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
);
201 #endif /* CONFIG_CGROUP_WRITEBACK */
203 void wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
204 bool range_cyclic
, enum wb_reason reason
)
206 struct wb_writeback_work
*work
;
208 if (!wb_has_dirty_io(wb
))
212 * This is WB_SYNC_NONE writeback, so if allocation fails just
213 * wakeup the thread for old dirty data writeback
215 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
217 trace_writeback_nowork(wb
->bdi
);
222 work
->sync_mode
= WB_SYNC_NONE
;
223 work
->nr_pages
= nr_pages
;
224 work
->range_cyclic
= range_cyclic
;
225 work
->reason
= reason
;
227 wb_queue_work(wb
, work
);
231 * bdi_start_background_writeback - start background writeback
232 * @bdi: the backing device to write from
235 * This makes sure WB_SYNC_NONE background writeback happens. When
236 * this function returns, it is only guaranteed that for given BDI
237 * some IO is happening if we are over background dirty threshold.
238 * Caller need not hold sb s_umount semaphore.
240 void bdi_start_background_writeback(struct backing_dev_info
*bdi
)
243 * We just wake up the flusher thread. It will perform background
244 * writeback as soon as there is no other work to do.
246 trace_writeback_wake_background(bdi
);
251 * Remove the inode from the writeback list it is on.
253 void inode_wb_list_del(struct inode
*inode
)
255 struct bdi_writeback
*wb
= inode_to_wb(inode
);
257 spin_lock(&wb
->list_lock
);
258 inode_wb_list_del_locked(inode
, wb
);
259 spin_unlock(&wb
->list_lock
);
263 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
264 * furthest end of its superblock's dirty-inode list.
266 * Before stamping the inode's ->dirtied_when, we check to see whether it is
267 * already the most-recently-dirtied inode on the b_dirty list. If that is
268 * the case then the inode must have been redirtied while it was being written
269 * out and we don't reset its dirtied_when.
271 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
273 if (!list_empty(&wb
->b_dirty
)) {
276 tail
= wb_inode(wb
->b_dirty
.next
);
277 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
278 inode
->dirtied_when
= jiffies
;
280 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty
);
284 * requeue inode for re-scanning after bdi->b_io list is exhausted.
286 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
288 inode_wb_list_move_locked(inode
, wb
, &wb
->b_more_io
);
291 static void inode_sync_complete(struct inode
*inode
)
293 inode
->i_state
&= ~I_SYNC
;
294 /* If inode is clean an unused, put it into LRU now... */
295 inode_add_lru(inode
);
296 /* Waiters must see I_SYNC cleared before being woken up */
298 wake_up_bit(&inode
->i_state
, __I_SYNC
);
301 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
303 bool ret
= time_after(inode
->dirtied_when
, t
);
306 * For inodes being constantly redirtied, dirtied_when can get stuck.
307 * It _appears_ to be in the future, but is actually in distant past.
308 * This test is necessary to prevent such wrapped-around relative times
309 * from permanently stopping the whole bdi writeback.
311 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
316 #define EXPIRE_DIRTY_ATIME 0x0001
319 * Move expired (dirtied before work->older_than_this) dirty inodes from
320 * @delaying_queue to @dispatch_queue.
322 static int move_expired_inodes(struct list_head
*delaying_queue
,
323 struct list_head
*dispatch_queue
,
325 struct wb_writeback_work
*work
)
327 unsigned long *older_than_this
= NULL
;
328 unsigned long expire_time
;
330 struct list_head
*pos
, *node
;
331 struct super_block
*sb
= NULL
;
336 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
337 older_than_this
= work
->older_than_this
;
338 else if (!work
->for_sync
) {
339 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
340 older_than_this
= &expire_time
;
342 while (!list_empty(delaying_queue
)) {
343 inode
= wb_inode(delaying_queue
->prev
);
344 if (older_than_this
&&
345 inode_dirtied_after(inode
, *older_than_this
))
347 list_move(&inode
->i_wb_list
, &tmp
);
349 if (flags
& EXPIRE_DIRTY_ATIME
)
350 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
351 if (sb_is_blkdev_sb(inode
->i_sb
))
353 if (sb
&& sb
!= inode
->i_sb
)
358 /* just one sb in list, splice to dispatch_queue and we're done */
360 list_splice(&tmp
, dispatch_queue
);
364 /* Move inodes from one superblock together */
365 while (!list_empty(&tmp
)) {
366 sb
= wb_inode(tmp
.prev
)->i_sb
;
367 list_for_each_prev_safe(pos
, node
, &tmp
) {
368 inode
= wb_inode(pos
);
369 if (inode
->i_sb
== sb
)
370 list_move(&inode
->i_wb_list
, dispatch_queue
);
378 * Queue all expired dirty inodes for io, eldest first.
