[XFS] Shutdown the filesystem if all device paths have gone. Made

shutdown vop flags consistent with sync vop flags declarations too.

SGI-PV: 939911
SGI-Modid: xfs-linux-melb:xfs-kern:26096a
Signed-off-by: Nathan Scott <nathans@sgi.com>

fs/xfs/linux-2.6/xfs_aops.c

@@ -136,9 +136,10 @@ xfs_destroy_ioend(
 
 	for (bh = ioend->io_buffer_head; bh; bh = next) {
 		next = bh->b_private;
-		bh->b_end_io(bh, ioend->io_uptodate);
+		bh->b_end_io(bh, !ioend->io_error);
 	}
-
+	if (unlikely(ioend->io_error))
+		vn_ioerror(ioend->io_vnode, ioend->io_error, __FILE__,__LINE__);
 	vn_iowake(ioend->io_vnode);
 	mempool_free(ioend, xfs_ioend_pool);
 }
@@ -185,7 +186,7 @@ xfs_end_bio_unwritten(
 	size_t			size = ioend->io_size;
 	int			error;
 
-	if (ioend->io_uptodate)
+	if (likely(!ioend->io_error))
 		VOP_BMAP(vp, offset, size, BMAPI_UNWRITTEN, NULL, NULL, error);
 	xfs_destroy_ioend(ioend);
 }
@@ -211,7 +212,7 @@ xfs_alloc_ioend(
 	 * all the I/O from calling the completion routine too early.
 	 */
 	atomic_set(&ioend->io_remaining, 1);
-	ioend->io_uptodate = 1; /* cleared if any I/O fails */
+	ioend->io_error = 0;
 	ioend->io_list = NULL;
 	ioend->io_type = type;
 	ioend->io_vnode = vn_from_inode(inode);
@@ -271,16 +272,14 @@ xfs_end_bio(
 	if (bio->bi_size)
 		return 1;
 
-	ASSERT(ioend);
 	ASSERT(atomic_read(&bio->bi_cnt) >= 1);
+	ioend->io_error = test_bit(BIO_UPTODATE, &bio->bi_flags) ? 0 : error;
 
 	/* Toss bio and pass work off to an xfsdatad thread */
-	if (!test_bit(BIO_UPTODATE, &bio->bi_flags))
-		ioend->io_uptodate = 0;
 	bio->bi_private = NULL;
 	bio->bi_end_io = NULL;
-
 	bio_put(bio);
+
 	xfs_finish_ioend(ioend);
 	return 0;
 }

fs/xfs/linux-2.6/xfs_aops.h

@@ -30,7 +30,7 @@ typedef void (*xfs_ioend_func_t)(void *);
 typedef struct xfs_ioend {
 	struct xfs_ioend	*io_list;	/* next ioend in chain */
 	unsigned int		io_type;	/* delalloc / unwritten */
-	unsigned int		io_uptodate;	/* I/O status register */
+	int			io_error;	/* I/O error code */
 	atomic_t		io_remaining;	/* hold count */
 	struct vnode		*io_vnode;	/* file being written to */
 	struct buffer_head	*io_buffer_head;/* buffer linked list head */

fs/xfs/linux-2.6/xfs_vfs.h

@@ -94,6 +94,13 @@ typedef enum {
 #define SYNC_REMOUNT		0x0080  /* remount readonly, no dummy LRs */
 #define SYNC_QUIESCE		0x0100  /* quiesce filesystem for a snapshot */
 
