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/module.h>
18 #include <linux/spinlock.h>
19 #include <linux/slab.h>
20 #include <linux/sched.h>
23 #include <linux/kthread.h>
24 #include <linux/freezer.h>
25 #include <linux/writeback.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/buffer_head.h>
29 #include <linux/tracepoint.h>
33 * Passed into wb_writeback(), essentially a subset of writeback_control
35 struct wb_writeback_work
{
37 struct super_block
*sb
;
38 enum writeback_sync_modes sync_mode
;
39 unsigned int tagged_writepages
:1;
40 unsigned int for_kupdate
:1;
41 unsigned int range_cyclic
:1;
42 unsigned int for_background
:1;
44 struct list_head list
; /* pending work list */
45 struct completion
*done
; /* set if the caller waits */
49 * Include the creation of the trace points after defining the
50 * wb_writeback_work structure so that the definition remains local to this
53 #define CREATE_TRACE_POINTS
54 #include <trace/events/writeback.h>
57 * We don't actually have pdflush, but this one is exported though /proc...
59 int nr_pdflush_threads
;
62 * writeback_in_progress - determine whether there is writeback in progress
63 * @bdi: the device's backing_dev_info structure.
65 * Determine whether there is writeback waiting to be handled against a
68 int writeback_in_progress(struct backing_dev_info
*bdi
)
70 return test_bit(BDI_writeback_running
, &bdi
->state
);
73 static inline struct backing_dev_info
*inode_to_bdi(struct inode
*inode
)
75 struct super_block
*sb
= inode
->i_sb
;
77 if (strcmp(sb
->s_type
->name
, "bdev") == 0)
78 return inode
->i_mapping
->backing_dev_info
;
83 static inline struct inode
*wb_inode(struct list_head
*head
)
85 return list_entry(head
, struct inode
, i_wb_list
);
88 /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
89 static void bdi_wakeup_flusher(struct backing_dev_info
*bdi
)
92 wake_up_process(bdi
->wb
.task
);
95 * The bdi thread isn't there, wake up the forker thread which
96 * will create and run it.
98 wake_up_process(default_backing_dev_info
.wb
.task
);
102 static void bdi_queue_work(struct backing_dev_info
*bdi
,
103 struct wb_writeback_work
*work
)
105 trace_writeback_queue(bdi
, work
);
107 spin_lock_bh(&bdi
->wb_lock
);
108 list_add_tail(&work
->list
, &bdi
->work_list
);
110 trace_writeback_nothread(bdi
, work
);
111 bdi_wakeup_flusher(bdi
);
112 spin_unlock_bh(&bdi
->wb_lock
);
116 __bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
119 struct wb_writeback_work
*work
;
122 * This is WB_SYNC_NONE writeback, so if allocation fails just
123 * wakeup the thread for old dirty data writeback
125 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
128 trace_writeback_nowork(bdi
);
129 wake_up_process(bdi
->wb
.task
);
134 work
->sync_mode
= WB_SYNC_NONE
;
135 work
->nr_pages
= nr_pages
;
136 work
->range_cyclic
= range_cyclic
;
138 bdi_queue_work(bdi
, work
);
142 * bdi_start_writeback - start writeback
143 * @bdi: the backing device to write from
144 * @nr_pages: the number of pages to write
147 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
148 * started when this function returns, we make no guarantees on
149 * completion. Caller need not hold sb s_umount semaphore.
152 void bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
)
154 __bdi_start_writeback(bdi
, nr_pages
, true);
158 * bdi_start_background_writeback - start background writeback
159 * @bdi: the backing device to write from
162 * This makes sure WB_SYNC_NONE background writeback happens. When
163 * this function returns, it is only guaranteed that for given BDI
164 * some IO is happening if we are over background dirty threshold.
165 * Caller need not hold sb s_umount semaphore.
167 void bdi_start_background_writeback(struct backing_dev_info
*bdi
)
170 * We just wake up the flusher thread. It will perform background
171 * writeback as soon as there is no other work to do.
173 trace_writeback_wake_background(bdi
);
174 spin_lock_bh(&bdi
->wb_lock
);
175 bdi_wakeup_flusher(bdi
);
176 spin_unlock_bh(&bdi
->wb_lock
);
180 * Remove the inode from the writeback list it is on.
182 void inode_wb_list_del(struct inode
*inode
)
184 struct backing_dev_info
*bdi
= inode_to_bdi(inode
);
186 spin_lock(&bdi
->wb
.list_lock
);
187 list_del_init(&inode
->i_wb_list
);
188 spin_unlock(&bdi
->wb
.list_lock
);
192 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
193 * furthest end of its superblock's dirty-inode list.
