Merge master.kernel.org:/home/hch/BK/xfs/linux-2.5

into home.transmeta.com:/home/torvalds/v2.5/linux

fs/xfs/pagebuf/page_buf.c

@@ -1299,7 +1299,6 @@ pagebuf_iorequest(			/* start real I/O		*/
 	int			status = 0;
 	int			i, map_i, total_nr_pages, nr_pages;
 	struct bio		*bio;
-	struct bio_vec		*bvec;
 	int			offset = pb->pb_offset;
 	int			size = pb->pb_count_desired;
 	sector_t		sector = pb->pb_bn;
@@ -1335,13 +1334,8 @@ pagebuf_iorequest(			/* start real I/O		*/
 		bio->bi_sector = sector - (offset >> BBSHIFT);
 		bio->bi_end_io = bio_end_io_pagebuf;
 		bio->bi_private = pb;
-		bio->bi_vcnt++;
-		bio->bi_size = PAGE_CACHE_SIZE;
 
-		bvec = bio->bi_io_vec;
-		bvec->bv_page = pb->pb_pages[0];
-		bvec->bv_len = PAGE_CACHE_SIZE;
-		bvec->bv_offset = 0;
+		bio_add_page(bio, pb->pb_pages[0], PAGE_CACHE_SIZE, 0);
 
 		atomic_inc(&pb->pb_io_remaining);
 		submit_bio(READ, bio);
@@ -1389,9 +1383,7 @@ pagebuf_iorequest(			/* start real I/O		*/
 	bio->bi_end_io = bio_end_io_pagebuf;
 	bio->bi_private = pb;
 
-	bvec = bio->bi_io_vec;
-
-	for (; size && nr_pages; nr_pages--, bvec++, map_i++) {
+	for (; size && nr_pages; nr_pages--, map_i++) {
 		int	nbytes = PAGE_CACHE_SIZE - offset;
 
 		if (nbytes > size)

fs/xfs/xfs_buf_item.c

@@ -379,7 +379,8 @@ xfs_buf_item_pin(
  */
 void
 xfs_buf_item_unpin(
-	xfs_buf_log_item_t	*bip)
+	xfs_buf_log_item_t	*bip,
+	int			stale)
 {
 	xfs_mount_t	*mp;
 	xfs_buf_t	*bp;
@@ -396,7 +397,8 @@ xfs_buf_item_unpin(
 	freed = atomic_dec_and_test(&bip->bli_refcount);
 	mp = bip->bli_item.li_mountp;
 	xfs_bunpin(bp);
-	if (freed && (bip->bli_flags & XFS_BLI_STALE)) {
+	if (freed && stale) {
+		ASSERT(bip->bli_flags & XFS_BLI_STALE);
 		ASSERT(XFS_BUF_VALUSEMA(bp) <= 0);
 		ASSERT(!(XFS_BUF_ISDELAYWRITE(bp)));
 		ASSERT(XFS_BUF_ISSTALE(bp));
@@ -418,7 +420,6 @@ xfs_buf_item_unpin(
 		ASSERT(XFS_BUF_FSPRIVATE(bp, void *) == NULL);
 		xfs_buf_relse(bp);
 	}
-
 }
 
 /*
@@ -435,6 +436,7 @@ xfs_buf_item_unpin_remove(
 {
 	xfs_buf_t		*bp;
 	xfs_log_item_desc_t	*lidp;
+	int			stale = 0;
 
 	bp = bip->bli_buf;
 	/*
@@ -454,6 +456,7 @@ xfs_buf_item_unpin_remove(
 		 * will be able to bump up the refcount.
 		 */
 		lidp = xfs_trans_find_item(tp, (xfs_log_item_t *) bip);
+		stale = lidp->lid_flags & XFS_LID_BUF_STALE;
 		xfs_trans_free_item(tp, lidp);
 		/*
 		 * Since the transaction no longer refers to the buffer,
@@ -462,7 +465,7 @@ xfs_buf_item_unpin_remove(
 		XFS_BUF_SET_FSPRIVATE2(bp, NULL);
 	}
 
