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_format.h"
10 #include "xfs_log_format.h"
11 #include "xfs_trans_resv.h"
12 #include "xfs_mount.h"
13 #include "xfs_trans.h"
14 #include "xfs_trans_priv.h"
15 #include "xfs_trace.h"
16 #include "xfs_errortag.h"
17 #include "xfs_error.h"
22 * Check that the list is sorted as it should be.
24 * Called with the ail lock held, but we don't want to assert fail with it
25 * held otherwise we'll lock everything up and won't be able to debug the
26 * cause. Hence we sample and check the state under the AIL lock and return if
27 * everything is fine, otherwise we drop the lock and run the ASSERT checks.
28 * Asserts may not be fatal, so pick the lock back up and continue onwards.
33 struct xfs_log_item *lip)
35 struct xfs_log_item *prev_lip;
36 struct xfs_log_item *next_lip;
37 xfs_lsn_t prev_lsn = NULLCOMMITLSN;
38 xfs_lsn_t next_lsn = NULLCOMMITLSN;
43 if (list_empty(&ailp->ail_head))
47 * Sample then check the next and previous entries are valid.
49 in_ail = test_bit(XFS_LI_IN_AIL, &lip->li_flags);
50 prev_lip = list_entry(lip->li_ail.prev, struct xfs_log_item, li_ail);
51 if (&prev_lip->li_ail != &ailp->ail_head)
52 prev_lsn = prev_lip->li_lsn;
53 next_lip = list_entry(lip->li_ail.next, struct xfs_log_item, li_ail);
54 if (&next_lip->li_ail != &ailp->ail_head)
55 next_lsn = next_lip->li_lsn;
59 (prev_lsn == NULLCOMMITLSN || XFS_LSN_CMP(prev_lsn, lsn) <= 0) &&
60 (next_lsn == NULLCOMMITLSN || XFS_LSN_CMP(next_lsn, lsn) >= 0))
63 spin_unlock(&ailp->ail_lock);
65 ASSERT(prev_lsn == NULLCOMMITLSN || XFS_LSN_CMP(prev_lsn, lsn) <= 0);
66 ASSERT(next_lsn == NULLCOMMITLSN || XFS_LSN_CMP(next_lsn, lsn) >= 0);
67 spin_lock(&ailp->ail_lock);
70 #define xfs_ail_check(a,l)
74 * Return a pointer to the last item in the AIL. If the AIL is empty, then
77 static xfs_log_item_t *
81 if (list_empty(&ailp->ail_head))
84 return list_entry(ailp->ail_head.prev, xfs_log_item_t, li_ail);
88 * Return a pointer to the item which follows the given item in the AIL. If
89 * the given item is the last item in the list, then return NULL.
91 static xfs_log_item_t *
96 if (lip->li_ail.next == &ailp->ail_head)
99 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
103 * This is called by the log manager code to determine the LSN of the tail of
104 * the log. This is exactly the LSN of the first item in the AIL. If the AIL
105 * is empty, then this function returns 0.
107 * We need the AIL lock in order to get a coherent read of the lsn of the last
112 struct xfs_ail *ailp)
117 spin_lock(&ailp->ail_lock);
118 lip = xfs_ail_min(ailp);
121 spin_unlock(&ailp->ail_lock);
127 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
131 struct xfs_ail *ailp)
136 spin_lock(&ailp->ail_lock);
137 lip = xfs_ail_max(ailp);
140 spin_unlock(&ailp->ail_lock);
146 * The cursor keeps track of where our current traversal is up to by tracking
147 * the next item in the list for us. However, for this to be safe, removing an
148 * object from the AIL needs to invalidate any cursor that points to it. hence
149 * the traversal cursor needs to be linked to the struct xfs_ail so that
150 * deletion can search all the active cursors for invalidation.
153 xfs_trans_ail_cursor_init(
154 struct xfs_ail *ailp,
155 struct xfs_ail_cursor *cur)
158 list_add_tail(&cur->list, &ailp->ail_cursors);
162 * Get the next item in the traversal and advance the cursor. If the cursor
163 * was invalidated (indicated by a lip of 1), restart the traversal.
165 struct xfs_log_item *
166 xfs_trans_ail_cursor_next(
167 struct xfs_ail *ailp,
168 struct xfs_ail_cursor *cur)
170 struct xfs_log_item *lip = cur->item;
172 if ((uintptr_t)lip & 1)
173 lip = xfs_ail_min(ailp);
175 cur->item = xfs_ail_next(ailp, lip);
180 * When the traversal is complete, we need to remove the cursor from the list
181 * of traversing cursors.
