Loading...
1/*
2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * Copyright (c) 2008 Dave Chinner
4 * All Rights Reserved.
5 *
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.
9 *
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.
14 *
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
18 */
19#include "xfs.h"
20#include "xfs_fs.h"
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"
29#include "xfs_log.h"
30
31#ifdef DEBUG
32/*
33 * Check that the list is sorted as it should be.
34 */
35STATIC void
36xfs_ail_check(
37 struct xfs_ail *ailp,
38 xfs_log_item_t *lip)
39{
40 xfs_log_item_t *prev_lip;
41
42 if (list_empty(&ailp->xa_ail))
43 return;
44
45 /*
46 * Check the next and previous entries are valid.
47 */
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);
52
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);
56
57
58}
59#else /* !DEBUG */
60#define xfs_ail_check(a,l)
61#endif /* DEBUG */
62
63/*
64 * Return a pointer to the last item in the AIL. If the AIL is empty, then
65 * return NULL.
66 */
67static xfs_log_item_t *
68xfs_ail_max(
69 struct xfs_ail *ailp)
70{
71 if (list_empty(&ailp->xa_ail))
72 return NULL;
73
74 return list_entry(ailp->xa_ail.prev, xfs_log_item_t, li_ail);
75}
76
77/*
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.
80 */
81static xfs_log_item_t *
82xfs_ail_next(
83 struct xfs_ail *ailp,
84 xfs_log_item_t *lip)
85{
86 if (lip->li_ail.next == &ailp->xa_ail)
87 return NULL;
88
89 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
90}
91
92/*
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.
96 *
97 * We need the AIL lock in order to get a coherent read of the lsn of the last
98 * item in the AIL.
99 */
100xfs_lsn_t
101xfs_ail_min_lsn(
102 struct xfs_ail *ailp)
103{
104 xfs_lsn_t lsn = 0;
105 xfs_log_item_t *lip;
106
107 spin_lock(&ailp->xa_lock);
108 lip = xfs_ail_min(ailp);
109 if (lip)
110 lsn = lip->li_lsn;
111 spin_unlock(&ailp->xa_lock);
112
113 return lsn;
114}
115
116/*
117 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
118 */
119static xfs_lsn_t
120xfs_ail_max_lsn(
121 struct xfs_ail *ailp)
122{
123 xfs_lsn_t lsn = 0;
124 xfs_log_item_t *lip;
125
126 spin_lock(&ailp->xa_lock);
127 lip = xfs_ail_max(ailp);
128 if (lip)
129 lsn = lip->li_lsn;
130 spin_unlock(&ailp->xa_lock);
131
132 return lsn;
133}
134
135/*
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.
141 */
142STATIC void
143xfs_trans_ail_cursor_init(
144 struct xfs_ail *ailp,
145 struct xfs_ail_cursor *cur)
146{
147 cur->item = NULL;
148 list_add_tail(&cur->list, &ailp->xa_cursors);
149}
150
151/*
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.
154 */
155struct xfs_log_item *
156xfs_trans_ail_cursor_next(
157 struct xfs_ail *ailp,
158 struct xfs_ail_cursor *cur)
159{
160 struct xfs_log_item *lip = cur->item;
161
162 if ((uintptr_t)lip & 1)
163 lip = xfs_ail_min(ailp);
164 if (lip)
165 cur->item = xfs_ail_next(ailp, lip);
166 return lip;
167}
168
169/*
170 * When the traversal is complete, we need to remove the cursor from the list
171 * of traversing cursors.
172 */
173void
174xfs_trans_ail_cursor_done(
175 struct xfs_ail_cursor *cur)
176{
177 cur->item = NULL;
178 list_del_init(&cur->list);
179}
180
181/*
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.
188 */
189STATIC void
190xfs_trans_ail_cursor_clear(
191 struct xfs_ail *ailp,
192 struct xfs_log_item *lip)
193{
194 struct xfs_ail_cursor *cur;
195
196 list_for_each_entry(cur, &ailp->xa_cursors, list) {
197 if (cur->item == lip)
198 cur->item = (struct xfs_log_item *)
199 ((uintptr_t)cur->item | 1);
200 }
201}
202
203/*
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.
208 */
209xfs_log_item_t *
210xfs_trans_ail_cursor_first(
211 struct xfs_ail *ailp,
212 struct xfs_ail_cursor *cur,
213 xfs_lsn_t lsn)
214{
215 xfs_log_item_t *lip;
216
217 xfs_trans_ail_cursor_init(ailp, cur);
218
219 if (lsn == 0) {
220 lip = xfs_ail_min(ailp);
221 goto out;
222 }
223
224 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
225 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
226 goto out;
227 }
228 return NULL;
229
230out:
231 if (lip)
232 cur->item = xfs_ail_next(ailp, lip);
233 return lip;
234}
235
236static struct xfs_log_item *
237__xfs_trans_ail_cursor_last(
238 struct xfs_ail *ailp,
239 xfs_lsn_t lsn)
240{
241 xfs_log_item_t *lip;
242
243 list_for_each_entry_reverse(lip, &ailp->xa_ail, li_ail) {
244 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
245 return lip;
246 }
247 return NULL;
248}
249
250/*
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.
