1 files changed, 133 insertions, 90 deletions

diff --git a/fs/xfs/xfs_log_recover.c b/fs/xfs/xfs_log_recover.c
index 9549188f5a36..87b1c331f9eb 100644
--- a/fs/xfs/xfs_log_recover.c
+++ b/fs/xfs/xfs_log_recover.c
@@ -85,17 +85,21 @@ struct xfs_buf_cancel {
  */
 
 /*
- * Verify the given count of basic blocks is valid number of blocks
- * to specify for an operation involving the given XFS log buffer.
- * Returns nonzero if the count is valid, 0 otherwise.
+ * Verify the log-relative block number and length in basic blocks are valid for
+ * an operation involving the given XFS log buffer. Returns true if the fields
+ * are valid, false otherwise.
  */
-
-static inline int
-xlog_buf_bbcount_valid(
+static inline bool
+xlog_verify_bp(
 	struct xlog	*log,
+	xfs_daddr_t	blk_no,
 	int		bbcount)
 {
-	return bbcount > 0 && bbcount <= log->l_logBBsize;
+	if (blk_no < 0 || blk_no >= log->l_logBBsize)
+		return false;
+	if (bbcount <= 0 || (blk_no + bbcount) > log->l_logBBsize)
+		return false;
+	return true;
 }
 
 /*
@@ -110,7 +114,11 @@ xlog_get_bp(
 {
 	struct xfs_buf	*bp;
 
-	if (!xlog_buf_bbcount_valid(log, nbblks)) {
+	/*
+	 * Pass log block 0 since we don't have an addr yet, buffer will be
+	 * verified on read.
+	 */
+	if (!xlog_verify_bp(log, 0, nbblks)) {
 		xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
 			nbblks);
 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
@@ -180,9 +188,10 @@ xlog_bread_noalign(
 {
 	int		error;
 
-	if (!xlog_buf_bbcount_valid(log, nbblks)) {
-		xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
-			nbblks);
+	if (!xlog_verify_bp(log, blk_no, nbblks)) {
+		xfs_warn(log->l_mp,
+			 "Invalid log block/length (0x%llx, 0x%x) for buffer",
+			 blk_no, nbblks);
 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
 		return -EFSCORRUPTED;
 	}
@@ -265,9 +274,10 @@ xlog_bwrite(
 {
 	int		error;
 
-	if (!xlog_buf_bbcount_valid(log, nbblks)) {
-		xfs_warn(log->l_mp, "Invalid block length (0x%x) for buffer",
-			nbblks);
+	if (!xlog_verify_bp(log, blk_no, nbblks)) {
+		xfs_warn(log->l_mp,
+			 "Invalid log block/length (0x%llx, 0x%x) for buffer",
+			 blk_no, nbblks);
 		XFS_ERROR_REPORT(__func__, XFS_ERRLEVEL_HIGH, log->l_mp);
 		return -EFSCORRUPTED;
 	}
@@ -753,7 +763,7 @@ xlog_find_head(
 	 * in the in-core log.  The following number can be made tighter if
 	 * we actually look at the block size of the filesystem.
 	 */
-	num_scan_bblks = XLOG_TOTAL_REC_SHIFT(log);
+	num_scan_bblks = min_t(int, log_bbnum, XLOG_TOTAL_REC_SHIFT(log));
 	if (head_blk >= num_scan_bblks) {
 		/*
 		 * We are guaranteed that the entire check can be performed
@@ -1029,61 +1039,106 @@ out_error:
 }
 
