. */
+
+static const struct xfs_scrub_meta_ops meta_scrub_ops[] = {
+};
+
+/* This isn't a stable feature, warn once per day. */
+static inline void
+xfs_scrub_experimental_warning(
+	struct xfs_mount	*mp)
+{
+	static struct ratelimit_state scrub_warning = RATELIMIT_STATE_INIT(
+			"xfs_scrub_warning", 86400 * HZ, 1);
+	ratelimit_set_flags(&scrub_warning, RATELIMIT_MSG_ON_RELEASE);
+
+	if (__ratelimit(&scrub_warning))
+		xfs_alert(mp,
+"EXPERIMENTAL online scrub feature in use. Use at your own risk!");
+}
+
 /* Dispatch metadata scrubbing. */
 int
 xfs_scrub_metadata(
 	struct xfs_inode		*ip,
 	struct xfs_scrub_metadata	*sm)
 {
-	return -EOPNOTSUPP;
+	struct xfs_scrub_context	sc;
+	struct xfs_mount		*mp = ip->i_mount;
+	const struct xfs_scrub_meta_ops	*ops;
+	bool				try_harder = false;
+	int				error = 0;
+
+	trace_xfs_scrub_start(ip, sm, error);
+
+	/* Forbidden if we are shut down or mounted norecovery. */
+	error = -ESHUTDOWN;
+	if (XFS_FORCED_SHUTDOWN(mp))
+		goto out;
+	error = -ENOTRECOVERABLE;
+	if (mp->m_flags & XFS_MOUNT_NORECOVERY)
+		goto out;
+
+	/* Check our inputs. */
+	error = -EINVAL;
+	sm->sm_flags &= ~XFS_SCRUB_FLAGS_OUT;
+	if (sm->sm_flags & ~XFS_SCRUB_FLAGS_IN)
+		goto out;
+	if (memchr_inv(sm->sm_reserved, 0, sizeof(sm->sm_reserved)))
+		goto out;
+
+	/* Do we know about this type of metadata? */
+	error = -ENOENT;
+	if (sm->sm_type >= XFS_SCRUB_TYPE_NR)
+		goto out;
+	ops = &meta_scrub_ops[sm->sm_type];
+	if (ops->scrub == NULL)
+		goto out;
+
+	/*
+	 * We won't scrub any filesystem that doesn't have the ability
+	 * to record unwritten extents.  The option was made default in
+	 * 2003, removed from mkfs in 2007, and cannot be disabled in
+	 * v5, so if we find a filesystem without this flag it's either
+	 * really old or totally unsupported.  Avoid it either way.
+	 * We also don't support v1-v3 filesystems, which aren't
+	 * mountable.
+	 */
+	error = -EOPNOTSUPP;
+	if (!xfs_sb_version_hasextflgbit(&mp->m_sb))
+		goto out;
+
+	/* Does this fs even support this type of metadata? */
+	error = -ENOENT;
+	if (ops->has && !ops->has(&mp->m_sb))
+		goto out;
+
+	/* We don't know how to repair anything yet. */
+	error = -EOPNOTSUPP;
+	if (sm->sm_flags & XFS_SCRUB_IFLAG_REPAIR)
+		goto out;
+
+	xfs_scrub_experimental_warning(mp);
+
+retry_op:
+	/* Set up for the operation. */
+	memset(&sc, 0, sizeof(sc));
+	sc.mp = ip->i_mount;
+	sc.sm = sm;
+	sc.ops = ops;
+	sc.try_harder = try_harder;
+	error = sc.ops->setup(&sc, ip);
+	if (error)
+		goto out_teardown;
+
+	/* Scrub for errors. */
+	error = sc.ops->scrub(&sc);
+	if (!try_harder && error == -EDEADLOCK) {
+		/*
+		 * Scrubbers return -EDEADLOCK to mean 'try harder'.
+		 * Tear down everything we hold, then set up again with
+		 * preparation for worst-case scenarios.
+		 */
+		error = xfs_scrub_teardown(&sc, 0);
+		if (error)
+			goto out;
+		try_harder = true;
+		goto retry_op;
+	} else if (error)
+		goto out_teardown;
+
+	if (sc.sm->sm_flags & (XFS_SCRUB_OFLAG_CORRUPT |
+			       XFS_SCRUB_OFLAG_XCORRUPT))
+		xfs_alert_ratelimited(mp, "Corruption detected during scrub.");
+
+out_teardown:
+	error = xfs_scrub_teardown(&sc, error);
+out:
+	trace_xfs_scrub_done(ip, sm, error);
+	return error;
 }

index eb1cd9d..ef7b50e 100644
@@ -20,6 +20,30 @@
 #ifndef __XFS_SCRUB_SCRUB_H__
 #define __XFS_SCRUB_SCRUB_H__
 
