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_types.h"
22#include "xfs_log.h"
23#include "xfs_inum.h"
24#include "xfs_trans.h"
25#include "xfs_sb.h"
26#include "xfs_ag.h"
27#include "xfs_mount.h"
28#include "xfs_trans_priv.h"
29#include "xfs_error.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#ifdef XFS_TRANS_DEBUG
59 /*
60 * Walk the list checking lsn ordering, and that every entry has the
61 * XFS_LI_IN_AIL flag set. This is really expensive, so only do it
62 * when specifically debugging the transaction subsystem.
63 */
64 prev_lip = list_entry(&ailp->xa_ail, xfs_log_item_t, li_ail);
65 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
66 if (&prev_lip->li_ail != &ailp->xa_ail)
67 ASSERT(XFS_LSN_CMP(prev_lip->li_lsn, lip->li_lsn) <= 0);
68 ASSERT((lip->li_flags & XFS_LI_IN_AIL) != 0);
69 prev_lip = lip;
70 }
71#endif /* XFS_TRANS_DEBUG */
72}
73#else /* !DEBUG */
74#define xfs_ail_check(a,l)
75#endif /* DEBUG */
76
77/*
78 * Return a pointer to the first item in the AIL. If the AIL is empty, then
79 * return NULL.
80 */
81static xfs_log_item_t *
82xfs_ail_min(
83 struct xfs_ail *ailp)
84{
85 if (list_empty(&ailp->xa_ail))
86 return NULL;
87
88 return list_first_entry(&ailp->xa_ail, xfs_log_item_t, li_ail);
89}
90
91 /*
92 * Return a pointer to the last item in the AIL. If the AIL is empty, then
93 * return NULL.
94 */
95static xfs_log_item_t *
96xfs_ail_max(
97 struct xfs_ail *ailp)
98{
99 if (list_empty(&ailp->xa_ail))
100 return NULL;
101
102 return list_entry(ailp->xa_ail.prev, xfs_log_item_t, li_ail);
103}
104
105/*
106 * Return a pointer to the item which follows the given item in the AIL. If
107 * the given item is the last item in the list, then return NULL.
108 */
109static xfs_log_item_t *
110xfs_ail_next(
111 struct xfs_ail *ailp,
112 xfs_log_item_t *lip)
113{
114 if (lip->li_ail.next == &ailp->xa_ail)
115 return NULL;
116
117 return list_first_entry(&lip->li_ail, xfs_log_item_t, li_ail);
118}
119
120/*
121 * This is called by the log manager code to determine the LSN of the tail of
122 * the log. This is exactly the LSN of the first item in the AIL. If the AIL
123 * is empty, then this function returns 0.
124 *
125 * We need the AIL lock in order to get a coherent read of the lsn of the last
126 * item in the AIL.
127 */
128xfs_lsn_t
129xfs_ail_min_lsn(
130 struct xfs_ail *ailp)
131{
132 xfs_lsn_t lsn = 0;
133 xfs_log_item_t *lip;
134
135 spin_lock(&ailp->xa_lock);
136 lip = xfs_ail_min(ailp);
137 if (lip)
138 lsn = lip->li_lsn;
139 spin_unlock(&ailp->xa_lock);
140
141 return lsn;
142}
143
144/*
145 * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
146 */
147static xfs_lsn_t
148xfs_ail_max_lsn(
149 struct xfs_ail *ailp)
150{
151 xfs_lsn_t lsn = 0;
152 xfs_log_item_t *lip;
153
154 spin_lock(&ailp->xa_lock);
155 lip = xfs_ail_max(ailp);
156 if (lip)
157 lsn = lip->li_lsn;
158 spin_unlock(&ailp->xa_lock);
159
160 return lsn;
161}
162
163/*
164 * The cursor keeps track of where our current traversal is up to by tracking
165 * the next item in the list for us. However, for this to be safe, removing an
166 * object from the AIL needs to invalidate any cursor that points to it. hence
167 * the traversal cursor needs to be linked to the struct xfs_ail so that
168 * deletion can search all the active cursors for invalidation.
