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 * If an inode is constantly having its pages dirtied, but then the
58 * updates stop dirtytime_expire_interval seconds in the past, it's
59 * possible for the worst case time between when an inode has its
60 * timestamps updated and when they finally get written out to be two
61 * dirtytime_expire_intervals. We set the default to 12 hours (in
62 * seconds), which means most of the time inodes will have their
63 * timestamps written to disk after 12 hours, but in the worst case a
64 * few inodes might not their timestamps updated for 24 hours.
66 unsigned int dirtytime_expire_interval
= 12 * 60 * 60;
69 * writeback_in_progress - determine whether there is writeback in progress
70 * @bdi: the device's backing_dev_info structure.
72 * Determine whether there is writeback waiting to be handled against a
75 int writeback_in_progress(struct backing_dev_info
*bdi
)
77 return test_bit(WB_writeback_running
, &bdi
->wb
.state
);
79 EXPORT_SYMBOL(writeback_in_progress
);
81 static inline struct inode
*wb_inode(struct list_head
*head
)
83 return list_entry(head
, struct inode
, i_wb_list
);
87 * Include the creation of the trace points after defining the
88 * wb_writeback_work structure and inline functions so that the definition
89 * remains local to this file.
91 #define CREATE_TRACE_POINTS
92 #include <trace/events/writeback.h>
94 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage
);
96 static bool wb_io_lists_populated(struct bdi_writeback
*wb
)
98 if (wb_has_dirty_io(wb
)) {
101 set_bit(WB_has_dirty_io
, &wb
->state
);
102 WARN_ON_ONCE(!wb
->avg_write_bandwidth
);
103 atomic_long_add(wb
->avg_write_bandwidth
,
104 &wb
->bdi
->tot_write_bandwidth
);
109 static void wb_io_lists_depopulated(struct bdi_writeback
*wb
)
111 if (wb_has_dirty_io(wb
) && list_empty(&wb
->b_dirty
) &&
112 list_empty(&wb
->b_io
) && list_empty(&wb
->b_more_io
)) {
113 clear_bit(WB_has_dirty_io
, &wb
->state
);
114 WARN_ON_ONCE(atomic_long_sub_return(wb
->avg_write_bandwidth
,
115 &wb
->bdi
->tot_write_bandwidth
) < 0);
120 * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list
121 * @inode: inode to be moved
122 * @wb: target bdi_writeback
123 * @head: one of @wb->b_{dirty|io|more_io}
125 * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io.
126 * Returns %true if @inode is the first occupant of the !dirty_time IO
127 * lists; otherwise, %false.
129 static bool inode_wb_list_move_locked(struct inode
*inode
,
130 struct bdi_writeback
*wb
,
131 struct list_head
*head
)
133 assert_spin_locked(&wb
->list_lock
);
135 list_move(&inode
->i_wb_list
, head
);
137 /* dirty_time doesn't count as dirty_io until expiration */
138 if (head
!= &wb
->b_dirty_time
)
139 return wb_io_lists_populated(wb
);
141 wb_io_lists_depopulated(wb
);
146 * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list
147 * @inode: inode to be removed
148 * @wb: bdi_writeback @inode is being removed from
150 * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and
151 * clear %WB_has_dirty_io if all are empty afterwards.
153 static void inode_wb_list_del_locked(struct inode
*inode
,
154 struct bdi_writeback
*wb
)
156 assert_spin_locked(&wb
->list_lock
);
158 list_del_init(&inode
->i_wb_list
);
159 wb_io_lists_depopulated(wb
);
162 static void wb_wakeup(struct bdi_writeback
*wb
)
164 spin_lock_bh(&wb
->work_lock
);
165 if (test_bit(WB_registered
, &wb
->state
))
166 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
167 spin_unlock_bh(&wb
->work_lock
);
170 static void wb_queue_work(struct bdi_writeback
*wb
,
171 struct wb_writeback_work
*work
)
173 trace_writeback_queue(wb
->bdi
, work
);
175 spin_lock_bh(&wb
->work_lock
);
176 if (!test_bit(WB_registered
, &wb
->state
)) {
178 complete(work
->done
);
181 list_add_tail(&work
->list
, &wb
->work_list
);
182 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
184 spin_unlock_bh(&wb
->work_lock
);
187 #ifdef CONFIG_CGROUP_WRITEBACK
190 * inode_congested - test whether an inode is congested
191 * @inode: inode to test for congestion
192 * @cong_bits: mask of WB_[a]sync_congested bits to test
194 * Tests whether @inode is congested. @cong_bits is the mask of congestion
195 * bits to test and the return value is the mask of set bits.
197 * If cgroup writeback is enabled for @inode, the congestion state is
198 * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
199 * associated with @inode is congested; otherwise, the root wb's congestion
202 int inode_congested(struct inode
*inode
, int cong_bits
)
205 struct bdi_writeback
*wb
= inode_to_wb(inode
);
207 return wb_congested(wb
, cong_bits
);
210 return wb_congested(&inode_to_bdi(inode
)->wb
, cong_bits
);
212 EXPORT_SYMBOL_GPL(inode_congested
);
214 #endif /* CONFIG_CGROUP_WRITEBACK */
216 void wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
217 bool range_cyclic
, enum wb_reason reason
)
219 struct wb_writeback_work
*work
;
221 if (!wb_has_dirty_io(wb
))
225 * This is WB_SYNC_NONE writeback, so if allocation fails just
226 * wakeup the thread for old dirty data writeback
228 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
230 trace_writeback_nowork(wb
->bdi
);
235 work
->sync_mode
= WB_SYNC_NONE
;
236 work
->nr_pages
= nr_pages
;
237 work
->range_cyclic
= range_cyclic
;
238 work
->reason
= reason
;
240 wb_queue_work(wb
, work
);
244 * bdi_start_background_writeback - start background writeback
245 * @bdi: the backing device to write from
248 * This makes sure WB_SYNC_NONE background writeback happens. When
249 * this function returns, it is only guaranteed that for given BDI
250 * some IO is happening if we are over background dirty threshold.
