1 // SPDX-License-Identifier: GPL-2.0
3 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4 * Copyright (c) 2008 Dave Chinner
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_trans.h"
15 #include "xfs_trans_priv.h"
16 #include "xfs_trace.h"
17 #include "xfs_errortag.h"
18 #include "xfs_error.h"
23 * Check that the list is sorted as it should be.
25 * Called with the ail lock held, but we don't want to assert fail with it
26 * held otherwise we'll lock everything up and won't be able to debug the
27 * cause. Hence we sample and check the state under the AIL lock and return if
28 * everything is fine, otherwise we drop the lock and run the ASSERT checks.
29 * Asserts may not be fatal, so pick the lock back up and continue onwards.
34 struct xfs_log_item *lip)
35 __must_hold(&ailp->ail_lock)
37 struct xfs_log_item *prev_lip;
38 struct xfs_log_item *next_lip;
39 xfs_lsn_t prev_lsn = NULLCOMMITLSN;
40 xfs_lsn_t next_lsn = NULLCOMMITLSN;
45 if (list_empty(&ailp->ail_head))
49 * Sample then check the next and previous entries are valid.
51 in_ail = test_bit(XFS_LI_IN_AIL, &lip->li_flags);
52 prev_lip = list_entry(lip->li_ail.prev, struct xfs_log_item, li_ail);
53 if (&prev_lip->li_ail != &ailp->ail_head)
54 prev_lsn = prev_lip->li_lsn;
55 next_lip = list_entry(lip->li_ail.next, struct xfs_log_item, li_ail);
56 if (&next_lip->li_ail != &ailp->ail_head)
57 next_lsn = next_lip->li_lsn;
61 (prev_lsn == NULLCOMMITLSN || XFS_LSN_CMP(prev_lsn, lsn) <= 0) &&
62 (next_lsn == NULLCOMMITLSN || XFS_LSN_CMP(next_lsn, lsn) >= 0))
65 spin_unlock(&ailp->ail_lock);
67 ASSERT(prev_lsn == NULLCOMMITLSN || XFS_LSN_CMP(prev_lsn, lsn) <= 0);
68 ASSERT(next_lsn == NULLCOMMITLSN || XFS_LSN_CMP(next_lsn, lsn) >= 0);
69 spin_lock(&ailp->ail_lock);
72 #define xfs_ail_check(a,l)
76 * Return a pointer to the last item in the AIL. If the AIL is empty, then
79 static struct xfs_log_item *
83 if (list_empty(&ailp->ail_head))
86 return list_entry(ailp->ail_head.prev, struct xfs_log_item, li_ail);
90 * Return a pointer to the item which follows the given item in the AIL. If
91 * the given item is the last item in the list, then return NULL.
93 static struct xfs_log_item *
96 struct xfs_log_item *lip)
98 if (lip->li_ail.next == &ailp->ail_head)
101 return list_first_entry(&lip->li_ail, struct xfs_log_item, li_ail);
105 * This is called by the log manager code to determine the LSN of the tail of
106 * the log. This is exactly the LSN of the first item in the AIL. If the AIL
107 * is empty, then this function returns 0.
109 * We need the AIL lock in order to get a coherent read of the lsn of the last
114 struct xfs_ail *ailp)
117 struct xfs_log_item *lip;
119 spin_lock(&ailp->ail_lock);
120 lip = xfs_ail_min(ailp);
123 spin_unlock(&ailp->ail_lock);
129 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
133 struct xfs_ail *ailp)
136 struct xfs_log_item *lip;
138 spin_lock(&ailp->ail_lock);
139 lip = xfs_ail_max(ailp);
142 spin_unlock(&ailp->ail_lock);
148 * The cursor keeps track of where our current traversal is up to by tracking
149 * the next item in the list for us. However, for this to be safe, removing an
150 * object from the AIL needs to invalidate any cursor that points to it. hence
151 * the traversal cursor needs to be linked to the struct xfs_ail so that
152 * deletion can search all the active cursors for invalidation.
