1 files changed, 161 insertions, 216 deletions

diff --git a/fs/xfs/xfs_log.c b/fs/xfs/xfs_log.c
index 3e5ba1ecc080..2fcd9ed5d075 100644
--- a/fs/xfs/xfs_log.c
+++ b/fs/xfs/xfs_log.c
@@ -560,7 +560,6 @@ xfs_log_done(
  */
 int
 xfs_log_notify(
-	struct xfs_mount	*mp,
 	struct xlog_in_core	*iclog,
 	xfs_log_callback_t	*cb)
 {
@@ -869,7 +868,7 @@ xfs_log_unmount_write(xfs_mount_t *mp)
 		return 0;
 	}
 
-	error = _xfs_log_force(mp, XFS_LOG_SYNC, NULL);
+	error = xfs_log_force(mp, XFS_LOG_SYNC);
 	ASSERT(error || !(XLOG_FORCED_SHUTDOWN(log)));
 
 #ifdef DEBUG
@@ -1149,7 +1148,7 @@ xlog_assign_tail_lsn_locked(
 	struct xfs_log_item	*lip;
 	xfs_lsn_t		tail_lsn;
 
-	assert_spin_locked(&mp->m_ail->xa_lock);
+	assert_spin_locked(&mp->m_ail->ail_lock);
 
 	/*
 	 * To make sure we always have a valid LSN for the log tail we keep
@@ -1172,9 +1171,9 @@ xlog_assign_tail_lsn(
 {
 	xfs_lsn_t		tail_lsn;
 
-	spin_lock(&mp->m_ail->xa_lock);
+	spin_lock(&mp->m_ail->ail_lock);
 	tail_lsn = xlog_assign_tail_lsn_locked(mp);
-	spin_unlock(&mp->m_ail->xa_lock);
+	spin_unlock(&mp->m_ail->ail_lock);
 
 	return tail_lsn;
 }
@@ -3304,269 +3303,215 @@ xlog_state_switch_iclogs(
  *		not in the active nor dirty state.
  */
 int
-_xfs_log_force(
+xfs_log_force(
 	struct xfs_mount	*mp,
-	uint			flags,
-	int			*log_flushed)
+	uint			flags)
 {
 	struct xlog		*log = mp->m_log;
 	struct xlog_in_core	*iclog;
 	xfs_lsn_t		lsn;
 
 	XFS_STATS_INC(mp, xs_log_force);
+	trace_xfs_log_force(mp, 0, _RET_IP_);
 
 	xlog_cil_force(log);
 
 	spin_lock(&log->l_icloglock);
-
 	iclog = log->l_iclog;
-	if (iclog->ic_state & XLOG_STATE_IOERROR) {
-		spin_unlock(&log->l_icloglock);
-		return -EIO;
-	}
+	if (iclog->ic_state & XLOG_STATE_IOERROR)
+		goto out_error;
 
-	/* If the head iclog is not active nor dirty, we just attach
-	 * ourselves to the head and go to sleep.
-	 */
-	if (iclog->ic_state == XLOG_STATE_ACTIVE ||
-	    iclog->ic_state == XLOG_STATE_DIRTY) {
+	if (iclog->ic_state == XLOG_STATE_DIRTY ||
+	    (iclog->ic_state == XLOG_STATE_ACTIVE &&
+	     atomic_read(&iclog->ic_refcnt) == 0 && iclog->ic_offset == 0)) {
 		/*
-		 * If the head is dirty or (active and empty), then
-		 * we need to look at the previous iclog.  If the previous
-		 * iclog is active or dirty we are done.  There is nothing
-		 * to sync out.  Otherwise, we attach ourselves to the
+		 * If the head is dirty or (active and empty), then we need to
+		 * look at the previous iclog.
+		 *
+		 * If the previous iclog is active or dirty we are done.  There
+		 * is nothing to sync out. Otherwise, we attach ourselves to the
 		 * previous iclog and go to sleep.
 		 */
-		if (iclog->ic_state == XLOG_STATE_DIRTY ||
-		    (atomic_read(&iclog->ic_refcnt) == 0
-		     && iclog->ic_offset == 0)) {
-			iclog = iclog->ic_prev;
-			if (iclog->ic_state == XLOG_STATE_ACTIVE ||
-			    iclog->ic_state == XLOG_STATE_DIRTY)
-				goto no_sleep;
-			else
-				goto maybe_sleep;
-		} else {
-			if (atomic_read(&iclog->ic_refcnt) == 0) {
-				/* We are the only one with access to this
-				 * iclog.  Flush it out now.  There should
-				 * be a roundoff of zero to show that someone
-				 * has already taken care of the roundoff from
-				 * the previous sync.
-				 */
-				atomic_inc(&iclog->ic_refcnt);
-				lsn = be64_to_cpu(iclog->ic_header.h_lsn);
-				xlog_state_switch_iclogs(log, iclog, 0);
-				spin_unlock(&log->l_icloglock);
+		iclog = iclog->ic_prev;
+		if (iclog->ic_state == XLOG_STATE_ACTIVE ||
+		    iclog->ic_state == XLOG_STATE_DIRTY)
+			goto out_unlock;
+	} else if (iclog->ic_state == XLOG_STATE_ACTIVE) {
+		if (atomic_read(&iclog->ic_refcnt) == 0) {
+			/*
+			 * We are the only one with access to this iclog.
+			 *
+			 * Flush it out now.  There should be a roundoff of zero
+			 * to show that someone has already taken care of the
+			 * roundoff from the previous sync.
+			 */
+			atomic_inc(&iclog->ic_refcnt);
+			lsn = be64_to_cpu(iclog->ic_header.h_lsn);
+			xlog_state_switch_iclogs(log, iclog, 0);
+			spin_unlock(&log->l_icloglock);
 