251 * Caller need not hold sb s_umount semaphore.
253 void bdi_start_background_writeback(struct backing_dev_info
*bdi
)
256 * We just wake up the flusher thread. It will perform background
257 * writeback as soon as there is no other work to do.
259 trace_writeback_wake_background(bdi
);
264 * Remove the inode from the writeback list it is on.
266 void inode_wb_list_del(struct inode
*inode
)
268 struct bdi_writeback
*wb
= inode_to_wb(inode
);
270 spin_lock(&wb
->list_lock
);
271 inode_wb_list_del_locked(inode
, wb
);
272 spin_unlock(&wb
->list_lock
);
276 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
277 * furthest end of its superblock's dirty-inode list.
279 * Before stamping the inode's ->dirtied_when, we check to see whether it is
280 * already the most-recently-dirtied inode on the b_dirty list. If that is
281 * the case then the inode must have been redirtied while it was being written
282 * out and we don't reset its dirtied_when.
284 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
286 if (!list_empty(&wb
->b_dirty
)) {
289 tail
= wb_inode(wb
->b_dirty
.next
);
290 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
291 inode
->dirtied_when
= jiffies
;
293 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty
);
297 * requeue inode for re-scanning after bdi->b_io list is exhausted.
299 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
301 inode_wb_list_move_locked(inode
, wb
, &wb
->b_more_io
);
304 static void inode_sync_complete(struct inode
*inode
)
306 inode
->i_state
&= ~I_SYNC
;
307 /* If inode is clean an unused, put it into LRU now... */
308 inode_add_lru(inode
);
309 /* Waiters must see I_SYNC cleared before being woken up */
311 wake_up_bit(&inode
->i_state
, __I_SYNC
);
314 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
316 bool ret
= time_after(inode
->dirtied_when
, t
);
319 * For inodes being constantly redirtied, dirtied_when can get stuck.
320 * It _appears_ to be in the future, but is actually in distant past.
321 * This test is necessary to prevent such wrapped-around relative times
322 * from permanently stopping the whole bdi writeback.
324 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
329 #define EXPIRE_DIRTY_ATIME 0x0001
332 * Move expired (dirtied before work->older_than_this) dirty inodes from
333 * @delaying_queue to @dispatch_queue.
335 static int move_expired_inodes(struct list_head
*delaying_queue
,
336 struct list_head
*dispatch_queue
,
338 struct wb_writeback_work
*work
)
340 unsigned long *older_than_this
= NULL
;
341 unsigned long expire_time
;
343 struct list_head
*pos
, *node
;
344 struct super_block
*sb
= NULL
;
349 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
350 older_than_this
= work
->older_than_this
;
351 else if (!work
->for_sync
) {
352 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
353 older_than_this
= &expire_time
;
355 while (!list_empty(delaying_queue
)) {
356 inode
= wb_inode(delaying_queue
->prev
);
357 if (older_than_this
&&
358 inode_dirtied_after(inode
, *older_than_this
))
360 list_move(&inode
->i_wb_list
, &tmp
);
362 if (flags
& EXPIRE_DIRTY_ATIME
)
363 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
364 if (sb_is_blkdev_sb(inode
->i_sb
))
366 if (sb
&& sb
!= inode
->i_sb
)
371 /* just one sb in list, splice to dispatch_queue and we're done */
373 list_splice(&tmp
, dispatch_queue
);
377 /* Move inodes from one superblock together */
378 while (!list_empty(&tmp
)) {
379 sb
= wb_inode(tmp
.prev
)->i_sb
;
380 list_for_each_prev_safe(pos
, node
, &tmp
) {
381 inode
= wb_inode(pos
);
382 if (inode
->i_sb
== sb
)
383 list_move(&inode
->i_wb_list
, dispatch_queue
);
391 * Queue all expired dirty inodes for io, eldest first.
393 * newly dirtied b_dirty b_io b_more_io
394 * =============> gf edc BA
396 * newly dirtied b_dirty b_io b_more_io
397 * =============> g fBAedc
399 * +--> dequeue for IO
401 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
405 assert_spin_locked(&wb
->list_lock
);
406 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
407 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
408 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
409 EXPIRE_DIRTY_ATIME
, work
);
411 wb_io_lists_populated(wb
);
412 trace_writeback_queue_io(wb
, work
, moved
);
415 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
419 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
420 trace_writeback_write_inode_start(inode
, wbc
);
421 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
422 trace_writeback_write_inode(inode
, wbc
);
429 * Wait for writeback on an inode to complete. Called with i_lock held.
430 * Caller must make sure inode cannot go away when we drop i_lock.