155 xfs_trans_ail_cursor_init(
156 struct xfs_ail *ailp,
157 struct xfs_ail_cursor *cur)
160 list_add_tail(&cur->list, &ailp->ail_cursors);
164 * Get the next item in the traversal and advance the cursor. If the cursor
165 * was invalidated (indicated by a lip of 1), restart the traversal.
167 struct xfs_log_item *
168 xfs_trans_ail_cursor_next(
169 struct xfs_ail *ailp,
170 struct xfs_ail_cursor *cur)
172 struct xfs_log_item *lip = cur->item;
174 if ((uintptr_t)lip & 1)
175 lip = xfs_ail_min(ailp);
177 cur->item = xfs_ail_next(ailp, lip);
182 * When the traversal is complete, we need to remove the cursor from the list
183 * of traversing cursors.
186 xfs_trans_ail_cursor_done(
187 struct xfs_ail_cursor *cur)
190 list_del_init(&cur->list);
194 * Invalidate any cursor that is pointing to this item. This is called when an
195 * item is removed from the AIL. Any cursor pointing to this object is now
196 * invalid and the traversal needs to be terminated so it doesn't reference a
197 * freed object. We set the low bit of the cursor item pointer so we can
198 * distinguish between an invalidation and the end of the list when getting the
199 * next item from the cursor.
202 xfs_trans_ail_cursor_clear(
203 struct xfs_ail *ailp,
204 struct xfs_log_item *lip)
206 struct xfs_ail_cursor *cur;
208 list_for_each_entry(cur, &ailp->ail_cursors, list) {
209 if (cur->item == lip)
210 cur->item = (struct xfs_log_item *)
211 ((uintptr_t)cur->item | 1);
216 * Find the first item in the AIL with the given @lsn by searching in ascending
217 * LSN order and initialise the cursor to point to the next item for a
218 * ascending traversal. Pass a @lsn of zero to initialise the cursor to the
219 * first item in the AIL. Returns NULL if the list is empty.
221 struct xfs_log_item *
222 xfs_trans_ail_cursor_first(
223 struct xfs_ail *ailp,
224 struct xfs_ail_cursor *cur,
227 struct xfs_log_item *lip;
229 xfs_trans_ail_cursor_init(ailp, cur);
232 lip = xfs_ail_min(ailp);
236 list_for_each_entry(lip, &ailp->ail_head, li_ail) {
237 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
244 cur->item = xfs_ail_next(ailp, lip);
248 static struct xfs_log_item *
249 __xfs_trans_ail_cursor_last(
250 struct xfs_ail *ailp,
253 struct xfs_log_item *lip;
255 list_for_each_entry_reverse(lip, &ailp->ail_head, li_ail) {
256 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
263 * Find the last item in the AIL with the given @lsn by searching in descending
264 * LSN order and initialise the cursor to point to that item. If there is no
265 * item with the value of @lsn, then it sets the cursor to the last item with an
266 * LSN lower than @lsn. Returns NULL if the list is empty.
268 struct xfs_log_item *
269 xfs_trans_ail_cursor_last(
270 struct xfs_ail *ailp,
271 struct xfs_ail_cursor *cur,
274 xfs_trans_ail_cursor_init(ailp, cur);
275 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
280 * Splice the log item list into the AIL at the given LSN. We splice to the
281 * tail of the given LSN to maintain insert order for push traversals. The
282 * cursor is optional, allowing repeated updates to the same LSN to avoid
283 * repeated traversals. This should not be called with an empty list.
287 struct xfs_ail *ailp,
288 struct xfs_ail_cursor *cur,
289 struct list_head *list,
292 struct xfs_log_item *lip;
294 ASSERT(!list_empty(list));
297 * Use the cursor to determine the insertion point if one is
298 * provided. If not, or if the one we got is not valid,
299 * find the place in the AIL where the items belong.
