xref: /openbmc/linux/fs/xfs/xfs_trans_ail.c (revision 51ad5b54)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (c) 2000-2002,2005 Silicon Graphics, Inc.
4  * Copyright (c) 2008 Dave Chinner
5  * All Rights Reserved.
6  */
7 #include "xfs.h"
8 #include "xfs_fs.h"
9 #include "xfs_shared.h"
10 #include "xfs_format.h"
11 #include "xfs_log_format.h"
12 #include "xfs_trans_resv.h"
13 #include "xfs_mount.h"
14 #include "xfs_trans.h"
15 #include "xfs_trans_priv.h"
16 #include "xfs_trace.h"
17 #include "xfs_errortag.h"
18 #include "xfs_error.h"
19 #include "xfs_log.h"
20 
21 #ifdef DEBUG
22 /*
23  * Check that the list is sorted as it should be.
24  *
25  * Called with the ail lock held, but we don't want to assert fail with it
26  * held otherwise we'll lock everything up and won't be able to debug the
27  * cause. Hence we sample and check the state under the AIL lock and return if
28  * everything is fine, otherwise we drop the lock and run the ASSERT checks.
29  * Asserts may not be fatal, so pick the lock back up and continue onwards.
30  */
31 STATIC void
32 xfs_ail_check(
33 	struct xfs_ail		*ailp,
34 	struct xfs_log_item	*lip)
35 	__must_hold(&ailp->ail_lock)
36 {
37 	struct xfs_log_item	*prev_lip;
38 	struct xfs_log_item	*next_lip;
39 	xfs_lsn_t		prev_lsn = NULLCOMMITLSN;
40 	xfs_lsn_t		next_lsn = NULLCOMMITLSN;
41 	xfs_lsn_t		lsn;
42 	bool			in_ail;
43 
44 
45 	if (list_empty(&ailp->ail_head))
46 		return;
47 
48 	/*
49 	 * Sample then check the next and previous entries are valid.
50 	 */
51 	in_ail = test_bit(XFS_LI_IN_AIL, &lip->li_flags);
52 	prev_lip = list_entry(lip->li_ail.prev, struct xfs_log_item, li_ail);
53 	if (&prev_lip->li_ail != &ailp->ail_head)
54 		prev_lsn = prev_lip->li_lsn;
55 	next_lip = list_entry(lip->li_ail.next, struct xfs_log_item, li_ail);
56 	if (&next_lip->li_ail != &ailp->ail_head)
57 		next_lsn = next_lip->li_lsn;
58 	lsn = lip->li_lsn;
59 
60 	if (in_ail &&
61 	    (prev_lsn == NULLCOMMITLSN || XFS_LSN_CMP(prev_lsn, lsn) <= 0) &&
62 	    (next_lsn == NULLCOMMITLSN || XFS_LSN_CMP(next_lsn, lsn) >= 0))
63 		return;
64 
65 	spin_unlock(&ailp->ail_lock);
66 	ASSERT(in_ail);
67 	ASSERT(prev_lsn == NULLCOMMITLSN || XFS_LSN_CMP(prev_lsn, lsn) <= 0);
68 	ASSERT(next_lsn == NULLCOMMITLSN || XFS_LSN_CMP(next_lsn, lsn) >= 0);
69 	spin_lock(&ailp->ail_lock);
70 }
71 #else /* !DEBUG */
72 #define	xfs_ail_check(a,l)
73 #endif /* DEBUG */
74 
75 /*
76  * Return a pointer to the last item in the AIL.  If the AIL is empty, then
77  * return NULL.
78  */
79 static struct xfs_log_item *
80 xfs_ail_max(
81 	struct xfs_ail  *ailp)
82 {
83 	if (list_empty(&ailp->ail_head))
84 		return NULL;
85 
86 	return list_entry(ailp->ail_head.prev, struct xfs_log_item, li_ail);
87 }
88 
89 /*
90  * Return a pointer to the item which follows the given item in the AIL.  If
91  * the given item is the last item in the list, then return NULL.
92  */
93 static struct xfs_log_item *
94 xfs_ail_next(
95 	struct xfs_ail		*ailp,
96 	struct xfs_log_item	*lip)
97 {
98 	if (lip->li_ail.next == &ailp->ail_head)
99 		return NULL;
100 
101 	return list_first_entry(&lip->li_ail, struct xfs_log_item, li_ail);
102 }
103 
104 /*
105  * This is called by the log manager code to determine the LSN of the tail of
106  * the log.  This is exactly the LSN of the first item in the AIL.  If the AIL
107  * is empty, then this function returns 0.