-				if (xlog_state_release_iclog(log, iclog))
-					return -EIO;
-
-				if (log_flushed)
-					*log_flushed = 1;
-				spin_lock(&log->l_icloglock);
-				if (be64_to_cpu(iclog->ic_header.h_lsn) == lsn &&
-				    iclog->ic_state != XLOG_STATE_DIRTY)
-					goto maybe_sleep;
-				else
-					goto no_sleep;
-			} else {
-				/* Someone else is writing to this iclog.
-				 * Use its call to flush out the data.  However,
-				 * the other thread may not force out this LR,
-				 * so we mark it WANT_SYNC.
-				 */
-				xlog_state_switch_iclogs(log, iclog, 0);
-				goto maybe_sleep;
-			}
-		}
-	}
+			if (xlog_state_release_iclog(log, iclog))
+				return -EIO;
 
-	/* By the time we come around again, the iclog could've been filled
-	 * which would give it another lsn.  If we have a new lsn, just
-	 * return because the relevant data has been flushed.
-	 */
-maybe_sleep:
-	if (flags & XFS_LOG_SYNC) {
-		/*
-		 * We must check if we're shutting down here, before
-		 * we wait, while we're holding the l_icloglock.
-		 * Then we check again after waking up, in case our
-		 * sleep was disturbed by a bad news.
-		 */
-		if (iclog->ic_state & XLOG_STATE_IOERROR) {
-			spin_unlock(&log->l_icloglock);
-			return -EIO;
+			spin_lock(&log->l_icloglock);
+			if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn ||
+			    iclog->ic_state == XLOG_STATE_DIRTY)
+				goto out_unlock;
+		} else {
+			/*
+			 * Someone else is writing to this iclog.
+			 *
+			 * Use its call to flush out the data.  However, the
+			 * other thread may not force out this LR, so we mark
+			 * it WANT_SYNC.
+			 */
+			xlog_state_switch_iclogs(log, iclog, 0);
 		}
-		XFS_STATS_INC(mp, xs_log_force_sleep);
-		xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
+	} else {
 		/*
-		 * No need to grab the log lock here since we're
-		 * only deciding whether or not to return EIO
-		 * and the memory read should be atomic.
+		 * If the head iclog is not active nor dirty, we just attach
+		 * ourselves to the head and go to sleep if necessary.
 		 */
-		if (iclog->ic_state & XLOG_STATE_IOERROR)
-			return -EIO;
-	} else {
-
-no_sleep:
-		spin_unlock(&log->l_icloglock);
+		;
 	}
+
+	if (!(flags & XFS_LOG_SYNC))
+		goto out_unlock;
+
+	if (iclog->ic_state & XLOG_STATE_IOERROR)
+		goto out_error;
+	XFS_STATS_INC(mp, xs_log_force_sleep);
+	xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
+	if (iclog->ic_state & XLOG_STATE_IOERROR)
+		return -EIO;
 	return 0;
-}
 
-/*
- * Wrapper for _xfs_log_force(), to be used when caller doesn't care
- * about errors or whether the log was flushed or not. This is the normal
- * interface to use when trying to unpin items or move the log forward.
- */
-void
-xfs_log_force(
-	xfs_mount_t	*mp,
-	uint		flags)
-{
-	trace_xfs_log_force(mp, 0, _RET_IP_);
-	_xfs_log_force(mp, flags, NULL);
+out_unlock:
+	spin_unlock(&log->l_icloglock);
+	return 0;
+out_error:
+	spin_unlock(&log->l_icloglock);
+	return -EIO;
 }
 
