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/tracepoint.h>
32 * Passed into wb_writeback(), essentially a subset of writeback_control
34 struct wb_writeback_work
{
36 struct super_block
*sb
;
37 unsigned long *older_than_this
;
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;
43 enum wb_reason reason
; /* why was writeback initiated? */
45 struct list_head list
; /* pending work list */
46 struct completion
*done
; /* set if the caller waits */
50 * Include the creation of the trace points after defining the
51 * wb_writeback_work structure so that the definition remains local to this
54 #define CREATE_TRACE_POINTS
55 #include <trace/events/writeback.h>
58 * We don't actually have pdflush, but this one is exported though /proc...
60 int nr_pdflush_threads
;
63 * writeback_in_progress - determine whether there is writeback in progress
64 * @bdi: the device's backing_dev_info structure.
66 * Determine whether there is writeback waiting to be handled against a
69 int writeback_in_progress(struct backing_dev_info
*bdi
)
71 return test_bit(BDI_writeback_running
, &bdi
->state
);
74 static inline struct backing_dev_info
*inode_to_bdi(struct inode
*inode
)
76 struct super_block
*sb
= inode
->i_sb
;
78 if (strcmp(sb
->s_type
->name
, "bdev") == 0)
79 return inode
->i_mapping
->backing_dev_info
;
84 static inline struct inode
*wb_inode(struct list_head
*head
)
86 return list_entry(head
, struct inode
, i_wb_list
);
89 /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
90 static void bdi_wakeup_flusher(struct backing_dev_info
*bdi
)
93 wake_up_process(bdi
->wb
.task
);
96 * The bdi thread isn't there, wake up the forker thread which
97 * will create and run it.
99 wake_up_process(default_backing_dev_info
.wb
.task
);
103 static void bdi_queue_work(struct backing_dev_info
*bdi
,
104 struct wb_writeback_work
*work
)
106 trace_writeback_queue(bdi
, work
);
108 spin_lock_bh(&bdi
->wb_lock
);
109 list_add_tail(&work
->list
, &bdi
->work_list
);
111 trace_writeback_nothread(bdi
, work
);
112 bdi_wakeup_flusher(bdi
);
113 spin_unlock_bh(&bdi
->wb_lock
);
117 __bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
118 bool range_cyclic
, enum wb_reason reason
)
120 struct wb_writeback_work
*work
;
123 * This is WB_SYNC_NONE writeback, so if allocation fails just
124 * wakeup the thread for old dirty data writeback
126 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
129 trace_writeback_nowork(bdi
);
130 wake_up_process(bdi
->wb
.task
);
135 work
->sync_mode
= WB_SYNC_NONE
;
136 work
->nr_pages
= nr_pages
;
137 work
->range_cyclic
= range_cyclic
;
138 work
->reason
= reason
;
140 bdi_queue_work(bdi
, work
);
144 * bdi_start_writeback - start writeback
145 * @bdi: the backing device to write from
146 * @nr_pages: the number of pages to write
147 * @reason: reason why some writeback work was initiated
150 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
151 * started when this function returns, we make no guarantees on
152 * completion. Caller need not hold sb s_umount semaphore.
155 void bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
156 enum wb_reason reason
)
158 __bdi_start_writeback(bdi
, nr_pages
, true, reason
);
162 * bdi_start_background_writeback - start background writeback
163 * @bdi: the backing device to write from
166 * This makes sure WB_SYNC_NONE background writeback happens. When
167 * this function returns, it is only guaranteed that for given BDI
168 * some IO is happening if we are over background dirty threshold.
169 * Caller need not hold sb s_umount semaphore.
171 void bdi_start_background_writeback(struct backing_dev_info
*bdi
)
174 * We just wake up the flusher thread. It will perform background
175 * writeback as soon as there is no other work to do.
177 trace_writeback_wake_background(bdi
);
178 spin_lock_bh(&bdi
->wb_lock
);
179 bdi_wakeup_flusher(bdi
);
180 spin_unlock_bh(&bdi
->wb_lock
);
184 * Remove the inode from the writeback list it is on.
186 void inode_wb_list_del(struct inode
*inode
)
188 struct backing_dev_info
*bdi
= inode_to_bdi(inode
);
190 spin_lock(&bdi
->wb
.list_lock
);
191 list_del_init(&inode
->i_wb_list
);
192 spin_unlock(&bdi
->wb
.list_lock
);
196 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
197 * furthest end of its superblock's dirty-inode list.