108  *
109  * We need the AIL lock in order to get a coherent read of the lsn of the last
110  * item in the AIL.
111  */
112 static xfs_lsn_t
113 __xfs_ail_min_lsn(
114 	struct xfs_ail		*ailp)
115 {
116 	struct xfs_log_item	*lip = xfs_ail_min(ailp);
117 
118 	if (lip)
119 		return lip->li_lsn;
120 	return 0;
121 }
122 
123 xfs_lsn_t
124 xfs_ail_min_lsn(
125 	struct xfs_ail		*ailp)
126 {
127 	xfs_lsn_t		lsn;
128 
129 	spin_lock(&ailp->ail_lock);
130 	lsn = __xfs_ail_min_lsn(ailp);
131 	spin_unlock(&ailp->ail_lock);
132 
133 	return lsn;
134 }
135 
136 /*
137  * Return the maximum lsn held in the AIL, or zero if the AIL is empty.
138  */
139 static xfs_lsn_t
140 xfs_ail_max_lsn(
141 	struct xfs_ail		*ailp)
142 {
143 	xfs_lsn_t       	lsn = 0;
144 	struct xfs_log_item	*lip;
145 
146 	spin_lock(&ailp->ail_lock);
147 	lip = xfs_ail_max(ailp);
148 	if (lip)
149 		lsn = lip->li_lsn;
150 	spin_unlock(&ailp->ail_lock);
151 
152 	return lsn;
153 }
154 
155 /*
156  * The cursor keeps track of where our current traversal is up to by tracking
157  * the next item in the list for us. However, for this to be safe, removing an
158  * object from the AIL needs to invalidate any cursor that points to it. hence
159  * the traversal cursor needs to be linked to the struct xfs_ail so that
160  * deletion can search all the active cursors for invalidation.
161  */
162 STATIC void
163 xfs_trans_ail_cursor_init(
164 	struct xfs_ail		*ailp,
165 	struct xfs_ail_cursor	*cur)
166 {
167 	cur->item = NULL;
168 	list_add_tail(&cur->list, &ailp->ail_cursors);
169 }
170 
171 /*
172  * Get the next item in the traversal and advance the cursor.  If the cursor
173  * was invalidated (indicated by a lip of 1), restart the traversal.
174  */
175 struct xfs_log_item *
176 xfs_trans_ail_cursor_next(
177 	struct xfs_ail		*ailp,
178 	struct xfs_ail_cursor	*cur)
179 {
180 	struct xfs_log_item	*lip = cur->item;
181 
182 	if ((uintptr_t)lip & 1)
183 		lip = xfs_ail_min(ailp);
184 	if (lip)
185 		cur->item = xfs_ail_next(ailp, lip);
186 	return lip;
187 }
188 
189 /*
190  * When the traversal is complete, we need to remove the cursor from the list
191  * of traversing cursors.
192  */
193 void
194 xfs_trans_ail_cursor_done(
195 	struct xfs_ail_cursor	*cur)
196 {
197 	cur->item = NULL;
198 	list_del_init(&cur->list);
199 }
200 
201 /*
202  * Invalidate any cursor that is pointing to this item. This is called when an
203  * item is removed from the AIL. Any cursor pointing to this object is now
204  * invalid and the traversal needs to be terminated so it doesn't reference a
205  * freed object. We set the low bit of the cursor item pointer so we can
206  * distinguish between an invalidation and the end of the list when getting the
207  * next item from the cursor.
208  */
209 STATIC void
210 xfs_trans_ail_cursor_clear(
211 	struct xfs_ail		*ailp,
212 	struct xfs_log_item	*lip)
213 {
214 	struct xfs_ail_cursor	*cur;
215 
216 	list_for_each_entry(cur, &ailp->ail_cursors, list) {
217 		if (cur->item == lip)
218 			cur->item = (struct xfs_log_item *)
219 					((uintptr_t)cur->item | 1);
220 	}
221 }
222 
223 /*
224  * Find the first item in the AIL with the given @lsn by searching in ascending
225  * LSN order and initialise the cursor to point to the next item for a
226  * ascending traversal.  Pass a @lsn of zero to initialise the cursor to the
227  * first item in the AIL. Returns NULL if the list is empty.