195 * Before stamping the inode's ->dirtied_when, we check to see whether it is
196 * already the most-recently-dirtied inode on the b_dirty list. If that is
197 * the case then the inode must have been redirtied while it was being written
198 * out and we don't reset its dirtied_when.
200 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
202 assert_spin_locked(&wb
->list_lock
);
203 if (!list_empty(&wb
->b_dirty
)) {
206 tail
= wb_inode(wb
->b_dirty
.next
);
207 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
208 inode
->dirtied_when
= jiffies
;
210 list_move(&inode
->i_wb_list
, &wb
->b_dirty
);
214 * requeue inode for re-scanning after bdi->b_io list is exhausted.
216 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
218 assert_spin_locked(&wb
->list_lock
);
219 list_move(&inode
->i_wb_list
, &wb
->b_more_io
);
222 static void inode_sync_complete(struct inode
*inode
)
225 * Prevent speculative execution through
226 * spin_unlock(&wb->list_lock);
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 dirty inodes from @delaying_queue to @dispatch_queue.
251 static int move_expired_inodes(struct list_head
*delaying_queue
,
252 struct list_head
*dispatch_queue
,
253 unsigned long *older_than_this
)
256 struct list_head
*pos
, *node
;
257 struct super_block
*sb
= NULL
;
262 while (!list_empty(delaying_queue
)) {
263 inode
= wb_inode(delaying_queue
->prev
);
264 if (older_than_this
&&
265 inode_dirtied_after(inode
, *older_than_this
))
267 if (sb
&& sb
!= inode
->i_sb
)
270 list_move(&inode
->i_wb_list
, &tmp
);
274 /* just one sb in list, splice to dispatch_queue and we're done */
276 list_splice(&tmp
, dispatch_queue
);
280 /* Move inodes from one superblock together */
281 while (!list_empty(&tmp
)) {
282 sb
= wb_inode(tmp
.prev
)->i_sb
;
283 list_for_each_prev_safe(pos
, node
, &tmp
) {
284 inode
= wb_inode(pos
);
285 if (inode
->i_sb
== sb
)
286 list_move(&inode
->i_wb_list
, dispatch_queue
);
294 * Queue all expired dirty inodes for io, eldest first.
296 * newly dirtied b_dirty b_io b_more_io
297 * =============> gf edc BA
299 * newly dirtied b_dirty b_io b_more_io
300 * =============> g fBAedc
302 * +--> dequeue for IO
304 static void queue_io(struct bdi_writeback
*wb
, unsigned long *older_than_this
)
307 assert_spin_locked(&wb
->list_lock
);
308 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
309 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, older_than_this
);
310 trace_writeback_queue_io(wb
, older_than_this
, moved
);
313 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
315 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
))
316 return inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
321 * Wait for writeback on an inode to complete.
323 static void inode_wait_for_writeback(struct inode
*inode
,
324 struct bdi_writeback
*wb
)
326 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
327 wait_queue_head_t
*wqh
;
329 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
330 while (inode
->i_state
& I_SYNC
) {
331 spin_unlock(&inode
->i_lock
);
332 spin_unlock(&wb
->list_lock
);
333 __wait_on_bit(wqh
, &wq
, inode_wait
, TASK_UNINTERRUPTIBLE
);
334 spin_lock(&wb
->list_lock
);
335 spin_lock(&inode
->i_lock
);
340 * Write out an inode's dirty pages. Called under wb->list_lock and
341 * inode->i_lock. Either the caller has an active reference on the inode or
342 * the inode has I_WILL_FREE set.
344 * If `wait' is set, wait on the writeout.
346 * The whole writeout design is quite complex and fragile. We want to avoid
347 * starvation of particular inodes when others are being redirtied, prevent
351 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
352 struct writeback_control
*wbc
)
354 struct address_space
*mapping
= inode
->i_mapping
;
355 long nr_to_write
= wbc
->nr_to_write
;
359 assert_spin_locked(&wb
->list_lock
);
360 assert_spin_locked(&inode
->i_lock
);
362 if (!atomic_read(&inode
->i_count
))
363 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
365 WARN_ON(inode
->i_state
& I_WILL_FREE
);
367 if (inode
->i_state
& I_SYNC
) {
369 * If this inode is locked for writeback and we are not doing
370 * writeback-for-data-integrity, move it to b_more_io so that
371 * writeback can proceed with the other inodes on s_io.