169 */
170STATIC void
171xfs_trans_ail_cursor_init(
172 struct xfs_ail *ailp,
173 struct xfs_ail_cursor *cur)
174{
175 cur->item = NULL;
176 list_add_tail(&cur->list, &ailp->xa_cursors);
177}
178
179/*
180 * Get the next item in the traversal and advance the cursor. If the cursor
181 * was invalidated (indicated by a lip of 1), restart the traversal.
182 */
183struct xfs_log_item *
184xfs_trans_ail_cursor_next(
185 struct xfs_ail *ailp,
186 struct xfs_ail_cursor *cur)
187{
188 struct xfs_log_item *lip = cur->item;
189
190 if ((__psint_t)lip & 1)
191 lip = xfs_ail_min(ailp);
192 if (lip)
193 cur->item = xfs_ail_next(ailp, lip);
194 return lip;
195}
196
197/*
198 * When the traversal is complete, we need to remove the cursor from the list
199 * of traversing cursors.
200 */
201void
202xfs_trans_ail_cursor_done(
203 struct xfs_ail *ailp,
204 struct xfs_ail_cursor *cur)
205{
206 cur->item = NULL;
207 list_del_init(&cur->list);
208}
209
210/*
211 * Invalidate any cursor that is pointing to this item. This is called when an
212 * item is removed from the AIL. Any cursor pointing to this object is now
213 * invalid and the traversal needs to be terminated so it doesn't reference a
214 * freed object. We set the low bit of the cursor item pointer so we can
215 * distinguish between an invalidation and the end of the list when getting the
216 * next item from the cursor.
217 */
218STATIC void
219xfs_trans_ail_cursor_clear(
220 struct xfs_ail *ailp,
221 struct xfs_log_item *lip)
222{
223 struct xfs_ail_cursor *cur;
224
225 list_for_each_entry(cur, &ailp->xa_cursors, list) {
226 if (cur->item == lip)
227 cur->item = (struct xfs_log_item *)
228 ((__psint_t)cur->item | 1);
229 }
230}
231
232/*
233 * Find the first item in the AIL with the given @lsn by searching in ascending
234 * LSN order and initialise the cursor to point to the next item for a
235 * ascending traversal. Pass a @lsn of zero to initialise the cursor to the
236 * first item in the AIL. Returns NULL if the list is empty.
237 */
238xfs_log_item_t *
239xfs_trans_ail_cursor_first(
240 struct xfs_ail *ailp,
241 struct xfs_ail_cursor *cur,
242 xfs_lsn_t lsn)
243{
244 xfs_log_item_t *lip;
245
246 xfs_trans_ail_cursor_init(ailp, cur);
247
248 if (lsn == 0) {
249 lip = xfs_ail_min(ailp);
250 goto out;
251 }
252
253 list_for_each_entry(lip, &ailp->xa_ail, li_ail) {
254 if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
255 goto out;
256 }
257 return NULL;
258
259out:
260 if (lip)
261 cur->item = xfs_ail_next(ailp, lip);
262 return lip;
263}
264
265static struct xfs_log_item *
266__xfs_trans_ail_cursor_last(
267 struct xfs_ail *ailp,
268 xfs_lsn_t lsn)
269{
270 xfs_log_item_t *lip;
271
272 list_for_each_entry_reverse(lip, &ailp->xa_ail, li_ail) {
273 if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
274 return lip;
275 }
276 return NULL;
277}
278
279/*
280 * Find the last item in the AIL with the given @lsn by searching in descending
281 * LSN order and initialise the cursor to point to that item. If there is no
282 * item with the value of @lsn, then it sets the cursor to the last item with an
283 * LSN lower than @lsn. Returns NULL if the list is empty.
