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 STATIC
void xfs_ail_splice(struct xfs_ail
*, struct list_head
*, xfs_lsn_t
);
32 STATIC
void xfs_ail_delete(struct xfs_ail
*, xfs_log_item_t
*);
33 STATIC xfs_log_item_t
* xfs_ail_min(struct xfs_ail
*);
34 STATIC xfs_log_item_t
* xfs_ail_next(struct xfs_ail
*, xfs_log_item_t
*);
37 STATIC
void xfs_ail_check(struct xfs_ail
*, xfs_log_item_t
*);
39 #define xfs_ail_check(a,l)
44 * This is called by the log manager code to determine the LSN
45 * of the tail of the log. This is exactly the LSN of the first
46 * item in the AIL. If the AIL is empty, then this function
49 * We need the AIL lock in order to get a coherent read of the
50 * lsn of the last item in the AIL.
59 spin_lock(&ailp
->xa_lock
);
60 lip
= xfs_ail_min(ailp
);
66 spin_unlock(&ailp
->xa_lock
);
74 * This routine is called to move the tail of the AIL forward. It does this by
75 * trying to flush items in the AIL whose lsns are below the given
78 * the push is run asynchronously in a separate thread, so we return the tail
79 * of the log right now instead of the tail after the push. This means we will
80 * either continue right away, or we will sleep waiting on the async thread to
83 * We do this unlocked - we only need to know whether there is anything in the
84 * AIL at the time we are called. We don't need to access the contents of
85 * any of the objects, so the lock is not needed.
90 xfs_lsn_t threshold_lsn
)
94 lip
= xfs_ail_min(ailp
);
95 if (lip
&& !XFS_FORCED_SHUTDOWN(ailp
->xa_mount
)) {
96 if (XFS_LSN_CMP(threshold_lsn
, ailp
->xa_target
) > 0)
97 xfsaild_wakeup(ailp
, threshold_lsn
);
102 * AIL traversal cursor initialisation.
104 * The cursor keeps track of where our current traversal is up
105 * to by tracking the next ƣtem in the list for us. However, for
106 * this to be safe, removing an object from the AIL needs to invalidate
107 * any cursor that points to it. hence the traversal cursor needs to
108 * be linked to the struct xfs_ail so that deletion can search all the
109 * active cursors for invalidation.
111 * We don't link the push cursor because it is embedded in the struct
112 * xfs_ail and hence easily findable.
115 xfs_trans_ail_cursor_init(
116 struct xfs_ail
*ailp
,
117 struct xfs_ail_cursor
*cur
)
120 if (cur
== &ailp
->xa_cursors
)
123 cur
->next
= ailp
->xa_cursors
.next
;
124 ailp
->xa_cursors
.next
= cur
;
128 * Set the cursor to the next item, because when we look
129 * up the cursor the current item may have been freed.
132 xfs_trans_ail_cursor_set(
133 struct xfs_ail
*ailp
,
134 struct xfs_ail_cursor
*cur
,
135 struct xfs_log_item
*lip
)
138 cur
->item
= xfs_ail_next(ailp
, lip
);
142 * Get the next item in the traversal and advance the cursor.
143 * If the cursor was invalidated (inidicated by a lip of 1),
144 * restart the traversal.
146 struct xfs_log_item
*
147 xfs_trans_ail_cursor_next(
148 struct xfs_ail
*ailp
,
149 struct xfs_ail_cursor
*cur
)
151 struct xfs_log_item
*lip
= cur
->item
;
153 if ((__psint_t
)lip
& 1)
154 lip
= xfs_ail_min(ailp
);
155 xfs_trans_ail_cursor_set(ailp
, cur
, lip
);
160 * Now that the traversal is complete, we need to remove the cursor
161 * from the list of traversing cursors. Avoid removing the embedded
162 * push cursor, but use the fact it is always present to make the
163 * list deletion simple.
