4 * Copyright (C) 2002, Linus Torvalds.
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the
9 * inode itself is not handled here.
11 * 10Apr2002 Andrew Morton
12 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/spinlock.h>
19 #include <linux/slab.h>
20 #include <linux/sched.h>
23 #include <linux/pagemap.h>
24 #include <linux/kthread.h>
25 #include <linux/writeback.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/tracepoint.h>
29 #include <linux/device.h>
33 * 4MB minimal write chunk size
35 #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
38 * Passed into wb_writeback(), essentially a subset of writeback_control
40 struct wb_writeback_work
{
42 struct super_block
*sb
;
43 unsigned long *older_than_this
;
44 enum writeback_sync_modes sync_mode
;
45 unsigned int tagged_writepages
:1;
46 unsigned int for_kupdate
:1;
47 unsigned int range_cyclic
:1;
48 unsigned int for_background
:1;
49 unsigned int for_sync
:1; /* sync(2) WB_SYNC_ALL writeback */
50 enum wb_reason reason
; /* why was writeback initiated? */
52 struct list_head list
; /* pending work list */
53 struct completion
*done
; /* set if the caller waits */
57 * writeback_in_progress - determine whether there is writeback in progress
58 * @bdi: the device's backing_dev_info structure.
60 * Determine whether there is writeback waiting to be handled against a
63 int writeback_in_progress(struct backing_dev_info
*bdi
)
65 return test_bit(BDI_writeback_running
, &bdi
->state
);
67 EXPORT_SYMBOL(writeback_in_progress
);
69 static inline struct backing_dev_info
*inode_to_bdi(struct inode
*inode
)
71 struct super_block
*sb
= inode
->i_sb
;
73 if (sb_is_blkdev_sb(sb
))
74 return inode
->i_mapping
->backing_dev_info
;
79 static inline struct inode
*wb_inode(struct list_head
*head
)
81 return list_entry(head
, struct inode
, i_wb_list
);
85 * Include the creation of the trace points after defining the
86 * wb_writeback_work structure and inline functions so that the definition
87 * remains local to this file.
89 #define CREATE_TRACE_POINTS
90 #include <trace/events/writeback.h>
92 static void bdi_wakeup_thread(struct backing_dev_info
*bdi
)
94 spin_lock_bh(&bdi
->wb_lock
);
95 if (test_bit(BDI_registered
, &bdi
->state
))
96 mod_delayed_work(bdi_wq
, &bdi
->wb
.dwork
, 0);
97 spin_unlock_bh(&bdi
->wb_lock
);
100 static void bdi_queue_work(struct backing_dev_info
*bdi
,
101 struct wb_writeback_work
*work
)
103 trace_writeback_queue(bdi
, work
);
105 spin_lock_bh(&bdi
->wb_lock
);
106 if (!test_bit(BDI_registered
, &bdi
->state
)) {
108 complete(work
->done
);
111 list_add_tail(&work
->list
, &bdi
->work_list
);
112 mod_delayed_work(bdi_wq
, &bdi
->wb
.dwork
, 0);
114 spin_unlock_bh(&bdi
->wb_lock
);
118 __bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
119 bool range_cyclic
, enum wb_reason reason
)
121 struct wb_writeback_work
*work
;
124 * This is WB_SYNC_NONE writeback, so if allocation fails just
125 * wakeup the thread for old dirty data writeback
127 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
129 trace_writeback_nowork(bdi
);
130 bdi_wakeup_thread(bdi
);
134 work
->sync_mode
= WB_SYNC_NONE
;
135 work
->nr_pages
= nr_pages
;
136 work
->range_cyclic
= range_cyclic
;
137 work
->reason
= reason
;
139 bdi_queue_work(bdi
, work
);
143 * bdi_start_writeback - start writeback
144 * @bdi: the backing device to write from
145 * @nr_pages: the number of pages to write
146 * @reason: reason why some writeback work was initiated
149 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
150 * started when this function returns, we make no guarantees on
151 * completion. Caller need not hold sb s_umount semaphore.
154 void bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
155 enum wb_reason reason
)
157 __bdi_start_writeback(bdi
, nr_pages
, true, reason
);
161 * bdi_start_background_writeback - start background writeback
162 * @bdi: the backing device to write from
165 * This makes sure WB_SYNC_NONE background writeback happens. When
166 * this function returns, it is only guaranteed that for given BDI
167 * some IO is happening if we are over background dirty threshold.
168 * Caller need not hold sb s_umount semaphore.
170 void bdi_start_background_writeback(struct backing_dev_info
*bdi
)
173 * We just wake up the flusher thread. It will perform background
174 * writeback as soon as there is no other work to do.
176 trace_writeback_wake_background(bdi
);
177 bdi_wakeup_thread(bdi
);
181 * Remove the inode from the writeback list it is on.