 /*
- * Check the log tail for torn writes. This is required when torn writes are
- * detected at the head and the head had to be walked back to a previous record.
- * The tail of the previous record must now be verified to ensure the torn
- * writes didn't corrupt the previous tail.
+ * Calculate distance from head to tail (i.e., unused space in the log).
+ */
+static inline int
+xlog_tail_distance(
+	struct xlog	*log,
+	xfs_daddr_t	head_blk,
+	xfs_daddr_t	tail_blk)
+{
+	if (head_blk < tail_blk)
+		return tail_blk - head_blk;
+
+	return tail_blk + (log->l_logBBsize - head_blk);
+}
+
+/*
+ * Verify the log tail. This is particularly important when torn or incomplete
+ * writes have been detected near the front of the log and the head has been
+ * walked back accordingly.
+ *
+ * We also have to handle the case where the tail was pinned and the head
+ * blocked behind the tail right before a crash. If the tail had been pushed
+ * immediately prior to the crash and the subsequent checkpoint was only
+ * partially written, it's possible it overwrote the last referenced tail in the
+ * log with garbage. This is not a coherency problem because the tail must have
+ * been pushed before it can be overwritten, but appears as log corruption to
+ * recovery because we have no way to know the tail was updated if the
+ * subsequent checkpoint didn't write successfully.
  *
- * Return an error if CRC verification fails as recovery cannot proceed.
+ * Therefore, CRC check the log from tail to head. If a failure occurs and the
+ * offending record is within max iclog bufs from the head, walk the tail
+ * forward and retry until a valid tail is found or corruption is detected out
+ * of the range of a possible overwrite.
  */
 STATIC int
 xlog_verify_tail(
 	struct xlog		*log,
 	xfs_daddr_t		head_blk,
-	xfs_daddr_t		tail_blk)
+	xfs_daddr_t		*tail_blk,
+	int			hsize)
 {
 	struct xlog_rec_header	*thead;
 	struct xfs_buf		*bp;
 	xfs_daddr_t		first_bad;
-	int			count;
 	int			error = 0;
 	bool			wrapped;
-	xfs_daddr_t		tmp_head;
+	xfs_daddr_t		tmp_tail;
+	xfs_daddr_t		orig_tail = *tail_blk;
 
 	bp = xlog_get_bp(log, 1);
 	if (!bp)
 		return -ENOMEM;
 
 	/*
-	 * Seek XLOG_MAX_ICLOGS + 1 records past the current tail record to get
-	 * a temporary head block that points after the last possible
-	 * concurrently written record of the tail.
+	 * Make sure the tail points to a record (returns positive count on
+	 * success).
 	 */
-	count = xlog_seek_logrec_hdr(log, head_blk, tail_blk,
-				     XLOG_MAX_ICLOGS + 1, bp, &tmp_head, &thead,
-				     &wrapped);
-	if (count < 0) {
-		error = count;
+	error = xlog_seek_logrec_hdr(log, head_blk, *tail_blk, 1, bp,
+			&tmp_tail, &thead, &wrapped);
+	if (error < 0)
 		goto out;
-	}
-
-	/*
-	 * If the call above didn't find XLOG_MAX_ICLOGS + 1 records, we ran
-	 * into the actual log head. tmp_head points to the start of the record
-	 * so update it to the actual head block.
-	 */
-	if (count < XLOG_MAX_ICLOGS + 1)
-		tmp_head = head_blk;
+	if (*tail_blk != tmp_tail)
+		*tail_blk = tmp_tail;
 
 	/*
-	 * We now have a tail and temporary head block that covers at least
-	 * XLOG_MAX_ICLOGS records from the tail. We need to verify that these
-	 * records were completely written. Run a CRC verification pass from
-	 * tail to head and return the result.
+	 * Run a CRC check from the tail to the head. We can't just check
+	 * MAX_ICLOGS records past the tail because the tail may point to stale
+	 * blocks cleared during the search for the head/tail. These blocks are
+	 * overwritten with zero-length records and thus record count is not a
+	 * reliable indicator of the iclog state before a crash.
 	 */
-	error = xlog_do_recovery_pass(log, tmp_head, tail_blk,
+	first_bad = 0;
+	error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
 				      XLOG_RECOVER_CRCPASS, &first_bad);
+	while ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
+		int	tail_distance;
+
+		/*
+		 * Is corruption within range of the head? If so, retry from
+		 * the next record. Otherwise return an error.
+		 */
+		tail_distance = xlog_tail_distance(log, head_blk, first_bad);
+		if (tail_distance > BTOBB(XLOG_MAX_ICLOGS * hsize))
+			break;
+
+		/* skip to the next record; returns positive count on success */
+		error = xlog_seek_logrec_hdr(log, head_blk, first_bad, 2, bp,
+				&tmp_tail, &thead, &wrapped);
+		if (error < 0)
+			goto out;
 