+#define SHUTDOWN_META_IO_ERROR	0x0001	/* write attempt to metadata failed */
+#define SHUTDOWN_LOG_IO_ERROR	0x0002	/* write attempt to the log failed */
+#define SHUTDOWN_FORCE_UMOUNT	0x0004	/* shutdown from a forced unmount */
+#define SHUTDOWN_CORRUPT_INCORE	0x0008	/* corrupt in-memory data structures */
+#define SHUTDOWN_REMOTE_REQ	0x0010	/* shutdown came from remote cell */
+#define SHUTDOWN_DEVICE_REQ	0x0020	/* failed all paths to the device */
+
 typedef int	(*vfs_mount_t)(bhv_desc_t *,
 				struct xfs_mount_args *, struct cred *);
 typedef int	(*vfs_parseargs_t)(bhv_desc_t *, char *,

fs/xfs/linux-2.6/xfs_vnode.c

@@ -54,6 +54,22 @@ vn_iowake(
 		wake_up(vptosync(vp));
 }
 
+/*
+ * Volume managers supporting multiple paths can send back ENODEV when the
+ * final path disappears.  In this case continuing to fill the page cache
+ * with dirty data which cannot be written out is evil, so prevent that.
+ */
+void
+vn_ioerror(
+	struct vnode	*vp,
+	int		error,
+	char		*f,
+	int		l)
+{
+	if (unlikely(error == -ENODEV))
+		VFS_FORCE_SHUTDOWN(vp->v_vfsp, SHUTDOWN_DEVICE_REQ, f, l);
+}
+
 struct vnode *
 vn_initialize(
 	struct inode	*inode)

fs/xfs/linux-2.6/xfs_vnode.h

@@ -514,6 +514,8 @@ extern void	vn_revalidate_core(struct vnode *, vattr_t *);
 extern void	vn_iowait(struct vnode *vp);
 extern void	vn_iowake(struct vnode *vp);
 
+extern void	vn_ioerror(struct vnode *vp, int error, char *f, int l);
+
 static inline int vn_count(struct vnode *vp)
 {
 	return atomic_read(&vn_to_inode(vp)->i_count);

fs/xfs/quota/xfs_dquot.c

@@ -1261,7 +1261,7 @@ xfs_qm_dqflush(
 
 	if (xfs_qm_dqcheck(&dqp->q_core, be32_to_cpu(ddqp->d_id),
 			   0, XFS_QMOPT_DOWARN, "dqflush (incore copy)")) {
-		xfs_force_shutdown(dqp->q_mount, XFS_CORRUPT_INCORE);
+		xfs_force_shutdown(dqp->q_mount, SHUTDOWN_CORRUPT_INCORE);
 		return XFS_ERROR(EIO);
 	}
 

fs/xfs/xfs_bmap.c

@@ -4263,8 +4263,8 @@ xfs_bmap_finish(
 			if (!XFS_FORCED_SHUTDOWN(mp))
 				xfs_force_shutdown(mp,
 						   (error == EFSCORRUPTED) ?
-						   XFS_CORRUPT_INCORE :
-						   XFS_METADATA_IO_ERROR);
+						   SHUTDOWN_CORRUPT_INCORE :
+						   SHUTDOWN_META_IO_ERROR);
 			return error;
 		}
 		xfs_trans_log_efd_extent(ntp, efd, free->xbfi_startblock,

fs/xfs/xfs_buf_item.c

@@ -1108,7 +1108,7 @@ xfs_buf_error_relse(
 	XFS_BUF_ERROR(bp,0);
 	xfs_buftrace("BUF_ERROR_RELSE", bp);
 	if (! XFS_FORCED_SHUTDOWN(mp))
-		xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
+		xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
 	/*
 	 * We have to unpin the pinned buffers so do the
 	 * callbacks.

fs/xfs/xfs_fsops.c

@@ -578,17 +578,18 @@ xfs_fs_goingdown(
 		struct super_block *sb = freeze_bdev(vfsp->vfs_super->s_bdev);
 
 		if (sb && !IS_ERR(sb)) {
-			xfs_force_shutdown(mp, XFS_FORCE_UMOUNT);
+			xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 			thaw_bdev(sb->s_bdev, sb);
 		}
 	