284 */
285struct xfs_log_item *
286xfs_trans_ail_cursor_last(
287 struct xfs_ail *ailp,
288 struct xfs_ail_cursor *cur,
289 xfs_lsn_t lsn)
290{
291 xfs_trans_ail_cursor_init(ailp, cur);
292 cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
293 return cur->item;
294}
295
296/*
297 * Splice the log item list into the AIL at the given LSN. We splice to the
298 * tail of the given LSN to maintain insert order for push traversals. The
299 * cursor is optional, allowing repeated updates to the same LSN to avoid
300 * repeated traversals. This should not be called with an empty list.
301 */
302static void
303xfs_ail_splice(
304 struct xfs_ail *ailp,
305 struct xfs_ail_cursor *cur,
306 struct list_head *list,
307 xfs_lsn_t lsn)
308{
309 struct xfs_log_item *lip;
310
311 ASSERT(!list_empty(list));
312
313 /*
314 * Use the cursor to determine the insertion point if one is
315 * provided. If not, or if the one we got is not valid,
316 * find the place in the AIL where the items belong.
317 */
318 lip = cur ? cur->item : NULL;
319 if (!lip || (__psint_t) lip & 1)
320 lip = __xfs_trans_ail_cursor_last(ailp, lsn);
321
322 /*
323 * If a cursor is provided, we know we're processing the AIL
324 * in lsn order, and future items to be spliced in will
325 * follow the last one being inserted now. Update the
326 * cursor to point to that last item, now while we have a
327 * reliable pointer to it.
328 */
329 if (cur)
330 cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
331
332 /*
333 * Finally perform the splice. Unless the AIL was empty,
334 * lip points to the item in the AIL _after_ which the new
335 * items should go. If lip is null the AIL was empty, so
336 * the new items go at the head of the AIL.
337 */
338 if (lip)
339 list_splice(list, &lip->li_ail);
340 else
341 list_splice(list, &ailp->xa_ail);
342}
343
344/*
345 * Delete the given item from the AIL. Return a pointer to the item.
346 */
347static void
348xfs_ail_delete(
349 struct xfs_ail *ailp,
350 xfs_log_item_t *lip)
351{
352 xfs_ail_check(ailp, lip);
353 list_del(&lip->li_ail);
354 xfs_trans_ail_cursor_clear(ailp, lip);
355}
356
357static long
358xfsaild_push(
359 struct xfs_ail *ailp)
360{
361 xfs_mount_t *mp = ailp->xa_mount;
362 struct xfs_ail_cursor cur;
363 xfs_log_item_t *lip;
364 xfs_lsn_t lsn;
365 xfs_lsn_t target;
366 long tout = 10;
367 int flush_log = 0;
368 int stuck = 0;
369 int count = 0;
370 int push_xfsbufd = 0;
371
372 spin_lock(&ailp->xa_lock);
373 target = ailp->xa_target;
374 lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->xa_last_pushed_lsn);
375 if (!lip || XFS_FORCED_SHUTDOWN(mp)) {
376 /*
377 * AIL is empty or our push has reached the end.
378 */
379 xfs_trans_ail_cursor_done(ailp, &cur);
380 spin_unlock(&ailp->xa_lock);
381 goto out_done;
382 }
383
384 XFS_STATS_INC(xs_push_ail);
385
386 /*
387 * While the item we are looking at is below the given threshold
388 * try to flush it out. We'd like not to stop until we've at least
389 * tried to push on everything in the AIL with an LSN less than
390 * the given threshold.
391 *
392 * However, we will stop after a certain number of pushes and wait
393 * for a reduced timeout to fire before pushing further. This
394 * prevents use from spinning when we can't do anything or there is
395 * lots of contention on the AIL lists.