228  */
229 struct xfs_log_item *
230 xfs_trans_ail_cursor_first(
231 	struct xfs_ail		*ailp,
232 	struct xfs_ail_cursor	*cur,
233 	xfs_lsn_t		lsn)
234 {
235 	struct xfs_log_item	*lip;
236 
237 	xfs_trans_ail_cursor_init(ailp, cur);
238 
239 	if (lsn == 0) {
240 		lip = xfs_ail_min(ailp);
241 		goto out;
242 	}
243 
244 	list_for_each_entry(lip, &ailp->ail_head, li_ail) {
245 		if (XFS_LSN_CMP(lip->li_lsn, lsn) >= 0)
246 			goto out;
247 	}
248 	return NULL;
249 
250 out:
251 	if (lip)
252 		cur->item = xfs_ail_next(ailp, lip);
253 	return lip;
254 }
255 
256 static struct xfs_log_item *
257 __xfs_trans_ail_cursor_last(
258 	struct xfs_ail		*ailp,
259 	xfs_lsn_t		lsn)
260 {
261 	struct xfs_log_item	*lip;
262 
263 	list_for_each_entry_reverse(lip, &ailp->ail_head, li_ail) {
264 		if (XFS_LSN_CMP(lip->li_lsn, lsn) <= 0)
265 			return lip;
266 	}
267 	return NULL;
268 }
269 
270 /*
271  * Find the last item in the AIL with the given @lsn by searching in descending
272  * LSN order and initialise the cursor to point to that item.  If there is no
273  * item with the value of @lsn, then it sets the cursor to the last item with an
274  * LSN lower than @lsn.  Returns NULL if the list is empty.
275  */
276 struct xfs_log_item *
277 xfs_trans_ail_cursor_last(
278 	struct xfs_ail		*ailp,
279 	struct xfs_ail_cursor	*cur,
280 	xfs_lsn_t		lsn)
281 {
282 	xfs_trans_ail_cursor_init(ailp, cur);
283 	cur->item = __xfs_trans_ail_cursor_last(ailp, lsn);
284 	return cur->item;
285 }
286 
287 /*
288  * Splice the log item list into the AIL at the given LSN. We splice to the
289  * tail of the given LSN to maintain insert order for push traversals. The
290  * cursor is optional, allowing repeated updates to the same LSN to avoid
291  * repeated traversals.  This should not be called with an empty list.
292  */
293 static void
294 xfs_ail_splice(
295 	struct xfs_ail		*ailp,
296 	struct xfs_ail_cursor	*cur,
297 	struct list_head	*list,
298 	xfs_lsn_t		lsn)
299 {
300 	struct xfs_log_item	*lip;
301 
302 	ASSERT(!list_empty(list));
303 
304 	/*
305 	 * Use the cursor to determine the insertion point if one is
306 	 * provided.  If not, or if the one we got is not valid,
307 	 * find the place in the AIL where the items belong.
308 	 */
309 	lip = cur ? cur->item : NULL;
310 	if (!lip || (uintptr_t)lip & 1)
311 		lip = __xfs_trans_ail_cursor_last(ailp, lsn);
312 
313 	/*
314 	 * If a cursor is provided, we know we're processing the AIL
315 	 * in lsn order, and future items to be spliced in will
316 	 * follow the last one being inserted now.  Update the
317 	 * cursor to point to that last item, now while we have a
318 	 * reliable pointer to it.
319 	 */
320 	if (cur)
321 		cur->item = list_entry(list->prev, struct xfs_log_item, li_ail);
322 
323 	/*
324 	 * Finally perform the splice.  Unless the AIL was empty,
325 	 * lip points to the item in the AIL _after_ which the new
326 	 * items should go.  If lip is null the AIL was empty, so
327 	 * the new items go at the head of the AIL.
328 	 */
329 	if (lip)
330 		list_splice(list, &lip->li_ail);
331 	else
332 		list_splice(list, &ailp->ail_head);
333 }
334 
335 /*
336  * Delete the given item from the AIL.  Return a pointer to the item.
337  */
338 static void
339 xfs_ail_delete(
340 	struct xfs_ail		*ailp,
341 	struct xfs_log_item	*lip)
342 {
343 	xfs_ail_check(ailp, lip);
344 	list_del(&lip->li_ail);
345 	xfs_trans_ail_cursor_clear(ailp, lip);
346 }
347 
348 /*
349  * Requeue a failed buffer for writeback.