 		break;
 	}
 	case XFS_FSOP_GOING_FLAGS_LOGFLUSH:
-		xfs_force_shutdown(mp, XFS_FORCE_UMOUNT);
+		xfs_force_shutdown(mp, SHUTDOWN_FORCE_UMOUNT);
 		break;
 	case XFS_FSOP_GOING_FLAGS_NOLOGFLUSH:
-		xfs_force_shutdown(mp, XFS_FORCE_UMOUNT|XFS_LOG_IO_ERROR);
+		xfs_force_shutdown(mp,
+				SHUTDOWN_FORCE_UMOUNT | SHUTDOWN_LOG_IO_ERROR);
 		break;
 	default:
 		return XFS_ERROR(EINVAL);

fs/xfs/xfs_inode.c

@@ -3206,7 +3206,7 @@ xfs_iflush(
 
 corrupt_out:
 	xfs_buf_relse(bp);
-	xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
+	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 	xfs_iflush_abort(ip);
 	/*
 	 * Unlocks the flush lock
@@ -3228,7 +3228,7 @@ xfs_iflush(
 		xfs_buf_relse(bp);
 	}
 
-	xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
+	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 
 	if(!bufwasdelwri)  {
 		/*

fs/xfs/xfs_log.c

@@ -402,7 +402,7 @@ xfs_log_release_iclog(xfs_mount_t *mp,
 	xlog_in_core_t	  *iclog = (xlog_in_core_t *)iclog_hndl;
 
 	if (xlog_state_release_iclog(log, iclog)) {
-		xfs_force_shutdown(mp, XFS_LOG_IO_ERROR);
+		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
 		return EIO;
 	}
 
@@ -726,7 +726,7 @@ xfs_log_write(xfs_mount_t *	mp,
 		return XFS_ERROR(EIO);
 
 	if ((error = xlog_write(mp, reg, nentries, tic, start_lsn, NULL, 0))) {
-		xfs_force_shutdown(mp, XFS_LOG_IO_ERROR);
+		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
 	}
 	return error;
 }	/* xfs_log_write */
@@ -956,7 +956,7 @@ xlog_iodone(xfs_buf_t *bp)
 			XFS_ERRTAG_IODONE_IOERR, XFS_RANDOM_IODONE_IOERR)) {
 		xfs_ioerror_alert("xlog_iodone", l->l_mp, bp, XFS_BUF_ADDR(bp));
 		XFS_BUF_STALE(bp);
-		xfs_force_shutdown(l->l_mp, XFS_LOG_IO_ERROR);
+		xfs_force_shutdown(l->l_mp, SHUTDOWN_LOG_IO_ERROR);
 		/*
 		 * This flag will be propagated to the trans-committed
 		 * callback routines to let them know that the log-commit
@@ -1261,7 +1261,7 @@ xlog_commit_record(xfs_mount_t  *mp,
 	ASSERT_ALWAYS(iclog);
 	if ((error = xlog_write(mp, reg, 1, ticket, commitlsnp,
 			       iclog, XLOG_COMMIT_TRANS))) {
-		xfs_force_shutdown(mp, XFS_LOG_IO_ERROR);
+		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
 	}
 	return error;
 }	/* xlog_commit_record */
@@ -1790,7 +1790,7 @@ xlog_write(xfs_mount_t *	mp,
 	xfs_cmn_err(XFS_PTAG_LOGRES, CE_ALERT, mp,
 		"xfs_log_write: reservation ran out. Need to up reservation");
 	/* If we did not panic, shutdown the filesystem */
-	xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
+	xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 #endif
     } else
 	ticket->t_curr_res -= len;

fs/xfs/xfs_log_recover.c

@@ -282,7 +282,7 @@ xlog_recover_iodone(
 		mp = XFS_BUF_FSPRIVATE(bp, xfs_mount_t *);
 		xfs_ioerror_alert("xlog_recover_iodone",
 				  mp, bp, XFS_BUF_ADDR(bp));
-		xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
+		xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
 	}
 	XFS_BUF_SET_FSPRIVATE(bp, NULL);
 	XFS_BUF_CLR_IODONE_FUNC(bp);

fs/xfs/xfs_mount.h

@@ -513,15 +513,6 @@ xfs_preferred_iosize(xfs_mount_t *mp)
 #define xfs_force_shutdown(m,f)	\
 	VFS_FORCE_SHUTDOWN((XFS_MTOVFS(m)), f, __FILE__, __LINE__)
 
