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_log_format.h"
22 #include "xfs_trans_resv.h"
25 #include "xfs_mount.h"
26 #include "xfs_trans.h"
27 #include "xfs_trans_priv.h"
28 #include "xfs_trace.h"
29 #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);
61 #define xfs_ail_check(a,l)
65 * Return a pointer to the last item in the AIL. If the AIL is empty, then
68 static xfs_log_item_t *
72 if (list_empty(&ailp->xa_ail))
75 return list_entry(ailp->xa_ail.prev, xfs_log_item_t, li_ail);
79 * Return a pointer to the item which follows the given item in the AIL. If
80 * the given item is the last item in the list, then return NULL.
82 static xfs_log_item_t *
87 if (lip->li_ail.next == &ailp->xa_ail)
90 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
94 * This is called by the log manager code to determine the LSN of the tail of
95 * the log. This is exactly the LSN of the first item in the AIL. If the AIL
96 * is empty, then this function returns 0.
98 * We need the AIL lock in order to get a coherent read of the lsn of the last
103 struct xfs_ail *ailp)
108 spin_lock(&ailp->xa_lock);
109 lip = xfs_ail_min(ailp);
112 spin_unlock(&ailp->xa_lock);
118 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
122 struct xfs_ail *ailp)
127 spin_lock(&ailp->xa_lock);
128 lip = xfs_ail_max(ailp);
131 spin_unlock(&ailp->xa_lock);
137 * The cursor keeps track of where our current traversal is up to by tracking
138 * the next item in the list for us. However, for this to be safe, removing an
139 * object from the AIL needs to invalidate any cursor that points to it. hence
140 * the traversal cursor needs to be linked to the struct xfs_ail so that
141 * deletion can search all the active cursors for invalidation.
144 xfs_trans_ail_cursor_init(
145 struct xfs_ail *ailp,
146 struct xfs_ail_cursor *cur)
149 list_add_tail(&cur->list, &ailp->xa_cursors);
153 * Get the next item in the traversal and advance the cursor. If the cursor
154 * was invalidated (indicated by a lip of 1), restart the traversal.
156 struct xfs_log_item *
157 xfs_trans_ail_cursor_next(
158 struct xfs_ail *ailp,
159 struct xfs_ail_cursor *cur)
161 struct xfs_log_item *lip = cur->item;
163 if ((__psint_t)lip & 1)
164 lip = xfs_ail_min(ailp);
166 cur->item = xfs_ail_next(ailp, lip);
171 * When the traversal is complete, we need to remove the cursor from the list
172 * of traversing cursors.
175 xfs_trans_ail_cursor_done(
176 struct xfs_ail_cursor *cur)
179 list_del_init(&cur->list);
183 * Invalidate any cursor that is pointing to this item. This is called when an
184 * item is removed from the AIL. Any cursor pointing to this object is now
185 * invalid and the traversal needs to be terminated so it doesn't reference a
186 * freed object. We set the low bit of the cursor item pointer so we can
187 * distinguish between an invalidation and the end of the list when getting the
188 * next item from the cursor.
191 xfs_trans_ail_cursor_clear(
192 struct xfs_ail *ailp,
193 struct xfs_log_item *lip)
195 struct xfs_ail_cursor *cur;
197 list_for_each_entry(cur, &ailp->xa_cursors, list) {
198 if (cur->item == lip)
199 cur->item = (struct xfs_log_item *)
200 ((__psint_t)cur->item | 1);
205 * Find the first item in the AIL with the given @lsn by searching in ascending
206 * LSN order and initialise the cursor to point to the next item for a
207 * ascending traversal. Pass a @lsn of zero to initialise the cursor to the
208 * first item in the AIL. Returns NULL if the list is empty.
211 xfs_trans_ail_cursor_first(
212 struct xfs_ail *ailp,
213 struct xfs_ail_cursor *cur,
218 xfs_trans_ail_cursor_init(ailp, cur);
221 lip = xfs_ail_min(ailp);
225 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
226 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
233 cur->item = xfs_ail_next(ailp, lip);
237 static struct xfs_log_item *
238 __xfs_trans_ail_cursor_last(
239 struct xfs_ail *ailp,
244 list_for_each_entry_reverse(lip, &ailp->xa_ail, li_ail) {
245 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
252 * Find the last item in the AIL with the given @lsn by searching in descending
253 * LSN order and initialise the cursor to point to that item. If there is no
254 * item with the value of @lsn, then it sets the cursor to the last item with an
255 * LSN lower than @lsn. Returns NULL if the list is empty.
257 struct xfs_log_item *
258 xfs_trans_ail_cursor_last(
259 struct xfs_ail *ailp,
260 struct xfs_ail_cursor *cur,
263 xfs_trans_ail_cursor_init(ailp, cur);
264 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
269 * Splice the log item list into the AIL at the given LSN. We splice to the
270 * tail of the given LSN to maintain insert order for push traversals. The
271 * cursor is optional, allowing repeated updates to the same LSN to avoid
272 * repeated traversals. This should not be called with an empty list.