350  *
351  * We clear the log item failed state here as well, but we have to be careful
352  * about reference counts because the only active reference counts on the buffer
353  * may be the failed log items. Hence if we clear the log item failed state
354  * before queuing the buffer for IO we can release all active references to
355  * the buffer and free it, leading to use after free problems in
356  * xfs_buf_delwri_queue. It makes no difference to the buffer or log items which
357  * order we process them in - the buffer is locked, and we own the buffer list
358  * so nothing on them is going to change while we are performing this action.
359  *
360  * Hence we can safely queue the buffer for IO before we clear the failed log
361  * item state, therefore  always having an active reference to the buffer and
362  * avoiding the transient zero-reference state that leads to use-after-free.
363  */
364 static inline int
365 xfsaild_resubmit_item(
366 	struct xfs_log_item	*lip,
367 	struct list_head	*buffer_list)
368 {
369 	struct xfs_buf		*bp = lip->li_buf;
370 
371 	if (!xfs_buf_trylock(bp))
372 		return XFS_ITEM_LOCKED;
373 
374 	if (!xfs_buf_delwri_queue(bp, buffer_list)) {
375 		xfs_buf_unlock(bp);
376 		return XFS_ITEM_FLUSHING;
377 	}
378 
379 	/* protected by ail_lock */
380 	list_for_each_entry(lip, &bp->b_li_list, li_bio_list)
381 		xfs_clear_li_failed(lip);
382 
383 	xfs_buf_unlock(bp);
384 	return XFS_ITEM_SUCCESS;
385 }
386 
387 static inline uint
388 xfsaild_push_item(
389 	struct xfs_ail		*ailp,
390 	struct xfs_log_item	*lip)
391 {
392 	/*
393 	 * If log item pinning is enabled, skip the push and track the item as
394 	 * pinned. This can help induce head-behind-tail conditions.
395 	 */
396 	if (XFS_TEST_ERROR(false, ailp->ail_mount, XFS_ERRTAG_LOG_ITEM_PIN))
397 		return XFS_ITEM_PINNED;
398 
399 	/*
400 	 * Consider the item pinned if a push callback is not defined so the
401 	 * caller will force the log. This should only happen for intent items
402 	 * as they are unpinned once the associated done item is committed to
403 	 * the on-disk log.
404 	 */
405 	if (!lip->li_ops->iop_push)
406 		return XFS_ITEM_PINNED;
407 	if (test_bit(XFS_LI_FAILED, &lip->li_flags))
408 		return xfsaild_resubmit_item(lip, &ailp->ail_buf_list);
409 	return lip->li_ops->iop_push(lip, &ailp->ail_buf_list);
410 }
411 
412 static long
413 xfsaild_push(
414 	struct xfs_ail		*ailp)
415 {
416 	xfs_mount_t		*mp = ailp->ail_mount;
417 	struct xfs_ail_cursor	cur;
418 	struct xfs_log_item	*lip;
419 	xfs_lsn_t		lsn;
420 	xfs_lsn_t		target;
421 	long			tout;
422 	int			stuck = 0;
423 	int			flushing = 0;
424 	int			count = 0;
425 
426 	/*
427 	 * If we encountered pinned items or did not finish writing out all
428 	 * buffers the last time we ran, force the log first and wait for it
429 	 * before pushing again.
430 	 */
431 	if (ailp->ail_log_flush && ailp->ail_last_pushed_lsn == 0 &&
432 	    (!list_empty_careful(&ailp->ail_buf_list) ||
433 	     xfs_ail_min_lsn(ailp))) {
434 		ailp->ail_log_flush = 0;
435 
436 		XFS_STATS_INC(mp, xs_push_ail_flush);
437 		xfs_log_force(mp, XFS_LOG_SYNC);
438 	}
439 
440 	spin_lock(&ailp->ail_lock);
441 
442 	/* barrier matches the ail_target update in xfs_ail_push() */
443 	smp_rmb();
444 	target = ailp->ail_target;
445 	ailp->ail_target_prev = target;
446 
447 	lip = xfs_trans_ail_cursor_first(ailp, &cur, ailp->ail_last_pushed_lsn);
448 	if (!lip) {
449 		/*
450 		 * If the AIL is empty or our push has reached the end we are
451 		 * done now.