199 * Before stamping the inode's ->dirtied_when, we check to see whether it is
200 * already the most-recently-dirtied inode on the b_dirty list. If that is
201 * the case then the inode must have been redirtied while it was being written
202 * out and we don't reset its dirtied_when.
204 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
206 assert_spin_locked(&wb
->list_lock
);
207 if (!list_empty(&wb
->b_dirty
)) {
210 tail
= wb_inode(wb
->b_dirty
.next
);
211 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
212 inode
->dirtied_when
= jiffies
;
214 list_move(&inode
->i_wb_list
, &wb
->b_dirty
);
218 * requeue inode for re-scanning after bdi->b_io list is exhausted.
220 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
222 assert_spin_locked(&wb
->list_lock
);
223 list_move(&inode
->i_wb_list
, &wb
->b_more_io
);
226 static void inode_sync_complete(struct inode
*inode
)
229 * Prevent speculative execution through
230 * spin_unlock(&wb->list_lock);
234 wake_up_bit(&inode
->i_state
, __I_SYNC
);
237 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
239 bool ret
= time_after(inode
->dirtied_when
, t
);
242 * For inodes being constantly redirtied, dirtied_when can get stuck.
243 * It _appears_ to be in the future, but is actually in distant past.
244 * This test is necessary to prevent such wrapped-around relative times
245 * from permanently stopping the whole bdi writeback.
247 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
253 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
255 static int move_expired_inodes(struct list_head
*delaying_queue
,
256 struct list_head
*dispatch_queue
,
257 struct wb_writeback_work
*work
)
260 struct list_head
*pos
, *node
;
261 struct super_block
*sb
= NULL
;
266 while (!list_empty(delaying_queue
)) {
267 inode
= wb_inode(delaying_queue
->prev
);
268 if (work
->older_than_this
&&
269 inode_dirtied_after(inode
, *work
->older_than_this
))
271 if (sb
&& sb
!= inode
->i_sb
)
274 list_move(&inode
->i_wb_list
, &tmp
);
278 /* just one sb in list, splice to dispatch_queue and we're done */
280 list_splice(&tmp
, dispatch_queue
);
284 /* Move inodes from one superblock together */
285 while (!list_empty(&tmp
)) {
286 sb
= wb_inode(tmp
.prev
)->i_sb
;
287 list_for_each_prev_safe(pos
, node
, &tmp
) {
288 inode
= wb_inode(pos
);
289 if (inode
->i_sb
== sb
)
290 list_move(&inode
->i_wb_list
, dispatch_queue
);
298 * Queue all expired dirty inodes for io, eldest first.
300 * newly dirtied b_dirty b_io b_more_io
301 * =============> gf edc BA
303 * newly dirtied b_dirty b_io b_more_io
304 * =============> g fBAedc
306 * +--> dequeue for IO
308 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
311 assert_spin_locked(&wb
->list_lock
);
312 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
313 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, work
);
314 trace_writeback_queue_io(wb
, work
, moved
);
317 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
319 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
))
320 return inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
325 * Wait for writeback on an inode to complete.
327 static void inode_wait_for_writeback(struct inode
*inode
,
328 struct bdi_writeback
*wb
)
330 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
331 wait_queue_head_t
*wqh
;
333 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
334 while (inode
->i_state
& I_SYNC
) {
335 spin_unlock(&inode
->i_lock
);
336 spin_unlock(&wb
->list_lock
);
337 __wait_on_bit(wqh
, &wq
, inode_wait
, TASK_UNINTERRUPTIBLE
);
338 spin_lock(&wb
->list_lock
);
339 spin_lock(&inode
->i_lock
);
344 * Write out an inode's dirty pages. Called under wb->list_lock and
345 * inode->i_lock. Either the caller has an active reference on the inode or
346 * the inode has I_WILL_FREE set.
348 * If `wait' is set, wait on the writeout.
350 * The whole writeout design is quite complex and fragile. We want to avoid
351 * starvation of particular inodes when others are being redirtied, prevent
355 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
356 struct writeback_control
*wbc
)
358 struct address_space
*mapping
= inode
->i_mapping
;
359 long nr_to_write
= wbc
->nr_to_write
;
363 assert_spin_locked(&wb
->list_lock
);
364 assert_spin_locked(&inode
->i_lock
);
366 if (!atomic_read(&inode
->i_count
))
367 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
369 WARN_ON(inode
->i_state
& I_WILL_FREE
);
371 if (inode
->i_state
& I_SYNC
) {
373 * If this inode is locked for writeback and we are not doing
374 * writeback-for-data-integrity, move it to b_more_io so that
375 * writeback can proceed with the other inodes on s_io.