373 * We'll have another go at writing back this inode when we
374 * completed a full scan of b_io.
376 if (wbc
->sync_mode
!= WB_SYNC_ALL
) {
377 requeue_io(inode
, wb
);
378 trace_writeback_single_inode_requeue(inode
, wbc
,
384 * It's a data-integrity sync. We must wait.
386 inode_wait_for_writeback(inode
, wb
);
389 BUG_ON(inode
->i_state
& I_SYNC
);
391 /* Set I_SYNC, reset I_DIRTY_PAGES */
392 inode
->i_state
|= I_SYNC
;
393 inode
->i_state
&= ~I_DIRTY_PAGES
;
394 spin_unlock(&inode
->i_lock
);
395 spin_unlock(&wb
->list_lock
);
397 ret
= do_writepages(mapping
, wbc
);
400 * Make sure to wait on the data before writing out the metadata.
401 * This is important for filesystems that modify metadata on data
404 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
405 int err
= filemap_fdatawait(mapping
);
411 * Some filesystems may redirty the inode during the writeback
412 * due to delalloc, clear dirty metadata flags right before
415 spin_lock(&inode
->i_lock
);
416 dirty
= inode
->i_state
& I_DIRTY
;
417 inode
->i_state
&= ~(I_DIRTY_SYNC
| I_DIRTY_DATASYNC
);
418 spin_unlock(&inode
->i_lock
);
419 /* Don't write the inode if only I_DIRTY_PAGES was set */
420 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
421 int err
= write_inode(inode
, wbc
);
426 spin_lock(&wb
->list_lock
);
427 spin_lock(&inode
->i_lock
);
428 inode
->i_state
&= ~I_SYNC
;
429 if (!(inode
->i_state
& I_FREEING
)) {
431 * Sync livelock prevention. Each inode is tagged and synced in
432 * one shot. If still dirty, it will be redirty_tail()'ed below.
433 * Update the dirty time to prevent enqueue and sync it again.
435 if ((inode
->i_state
& I_DIRTY
) &&
436 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
437 inode
->dirtied_when
= jiffies
;
439 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
441 * We didn't write back all the pages. nfs_writepages()
442 * sometimes bales out without doing anything.
444 inode
->i_state
|= I_DIRTY_PAGES
;
445 if (wbc
->nr_to_write
<= 0) {
447 * slice used up: queue for next turn
449 requeue_io(inode
, wb
);
452 * Writeback blocked by something other than
453 * congestion. Delay the inode for some time to
454 * avoid spinning on the CPU (100% iowait)
455 * retrying writeback of the dirty page/inode
456 * that cannot be performed immediately.
458 redirty_tail(inode
, wb
);
460 } else if (inode
->i_state
& I_DIRTY
) {
462 * Filesystems can dirty the inode during writeback
463 * operations, such as delayed allocation during
464 * submission or metadata updates after data IO
467 redirty_tail(inode
, wb
);
470 * The inode is clean. At this point we either have
471 * a reference to the inode or it's on it's way out.
472 * No need to add it back to the LRU.
474 list_del_init(&inode
->i_wb_list
);
475 wbc
->inodes_written
++;
478 inode_sync_complete(inode
);
479 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
484 * For background writeback the caller does not have the sb pinned
485 * before calling writeback. So make sure that we do pin it, so it doesn't
486 * go away while we are writing inodes from it.
488 static bool pin_sb_for_writeback(struct super_block
*sb
)
491 if (list_empty(&sb
->s_instances
)) {
492 spin_unlock(&sb_lock
);
497 spin_unlock(&sb_lock
);
499 if (down_read_trylock(&sb
->s_umount
)) {
502 up_read(&sb
->s_umount
);
510 * Write a portion of b_io inodes which belong to @sb.
512 * If @only_this_sb is true, then find and write all such
513 * inodes. Otherwise write only ones which go sequentially
516 * Return 1, if the caller writeback routine should be
517 * interrupted. Otherwise return 0.
519 static int writeback_sb_inodes(struct super_block
*sb
, struct bdi_writeback
*wb
,
520 struct writeback_control
*wbc
, bool only_this_sb
)
522 while (!list_empty(&wb
->b_io
)) {
524 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
526 if (inode
->i_sb
!= sb
) {
529 * We only want to write back data for this
530 * superblock, move all inodes not belonging
531 * to it back onto the dirty list.