+struct xfs_scrub_context;
+
+struct xfs_scrub_meta_ops {
+	/* Acquire whatever resources are needed for the operation. */
+	int		(*setup)(struct xfs_scrub_context *,
+				 struct xfs_inode *);
+
+	/* Examine metadata for errors. */
+	int		(*scrub)(struct xfs_scrub_context *);
+
+	/* Decide if we even have this piece of metadata. */
+	bool		(*has)(struct xfs_sb *);
+};
+
+struct xfs_scrub_context {
+	/* General scrub state. */
+	struct xfs_mount		*mp;
+	struct xfs_scrub_metadata	*sm;
+	const struct xfs_scrub_meta_ops	*ops;
+	struct xfs_trans		*tp;
+	struct xfs_inode		*ip;
+	bool				try_harder;
+};
+
 /* Metadata scrubbers */
 
 #endif	/* __XFS_SCRUB_SCRUB_H__ */

49 lines not on this page · raw message

index 3b4a36e..765f91e 100644
@@ -483,9 +483,10 @@ struct xfs_scrub_metadata {
  */
 
 /* Scrub subcommands. */
+#define XFS_SCRUB_TYPE_PROBE	0	/* presence test ioctl */
 
 /* Number of scrub subcommands. */
-#define XFS_SCRUB_TYPE_NR	0
+#define XFS_SCRUB_TYPE_NR	1
 
 /* i: Repair this metadata. */
 #define XFS_SCRUB_IFLAG_REPAIR		(1 << 0)

59 lines not on this page · raw message

new file mode 100644
index 0000000..f3d5865
@@ -0,0 +1,39 @@
+/*
+ * Copyright (C) 2017 Oracle.  All Rights Reserved.
+ *
+ * Author: Darrick J. Wong <[email protected]>
+ *
+ * This program is free software; you can redistribute it and/or
+ * modify it under the terms of the GNU General Public License
+ * as published by the Free Software Foundation; either version 2
+ * of the License, or (at your option) any later version.
+ *
+ * This program is distributed in the hope that it would be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write the Free Software Foundation,
+ * Inc.,  51 Franklin St, Fifth Floor, Boston, MA  02110-1301, USA.
+ */
+#ifndef __XFS_SCRUB_COMMON_H__
+#define __XFS_SCRUB_COMMON_H__
+
+/*
+ * Grab an empty transaction so that we can re-grab locked buffers if
+ * one of our btrees turns out to be cyclic.
+ */
+static inline int
+xfs_scrub_trans_alloc(
+	struct xfs_scrub_metadata	*sm,
+	struct xfs_mount		*mp,
+	struct xfs_trans		**tpp)
+{
+	return xfs_trans_alloc_empty(mp, tpp);
+}
+
+/* Setup functions */
+int xfs_scrub_setup_fs(struct xfs_scrub_context *sc, struct xfs_inode *ip);
+
+#endif	/* __XFS_SCRUB_COMMON_H__ */

index 75c318b..92eac98 100644
@@ -42,6 +42,7 @@
 #include "xfs_rmap_btree.h"
 #include "scrub/xfs_scrub.h"
 #include "scrub/scrub.h"
+#include "scrub/common.h"
 #include "scrub/trace.h"
 
 /*

@@ -108,6 +109,20 @@
  * will be set.
  */
 
+/*
+ * Scrub probe -- userspace uses this to probe if we're willing to
+ * scrub or repair a given mountpoint.
+ */
+int
+xfs_scrub_probe(
+	struct xfs_scrub_context	*sc)
+{
+	if (sc->sm->sm_ino || sc->sm->sm_agno)
+		return -EINVAL;
+
+	return 0;
+}
+
 /* Scrub setup and teardown */
 
 /* Free all the resources and finish the transactions. */

@@ -126,6 +141,10 @@ xfs_scrub_teardown(
 /* Scrubbing dispatch. */
 
 static const struct xfs_scrub_meta_ops meta_scrub_ops[] = {
+	{ /* ioctl presence test */
+		.setup	= xfs_scrub_setup_fs,
+		.scrub	= xfs_scrub_probe,
+	},
 };
 
 /* This isn't a stable feature, warn once per day. */

index ef7b50e..b7b9422 100644
@@ -45,5 +45,6 @@ struct xfs_scrub_context {
 };
 