377 * We'll have another go at writing back this inode when we
378 * completed a full scan of b_io.
380 if (wbc
->sync_mode
!= WB_SYNC_ALL
) {
381 requeue_io(inode
, wb
);
382 trace_writeback_single_inode_requeue(inode
, wbc
,
388 * It's a data-integrity sync. We must wait.
390 inode_wait_for_writeback(inode
, wb
);
393 BUG_ON(inode
->i_state
& I_SYNC
);
395 /* Set I_SYNC, reset I_DIRTY_PAGES */
396 inode
->i_state
|= I_SYNC
;
397 inode
->i_state
&= ~I_DIRTY_PAGES
;
398 spin_unlock(&inode
->i_lock
);
399 spin_unlock(&wb
->list_lock
);
401 ret
= do_writepages(mapping
, wbc
);
404 * Make sure to wait on the data before writing out the metadata.
405 * This is important for filesystems that modify metadata on data
408 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
409 int err
= filemap_fdatawait(mapping
);
415 * Some filesystems may redirty the inode during the writeback
416 * due to delalloc, clear dirty metadata flags right before
419 spin_lock(&inode
->i_lock
);
420 dirty
= inode
->i_state
& I_DIRTY
;
421 inode
->i_state
&= ~(I_DIRTY_SYNC
| I_DIRTY_DATASYNC
);
422 spin_unlock(&inode
->i_lock
);
423 /* Don't write the inode if only I_DIRTY_PAGES was set */
424 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
425 int err
= write_inode(inode
, wbc
);
430 spin_lock(&wb
->list_lock
);
431 spin_lock(&inode
->i_lock
);
432 inode
->i_state
&= ~I_SYNC
;
433 if (!(inode
->i_state
& I_FREEING
)) {
435 * Sync livelock prevention. Each inode is tagged and synced in
436 * one shot. If still dirty, it will be redirty_tail()'ed below.
437 * Update the dirty time to prevent enqueue and sync it again.
439 if ((inode
->i_state
& I_DIRTY
) &&
440 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
441 inode
->dirtied_when
= jiffies
;
443 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
445 * We didn't write back all the pages. nfs_writepages()
446 * sometimes bales out without doing anything.
448 inode
->i_state
|= I_DIRTY_PAGES
;
449 if (wbc
->nr_to_write
<= 0) {
451 * slice used up: queue for next turn
453 requeue_io(inode
, wb
);
456 * Writeback blocked by something other than
457 * congestion. Delay the inode for some time to
458 * avoid spinning on the CPU (100% iowait)
459 * retrying writeback of the dirty page/inode
460 * that cannot be performed immediately.
462 redirty_tail(inode
, wb
);
464 } else if (inode
->i_state
& I_DIRTY
) {
466 * Filesystems can dirty the inode during writeback
467 * operations, such as delayed allocation during
468 * submission or metadata updates after data IO
471 redirty_tail(inode
, wb
);
474 * The inode is clean. At this point we either have
475 * a reference to the inode or it's on it's way out.
476 * No need to add it back to the LRU.
478 list_del_init(&inode
->i_wb_list
);
481 inode_sync_complete(inode
);
482 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
486 static long writeback_chunk_size(struct backing_dev_info
*bdi
,
487 struct wb_writeback_work
*work
)
492 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
493 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
494 * here avoids calling into writeback_inodes_wb() more than once.
496 * The intended call sequence for WB_SYNC_ALL writeback is:
499 * writeback_sb_inodes() <== called only once
500 * write_cache_pages() <== called once for each inode
501 * (quickly) tag currently dirty pages
502 * (maybe slowly) sync all tagged pages
504 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
507 pages
= min(bdi
->avg_write_bandwidth
/ 2,
508 global_dirty_limit
/ DIRTY_SCOPE
);
509 pages
= min(pages
, work
->nr_pages
);
510 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
511 MIN_WRITEBACK_PAGES
);
518 * Write a portion of b_io inodes which belong to @sb.
520 * If @only_this_sb is true, then find and write all such
521 * inodes. Otherwise write only ones which go sequentially
524 * Return the number of pages and/or inodes written.
526 static long writeback_sb_inodes(struct super_block
*sb
,
527 struct bdi_writeback
*wb
,
528 struct wb_writeback_work
*work
)
530 struct writeback_control wbc
= {
531 .sync_mode
= work
->sync_mode
,
532 .tagged_writepages
= work
->tagged_writepages
,
533 .for_kupdate
= work
->for_kupdate
,
534 .for_background
= work
->for_background
,
535 .range_cyclic
= work
->range_cyclic
,
537 .range_end
= LLONG_MAX
,
539 unsigned long start_time
= jiffies
;
541 long wrote
= 0; /* count both pages and inodes */
543 while (!list_empty(&wb
->b_io
)) {
544 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
546 if (inode
->i_sb
!= sb
) {
549 * We only want to write back data for this
550 * superblock, move all inodes not belonging
551 * to it back onto the dirty list.
