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 * AIL traversal cursor initialisation.
168 * The cursor keeps track of where our current traversal is up
169 * to by tracking the next ƣtem in the list for us. However, for
170 * this to be safe, removing an object from the AIL needs to invalidate
171 * any cursor that points to it. hence the traversal cursor needs to
172 * be linked to the struct xfs_ail so that deletion can search all the
173 * active cursors for invalidation.
175 * We don't link the push cursor because it is embedded in the struct
176 * xfs_ail and hence easily findable.
179 xfs_trans_ail_cursor_init(
180 struct xfs_ail *ailp,
181 struct xfs_ail_cursor *cur)
184 if (cur == &ailp->xa_cursors)
187 cur->next = ailp->xa_cursors.next;
188 ailp->xa_cursors.next = cur;
192 * Set the cursor to the next item, because when we look
193 * up the cursor the current item may have been freed.
196 xfs_trans_ail_cursor_set(
197 struct xfs_ail *ailp,
198 struct xfs_ail_cursor *cur,
199 struct xfs_log_item *lip)
202 cur->item = xfs_ail_next(ailp, lip);
206 * Get the next item in the traversal and advance the cursor.
207 * If the cursor was invalidated (inidicated by a lip of 1),
208 * restart the traversal.
210 struct xfs_log_item *
211 xfs_trans_ail_cursor_next(
212 struct xfs_ail *ailp,
213 struct xfs_ail_cursor *cur)
215 struct xfs_log_item *lip = cur->item;
217 if ((__psint_t)lip & 1)
218 lip = xfs_ail_min(ailp);
219 xfs_trans_ail_cursor_set(ailp, cur, lip);
224 * Now that the traversal is complete, we need to remove the cursor
225 * from the list of traversing cursors. Avoid removing the embedded
226 * push cursor, but use the fact it is always present to make the
227 * list deletion simple.
230 xfs_trans_ail_cursor_done(
231 struct xfs_ail *ailp,
232 struct xfs_ail_cursor *done)
234 struct xfs_ail_cursor *prev = NULL;
235 struct xfs_ail_cursor *cur;
238 if (done == &ailp->xa_cursors)
240 prev = &ailp->xa_cursors;
241 for (cur = prev->next; cur; prev = cur, cur = prev->next) {
243 prev->next = cur->next;
251 * Invalidate any cursor that is pointing to this item. This is
252 * called when an item is removed from the AIL. Any cursor pointing
253 * to this object is now invalid and the traversal needs to be
254 * terminated so it doesn't reference a freed object. We set the
255 * cursor item to a value of 1 so we can distinguish between an
256 * invalidation and the end of the list when getting the next item
260 xfs_trans_ail_cursor_clear(
261 struct xfs_ail *ailp,
262 struct xfs_log_item *lip)
264 struct xfs_ail_cursor *cur;
266 /* need to search all cursors */
267 for (cur = &ailp->xa_cursors; cur; cur = cur->next) {
268 if (cur->item == lip)
269 cur->item = (struct xfs_log_item *)
270 ((__psint_t)cur->item | 1);
275 * Initialise the cursor to the first item in the AIL with the given @lsn.
276 * This searches the list from lowest LSN to highest. Pass a @lsn of zero
277 * to initialise the cursor to the first item in the AIL.
280 xfs_trans_ail_cursor_first(
281 struct xfs_ail *ailp,
282 struct xfs_ail_cursor *cur,
287 xfs_trans_ail_cursor_init(ailp, cur);
288 lip = xfs_ail_min(ailp);
292 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
293 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
298 xfs_trans_ail_cursor_set(ailp, cur, lip);
303 * Initialise the cursor to the last item in the AIL with the given @lsn.
304 * This searches the list from highest LSN to lowest. If there is no item with
305 * the value of @lsn, then it sets the cursor to the last item with an LSN lower
308 static struct xfs_log_item *
309 __xfs_trans_ail_cursor_last(
310 struct xfs_ail *ailp,
315 list_for_each_entry_reverse(lip, &ailp->xa_ail, li_ail) {
316 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
323 * Initialise the cursor to the last item in the AIL with the given @lsn.
324 * This searches the list from highest LSN to lowest.
326 struct xfs_log_item *
327 xfs_trans_ail_cursor_last(
328 struct xfs_ail *ailp,
329 struct xfs_ail_cursor *cur,
332 xfs_trans_ail_cursor_init(ailp, cur);
333 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
338 * splice the log item list into the AIL at the given LSN. We splice to the
339 * tail of the given LSN to maintain insert order for push traversals. The
340 * cursor is optional, allowing repeated updates to the same LSN to avoid
341 * repeated traversals.
345 struct xfs_ail *ailp,
346 struct xfs_ail_cursor *cur,
347 struct list_head *list,
350 struct xfs_log_item *lip = cur ? cur->item : NULL;
351 struct xfs_log_item *next_lip;
354 * Get a new cursor if we don't have a placeholder or the existing one
355 * has been invalidated.