-	xfs_buf_item_unpin(bip);
+	xfs_buf_item_unpin(bip, stale);
 
 	return;
 }
@@ -686,7 +689,7 @@ struct xfs_item_ops xfs_buf_item_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_buf_item_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_buf_item_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_buf_item_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t*, int))xfs_buf_item_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *))
 					xfs_buf_item_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_buf_item_trylock,

fs/xfs/xfs_dquot_item.c

@@ -93,9 +93,11 @@ xfs_qm_dquot_logitem_pin(
  * anyone in xfs_dqwait_unpin() if the count goes to 0.	 The
  * dquot must have been previously pinned with a call to xfs_dqpin().
  */
+/* ARGSUSED */
 STATIC void
 xfs_qm_dquot_logitem_unpin(
-	xfs_dq_logitem_t *logitem)
+	xfs_dq_logitem_t *logitem,
+	int		  stale)
 {
 	unsigned long	s;
 	xfs_dquot_t *dqp;
@@ -116,7 +118,7 @@ xfs_qm_dquot_logitem_unpin_remove(
 	xfs_dq_logitem_t *logitem,
 	xfs_trans_t	 *tp)
 {
-	xfs_qm_dquot_logitem_unpin(logitem);
+	xfs_qm_dquot_logitem_unpin(logitem, 0);
 }
 
 /*
@@ -396,7 +398,8 @@ struct xfs_item_ops xfs_dquot_item_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_qm_dquot_logitem_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_qm_dquot_logitem_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_qm_dquot_logitem_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t*, int))
+					xfs_qm_dquot_logitem_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t*))
 					xfs_qm_dquot_logitem_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))
@@ -492,7 +495,7 @@ xfs_qm_qoff_logitem_pin(xfs_qoff_logitem_t *qf)
  */
 /*ARGSUSED*/
 STATIC void
-xfs_qm_qoff_logitem_unpin(xfs_qoff_logitem_t *qf)
+xfs_qm_qoff_logitem_unpin(xfs_qoff_logitem_t *qf, int stale)
 {
 	return;
 }
@@ -613,7 +616,8 @@ struct xfs_item_ops xfs_qm_qoffend_logitem_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_qm_qoff_logitem_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t* ,int))
+					xfs_qm_qoff_logitem_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*,xfs_trans_t*))
 					xfs_qm_qoff_logitem_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_trylock,
@@ -635,7 +639,8 @@ struct xfs_item_ops xfs_qm_qoff_logitem_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_qm_qoff_logitem_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t*, int))
+					xfs_qm_qoff_logitem_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*,xfs_trans_t*))
 					xfs_qm_qoff_logitem_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_qm_qoff_logitem_trylock,

fs/xfs/xfs_extfree_item.c

@@ -105,7 +105,7 @@ xfs_efi_item_pin(xfs_efi_log_item_t *efip)
  */
 /*ARGSUSED*/
 STATIC void
-xfs_efi_item_unpin(xfs_efi_log_item_t *efip)
+xfs_efi_item_unpin(xfs_efi_log_item_t *efip, int stale)
 {
 	int		nexts;
 	int		size;
@@ -280,7 +280,7 @@ struct xfs_item_ops xfs_efi_item_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_efi_item_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_efi_item_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_efi_item_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t*, int))xfs_efi_item_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t *))
 					xfs_efi_item_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_efi_item_trylock,
@@ -474,7 +474,7 @@ xfs_efd_item_pin(xfs_efd_log_item_t *efdp)
  */
 /*ARGSUSED*/
 STATIC void
-xfs_efd_item_unpin(xfs_efd_log_item_t *efdp)
+xfs_efd_item_unpin(xfs_efd_log_item_t *efdp, int stale)
 {
 	return;
 }
@@ -607,7 +607,7 @@ struct xfs_item_ops xfs_efd_item_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_efd_item_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_efd_item_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_efd_item_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t*, int))xfs_efd_item_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t*))
 					xfs_efd_item_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_efd_item_trylock,

fs/xfs/xfs_inode_item.c

@@ -534,9 +534,11 @@ xfs_inode_item_pin(
  * item which was previously pinned with a call to xfs_inode_item_pin().
  * Just call xfs_iunpin() on the inode to do this.
  */
+/* ARGSUSED */
 STATIC void
 xfs_inode_item_unpin(
-	xfs_inode_log_item_t	*iip)
+	xfs_inode_log_item_t	*iip,
+	int			stale)
 {
 	xfs_iunpin(iip->ili_inode);
 }
@@ -880,7 +882,7 @@ struct xfs_item_ops xfs_inode_item_ops = {
 	.iop_format	= (void(*)(xfs_log_item_t*, xfs_log_iovec_t*))
 					xfs_inode_item_format,
 	.iop_pin	= (void(*)(xfs_log_item_t*))xfs_inode_item_pin,
-	.iop_unpin	= (void(*)(xfs_log_item_t*))xfs_inode_item_unpin,
+	.iop_unpin	= (void(*)(xfs_log_item_t*, int))xfs_inode_item_unpin,
 	.iop_unpin_remove = (void(*)(xfs_log_item_t*, xfs_trans_t*))
 					xfs_inode_item_unpin_remove,
 	.iop_trylock	= (uint(*)(xfs_log_item_t*))xfs_inode_item_trylock,