553 redirty_tail(inode
, wb
);
558 * The inode belongs to a different superblock.
559 * Bounce back to the caller to unpin this and
560 * pin the next superblock.
566 * Don't bother with new inodes or inodes beeing freed, first
567 * kind does not need peridic writeout yet, and for the latter
568 * kind writeout is handled by the freer.
570 spin_lock(&inode
->i_lock
);
571 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
572 spin_unlock(&inode
->i_lock
);
573 redirty_tail(inode
, wb
);
577 write_chunk
= writeback_chunk_size(wb
->bdi
, work
);
578 wbc
.nr_to_write
= write_chunk
;
579 wbc
.pages_skipped
= 0;
581 writeback_single_inode(inode
, wb
, &wbc
);
583 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
584 wrote
+= write_chunk
- wbc
.nr_to_write
;
585 if (!(inode
->i_state
& I_DIRTY
))
587 if (wbc
.pages_skipped
) {
589 * writeback is not making progress due to locked
590 * buffers. Skip this inode for now.
592 redirty_tail(inode
, wb
);
594 spin_unlock(&inode
->i_lock
);
595 spin_unlock(&wb
->list_lock
);
598 spin_lock(&wb
->list_lock
);
600 * bail out to wb_writeback() often enough to check
601 * background threshold and other termination conditions.
604 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
606 if (work
->nr_pages
<= 0)
613 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
614 struct wb_writeback_work
*work
)
616 unsigned long start_time
= jiffies
;
619 while (!list_empty(&wb
->b_io
)) {
620 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
621 struct super_block
*sb
= inode
->i_sb
;
623 if (!grab_super_passive(sb
)) {
625 * grab_super_passive() may fail consistently due to
626 * s_umount being grabbed by someone else. Don't use
627 * requeue_io() to avoid busy retrying the inode/sb.
629 redirty_tail(inode
, wb
);
632 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
635 /* refer to the same tests at the end of writeback_sb_inodes */
637 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
639 if (work
->nr_pages
<= 0)
643 /* Leave any unwritten inodes on b_io */
647 long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
648 enum wb_reason reason
)
650 struct wb_writeback_work work
= {
651 .nr_pages
= nr_pages
,
652 .sync_mode
= WB_SYNC_NONE
,
657 spin_lock(&wb
->list_lock
);
658 if (list_empty(&wb
->b_io
))
660 __writeback_inodes_wb(wb
, &work
);
661 spin_unlock(&wb
->list_lock
);
663 return nr_pages
- work
.nr_pages
;
666 static bool over_bground_thresh(struct backing_dev_info
*bdi
)
668 unsigned long background_thresh
, dirty_thresh
;
670 global_dirty_limits(&background_thresh
, &dirty_thresh
);
672 if (global_page_state(NR_FILE_DIRTY
) +
673 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
)
676 if (bdi_stat(bdi
, BDI_RECLAIMABLE
) >
677 bdi_dirty_limit(bdi
, background_thresh
))
684 * Called under wb->list_lock. If there are multiple wb per bdi,
685 * only the flusher working on the first wb should do it.
687 static void wb_update_bandwidth(struct bdi_writeback
*wb
,
688 unsigned long start_time
)
690 __bdi_update_bandwidth(wb
->bdi
, 0, 0, 0, 0, 0, start_time
);
694 * Explicit flushing or periodic writeback of "old" data.
696 * Define "old": the first time one of an inode's pages is dirtied, we mark the
697 * dirtying-time in the inode's address_space. So this periodic writeback code
698 * just walks the superblock inode list, writing back any inodes which are
699 * older than a specific point in time.
701 * Try to run once per dirty_writeback_interval. But if a writeback event
702 * takes longer than a dirty_writeback_interval interval, then leave a
705 * older_than_this takes precedence over nr_to_write. So we'll only write back
706 * all dirty pages if they are all attached to "old" mappings.