301 lip = cur ? cur->item : NULL;
302 if (!lip || (uintptr_t)lip & 1)
303 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
306 * If a cursor is provided, we know we're processing the AIL
307 * in lsn order, and future items to be spliced in will
308 * follow the last one being inserted now. Update the
309 * cursor to point to that last item, now while we have a
310 * reliable pointer to it.
313 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
316 * Finally perform the splice. Unless the AIL was empty,
317 * lip points to the item in the AIL _after_ which the new
318 * items should go. If lip is null the AIL was empty, so
319 * the new items go at the head of the AIL.
322 list_splice(list, &lip->li_ail);
324 list_splice(list, &ailp->ail_head);
328 * Delete the given item from the AIL. Return a pointer to the item.
332 struct xfs_ail *ailp,
333 struct xfs_log_item *lip)
335 xfs_ail_check(ailp, lip);
336 list_del(&lip->li_ail);
337 xfs_trans_ail_cursor_clear(ailp, lip);
342 struct xfs_ail *ailp,
343 struct xfs_log_item *lip)
346 * If log item pinning is enabled, skip the push and track the item as
347 * pinned. This can help induce head-behind-tail conditions.
349 if (XFS_TEST_ERROR(false, ailp->ail_mount, XFS_ERRTAG_LOG_ITEM_PIN))
350 return XFS_ITEM_PINNED;
353 * Consider the item pinned if a push callback is not defined so the
354 * caller will force the log. This should only happen for intent items
355 * as they are unpinned once the associated done item is committed to
358 if (!lip->li_ops->iop_push)
359 return XFS_ITEM_PINNED;
360 return lip->li_ops->iop_push(lip, &ailp->ail_buf_list);
365 struct xfs_ail *ailp)
367 xfs_mount_t *mp = ailp->ail_mount;
368 struct xfs_ail_cursor cur;
369 struct xfs_log_item *lip;
378 * If we encountered pinned items or did not finish writing out all
379 * buffers the last time we ran, force the log first and wait for it
380 * before pushing again.
382 if (ailp->ail_log_flush && ailp->ail_last_pushed_lsn == 0 &&
383 (!list_empty_careful(&ailp->ail_buf_list) ||
384 xfs_ail_min_lsn(ailp))) {
385 ailp->ail_log_flush = 0;
387 XFS_STATS_INC(mp, xs_push_ail_flush);
388 xfs_log_force(mp, XFS_LOG_SYNC);
391 spin_lock(&ailp->ail_lock);
393 /* barrier matches the ail_target update in xfs_ail_push() */
395 target = ailp->ail_target;
396 ailp->ail_target_prev = target;
398 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->ail_last_pushed_lsn);
401 * If the AIL is empty or our push has reached the end we are
404 xfs_trans_ail_cursor_done(&cur);
405 spin_unlock(&ailp->ail_lock);
409 XFS_STATS_INC(mp, xs_push_ail);
412 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
416 * Note that iop_push may unlock and reacquire the AIL lock. We
417 * rely on the AIL cursor implementation to be able to deal with
420 lock_result = xfsaild_push_item(ailp, lip);
421 switch (lock_result) {
422 case XFS_ITEM_SUCCESS:
423 XFS_STATS_INC(mp, xs_push_ail_success);
424 trace_xfs_ail_push(lip);
426 ailp->ail_last_pushed_lsn = lsn;
429 case XFS_ITEM_FLUSHING:
431 * The item or its backing buffer is already being
432 * flushed. The typical reason for that is that an
433 * inode buffer is locked because we already pushed the
434 * updates to it as part of inode clustering.
436 * We do not want to to stop flushing just because lots
437 * of items are already being flushed, but we need to
438 * re-try the flushing relatively soon if most of the
439 * AIL is being flushed.