396 */
397 lsn = lip->li_lsn;
398 while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
399 int lock_result;
400 /*
401 * If we can lock the item without sleeping, unlock the AIL
402 * lock and flush the item. Then re-grab the AIL lock so we
403 * can look for the next item on the AIL. List changes are
404 * handled by the AIL lookup functions internally
405 *
406 * If we can't lock the item, either its holder will flush it
407 * or it is already being flushed or it is being relogged. In
408 * any of these case it is being taken care of and we can just
409 * skip to the next item in the list.
410 */
411 lock_result = IOP_TRYLOCK(lip);
412 spin_unlock(&ailp->xa_lock);
413 switch (lock_result) {
414 case XFS_ITEM_SUCCESS:
415 XFS_STATS_INC(xs_push_ail_success);
416 IOP_PUSH(lip);
417 ailp->xa_last_pushed_lsn = lsn;
418 break;
419
420 case XFS_ITEM_PUSHBUF:
421 XFS_STATS_INC(xs_push_ail_pushbuf);
422
423 if (!IOP_PUSHBUF(lip)) {
424 stuck++;
425 flush_log = 1;
426 } else {
427 ailp->xa_last_pushed_lsn = lsn;
428 }
429 push_xfsbufd = 1;
430 break;
431
432 case XFS_ITEM_PINNED:
433 XFS_STATS_INC(xs_push_ail_pinned);
434 stuck++;
435 flush_log = 1;
436 break;
437
438 case XFS_ITEM_LOCKED:
439 XFS_STATS_INC(xs_push_ail_locked);
440 stuck++;
441 break;
442
443 default:
444 ASSERT(0);
445 break;
446 }
447
448 spin_lock(&ailp->xa_lock);
449 /* should we bother continuing? */
450 if (XFS_FORCED_SHUTDOWN(mp))
451 break;
452 ASSERT(mp->m_log);
453
454 count++;
455
456 /*
457 * Are there too many items we can't do anything with?
458 * If we we are skipping too many items because we can't flush
459 * them or they are already being flushed, we back off and
460 * given them time to complete whatever operation is being
461 * done. i.e. remove pressure from the AIL while we can't make
462 * progress so traversals don't slow down further inserts and
463 * removals to/from the AIL.
464 *
465 * The value of 100 is an arbitrary magic number based on
466 * observation.
467 */
468 if (stuck > 100)
469 break;
470
471 lip = xfs_trans_ail_cursor_next(ailp, &cur);
472 if (lip == NULL)
473 break;
474 lsn = lip->li_lsn;
475 }
476 xfs_trans_ail_cursor_done(ailp, &cur);
477 spin_unlock(&ailp->xa_lock);
478
479 if (flush_log) {
480 /*
481 * If something we need to push out was pinned, then
482 * push out the log so it will become unpinned and
483 * move forward in the AIL.
484 */
485 XFS_STATS_INC(xs_push_ail_flush);
486 xfs_log_force(mp, 0);
487 }
488
489 if (push_xfsbufd) {
490 /* we've got delayed write buffers to flush */
491 wake_up_process(mp->m_ddev_targp->bt_task);
492 }
493
494 /* assume we have more work to do in a short while */
495out_done:
496 if (!count) {
497 /* We're past our target or empty, so idle */
498 ailp->xa_last_pushed_lsn = 0;
499
500 tout = 50;
501 } else if (XFS_LSN_CMP(lsn, target) >= 0) {
502 /*
503 * We reached the target so wait a bit longer for I/O to
504 * complete and remove pushed items from the AIL before we
505 * start the next scan from the start of the AIL.
506 */
507 tout = 50;
508 ailp->xa_last_pushed_lsn = 0;
509 } else if ((stuck * 100) / count > 90) {
510 /*
511 * Either there is a lot of contention on the AIL or we
512 * are stuck due to operations in progress. "Stuck" in this
513 * case is defined as >90% of the items we tried to push
514 * were stuck.
515 *
516 * Backoff a bit more to allow some I/O to complete before
517 * continuing from where we were.