708 static long wb_writeback(struct bdi_writeback
*wb
,
709 struct wb_writeback_work
*work
)
711 unsigned long wb_start
= jiffies
;
712 long nr_pages
= work
->nr_pages
;
713 unsigned long oldest_jif
;
717 oldest_jif
= jiffies
;
718 work
->older_than_this
= &oldest_jif
;
720 spin_lock(&wb
->list_lock
);
723 * Stop writeback when nr_pages has been consumed
725 if (work
->nr_pages
<= 0)
729 * Background writeout and kupdate-style writeback may
730 * run forever. Stop them if there is other work to do
731 * so that e.g. sync can proceed. They'll be restarted
732 * after the other works are all done.
734 if ((work
->for_background
|| work
->for_kupdate
) &&
735 !list_empty(&wb
->bdi
->work_list
))
739 * For background writeout, stop when we are below the
740 * background dirty threshold
742 if (work
->for_background
&& !over_bground_thresh(wb
->bdi
))
745 if (work
->for_kupdate
) {
746 oldest_jif
= jiffies
-
747 msecs_to_jiffies(dirty_expire_interval
* 10);
748 work
->older_than_this
= &oldest_jif
;
751 trace_writeback_start(wb
->bdi
, work
);
752 if (list_empty(&wb
->b_io
))
755 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
757 progress
= __writeback_inodes_wb(wb
, work
);
758 trace_writeback_written(wb
->bdi
, work
);
760 wb_update_bandwidth(wb
, wb_start
);
763 * Did we write something? Try for more
765 * Dirty inodes are moved to b_io for writeback in batches.
766 * The completion of the current batch does not necessarily
767 * mean the overall work is done. So we keep looping as long
768 * as made some progress on cleaning pages or inodes.
773 * No more inodes for IO, bail
775 if (list_empty(&wb
->b_more_io
))
778 * Nothing written. Wait for some inode to
779 * become available for writeback. Otherwise
780 * we'll just busyloop.
782 if (!list_empty(&wb
->b_more_io
)) {
783 trace_writeback_wait(wb
->bdi
, work
);
784 inode
= wb_inode(wb
->b_more_io
.prev
);
785 spin_lock(&inode
->i_lock
);
786 inode_wait_for_writeback(inode
, wb
);
787 spin_unlock(&inode
->i_lock
);
790 spin_unlock(&wb
->list_lock
);
792 return nr_pages
- work
->nr_pages
;
796 * Return the next wb_writeback_work struct that hasn't been processed yet.
798 static struct wb_writeback_work
*
799 get_next_work_item(struct backing_dev_info
*bdi
)
801 struct wb_writeback_work
*work
= NULL
;
803 spin_lock_bh(&bdi
->wb_lock
);
804 if (!list_empty(&bdi
->work_list
)) {
805 work
= list_entry(bdi
->work_list
.next
,
806 struct wb_writeback_work
, list
);
807 list_del_init(&work
->list
);
809 spin_unlock_bh(&bdi
->wb_lock
);
814 * Add in the number of potentially dirty inodes, because each inode
815 * write can dirty pagecache in the underlying blockdev.
817 static unsigned long get_nr_dirty_pages(void)
819 return global_page_state(NR_FILE_DIRTY
) +
820 global_page_state(NR_UNSTABLE_NFS
) +
821 get_nr_dirty_inodes();
824 static long wb_check_background_flush(struct bdi_writeback
*wb
)
826 if (over_bground_thresh(wb
->bdi
)) {
828 struct wb_writeback_work work
= {
829 .nr_pages
= LONG_MAX
,
830 .sync_mode
= WB_SYNC_NONE
,
833 .reason
= WB_REASON_BACKGROUND
,
836 return wb_writeback(wb
, &work
);
842 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
844 unsigned long expired
;
848 * When set to zero, disable periodic writeback
850 if (!dirty_writeback_interval
)
853 expired
= wb
->last_old_flush
+
854 msecs_to_jiffies(dirty_writeback_interval
* 10);
855 if (time_before(jiffies
, expired
))
858 wb
->last_old_flush
= jiffies
;
859 nr_pages
= get_nr_dirty_pages();
862 struct wb_writeback_work work
= {
863 .nr_pages
= nr_pages
,
864 .sync_mode
= WB_SYNC_NONE
,
867 .reason
= WB_REASON_PERIODIC
,
870 return wb_writeback(wb
, &work
);
877 * Retrieve work items and do the writeback they describe
879 long wb_do_writeback(struct bdi_writeback
*wb
, int force_wait
)
881 struct backing_dev_info
*bdi
= wb
->bdi
;
882 struct wb_writeback_work
*work
;
885 set_bit(BDI_writeback_running
, &wb
->bdi
->state
);
886 while ((work
= get_next_work_item(bdi
)) != NULL
) {
888 * Override sync mode, in case we must wait for completion
889 * because this thread is exiting now.
