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"
24 #include "xfs_trans.h"
27 #include "xfs_mount.h"
28 #include "xfs_trans_priv.h"
29 #include "xfs_trace.h"
30 #include "xfs_error.h"
34 * Check that the list is sorted as it should be.
41 xfs_log_item_t *prev_lip;
43 if (list_empty(&ailp->xa_ail))
47 * Check the next and previous entries are valid.
49 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
50 prev_lip = list_entry(lip->li_ail.prev, xfs_log_item_t, li_ail);
51 if (&prev_lip->li_ail != &ailp->xa_ail)
52 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
54 prev_lip = list_entry(lip->li_ail.next, xfs_log_item_t, li_ail);
55 if (&prev_lip->li_ail != &ailp->xa_ail)
56 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) >= 0);
59 #ifdef XFS_TRANS_DEBUG
61 * Walk the list checking lsn ordering, and that every entry has the
62 * XFS_LI_IN_AIL flag set. This is really expensive, so only do it
63 * when specifically debugging the transaction subsystem.
65 prev_lip = list_entry(&ailp->xa_ail, xfs_log_item_t, li_ail);
66 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
67 if (&prev_lip->li_ail != &ailp->xa_ail)
68 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
69 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
72 #endif /* XFS_TRANS_DEBUG */
75 #define xfs_ail_check(a,l)
79 * Return a pointer to the first item in the AIL. If the AIL is empty, then
86 if (list_empty(&ailp->xa_ail))
89 return list_first_entry(&ailp->xa_ail, xfs_log_item_t, li_ail);
93 * Return a pointer to the last item in the AIL. If the AIL is empty, then
96 static xfs_log_item_t *
100 if (list_empty(&ailp->xa_ail))
103 return list_entry(ailp->xa_ail.prev, xfs_log_item_t, li_ail);
107 * Return a pointer to the item which follows the given item in the AIL. If
108 * the given item is the last item in the list, then return NULL.
110 static xfs_log_item_t *
112 struct xfs_ail *ailp,
115 if (lip->li_ail.next == &ailp->xa_ail)
118 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
122 * This is called by the log manager code to determine the LSN of the tail of
123 * the log. This is exactly the LSN of the first item in the AIL. If the AIL
124 * is empty, then this function returns 0.
126 * We need the AIL lock in order to get a coherent read of the lsn of the last
131 struct xfs_ail *ailp)
136 spin_lock(&ailp->xa_lock);
137 lip = xfs_ail_min(ailp);
140 spin_unlock(&ailp->xa_lock);
146 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
150 struct xfs_ail *ailp)
155 spin_lock(&ailp->xa_lock);
156 lip = xfs_ail_max(ailp);
159 spin_unlock(&ailp->xa_lock);
165 * The cursor keeps track of where our current traversal is up to by tracking
166 * the next item in the list for us. However, for this to be safe, removing an
167 * object from the AIL needs to invalidate any cursor that points to it. hence
168 * the traversal cursor needs to be linked to the struct xfs_ail so that
169 * deletion can search all the active cursors for invalidation.
172 xfs_trans_ail_cursor_init(
173 struct xfs_ail *ailp,
174 struct xfs_ail_cursor *cur)
177 list_add_tail(&cur->list, &ailp->xa_cursors);
181 * Get the next item in the traversal and advance the cursor. If the cursor
182 * was invalidated (indicated by a lip of 1), restart the traversal.
184 struct xfs_log_item *
185 xfs_trans_ail_cursor_next(
186 struct xfs_ail *ailp,
187 struct xfs_ail_cursor *cur)
189 struct xfs_log_item *lip = cur->item;
191 if ((__psint_t)lip & 1)
192 lip = xfs_ail_min(ailp);
194 cur->item = xfs_ail_next(ailp, lip);
199 * When the traversal is complete, we need to remove the cursor from the list
200 * of traversing cursors.
203 xfs_trans_ail_cursor_done(
204 struct xfs_ail *ailp,
205 struct xfs_ail_cursor *cur)
208 list_del_init(&cur->list);
212 * Invalidate any cursor that is pointing to this item. This is called when an
213 * item is removed from the AIL. Any cursor pointing to this object is now
214 * invalid and the traversal needs to be terminated so it doesn't reference a
215 * freed object. We set the low bit of the cursor item pointer so we can
216 * distinguish between an invalidation and the end of the list when getting the
217 * next item from the cursor.