 /* Metadata scrubbers */
+int xfs_scrub_tester(struct xfs_scrub_context *sc);
 
 #endif	/* __XFS_SCRUB_SCRUB_H__ */

index c59fd41..88b5ccb 100644
@@ -32,6 +32,7 @@
 #include "xfs_trans.h"
 #include "scrub/xfs_scrub.h"
 #include "scrub/scrub.h"
+#include "scrub/common.h"
 
 /*
  * We include this last to have the helpers above available for the trace

index d2c8f94..709d491 100644
@@ -47,6 +47,196 @@
 
 /* Common code for the metadata scrubbers. */
 
+/*
+ * Handling operational errors.
+ *
+ * The *_process_error() family of functions are used to process error return
+ * codes from functions called as part of a scrub operation.
+ *
+ * If there's no error, we return true to tell the caller that it's ok
+ * to move on to the next check in its list.
+ *
+ * For non-verifier errors (e.g. ENOMEM) we return false to tell the
+ * caller that something bad happened, and we preserve *error so that
+ * the caller can return the *error up the stack to userspace.
+ *
+ * Verifier errors (EFSBADCRC/EFSCORRUPTED) are recorded by setting
+ * OFLAG_CORRUPT in sm_flags and the *error is cleared.  In other words,
+ * we track verifier errors (and failed scrub checks) via OFLAG_CORRUPT,
+ * not via return codes.  We return false to tell the caller that
+ * something bad happened.  Since the error has been cleared, the caller
+ * will (presumably) return that zero and scrubbing will move on to
+ * whatever's next.
+ *
+ * ftrace can be used to record the precise metadata location and the
+ * approximate code location of the failed operation.
+ */
+
+/* Check for operational errors. */
+bool
+xfs_scrub_process_error(
+	struct xfs_scrub_context	*sc,
+	xfs_agnumber_t			agno,
+	xfs_agblock_t			bno,
+	int				*error)
+{
+	switch (*error) {
+	case 0:
+		return true;
+	case -EDEADLOCK:
+		/* Used to restart an op with deadlock avoidance. */
+		trace_xfs_scrub_deadlock_retry(sc->ip, sc->sm, *error);
+		break;
+	case -EFSBADCRC:
+	case -EFSCORRUPTED:
+		/* Note the badness but don't abort. */
+		sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
+		*error = 0;
+		/* fall through */
+	default:
+		trace_xfs_scrub_op_error(sc, agno, bno, *error,
+				__return_address);
+		break;
+	}
+	return false;
+}
+
+/* Check for operational errors for a file offset. */
+bool
+xfs_scrub_fblock_process_error(
+	struct xfs_scrub_context	*sc,
+	int				whichfork,
+	xfs_fileoff_t			offset,
+	int				*error)
+{
+	switch (*error) {
+	case 0:
+		return true;
+	case -EDEADLOCK:
+		/* Used to restart an op with deadlock avoidance. */
+		trace_xfs_scrub_deadlock_retry(sc->ip, sc->sm, *error);
+		break;
+	case -EFSBADCRC:
+	case -EFSCORRUPTED:
+		/* Note the badness but don't abort. */
+		sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
+		*error = 0;
+		/* fall through */
+	default:
+		trace_xfs_scrub_file_op_error(sc, whichfork, offset, *error,
+				__return_address);
+		break;
+	}
+	return false;
+}
+
+/*
+ * Handling scrub corruption/optimization/warning checks.
+ *
+ * The *_set_{corrupt,preen,warning}() family of functions are used to
+ * record the presence of metadata that is incorrect (corrupt), could be
+ * optimized somehow (preen), or should be flagged for administrative
+ * review but is not incorrect (warn).
+ *
+ * ftrace can be used to record the precise metadata location and
+ * approximate code location of the failed check.
+ */
+
+/* Record a block which could be optimized. */
+void
+xfs_scrub_block_set_preen(
+	struct xfs_scrub_context	*sc,
+	struct xfs_buf			*bp)
+{
+	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_PREEN;
+	trace_xfs_scrub_block_preen(sc, bp->b_bn, __return_address);
+}
+
+/*
+ * Record an inode which could be optimized.  The trace data will
+ * include the block given by bp if bp is given; otherwise it will use
+ * the block location of the inode record itself.
+ */
+void
+xfs_scrub_ino_set_preen(
+	struct xfs_scrub_context	*sc,
+	struct xfs_buf			*bp)
+{
+	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_PREEN;
+	trace_xfs_scrub_ino_preen(sc, sc->ip->i_ino, bp ? bp->b_bn : 0,
+			__return_address);
+}
+
+/* Record a corrupt block. */
+void
+xfs_scrub_block_set_corrupt(
+	struct xfs_scrub_context	*sc,
+	struct xfs_buf			*bp)
+{
+	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
+	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
+ * inode record itself.
+ */
+void
+xfs_scrub_ino_set_corrupt(
+	struct xfs_scrub_context	*sc,
+	xfs_ino_t			ino,
+	struct xfs_buf			*bp)
+{
+	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
+	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(
+	struct xfs_scrub_context	*sc,
+	int				whichfork,
+	xfs_fileoff_t			offset)
+{
+	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_CORRUPT;
+	trace_xfs_scrub_fblock_error(sc, whichfork, offset, __return_address);
+}
+
+/*
+ * Warn about inodes that need administrative review but is not
+ * incorrect.
+ */
+void
+xfs_scrub_ino_set_warning(
+	struct xfs_scrub_context	*sc,
+	struct xfs_buf			*bp)
+{
+	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_WARNING;
+	trace_xfs_scrub_ino_warning(sc, sc->ip->i_ino, bp ? bp->b_bn : 0,
+			__return_address);
+}
+
+/* Warn about a block indexed by a file fork that needs review. */
+void
+xfs_scrub_fblock_set_warning(
+	struct xfs_scrub_context	*sc,
+	int				whichfork,
+	xfs_fileoff_t			offset)
+{
+	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_WARNING;
+	trace_xfs_scrub_fblock_warning(sc, whichfork, offset, __return_address);
+}
+
+/* Signal an incomplete scrub. */
+void
+xfs_scrub_set_incomplete(
+	struct xfs_scrub_context	*sc)
+{
+	sc->sm->sm_flags |= XFS_SCRUB_OFLAG_INCOMPLETE;
+	trace_xfs_scrub_incomplete(sc, __return_address);
+}
+
 /* Per-scrubber setup functions */
 