357 if (!lip || (__psint_t)lip & 1) {
358 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
361 /* The list is empty, so just splice and return. */
364 list_splice(list, &ailp->xa_ail);
370 * Our cursor points to the item we want to insert _after_, so we have
371 * to update the cursor to point to the end of the list we are splicing
372 * in so that it points to the correct location for the next splice.
373 * i.e. before the splice
375 * lsn -> lsn -> lsn + x -> lsn + x ...
377 * | cursor points here
379 * After the splice we have:
381 * lsn -> lsn -> lsn -> lsn -> .... -> lsn -> lsn + x -> lsn + x ...
383 * | cursor points here | needs to move here
385 * So we set the cursor to the last item in the list to be spliced
386 * before we execute the splice, resulting in the cursor pointing to
387 * the correct item after the splice occurs.
390 next_lip = list_entry(list->prev, struct xfs_log_item, li_ail);
391 cur->item = next_lip;
393 list_splice(list, &lip->li_ail);
397 * Delete the given item from the AIL. Return a pointer to the item.
401 struct xfs_ail *ailp,
404 xfs_ail_check(ailp, lip);
405 list_del(&lip->li_ail);
406 xfs_trans_ail_cursor_clear(ailp, lip);
410 * xfs_ail_worker does the work of pushing on the AIL. It will requeue itself
411 * to run at a later time if there is more work to do to complete the push.
415 struct work_struct *work)
417 struct xfs_ail *ailp = container_of(to_delayed_work(work),
418 struct xfs_ail, xa_work);
419 xfs_mount_t *mp = ailp->xa_mount;
420 struct xfs_ail_cursor *cur = &ailp->xa_cursors;
428 int push_xfsbufd = 0;
430 spin_lock(&ailp->xa_lock);
431 target = ailp->xa_target;
432 xfs_trans_ail_cursor_init(ailp, cur);
433 lip = xfs_trans_ail_cursor_first(ailp, cur, ailp->xa_last_pushed_lsn);
434 if (!lip || XFS_FORCED_SHUTDOWN(mp)) {
436 * AIL is empty or our push has reached the end.
438 xfs_trans_ail_cursor_done(ailp, cur);
439 spin_unlock(&ailp->xa_lock);
443 XFS_STATS_INC(xs_push_ail);
446 * While the item we are looking at is below the given threshold
447 * try to flush it out. We'd like not to stop until we've at least
448 * tried to push on everything in the AIL with an LSN less than
449 * the given threshold.
451 * However, we will stop after a certain number of pushes and wait
452 * for a reduced timeout to fire before pushing further. This
453 * prevents use from spinning when we can't do anything or there is
454 * lots of contention on the AIL lists.
457 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
460 * If we can lock the item without sleeping, unlock the AIL
461 * lock and flush the item. Then re-grab the AIL lock so we
462 * can look for the next item on the AIL. List changes are
463 * handled by the AIL lookup functions internally
465 * If we can't lock the item, either its holder will flush it
466 * or it is already being flushed or it is being relogged. In
467 * any of these case it is being taken care of and we can just
468 * skip to the next item in the list.
470 lock_result = IOP_TRYLOCK(lip);
471 spin_unlock(&ailp->xa_lock);
472 switch (lock_result) {
473 case XFS_ITEM_SUCCESS:
474 XFS_STATS_INC(xs_push_ail_success);
476 ailp->xa_last_pushed_lsn = lsn;
479 case XFS_ITEM_PUSHBUF:
480 XFS_STATS_INC(xs_push_ail_pushbuf);
482 ailp->xa_last_pushed_lsn = lsn;
486 case XFS_ITEM_PINNED:
487 XFS_STATS_INC(xs_push_ail_pinned);
492 case XFS_ITEM_LOCKED:
493 XFS_STATS_INC(xs_push_ail_locked);
502 spin_lock(&ailp->xa_lock);
503 /* should we bother continuing? */
504 if (XFS_FORCED_SHUTDOWN(mp))
511 * Are there too many items we can't do anything with?
512 * If we we are skipping too many items because we can't flush
513 * them or they are already being flushed, we back off and
514 * given them time to complete whatever operation is being
515 * done. i.e. remove pressure from the AIL while we can't make
516 * progress so traversals don't slow down further inserts and
517 * removals to/from the AIL.
519 * The value of 100 is an arbitrary magic number based on
525 lip = xfs_trans_ail_cursor_next(ailp, cur);
530 xfs_trans_ail_cursor_done(ailp, cur);
531 spin_unlock(&ailp->xa_lock);
535 * If something we need to push out was pinned, then
536 * push out the log so it will become unpinned and
537 * move forward in the AIL.