276 struct xfs_ail *ailp,
277 struct xfs_ail_cursor *cur,
278 struct list_head *list,
281 struct xfs_log_item *lip;
283 ASSERT(!list_empty(list));
286 * Use the cursor to determine the insertion point if one is
287 * provided. If not, or if the one we got is not valid,
288 * find the place in the AIL where the items belong.
290 lip = cur ? cur->item : NULL;
291 if (!lip || (__psint_t) lip & 1)
292 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
295 * If a cursor is provided, we know we're processing the AIL
296 * in lsn order, and future items to be spliced in will
297 * follow the last one being inserted now. Update the
298 * cursor to point to that last item, now while we have a
299 * reliable pointer to it.
302 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
305 * Finally perform the splice. Unless the AIL was empty,
306 * lip points to the item in the AIL _after_ which the new
307 * items should go. If lip is null the AIL was empty, so
308 * the new items go at the head of the AIL.
311 list_splice(list, &lip->li_ail);
313 list_splice(list, &ailp->xa_ail);
317 * Delete the given item from the AIL. Return a pointer to the item.
321 struct xfs_ail *ailp,
324 xfs_ail_check(ailp, lip);
325 list_del(&lip->li_ail);
326 xfs_trans_ail_cursor_clear(ailp, lip);
331 struct xfs_ail *ailp)
333 xfs_mount_t *mp = ailp->xa_mount;
334 struct xfs_ail_cursor cur;
344 * If we encountered pinned items or did not finish writing out all
345 * buffers the last time we ran, force the log first and wait for it
346 * before pushing again.
348 if (ailp->xa_log_flush && ailp->xa_last_pushed_lsn == 0 &&
349 (!list_empty_careful(&ailp->xa_buf_list) ||
350 xfs_ail_min_lsn(ailp))) {
351 ailp->xa_log_flush = 0;
353 XFS_STATS_INC(xs_push_ail_flush);
354 xfs_log_force(mp, XFS_LOG_SYNC);
357 spin_lock(&ailp->xa_lock);
359 /* barrier matches the xa_target update in xfs_ail_push() */
361 target = ailp->xa_target;
362 ailp->xa_target_prev = target;
364 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->xa_last_pushed_lsn);
367 * If the AIL is empty or our push has reached the end we are
370 xfs_trans_ail_cursor_done(&cur);
371 spin_unlock(&ailp->xa_lock);
375 XFS_STATS_INC(xs_push_ail);
378 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
382 * Note that iop_push may unlock and reacquire the AIL lock. We
383 * rely on the AIL cursor implementation to be able to deal with
386 lock_result = lip->li_ops->iop_push(lip, &ailp->xa_buf_list);
387 switch (lock_result) {
388 case XFS_ITEM_SUCCESS:
389 XFS_STATS_INC(xs_push_ail_success);
390 trace_xfs_ail_push(lip);
392 ailp->xa_last_pushed_lsn = lsn;
395 case XFS_ITEM_FLUSHING:
397 * The item or its backing buffer is already beeing
398 * flushed. The typical reason for that is that an
399 * inode buffer is locked because we already pushed the
400 * updates to it as part of inode clustering.
402 * We do not want to to stop flushing just because lots
403 * of items are already beeing flushed, but we need to
404 * re-try the flushing relatively soon if most of the
405 * AIL is beeing flushed.
407 XFS_STATS_INC(xs_push_ail_flushing);
408 trace_xfs_ail_flushing(lip);
411 ailp->xa_last_pushed_lsn = lsn;
414 case XFS_ITEM_PINNED:
415 XFS_STATS_INC(xs_push_ail_pinned);
416 trace_xfs_ail_pinned(lip);
419 ailp->xa_log_flush++;
421 case XFS_ITEM_LOCKED:
422 XFS_STATS_INC(xs_push_ail_locked);
423 trace_xfs_ail_locked(lip);
435 * Are there too many items we can't do anything with?
437 * If we we are skipping too many items because we can't flush
438 * them or they are already being flushed, we back off and
439 * given them time to complete whatever operation is being
440 * done. i.e. remove pressure from the AIL while we can't make
441 * progress so traversals don't slow down further inserts and
442 * removals to/from the AIL.
444 * The value of 100 is an arbitrary magic number based on
450 lip = xfs_trans_ail_cursor_next(ailp, &cur);
455 xfs_trans_ail_cursor_done(&cur);
456 spin_unlock(&ailp->xa_lock);
458 if (xfs_buf_delwri_submit_nowait(&ailp->xa_buf_list))
459 ailp->xa_log_flush++;
461 if (!count || XFS_LSN_CMP(lsn, target) >= 0) {
464 * We reached the target or the AIL is empty, so wait a bit
465 * longer for I/O to complete and remove pushed items from the
466 * AIL before we start the next scan from the start of the AIL.
469 ailp->xa_last_pushed_lsn = 0;
470 } else if (((stuck + flushing) * 100) / count > 90) {
472 * Either there is a lot of contention on the AIL or we are
473 * stuck due to operations in progress. "Stuck" in this case
474 * is defined as >90% of the items we tried to push were stuck.