 /* Set us up with a transaction and an empty context. */

index f3d5865..a7c3361 100644
@@ -33,6 +33,29 @@ xfs_scrub_trans_alloc(
 	return xfs_trans_alloc_empty(mp, tpp);
 }
 
+bool xfs_scrub_process_error(struct xfs_scrub_context *sc, xfs_agnumber_t agno,
+		xfs_agblock_t bno, int *error);
+bool xfs_scrub_fblock_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, struct xfs_buf *bp);
+
+void xfs_scrub_block_set_corrupt(struct xfs_scrub_context *sc,
+		struct xfs_buf *bp);
+void xfs_scrub_ino_set_corrupt(struct xfs_scrub_context *sc, xfs_ino_t ino,
+		struct xfs_buf *bp);
+void xfs_scrub_fblock_set_corrupt(struct xfs_scrub_context *sc, int whichfork,
+		xfs_fileoff_t offset);
+
+void xfs_scrub_ino_set_warning(struct xfs_scrub_context *sc,
+		struct xfs_buf *bp);
+void xfs_scrub_fblock_set_warning(struct xfs_scrub_context *sc, int whichfork,
+		xfs_fileoff_t offset);
+
+void xfs_scrub_set_incomplete(struct xfs_scrub_context *sc);
+
 /* Setup functions */
 int xfs_scrub_setup_fs(struct xfs_scrub_context *sc, struct xfs_inode *ip);
 

227 lines not on this page · raw message

114 lines not on this page · raw message

57 lines not on this page · raw message

index 88b5ccb..472080e 100644
@@ -30,10 +30,27 @@
 #include "xfs_inode.h"
 #include "xfs_btree.h"
 #include "xfs_trans.h"
+#include "xfs_bit.h"
 #include "scrub/xfs_scrub.h"
 #include "scrub/scrub.h"
 #include "scrub/common.h"
 
+/* Figure out which block the btree cursor was pointing to. */
+static inline xfs_fsblock_t
+xfs_scrub_btree_cur_fsbno(
+	struct xfs_btree_cur		*cur,
+	int				level)
+{
+	if (level < cur->bc_nlevels && cur->bc_bufs[level])
+		return XFS_DADDR_TO_FSB(cur->bc_mp, cur->bc_bufs[level]->b_bn);
+	else if (level == cur->bc_nlevels - 1 &&
+		 cur->bc_flags & XFS_BTREE_LONG_PTRS)
+		return XFS_INO_TO_FSB(cur->bc_mp, cur->bc_private.b.ip->i_ino);
+	else if (!(cur->bc_flags & XFS_BTREE_LONG_PTRS))
+		return XFS_AGB_TO_FSB(cur->bc_mp, cur->bc_private.a.agno, 0);
+	return NULLFSBLOCK;
+}
+
 /*
  * We include this last to have the helpers above available for the trace
  * event implementations.