fs/xfs/xfs_log.c

@@ -45,7 +45,7 @@
 /* Local miscellaneous function prototypes */
 STATIC int	 xlog_bdstrat_cb(struct xfs_buf *);
 STATIC int	 xlog_commit_record(xfs_mount_t *mp, xlog_ticket_t *ticket,
-				    xfs_lsn_t *);
+				    xlog_in_core_t **, xfs_lsn_t *);
 STATIC xlog_t *	 xlog_alloc_log(xfs_mount_t	*mp,
 				dev_t		log_dev,
 				xfs_daddr_t	blk_offset,
@@ -55,7 +55,9 @@ STATIC int	 xlog_sync(xlog_t *log, xlog_in_core_t *iclog);
 STATIC void	 xlog_unalloc_log(xlog_t *log);
 STATIC int	 xlog_write(xfs_mount_t *mp, xfs_log_iovec_t region[],
 			    int nentries, xfs_log_ticket_t tic,
-			    xfs_lsn_t *start_lsn, uint flags);
+			    xfs_lsn_t *start_lsn,
+			    xlog_in_core_t **commit_iclog,
+			    uint flags);
 
 /* local state machine functions */
 STATIC void xlog_state_done_syncing(xlog_in_core_t *iclog, int);
@@ -70,10 +72,6 @@ STATIC int  xlog_state_get_iclog_space(xlog_t		*log,
 				       xlog_ticket_t	*ticket,
 				       int		*continued_write,
 				       int		*logoffsetp);
-STATIC int  xlog_state_lsn_is_synced(xlog_t		*log,
-				     xfs_lsn_t		lsn,
-				     xfs_log_callback_t *cb,
-				     int		*abortflg);
 STATIC void xlog_state_put_ticket(xlog_t	*log,
 				  xlog_ticket_t *tic);
 STATIC int  xlog_state_release_iclog(xlog_t		*log,
@@ -254,6 +252,7 @@ xlog_trace_iclog(xlog_in_core_t *iclog, uint state)
 xfs_lsn_t
 xfs_log_done(xfs_mount_t	*mp,
 	     xfs_log_ticket_t	xtic,
+	     void		**iclog,
 	     uint		flags)
 {
 	xlog_t		*log	= mp->m_log;
@@ -271,7 +270,8 @@ xfs_log_done(xfs_mount_t	*mp,
 	     * If we get an error, just continue and give back the log ticket.
 	     */
 	    (((ticket->t_flags & XLOG_TIC_INITED) == 0) &&
-	     (xlog_commit_record(mp, ticket, &lsn)))) {
+	     (xlog_commit_record(mp, ticket,
+				 (xlog_in_core_t **)iclog, &lsn)))) {
 		lsn = (xfs_lsn_t) -1;
 		if (ticket->t_flags & XLOG_TIC_PERM_RESERV) {
 			flags |= XFS_LOG_REL_PERM_RESERV;
@@ -354,21 +354,39 @@ xfs_log_force(xfs_mount_t *mp,
  * been synced to disk, we add the callback to the callback list of the
  * in-core log.
  */
-void
+int
 xfs_log_notify(xfs_mount_t	  *mp,		/* mount of partition */
-	       xfs_lsn_t	  lsn,		/* lsn looking for */
+	       void		  *iclog_hndl,	/* iclog to hang callback off */
 	       xfs_log_callback_t *cb)
 {
 	xlog_t *log = mp->m_log;
-	int	abortflg;
+	xlog_in_core_t	  *iclog = (xlog_in_core_t *)iclog_hndl;
+	int	abortflg, spl;
 
