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_format.h"
22 #include "xfs_log_format.h"
23 #include "xfs_trans_resv.h"
24 #include "xfs_mount.h"
25 #include "xfs_trans.h"
26 #include "xfs_trans_priv.h"
27 #include "xfs_trace.h"
28 #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);
60 #define xfs_ail_check(a,l)
64 * Return a pointer to the last item in the AIL. If the AIL is empty, then
67 static xfs_log_item_t *
71 if (list_empty(&ailp->xa_ail))
74 return list_entry(ailp->xa_ail.prev, xfs_log_item_t, li_ail);
78 * Return a pointer to the item which follows the given item in the AIL. If
79 * the given item is the last item in the list, then return NULL.
81 static xfs_log_item_t *
86 if (lip->li_ail.next == &ailp->xa_ail)
89 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
93 * This is called by the log manager code to determine the LSN of the tail of
94 * the log. This is exactly the LSN of the first item in the AIL. If the AIL
95 * is empty, then this function returns 0.
97 * We need the AIL lock in order to get a coherent read of the lsn of the last
102 struct xfs_ail *ailp)
107 spin_lock(&ailp->xa_lock);
108 lip = xfs_ail_min(ailp);
111 spin_unlock(&ailp->xa_lock);
117 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
121 struct xfs_ail *ailp)
126 spin_lock(&ailp->xa_lock);
127 lip = xfs_ail_max(ailp);
130 spin_unlock(&ailp->xa_lock);
136 * The cursor keeps track of where our current traversal is up to by tracking
137 * the next item in the list for us. However, for this to be safe, removing an
138 * object from the AIL needs to invalidate any cursor that points to it. hence
139 * the traversal cursor needs to be linked to the struct xfs_ail so that
140 * deletion can search all the active cursors for invalidation.
143 xfs_trans_ail_cursor_init(
144 struct xfs_ail *ailp,
145 struct xfs_ail_cursor *cur)
148 list_add_tail(&cur->list, &ailp->xa_cursors);
152 * Get the next item in the traversal and advance the cursor. If the cursor
153 * was invalidated (indicated by a lip of 1), restart the traversal.
155 struct xfs_log_item *
156 xfs_trans_ail_cursor_next(
157 struct xfs_ail *ailp,
158 struct xfs_ail_cursor *cur)
160 struct xfs_log_item *lip = cur->item;
162 if ((__psint_t)lip & 1)
163 lip = xfs_ail_min(ailp);
165 cur->item = xfs_ail_next(ailp, lip);
170 * When the traversal is complete, we need to remove the cursor from the list
171 * of traversing cursors.
174 xfs_trans_ail_cursor_done(
175 struct xfs_ail_cursor *cur)
178 list_del_init(&cur->list);
182 * Invalidate any cursor that is pointing to this item. This is called when an
183 * item is removed from the AIL. Any cursor pointing to this object is now
184 * invalid and the traversal needs to be terminated so it doesn't reference a
185 * freed object. We set the low bit of the cursor item pointer so we can
186 * distinguish between an invalidation and the end of the list when getting the
187 * next item from the cursor.
190 xfs_trans_ail_cursor_clear(
191 struct xfs_ail *ailp,
192 struct xfs_log_item *lip)
194 struct xfs_ail_cursor *cur;
196 list_for_each_entry(cur, &ailp->xa_cursors, list) {
197 if (cur->item == lip)
198 cur->item = (struct xfs_log_item *)
199 ((__psint_t)cur->item | 1);
204 * Find the first item in the AIL with the given @lsn by searching in ascending
205 * LSN order and initialise the cursor to point to the next item for a
206 * ascending traversal. Pass a @lsn of zero to initialise the cursor to the
207 * first item in the AIL. Returns NULL if the list is empty.
210 xfs_trans_ail_cursor_first(
211 struct xfs_ail *ailp,
212 struct xfs_ail_cursor *cur,
217 xfs_trans_ail_cursor_init(ailp, cur);
220 lip = xfs_ail_min(ailp);
224 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
225 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
232 cur->item = xfs_ail_next(ailp, lip);
236 static struct xfs_log_item *
237 __xfs_trans_ail_cursor_last(
238 struct xfs_ail *ailp,
243 list_for_each_entry_reverse(lip, &ailp->xa_ail, li_ail) {
244 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
251 * Find the last item in the AIL with the given @lsn by searching in descending
252 * LSN order and initialise the cursor to point to that item. If there is no
253 * item with the value of @lsn, then it sets the cursor to the last item with an
254 * LSN lower than @lsn. Returns NULL if the list is empty.
