2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * Copyright (c) 2008 Dave Chinner
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License as
8 * published by the Free Software Foundation.
10 * This program is distributed in the hope that it would be useful,
11 * but WITHOUT ANY WARRANTY; without even the implied warranty of
12 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
13 * GNU General Public License for more details.
15 * You should have received a copy of the GNU General Public License
16 * along with this program; if not, write the Free Software Foundation,
17 * Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA
21 #include "xfs_types.h"
23 #include "xfs_trans.h"
26 #include "xfs_mount.h"
27 #include "xfs_trans_priv.h"
28 #include "xfs_trace.h"
29 #include "xfs_error.h"
33 * Check that the list is sorted as it should be.
40 xfs_log_item_t
*prev_lip
;
42 if (list_empty(&ailp
->xa_ail
))
46 * Check the next and previous entries are valid.
48 ASSERT((lip
->li_flags
& XFS_LI_IN_AIL
) != 0);
49 prev_lip
= list_entry(lip
->li_ail
.prev
, xfs_log_item_t
, li_ail
);
50 if (&prev_lip
->li_ail
!= &ailp
->xa_ail
)
51 ASSERT(XFS_LSN_CMP(prev_lip
->li_lsn
, lip
->li_lsn
) <= 0);
53 prev_lip
= list_entry(lip
->li_ail
.next
, xfs_log_item_t
, li_ail
);
54 if (&prev_lip
->li_ail
!= &ailp
->xa_ail
)
55 ASSERT(XFS_LSN_CMP(prev_lip
->li_lsn
, lip
->li_lsn
) >= 0);
58 #ifdef XFS_TRANS_DEBUG
60 * Walk the list checking lsn ordering, and that every entry has the
61 * XFS_LI_IN_AIL flag set. This is really expensive, so only do it
62 * when specifically debugging the transaction subsystem.
64 prev_lip
= list_entry(&ailp
->xa_ail
, xfs_log_item_t
, li_ail
);
65 list_for_each_entry(lip
, &ailp
->xa_ail
, li_ail
) {
66 if (&prev_lip
->li_ail
!= &ailp
->xa_ail
)
67 ASSERT(XFS_LSN_CMP(prev_lip
->li_lsn
, lip
->li_lsn
) <= 0);
68 ASSERT((lip
->li_flags
& XFS_LI_IN_AIL
) != 0);
71 #endif /* XFS_TRANS_DEBUG */
74 #define xfs_ail_check(a,l)
78 * Return a pointer to the first item in the AIL. If the AIL is empty, then
85 if (list_empty(&ailp
->xa_ail
))
88 return list_first_entry(&ailp
->xa_ail
, xfs_log_item_t
, li_ail
);
92 * Return a pointer to the last item in the AIL. If the AIL is empty, then
95 static xfs_log_item_t
*
99 if (list_empty(&ailp
->xa_ail
))
102 return list_entry(ailp
->xa_ail
.prev
, xfs_log_item_t
, li_ail
);
106 * Return a pointer to the item which follows the given item in the AIL. If
107 * the given item is the last item in the list, then return NULL.
109 static xfs_log_item_t
*
111 struct xfs_ail
*ailp
,
114 if (lip
->li_ail
.next
== &ailp
->xa_ail
)
117 return list_first_entry(&lip
->li_ail
, xfs_log_item_t
, li_ail
);
121 * This is called by the log manager code to determine the LSN of the tail of
122 * the log. This is exactly the LSN of the first item in the AIL. If the AIL
123 * is empty, then this function returns 0.