380 * newly dirtied b_dirty b_io b_more_io
381 * =============> gf edc BA
383 * newly dirtied b_dirty b_io b_more_io
384 * =============> g fBAedc
386 * +--> dequeue for IO
388 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
392 assert_spin_locked(&wb
->list_lock
);
393 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
394 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
395 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
396 EXPIRE_DIRTY_ATIME
, work
);
398 wb_io_lists_populated(wb
);
399 trace_writeback_queue_io(wb
, work
, moved
);
402 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
406 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
407 trace_writeback_write_inode_start(inode
, wbc
);
408 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
409 trace_writeback_write_inode(inode
, wbc
);
416 * Wait for writeback on an inode to complete. Called with i_lock held.
417 * Caller must make sure inode cannot go away when we drop i_lock.
419 static void __inode_wait_for_writeback(struct inode
*inode
)
420 __releases(inode
->i_lock
)
421 __acquires(inode
->i_lock
)
423 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
424 wait_queue_head_t
*wqh
;
426 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
427 while (inode
->i_state
& I_SYNC
) {
428 spin_unlock(&inode
->i_lock
);
429 __wait_on_bit(wqh
, &wq
, bit_wait
,
430 TASK_UNINTERRUPTIBLE
);
431 spin_lock(&inode
->i_lock
);
436 * Wait for writeback on an inode to complete. Caller must have inode pinned.
438 void inode_wait_for_writeback(struct inode
*inode
)
440 spin_lock(&inode
->i_lock
);
441 __inode_wait_for_writeback(inode
);
442 spin_unlock(&inode
->i_lock
);
446 * Sleep until I_SYNC is cleared. This function must be called with i_lock
447 * held and drops it. It is aimed for callers not holding any inode reference
448 * so once i_lock is dropped, inode can go away.
450 static void inode_sleep_on_writeback(struct inode
*inode
)
451 __releases(inode
->i_lock
)
454 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
457 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
458 sleep
= inode
->i_state
& I_SYNC
;
459 spin_unlock(&inode
->i_lock
);
462 finish_wait(wqh
, &wait
);
466 * Find proper writeback list for the inode depending on its current state and
467 * possibly also change of its state while we were doing writeback. Here we
468 * handle things such as livelock prevention or fairness of writeback among
469 * inodes. This function can be called only by flusher thread - noone else
470 * processes all inodes in writeback lists and requeueing inodes behind flusher
471 * thread's back can have unexpected consequences.
473 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
474 struct writeback_control
*wbc
)
476 if (inode
->i_state
& I_FREEING
)
480 * Sync livelock prevention. Each inode is tagged and synced in one
481 * shot. If still dirty, it will be redirty_tail()'ed below. Update
482 * the dirty time to prevent enqueue and sync it again.
484 if ((inode
->i_state
& I_DIRTY
) &&
485 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
486 inode
->dirtied_when
= jiffies
;
488 if (wbc
->pages_skipped
) {
490 * writeback is not making progress due to locked
491 * buffers. Skip this inode for now.
493 redirty_tail(inode
, wb
);
497 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
499 * We didn't write back all the pages. nfs_writepages()
500 * sometimes bales out without doing anything.
502 if (wbc
->nr_to_write
<= 0) {
503 /* Slice used up. Queue for next turn. */
504 requeue_io(inode
, wb
);
507 * Writeback blocked by something other than
508 * congestion. Delay the inode for some time to
509 * avoid spinning on the CPU (100% iowait)
510 * retrying writeback of the dirty page/inode
511 * that cannot be performed immediately.
513 redirty_tail(inode
, wb
);
515 } else if (inode
->i_state
& I_DIRTY
) {
517 * Filesystems can dirty the inode during writeback operations,
518 * such as delayed allocation during submission or metadata
519 * updates after data IO completion.
521 redirty_tail(inode
, wb
);
522 } else if (inode
->i_state
& I_DIRTY_TIME
) {
523 inode
->dirtied_when
= jiffies
;
524 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty_time
);
526 /* The inode is clean. Remove from writeback lists. */
527 inode_wb_list_del_locked(inode
, wb
);
532 * Write out an inode and its dirty pages. Do not update the writeback list
533 * linkage. That is left to the caller. The caller is also responsible for
534 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
537 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
539 struct address_space
*mapping
= inode
->i_mapping
;
540 long nr_to_write
= wbc
->nr_to_write
;
544 WARN_ON(!(inode
->i_state
& I_SYNC
));
546 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
548 ret
= do_writepages(mapping
, wbc
);
551 * Make sure to wait on the data before writing out the metadata.
552 * This is important for filesystems that modify metadata on data
553 * I/O completion. We don't do it for sync(2) writeback because it has a
554 * separate, external IO completion path and ->sync_fs for guaranteeing
555 * inode metadata is written back correctly.