441 XFS_STATS_INC(mp, xs_push_ail_flushing);
442 trace_xfs_ail_flushing(lip);
445 ailp->ail_last_pushed_lsn = lsn;
448 case XFS_ITEM_PINNED:
449 XFS_STATS_INC(mp, xs_push_ail_pinned);
450 trace_xfs_ail_pinned(lip);
453 ailp->ail_log_flush++;
455 case XFS_ITEM_LOCKED:
456 XFS_STATS_INC(mp, xs_push_ail_locked);
457 trace_xfs_ail_locked(lip);
469 * Are there too many items we can't do anything with?
471 * If we we are skipping too many items because we can't flush
472 * them or they are already being flushed, we back off and
473 * given them time to complete whatever operation is being
474 * done. i.e. remove pressure from the AIL while we can't make
475 * progress so traversals don't slow down further inserts and
476 * removals to/from the AIL.
478 * The value of 100 is an arbitrary magic number based on
484 lip = xfs_trans_ail_cursor_next(ailp, &cur);
489 xfs_trans_ail_cursor_done(&cur);
490 spin_unlock(&ailp->ail_lock);
492 if (xfs_buf_delwri_submit_nowait(&ailp->ail_buf_list))
493 ailp->ail_log_flush++;
495 if (!count || XFS_LSN_CMP(lsn, target) >= 0) {
498 * We reached the target or the AIL is empty, so wait a bit
499 * longer for I/O to complete and remove pushed items from the
500 * AIL before we start the next scan from the start of the AIL.
503 ailp->ail_last_pushed_lsn = 0;
504 } else if (((stuck + flushing) * 100) / count > 90) {
506 * Either there is a lot of contention on the AIL or we are
507 * stuck due to operations in progress. "Stuck" in this case
508 * is defined as >90% of the items we tried to push were stuck.
510 * Backoff a bit more to allow some I/O to complete before
511 * restarting from the start of the AIL. This prevents us from
512 * spinning on the same items, and if they are pinned will all
513 * the restart to issue a log force to unpin the stuck items.
516 ailp->ail_last_pushed_lsn = 0;
519 * Assume we have more work to do in a short while.
531 struct xfs_ail *ailp = data;
532 long tout = 0; /* milliseconds */
533 unsigned int noreclaim_flag;
535 noreclaim_flag = memalloc_noreclaim_save();
539 if (tout && tout <= 20)
540 set_current_state(TASK_KILLABLE);
542 set_current_state(TASK_INTERRUPTIBLE);
545 * Check kthread_should_stop() after we set the task state to
546 * guarantee that we either see the stop bit and exit or the
547 * task state is reset to runnable such that it's not scheduled
548 * out indefinitely and detects the stop bit at next iteration.
549 * A memory barrier is included in above task state set to
550 * serialize again kthread_stop().
552 if (kthread_should_stop()) {
553 __set_current_state(TASK_RUNNING);
556 * The caller forces out the AIL before stopping the
557 * thread in the common case, which means the delwri
558 * queue is drained. In the shutdown case, the queue may
559 * still hold relogged buffers that haven't been
560 * submitted because they were pinned since added to the
563 * Log I/O error processing stales the underlying buffer
564 * and clears the delwri state, expecting the buf to be
565 * removed on the next submission attempt. That won't
566 * happen if we're shutting down, so this is the last
567 * opportunity to release such buffers from the queue.
569 ASSERT(list_empty(&ailp->ail_buf_list) ||
570 XFS_FORCED_SHUTDOWN(ailp->ail_mount));
571 xfs_buf_delwri_cancel(&ailp->ail_buf_list);
575 spin_lock(&ailp->ail_lock);
578 * Idle if the AIL is empty and we are not racing with a target
579 * update. We check the AIL after we set the task to a sleep
580 * state to guarantee that we either catch an ail_target update
581 * or that a wake_up resets the state to TASK_RUNNING.
582 * Otherwise, we run the risk of sleeping indefinitely.
