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"
31 struct workqueue_struct *xfs_ail_wq; /* AIL workqueue */
35 * Check that the list is sorted as it should be.
42 xfs_log_item_t *prev_lip;
44 if (list_empty(&ailp->xa_ail))
48 * Check the next and previous entries are valid.
50 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
51 prev_lip = list_entry(lip->li_ail.prev, xfs_log_item_t, li_ail);
52 if (&prev_lip->li_ail != &ailp->xa_ail)
53 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
55 prev_lip = list_entry(lip->li_ail.next, xfs_log_item_t, li_ail);
56 if (&prev_lip->li_ail != &ailp->xa_ail)
57 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) >= 0);
60 #ifdef XFS_TRANS_DEBUG
62 * Walk the list checking lsn ordering, and that every entry has the
63 * XFS_LI_IN_AIL flag set. This is really expensive, so only do it
64 * when specifically debugging the transaction subsystem.
66 prev_lip = list_entry(&ailp->xa_ail, xfs_log_item_t, li_ail);
67 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
68 if (&prev_lip->li_ail != &ailp->xa_ail)
69 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
70 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
73 #endif /* XFS_TRANS_DEBUG */
76 #define xfs_ail_check(a,l)
80 * Return a pointer to the first item in the AIL. If the AIL is empty, then
83 static xfs_log_item_t *
87 if (list_empty(&ailp->xa_ail))
90 return list_first_entry(&ailp->xa_ail, xfs_log_item_t, li_ail);
94 * Return a pointer to the last item in the AIL. If the AIL is empty, then
97 static xfs_log_item_t *
101 if (list_empty(&ailp->xa_ail))
104 return list_entry(ailp->xa_ail.prev, xfs_log_item_t, li_ail);
108 * Return a pointer to the item which follows the given item in the AIL. If
109 * the given item is the last item in the list, then return NULL.
111 static xfs_log_item_t *
113 struct xfs_ail *ailp,
116 if (lip->li_ail.next == &ailp->xa_ail)
119 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
123 * This is called by the log manager code to determine the LSN of the tail of
124 * the log. This is exactly the LSN of the first item in the AIL. If the AIL
125 * is empty, then this function returns 0.
127 * We need the AIL lock in order to get a coherent read of the lsn of the last
132 struct xfs_ail *ailp)
137 spin_lock(&ailp->xa_lock);
138 lip = xfs_ail_min(ailp);
141 spin_unlock(&ailp->xa_lock);
147 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
151 struct xfs_ail *ailp)
156 spin_lock(&ailp->xa_lock);
157 lip = xfs_ail_max(ailp);
160 spin_unlock(&ailp->xa_lock);
166 * The cursor keeps track of where our current traversal is up to by tracking
167 * the next item in the list for us. However, for this to be safe, removing an
168 * object from the AIL needs to invalidate any cursor that points to it. hence
169 * the traversal cursor needs to be linked to the struct xfs_ail so that
170 * deletion can search all the active cursors for invalidation.
173 xfs_trans_ail_cursor_init(
174 struct xfs_ail *ailp,
175 struct xfs_ail_cursor *cur)
178 list_add_tail(&cur->list, &ailp->xa_cursors);
182 * Get the next item in the traversal and advance the cursor. If the cursor
183 * was invalidated (indicated by a lip of 1), restart the traversal.
185 struct xfs_log_item *
186 xfs_trans_ail_cursor_next(
187 struct xfs_ail *ailp,
188 struct xfs_ail_cursor *cur)
190 struct xfs_log_item *lip = cur->item;
192 if ((__psint_t)lip & 1)
193 lip = xfs_ail_min(ailp);
195 cur->item = xfs_ail_next(ailp, lip);
200 * When the traversal is complete, we need to remove the cursor from the list
201 * of traversing cursors.
204 xfs_trans_ail_cursor_done(
205 struct xfs_ail *ailp,
206 struct xfs_ail_cursor *cur)
209 list_del_init(&cur->list);
213 * Invalidate any cursor that is pointing to this item. This is called when an
214 * item is removed from the AIL. Any cursor pointing to this object is now
215 * invalid and the traversal needs to be terminated so it doesn't reference a
216 * freed object. We set the low bit of the cursor item pointer so we can
217 * distinguish between an invalidation and the end of the list when getting the
218 * next item from the cursor.