432 static void __inode_wait_for_writeback(struct inode
*inode
)
433 __releases(inode
->i_lock
)
434 __acquires(inode
->i_lock
)
436 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
437 wait_queue_head_t
*wqh
;
439 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
440 while (inode
->i_state
& I_SYNC
) {
441 spin_unlock(&inode
->i_lock
);
442 __wait_on_bit(wqh
, &wq
, bit_wait
,
443 TASK_UNINTERRUPTIBLE
);
444 spin_lock(&inode
->i_lock
);
449 * Wait for writeback on an inode to complete. Caller must have inode pinned.
451 void inode_wait_for_writeback(struct inode
*inode
)
453 spin_lock(&inode
->i_lock
);
454 __inode_wait_for_writeback(inode
);
455 spin_unlock(&inode
->i_lock
);
459 * Sleep until I_SYNC is cleared. This function must be called with i_lock
460 * held and drops it. It is aimed for callers not holding any inode reference
461 * so once i_lock is dropped, inode can go away.
463 static void inode_sleep_on_writeback(struct inode
*inode
)
464 __releases(inode
->i_lock
)
467 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
470 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
471 sleep
= inode
->i_state
& I_SYNC
;
472 spin_unlock(&inode
->i_lock
);
475 finish_wait(wqh
, &wait
);
479 * Find proper writeback list for the inode depending on its current state and
480 * possibly also change of its state while we were doing writeback. Here we
481 * handle things such as livelock prevention or fairness of writeback among
482 * inodes. This function can be called only by flusher thread - noone else
483 * processes all inodes in writeback lists and requeueing inodes behind flusher
484 * thread's back can have unexpected consequences.
486 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
487 struct writeback_control
*wbc
)
489 if (inode
->i_state
& I_FREEING
)
493 * Sync livelock prevention. Each inode is tagged and synced in one
494 * shot. If still dirty, it will be redirty_tail()'ed below. Update
495 * the dirty time to prevent enqueue and sync it again.
497 if ((inode
->i_state
& I_DIRTY
) &&
498 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
499 inode
->dirtied_when
= jiffies
;
501 if (wbc
->pages_skipped
) {
503 * writeback is not making progress due to locked
504 * buffers. Skip this inode for now.
506 redirty_tail(inode
, wb
);
510 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
512 * We didn't write back all the pages. nfs_writepages()
513 * sometimes bales out without doing anything.
515 if (wbc
->nr_to_write
<= 0) {
516 /* Slice used up. Queue for next turn. */
517 requeue_io(inode
, wb
);
520 * Writeback blocked by something other than
521 * congestion. Delay the inode for some time to
522 * avoid spinning on the CPU (100% iowait)
523 * retrying writeback of the dirty page/inode
524 * that cannot be performed immediately.
526 redirty_tail(inode
, wb
);
528 } else if (inode
->i_state
& I_DIRTY
) {
530 * Filesystems can dirty the inode during writeback operations,
531 * such as delayed allocation during submission or metadata
532 * updates after data IO completion.
534 redirty_tail(inode
, wb
);
535 } else if (inode
->i_state
& I_DIRTY_TIME
) {
536 inode
->dirtied_when
= jiffies
;
537 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty_time
);
539 /* The inode is clean. Remove from writeback lists. */
540 inode_wb_list_del_locked(inode
, wb
);
545 * Write out an inode and its dirty pages. Do not update the writeback list
546 * linkage. That is left to the caller. The caller is also responsible for
547 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
550 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
552 struct address_space
*mapping
= inode
->i_mapping
;
553 long nr_to_write
= wbc
->nr_to_write
;
557 WARN_ON(!(inode
->i_state
& I_SYNC
));
559 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
561 ret
= do_writepages(mapping
, wbc
);
564 * Make sure to wait on the data before writing out the metadata.
565 * This is important for filesystems that modify metadata on data
566 * I/O completion. We don't do it for sync(2) writeback because it has a
567 * separate, external IO completion path and ->sync_fs for guaranteeing
568 * inode metadata is written back correctly.
570 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
571 int err
= filemap_fdatawait(mapping
);
577 * Some filesystems may redirty the inode during the writeback
578 * due to delalloc, clear dirty metadata flags right before
581 spin_lock(&inode
->i_lock
);
583 dirty
= inode
->i_state
& I_DIRTY
;
584 if (inode
->i_state
& I_DIRTY_TIME
) {
585 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
586 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
587 unlikely(time_after(jiffies
,
588 (inode
->dirtied_time_when
+
589 dirtytime_expire_interval
* HZ
)))) {
590 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
591 trace_writeback_lazytime(inode
);
594 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
595 inode
->i_state
&= ~dirty
;
598 * Paired with smp_mb() in __mark_inode_dirty(). This allows
599 * __mark_inode_dirty() to test i_state without grabbing i_lock -
600 * either they see the I_DIRTY bits cleared or we see the dirtied
603 * I_DIRTY_PAGES is always cleared together above even if @mapping
604 * still has dirty pages. The flag is reinstated after smp_mb() if
605 * necessary. This guarantees that either __mark_inode_dirty()
606 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
610 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
611 inode
->i_state
|= I_DIRTY_PAGES
;
613 spin_unlock(&inode
->i_lock
);
615 if (dirty
& I_DIRTY_TIME
)
616 mark_inode_dirty_sync(inode
);
617 /* Don't write the inode if only I_DIRTY_PAGES was set */
618 if (dirty
& ~I_DIRTY_PAGES
) {
619 int err
= write_inode(inode
, wbc
);
623 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
628 * Write out an inode's dirty pages. Either the caller has an active reference
629 * on the inode or the inode has I_WILL_FREE set.
