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 struct backing_dev_info
*inode_to_bdi(struct inode
*inode
)
83 struct super_block
*sb
;
86 return &noop_backing_dev_info
;
90 if (sb_is_blkdev_sb(sb
))
91 return blk_get_backing_dev_info(I_BDEV(inode
));
95 EXPORT_SYMBOL_GPL(inode_to_bdi
);
97 static inline struct inode
*wb_inode(struct list_head
*head
)
99 return list_entry(head
, struct inode
, i_wb_list
);
103 * Include the creation of the trace points after defining the
104 * wb_writeback_work structure and inline functions so that the definition
105 * remains local to this file.
107 #define CREATE_TRACE_POINTS
108 #include <trace/events/writeback.h>
110 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage
);
112 static void bdi_wakeup_thread(struct backing_dev_info
*bdi
)
114 spin_lock_bh(&bdi
->wb_lock
);
115 if (test_bit(WB_registered
, &bdi
->wb
.state
))
116 mod_delayed_work(bdi_wq
, &bdi
->wb
.dwork
, 0);
117 spin_unlock_bh(&bdi
->wb_lock
);
120 static void bdi_queue_work(struct backing_dev_info
*bdi
,
121 struct wb_writeback_work
*work
)
123 trace_writeback_queue(bdi
, work
);
125 spin_lock_bh(&bdi
->wb_lock
);
126 if (!test_bit(WB_registered
, &bdi
->wb
.state
)) {
128 complete(work
->done
);
131 list_add_tail(&work
->list
, &bdi
->work_list
);
132 mod_delayed_work(bdi_wq
, &bdi
->wb
.dwork
, 0);
134 spin_unlock_bh(&bdi
->wb_lock
);
138 __bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
139 bool range_cyclic
, enum wb_reason reason
)
141 struct wb_writeback_work
*work
;
144 * This is WB_SYNC_NONE writeback, so if allocation fails just
145 * wakeup the thread for old dirty data writeback
147 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
149 trace_writeback_nowork(bdi
);
150 bdi_wakeup_thread(bdi
);
154 work
->sync_mode
= WB_SYNC_NONE
;
155 work
->nr_pages
= nr_pages
;
156 work
->range_cyclic
= range_cyclic
;
157 work
->reason
= reason
;
159 bdi_queue_work(bdi
, work
);
163 * bdi_start_writeback - start writeback
164 * @bdi: the backing device to write from
165 * @nr_pages: the number of pages to write
166 * @reason: reason why some writeback work was initiated
169 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
170 * started when this function returns, we make no guarantees on
171 * completion. Caller need not hold sb s_umount semaphore.
174 void bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
175 enum wb_reason reason
)
177 __bdi_start_writeback(bdi
, nr_pages
, true, reason
);
181 * bdi_start_background_writeback - start background writeback
182 * @bdi: the backing device to write from
185 * This makes sure WB_SYNC_NONE background writeback happens. When
186 * this function returns, it is only guaranteed that for given BDI
187 * some IO is happening if we are over background dirty threshold.
188 * Caller need not hold sb s_umount semaphore.
190 void bdi_start_background_writeback(struct backing_dev_info
*bdi
)
193 * We just wake up the flusher thread. It will perform background
194 * writeback as soon as there is no other work to do.
196 trace_writeback_wake_background(bdi
);
197 bdi_wakeup_thread(bdi
);
201 * Remove the inode from the writeback list it is on.
203 void inode_wb_list_del(struct inode
*inode
)
205 struct backing_dev_info
*bdi
= inode_to_bdi(inode
);
207 spin_lock(&bdi
->wb
.list_lock
);
208 list_del_init(&inode
->i_wb_list
);
209 spin_unlock(&bdi
->wb
.list_lock
);
213 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
214 * furthest end of its superblock's dirty-inode list.
216 * Before stamping the inode's ->dirtied_when, we check to see whether it is
217 * already the most-recently-dirtied inode on the b_dirty list. If that is
218 * the case then the inode must have been redirtied while it was being written
219 * out and we don't reset its dirtied_when.
221 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
223 assert_spin_locked(&wb
->list_lock
);
224 if (!list_empty(&wb
->b_dirty
)) {
227 tail
= wb_inode(wb
->b_dirty
.next
);
228 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
229 inode
->dirtied_when
= jiffies
;
231 list_move(&inode
->i_wb_list
, &wb
->b_dirty
);
235 * requeue inode for re-scanning after bdi->b_io list is exhausted.