125 * We need the AIL lock in order to get a coherent read of the lsn of the last
130 struct xfs_ail
*ailp
)
135 spin_lock(&ailp
->xa_lock
);
136 lip
= xfs_ail_min(ailp
);
139 spin_unlock(&ailp
->xa_lock
);
145 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
149 struct xfs_ail
*ailp
)
154 spin_lock(&ailp
->xa_lock
);
155 lip
= xfs_ail_max(ailp
);
158 spin_unlock(&ailp
->xa_lock
);
164 * The cursor keeps track of where our current traversal is up to by tracking
165 * the next item in the list for us. However, for this to be safe, removing an
166 * object from the AIL needs to invalidate any cursor that points to it. hence
167 * the traversal cursor needs to be linked to the struct xfs_ail so that
168 * deletion can search all the active cursors for invalidation.
171 xfs_trans_ail_cursor_init(
172 struct xfs_ail
*ailp
,
173 struct xfs_ail_cursor
*cur
)
176 list_add_tail(&cur
->list
, &ailp
->xa_cursors
);
180 * Get the next item in the traversal and advance the cursor. If the cursor
181 * was invalidated (indicated by a lip of 1), restart the traversal.
183 struct xfs_log_item
*
184 xfs_trans_ail_cursor_next(
185 struct xfs_ail
*ailp
,
186 struct xfs_ail_cursor
*cur
)
188 struct xfs_log_item
*lip
= cur
->item
;
190 if ((__psint_t
)lip
& 1)
191 lip
= xfs_ail_min(ailp
);
193 cur
->item
= xfs_ail_next(ailp
, lip
);
198 * When the traversal is complete, we need to remove the cursor from the list
199 * of traversing cursors.
202 xfs_trans_ail_cursor_done(
203 struct xfs_ail
*ailp
,
204 struct xfs_ail_cursor
*cur
)
207 list_del_init(&cur
->list
);
211 * Invalidate any cursor that is pointing to this item. This is called when an
212 * item is removed from the AIL. Any cursor pointing to this object is now
213 * invalid and the traversal needs to be terminated so it doesn't reference a
214 * freed object. We set the low bit of the cursor item pointer so we can
215 * distinguish between an invalidation and the end of the list when getting the
216 * next item from the cursor.
219 xfs_trans_ail_cursor_clear(
220 struct xfs_ail
*ailp
,
221 struct xfs_log_item
*lip
)
223 struct xfs_ail_cursor
*cur
;
225 list_for_each_entry(cur
, &ailp
->xa_cursors
, list
) {
226 if (cur
->item
== lip
)
227 cur
->item
= (struct xfs_log_item
*)
228 ((__psint_t
)cur
->item
| 1);
233 * Find the first item in the AIL with the given @lsn by searching in ascending
234 * LSN order and initialise the cursor to point to the next item for a
235 * ascending traversal. Pass a @lsn of zero to initialise the cursor to the
236 * first item in the AIL. Returns NULL if the list is empty.
239 xfs_trans_ail_cursor_first(
240 struct xfs_ail
*ailp
,
241 struct xfs_ail_cursor
*cur
,
246 xfs_trans_ail_cursor_init(ailp
, cur
);
249 lip
= xfs_ail_min(ailp
);
253 list_for_each_entry(lip
, &ailp
->xa_ail
, li_ail
) {
254 if (XFS_LSN_CMP(lip
->li_lsn
, lsn
) >= 0)
261 cur
->item
= xfs_ail_next(ailp
, lip
);
265 static struct xfs_log_item
*
266 __xfs_trans_ail_cursor_last(
267 struct xfs_ail
*ailp
,
272 list_for_each_entry_reverse(lip
, &ailp
->xa_ail
, li_ail
) {
273 if (XFS_LSN_CMP(lip
->li_lsn
, lsn
) <= 0)
280 * Find the last item in the AIL with the given @lsn by searching in descending
281 * LSN order and initialise the cursor to point to that item. If there is no
282 * item with the value of @lsn, then it sets the cursor to the last item with an
283 * LSN lower than @lsn. Returns NULL if the list is empty.