169 lines not on this page · raw message

index 2266a5a..dc23407 100644
@@ -1051,7 +1051,7 @@ xfs_btree_setbuf(
 	}
 }
 
-STATIC int
+bool
 xfs_btree_ptr_is_null(
 	struct xfs_btree_cur	*cur,
 	union xfs_btree_ptr	*ptr)

@@ -1076,7 +1076,7 @@ xfs_btree_set_ptr_null(
 /*
  * Get/set/init sibling pointers
  */
-STATIC void
+void
 xfs_btree_get_sibling(
 	struct xfs_btree_cur	*cur,
 	struct xfs_btree_block	*block,

@@ -4938,3 +4938,15 @@ xfs_btree_count_blocks(
 	return xfs_btree_visit_blocks(cur, xfs_btree_count_blocks_helper,
 			blocks);
 }
+
+/* Compare two btree pointers. */
+int64_t
+xfs_btree_diff_two_ptrs(
+	struct xfs_btree_cur		*cur,
+	const union xfs_btree_ptr	*a,
+	const union xfs_btree_ptr	*b)
+{
+	if (cur->bc_flags & XFS_BTREE_LONG_PTRS)
+		return (int64_t)be64_to_cpu(a->l) - be64_to_cpu(b->l);
+	return (int64_t)be32_to_cpu(a->s) - be32_to_cpu(b->s);
+}

index baf7064..a8431bc 100644
@@ -531,5 +531,12 @@ int xfs_btree_lookup_get_block(struct xfs_btree_cur *cur, int level,
 		union xfs_btree_ptr *pp, struct xfs_btree_block **blkp);
 struct xfs_btree_block *xfs_btree_get_block(struct xfs_btree_cur *cur,
 		int level, struct xfs_buf **bpp);
+bool xfs_btree_ptr_is_null(struct xfs_btree_cur *cur, union xfs_btree_ptr *ptr);
+int64_t xfs_btree_diff_two_ptrs(struct xfs_btree_cur *cur,
+				const union xfs_btree_ptr *a,
+				const union xfs_btree_ptr *b);
+void xfs_btree_get_sibling(struct xfs_btree_cur *cur,
+			   struct xfs_btree_block *block,
+			   union xfs_btree_ptr *ptr, int lr);
 
 #endif	/* __XFS_BTREE_H__ */

263 lines not on this page · raw message

index a7c3361..414bbb8 100644
@@ -21,6 +21,24 @@
 #define __XFS_SCRUB_COMMON_H__
 
 /*
+ * We /could/ terminate a scrub/repair operation early.  If we're not
+ * in a good place to continue (fatal signal, etc.) then bail out.
+ * Note that we're careful not to make any judgements about *error.
+ */
+static inline bool
+xfs_scrub_should_terminate(
+	struct xfs_scrub_context	*sc,
+	int				*error)
+{
+	if (fatal_signal_pending(current)) {
+		if (*error == 0)
+			*error = -EAGAIN;
+		return true;
+	}
+	return false;
+}
+
+/*
  * Grab an empty transaction so that we can re-grab locked buffers if
  * one of our btrees turns out to be cyclic.
  */

index 68dec6a..0cd591f 100644
@@ -93,6 +93,101 @@ xfs_scrub_btree_set_corrupt(
 }
 