237 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
239 assert_spin_locked(&wb
->list_lock
);
240 list_move(&inode
->i_wb_list
, &wb
->b_more_io
);
243 static void inode_sync_complete(struct inode
*inode
)
245 inode
->i_state
&= ~I_SYNC
;
246 /* If inode is clean an unused, put it into LRU now... */
247 inode_add_lru(inode
);
248 /* Waiters must see I_SYNC cleared before being woken up */
250 wake_up_bit(&inode
->i_state
, __I_SYNC
);
253 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
255 bool ret
= time_after(inode
->dirtied_when
, t
);
258 * For inodes being constantly redirtied, dirtied_when can get stuck.
259 * It _appears_ to be in the future, but is actually in distant past.
260 * This test is necessary to prevent such wrapped-around relative times
261 * from permanently stopping the whole bdi writeback.
263 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
268 #define EXPIRE_DIRTY_ATIME 0x0001
271 * Move expired (dirtied before work->older_than_this) dirty inodes from
272 * @delaying_queue to @dispatch_queue.
274 static int move_expired_inodes(struct list_head
*delaying_queue
,
275 struct list_head
*dispatch_queue
,
277 struct wb_writeback_work
*work
)
279 unsigned long *older_than_this
= NULL
;
280 unsigned long expire_time
;
282 struct list_head
*pos
, *node
;
283 struct super_block
*sb
= NULL
;
288 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
289 older_than_this
= work
->older_than_this
;
290 else if (!work
->for_sync
) {
291 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
292 older_than_this
= &expire_time
;
294 while (!list_empty(delaying_queue
)) {
295 inode
= wb_inode(delaying_queue
->prev
);
296 if (older_than_this
&&
297 inode_dirtied_after(inode
, *older_than_this
))
299 list_move(&inode
->i_wb_list
, &tmp
);
301 if (flags
& EXPIRE_DIRTY_ATIME
)
302 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
303 if (sb_is_blkdev_sb(inode
->i_sb
))
305 if (sb
&& sb
!= inode
->i_sb
)
310 /* just one sb in list, splice to dispatch_queue and we're done */
312 list_splice(&tmp
, dispatch_queue
);
316 /* Move inodes from one superblock together */
317 while (!list_empty(&tmp
)) {
318 sb
= wb_inode(tmp
.prev
)->i_sb
;
319 list_for_each_prev_safe(pos
, node
, &tmp
) {
320 inode
= wb_inode(pos
);
321 if (inode
->i_sb
== sb
)
322 list_move(&inode
->i_wb_list
, dispatch_queue
);
330 * Queue all expired dirty inodes for io, eldest first.
332 * newly dirtied b_dirty b_io b_more_io
333 * =============> gf edc BA
335 * newly dirtied b_dirty b_io b_more_io
336 * =============> g fBAedc
338 * +--> dequeue for IO
340 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
344 assert_spin_locked(&wb
->list_lock
);
345 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
346 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
347 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
348 EXPIRE_DIRTY_ATIME
, work
);
349 trace_writeback_queue_io(wb
, work
, moved
);
352 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
356 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
357 trace_writeback_write_inode_start(inode
, wbc
);
358 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
359 trace_writeback_write_inode(inode
, wbc
);
366 * Wait for writeback on an inode to complete. Called with i_lock held.
367 * Caller must make sure inode cannot go away when we drop i_lock.
369 static void __inode_wait_for_writeback(struct inode
*inode
)
370 __releases(inode
->i_lock
)
371 __acquires(inode
->i_lock
)
373 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
374 wait_queue_head_t
*wqh
;
376 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
377 while (inode
->i_state
& I_SYNC
) {
378 spin_unlock(&inode
->i_lock
);
379 __wait_on_bit(wqh
, &wq
, bit_wait
,
380 TASK_UNINTERRUPTIBLE
);
381 spin_lock(&inode
->i_lock
);
386 * Wait for writeback on an inode to complete. Caller must have inode pinned.
388 void inode_wait_for_writeback(struct inode
*inode
)
390 spin_lock(&inode
->i_lock
);
391 __inode_wait_for_writeback(inode
);
392 spin_unlock(&inode
->i_lock
);
396 * Sleep until I_SYNC is cleared. This function must be called with i_lock
397 * held and drops it. It is aimed for callers not holding any inode reference
398 * so once i_lock is dropped, inode can go away.
400 static void inode_sleep_on_writeback(struct inode
*inode
)
401 __releases(inode
->i_lock
)
404 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
407 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
408 sleep
= inode
->i_state
& I_SYNC
;
409 spin_unlock(&inode
->i_lock
);
412 finish_wait(wqh
, &wait
);
416 * Find proper writeback list for the inode depending on its current state and
417 * possibly also change of its state while we were doing writeback. Here we
418 * handle things such as livelock prevention or fairness of writeback among
419 * inodes. This function can be called only by flusher thread - noone else
420 * processes all inodes in writeback lists and requeueing inodes behind flusher
421 * thread's back can have unexpected consequences.
