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)
116 struct xfs_log_item *lip = xfs_ail_min(ailp);
125 struct xfs_ail *ailp)
129 spin_lock(&ailp->ail_lock);
130 lsn = __xfs_ail_min_lsn(ailp);
131 spin_unlock(&ailp->ail_lock);
137 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
141 struct xfs_ail *ailp)
144 struct xfs_log_item *lip;
146 spin_lock(&ailp->ail_lock);
147 lip = xfs_ail_max(ailp);
150 spin_unlock(&ailp->ail_lock);
156 * The cursor keeps track of where our current traversal is up to by tracking
157 * the next item in the list for us. However, for this to be safe, removing an
158 * object from the AIL needs to invalidate any cursor that points to it. hence
159 * the traversal cursor needs to be linked to the struct xfs_ail so that
160 * deletion can search all the active cursors for invalidation.
163 xfs_trans_ail_cursor_init(
164 struct xfs_ail *ailp,
165 struct xfs_ail_cursor *cur)
168 list_add_tail(&cur->list, &ailp->ail_cursors);
172 * Get the next item in the traversal and advance the cursor. If the cursor
173 * was invalidated (indicated by a lip of 1), restart the traversal.
175 struct xfs_log_item *
176 xfs_trans_ail_cursor_next(
177 struct xfs_ail *ailp,
178 struct xfs_ail_cursor *cur)
180 struct xfs_log_item *lip = cur->item;
182 if ((uintptr_t)lip & 1)
183 lip = xfs_ail_min(ailp);
185 cur->item = xfs_ail_next(ailp, lip);
190 * When the traversal is complete, we need to remove the cursor from the list
191 * of traversing cursors.
194 xfs_trans_ail_cursor_done(
195 struct xfs_ail_cursor *cur)
198 list_del_init(&cur->list);
202 * Invalidate any cursor that is pointing to this item. This is called when an
203 * item is removed from the AIL. Any cursor pointing to this object is now
204 * invalid and the traversal needs to be terminated so it doesn't reference a
205 * freed object. We set the low bit of the cursor item pointer so we can
206 * distinguish between an invalidation and the end of the list when getting the
207 * next item from the cursor.
210 xfs_trans_ail_cursor_clear(
211 struct xfs_ail *ailp,
212 struct xfs_log_item *lip)
214 struct xfs_ail_cursor *cur;
216 list_for_each_entry(cur, &ailp->ail_cursors, list) {
217 if (cur->item == lip)
218 cur->item = (struct xfs_log_item *)
219 ((uintptr_t)cur->item | 1);
224 * Find the first item in the AIL with the given @lsn by searching in ascending
225 * LSN order and initialise the cursor to point to the next item for a
226 * ascending traversal. Pass a @lsn of zero to initialise the cursor to the
227 * first item in the AIL. Returns NULL if the list is empty.
229 struct xfs_log_item *
230 xfs_trans_ail_cursor_first(
231 struct xfs_ail *ailp,
232 struct xfs_ail_cursor *cur,
235 struct xfs_log_item *lip;
237 xfs_trans_ail_cursor_init(ailp, cur);
240 lip = xfs_ail_min(ailp);
244 list_for_each_entry(lip, &ailp->ail_head, li_ail) {
245 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
252 cur->item = xfs_ail_next(ailp, lip);
256 static struct xfs_log_item *
257 __xfs_trans_ail_cursor_last(
258 struct xfs_ail *ailp,
261 struct xfs_log_item *lip;
263 list_for_each_entry_reverse(lip, &ailp->ail_head, li_ail) {
264 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
271 * Find the last item in the AIL with the given @lsn by searching in descending
272 * LSN order and initialise the cursor to point to that item. If there is no
273 * item with the value of @lsn, then it sets the cursor to the last item with an
274 * LSN lower than @lsn. Returns NULL if the list is empty.
276 struct xfs_log_item *
277 xfs_trans_ail_cursor_last(
278 struct xfs_ail *ailp,
279 struct xfs_ail_cursor *cur,
282 xfs_trans_ail_cursor_init(ailp, cur);
283 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
288 * Splice the log item list into the AIL at the given LSN. We splice to the
289 * tail of the given LSN to maintain insert order for push traversals. The
290 * cursor is optional, allowing repeated updates to the same LSN to avoid
291 * repeated traversals. This should not be called with an empty list.