 /*
+ * Make sure this record is in order and doesn't stray outside of the parent
+ * keys.
+ */
+STATIC void
+xfs_scrub_btree_rec(
+	struct xfs_scrub_btree	*bs)
+{
+	struct xfs_btree_cur	*cur = bs->cur;
+	union xfs_btree_rec	*rec;
+	union xfs_btree_key	key;
+	union xfs_btree_key	hkey;
+	union xfs_btree_key	*keyp;
+	struct xfs_btree_block	*block;
+	struct xfs_btree_block	*keyblock;
+	struct xfs_buf		*bp;
+
+	block = xfs_btree_get_block(cur, 0, &bp);
+	rec = xfs_btree_rec_addr(cur, cur->bc_ptrs[0], block);
+
+	trace_xfs_scrub_btree_rec(bs->sc, cur, 0);
+
+	/* If this isn't the first record, are they in order? */
+	if (!bs->firstrec && !cur->bc_ops->recs_inorder(cur, &bs->lastrec, rec))
+		xfs_scrub_btree_set_corrupt(bs->sc, cur, 0);
+	bs->firstrec = false;
+	memcpy(&bs->lastrec, rec, cur->bc_ops->rec_len);
+
+	if (cur->bc_nlevels == 1)
+		return;
+
+	/* Is this at least as large as the parent low key? */
+	cur->bc_ops->init_key_from_rec(&key, rec);
+	keyblock = xfs_btree_get_block(cur, 1, &bp);
+	keyp = xfs_btree_key_addr(cur, cur->bc_ptrs[1], keyblock);
+	if (cur->bc_ops->diff_two_keys(cur, &key, keyp) < 0)
+		xfs_scrub_btree_set_corrupt(bs->sc, cur, 1);
+
+	if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING))
+		return;
+
+	/* Is this no larger than the parent high key? */
+	cur->bc_ops->init_high_key_from_rec(&hkey, rec);
+	keyp = xfs_btree_high_key_addr(cur, cur->bc_ptrs[1], keyblock);
+	if (cur->bc_ops->diff_two_keys(cur, keyp, &hkey) < 0)
+		xfs_scrub_btree_set_corrupt(bs->sc, cur, 1);
+}
+
+/*
+ * Make sure this key is in order and doesn't stray outside of the parent
+ * keys.
+ */
+STATIC void
+xfs_scrub_btree_key(
+	struct xfs_scrub_btree	*bs,
+	int			level)
+{
+	struct xfs_btree_cur	*cur = bs->cur;
+	union xfs_btree_key	*key;
+	union xfs_btree_key	*keyp;
+	struct xfs_btree_block	*block;
+	struct xfs_btree_block	*keyblock;
+	struct xfs_buf		*bp;
+
+	block = xfs_btree_get_block(cur, level, &bp);
+	key = xfs_btree_key_addr(cur, cur->bc_ptrs[level], block);
+
+	trace_xfs_scrub_btree_key(bs->sc, cur, level);
+
+	/* If this isn't the first key, are they in order? */
+	if (!bs->firstkey[level] &&
+	    !cur->bc_ops->keys_inorder(cur, &bs->lastkey[level], key))
+		xfs_scrub_btree_set_corrupt(bs->sc, cur, level);
+	bs->firstkey[level] = false;
+	memcpy(&bs->lastkey[level], key, cur->bc_ops->key_len);
+
+	if (level + 1 >= cur->bc_nlevels)
+		return;
+
+	/* Is this at least as large as the parent low key? */
+	keyblock = xfs_btree_get_block(cur, level + 1, &bp);
+	keyp = xfs_btree_key_addr(cur, cur->bc_ptrs[level + 1], keyblock);
+	if (cur->bc_ops->diff_two_keys(cur, key, keyp) < 0)
+		xfs_scrub_btree_set_corrupt(bs->sc, cur, level);
+
+	if (!(cur->bc_flags & XFS_BTREE_OVERLAPPING))
+		return;
+
+	/* Is this no larger than the parent high key? */
+	key = xfs_btree_high_key_addr(cur, cur->bc_ptrs[level], block);
+	keyp = xfs_btree_high_key_addr(cur, cur->bc_ptrs[level + 1], keyblock);
+	if (cur->bc_ops->diff_two_keys(cur, keyp, key) < 0)
+		xfs_scrub_btree_set_corrupt(bs->sc, cur, level);
+}
+
+/*
  * Check a btree pointer.  Returns true if it's ok to use this pointer.
  * Callers do not need to set the corrupt flag.
  */

@@ -269,6 +364,7 @@ xfs_scrub_btree(
 	struct xfs_scrub_btree		bs = {0};
 	union xfs_btree_ptr		ptr;
 	union xfs_btree_ptr		*pp;
+	union xfs_btree_rec		*recp;
 	struct xfs_btree_block		*block;
 	int				level;
 	struct xfs_buf			*bp;

@@ -319,7 +415,16 @@ xfs_scrub_btree(
 				continue;
 			}
 
-			if (xfs_scrub_should_terminate(sc, &error))
+			/* Records in order for scrub? */
+			xfs_scrub_btree_rec(&bs);
+
+			/* Call out to the record checker. */
+			recp = xfs_btree_rec_addr(cur, cur->bc_ptrs[0], block);
+			error = bs.scrub_rec(&bs, recp);
+			if (error)
+				break;
+			if (xfs_scrub_should_terminate(sc, &error) ||
+			    (sc->sm->sm_flags & XFS_SCRUB_OFLAG_CORRUPT))
 				break;
 
 			cur->bc_ptrs[level]++;