476 * Backoff a bit more to allow some I/O to complete before
477 * restarting from the start of the AIL. This prevents us from
478 * spinning on the same items, and if they are pinned will all
479 * the restart to issue a log force to unpin the stuck items.
482 ailp->xa_last_pushed_lsn = 0;
485 * Assume we have more work to do in a short while.
497 struct xfs_ail *ailp = data;
498 long tout = 0; /* milliseconds */
500 current->flags |= PF_MEMALLOC;
502 while (!kthread_should_stop()) {
503 if (tout && tout <= 20)
504 __set_current_state(TASK_KILLABLE);
506 __set_current_state(TASK_INTERRUPTIBLE);
508 spin_lock(&ailp->xa_lock);
511 * Idle if the AIL is empty and we are not racing with a target
512 * update. We check the AIL after we set the task to a sleep
513 * state to guarantee that we either catch an xa_target update
514 * or that a wake_up resets the state to TASK_RUNNING.
515 * Otherwise, we run the risk of sleeping indefinitely.
517 * The barrier matches the xa_target update in xfs_ail_push().
520 if (!xfs_ail_min(ailp) &&
521 ailp->xa_target == ailp->xa_target_prev) {
522 spin_unlock(&ailp->xa_lock);
527 spin_unlock(&ailp->xa_lock);
530 schedule_timeout(msecs_to_jiffies(tout));
532 __set_current_state(TASK_RUNNING);
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 trace_xfs_ail_move(lip, lip->li_lsn, lsn);
662 xfs_ail_delete(ailp, lip);
666 lip->li_flags |= XFS_LI_IN_AIL;
667 trace_xfs_ail_insert(lip, 0, lsn);
670 list_add(&lip->li_ail, &tmp);
673 if (!list_empty(&tmp))
674 xfs_ail_splice(ailp, cur, &tmp, lsn);
677 if (!XFS_FORCED_SHUTDOWN(ailp->xa_mount))
678 xlog_assign_tail_lsn_locked(ailp->xa_mount);
679 spin_unlock(&ailp->xa_lock);
681 xfs_log_space_wake(ailp->xa_mount);
683 spin_unlock(&ailp->xa_lock);
688 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
690 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
691 * removed from the AIL. The caller is already holding the AIL lock, and done
692 * all the checks necessary to ensure the items passed in via @log_items are
693 * ready for deletion. This includes checking that the items are in the AIL.
695 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
696 * flag from the item and reset the item's lsn to 0. If we remove the first
697 * item in the AIL, update the log tail to match the new minimum LSN in the
700 * This function will not drop the AIL lock until all items are removed from
701 * the AIL to minimise the amount of lock traffic on the AIL. This does not
702 * greatly increase the AIL hold time, but does significantly reduce the amount
703 * of traffic on the lock, especially during IO completion.
705 * This function must be called with the AIL lock held. The lock is dropped
709 xfs_trans_ail_delete_bulk(
710 struct xfs_ail *ailp,
711 struct xfs_log_item **log_items,
713 int shutdown_type) __releases(ailp->xa_lock)
715 xfs_log_item_t *mlip;
716 int mlip_changed = 0;
719 mlip = xfs_ail_min(ailp);
721 for (i = 0; i < nr_items; i++) {
722 struct xfs_log_item *lip = log_items[i];
723 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
724 struct xfs_mount *mp = ailp->xa_mount;
726 spin_unlock(&ailp->xa_lock);
727 if (!XFS_FORCED_SHUTDOWN(mp)) {
728 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
729 "%s: attempting to delete a log item that is not in the AIL",
731 xfs_force_shutdown(mp, shutdown_type);
736 trace_xfs_ail_delete(lip, mlip->li_lsn, lip->li_lsn);
737 xfs_ail_delete(ailp, lip);
738 lip->li_flags &= ~XFS_LI_IN_AIL;
745 if (!XFS_FORCED_SHUTDOWN(ailp->xa_mount))
746 xlog_assign_tail_lsn_locked(ailp->xa_mount);
747 if (list_empty(&ailp->xa_ail))
748 wake_up_all(&ailp->xa_empty);
749 spin_unlock(&ailp->xa_lock);
751 xfs_log_space_wake(ailp->xa_mount);
753 spin_unlock(&ailp->xa_lock);
761 struct xfs_ail *ailp;
763 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
768 INIT_LIST_HEAD(&ailp->xa_ail);
769 INIT_LIST_HEAD(&ailp->xa_cursors);
770 spin_lock_init(&ailp->xa_lock);
771 INIT_LIST_HEAD(&ailp->xa_buf_list);
772 init_waitqueue_head(&ailp->xa_empty);
774 ailp->xa_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
775 ailp->xa_mount->m_fsname);
776 if (IS_ERR(ailp->xa_task))
788 xfs_trans_ail_destroy(
791 struct xfs_ail *ailp = mp->m_ail;
793 kthread_stop(ailp->xa_task);