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_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
81 static xfs_log_item_t *
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;
369 int push_xfsbufd = 0;
372 * If last time we ran we encountered pinned items, force the log first
373 * and wait for it before pushing again.
375 spin_lock(&ailp->xa_lock);
376 if (ailp->xa_last_pushed_lsn == 0 && ailp->xa_log_flush &&
377 !list_empty(&ailp->xa_ail)) {
378 ailp->xa_log_flush = 0;
379 spin_unlock(&ailp->xa_lock);
380 XFS_STATS_INC(xs_push_ail_flush);
381 xfs_log_force(mp, XFS_LOG_SYNC);
382 spin_lock(&ailp->xa_lock);
385 target = ailp->xa_target;
386 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->xa_last_pushed_lsn);
387 if (!lip || XFS_FORCED_SHUTDOWN(mp)) {
389 * AIL is empty or our push has reached the end.
391 xfs_trans_ail_cursor_done(ailp, &cur);
392 spin_unlock(&ailp->xa_lock);
396 XFS_STATS_INC(xs_push_ail);
399 * While the item we are looking at is below the given threshold
400 * try to flush it out. We'd like not to stop until we've at least
401 * tried to push on everything in the AIL with an LSN less than
402 * the given threshold.
404 * However, we will stop after a certain number of pushes and wait
405 * for a reduced timeout to fire before pushing further. This
406 * prevents use from spinning when we can't do anything or there is
407 * lots of contention on the AIL lists.
410 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
413 * If we can lock the item without sleeping, unlock the AIL
414 * lock and flush the item. Then re-grab the AIL lock so we
415 * can look for the next item on the AIL. List changes are
416 * handled by the AIL lookup functions internally
418 * If we can't lock the item, either its holder will flush it
419 * or it is already being flushed or it is being relogged. In
420 * any of these case it is being taken care of and we can just
421 * skip to the next item in the list.
423 lock_result = IOP_TRYLOCK(lip);
424 spin_unlock(&ailp->xa_lock);
425 switch (lock_result) {
426 case XFS_ITEM_SUCCESS:
427 XFS_STATS_INC(xs_push_ail_success);
429 ailp->xa_last_pushed_lsn = lsn;
432 case XFS_ITEM_PUSHBUF:
433 XFS_STATS_INC(xs_push_ail_pushbuf);
435 if (!IOP_PUSHBUF(lip)) {
439 ailp->xa_last_pushed_lsn = lsn;
444 case XFS_ITEM_PINNED:
445 XFS_STATS_INC(xs_push_ail_pinned);
447 ailp->xa_log_flush++;
450 case XFS_ITEM_LOCKED:
451 XFS_STATS_INC(xs_push_ail_locked);
460 spin_lock(&ailp->xa_lock);
461 /* should we bother continuing? */
462 if (XFS_FORCED_SHUTDOWN(mp))
469 * Are there too many items we can't do anything with?
470 * If we we are skipping too many items because we can't flush
471 * them or they are already being flushed, we back off and
472 * given them time to complete whatever operation is being
473 * done. i.e. remove pressure from the AIL while we can't make
474 * progress so traversals don't slow down further inserts and
475 * removals to/from the AIL.
477 * The value of 100 is an arbitrary magic number based on
483 lip = xfs_trans_ail_cursor_next(ailp, &cur);
488 xfs_trans_ail_cursor_done(ailp, &cur);
489 spin_unlock(&ailp->xa_lock);
492 /* we've got delayed write buffers to flush */
493 wake_up_process(mp->m_ddev_targp->bt_task);
496 /* assume we have more work to do in a short while */
499 /* We're past our target or empty, so idle */
500 ailp->xa_last_pushed_lsn = 0;
501 ailp->xa_log_flush = 0;
504 } else if (XFS_LSN_CMP(lsn, target) >= 0) {
506 * We reached the target so wait a bit longer for I/O to
507 * complete and remove pushed items from the AIL before we
508 * start the next scan from the start of the AIL.