@@ -334,6 +439,9 @@ xfs_scrub_btree(
 			continue;
 		}
 
+		/* Keys in order for scrub? */
+		xfs_scrub_btree_key(&bs, level);
+
 		/* Drill another level deeper. */
 		pp = xfs_btree_ptr_addr(cur, cur->bc_ptrs[level], block);
 		if (!xfs_scrub_btree_ptr_ok(&bs, level, pp)) {

51 lines not on this page · raw message

index 709d491..cd6fada 100644
@@ -44,6 +44,7 @@
 #include "scrub/scrub.h"
 #include "scrub/common.h"
 #include "scrub/trace.h"
+#include "scrub/btree.h"
 
 /* Common code for the metadata scrubbers. */
 

@@ -237,6 +238,184 @@ xfs_scrub_set_incomplete(
 	trace_xfs_scrub_incomplete(sc, __return_address);
 }
 
+/*
+ * AG scrubbing
+ *
+ * These helpers facilitate locking an allocation group's header
+ * buffers, setting up cursors for all btrees that are present, and
+ * cleaning everything up once we're through.
+ */
+
+/*
+ * Grab all the headers for an AG.
+ *
+ * The headers should be released by xfs_scrub_ag_free, but as a fail
+ * safe we attach all the buffers we grab to the scrub transaction so
+ * they'll all be freed when we cancel it.
+ */
+int
+xfs_scrub_ag_read_headers(
+	struct xfs_scrub_context	*sc,
+	xfs_agnumber_t			agno,
+	struct xfs_buf			**agi,
+	struct xfs_buf			**agf,
+	struct xfs_buf			**agfl)
+{
+	struct xfs_mount		*mp = sc->mp;
+	int				error;
+
+	error = xfs_ialloc_read_agi(mp, sc->tp, agno, agi);
+	if (error)
+		goto out;
+
+	error = xfs_alloc_read_agf(mp, sc->tp, agno, 0, agf);
+	if (error)
+		goto out;
+	if (!*agf) {
+		error = -ENOMEM;
+		goto out;
+	}
+
+	error = xfs_alloc_read_agfl(mp, sc->tp, agno, agfl);
+	if (error)
+		goto out;
+
+out:
+	return error;
+}
+
+/* Release all the AG btree cursors. */
+void
+xfs_scrub_ag_btcur_free(
+	struct xfs_scrub_ag		*sa)
+{
+	if (sa->refc_cur)
+		xfs_btree_del_cursor(sa->refc_cur, XFS_BTREE_ERROR);
+	if (sa->rmap_cur)
+		xfs_btree_del_cursor(sa->rmap_cur, XFS_BTREE_ERROR);
+	if (sa->fino_cur)
+		xfs_btree_del_cursor(sa->fino_cur, XFS_BTREE_ERROR);
+	if (sa->ino_cur)
+		xfs_btree_del_cursor(sa->ino_cur, XFS_BTREE_ERROR);
+	if (sa->cnt_cur)
+		xfs_btree_del_cursor(sa->cnt_cur, XFS_BTREE_ERROR);
+	if (sa->bno_cur)
+		xfs_btree_del_cursor(sa->bno_cur, XFS_BTREE_ERROR);
+
+	sa->refc_cur = NULL;
+	sa->rmap_cur = NULL;
+	sa->fino_cur = NULL;
+	sa->ino_cur = NULL;
+	sa->bno_cur = NULL;
+	sa->cnt_cur = NULL;
+}
+
+/* Initialize all the btree cursors for an AG. */
+int
+xfs_scrub_ag_btcur_init(
+	struct xfs_scrub_context	*sc,
+	struct xfs_scrub_ag		*sa)
+{
+	struct xfs_mount		*mp = sc->mp;
+	xfs_agnumber_t			agno = sa->agno;
+
+	if (sa->agf_bp) {
+		/* Set up a bnobt cursor for cross-referencing. */
+		sa->bno_cur = xfs_allocbt_init_cursor(mp, sc->tp, sa->agf_bp,
+				agno, XFS_BTNUM_BNO);
+		if (!sa->bno_cur)
+			goto err;
+
+		/* Set up a cntbt cursor for cross-referencing. */
+		sa->cnt_cur = xfs_allocbt_init_cursor(mp, sc->tp, sa->agf_bp,
+				agno, XFS_BTNUM_CNT);
+		if (!sa->cnt_cur)
+			goto err;
+	}
+