 #if defined(DEBUG) || defined(XLOG_NOLOG)
 	if (! xlog_debug && xlog_devt == log->l_dev)
-		return;
+		return 0;
 #endif
 	cb->cb_next = 0;
-	if (xlog_state_lsn_is_synced(log, lsn, cb, &abortflg))
+	spl = LOG_LOCK(log);
+	abortflg = (iclog->ic_state & XLOG_STATE_IOERROR);
+	if (!abortflg) {
+		ASSERT_ALWAYS((iclog->ic_state == XLOG_STATE_ACTIVE) ||
+			      (iclog->ic_state == XLOG_STATE_WANT_SYNC));
+		cb->cb_next = 0;
+		*(iclog->ic_callback_tail) = cb;
+		iclog->ic_callback_tail = &(cb->cb_next);
+	}
+	LOG_UNLOCK(log, spl);
+	if (!abortflg) {
+		if (xlog_state_release_iclog(log, iclog)) {
+			xfs_force_shutdown(mp, XFS_LOG_IO_ERROR);
+			return EIO;
+		}
+	} else {
 		cb->cb_func(cb->cb_arg, abortflg);
+	}
+	return 0;
 }	/* xfs_log_notify */
 
 
@@ -611,7 +629,7 @@ xfs_log_unmount_write(xfs_mount_t *mp)
 			/* remove inited flag */
 			((xlog_ticket_t *)tic)->t_flags = 0;
 			error = xlog_write(mp, reg, 1, tic, &lsn,
-					   XLOG_UNMOUNT_TRANS);
+					   NULL, XLOG_UNMOUNT_TRANS);
 			/*
 			 * At this point, we're umounting anyway,
 			 * so there's no point in transitioning log state
@@ -717,7 +735,7 @@ xfs_log_write(xfs_mount_t *	mp,
 	if (XLOG_FORCED_SHUTDOWN(log))
 		return XFS_ERROR(EIO);
 
-	if ((error = xlog_write(mp, reg, nentries, tic, start_lsn, 0))) {
+	if ((error = xlog_write(mp, reg, nentries, tic, start_lsn, NULL, 0))) {
 		xfs_force_shutdown(mp, XFS_LOG_IO_ERROR);
 	}
 	return (error);
@@ -1259,6 +1277,7 @@ xlog_alloc_log(xfs_mount_t	*mp,
 STATIC int
 xlog_commit_record(xfs_mount_t	*mp,
 		   xlog_ticket_t *ticket,
+		   xlog_in_core_t **iclog,
 		   xfs_lsn_t	*commitlsnp)
 {
 	int		error;
@@ -1267,8 +1286,9 @@ xlog_commit_record(xfs_mount_t	*mp,
 	reg[0].i_addr = 0;
 	reg[0].i_len = 0;
 
+	ASSERT_ALWAYS(iclog);
 	if ((error = xlog_write(mp, reg, 1, ticket, commitlsnp,
-			       XLOG_COMMIT_TRANS))) {
+			       iclog, XLOG_COMMIT_TRANS))) {
 		xfs_force_shutdown(mp, XFS_LOG_IO_ERROR);
 	}
 	return (error);
@@ -1614,6 +1634,7 @@ xlog_write(xfs_mount_t *	mp,
 	   int			nentries,
 	   xfs_log_ticket_t	tic,
 	   xfs_lsn_t		*start_lsn,
+	   xlog_in_core_t	**commit_iclog,
 	   uint			flags)
 {
     xlog_t	     *log    = mp->m_log;
@@ -1776,7 +1797,10 @@ xlog_write(xfs_mount_t *	mp,
 
 		if (iclog->ic_size - log_offset <= sizeof(xlog_op_header_t)) {
 		    xlog_state_want_sync(log, iclog);
-		    if ((error = xlog_state_release_iclog(log, iclog)))
+		    if (commit_iclog) {
+			ASSERT(flags & XLOG_COMMIT_TRANS);
+			*commit_iclog = iclog;
+		    } else if ((error = xlog_state_release_iclog(log, iclog)))
 			   return (error);
 		    if (index == nentries)
 			    return 0;		/* we are done */
@@ -1788,6 +1812,11 @@ xlog_write(xfs_mount_t *	mp,
     } /* for (index = 0; index < nentries; ) */
     ASSERT(len == 0);
 
+    if (commit_iclog) {
+	ASSERT(flags & XLOG_COMMIT_TRANS);
+	*commit_iclog = iclog;
+	return 0;
+    }
     return (xlog_state_release_iclog(log, iclog));
 }	/* xlog_write */
 