631 * This function is designed to be called for writing back one inode which
632 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
633 * and does more profound writeback list handling in writeback_sb_inodes().
636 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
637 struct writeback_control
*wbc
)
641 spin_lock(&inode
->i_lock
);
642 if (!atomic_read(&inode
->i_count
))
643 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
645 WARN_ON(inode
->i_state
& I_WILL_FREE
);
647 if (inode
->i_state
& I_SYNC
) {
648 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
651 * It's a data-integrity sync. We must wait. Since callers hold
652 * inode reference or inode has I_WILL_FREE set, it cannot go
655 __inode_wait_for_writeback(inode
);
657 WARN_ON(inode
->i_state
& I_SYNC
);
659 * Skip inode if it is clean and we have no outstanding writeback in
660 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
661 * function since flusher thread may be doing for example sync in
662 * parallel and if we move the inode, it could get skipped. So here we
663 * make sure inode is on some writeback list and leave it there unless
664 * we have completely cleaned the inode.
666 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
667 (wbc
->sync_mode
!= WB_SYNC_ALL
||
668 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
670 inode
->i_state
|= I_SYNC
;
671 spin_unlock(&inode
->i_lock
);
673 ret
= __writeback_single_inode(inode
, wbc
);
675 spin_lock(&wb
->list_lock
);
676 spin_lock(&inode
->i_lock
);
678 * If inode is clean, remove it from writeback lists. Otherwise don't
679 * touch it. See comment above for explanation.
681 if (!(inode
->i_state
& I_DIRTY_ALL
))
682 inode_wb_list_del_locked(inode
, wb
);
683 spin_unlock(&wb
->list_lock
);
684 inode_sync_complete(inode
);
686 spin_unlock(&inode
->i_lock
);
690 static long writeback_chunk_size(struct bdi_writeback
*wb
,
691 struct wb_writeback_work
*work
)
696 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
697 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
698 * here avoids calling into writeback_inodes_wb() more than once.
700 * The intended call sequence for WB_SYNC_ALL writeback is:
703 * writeback_sb_inodes() <== called only once
704 * write_cache_pages() <== called once for each inode
705 * (quickly) tag currently dirty pages
706 * (maybe slowly) sync all tagged pages
708 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
711 pages
= min(wb
->avg_write_bandwidth
/ 2,
712 global_dirty_limit
/ DIRTY_SCOPE
);
713 pages
= min(pages
, work
->nr_pages
);
714 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
715 MIN_WRITEBACK_PAGES
);
722 * Write a portion of b_io inodes which belong to @sb.
724 * Return the number of pages and/or inodes written.
726 static long writeback_sb_inodes(struct super_block
*sb
,
727 struct bdi_writeback
*wb
,
728 struct wb_writeback_work
*work
)
730 struct writeback_control wbc
= {
731 .sync_mode
= work
->sync_mode
,
732 .tagged_writepages
= work
->tagged_writepages
,
733 .for_kupdate
= work
->for_kupdate
,
734 .for_background
= work
->for_background
,
735 .for_sync
= work
->for_sync
,
736 .range_cyclic
= work
->range_cyclic
,
738 .range_end
= LLONG_MAX
,
740 unsigned long start_time
= jiffies
;
742 long wrote
= 0; /* count both pages and inodes */
744 while (!list_empty(&wb
->b_io
)) {
745 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
747 if (inode
->i_sb
!= sb
) {
750 * We only want to write back data for this
751 * superblock, move all inodes not belonging
752 * to it back onto the dirty list.
754 redirty_tail(inode
, wb
);
759 * The inode belongs to a different superblock.
760 * Bounce back to the caller to unpin this and
761 * pin the next superblock.
767 * Don't bother with new inodes or inodes being freed, first
768 * kind does not need periodic writeout yet, and for the latter
769 * kind writeout is handled by the freer.
771 spin_lock(&inode
->i_lock
);
772 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
773 spin_unlock(&inode
->i_lock
);
774 redirty_tail(inode
, wb
);
777 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
779 * If this inode is locked for writeback and we are not
780 * doing writeback-for-data-integrity, move it to
781 * b_more_io so that writeback can proceed with the
782 * other inodes on s_io.
784 * We'll have another go at writing back this inode
785 * when we completed a full scan of b_io.
787 spin_unlock(&inode
->i_lock
);
788 requeue_io(inode
, wb
);
789 trace_writeback_sb_inodes_requeue(inode
);
792 spin_unlock(&wb
->list_lock
);
795 * We already requeued the inode if it had I_SYNC set and we
796 * are doing WB_SYNC_NONE writeback. So this catches only the
799 if (inode
->i_state
& I_SYNC
) {
800 /* Wait for I_SYNC. This function drops i_lock... */
801 inode_sleep_on_writeback(inode
);
802 /* Inode may be gone, start again */
803 spin_lock(&wb
->list_lock
);
806 inode
->i_state
|= I_SYNC
;
807 spin_unlock(&inode
->i_lock
);
809 write_chunk
= writeback_chunk_size(wb
, work
);
810 wbc
.nr_to_write
= write_chunk
;
811 wbc
.pages_skipped
= 0;
814 * We use I_SYNC to pin the inode in memory. While it is set
815 * evict_inode() will wait so the inode cannot be freed.