-/*
- * Flags sent to xfs_force_shutdown.
- */
-#define XFS_METADATA_IO_ERROR	0x1
-#define XFS_LOG_IO_ERROR	0x2
-#define XFS_FORCE_UMOUNT	0x4
-#define XFS_CORRUPT_INCORE	0x8	/* Corrupt in-memory data structures */
-#define XFS_SHUTDOWN_REMOTE_REQ 0x10	/* Shutdown came from remote cell */
-
 /*
  * Flags for xfs_mountfs
  */

fs/xfs/xfs_rw.c

@@ -109,12 +109,12 @@ xfs_do_force_shutdown(
 	xfs_mount_t	*mp;
 
 	mp = XFS_BHVTOM(bdp);
-	logerror = flags & XFS_LOG_IO_ERROR;
+	logerror = flags & SHUTDOWN_LOG_IO_ERROR;
 
-	if (!(flags & XFS_FORCE_UMOUNT)) {
-		cmn_err(CE_NOTE,
-		"xfs_force_shutdown(%s,0x%x) called from line %d of file %s.  Return address = 0x%p",
-			mp->m_fsname,flags,lnnum,fname,__return_address);
+	if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
+		cmn_err(CE_NOTE, "xfs_force_shutdown(%s,0x%x) called from "
+				 "line %d of file %s.  Return address = 0x%p",
+			mp->m_fsname, flags, lnnum, fname, __return_address);
 	}
 	/*
 	 * No need to duplicate efforts.
@@ -125,33 +125,37 @@ xfs_do_force_shutdown(
 	/*
 	 * This flags XFS_MOUNT_FS_SHUTDOWN, makes sure that we don't
 	 * queue up anybody new on the log reservations, and wakes up
-	 * everybody who's sleeping on log reservations and tells
-	 * them the bad news.
+	 * everybody who's sleeping on log reservations to tell them
+	 * the bad news.
 	 */
 	if (xfs_log_force_umount(mp, logerror))
 		return;
 
-	if (flags & XFS_CORRUPT_INCORE) {
+	if (flags & SHUTDOWN_CORRUPT_INCORE) {
 		xfs_cmn_err(XFS_PTAG_SHUTDOWN_CORRUPT, CE_ALERT, mp,
     "Corruption of in-memory data detected.  Shutting down filesystem: %s",
 			mp->m_fsname);
 		if (XFS_ERRLEVEL_HIGH <= xfs_error_level) {
 			xfs_stack_trace();
 		}
-	} else if (!(flags & XFS_FORCE_UMOUNT)) {
+	} else if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
 		if (logerror) {
 			xfs_cmn_err(XFS_PTAG_SHUTDOWN_LOGERROR, CE_ALERT, mp,
-			"Log I/O Error Detected.  Shutting down filesystem: %s",
+		"Log I/O Error Detected.  Shutting down filesystem: %s",
 				mp->m_fsname);
-		} else if (!(flags & XFS_SHUTDOWN_REMOTE_REQ)) {
+		} else if (flags & SHUTDOWN_DEVICE_REQ) {
 			xfs_cmn_err(XFS_PTAG_SHUTDOWN_IOERROR, CE_ALERT, mp,
-				"I/O Error Detected.  Shutting down filesystem: %s",
+		"All device paths lost.  Shutting down filesystem: %s",
+				mp->m_fsname);
+		} else if (!(flags & SHUTDOWN_REMOTE_REQ)) {
+			xfs_cmn_err(XFS_PTAG_SHUTDOWN_IOERROR, CE_ALERT, mp,
+		"I/O Error Detected.  Shutting down filesystem: %s",
 				mp->m_fsname);
 		}
 	}
-	if (!(flags & XFS_FORCE_UMOUNT)) {
-		cmn_err(CE_ALERT,
-		"Please umount the filesystem, and rectify the problem(s)");
+	if (!(flags & SHUTDOWN_FORCE_UMOUNT)) {
+		cmn_err(CE_ALERT, "Please umount the filesystem, "
+				  "and rectify the problem(s)");
 	}
 }
 