183 void inode_wb_list_del(struct inode
*inode
)
185 struct backing_dev_info
*bdi
= inode_to_bdi(inode
);
187 spin_lock(&bdi
->wb
.list_lock
);
188 list_del_init(&inode
->i_wb_list
);
189 spin_unlock(&bdi
->wb
.list_lock
);
193 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
194 * furthest end of its superblock's dirty-inode list.
196 * Before stamping the inode's ->dirtied_when, we check to see whether it is
197 * already the most-recently-dirtied inode on the b_dirty list. If that is
198 * the case then the inode must have been redirtied while it was being written
199 * out and we don't reset its dirtied_when.
201 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
203 assert_spin_locked(&wb
->list_lock
);
204 if (!list_empty(&wb
->b_dirty
)) {
207 tail
= wb_inode(wb
->b_dirty
.next
);
208 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
209 inode
->dirtied_when
= jiffies
;
211 list_move(&inode
->i_wb_list
, &wb
->b_dirty
);
215 * requeue inode for re-scanning after bdi->b_io list is exhausted.
217 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
219 assert_spin_locked(&wb
->list_lock
);
220 list_move(&inode
->i_wb_list
, &wb
->b_more_io
);
223 static void inode_sync_complete(struct inode
*inode
)
225 inode
->i_state
&= ~I_SYNC
;
226 /* If inode is clean an unused, put it into LRU now... */
227 inode_add_lru(inode
);
228 /* Waiters must see I_SYNC cleared before being woken up */
230 wake_up_bit(&inode
->i_state
, __I_SYNC
);
233 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
235 bool ret
= time_after(inode
->dirtied_when
, t
);
238 * For inodes being constantly redirtied, dirtied_when can get stuck.
239 * It _appears_ to be in the future, but is actually in distant past.
240 * This test is necessary to prevent such wrapped-around relative times
241 * from permanently stopping the whole bdi writeback.
243 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
249 * Move expired (dirtied before work->older_than_this) dirty inodes from
250 * @delaying_queue to @dispatch_queue.
252 static int move_expired_inodes(struct list_head
*delaying_queue
,
253 struct list_head
*dispatch_queue
,
254 struct wb_writeback_work
*work
)
257 struct list_head
*pos
, *node
;
258 struct super_block
*sb
= NULL
;
263 while (!list_empty(delaying_queue
)) {
264 inode
= wb_inode(delaying_queue
->prev
);
265 if (work
->older_than_this
&&
266 inode_dirtied_after(inode
, *work
->older_than_this
))
268 list_move(&inode
->i_wb_list
, &tmp
);
270 if (sb_is_blkdev_sb(inode
->i_sb
))
272 if (sb
&& sb
!= inode
->i_sb
)
277 /* just one sb in list, splice to dispatch_queue and we're done */
279 list_splice(&tmp
, dispatch_queue
);
283 /* Move inodes from one superblock together */
284 while (!list_empty(&tmp
)) {
285 sb
= wb_inode(tmp
.prev
)->i_sb
;
286 list_for_each_prev_safe(pos
, node
, &tmp
) {
287 inode
= wb_inode(pos
);
288 if (inode
->i_sb
== sb
)
289 list_move(&inode
->i_wb_list
, dispatch_queue
);
297 * Queue all expired dirty inodes for io, eldest first.
299 * newly dirtied b_dirty b_io b_more_io
300 * =============> gf edc BA
302 * newly dirtied b_dirty b_io b_more_io
303 * =============> g fBAedc
305 * +--> dequeue for IO
307 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
310 assert_spin_locked(&wb
->list_lock
);
311 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
312 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, work
);
313 trace_writeback_queue_io(wb
, work
, moved
);
316 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
320 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
321 trace_writeback_write_inode_start(inode
, wbc
);
322 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
323 trace_writeback_write_inode(inode
, wbc
);
330 * Wait for writeback on an inode to complete. Called with i_lock held.
331 * Caller must make sure inode cannot go away when we drop i_lock.
333 static void __inode_wait_for_writeback(struct inode
*inode
)
334 __releases(inode
->i_lock
)
335 __acquires(inode
->i_lock
)
337 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
338 wait_queue_head_t
*wqh
;
340 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
341 while (inode
->i_state
& I_SYNC
) {
342 spin_unlock(&inode
->i_lock
);
343 __wait_on_bit(wqh
, &wq
, inode_wait
, TASK_UNINTERRUPTIBLE
);
344 spin_lock(&inode
->i_lock
);
349 * Wait for writeback on an inode to complete. Caller must have inode pinned.
351 void inode_wait_for_writeback(struct inode
*inode
)
353 spin_lock(&inode
->i_lock
);
354 __inode_wait_for_writeback(inode
);
355 spin_unlock(&inode
->i_lock
);
359 * Sleep until I_SYNC is cleared. This function must be called with i_lock
360 * held and drops it. It is aimed for callers not holding any inode reference
361 * so once i_lock is dropped, inode can go away.