166 xfs_trans_ail_cursor_done(
167 struct xfs_ail
*ailp
,
168 struct xfs_ail_cursor
*done
)
170 struct xfs_ail_cursor
*prev
= NULL
;
171 struct xfs_ail_cursor
*cur
;
174 if (done
== &ailp
->xa_cursors
)
176 prev
= &ailp
->xa_cursors
;
177 for (cur
= prev
->next
; cur
; prev
= cur
, cur
= prev
->next
) {
179 prev
->next
= cur
->next
;
187 * Invalidate any cursor that is pointing to this item. This is
188 * called when an item is removed from the AIL. Any cursor pointing
189 * to this object is now invalid and the traversal needs to be
190 * terminated so it doesn't reference a freed object. We set the
191 * cursor item to a value of 1 so we can distinguish between an
192 * invalidation and the end of the list when getting the next item
196 xfs_trans_ail_cursor_clear(
197 struct xfs_ail
*ailp
,
198 struct xfs_log_item
*lip
)
200 struct xfs_ail_cursor
*cur
;
202 /* need to search all cursors */
203 for (cur
= &ailp
->xa_cursors
; cur
; cur
= cur
->next
) {
204 if (cur
->item
== lip
)
205 cur
->item
= (struct xfs_log_item
*)
206 ((__psint_t
)cur
->item
| 1);
211 * Return the item in the AIL with the current lsn.
212 * Return the current tree generation number for use
213 * in calls to xfs_trans_next_ail().
216 xfs_trans_ail_cursor_first(
217 struct xfs_ail
*ailp
,
218 struct xfs_ail_cursor
*cur
,
223 xfs_trans_ail_cursor_init(ailp
, cur
);
224 lip
= xfs_ail_min(ailp
);
228 list_for_each_entry(lip
, &ailp
->xa_ail
, li_ail
) {
229 if (XFS_LSN_CMP(lip
->li_lsn
, lsn
) >= 0)
234 xfs_trans_ail_cursor_set(ailp
, cur
, lip
);
239 * xfsaild_push does the work of pushing on the AIL. Returning a timeout of
240 * zero indicates that the caller should sleep until woken.
244 struct xfs_ail
*ailp
,
248 xfs_lsn_t last_pushed_lsn
= *last_lsn
;
249 xfs_lsn_t target
= ailp
->xa_target
;
252 int flush_log
, count
, stuck
;
253 xfs_mount_t
*mp
= ailp
->xa_mount
;
254 struct xfs_ail_cursor
*cur
= &ailp
->xa_cursors
;
255 int push_xfsbufd
= 0;
257 spin_lock(&ailp
->xa_lock
);
258 xfs_trans_ail_cursor_init(ailp
, cur
);
259 lip
= xfs_trans_ail_cursor_first(ailp
, cur
, *last_lsn
);
260 if (!lip
|| XFS_FORCED_SHUTDOWN(mp
)) {
262 * AIL is empty or our push has reached the end.
264 xfs_trans_ail_cursor_done(ailp
, cur
);
265 spin_unlock(&ailp
->xa_lock
);
270 XFS_STATS_INC(xs_push_ail
);
273 * While the item we are looking at is below the given threshold
274 * try to flush it out. We'd like not to stop until we've at least
275 * tried to push on everything in the AIL with an LSN less than
276 * the given threshold.
278 * However, we will stop after a certain number of pushes and wait
279 * for a reduced timeout to fire before pushing further. This
280 * prevents use from spinning when we can't do anything or there is
281 * lots of contention on the AIL lists.
284 flush_log
= stuck
= count
= 0;
285 while ((XFS_LSN_CMP(lip
->li_lsn
, target
) < 0)) {
288 * If we can lock the item without sleeping, unlock the AIL
289 * lock and flush the item. Then re-grab the AIL lock so we
290 * can look for the next item on the AIL. List changes are
291 * handled by the AIL lookup functions internally
293 * If we can't lock the item, either its holder will flush it
294 * or it is already being flushed or it is being relogged. In
295 * any of these case it is being taken care of and we can just
296 * skip to the next item in the list.
298 lock_result
= IOP_TRYLOCK(lip
);
299 spin_unlock(&ailp
->xa_lock
);
300 switch (lock_result
) {
301 case XFS_ITEM_SUCCESS
:
302 XFS_STATS_INC(xs_push_ail_success
);
304 last_pushed_lsn
= lsn
;
307 case XFS_ITEM_PUSHBUF
:
308 XFS_STATS_INC(xs_push_ail_pushbuf
);
310 last_pushed_lsn
= lsn
;
314 case XFS_ITEM_PINNED
:
315 XFS_STATS_INC(xs_push_ail_pinned
);
320 case XFS_ITEM_LOCKED
:
321 XFS_STATS_INC(xs_push_ail_locked
);
322 last_pushed_lsn
= lsn
;
331 spin_lock(&ailp
->xa_lock
);
332 /* should we bother continuing? */
333 if (XFS_FORCED_SHUTDOWN(mp
))
340 * Are there too many items we can't do anything with?