518 */
519 tout = 20;
520 }
521
522 return tout;
523}
524
525static int
526xfsaild(
527 void *data)
528{
529 struct xfs_ail *ailp = data;
530 long tout = 0; /* milliseconds */
531
532 while (!kthread_should_stop()) {
533 if (tout && tout <= 20)
534 __set_current_state(TASK_KILLABLE);
535 else
536 __set_current_state(TASK_INTERRUPTIBLE);
537 schedule_timeout(tout ?
538 msecs_to_jiffies(tout) : MAX_SCHEDULE_TIMEOUT);
539
540 try_to_freeze();
541
542 tout = xfsaild_push(ailp);
543 }
544
545 return 0;
546}
547
548/*
549 * This routine is called to move the tail of the AIL forward. It does this by
550 * trying to flush items in the AIL whose lsns are below the given
551 * threshold_lsn.
552 *
553 * The push is run asynchronously in a workqueue, which means the caller needs
554 * to handle waiting on the async flush for space to become available.
555 * We don't want to interrupt any push that is in progress, hence we only queue
556 * work if we set the pushing bit approriately.
557 *
558 * We do this unlocked - we only need to know whether there is anything in the
559 * AIL at the time we are called. We don't need to access the contents of
560 * any of the objects, so the lock is not needed.
561 */
562void
563xfs_ail_push(
564 struct xfs_ail *ailp,
565 xfs_lsn_t threshold_lsn)
566{
567 xfs_log_item_t *lip;
568
569 lip = xfs_ail_min(ailp);
570 if (!lip || XFS_FORCED_SHUTDOWN(ailp->xa_mount) ||
571 XFS_LSN_CMP(threshold_lsn, ailp->xa_target) <= 0)
572 return;
573
574 /*
575 * Ensure that the new target is noticed in push code before it clears
576 * the XFS_AIL_PUSHING_BIT.
577 */
578 smp_wmb();
579 xfs_trans_ail_copy_lsn(ailp, &ailp->xa_target, &threshold_lsn);
580 smp_wmb();
581
582 wake_up_process(ailp->xa_task);
583}
584
585/*
586 * Push out all items in the AIL immediately
587 */
588void
589xfs_ail_push_all(
590 struct xfs_ail *ailp)
591{
592 xfs_lsn_t threshold_lsn = xfs_ail_max_lsn(ailp);
593
594 if (threshold_lsn)
595 xfs_ail_push(ailp, threshold_lsn);
596}
597
598/*
599 * This is to be called when an item is unlocked that may have
600 * been in the AIL. It will wake up the first member of the AIL
601 * wait list if this item's unlocking might allow it to progress.
602 * If the item is in the AIL, then we need to get the AIL lock
603 * while doing our checking so we don't race with someone going
604 * to sleep waiting for this event in xfs_trans_push_ail().
605 */
606void
607xfs_trans_unlocked_item(
608 struct xfs_ail *ailp,
609 xfs_log_item_t *lip)
610{
611 xfs_log_item_t *min_lip;
612
613 /*
614 * If we're forcibly shutting down, we may have
615 * unlocked log items arbitrarily. The last thing
616 * we want to do is to move the tail of the log
617 * over some potentially valid data.
618 */
619 if (!(lip->li_flags & XFS_LI_IN_AIL) ||
620 XFS_FORCED_SHUTDOWN(ailp->xa_mount)) {
621 return;
622 }
623
624 /*
625 * This is the one case where we can call into xfs_ail_min()
626 * without holding the AIL lock because we only care about the
627 * case where we are at the tail of the AIL. If the object isn't
628 * at the tail, it doesn't matter what result we get back. This
629 * is slightly racy because since we were just unlocked, we could
630 * go to sleep between the call to xfs_ail_min and the call to
631 * xfs_log_move_tail, have someone else lock us, commit to us disk,
632 * move us out of the tail of the AIL, and then we wake up. However,
633 * the call to xfs_log_move_tail() doesn't do anything if there's
634 * not enough free space to wake people up so we're safe calling it.