557 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
558 int err
= filemap_fdatawait(mapping
);
564 * Some filesystems may redirty the inode during the writeback
565 * due to delalloc, clear dirty metadata flags right before
568 spin_lock(&inode
->i_lock
);
570 dirty
= inode
->i_state
& I_DIRTY
;
571 if (inode
->i_state
& I_DIRTY_TIME
) {
572 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
573 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
574 unlikely(time_after(jiffies
,
575 (inode
->dirtied_time_when
+
576 dirtytime_expire_interval
* HZ
)))) {
577 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
578 trace_writeback_lazytime(inode
);
581 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
582 inode
->i_state
&= ~dirty
;
585 * Paired with smp_mb() in __mark_inode_dirty(). This allows
586 * __mark_inode_dirty() to test i_state without grabbing i_lock -
587 * either they see the I_DIRTY bits cleared or we see the dirtied
590 * I_DIRTY_PAGES is always cleared together above even if @mapping
591 * still has dirty pages. The flag is reinstated after smp_mb() if
592 * necessary. This guarantees that either __mark_inode_dirty()
593 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
597 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
598 inode
->i_state
|= I_DIRTY_PAGES
;
600 spin_unlock(&inode
->i_lock
);
602 if (dirty
& I_DIRTY_TIME
)
603 mark_inode_dirty_sync(inode
);
604 /* Don't write the inode if only I_DIRTY_PAGES was set */
605 if (dirty
& ~I_DIRTY_PAGES
) {
606 int err
= write_inode(inode
, wbc
);
610 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
615 * Write out an inode's dirty pages. Either the caller has an active reference
616 * on the inode or the inode has I_WILL_FREE set.
618 * This function is designed to be called for writing back one inode which
619 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
620 * and does more profound writeback list handling in writeback_sb_inodes().
623 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
624 struct writeback_control
*wbc
)
628 spin_lock(&inode
->i_lock
);
629 if (!atomic_read(&inode
->i_count
))
630 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
632 WARN_ON(inode
->i_state
& I_WILL_FREE
);
634 if (inode
->i_state
& I_SYNC
) {
635 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
638 * It's a data-integrity sync. We must wait. Since callers hold
639 * inode reference or inode has I_WILL_FREE set, it cannot go
642 __inode_wait_for_writeback(inode
);
644 WARN_ON(inode
->i_state
& I_SYNC
);
646 * Skip inode if it is clean and we have no outstanding writeback in
647 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
648 * function since flusher thread may be doing for example sync in
649 * parallel and if we move the inode, it could get skipped. So here we
650 * make sure inode is on some writeback list and leave it there unless
651 * we have completely cleaned the inode.
653 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
654 (wbc
->sync_mode
!= WB_SYNC_ALL
||
655 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
657 inode
->i_state
|= I_SYNC
;
658 spin_unlock(&inode
->i_lock
);
660 ret
= __writeback_single_inode(inode
, wbc
);
662 spin_lock(&wb
->list_lock
);
663 spin_lock(&inode
->i_lock
);
665 * If inode is clean, remove it from writeback lists. Otherwise don't
666 * touch it. See comment above for explanation.
668 if (!(inode
->i_state
& I_DIRTY_ALL
))
669 inode_wb_list_del_locked(inode
, wb
);
670 spin_unlock(&wb
->list_lock
);
671 inode_sync_complete(inode
);
673 spin_unlock(&inode
->i_lock
);
677 static long writeback_chunk_size(struct bdi_writeback
*wb
,
678 struct wb_writeback_work
*work
)
683 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
684 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
685 * here avoids calling into writeback_inodes_wb() more than once.
687 * The intended call sequence for WB_SYNC_ALL writeback is:
690 * writeback_sb_inodes() <== called only once
691 * write_cache_pages() <== called once for each inode
692 * (quickly) tag currently dirty pages
693 * (maybe slowly) sync all tagged pages
695 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
698 pages
= min(wb
->avg_write_bandwidth
/ 2,
699 global_dirty_limit
/ DIRTY_SCOPE
);
700 pages
= min(pages
, work
->nr_pages
);
701 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
702 MIN_WRITEBACK_PAGES
);
709 * Write a portion of b_io inodes which belong to @sb.
711 * Return the number of pages and/or inodes written.
713 static long writeback_sb_inodes(struct super_block
*sb
,
714 struct bdi_writeback
*wb
,
715 struct wb_writeback_work
*work
)
717 struct writeback_control wbc
= {
718 .sync_mode
= work
->sync_mode
,
719 .tagged_writepages
= work
->tagged_writepages
,
720 .for_kupdate
= work
->for_kupdate
,
721 .for_background
= work
->for_background
,
722 .for_sync
= work
->for_sync
,
723 .range_cyclic
= work
->range_cyclic
,
725 .range_end
= LLONG_MAX
,
727 unsigned long start_time
= jiffies
;
729 long wrote
= 0; /* count both pages and inodes */
731 while (!list_empty(&wb
->b_io
)) {
732 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
734 if (inode
->i_sb
!= sb
) {
737 * We only want to write back data for this
738 * superblock, move all inodes not belonging
739 * to it back onto the dirty list.