@@ -2060,6 +2089,12 @@ xlog_state_do_callback(
 			if (!(iclog->ic_state & XLOG_STATE_IOERROR))
 				iclog->ic_state = XLOG_STATE_DIRTY;
 
+			/*
+			 * Transition from DIRTY to ACTIVE if applicable.
+			 * NOP if STATE_IOERROR.
+			 */
+			xlog_state_clean_log(log);
+
 			/* wake up threads waiting in xfs_log_force() */
 			sv_broadcast(&iclog->ic_forcesema);
 
@@ -2098,12 +2133,6 @@ xlog_state_do_callback(
 	}
 #endif
 
-	/*
-	 * Transition from DIRTY to ACTIVE if applicable. NOP if
-	 * STATE_IOERROR.
-	 */
-	xlog_state_clean_log(log);
-
 	if (log->l_iclog->ic_state & (XLOG_STATE_ACTIVE|XLOG_STATE_IOERROR)) {
 		flushcnt = log->l_flushcnt;
 		log->l_flushcnt = 0;
@@ -2654,52 +2683,6 @@ xlog_ungrant_log_space(xlog_t	     *log,
 }	/* xlog_ungrant_log_space */
 
 
-/*
- * If the lsn is not found or the iclog with the lsn is in the callback
- * state, we need to call the function directly.  This is done outside
- * this function's scope.  Otherwise, we insert the callback at the end
- * of the iclog's callback list.
- */
-int
-xlog_state_lsn_is_synced(xlog_t		    *log,
-			 xfs_lsn_t	    lsn,
-			 xfs_log_callback_t *cb,
-			 int		    *abortflg)
-{
-	xlog_in_core_t *iclog;
-	SPLDECL(s);
-	int	      lsn_is_synced = 1;
-
-	*abortflg = 0;
-	s = LOG_LOCK(log);
-
-	iclog = log->l_iclog;
-	do {
-		if (INT_GET(iclog->ic_header.h_lsn, ARCH_CONVERT) != lsn) {
-			iclog = iclog->ic_next;
-			continue;
-		} else {
-			if (iclog->ic_state & XLOG_STATE_DIRTY) /* call it*/
-				break;
-
-			if (iclog->ic_state & XLOG_STATE_IOERROR) {
-				*abortflg = XFS_LI_ABORTED;
-				break;
-			}
-			/* insert callback onto end of list */
-			cb->cb_next = 0;
-			*(iclog->ic_callback_tail) = cb;
-			iclog->ic_callback_tail = &(cb->cb_next);
-			lsn_is_synced = 0;
-			break;
-		}
-	} while (iclog != log->l_iclog);
-
-	LOG_UNLOCK(log, s);
-	return lsn_is_synced;
-}	/* xlog_state_lsn_is_synced */
-
-
 /*
  * Atomically put back used ticket.
  */

fs/xfs/xfs_log.h

@@ -148,6 +148,7 @@ typedef struct xfs_log_callback {
 struct xfs_mount;
 xfs_lsn_t xfs_log_done(struct xfs_mount *mp,
 		       xfs_log_ticket_t ticket,
+		       void		**iclog,
 		       uint		flags);
 int	  xfs_log_force(struct xfs_mount *mp,
 			xfs_lsn_t	 lsn,
@@ -160,8 +161,8 @@ int	  xfs_log_mount(struct xfs_mount *mp,
 int	  xfs_log_mount_finish(struct xfs_mount *mp, int);
 void	  xfs_log_move_tail(struct xfs_mount	*mp,
 			    xfs_lsn_t		tail_lsn);
-void	  xfs_log_notify(struct xfs_mount	*mp,
-			 xfs_lsn_t		lsn,
+int	  xfs_log_notify(struct xfs_mount	*mp,
+			 void			*iclog,
 			 xfs_log_callback_t	*callback_entry);
 int	  xfs_log_reserve(struct xfs_mount *mp,
 			  int		   length,

fs/xfs/xfs_trans.c

@@ -285,7 +285,7 @@ xfs_trans_reserve(
 		} else {
 			log_flags = 0;
 		}
-		xfs_log_done(tp->t_mountp, tp->t_ticket, log_flags);
+		xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
 		tp->t_ticket = NULL;
 		tp->t_log_res = 0;
 		tp->t_flags &= ~XFS_TRANS_PERM_LOG_RES;
@@ -669,6 +669,7 @@ xfs_trans_commit(
 #if defined(XLOG_NOLOG) || defined(DEBUG)
 	static xfs_lsn_t	trans_lsn = 1;
 #endif
+	void			*commit_iclog;
 	int			shutdown;
 