363 static void inode_sleep_on_writeback(struct inode
*inode
)
364 __releases(inode
->i_lock
)
367 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
370 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
371 sleep
= inode
->i_state
& I_SYNC
;
372 spin_unlock(&inode
->i_lock
);
375 finish_wait(wqh
, &wait
);
379 * Find proper writeback list for the inode depending on its current state and
380 * possibly also change of its state while we were doing writeback. Here we
381 * handle things such as livelock prevention or fairness of writeback among
382 * inodes. This function can be called only by flusher thread - noone else
383 * processes all inodes in writeback lists and requeueing inodes behind flusher
384 * thread's back can have unexpected consequences.
386 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
387 struct writeback_control
*wbc
)
389 if (inode
->i_state
& I_FREEING
)
393 * Sync livelock prevention. Each inode is tagged and synced in one
394 * shot. If still dirty, it will be redirty_tail()'ed below. Update
395 * the dirty time to prevent enqueue and sync it again.
397 if ((inode
->i_state
& I_DIRTY
) &&
398 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
399 inode
->dirtied_when
= jiffies
;
401 if (wbc
->pages_skipped
) {
403 * writeback is not making progress due to locked
404 * buffers. Skip this inode for now.
406 redirty_tail(inode
, wb
);
410 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
412 * We didn't write back all the pages. nfs_writepages()
413 * sometimes bales out without doing anything.
415 if (wbc
->nr_to_write
<= 0) {
416 /* Slice used up. Queue for next turn. */
417 requeue_io(inode
, wb
);
420 * Writeback blocked by something other than
421 * congestion. Delay the inode for some time to
422 * avoid spinning on the CPU (100% iowait)
423 * retrying writeback of the dirty page/inode
424 * that cannot be performed immediately.
426 redirty_tail(inode
, wb
);
428 } else if (inode
->i_state
& I_DIRTY
) {
430 * Filesystems can dirty the inode during writeback operations,
431 * such as delayed allocation during submission or metadata
432 * updates after data IO completion.
434 redirty_tail(inode
, wb
);
436 /* The inode is clean. Remove from writeback lists. */
437 list_del_init(&inode
->i_wb_list
);
442 * Write out an inode and its dirty pages. Do not update the writeback list
443 * linkage. That is left to the caller. The caller is also responsible for
444 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
447 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
449 struct address_space
*mapping
= inode
->i_mapping
;
450 long nr_to_write
= wbc
->nr_to_write
;
454 WARN_ON(!(inode
->i_state
& I_SYNC
));
456 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
458 ret
= do_writepages(mapping
, wbc
);
461 * Make sure to wait on the data before writing out the metadata.
462 * This is important for filesystems that modify metadata on data
463 * I/O completion. We don't do it for sync(2) writeback because it has a
464 * separate, external IO completion path and ->sync_fs for guaranteeing
465 * inode metadata is written back correctly.
467 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
468 int err
= filemap_fdatawait(mapping
);
474 * Some filesystems may redirty the inode during the writeback
475 * due to delalloc, clear dirty metadata flags right before
478 spin_lock(&inode
->i_lock
);
479 /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
480 if (!mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
481 inode
->i_state
&= ~I_DIRTY_PAGES
;
482 dirty
= inode
->i_state
& I_DIRTY
;
483 inode
->i_state
&= ~(I_DIRTY_SYNC
| I_DIRTY_DATASYNC
);
484 spin_unlock(&inode
->i_lock
);
485 /* Don't write the inode if only I_DIRTY_PAGES was set */
486 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
487 int err
= write_inode(inode
, wbc
);
491 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
496 * Write out an inode's dirty pages. Either the caller has an active reference
497 * on the inode or the inode has I_WILL_FREE set.
499 * This function is designed to be called for writing back one inode which
500 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
501 * and does more profound writeback list handling in writeback_sb_inodes().
504 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
505 struct writeback_control
*wbc
)
509 spin_lock(&inode
->i_lock
);
510 if (!atomic_read(&inode
->i_count
))
511 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
513 WARN_ON(inode
->i_state
& I_WILL_FREE
);
515 if (inode
->i_state
& I_SYNC
) {
516 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
519 * It's a data-integrity sync. We must wait. Since callers hold
520 * inode reference or inode has I_WILL_FREE set, it cannot go
523 __inode_wait_for_writeback(inode
);
525 WARN_ON(inode
->i_state
& I_SYNC
);
527 * Skip inode if it is clean and we have no outstanding writeback in
528 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
529 * function since flusher thread may be doing for example sync in
530 * parallel and if we move the inode, it could get skipped. So here we
531 * make sure inode is on some writeback list and leave it there unless
532 * we have completely cleaned the inode.