741 redirty_tail(inode
, wb
);
746 * The inode belongs to a different superblock.
747 * Bounce back to the caller to unpin this and
748 * pin the next superblock.
754 * Don't bother with new inodes or inodes being freed, first
755 * kind does not need periodic writeout yet, and for the latter
756 * kind writeout is handled by the freer.
758 spin_lock(&inode
->i_lock
);
759 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
760 spin_unlock(&inode
->i_lock
);
761 redirty_tail(inode
, wb
);
764 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
766 * If this inode is locked for writeback and we are not
767 * doing writeback-for-data-integrity, move it to
768 * b_more_io so that writeback can proceed with the
769 * other inodes on s_io.
771 * We'll have another go at writing back this inode
772 * when we completed a full scan of b_io.
774 spin_unlock(&inode
->i_lock
);
775 requeue_io(inode
, wb
);
776 trace_writeback_sb_inodes_requeue(inode
);
779 spin_unlock(&wb
->list_lock
);
782 * We already requeued the inode if it had I_SYNC set and we
783 * are doing WB_SYNC_NONE writeback. So this catches only the
786 if (inode
->i_state
& I_SYNC
) {
787 /* Wait for I_SYNC. This function drops i_lock... */
788 inode_sleep_on_writeback(inode
);
789 /* Inode may be gone, start again */
790 spin_lock(&wb
->list_lock
);
793 inode
->i_state
|= I_SYNC
;
794 spin_unlock(&inode
->i_lock
);
796 write_chunk
= writeback_chunk_size(wb
, work
);
797 wbc
.nr_to_write
= write_chunk
;
798 wbc
.pages_skipped
= 0;
801 * We use I_SYNC to pin the inode in memory. While it is set
802 * evict_inode() will wait so the inode cannot be freed.
804 __writeback_single_inode(inode
, &wbc
);
806 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
807 wrote
+= write_chunk
- wbc
.nr_to_write
;
808 spin_lock(&wb
->list_lock
);
809 spin_lock(&inode
->i_lock
);
810 if (!(inode
->i_state
& I_DIRTY_ALL
))
812 requeue_inode(inode
, wb
, &wbc
);
813 inode_sync_complete(inode
);
814 spin_unlock(&inode
->i_lock
);
815 cond_resched_lock(&wb
->list_lock
);
817 * bail out to wb_writeback() often enough to check
818 * background threshold and other termination conditions.
821 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
823 if (work
->nr_pages
<= 0)
830 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
831 struct wb_writeback_work
*work
)
833 unsigned long start_time
= jiffies
;
836 while (!list_empty(&wb
->b_io
)) {
837 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
838 struct super_block
*sb
= inode
->i_sb
;
840 if (!trylock_super(sb
)) {
842 * trylock_super() may fail consistently due to
843 * s_umount being grabbed by someone else. Don't use
844 * requeue_io() to avoid busy retrying the inode/sb.
846 redirty_tail(inode
, wb
);
849 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
850 up_read(&sb
->s_umount
);
852 /* refer to the same tests at the end of writeback_sb_inodes */
854 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
856 if (work
->nr_pages
<= 0)
860 /* Leave any unwritten inodes on b_io */
864 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
865 enum wb_reason reason
)
867 struct wb_writeback_work work
= {
868 .nr_pages
= nr_pages
,
869 .sync_mode
= WB_SYNC_NONE
,
874 spin_lock(&wb
->list_lock
);
875 if (list_empty(&wb
->b_io
))
877 __writeback_inodes_wb(wb
, &work
);
878 spin_unlock(&wb
->list_lock
);
880 return nr_pages
- work
.nr_pages
;
883 static bool over_bground_thresh(struct bdi_writeback
*wb
)
885 unsigned long background_thresh
, dirty_thresh
;
887 global_dirty_limits(&background_thresh
, &dirty_thresh
);
889 if (global_page_state(NR_FILE_DIRTY
) +
890 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
)
893 if (wb_stat(wb
, WB_RECLAIMABLE
) > wb_dirty_limit(wb
, background_thresh
))
900 * Called under wb->list_lock. If there are multiple wb per bdi,
901 * only the flusher working on the first wb should do it.
903 static void wb_update_bandwidth(struct bdi_writeback
*wb
,
904 unsigned long start_time
)
906 __wb_update_bandwidth(wb
, 0, 0, 0, 0, 0, start_time
);
910 * Explicit flushing or periodic writeback of "old" data.
