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 void wb_wakeup(struct bdi_writeback
*wb
)
98 spin_lock_bh(&wb
->work_lock
);
99 if (test_bit(WB_registered
, &wb
->state
))
100 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
101 spin_unlock_bh(&wb
->work_lock
);
104 static void wb_queue_work(struct bdi_writeback
*wb
,
105 struct wb_writeback_work
*work
)
107 trace_writeback_queue(wb
->bdi
, work
);
109 spin_lock_bh(&wb
->work_lock
);
110 if (!test_bit(WB_registered
, &wb
->state
)) {
112 complete(work
->done
);
115 list_add_tail(&work
->list
, &wb
->work_list
);
116 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
118 spin_unlock_bh(&wb
->work_lock
);
121 static void __wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
122 bool range_cyclic
, enum wb_reason reason
)
124 struct wb_writeback_work
*work
;
127 * This is WB_SYNC_NONE writeback, so if allocation fails just
128 * wakeup the thread for old dirty data writeback
130 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
132 trace_writeback_nowork(wb
->bdi
);
137 work
->sync_mode
= WB_SYNC_NONE
;
138 work
->nr_pages
= nr_pages
;
139 work
->range_cyclic
= range_cyclic
;
140 work
->reason
= reason
;
142 wb_queue_work(wb
, work
);
146 * bdi_start_writeback - start writeback
147 * @bdi: the backing device to write from
148 * @nr_pages: the number of pages to write
149 * @reason: reason why some writeback work was initiated
152 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
153 * started when this function returns, we make no guarantees on
154 * completion. Caller need not hold sb s_umount semaphore.
157 void bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
158 enum wb_reason reason
)
160 __wb_start_writeback(&bdi
->wb
, nr_pages
, true, reason
);
164 * bdi_start_background_writeback - start background writeback
165 * @bdi: the backing device to write from
168 * This makes sure WB_SYNC_NONE background writeback happens. When
169 * this function returns, it is only guaranteed that for given BDI
170 * some IO is happening if we are over background dirty threshold.
171 * Caller need not hold sb s_umount semaphore.
173 void bdi_start_background_writeback(struct backing_dev_info
*bdi
)
176 * We just wake up the flusher thread. It will perform background
177 * writeback as soon as there is no other work to do.
179 trace_writeback_wake_background(bdi
);
184 * Remove the inode from the writeback list it is on.
186 void inode_wb_list_del(struct inode
*inode
)
188 struct bdi_writeback
*wb
= inode_to_wb(inode
);
190 spin_lock(&wb
->list_lock
);
191 list_del_init(&inode
->i_wb_list
);
192 spin_unlock(&wb
->list_lock
);
196 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
197 * furthest end of its superblock's dirty-inode list.
199 * Before stamping the inode's ->dirtied_when, we check to see whether it is
200 * already the most-recently-dirtied inode on the b_dirty list. If that is
201 * the case then the inode must have been redirtied while it was being written
202 * out and we don't reset its dirtied_when.
204 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
206 assert_spin_locked(&wb
->list_lock
);
207 if (!list_empty(&wb
->b_dirty
)) {
210 tail
= wb_inode(wb
->b_dirty
.next
);
211 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
212 inode
->dirtied_when
= jiffies
;
214 list_move(&inode
->i_wb_list
, &wb
->b_dirty
);
218 * requeue inode for re-scanning after bdi->b_io list is exhausted.
220 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
222 assert_spin_locked(&wb
->list_lock
);
223 list_move(&inode
->i_wb_list
, &wb
->b_more_io
);
226 static void inode_sync_complete(struct inode
*inode
)
228 inode
->i_state
&= ~I_SYNC
;
229 /* If inode is clean an unused, put it into LRU now... */
230 inode_add_lru(inode
);
231 /* Waiters must see I_SYNC cleared before being woken up */
233 wake_up_bit(&inode
->i_state
, __I_SYNC
);
236 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
238 bool ret
= time_after(inode
->dirtied_when
, t
);
241 * For inodes being constantly redirtied, dirtied_when can get stuck.
242 * It _appears_ to be in the future, but is actually in distant past.
243 * This test is necessary to prevent such wrapped-around relative times
244 * from permanently stopping the whole bdi writeback.
246 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
251 #define EXPIRE_DIRTY_ATIME 0x0001
254 * Move expired (dirtied before work->older_than_this) dirty inodes from
255 * @delaying_queue to @dispatch_queue.
