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;
68 static inline struct inode
*wb_inode(struct list_head
*head
)
70 return list_entry(head
, struct inode
, i_wb_list
);
74 * Include the creation of the trace points after defining the
75 * wb_writeback_work structure and inline functions so that the definition
76 * remains local to this file.
78 #define CREATE_TRACE_POINTS
79 #include <trace/events/writeback.h>
81 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage
);
83 static bool wb_io_lists_populated(struct bdi_writeback
*wb
)
85 if (wb_has_dirty_io(wb
)) {
88 set_bit(WB_has_dirty_io
, &wb
->state
);
89 WARN_ON_ONCE(!wb
->avg_write_bandwidth
);
90 atomic_long_add(wb
->avg_write_bandwidth
,
91 &wb
->bdi
->tot_write_bandwidth
);
96 static void wb_io_lists_depopulated(struct bdi_writeback
*wb
)
98 if (wb_has_dirty_io(wb
) && list_empty(&wb
->b_dirty
) &&
99 list_empty(&wb
->b_io
) && list_empty(&wb
->b_more_io
)) {
100 clear_bit(WB_has_dirty_io
, &wb
->state
);
101 WARN_ON_ONCE(atomic_long_sub_return(wb
->avg_write_bandwidth
,
102 &wb
->bdi
->tot_write_bandwidth
) < 0);
107 * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list
108 * @inode: inode to be moved
109 * @wb: target bdi_writeback
110 * @head: one of @wb->b_{dirty|io|more_io}
112 * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io.
113 * Returns %true if @inode is the first occupant of the !dirty_time IO
114 * lists; otherwise, %false.
116 static bool inode_wb_list_move_locked(struct inode
*inode
,
117 struct bdi_writeback
*wb
,
118 struct list_head
*head
)
120 assert_spin_locked(&wb
->list_lock
);
122 list_move(&inode
->i_wb_list
, head
);
124 /* dirty_time doesn't count as dirty_io until expiration */
125 if (head
!= &wb
->b_dirty_time
)
126 return wb_io_lists_populated(wb
);
128 wb_io_lists_depopulated(wb
);
133 * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list
134 * @inode: inode to be removed
135 * @wb: bdi_writeback @inode is being removed from
137 * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and
138 * clear %WB_has_dirty_io if all are empty afterwards.
140 static void inode_wb_list_del_locked(struct inode
*inode
,
141 struct bdi_writeback
*wb
)
143 assert_spin_locked(&wb
->list_lock
);
145 list_del_init(&inode
->i_wb_list
);
146 wb_io_lists_depopulated(wb
);
149 static void wb_wakeup(struct bdi_writeback
*wb
)
151 spin_lock_bh(&wb
->work_lock
);
152 if (test_bit(WB_registered
, &wb
->state
))
153 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
154 spin_unlock_bh(&wb
->work_lock
);
157 static void wb_queue_work(struct bdi_writeback
*wb
,
158 struct wb_writeback_work
*work
)
160 trace_writeback_queue(wb
->bdi
, work
);
162 spin_lock_bh(&wb
->work_lock
);
163 if (!test_bit(WB_registered
, &wb
->state
)) {
165 complete(work
->done
);
168 list_add_tail(&work
->list
, &wb
->work_list
);
169 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
171 spin_unlock_bh(&wb
->work_lock
);
174 #ifdef CONFIG_CGROUP_WRITEBACK
177 * inode_congested - test whether an inode is congested
178 * @inode: inode to test for congestion
179 * @cong_bits: mask of WB_[a]sync_congested bits to test
181 * Tests whether @inode is congested. @cong_bits is the mask of congestion
182 * bits to test and the return value is the mask of set bits.
184 * If cgroup writeback is enabled for @inode, the congestion state is
185 * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
186 * associated with @inode is congested; otherwise, the root wb's congestion
189 int inode_congested(struct inode
*inode
, int cong_bits
)
192 struct bdi_writeback
*wb
= inode_to_wb(inode
);
194 return wb_congested(wb
, cong_bits
);
197 return wb_congested(&inode_to_bdi(inode
)->wb
, cong_bits
);
199 EXPORT_SYMBOL_GPL(inode_congested
);
202 * wb_split_bdi_pages - split nr_pages to write according to bandwidth
203 * @wb: target bdi_writeback to split @nr_pages to
204 * @nr_pages: number of pages to write for the whole bdi
206 * Split @wb's portion of @nr_pages according to @wb's write bandwidth in
207 * relation to the total write bandwidth of all wb's w/ dirty inodes on
210 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
212 unsigned long this_bw
= wb
->avg_write_bandwidth
;
213 unsigned long tot_bw
= atomic_long_read(&wb
->bdi
->tot_write_bandwidth
);
215 if (nr_pages
== LONG_MAX
)
219 * This may be called on clean wb's and proportional distribution
220 * may not make sense, just use the original @nr_pages in those
221 * cases. In general, we wanna err on the side of writing more.