912 * Define "old": the first time one of an inode's pages is dirtied, we mark the
913 * dirtying-time in the inode's address_space. So this periodic writeback code
914 * just walks the superblock inode list, writing back any inodes which are
915 * older than a specific point in time.
917 * Try to run once per dirty_writeback_interval. But if a writeback event
918 * takes longer than a dirty_writeback_interval interval, then leave a
921 * older_than_this takes precedence over nr_to_write. So we'll only write back
922 * all dirty pages if they are all attached to "old" mappings.
924 static long wb_writeback(struct bdi_writeback
*wb
,
925 struct wb_writeback_work
*work
)
927 unsigned long wb_start
= jiffies
;
928 long nr_pages
= work
->nr_pages
;
929 unsigned long oldest_jif
;
933 oldest_jif
= jiffies
;
934 work
->older_than_this
= &oldest_jif
;
936 spin_lock(&wb
->list_lock
);
939 * Stop writeback when nr_pages has been consumed
941 if (work
->nr_pages
<= 0)
945 * Background writeout and kupdate-style writeback may
946 * run forever. Stop them if there is other work to do
947 * so that e.g. sync can proceed. They'll be restarted
948 * after the other works are all done.
950 if ((work
->for_background
|| work
->for_kupdate
) &&
951 !list_empty(&wb
->work_list
))
955 * For background writeout, stop when we are below the
956 * background dirty threshold
958 if (work
->for_background
&& !over_bground_thresh(wb
))
962 * Kupdate and background works are special and we want to
963 * include all inodes that need writing. Livelock avoidance is
964 * handled by these works yielding to any other work so we are
967 if (work
->for_kupdate
) {
968 oldest_jif
= jiffies
-
969 msecs_to_jiffies(dirty_expire_interval
* 10);
970 } else if (work
->for_background
)
971 oldest_jif
= jiffies
;
973 trace_writeback_start(wb
->bdi
, work
);
974 if (list_empty(&wb
->b_io
))
977 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
979 progress
= __writeback_inodes_wb(wb
, work
);
980 trace_writeback_written(wb
->bdi
, work
);
982 wb_update_bandwidth(wb
, wb_start
);
985 * Did we write something? Try for more
987 * Dirty inodes are moved to b_io for writeback in batches.
988 * The completion of the current batch does not necessarily
989 * mean the overall work is done. So we keep looping as long
990 * as made some progress on cleaning pages or inodes.
995 * No more inodes for IO, bail
997 if (list_empty(&wb
->b_more_io
))
1000 * Nothing written. Wait for some inode to
1001 * become available for writeback. Otherwise
1002 * we'll just busyloop.
1004 if (!list_empty(&wb
->b_more_io
)) {
1005 trace_writeback_wait(wb
->bdi
, work
);
1006 inode
= wb_inode(wb
->b_more_io
.prev
);
1007 spin_lock(&inode
->i_lock
);
1008 spin_unlock(&wb
->list_lock
);
1009 /* This function drops i_lock... */
1010 inode_sleep_on_writeback(inode
);
1011 spin_lock(&wb
->list_lock
);
1014 spin_unlock(&wb
->list_lock
);
1016 return nr_pages
- work
->nr_pages
;
1020 * Return the next wb_writeback_work struct that hasn't been processed yet.
1022 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
1024 struct wb_writeback_work
*work
= NULL
;
1026 spin_lock_bh(&wb
->work_lock
);
1027 if (!list_empty(&wb
->work_list
)) {
1028 work
= list_entry(wb
->work_list
.next
,
1029 struct wb_writeback_work
, list
);
1030 list_del_init(&work
->list
);
1032 spin_unlock_bh(&wb
->work_lock
);
1037 * Add in the number of potentially dirty inodes, because each inode
1038 * write can dirty pagecache in the underlying blockdev.
1040 static unsigned long get_nr_dirty_pages(void)
1042 return global_page_state(NR_FILE_DIRTY
) +
1043 global_page_state(NR_UNSTABLE_NFS
) +
1044 get_nr_dirty_inodes();
1047 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1049 if (over_bground_thresh(wb
)) {
1051 struct wb_writeback_work work
= {
1052 .nr_pages
= LONG_MAX
,
1053 .sync_mode
= WB_SYNC_NONE
,
1054 .for_background
= 1,
1056 .reason
= WB_REASON_BACKGROUND
,
1059 return wb_writeback(wb
, &work
);
1065 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1067 unsigned long expired
;
1071 * When set to zero, disable periodic writeback
1073 if (!dirty_writeback_interval
)
1076 expired
= wb
->last_old_flush
+
1077 msecs_to_jiffies(dirty_writeback_interval
* 10);
1078 if (time_before(jiffies
, expired
))
1081 wb
->last_old_flush
= jiffies
;
1082 nr_pages
= get_nr_dirty_pages();
1085 struct wb_writeback_work work
= {
1086 .nr_pages
= nr_pages
,
1087 .sync_mode
= WB_SYNC_NONE
,
1090 .reason
= WB_REASON_PERIODIC
,
1093 return wb_writeback(wb
, &work
);
1100 * Retrieve work items and do the writeback they describe
1102 static long wb_do_writeback(struct bdi_writeback
*wb
)
1104 struct wb_writeback_work
*work
;
1107 set_bit(WB_writeback_running
, &wb
->state
);
1108 while ((work
= get_next_work_item(wb
)) != NULL
) {
1110 trace_writeback_exec(wb
->bdi
, work
);
1112 wrote
+= wb_writeback(wb
, work
);
1115 * Notify the caller of completion if this is a synchronous
1116 * work item, otherwise just free it.