534 if (!(inode
->i_state
& I_DIRTY
) &&
535 (wbc
->sync_mode
!= WB_SYNC_ALL
||
536 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
538 inode
->i_state
|= I_SYNC
;
539 spin_unlock(&inode
->i_lock
);
541 ret
= __writeback_single_inode(inode
, wbc
);
543 spin_lock(&wb
->list_lock
);
544 spin_lock(&inode
->i_lock
);
546 * If inode is clean, remove it from writeback lists. Otherwise don't
547 * touch it. See comment above for explanation.
549 if (!(inode
->i_state
& I_DIRTY
))
550 list_del_init(&inode
->i_wb_list
);
551 spin_unlock(&wb
->list_lock
);
552 inode_sync_complete(inode
);
554 spin_unlock(&inode
->i_lock
);
558 static long writeback_chunk_size(struct backing_dev_info
*bdi
,
559 struct wb_writeback_work
*work
)
564 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
565 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
566 * here avoids calling into writeback_inodes_wb() more than once.
568 * The intended call sequence for WB_SYNC_ALL writeback is:
571 * writeback_sb_inodes() <== called only once
572 * write_cache_pages() <== called once for each inode
573 * (quickly) tag currently dirty pages
574 * (maybe slowly) sync all tagged pages
576 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
579 pages
= min(bdi
->avg_write_bandwidth
/ 2,
580 global_dirty_limit
/ DIRTY_SCOPE
);
581 pages
= min(pages
, work
->nr_pages
);
582 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
583 MIN_WRITEBACK_PAGES
);
590 * Write a portion of b_io inodes which belong to @sb.
592 * Return the number of pages and/or inodes written.
594 static long writeback_sb_inodes(struct super_block
*sb
,
595 struct bdi_writeback
*wb
,
596 struct wb_writeback_work
*work
)
598 struct writeback_control wbc
= {
599 .sync_mode
= work
->sync_mode
,
600 .tagged_writepages
= work
->tagged_writepages
,
601 .for_kupdate
= work
->for_kupdate
,
602 .for_background
= work
->for_background
,
603 .for_sync
= work
->for_sync
,
604 .range_cyclic
= work
->range_cyclic
,
606 .range_end
= LLONG_MAX
,
608 unsigned long start_time
= jiffies
;
610 long wrote
= 0; /* count both pages and inodes */
612 while (!list_empty(&wb
->b_io
)) {
613 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
615 if (inode
->i_sb
!= sb
) {
618 * We only want to write back data for this
619 * superblock, move all inodes not belonging
620 * to it back onto the dirty list.
622 redirty_tail(inode
, wb
);
627 * The inode belongs to a different superblock.
628 * Bounce back to the caller to unpin this and
629 * pin the next superblock.
635 * Don't bother with new inodes or inodes being freed, first
636 * kind does not need periodic writeout yet, and for the latter
637 * kind writeout is handled by the freer.
639 spin_lock(&inode
->i_lock
);
640 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
641 spin_unlock(&inode
->i_lock
);
642 redirty_tail(inode
, wb
);
645 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
647 * If this inode is locked for writeback and we are not
648 * doing writeback-for-data-integrity, move it to
649 * b_more_io so that writeback can proceed with the
650 * other inodes on s_io.
652 * We'll have another go at writing back this inode
653 * when we completed a full scan of b_io.
655 spin_unlock(&inode
->i_lock
);
656 requeue_io(inode
, wb
);
657 trace_writeback_sb_inodes_requeue(inode
);
660 spin_unlock(&wb
->list_lock
);
663 * We already requeued the inode if it had I_SYNC set and we
664 * are doing WB_SYNC_NONE writeback. So this catches only the
667 if (inode
->i_state
& I_SYNC
) {
668 /* Wait for I_SYNC. This function drops i_lock... */
669 inode_sleep_on_writeback(inode
);
670 /* Inode may be gone, start again */
671 spin_lock(&wb
->list_lock
);
674 inode
->i_state
|= I_SYNC
;
675 spin_unlock(&inode
->i_lock
);
677 write_chunk
= writeback_chunk_size(wb
->bdi
, work
);
678 wbc
.nr_to_write
= write_chunk
;
679 wbc
.pages_skipped
= 0;
682 * We use I_SYNC to pin the inode in memory. While it is set
683 * evict_inode() will wait so the inode cannot be freed.
685 __writeback_single_inode(inode
, &wbc
);
687 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
688 wrote
+= write_chunk
- wbc
.nr_to_write
;
689 spin_lock(&wb
->list_lock
);
690 spin_lock(&inode
->i_lock
);
691 if (!(inode
->i_state
& I_DIRTY
))
693 requeue_inode(inode
, wb
, &wbc
);
694 inode_sync_complete(inode
);
695 spin_unlock(&inode
->i_lock
);
696 cond_resched_lock(&wb
->list_lock
);
698 * bail out to wb_writeback() often enough to check
699 * background threshold and other termination conditions.