423 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
424 struct writeback_control
*wbc
)
426 if (inode
->i_state
& I_FREEING
)
430 * Sync livelock prevention. Each inode is tagged and synced in one
431 * shot. If still dirty, it will be redirty_tail()'ed below. Update
432 * the dirty time to prevent enqueue and sync it again.
434 if ((inode
->i_state
& I_DIRTY
) &&
435 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
436 inode
->dirtied_when
= jiffies
;
438 if (wbc
->pages_skipped
) {
440 * writeback is not making progress due to locked
441 * buffers. Skip this inode for now.
443 redirty_tail(inode
, wb
);
447 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
449 * We didn't write back all the pages. nfs_writepages()
450 * sometimes bales out without doing anything.
452 if (wbc
->nr_to_write
<= 0) {
453 /* Slice used up. Queue for next turn. */
454 requeue_io(inode
, wb
);
457 * Writeback blocked by something other than
458 * congestion. Delay the inode for some time to
459 * avoid spinning on the CPU (100% iowait)
460 * retrying writeback of the dirty page/inode
461 * that cannot be performed immediately.
463 redirty_tail(inode
, wb
);
465 } else if (inode
->i_state
& I_DIRTY
) {
467 * Filesystems can dirty the inode during writeback operations,
468 * such as delayed allocation during submission or metadata
469 * updates after data IO completion.
471 redirty_tail(inode
, wb
);
472 } else if (inode
->i_state
& I_DIRTY_TIME
) {
473 inode
->dirtied_when
= jiffies
;
474 list_move(&inode
->i_wb_list
, &wb
->b_dirty_time
);
476 /* The inode is clean. Remove from writeback lists. */
477 list_del_init(&inode
->i_wb_list
);
482 * Write out an inode and its dirty pages. Do not update the writeback list
483 * linkage. That is left to the caller. The caller is also responsible for
484 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
487 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
489 struct address_space
*mapping
= inode
->i_mapping
;
490 long nr_to_write
= wbc
->nr_to_write
;
494 WARN_ON(!(inode
->i_state
& I_SYNC
));
496 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
498 ret
= do_writepages(mapping
, wbc
);
501 * Make sure to wait on the data before writing out the metadata.
502 * This is important for filesystems that modify metadata on data
503 * I/O completion. We don't do it for sync(2) writeback because it has a
504 * separate, external IO completion path and ->sync_fs for guaranteeing
505 * inode metadata is written back correctly.
507 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
508 int err
= filemap_fdatawait(mapping
);
514 * Some filesystems may redirty the inode during the writeback
515 * due to delalloc, clear dirty metadata flags right before
518 spin_lock(&inode
->i_lock
);
520 dirty
= inode
->i_state
& I_DIRTY
;
521 if (inode
->i_state
& I_DIRTY_TIME
) {
522 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
523 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
524 unlikely(time_after(jiffies
,
525 (inode
->dirtied_time_when
+
526 dirtytime_expire_interval
* HZ
)))) {
527 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
528 trace_writeback_lazytime(inode
);
531 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
532 inode
->i_state
&= ~dirty
;
535 * Paired with smp_mb() in __mark_inode_dirty(). This allows
536 * __mark_inode_dirty() to test i_state without grabbing i_lock -
537 * either they see the I_DIRTY bits cleared or we see the dirtied
540 * I_DIRTY_PAGES is always cleared together above even if @mapping
541 * still has dirty pages. The flag is reinstated after smp_mb() if
542 * necessary. This guarantees that either __mark_inode_dirty()
543 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
547 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
548 inode
->i_state
|= I_DIRTY_PAGES
;
550 spin_unlock(&inode
->i_lock
);
552 if (dirty
& I_DIRTY_TIME
)
553 mark_inode_dirty_sync(inode
);
554 /* Don't write the inode if only I_DIRTY_PAGES was set */
555 if (dirty
& ~I_DIRTY_PAGES
) {
556 int err
= write_inode(inode
, wbc
);
560 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
565 * Write out an inode's dirty pages. Either the caller has an active reference
566 * on the inode or the inode has I_WILL_FREE set.
568 * This function is designed to be called for writing back one inode which
569 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
570 * and does more profound writeback list handling in writeback_sb_inodes().