1119 complete(work
->done
);
1125 * Check for periodic writeback, kupdated() style
1127 wrote
+= wb_check_old_data_flush(wb
);
1128 wrote
+= wb_check_background_flush(wb
);
1129 clear_bit(WB_writeback_running
, &wb
->state
);
1135 * Handle writeback of dirty data for the device backed by this bdi. Also
1136 * reschedules periodically and does kupdated style flushing.
1138 void wb_workfn(struct work_struct
*work
)
1140 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1141 struct bdi_writeback
, dwork
);
1144 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1145 current
->flags
|= PF_SWAPWRITE
;
1147 if (likely(!current_is_workqueue_rescuer() ||
1148 !test_bit(WB_registered
, &wb
->state
))) {
1150 * The normal path. Keep writing back @wb until its
1151 * work_list is empty. Note that this path is also taken
1152 * if @wb is shutting down even when we're running off the
1153 * rescuer as work_list needs to be drained.
1156 pages_written
= wb_do_writeback(wb
);
1157 trace_writeback_pages_written(pages_written
);
1158 } while (!list_empty(&wb
->work_list
));
1161 * bdi_wq can't get enough workers and we're running off
1162 * the emergency worker. Don't hog it. Hopefully, 1024 is
1163 * enough for efficient IO.
1165 pages_written
= writeback_inodes_wb(wb
, 1024,
1166 WB_REASON_FORKER_THREAD
);
1167 trace_writeback_pages_written(pages_written
);
1170 if (!list_empty(&wb
->work_list
))
1171 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1172 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1173 wb_wakeup_delayed(wb
);
1175 current
->flags
&= ~PF_SWAPWRITE
;
1179 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1182 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1184 struct backing_dev_info
*bdi
;
1187 nr_pages
= get_nr_dirty_pages();
1190 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
)
1191 wb_start_writeback(&bdi
->wb
, nr_pages
, false, reason
);
1196 * Wake up bdi's periodically to make sure dirtytime inodes gets
1197 * written back periodically. We deliberately do *not* check the
1198 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1199 * kernel to be constantly waking up once there are any dirtytime
1200 * inodes on the system. So instead we define a separate delayed work
1201 * function which gets called much more rarely. (By default, only
1202 * once every 12 hours.)
1204 * If there is any other write activity going on in the file system,
1205 * this function won't be necessary. But if the only thing that has
1206 * happened on the file system is a dirtytime inode caused by an atime
1207 * update, we need this infrastructure below to make sure that inode
1208 * eventually gets pushed out to disk.
1210 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1211 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1213 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1215 struct backing_dev_info
*bdi
;
1218 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1219 if (list_empty(&bdi
->wb
.b_dirty_time
))
1221 wb_wakeup(&bdi
->wb
);
1224 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1227 static int __init
start_dirtytime_writeback(void)
1229 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1232 __initcall(start_dirtytime_writeback
);
1234 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1235 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1239 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1240 if (ret
== 0 && write
)
1241 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1245 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1247 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1248 struct dentry
*dentry
;
1249 const char *name
= "?";
1251 dentry
= d_find_alias(inode
);
1253 spin_lock(&dentry
->d_lock
);
1254 name
= (const char *) dentry
->d_name
.name
;
1257 "%s(%d): dirtied inode %lu (%s) on %s\n",
1258 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1259 name
, inode
->i_sb
->s_id
);
1261 spin_unlock(&dentry
->d_lock
);
1268 * __mark_inode_dirty - internal function
1269 * @inode: inode to mark
1270 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1271 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1272 * mark_inode_dirty_sync.
1274 * Put the inode on the super block's dirty list.
1276 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1277 * dirty list only if it is hashed or if it refers to a blockdev.
1278 * If it was not hashed, it will never be added to the dirty list
1279 * even if it is later hashed, as it will have been marked dirty already.