511 ailp->xa_last_pushed_lsn = 0;
512 } else if ((stuck * 100) / count > 90) {
514 * Either there is a lot of contention on the AIL or we
515 * are stuck due to operations in progress. "Stuck" in this
516 * case is defined as >90% of the items we tried to push
519 * Backoff a bit more to allow some I/O to complete before
520 * restarting from the start of the AIL. This prevents us
521 * from spinning on the same items, and if they are pinned will
522 * all the restart to issue a log force to unpin the stuck
526 ailp->xa_last_pushed_lsn = 0;
536 struct xfs_ail *ailp = data;
537 long tout = 0; /* milliseconds */
539 while (!kthread_should_stop()) {
540 if (tout && tout <= 20)
541 __set_current_state(TASK_KILLABLE);
543 __set_current_state(TASK_INTERRUPTIBLE);
544 schedule_timeout(tout ?
545 msecs_to_jiffies(tout) : MAX_SCHEDULE_TIMEOUT);
549 tout = xfsaild_push(ailp);
556 * This routine is called to move the tail of the AIL forward. It does this by
557 * trying to flush items in the AIL whose lsns are below the given
560 * The push is run asynchronously in a workqueue, which means the caller needs
561 * to handle waiting on the async flush for space to become available.
562 * We don't want to interrupt any push that is in progress, hence we only queue
563 * work if we set the pushing bit approriately.
565 * We do this unlocked - we only need to know whether there is anything in the
566 * AIL at the time we are called. We don't need to access the contents of
567 * any of the objects, so the lock is not needed.
571 struct xfs_ail *ailp,
572 xfs_lsn_t threshold_lsn)
576 lip = xfs_ail_min(ailp);
577 if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) ||
578 XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0)
582 * Ensure that the new target is noticed in push code before it clears
583 * the XFS_AIL_PUSHING_BIT.
586 xfs_trans_ail_copy_lsn(ailp, &ailp->xa_target, &threshold_lsn);
589 wake_up_process(ailp->xa_task);
593 * Push out all items in the AIL immediately
597 struct xfs_ail *ailp)
599 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
602 xfs_ail_push(ailp, threshold_lsn);
606 * This is to be called when an item is unlocked that may have
607 * been in the AIL. It will wake up the first member of the AIL
608 * wait list if this item's unlocking might allow it to progress.
609 * If the item is in the AIL, then we need to get the AIL lock
610 * while doing our checking so we don't race with someone going
611 * to sleep waiting for this event in xfs_trans_push_ail().
614 xfs_trans_unlocked_item(
615 struct xfs_ail *ailp,
618 xfs_log_item_t *min_lip;
621 * If we're forcibly shutting down, we may have
622 * unlocked log items arbitrarily. The last thing
623 * we want to do is to move the tail of the log
624 * over some potentially valid data.
626 if (!(lip->li_flags & XFS_LI_IN_AIL) ||
627 XFS_FORCED_SHUTDOWN(ailp->xa_mount)) {
632 * This is the one case where we can call into xfs_ail_min()
633 * without holding the AIL lock because we only care about the
634 * case where we are at the tail of the AIL. If the object isn't
635 * at the tail, it doesn't matter what result we get back. This
636 * is slightly racy because since we were just unlocked, we could
637 * go to sleep between the call to xfs_ail_min and the call to
638 * xfs_log_move_tail, have someone else lock us, commit to us disk,
639 * move us out of the tail of the AIL, and then we wake up. However,
640 * the call to xfs_log_move_tail() doesn't do anything if there's
641 * not enough free space to wake people up so we're safe calling it.
643 min_lip = xfs_ail_min(ailp);
646 xfs_log_move_tail(ailp->xa_mount, 1);
647 } /* xfs_trans_unlocked_item */
650 * xfs_trans_ail_update - bulk AIL insertion operation.
652 * @xfs_trans_ail_update takes an array of log items that all need to be
653 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
654 * be added. Otherwise, it will be repositioned by removing it and re-adding
655 * it to the AIL. If we move the first item in the AIL, update the log tail to
656 * match the new minimum LSN in the AIL.
658 * This function takes the AIL lock once to execute the update operations on
659 * all the items in the array, and as such should not be called with the AIL
660 * lock held. As a result, once we have the AIL lock, we need to check each log
661 * item LSN to confirm it needs to be moved forward in the AIL.
663 * To optimise the insert operation, we delete all the items from the AIL in
664 * the first pass, moving them into a temporary list, then splice the temporary
665 * list into the correct position in the AIL. This avoids needing to do an
666 * insert operation on every item.