635 */
636 min_lip = xfs_ail_min(ailp);
637
638 if (min_lip == lip)
639 xfs_log_move_tail(ailp->xa_mount, 1);
640} /* xfs_trans_unlocked_item */
641
642/*
643 * xfs_trans_ail_update - bulk AIL insertion operation.
644 *
645 * @xfs_trans_ail_update takes an array of log items that all need to be
646 * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
647 * be added. Otherwise, it will be repositioned by removing it and re-adding
648 * it to the AIL. If we move the first item in the AIL, update the log tail to
649 * match the new minimum LSN in the AIL.
650 *
651 * This function takes the AIL lock once to execute the update operations on
652 * all the items in the array, and as such should not be called with the AIL
653 * lock held. As a result, once we have the AIL lock, we need to check each log
654 * item LSN to confirm it needs to be moved forward in the AIL.
655 *
656 * To optimise the insert operation, we delete all the items from the AIL in
657 * the first pass, moving them into a temporary list, then splice the temporary
658 * list into the correct position in the AIL. This avoids needing to do an
659 * insert operation on every item.
660 *
661 * This function must be called with the AIL lock held. The lock is dropped
662 * before returning.
663 */
664void
665xfs_trans_ail_update_bulk(
666 struct xfs_ail *ailp,
667 struct xfs_ail_cursor *cur,
668 struct xfs_log_item **log_items,
669 int nr_items,
670 xfs_lsn_t lsn) __releases(ailp->xa_lock)
671{
672 xfs_log_item_t *mlip;
673 xfs_lsn_t tail_lsn;
674 int mlip_changed = 0;
675 int i;
676 LIST_HEAD(tmp);
677
678 ASSERT(nr_items > 0); /* Not required, but true. */
679 mlip = xfs_ail_min(ailp);
680
681 for (i = 0; i < nr_items; i++) {
682 struct xfs_log_item *lip = log_items[i];
683 if (lip->li_flags & XFS_LI_IN_AIL) {
684 /* check if we really need to move the item */
685 if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
686 continue;
687
688 xfs_ail_delete(ailp, lip);
689 if (mlip == lip)
690 mlip_changed = 1;
691 } else {
692 lip->li_flags |= XFS_LI_IN_AIL;
693 }
694 lip->li_lsn = lsn;
695 list_add(&lip->li_ail, &tmp);
696 }
697
698 if (!list_empty(&tmp))
699 xfs_ail_splice(ailp, cur, &tmp, lsn);
700
701 if (!mlip_changed) {
702 spin_unlock(&ailp->xa_lock);
703 return;
704 }
705
706 /*
707 * It is not safe to access mlip after the AIL lock is dropped, so we
708 * must get a copy of li_lsn before we do so. This is especially
709 * important on 32-bit platforms where accessing and updating 64-bit
710 * values like li_lsn is not atomic.
711 */
712 mlip = xfs_ail_min(ailp);
713 tail_lsn = mlip->li_lsn;
714 spin_unlock(&ailp->xa_lock);
715 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
716}
717
718/*
719 * xfs_trans_ail_delete_bulk - remove multiple log items from the AIL
720 *
721 * @xfs_trans_ail_delete_bulk takes an array of log items that all need to
722 * removed from the AIL. The caller is already holding the AIL lock, and done
723 * all the checks necessary to ensure the items passed in via @log_items are
724 * ready for deletion. This includes checking that the items are in the AIL.
725 *
726 * For each log item to be removed, unlink it from the AIL, clear the IN_AIL
727 * flag from the item and reset the item's lsn to 0. If we remove the first
728 * item in the AIL, update the log tail to match the new minimum LSN in the
729 * AIL.