533 redirty_tail(inode
, wb
);
538 * The inode belongs to a different superblock.
539 * Bounce back to the caller to unpin this and
540 * pin the next superblock.
546 * Don't bother with new inodes or inodes beeing freed, first
547 * kind does not need peridic writeout yet, and for the latter
548 * kind writeout is handled by the freer.
550 spin_lock(&inode
->i_lock
);
551 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
552 spin_unlock(&inode
->i_lock
);
553 requeue_io(inode
, wb
);
559 pages_skipped
= wbc
->pages_skipped
;
560 writeback_single_inode(inode
, wb
, wbc
);
561 if (wbc
->pages_skipped
!= pages_skipped
) {
563 * writeback is not making progress due to locked
564 * buffers. Skip this inode for now.
566 redirty_tail(inode
, wb
);
568 spin_unlock(&inode
->i_lock
);
569 spin_unlock(&wb
->list_lock
);
572 spin_lock(&wb
->list_lock
);
573 if (wbc
->nr_to_write
<= 0)
580 static void __writeback_inodes_wb(struct bdi_writeback
*wb
,
581 struct writeback_control
*wbc
)
585 while (!list_empty(&wb
->b_io
)) {
586 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
587 struct super_block
*sb
= inode
->i_sb
;
589 if (!pin_sb_for_writeback(sb
)) {
590 requeue_io(inode
, wb
);
593 ret
= writeback_sb_inodes(sb
, wb
, wbc
, false);
599 /* Leave any unwritten inodes on b_io */
602 void writeback_inodes_wb(struct bdi_writeback
*wb
,
603 struct writeback_control
*wbc
)
605 spin_lock(&wb
->list_lock
);
606 if (list_empty(&wb
->b_io
))
607 queue_io(wb
, wbc
->older_than_this
);
608 __writeback_inodes_wb(wb
, wbc
);
609 spin_unlock(&wb
->list_lock
);
613 * The maximum number of pages to writeout in a single bdi flush/kupdate
614 * operation. We do this so we don't hold I_SYNC against an inode for
615 * enormous amounts of time, which would block a userspace task which has
616 * been forced to throttle against that inode. Also, the code reevaluates
617 * the dirty each time it has written this many pages.
619 #define MAX_WRITEBACK_PAGES 1024
621 static inline bool over_bground_thresh(void)
623 unsigned long background_thresh
, dirty_thresh
;
625 global_dirty_limits(&background_thresh
, &dirty_thresh
);
627 return (global_page_state(NR_FILE_DIRTY
) +
628 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
);
632 * Explicit flushing or periodic writeback of "old" data.
634 * Define "old": the first time one of an inode's pages is dirtied, we mark the
635 * dirtying-time in the inode's address_space. So this periodic writeback code
636 * just walks the superblock inode list, writing back any inodes which are
637 * older than a specific point in time.
639 * Try to run once per dirty_writeback_interval. But if a writeback event
640 * takes longer than a dirty_writeback_interval interval, then leave a
643 * older_than_this takes precedence over nr_to_write. So we'll only write back
644 * all dirty pages if they are all attached to "old" mappings.
646 static long wb_writeback(struct bdi_writeback
*wb
,
647 struct wb_writeback_work
*work
)
649 struct writeback_control wbc
= {
650 .sync_mode
= work
->sync_mode
,
651 .tagged_writepages
= work
->tagged_writepages
,
652 .older_than_this
= NULL
,
653 .for_kupdate
= work
->for_kupdate
,
654 .for_background
= work
->for_background
,
655 .range_cyclic
= work
->range_cyclic
,
657 unsigned long oldest_jif
;
659 long write_chunk
= MAX_WRITEBACK_PAGES
;
662 if (!wbc
.range_cyclic
) {
664 wbc
.range_end
= LLONG_MAX
;
668 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
669 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
670 * here avoids calling into writeback_inodes_wb() more than once.
672 * The intended call sequence for WB_SYNC_ALL writeback is:
675 * writeback_sb_inodes() <== called only once
676 * write_cache_pages() <== called once for each inode
677 * (quickly) tag currently dirty pages
678 * (maybe slowly) sync all tagged pages
680 if (wbc
.sync_mode
== WB_SYNC_ALL
|| wbc
.tagged_writepages
)
681 write_chunk
= LONG_MAX
;
683 oldest_jif
= jiffies
;
684 wbc
.older_than_this
= &oldest_jif
;
686 spin_lock(&wb
->list_lock
);
689 * Stop writeback when nr_pages has been consumed
691 if (work
->nr_pages
<= 0)
695 * Background writeout and kupdate-style writeback may
696 * run forever. Stop them if there is other work to do
697 * so that e.g. sync can proceed. They'll be restarted
698 * after the other works are all done.