285 struct xfs_log_item
*
286 xfs_trans_ail_cursor_last(
287 struct xfs_ail
*ailp
,
288 struct xfs_ail_cursor
*cur
,
291 xfs_trans_ail_cursor_init(ailp
, cur
);
292 cur
->item
= __xfs_trans_ail_cursor_last(ailp
, lsn
);
297 * Splice the log item list into the AIL at the given LSN. We splice to the
298 * tail of the given LSN to maintain insert order for push traversals. The
299 * cursor is optional, allowing repeated updates to the same LSN to avoid
300 * repeated traversals. This should not be called with an empty list.
304 struct xfs_ail
*ailp
,
305 struct xfs_ail_cursor
*cur
,
306 struct list_head
*list
,
309 struct xfs_log_item
*lip
;
311 ASSERT(!list_empty(list
));
314 * Use the cursor to determine the insertion point if one is
315 * provided. If not, or if the one we got is not valid,
316 * find the place in the AIL where the items belong.
318 lip
= cur
? cur
->item
: NULL
;
319 if (!lip
|| (__psint_t
) lip
& 1)
320 lip
= __xfs_trans_ail_cursor_last(ailp
, lsn
);
323 * If a cursor is provided, we know we're processing the AIL
324 * in lsn order, and future items to be spliced in will
325 * follow the last one being inserted now. Update the
326 * cursor to point to that last item, now while we have a
327 * reliable pointer to it.
330 cur
->item
= list_entry(list
->prev
, struct xfs_log_item
, li_ail
);
333 * Finally perform the splice. Unless the AIL was empty,
334 * lip points to the item in the AIL _after_ which the new
335 * items should go. If lip is null the AIL was empty, so
336 * the new items go at the head of the AIL.
339 list_splice(list
, &lip
->li_ail
);
341 list_splice(list
, &ailp
->xa_ail
);
345 * Delete the given item from the AIL. Return a pointer to the item.
349 struct xfs_ail
*ailp
,
352 xfs_ail_check(ailp
, lip
);
353 list_del(&lip
->li_ail
);
354 xfs_trans_ail_cursor_clear(ailp
, lip
);
359 struct xfs_ail
*ailp
)
361 xfs_mount_t
*mp
= ailp
->xa_mount
;
362 struct xfs_ail_cursor cur
;
372 * If we encountered pinned items or did not finish writing out all
373 * buffers the last time we ran, force the log first and wait for it
374 * before pushing again.
376 if (ailp
->xa_log_flush
&& ailp
->xa_last_pushed_lsn
== 0 &&
377 (!list_empty_careful(&ailp
->xa_buf_list
) ||
378 xfs_ail_min_lsn(ailp
))) {
379 ailp
->xa_log_flush
= 0;
381 XFS_STATS_INC(xs_push_ail_flush
);
382 xfs_log_force(mp
, XFS_LOG_SYNC
);
385 spin_lock(&ailp
->xa_lock
);
387 /* barrier matches the xa_target update in xfs_ail_push() */
389 target
= ailp
->xa_target
;
390 ailp
->xa_target_prev
= target
;
392 lip
= xfs_trans_ail_cursor_first(ailp
, &cur
, ailp
->xa_last_pushed_lsn
);
395 * If the AIL is empty or our push has reached the end we are
398 xfs_trans_ail_cursor_done(ailp
, &cur
);
399 spin_unlock(&ailp
->xa_lock
);
403 XFS_STATS_INC(xs_push_ail
);
406 while ((XFS_LSN_CMP(lip
->li_lsn
, target
) <= 0)) {
410 * Note that IOP_PUSH may unlock and reacquire the AIL lock. We
411 * rely on the AIL cursor implementation to be able to deal with
414 lock_result
= IOP_PUSH(lip
, &ailp
->xa_buf_list
);
415 switch (lock_result
) {
416 case XFS_ITEM_SUCCESS
:
417 XFS_STATS_INC(xs_push_ail_success
);
418 trace_xfs_ail_push(lip
);
420 ailp
->xa_last_pushed_lsn
= lsn
;
423 case XFS_ITEM_FLUSHING
:
425 * The item or its backing buffer is already beeing
426 * flushed. The typical reason for that is that an
427 * inode buffer is locked because we already pushed the
428 * updates to it as part of inode clustering.