817 __writeback_single_inode(inode
, &wbc
);
819 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
820 wrote
+= write_chunk
- wbc
.nr_to_write
;
821 spin_lock(&wb
->list_lock
);
822 spin_lock(&inode
->i_lock
);
823 if (!(inode
->i_state
& I_DIRTY_ALL
))
825 requeue_inode(inode
, wb
, &wbc
);
826 inode_sync_complete(inode
);
827 spin_unlock(&inode
->i_lock
);
828 cond_resched_lock(&wb
->list_lock
);
830 * bail out to wb_writeback() often enough to check
831 * background threshold and other termination conditions.
834 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
836 if (work
->nr_pages
<= 0)
843 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
844 struct wb_writeback_work
*work
)
846 unsigned long start_time
= jiffies
;
849 while (!list_empty(&wb
->b_io
)) {
850 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
851 struct super_block
*sb
= inode
->i_sb
;
853 if (!trylock_super(sb
)) {
855 * trylock_super() may fail consistently due to
856 * s_umount being grabbed by someone else. Don't use
857 * requeue_io() to avoid busy retrying the inode/sb.
859 redirty_tail(inode
, wb
);
862 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
863 up_read(&sb
->s_umount
);
865 /* refer to the same tests at the end of writeback_sb_inodes */
867 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
869 if (work
->nr_pages
<= 0)
873 /* Leave any unwritten inodes on b_io */
877 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
878 enum wb_reason reason
)
880 struct wb_writeback_work work
= {
881 .nr_pages
= nr_pages
,
882 .sync_mode
= WB_SYNC_NONE
,
887 spin_lock(&wb
->list_lock
);
888 if (list_empty(&wb
->b_io
))
890 __writeback_inodes_wb(wb
, &work
);
891 spin_unlock(&wb
->list_lock
);
893 return nr_pages
- work
.nr_pages
;
896 static bool over_bground_thresh(struct bdi_writeback
*wb
)
898 unsigned long background_thresh
, dirty_thresh
;
900 global_dirty_limits(&background_thresh
, &dirty_thresh
);
902 if (global_page_state(NR_FILE_DIRTY
) +
903 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
)
906 if (wb_stat(wb
, WB_RECLAIMABLE
) > wb_dirty_limit(wb
, background_thresh
))
913 * Called under wb->list_lock. If there are multiple wb per bdi,
914 * only the flusher working on the first wb should do it.
916 static void wb_update_bandwidth(struct bdi_writeback
*wb
,
917 unsigned long start_time
)
919 __wb_update_bandwidth(wb
, 0, 0, 0, 0, 0, start_time
);
923 * Explicit flushing or periodic writeback of "old" data.
925 * Define "old": the first time one of an inode's pages is dirtied, we mark the
926 * dirtying-time in the inode's address_space. So this periodic writeback code
927 * just walks the superblock inode list, writing back any inodes which are
928 * older than a specific point in time.
930 * Try to run once per dirty_writeback_interval. But if a writeback event
931 * takes longer than a dirty_writeback_interval interval, then leave a
934 * older_than_this takes precedence over nr_to_write. So we'll only write back
935 * all dirty pages if they are all attached to "old" mappings.
937 static long wb_writeback(struct bdi_writeback
*wb
,
938 struct wb_writeback_work
*work
)
940 unsigned long wb_start
= jiffies
;
941 long nr_pages
= work
->nr_pages
;
942 unsigned long oldest_jif
;
946 oldest_jif
= jiffies
;
947 work
->older_than_this
= &oldest_jif
;
949 spin_lock(&wb
->list_lock
);
952 * Stop writeback when nr_pages has been consumed
954 if (work
->nr_pages
<= 0)
958 * Background writeout and kupdate-style writeback may
959 * run forever. Stop them if there is other work to do
960 * so that e.g. sync can proceed. They'll be restarted
961 * after the other works are all done.
963 if ((work
->for_background
|| work
->for_kupdate
) &&
964 !list_empty(&wb
->work_list
))
968 * For background writeout, stop when we are below the
969 * background dirty threshold
971 if (work
->for_background
&& !over_bground_thresh(wb
))
975 * Kupdate and background works are special and we want to
976 * include all inodes that need writing. Livelock avoidance is
977 * handled by these works yielding to any other work so we are
980 if (work
->for_kupdate
) {
981 oldest_jif
= jiffies
-
982 msecs_to_jiffies(dirty_expire_interval
* 10);
983 } else if (work
->for_background
)
984 oldest_jif
= jiffies
;
986 trace_writeback_start(wb
->bdi
, work
);
987 if (list_empty(&wb
->b_io
))
990 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
992 progress
= __writeback_inodes_wb(wb
, work
);
993 trace_writeback_written(wb
->bdi
, work
);
995 wb_update_bandwidth(wb
, wb_start
);
998 * Did we write something? Try for more
1000 * Dirty inodes are moved to b_io for writeback in batches.
1001 * The completion of the current batch does not necessarily
1002 * mean the overall work is done. So we keep looping as long
1003 * as made some progress on cleaning pages or inodes.
1008 * No more inodes for IO, bail
1010 if (list_empty(&wb
->b_more_io
))
1013 * Nothing written. Wait for some inode to
1014 * become available for writeback. Otherwise
1015 * we'll just busyloop.