223 if (!tot_bw
|| this_bw
>= tot_bw
)
226 return DIV_ROUND_UP_ULL((u64
)nr_pages
* this_bw
, tot_bw
);
229 #else /* CONFIG_CGROUP_WRITEBACK */
231 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
236 #endif /* CONFIG_CGROUP_WRITEBACK */
238 void wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
239 bool range_cyclic
, enum wb_reason reason
)
241 struct wb_writeback_work
*work
;
243 if (!wb_has_dirty_io(wb
))
247 * This is WB_SYNC_NONE writeback, so if allocation fails just
248 * wakeup the thread for old dirty data writeback
250 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
252 trace_writeback_nowork(wb
->bdi
);
257 work
->sync_mode
= WB_SYNC_NONE
;
258 work
->nr_pages
= nr_pages
;
259 work
->range_cyclic
= range_cyclic
;
260 work
->reason
= reason
;
262 wb_queue_work(wb
, work
);
266 * wb_start_background_writeback - start background writeback
267 * @wb: bdi_writback to write from
270 * This makes sure WB_SYNC_NONE background writeback happens. When
271 * this function returns, it is only guaranteed that for given wb
272 * some IO is happening if we are over background dirty threshold.
273 * Caller need not hold sb s_umount semaphore.
275 void wb_start_background_writeback(struct bdi_writeback
*wb
)
278 * We just wake up the flusher thread. It will perform background
279 * writeback as soon as there is no other work to do.
281 trace_writeback_wake_background(wb
->bdi
);
286 * Remove the inode from the writeback list it is on.
288 void inode_wb_list_del(struct inode
*inode
)
290 struct bdi_writeback
*wb
= inode_to_wb(inode
);
292 spin_lock(&wb
->list_lock
);
293 inode_wb_list_del_locked(inode
, wb
);
294 spin_unlock(&wb
->list_lock
);
298 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
299 * furthest end of its superblock's dirty-inode list.
301 * Before stamping the inode's ->dirtied_when, we check to see whether it is
302 * already the most-recently-dirtied inode on the b_dirty list. If that is
303 * the case then the inode must have been redirtied while it was being written
304 * out and we don't reset its dirtied_when.
306 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
308 if (!list_empty(&wb
->b_dirty
)) {
311 tail
= wb_inode(wb
->b_dirty
.next
);
312 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
313 inode
->dirtied_when
= jiffies
;
315 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty
);
319 * requeue inode for re-scanning after bdi->b_io list is exhausted.
321 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
323 inode_wb_list_move_locked(inode
, wb
, &wb
->b_more_io
);
326 static void inode_sync_complete(struct inode
*inode
)
328 inode
->i_state
&= ~I_SYNC
;
329 /* If inode is clean an unused, put it into LRU now... */
330 inode_add_lru(inode
);
331 /* Waiters must see I_SYNC cleared before being woken up */
333 wake_up_bit(&inode
->i_state
, __I_SYNC
);
336 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
338 bool ret
= time_after(inode
->dirtied_when
, t
);
341 * For inodes being constantly redirtied, dirtied_when can get stuck.
342 * It _appears_ to be in the future, but is actually in distant past.
343 * This test is necessary to prevent such wrapped-around relative times
344 * from permanently stopping the whole bdi writeback.
346 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
351 #define EXPIRE_DIRTY_ATIME 0x0001
354 * Move expired (dirtied before work->older_than_this) dirty inodes from
355 * @delaying_queue to @dispatch_queue.
357 static int move_expired_inodes(struct list_head
*delaying_queue
,
358 struct list_head
*dispatch_queue
,
360 struct wb_writeback_work
*work
)
362 unsigned long *older_than_this
= NULL
;
363 unsigned long expire_time
;
365 struct list_head
*pos
, *node
;
366 struct super_block
*sb
= NULL
;
371 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
372 older_than_this
= work
->older_than_this
;
373 else if (!work
->for_sync
) {
374 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
375 older_than_this
= &expire_time
;
377 while (!list_empty(delaying_queue
)) {
378 inode
= wb_inode(delaying_queue
->prev
);
379 if (older_than_this
&&
380 inode_dirtied_after(inode
, *older_than_this
))
382 list_move(&inode
->i_wb_list
, &tmp
);
384 if (flags
& EXPIRE_DIRTY_ATIME
)
385 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
386 if (sb_is_blkdev_sb(inode
->i_sb
))
388 if (sb
&& sb
!= inode
->i_sb
)
393 /* just one sb in list, splice to dispatch_queue and we're done */
395 list_splice(&tmp
, dispatch_queue
);
399 /* Move inodes from one superblock together */
400 while (!list_empty(&tmp
)) {
401 sb
= wb_inode(tmp
.prev
)->i_sb
;
402 list_for_each_prev_safe(pos
, node
, &tmp
) {
403 inode
= wb_inode(pos
);
404 if (inode
->i_sb
== sb
)
405 list_move(&inode
->i_wb_list
, dispatch_queue
);
413 * Queue all expired dirty inodes for io, eldest first.
415 * newly dirtied b_dirty b_io b_more_io
416 * =============> gf edc BA
418 * newly dirtied b_dirty b_io b_more_io
419 * =============> g fBAedc
421 * +--> dequeue for IO
423 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
427 assert_spin_locked(&wb
->list_lock
);
428 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
429 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
430 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
431 EXPIRE_DIRTY_ATIME
, work
);
433 wb_io_lists_populated(wb
);
434 trace_writeback_queue_io(wb
, work
, moved
);
437 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
441 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
442 trace_writeback_write_inode_start(inode
, wbc
);
443 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
444 trace_writeback_write_inode(inode
, wbc
);
451 * Wait for writeback on an inode to complete. Called with i_lock held.
452 * Caller must make sure inode cannot go away when we drop i_lock.