220 xfs_trans_ail_cursor_clear(
221 struct xfs_ail *ailp,
222 struct xfs_log_item *lip)
224 struct xfs_ail_cursor *cur;
226 list_for_each_entry(cur, &ailp->xa_cursors, list) {
227 if (cur->item == lip)
228 cur->item = (struct xfs_log_item *)
229 ((__psint_t)cur->item | 1);
234 * Find the first item in the AIL with the given @lsn by searching in ascending
235 * LSN order and initialise the cursor to point to the next item for a
236 * ascending traversal. Pass a @lsn of zero to initialise the cursor to the
237 * first item in the AIL. Returns NULL if the list is empty.
240 xfs_trans_ail_cursor_first(
241 struct xfs_ail *ailp,
242 struct xfs_ail_cursor *cur,
247 xfs_trans_ail_cursor_init(ailp, cur);
250 lip = xfs_ail_min(ailp);
254 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
255 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
262 cur->item = xfs_ail_next(ailp, lip);
266 static struct xfs_log_item *
267 __xfs_trans_ail_cursor_last(
268 struct xfs_ail *ailp,
273 list_for_each_entry_reverse(lip, &ailp->xa_ail, li_ail) {
274 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
281 * Find the last item in the AIL with the given @lsn by searching in descending
282 * LSN order and initialise the cursor to point to that item. If there is no
283 * item with the value of @lsn, then it sets the cursor to the last item with an
284 * LSN lower than @lsn. Returns NULL if the list is empty.
286 struct xfs_log_item *
287 xfs_trans_ail_cursor_last(
288 struct xfs_ail *ailp,
289 struct xfs_ail_cursor *cur,
292 xfs_trans_ail_cursor_init(ailp, cur);
293 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
298 * Splice the log item list into the AIL at the given LSN. We splice to the
299 * tail of the given LSN to maintain insert order for push traversals. The
300 * cursor is optional, allowing repeated updates to the same LSN to avoid
301 * repeated traversals. This should not be called with an empty list.
305 struct xfs_ail *ailp,
306 struct xfs_ail_cursor *cur,
307 struct list_head *list,
310 struct xfs_log_item *lip;
312 ASSERT(!list_empty(list));
315 * Use the cursor to determine the insertion point if one is
316 * provided. If not, or if the one we got is not valid,
317 * find the place in the AIL where the items belong.
319 lip = cur ? cur->item : NULL;
320 if (!lip || (__psint_t) lip & 1)
321 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
324 * If a cursor is provided, we know we're processing the AIL
325 * in lsn order, and future items to be spliced in will
326 * follow the last one being inserted now. Update the
327 * cursor to point to that last item, now while we have a
328 * reliable pointer to it.
331 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
334 * Finally perform the splice. Unless the AIL was empty,
335 * lip points to the item in the AIL _after_ which the new
336 * items should go. If lip is null the AIL was empty, so
337 * the new items go at the head of the AIL.
340 list_splice(list, &lip->li_ail);
342 list_splice(list, &ailp->xa_ail);
346 * Delete the given item from the AIL. Return a pointer to the item.
350 struct xfs_ail *ailp,
353 xfs_ail_check(ailp, lip);
354 list_del(&lip->li_ail);
355 xfs_trans_ail_cursor_clear(ailp, lip);
360 struct xfs_ail *ailp)
362 xfs_mount_t *mp = ailp->xa_mount;
363 struct xfs_ail_cursor cur;
373 * If we encountered pinned items or did not finish writing out all
374 * buffers the last time we ran, force the log first and wait for it
375 * before pushing again.