430 * We do not want to to stop flushing just because lots
431 * of items are already beeing flushed, but we need to
432 * re-try the flushing relatively soon if most of the
433 * AIL is beeing flushed.
435 XFS_STATS_INC(xs_push_ail_flushing
);
436 trace_xfs_ail_flushing(lip
);
439 ailp
->xa_last_pushed_lsn
= lsn
;
442 case XFS_ITEM_PINNED
:
443 XFS_STATS_INC(xs_push_ail_pinned
);
444 trace_xfs_ail_pinned(lip
);
447 ailp
->xa_log_flush
++;
449 case XFS_ITEM_LOCKED
:
450 XFS_STATS_INC(xs_push_ail_locked
);
451 trace_xfs_ail_locked(lip
);
463 * Are there too many items we can't do anything with?
465 * If we we are skipping too many items because we can't flush
466 * them or they are already being flushed, we back off and
467 * given them time to complete whatever operation is being
468 * done. i.e. remove pressure from the AIL while we can't make
469 * progress so traversals don't slow down further inserts and
470 * removals to/from the AIL.
472 * The value of 100 is an arbitrary magic number based on
478 lip
= xfs_trans_ail_cursor_next(ailp
, &cur
);
483 xfs_trans_ail_cursor_done(ailp
, &cur
);
484 spin_unlock(&ailp
->xa_lock
);
486 if (xfs_buf_delwri_submit_nowait(&ailp
->xa_buf_list
))
487 ailp
->xa_log_flush
++;
489 if (!count
|| XFS_LSN_CMP(lsn
, target
) >= 0) {
492 * We reached the target or the AIL is empty, so wait a bit
493 * longer for I/O to complete and remove pushed items from the
494 * AIL before we start the next scan from the start of the AIL.
497 ailp
->xa_last_pushed_lsn
= 0;
498 } else if (((stuck
+ flushing
) * 100) / count
> 90) {
500 * Either there is a lot of contention on the AIL or we are
501 * stuck due to operations in progress. "Stuck" in this case
502 * is defined as >90% of the items we tried to push were stuck.
504 * Backoff a bit more to allow some I/O to complete before
505 * restarting from the start of the AIL. This prevents us from
506 * spinning on the same items, and if they are pinned will all
507 * the restart to issue a log force to unpin the stuck items.
510 ailp
->xa_last_pushed_lsn
= 0;
513 * Assume we have more work to do in a short while.
525 struct xfs_ail
*ailp
= data
;
526 long tout
= 0; /* milliseconds */
528 current
->flags
|= PF_MEMALLOC
;
530 while (!kthread_should_stop()) {
531 if (tout
&& tout
<= 20)
532 __set_current_state(TASK_KILLABLE
);
534 __set_current_state(TASK_INTERRUPTIBLE
);
536 spin_lock(&ailp
->xa_lock
);
539 * Idle if the AIL is empty and we are not racing with a target
540 * update. We check the AIL after we set the task to a sleep
541 * state to guarantee that we either catch an xa_target update
542 * or that a wake_up resets the state to TASK_RUNNING.
543 * Otherwise, we run the risk of sleeping indefinitely.
545 * The barrier matches the xa_target update in xfs_ail_push().
548 if (!xfs_ail_min(ailp
) &&
549 ailp
->xa_target
== ailp
->xa_target_prev
) {
550 spin_unlock(&ailp
->xa_lock
);
555 spin_unlock(&ailp
->xa_lock
);
558 schedule_timeout(msecs_to_jiffies(tout
));
560 __set_current_state(TASK_RUNNING
);
564 tout
= xfsaild_push(ailp
);
571 * This routine is called to move the tail of the AIL forward. It does this by
572 * trying to flush items in the AIL whose lsns are below the given
575 * The push is run asynchronously in a workqueue, which means the caller needs
576 * to handle waiting on the async flush for space to become available.