700 if ((work
->for_background
|| work
->for_kupdate
) &&
701 !list_empty(&wb
->bdi
->work_list
))
705 * For background writeout, stop when we are below the
706 * background dirty threshold
708 if (work
->for_background
&& !over_bground_thresh())
711 if (work
->for_kupdate
) {
712 oldest_jif
= jiffies
-
713 msecs_to_jiffies(dirty_expire_interval
* 10);
714 wbc
.older_than_this
= &oldest_jif
;
717 wbc
.nr_to_write
= write_chunk
;
718 wbc
.pages_skipped
= 0;
719 wbc
.inodes_written
= 0;
721 trace_wbc_writeback_start(&wbc
, wb
->bdi
);
722 if (list_empty(&wb
->b_io
))
723 queue_io(wb
, wbc
.older_than_this
);
725 writeback_sb_inodes(work
->sb
, wb
, &wbc
, true);
727 __writeback_inodes_wb(wb
, &wbc
);
728 trace_wbc_writeback_written(&wbc
, wb
->bdi
);
730 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
731 wrote
+= write_chunk
- wbc
.nr_to_write
;
734 * Did we write something? Try for more
736 * Dirty inodes are moved to b_io for writeback in batches.
737 * The completion of the current batch does not necessarily
738 * mean the overall work is done. So we keep looping as long
739 * as made some progress on cleaning pages or inodes.
741 if (wbc
.nr_to_write
< write_chunk
)
743 if (wbc
.inodes_written
)
746 * No more inodes for IO, bail
748 if (list_empty(&wb
->b_more_io
))
751 * Nothing written. Wait for some inode to
752 * become available for writeback. Otherwise
753 * we'll just busyloop.
755 if (!list_empty(&wb
->b_more_io
)) {
756 inode
= wb_inode(wb
->b_more_io
.prev
);
757 trace_wbc_writeback_wait(&wbc
, wb
->bdi
);
758 spin_lock(&inode
->i_lock
);
759 inode_wait_for_writeback(inode
, wb
);
760 spin_unlock(&inode
->i_lock
);
763 spin_unlock(&wb
->list_lock
);
769 * Return the next wb_writeback_work struct that hasn't been processed yet.
771 static struct wb_writeback_work
*
772 get_next_work_item(struct backing_dev_info
*bdi
)
774 struct wb_writeback_work
*work
= NULL
;
776 spin_lock_bh(&bdi
->wb_lock
);
777 if (!list_empty(&bdi
->work_list
)) {
778 work
= list_entry(bdi
->work_list
.next
,
779 struct wb_writeback_work
, list
);
780 list_del_init(&work
->list
);
782 spin_unlock_bh(&bdi
->wb_lock
);
787 * Add in the number of potentially dirty inodes, because each inode
788 * write can dirty pagecache in the underlying blockdev.
790 static unsigned long get_nr_dirty_pages(void)
792 return global_page_state(NR_FILE_DIRTY
) +
793 global_page_state(NR_UNSTABLE_NFS
) +
794 get_nr_dirty_inodes();
797 static long wb_check_background_flush(struct bdi_writeback
*wb
)
799 if (over_bground_thresh()) {
801 struct wb_writeback_work work
= {
802 .nr_pages
= LONG_MAX
,
803 .sync_mode
= WB_SYNC_NONE
,
808 return wb_writeback(wb
, &work
);
814 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
816 unsigned long expired
;
820 * When set to zero, disable periodic writeback
822 if (!dirty_writeback_interval
)
825 expired
= wb
->last_old_flush
+
826 msecs_to_jiffies(dirty_writeback_interval
* 10);
827 if (time_before(jiffies
, expired
))
830 wb
->last_old_flush
= jiffies
;
831 nr_pages
= get_nr_dirty_pages();
834 struct wb_writeback_work work
= {
835 .nr_pages
= nr_pages
,
836 .sync_mode
= WB_SYNC_NONE
,
841 return wb_writeback(wb
, &work
);
848 * Retrieve work items and do the writeback they describe
850 long wb_do_writeback(struct bdi_writeback
*wb
, int force_wait
)
852 struct backing_dev_info
*bdi
= wb
->bdi
;
853 struct wb_writeback_work
*work
;
856 set_bit(BDI_writeback_running
, &wb
->bdi
->state
);
857 while ((work
= get_next_work_item(bdi
)) != NULL
) {
859 * Override sync mode, in case we must wait for completion
860 * because this thread is exiting now.