-/*
- * Force the in-core log to disk for a specific LSN.
- *
- * Find in-core log with lsn.
- *	If it is in the DIRTY state, just return.
- *	If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
- *		state and go to sleep or return.
- *	If it is in any other state, go to sleep or return.
- *
- * Synchronous forces are implemented with a signal variable. All callers
- * to force a given lsn to disk will wait on a the sv attached to the
- * specific in-core log.  When given in-core log finally completes its
- * write to disk, that thread will wake up all threads waiting on the
- * sv.
- */
-int
-_xfs_log_force_lsn(
+static int
+__xfs_log_force_lsn(
 	struct xfs_mount	*mp,
 	xfs_lsn_t		lsn,
 	uint			flags,
-	int			*log_flushed)
+	int			*log_flushed,
+	bool			already_slept)
 {
 	struct xlog		*log = mp->m_log;
 	struct xlog_in_core	*iclog;
-	int			already_slept = 0;
-
-	ASSERT(lsn != 0);
-
-	XFS_STATS_INC(mp, xs_log_force);
 
-	lsn = xlog_cil_force_lsn(log, lsn);
-	if (lsn == NULLCOMMITLSN)
-		return 0;
-
-try_again:
 	spin_lock(&log->l_icloglock);
 	iclog = log->l_iclog;
-	if (iclog->ic_state & XLOG_STATE_IOERROR) {
-		spin_unlock(&log->l_icloglock);
-		return -EIO;
-	}
+	if (iclog->ic_state & XLOG_STATE_IOERROR)
+		goto out_error;
 
-	do {
-		if (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
-			iclog = iclog->ic_next;
-			continue;
-		}
+	while (be64_to_cpu(iclog->ic_header.h_lsn) != lsn) {
+		iclog = iclog->ic_next;
+		if (iclog == log->l_iclog)
+			goto out_unlock;
+	}
 
-		if (iclog->ic_state == XLOG_STATE_DIRTY) {
-			spin_unlock(&log->l_icloglock);
-			return 0;
-		}
+	if (iclog->ic_state == XLOG_STATE_DIRTY)
+		goto out_unlock;
 
-		if (iclog->ic_state == XLOG_STATE_ACTIVE) {
-			/*
-			 * We sleep here if we haven't already slept (e.g.
-			 * this is the first time we've looked at the correct
-			 * iclog buf) and the buffer before us is going to
-			 * be sync'ed. The reason for this is that if we
-			 * are doing sync transactions here, by waiting for
-			 * the previous I/O to complete, we can allow a few
-			 * more transactions into this iclog before we close
-			 * it down.
-			 *
-			 * Otherwise, we mark the buffer WANT_SYNC, and bump
-			 * up the refcnt so we can release the log (which
-			 * drops the ref count).  The state switch keeps new
-			 * transaction commits from using this buffer.  When
-			 * the current commits finish writing into the buffer,
-			 * the refcount will drop to zero and the buffer will
-			 * go out then.
-			 */
-			if (!already_slept &&
-			    (iclog->ic_prev->ic_state &
-			     (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) {
-				ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
+	if (iclog->ic_state == XLOG_STATE_ACTIVE) {
+		/*
+		 * We sleep here if we haven't already slept (e.g. this is the
+		 * first time we've looked at the correct iclog buf) and the
+		 * buffer before us is going to be sync'ed.  The reason for this
+		 * is that if we are doing sync transactions here, by waiting
+		 * for the previous I/O to complete, we can allow a few more
+		 * transactions into this iclog before we close it down.
+		 *
+		 * Otherwise, we mark the buffer WANT_SYNC, and bump up the
+		 * refcnt so we can release the log (which drops the ref count).
+		 * The state switch keeps new transaction commits from using
+		 * this buffer.  When the current commits finish writing into
+		 * the buffer, the refcount will drop to zero and the buffer
+		 * will go out then.
+		 */
+		if (!already_slept &&
+		    (iclog->ic_prev->ic_state &
+		     (XLOG_STATE_WANT_SYNC | XLOG_STATE_SYNCING))) {
+			ASSERT(!(iclog->ic_state & XLOG_STATE_IOERROR));
 
-				XFS_STATS_INC(mp, xs_log_force_sleep);
+			XFS_STATS_INC(mp, xs_log_force_sleep);
 