257 static int move_expired_inodes(struct list_head
*delaying_queue
,
258 struct list_head
*dispatch_queue
,
260 struct wb_writeback_work
*work
)
262 unsigned long *older_than_this
= NULL
;
263 unsigned long expire_time
;
265 struct list_head
*pos
, *node
;
266 struct super_block
*sb
= NULL
;
271 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
272 older_than_this
= work
->older_than_this
;
273 else if (!work
->for_sync
) {
274 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
275 older_than_this
= &expire_time
;
277 while (!list_empty(delaying_queue
)) {
278 inode
= wb_inode(delaying_queue
->prev
);
279 if (older_than_this
&&
280 inode_dirtied_after(inode
, *older_than_this
))
282 list_move(&inode
->i_wb_list
, &tmp
);
284 if (flags
& EXPIRE_DIRTY_ATIME
)
285 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
286 if (sb_is_blkdev_sb(inode
->i_sb
))
288 if (sb
&& sb
!= inode
->i_sb
)
293 /* just one sb in list, splice to dispatch_queue and we're done */
295 list_splice(&tmp
, dispatch_queue
);
299 /* Move inodes from one superblock together */
300 while (!list_empty(&tmp
)) {
301 sb
= wb_inode(tmp
.prev
)->i_sb
;
302 list_for_each_prev_safe(pos
, node
, &tmp
) {
303 inode
= wb_inode(pos
);
304 if (inode
->i_sb
== sb
)
305 list_move(&inode
->i_wb_list
, dispatch_queue
);
313 * Queue all expired dirty inodes for io, eldest first.
315 * newly dirtied b_dirty b_io b_more_io
316 * =============> gf edc BA
318 * newly dirtied b_dirty b_io b_more_io
319 * =============> g fBAedc
321 * +--> dequeue for IO
323 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
327 assert_spin_locked(&wb
->list_lock
);
328 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
329 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
330 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
331 EXPIRE_DIRTY_ATIME
, work
);
332 trace_writeback_queue_io(wb
, work
, moved
);
335 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
339 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
340 trace_writeback_write_inode_start(inode
, wbc
);
341 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
342 trace_writeback_write_inode(inode
, wbc
);
349 * Wait for writeback on an inode to complete. Called with i_lock held.
350 * Caller must make sure inode cannot go away when we drop i_lock.
352 static void __inode_wait_for_writeback(struct inode
*inode
)
353 __releases(inode
->i_lock
)
354 __acquires(inode
->i_lock
)
356 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
357 wait_queue_head_t
*wqh
;
359 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
360 while (inode
->i_state
& I_SYNC
) {
361 spin_unlock(&inode
->i_lock
);
362 __wait_on_bit(wqh
, &wq
, bit_wait
,
363 TASK_UNINTERRUPTIBLE
);
364 spin_lock(&inode
->i_lock
);
369 * Wait for writeback on an inode to complete. Caller must have inode pinned.
371 void inode_wait_for_writeback(struct inode
*inode
)
373 spin_lock(&inode
->i_lock
);
374 __inode_wait_for_writeback(inode
);
375 spin_unlock(&inode
->i_lock
);
379 * Sleep until I_SYNC is cleared. This function must be called with i_lock
380 * held and drops it. It is aimed for callers not holding any inode reference
381 * so once i_lock is dropped, inode can go away.
383 static void inode_sleep_on_writeback(struct inode
*inode
)
384 __releases(inode
->i_lock
)
387 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
390 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
391 sleep
= inode
->i_state
& I_SYNC
;
392 spin_unlock(&inode
->i_lock
);
395 finish_wait(wqh
, &wait
);
399 * Find proper writeback list for the inode depending on its current state and
400 * possibly also change of its state while we were doing writeback. Here we
401 * handle things such as livelock prevention or fairness of writeback among
402 * inodes. This function can be called only by flusher thread - noone else
403 * processes all inodes in writeback lists and requeueing inodes behind flusher
404 * thread's back can have unexpected consequences.
406 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
407 struct writeback_control
*wbc
)
409 if (inode
->i_state
& I_FREEING
)
413 * Sync livelock prevention. Each inode is tagged and synced in one
414 * shot. If still dirty, it will be redirty_tail()'ed below. Update
415 * the dirty time to prevent enqueue and sync it again.
417 if ((inode
->i_state
& I_DIRTY
) &&
418 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
419 inode
->dirtied_when
= jiffies
;
421 if (wbc
->pages_skipped
) {
423 * writeback is not making progress due to locked
424 * buffers. Skip this inode for now.
426 redirty_tail(inode
, wb
);
430 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
432 * We didn't write back all the pages. nfs_writepages()
433 * sometimes bales out without doing anything.
435 if (wbc
->nr_to_write
<= 0) {
436 /* Slice used up. Queue for next turn. */
437 requeue_io(inode
, wb
);
440 * Writeback blocked by something other than
441 * congestion. Delay the inode for some time to
442 * avoid spinning on the CPU (100% iowait)
443 * retrying writeback of the dirty page/inode
444 * that cannot be performed immediately.
446 redirty_tail(inode
, wb
);
448 } else if (inode
->i_state
& I_DIRTY
) {
450 * Filesystems can dirty the inode during writeback operations,
451 * such as delayed allocation during submission or metadata
452 * updates after data IO completion.