892 work
->sync_mode
= WB_SYNC_ALL
;
894 trace_writeback_exec(bdi
, work
);
896 wrote
+= wb_writeback(wb
, work
);
899 * Notify the caller of completion if this is a synchronous
900 * work item, otherwise just free it.
903 complete(work
->done
);
909 * Check for periodic writeback, kupdated() style
911 wrote
+= wb_check_old_data_flush(wb
);
912 wrote
+= wb_check_background_flush(wb
);
913 clear_bit(BDI_writeback_running
, &wb
->bdi
->state
);
919 * Handle writeback of dirty data for the device backed by this bdi. Also
920 * wakes up periodically and does kupdated style flushing.
922 int bdi_writeback_thread(void *data
)
924 struct bdi_writeback
*wb
= data
;
925 struct backing_dev_info
*bdi
= wb
->bdi
;
928 current
->flags
|= PF_SWAPWRITE
;
930 wb
->last_active
= jiffies
;
933 * Our parent may run at a different priority, just set us to normal
935 set_user_nice(current
, 0);
937 trace_writeback_thread_start(bdi
);
939 while (!kthread_freezable_should_stop(NULL
)) {
941 * Remove own delayed wake-up timer, since we are already awake
942 * and we'll take care of the preriodic write-back.
944 del_timer(&wb
->wakeup_timer
);
946 pages_written
= wb_do_writeback(wb
, 0);
948 trace_writeback_pages_written(pages_written
);
951 wb
->last_active
= jiffies
;
953 set_current_state(TASK_INTERRUPTIBLE
);
954 if (!list_empty(&bdi
->work_list
) || kthread_should_stop()) {
955 __set_current_state(TASK_RUNNING
);
959 if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
960 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval
* 10));
963 * We have nothing to do, so can go sleep without any
964 * timeout and save power. When a work is queued or
965 * something is made dirty - we will be woken up.
971 /* Flush any work that raced with us exiting */
972 if (!list_empty(&bdi
->work_list
))
973 wb_do_writeback(wb
, 1);
975 trace_writeback_thread_stop(bdi
);
981 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
984 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
986 struct backing_dev_info
*bdi
;
989 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
990 global_page_state(NR_UNSTABLE_NFS
);
994 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
995 if (!bdi_has_dirty_io(bdi
))
997 __bdi_start_writeback(bdi
, nr_pages
, false, reason
);
1002 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1004 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1005 struct dentry
*dentry
;
1006 const char *name
= "?";
1008 dentry
= d_find_alias(inode
);
1010 spin_lock(&dentry
->d_lock
);
1011 name
= (const char *) dentry
->d_name
.name
;
1014 "%s(%d): dirtied inode %lu (%s) on %s\n",
1015 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1016 name
, inode
->i_sb
->s_id
);
1018 spin_unlock(&dentry
->d_lock
);
1025 * __mark_inode_dirty - internal function
1026 * @inode: inode to mark
1027 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1028 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1029 * mark_inode_dirty_sync.
1031 * Put the inode on the super block's dirty list.
1033 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1034 * dirty list only if it is hashed or if it refers to a blockdev.
1035 * If it was not hashed, it will never be added to the dirty list
1036 * even if it is later hashed, as it will have been marked dirty already.
1038 * In short, make sure you hash any inodes _before_ you start marking
1041 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1042 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1043 * the kernel-internal blockdev inode represents the dirtying time of the
1044 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1045 * page->mapping->host, so the page-dirtying time is recorded in the internal
1048 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1050 struct super_block
*sb
= inode
->i_sb
;
1051 struct backing_dev_info
*bdi
= NULL
;
1054 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1055 * dirty the inode itself
1057 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
1058 if (sb
->s_op
->dirty_inode
)
1059 sb
->s_op
->dirty_inode(inode
, flags
);
1063 * make sure that changes are seen by all cpus before we test i_state
1068 /* avoid the locking if we can */
1069 if ((inode
->i_state
& flags
) == flags
)
1072 if (unlikely(block_dump
))
1073 block_dump___mark_inode_dirty(inode
);
1075 spin_lock(&inode
->i_lock
);
1076 if ((inode
->i_state
& flags
) != flags
) {
1077 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1079 inode
->i_state
|= flags
;
1082 * If the inode is being synced, just update its dirty state.
1083 * The unlocker will place the inode on the appropriate
1084 * superblock list, based upon its state.
1086 if (inode
->i_state
& I_SYNC
)
1087 goto out_unlock_inode
;
1090 * Only add valid (hashed) inodes to the superblock's
1091 * dirty list. Add blockdev inodes as well.