295 struct xfs_ail *ailp,
296 struct xfs_ail_cursor *cur,
297 struct list_head *list,
300 struct xfs_log_item *lip;
302 ASSERT(!list_empty(list));
305 * Use the cursor to determine the insertion point if one is
306 * provided. If not, or if the one we got is not valid,
307 * find the place in the AIL where the items belong.
309 lip = cur ? cur->item : NULL;
310 if (!lip || (uintptr_t)lip & 1)
311 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
314 * If a cursor is provided, we know we're processing the AIL
315 * in lsn order, and future items to be spliced in will
316 * follow the last one being inserted now. Update the
317 * cursor to point to that last item, now while we have a
318 * reliable pointer to it.
321 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
324 * Finally perform the splice. Unless the AIL was empty,
325 * lip points to the item in the AIL _after_ which the new
326 * items should go. If lip is null the AIL was empty, so
327 * the new items go at the head of the AIL.
330 list_splice(list, &lip->li_ail);
332 list_splice(list, &ailp->ail_head);
336 * Delete the given item from the AIL. Return a pointer to the item.
340 struct xfs_ail *ailp,
341 struct xfs_log_item *lip)
343 xfs_ail_check(ailp, lip);
344 list_del(&lip->li_ail);
345 xfs_trans_ail_cursor_clear(ailp, lip);
350 struct xfs_ail *ailp,
351 struct xfs_log_item *lip)
354 * If log item pinning is enabled, skip the push and track the item as
355 * pinned. This can help induce head-behind-tail conditions.
357 if (XFS_TEST_ERROR(false, ailp->ail_mount, XFS_ERRTAG_LOG_ITEM_PIN))
358 return XFS_ITEM_PINNED;
361 * Consider the item pinned if a push callback is not defined so the
362 * caller will force the log. This should only happen for intent items
363 * as they are unpinned once the associated done item is committed to
366 if (!lip->li_ops->iop_push)
367 return XFS_ITEM_PINNED;
368 return lip->li_ops->iop_push(lip, &ailp->ail_buf_list);
373 struct xfs_ail *ailp)
375 xfs_mount_t *mp = ailp->ail_mount;
376 struct xfs_ail_cursor cur;
377 struct xfs_log_item *lip;
386 * If we encountered pinned items or did not finish writing out all
387 * buffers the last time we ran, force the log first and wait for it
388 * before pushing again.
390 if (ailp->ail_log_flush && ailp->ail_last_pushed_lsn == 0 &&
391 (!list_empty_careful(&ailp->ail_buf_list) ||
392 xfs_ail_min_lsn(ailp))) {
393 ailp->ail_log_flush = 0;
395 XFS_STATS_INC(mp, xs_push_ail_flush);
396 xfs_log_force(mp, XFS_LOG_SYNC);
399 spin_lock(&ailp->ail_lock);
401 /* barrier matches the ail_target update in xfs_ail_push() */
403 target = ailp->ail_target;
404 ailp->ail_target_prev = target;
406 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->ail_last_pushed_lsn);
409 * If the AIL is empty or our push has reached the end we are
412 xfs_trans_ail_cursor_done(&cur);
413 spin_unlock(&ailp->ail_lock);
417 XFS_STATS_INC(mp, xs_push_ail);
420 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
424 * Note that iop_push may unlock and reacquire the AIL lock. We
425 * rely on the AIL cursor implementation to be able to deal with
428 lock_result = xfsaild_push_item(ailp, lip);
429 switch (lock_result) {
430 case XFS_ITEM_SUCCESS:
431 XFS_STATS_INC(mp, xs_push_ail_success);
432 trace_xfs_ail_push(lip);
434 ailp->ail_last_pushed_lsn = lsn;
437 case XFS_ITEM_FLUSHING:
439 * The item or its backing buffer is already being
440 * flushed. The typical reason for that is that an
441 * inode buffer is locked because we already pushed the
442 * updates to it as part of inode clustering.