863 work
->sync_mode
= WB_SYNC_ALL
;
865 trace_writeback_exec(bdi
, work
);
867 wrote
+= wb_writeback(wb
, work
);
870 * Notify the caller of completion if this is a synchronous
871 * work item, otherwise just free it.
874 complete(work
->done
);
880 * Check for periodic writeback, kupdated() style
882 wrote
+= wb_check_old_data_flush(wb
);
883 wrote
+= wb_check_background_flush(wb
);
884 clear_bit(BDI_writeback_running
, &wb
->bdi
->state
);
890 * Handle writeback of dirty data for the device backed by this bdi. Also
891 * wakes up periodically and does kupdated style flushing.
893 int bdi_writeback_thread(void *data
)
895 struct bdi_writeback
*wb
= data
;
896 struct backing_dev_info
*bdi
= wb
->bdi
;
899 current
->flags
|= PF_SWAPWRITE
;
901 wb
->last_active
= jiffies
;
904 * Our parent may run at a different priority, just set us to normal
906 set_user_nice(current
, 0);
908 trace_writeback_thread_start(bdi
);
910 while (!kthread_should_stop()) {
912 * Remove own delayed wake-up timer, since we are already awake
913 * and we'll take care of the preriodic write-back.
915 del_timer(&wb
->wakeup_timer
);
917 pages_written
= wb_do_writeback(wb
, 0);
919 trace_writeback_pages_written(pages_written
);
922 wb
->last_active
= jiffies
;
924 set_current_state(TASK_INTERRUPTIBLE
);
925 if (!list_empty(&bdi
->work_list
) || kthread_should_stop()) {
926 __set_current_state(TASK_RUNNING
);
930 if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
931 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval
* 10));
934 * We have nothing to do, so can go sleep without any
935 * timeout and save power. When a work is queued or
936 * something is made dirty - we will be woken up.
944 /* Flush any work that raced with us exiting */
945 if (!list_empty(&bdi
->work_list
))
946 wb_do_writeback(wb
, 1);
948 trace_writeback_thread_stop(bdi
);
954 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
957 void wakeup_flusher_threads(long nr_pages
)
959 struct backing_dev_info
*bdi
;
962 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
963 global_page_state(NR_UNSTABLE_NFS
);
967 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
968 if (!bdi_has_dirty_io(bdi
))
970 __bdi_start_writeback(bdi
, nr_pages
, false);
975 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
977 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
978 struct dentry
*dentry
;
979 const char *name
= "?";
981 dentry
= d_find_alias(inode
);
983 spin_lock(&dentry
->d_lock
);
984 name
= (const char *) dentry
->d_name
.name
;
987 "%s(%d): dirtied inode %lu (%s) on %s\n",
988 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
989 name
, inode
->i_sb
->s_id
);
991 spin_unlock(&dentry
->d_lock
);
998 * __mark_inode_dirty - internal function
999 * @inode: inode to mark
1000 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1001 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1002 * mark_inode_dirty_sync.
1004 * Put the inode on the super block's dirty list.
1006 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1007 * dirty list only if it is hashed or if it refers to a blockdev.
1008 * If it was not hashed, it will never be added to the dirty list
1009 * even if it is later hashed, as it will have been marked dirty already.
1011 * In short, make sure you hash any inodes _before_ you start marking
1014 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1015 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1016 * the kernel-internal blockdev inode represents the dirtying time of the
1017 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1018 * page->mapping->host, so the page-dirtying time is recorded in the internal
1021 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1023 struct super_block
*sb
= inode
->i_sb
;
1024 struct backing_dev_info
*bdi
= NULL
;
1027 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1028 * dirty the inode itself
1030 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
1031 if (sb
->s_op
->dirty_inode
)
1032 sb
->s_op
->dirty_inode(inode
, flags
);
1036 * make sure that changes are seen by all cpus before we test i_state
1041 /* avoid the locking if we can */
1042 if ((inode
->i_state
& flags
) == flags
)
1045 if (unlikely(block_dump
))
1046 block_dump___mark_inode_dirty(inode
);
1048 spin_lock(&inode
->i_lock
);
1049 if ((inode
->i_state
& flags
) != flags
) {
1050 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1052 inode
->i_state
|= flags
;
1055 * If the inode is being synced, just update its dirty state.