730 *
731 * This function will not drop the AIL lock until all items are removed from
732 * the AIL to minimise the amount of lock traffic on the AIL. This does not
733 * greatly increase the AIL hold time, but does significantly reduce the amount
734 * of traffic on the lock, especially during IO completion.
735 *
736 * This function must be called with the AIL lock held. The lock is dropped
737 * before returning.
738 */
739void
740xfs_trans_ail_delete_bulk(
741 struct xfs_ail *ailp,
742 struct xfs_log_item **log_items,
743 int nr_items) __releases(ailp->xa_lock)
744{
745 xfs_log_item_t *mlip;
746 xfs_lsn_t tail_lsn;
747 int mlip_changed = 0;
748 int i;
749
750 mlip = xfs_ail_min(ailp);
751
752 for (i = 0; i < nr_items; i++) {
753 struct xfs_log_item *lip = log_items[i];
754 if (!(lip->li_flags & XFS_LI_IN_AIL)) {
755 struct xfs_mount *mp = ailp->xa_mount;
756
757 spin_unlock(&ailp->xa_lock);
758 if (!XFS_FORCED_SHUTDOWN(mp)) {
759 xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
760 "%s: attempting to delete a log item that is not in the AIL",
761 __func__);
762 xfs_force_shutdown(mp, SHUTDOWN_CORRUPT_INCORE);
763 }
764 return;
765 }
766
767 xfs_ail_delete(ailp, lip);
768 lip->li_flags &= ~XFS_LI_IN_AIL;
769 lip->li_lsn = 0;
770 if (mlip == lip)
771 mlip_changed = 1;
772 }
773
774 if (!mlip_changed) {
775 spin_unlock(&ailp->xa_lock);
776 return;
777 }
778
779 /*
780 * It is not safe to access mlip after the AIL lock is dropped, so we
781 * must get a copy of li_lsn before we do so. This is especially
782 * important on 32-bit platforms where accessing and updating 64-bit
783 * values like li_lsn is not atomic. It is possible we've emptied the
784 * AIL here, so if that is the case, pass an LSN of 0 to the tail move.
785 */
786 mlip = xfs_ail_min(ailp);
787 tail_lsn = mlip ? mlip->li_lsn : 0;
788 spin_unlock(&ailp->xa_lock);
789 xfs_log_move_tail(ailp->xa_mount, tail_lsn);
790}
791
792/*
793 * The active item list (AIL) is a doubly linked list of log
794 * items sorted by ascending lsn. The base of the list is
795 * a forw/back pointer pair embedded in the xfs mount structure.
796 * The base is initialized with both pointers pointing to the
797 * base. This case always needs to be distinguished, because
798 * the base has no lsn to look at. We almost always insert
799 * at the end of the list, so on inserts we search from the
800 * end of the list to find where the new item belongs.
801 */
802
803/*
804 * Initialize the doubly linked list to point only to itself.
805 */
806int
807xfs_trans_ail_init(
808 xfs_mount_t *mp)
809{
810 struct xfs_ail *ailp;
811
812 ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
813 if (!ailp)
814 return ENOMEM;
815
816 ailp->xa_mount = mp;
817 INIT_LIST_HEAD(&ailp->xa_ail);
818 INIT_LIST_HEAD(&ailp->xa_cursors);
819 spin_lock_init(&ailp->xa_lock);
820
821 ailp->xa_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
822 ailp->xa_mount->m_fsname);
823 if (IS_ERR(ailp->xa_task))
824 goto out_free_ailp;
825
826 mp->m_ail = ailp;
827 return 0;
828
829out_free_ailp:
830 kmem_free(ailp);
831 return ENOMEM;
832}
833
834void
835xfs_trans_ail_destroy(
836 xfs_mount_t *mp)
837{
838 struct xfs_ail *ailp = mp->m_ail;
839
840 kthread_stop(ailp->xa_task);
841 kmem_free(ailp);
842}