444 * We do not want to to stop flushing just because lots
445 * of items are already being flushed, but we need to
446 * re-try the flushing relatively soon if most of the
447 * AIL is being flushed.
449 XFS_STATS_INC(mp, xs_push_ail_flushing);
450 trace_xfs_ail_flushing(lip);
453 ailp->ail_last_pushed_lsn = lsn;
456 case XFS_ITEM_PINNED:
457 XFS_STATS_INC(mp, xs_push_ail_pinned);
458 trace_xfs_ail_pinned(lip);
461 ailp->ail_log_flush++;
463 case XFS_ITEM_LOCKED:
464 XFS_STATS_INC(mp, xs_push_ail_locked);
465 trace_xfs_ail_locked(lip);
477 * Are there too many items we can't do anything with?
479 * If we we are skipping too many items because we can't flush
480 * them or they are already being flushed, we back off and
481 * given them time to complete whatever operation is being
482 * done. i.e. remove pressure from the AIL while we can't make
483 * progress so traversals don't slow down further inserts and
484 * removals to/from the AIL.
486 * The value of 100 is an arbitrary magic number based on
492 lip = xfs_trans_ail_cursor_next(ailp, &cur);
497 xfs_trans_ail_cursor_done(&cur);
498 spin_unlock(&ailp->ail_lock);
500 if (xfs_buf_delwri_submit_nowait(&ailp->ail_buf_list))
501 ailp->ail_log_flush++;
503 if (!count || XFS_LSN_CMP(lsn, target) >= 0) {
506 * We reached the target or the AIL is empty, so wait a bit
507 * longer for I/O to complete and remove pushed items from the
508 * AIL before we start the next scan from the start of the AIL.
511 ailp->ail_last_pushed_lsn = 0;
512 } else if (((stuck + flushing) * 100) / count > 90) {
514 * Either there is a lot of contention on the AIL or we are
515 * stuck due to operations in progress. "Stuck" in this case
516 * is defined as >90% of the items we tried to push were stuck.
518 * Backoff a bit more to allow some I/O to complete before
519 * restarting from the start of the AIL. This prevents us from
520 * spinning on the same items, and if they are pinned will all
521 * the restart to issue a log force to unpin the stuck items.
524 ailp->ail_last_pushed_lsn = 0;
527 * Assume we have more work to do in a short while.
539 struct xfs_ail *ailp = data;
540 long tout = 0; /* milliseconds */
541 unsigned int noreclaim_flag;
543 noreclaim_flag = memalloc_noreclaim_save();
547 if (tout && tout <= 20)
548 set_current_state(TASK_KILLABLE);
550 set_current_state(TASK_INTERRUPTIBLE);
553 * Check kthread_should_stop() after we set the task state to
554 * guarantee that we either see the stop bit and exit or the
555 * task state is reset to runnable such that it's not scheduled
556 * out indefinitely and detects the stop bit at next iteration.
557 * A memory barrier is included in above task state set to
558 * serialize again kthread_stop().
560 if (kthread_should_stop()) {
561 __set_current_state(TASK_RUNNING);
564 * The caller forces out the AIL before stopping the
565 * thread in the common case, which means the delwri
566 * queue is drained. In the shutdown case, the queue may
567 * still hold relogged buffers that haven't been
568 * submitted because they were pinned since added to the
571 * Log I/O error processing stales the underlying buffer
572 * and clears the delwri state, expecting the buf to be
573 * removed on the next submission attempt. That won't
574 * happen if we're shutting down, so this is the last
575 * opportunity to release such buffers from the queue.
577 ASSERT(list_empty(&ailp->ail_buf_list) ||
578 XFS_FORCED_SHUTDOWN(ailp->ail_mount));
579 xfs_buf_delwri_cancel(&ailp->ail_buf_list);
583 spin_lock(&ailp->ail_lock);
586 * Idle if the AIL is empty and we are not racing with a target
587 * update. We check the AIL after we set the task to a sleep
588 * state to guarantee that we either catch an ail_target update
589 * or that a wake_up resets the state to TASK_RUNNING.
590 * Otherwise, we run the risk of sleeping indefinitely.
592 * The barrier matches the ail_target update in xfs_ail_push().