255 */
256struct xfs_log_item *
257xfs_trans_ail_cursor_last(
258 struct xfs_ail *ailp,
259 struct xfs_ail_cursor *cur,
260 xfs_lsn_t lsn)
261{
262 xfs_trans_ail_cursor_init(ailp, cur);
263 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
264 return cur->item;
265}
266
267/*
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.
272 */
273static void
274xfs_ail_splice(
275 struct xfs_ail *ailp,
276 struct xfs_ail_cursor *cur,
277 struct list_head *list,
278 xfs_lsn_t lsn)
279{
280 struct xfs_log_item *lip;
281
282 ASSERT(!list_empty(list));
283
284 /*
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.
288 */
289 lip = cur ? cur->item : NULL;
290 if (!lip || (uintptr_t)lip & 1)
291 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
292
293 /*
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.
299 */
300 if (cur)
301 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
302
303 /*
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.
308 */
309 if (lip)
310 list_splice(list, &lip->li_ail);
311 else
312 list_splice(list, &ailp->xa_ail);
313}
314
315/*
316 * Delete the given item from the AIL. Return a pointer to the item.
317 */
318static void
319xfs_ail_delete(
320 struct xfs_ail *ailp,
321 xfs_log_item_t *lip)
322{
323 xfs_ail_check(ailp, lip);
324 list_del(&lip->li_ail);
325 xfs_trans_ail_cursor_clear(ailp, lip);
326}
327
328static long
329xfsaild_push(
330 struct xfs_ail *ailp)
331{
332 xfs_mount_t *mp = ailp->xa_mount;
333 struct xfs_ail_cursor cur;
334 xfs_log_item_t *lip;
335 xfs_lsn_t lsn;
336 xfs_lsn_t target;
337 long tout;
338 int stuck = 0;
339 int flushing = 0;
340 int count = 0;
341
342 /*
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.
346 */
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;
351
352 XFS_STATS_INC(mp, xs_push_ail_flush);
353 xfs_log_force(mp, XFS_LOG_SYNC);
354 }
355
356 spin_lock(&ailp->xa_lock);
357
358 /* barrier matches the xa_target update in xfs_ail_push() */
359 smp_rmb();
360 target = ailp->xa_target;
361 ailp->xa_target_prev = target;
362
363 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->xa_last_pushed_lsn);
364 if (!lip) {
365 /*
366 * If the AIL is empty or our push has reached the end we are
367 * done now.
368 */
369 xfs_trans_ail_cursor_done(&cur);
370 spin_unlock(&ailp->xa_lock);
371 goto out_done;
372 }
373
374 XFS_STATS_INC(mp, xs_push_ail);
375
376 lsn = lip->li_lsn;
377 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
378 int lock_result;
379
380 /*
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
383 * the dropped lock.
384 */
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(mp, xs_push_ail_success);
389 trace_xfs_ail_push(lip);
390
391 ailp->xa_last_pushed_lsn = lsn;
392 break;
393
394 case XFS_ITEM_FLUSHING:
395 /*
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.
400 *
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.
405 */
406 XFS_STATS_INC(mp, xs_push_ail_flushing);
407 trace_xfs_ail_flushing(lip);
408
409 flushing++;
410 ailp->xa_last_pushed_lsn = lsn;
411 break;
412
413 case XFS_ITEM_PINNED:
414 XFS_STATS_INC(mp, xs_push_ail_pinned);
415 trace_xfs_ail_pinned(lip);
416
417 stuck++;
418 ailp->xa_log_flush++;
419 break;
420 case XFS_ITEM_LOCKED:
421 XFS_STATS_INC(mp, xs_push_ail_locked);
422 trace_xfs_ail_locked(lip);
423
424 stuck++;
425 break;
426 default:
427 ASSERT(0);
428 break;
429 }
430
431 count++;
432
433 /*
434 * Are there too many items we can't do anything with?
435 *
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.
442 *
443 * The value of 100 is an arbitrary magic number based on
444 * observation.
445 */
446 if (stuck > 100)
447 break;
448
449 lip = xfs_trans_ail_cursor_next(ailp, &cur);
450 if (lip == NULL)
451 break;
452 lsn = lip->li_lsn;
453 }
454 xfs_trans_ail_cursor_done(&cur);
455 spin_unlock(&ailp->xa_lock);
456
457 if (xfs_buf_delwri_submit_nowait(&ailp->xa_buf_list))
458 ailp->xa_log_flush++;
459
460 if (!count || XFS_LSN_CMP(lsn, target) >= 0) {
461out_done:
462 /*
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.
466 */
467 tout = 50;
468 ailp->xa_last_pushed_lsn = 0;
469 } else if (((stuck + flushing) * 100) / count > 90) {
470 /*
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.
474 *
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.
479 */
480 tout = 20;
481 ailp->xa_last_pushed_lsn = 0;
482 } else {
483 /*
484 * Assume we have more work to do in a short while.
