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 * Return the item in the AIL with the current lsn.
276 * Return the current tree generation number for use
277 * in calls to xfs_trans_next_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 * splice the log item list into the AIL at the given LSN.
307 struct xfs_ail
*ailp
,
308 struct list_head
*list
,
311 xfs_log_item_t
*next_lip
;
313 /* If the list is empty, just insert the item. */
314 if (list_empty(&ailp
->xa_ail
)) {
315 list_splice(list
, &ailp
->xa_ail
);
319 list_for_each_entry_reverse(next_lip
, &ailp
->xa_ail
, li_ail
) {
320 if (XFS_LSN_CMP(next_lip
->li_lsn
, lsn
) <= 0)
324 ASSERT(&next_lip
->li_ail
== &ailp
->xa_ail
||
325 XFS_LSN_CMP(next_lip
->li_lsn
, lsn
) <= 0);
327 list_splice_init(list
, &next_lip
->li_ail
);
331 * Delete the given item from the AIL. Return a pointer to the item.
335 struct xfs_ail
*ailp
,
338 xfs_ail_check(ailp
, lip
);
339 list_del(&lip
->li_ail
);
340 xfs_trans_ail_cursor_clear(ailp
, lip
);
344 * xfs_ail_worker does the work of pushing on the AIL. It will requeue itself
345 * to run at a later time if there is more work to do to complete the push.
349 struct work_struct
*work
)
351 struct xfs_ail
*ailp
= container_of(to_delayed_work(work
),
352 struct xfs_ail
, xa_work
);
353 xfs_mount_t
*mp
= ailp
->xa_mount
;
354 struct xfs_ail_cursor
*cur
= &ailp
->xa_cursors
;
362 int push_xfsbufd
= 0;
364 spin_lock(&ailp
->xa_lock
);
365 target
= ailp
->xa_target
;
366 xfs_trans_ail_cursor_init(ailp
, cur
);
367 lip
= xfs_trans_ail_cursor_first(ailp
, cur
, ailp
->xa_last_pushed_lsn
);
368 if (!lip
|| XFS_FORCED_SHUTDOWN(mp
)) {
370 * AIL is empty or our push has reached the end.
372 xfs_trans_ail_cursor_done(ailp
, cur
);
373 spin_unlock(&ailp
->xa_lock
);
377 XFS_STATS_INC(xs_push_ail
);
380 * While the item we are looking at is below the given threshold
381 * try to flush it out. We'd like not to stop until we've at least
382 * tried to push on everything in the AIL with an LSN less than
383 * the given threshold.
385 * However, we will stop after a certain number of pushes and wait
386 * for a reduced timeout to fire before pushing further. This
387 * prevents use from spinning when we can't do anything or there is
388 * lots of contention on the AIL lists.
391 while ((XFS_LSN_CMP(lip
->li_lsn
, target
) <= 0)) {
394 * If we can lock the item without sleeping, unlock the AIL
395 * lock and flush the item. Then re-grab the AIL lock so we
396 * can look for the next item on the AIL. List changes are
397 * handled by the AIL lookup functions internally
399 * If we can't lock the item, either its holder will flush it
400 * or it is already being flushed or it is being relogged. In
401 * any of these case it is being taken care of and we can just
402 * skip to the next item in the list.
404 lock_result
= IOP_TRYLOCK(lip
);
405 spin_unlock(&ailp
->xa_lock
);
406 switch (lock_result
) {
407 case XFS_ITEM_SUCCESS
:
408 XFS_STATS_INC(xs_push_ail_success
);
410 ailp
->xa_last_pushed_lsn
= lsn
;
413 case XFS_ITEM_PUSHBUF
:
414 XFS_STATS_INC(xs_push_ail_pushbuf
);
416 ailp
->xa_last_pushed_lsn
= lsn
;
420 case XFS_ITEM_PINNED
:
421 XFS_STATS_INC(xs_push_ail_pinned
);
426 case XFS_ITEM_LOCKED
:
427 XFS_STATS_INC(xs_push_ail_locked
);
428 ailp
->xa_last_pushed_lsn
= lsn
;
437 spin_lock(&ailp
->xa_lock
);
438 /* should we bother continuing? */
439 if (XFS_FORCED_SHUTDOWN(mp
))
446 * Are there too many items we can't do anything with?
447 * If we we are skipping too many items because we can't flush
448 * them or they are already being flushed, we back off and
449 * given them time to complete whatever operation is being
450 * done. i.e. remove pressure from the AIL while we can't make
451 * progress so traversals don't slow down further inserts and
452 * removals to/from the AIL.
454 * The value of 100 is an arbitrary magic number based on
460 lip
= xfs_trans_ail_cursor_next(ailp
, cur
);
465 xfs_trans_ail_cursor_done(ailp
, cur
);
466 spin_unlock(&ailp
->xa_lock
);
470 * If something we need to push out was pinned, then
471 * push out the log so it will become unpinned and
472 * move forward in the AIL.