+		*tail_blk = tmp_tail;
+		first_bad = 0;
+		error = xlog_do_recovery_pass(log, head_blk, *tail_blk,
+					      XLOG_RECOVER_CRCPASS, &first_bad);
+	}
+
+	if (!error && *tail_blk != orig_tail)
+		xfs_warn(log->l_mp,
+		"Tail block (0x%llx) overwrite detected. Updated to 0x%llx",
+			 orig_tail, *tail_blk);
 out:
 	xlog_put_bp(bp);
 	return error;
@@ -1143,7 +1198,7 @@ xlog_verify_head(
 	 */
 	error = xlog_do_recovery_pass(log, *head_blk, tmp_rhead_blk,
 				      XLOG_RECOVER_CRCPASS, &first_bad);
-	if (error == -EFSBADCRC) {
+	if ((error == -EFSBADCRC || error == -EFSCORRUPTED) && first_bad) {
 		/*
 		 * We've hit a potential torn write. Reset the error and warn
 		 * about it.
@@ -1183,31 +1238,12 @@ xlog_verify_head(
 			ASSERT(0);
 			return 0;
 		}
-
-		/*
-		 * Now verify the tail based on the updated head. This is
-		 * required because the torn writes trimmed from the head could
-		 * have been written over the tail of a previous record. Return
-		 * any errors since recovery cannot proceed if the tail is
-		 * corrupt.
-		 *
-		 * XXX: This leaves a gap in truly robust protection from torn
-		 * writes in the log. If the head is behind the tail, the tail
-		 * pushes forward to create some space and then a crash occurs
-		 * causing the writes into the previous record's tail region to
-		 * tear, log recovery isn't able to recover.
-		 *
-		 * How likely is this to occur? If possible, can we do something
-		 * more intelligent here? Is it safe to push the tail forward if
-		 * we can determine that the tail is within the range of the
-		 * torn write (e.g., the kernel can only overwrite the tail if
-		 * it has actually been pushed forward)? Alternatively, could we
-		 * somehow prevent this condition at runtime?
-		 */
-		error = xlog_verify_tail(log, *head_blk, *tail_blk);
 	}
+	if (error)
+		return error;
 
-	return error;
+	return xlog_verify_tail(log, *head_blk, tail_blk,
+				be32_to_cpu((*rhead)->h_size));
 }
 
 /*
@@ -2949,7 +2985,7 @@ xlog_recover_inode_pass2(
 	struct xlog_recover_item	*item,
 	xfs_lsn_t			current_lsn)
 {
-	xfs_inode_log_format_t	*in_f;
+	struct xfs_inode_log_format	*in_f;
 	xfs_mount_t		*mp = log->l_mp;
 	xfs_buf_t		*bp;
 	xfs_dinode_t		*dip;
@@ -2963,10 +2999,10 @@ xlog_recover_inode_pass2(
 	uint			isize;
 	int			need_free = 0;
 
-	if (item->ri_buf[0].i_len == sizeof(xfs_inode_log_format_t)) {
+	if (item->ri_buf[0].i_len == sizeof(struct xfs_inode_log_format)) {
 		in_f = item->ri_buf[0].i_addr;
 	} else {
-		in_f = kmem_alloc(sizeof(xfs_inode_log_format_t), KM_SLEEP);
+		in_f = kmem_alloc(sizeof(struct xfs_inode_log_format), KM_SLEEP);
 		need_free = 1;
 		error = xfs_inode_item_format_convert(&item->ri_buf[0], in_f);
 		if (error)
@@ -3137,16 +3173,8 @@ xlog_recover_inode_pass2(
 	}
 
 	fields = in_f->ilf_fields;
-	switch (fields & (XFS_ILOG_DEV | XFS_ILOG_UUID)) {
-	case XFS_ILOG_DEV:
+	if (fields & XFS_ILOG_DEV)
 		xfs_dinode_put_rdev(dip, in_f->ilf_u.ilfu_rdev);
-		break;
-	case XFS_ILOG_UUID:
-		memcpy(XFS_DFORK_DPTR(dip),
-		       &in_f->ilf_u.ilfu_uuid,
-		       sizeof(uuid_t));
-		break;
-	}
 
 	if (in_f->ilf_size == 2)
 		goto out_owner_change;
@@ -4271,7 +4299,7 @@ xlog_recover_add_to_trans(
 	char			*dp,
 	int			len)
 {
-	xfs_inode_log_format_t	*in_f;			/* any will do */
+	struct xfs_inode_log_format	*in_f;			/* any will do */
 	xlog_recover_item_t	*item;
 	char			*ptr;
 