256 struct xfs_log_item *
257 xfs_trans_ail_cursor_last(
258 struct xfs_ail *ailp,
259 struct xfs_ail_cursor *cur,
262 xfs_trans_ail_cursor_init(ailp, cur);
263 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
268 * Splice the log item list into the AIL at the given LSN. We splice to the
269 * tail of the given LSN to maintain insert order for push traversals. The
270 * cursor is optional, allowing repeated updates to the same LSN to avoid
271 * repeated traversals. This should not be called with an empty list.
275 struct xfs_ail *ailp,
276 struct xfs_ail_cursor *cur,
277 struct list_head *list,
280 struct xfs_log_item *lip;
282 ASSERT(!list_empty(list));
285 * Use the cursor to determine the insertion point if one is
286 * provided. If not, or if the one we got is not valid,
287 * find the place in the AIL where the items belong.
289 lip = cur ? cur->item : NULL;
290 if (!lip || (__psint_t) lip & 1)
291 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
294 * If a cursor is provided, we know we're processing the AIL
295 * in lsn order, and future items to be spliced in will
296 * follow the last one being inserted now. Update the
297 * cursor to point to that last item, now while we have a
298 * reliable pointer to it.
301 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
304 * Finally perform the splice. Unless the AIL was empty,
305 * lip points to the item in the AIL _after_ which the new
306 * items should go. If lip is null the AIL was empty, so
307 * the new items go at the head of the AIL.
310 list_splice(list, &lip->li_ail);
312 list_splice(list, &ailp->xa_ail);
316 * Delete the given item from the AIL. Return a pointer to the item.
320 struct xfs_ail *ailp,
323 xfs_ail_check(ailp, lip);
324 list_del(&lip->li_ail);
325 xfs_trans_ail_cursor_clear(ailp, lip);
330 struct xfs_ail *ailp)
332 xfs_mount_t *mp = ailp->xa_mount;
333 struct xfs_ail_cursor cur;
343 * If we encountered pinned items or did not finish writing out all
344 * buffers the last time we ran, force the log first and wait for it
345 * before pushing again.
347 if (ailp->xa_log_flush && ailp->xa_last_pushed_lsn == 0 &&
348 (!list_empty_careful(&ailp->xa_buf_list) ||
349 xfs_ail_min_lsn(ailp))) {
350 ailp->xa_log_flush = 0;
352 XFS_STATS_INC(xs_push_ail_flush);
353 xfs_log_force(mp, XFS_LOG_SYNC);
356 spin_lock(&ailp->xa_lock);
358 /* barrier matches the xa_target update in xfs_ail_push() */
360 target = ailp->xa_target;
361 ailp->xa_target_prev = target;
363 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->xa_last_pushed_lsn);
366 * If the AIL is empty or our push has reached the end we are
369 xfs_trans_ail_cursor_done(&cur);
370 spin_unlock(&ailp->xa_lock);
374 XFS_STATS_INC(xs_push_ail);
377 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
381 * Note that iop_push may unlock and reacquire the AIL lock. We
382 * rely on the AIL cursor implementation to be able to deal with
385 lock_result = lip->li_ops->iop_push(lip, &ailp->xa_buf_list);
386 switch (lock_result) {
387 case XFS_ITEM_SUCCESS:
388 XFS_STATS_INC(xs_push_ail_success);
389 trace_xfs_ail_push(lip);
391 ailp->xa_last_pushed_lsn = lsn;
394 case XFS_ITEM_FLUSHING:
396 * The item or its backing buffer is already beeing
397 * flushed. The typical reason for that is that an
398 * inode buffer is locked because we already pushed the
399 * updates to it as part of inode clustering.
401 * We do not want to to stop flushing just because lots
402 * of items are already beeing flushed, but we need to
403 * re-try the flushing relatively soon if most of the
404 * AIL is beeing flushed.
406 XFS_STATS_INC(xs_push_ail_flushing);
407 trace_xfs_ail_flushing(lip);
410 ailp->xa_last_pushed_lsn = lsn;
413 case XFS_ITEM_PINNED:
414 XFS_STATS_INC(xs_push_ail_pinned);
415 trace_xfs_ail_pinned(lip);
418 ailp->xa_log_flush++;
420 case XFS_ITEM_LOCKED:
421 XFS_STATS_INC(xs_push_ail_locked);
422 trace_xfs_ail_locked(lip);
434 * Are there too many items we can't do anything with?
436 * If we we are skipping too many items because we can't flush
437 * them or they are already being flushed, we back off and
438 * given them time to complete whatever operation is being
439 * done. i.e. remove pressure from the AIL while we can't make
440 * progress so traversals don't slow down further inserts and
441 * removals to/from the AIL.