595 if (!xfs_ail_min(ailp) &&
596 ailp->ail_target == ailp->ail_target_prev) {
597 spin_unlock(&ailp->ail_lock);
598 freezable_schedule();
602 spin_unlock(&ailp->ail_lock);
605 freezable_schedule_timeout(msecs_to_jiffies(tout));
607 __set_current_state(TASK_RUNNING);
611 tout = xfsaild_push(ailp);
614 memalloc_noreclaim_restore(noreclaim_flag);
619 * This routine is called to move the tail of the AIL forward. It does this by
620 * trying to flush items in the AIL whose lsns are below the given
623 * The push is run asynchronously in a workqueue, which means the caller needs
624 * to handle waiting on the async flush for space to become available.
625 * We don't want to interrupt any push that is in progress, hence we only queue
626 * work if we set the pushing bit appropriately.
628 * We do this unlocked - we only need to know whether there is anything in the
629 * AIL at the time we are called. We don't need to access the contents of
630 * any of the objects, so the lock is not needed.
634 struct xfs_ail *ailp,
635 xfs_lsn_t threshold_lsn)
637 struct xfs_log_item *lip;
639 lip = xfs_ail_min(ailp);
640 if (!lip || XFS_FORCED_SHUTDOWN(ailp->ail_mount) ||
641 XFS_LSN_CMP(threshold_lsn, ailp->ail_target) <= 0)
645 * Ensure that the new target is noticed in push code before it clears
646 * the XFS_AIL_PUSHING_BIT.
649 xfs_trans_ail_copy_lsn(ailp, &ailp->ail_target, &threshold_lsn);
652 wake_up_process(ailp->ail_task);
656 * Push out all items in the AIL immediately
660 struct xfs_ail *ailp)
662 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
665 xfs_ail_push(ailp, threshold_lsn);
669 * Push out all items in the AIL immediately and wait until the AIL is empty.
672 xfs_ail_push_all_sync(
673 struct xfs_ail *ailp)
675 struct xfs_log_item *lip;
678 spin_lock(&ailp->ail_lock);
679 while ((lip = xfs_ail_max(ailp)) != NULL) {
680 prepare_to_wait(&ailp->ail_empty, &wait, TASK_UNINTERRUPTIBLE);
681 ailp->ail_target = lip->li_lsn;
682 wake_up_process(ailp->ail_task);
683 spin_unlock(&ailp->ail_lock);
685 spin_lock(&ailp->ail_lock);
687 spin_unlock(&ailp->ail_lock);
689 finish_wait(&ailp->ail_empty, &wait);
693 xfs_ail_update_finish(
694 struct xfs_ail *ailp,
695 xfs_lsn_t old_lsn) __releases(ailp->ail_lock)
697 struct xfs_mount *mp = ailp->ail_mount;
699 /* if the tail lsn hasn't changed, don't do updates or wakeups. */
700 if (!old_lsn || old_lsn == __xfs_ail_min_lsn(ailp)) {
701 spin_unlock(&ailp->ail_lock);
705 if (!XFS_FORCED_SHUTDOWN(mp))
706 xlog_assign_tail_lsn_locked(mp);
708 if (list_empty(&ailp->ail_head))
709 wake_up_all(&ailp->ail_empty);
710 spin_unlock(&ailp->ail_lock);
711 xfs_log_space_wake(mp);
715 * xfs_trans_ail_update - bulk AIL insertion operation.
717 * @xfs_trans_ail_update takes an array of log items that all need to be
718 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
719 * be added. Otherwise, it will be repositioned by removing it and re-adding
720 * it to the AIL. If we move the first item in the AIL, update the log tail to
721 * match the new minimum LSN in the AIL.
723 * This function takes the AIL lock once to execute the update operations on
724 * all the items in the array, and as such should not be called with the AIL
725 * lock held. As a result, once we have the AIL lock, we need to check each log
726 * item LSN to confirm it needs to be moved forward in the AIL.
728 * To optimise the insert operation, we delete all the items from the AIL in
729 * the first pass, moving them into a temporary list, then splice the temporary
730 * list into the correct position in the AIL. This avoids needing to do an
731 * insert operation on every item.