1056 * The unlocker will place the inode on the appropriate
1057 * superblock list, based upon its state.
1059 if (inode
->i_state
& I_SYNC
)
1060 goto out_unlock_inode
;
1063 * Only add valid (hashed) inodes to the superblock's
1064 * dirty list. Add blockdev inodes as well.
1066 if (!S_ISBLK(inode
->i_mode
)) {
1067 if (inode_unhashed(inode
))
1068 goto out_unlock_inode
;
1070 if (inode
->i_state
& I_FREEING
)
1071 goto out_unlock_inode
;
1074 * If the inode was already on b_dirty/b_io/b_more_io, don't
1075 * reposition it (that would break b_dirty time-ordering).
1078 bool wakeup_bdi
= false;
1079 bdi
= inode_to_bdi(inode
);
1081 if (bdi_cap_writeback_dirty(bdi
)) {
1082 WARN(!test_bit(BDI_registered
, &bdi
->state
),
1083 "bdi-%s not registered\n", bdi
->name
);
1086 * If this is the first dirty inode for this
1087 * bdi, we have to wake-up the corresponding
1088 * bdi thread to make sure background
1089 * write-back happens later.
1091 if (!wb_has_dirty_io(&bdi
->wb
))
1095 spin_unlock(&inode
->i_lock
);
1096 spin_lock(&bdi
->wb
.list_lock
);
1097 inode
->dirtied_when
= jiffies
;
1098 list_move(&inode
->i_wb_list
, &bdi
->wb
.b_dirty
);
1099 spin_unlock(&bdi
->wb
.list_lock
);
1102 bdi_wakeup_thread_delayed(bdi
);
1107 spin_unlock(&inode
->i_lock
);
1110 EXPORT_SYMBOL(__mark_inode_dirty
);
1113 * Write out a superblock's list of dirty inodes. A wait will be performed
1114 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1116 * If older_than_this is non-NULL, then only write out inodes which
1117 * had their first dirtying at a time earlier than *older_than_this.
1119 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1120 * This function assumes that the blockdev superblock's inodes are backed by
1121 * a variety of queues, so all inodes are searched. For other superblocks,
1122 * assume that all inodes are backed by the same queue.
1124 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1125 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1126 * on the writer throttling path, and we get decent balancing between many
1127 * throttled threads: we don't want them all piling up on inode_sync_wait.
1129 static void wait_sb_inodes(struct super_block
*sb
)
1131 struct inode
*inode
, *old_inode
= NULL
;
1134 * We need to be protected against the filesystem going from
1135 * r/o to r/w or vice versa.
1137 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1139 spin_lock(&inode_sb_list_lock
);
1142 * Data integrity sync. Must wait for all pages under writeback,
1143 * because there may have been pages dirtied before our sync
1144 * call, but which had writeout started before we write it out.
1145 * In which case, the inode may not be on the dirty list, but
1146 * we still have to wait for that writeout.
1148 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1149 struct address_space
*mapping
= inode
->i_mapping
;
1151 spin_lock(&inode
->i_lock
);
1152 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1153 (mapping
->nrpages
== 0)) {
1154 spin_unlock(&inode
->i_lock
);
1158 spin_unlock(&inode
->i_lock
);
1159 spin_unlock(&inode_sb_list_lock
);
1162 * We hold a reference to 'inode' so it couldn't have been
1163 * removed from s_inodes list while we dropped the
1164 * inode_sb_list_lock. We cannot iput the inode now as we can
1165 * be holding the last reference and we cannot iput it under
1166 * inode_sb_list_lock. So we keep the reference and iput it
1172 filemap_fdatawait(mapping
);
1176 spin_lock(&inode_sb_list_lock
);
1178 spin_unlock(&inode_sb_list_lock
);
1183 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1184 * @sb: the superblock
1185 * @nr: the number of pages to write
1187 * Start writeback on some inodes on this super_block. No guarantees are made
1188 * on how many (if any) will be written, and this function does not wait
1189 * for IO completion of submitted IO.
1191 void writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
)
1193 DECLARE_COMPLETION_ONSTACK(done
);
1194 struct wb_writeback_work work
= {
1196 .sync_mode
= WB_SYNC_NONE
,
1197 .tagged_writepages
= 1,
1202 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1203 bdi_queue_work(sb
->s_bdi
, &work
);
1204 wait_for_completion(&done
);
1206 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1209 * writeback_inodes_sb - writeback dirty inodes from given super_block
1210 * @sb: the superblock
1212 * Start writeback on some inodes on this super_block. No guarantees are made
1213 * on how many (if any) will be written, and this function does not wait
1214 * for IO completion of submitted IO.