1017 if (!list_empty(&wb
->b_more_io
)) {
1018 trace_writeback_wait(wb
->bdi
, work
);
1019 inode
= wb_inode(wb
->b_more_io
.prev
);
1020 spin_lock(&inode
->i_lock
);
1021 spin_unlock(&wb
->list_lock
);
1022 /* This function drops i_lock... */
1023 inode_sleep_on_writeback(inode
);
1024 spin_lock(&wb
->list_lock
);
1027 spin_unlock(&wb
->list_lock
);
1029 return nr_pages
- work
->nr_pages
;
1033 * Return the next wb_writeback_work struct that hasn't been processed yet.
1035 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
1037 struct wb_writeback_work
*work
= NULL
;
1039 spin_lock_bh(&wb
->work_lock
);
1040 if (!list_empty(&wb
->work_list
)) {
1041 work
= list_entry(wb
->work_list
.next
,
1042 struct wb_writeback_work
, list
);
1043 list_del_init(&work
->list
);
1045 spin_unlock_bh(&wb
->work_lock
);
1050 * Add in the number of potentially dirty inodes, because each inode
1051 * write can dirty pagecache in the underlying blockdev.
1053 static unsigned long get_nr_dirty_pages(void)
1055 return global_page_state(NR_FILE_DIRTY
) +
1056 global_page_state(NR_UNSTABLE_NFS
) +
1057 get_nr_dirty_inodes();
1060 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1062 if (over_bground_thresh(wb
)) {
1064 struct wb_writeback_work work
= {
1065 .nr_pages
= LONG_MAX
,
1066 .sync_mode
= WB_SYNC_NONE
,
1067 .for_background
= 1,
1069 .reason
= WB_REASON_BACKGROUND
,
1072 return wb_writeback(wb
, &work
);
1078 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1080 unsigned long expired
;
1084 * When set to zero, disable periodic writeback
1086 if (!dirty_writeback_interval
)
1089 expired
= wb
->last_old_flush
+
1090 msecs_to_jiffies(dirty_writeback_interval
* 10);
1091 if (time_before(jiffies
, expired
))
1094 wb
->last_old_flush
= jiffies
;
1095 nr_pages
= get_nr_dirty_pages();
1098 struct wb_writeback_work work
= {
1099 .nr_pages
= nr_pages
,
1100 .sync_mode
= WB_SYNC_NONE
,
1103 .reason
= WB_REASON_PERIODIC
,
1106 return wb_writeback(wb
, &work
);
1113 * Retrieve work items and do the writeback they describe
1115 static long wb_do_writeback(struct bdi_writeback
*wb
)
1117 struct wb_writeback_work
*work
;
1120 set_bit(WB_writeback_running
, &wb
->state
);
1121 while ((work
= get_next_work_item(wb
)) != NULL
) {
1123 trace_writeback_exec(wb
->bdi
, work
);
1125 wrote
+= wb_writeback(wb
, work
);
1128 * Notify the caller of completion if this is a synchronous
1129 * work item, otherwise just free it.
1132 complete(work
->done
);
1138 * Check for periodic writeback, kupdated() style
1140 wrote
+= wb_check_old_data_flush(wb
);
1141 wrote
+= wb_check_background_flush(wb
);
1142 clear_bit(WB_writeback_running
, &wb
->state
);
1148 * Handle writeback of dirty data for the device backed by this bdi. Also
1149 * reschedules periodically and does kupdated style flushing.
1151 void wb_workfn(struct work_struct
*work
)
1153 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1154 struct bdi_writeback
, dwork
);
1157 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1158 current
->flags
|= PF_SWAPWRITE
;
1160 if (likely(!current_is_workqueue_rescuer() ||
1161 !test_bit(WB_registered
, &wb
->state
))) {
1163 * The normal path. Keep writing back @wb until its
1164 * work_list is empty. Note that this path is also taken
1165 * if @wb is shutting down even when we're running off the
1166 * rescuer as work_list needs to be drained.
1169 pages_written
= wb_do_writeback(wb
);
1170 trace_writeback_pages_written(pages_written
);
1171 } while (!list_empty(&wb
->work_list
));
1174 * bdi_wq can't get enough workers and we're running off
1175 * the emergency worker. Don't hog it. Hopefully, 1024 is
1176 * enough for efficient IO.
1178 pages_written
= writeback_inodes_wb(wb
, 1024,
1179 WB_REASON_FORKER_THREAD
);
1180 trace_writeback_pages_written(pages_written
);
1183 if (!list_empty(&wb
->work_list
))
1184 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1185 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1186 wb_wakeup_delayed(wb
);
1188 current
->flags
&= ~PF_SWAPWRITE
;
1192 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1195 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1197 struct backing_dev_info
*bdi
;
1200 nr_pages
= get_nr_dirty_pages();
1203 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
)
1204 wb_start_writeback(&bdi
->wb
, nr_pages
, false, reason
);
1209 * Wake up bdi's periodically to make sure dirtytime inodes gets
1210 * written back periodically. We deliberately do *not* check the
1211 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1212 * kernel to be constantly waking up once there are any dirtytime
1213 * inodes on the system. So instead we define a separate delayed work
1214 * function which gets called much more rarely. (By default, only
1215 * once every 12 hours.)
1217 * If there is any other write activity going on in the file system,
1218 * this function won't be necessary. But if the only thing that has
1219 * happened on the file system is a dirtytime inode caused by an atime
1220 * update, we need this infrastructure below to make sure that inode
1221 * eventually gets pushed out to disk.