454 static void __inode_wait_for_writeback(struct inode
*inode
)
455 __releases(inode
->i_lock
)
456 __acquires(inode
->i_lock
)
458 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
459 wait_queue_head_t
*wqh
;
461 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
462 while (inode
->i_state
& I_SYNC
) {
463 spin_unlock(&inode
->i_lock
);
464 __wait_on_bit(wqh
, &wq
, bit_wait
,
465 TASK_UNINTERRUPTIBLE
);
466 spin_lock(&inode
->i_lock
);
471 * Wait for writeback on an inode to complete. Caller must have inode pinned.
473 void inode_wait_for_writeback(struct inode
*inode
)
475 spin_lock(&inode
->i_lock
);
476 __inode_wait_for_writeback(inode
);
477 spin_unlock(&inode
->i_lock
);
481 * Sleep until I_SYNC is cleared. This function must be called with i_lock
482 * held and drops it. It is aimed for callers not holding any inode reference
483 * so once i_lock is dropped, inode can go away.
485 static void inode_sleep_on_writeback(struct inode
*inode
)
486 __releases(inode
->i_lock
)
489 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
492 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
493 sleep
= inode
->i_state
& I_SYNC
;
494 spin_unlock(&inode
->i_lock
);
497 finish_wait(wqh
, &wait
);
501 * Find proper writeback list for the inode depending on its current state and
502 * possibly also change of its state while we were doing writeback. Here we
503 * handle things such as livelock prevention or fairness of writeback among
504 * inodes. This function can be called only by flusher thread - noone else
505 * processes all inodes in writeback lists and requeueing inodes behind flusher
506 * thread's back can have unexpected consequences.
508 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
509 struct writeback_control
*wbc
)
511 if (inode
->i_state
& I_FREEING
)
515 * Sync livelock prevention. Each inode is tagged and synced in one
516 * shot. If still dirty, it will be redirty_tail()'ed below. Update
517 * the dirty time to prevent enqueue and sync it again.
519 if ((inode
->i_state
& I_DIRTY
) &&
520 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
521 inode
->dirtied_when
= jiffies
;
523 if (wbc
->pages_skipped
) {
525 * writeback is not making progress due to locked
526 * buffers. Skip this inode for now.
528 redirty_tail(inode
, wb
);
532 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
534 * We didn't write back all the pages. nfs_writepages()
535 * sometimes bales out without doing anything.
537 if (wbc
->nr_to_write
<= 0) {
538 /* Slice used up. Queue for next turn. */
539 requeue_io(inode
, wb
);
542 * Writeback blocked by something other than
543 * congestion. Delay the inode for some time to
544 * avoid spinning on the CPU (100% iowait)
545 * retrying writeback of the dirty page/inode
546 * that cannot be performed immediately.
548 redirty_tail(inode
, wb
);
550 } else if (inode
->i_state
& I_DIRTY
) {
552 * Filesystems can dirty the inode during writeback operations,
553 * such as delayed allocation during submission or metadata
554 * updates after data IO completion.
556 redirty_tail(inode
, wb
);
557 } else if (inode
->i_state
& I_DIRTY_TIME
) {
558 inode
->dirtied_when
= jiffies
;
559 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty_time
);
561 /* The inode is clean. Remove from writeback lists. */
562 inode_wb_list_del_locked(inode
, wb
);
567 * Write out an inode and its dirty pages. Do not update the writeback list
568 * linkage. That is left to the caller. The caller is also responsible for
569 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
572 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
574 struct address_space
*mapping
= inode
->i_mapping
;
575 long nr_to_write
= wbc
->nr_to_write
;
579 WARN_ON(!(inode
->i_state
& I_SYNC
));
581 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
583 ret
= do_writepages(mapping
, wbc
);
586 * Make sure to wait on the data before writing out the metadata.
587 * This is important for filesystems that modify metadata on data
588 * I/O completion. We don't do it for sync(2) writeback because it has a
589 * separate, external IO completion path and ->sync_fs for guaranteeing
590 * inode metadata is written back correctly.
592 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
593 int err
= filemap_fdatawait(mapping
);
599 * Some filesystems may redirty the inode during the writeback
600 * due to delalloc, clear dirty metadata flags right before
603 spin_lock(&inode
->i_lock
);
605 dirty
= inode
->i_state
& I_DIRTY
;
606 if (inode
->i_state
& I_DIRTY_TIME
) {
607 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
608 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
609 unlikely(time_after(jiffies
,
610 (inode
->dirtied_time_when
+
611 dirtytime_expire_interval
* HZ
)))) {
612 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
613 trace_writeback_lazytime(inode
);
616 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
617 inode
->i_state
&= ~dirty
;
620 * Paired with smp_mb() in __mark_inode_dirty(). This allows
621 * __mark_inode_dirty() to test i_state without grabbing i_lock -
622 * either they see the I_DIRTY bits cleared or we see the dirtied
625 * I_DIRTY_PAGES is always cleared together above even if @mapping
626 * still has dirty pages. The flag is reinstated after smp_mb() if
627 * necessary. This guarantees that either __mark_inode_dirty()
628 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
632 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
633 inode
->i_state
|= I_DIRTY_PAGES
;
635 spin_unlock(&inode
->i_lock
);
637 if (dirty
& I_DIRTY_TIME
)
638 mark_inode_dirty_sync(inode
);
639 /* Don't write the inode if only I_DIRTY_PAGES was set */
640 if (dirty
& ~I_DIRTY_PAGES
) {
641 int err
= write_inode(inode
, wbc
);
645 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
650 * Write out an inode's dirty pages. Either the caller has an active reference
651 * on the inode or the inode has I_WILL_FREE set.