-				xlog_wait(&iclog->ic_prev->ic_write_wait,
-							&log->l_icloglock);
-				already_slept = 1;
-				goto try_again;
-			}
-			atomic_inc(&iclog->ic_refcnt);
-			xlog_state_switch_iclogs(log, iclog, 0);
-			spin_unlock(&log->l_icloglock);
-			if (xlog_state_release_iclog(log, iclog))
-				return -EIO;
-			if (log_flushed)
-				*log_flushed = 1;
-			spin_lock(&log->l_icloglock);
+			xlog_wait(&iclog->ic_prev->ic_write_wait,
+					&log->l_icloglock);
+			return -EAGAIN;
 		}
+		atomic_inc(&iclog->ic_refcnt);
+		xlog_state_switch_iclogs(log, iclog, 0);
+		spin_unlock(&log->l_icloglock);
+		if (xlog_state_release_iclog(log, iclog))
+			return -EIO;
+		if (log_flushed)
+			*log_flushed = 1;
+		spin_lock(&log->l_icloglock);
+	}
 
-		if ((flags & XFS_LOG_SYNC) && /* sleep */
-		    !(iclog->ic_state &
-		      (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY))) {
-			/*
-			 * Don't wait on completion if we know that we've
-			 * gotten a log write error.
-			 */
-			if (iclog->ic_state & XLOG_STATE_IOERROR) {
-				spin_unlock(&log->l_icloglock);
-				return -EIO;
-			}
-			XFS_STATS_INC(mp, xs_log_force_sleep);
-			xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
-			/*
-			 * No need to grab the log lock here since we're
-			 * only deciding whether or not to return EIO
-			 * and the memory read should be atomic.
-			 */
-			if (iclog->ic_state & XLOG_STATE_IOERROR)
-				return -EIO;
-		} else {		/* just return */
-			spin_unlock(&log->l_icloglock);
-		}
+	if (!(flags & XFS_LOG_SYNC) ||
+	    (iclog->ic_state & (XLOG_STATE_ACTIVE | XLOG_STATE_DIRTY)))
+		goto out_unlock;
 
-		return 0;
-	} while (iclog != log->l_iclog);
+	if (iclog->ic_state & XLOG_STATE_IOERROR)
+		goto out_error;
+
+	XFS_STATS_INC(mp, xs_log_force_sleep);
+	xlog_wait(&iclog->ic_force_wait, &log->l_icloglock);
+	if (iclog->ic_state & XLOG_STATE_IOERROR)
+		return -EIO;
+	return 0;
 
+out_unlock:
 	spin_unlock(&log->l_icloglock);
 	return 0;
+out_error:
+	spin_unlock(&log->l_icloglock);
+	return -EIO;
 }
 
 /*
- * Wrapper for _xfs_log_force_lsn(), to be used when caller doesn't care
- * about errors or whether the log was flushed or not. This is the normal
- * interface to use when trying to unpin items or move the log forward.
+ * Force the in-core log to disk for a specific LSN.
+ *
+ * Find in-core log with lsn.
+ *	If it is in the DIRTY state, just return.
+ *	If it is in the ACTIVE state, move the in-core log into the WANT_SYNC
+ *		state and go to sleep or return.
+ *	If it is in any other state, go to sleep or return.
+ *
+ * Synchronous forces are implemented with a wait queue.  All callers trying
+ * to force a given lsn to disk must wait on the queue attached to the
+ * specific in-core log.  When given in-core log finally completes its write
+ * to disk, that thread will wake up all threads waiting on the queue.
  */
-void
+int
 xfs_log_force_lsn(
-	xfs_mount_t	*mp,
-	xfs_lsn_t	lsn,
-	uint		flags)
+	struct xfs_mount	*mp,
+	xfs_lsn_t		lsn,
+	uint			flags,
+	int			*log_flushed)
 {
+	int			ret;
+	ASSERT(lsn != 0);
+
+	XFS_STATS_INC(mp, xs_log_force);
 	trace_xfs_log_force(mp, lsn, _RET_IP_);
-	_xfs_log_force_lsn(mp, lsn, flags, NULL);
+
+	lsn = xlog_cil_force_lsn(mp->m_log, lsn);
+	if (lsn == NULLCOMMITLSN)
+		return 0;
+
+	ret = __xfs_log_force_lsn(mp, lsn, flags, log_flushed, false);
+	if (ret == -EAGAIN)
+		ret = __xfs_log_force_lsn(mp, lsn, flags, log_flushed, true);
+	return ret;
 }
 
 /*
@@ -4035,7 +3980,7 @@ xfs_log_force_umount(
 	 * to guarantee this.
 	 */
 	if (!logerror)
-		_xfs_log_force(mp, XFS_LOG_SYNC, NULL);
+		xfs_log_force(mp, XFS_LOG_SYNC);
 
 	/*
 	 * mark the filesystem and the as in a shutdown state and wake