452 		 */
453 		xfs_trans_ail_cursor_done(&cur);
454 		spin_unlock(&ailp->ail_lock);
455 		goto out_done;
456 	}
457 
458 	XFS_STATS_INC(mp, xs_push_ail);
459 
460 	lsn = lip->li_lsn;
461 	while ((XFS_LSN_CMP(lip->li_lsn, target) <= 0)) {
462 		int	lock_result;
463 
464 		/*
465 		 * Note that iop_push may unlock and reacquire the AIL lock.  We
466 		 * rely on the AIL cursor implementation to be able to deal with
467 		 * the dropped lock.
468 		 */
469 		lock_result = xfsaild_push_item(ailp, lip);
470 		switch (lock_result) {
471 		case XFS_ITEM_SUCCESS:
472 			XFS_STATS_INC(mp, xs_push_ail_success);
473 			trace_xfs_ail_push(lip);
474 
475 			ailp->ail_last_pushed_lsn = lsn;
476 			break;
477 
478 		case XFS_ITEM_FLUSHING:
479 			/*
480 			 * The item or its backing buffer is already being
481 			 * flushed.  The typical reason for that is that an
482 			 * inode buffer is locked because we already pushed the
483 			 * updates to it as part of inode clustering.
484 			 *
485 			 * We do not want to to stop flushing just because lots
486 			 * of items are already being flushed, but we need to
487 			 * re-try the flushing relatively soon if most of the
488 			 * AIL is being flushed.
489 			 */
490 			XFS_STATS_INC(mp, xs_push_ail_flushing);
491 			trace_xfs_ail_flushing(lip);
492 
493 			flushing++;
494 			ailp->ail_last_pushed_lsn = lsn;
495 			break;
496 
497 		case XFS_ITEM_PINNED:
498 			XFS_STATS_INC(mp, xs_push_ail_pinned);
499 			trace_xfs_ail_pinned(lip);
500 
501 			stuck++;
502 			ailp->ail_log_flush++;
503 			break;
504 		case XFS_ITEM_LOCKED:
505 			XFS_STATS_INC(mp, xs_push_ail_locked);
506 			trace_xfs_ail_locked(lip);
507 
508 			stuck++;
509 			break;
510 		default:
511 			ASSERT(0);
512 			break;
513 		}
514 
515 		count++;
516 
517 		/*
518 		 * Are there too many items we can't do anything with?
519 		 *
520 		 * If we we are skipping too many items because we can't flush
521 		 * them or they are already being flushed, we back off and
522 		 * given them time to complete whatever operation is being
523 		 * done. i.e. remove pressure from the AIL while we can't make
524 		 * progress so traversals don't slow down further inserts and
525 		 * removals to/from the AIL.
526 		 *
527 		 * The value of 100 is an arbitrary magic number based on
528 		 * observation.
529 		 */
530 		if (stuck > 100)
531 			break;
532 
533 		lip = xfs_trans_ail_cursor_next(ailp, &cur);
534 		if (lip == NULL)
535 			break;
536 		lsn = lip->li_lsn;
537 	}
538 	xfs_trans_ail_cursor_done(&cur);
539 	spin_unlock(&ailp->ail_lock);
540 
541 	if (xfs_buf_delwri_submit_nowait(&ailp->ail_buf_list))
542 		ailp->ail_log_flush++;
543 
544 	if (!count || XFS_LSN_CMP(lsn, target) >= 0) {
545 out_done:
546 		/*
547 		 * We reached the target or the AIL is empty, so wait a bit
548 		 * longer for I/O to complete and remove pushed items from the
549 		 * AIL before we start the next scan from the start of the AIL.
550 		 */
551 		tout = 50;
552 		ailp->ail_last_pushed_lsn = 0;
553 	} else if (((stuck + flushing) * 100) / count > 90) {
554 		/*
555 		 * Either there is a lot of contention on the AIL or we are
556 		 * stuck due to operations in progress. "Stuck" in this case
557 		 * is defined as >90% of the items we tried to push were stuck.
558 		 *
559 		 * Backoff a bit more to allow some I/O to complete before
560 		 * restarting from the start of the AIL. This prevents us from
561 		 * spinning on the same items, and if they are pinned will all
562 		 * the restart to issue a log force to unpin the stuck items.
563 		 */
564 		tout = 20;
565 		ailp->ail_last_pushed_lsn = 0;
566 	} else {
567 		/*
568 		 * Assume we have more work to do in a short while.