184 xfs_trans_ail_cursor_done(
185 struct xfs_ail_cursor *cur)
188 list_del_init(&cur->list);
192 * Invalidate any cursor that is pointing to this item. This is called when an
193 * item is removed from the AIL. Any cursor pointing to this object is now
194 * invalid and the traversal needs to be terminated so it doesn't reference a
195 * freed object. We set the low bit of the cursor item pointer so we can
196 * distinguish between an invalidation and the end of the list when getting the
197 * next item from the cursor.
200 xfs_trans_ail_cursor_clear(
201 struct xfs_ail *ailp,
202 struct xfs_log_item *lip)
204 struct xfs_ail_cursor *cur;
206 list_for_each_entry(cur, &ailp->ail_cursors, list) {
207 if (cur->item == lip)
208 cur->item = (struct xfs_log_item *)
209 ((uintptr_t)cur->item | 1);
214 * Find the first item in the AIL with the given @lsn by searching in ascending
215 * LSN order and initialise the cursor to point to the next item for a
216 * ascending traversal. Pass a @lsn of zero to initialise the cursor to the
217 * first item in the AIL. Returns NULL if the list is empty.
220 xfs_trans_ail_cursor_first(
221 struct xfs_ail *ailp,
222 struct xfs_ail_cursor *cur,
227 xfs_trans_ail_cursor_init(ailp, cur);
230 lip = xfs_ail_min(ailp);
234 list_for_each_entry(lip, &ailp->ail_head, li_ail) {
235 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
242 cur->item = xfs_ail_next(ailp, lip);
246 static struct xfs_log_item *
247 __xfs_trans_ail_cursor_last(
248 struct xfs_ail *ailp,
253 list_for_each_entry_reverse(lip, &ailp->ail_head, li_ail) {
254 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
261 * Find the last item in the AIL with the given @lsn by searching in descending
262 * LSN order and initialise the cursor to point to that item. If there is no
263 * item with the value of @lsn, then it sets the cursor to the last item with an
264 * LSN lower than @lsn. Returns NULL if the list is empty.
266 struct xfs_log_item *
267 xfs_trans_ail_cursor_last(
268 struct xfs_ail *ailp,
269 struct xfs_ail_cursor *cur,
272 xfs_trans_ail_cursor_init(ailp, cur);
273 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
278 * Splice the log item list into the AIL at the given LSN. We splice to the
279 * tail of the given LSN to maintain insert order for push traversals. The
280 * cursor is optional, allowing repeated updates to the same LSN to avoid
281 * repeated traversals. This should not be called with an empty list.
285 struct xfs_ail *ailp,
286 struct xfs_ail_cursor *cur,
287 struct list_head *list,
290 struct xfs_log_item *lip;
292 ASSERT(!list_empty(list));
295 * Use the cursor to determine the insertion point if one is
296 * provided. If not, or if the one we got is not valid,
297 * find the place in the AIL where the items belong.
299 lip = cur ? cur->item : NULL;
300 if (!lip || (uintptr_t)lip & 1)
301 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
304 * If a cursor is provided, we know we're processing the AIL
305 * in lsn order, and future items to be spliced in will
306 * follow the last one being inserted now. Update the
307 * cursor to point to that last item, now while we have a
308 * reliable pointer to it.
311 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
314 * Finally perform the splice. Unless the AIL was empty,
315 * lip points to the item in the AIL _after_ which the new
316 * items should go. If lip is null the AIL was empty, so
317 * the new items go at the head of the AIL.
320 list_splice(list, &lip->li_ail);
322 list_splice(list, &ailp->ail_head);
326 * Delete the given item from the AIL. Return a pointer to the item.
330 struct xfs_ail *ailp,
333 xfs_ail_check(ailp, lip);
334 list_del(&lip->li_ail);
335 xfs_trans_ail_cursor_clear(ailp, lip);
340 struct xfs_ail *ailp,
341 struct xfs_log_item *lip)
344 * If log item pinning is enabled, skip the push and track the item as
345 * pinned. This can help induce head-behind-tail conditions.
347 if (XFS_TEST_ERROR(false, ailp->ail_mount, XFS_ERRTAG_LOG_ITEM_PIN))
348 return XFS_ITEM_PINNED;
350 return lip->li_ops->iop_push(lip, &ailp->ail_buf_list);
355 struct xfs_ail *ailp)
357 xfs_mount_t *mp = ailp->ail_mount;
358 struct xfs_ail_cursor cur;
368 * If we encountered pinned items or did not finish writing out all
369 * buffers the last time we ran, force the log first and wait for it
370 * before pushing again.