577 * We don't want to interrupt any push that is in progress, hence we only queue
578 * work if we set the pushing bit approriately.
580 * We do this unlocked - we only need to know whether there is anything in the
581 * AIL at the time we are called. We don't need to access the contents of
582 * any of the objects, so the lock is not needed.
586 struct xfs_ail
*ailp
,
587 xfs_lsn_t threshold_lsn
)
591 lip
= xfs_ail_min(ailp
);
592 if (!lip
|| XFS_FORCED_SHUTDOWN(ailp
->xa_mount
) ||
593 XFS_LSN_CMP(threshold_lsn
, ailp
->xa_target
) <= 0)
597 * Ensure that the new target is noticed in push code before it clears
598 * the XFS_AIL_PUSHING_BIT.
601 xfs_trans_ail_copy_lsn(ailp
, &ailp
->xa_target
, &threshold_lsn
);
604 wake_up_process(ailp
->xa_task
);
608 * Push out all items in the AIL immediately
612 struct xfs_ail
*ailp
)
614 xfs_lsn_t threshold_lsn
= xfs_ail_max_lsn(ailp
);
617 xfs_ail_push(ailp
, threshold_lsn
);
621 * Push out all items in the AIL immediately and wait until the AIL is empty.
624 xfs_ail_push_all_sync(
625 struct xfs_ail
*ailp
)
627 struct xfs_log_item
*lip
;
630 spin_lock(&ailp
->xa_lock
);
631 while ((lip
= xfs_ail_max(ailp
)) != NULL
) {
632 prepare_to_wait(&ailp
->xa_empty
, &wait
, TASK_UNINTERRUPTIBLE
);
633 ailp
->xa_target
= lip
->li_lsn
;
634 wake_up_process(ailp
->xa_task
);
635 spin_unlock(&ailp
->xa_lock
);
637 spin_lock(&ailp
->xa_lock
);
639 spin_unlock(&ailp
->xa_lock
);
641 finish_wait(&ailp
->xa_empty
, &wait
);
645 * xfs_trans_ail_update - bulk AIL insertion operation.
647 * @xfs_trans_ail_update takes an array of log items that all need to be
648 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
649 * be added. Otherwise, it will be repositioned by removing it and re-adding
650 * it to the AIL. If we move the first item in the AIL, update the log tail to
651 * match the new minimum LSN in the AIL.
653 * This function takes the AIL lock once to execute the update operations on
654 * all the items in the array, and as such should not be called with the AIL
655 * lock held. As a result, once we have the AIL lock, we need to check each log
656 * item LSN to confirm it needs to be moved forward in the AIL.
658 * To optimise the insert operation, we delete all the items from the AIL in
659 * the first pass, moving them into a temporary list, then splice the temporary
660 * list into the correct position in the AIL. This avoids needing to do an
661 * insert operation on every item.