221 xfs_trans_ail_cursor_clear(
222 struct xfs_ail *ailp,
223 struct xfs_log_item *lip)
225 struct xfs_ail_cursor *cur;
227 list_for_each_entry(cur, &ailp->xa_cursors, list) {
228 if (cur->item == lip)
229 cur->item = (struct xfs_log_item *)
230 ((__psint_t)cur->item | 1);
235 * Find the first item in the AIL with the given @lsn by searching in ascending
236 * LSN order and initialise the cursor to point to the next item for a
237 * ascending traversal. Pass a @lsn of zero to initialise the cursor to the
238 * first item in the AIL. Returns NULL if the list is empty.
241 xfs_trans_ail_cursor_first(
242 struct xfs_ail *ailp,
243 struct xfs_ail_cursor *cur,
248 xfs_trans_ail_cursor_init(ailp, cur);
251 lip = xfs_ail_min(ailp);
255 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
256 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
263 cur->item = xfs_ail_next(ailp, lip);
267 static struct xfs_log_item *
268 __xfs_trans_ail_cursor_last(
269 struct xfs_ail *ailp,
274 list_for_each_entry_reverse(lip, &ailp->xa_ail, li_ail) {
275 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
282 * Find the last item in the AIL with the given @lsn by searching in descending
283 * LSN order and initialise the cursor to point to that item. If there is no
284 * item with the value of @lsn, then it sets the cursor to the last item with an
285 * LSN lower than @lsn. Returns NULL if the list is empty.
287 struct xfs_log_item *
288 xfs_trans_ail_cursor_last(
289 struct xfs_ail *ailp,
290 struct xfs_ail_cursor *cur,
293 xfs_trans_ail_cursor_init(ailp, cur);
294 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
299 * Splice the log item list into the AIL at the given LSN. We splice to the
300 * tail of the given LSN to maintain insert order for push traversals. The
301 * cursor is optional, allowing repeated updates to the same LSN to avoid
302 * repeated traversals. This should not be called with an empty list.
306 struct xfs_ail *ailp,
307 struct xfs_ail_cursor *cur,
308 struct list_head *list,
311 struct xfs_log_item *lip;
313 ASSERT(!list_empty(list));
316 * Use the cursor to determine the insertion point if one is
317 * provided. If not, or if the one we got is not valid,
318 * find the place in the AIL where the items belong.
320 lip = cur ? cur->item : NULL;
321 if (!lip || (__psint_t) lip & 1)
322 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
325 * If a cursor is provided, we know we're processing the AIL
326 * in lsn order, and future items to be spliced in will
327 * follow the last one being inserted now. Update the
328 * cursor to point to that last item, now while we have a
329 * reliable pointer to it.
332 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
335 * Finally perform the splice. Unless the AIL was empty,
336 * lip points to the item in the AIL _after_ which the new
337 * items should go. If lip is null the AIL was empty, so
338 * the new items go at the head of the AIL.
341 list_splice(list, &lip->li_ail);
343 list_splice(list, &ailp->xa_ail);
347 * Delete the given item from the AIL. Return a pointer to the item.
351 struct xfs_ail *ailp,
354 xfs_ail_check(ailp, lip);
355 list_del(&lip->li_ail);
356 xfs_trans_ail_cursor_clear(ailp, lip);
360 * xfs_ail_worker does the work of pushing on the AIL. It will requeue itself
361 * to run at a later time if there is more work to do to complete the push.
365 struct work_struct *work)
367 struct xfs_ail *ailp = container_of(to_delayed_work(work),
368 struct xfs_ail, xa_work);
369 xfs_mount_t *mp = ailp->xa_mount;
370 struct xfs_ail_cursor cur;
377 int push_xfsbufd = 0;
380 * If last time we ran we encountered pinned items, force the log first
381 * and wait for it before pushing again.
383 spin_lock(&ailp->xa_lock);
384 if (ailp->xa_last_pushed_lsn == 0 && ailp->xa_log_flush &&
385 !list_empty(&ailp->xa_ail)) {
386 ailp->xa_log_flush = 0;
387 spin_unlock(&ailp->xa_lock);
388 XFS_STATS_INC(xs_push_ail_flush);
389 xfs_log_force(mp, XFS_LOG_SYNC);
390 spin_lock(&ailp->xa_lock);
393 target = ailp->xa_target;
394 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->xa_last_pushed_lsn);
395 if (!lip || XFS_FORCED_SHUTDOWN(mp)) {
397 * AIL is empty or our push has reached the end.