443 * The value of 100 is an arbitrary magic number based on
449 lip = xfs_trans_ail_cursor_next(ailp, &cur);
454 xfs_trans_ail_cursor_done(&cur);
455 spin_unlock(&ailp->xa_lock);
457 if (xfs_buf_delwri_submit_nowait(&ailp->xa_buf_list))
458 ailp->xa_log_flush++;
460 if (!count || XFS_LSN_CMP(lsn, target) >= 0) {
463 * We reached the target or the AIL is empty, so wait a bit
464 * longer for I/O to complete and remove pushed items from the
465 * AIL before we start the next scan from the start of the AIL.
468 ailp->xa_last_pushed_lsn = 0;
469 } else if (((stuck + flushing) * 100) / count > 90) {
471 * Either there is a lot of contention on the AIL or we are
472 * stuck due to operations in progress. "Stuck" in this case
473 * is defined as >90% of the items we tried to push were stuck.
475 * Backoff a bit more to allow some I/O to complete before
476 * restarting from the start of the AIL. This prevents us from
477 * spinning on the same items, and if they are pinned will all
478 * the restart to issue a log force to unpin the stuck items.
481 ailp->xa_last_pushed_lsn = 0;
484 * Assume we have more work to do in a short while.
496 struct xfs_ail *ailp = data;
497 long tout = 0; /* milliseconds */
499 current->flags |= PF_MEMALLOC;
501 while (!kthread_should_stop()) {
502 if (tout && tout <= 20)
503 __set_current_state(TASK_KILLABLE);
505 __set_current_state(TASK_INTERRUPTIBLE);
507 spin_lock(&ailp->xa_lock);
510 * Idle if the AIL is empty and we are not racing with a target
511 * update. We check the AIL after we set the task to a sleep
512 * state to guarantee that we either catch an xa_target update
513 * or that a wake_up resets the state to TASK_RUNNING.
514 * Otherwise, we run the risk of sleeping indefinitely.
516 * The barrier matches the xa_target update in xfs_ail_push().
519 if (!xfs_ail_min(ailp) &&
520 ailp->xa_target == ailp->xa_target_prev) {
521 spin_unlock(&ailp->xa_lock);
526 spin_unlock(&ailp->xa_lock);
529 schedule_timeout(msecs_to_jiffies(tout));
531 __set_current_state(TASK_RUNNING);
535 tout = xfsaild_push(ailp);
542 * This routine is called to move the tail of the AIL forward. It does this by
543 * trying to flush items in the AIL whose lsns are below the given
546 * The push is run asynchronously in a workqueue, which means the caller needs
547 * to handle waiting on the async flush for space to become available.
548 * We don't want to interrupt any push that is in progress, hence we only queue
549 * work if we set the pushing bit approriately.
551 * We do this unlocked - we only need to know whether there is anything in the
552 * AIL at the time we are called. We don't need to access the contents of
553 * any of the objects, so the lock is not needed.
557 struct xfs_ail *ailp,
558 xfs_lsn_t threshold_lsn)
562 lip = xfs_ail_min(ailp);
563 if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) ||
564 XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0)
568 * Ensure that the new target is noticed in push code before it clears
569 * the XFS_AIL_PUSHING_BIT.
572 xfs_trans_ail_copy_lsn(ailp, &ailp->xa_target, &threshold_lsn);
575 wake_up_process(ailp->xa_task);
579 * Push out all items in the AIL immediately
583 struct xfs_ail *ailp)
585 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
588 xfs_ail_push(ailp, threshold_lsn);
592 * Push out all items in the AIL immediately and wait until the AIL is empty.
595 xfs_ail_push_all_sync(
596 struct xfs_ail *ailp)
598 struct xfs_log_item *lip;
601 spin_lock(&ailp->xa_lock);
602 while ((lip = xfs_ail_max(ailp)) != NULL) {
603 prepare_to_wait(&ailp->xa_empty, &wait, TASK_UNINTERRUPTIBLE);
604 ailp->xa_target = lip->li_lsn;
605 wake_up_process(ailp->xa_task);
606 spin_unlock(&ailp->xa_lock);
608 spin_lock(&ailp->xa_lock);
610 spin_unlock(&ailp->xa_lock);
612 finish_wait(&ailp->xa_empty, &wait);
616 * xfs_trans_ail_update - bulk AIL insertion operation.
618 * @xfs_trans_ail_update takes an array of log items that all need to be
619 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
620 * be added. Otherwise, it will be repositioned by removing it and re-adding
621 * it to the AIL. If we move the first item in the AIL, update the log tail to
622 * match the new minimum LSN in the AIL.