454 redirty_tail(inode
, wb
);
455 } else if (inode
->i_state
& I_DIRTY_TIME
) {
456 inode
->dirtied_when
= jiffies
;
457 list_move(&inode
->i_wb_list
, &wb
->b_dirty_time
);
459 /* The inode is clean. Remove from writeback lists. */
460 list_del_init(&inode
->i_wb_list
);
465 * Write out an inode and its dirty pages. Do not update the writeback list
466 * linkage. That is left to the caller. The caller is also responsible for
467 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
470 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
472 struct address_space
*mapping
= inode
->i_mapping
;
473 long nr_to_write
= wbc
->nr_to_write
;
477 WARN_ON(!(inode
->i_state
& I_SYNC
));
479 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
481 ret
= do_writepages(mapping
, wbc
);
484 * Make sure to wait on the data before writing out the metadata.
485 * This is important for filesystems that modify metadata on data
486 * I/O completion. We don't do it for sync(2) writeback because it has a
487 * separate, external IO completion path and ->sync_fs for guaranteeing
488 * inode metadata is written back correctly.
490 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
491 int err
= filemap_fdatawait(mapping
);
497 * Some filesystems may redirty the inode during the writeback
498 * due to delalloc, clear dirty metadata flags right before
501 spin_lock(&inode
->i_lock
);
503 dirty
= inode
->i_state
& I_DIRTY
;
504 if (inode
->i_state
& I_DIRTY_TIME
) {
505 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
506 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
507 unlikely(time_after(jiffies
,
508 (inode
->dirtied_time_when
+
509 dirtytime_expire_interval
* HZ
)))) {
510 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
511 trace_writeback_lazytime(inode
);
514 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
515 inode
->i_state
&= ~dirty
;
518 * Paired with smp_mb() in __mark_inode_dirty(). This allows
519 * __mark_inode_dirty() to test i_state without grabbing i_lock -
520 * either they see the I_DIRTY bits cleared or we see the dirtied
523 * I_DIRTY_PAGES is always cleared together above even if @mapping
524 * still has dirty pages. The flag is reinstated after smp_mb() if
525 * necessary. This guarantees that either __mark_inode_dirty()
526 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
530 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
531 inode
->i_state
|= I_DIRTY_PAGES
;
533 spin_unlock(&inode
->i_lock
);
535 if (dirty
& I_DIRTY_TIME
)
536 mark_inode_dirty_sync(inode
);
537 /* Don't write the inode if only I_DIRTY_PAGES was set */
538 if (dirty
& ~I_DIRTY_PAGES
) {
539 int err
= write_inode(inode
, wbc
);
543 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
548 * Write out an inode's dirty pages. Either the caller has an active reference
549 * on the inode or the inode has I_WILL_FREE set.
551 * This function is designed to be called for writing back one inode which
552 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
553 * and does more profound writeback list handling in writeback_sb_inodes().
556 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
557 struct writeback_control
*wbc
)
561 spin_lock(&inode
->i_lock
);
562 if (!atomic_read(&inode
->i_count
))
563 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
565 WARN_ON(inode
->i_state
& I_WILL_FREE
);
567 if (inode
->i_state
& I_SYNC
) {
568 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
571 * It's a data-integrity sync. We must wait. Since callers hold
572 * inode reference or inode has I_WILL_FREE set, it cannot go
575 __inode_wait_for_writeback(inode
);
577 WARN_ON(inode
->i_state
& I_SYNC
);
579 * Skip inode if it is clean and we have no outstanding writeback in
580 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
581 * function since flusher thread may be doing for example sync in
582 * parallel and if we move the inode, it could get skipped. So here we
583 * make sure inode is on some writeback list and leave it there unless
584 * we have completely cleaned the inode.
586 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
587 (wbc
->sync_mode
!= WB_SYNC_ALL
||
588 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
590 inode
->i_state
|= I_SYNC
;
591 spin_unlock(&inode
->i_lock
);
593 ret
= __writeback_single_inode(inode
, wbc
);
595 spin_lock(&wb
->list_lock
);
596 spin_lock(&inode
->i_lock
);
598 * If inode is clean, remove it from writeback lists. Otherwise don't
599 * touch it. See comment above for explanation.
601 if (!(inode
->i_state
& I_DIRTY_ALL
))
602 list_del_init(&inode
->i_wb_list
);
603 spin_unlock(&wb
->list_lock
);
604 inode_sync_complete(inode
);
606 spin_unlock(&inode
->i_lock
);
610 static long writeback_chunk_size(struct bdi_writeback
*wb
,
611 struct wb_writeback_work
*work
)
616 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
617 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
618 * here avoids calling into writeback_inodes_wb() more than once.
620 * The intended call sequence for WB_SYNC_ALL writeback is:
623 * writeback_sb_inodes() <== called only once
624 * write_cache_pages() <== called once for each inode
625 * (quickly) tag currently dirty pages
626 * (maybe slowly) sync all tagged pages
628 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
631 pages
= min(wb
->avg_write_bandwidth
/ 2,
632 global_dirty_limit
/ DIRTY_SCOPE
);
633 pages
= min(pages
, work
->nr_pages
);
634 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
635 MIN_WRITEBACK_PAGES
);
642 * Write a portion of b_io inodes which belong to @sb.