1281 * In short, make sure you hash any inodes _before_ you start marking
1284 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1285 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1286 * the kernel-internal blockdev inode represents the dirtying time of the
1287 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1288 * page->mapping->host, so the page-dirtying time is recorded in the internal
1291 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1292 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1294 struct super_block
*sb
= inode
->i_sb
;
1295 struct backing_dev_info
*bdi
= NULL
;
1298 trace_writeback_mark_inode_dirty(inode
, flags
);
1301 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1302 * dirty the inode itself
1304 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
1305 trace_writeback_dirty_inode_start(inode
, flags
);
1307 if (sb
->s_op
->dirty_inode
)
1308 sb
->s_op
->dirty_inode(inode
, flags
);
1310 trace_writeback_dirty_inode(inode
, flags
);
1312 if (flags
& I_DIRTY_INODE
)
1313 flags
&= ~I_DIRTY_TIME
;
1314 dirtytime
= flags
& I_DIRTY_TIME
;
1317 * Paired with smp_mb() in __writeback_single_inode() for the
1318 * following lockless i_state test. See there for details.
1322 if (((inode
->i_state
& flags
) == flags
) ||
1323 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
1326 if (unlikely(block_dump
))
1327 block_dump___mark_inode_dirty(inode
);
1329 spin_lock(&inode
->i_lock
);
1330 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
1331 goto out_unlock_inode
;
1332 if ((inode
->i_state
& flags
) != flags
) {
1333 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1335 inode_attach_wb(inode
, NULL
);
1337 if (flags
& I_DIRTY_INODE
)
1338 inode
->i_state
&= ~I_DIRTY_TIME
;
1339 inode
->i_state
|= flags
;
1342 * If the inode is being synced, just update its dirty state.
1343 * The unlocker will place the inode on the appropriate
1344 * superblock list, based upon its state.
1346 if (inode
->i_state
& I_SYNC
)
1347 goto out_unlock_inode
;
1350 * Only add valid (hashed) inodes to the superblock's
1351 * dirty list. Add blockdev inodes as well.
1353 if (!S_ISBLK(inode
->i_mode
)) {
1354 if (inode_unhashed(inode
))
1355 goto out_unlock_inode
;
1357 if (inode
->i_state
& I_FREEING
)
1358 goto out_unlock_inode
;
1361 * If the inode was already on b_dirty/b_io/b_more_io, don't
1362 * reposition it (that would break b_dirty time-ordering).
1365 struct list_head
*dirty_list
;
1366 bool wakeup_bdi
= false;
1367 bdi
= inode_to_bdi(inode
);
1369 spin_unlock(&inode
->i_lock
);
1370 spin_lock(&bdi
->wb
.list_lock
);
1372 WARN(bdi_cap_writeback_dirty(bdi
) &&
1373 !test_bit(WB_registered
, &bdi
->wb
.state
),
1374 "bdi-%s not registered\n", bdi
->name
);
1376 inode
->dirtied_when
= jiffies
;
1378 inode
->dirtied_time_when
= jiffies
;
1380 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
1381 dirty_list
= &bdi
->wb
.b_dirty
;
1383 dirty_list
= &bdi
->wb
.b_dirty_time
;
1385 wakeup_bdi
= inode_wb_list_move_locked(inode
, &bdi
->wb
,
1388 spin_unlock(&bdi
->wb
.list_lock
);
1389 trace_writeback_dirty_inode_enqueue(inode
);
1392 * If this is the first dirty inode for this bdi,
1393 * we have to wake-up the corresponding bdi thread
1394 * to make sure background write-back happens
1397 if (bdi_cap_writeback_dirty(bdi
) && wakeup_bdi
)
1398 wb_wakeup_delayed(&bdi
->wb
);
1403 spin_unlock(&inode
->i_lock
);
1406 EXPORT_SYMBOL(__mark_inode_dirty
);
1408 static void wait_sb_inodes(struct super_block
*sb
)
1410 struct inode
*inode
, *old_inode
= NULL
;
1413 * We need to be protected against the filesystem going from
1414 * r/o to r/w or vice versa.
1416 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1418 spin_lock(&inode_sb_list_lock
);
1421 * Data integrity sync. Must wait for all pages under writeback,
1422 * because there may have been pages dirtied before our sync
1423 * call, but which had writeout started before we write it out.
1424 * In which case, the inode may not be on the dirty list, but
1425 * we still have to wait for that writeout.
1427 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1428 struct address_space
*mapping
= inode
->i_mapping
;
1430 spin_lock(&inode
->i_lock
);
1431 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1432 (mapping
->nrpages
== 0)) {
1433 spin_unlock(&inode
->i_lock
);
1437 spin_unlock(&inode
->i_lock
);
1438 spin_unlock(&inode_sb_list_lock
);
1441 * We hold a reference to 'inode' so it couldn't have been
1442 * removed from s_inodes list while we dropped the
1443 * inode_sb_list_lock. We cannot iput the inode now as we can
1444 * be holding the last reference and we cannot iput it under
1445 * inode_sb_list_lock. So we keep the reference and iput it
1451 filemap_fdatawait(mapping
);
1455 spin_lock(&inode_sb_list_lock
);
1457 spin_unlock(&inode_sb_list_lock
);
1462 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1463 * @sb: the superblock
1464 * @nr: the number of pages to write
1465 * @reason: reason why some writeback work initiated
1467 * Start writeback on some inodes on this super_block. No guarantees are made
1468 * on how many (if any) will be written, and this function does not wait
1469 * for IO completion of submitted IO.