485 */
486 tout = 10;
487 }
488
489 return tout;
490}
491
492static int
493xfsaild(
494 void *data)
495{
496 struct xfs_ail *ailp = data;
497 long tout = 0; /* milliseconds */
498
499 current->flags |= PF_MEMALLOC;
500 set_freezable();
501
502 while (!kthread_should_stop()) {
503 if (tout && tout <= 20)
504 __set_current_state(TASK_KILLABLE);
505 else
506 __set_current_state(TASK_INTERRUPTIBLE);
507
508 spin_lock(&ailp->xa_lock);
509
510 /*
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.
516 *
517 * The barrier matches the xa_target update in xfs_ail_push().
518 */
519 smp_rmb();
520 if (!xfs_ail_min(ailp) &&
521 ailp->xa_target == ailp->xa_target_prev) {
522 spin_unlock(&ailp->xa_lock);
523 freezable_schedule();
524 tout = 0;
525 continue;
526 }
527 spin_unlock(&ailp->xa_lock);
528
529 if (tout)
530 freezable_schedule_timeout(msecs_to_jiffies(tout));
531
532 __set_current_state(TASK_RUNNING);
533
534 try_to_freeze();
535
536 tout = xfsaild_push(ailp);
537 }
538
539 return 0;
540}
541
542/*
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
545 * threshold_lsn.
546 *
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.
551 *
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.
555 */
556void
557xfs_ail_push(
558 struct xfs_ail *ailp,
559 xfs_lsn_t threshold_lsn)
560{
561 xfs_log_item_t *lip;
562
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)
566 return;
567
568 /*
569 * Ensure that the new target is noticed in push code before it clears
570 * the XFS_AIL_PUSHING_BIT.
571 */
572 smp_wmb();
573 xfs_trans_ail_copy_lsn(ailp, &ailp->xa_target, &threshold_lsn);
574 smp_wmb();
575
576 wake_up_process(ailp->xa_task);
577}
578
579/*
580 * Push out all items in the AIL immediately
581 */
582void
583xfs_ail_push_all(
584 struct xfs_ail *ailp)
585{
586 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
587
588 if (threshold_lsn)
589 xfs_ail_push(ailp, threshold_lsn);
590}
591
592/*
593 * Push out all items in the AIL immediately and wait until the AIL is empty.
594 */
595void
596xfs_ail_push_all_sync(
597 struct xfs_ail *ailp)
598{
599 struct xfs_log_item *lip;
600 DEFINE_WAIT(wait);
601
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);
608 schedule();
609 spin_lock(&ailp->xa_lock);
610 }
611 spin_unlock(&ailp->xa_lock);
612
613 finish_wait(&ailp->xa_empty, &wait);
614}
615
616/*
617 * xfs_trans_ail_update - bulk AIL insertion operation.
618 *
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.
624 *
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.
629 *
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.
634 *
635 * This function must be called with the AIL lock held. The lock is dropped
636 * before returning.
637 */
638void
639xfs_trans_ail_update_bulk(
640 struct xfs_ail *ailp,
641 struct xfs_ail_cursor *cur,
642 struct xfs_log_item **log_items,
643 int nr_items,
644 xfs_lsn_t lsn) __releases(ailp->xa_lock)
645{
646 xfs_log_item_t *mlip;
647 int mlip_changed = 0;
648 int i;
649 LIST_HEAD(tmp);
650
651 ASSERT(nr_items > 0); /* Not required, but true. */
652 mlip = xfs_ail_min(ailp);
653
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)
659 continue;
660
661 trace_xfs_ail_move(lip, lip->li_lsn, lsn);
662 xfs_ail_delete(ailp, lip);
663 if (mlip == lip)
664 mlip_changed = 1;
665 } else {
666 lip->li_flags |= XFS_LI_IN_AIL;
667 trace_xfs_ail_insert(lip, 0, lsn);
668 }
669 lip->li_lsn = lsn;
670 list_add(&lip->li_ail, &tmp);
671 }
672
673 if (!list_empty(&tmp))
674 xfs_ail_splice(ailp, cur, &tmp, lsn);
675
676 if (mlip_changed) {
677 if (!XFS_FORCED_SHUTDOWN(ailp->xa_mount))
678 xlog_assign_tail_lsn_locked(ailp->xa_mount);
679 spin_unlock(&ailp->xa_lock);
680
681 xfs_log_space_wake(ailp->xa_mount);
682 } else {
683 spin_unlock(&ailp->xa_lock);
684 }
685}
686
687/*
688 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
689 *
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.
694 *
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
698 * AIL.
699 *
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.
704 *
705 * This function must be called with the AIL lock held. The lock is dropped
706 * before returning.