@@ -4305,7 +4333,7 @@ xlog_recover_add_to_trans(
 
 	ptr = kmem_alloc(len, KM_SLEEP);
 	memcpy(ptr, dp, len);
-	in_f = (xfs_inode_log_format_t *)ptr;
+	in_f = (struct xfs_inode_log_format *)ptr;
 
 	/* take the tail entry */
 	item = list_entry(trans->r_itemq.prev, xlog_recover_item_t, ri_list);
@@ -4801,12 +4829,16 @@ xlog_recover_process_intents(
 	int			error = 0;
 	struct xfs_ail_cursor	cur;
 	struct xfs_ail		*ailp;
+#if defined(DEBUG) || defined(XFS_WARN)
 	xfs_lsn_t		last_lsn;
+#endif
 
 	ailp = log->l_ailp;
 	spin_lock(&ailp->xa_lock);
 	lip = xfs_trans_ail_cursor_first(ailp, &cur, 0);
+#if defined(DEBUG) || defined(XFS_WARN)
 	last_lsn = xlog_assign_lsn(log->l_curr_cycle, log->l_curr_block);
+#endif
 	while (lip != NULL) {
 		/*
 		 * We're done when we see something other than an intent.
@@ -5218,7 +5250,7 @@ xlog_do_recovery_pass(
 	xfs_daddr_t		*first_bad)	/* out: first bad log rec */
 {
 	xlog_rec_header_t	*rhead;
-	xfs_daddr_t		blk_no;
+	xfs_daddr_t		blk_no, rblk_no;
 	xfs_daddr_t		rhead_blk;
 	char			*offset;
 	xfs_buf_t		*hbp, *dbp;
@@ -5231,7 +5263,7 @@ xlog_do_recovery_pass(
 	LIST_HEAD		(buffer_list);
 
 	ASSERT(head_blk != tail_blk);
-	rhead_blk = 0;
+	blk_no = rhead_blk = tail_blk;
 
 	for (i = 0; i < XLOG_RHASH_SIZE; i++)
 		INIT_HLIST_HEAD(&rhash[i]);
@@ -5309,7 +5341,6 @@ xlog_do_recovery_pass(
 	}
 
 	memset(rhash, 0, sizeof(rhash));
-	blk_no = rhead_blk = tail_blk;
 	if (tail_blk > head_blk) {
 		/*
 		 * Perform recovery around the end of the physical log.
@@ -5371,9 +5402,19 @@ xlog_do_recovery_pass(
 			bblks = (int)BTOBB(be32_to_cpu(rhead->h_len));
 			blk_no += hblks;
 
-			/* Read in data for log record */
-			if (blk_no + bblks <= log->l_logBBsize) {
-				error = xlog_bread(log, blk_no, bblks, dbp,
+			/*
+			 * Read the log record data in multiple reads if it
+			 * wraps around the end of the log. Note that if the
+			 * header already wrapped, blk_no could point past the
+			 * end of the log. The record data is contiguous in
+			 * that case.
+			 */
+			if (blk_no + bblks <= log->l_logBBsize ||
+			    blk_no >= log->l_logBBsize) {
+				/* mod blk_no in case the header wrapped and
+				 * pushed it beyond the end of the log */
+				rblk_no = do_mod(blk_no, log->l_logBBsize);
+				error = xlog_bread(log, rblk_no, bblks, dbp,
 						   &offset);
 				if (error)
 					goto bread_err2;
@@ -5563,6 +5604,8 @@ xlog_do_recover(
 	xfs_buf_t	*bp;
 	xfs_sb_t	*sbp;
 
+	trace_xfs_log_recover(log, head_blk, tail_blk);
+
 	/*
 	 * First replay the images in the log.
 	 */
@@ -5782,7 +5825,7 @@ xlog_recover_cancel(
  * Read all of the agf and agi counters and check that they
  * are consistent with the superblock counters.
  */
-void
+STATIC void
 xlog_recover_check_summary(
 	struct xlog	*log)
 {