1223 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1224 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1226 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1228 struct backing_dev_info
*bdi
;
1231 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1232 if (list_empty(&bdi
->wb
.b_dirty_time
))
1234 wb_wakeup(&bdi
->wb
);
1237 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1240 static int __init
start_dirtytime_writeback(void)
1242 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1245 __initcall(start_dirtytime_writeback
);
1247 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1248 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1252 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1253 if (ret
== 0 && write
)
1254 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1258 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1260 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1261 struct dentry
*dentry
;
1262 const char *name
= "?";
1264 dentry
= d_find_alias(inode
);
1266 spin_lock(&dentry
->d_lock
);
1267 name
= (const char *) dentry
->d_name
.name
;
1270 "%s(%d): dirtied inode %lu (%s) on %s\n",
1271 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1272 name
, inode
->i_sb
->s_id
);
1274 spin_unlock(&dentry
->d_lock
);
1281 * __mark_inode_dirty - internal function
1282 * @inode: inode to mark
1283 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1284 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1285 * mark_inode_dirty_sync.
1287 * Put the inode on the super block's dirty list.
1289 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1290 * dirty list only if it is hashed or if it refers to a blockdev.
1291 * If it was not hashed, it will never be added to the dirty list
1292 * even if it is later hashed, as it will have been marked dirty already.
1294 * In short, make sure you hash any inodes _before_ you start marking
1297 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1298 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1299 * the kernel-internal blockdev inode represents the dirtying time of the
1300 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1301 * page->mapping->host, so the page-dirtying time is recorded in the internal
1304 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1305 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1307 struct super_block
*sb
= inode
->i_sb
;
1308 struct backing_dev_info
*bdi
= NULL
;
1311 trace_writeback_mark_inode_dirty(inode
, flags
);
1314 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1315 * dirty the inode itself
1317 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
1318 trace_writeback_dirty_inode_start(inode
, flags
);
1320 if (sb
->s_op
->dirty_inode
)
1321 sb
->s_op
->dirty_inode(inode
, flags
);
1323 trace_writeback_dirty_inode(inode
, flags
);
1325 if (flags
& I_DIRTY_INODE
)
1326 flags
&= ~I_DIRTY_TIME
;
1327 dirtytime
= flags
& I_DIRTY_TIME
;
1330 * Paired with smp_mb() in __writeback_single_inode() for the
1331 * following lockless i_state test. See there for details.
1335 if (((inode
->i_state
& flags
) == flags
) ||
1336 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
1339 if (unlikely(block_dump
))
1340 block_dump___mark_inode_dirty(inode
);
1342 spin_lock(&inode
->i_lock
);
1343 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
1344 goto out_unlock_inode
;
1345 if ((inode
->i_state
& flags
) != flags
) {
1346 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1348 inode_attach_wb(inode
, NULL
);
1350 if (flags
& I_DIRTY_INODE
)
1351 inode
->i_state
&= ~I_DIRTY_TIME
;
1352 inode
->i_state
|= flags
;
1355 * If the inode is being synced, just update its dirty state.
1356 * The unlocker will place the inode on the appropriate
1357 * superblock list, based upon its state.
1359 if (inode
->i_state
& I_SYNC
)
1360 goto out_unlock_inode
;
1363 * Only add valid (hashed) inodes to the superblock's
1364 * dirty list. Add blockdev inodes as well.
1366 if (!S_ISBLK(inode
->i_mode
)) {
1367 if (inode_unhashed(inode
))
1368 goto out_unlock_inode
;
1370 if (inode
->i_state
& I_FREEING
)
1371 goto out_unlock_inode
;
1374 * If the inode was already on b_dirty/b_io/b_more_io, don't
1375 * reposition it (that would break b_dirty time-ordering).
1378 struct list_head
*dirty_list
;
1379 bool wakeup_bdi
= false;
1380 bdi
= inode_to_bdi(inode
);
1382 spin_unlock(&inode
->i_lock
);
1383 spin_lock(&bdi
->wb
.list_lock
);
1385 WARN(bdi_cap_writeback_dirty(bdi
) &&
1386 !test_bit(WB_registered
, &bdi
->wb
.state
),
1387 "bdi-%s not registered\n", bdi
->name
);
1389 inode
->dirtied_when
= jiffies
;
1391 inode
->dirtied_time_when
= jiffies
;
1393 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
1394 dirty_list
= &bdi
->wb
.b_dirty
;
1396 dirty_list
= &bdi
->wb
.b_dirty_time
;
1398 wakeup_bdi
= inode_wb_list_move_locked(inode
, &bdi
->wb
,
1401 spin_unlock(&bdi
->wb
.list_lock
);
1402 trace_writeback_dirty_inode_enqueue(inode
);
1405 * If this is the first dirty inode for this bdi,
1406 * we have to wake-up the corresponding bdi thread
1407 * to make sure background write-back happens
1410 if (bdi_cap_writeback_dirty(bdi
) && wakeup_bdi
)
1411 wb_wakeup_delayed(&bdi
->wb
);
1416 spin_unlock(&inode
->i_lock
);
1419 EXPORT_SYMBOL(__mark_inode_dirty
);
1421 static void wait_sb_inodes(struct super_block
*sb
)
1423 struct inode
*inode
, *old_inode
= NULL
;
1426 * We need to be protected against the filesystem going from
1427 * r/o to r/w or vice versa.
1429 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1431 spin_lock(&inode_sb_list_lock
);
1434 * Data integrity sync. Must wait for all pages under writeback,
1435 * because there may have been pages dirtied before our sync
1436 * call, but which had writeout started before we write it out.