707 */
708void
709xfs_trans_ail_delete_bulk(
710 struct xfs_ail *ailp,
711 struct xfs_log_item **log_items,
712 int nr_items,
713 int shutdown_type) __releases(ailp->xa_lock)
714{
715 xfs_log_item_t *mlip;
716 int mlip_changed = 0;
717 int i;
718
719 mlip = xfs_ail_min(ailp);
720
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;
725
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",
730 __func__);
731 xfs_force_shutdown(mp, shutdown_type);
732 }
733 return;
734 }
735
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;
739 lip->li_lsn = 0;
740 if (mlip == lip)
741 mlip_changed = 1;
742 }
743
744 if (mlip_changed) {
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);
750
751 xfs_log_space_wake(ailp->xa_mount);
752 } else {
753 spin_unlock(&ailp->xa_lock);
754 }
755}
756
757int
758xfs_trans_ail_init(
759 xfs_mount_t *mp)
760{
761 struct xfs_ail *ailp;
762
763 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
764 if (!ailp)
765 return -ENOMEM;
766
767 ailp->xa_mount = mp;
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);
773
774 ailp->xa_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
775 ailp->xa_mount->m_fsname);
776 if (IS_ERR(ailp->xa_task))
777 goto out_free_ailp;
778
779 mp->m_ail = ailp;
780 return 0;
781
782out_free_ailp:
783 kmem_free(ailp);
784 return -ENOMEM;
785}
786
787void
788xfs_trans_ail_destroy(
789 xfs_mount_t *mp)
790{
791 struct xfs_ail *ailp = mp->m_ail;
792
793 kthread_stop(ailp->xa_task);
794 kmem_free(ailp);
795}
1/*
2 * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
3 * Copyright (c) 2008 Dave Chinner
4 * All Rights Reserved.
5 *
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.
9 *
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.
14 *
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
18 */
19#include "xfs.h"
20#include "xfs_fs.h"
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_errortag.h"
29#include "xfs_error.h"
30#include "xfs_log.h"
31
32#ifdef DEBUG
33/*
34 * Check that the list is sorted as it should be.
35 */
36STATIC void
37xfs_ail_check(
38 struct xfs_ail *ailp,
39 xfs_log_item_t *lip)
40{
41 xfs_log_item_t *prev_lip;
42
43 if (list_empty(&ailp->ail_head))
44 return;
45
46 /*
47 * Check the next and previous entries are valid.
48 */
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->ail_head)
52 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
53
54 prev_lip = list_entry(lip->li_ail.next, xfs_log_item_t, li_ail);
55 if (&prev_lip->li_ail != &ailp->ail_head)
56 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) >= 0);
57
58
59}
60#else /* !DEBUG */
61#define xfs_ail_check(a,l)
62#endif /* DEBUG */
63
64/*
65 * Return a pointer to the last item in the AIL. If the AIL is empty, then
66 * return NULL.
67 */
68static xfs_log_item_t *
69xfs_ail_max(
70 struct xfs_ail *ailp)
71{
72 if (list_empty(&ailp->ail_head))
73 return NULL;
74
75 return list_entry(ailp->ail_head.prev, xfs_log_item_t, li_ail);
76}
77
78/*
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.
81 */
82static xfs_log_item_t *
83xfs_ail_next(
84 struct xfs_ail *ailp,
85 xfs_log_item_t *lip)
86{
87 if (lip->li_ail.next == &ailp->ail_head)
88 return NULL;
89
90 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
91}
92
93/*
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.
97 *
98 * We need the AIL lock in order to get a coherent read of the lsn of the last
99 * item in the AIL.
100 */
101xfs_lsn_t
102xfs_ail_min_lsn(
103 struct xfs_ail *ailp)
104{
105 xfs_lsn_t lsn = 0;
106 xfs_log_item_t *lip;
107
108 spin_lock(&ailp->ail_lock);
109 lip = xfs_ail_min(ailp);
110 if (lip)
111 lsn = lip->li_lsn;
112 spin_unlock(&ailp->ail_lock);
113
114 return lsn;
115}
116
117/*
118 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
119 */
120static xfs_lsn_t
121xfs_ail_max_lsn(
122 struct xfs_ail *ailp)
123{
124 xfs_lsn_t lsn = 0;
125 xfs_log_item_t *lip;
126
127 spin_lock(&ailp->ail_lock);
128 lip = xfs_ail_max(ailp);
129 if (lip)
130 lsn = lip->li_lsn;
131 spin_unlock(&ailp->ail_lock);
132
133 return lsn;
134}
135
136/*
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.
142 */
143STATIC void
144xfs_trans_ail_cursor_init(
145 struct xfs_ail *ailp,
146 struct xfs_ail_cursor *cur)
147{
148 cur->item = NULL;
149 list_add_tail(&cur->list, &ailp->ail_cursors);
150}
151
152/*
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.
155 */
156struct xfs_log_item *
157xfs_trans_ail_cursor_next(
158 struct xfs_ail *ailp,
159 struct xfs_ail_cursor *cur)
160{
161 struct xfs_log_item *lip = cur->item;
162
163 if ((uintptr_t)lip & 1)
164 lip = xfs_ail_min(ailp);
165 if (lip)
166 cur->item = xfs_ail_next(ailp, lip);
167 return lip;
168}
169
170/*
171 * When the traversal is complete, we need to remove the cursor from the list
172 * of traversing cursors.