341 * If we we are skipping too many items because we can't flush
342 * them or they are already being flushed, we back off and
343 * given them time to complete whatever operation is being
344 * done. i.e. remove pressure from the AIL while we can't make
345 * progress so traversals don't slow down further inserts and
346 * removals to/from the AIL.
348 * The value of 100 is an arbitrary magic number based on
354 lip
= xfs_trans_ail_cursor_next(ailp
, cur
);
359 xfs_trans_ail_cursor_done(ailp
, cur
);
360 spin_unlock(&ailp
->xa_lock
);
364 * If something we need to push out was pinned, then
365 * push out the log so it will become unpinned and
366 * move forward in the AIL.
368 XFS_STATS_INC(xs_push_ail_flush
);
369 xfs_log_force(mp
, 0);
373 /* we've got delayed write buffers to flush */
374 wake_up_process(mp
->m_ddev_targp
->bt_task
);
378 /* We're past our target or empty, so idle */
380 } else if (XFS_LSN_CMP(lsn
, target
) >= 0) {
382 * We reached the target so wait a bit longer for I/O to
383 * complete and remove pushed items from the AIL before we
384 * start the next scan from the start of the AIL.
388 } else if ((stuck
* 100) / count
> 90) {
390 * Either there is a lot of contention on the AIL or we
391 * are stuck due to operations in progress. "Stuck" in this
392 * case is defined as >90% of the items we tried to push
395 * Backoff a bit more to allow some I/O to complete before
396 * continuing from where we were.
400 /* more to do, but wait a short while before continuing */
403 *last_lsn
= last_pushed_lsn
;
409 * This is to be called when an item is unlocked that may have
410 * been in the AIL. It will wake up the first member of the AIL
411 * wait list if this item's unlocking might allow it to progress.
412 * If the item is in the AIL, then we need to get the AIL lock
413 * while doing our checking so we don't race with someone going
414 * to sleep waiting for this event in xfs_trans_push_ail().
417 xfs_trans_unlocked_item(
418 struct xfs_ail
*ailp
,
421 xfs_log_item_t
*min_lip
;
424 * If we're forcibly shutting down, we may have
425 * unlocked log items arbitrarily. The last thing
426 * we want to do is to move the tail of the log
427 * over some potentially valid data.
429 if (!(lip
->li_flags
& XFS_LI_IN_AIL
) ||
430 XFS_FORCED_SHUTDOWN(ailp
->xa_mount
)) {
435 * This is the one case where we can call into xfs_ail_min()
436 * without holding the AIL lock because we only care about the
437 * case where we are at the tail of the AIL. If the object isn't
438 * at the tail, it doesn't matter what result we get back. This
439 * is slightly racy because since we were just unlocked, we could
440 * go to sleep between the call to xfs_ail_min and the call to
441 * xfs_log_move_tail, have someone else lock us, commit to us disk,
442 * move us out of the tail of the AIL, and then we wake up. However,
443 * the call to xfs_log_move_tail() doesn't do anything if there's
444 * not enough free space to wake people up so we're safe calling it.
446 min_lip
= xfs_ail_min(ailp
);
449 xfs_log_move_tail(ailp
->xa_mount
, 1);
450 } /* xfs_trans_unlocked_item */
453 * xfs_trans_ail_update - bulk AIL insertion operation.
455 * @xfs_trans_ail_update takes an array of log items that all need to be
456 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
457 * be added. Otherwise, it will be repositioned by removing it and re-adding
458 * it to the AIL. If we move the first item in the AIL, update the log tail to
459 * match the new minimum LSN in the AIL.
461 * This function takes the AIL lock once to execute the update operations on
462 * all the items in the array, and as such should not be called with the AIL
463 * lock held. As a result, once we have the AIL lock, we need to check each log
464 * item LSN to confirm it needs to be moved forward in the AIL.
466 * To optimise the insert operation, we delete all the items from the AIL in
467 * the first pass, moving them into a temporary list, then splice the temporary
468 * list into the correct position in the AIL. This avoids needing to do an
469 * insert operation on every item.