399 xfs_trans_ail_cursor_done(ailp, &cur);
400 spin_unlock(&ailp->xa_lock);
404 XFS_STATS_INC(xs_push_ail);
407 * While the item we are looking at is below the given threshold
408 * try to flush it out. We'd like not to stop until we've at least
409 * tried to push on everything in the AIL with an LSN less than
410 * the given threshold.
412 * However, we will stop after a certain number of pushes and wait
413 * for a reduced timeout to fire before pushing further. This
414 * prevents use from spinning when we can't do anything or there is
415 * lots of contention on the AIL lists.
418 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
421 * If we can lock the item without sleeping, unlock the AIL
422 * lock and flush the item. Then re-grab the AIL lock so we
423 * can look for the next item on the AIL. List changes are
424 * handled by the AIL lookup functions internally
426 * If we can't lock the item, either its holder will flush it
427 * or it is already being flushed or it is being relogged. In
428 * any of these case it is being taken care of and we can just
429 * skip to the next item in the list.
431 lock_result = IOP_TRYLOCK(lip);
432 spin_unlock(&ailp->xa_lock);
433 switch (lock_result) {
434 case XFS_ITEM_SUCCESS:
435 XFS_STATS_INC(xs_push_ail_success);
437 ailp->xa_last_pushed_lsn = lsn;
440 case XFS_ITEM_PUSHBUF:
441 XFS_STATS_INC(xs_push_ail_pushbuf);
443 ailp->xa_last_pushed_lsn = lsn;
447 case XFS_ITEM_PINNED:
448 XFS_STATS_INC(xs_push_ail_pinned);
450 ailp->xa_log_flush++;
453 case XFS_ITEM_LOCKED:
454 XFS_STATS_INC(xs_push_ail_locked);
455 ailp->xa_last_pushed_lsn = lsn;
464 spin_lock(&ailp->xa_lock);
465 /* should we bother continuing? */
466 if (XFS_FORCED_SHUTDOWN(mp))
473 * Are there too many items we can't do anything with?
474 * If we we are skipping too many items because we can't flush
475 * them or they are already being flushed, we back off and
476 * given them time to complete whatever operation is being
477 * done. i.e. remove pressure from the AIL while we can't make
478 * progress so traversals don't slow down further inserts and
479 * removals to/from the AIL.
481 * The value of 100 is an arbitrary magic number based on
487 lip = xfs_trans_ail_cursor_next(ailp, &cur);
492 xfs_trans_ail_cursor_done(ailp, &cur);
493 spin_unlock(&ailp->xa_lock);
496 /* we've got delayed write buffers to flush */
497 wake_up_process(mp->m_ddev_targp->bt_task);
500 /* assume we have more work to do in a short while */
503 /* We're past our target or empty, so idle */
504 ailp->xa_last_pushed_lsn = 0;
505 ailp->xa_log_flush = 0;
508 * We clear the XFS_AIL_PUSHING_BIT first before checking
509 * whether the target has changed. If the target has changed,
510 * this pushes the requeue race directly onto the result of the
511 * atomic test/set bit, so we are guaranteed that either the
512 * the pusher that changed the target or ourselves will requeue
513 * the work (but not both).
515 clear_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags);
517 if (XFS_LSN_CMP(ailp->xa_target, target) == 0 ||
518 test_and_set_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags))
522 } else if (XFS_LSN_CMP(lsn, target) >= 0) {
524 * We reached the target so wait a bit longer for I/O to
525 * complete and remove pushed items from the AIL before we
526 * start the next scan from the start of the AIL.