644 * Return the number of pages and/or inodes written.
646 static long writeback_sb_inodes(struct super_block
*sb
,
647 struct bdi_writeback
*wb
,
648 struct wb_writeback_work
*work
)
650 struct writeback_control wbc
= {
651 .sync_mode
= work
->sync_mode
,
652 .tagged_writepages
= work
->tagged_writepages
,
653 .for_kupdate
= work
->for_kupdate
,
654 .for_background
= work
->for_background
,
655 .for_sync
= work
->for_sync
,
656 .range_cyclic
= work
->range_cyclic
,
658 .range_end
= LLONG_MAX
,
660 unsigned long start_time
= jiffies
;
662 long wrote
= 0; /* count both pages and inodes */
664 while (!list_empty(&wb
->b_io
)) {
665 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
667 if (inode
->i_sb
!= sb
) {
670 * We only want to write back data for this
671 * superblock, move all inodes not belonging
672 * to it back onto the dirty list.
674 redirty_tail(inode
, wb
);
679 * The inode belongs to a different superblock.
680 * Bounce back to the caller to unpin this and
681 * pin the next superblock.
687 * Don't bother with new inodes or inodes being freed, first
688 * kind does not need periodic writeout yet, and for the latter
689 * kind writeout is handled by the freer.
691 spin_lock(&inode
->i_lock
);
692 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
693 spin_unlock(&inode
->i_lock
);
694 redirty_tail(inode
, wb
);
697 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
699 * If this inode is locked for writeback and we are not
700 * doing writeback-for-data-integrity, move it to
701 * b_more_io so that writeback can proceed with the
702 * other inodes on s_io.
704 * We'll have another go at writing back this inode
705 * when we completed a full scan of b_io.
707 spin_unlock(&inode
->i_lock
);
708 requeue_io(inode
, wb
);
709 trace_writeback_sb_inodes_requeue(inode
);
712 spin_unlock(&wb
->list_lock
);
715 * We already requeued the inode if it had I_SYNC set and we
716 * are doing WB_SYNC_NONE writeback. So this catches only the
719 if (inode
->i_state
& I_SYNC
) {
720 /* Wait for I_SYNC. This function drops i_lock... */
721 inode_sleep_on_writeback(inode
);
722 /* Inode may be gone, start again */
723 spin_lock(&wb
->list_lock
);
726 inode
->i_state
|= I_SYNC
;
727 spin_unlock(&inode
->i_lock
);
729 write_chunk
= writeback_chunk_size(wb
, work
);
730 wbc
.nr_to_write
= write_chunk
;
731 wbc
.pages_skipped
= 0;
734 * We use I_SYNC to pin the inode in memory. While it is set
735 * evict_inode() will wait so the inode cannot be freed.
737 __writeback_single_inode(inode
, &wbc
);
739 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
740 wrote
+= write_chunk
- wbc
.nr_to_write
;
741 spin_lock(&wb
->list_lock
);
742 spin_lock(&inode
->i_lock
);
743 if (!(inode
->i_state
& I_DIRTY_ALL
))
745 requeue_inode(inode
, wb
, &wbc
);
746 inode_sync_complete(inode
);
747 spin_unlock(&inode
->i_lock
);
748 cond_resched_lock(&wb
->list_lock
);
750 * bail out to wb_writeback() often enough to check
751 * background threshold and other termination conditions.
754 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
756 if (work
->nr_pages
<= 0)
763 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
764 struct wb_writeback_work
*work
)
766 unsigned long start_time
= jiffies
;
769 while (!list_empty(&wb
->b_io
)) {
770 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
771 struct super_block
*sb
= inode
->i_sb
;
773 if (!trylock_super(sb
)) {
775 * trylock_super() may fail consistently due to
776 * s_umount being grabbed by someone else. Don't use
777 * requeue_io() to avoid busy retrying the inode/sb.
779 redirty_tail(inode
, wb
);
782 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
783 up_read(&sb
->s_umount
);
785 /* refer to the same tests at the end of writeback_sb_inodes */
787 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
789 if (work
->nr_pages
<= 0)
793 /* Leave any unwritten inodes on b_io */
797 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
798 enum wb_reason reason
)
800 struct wb_writeback_work work
= {
801 .nr_pages
= nr_pages
,
802 .sync_mode
= WB_SYNC_NONE
,
807 spin_lock(&wb
->list_lock
);
808 if (list_empty(&wb
->b_io
))
810 __writeback_inodes_wb(wb
, &work
);
811 spin_unlock(&wb
->list_lock
);
813 return nr_pages
- work
.nr_pages
;
816 static bool over_bground_thresh(struct bdi_writeback
*wb
)
818 unsigned long background_thresh
, dirty_thresh
;
820 global_dirty_limits(&background_thresh
, &dirty_thresh
);
822 if (global_page_state(NR_FILE_DIRTY
) +
823 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
)
826 if (wb_stat(wb
, WB_RECLAIMABLE
) > wb_dirty_limit(wb
, background_thresh
))
833 * Called under wb->list_lock. If there are multiple wb per bdi,
834 * only the flusher working on the first wb should do it.