653 * This function is designed to be called for writing back one inode which
654 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
655 * and does more profound writeback list handling in writeback_sb_inodes().
658 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
659 struct writeback_control
*wbc
)
663 spin_lock(&inode
->i_lock
);
664 if (!atomic_read(&inode
->i_count
))
665 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
667 WARN_ON(inode
->i_state
& I_WILL_FREE
);
669 if (inode
->i_state
& I_SYNC
) {
670 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
673 * It's a data-integrity sync. We must wait. Since callers hold
674 * inode reference or inode has I_WILL_FREE set, it cannot go
677 __inode_wait_for_writeback(inode
);
679 WARN_ON(inode
->i_state
& I_SYNC
);
681 * Skip inode if it is clean and we have no outstanding writeback in
682 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
683 * function since flusher thread may be doing for example sync in
684 * parallel and if we move the inode, it could get skipped. So here we
685 * make sure inode is on some writeback list and leave it there unless
686 * we have completely cleaned the inode.
688 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
689 (wbc
->sync_mode
!= WB_SYNC_ALL
||
690 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
692 inode
->i_state
|= I_SYNC
;
693 spin_unlock(&inode
->i_lock
);
695 ret
= __writeback_single_inode(inode
, wbc
);
697 spin_lock(&wb
->list_lock
);
698 spin_lock(&inode
->i_lock
);
700 * If inode is clean, remove it from writeback lists. Otherwise don't
701 * touch it. See comment above for explanation.
703 if (!(inode
->i_state
& I_DIRTY_ALL
))
704 inode_wb_list_del_locked(inode
, wb
);
705 spin_unlock(&wb
->list_lock
);
706 inode_sync_complete(inode
);
708 spin_unlock(&inode
->i_lock
);
712 static long writeback_chunk_size(struct bdi_writeback
*wb
,
713 struct wb_writeback_work
*work
)
718 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
719 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
720 * here avoids calling into writeback_inodes_wb() more than once.
722 * The intended call sequence for WB_SYNC_ALL writeback is:
725 * writeback_sb_inodes() <== called only once
726 * write_cache_pages() <== called once for each inode
727 * (quickly) tag currently dirty pages
728 * (maybe slowly) sync all tagged pages
730 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
733 pages
= min(wb
->avg_write_bandwidth
/ 2,
734 global_dirty_limit
/ DIRTY_SCOPE
);
735 pages
= min(pages
, work
->nr_pages
);
736 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
737 MIN_WRITEBACK_PAGES
);
744 * Write a portion of b_io inodes which belong to @sb.
746 * Return the number of pages and/or inodes written.
748 static long writeback_sb_inodes(struct super_block
*sb
,
749 struct bdi_writeback
*wb
,
750 struct wb_writeback_work
*work
)
752 struct writeback_control wbc
= {
753 .sync_mode
= work
->sync_mode
,
754 .tagged_writepages
= work
->tagged_writepages
,
755 .for_kupdate
= work
->for_kupdate
,
756 .for_background
= work
->for_background
,
757 .for_sync
= work
->for_sync
,
758 .range_cyclic
= work
->range_cyclic
,
760 .range_end
= LLONG_MAX
,
762 unsigned long start_time
= jiffies
;
764 long wrote
= 0; /* count both pages and inodes */
766 while (!list_empty(&wb
->b_io
)) {
767 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
769 if (inode
->i_sb
!= sb
) {
772 * We only want to write back data for this
773 * superblock, move all inodes not belonging
774 * to it back onto the dirty list.
776 redirty_tail(inode
, wb
);
781 * The inode belongs to a different superblock.
782 * Bounce back to the caller to unpin this and
783 * pin the next superblock.
789 * Don't bother with new inodes or inodes being freed, first
790 * kind does not need periodic writeout yet, and for the latter
791 * kind writeout is handled by the freer.
793 spin_lock(&inode
->i_lock
);
794 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
795 spin_unlock(&inode
->i_lock
);
796 redirty_tail(inode
, wb
);
799 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
801 * If this inode is locked for writeback and we are not
802 * doing writeback-for-data-integrity, move it to
803 * b_more_io so that writeback can proceed with the
804 * other inodes on s_io.
806 * We'll have another go at writing back this inode
807 * when we completed a full scan of b_io.
809 spin_unlock(&inode
->i_lock
);
810 requeue_io(inode
, wb
);
811 trace_writeback_sb_inodes_requeue(inode
);
814 spin_unlock(&wb
->list_lock
);
817 * We already requeued the inode if it had I_SYNC set and we
818 * are doing WB_SYNC_NONE writeback. So this catches only the
821 if (inode
->i_state
& I_SYNC
) {
822 /* Wait for I_SYNC. This function drops i_lock... */
823 inode_sleep_on_writeback(inode
);
824 /* Inode may be gone, start again */
825 spin_lock(&wb
->list_lock
);
828 inode
->i_state
|= I_SYNC
;
829 spin_unlock(&inode
->i_lock
);
831 write_chunk
= writeback_chunk_size(wb
, work
);
832 wbc
.nr_to_write
= write_chunk
;
833 wbc
.pages_skipped
= 0;
836 * We use I_SYNC to pin the inode in memory. While it is set
837 * evict_inode() will wait so the inode cannot be freed.