668 * This function must be called with the AIL lock held. The lock is dropped
672 xfs_trans_ail_update_bulk(
673 struct xfs_ail *ailp,
674 struct xfs_ail_cursor *cur,
675 struct xfs_log_item **log_items,
677 xfs_lsn_t lsn) __releases(ailp->xa_lock)
679 xfs_log_item_t *mlip;
681 int mlip_changed = 0;
685 ASSERT(nr_items > 0); /* Not required, but true. */
686 mlip = xfs_ail_min(ailp);
688 for (i = 0; i < nr_items; i++) {
689 struct xfs_log_item *lip = log_items[i];
690 if (lip->li_flags & XFS_LI_IN_AIL) {
691 /* check if we really need to move the item */
692 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
695 xfs_ail_delete(ailp, lip);
699 lip->li_flags |= XFS_LI_IN_AIL;
702 list_add(&lip->li_ail, &tmp);
705 if (!list_empty(&tmp))
706 xfs_ail_splice(ailp, cur, &tmp, lsn);
709 spin_unlock(&ailp->xa_lock);
714 * It is not safe to access mlip after the AIL lock is dropped, so we
715 * must get a copy of li_lsn before we do so. This is especially
716 * important on 32-bit platforms where accessing and updating 64-bit
717 * values like li_lsn is not atomic.
719 mlip = xfs_ail_min(ailp);
720 tail_lsn = mlip->li_lsn;
721 spin_unlock(&ailp->xa_lock);
722 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
726 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
728 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
729 * removed from the AIL. The caller is already holding the AIL lock, and done
730 * all the checks necessary to ensure the items passed in via @log_items are
731 * ready for deletion. This includes checking that the items are in the AIL.
733 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
734 * flag from the item and reset the item's lsn to 0. If we remove the first
735 * item in the AIL, update the log tail to match the new minimum LSN in the
738 * This function will not drop the AIL lock until all items are removed from
739 * the AIL to minimise the amount of lock traffic on the AIL. This does not
740 * greatly increase the AIL hold time, but does significantly reduce the amount
741 * of traffic on the lock, especially during IO completion.
743 * This function must be called with the AIL lock held. The lock is dropped
747 xfs_trans_ail_delete_bulk(
748 struct xfs_ail *ailp,
749 struct xfs_log_item **log_items,
750 int nr_items) __releases(ailp->xa_lock)
752 xfs_log_item_t *mlip;
754 int mlip_changed = 0;
757 mlip = xfs_ail_min(ailp);
759 for (i = 0; i < nr_items; i++) {
760 struct xfs_log_item *lip = log_items[i];
761 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
762 struct xfs_mount *mp = ailp->xa_mount;
764 spin_unlock(&ailp->xa_lock);
765 if (!XFS_FORCED_SHUTDOWN(mp)) {
766 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
767 "%s: attempting to delete a log item that is not in the AIL",
769 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
774 xfs_ail_delete(ailp, lip);
775 lip->li_flags &= ~XFS_LI_IN_AIL;
782 spin_unlock(&ailp->xa_lock);
787 * It is not safe to access mlip after the AIL lock is dropped, so we
788 * must get a copy of li_lsn before we do so. This is especially
789 * important on 32-bit platforms where accessing and updating 64-bit
790 * values like li_lsn is not atomic. It is possible we've emptied the
791 * AIL here, so if that is the case, pass an LSN of 0 to the tail move.
793 mlip = xfs_ail_min(ailp);
794 tail_lsn = mlip ? mlip->li_lsn : 0;
795 spin_unlock(&ailp->xa_lock);
796 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
800 * The active item list (AIL) is a doubly linked list of log
801 * items sorted by ascending lsn. The base of the list is
802 * a forw/back pointer pair embedded in the xfs mount structure.
803 * The base is initialized with both pointers pointing to the
804 * base. This case always needs to be distinguished, because
805 * the base has no lsn to look at. We almost always insert
806 * at the end of the list, so on inserts we search from the
807 * end of the list to find where the new item belongs.
811 * Initialize the doubly linked list to point only to itself.
817 struct xfs_ail *ailp;
819 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
824 INIT_LIST_HEAD(&ailp->xa_ail);
825 INIT_LIST_HEAD(&ailp->xa_cursors);
826 spin_lock_init(&ailp->xa_lock);
828 ailp->xa_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
829 ailp->xa_mount->m_fsname);
830 if (IS_ERR(ailp->xa_task))
842 xfs_trans_ail_destroy(
845 struct xfs_ail *ailp = mp->m_ail;
847 kthread_stop(ailp->xa_task);