569 		 */
570 		tout = 10;
571 	}
572 
573 	return tout;
574 }
575 
576 static int
577 xfsaild(
578 	void		*data)
579 {
580 	struct xfs_ail	*ailp = data;
581 	long		tout = 0;	/* milliseconds */
582 	unsigned int	noreclaim_flag;
583 
584 	noreclaim_flag = memalloc_noreclaim_save();
585 	set_freezable();
586 
587 	while (1) {
588 		if (tout && tout <= 20)
589 			set_current_state(TASK_KILLABLE);
590 		else
591 			set_current_state(TASK_INTERRUPTIBLE);
592 
593 		/*
594 		 * Check kthread_should_stop() after we set the task state to
595 		 * guarantee that we either see the stop bit and exit or the
596 		 * task state is reset to runnable such that it's not scheduled
597 		 * out indefinitely and detects the stop bit at next iteration.
598 		 * A memory barrier is included in above task state set to
599 		 * serialize again kthread_stop().
600 		 */
601 		if (kthread_should_stop()) {
602 			__set_current_state(TASK_RUNNING);
603 
604 			/*
605 			 * The caller forces out the AIL before stopping the
606 			 * thread in the common case, which means the delwri
607 			 * queue is drained. In the shutdown case, the queue may
608 			 * still hold relogged buffers that haven't been
609 			 * submitted because they were pinned since added to the
610 			 * queue.
611 			 *
612 			 * Log I/O error processing stales the underlying buffer
613 			 * and clears the delwri state, expecting the buf to be
614 			 * removed on the next submission attempt. That won't
615 			 * happen if we're shutting down, so this is the last
616 			 * opportunity to release such buffers from the queue.
617 			 */
618 			ASSERT(list_empty(&ailp->ail_buf_list) ||
619 			       XFS_FORCED_SHUTDOWN(ailp->ail_mount));
620 			xfs_buf_delwri_cancel(&ailp->ail_buf_list);
621 			break;
622 		}
623 
624 		spin_lock(&ailp->ail_lock);
625 
626 		/*
627 		 * Idle if the AIL is empty and we are not racing with a target
628 		 * update. We check the AIL after we set the task to a sleep
629 		 * state to guarantee that we either catch an ail_target update
630 		 * or that a wake_up resets the state to TASK_RUNNING.
631 		 * Otherwise, we run the risk of sleeping indefinitely.
632 		 *
633 		 * The barrier matches the ail_target update in xfs_ail_push().
634 		 */
635 		smp_rmb();
636 		if (!xfs_ail_min(ailp) &&
637 		    ailp->ail_target == ailp->ail_target_prev) {
638 			spin_unlock(&ailp->ail_lock);
639 			freezable_schedule();
640 			tout = 0;
641 			continue;
642 		}
643 		spin_unlock(&ailp->ail_lock);
644 
645 		if (tout)
646 			freezable_schedule_timeout(msecs_to_jiffies(tout));
647 
648 		__set_current_state(TASK_RUNNING);
649 
650 		try_to_freeze();
651 
652 		tout = xfsaild_push(ailp);
653 	}
654 
655 	memalloc_noreclaim_restore(noreclaim_flag);
656 	return 0;
657 }
658 
659 /*
660  * This routine is called to move the tail of the AIL forward.  It does this by
661  * trying to flush items in the AIL whose lsns are below the given
662  * threshold_lsn.
663  *
664  * The push is run asynchronously in a workqueue, which means the caller needs
665  * to handle waiting on the async flush for space to become available.
666  * We don't want to interrupt any push that is in progress, hence we only queue
667  * work if we set the pushing bit appropriately.
668  *
669  * We do this unlocked - we only need to know whether there is anything in the
670  * AIL at the time we are called. We don't need to access the contents of
671  * any of the objects, so the lock is not needed.
672  */
673 void
674 xfs_ail_push(
675 	struct xfs_ail		*ailp,
676 	xfs_lsn_t		threshold_lsn)
677 {
678 	struct xfs_log_item	*lip;
679 
680 	lip = xfs_ail_min(ailp);
681 	if (!lip || XFS_FORCED_SHUTDOWN(ailp->ail_mount) ||
682 	    XFS_LSN_CMP(threshold_lsn, ailp->ail_target) <= 0)
683 		return;
684 
685 	/*
686 	 * Ensure that the new target is noticed in push code before it clears
687 	 * the XFS_AIL_PUSHING_BIT.