836 static void wb_update_bandwidth(struct bdi_writeback
*wb
,
837 unsigned long start_time
)
839 __wb_update_bandwidth(wb
, 0, 0, 0, 0, 0, start_time
);
843 * Explicit flushing or periodic writeback of "old" data.
845 * Define "old": the first time one of an inode's pages is dirtied, we mark the
846 * dirtying-time in the inode's address_space. So this periodic writeback code
847 * just walks the superblock inode list, writing back any inodes which are
848 * older than a specific point in time.
850 * Try to run once per dirty_writeback_interval. But if a writeback event
851 * takes longer than a dirty_writeback_interval interval, then leave a
854 * older_than_this takes precedence over nr_to_write. So we'll only write back
855 * all dirty pages if they are all attached to "old" mappings.
857 static long wb_writeback(struct bdi_writeback
*wb
,
858 struct wb_writeback_work
*work
)
860 unsigned long wb_start
= jiffies
;
861 long nr_pages
= work
->nr_pages
;
862 unsigned long oldest_jif
;
866 oldest_jif
= jiffies
;
867 work
->older_than_this
= &oldest_jif
;
869 spin_lock(&wb
->list_lock
);
872 * Stop writeback when nr_pages has been consumed
874 if (work
->nr_pages
<= 0)
878 * Background writeout and kupdate-style writeback may
879 * run forever. Stop them if there is other work to do
880 * so that e.g. sync can proceed. They'll be restarted
881 * after the other works are all done.
883 if ((work
->for_background
|| work
->for_kupdate
) &&
884 !list_empty(&wb
->work_list
))
888 * For background writeout, stop when we are below the
889 * background dirty threshold
891 if (work
->for_background
&& !over_bground_thresh(wb
))
895 * Kupdate and background works are special and we want to
896 * include all inodes that need writing. Livelock avoidance is
897 * handled by these works yielding to any other work so we are
900 if (work
->for_kupdate
) {
901 oldest_jif
= jiffies
-
902 msecs_to_jiffies(dirty_expire_interval
* 10);
903 } else if (work
->for_background
)
904 oldest_jif
= jiffies
;
906 trace_writeback_start(wb
->bdi
, work
);
907 if (list_empty(&wb
->b_io
))
910 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
912 progress
= __writeback_inodes_wb(wb
, work
);
913 trace_writeback_written(wb
->bdi
, work
);
915 wb_update_bandwidth(wb
, wb_start
);
918 * Did we write something? Try for more
920 * Dirty inodes are moved to b_io for writeback in batches.
921 * The completion of the current batch does not necessarily
922 * mean the overall work is done. So we keep looping as long
923 * as made some progress on cleaning pages or inodes.
928 * No more inodes for IO, bail
930 if (list_empty(&wb
->b_more_io
))
933 * Nothing written. Wait for some inode to
934 * become available for writeback. Otherwise
935 * we'll just busyloop.
937 if (!list_empty(&wb
->b_more_io
)) {
938 trace_writeback_wait(wb
->bdi
, work
);
939 inode
= wb_inode(wb
->b_more_io
.prev
);
940 spin_lock(&inode
->i_lock
);
941 spin_unlock(&wb
->list_lock
);
942 /* This function drops i_lock... */
943 inode_sleep_on_writeback(inode
);
944 spin_lock(&wb
->list_lock
);
947 spin_unlock(&wb
->list_lock
);
949 return nr_pages
- work
->nr_pages
;
953 * Return the next wb_writeback_work struct that hasn't been processed yet.
955 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
957 struct wb_writeback_work
*work
= NULL
;
959 spin_lock_bh(&wb
->work_lock
);
960 if (!list_empty(&wb
->work_list
)) {
961 work
= list_entry(wb
->work_list
.next
,
962 struct wb_writeback_work
, list
);
963 list_del_init(&work
->list
);
965 spin_unlock_bh(&wb
->work_lock
);
970 * Add in the number of potentially dirty inodes, because each inode
971 * write can dirty pagecache in the underlying blockdev.