839 __writeback_single_inode(inode
, &wbc
);
841 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
842 wrote
+= write_chunk
- wbc
.nr_to_write
;
843 spin_lock(&wb
->list_lock
);
844 spin_lock(&inode
->i_lock
);
845 if (!(inode
->i_state
& I_DIRTY_ALL
))
847 requeue_inode(inode
, wb
, &wbc
);
848 inode_sync_complete(inode
);
849 spin_unlock(&inode
->i_lock
);
850 cond_resched_lock(&wb
->list_lock
);
852 * bail out to wb_writeback() often enough to check
853 * background threshold and other termination conditions.
856 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
858 if (work
->nr_pages
<= 0)
865 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
866 struct wb_writeback_work
*work
)
868 unsigned long start_time
= jiffies
;
871 while (!list_empty(&wb
->b_io
)) {
872 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
873 struct super_block
*sb
= inode
->i_sb
;
875 if (!trylock_super(sb
)) {
877 * trylock_super() may fail consistently due to
878 * s_umount being grabbed by someone else. Don't use
879 * requeue_io() to avoid busy retrying the inode/sb.
881 redirty_tail(inode
, wb
);
884 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
885 up_read(&sb
->s_umount
);
887 /* refer to the same tests at the end of writeback_sb_inodes */
889 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
891 if (work
->nr_pages
<= 0)
895 /* Leave any unwritten inodes on b_io */
899 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
900 enum wb_reason reason
)
902 struct wb_writeback_work work
= {
903 .nr_pages
= nr_pages
,
904 .sync_mode
= WB_SYNC_NONE
,
909 spin_lock(&wb
->list_lock
);
910 if (list_empty(&wb
->b_io
))
912 __writeback_inodes_wb(wb
, &work
);
913 spin_unlock(&wb
->list_lock
);
915 return nr_pages
- work
.nr_pages
;
918 static bool over_bground_thresh(struct bdi_writeback
*wb
)
920 unsigned long background_thresh
, dirty_thresh
;
922 global_dirty_limits(&background_thresh
, &dirty_thresh
);
924 if (global_page_state(NR_FILE_DIRTY
) +
925 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
)
928 if (wb_stat(wb
, WB_RECLAIMABLE
) > wb_dirty_limit(wb
, background_thresh
))
935 * Called under wb->list_lock. If there are multiple wb per bdi,
936 * only the flusher working on the first wb should do it.
938 static void wb_update_bandwidth(struct bdi_writeback
*wb
,
939 unsigned long start_time
)
941 __wb_update_bandwidth(wb
, 0, 0, 0, 0, 0, start_time
);
945 * Explicit flushing or periodic writeback of "old" data.
947 * Define "old": the first time one of an inode's pages is dirtied, we mark the
948 * dirtying-time in the inode's address_space. So this periodic writeback code
949 * just walks the superblock inode list, writing back any inodes which are
950 * older than a specific point in time.
952 * Try to run once per dirty_writeback_interval. But if a writeback event
953 * takes longer than a dirty_writeback_interval interval, then leave a
956 * older_than_this takes precedence over nr_to_write. So we'll only write back
957 * all dirty pages if they are all attached to "old" mappings.
959 static long wb_writeback(struct bdi_writeback
*wb
,
960 struct wb_writeback_work
*work
)
962 unsigned long wb_start
= jiffies
;
963 long nr_pages
= work
->nr_pages
;
964 unsigned long oldest_jif
;
968 oldest_jif
= jiffies
;
969 work
->older_than_this
= &oldest_jif
;
971 spin_lock(&wb
->list_lock
);
974 * Stop writeback when nr_pages has been consumed
976 if (work
->nr_pages
<= 0)
980 * Background writeout and kupdate-style writeback may
981 * run forever. Stop them if there is other work to do
982 * so that e.g. sync can proceed. They'll be restarted
983 * after the other works are all done.
985 if ((work
->for_background
|| work
->for_kupdate
) &&
986 !list_empty(&wb
->work_list
))
990 * For background writeout, stop when we are below the
991 * background dirty threshold
993 if (work
->for_background
&& !over_bground_thresh(wb
))
997 * Kupdate and background works are special and we want to
998 * include all inodes that need writing. Livelock avoidance is
999 * handled by these works yielding to any other work so we are
1002 if (work
->for_kupdate
) {
1003 oldest_jif
= jiffies
-
1004 msecs_to_jiffies(dirty_expire_interval
* 10);
1005 } else if (work
->for_background
)
1006 oldest_jif
= jiffies
;
1008 trace_writeback_start(wb
->bdi
, work
);
1009 if (list_empty(&wb
->b_io
))
1012 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
1014 progress
= __writeback_inodes_wb(wb
, work
);
1015 trace_writeback_written(wb
->bdi
, work
);
1017 wb_update_bandwidth(wb
, wb_start
);
1020 * Did we write something? Try for more
1022 * Dirty inodes are moved to b_io for writeback in batches.
1023 * The completion of the current batch does not necessarily
1024 * mean the overall work is done. So we keep looping as long
1025 * as made some progress on cleaning pages or inodes.
1030 * No more inodes for IO, bail
1032 if (list_empty(&wb
->b_more_io
))
1035 * Nothing written. Wait for some inode to
1036 * become available for writeback. Otherwise
1037 * we'll just busyloop.
1039 if (!list_empty(&wb
->b_more_io
)) {
1040 trace_writeback_wait(wb
->bdi
, work
);
1041 inode
= wb_inode(wb
->b_more_io
.prev
);
1042 spin_lock(&inode
->i_lock
);
1043 spin_unlock(&wb
->list_lock
);
1044 /* This function drops i_lock... */
1045 inode_sleep_on_writeback(inode
);
1046 spin_lock(&wb
->list_lock
);
1049 spin_unlock(&wb
->list_lock
);
1051 return nr_pages
- work
->nr_pages
;
1055 * Return the next wb_writeback_work struct that hasn't been processed yet.