471 * This function must be called with the AIL lock held. The lock is dropped
475 xfs_trans_ail_update_bulk(
476 struct xfs_ail
*ailp
,
477 struct xfs_log_item
**log_items
,
479 xfs_lsn_t lsn
) __releases(ailp
->xa_lock
)
481 xfs_log_item_t
*mlip
;
483 int mlip_changed
= 0;
487 mlip
= xfs_ail_min(ailp
);
489 for (i
= 0; i
< nr_items
; i
++) {
490 struct xfs_log_item
*lip
= log_items
[i
];
491 if (lip
->li_flags
& XFS_LI_IN_AIL
) {
492 /* check if we really need to move the item */
493 if (XFS_LSN_CMP(lsn
, lip
->li_lsn
) <= 0)
496 xfs_ail_delete(ailp
, lip
);
500 lip
->li_flags
|= XFS_LI_IN_AIL
;
503 list_add(&lip
->li_ail
, &tmp
);
506 xfs_ail_splice(ailp
, &tmp
, lsn
);
509 spin_unlock(&ailp
->xa_lock
);
514 * It is not safe to access mlip after the AIL lock is dropped, so we
515 * must get a copy of li_lsn before we do so. This is especially
516 * important on 32-bit platforms where accessing and updating 64-bit
517 * values like li_lsn is not atomic.
519 mlip
= xfs_ail_min(ailp
);
520 tail_lsn
= mlip
->li_lsn
;
521 spin_unlock(&ailp
->xa_lock
);
522 xfs_log_move_tail(ailp
->xa_mount
, tail_lsn
);
526 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
528 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
529 * removed from the AIL. The caller is already holding the AIL lock, and done
530 * all the checks necessary to ensure the items passed in via @log_items are
531 * ready for deletion. This includes checking that the items are in the AIL.
533 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
534 * flag from the item and reset the item's lsn to 0. If we remove the first
535 * item in the AIL, update the log tail to match the new minimum LSN in the
538 * This function will not drop the AIL lock until all items are removed from
539 * the AIL to minimise the amount of lock traffic on the AIL. This does not
540 * greatly increase the AIL hold time, but does significantly reduce the amount
541 * of traffic on the lock, especially during IO completion.
543 * This function must be called with the AIL lock held. The lock is dropped
547 xfs_trans_ail_delete_bulk(
548 struct xfs_ail
*ailp
,
549 struct xfs_log_item
**log_items
,
550 int nr_items
) __releases(ailp
->xa_lock
)
552 xfs_log_item_t
*mlip
;
554 int mlip_changed
= 0;
557 mlip
= xfs_ail_min(ailp
);
559 for (i
= 0; i
< nr_items
; i
++) {
560 struct xfs_log_item
*lip
= log_items
[i
];
561 if (!(lip
->li_flags
& XFS_LI_IN_AIL
)) {
562 struct xfs_mount
*mp
= ailp
->xa_mount
;
564 spin_unlock(&ailp
->xa_lock
);
565 if (!XFS_FORCED_SHUTDOWN(mp
)) {
566 xfs_cmn_err(XFS_PTAG_AILDELETE
, CE_ALERT
, mp
,
567 "%s: attempting to delete a log item that is not in the AIL",
569 xfs_force_shutdown(mp
, SHUTDOWN_CORRUPT_INCORE
);
574 xfs_ail_delete(ailp
, lip
);
575 lip
->li_flags
&= ~XFS_LI_IN_AIL
;
582 spin_unlock(&ailp
->xa_lock
);
587 * It is not safe to access mlip after the AIL lock is dropped, so we
588 * must get a copy of li_lsn before we do so. This is especially
589 * important on 32-bit platforms where accessing and updating 64-bit
590 * values like li_lsn is not atomic. It is possible we've emptied the
591 * AIL here, so if that is the case, pass an LSN of 0 to the tail move.
593 mlip
= xfs_ail_min(ailp
);
594 tail_lsn
= mlip
? mlip
->li_lsn
: 0;
595 spin_unlock(&ailp
->xa_lock
);
596 xfs_log_move_tail(ailp
->xa_mount
, tail_lsn
);
600 * The active item list (AIL) is a doubly linked list of log
601 * items sorted by ascending lsn. The base of the list is
602 * a forw/back pointer pair embedded in the xfs mount structure.