973 static unsigned long get_nr_dirty_pages(void)
975 return global_page_state(NR_FILE_DIRTY
) +
976 global_page_state(NR_UNSTABLE_NFS
) +
977 get_nr_dirty_inodes();
980 static long wb_check_background_flush(struct bdi_writeback
*wb
)
982 if (over_bground_thresh(wb
)) {
984 struct wb_writeback_work work
= {
985 .nr_pages
= LONG_MAX
,
986 .sync_mode
= WB_SYNC_NONE
,
989 .reason
= WB_REASON_BACKGROUND
,
992 return wb_writeback(wb
, &work
);
998 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1000 unsigned long expired
;
1004 * When set to zero, disable periodic writeback
1006 if (!dirty_writeback_interval
)
1009 expired
= wb
->last_old_flush
+
1010 msecs_to_jiffies(dirty_writeback_interval
* 10);
1011 if (time_before(jiffies
, expired
))
1014 wb
->last_old_flush
= jiffies
;
1015 nr_pages
= get_nr_dirty_pages();
1018 struct wb_writeback_work work
= {
1019 .nr_pages
= nr_pages
,
1020 .sync_mode
= WB_SYNC_NONE
,
1023 .reason
= WB_REASON_PERIODIC
,
1026 return wb_writeback(wb
, &work
);
1033 * Retrieve work items and do the writeback they describe
1035 static long wb_do_writeback(struct bdi_writeback
*wb
)
1037 struct wb_writeback_work
*work
;
1040 set_bit(WB_writeback_running
, &wb
->state
);
1041 while ((work
= get_next_work_item(wb
)) != NULL
) {
1043 trace_writeback_exec(wb
->bdi
, work
);
1045 wrote
+= wb_writeback(wb
, work
);
1048 * Notify the caller of completion if this is a synchronous
1049 * work item, otherwise just free it.
1052 complete(work
->done
);
1058 * Check for periodic writeback, kupdated() style
1060 wrote
+= wb_check_old_data_flush(wb
);
1061 wrote
+= wb_check_background_flush(wb
);
1062 clear_bit(WB_writeback_running
, &wb
->state
);
1068 * Handle writeback of dirty data for the device backed by this bdi. Also
1069 * reschedules periodically and does kupdated style flushing.
1071 void wb_workfn(struct work_struct
*work
)
1073 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1074 struct bdi_writeback
, dwork
);
1077 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1078 current
->flags
|= PF_SWAPWRITE
;
1080 if (likely(!current_is_workqueue_rescuer() ||
1081 !test_bit(WB_registered
, &wb
->state
))) {
1083 * The normal path. Keep writing back @wb until its
1084 * work_list is empty. Note that this path is also taken
1085 * if @wb is shutting down even when we're running off the
1086 * rescuer as work_list needs to be drained.
1089 pages_written
= wb_do_writeback(wb
);
1090 trace_writeback_pages_written(pages_written
);
1091 } while (!list_empty(&wb
->work_list
));
1094 * bdi_wq can't get enough workers and we're running off
1095 * the emergency worker. Don't hog it. Hopefully, 1024 is
1096 * enough for efficient IO.
1098 pages_written
= writeback_inodes_wb(wb
, 1024,
1099 WB_REASON_FORKER_THREAD
);
1100 trace_writeback_pages_written(pages_written
);
1103 if (!list_empty(&wb
->work_list
))
1104 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1105 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1106 wb_wakeup_delayed(wb
);
1108 current
->flags
&= ~PF_SWAPWRITE
;
1112 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1115 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1117 struct backing_dev_info
*bdi
;
1120 nr_pages
= get_nr_dirty_pages();
1123 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1124 if (!bdi_has_dirty_io(bdi
))
1126 __wb_start_writeback(&bdi
->wb
, nr_pages
, false, reason
);
1132 * Wake up bdi's periodically to make sure dirtytime inodes gets
1133 * written back periodically. We deliberately do *not* check the
1134 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1135 * kernel to be constantly waking up once there are any dirtytime
1136 * inodes on the system. So instead we define a separate delayed work
1137 * function which gets called much more rarely. (By default, only
1138 * once every 12 hours.)
1140 * If there is any other write activity going on in the file system,
1141 * this function won't be necessary. But if the only thing that has
1142 * happened on the file system is a dirtytime inode caused by an atime
1143 * update, we need this infrastructure below to make sure that inode
1144 * eventually gets pushed out to disk.
1146 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1147 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1149 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1151 struct backing_dev_info
*bdi
;
1154 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1155 if (list_empty(&bdi
->wb
.b_dirty_time
))
1157 wb_wakeup(&bdi
->wb
);
1160 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1163 static int __init
start_dirtytime_writeback(void)
1165 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1168 __initcall(start_dirtytime_writeback
);
1170 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1171 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1175 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1176 if (ret
== 0 && write
)
1177 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1181 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1183 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1184 struct dentry
*dentry
;
1185 const char *name
= "?";
1187 dentry
= d_find_alias(inode
);
1189 spin_lock(&dentry
->d_lock
);
1190 name
= (const char *) dentry
->d_name
.name
;
1193 "%s(%d): dirtied inode %lu (%s) on %s\n",
1194 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1195 name
, inode
->i_sb
->s_id
);
1197 spin_unlock(&dentry
->d_lock
);
1204 * __mark_inode_dirty - internal function
1205 * @inode: inode to mark
1206 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1207 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1208 * mark_inode_dirty_sync.
1210 * Put the inode on the super block's dirty list.
1212 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1213 * dirty list only if it is hashed or if it refers to a blockdev.