+	/* Set up a inobt cursor for cross-referencing. */
+	if (sa->agi_bp) {
+		sa->ino_cur = xfs_inobt_init_cursor(mp, sc->tp, sa->agi_bp,
+					agno, XFS_BTNUM_INO);
+		if (!sa->ino_cur)
+			goto err;
+	}
+
+	/* Set up a finobt cursor for cross-referencing. */
+	if (sa->agi_bp && xfs_sb_version_hasfinobt(&mp->m_sb)) {
+		sa->fino_cur = xfs_inobt_init_cursor(mp, sc->tp, sa->agi_bp,
+				agno, XFS_BTNUM_FINO);
+		if (!sa->fino_cur)
+			goto err;
+	}
+
+	/* Set up a rmapbt cursor for cross-referencing. */
+	if (sa->agf_bp && xfs_sb_version_hasrmapbt(&mp->m_sb)) {
+		sa->rmap_cur = xfs_rmapbt_init_cursor(mp, sc->tp, sa->agf_bp,
+				agno);
+		if (!sa->rmap_cur)
+			goto err;
+	}
+
+	/* Set up a refcountbt cursor for cross-referencing. */
+	if (sa->agf_bp && xfs_sb_version_hasreflink(&mp->m_sb)) {
+		sa->refc_cur = xfs_refcountbt_init_cursor(mp, sc->tp,
+				sa->agf_bp, agno, NULL);
+		if (!sa->refc_cur)
+			goto err;
+	}
+
+	return 0;
+err:
+	return -ENOMEM;
+}
+
+/* Release the AG header context and btree cursors. */
+void
+xfs_scrub_ag_free(
+	struct xfs_scrub_context	*sc,
+	struct xfs_scrub_ag		*sa)
+{
+	xfs_scrub_ag_btcur_free(sa);
+	if (sa->agfl_bp) {
+		xfs_trans_brelse(sc->tp, sa->agfl_bp);
+		sa->agfl_bp = NULL;
+	}
+	if (sa->agf_bp) {
+		xfs_trans_brelse(sc->tp, sa->agf_bp);
+		sa->agf_bp = NULL;
+	}
+	if (sa->agi_bp) {
+		xfs_trans_brelse(sc->tp, sa->agi_bp);
+		sa->agi_bp = NULL;
+	}
+	sa->agno = NULLAGNUMBER;
+}
+
+/*
+ * For scrub, grab the AGI and the AGF headers, in that order.  Locking
+ * order requires us to get the AGI before the AGF.  We use the
+ * transaction to avoid deadlocking on crosslinked metadata buffers;
+ * either the caller passes one in (bmap scrub) or we have to create a
+ * transaction ourselves.
+ */
+int
+xfs_scrub_ag_init(
+	struct xfs_scrub_context	*sc,
+	xfs_agnumber_t			agno,
+	struct xfs_scrub_ag		*sa)
+{
+	int				error;
+
+	sa->agno = agno;
+	error = xfs_scrub_ag_read_headers(sc, agno, &sa->agi_bp,
+			&sa->agf_bp, &sa->agfl_bp);
+	if (error)
+		return error;
+
+	return xfs_scrub_ag_btcur_init(sc, sa);
+}
+
 /* Per-scrubber setup functions */
 
 /* Set us up with a transaction and an empty context. */

index 414bbb8..aca39b5 100644
@@ -77,4 +77,14 @@ void xfs_scrub_set_incomplete(struct xfs_scrub_context *sc);
 /* Setup functions */
 int xfs_scrub_setup_fs(struct xfs_scrub_context *sc, struct xfs_inode *ip);
 
+void xfs_scrub_ag_free(struct xfs_scrub_context *sc, struct xfs_scrub_ag *sa);
+int xfs_scrub_ag_init(struct xfs_scrub_context *sc, xfs_agnumber_t agno,
+		      struct xfs_scrub_ag *sa);
+int xfs_scrub_ag_read_headers(struct xfs_scrub_context *sc, xfs_agnumber_t agno,
+			      struct xfs_buf **agi, struct xfs_buf **agf,
+			      struct xfs_buf **agfl);
+void xfs_scrub_ag_btcur_free(struct xfs_scrub_ag *sa);
+int xfs_scrub_ag_btcur_init(struct xfs_scrub_context *sc,
+			    struct xfs_scrub_ag *sa);
+
 #endif	/* __XFS_SCRUB_COMMON_H__ */