474 XFS_STATS_INC(xs_push_ail_flush
);
475 xfs_log_force(mp
, 0);
479 /* we've got delayed write buffers to flush */
480 wake_up_process(mp
->m_ddev_targp
->bt_task
);
483 /* assume we have more work to do in a short while */
486 /* We're past our target or empty, so idle */
487 ailp
->xa_last_pushed_lsn
= 0;
490 * We clear the XFS_AIL_PUSHING_BIT first before checking
491 * whether the target has changed. If the target has changed,
492 * this pushes the requeue race directly onto the result of the
493 * atomic test/set bit, so we are guaranteed that either the
494 * the pusher that changed the target or ourselves will requeue
495 * the work (but not both).
497 clear_bit(XFS_AIL_PUSHING_BIT
, &ailp
->xa_flags
);
499 if (XFS_LSN_CMP(ailp
->xa_target
, target
) == 0 ||
500 test_and_set_bit(XFS_AIL_PUSHING_BIT
, &ailp
->xa_flags
))
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 * continuing from where we were.
525 /* There is more to do, requeue us. */
526 queue_delayed_work(xfs_syncd_wq
, &ailp
->xa_work
,
527 msecs_to_jiffies(tout
));
531 * This routine is called to move the tail of the AIL forward. It does this by
532 * trying to flush items in the AIL whose lsns are below the given
535 * The push is run asynchronously in a workqueue, which means the caller needs
536 * to handle waiting on the async flush for space to become available.
537 * We don't want to interrupt any push that is in progress, hence we only queue
538 * work if we set the pushing bit approriately.
540 * We do this unlocked - we only need to know whether there is anything in the
541 * AIL at the time we are called. We don't need to access the contents of
542 * any of the objects, so the lock is not needed.
546 struct xfs_ail
*ailp
,
547 xfs_lsn_t threshold_lsn
)
551 lip
= xfs_ail_min(ailp
);
552 if (!lip
|| XFS_FORCED_SHUTDOWN(ailp
->xa_mount
) ||
553 XFS_LSN_CMP(threshold_lsn
, ailp
->xa_target
) <= 0)
557 * Ensure that the new target is noticed in push code before it clears
558 * the XFS_AIL_PUSHING_BIT.
561 xfs_trans_ail_copy_lsn(ailp
, &ailp
->xa_target
, &threshold_lsn
);
562 if (!test_and_set_bit(XFS_AIL_PUSHING_BIT
, &ailp
->xa_flags
))
563 queue_delayed_work(xfs_syncd_wq
, &ailp
->xa_work
, 0);
567 * Push out all items in the AIL immediately
571 struct xfs_ail
*ailp
)
573 xfs_lsn_t threshold_lsn
= xfs_ail_max_lsn(ailp
);
576 xfs_ail_push(ailp
, threshold_lsn
);
580 * This is to be called when an item is unlocked that may have
581 * been in the AIL. It will wake up the first member of the AIL
582 * wait list if this item's unlocking might allow it to progress.
583 * If the item is in the AIL, then we need to get the AIL lock
584 * while doing our checking so we don't race with someone going
585 * to sleep waiting for this event in xfs_trans_push_ail().
588 xfs_trans_unlocked_item(
589 struct xfs_ail
*ailp
,
592 xfs_log_item_t
*min_lip
;
595 * If we're forcibly shutting down, we may have
596 * unlocked log items arbitrarily. The last thing
597 * we want to do is to move the tail of the log
598 * over some potentially valid data.
600 if (!(lip
->li_flags
& XFS_LI_IN_AIL
) ||
601 XFS_FORCED_SHUTDOWN(ailp
->xa_mount
)) {
606 * This is the one case where we can call into xfs_ail_min()
607 * without holding the AIL lock because we only care about the
608 * case where we are at the tail of the AIL. If the object isn't
609 * at the tail, it doesn't matter what result we get back. This
610 * is slightly racy because since we were just unlocked, we could
611 * go to sleep between the call to xfs_ail_min and the call to
612 * xfs_log_move_tail, have someone else lock us, commit to us disk,
613 * move us out of the tail of the AIL, and then we wake up. However,
614 * the call to xfs_log_move_tail() doesn't do anything if there's
615 * not enough free space to wake people up so we're safe calling it.
617 min_lip
= xfs_ail_min(ailp
);
620 xfs_log_move_tail(ailp
->xa_mount
, 1);
621 } /* xfs_trans_unlocked_item */
624 * xfs_trans_ail_update - bulk AIL insertion operation.
626 * @xfs_trans_ail_update takes an array of log items that all need to be
627 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
628 * be added. Otherwise, it will be repositioned by removing it and re-adding
629 * it to the AIL. If we move the first item in the AIL, update the log tail to
630 * match the new minimum LSN in the AIL.
632 * This function takes the AIL lock once to execute the update operations on
633 * all the items in the array, and as such should not be called with the AIL
634 * lock held. As a result, once we have the AIL lock, we need to check each log
635 * item LSN to confirm it needs to be moved forward in the AIL.
637 * To optimise the insert operation, we delete all the items from the AIL in
638 * the first pass, moving them into a temporary list, then splice the temporary
639 * list into the correct position in the AIL. This avoids needing to do an
640 * insert operation on every item.