1093 if (!S_ISBLK(inode
->i_mode
)) {
1094 if (inode_unhashed(inode
))
1095 goto out_unlock_inode
;
1097 if (inode
->i_state
& I_FREEING
)
1098 goto out_unlock_inode
;
1101 * If the inode was already on b_dirty/b_io/b_more_io, don't
1102 * reposition it (that would break b_dirty time-ordering).
1105 bool wakeup_bdi
= false;
1106 bdi
= inode_to_bdi(inode
);
1108 if (bdi_cap_writeback_dirty(bdi
)) {
1109 WARN(!test_bit(BDI_registered
, &bdi
->state
),
1110 "bdi-%s not registered\n", bdi
->name
);
1113 * If this is the first dirty inode for this
1114 * bdi, we have to wake-up the corresponding
1115 * bdi thread to make sure background
1116 * write-back happens later.
1118 if (!wb_has_dirty_io(&bdi
->wb
))
1122 spin_unlock(&inode
->i_lock
);
1123 spin_lock(&bdi
->wb
.list_lock
);
1124 inode
->dirtied_when
= jiffies
;
1125 list_move(&inode
->i_wb_list
, &bdi
->wb
.b_dirty
);
1126 spin_unlock(&bdi
->wb
.list_lock
);
1129 bdi_wakeup_thread_delayed(bdi
);
1134 spin_unlock(&inode
->i_lock
);
1137 EXPORT_SYMBOL(__mark_inode_dirty
);
1140 * Write out a superblock's list of dirty inodes. A wait will be performed
1141 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1143 * If older_than_this is non-NULL, then only write out inodes which
1144 * had their first dirtying at a time earlier than *older_than_this.
1146 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1147 * This function assumes that the blockdev superblock's inodes are backed by
1148 * a variety of queues, so all inodes are searched. For other superblocks,
1149 * assume that all inodes are backed by the same queue.
1151 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1152 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1153 * on the writer throttling path, and we get decent balancing between many
1154 * throttled threads: we don't want them all piling up on inode_sync_wait.
1156 static void wait_sb_inodes(struct super_block
*sb
)
1158 struct inode
*inode
, *old_inode
= NULL
;
1161 * We need to be protected against the filesystem going from
1162 * r/o to r/w or vice versa.
1164 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1166 spin_lock(&inode_sb_list_lock
);
1169 * Data integrity sync. Must wait for all pages under writeback,
1170 * because there may have been pages dirtied before our sync
1171 * call, but which had writeout started before we write it out.
1172 * In which case, the inode may not be on the dirty list, but
1173 * we still have to wait for that writeout.
1175 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1176 struct address_space
*mapping
= inode
->i_mapping
;
1178 spin_lock(&inode
->i_lock
);
1179 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1180 (mapping
->nrpages
== 0)) {
1181 spin_unlock(&inode
->i_lock
);
1185 spin_unlock(&inode
->i_lock
);
1186 spin_unlock(&inode_sb_list_lock
);
1189 * We hold a reference to 'inode' so it couldn't have been
1190 * removed from s_inodes list while we dropped the
1191 * inode_sb_list_lock. We cannot iput the inode now as we can
1192 * be holding the last reference and we cannot iput it under
1193 * inode_sb_list_lock. So we keep the reference and iput it
1199 filemap_fdatawait(mapping
);
1203 spin_lock(&inode_sb_list_lock
);
1205 spin_unlock(&inode_sb_list_lock
);
1210 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1211 * @sb: the superblock
1212 * @nr: the number of pages to write
1213 * @reason: reason why some writeback work initiated
1215 * Start writeback on some inodes on this super_block. No guarantees are made
1216 * on how many (if any) will be written, and this function does not wait
1217 * for IO completion of submitted IO.
1219 void writeback_inodes_sb_nr(struct super_block
*sb
,
1221 enum wb_reason reason
)
1223 DECLARE_COMPLETION_ONSTACK(done
);
1224 struct wb_writeback_work work
= {
1226 .sync_mode
= WB_SYNC_NONE
,
1227 .tagged_writepages
= 1,
1233 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1234 bdi_queue_work(sb
->s_bdi
, &work
);
1235 wait_for_completion(&done
);
1237 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1240 * writeback_inodes_sb - writeback dirty inodes from given super_block
1241 * @sb: the superblock
1242 * @reason: reason why some writeback work was initiated
1244 * Start writeback on some inodes on this super_block. No guarantees are made
1245 * on how many (if any) will be written, and this function does not wait
1246 * for IO completion of submitted IO.