1214 * If it was not hashed, it will never be added to the dirty list
1215 * even if it is later hashed, as it will have been marked dirty already.
1217 * In short, make sure you hash any inodes _before_ you start marking
1220 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1221 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1222 * the kernel-internal blockdev inode represents the dirtying time of the
1223 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1224 * page->mapping->host, so the page-dirtying time is recorded in the internal
1227 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1228 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1230 struct super_block
*sb
= inode
->i_sb
;
1231 struct backing_dev_info
*bdi
= NULL
;
1234 trace_writeback_mark_inode_dirty(inode
, flags
);
1237 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1238 * dirty the inode itself
1240 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
1241 trace_writeback_dirty_inode_start(inode
, flags
);
1243 if (sb
->s_op
->dirty_inode
)
1244 sb
->s_op
->dirty_inode(inode
, flags
);
1246 trace_writeback_dirty_inode(inode
, flags
);
1248 if (flags
& I_DIRTY_INODE
)
1249 flags
&= ~I_DIRTY_TIME
;
1250 dirtytime
= flags
& I_DIRTY_TIME
;
1253 * Paired with smp_mb() in __writeback_single_inode() for the
1254 * following lockless i_state test. See there for details.
1258 if (((inode
->i_state
& flags
) == flags
) ||
1259 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
1262 if (unlikely(block_dump
))
1263 block_dump___mark_inode_dirty(inode
);
1265 spin_lock(&inode
->i_lock
);
1266 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
1267 goto out_unlock_inode
;
1268 if ((inode
->i_state
& flags
) != flags
) {
1269 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1271 inode_attach_wb(inode
, NULL
);
1273 if (flags
& I_DIRTY_INODE
)
1274 inode
->i_state
&= ~I_DIRTY_TIME
;
1275 inode
->i_state
|= flags
;
1278 * If the inode is being synced, just update its dirty state.
1279 * The unlocker will place the inode on the appropriate
1280 * superblock list, based upon its state.
1282 if (inode
->i_state
& I_SYNC
)
1283 goto out_unlock_inode
;
1286 * Only add valid (hashed) inodes to the superblock's
1287 * dirty list. Add blockdev inodes as well.
1289 if (!S_ISBLK(inode
->i_mode
)) {
1290 if (inode_unhashed(inode
))
1291 goto out_unlock_inode
;
1293 if (inode
->i_state
& I_FREEING
)
1294 goto out_unlock_inode
;
1297 * If the inode was already on b_dirty/b_io/b_more_io, don't
1298 * reposition it (that would break b_dirty time-ordering).
1301 bool wakeup_bdi
= false;
1302 bdi
= inode_to_bdi(inode
);
1304 spin_unlock(&inode
->i_lock
);
1305 spin_lock(&bdi
->wb
.list_lock
);
1306 if (bdi_cap_writeback_dirty(bdi
)) {
1307 WARN(!test_bit(WB_registered
, &bdi
->wb
.state
),
1308 "bdi-%s not registered\n", bdi
->name
);
1311 * If this is the first dirty inode for this
1312 * bdi, we have to wake-up the corresponding
1313 * bdi thread to make sure background
1314 * write-back happens later.
1316 if (!wb_has_dirty_io(&bdi
->wb
))
1320 inode
->dirtied_when
= jiffies
;
1322 inode
->dirtied_time_when
= jiffies
;
1323 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
1324 list_move(&inode
->i_wb_list
, &bdi
->wb
.b_dirty
);
1326 list_move(&inode
->i_wb_list
,
1327 &bdi
->wb
.b_dirty_time
);
1328 spin_unlock(&bdi
->wb
.list_lock
);
1329 trace_writeback_dirty_inode_enqueue(inode
);
1332 wb_wakeup_delayed(&bdi
->wb
);
1337 spin_unlock(&inode
->i_lock
);
1340 EXPORT_SYMBOL(__mark_inode_dirty
);
1342 static void wait_sb_inodes(struct super_block
*sb
)
1344 struct inode
*inode
, *old_inode
= NULL
;
1347 * We need to be protected against the filesystem going from
1348 * r/o to r/w or vice versa.
1350 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1352 spin_lock(&inode_sb_list_lock
);
1355 * Data integrity sync. Must wait for all pages under writeback,
1356 * because there may have been pages dirtied before our sync
1357 * call, but which had writeout started before we write it out.
1358 * In which case, the inode may not be on the dirty list, but
1359 * we still have to wait for that writeout.