539 XFS_STATS_INC(xs_push_ail_flush);
540 xfs_log_force(mp, 0);
544 /* we've got delayed write buffers to flush */
545 wake_up_process(mp->m_ddev_targp->bt_task);
548 /* assume we have more work to do in a short while */
551 /* We're past our target or empty, so idle */
552 ailp->xa_last_pushed_lsn = 0;
555 * We clear the XFS_AIL_PUSHING_BIT first before checking
556 * whether the target has changed. If the target has changed,
557 * this pushes the requeue race directly onto the result of the
558 * atomic test/set bit, so we are guaranteed that either the
559 * the pusher that changed the target or ourselves will requeue
560 * the work (but not both).
562 clear_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags);
564 if (XFS_LSN_CMP(ailp->xa_target, target) == 0 ||
565 test_and_set_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags))
569 } else if (XFS_LSN_CMP(lsn, target) >= 0) {
571 * We reached the target so wait a bit longer for I/O to
572 * complete and remove pushed items from the AIL before we
573 * start the next scan from the start of the AIL.
576 ailp->xa_last_pushed_lsn = 0;
577 } else if ((stuck * 100) / count > 90) {
579 * Either there is a lot of contention on the AIL or we
580 * are stuck due to operations in progress. "Stuck" in this
581 * case is defined as >90% of the items we tried to push
584 * Backoff a bit more to allow some I/O to complete before
585 * continuing from where we were.
590 /* There is more to do, requeue us. */
591 queue_delayed_work(xfs_syncd_wq, &ailp->xa_work,
592 msecs_to_jiffies(tout));
596 * This routine is called to move the tail of the AIL forward. It does this by
597 * trying to flush items in the AIL whose lsns are below the given
600 * The push is run asynchronously in a workqueue, which means the caller needs
601 * to handle waiting on the async flush for space to become available.
602 * We don't want to interrupt any push that is in progress, hence we only queue
603 * work if we set the pushing bit approriately.
605 * We do this unlocked - we only need to know whether there is anything in the
606 * AIL at the time we are called. We don't need to access the contents of
607 * any of the objects, so the lock is not needed.
611 struct xfs_ail *ailp,
612 xfs_lsn_t threshold_lsn)
616 lip = xfs_ail_min(ailp);
617 if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) ||
618 XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0)
622 * Ensure that the new target is noticed in push code before it clears
623 * the XFS_AIL_PUSHING_BIT.
626 xfs_trans_ail_copy_lsn(ailp, &ailp->xa_target, &threshold_lsn);
627 if (!test_and_set_bit(XFS_AIL_PUSHING_BIT, &ailp->xa_flags))
628 queue_delayed_work(xfs_syncd_wq, &ailp->xa_work, 0);
632 * Push out all items in the AIL immediately
636 struct xfs_ail *ailp)
638 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
641 xfs_ail_push(ailp, threshold_lsn);
645 * This is to be called when an item is unlocked that may have
646 * been in the AIL. It will wake up the first member of the AIL
647 * wait list if this item's unlocking might allow it to progress.
648 * If the item is in the AIL, then we need to get the AIL lock
649 * while doing our checking so we don't race with someone going
650 * to sleep waiting for this event in xfs_trans_push_ail().
653 xfs_trans_unlocked_item(
654 struct xfs_ail *ailp,
657 xfs_log_item_t *min_lip;
660 * If we're forcibly shutting down, we may have
661 * unlocked log items arbitrarily. The last thing
662 * we want to do is to move the tail of the log
663 * over some potentially valid data.
665 if (!(lip->li_flags & XFS_LI_IN_AIL) ||
666 XFS_FORCED_SHUTDOWN(ailp->xa_mount)) {
671 * This is the one case where we can call into xfs_ail_min()
672 * without holding the AIL lock because we only care about the
673 * case where we are at the tail of the AIL. If the object isn't
674 * at the tail, it doesn't matter what result we get back. This
675 * is slightly racy because since we were just unlocked, we could
676 * go to sleep between the call to xfs_ail_min and the call to
677 * xfs_log_move_tail, have someone else lock us, commit to us disk,
678 * move us out of the tail of the AIL, and then we wake up. However,
679 * the call to xfs_log_move_tail() doesn't do anything if there's
680 * not enough free space to wake people up so we're safe calling it.
682 min_lip = xfs_ail_min(ailp);
685 xfs_log_move_tail(ailp->xa_mount, 1);
686 } /* xfs_trans_unlocked_item */
689 * xfs_trans_ail_update - bulk AIL insertion operation.
691 * @xfs_trans_ail_update takes an array of log items that all need to be
692 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
693 * be added. Otherwise, it will be repositioned by removing it and re-adding
694 * it to the AIL. If we move the first item in the AIL, update the log tail to
695 * match the new minimum LSN in the AIL.