372 if (ailp->ail_log_flush && ailp->ail_last_pushed_lsn == 0 &&
373 (!list_empty_careful(&ailp->ail_buf_list) ||
374 xfs_ail_min_lsn(ailp))) {
375 ailp->ail_log_flush = 0;
377 XFS_STATS_INC(mp, xs_push_ail_flush);
378 xfs_log_force(mp, XFS_LOG_SYNC);
381 spin_lock(&ailp->ail_lock);
383 /* barrier matches the ail_target update in xfs_ail_push() */
385 target = ailp->ail_target;
386 ailp->ail_target_prev = target;
388 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->ail_last_pushed_lsn);
391 * If the AIL is empty or our push has reached the end we are
394 xfs_trans_ail_cursor_done(&cur);
395 spin_unlock(&ailp->ail_lock);
399 XFS_STATS_INC(mp, xs_push_ail);
402 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
406 * Note that iop_push may unlock and reacquire the AIL lock. We
407 * rely on the AIL cursor implementation to be able to deal with
410 lock_result = xfsaild_push_item(ailp, lip);
411 switch (lock_result) {
412 case XFS_ITEM_SUCCESS:
413 XFS_STATS_INC(mp, xs_push_ail_success);
414 trace_xfs_ail_push(lip);
416 ailp->ail_last_pushed_lsn = lsn;
419 case XFS_ITEM_FLUSHING:
421 * The item or its backing buffer is already beeing
422 * flushed. The typical reason for that is that an
423 * inode buffer is locked because we already pushed the
424 * updates to it as part of inode clustering.
426 * We do not want to to stop flushing just because lots
427 * of items are already beeing flushed, but we need to
428 * re-try the flushing relatively soon if most of the
429 * AIL is beeing flushed.
431 XFS_STATS_INC(mp, xs_push_ail_flushing);
432 trace_xfs_ail_flushing(lip);
435 ailp->ail_last_pushed_lsn = lsn;
438 case XFS_ITEM_PINNED:
439 XFS_STATS_INC(mp, xs_push_ail_pinned);
440 trace_xfs_ail_pinned(lip);
443 ailp->ail_log_flush++;
445 case XFS_ITEM_LOCKED:
446 XFS_STATS_INC(mp, xs_push_ail_locked);
447 trace_xfs_ail_locked(lip);
459 * Are there too many items we can't do anything with?
461 * If we we are skipping too many items because we can't flush
462 * them or they are already being flushed, we back off and
463 * given them time to complete whatever operation is being
464 * done. i.e. remove pressure from the AIL while we can't make
465 * progress so traversals don't slow down further inserts and
466 * removals to/from the AIL.
468 * The value of 100 is an arbitrary magic number based on
474 lip = xfs_trans_ail_cursor_next(ailp, &cur);
479 xfs_trans_ail_cursor_done(&cur);
480 spin_unlock(&ailp->ail_lock);
482 if (xfs_buf_delwri_submit_nowait(&ailp->ail_buf_list))
483 ailp->ail_log_flush++;
485 if (!count || XFS_LSN_CMP(lsn, target) >= 0) {
488 * We reached the target or the AIL is empty, so wait a bit
489 * longer for I/O to complete and remove pushed items from the
490 * AIL before we start the next scan from the start of the AIL.
493 ailp->ail_last_pushed_lsn = 0;
494 } else if (((stuck + flushing) * 100) / count > 90) {
496 * Either there is a lot of contention on the AIL or we are
497 * stuck due to operations in progress. "Stuck" in this case
498 * is defined as >90% of the items we tried to push were stuck.
500 * Backoff a bit more to allow some I/O to complete before
501 * restarting from the start of the AIL. This prevents us from
502 * spinning on the same items, and if they are pinned will all
503 * the restart to issue a log force to unpin the stuck items.
506 ailp->ail_last_pushed_lsn = 0;
509 * Assume we have more work to do in a short while.
521 struct xfs_ail *ailp = data;
522 long tout = 0; /* milliseconds */
523 unsigned int noreclaim_flag;
525 noreclaim_flag = memalloc_noreclaim_save();
529 if (tout && tout <= 20)
530 set_current_state(TASK_KILLABLE);
532 set_current_state(TASK_INTERRUPTIBLE);
535 * Check kthread_should_stop() after we set the task state to
536 * guarantee that we either see the stop bit and exit or the
537 * task state is reset to runnable such that it's not scheduled
538 * out indefinitely and detects the stop bit at next iteration.