173 */
174void
175xfs_trans_ail_cursor_done(
176 struct xfs_ail_cursor *cur)
177{
178 cur->item = NULL;
179 list_del_init(&cur->list);
180}
181
182/*
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.
189 */
190STATIC void
191xfs_trans_ail_cursor_clear(
192 struct xfs_ail *ailp,
193 struct xfs_log_item *lip)
194{
195 struct xfs_ail_cursor *cur;
196
197 list_for_each_entry(cur, &ailp->ail_cursors, list) {
198 if (cur->item == lip)
199 cur->item = (struct xfs_log_item *)
200 ((uintptr_t)cur->item | 1);
201 }
202}
203
204/*
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.
209 */
210xfs_log_item_t *
211xfs_trans_ail_cursor_first(
212 struct xfs_ail *ailp,
213 struct xfs_ail_cursor *cur,
214 xfs_lsn_t lsn)
215{
216 xfs_log_item_t *lip;
217
218 xfs_trans_ail_cursor_init(ailp, cur);
219
220 if (lsn == 0) {
221 lip = xfs_ail_min(ailp);
222 goto out;
223 }
224
225 list_for_each_entry(lip, &ailp->ail_head, li_ail) {
226 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
227 goto out;
228 }
229 return NULL;
230
231out:
232 if (lip)
233 cur->item = xfs_ail_next(ailp, lip);
234 return lip;
235}
236
237static struct xfs_log_item *
238__xfs_trans_ail_cursor_last(
239 struct xfs_ail *ailp,
240 xfs_lsn_t lsn)
241{
242 xfs_log_item_t *lip;
243
244 list_for_each_entry_reverse(lip, &ailp->ail_head, li_ail) {
245 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
246 return lip;
247 }
248 return NULL;
249}
250
251/*
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.
256 */
257struct xfs_log_item *
258xfs_trans_ail_cursor_last(
259 struct xfs_ail *ailp,
260 struct xfs_ail_cursor *cur,
261 xfs_lsn_t lsn)
262{
263 xfs_trans_ail_cursor_init(ailp, cur);
264 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
265 return cur->item;
266}
267
268/*
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.
273 */
274static void
275xfs_ail_splice(
276 struct xfs_ail *ailp,
277 struct xfs_ail_cursor *cur,
278 struct list_head *list,
279 xfs_lsn_t lsn)
280{
281 struct xfs_log_item *lip;
282
283 ASSERT(!list_empty(list));
284
285 /*
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.
289 */
290 lip = cur ? cur->item : NULL;
291 if (!lip || (uintptr_t)lip & 1)
292 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
293
294 /*
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.
300 */
301 if (cur)
302 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
303
304 /*
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.
309 */
310 if (lip)
311 list_splice(list, &lip->li_ail);
312 else
313 list_splice(list, &ailp->ail_head);
314}
315
316/*
317 * Delete the given item from the AIL. Return a pointer to the item.
318 */
319static void
320xfs_ail_delete(
321 struct xfs_ail *ailp,
322 xfs_log_item_t *lip)
323{
324 xfs_ail_check(ailp, lip);
325 list_del(&lip->li_ail);
326 xfs_trans_ail_cursor_clear(ailp, lip);
327}
328
329static inline uint
330xfsaild_push_item(
331 struct xfs_ail *ailp,
332 struct xfs_log_item *lip)
333{
334 /*
335 * If log item pinning is enabled, skip the push and track the item as
336 * pinned. This can help induce head-behind-tail conditions.
337 */
338 if (XFS_TEST_ERROR(false, ailp->ail_mount, XFS_ERRTAG_LOG_ITEM_PIN))
339 return XFS_ITEM_PINNED;
340
341 return lip->li_ops->iop_push(lip, &ailp->ail_buf_list);
342}
343
344static long
345xfsaild_push(
346 struct xfs_ail *ailp)
347{
348 xfs_mount_t *mp = ailp->ail_mount;
349 struct xfs_ail_cursor cur;
350 xfs_log_item_t *lip;
351 xfs_lsn_t lsn;
352 xfs_lsn_t target;
353 long tout;
354 int stuck = 0;
355 int flushing = 0;
356 int count = 0;
357
358 /*
359 * If we encountered pinned items or did not finish writing out all
360 * buffers the last time we ran, force the log first and wait for it
361 * before pushing again.
362 */
363 if (ailp->ail_log_flush && ailp->ail_last_pushed_lsn == 0 &&
364 (!list_empty_careful(&ailp->ail_buf_list) ||
365 xfs_ail_min_lsn(ailp))) {
366 ailp->ail_log_flush = 0;
367
368 XFS_STATS_INC(mp, xs_push_ail_flush);
369 xfs_log_force(mp, XFS_LOG_SYNC);
370 }
371
372 spin_lock(&ailp->ail_lock);
373
374 /* barrier matches the ail_target update in xfs_ail_push() */
375 smp_rmb();
376 target = ailp->ail_target;
377 ailp->ail_target_prev = target;
378
379 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->ail_last_pushed_lsn);
380 if (!lip) {
381 /*
382 * If the AIL is empty or our push has reached the end we are
383 * done now.