688 	 */
689 	smp_wmb();
690 	xfs_trans_ail_copy_lsn(ailp, &ailp->ail_target, &threshold_lsn);
691 	smp_wmb();
692 
693 	wake_up_process(ailp->ail_task);
694 }
695 
696 /*
697  * Push out all items in the AIL immediately
698  */
699 void
700 xfs_ail_push_all(
701 	struct xfs_ail  *ailp)
702 {
703 	xfs_lsn_t       threshold_lsn = xfs_ail_max_lsn(ailp);
704 
705 	if (threshold_lsn)
706 		xfs_ail_push(ailp, threshold_lsn);
707 }
708 
709 /*
710  * Push out all items in the AIL immediately and wait until the AIL is empty.
711  */
712 void
713 xfs_ail_push_all_sync(
714 	struct xfs_ail  *ailp)
715 {
716 	struct xfs_log_item	*lip;
717 	DEFINE_WAIT(wait);
718 
719 	spin_lock(&ailp->ail_lock);
720 	while ((lip = xfs_ail_max(ailp)) != NULL) {
721 		prepare_to_wait(&ailp->ail_empty, &wait, TASK_UNINTERRUPTIBLE);
722 		ailp->ail_target = lip->li_lsn;
723 		wake_up_process(ailp->ail_task);
724 		spin_unlock(&ailp->ail_lock);
725 		schedule();
726 		spin_lock(&ailp->ail_lock);
727 	}
728 	spin_unlock(&ailp->ail_lock);
729 
730 	finish_wait(&ailp->ail_empty, &wait);
731 }
732 
733 void
734 xfs_ail_update_finish(
735 	struct xfs_ail		*ailp,
736 	xfs_lsn_t		old_lsn) __releases(ailp->ail_lock)
737 {
738 	struct xfs_mount	*mp = ailp->ail_mount;
739 
740 	/* if the tail lsn hasn't changed, don't do updates or wakeups. */
741 	if (!old_lsn || old_lsn == __xfs_ail_min_lsn(ailp)) {
742 		spin_unlock(&ailp->ail_lock);
743 		return;
744 	}
745 
746 	if (!XFS_FORCED_SHUTDOWN(mp))
747 		xlog_assign_tail_lsn_locked(mp);
748 
749 	if (list_empty(&ailp->ail_head))
750 		wake_up_all(&ailp->ail_empty);
751 	spin_unlock(&ailp->ail_lock);
752 	xfs_log_space_wake(mp);
753 }
754 
755 /*
756  * xfs_trans_ail_update - bulk AIL insertion operation.
757  *
758  * @xfs_trans_ail_update takes an array of log items that all need to be
759  * positioned at the same LSN in the AIL. If an item is not in the AIL, it will
760  * be added.  Otherwise, it will be repositioned  by removing it and re-adding
761  * it to the AIL. If we move the first item in the AIL, update the log tail to
762  * match the new minimum LSN in the AIL.
763  *
764  * This function takes the AIL lock once to execute the update operations on
765  * all the items in the array, and as such should not be called with the AIL
766  * lock held. As a result, once we have the AIL lock, we need to check each log
767  * item LSN to confirm it needs to be moved forward in the AIL.
768  *
769  * To optimise the insert operation, we delete all the items from the AIL in
770  * the first pass, moving them into a temporary list, then splice the temporary
771  * list into the correct position in the AIL. This avoids needing to do an
772  * insert operation on every item.
773  *
774  * This function must be called with the AIL lock held.  The lock is dropped
775  * before returning.
776  */
777 void
778 xfs_trans_ail_update_bulk(
779 	struct xfs_ail		*ailp,
780 	struct xfs_ail_cursor	*cur,
781 	struct xfs_log_item	**log_items,
782 	int			nr_items,
783 	xfs_lsn_t		lsn) __releases(ailp->ail_lock)
784 {
785 	struct xfs_log_item	*mlip;
786 	xfs_lsn_t		tail_lsn = 0;
787 	int			i;
788 	LIST_HEAD(tmp);
789 
790 	ASSERT(nr_items > 0);		/* Not required, but true. */
791 	mlip = xfs_ail_min(ailp);
792 
793 	for (i = 0; i < nr_items; i++) {
794 		struct xfs_log_item *lip = log_items[i];
795 		if (test_and_set_bit(XFS_LI_IN_AIL, &lip->li_flags)) {
796 			/* check if we really need to move the item */
797 			if (XFS_LSN_CMP(lsn, lip->li_lsn) <= 0)
798 				continue;
799 
800 			trace_xfs_ail_move(lip, lip->li_lsn, lsn);
801 			if (mlip == lip && !tail_lsn)
802 				tail_lsn = lip->li_lsn;
803 
804 			xfs_ail_delete(ailp, lip);
805 		} else {
806 			trace_xfs_ail_insert(lip, 0, lsn);
807 		}
808 		lip->li_lsn = lsn;
809 		list_add(&lip->li_ail, &tmp);
810 	}
811 
812 	if (!list_empty(&tmp))
813 		xfs_ail_splice(ailp, cur, &tmp, lsn);
814 
815 	xfs_ail_update_finish(ailp, tail_lsn);
816 }
817 
818 /* Insert a log item into the AIL. */
819 void
820 xfs_trans_ail_insert(
821 	struct xfs_ail		*ailp,
822 	struct xfs_log_item	*lip,
823 	xfs_lsn_t		lsn)
824 {
825 	spin_lock(&ailp->ail_lock);
826 	xfs_trans_ail_update_bulk(ailp, NULL, &lip, 1, lsn);
827 }
828 
829 /*
830  * Delete one log item from the AIL.