1248 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1250 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1252 EXPORT_SYMBOL(writeback_inodes_sb
);
1255 * writeback_inodes_sb_if_idle - start writeback if none underway
1256 * @sb: the superblock
1257 * @reason: reason why some writeback work was initiated
1259 * Invoke writeback_inodes_sb if no writeback is currently underway.
1260 * Returns 1 if writeback was started, 0 if not.
1262 int writeback_inodes_sb_if_idle(struct super_block
*sb
, enum wb_reason reason
)
1264 if (!writeback_in_progress(sb
->s_bdi
)) {
1265 down_read(&sb
->s_umount
);
1266 writeback_inodes_sb(sb
, reason
);
1267 up_read(&sb
->s_umount
);
1272 EXPORT_SYMBOL(writeback_inodes_sb_if_idle
);
1275 * writeback_inodes_sb_if_idle - start writeback if none underway
1276 * @sb: the superblock
1277 * @nr: the number of pages to write
1278 * @reason: reason why some writeback work was initiated
1280 * Invoke writeback_inodes_sb if no writeback is currently underway.
1281 * Returns 1 if writeback was started, 0 if not.
1283 int writeback_inodes_sb_nr_if_idle(struct super_block
*sb
,
1285 enum wb_reason reason
)
1287 if (!writeback_in_progress(sb
->s_bdi
)) {
1288 down_read(&sb
->s_umount
);
1289 writeback_inodes_sb_nr(sb
, nr
, reason
);
1290 up_read(&sb
->s_umount
);
1295 EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle
);
1298 * sync_inodes_sb - sync sb inode pages
1299 * @sb: the superblock
1301 * This function writes and waits on any dirty inode belonging to this
1304 void sync_inodes_sb(struct super_block
*sb
)
1306 DECLARE_COMPLETION_ONSTACK(done
);
1307 struct wb_writeback_work work
= {
1309 .sync_mode
= WB_SYNC_ALL
,
1310 .nr_pages
= LONG_MAX
,
1313 .reason
= WB_REASON_SYNC
,
1316 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1318 bdi_queue_work(sb
->s_bdi
, &work
);
1319 wait_for_completion(&done
);
1323 EXPORT_SYMBOL(sync_inodes_sb
);
1326 * write_inode_now - write an inode to disk
1327 * @inode: inode to write to disk
1328 * @sync: whether the write should be synchronous or not
1330 * This function commits an inode to disk immediately if it is dirty. This is
1331 * primarily needed by knfsd.
1333 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1335 int write_inode_now(struct inode
*inode
, int sync
)
1337 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1339 struct writeback_control wbc
= {
1340 .nr_to_write
= LONG_MAX
,
1341 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1343 .range_end
= LLONG_MAX
,
1346 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1347 wbc
.nr_to_write
= 0;
1350 spin_lock(&wb
->list_lock
);
1351 spin_lock(&inode
->i_lock
);
1352 ret
= writeback_single_inode(inode
, wb
, &wbc
);
1353 spin_unlock(&inode
->i_lock
);
1354 spin_unlock(&wb
->list_lock
);
1356 inode_sync_wait(inode
);
1359 EXPORT_SYMBOL(write_inode_now
);
1362 * sync_inode - write an inode and its pages to disk.
1363 * @inode: the inode to sync
1364 * @wbc: controls the writeback mode
1366 * sync_inode() will write an inode and its pages to disk. It will also
1367 * correctly update the inode on its superblock's dirty inode lists and will
1368 * update inode->i_state.
1370 * The caller must have a ref on the inode.
1372 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1374 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1377 spin_lock(&wb
->list_lock
);
1378 spin_lock(&inode
->i_lock
);
1379 ret
= writeback_single_inode(inode
, wb
, wbc
);
1380 spin_unlock(&inode
->i_lock
);
1381 spin_unlock(&wb
->list_lock
);
1384 EXPORT_SYMBOL(sync_inode
);
1387 * sync_inode_metadata - write an inode to disk
1388 * @inode: the inode to sync
1389 * @wait: wait for I/O to complete.
1391 * Write an inode to disk and adjust its dirty state after completion.
1393 * Note: only writes the actual inode, no associated data or other metadata.
1395 int sync_inode_metadata(struct inode
*inode
, int wait
)
1397 struct writeback_control wbc
= {
1398 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1399 .nr_to_write
= 0, /* metadata-only */
1402 return sync_inode(inode
, &wbc
);
1404 EXPORT_SYMBOL(sync_inode_metadata
);