1216 void writeback_inodes_sb(struct super_block
*sb
)
1218 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages());
1220 EXPORT_SYMBOL(writeback_inodes_sb
);
1223 * writeback_inodes_sb_if_idle - start writeback if none underway
1224 * @sb: the superblock
1226 * Invoke writeback_inodes_sb if no writeback is currently underway.
1227 * Returns 1 if writeback was started, 0 if not.
1229 int writeback_inodes_sb_if_idle(struct super_block
*sb
)
1231 if (!writeback_in_progress(sb
->s_bdi
)) {
1232 down_read(&sb
->s_umount
);
1233 writeback_inodes_sb(sb
);
1234 up_read(&sb
->s_umount
);
1239 EXPORT_SYMBOL(writeback_inodes_sb_if_idle
);
1242 * writeback_inodes_sb_if_idle - start writeback if none underway
1243 * @sb: the superblock
1244 * @nr: the number of pages to write
1246 * Invoke writeback_inodes_sb if no writeback is currently underway.
1247 * Returns 1 if writeback was started, 0 if not.
1249 int writeback_inodes_sb_nr_if_idle(struct super_block
*sb
,
1252 if (!writeback_in_progress(sb
->s_bdi
)) {
1253 down_read(&sb
->s_umount
);
1254 writeback_inodes_sb_nr(sb
, nr
);
1255 up_read(&sb
->s_umount
);
1260 EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle
);
1263 * sync_inodes_sb - sync sb inode pages
1264 * @sb: the superblock
1266 * This function writes and waits on any dirty inode belonging to this
1269 void sync_inodes_sb(struct super_block
*sb
)
1271 DECLARE_COMPLETION_ONSTACK(done
);
1272 struct wb_writeback_work work
= {
1274 .sync_mode
= WB_SYNC_ALL
,
1275 .nr_pages
= LONG_MAX
,
1280 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1282 bdi_queue_work(sb
->s_bdi
, &work
);
1283 wait_for_completion(&done
);
1287 EXPORT_SYMBOL(sync_inodes_sb
);
1290 * write_inode_now - write an inode to disk
1291 * @inode: inode to write to disk
1292 * @sync: whether the write should be synchronous or not
1294 * This function commits an inode to disk immediately if it is dirty. This is
1295 * primarily needed by knfsd.
1297 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1299 int write_inode_now(struct inode
*inode
, int sync
)
1301 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1303 struct writeback_control wbc
= {
1304 .nr_to_write
= LONG_MAX
,
1305 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1307 .range_end
= LLONG_MAX
,
1310 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1311 wbc
.nr_to_write
= 0;
1314 spin_lock(&wb
->list_lock
);
1315 spin_lock(&inode
->i_lock
);
1316 ret
= writeback_single_inode(inode
, wb
, &wbc
);
1317 spin_unlock(&inode
->i_lock
);
1318 spin_unlock(&wb
->list_lock
);
1320 inode_sync_wait(inode
);
1323 EXPORT_SYMBOL(write_inode_now
);
1326 * sync_inode - write an inode and its pages to disk.
1327 * @inode: the inode to sync
1328 * @wbc: controls the writeback mode
1330 * sync_inode() will write an inode and its pages to disk. It will also
1331 * correctly update the inode on its superblock's dirty inode lists and will
1332 * update inode->i_state.
1334 * The caller must have a ref on the inode.
1336 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1338 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1341 spin_lock(&wb
->list_lock
);
1342 spin_lock(&inode
->i_lock
);
1343 ret
= writeback_single_inode(inode
, wb
, wbc
);
1344 spin_unlock(&inode
->i_lock
);
1345 spin_unlock(&wb
->list_lock
);
1348 EXPORT_SYMBOL(sync_inode
);
1351 * sync_inode_metadata - write an inode to disk
1352 * @inode: the inode to sync
1353 * @wait: wait for I/O to complete.
1355 * Write an inode to disk and adjust its dirty state after completion.
1357 * Note: only writes the actual inode, no associated data or other metadata.
1359 int sync_inode_metadata(struct inode
*inode
, int wait
)
1361 struct writeback_control wbc
= {
1362 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1363 .nr_to_write
= 0, /* metadata-only */
1366 return sync_inode(inode
, &wbc
);
1368 EXPORT_SYMBOL(sync_inode_metadata
);