663 * This function must be called with the AIL lock held. The lock is dropped
667 xfs_trans_ail_update_bulk(
668 struct xfs_ail
*ailp
,
669 struct xfs_ail_cursor
*cur
,
670 struct xfs_log_item
**log_items
,
672 xfs_lsn_t lsn
) __releases(ailp
->xa_lock
)
674 xfs_log_item_t
*mlip
;
675 int mlip_changed
= 0;
679 ASSERT(nr_items
> 0); /* Not required, but true. */
680 mlip
= xfs_ail_min(ailp
);
682 for (i
= 0; i
< nr_items
; i
++) {
683 struct xfs_log_item
*lip
= log_items
[i
];
684 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
685 /* check if we really need to move the item */
686 if (XFS_LSN_CMP(lsn
, lip
->li_lsn
) <= 0)
689 xfs_ail_delete(ailp
, lip
);
693 lip
->li_flags
|= XFS_LI_IN_AIL
;
696 list_add(&lip
->li_ail
, &tmp
);
699 if (!list_empty(&tmp
))
700 xfs_ail_splice(ailp
, cur
, &tmp
, lsn
);
703 if (!XFS_FORCED_SHUTDOWN(ailp
->xa_mount
))
704 xlog_assign_tail_lsn_locked(ailp
->xa_mount
);
705 spin_unlock(&ailp
->xa_lock
);
707 xfs_log_space_wake(ailp
->xa_mount
);
709 spin_unlock(&ailp
->xa_lock
);
714 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
716 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
717 * removed from the AIL. The caller is already holding the AIL lock, and done
718 * all the checks necessary to ensure the items passed in via @log_items are
719 * ready for deletion. This includes checking that the items are in the AIL.
721 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
722 * flag from the item and reset the item's lsn to 0. If we remove the first
723 * item in the AIL, update the log tail to match the new minimum LSN in the
726 * This function will not drop the AIL lock until all items are removed from
727 * the AIL to minimise the amount of lock traffic on the AIL. This does not
728 * greatly increase the AIL hold time, but does significantly reduce the amount
729 * of traffic on the lock, especially during IO completion.
731 * This function must be called with the AIL lock held. The lock is dropped
735 xfs_trans_ail_delete_bulk(
736 struct xfs_ail
*ailp
,
737 struct xfs_log_item
**log_items
,
739 int shutdown_type
) __releases(ailp
->xa_lock
)
741 xfs_log_item_t
*mlip
;
742 int mlip_changed
= 0;
745 mlip
= xfs_ail_min(ailp
);
747 for (i
= 0; i
< nr_items
; i
++) {
748 struct xfs_log_item
*lip
= log_items
[i
];
749 if (!(lip
->li_flags
& XFS_LI_IN_AIL
)) {
750 struct xfs_mount
*mp
= ailp
->xa_mount
;
752 spin_unlock(&ailp
->xa_lock
);
753 if (!XFS_FORCED_SHUTDOWN(mp
)) {
754 xfs_alert_tag(mp
, XFS_PTAG_AILDELETE
,
755 "%s: attempting to delete a log item that is not in the AIL",
757 xfs_force_shutdown(mp
, shutdown_type
);
762 xfs_ail_delete(ailp
, lip
);
763 lip
->li_flags
&= ~XFS_LI_IN_AIL
;
770 if (!XFS_FORCED_SHUTDOWN(ailp
->xa_mount
))
771 xlog_assign_tail_lsn_locked(ailp
->xa_mount
);
772 if (list_empty(&ailp
->xa_ail
))
773 wake_up_all(&ailp
->xa_empty
);
774 spin_unlock(&ailp
->xa_lock
);
776 xfs_log_space_wake(ailp
->xa_mount
);
778 spin_unlock(&ailp
->xa_lock
);
786 struct xfs_ail
*ailp
;
788 ailp
= kmem_zalloc(sizeof(struct xfs_ail
), KM_MAYFAIL
);
793 INIT_LIST_HEAD(&ailp
->xa_ail
);
794 INIT_LIST_HEAD(&ailp
->xa_cursors
);
795 spin_lock_init(&ailp
->xa_lock
);
796 INIT_LIST_HEAD(&ailp
->xa_buf_list
);
797 init_waitqueue_head(&ailp
->xa_empty
);
799 ailp
->xa_task
= kthread_run(xfsaild
, ailp
, "xfsaild/%s",
800 ailp
->xa_mount
->m_fsname
);
801 if (IS_ERR(ailp
->xa_task
))
813 xfs_trans_ail_destroy(
816 struct xfs_ail
*ailp
= mp
->m_ail
;
818 kthread_stop(ailp
->xa_task
);