733 * This function must be called with the AIL lock held. The lock is dropped
737 xfs_trans_ail_update_bulk(
738 struct xfs_ail *ailp,
739 struct xfs_ail_cursor *cur,
740 struct xfs_log_item **log_items,
742 xfs_lsn_t lsn) __releases(ailp->ail_lock)
744 struct xfs_log_item *mlip;
745 xfs_lsn_t tail_lsn = 0;
749 ASSERT(nr_items > 0); /* Not required, but true. */
750 mlip = xfs_ail_min(ailp);
752 for (i = 0; i < nr_items; i++) {
753 struct xfs_log_item *lip = log_items[i];
754 if (test_and_set_bit(XFS_LI_IN_AIL, &lip->li_flags)) {
755 /* check if we really need to move the item */
756 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
759 trace_xfs_ail_move(lip, lip->li_lsn, lsn);
760 if (mlip == lip && !tail_lsn)
761 tail_lsn = lip->li_lsn;
763 xfs_ail_delete(ailp, lip);
765 trace_xfs_ail_insert(lip, 0, lsn);
768 list_add(&lip->li_ail, &tmp);
771 if (!list_empty(&tmp))
772 xfs_ail_splice(ailp, cur, &tmp, lsn);
774 xfs_ail_update_finish(ailp, tail_lsn);
778 * Delete one log item from the AIL.
780 * If this item was at the tail of the AIL, return the LSN of the log item so
781 * that we can use it to check if the LSN of the tail of the log has moved
782 * when finishing up the AIL delete process in xfs_ail_update_finish().
786 struct xfs_ail *ailp,
787 struct xfs_log_item *lip)
789 struct xfs_log_item *mlip = xfs_ail_min(ailp);
790 xfs_lsn_t lsn = lip->li_lsn;
792 trace_xfs_ail_delete(lip, mlip->li_lsn, lip->li_lsn);
793 xfs_ail_delete(ailp, lip);
794 xfs_clear_li_failed(lip);
795 clear_bit(XFS_LI_IN_AIL, &lip->li_flags);
804 * Remove a log items from the AIL
806 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
807 * removed from the AIL. The caller is already holding the AIL lock, and done
808 * all the checks necessary to ensure the items passed in via @log_items are
809 * ready for deletion. This includes checking that the items are in the AIL.
811 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
812 * flag from the item and reset the item's lsn to 0. If we remove the first
813 * item in the AIL, update the log tail to match the new minimum LSN in the
816 * This function will not drop the AIL lock until all items are removed from
817 * the AIL to minimise the amount of lock traffic on the AIL. This does not
818 * greatly increase the AIL hold time, but does significantly reduce the amount
819 * of traffic on the lock, especially during IO completion.
821 * This function must be called with the AIL lock held. The lock is dropped
825 xfs_trans_ail_delete(
826 struct xfs_ail *ailp,
827 struct xfs_log_item *lip,
830 struct xfs_mount *mp = ailp->ail_mount;
833 if (!test_bit(XFS_LI_IN_AIL, &lip->li_flags)) {
834 spin_unlock(&ailp->ail_lock);
835 if (!XFS_FORCED_SHUTDOWN(mp)) {
836 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
837 "%s: attempting to delete a log item that is not in the AIL",
839 xfs_force_shutdown(mp, shutdown_type);
844 tail_lsn = xfs_ail_delete_one(ailp, lip);
845 xfs_ail_update_finish(ailp, tail_lsn);
852 struct xfs_ail *ailp;
854 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
858 ailp->ail_mount = mp;
859 INIT_LIST_HEAD(&ailp->ail_head);
860 INIT_LIST_HEAD(&ailp->ail_cursors);
861 spin_lock_init(&ailp->ail_lock);
862 INIT_LIST_HEAD(&ailp->ail_buf_list);
863 init_waitqueue_head(&ailp->ail_empty);
865 ailp->ail_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
866 ailp->ail_mount->m_super->s_id);
867 if (IS_ERR(ailp->ail_task))
879 xfs_trans_ail_destroy(
882 struct xfs_ail *ailp = mp->m_ail;
884 kthread_stop(ailp->ail_task);