1471 void writeback_inodes_sb_nr(struct super_block
*sb
,
1473 enum wb_reason reason
)
1475 DECLARE_COMPLETION_ONSTACK(done
);
1476 struct wb_writeback_work work
= {
1478 .sync_mode
= WB_SYNC_NONE
,
1479 .tagged_writepages
= 1,
1484 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1486 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1488 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1489 wb_queue_work(&bdi
->wb
, &work
);
1490 wait_for_completion(&done
);
1492 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1495 * writeback_inodes_sb - writeback dirty inodes from given super_block
1496 * @sb: the superblock
1497 * @reason: reason why some writeback work was initiated
1499 * Start writeback on some inodes on this super_block. No guarantees are made
1500 * on how many (if any) will be written, and this function does not wait
1501 * for IO completion of submitted IO.
1503 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1505 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1507 EXPORT_SYMBOL(writeback_inodes_sb
);
1510 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1511 * @sb: the superblock
1512 * @nr: the number of pages to write
1513 * @reason: the reason of writeback
1515 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1516 * Returns 1 if writeback was started, 0 if not.
1518 int try_to_writeback_inodes_sb_nr(struct super_block
*sb
,
1520 enum wb_reason reason
)
1522 if (writeback_in_progress(&sb
->s_bdi
->wb
))
1525 if (!down_read_trylock(&sb
->s_umount
))
1528 writeback_inodes_sb_nr(sb
, nr
, reason
);
1529 up_read(&sb
->s_umount
);
1532 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1535 * try_to_writeback_inodes_sb - try to start writeback if none underway
1536 * @sb: the superblock
1537 * @reason: reason why some writeback work was initiated
1539 * Implement by try_to_writeback_inodes_sb_nr()
1540 * Returns 1 if writeback was started, 0 if not.
1542 int try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1544 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1546 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1549 * sync_inodes_sb - sync sb inode pages
1550 * @sb: the superblock
1552 * This function writes and waits on any dirty inode belonging to this
1555 void sync_inodes_sb(struct super_block
*sb
)
1557 DECLARE_COMPLETION_ONSTACK(done
);
1558 struct wb_writeback_work work
= {
1560 .sync_mode
= WB_SYNC_ALL
,
1561 .nr_pages
= LONG_MAX
,
1564 .reason
= WB_REASON_SYNC
,
1567 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1569 /* Nothing to do? */
1570 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1572 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1574 wb_queue_work(&bdi
->wb
, &work
);
1575 wait_for_completion(&done
);
1579 EXPORT_SYMBOL(sync_inodes_sb
);
1582 * write_inode_now - write an inode to disk
1583 * @inode: inode to write to disk
1584 * @sync: whether the write should be synchronous or not
1586 * This function commits an inode to disk immediately if it is dirty. This is
1587 * primarily needed by knfsd.
1589 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1591 int write_inode_now(struct inode
*inode
, int sync
)
1593 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1594 struct writeback_control wbc
= {
1595 .nr_to_write
= LONG_MAX
,
1596 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1598 .range_end
= LLONG_MAX
,
1601 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1602 wbc
.nr_to_write
= 0;
1605 return writeback_single_inode(inode
, wb
, &wbc
);
1607 EXPORT_SYMBOL(write_inode_now
);
1610 * sync_inode - write an inode and its pages to disk.
1611 * @inode: the inode to sync
1612 * @wbc: controls the writeback mode
1614 * sync_inode() will write an inode and its pages to disk. It will also
1615 * correctly update the inode on its superblock's dirty inode lists and will
1616 * update inode->i_state.
1618 * The caller must have a ref on the inode.
1620 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1622 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1624 EXPORT_SYMBOL(sync_inode
);
1627 * sync_inode_metadata - write an inode to disk
1628 * @inode: the inode to sync
1629 * @wait: wait for I/O to complete.
1631 * Write an inode to disk and adjust its dirty state after completion.
1633 * Note: only writes the actual inode, no associated data or other metadata.
1635 int sync_inode_metadata(struct inode
*inode
, int wait
)
1637 struct writeback_control wbc
= {
1638 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1639 .nr_to_write
= 0, /* metadata-only */
1642 return sync_inode(inode
, &wbc
);
1644 EXPORT_SYMBOL(sync_inode_metadata
);