624 * This function takes the AIL lock once to execute the update operations on
625 * all the items in the array, and as such should not be called with the AIL
626 * lock held. As a result, once we have the AIL lock, we need to check each log
627 * item LSN to confirm it needs to be moved forward in the AIL.
629 * To optimise the insert operation, we delete all the items from the AIL in
630 * the first pass, moving them into a temporary list, then splice the temporary
631 * list into the correct position in the AIL. This avoids needing to do an
632 * insert operation on every item.
634 * This function must be called with the AIL lock held. The lock is dropped
638 xfs_trans_ail_update_bulk(
639 struct xfs_ail *ailp,
640 struct xfs_ail_cursor *cur,
641 struct xfs_log_item **log_items,
643 xfs_lsn_t lsn) __releases(ailp->xa_lock)
645 xfs_log_item_t *mlip;
646 int mlip_changed = 0;
650 ASSERT(nr_items > 0); /* Not required, but true. */
651 mlip = xfs_ail_min(ailp);
653 for (i = 0; i < nr_items; i++) {
654 struct xfs_log_item *lip = log_items[i];
655 if (lip->li_flags & XFS_LI_IN_AIL) {
656 /* check if we really need to move the item */
657 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
660 trace_xfs_ail_move(lip, lip->li_lsn, lsn);
661 xfs_ail_delete(ailp, lip);
665 lip->li_flags |= XFS_LI_IN_AIL;
666 trace_xfs_ail_insert(lip, 0, lsn);
669 list_add(&lip->li_ail, &tmp);
672 if (!list_empty(&tmp))
673 xfs_ail_splice(ailp, cur, &tmp, lsn);
676 if (!XFS_FORCED_SHUTDOWN(ailp->xa_mount))
677 xlog_assign_tail_lsn_locked(ailp->xa_mount);
678 spin_unlock(&ailp->xa_lock);
680 xfs_log_space_wake(ailp->xa_mount);
682 spin_unlock(&ailp->xa_lock);
687 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
689 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
690 * removed from the AIL. The caller is already holding the AIL lock, and done
691 * all the checks necessary to ensure the items passed in via @log_items are
692 * ready for deletion. This includes checking that the items are in the AIL.
694 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
695 * flag from the item and reset the item's lsn to 0. If we remove the first
696 * item in the AIL, update the log tail to match the new minimum LSN in the
699 * This function will not drop the AIL lock until all items are removed from
700 * the AIL to minimise the amount of lock traffic on the AIL. This does not
701 * greatly increase the AIL hold time, but does significantly reduce the amount
702 * of traffic on the lock, especially during IO completion.
704 * This function must be called with the AIL lock held. The lock is dropped
708 xfs_trans_ail_delete_bulk(
709 struct xfs_ail *ailp,
710 struct xfs_log_item **log_items,
712 int shutdown_type) __releases(ailp->xa_lock)
714 xfs_log_item_t *mlip;
715 int mlip_changed = 0;
718 mlip = xfs_ail_min(ailp);
720 for (i = 0; i < nr_items; i++) {
721 struct xfs_log_item *lip = log_items[i];
722 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
723 struct xfs_mount *mp = ailp->xa_mount;
725 spin_unlock(&ailp->xa_lock);
726 if (!XFS_FORCED_SHUTDOWN(mp)) {
727 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
728 "%s: attempting to delete a log item that is not in the AIL",
730 xfs_force_shutdown(mp, shutdown_type);
735 trace_xfs_ail_delete(lip, mlip->li_lsn, lip->li_lsn);
736 xfs_ail_delete(ailp, lip);
737 lip->li_flags &= ~XFS_LI_IN_AIL;
744 if (!XFS_FORCED_SHUTDOWN(ailp->xa_mount))
745 xlog_assign_tail_lsn_locked(ailp->xa_mount);
746 if (list_empty(&ailp->xa_ail))
747 wake_up_all(&ailp->xa_empty);
748 spin_unlock(&ailp->xa_lock);
750 xfs_log_space_wake(ailp->xa_mount);
752 spin_unlock(&ailp->xa_lock);
760 struct xfs_ail *ailp;
762 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
767 INIT_LIST_HEAD(&ailp->xa_ail);
768 INIT_LIST_HEAD(&ailp->xa_cursors);
769 spin_lock_init(&ailp->xa_lock);
770 INIT_LIST_HEAD(&ailp->xa_buf_list);
771 init_waitqueue_head(&ailp->xa_empty);
773 ailp->xa_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
774 ailp->xa_mount->m_fsname);
775 if (IS_ERR(ailp->xa_task))
787 xfs_trans_ail_destroy(
790 struct xfs_ail *ailp = mp->m_ail;
792 kthread_stop(ailp->xa_task);