377 if (ailp->xa_log_flush && ailp->xa_last_pushed_lsn == 0 &&
378 (!list_empty_careful(&ailp->xa_buf_list) ||
379 xfs_ail_min_lsn(ailp))) {
380 ailp->xa_log_flush = 0;
382 XFS_STATS_INC(xs_push_ail_flush);
383 xfs_log_force(mp, XFS_LOG_SYNC);
386 spin_lock(&ailp->xa_lock);
387 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->xa_last_pushed_lsn);
390 * If the AIL is empty or our push has reached the end we are
393 xfs_trans_ail_cursor_done(ailp, &cur);
394 spin_unlock(&ailp->xa_lock);
398 XFS_STATS_INC(xs_push_ail);
401 target = ailp->xa_target;
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 = IOP_PUSH(lip, &ailp->xa_buf_list);
411 switch (lock_result) {
412 case XFS_ITEM_SUCCESS:
413 XFS_STATS_INC(xs_push_ail_success);
414 trace_xfs_ail_push(lip);
416 ailp->xa_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(xs_push_ail_flushing);
432 trace_xfs_ail_flushing(lip);
435 ailp->xa_last_pushed_lsn = lsn;
438 case XFS_ITEM_PINNED:
439 XFS_STATS_INC(xs_push_ail_pinned);
440 trace_xfs_ail_pinned(lip);
443 ailp->xa_log_flush++;
445 case XFS_ITEM_LOCKED:
446 XFS_STATS_INC(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(ailp, &cur);
480 spin_unlock(&ailp->xa_lock);
482 if (xfs_buf_delwri_submit_nowait(&ailp->xa_buf_list))
483 ailp->xa_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->xa_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->xa_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 */
524 current->flags |= PF_MEMALLOC;
526 while (!kthread_should_stop()) {
527 if (tout && tout <= 20)
528 __set_current_state(TASK_KILLABLE);
530 __set_current_state(TASK_INTERRUPTIBLE);
531 schedule_timeout(tout ?
532 msecs_to_jiffies(tout) : MAX_SCHEDULE_TIMEOUT);
536 tout = xfsaild_push(ailp);
543 * This routine is called to move the tail of the AIL forward. It does this by
544 * trying to flush items in the AIL whose lsns are below the given
547 * The push is run asynchronously in a workqueue, which means the caller needs
548 * to handle waiting on the async flush for space to become available.
549 * We don't want to interrupt any push that is in progress, hence we only queue
550 * work if we set the pushing bit approriately.
552 * We do this unlocked - we only need to know whether there is anything in the
553 * AIL at the time we are called. We don't need to access the contents of
554 * any of the objects, so the lock is not needed.
558 struct xfs_ail *ailp,
559 xfs_lsn_t threshold_lsn)
563 lip = xfs_ail_min(ailp);
564 if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) ||
565 XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0)
569 * Ensure that the new target is noticed in push code before it clears
570 * the XFS_AIL_PUSHING_BIT.
573 xfs_trans_ail_copy_lsn(ailp, &ailp->xa_target, &threshold_lsn);
576 wake_up_process(ailp->xa_task);
580 * Push out all items in the AIL immediately
584 struct xfs_ail *ailp)
586 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
589 xfs_ail_push(ailp, threshold_lsn);
593 * Push out all items in the AIL immediately and wait until the AIL is empty.
596 xfs_ail_push_all_sync(
597 struct xfs_ail *ailp)
599 struct xfs_log_item *lip;
602 spin_lock(&ailp->xa_lock);
603 while ((lip = xfs_ail_max(ailp)) != NULL) {
604 prepare_to_wait(&ailp->xa_empty, &wait, TASK_UNINTERRUPTIBLE);
605 ailp->xa_target = lip->li_lsn;
606 wake_up_process(ailp->xa_task);
607 spin_unlock(&ailp->xa_lock);
609 spin_lock(&ailp->xa_lock);
611 spin_unlock(&ailp->xa_lock);
613 finish_wait(&ailp->xa_empty, &wait);
617 * xfs_trans_ail_update - bulk AIL insertion operation.
619 * @xfs_trans_ail_update takes an array of log items that all need to be
620 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
621 * be added. Otherwise, it will be repositioned by removing it and re-adding
622 * it to the AIL. If we move the first item in the AIL, update the log tail to
623 * match the new minimum LSN in the AIL.
625 * This function takes the AIL lock once to execute the update operations on
626 * all the items in the array, and as such should not be called with the AIL
627 * lock held. As a result, once we have the AIL lock, we need to check each log
628 * item LSN to confirm it needs to be moved forward in the AIL.
630 * To optimise the insert operation, we delete all the items from the AIL in
631 * the first pass, moving them into a temporary list, then splice the temporary
632 * list into the correct position in the AIL. This avoids needing to do an
633 * insert operation on every item.