702 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
704 if (work
->nr_pages
<= 0)
711 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
712 struct wb_writeback_work
*work
)
714 unsigned long start_time
= jiffies
;
717 while (!list_empty(&wb
->b_io
)) {
718 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
719 struct super_block
*sb
= inode
->i_sb
;
721 if (!grab_super_passive(sb
)) {
723 * grab_super_passive() may fail consistently due to
724 * s_umount being grabbed by someone else. Don't use
725 * requeue_io() to avoid busy retrying the inode/sb.
727 redirty_tail(inode
, wb
);
730 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
733 /* refer to the same tests at the end of writeback_sb_inodes */
735 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
737 if (work
->nr_pages
<= 0)
741 /* Leave any unwritten inodes on b_io */
745 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
746 enum wb_reason reason
)
748 struct wb_writeback_work work
= {
749 .nr_pages
= nr_pages
,
750 .sync_mode
= WB_SYNC_NONE
,
755 spin_lock(&wb
->list_lock
);
756 if (list_empty(&wb
->b_io
))
758 __writeback_inodes_wb(wb
, &work
);
759 spin_unlock(&wb
->list_lock
);
761 return nr_pages
- work
.nr_pages
;
764 static bool over_bground_thresh(struct backing_dev_info
*bdi
)
766 unsigned long background_thresh
, dirty_thresh
;
768 global_dirty_limits(&background_thresh
, &dirty_thresh
);
770 if (global_page_state(NR_FILE_DIRTY
) +
771 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
)
774 if (bdi_stat(bdi
, BDI_RECLAIMABLE
) >
775 bdi_dirty_limit(bdi
, background_thresh
))
782 * Called under wb->list_lock. If there are multiple wb per bdi,
783 * only the flusher working on the first wb should do it.
785 static void wb_update_bandwidth(struct bdi_writeback
*wb
,
786 unsigned long start_time
)
788 __bdi_update_bandwidth(wb
->bdi
, 0, 0, 0, 0, 0, start_time
);
792 * Explicit flushing or periodic writeback of "old" data.
794 * Define "old": the first time one of an inode's pages is dirtied, we mark the
795 * dirtying-time in the inode's address_space. So this periodic writeback code
796 * just walks the superblock inode list, writing back any inodes which are
797 * older than a specific point in time.
799 * Try to run once per dirty_writeback_interval. But if a writeback event
800 * takes longer than a dirty_writeback_interval interval, then leave a
803 * older_than_this takes precedence over nr_to_write. So we'll only write back
804 * all dirty pages if they are all attached to "old" mappings.
806 static long wb_writeback(struct bdi_writeback
*wb
,
807 struct wb_writeback_work
*work
)
809 unsigned long wb_start
= jiffies
;
810 long nr_pages
= work
->nr_pages
;
811 unsigned long oldest_jif
;
815 oldest_jif
= jiffies
;
816 work
->older_than_this
= &oldest_jif
;
818 spin_lock(&wb
->list_lock
);
821 * Stop writeback when nr_pages has been consumed
823 if (work
->nr_pages
<= 0)
827 * Background writeout and kupdate-style writeback may
828 * run forever. Stop them if there is other work to do
829 * so that e.g. sync can proceed. They'll be restarted
830 * after the other works are all done.
832 if ((work
->for_background
|| work
->for_kupdate
) &&
833 !list_empty(&wb
->bdi
->work_list
))
837 * For background writeout, stop when we are below the
838 * background dirty threshold
840 if (work
->for_background
&& !over_bground_thresh(wb
->bdi
))
844 * Kupdate and background works are special and we want to
845 * include all inodes that need writing. Livelock avoidance is
846 * handled by these works yielding to any other work so we are
849 if (work
->for_kupdate
) {
850 oldest_jif
= jiffies
-
851 msecs_to_jiffies(dirty_expire_interval
* 10);
852 } else if (work
->for_background
)
853 oldest_jif
= jiffies
;
855 trace_writeback_start(wb
->bdi
, work
);
856 if (list_empty(&wb
->b_io
))
859 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
861 progress
= __writeback_inodes_wb(wb
, work
);
862 trace_writeback_written(wb
->bdi
, work
);
864 wb_update_bandwidth(wb
, wb_start
);
867 * Did we write something? Try for more
869 * Dirty inodes are moved to b_io for writeback in batches.
870 * The completion of the current batch does not necessarily
871 * mean the overall work is done. So we keep looping as long
872 * as made some progress on cleaning pages or inodes.
877 * No more inodes for IO, bail
879 if (list_empty(&wb
->b_more_io
))
882 * Nothing written. Wait for some inode to
883 * become available for writeback. Otherwise
884 * we'll just busyloop.
886 if (!list_empty(&wb
->b_more_io
)) {
887 trace_writeback_wait(wb
->bdi
, work
);
888 inode
= wb_inode(wb
->b_more_io
.prev
);
889 spin_lock(&inode
->i_lock
);
890 spin_unlock(&wb
->list_lock
);
891 /* This function drops i_lock... */
892 inode_sleep_on_writeback(inode
);
893 spin_lock(&wb
->list_lock
);
896 spin_unlock(&wb
->list_lock
);
898 return nr_pages
- work
->nr_pages
;
902 * Return the next wb_writeback_work struct that hasn't been processed yet.