384 */
385 xfs_trans_ail_cursor_done(&cur);
386 spin_unlock(&ailp->ail_lock);
387 goto out_done;
388 }
389
390 XFS_STATS_INC(mp, xs_push_ail);
391
392 lsn = lip->li_lsn;
393 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
394 int lock_result;
395
396 /*
397 * Note that iop_push may unlock and reacquire the AIL lock. We
398 * rely on the AIL cursor implementation to be able to deal with
399 * the dropped lock.
400 */
401 lock_result = xfsaild_push_item(ailp, lip);
402 switch (lock_result) {
403 case XFS_ITEM_SUCCESS:
404 XFS_STATS_INC(mp, xs_push_ail_success);
405 trace_xfs_ail_push(lip);
406
407 ailp->ail_last_pushed_lsn = lsn;
408 break;
409
410 case XFS_ITEM_FLUSHING:
411 /*
412 * The item or its backing buffer is already beeing
413 * flushed. The typical reason for that is that an
414 * inode buffer is locked because we already pushed the
415 * updates to it as part of inode clustering.
416 *
417 * We do not want to to stop flushing just because lots
418 * of items are already beeing flushed, but we need to
419 * re-try the flushing relatively soon if most of the
420 * AIL is beeing flushed.
421 */
422 XFS_STATS_INC(mp, xs_push_ail_flushing);
423 trace_xfs_ail_flushing(lip);
424
425 flushing++;
426 ailp->ail_last_pushed_lsn = lsn;
427 break;
428
429 case XFS_ITEM_PINNED:
430 XFS_STATS_INC(mp, xs_push_ail_pinned);
431 trace_xfs_ail_pinned(lip);
432
433 stuck++;
434 ailp->ail_log_flush++;
435 break;
436 case XFS_ITEM_LOCKED:
437 XFS_STATS_INC(mp, xs_push_ail_locked);
438 trace_xfs_ail_locked(lip);
439
440 stuck++;
441 break;
442 default:
443 ASSERT(0);
444 break;
445 }
446
447 count++;
448
449 /*
450 * Are there too many items we can't do anything with?
451 *
452 * If we we are skipping too many items because we can't flush
453 * them or they are already being flushed, we back off and
454 * given them time to complete whatever operation is being
455 * done. i.e. remove pressure from the AIL while we can't make
456 * progress so traversals don't slow down further inserts and
457 * removals to/from the AIL.
458 *
459 * The value of 100 is an arbitrary magic number based on
460 * observation.
461 */
462 if (stuck > 100)
463 break;
464
465 lip = xfs_trans_ail_cursor_next(ailp, &cur);
466 if (lip == NULL)
467 break;
468 lsn = lip->li_lsn;
469 }
470 xfs_trans_ail_cursor_done(&cur);
471 spin_unlock(&ailp->ail_lock);
472
473 if (xfs_buf_delwri_submit_nowait(&ailp->ail_buf_list))
474 ailp->ail_log_flush++;
475
476 if (!count || XFS_LSN_CMP(lsn, target) >= 0) {
477out_done:
478 /*
479 * We reached the target or the AIL is empty, so wait a bit
480 * longer for I/O to complete and remove pushed items from the
481 * AIL before we start the next scan from the start of the AIL.
482 */
483 tout = 50;
484 ailp->ail_last_pushed_lsn = 0;
485 } else if (((stuck + flushing) * 100) / count > 90) {
486 /*
487 * Either there is a lot of contention on the AIL or we are
488 * stuck due to operations in progress. "Stuck" in this case
489 * is defined as >90% of the items we tried to push were stuck.
490 *
491 * Backoff a bit more to allow some I/O to complete before
492 * restarting from the start of the AIL. This prevents us from
493 * spinning on the same items, and if they are pinned will all
494 * the restart to issue a log force to unpin the stuck items.
495 */
496 tout = 20;
497 ailp->ail_last_pushed_lsn = 0;
498 } else {
499 /*
500 * Assume we have more work to do in a short while.
501 */
502 tout = 10;
503 }
504
505 return tout;
506}
507
508static int
509xfsaild(
510 void *data)
511{
512 struct xfs_ail *ailp = data;
513 long tout = 0; /* milliseconds */
514
515 current->flags |= PF_MEMALLOC;
516 set_freezable();
517
518 while (1) {
519 if (tout && tout <= 20)
520 set_current_state(TASK_KILLABLE);
521 else
522 set_current_state(TASK_INTERRUPTIBLE);
523
524 /*
525 * Check kthread_should_stop() after we set the task state
526 * to guarantee that we either see the stop bit and exit or
527 * the task state is reset to runnable such that it's not
528 * scheduled out indefinitely and detects the stop bit at
529 * next iteration.
530 *
531 * A memory barrier is included in above task state set to
532 * serialize again kthread_stop().