1057 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
1059 struct wb_writeback_work
*work
= NULL
;
1061 spin_lock_bh(&wb
->work_lock
);
1062 if (!list_empty(&wb
->work_list
)) {
1063 work
= list_entry(wb
->work_list
.next
,
1064 struct wb_writeback_work
, list
);
1065 list_del_init(&work
->list
);
1067 spin_unlock_bh(&wb
->work_lock
);
1072 * Add in the number of potentially dirty inodes, because each inode
1073 * write can dirty pagecache in the underlying blockdev.
1075 static unsigned long get_nr_dirty_pages(void)
1077 return global_page_state(NR_FILE_DIRTY
) +
1078 global_page_state(NR_UNSTABLE_NFS
) +
1079 get_nr_dirty_inodes();
1082 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1084 if (over_bground_thresh(wb
)) {
1086 struct wb_writeback_work work
= {
1087 .nr_pages
= LONG_MAX
,
1088 .sync_mode
= WB_SYNC_NONE
,
1089 .for_background
= 1,
1091 .reason
= WB_REASON_BACKGROUND
,
1094 return wb_writeback(wb
, &work
);
1100 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1102 unsigned long expired
;
1106 * When set to zero, disable periodic writeback
1108 if (!dirty_writeback_interval
)
1111 expired
= wb
->last_old_flush
+
1112 msecs_to_jiffies(dirty_writeback_interval
* 10);
1113 if (time_before(jiffies
, expired
))
1116 wb
->last_old_flush
= jiffies
;
1117 nr_pages
= get_nr_dirty_pages();
1120 struct wb_writeback_work work
= {
1121 .nr_pages
= nr_pages
,
1122 .sync_mode
= WB_SYNC_NONE
,
1125 .reason
= WB_REASON_PERIODIC
,
1128 return wb_writeback(wb
, &work
);
1135 * Retrieve work items and do the writeback they describe
1137 static long wb_do_writeback(struct bdi_writeback
*wb
)
1139 struct wb_writeback_work
*work
;
1142 set_bit(WB_writeback_running
, &wb
->state
);
1143 while ((work
= get_next_work_item(wb
)) != NULL
) {
1145 trace_writeback_exec(wb
->bdi
, work
);
1147 wrote
+= wb_writeback(wb
, work
);
1150 * Notify the caller of completion if this is a synchronous
1151 * work item, otherwise just free it.
1154 complete(work
->done
);
1160 * Check for periodic writeback, kupdated() style
1162 wrote
+= wb_check_old_data_flush(wb
);
1163 wrote
+= wb_check_background_flush(wb
);
1164 clear_bit(WB_writeback_running
, &wb
->state
);
1170 * Handle writeback of dirty data for the device backed by this bdi. Also
1171 * reschedules periodically and does kupdated style flushing.
1173 void wb_workfn(struct work_struct
*work
)
1175 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1176 struct bdi_writeback
, dwork
);
1179 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1180 current
->flags
|= PF_SWAPWRITE
;
1182 if (likely(!current_is_workqueue_rescuer() ||
1183 !test_bit(WB_registered
, &wb
->state
))) {
1185 * The normal path. Keep writing back @wb until its
1186 * work_list is empty. Note that this path is also taken
1187 * if @wb is shutting down even when we're running off the
1188 * rescuer as work_list needs to be drained.
1191 pages_written
= wb_do_writeback(wb
);
1192 trace_writeback_pages_written(pages_written
);
1193 } while (!list_empty(&wb
->work_list
));
1196 * bdi_wq can't get enough workers and we're running off
1197 * the emergency worker. Don't hog it. Hopefully, 1024 is
1198 * enough for efficient IO.
1200 pages_written
= writeback_inodes_wb(wb
, 1024,
1201 WB_REASON_FORKER_THREAD
);
1202 trace_writeback_pages_written(pages_written
);
1205 if (!list_empty(&wb
->work_list
))
1206 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1207 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1208 wb_wakeup_delayed(wb
);
1210 current
->flags
&= ~PF_SWAPWRITE
;
1214 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1217 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1219 struct backing_dev_info
*bdi
;
1222 nr_pages
= get_nr_dirty_pages();
1225 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1226 struct bdi_writeback
*wb
;
1227 struct wb_iter iter
;
1229 if (!bdi_has_dirty_io(bdi
))
1232 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1233 wb_start_writeback(wb
, wb_split_bdi_pages(wb
, nr_pages
),
1240 * Wake up bdi's periodically to make sure dirtytime inodes gets
1241 * written back periodically. We deliberately do *not* check the
1242 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1243 * kernel to be constantly waking up once there are any dirtytime
1244 * inodes on the system. So instead we define a separate delayed work
1245 * function which gets called much more rarely. (By default, only
1246 * once every 12 hours.)
1248 * If there is any other write activity going on in the file system,
1249 * this function won't be necessary. But if the only thing that has
1250 * happened on the file system is a dirtytime inode caused by an atime
1251 * update, we need this infrastructure below to make sure that inode
1252 * eventually gets pushed out to disk.