904 static struct wb_writeback_work
*
905 get_next_work_item(struct backing_dev_info
*bdi
)
907 struct wb_writeback_work
*work
= NULL
;
909 spin_lock_bh(&bdi
->wb_lock
);
910 if (!list_empty(&bdi
->work_list
)) {
911 work
= list_entry(bdi
->work_list
.next
,
912 struct wb_writeback_work
, list
);
913 list_del_init(&work
->list
);
915 spin_unlock_bh(&bdi
->wb_lock
);
920 * Add in the number of potentially dirty inodes, because each inode
921 * write can dirty pagecache in the underlying blockdev.
923 static unsigned long get_nr_dirty_pages(void)
925 return global_page_state(NR_FILE_DIRTY
) +
926 global_page_state(NR_UNSTABLE_NFS
) +
927 get_nr_dirty_inodes();
930 static long wb_check_background_flush(struct bdi_writeback
*wb
)
932 if (over_bground_thresh(wb
->bdi
)) {
934 struct wb_writeback_work work
= {
935 .nr_pages
= LONG_MAX
,
936 .sync_mode
= WB_SYNC_NONE
,
939 .reason
= WB_REASON_BACKGROUND
,
942 return wb_writeback(wb
, &work
);
948 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
950 unsigned long expired
;
954 * When set to zero, disable periodic writeback
956 if (!dirty_writeback_interval
)
959 expired
= wb
->last_old_flush
+
960 msecs_to_jiffies(dirty_writeback_interval
* 10);
961 if (time_before(jiffies
, expired
))
964 wb
->last_old_flush
= jiffies
;
965 nr_pages
= get_nr_dirty_pages();
968 struct wb_writeback_work work
= {
969 .nr_pages
= nr_pages
,
970 .sync_mode
= WB_SYNC_NONE
,
973 .reason
= WB_REASON_PERIODIC
,
976 return wb_writeback(wb
, &work
);
983 * Retrieve work items and do the writeback they describe
985 static long wb_do_writeback(struct bdi_writeback
*wb
)
987 struct backing_dev_info
*bdi
= wb
->bdi
;
988 struct wb_writeback_work
*work
;
991 set_bit(BDI_writeback_running
, &wb
->bdi
->state
);
992 while ((work
= get_next_work_item(bdi
)) != NULL
) {
994 trace_writeback_exec(bdi
, work
);
996 wrote
+= wb_writeback(wb
, work
);
999 * Notify the caller of completion if this is a synchronous
1000 * work item, otherwise just free it.
1003 complete(work
->done
);
1009 * Check for periodic writeback, kupdated() style
1011 wrote
+= wb_check_old_data_flush(wb
);
1012 wrote
+= wb_check_background_flush(wb
);
1013 clear_bit(BDI_writeback_running
, &wb
->bdi
->state
);
1019 * Handle writeback of dirty data for the device backed by this bdi. Also
1020 * reschedules periodically and does kupdated style flushing.
1022 void bdi_writeback_workfn(struct work_struct
*work
)
1024 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1025 struct bdi_writeback
, dwork
);
1026 struct backing_dev_info
*bdi
= wb
->bdi
;
1029 set_worker_desc("flush-%s", dev_name(bdi
->dev
));
1030 current
->flags
|= PF_SWAPWRITE
;
1032 if (likely(!current_is_workqueue_rescuer() ||
1033 !test_bit(BDI_registered
, &bdi
->state
))) {
1035 * The normal path. Keep writing back @bdi until its
1036 * work_list is empty. Note that this path is also taken
1037 * if @bdi is shutting down even when we're running off the
1038 * rescuer as work_list needs to be drained.
1041 pages_written
= wb_do_writeback(wb
);
1042 trace_writeback_pages_written(pages_written
);
1043 } while (!list_empty(&bdi
->work_list
));
1046 * bdi_wq can't get enough workers and we're running off
1047 * the emergency worker. Don't hog it. Hopefully, 1024 is
1048 * enough for efficient IO.
1050 pages_written
= writeback_inodes_wb(&bdi
->wb
, 1024,
1051 WB_REASON_FORKER_THREAD
);
1052 trace_writeback_pages_written(pages_written
);
1055 if (!list_empty(&bdi
->work_list
))
1056 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1057 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1058 bdi_wakeup_thread_delayed(bdi
);
1060 current
->flags
&= ~PF_SWAPWRITE
;
1064 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1067 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1069 struct backing_dev_info
*bdi
;
1072 nr_pages
= get_nr_dirty_pages();
1075 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1076 if (!bdi_has_dirty_io(bdi
))
1078 __bdi_start_writeback(bdi
, nr_pages
, false, reason
);
1083 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1085 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1086 struct dentry
*dentry
;
1087 const char *name
= "?";
1089 dentry
= d_find_alias(inode
);
1091 spin_lock(&dentry
->d_lock
);
1092 name
= (const char *) dentry
->d_name
.name
;
1095 "%s(%d): dirtied inode %lu (%s) on %s\n",
1096 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1097 name
, inode
->i_sb
->s_id
);
1099 spin_unlock(&dentry
->d_lock
);
1106 * __mark_inode_dirty - internal function
1107 * @inode: inode to mark
1108 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1109 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1110 * mark_inode_dirty_sync.