533 */
534 if (kthread_should_stop()) {
535 __set_current_state(TASK_RUNNING);
536 break;
537 }
538
539 spin_lock(&ailp->ail_lock);
540
541 /*
542 * Idle if the AIL is empty and we are not racing with a target
543 * update. We check the AIL after we set the task to a sleep
544 * state to guarantee that we either catch an ail_target update
545 * or that a wake_up resets the state to TASK_RUNNING.
546 * Otherwise, we run the risk of sleeping indefinitely.
547 *
548 * The barrier matches the ail_target update in xfs_ail_push().
549 */
550 smp_rmb();
551 if (!xfs_ail_min(ailp) &&
552 ailp->ail_target == ailp->ail_target_prev) {
553 spin_unlock(&ailp->ail_lock);
554 freezable_schedule();
555 tout = 0;
556 continue;
557 }
558 spin_unlock(&ailp->ail_lock);
559
560 if (tout)
561 freezable_schedule_timeout(msecs_to_jiffies(tout));
562
563 __set_current_state(TASK_RUNNING);
564
565 try_to_freeze();
566
567 tout = xfsaild_push(ailp);
568 }
569
570 return 0;
571}
572
573/*
574 * This routine is called to move the tail of the AIL forward. It does this by
575 * trying to flush items in the AIL whose lsns are below the given
576 * threshold_lsn.
577 *
578 * The push is run asynchronously in a workqueue, which means the caller needs
579 * to handle waiting on the async flush for space to become available.
580 * We don't want to interrupt any push that is in progress, hence we only queue
581 * work if we set the pushing bit approriately.
582 *
583 * We do this unlocked - we only need to know whether there is anything in the
584 * AIL at the time we are called. We don't need to access the contents of
585 * any of the objects, so the lock is not needed.
586 */
587void
588xfs_ail_push(
589 struct xfs_ail *ailp,
590 xfs_lsn_t threshold_lsn)
591{
592 xfs_log_item_t *lip;
593
594 lip = xfs_ail_min(ailp);
595 if (!lip || XFS_FORCED_SHUTDOWN(ailp->ail_mount) ||
596 XFS_LSN_CMP(threshold_lsn, ailp->ail_target) <= 0)
597 return;
598
599 /*
600 * Ensure that the new target is noticed in push code before it clears
601 * the XFS_AIL_PUSHING_BIT.
602 */
603 smp_wmb();
604 xfs_trans_ail_copy_lsn(ailp, &ailp->ail_target, &threshold_lsn);
605 smp_wmb();
606
607 wake_up_process(ailp->ail_task);
608}
609
610/*
611 * Push out all items in the AIL immediately
612 */
613void
614xfs_ail_push_all(
615 struct xfs_ail *ailp)
616{
617 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
618
619 if (threshold_lsn)
620 xfs_ail_push(ailp, threshold_lsn);
621}
622
623/*
624 * Push out all items in the AIL immediately and wait until the AIL is empty.
625 */
626void
627xfs_ail_push_all_sync(
628 struct xfs_ail *ailp)
629{
630 struct xfs_log_item *lip;
631 DEFINE_WAIT(wait);
632
633 spin_lock(&ailp->ail_lock);
634 while ((lip = xfs_ail_max(ailp)) != NULL) {
635 prepare_to_wait(&ailp->ail_empty, &wait, TASK_UNINTERRUPTIBLE);
636 ailp->ail_target = lip->li_lsn;
637 wake_up_process(ailp->ail_task);
638 spin_unlock(&ailp->ail_lock);
639 schedule();
640 spin_lock(&ailp->ail_lock);
641 }
642 spin_unlock(&ailp->ail_lock);
643
644 finish_wait(&ailp->ail_empty, &wait);
645}
646
647/*
648 * xfs_trans_ail_update - bulk AIL insertion operation.
649 *
650 * @xfs_trans_ail_update takes an array of log items that all need to be
651 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
652 * be added. Otherwise, it will be repositioned by removing it and re-adding
653 * it to the AIL. If we move the first item in the AIL, update the log tail to
654 * match the new minimum LSN in the AIL.
655 *
656 * This function takes the AIL lock once to execute the update operations on
657 * all the items in the array, and as such should not be called with the AIL
658 * lock held. As a result, once we have the AIL lock, we need to check each log
659 * item LSN to confirm it needs to be moved forward in the AIL.
660 *
661 * To optimise the insert operation, we delete all the items from the AIL in
662 * the first pass, moving them into a temporary list, then splice the temporary
663 * list into the correct position in the AIL. This avoids needing to do an
664 * insert operation on every item.
665 *
666 * This function must be called with the AIL lock held. The lock is dropped
667 * before returning.