 	commit_lsn = -1;
@@ -706,7 +707,8 @@ xfs_trans_commit(
 			xfs_trans_unreserve_and_mod_dquots(tp);
 		}
 		if (tp->t_ticket) {
-			commit_lsn = xfs_log_done(mp, tp->t_ticket, log_flags);
+			commit_lsn = xfs_log_done(mp, tp->t_ticket,
+							NULL, log_flags);
 			if (commit_lsn == -1 && !shutdown)
 				shutdown = XFS_ERROR(EIO);
 		}
@@ -773,7 +775,7 @@ xfs_trans_commit(
 #if defined(XLOG_NOLOG) || defined(DEBUG)
 	if (xlog_debug) {
 		commit_lsn = xfs_log_done(mp, tp->t_ticket,
-					  log_flags);
+					  &commit_iclog, log_flags);
 	} else {
 		commit_lsn = 0;
 		tp->t_lsn = trans_lsn++;
@@ -785,7 +787,7 @@ xfs_trans_commit(
 	 * any time.  However, all the items associated with the transaction
 	 * are still locked and pinned in memory.
 	 */
-	commit_lsn = xfs_log_done(mp, tp->t_ticket, log_flags);
+	commit_lsn = xfs_log_done(mp, tp->t_ticket, &commit_iclog, log_flags);
 #endif
 
 	tp->t_commit_lsn = commit_lsn;
@@ -845,20 +847,27 @@ xfs_trans_commit(
 	if (xlog_debug) {
 		tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed;
 		tp->t_logcb.cb_arg = tp;
-		xfs_log_notify(mp, commit_lsn, &(tp->t_logcb));
+		error = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb));
 	} else {
 		xfs_trans_committed(tp, 0);
 	}
 #else
 	tp->t_logcb.cb_func = (void(*)(void*, int))xfs_trans_committed;
 	tp->t_logcb.cb_arg = tp;
-	xfs_log_notify(mp, commit_lsn, &(tp->t_logcb));
+
+	/* We need to pass the iclog buffer which was used for the
+	 * transaction commit record into this function, attach
+	 * the callback to it, and then release it. This will guarantee
+	 * that we do callbacks on the transaction in the correct order.
+	 */
+	error = xfs_log_notify(mp, commit_iclog, &(tp->t_logcb));
 #endif
 	/*
 	 * If the transaction needs to be synchronous, then force the
 	 * log out now and wait for it.
 	 */
 	if (sync) {
+		if (!error)
 			error = xfs_log_force(mp, commit_lsn,
 				      XFS_LOG_FORCE | XFS_LOG_SYNC);
 		XFS_STATS_INC(xfsstats.xs_trans_sync);
@@ -1070,7 +1079,7 @@ xfs_trans_cancel(
 		} else {
 			log_flags = 0;
 		}
-		xfs_log_done(tp->t_mountp, tp->t_ticket, log_flags);
+		xfs_log_done(tp->t_mountp, tp->t_ticket, NULL, log_flags);
 	}
 	xfs_trans_free_items(tp, flags);
 	xfs_trans_free_busy(tp);
@@ -1262,8 +1271,11 @@ xfs_trans_chunk_committed(
 
 		/*
 		 * Now that we've repositioned the item in the AIL,
-		 * unpin it so it can be flushed.
+		 * unpin it so it can be flushed. Pass information
+		 * about buffer stale state down from the log item
+		 * flags, if anyone else stales the buffer we do not
+		 * want to pay any attention to it.
 		 */
-		IOP_UNPIN(lip);
+		IOP_UNPIN(lip, lidp->lid_flags & XFS_LID_BUF_STALE);
 	}
 }