1112 * Put the inode on the super block's dirty list.
1114 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1115 * dirty list only if it is hashed or if it refers to a blockdev.
1116 * If it was not hashed, it will never be added to the dirty list
1117 * even if it is later hashed, as it will have been marked dirty already.
1119 * In short, make sure you hash any inodes _before_ you start marking
1122 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1123 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1124 * the kernel-internal blockdev inode represents the dirtying time of the
1125 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1126 * page->mapping->host, so the page-dirtying time is recorded in the internal
1129 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1131 struct super_block
*sb
= inode
->i_sb
;
1132 struct backing_dev_info
*bdi
= NULL
;
1135 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1136 * dirty the inode itself
1138 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
1139 trace_writeback_dirty_inode_start(inode
, flags
);
1141 if (sb
->s_op
->dirty_inode
)
1142 sb
->s_op
->dirty_inode(inode
, flags
);
1144 trace_writeback_dirty_inode(inode
, flags
);
1148 * make sure that changes are seen by all cpus before we test i_state
1153 /* avoid the locking if we can */
1154 if ((inode
->i_state
& flags
) == flags
)
1157 if (unlikely(block_dump
))
1158 block_dump___mark_inode_dirty(inode
);
1160 spin_lock(&inode
->i_lock
);
1161 if ((inode
->i_state
& flags
) != flags
) {
1162 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1164 inode
->i_state
|= flags
;
1167 * If the inode is being synced, just update its dirty state.
1168 * The unlocker will place the inode on the appropriate
1169 * superblock list, based upon its state.
1171 if (inode
->i_state
& I_SYNC
)
1172 goto out_unlock_inode
;
1175 * Only add valid (hashed) inodes to the superblock's
1176 * dirty list. Add blockdev inodes as well.
1178 if (!S_ISBLK(inode
->i_mode
)) {
1179 if (inode_unhashed(inode
))
1180 goto out_unlock_inode
;
1182 if (inode
->i_state
& I_FREEING
)
1183 goto out_unlock_inode
;
1186 * If the inode was already on b_dirty/b_io/b_more_io, don't
1187 * reposition it (that would break b_dirty time-ordering).
1190 bool wakeup_bdi
= false;
1191 bdi
= inode_to_bdi(inode
);
1193 spin_unlock(&inode
->i_lock
);
1194 spin_lock(&bdi
->wb
.list_lock
);
1195 if (bdi_cap_writeback_dirty(bdi
)) {
1196 WARN(!test_bit(BDI_registered
, &bdi
->state
),
1197 "bdi-%s not registered\n", bdi
->name
);
1200 * If this is the first dirty inode for this
1201 * bdi, we have to wake-up the corresponding
1202 * bdi thread to make sure background
1203 * write-back happens later.
1205 if (!wb_has_dirty_io(&bdi
->wb
))
1209 inode
->dirtied_when
= jiffies
;
1210 list_move(&inode
->i_wb_list
, &bdi
->wb
.b_dirty
);
1211 spin_unlock(&bdi
->wb
.list_lock
);
1214 bdi_wakeup_thread_delayed(bdi
);
1219 spin_unlock(&inode
->i_lock
);
1222 EXPORT_SYMBOL(__mark_inode_dirty
);
1224 static void wait_sb_inodes(struct super_block
*sb
)
1226 struct inode
*inode
, *old_inode
= NULL
;
1229 * We need to be protected against the filesystem going from
1230 * r/o to r/w or vice versa.
1232 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1234 spin_lock(&inode_sb_list_lock
);
1237 * Data integrity sync. Must wait for all pages under writeback,
1238 * because there may have been pages dirtied before our sync
1239 * call, but which had writeout started before we write it out.
1240 * In which case, the inode may not be on the dirty list, but
1241 * we still have to wait for that writeout.
1243 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1244 struct address_space
*mapping
= inode
->i_mapping
;
1246 spin_lock(&inode
->i_lock
);
1247 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1248 (mapping
->nrpages
== 0)) {
1249 spin_unlock(&inode
->i_lock
);
1253 spin_unlock(&inode
->i_lock
);
1254 spin_unlock(&inode_sb_list_lock
);
1257 * We hold a reference to 'inode' so it couldn't have been
1258 * removed from s_inodes list while we dropped the
1259 * inode_sb_list_lock. We cannot iput the inode now as we can
1260 * be holding the last reference and we cannot iput it under
1261 * inode_sb_list_lock. So we keep the reference and iput it
1267 filemap_fdatawait(mapping
);
1271 spin_lock(&inode_sb_list_lock
);
1273 spin_unlock(&inode_sb_list_lock
);
1278 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1279 * @sb: the superblock
1280 * @nr: the number of pages to write
1281 * @reason: reason why some writeback work initiated
1283 * Start writeback on some inodes on this super_block. No guarantees are made
1284 * on how many (if any) will be written, and this function does not wait
1285 * for IO completion of submitted IO.