642 * This function must be called with the AIL lock held. The lock is dropped
646 xfs_trans_ail_update_bulk(
647 struct xfs_ail
*ailp
,
648 struct xfs_log_item
**log_items
,
650 xfs_lsn_t lsn
) __releases(ailp
->xa_lock
)
652 xfs_log_item_t
*mlip
;
654 int mlip_changed
= 0;
658 mlip
= xfs_ail_min(ailp
);
660 for (i
= 0; i
< nr_items
; i
++) {
661 struct xfs_log_item
*lip
= log_items
[i
];
662 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
663 /* check if we really need to move the item */
664 if (XFS_LSN_CMP(lsn
, lip
->li_lsn
) <= 0)
667 xfs_ail_delete(ailp
, lip
);
671 lip
->li_flags
|= XFS_LI_IN_AIL
;
674 list_add(&lip
->li_ail
, &tmp
);
677 xfs_ail_splice(ailp
, &tmp
, lsn
);
680 spin_unlock(&ailp
->xa_lock
);
685 * It is not safe to access mlip after the AIL lock is dropped, so we
686 * must get a copy of li_lsn before we do so. This is especially
687 * important on 32-bit platforms where accessing and updating 64-bit
688 * values like li_lsn is not atomic.
690 mlip
= xfs_ail_min(ailp
);
691 tail_lsn
= mlip
->li_lsn
;
692 spin_unlock(&ailp
->xa_lock
);
693 xfs_log_move_tail(ailp
->xa_mount
, tail_lsn
);
697 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
699 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
700 * removed from the AIL. The caller is already holding the AIL lock, and done
701 * all the checks necessary to ensure the items passed in via @log_items are
702 * ready for deletion. This includes checking that the items are in the AIL.
704 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
705 * flag from the item and reset the item's lsn to 0. If we remove the first
706 * item in the AIL, update the log tail to match the new minimum LSN in the
709 * This function will not drop the AIL lock until all items are removed from
710 * the AIL to minimise the amount of lock traffic on the AIL. This does not
711 * greatly increase the AIL hold time, but does significantly reduce the amount
712 * of traffic on the lock, especially during IO completion.
714 * This function must be called with the AIL lock held. The lock is dropped
718 xfs_trans_ail_delete_bulk(
719 struct xfs_ail
*ailp
,
720 struct xfs_log_item
**log_items
,
721 int nr_items
) __releases(ailp
->xa_lock
)
723 xfs_log_item_t
*mlip
;
725 int mlip_changed
= 0;
728 mlip
= xfs_ail_min(ailp
);
730 for (i
= 0; i
< nr_items
; i
++) {
731 struct xfs_log_item
*lip
= log_items
[i
];
732 if (!(lip
->li_flags
& XFS_LI_IN_AIL
)) {
733 struct xfs_mount
*mp
= ailp
->xa_mount
;
735 spin_unlock(&ailp
->xa_lock
);
736 if (!XFS_FORCED_SHUTDOWN(mp
)) {
737 xfs_alert_tag(mp
, XFS_PTAG_AILDELETE
,
738 "%s: attempting to delete a log item that is not in the AIL",
740 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
745 xfs_ail_delete(ailp
, lip
);
746 lip
->li_flags
&= ~XFS_LI_IN_AIL
;
753 spin_unlock(&ailp
->xa_lock
);
758 * It is not safe to access mlip after the AIL lock is dropped, so we
759 * must get a copy of li_lsn before we do so. This is especially
760 * important on 32-bit platforms where accessing and updating 64-bit
761 * values like li_lsn is not atomic. It is possible we've emptied the
762 * AIL here, so if that is the case, pass an LSN of 0 to the tail move.
764 mlip
= xfs_ail_min(ailp
);
765 tail_lsn
= mlip
? mlip
->li_lsn
: 0;
766 spin_unlock(&ailp
->xa_lock
);
767 xfs_log_move_tail(ailp
->xa_mount
, tail_lsn
);
771 * The active item list (AIL) is a doubly linked list of log
772 * items sorted by ascending lsn. The base of the list is
773 * a forw/back pointer pair embedded in the xfs mount structure.
774 * The base is initialized with both pointers pointing to the
775 * base. This case always needs to be distinguished, because
776 * the base has no lsn to look at. We almost always insert
777 * at the end of the list, so on inserts we search from the
778 * end of the list to find where the new item belongs.
782 * Initialize the doubly linked list to point only to itself.
788 struct xfs_ail
*ailp
;
790 ailp
= kmem_zalloc(sizeof(struct xfs_ail
), KM_MAYFAIL
);
795 INIT_LIST_HEAD(&ailp
->xa_ail
);
796 spin_lock_init(&ailp
->xa_lock
);
797 INIT_DELAYED_WORK(&ailp
->xa_work
, xfs_ail_worker
);
803 xfs_trans_ail_destroy(
806 struct xfs_ail
*ailp
= mp
->m_ail
;
808 cancel_delayed_work_sync(&ailp
->xa_work
);