831  *
832  * If this item was at the tail of the AIL, return the LSN of the log item so
833  * that we can use it to check if the LSN of the tail of the log has moved
834  * when finishing up the AIL delete process in xfs_ail_update_finish().
835  */
836 xfs_lsn_t
837 xfs_ail_delete_one(
838 	struct xfs_ail		*ailp,
839 	struct xfs_log_item	*lip)
840 {
841 	struct xfs_log_item	*mlip = xfs_ail_min(ailp);
842 	xfs_lsn_t		lsn = lip->li_lsn;
843 
844 	trace_xfs_ail_delete(lip, mlip->li_lsn, lip->li_lsn);
845 	xfs_ail_delete(ailp, lip);
846 	xfs_clear_li_failed(lip);
847 	clear_bit(XFS_LI_IN_AIL, &lip->li_flags);
848 	lip->li_lsn = 0;
849 
850 	if (mlip == lip)
851 		return lsn;
852 	return 0;
853 }
854 
855 void
856 xfs_trans_ail_delete(
857 	struct xfs_log_item	*lip,
858 	int			shutdown_type)
859 {
860 	struct xfs_ail		*ailp = lip->li_ailp;
861 	struct xfs_mount	*mp = ailp->ail_mount;
862 	xfs_lsn_t		tail_lsn;
863 
864 	spin_lock(&ailp->ail_lock);
865 	if (!test_bit(XFS_LI_IN_AIL, &lip->li_flags)) {
866 		spin_unlock(&ailp->ail_lock);
867 		if (shutdown_type && !XFS_FORCED_SHUTDOWN(mp)) {
868 			xfs_alert_tag(mp, XFS_PTAG_AILDELETE,
869 	"%s: attempting to delete a log item that is not in the AIL",
870 					__func__);
871 			xfs_force_shutdown(mp, shutdown_type);
872 		}
873 		return;
874 	}
875 
876 	/* xfs_ail_update_finish() drops the AIL lock */
877 	tail_lsn = xfs_ail_delete_one(ailp, lip);
878 	xfs_ail_update_finish(ailp, tail_lsn);
879 }
880 
881 int
882 xfs_trans_ail_init(
883 	xfs_mount_t	*mp)
884 {
885 	struct xfs_ail	*ailp;
886 
887 	ailp = kmem_zalloc(sizeof(struct xfs_ail), KM_MAYFAIL);
888 	if (!ailp)
889 		return -ENOMEM;
890 
891 	ailp->ail_mount = mp;
892 	INIT_LIST_HEAD(&ailp->ail_head);
893 	INIT_LIST_HEAD(&ailp->ail_cursors);
894 	spin_lock_init(&ailp->ail_lock);
895 	INIT_LIST_HEAD(&ailp->ail_buf_list);
896 	init_waitqueue_head(&ailp->ail_empty);
897 
898 	ailp->ail_task = kthread_run(xfsaild, ailp, "xfsaild/%s",
899 			ailp->ail_mount->m_super->s_id);
900 	if (IS_ERR(ailp->ail_task))
901 		goto out_free_ailp;
902 
903 	mp->m_ail = ailp;
904 	return 0;
905 
906 out_free_ailp:
907 	kmem_free(ailp);
908 	return -ENOMEM;
909 }
910 
911 void
912 xfs_trans_ail_destroy(
913 	xfs_mount_t	*mp)
914 {
915 	struct xfs_ail	*ailp = mp->m_ail;
916 
917 	kthread_stop(ailp->ail_task);
918 	kmem_free(ailp);
919 }
920