1254 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1255 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1257 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1259 struct backing_dev_info
*bdi
;
1262 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1263 if (list_empty(&bdi
->wb
.b_dirty_time
))
1265 wb_wakeup(&bdi
->wb
);
1268 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1271 static int __init
start_dirtytime_writeback(void)
1273 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1276 __initcall(start_dirtytime_writeback
);
1278 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1279 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1283 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1284 if (ret
== 0 && write
)
1285 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1289 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1291 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1292 struct dentry
*dentry
;
1293 const char *name
= "?";
1295 dentry
= d_find_alias(inode
);
1297 spin_lock(&dentry
->d_lock
);
1298 name
= (const char *) dentry
->d_name
.name
;
1301 "%s(%d): dirtied inode %lu (%s) on %s\n",
1302 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1303 name
, inode
->i_sb
->s_id
);
1305 spin_unlock(&dentry
->d_lock
);
1312 * __mark_inode_dirty - internal function
1313 * @inode: inode to mark
1314 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1315 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1316 * mark_inode_dirty_sync.
1318 * Put the inode on the super block's dirty list.
1320 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1321 * dirty list only if it is hashed or if it refers to a blockdev.
1322 * If it was not hashed, it will never be added to the dirty list
1323 * even if it is later hashed, as it will have been marked dirty already.
1325 * In short, make sure you hash any inodes _before_ you start marking
1328 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1329 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1330 * the kernel-internal blockdev inode represents the dirtying time of the
1331 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1332 * page->mapping->host, so the page-dirtying time is recorded in the internal
1335 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1336 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1338 struct super_block
*sb
= inode
->i_sb
;
1339 struct backing_dev_info
*bdi
= NULL
;
1342 trace_writeback_mark_inode_dirty(inode
, flags
);
1345 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1346 * dirty the inode itself
1348 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
1349 trace_writeback_dirty_inode_start(inode
, flags
);
1351 if (sb
->s_op
->dirty_inode
)
1352 sb
->s_op
->dirty_inode(inode
, flags
);
1354 trace_writeback_dirty_inode(inode
, flags
);
1356 if (flags
& I_DIRTY_INODE
)
1357 flags
&= ~I_DIRTY_TIME
;
1358 dirtytime
= flags
& I_DIRTY_TIME
;
1361 * Paired with smp_mb() in __writeback_single_inode() for the
1362 * following lockless i_state test. See there for details.
1366 if (((inode
->i_state
& flags
) == flags
) ||
1367 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
1370 if (unlikely(block_dump
))
1371 block_dump___mark_inode_dirty(inode
);
1373 spin_lock(&inode
->i_lock
);
1374 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
1375 goto out_unlock_inode
;
1376 if ((inode
->i_state
& flags
) != flags
) {
1377 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1379 inode_attach_wb(inode
, NULL
);
1381 if (flags
& I_DIRTY_INODE
)
1382 inode
->i_state
&= ~I_DIRTY_TIME
;
1383 inode
->i_state
|= flags
;
1386 * If the inode is being synced, just update its dirty state.
1387 * The unlocker will place the inode on the appropriate
1388 * superblock list, based upon its state.
1390 if (inode
->i_state
& I_SYNC
)
1391 goto out_unlock_inode
;
1394 * Only add valid (hashed) inodes to the superblock's
1395 * dirty list. Add blockdev inodes as well.
1397 if (!S_ISBLK(inode
->i_mode
)) {
1398 if (inode_unhashed(inode
))
1399 goto out_unlock_inode
;
1401 if (inode
->i_state
& I_FREEING
)
1402 goto out_unlock_inode
;
1405 * If the inode was already on b_dirty/b_io/b_more_io, don't
1406 * reposition it (that would break b_dirty time-ordering).
1409 struct list_head
*dirty_list
;
1410 bool wakeup_bdi
= false;
1411 bdi
= inode_to_bdi(inode
);
1413 spin_unlock(&inode
->i_lock
);
1414 spin_lock(&bdi
->wb
.list_lock
);
1416 WARN(bdi_cap_writeback_dirty(bdi
) &&
1417 !test_bit(WB_registered
, &bdi
->wb
.state
),
1418 "bdi-%s not registered\n", bdi
->name
);
1420 inode
->dirtied_when
= jiffies
;
1422 inode
->dirtied_time_when
= jiffies
;
1424 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
1425 dirty_list
= &bdi
->wb
.b_dirty
;
1427 dirty_list
= &bdi
->wb
.b_dirty_time
;
1429 wakeup_bdi
= inode_wb_list_move_locked(inode
, &bdi
->wb
,
1432 spin_unlock(&bdi
->wb
.list_lock
);
1433 trace_writeback_dirty_inode_enqueue(inode
);
1436 * If this is the first dirty inode for this bdi,
1437 * we have to wake-up the corresponding bdi thread
1438 * to make sure background write-back happens
1441 if (bdi_cap_writeback_dirty(bdi
) && wakeup_bdi
)
1442 wb_wakeup_delayed(&bdi
->wb
);
1447 spin_unlock(&inode
->i_lock
);
1450 EXPORT_SYMBOL(__mark_inode_dirty
);
1452 static void wait_sb_inodes(struct super_block
*sb
)
1454 struct inode
*inode
, *old_inode
= NULL
;
1457 * We need to be protected against the filesystem going from
1458 * r/o to r/w or vice versa.
1460 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1462 spin_lock(&inode_sb_list_lock
);
1465 * Data integrity sync. Must wait for all pages under writeback,
1466 * because there may have been pages dirtied before our sync
1467 * call, but which had writeout started before we write it out.