584 * The barrier matches the ail_target update in xfs_ail_push().
587 if (!xfs_ail_min(ailp) &&
588 ailp->ail_target == ailp->ail_target_prev) {
589 spin_unlock(&ailp->ail_lock);
590 freezable_schedule();
594 spin_unlock(&ailp->ail_lock);
597 freezable_schedule_timeout(msecs_to_jiffies(tout));
599 __set_current_state(TASK_RUNNING);
603 tout = xfsaild_push(ailp);
606 memalloc_noreclaim_restore(noreclaim_flag);
611 * This routine is called to move the tail of the AIL forward. It does this by
612 * trying to flush items in the AIL whose lsns are below the given
615 * The push is run asynchronously in a workqueue, which means the caller needs
616 * to handle waiting on the async flush for space to become available.
617 * We don't want to interrupt any push that is in progress, hence we only queue
618 * work if we set the pushing bit appropriately.
620 * We do this unlocked - we only need to know whether there is anything in the
621 * AIL at the time we are called. We don't need to access the contents of
622 * any of the objects, so the lock is not needed.
626 struct xfs_ail *ailp,
627 xfs_lsn_t threshold_lsn)
629 struct xfs_log_item *lip;
631 lip = xfs_ail_min(ailp);
632 if (!lip || XFS_FORCED_SHUTDOWN(ailp->ail_mount) ||
633 XFS_LSN_CMP(threshold_lsn, ailp->ail_target) <= 0)
637 * Ensure that the new target is noticed in push code before it clears
638 * the XFS_AIL_PUSHING_BIT.
641 xfs_trans_ail_copy_lsn(ailp, &ailp->ail_target, &threshold_lsn);
644 wake_up_process(ailp->ail_task);
648 * Push out all items in the AIL immediately
652 struct xfs_ail *ailp)
654 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
657 xfs_ail_push(ailp, threshold_lsn);
661 * Push out all items in the AIL immediately and wait until the AIL is empty.
664 xfs_ail_push_all_sync(
665 struct xfs_ail *ailp)
667 struct xfs_log_item *lip;
670 spin_lock(&ailp->ail_lock);
671 while ((lip = xfs_ail_max(ailp)) != NULL) {
672 prepare_to_wait(&ailp->ail_empty, &wait, TASK_UNINTERRUPTIBLE);
673 ailp->ail_target = lip->li_lsn;
674 wake_up_process(ailp->ail_task);
675 spin_unlock(&ailp->ail_lock);
677 spin_lock(&ailp->ail_lock);
679 spin_unlock(&ailp->ail_lock);
681 finish_wait(&ailp->ail_empty, &wait);
685 * xfs_trans_ail_update - bulk AIL insertion operation.
687 * @xfs_trans_ail_update takes an array of log items that all need to be
688 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
689 * be added. Otherwise, it will be repositioned by removing it and re-adding
690 * it to the AIL. If we move the first item in the AIL, update the log tail to
691 * match the new minimum LSN in the AIL.
693 * This function takes the AIL lock once to execute the update operations on
694 * all the items in the array, and as such should not be called with the AIL
695 * lock held. As a result, once we have the AIL lock, we need to check each log
696 * item LSN to confirm it needs to be moved forward in the AIL.
698 * To optimise the insert operation, we delete all the items from the AIL in
699 * the first pass, moving them into a temporary list, then splice the temporary
700 * list into the correct position in the AIL. This avoids needing to do an
701 * insert operation on every item.