@@ -335,7 +339,7 @@ xfs_bwrite(
 		 * from bwrite and we could be tracing a buffer that has
 		 * been reused.
 		 */
-		xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
+		xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
 	}
 	return (error);
 }

fs/xfs/xfs_trans.c

@@ -846,7 +846,7 @@ _xfs_trans_commit(
 	 */
 	nvec = xfs_trans_count_vecs(tp);
 	if (nvec == 0) {
-		xfs_force_shutdown(mp, XFS_LOG_IO_ERROR);
+		xfs_force_shutdown(mp, SHUTDOWN_LOG_IO_ERROR);
 		goto shut_us_down;
 	} else if (nvec <= XFS_TRANS_LOGVEC_COUNT) {
 		log_vector = log_vector_fast;
@@ -1148,7 +1148,7 @@ xfs_trans_cancel(
 	 */
 	if ((tp->t_flags & XFS_TRANS_DIRTY) && !XFS_FORCED_SHUTDOWN(mp)) {
 		XFS_ERROR_REPORT("xfs_trans_cancel", XFS_ERRLEVEL_LOW, mp);
-		xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
+		xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 	}
 #ifdef DEBUG
 	if (!(flags & XFS_TRANS_ABORT)) {

fs/xfs/xfs_trans_ail.c

@@ -363,9 +363,10 @@ xfs_trans_delete_ail(
 			AIL_UNLOCK(mp, s);
 		else {
 			xfs_cmn_err(XFS_PTAG_AILDELETE, CE_ALERT, mp,
-				"xfs_trans_delete_ail: attempting to delete a log item that is not in the AIL");
+		"%s: attempting to delete a log item that is not in the AIL",
+					__FUNCTION__);
 			AIL_UNLOCK(mp, s);
-			xfs_force_shutdown(mp, XFS_CORRUPT_INCORE);
+			xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
 		}
 	}
 }

fs/xfs/xfs_trans_buf.c

@@ -369,7 +369,7 @@ xfs_trans_read_buf(
 				 */
 				if (tp->t_flags & XFS_TRANS_DIRTY)
 					xfs_force_shutdown(tp->t_mountp,
-							   XFS_METADATA_IO_ERROR);
+							SHUTDOWN_META_IO_ERROR);
 				return error;
 			}
 		}
@@ -414,7 +414,7 @@ xfs_trans_read_buf(
 		xfs_ioerror_alert("xfs_trans_read_buf", mp,
 				  bp, blkno);
 		if (tp->t_flags & XFS_TRANS_DIRTY)
-			xfs_force_shutdown(tp->t_mountp, XFS_METADATA_IO_ERROR);
+			xfs_force_shutdown(tp->t_mountp, SHUTDOWN_META_IO_ERROR);
 		xfs_buf_relse(bp);
 		return error;
 	}
@@ -423,7 +423,7 @@ xfs_trans_read_buf(
 		if (xfs_error_target == target) {
 			if (((xfs_req_num++) % xfs_error_mod) == 0) {
 				xfs_force_shutdown(tp->t_mountp,
-						   XFS_METADATA_IO_ERROR);
+						   SHUTDOWN_META_IO_ERROR);
 				xfs_buf_relse(bp);
 				printk("Returning error in trans!\n");
 				return XFS_ERROR(EIO);

fs/xfs/xfs_vnodeops.c

@@ -1761,7 +1761,7 @@ xfs_inactive(
 			cmn_err(CE_NOTE,
 		"xfs_inactive:	xfs_ifree() returned an error = %d on %s",
 				error, mp->m_fsname);
-			xfs_force_shutdown(mp, XFS_METADATA_IO_ERROR);
+			xfs_force_shutdown(mp, SHUTDOWN_META_IO_ERROR);
 		}
 		xfs_trans_cancel(tp, XFS_TRANS_RELEASE_LOG_RES|XFS_TRANS_ABORT);
 	} else {