1361 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1362 struct address_space
*mapping
= inode
->i_mapping
;
1364 spin_lock(&inode
->i_lock
);
1365 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1366 (mapping
->nrpages
== 0)) {
1367 spin_unlock(&inode
->i_lock
);
1371 spin_unlock(&inode
->i_lock
);
1372 spin_unlock(&inode_sb_list_lock
);
1375 * We hold a reference to 'inode' so it couldn't have been
1376 * removed from s_inodes list while we dropped the
1377 * inode_sb_list_lock. We cannot iput the inode now as we can
1378 * be holding the last reference and we cannot iput it under
1379 * inode_sb_list_lock. So we keep the reference and iput it
1385 filemap_fdatawait(mapping
);
1389 spin_lock(&inode_sb_list_lock
);
1391 spin_unlock(&inode_sb_list_lock
);
1396 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1397 * @sb: the superblock
1398 * @nr: the number of pages to write
1399 * @reason: reason why some writeback work initiated
1401 * Start writeback on some inodes on this super_block. No guarantees are made
1402 * on how many (if any) will be written, and this function does not wait
1403 * for IO completion of submitted IO.
1405 void writeback_inodes_sb_nr(struct super_block
*sb
,
1407 enum wb_reason reason
)
1409 DECLARE_COMPLETION_ONSTACK(done
);
1410 struct wb_writeback_work work
= {
1412 .sync_mode
= WB_SYNC_NONE
,
1413 .tagged_writepages
= 1,
1419 if (sb
->s_bdi
== &noop_backing_dev_info
)
1421 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1422 wb_queue_work(&sb
->s_bdi
->wb
, &work
);
1423 wait_for_completion(&done
);
1425 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1428 * writeback_inodes_sb - writeback dirty inodes from given super_block
1429 * @sb: the superblock
1430 * @reason: reason why some writeback work was initiated
1432 * Start writeback on some inodes on this super_block. No guarantees are made
1433 * on how many (if any) will be written, and this function does not wait
1434 * for IO completion of submitted IO.
1436 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1438 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1440 EXPORT_SYMBOL(writeback_inodes_sb
);
1443 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1444 * @sb: the superblock
1445 * @nr: the number of pages to write
1446 * @reason: the reason of writeback
1448 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1449 * Returns 1 if writeback was started, 0 if not.
1451 int try_to_writeback_inodes_sb_nr(struct super_block
*sb
,
1453 enum wb_reason reason
)
1455 if (writeback_in_progress(sb
->s_bdi
))
1458 if (!down_read_trylock(&sb
->s_umount
))
1461 writeback_inodes_sb_nr(sb
, nr
, reason
);
1462 up_read(&sb
->s_umount
);
1465 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1468 * try_to_writeback_inodes_sb - try to start writeback if none underway
1469 * @sb: the superblock
1470 * @reason: reason why some writeback work was initiated
1472 * Implement by try_to_writeback_inodes_sb_nr()
1473 * Returns 1 if writeback was started, 0 if not.
1475 int try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1477 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1479 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1482 * sync_inodes_sb - sync sb inode pages
1483 * @sb: the superblock
1485 * This function writes and waits on any dirty inode belonging to this
1488 void sync_inodes_sb(struct super_block
*sb
)
1490 DECLARE_COMPLETION_ONSTACK(done
);
1491 struct wb_writeback_work work
= {
1493 .sync_mode
= WB_SYNC_ALL
,
1494 .nr_pages
= LONG_MAX
,
1497 .reason
= WB_REASON_SYNC
,
1501 /* Nothing to do? */
1502 if (sb
->s_bdi
== &noop_backing_dev_info
)
1504 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1506 wb_queue_work(&sb
->s_bdi
->wb
, &work
);
1507 wait_for_completion(&done
);
1511 EXPORT_SYMBOL(sync_inodes_sb
);
1514 * write_inode_now - write an inode to disk
1515 * @inode: inode to write to disk
1516 * @sync: whether the write should be synchronous or not
1518 * This function commits an inode to disk immediately if it is dirty. This is
1519 * primarily needed by knfsd.
1521 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1523 int write_inode_now(struct inode
*inode
, int sync
)
1525 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1526 struct writeback_control wbc
= {
1527 .nr_to_write
= LONG_MAX
,
1528 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1530 .range_end
= LLONG_MAX
,
1533 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1534 wbc
.nr_to_write
= 0;
1537 return writeback_single_inode(inode
, wb
, &wbc
);
1539 EXPORT_SYMBOL(write_inode_now
);
1542 * sync_inode - write an inode and its pages to disk.
1543 * @inode: the inode to sync
1544 * @wbc: controls the writeback mode
1546 * sync_inode() will write an inode and its pages to disk. It will also
1547 * correctly update the inode on its superblock's dirty inode lists and will
1548 * update inode->i_state.
1550 * The caller must have a ref on the inode.
1552 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1554 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1556 EXPORT_SYMBOL(sync_inode
);
1559 * sync_inode_metadata - write an inode to disk
1560 * @inode: the inode to sync
1561 * @wait: wait for I/O to complete.
1563 * Write an inode to disk and adjust its dirty state after completion.
1565 * Note: only writes the actual inode, no associated data or other metadata.
1567 int sync_inode_metadata(struct inode
*inode
, int wait
)
1569 struct writeback_control wbc
= {
1570 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1571 .nr_to_write
= 0, /* metadata-only */
1574 return sync_inode(inode
, &wbc
);
1576 EXPORT_SYMBOL(sync_inode_metadata
);