1468 * In which case, the inode may not be on the dirty list, but
1469 * we still have to wait for that writeout.
1471 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1472 struct address_space
*mapping
= inode
->i_mapping
;
1474 spin_lock(&inode
->i_lock
);
1475 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1476 (mapping
->nrpages
== 0)) {
1477 spin_unlock(&inode
->i_lock
);
1481 spin_unlock(&inode
->i_lock
);
1482 spin_unlock(&inode_sb_list_lock
);
1485 * We hold a reference to 'inode' so it couldn't have been
1486 * removed from s_inodes list while we dropped the
1487 * inode_sb_list_lock. We cannot iput the inode now as we can
1488 * be holding the last reference and we cannot iput it under
1489 * inode_sb_list_lock. So we keep the reference and iput it
1495 filemap_fdatawait(mapping
);
1499 spin_lock(&inode_sb_list_lock
);
1501 spin_unlock(&inode_sb_list_lock
);
1506 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1507 * @sb: the superblock
1508 * @nr: the number of pages to write
1509 * @reason: reason why some writeback work initiated
1511 * Start writeback on some inodes on this super_block. No guarantees are made
1512 * on how many (if any) will be written, and this function does not wait
1513 * for IO completion of submitted IO.
1515 void writeback_inodes_sb_nr(struct super_block
*sb
,
1517 enum wb_reason reason
)
1519 DECLARE_COMPLETION_ONSTACK(done
);
1520 struct wb_writeback_work work
= {
1522 .sync_mode
= WB_SYNC_NONE
,
1523 .tagged_writepages
= 1,
1528 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1530 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1532 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1533 wb_queue_work(&bdi
->wb
, &work
);
1534 wait_for_completion(&done
);
1536 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1539 * writeback_inodes_sb - writeback dirty inodes from given super_block
1540 * @sb: the superblock
1541 * @reason: reason why some writeback work was initiated
1543 * Start writeback on some inodes on this super_block. No guarantees are made
1544 * on how many (if any) will be written, and this function does not wait
1545 * for IO completion of submitted IO.
1547 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1549 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1551 EXPORT_SYMBOL(writeback_inodes_sb
);
1554 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1555 * @sb: the superblock
1556 * @nr: the number of pages to write
1557 * @reason: the reason of writeback
1559 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1560 * Returns 1 if writeback was started, 0 if not.
1562 int try_to_writeback_inodes_sb_nr(struct super_block
*sb
,
1564 enum wb_reason reason
)
1566 if (writeback_in_progress(&sb
->s_bdi
->wb
))
1569 if (!down_read_trylock(&sb
->s_umount
))
1572 writeback_inodes_sb_nr(sb
, nr
, reason
);
1573 up_read(&sb
->s_umount
);
1576 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1579 * try_to_writeback_inodes_sb - try to start writeback if none underway
1580 * @sb: the superblock
1581 * @reason: reason why some writeback work was initiated
1583 * Implement by try_to_writeback_inodes_sb_nr()
1584 * Returns 1 if writeback was started, 0 if not.
1586 int try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1588 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1590 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1593 * sync_inodes_sb - sync sb inode pages
1594 * @sb: the superblock
1596 * This function writes and waits on any dirty inode belonging to this
1599 void sync_inodes_sb(struct super_block
*sb
)
1601 DECLARE_COMPLETION_ONSTACK(done
);
1602 struct wb_writeback_work work
= {
1604 .sync_mode
= WB_SYNC_ALL
,
1605 .nr_pages
= LONG_MAX
,
1608 .reason
= WB_REASON_SYNC
,
1611 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1613 /* Nothing to do? */
1614 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1616 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1618 wb_queue_work(&bdi
->wb
, &work
);
1619 wait_for_completion(&done
);
1623 EXPORT_SYMBOL(sync_inodes_sb
);
1626 * write_inode_now - write an inode to disk
1627 * @inode: inode to write to disk
1628 * @sync: whether the write should be synchronous or not
1630 * This function commits an inode to disk immediately if it is dirty. This is
1631 * primarily needed by knfsd.
1633 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1635 int write_inode_now(struct inode
*inode
, int sync
)
1637 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1638 struct writeback_control wbc
= {
1639 .nr_to_write
= LONG_MAX
,
1640 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1642 .range_end
= LLONG_MAX
,
1645 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1646 wbc
.nr_to_write
= 0;
1649 return writeback_single_inode(inode
, wb
, &wbc
);
1651 EXPORT_SYMBOL(write_inode_now
);
1654 * sync_inode - write an inode and its pages to disk.
1655 * @inode: the inode to sync
1656 * @wbc: controls the writeback mode
1658 * sync_inode() will write an inode and its pages to disk. It will also
1659 * correctly update the inode on its superblock's dirty inode lists and will
1660 * update inode->i_state.
1662 * The caller must have a ref on the inode.
1664 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1666 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1668 EXPORT_SYMBOL(sync_inode
);
1671 * sync_inode_metadata - write an inode to disk
1672 * @inode: the inode to sync
1673 * @wait: wait for I/O to complete.
1675 * Write an inode to disk and adjust its dirty state after completion.
1677 * Note: only writes the actual inode, no associated data or other metadata.
1679 int sync_inode_metadata(struct inode
*inode
, int wait
)
1681 struct writeback_control wbc
= {
1682 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1683 .nr_to_write
= 0, /* metadata-only */
1686 return sync_inode(inode
, &wbc
);
1688 EXPORT_SYMBOL(sync_inode_metadata
);