1437 * In which case, the inode may not be on the dirty list, but
1438 * we still have to wait for that writeout.
1440 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1441 struct address_space
*mapping
= inode
->i_mapping
;
1443 spin_lock(&inode
->i_lock
);
1444 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1445 (mapping
->nrpages
== 0)) {
1446 spin_unlock(&inode
->i_lock
);
1450 spin_unlock(&inode
->i_lock
);
1451 spin_unlock(&inode_sb_list_lock
);
1454 * We hold a reference to 'inode' so it couldn't have been
1455 * removed from s_inodes list while we dropped the
1456 * inode_sb_list_lock. We cannot iput the inode now as we can
1457 * be holding the last reference and we cannot iput it under
1458 * inode_sb_list_lock. So we keep the reference and iput it
1464 filemap_fdatawait(mapping
);
1468 spin_lock(&inode_sb_list_lock
);
1470 spin_unlock(&inode_sb_list_lock
);
1475 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1476 * @sb: the superblock
1477 * @nr: the number of pages to write
1478 * @reason: reason why some writeback work initiated
1480 * Start writeback on some inodes on this super_block. No guarantees are made
1481 * on how many (if any) will be written, and this function does not wait
1482 * for IO completion of submitted IO.
1484 void writeback_inodes_sb_nr(struct super_block
*sb
,
1486 enum wb_reason reason
)
1488 DECLARE_COMPLETION_ONSTACK(done
);
1489 struct wb_writeback_work work
= {
1491 .sync_mode
= WB_SYNC_NONE
,
1492 .tagged_writepages
= 1,
1497 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1499 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1501 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1502 wb_queue_work(&bdi
->wb
, &work
);
1503 wait_for_completion(&done
);
1505 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1508 * writeback_inodes_sb - writeback dirty inodes from given super_block
1509 * @sb: the superblock
1510 * @reason: reason why some writeback work was initiated
1512 * Start writeback on some inodes on this super_block. No guarantees are made
1513 * on how many (if any) will be written, and this function does not wait
1514 * for IO completion of submitted IO.
1516 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1518 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1520 EXPORT_SYMBOL(writeback_inodes_sb
);
1523 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1524 * @sb: the superblock
1525 * @nr: the number of pages to write
1526 * @reason: the reason of writeback
1528 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1529 * Returns 1 if writeback was started, 0 if not.
1531 int try_to_writeback_inodes_sb_nr(struct super_block
*sb
,
1533 enum wb_reason reason
)
1535 if (writeback_in_progress(sb
->s_bdi
))
1538 if (!down_read_trylock(&sb
->s_umount
))
1541 writeback_inodes_sb_nr(sb
, nr
, reason
);
1542 up_read(&sb
->s_umount
);
1545 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1548 * try_to_writeback_inodes_sb - try to start writeback if none underway
1549 * @sb: the superblock
1550 * @reason: reason why some writeback work was initiated
1552 * Implement by try_to_writeback_inodes_sb_nr()
1553 * Returns 1 if writeback was started, 0 if not.
1555 int try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1557 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1559 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1562 * sync_inodes_sb - sync sb inode pages
1563 * @sb: the superblock
1565 * This function writes and waits on any dirty inode belonging to this
1568 void sync_inodes_sb(struct super_block
*sb
)
1570 DECLARE_COMPLETION_ONSTACK(done
);
1571 struct wb_writeback_work work
= {
1573 .sync_mode
= WB_SYNC_ALL
,
1574 .nr_pages
= LONG_MAX
,
1577 .reason
= WB_REASON_SYNC
,
1580 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1582 /* Nothing to do? */
1583 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1585 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1587 wb_queue_work(&bdi
->wb
, &work
);
1588 wait_for_completion(&done
);
1592 EXPORT_SYMBOL(sync_inodes_sb
);
1595 * write_inode_now - write an inode to disk
1596 * @inode: inode to write to disk
1597 * @sync: whether the write should be synchronous or not
1599 * This function commits an inode to disk immediately if it is dirty. This is
1600 * primarily needed by knfsd.
1602 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1604 int write_inode_now(struct inode
*inode
, int sync
)
1606 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1607 struct writeback_control wbc
= {
1608 .nr_to_write
= LONG_MAX
,
1609 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1611 .range_end
= LLONG_MAX
,
1614 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1615 wbc
.nr_to_write
= 0;
1618 return writeback_single_inode(inode
, wb
, &wbc
);
1620 EXPORT_SYMBOL(write_inode_now
);
1623 * sync_inode - write an inode and its pages to disk.
1624 * @inode: the inode to sync
1625 * @wbc: controls the writeback mode
1627 * sync_inode() will write an inode and its pages to disk. It will also
1628 * correctly update the inode on its superblock's dirty inode lists and will
1629 * update inode->i_state.
1631 * The caller must have a ref on the inode.
1633 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1635 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1637 EXPORT_SYMBOL(sync_inode
);
1640 * sync_inode_metadata - write an inode to disk
1641 * @inode: the inode to sync
1642 * @wait: wait for I/O to complete.
1644 * Write an inode to disk and adjust its dirty state after completion.
1646 * Note: only writes the actual inode, no associated data or other metadata.
1648 int sync_inode_metadata(struct inode
*inode
, int wait
)
1650 struct writeback_control wbc
= {
1651 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1652 .nr_to_write
= 0, /* metadata-only */
1655 return sync_inode(inode
, &wbc
);
1657 EXPORT_SYMBOL(sync_inode_metadata
);