[XFS] Do not release the last iclog of a transaction before we get our

callbacks attached to it. Otherwise we can end up executing the
callback out of order.

SGI Modid: 2.5.x-xfs:slinx:137750a

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;
@@ -845,21 +847,28 @@ 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) {
-		error = xfs_log_force(mp, commit_lsn,
+		if (!error)
+			error = xfs_log_force(mp, commit_lsn,
 				      XFS_LOG_FORCE | XFS_LOG_SYNC);
 		XFS_STATS_INC(xfsstats.xs_trans_sync);
 	} else {
@@ -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);