529 ailp->xa_last_pushed_lsn = 0;
530 } else if ((stuck * 100) / count > 90) {
532 * Either there is a lot of contention on the AIL or we
533 * are stuck due to operations in progress. "Stuck" in this
534 * case is defined as >90% of the items we tried to push
537 * Backoff a bit more to allow some I/O to complete before
538 * restarting from the start of the AIL. This prevents us
539 * from spinning on the same items, and if they are pinned will
540 * all the restart to issue a log force to unpin the stuck
544 ailp->xa_last_pushed_lsn = 0;
547 /* There is more to do, requeue us. */
548 queue_delayed_work(xfs_syncd_wq, &ailp->xa_work,
549 msecs_to_jiffies(tout));
553 * This routine is called to move the tail of the AIL forward. It does this by
554 * trying to flush items in the AIL whose lsns are below the given
557 * The push is run asynchronously in a workqueue, which means the caller needs
558 * to handle waiting on the async flush for space to become available.
559 * We don't want to interrupt any push that is in progress, hence we only queue
560 * work if we set the pushing bit approriately.
562 * We do this unlocked - we only need to know whether there is anything in the
563 * AIL at the time we are called. We don't need to access the contents of
564 * any of the objects, so the lock is not needed.
568 struct xfs_ail *ailp,
569 xfs_lsn_t threshold_lsn)
573 lip = xfs_ail_min(ailp);
574 if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) ||
575 XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0)
579 * Ensure that the new target is noticed in push code before it clears
580 * the XFS_AIL_PUSHING_BIT.
583 xfs_trans_ail_copy_lsn(ailp, &ailp->xa_target, &threshold_lsn);
584 if (!test_and_set_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags))
585 queue_delayed_work(xfs_syncd_wq, &ailp->xa_work, 0);
589 * Push out all items in the AIL immediately
593 struct xfs_ail *ailp)
595 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
598 xfs_ail_push(ailp, threshold_lsn);
602 * This is to be called when an item is unlocked that may have
603 * been in the AIL. It will wake up the first member of the AIL
604 * wait list if this item's unlocking might allow it to progress.
605 * If the item is in the AIL, then we need to get the AIL lock
606 * while doing our checking so we don't race with someone going
607 * to sleep waiting for this event in xfs_trans_push_ail().
610 xfs_trans_unlocked_item(
611 struct xfs_ail *ailp,
614 xfs_log_item_t *min_lip;
617 * If we're forcibly shutting down, we may have
618 * unlocked log items arbitrarily. The last thing
619 * we want to do is to move the tail of the log
620 * over some potentially valid data.
622 if (!(lip->li_flags & XFS_LI_IN_AIL) ||
623 XFS_FORCED_SHUTDOWN(ailp->xa_mount)) {
628 * This is the one case where we can call into xfs_ail_min()
629 * without holding the AIL lock because we only care about the
630 * case where we are at the tail of the AIL. If the object isn't
631 * at the tail, it doesn't matter what result we get back. This
632 * is slightly racy because since we were just unlocked, we could
633 * go to sleep between the call to xfs_ail_min and the call to
634 * xfs_log_move_tail, have someone else lock us, commit to us disk,
635 * move us out of the tail of the AIL, and then we wake up. However,
636 * the call to xfs_log_move_tail() doesn't do anything if there's
637 * not enough free space to wake people up so we're safe calling it.
639 min_lip = xfs_ail_min(ailp);
642 xfs_log_move_tail(ailp->xa_mount, 1);
643 } /* xfs_trans_unlocked_item */
646 * xfs_trans_ail_update - bulk AIL insertion operation.
648 * @xfs_trans_ail_update takes an array of log items that all need to be
649 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
650 * be added. Otherwise, it will be repositioned by removing it and re-adding
651 * it to the AIL. If we move the first item in the AIL, update the log tail to
652 * match the new minimum LSN in the AIL.
654 * This function takes the AIL lock once to execute the update operations on
655 * all the items in the array, and as such should not be called with the AIL
656 * lock held. As a result, once we have the AIL lock, we need to check each log
657 * item LSN to confirm it needs to be moved forward in the AIL.
659 * To optimise the insert operation, we delete all the items from the AIL in
660 * the first pass, moving them into a temporary list, then splice the temporary
661 * list into the correct position in the AIL. This avoids needing to do an
662 * insert operation on every item.