1287 void writeback_inodes_sb_nr(struct super_block
*sb
,
1289 enum wb_reason reason
)
1291 DECLARE_COMPLETION_ONSTACK(done
);
1292 struct wb_writeback_work work
= {
1294 .sync_mode
= WB_SYNC_NONE
,
1295 .tagged_writepages
= 1,
1301 if (sb
->s_bdi
== &noop_backing_dev_info
)
1303 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1304 bdi_queue_work(sb
->s_bdi
, &work
);
1305 wait_for_completion(&done
);
1307 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1310 * writeback_inodes_sb - writeback dirty inodes from given super_block
1311 * @sb: the superblock
1312 * @reason: reason why some writeback work was initiated
1314 * Start writeback on some inodes on this super_block. No guarantees are made
1315 * on how many (if any) will be written, and this function does not wait
1316 * for IO completion of submitted IO.
1318 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1320 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1322 EXPORT_SYMBOL(writeback_inodes_sb
);
1325 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1326 * @sb: the superblock
1327 * @nr: the number of pages to write
1328 * @reason: the reason of writeback
1330 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1331 * Returns 1 if writeback was started, 0 if not.
1333 int try_to_writeback_inodes_sb_nr(struct super_block
*sb
,
1335 enum wb_reason reason
)
1337 if (writeback_in_progress(sb
->s_bdi
))
1340 if (!down_read_trylock(&sb
->s_umount
))
1343 writeback_inodes_sb_nr(sb
, nr
, reason
);
1344 up_read(&sb
->s_umount
);
1347 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1350 * try_to_writeback_inodes_sb - try to start writeback if none underway
1351 * @sb: the superblock
1352 * @reason: reason why some writeback work was initiated
1354 * Implement by try_to_writeback_inodes_sb_nr()
1355 * Returns 1 if writeback was started, 0 if not.
1357 int try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1359 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1361 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1364 * sync_inodes_sb - sync sb inode pages
1365 * @sb: the superblock
1367 * This function writes and waits on any dirty inode belonging to this
1370 void sync_inodes_sb(struct super_block
*sb
)
1372 DECLARE_COMPLETION_ONSTACK(done
);
1373 struct wb_writeback_work work
= {
1375 .sync_mode
= WB_SYNC_ALL
,
1376 .nr_pages
= LONG_MAX
,
1379 .reason
= WB_REASON_SYNC
,
1383 /* Nothing to do? */
1384 if (sb
->s_bdi
== &noop_backing_dev_info
)
1386 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1388 bdi_queue_work(sb
->s_bdi
, &work
);
1389 wait_for_completion(&done
);
1393 EXPORT_SYMBOL(sync_inodes_sb
);
1396 * write_inode_now - write an inode to disk
1397 * @inode: inode to write to disk
1398 * @sync: whether the write should be synchronous or not
1400 * This function commits an inode to disk immediately if it is dirty. This is
1401 * primarily needed by knfsd.
1403 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1405 int write_inode_now(struct inode
*inode
, int sync
)
1407 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1408 struct writeback_control wbc
= {
1409 .nr_to_write
= LONG_MAX
,
1410 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1412 .range_end
= LLONG_MAX
,
1415 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1416 wbc
.nr_to_write
= 0;
1419 return writeback_single_inode(inode
, wb
, &wbc
);
1421 EXPORT_SYMBOL(write_inode_now
);
1424 * sync_inode - write an inode and its pages to disk.
1425 * @inode: the inode to sync
1426 * @wbc: controls the writeback mode
1428 * sync_inode() will write an inode and its pages to disk. It will also
1429 * correctly update the inode on its superblock's dirty inode lists and will
1430 * update inode->i_state.
1432 * The caller must have a ref on the inode.
1434 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1436 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1438 EXPORT_SYMBOL(sync_inode
);
1441 * sync_inode_metadata - write an inode to disk
1442 * @inode: the inode to sync
1443 * @wait: wait for I/O to complete.
1445 * Write an inode to disk and adjust its dirty state after completion.
1447 * Note: only writes the actual inode, no associated data or other metadata.
1449 int sync_inode_metadata(struct inode
*inode
, int wait
)
1451 struct writeback_control wbc
= {
1452 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1453 .nr_to_write
= 0, /* metadata-only */
1456 return sync_inode(inode
, &wbc
);
1458 EXPORT_SYMBOL(sync_inode_metadata
);