539 * A memory barrier is included in above task state set to
540 * serialize again kthread_stop().
542 if (kthread_should_stop()) {
543 __set_current_state(TASK_RUNNING);
546 * The caller forces out the AIL before stopping the
547 * thread in the common case, which means the delwri
548 * queue is drained. In the shutdown case, the queue may
549 * still hold relogged buffers that haven't been
550 * submitted because they were pinned since added to the
553 * Log I/O error processing stales the underlying buffer
554 * and clears the delwri state, expecting the buf to be
555 * removed on the next submission attempt. That won't
556 * happen if we're shutting down, so this is the last
557 * opportunity to release such buffers from the queue.
559 ASSERT(list_empty(&ailp->ail_buf_list) ||
560 XFS_FORCED_SHUTDOWN(ailp->ail_mount));
561 xfs_buf_delwri_cancel(&ailp->ail_buf_list);
565 spin_lock(&ailp->ail_lock);
568 * Idle if the AIL is empty and we are not racing with a target
569 * update. We check the AIL after we set the task to a sleep
570 * state to guarantee that we either catch an ail_target update
571 * or that a wake_up resets the state to TASK_RUNNING.
572 * Otherwise, we run the risk of sleeping indefinitely.
574 * The barrier matches the ail_target update in xfs_ail_push().
577 if (!xfs_ail_min(ailp) &&
578 ailp->ail_target == ailp->ail_target_prev) {
579 spin_unlock(&ailp->ail_lock);
580 freezable_schedule();
584 spin_unlock(&ailp->ail_lock);
587 freezable_schedule_timeout(msecs_to_jiffies(tout));
589 __set_current_state(TASK_RUNNING);
593 tout = xfsaild_push(ailp);
596 memalloc_noreclaim_restore(noreclaim_flag);
601 * This routine is called to move the tail of the AIL forward. It does this by
602 * trying to flush items in the AIL whose lsns are below the given
605 * The push is run asynchronously in a workqueue, which means the caller needs
606 * to handle waiting on the async flush for space to become available.
607 * We don't want to interrupt any push that is in progress, hence we only queue
608 * work if we set the pushing bit approriately.
610 * We do this unlocked - we only need to know whether there is anything in the
611 * AIL at the time we are called. We don't need to access the contents of
612 * any of the objects, so the lock is not needed.
616 struct xfs_ail *ailp,
617 xfs_lsn_t threshold_lsn)
621 lip = xfs_ail_min(ailp);
622 if (!lip || XFS_FORCED_SHUTDOWN(ailp->ail_mount) ||
623 XFS_LSN_CMP(threshold_lsn, ailp->ail_target) <= 0)
627 * Ensure that the new target is noticed in push code before it clears
628 * the XFS_AIL_PUSHING_BIT.
631 xfs_trans_ail_copy_lsn(ailp, &ailp->ail_target, &threshold_lsn);
634 wake_up_process(ailp->ail_task);
638 * Push out all items in the AIL immediately
642 struct xfs_ail *ailp)
644 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
647 xfs_ail_push(ailp, threshold_lsn);
651 * Push out all items in the AIL immediately and wait until the AIL is empty.
654 xfs_ail_push_all_sync(
655 struct xfs_ail *ailp)
657 struct xfs_log_item *lip;
660 spin_lock(&ailp->ail_lock);
661 while ((lip = xfs_ail_max(ailp)) != NULL) {
662 prepare_to_wait(&ailp->ail_empty, &wait, TASK_UNINTERRUPTIBLE);
663 ailp->ail_target = lip->li_lsn;
664 wake_up_process(ailp->ail_task);
665 spin_unlock(&ailp->ail_lock);
667 spin_lock(&ailp->ail_lock);
669 spin_unlock(&ailp->ail_lock);
671 finish_wait(&ailp->ail_empty, &wait);
675 * xfs_trans_ail_update - bulk AIL insertion operation.
677 * @xfs_trans_ail_update takes an array of log items that all need to be
678 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
679 * be added. Otherwise, it will be repositioned by removing it and re-adding
680 * it to the AIL. If we move the first item in the AIL, update the log tail to
681 * match the new minimum LSN in the AIL.
683 * This function takes the AIL lock once to execute the update operations on
684 * all the items in the array, and as such should not be called with the AIL
685 * lock held. As a result, once we have the AIL lock, we need to check each log
686 * item LSN to confirm it needs to be moved forward in the AIL.