664 * This function must be called with the AIL lock held. The lock is dropped
668 xfs_trans_ail_update_bulk(
669 struct xfs_ail *ailp,
670 struct xfs_ail_cursor *cur,
671 struct xfs_log_item **log_items,
673 xfs_lsn_t lsn) __releases(ailp->xa_lock)
675 xfs_log_item_t *mlip;
677 int mlip_changed = 0;
681 ASSERT(nr_items > 0); /* Not required, but true. */
682 mlip = xfs_ail_min(ailp);
684 for (i = 0; i < nr_items; i++) {
685 struct xfs_log_item *lip = log_items[i];
686 if (lip->li_flags & XFS_LI_IN_AIL) {
687 /* check if we really need to move the item */
688 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
691 xfs_ail_delete(ailp, lip);
695 lip->li_flags |= XFS_LI_IN_AIL;
698 list_add(&lip->li_ail, &tmp);
701 if (!list_empty(&tmp))
702 xfs_ail_splice(ailp, cur, &tmp, lsn);
705 spin_unlock(&ailp->xa_lock);
710 * It is not safe to access mlip after the AIL lock is dropped, so we
711 * must get a copy of li_lsn before we do so. This is especially
712 * important on 32-bit platforms where accessing and updating 64-bit
713 * values like li_lsn is not atomic.
715 mlip = xfs_ail_min(ailp);
716 tail_lsn = mlip->li_lsn;
717 spin_unlock(&ailp->xa_lock);
718 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
722 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
724 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
725 * removed from the AIL. The caller is already holding the AIL lock, and done
726 * all the checks necessary to ensure the items passed in via @log_items are
727 * ready for deletion. This includes checking that the items are in the AIL.
729 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
730 * flag from the item and reset the item's lsn to 0. If we remove the first
731 * item in the AIL, update the log tail to match the new minimum LSN in the
734 * This function will not drop the AIL lock until all items are removed from
735 * the AIL to minimise the amount of lock traffic on the AIL. This does not
736 * greatly increase the AIL hold time, but does significantly reduce the amount
737 * of traffic on the lock, especially during IO completion.
739 * This function must be called with the AIL lock held. The lock is dropped
743 xfs_trans_ail_delete_bulk(
744 struct xfs_ail *ailp,
745 struct xfs_log_item **log_items,
746 int nr_items) __releases(ailp->xa_lock)
748 xfs_log_item_t *mlip;
750 int mlip_changed = 0;
753 mlip = xfs_ail_min(ailp);
755 for (i = 0; i < nr_items; i++) {
756 struct xfs_log_item *lip = log_items[i];
757 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
758 struct xfs_mount *mp = ailp->xa_mount;
760 spin_unlock(&ailp->xa_lock);
761 if (!XFS_FORCED_SHUTDOWN(mp)) {
762 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
763 "%s: attempting to delete a log item that is not in the AIL",
765 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
770 xfs_ail_delete(ailp, lip);
771 lip->li_flags &= ~XFS_LI_IN_AIL;
778 spin_unlock(&ailp->xa_lock);
783 * It is not safe to access mlip after the AIL lock is dropped, so we
784 * must get a copy of li_lsn before we do so. This is especially
785 * important on 32-bit platforms where accessing and updating 64-bit
786 * values like li_lsn is not atomic. It is possible we've emptied the
787 * AIL here, so if that is the case, pass an LSN of 0 to the tail move.
789 mlip = xfs_ail_min(ailp);
790 tail_lsn = mlip ? mlip->li_lsn : 0;
791 spin_unlock(&ailp->xa_lock);
792 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
796 * The active item list (AIL) is a doubly linked list of log
797 * items sorted by ascending lsn. The base of the list is
798 * a forw/back pointer pair embedded in the xfs mount structure.
799 * The base is initialized with both pointers pointing to the
800 * base. This case always needs to be distinguished, because
801 * the base has no lsn to look at. We almost always insert
802 * at the end of the list, so on inserts we search from the
803 * end of the list to find where the new item belongs.
807 * Initialize the doubly linked list to point only to itself.
813 struct xfs_ail *ailp;
815 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
820 INIT_LIST_HEAD(&ailp->xa_ail);
821 INIT_LIST_HEAD(&ailp->xa_cursors);
822 spin_lock_init(&ailp->xa_lock);
823 INIT_DELAYED_WORK(&ailp->xa_work, xfs_ail_worker);
829 xfs_trans_ail_destroy(
832 struct xfs_ail *ailp = mp->m_ail;
834 cancel_delayed_work_sync(&ailp->xa_work);