703 * This function must be called with the AIL lock held. The lock is dropped
707 xfs_trans_ail_update_bulk(
708 struct xfs_ail *ailp,
709 struct xfs_ail_cursor *cur,
710 struct xfs_log_item **log_items,
712 xfs_lsn_t lsn) __releases(ailp->ail_lock)
714 struct xfs_log_item *mlip;
715 int mlip_changed = 0;
719 ASSERT(nr_items > 0); /* Not required, but true. */
720 mlip = xfs_ail_min(ailp);
722 for (i = 0; i < nr_items; i++) {
723 struct xfs_log_item *lip = log_items[i];
724 if (test_and_set_bit(XFS_LI_IN_AIL, &lip->li_flags)) {
725 /* check if we really need to move the item */
726 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
729 trace_xfs_ail_move(lip, lip->li_lsn, lsn);
730 xfs_ail_delete(ailp, lip);
734 trace_xfs_ail_insert(lip, 0, lsn);
737 list_add(&lip->li_ail, &tmp);
740 if (!list_empty(&tmp))
741 xfs_ail_splice(ailp, cur, &tmp, lsn);
744 if (!XFS_FORCED_SHUTDOWN(ailp->ail_mount))
745 xlog_assign_tail_lsn_locked(ailp->ail_mount);
746 spin_unlock(&ailp->ail_lock);
748 xfs_log_space_wake(ailp->ail_mount);
750 spin_unlock(&ailp->ail_lock);
756 struct xfs_ail *ailp,
757 struct xfs_log_item *lip)
759 struct xfs_log_item *mlip = xfs_ail_min(ailp);
761 trace_xfs_ail_delete(lip, mlip->li_lsn, lip->li_lsn);
762 xfs_ail_delete(ailp, lip);
763 xfs_clear_li_failed(lip);
764 clear_bit(XFS_LI_IN_AIL, &lip->li_flags);
771 * Remove a log items from the AIL
773 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
774 * removed from the AIL. The caller is already holding the AIL lock, and done
775 * all the checks necessary to ensure the items passed in via @log_items are
776 * ready for deletion. This includes checking that the items are in the AIL.
778 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
779 * flag from the item and reset the item's lsn to 0. If we remove the first
780 * item in the AIL, update the log tail to match the new minimum LSN in the
783 * This function will not drop the AIL lock until all items are removed from
784 * the AIL to minimise the amount of lock traffic on the AIL. This does not
785 * greatly increase the AIL hold time, but does significantly reduce the amount
786 * of traffic on the lock, especially during IO completion.
788 * This function must be called with the AIL lock held. The lock is dropped
792 xfs_trans_ail_delete(
793 struct xfs_ail *ailp,
794 struct xfs_log_item *lip,
795 int shutdown_type) __releases(ailp->ail_lock)
797 struct xfs_mount *mp = ailp->ail_mount;
800 if (!test_bit(XFS_LI_IN_AIL, &lip->li_flags)) {
801 spin_unlock(&ailp->ail_lock);
802 if (!XFS_FORCED_SHUTDOWN(mp)) {
803 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
804 "%s: attempting to delete a log item that is not in the AIL",
806 xfs_force_shutdown(mp, shutdown_type);
811 mlip_changed = xfs_ail_delete_one(ailp, lip);
813 if (!XFS_FORCED_SHUTDOWN(mp))
814 xlog_assign_tail_lsn_locked(mp);
815 if (list_empty(&ailp->ail_head))
816 wake_up_all(&ailp->ail_empty);
819 spin_unlock(&ailp->ail_lock);
821 xfs_log_space_wake(ailp->ail_mount);
828 struct xfs_ail *ailp;
830 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
834 ailp->ail_mount = mp;
835 INIT_LIST_HEAD(&ailp->ail_head);
836 INIT_LIST_HEAD(&ailp->ail_cursors);
837 spin_lock_init(&ailp->ail_lock);
838 INIT_LIST_HEAD(&ailp->ail_buf_list);
839 init_waitqueue_head(&ailp->ail_empty);
841 ailp->ail_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
842 ailp->ail_mount->m_super->s_id);
843 if (IS_ERR(ailp->ail_task))
855 xfs_trans_ail_destroy(
858 struct xfs_ail *ailp = mp->m_ail;
860 kthread_stop(ailp->ail_task);