fs/xfs/xfs_trans.h

@@ -170,7 +170,7 @@ typedef struct xfs_item_ops {
 	uint (*iop_size)(xfs_log_item_t *);
 	void (*iop_format)(xfs_log_item_t *, struct xfs_log_iovec *);
 	void (*iop_pin)(xfs_log_item_t *);
-	void (*iop_unpin)(xfs_log_item_t *);
+	void (*iop_unpin)(xfs_log_item_t *, int);
 	void (*iop_unpin_remove)(xfs_log_item_t *, struct xfs_trans *);
 	uint (*iop_trylock)(xfs_log_item_t *);
 	void (*iop_unlock)(xfs_log_item_t *);
@@ -184,7 +184,7 @@ typedef struct xfs_item_ops {
 #define IOP_SIZE(ip)		(*(ip)->li_ops->iop_size)(ip)
 #define IOP_FORMAT(ip,vp)	(*(ip)->li_ops->iop_format)(ip, vp)
 #define IOP_PIN(ip)		(*(ip)->li_ops->iop_pin)(ip)
-#define IOP_UNPIN(ip)		(*(ip)->li_ops->iop_unpin)(ip)
+#define IOP_UNPIN(ip, flags)	(*(ip)->li_ops->iop_unpin)(ip, flags)
 #define IOP_UNPIN_REMOVE(ip,tp) (*(ip)->li_ops->iop_unpin_remove)(ip, tp)
 #define IOP_TRYLOCK(ip)		(*(ip)->li_ops->iop_trylock)(ip)
 #define IOP_UNLOCK(ip)		(*(ip)->li_ops->iop_unlock)(ip)
@@ -222,6 +222,7 @@ typedef struct xfs_log_item_desc {
 #define XFS_LID_DIRTY		0x1
 #define XFS_LID_PINNED		0x2
 #define XFS_LID_SYNC_UNLOCK	0x4
+#define XFS_LID_BUF_STALE	0x8
 
 /*
  * This structure is used to maintain a chunk list of log_item_desc

fs/xfs/xfs_trans_buf.c

@@ -796,6 +796,7 @@ xfs_trans_log_buf(xfs_trans_t	*tp,
 
 	tp->t_flags |= XFS_TRANS_DIRTY;
 	lidp->lid_flags |= XFS_LID_DIRTY;
+	lidp->lid_flags &= ~XFS_LID_BUF_STALE;
 	bip->bli_flags |= XFS_BLI_LOGGED;
 	xfs_buf_item_log(bip, first, last);
 	xfs_buf_item_trace("BLOG", bip);
@@ -882,7 +883,7 @@ xfs_trans_binval(
 	bip->bli_format.blf_flags |= XFS_BLI_CANCEL;
 	memset((char *)(bip->bli_format.blf_data_map), 0,
 	      (bip->bli_format.blf_map_size * sizeof(uint)));
-	lidp->lid_flags |= XFS_LID_DIRTY;
+	lidp->lid_flags |= XFS_LID_DIRTY|XFS_LID_BUF_STALE;
 	tp->t_flags |= XFS_TRANS_DIRTY;
 	xfs_buftrace("XFS_BINVAL", bp);
 	xfs_buf_item_trace("BINVAL", bip);

fs/xfs/xfs_vfsops.c

@@ -873,7 +873,6 @@ xfs_syncsub(
 	boolean_t	mount_locked;
 	boolean_t	vnode_refed;
 	int		preempt;
-	int		do_mmap_flush;
 	xfs_dinode_t	*dip;
 	xfs_buf_log_item_t	*bip;
 	xfs_iptr_t	*ipointer;
@@ -944,8 +943,6 @@ xfs_syncsub(
 		fflag = XFS_B_DELWRI;
 	if (flags & SYNC_WAIT)
 		fflag = 0;		/* synchronous overrides all */
-	do_mmap_flush = (flags & (SYNC_DELWRI|SYNC_BDFLUSH)) !=
-						(SYNC_DELWRI|SYNC_BDFLUSH);
 
 	base_lock_flags = XFS_ILOCK_SHARED;
 	if (flags & (SYNC_DELWRI | SYNC_CLOSE)) {
@@ -1177,12 +1174,8 @@ xfs_syncsub(
 				 * across calls to the buffer cache.
 				 */
 				xfs_iunlock(ip, XFS_ILOCK_SHARED);
-				if (do_mmap_flush) {
 				VOP_FLUSH_PAGES(vp, (xfs_off_t)0, -1,
 							fflag, FI_NONE, error);
-				} else {
-					filemap_fdatawrite(LINVFS_GET_IP(vp)->i_mapping);
-				}
 				xfs_ilock(ip, XFS_ILOCK_SHARED);
 			}
 

fs/xfs/xfsidbg.c

@@ -5248,6 +5248,7 @@ xfsidbg_xtp(xfs_trans_t *tp)
 		"dirty",	/* 0x1 */
 		"pinned",	/* 0x2 */
 		"sync unlock",	/* 0x4 */
+		"buf stale",	/* 0x8 */
 		0
 		};