697 * This function takes the AIL lock once to execute the update operations on
698 * all the items in the array, and as such should not be called with the AIL
699 * lock held. As a result, once we have the AIL lock, we need to check each log
700 * item LSN to confirm it needs to be moved forward in the AIL.
702 * To optimise the insert operation, we delete all the items from the AIL in
703 * the first pass, moving them into a temporary list, then splice the temporary
704 * list into the correct position in the AIL. This avoids needing to do an
705 * insert operation on every item.
707 * This function must be called with the AIL lock held. The lock is dropped
711 xfs_trans_ail_update_bulk(
712 struct xfs_ail *ailp,
713 struct xfs_ail_cursor *cur,
714 struct xfs_log_item **log_items,
716 xfs_lsn_t lsn) __releases(ailp->xa_lock)
718 xfs_log_item_t *mlip;
720 int mlip_changed = 0;
724 mlip = xfs_ail_min(ailp);
726 for (i = 0; i < nr_items; i++) {
727 struct xfs_log_item *lip = log_items[i];
728 if (lip->li_flags & XFS_LI_IN_AIL) {
729 /* check if we really need to move the item */
730 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
733 xfs_ail_delete(ailp, lip);
737 lip->li_flags |= XFS_LI_IN_AIL;
740 list_add(&lip->li_ail, &tmp);
743 xfs_ail_splice(ailp, cur, &tmp, lsn);
746 spin_unlock(&ailp->xa_lock);
751 * It is not safe to access mlip after the AIL lock is dropped, so we
752 * must get a copy of li_lsn before we do so. This is especially
753 * important on 32-bit platforms where accessing and updating 64-bit
754 * values like li_lsn is not atomic.
756 mlip = xfs_ail_min(ailp);
757 tail_lsn = mlip->li_lsn;
758 spin_unlock(&ailp->xa_lock);
759 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
763 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
765 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
766 * removed from the AIL. The caller is already holding the AIL lock, and done
767 * all the checks necessary to ensure the items passed in via @log_items are
768 * ready for deletion. This includes checking that the items are in the AIL.
770 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
771 * flag from the item and reset the item's lsn to 0. If we remove the first
772 * item in the AIL, update the log tail to match the new minimum LSN in the
775 * This function will not drop the AIL lock until all items are removed from
776 * the AIL to minimise the amount of lock traffic on the AIL. This does not
777 * greatly increase the AIL hold time, but does significantly reduce the amount
778 * of traffic on the lock, especially during IO completion.
780 * This function must be called with the AIL lock held. The lock is dropped
784 xfs_trans_ail_delete_bulk(
785 struct xfs_ail *ailp,
786 struct xfs_log_item **log_items,
787 int nr_items) __releases(ailp->xa_lock)
789 xfs_log_item_t *mlip;
791 int mlip_changed = 0;
794 mlip = xfs_ail_min(ailp);
796 for (i = 0; i < nr_items; i++) {
797 struct xfs_log_item *lip = log_items[i];
798 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
799 struct xfs_mount *mp = ailp->xa_mount;
801 spin_unlock(&ailp->xa_lock);
802 if (!XFS_FORCED_SHUTDOWN(mp)) {
803 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
804 "%s: attempting to delete a log item that is not in the AIL",
806 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
811 xfs_ail_delete(ailp, lip);
812 lip->li_flags &= ~XFS_LI_IN_AIL;
819 spin_unlock(&ailp->xa_lock);
824 * It is not safe to access mlip after the AIL lock is dropped, so we
825 * must get a copy of li_lsn before we do so. This is especially
826 * important on 32-bit platforms where accessing and updating 64-bit
827 * values like li_lsn is not atomic. It is possible we've emptied the
828 * AIL here, so if that is the case, pass an LSN of 0 to the tail move.
830 mlip = xfs_ail_min(ailp);
831 tail_lsn = mlip ? mlip->li_lsn : 0;
832 spin_unlock(&ailp->xa_lock);
833 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
837 * The active item list (AIL) is a doubly linked list of log
838 * items sorted by ascending lsn. The base of the list is
839 * a forw/back pointer pair embedded in the xfs mount structure.
840 * The base is initialized with both pointers pointing to the
841 * base. This case always needs to be distinguished, because
842 * the base has no lsn to look at. We almost always insert
843 * at the end of the list, so on inserts we search from the
844 * end of the list to find where the new item belongs.
848 * Initialize the doubly linked list to point only to itself.
854 struct xfs_ail *ailp;
856 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
861 INIT_LIST_HEAD(&ailp->xa_ail);
862 spin_lock_init(&ailp->xa_lock);
863 INIT_DELAYED_WORK(&ailp->xa_work, xfs_ail_worker);
869 xfs_trans_ail_destroy(
872 struct xfs_ail *ailp = mp->m_ail;
874 cancel_delayed_work_sync(&ailp->xa_work);