688 * To optimise the insert operation, we delete all the items from the AIL in
689 * the first pass, moving them into a temporary list, then splice the temporary
690 * list into the correct position in the AIL. This avoids needing to do an
691 * insert operation on every item.
693 * This function must be called with the AIL lock held. The lock is dropped
697 xfs_trans_ail_update_bulk(
698 struct xfs_ail *ailp,
699 struct xfs_ail_cursor *cur,
700 struct xfs_log_item **log_items,
702 xfs_lsn_t lsn) __releases(ailp->ail_lock)
704 xfs_log_item_t *mlip;
705 int mlip_changed = 0;
709 ASSERT(nr_items > 0); /* Not required, but true. */
710 mlip = xfs_ail_min(ailp);
712 for (i = 0; i < nr_items; i++) {
713 struct xfs_log_item *lip = log_items[i];
714 if (test_and_set_bit(XFS_LI_IN_AIL, &lip->li_flags)) {
715 /* check if we really need to move the item */
716 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
719 trace_xfs_ail_move(lip, lip->li_lsn, lsn);
720 xfs_ail_delete(ailp, lip);
724 trace_xfs_ail_insert(lip, 0, lsn);
727 list_add(&lip->li_ail, &tmp);
730 if (!list_empty(&tmp))
731 xfs_ail_splice(ailp, cur, &tmp, lsn);
734 if (!XFS_FORCED_SHUTDOWN(ailp->ail_mount))
735 xlog_assign_tail_lsn_locked(ailp->ail_mount);
736 spin_unlock(&ailp->ail_lock);
738 xfs_log_space_wake(ailp->ail_mount);
740 spin_unlock(&ailp->ail_lock);
746 struct xfs_ail *ailp,
747 struct xfs_log_item *lip)
749 struct xfs_log_item *mlip = xfs_ail_min(ailp);
751 trace_xfs_ail_delete(lip, mlip->li_lsn, lip->li_lsn);
752 xfs_ail_delete(ailp, lip);
753 xfs_clear_li_failed(lip);
754 clear_bit(XFS_LI_IN_AIL, &lip->li_flags);
761 * Remove a log items from the AIL
763 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
764 * removed from the AIL. The caller is already holding the AIL lock, and done
765 * all the checks necessary to ensure the items passed in via @log_items are
766 * ready for deletion. This includes checking that the items are in the AIL.
768 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
769 * flag from the item and reset the item's lsn to 0. If we remove the first
770 * item in the AIL, update the log tail to match the new minimum LSN in the
773 * This function will not drop the AIL lock until all items are removed from
774 * the AIL to minimise the amount of lock traffic on the AIL. This does not
775 * greatly increase the AIL hold time, but does significantly reduce the amount
776 * of traffic on the lock, especially during IO completion.
778 * This function must be called with the AIL lock held. The lock is dropped
782 xfs_trans_ail_delete(
783 struct xfs_ail *ailp,
784 struct xfs_log_item *lip,
785 int shutdown_type) __releases(ailp->ail_lock)
787 struct xfs_mount *mp = ailp->ail_mount;
790 if (!test_bit(XFS_LI_IN_AIL, &lip->li_flags)) {
791 spin_unlock(&ailp->ail_lock);
792 if (!XFS_FORCED_SHUTDOWN(mp)) {
793 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
794 "%s: attempting to delete a log item that is not in the AIL",
796 xfs_force_shutdown(mp, shutdown_type);
801 mlip_changed = xfs_ail_delete_one(ailp, lip);
803 if (!XFS_FORCED_SHUTDOWN(mp))
804 xlog_assign_tail_lsn_locked(mp);
805 if (list_empty(&ailp->ail_head))
806 wake_up_all(&ailp->ail_empty);
809 spin_unlock(&ailp->ail_lock);
811 xfs_log_space_wake(ailp->ail_mount);
818 struct xfs_ail *ailp;
820 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
824 ailp->ail_mount = mp;
825 INIT_LIST_HEAD(&ailp->ail_head);
826 INIT_LIST_HEAD(&ailp->ail_cursors);
827 spin_lock_init(&ailp->ail_lock);
828 INIT_LIST_HEAD(&ailp->ail_buf_list);
829 init_waitqueue_head(&ailp->ail_empty);
831 ailp->ail_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
832 ailp->ail_mount->m_fsname);
833 if (IS_ERR(ailp->ail_task))
845 xfs_trans_ail_destroy(
848 struct xfs_ail *ailp = mp->m_ail;
850 kthread_stop(ailp->ail_task);