Merge tag 'xfs-for-linus-4.5' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs

Pull xfs updates from Dave Chinner:
 "There's not a lot in this - the main addition is the CRC validation of
  the entire region of the log that the will be recovered, along with
  several log recovery fixes.  Most of the rest is small bug fixes and
  cleanups.

  I have three bug fixes still pending, all that address recently fixed
  regressions that I will send to next week after they've had some time
  in for-next.

  Summary:
   - extensive CRC validation during log recovery
   - several log recovery bug fixes
   - Various DAX support fixes
   - AGFL size calculation fix
   - various cleanups in preparation for new functionality
   - project quota ENOSPC notification via netlink
   - tracing and debug improvements"

* tag 'xfs-for-linus-4.5' of git://git.kernel.org/pub/scm/linux/kernel/git/dgc/linux-xfs: (26 commits)
  xfs: handle dquot buffer readahead in log recovery correctly
  xfs: inode recovery readahead can race with inode buffer creation
  xfs: eliminate committed arg from xfs_bmap_finish
  xfs: bmapbt checking on debug kernels too expensive
  xfs: add tracepoints to readpage calls
  xfs: debug mode log record crc error injection
  xfs: detect and trim torn writes during log recovery
  xfs: fix recursive splice read locking with DAX
  xfs: Don't use reserved blocks for data blocks with DAX
  XFS: Use a signed return type for suffix_kstrtoint()
  libxfs: refactor short btree block verification
  libxfs: pack the agfl header structure so XFS_AGFL_SIZE is correct
  libxfs: use a convenience variable instead of open-coding the fork
  xfs: fix log ticket type printing
  libxfs: make xfs_alloc_fix_freelist non-static
  xfs: make xfs_buf_ioend_async() static
  xfs: send warning of project quota to userspace via netlink
  xfs: get mp from bma->ip in xfs_bmap code
  xfs: print name of verifier if it fails
  libxfs: Optimize the loop for xfs_bitmap_empty
  ...

1 files changed, 42 insertions, 9 deletions

diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c
index f52c72a1a06f..9c9a1c9bcc7f 100644
--- a/fs/xfs/xfs_log.c
+++ b/fs/xfs/xfs_log.c
@@ -1188,10 +1188,16 @@ xlog_iodone(xfs_buf_t *bp)
 	int			aborted = 0;
 
 	/*
-	 * Race to shutdown the filesystem if we see an error.
+	 * Race to shutdown the filesystem if we see an error or the iclog is in
+	 * IOABORT state. The IOABORT state is only set in DEBUG mode to inject
+	 * CRC errors into log recovery.
 	 */
-	if (XFS_TEST_ERROR(bp->b_error, l->l_mp,
-			XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
+	if (XFS_TEST_ERROR(bp->b_error, l->l_mp, XFS_ERRTAG_IODONE_IOERR,
+			   XFS_RANDOM_IODONE_IOERR) ||
+	    iclog->ic_state & XLOG_STATE_IOABORT) {
+		if (iclog->ic_state & XLOG_STATE_IOABORT)
+			iclog->ic_state &= ~XLOG_STATE_IOABORT;
+
 		xfs_buf_ioerror_alert(bp, __func__);
 		xfs_buf_stale(bp);
 		xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
@@ -1838,6 +1844,23 @@ xlog_sync(
 	/* calculcate the checksum */
 	iclog->ic_header.h_crc = xlog_cksum(log, &iclog->ic_header,
 					    iclog->ic_datap, size);
+#ifdef DEBUG
+	/*
+	 * Intentionally corrupt the log record CRC based on the error injection
+	 * frequency, if defined. This facilitates testing log recovery in the
+	 * event of torn writes. Hence, set the IOABORT state to abort the log
+	 * write on I/O completion and shutdown the fs. The subsequent mount
+	 * detects the bad CRC and attempts to recover.
+	 */
+	if (log->l_badcrc_factor &&
+	    (prandom_u32() % log->l_badcrc_factor == 0)) {
+		iclog->ic_header.h_crc &= 0xAAAAAAAA;
+		iclog->ic_state |= XLOG_STATE_IOABORT;
+		xfs_warn(log->l_mp,
+	"Intentionally corrupted log record at LSN 0x%llx. Shutdown imminent.",
+			 be64_to_cpu(iclog->ic_header.h_lsn));
+	}
+#endif
 
 	bp->b_io_length = BTOBB(count);
 	bp->b_fspriv = iclog;
@@ -2045,12 +2068,14 @@ xlog_print_tic_res(
 	    "QM_DQCLUSTER",
 	    "QM_QINOCREATE",
 	    "QM_QUOTAOFF_END",
-	    "SB_UNIT",
 	    "FSYNC_TS",
 	    "GROWFSRT_ALLOC",
 	    "GROWFSRT_ZERO",
 	    "GROWFSRT_FREE",
-	    "SWAPEXT"
+	    "SWAPEXT",
+	    "CHECKPOINT",
+	    "ICREATE",
+	    "CREATE_TMPFILE"
 	};
 
 	xfs_warn(mp, "xlog_write: reservation summary:");
@@ -2791,11 +2816,19 @@ xlog_state_do_callback(
 		}
 	} while (!ioerrors && loopdidcallbacks);
 
+#ifdef DEBUG
 	/*
-	 * make one last gasp attempt to see if iclogs are being left in
-	 * limbo..
+	 * Make one last gasp attempt to see if iclogs are being left in limbo.
+	 * If the above loop finds an iclog earlier than the current iclog and
+	 * in one of the syncing states, the current iclog is put into
+	 * DO_CALLBACK and the callbacks are deferred to the completion of the
+	 * earlier iclog. Walk the iclogs in order and make sure that no iclog
+	 * is in DO_CALLBACK unless an earlier iclog is in one of the syncing
+	 * states.
+	 *
+	 * Note that SYNCING|IOABORT is a valid state so we cannot just check
+	 * for ic_state == SYNCING.
 	 */
-#ifdef DEBUG
 	if (funcdidcallbacks) {
 		first_iclog = iclog = log->l_iclog;
 		do {
@@ -2810,7 +2843,7 @@ xlog_state_do_callback(
 			 * IOERROR - give up hope all ye who enter here
 			 */
 			if (iclog->ic_state == XLOG_STATE_WANT_SYNC ||
-			    iclog->ic_state == XLOG_STATE_SYNCING ||
+			    iclog->ic_state & XLOG_STATE_SYNCING ||
 			    iclog->ic_state == XLOG_STATE_DONE_SYNC ||
 			    iclog->ic_state == XLOG_STATE_IOERROR )
 				break;