635 * This function must be called with the AIL lock held. The lock is dropped
639 xfs_trans_ail_update_bulk(
640 struct xfs_ail *ailp,
641 struct xfs_ail_cursor *cur,
642 struct xfs_log_item **log_items,
644 xfs_lsn_t lsn) __releases(ailp->xa_lock)
646 xfs_log_item_t *mlip;
647 int mlip_changed = 0;
651 ASSERT(nr_items > 0); /* Not required, but true. */
652 mlip = xfs_ail_min(ailp);
654 for (i = 0; i < nr_items; i++) {
655 struct xfs_log_item *lip = log_items[i];
656 if (lip->li_flags & XFS_LI_IN_AIL) {
657 /* check if we really need to move the item */
658 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
661 xfs_ail_delete(ailp, lip);
665 lip->li_flags |= XFS_LI_IN_AIL;
668 list_add(&lip->li_ail, &tmp);
671 if (!list_empty(&tmp))
672 xfs_ail_splice(ailp, cur, &tmp, lsn);
675 if (!XFS_FORCED_SHUTDOWN(ailp->xa_mount))
676 xlog_assign_tail_lsn_locked(ailp->xa_mount);
677 spin_unlock(&ailp->xa_lock);
679 xfs_log_space_wake(ailp->xa_mount);
681 spin_unlock(&ailp->xa_lock);
686 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
688 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
689 * removed from the AIL. The caller is already holding the AIL lock, and done
690 * all the checks necessary to ensure the items passed in via @log_items are
691 * ready for deletion. This includes checking that the items are in the AIL.
693 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
694 * flag from the item and reset the item's lsn to 0. If we remove the first
695 * item in the AIL, update the log tail to match the new minimum LSN in the
698 * This function will not drop the AIL lock until all items are removed from
699 * the AIL to minimise the amount of lock traffic on the AIL. This does not
700 * greatly increase the AIL hold time, but does significantly reduce the amount
701 * of traffic on the lock, especially during IO completion.
703 * This function must be called with the AIL lock held. The lock is dropped
707 xfs_trans_ail_delete_bulk(
708 struct xfs_ail *ailp,
709 struct xfs_log_item **log_items,
710 int nr_items) __releases(ailp->xa_lock)
712 xfs_log_item_t *mlip;
713 int mlip_changed = 0;
716 mlip = xfs_ail_min(ailp);
718 for (i = 0; i < nr_items; i++) {
719 struct xfs_log_item *lip = log_items[i];
720 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
721 struct xfs_mount *mp = ailp->xa_mount;
723 spin_unlock(&ailp->xa_lock);
724 if (!XFS_FORCED_SHUTDOWN(mp)) {
725 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
726 "%s: attempting to delete a log item that is not in the AIL",
728 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
733 xfs_ail_delete(ailp, lip);
734 lip->li_flags &= ~XFS_LI_IN_AIL;
741 if (!XFS_FORCED_SHUTDOWN(ailp->xa_mount))
742 xlog_assign_tail_lsn_locked(ailp->xa_mount);
743 if (list_empty(&ailp->xa_ail))
744 wake_up_all(&ailp->xa_empty);
745 spin_unlock(&ailp->xa_lock);
747 xfs_log_space_wake(ailp->xa_mount);
749 spin_unlock(&ailp->xa_lock);
754 * The active item list (AIL) is a doubly linked list of log
755 * items sorted by ascending lsn. The base of the list is
756 * a forw/back pointer pair embedded in the xfs mount structure.
757 * The base is initialized with both pointers pointing to the
758 * base. This case always needs to be distinguished, because
759 * the base has no lsn to look at. We almost always insert
760 * at the end of the list, so on inserts we search from the
761 * end of the list to find where the new item belongs.
765 * Initialize the doubly linked list to point only to itself.
771 struct xfs_ail *ailp;
773 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
778 INIT_LIST_HEAD(&ailp->xa_ail);
779 INIT_LIST_HEAD(&ailp->xa_cursors);
780 spin_lock_init(&ailp->xa_lock);
781 INIT_LIST_HEAD(&ailp->xa_buf_list);
782 init_waitqueue_head(&ailp->xa_empty);
784 ailp->xa_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
785 ailp->xa_mount->m_fsname);
786 if (IS_ERR(ailp->xa_task))
798 xfs_trans_ail_destroy(
801 struct xfs_ail *ailp = mp->m_ail;
803 kthread_stop(ailp->xa_task);