668 */
669void
670xfs_trans_ail_update_bulk(
671 struct xfs_ail *ailp,
672 struct xfs_ail_cursor *cur,
673 struct xfs_log_item **log_items,
674 int nr_items,
675 xfs_lsn_t lsn) __releases(ailp->ail_lock)
676{
677 xfs_log_item_t *mlip;
678 int mlip_changed = 0;
679 int i;
680 LIST_HEAD(tmp);
681
682 ASSERT(nr_items > 0); /* Not required, but true. */
683 mlip = xfs_ail_min(ailp);
684
685 for (i = 0; i < nr_items; i++) {
686 struct xfs_log_item *lip = log_items[i];
687 if (lip->li_flags & XFS_LI_IN_AIL) {
688 /* check if we really need to move the item */
689 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
690 continue;
691
692 trace_xfs_ail_move(lip, lip->li_lsn, lsn);
693 xfs_ail_delete(ailp, lip);
694 if (mlip == lip)
695 mlip_changed = 1;
696 } else {
697 lip->li_flags |= XFS_LI_IN_AIL;
698 trace_xfs_ail_insert(lip, 0, lsn);
699 }
700 lip->li_lsn = lsn;
701 list_add(&lip->li_ail, &tmp);
702 }
703
704 if (!list_empty(&tmp))
705 xfs_ail_splice(ailp, cur, &tmp, lsn);
706
707 if (mlip_changed) {
708 if (!XFS_FORCED_SHUTDOWN(ailp->ail_mount))
709 xlog_assign_tail_lsn_locked(ailp->ail_mount);
710 spin_unlock(&ailp->ail_lock);
711
712 xfs_log_space_wake(ailp->ail_mount);
713 } else {
714 spin_unlock(&ailp->ail_lock);
715 }
716}
717
718bool
719xfs_ail_delete_one(
720 struct xfs_ail *ailp,
721 struct xfs_log_item *lip)
722{
723 struct xfs_log_item *mlip = xfs_ail_min(ailp);
724
725 trace_xfs_ail_delete(lip, mlip->li_lsn, lip->li_lsn);
726 xfs_ail_delete(ailp, lip);
727 xfs_clear_li_failed(lip);
728 lip->li_flags &= ~XFS_LI_IN_AIL;
729 lip->li_lsn = 0;
730
731 return mlip == lip;
732}
733
734/**
735 * Remove a log items from the AIL
736 *
737 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
738 * removed from the AIL. The caller is already holding the AIL lock, and done
739 * all the checks necessary to ensure the items passed in via @log_items are
740 * ready for deletion. This includes checking that the items are in the AIL.
741 *
742 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
743 * flag from the item and reset the item's lsn to 0. If we remove the first
744 * item in the AIL, update the log tail to match the new minimum LSN in the
745 * AIL.
746 *
747 * This function will not drop the AIL lock until all items are removed from
748 * the AIL to minimise the amount of lock traffic on the AIL. This does not
749 * greatly increase the AIL hold time, but does significantly reduce the amount
750 * of traffic on the lock, especially during IO completion.
751 *
752 * This function must be called with the AIL lock held. The lock is dropped
753 * before returning.
754 */
755void
756xfs_trans_ail_delete(
757 struct xfs_ail *ailp,
758 struct xfs_log_item *lip,
759 int shutdown_type) __releases(ailp->ail_lock)
760{
761 struct xfs_mount *mp = ailp->ail_mount;
762 bool mlip_changed;
763
764 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
765 spin_unlock(&ailp->ail_lock);
766 if (!XFS_FORCED_SHUTDOWN(mp)) {
767 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
768 "%s: attempting to delete a log item that is not in the AIL",
769 __func__);
770 xfs_force_shutdown(mp, shutdown_type);
771 }
772 return;
773 }
774
775 mlip_changed = xfs_ail_delete_one(ailp, lip);
776 if (mlip_changed) {
777 if (!XFS_FORCED_SHUTDOWN(mp))
778 xlog_assign_tail_lsn_locked(mp);
779 if (list_empty(&ailp->ail_head))
780 wake_up_all(&ailp->ail_empty);
781 }
782
783 spin_unlock(&ailp->ail_lock);
784 if (mlip_changed)
785 xfs_log_space_wake(ailp->ail_mount);
786}
787
788int
789xfs_trans_ail_init(
790 xfs_mount_t *mp)
791{
792 struct xfs_ail *ailp;
793
794 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
795 if (!ailp)
796 return -ENOMEM;
797
798 ailp->ail_mount = mp;
799 INIT_LIST_HEAD(&ailp->ail_head);
800 INIT_LIST_HEAD(&ailp->ail_cursors);
801 spin_lock_init(&ailp->ail_lock);
802 INIT_LIST_HEAD(&ailp->ail_buf_list);
803 init_waitqueue_head(&ailp->ail_empty);
804
805 ailp->ail_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
806 ailp->ail_mount->m_fsname);
807 if (IS_ERR(ailp->ail_task))
808 goto out_free_ailp;
809
810 mp->m_ail = ailp;
811 return 0;
812
813out_free_ailp:
814 kmem_free(ailp);
815 return -ENOMEM;
816}
817
818void
819xfs_trans_ail_destroy(
820 xfs_mount_t *mp)
821{
822 struct xfs_ail *ailp = mp->m_ail;
823
824 kthread_stop(ailp->ail_task);
825 kmem_free(ailp);
826}