603 * The base is initialized with both pointers pointing to the
604 * base. This case always needs to be distinguished, because
605 * the base has no lsn to look at. We almost always insert
606 * at the end of the list, so on inserts we search from the
607 * end of the list to find where the new item belongs.
611 * Initialize the doubly linked list to point only to itself.
617 struct xfs_ail
*ailp
;
620 ailp
= kmem_zalloc(sizeof(struct xfs_ail
), KM_MAYFAIL
);
625 INIT_LIST_HEAD(&ailp
->xa_ail
);
626 spin_lock_init(&ailp
->xa_lock
);
627 error
= xfsaild_start(ailp
);
639 xfs_trans_ail_destroy(
642 struct xfs_ail
*ailp
= mp
->m_ail
;
649 * splice the log item list into the AIL at the given LSN.
653 struct xfs_ail
*ailp
,
654 struct list_head
*list
,
657 xfs_log_item_t
*next_lip
;
660 * If the list is empty, just insert the item.
662 if (list_empty(&ailp
->xa_ail
)) {
663 list_splice(list
, &ailp
->xa_ail
);
667 list_for_each_entry_reverse(next_lip
, &ailp
->xa_ail
, li_ail
) {
668 if (XFS_LSN_CMP(next_lip
->li_lsn
, lsn
) <= 0)
672 ASSERT((&next_lip
->li_ail
== &ailp
->xa_ail
) ||
673 (XFS_LSN_CMP(next_lip
->li_lsn
, lsn
) <= 0));
675 list_splice_init(list
, &next_lip
->li_ail
);
680 * Delete the given item from the AIL. Return a pointer to the item.
684 struct xfs_ail
*ailp
,
687 xfs_ail_check(ailp
, lip
);
688 list_del(&lip
->li_ail
);
689 xfs_trans_ail_cursor_clear(ailp
, lip
);
693 * Return a pointer to the first item in the AIL.
694 * If the AIL is empty, then return NULL.
696 STATIC xfs_log_item_t
*
698 struct xfs_ail
*ailp
)
700 if (list_empty(&ailp
->xa_ail
))
703 return list_first_entry(&ailp
->xa_ail
, xfs_log_item_t
, li_ail
);
707 * Return a pointer to the item which follows
708 * the given item in the AIL. If the given item
709 * is the last item in the list, then return NULL.
711 STATIC xfs_log_item_t
*
713 struct xfs_ail
*ailp
,
716 if (lip
->li_ail
.next
== &ailp
->xa_ail
)
719 return list_first_entry(&lip
->li_ail
, xfs_log_item_t
, li_ail
);
724 * Check that the list is sorted as it should be.
728 struct xfs_ail
*ailp
,
731 xfs_log_item_t
*prev_lip
;
733 if (list_empty(&ailp
->xa_ail
))
737 * Check the next and previous entries are valid.
739 ASSERT((lip
->li_flags
& XFS_LI_IN_AIL
) != 0);
740 prev_lip
= list_entry(lip
->li_ail
.prev
, xfs_log_item_t
, li_ail
);
741 if (&prev_lip
->li_ail
!= &ailp
->xa_ail
)
742 ASSERT(XFS_LSN_CMP(prev_lip
->li_lsn
, lip
->li_lsn
) <= 0);
744 prev_lip
= list_entry(lip
->li_ail
.next
, xfs_log_item_t
, li_ail
);
745 if (&prev_lip
->li_ail
!= &ailp
->xa_ail
)
746 ASSERT(XFS_LSN_CMP(prev_lip
->li_lsn
, lip
->li_lsn
) >= 0);
749 #ifdef XFS_TRANS_DEBUG
751 * Walk the list checking lsn ordering, and that every entry has the
752 * XFS_LI_IN_AIL flag set. This is really expensive, so only do it
753 * when specifically debugging the transaction subsystem.
755 prev_lip
= list_entry(&ailp
->xa_ail
, xfs_log_item_t
, li_ail
);
756 list_for_each_entry(lip
, &ailp
->xa_ail
, li_ail
) {
757 if (&prev_lip
->li_ail
!= &ailp
->xa_ail
)
758 ASSERT(XFS_LSN_CMP(prev_lip
->li_lsn
, lip
->li_lsn
) <= 0);
759 ASSERT((lip
->li_flags
& XFS_LI_IN_AIL
) != 0);
762 #endif /* XFS_TRANS_DEBUG */