573 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
574 struct writeback_control
*wbc
)
578 spin_lock(&inode
->i_lock
);
579 if (!atomic_read(&inode
->i_count
))
580 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
582 WARN_ON(inode
->i_state
& I_WILL_FREE
);
584 if (inode
->i_state
& I_SYNC
) {
585 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
588 * It's a data-integrity sync. We must wait. Since callers hold
589 * inode reference or inode has I_WILL_FREE set, it cannot go
592 __inode_wait_for_writeback(inode
);
594 WARN_ON(inode
->i_state
& I_SYNC
);
596 * Skip inode if it is clean and we have no outstanding writeback in
597 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
598 * function since flusher thread may be doing for example sync in
599 * parallel and if we move the inode, it could get skipped. So here we
600 * make sure inode is on some writeback list and leave it there unless
601 * we have completely cleaned the inode.
603 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
604 (wbc
->sync_mode
!= WB_SYNC_ALL
||
605 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
607 inode
->i_state
|= I_SYNC
;
608 spin_unlock(&inode
->i_lock
);
610 ret
= __writeback_single_inode(inode
, wbc
);
612 spin_lock(&wb
->list_lock
);
613 spin_lock(&inode
->i_lock
);
615 * If inode is clean, remove it from writeback lists. Otherwise don't
616 * touch it. See comment above for explanation.
618 if (!(inode
->i_state
& I_DIRTY_ALL
))
619 list_del_init(&inode
->i_wb_list
);
620 spin_unlock(&wb
->list_lock
);
621 inode_sync_complete(inode
);
623 spin_unlock(&inode
->i_lock
);
627 static long writeback_chunk_size(struct bdi_writeback
*wb
,
628 struct wb_writeback_work
*work
)
633 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
634 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
635 * here avoids calling into writeback_inodes_wb() more than once.
637 * The intended call sequence for WB_SYNC_ALL writeback is:
640 * writeback_sb_inodes() <== called only once
641 * write_cache_pages() <== called once for each inode
642 * (quickly) tag currently dirty pages
643 * (maybe slowly) sync all tagged pages
645 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
648 pages
= min(wb
->avg_write_bandwidth
/ 2,
649 global_dirty_limit
/ DIRTY_SCOPE
);
650 pages
= min(pages
, work
->nr_pages
);
651 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
652 MIN_WRITEBACK_PAGES
);
659 * Write a portion of b_io inodes which belong to @sb.
661 * Return the number of pages and/or inodes written.
663 static long writeback_sb_inodes(struct super_block
*sb
,
664 struct bdi_writeback
*wb
,
665 struct wb_writeback_work
*work
)
667 struct writeback_control wbc
= {
668 .sync_mode
= work
->sync_mode
,
669 .tagged_writepages
= work
->tagged_writepages
,
670 .for_kupdate
= work
->for_kupdate
,
671 .for_background
= work
->for_background
,
672 .for_sync
= work
->for_sync
,
673 .range_cyclic
= work
->range_cyclic
,
675 .range_end
= LLONG_MAX
,
677 unsigned long start_time
= jiffies
;
679 long wrote
= 0; /* count both pages and inodes */
681 while (!list_empty(&wb
->b_io
)) {
682 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
684 if (inode
->i_sb
!= sb
) {
687 * We only want to write back data for this
688 * superblock, move all inodes not belonging
689 * to it back onto the dirty list.
691 redirty_tail(inode
, wb
);
696 * The inode belongs to a different superblock.
697 * Bounce back to the caller to unpin this and
698 * pin the next superblock.
704 * Don't bother with new inodes or inodes being freed, first
705 * kind does not need periodic writeout yet, and for the latter
706 * kind writeout is handled by the freer.
708 spin_lock(&inode
->i_lock
);
709 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
710 spin_unlock(&inode
->i_lock
);
711 redirty_tail(inode
, wb
);
714 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
716 * If this inode is locked for writeback and we are not
717 * doing writeback-for-data-integrity, move it to
718 * b_more_io so that writeback can proceed with the
719 * other inodes on s_io.
721 * We'll have another go at writing back this inode
722 * when we completed a full scan of b_io.
724 spin_unlock(&inode
->i_lock
);
725 requeue_io(inode
, wb
);
726 trace_writeback_sb_inodes_requeue(inode
);
729 spin_unlock(&wb
->list_lock
);
732 * We already requeued the inode if it had I_SYNC set and we
733 * are doing WB_SYNC_NONE writeback. So this catches only the
736 if (inode
->i_state
& I_SYNC
) {
737 /* Wait for I_SYNC. This function drops i_lock... */
738 inode_sleep_on_writeback(inode
);
739 /* Inode may be gone, start again */
740 spin_lock(&wb
->list_lock
);
743 inode
->i_state
|= I_SYNC
;
744 spin_unlock(&inode
->i_lock
);
746 write_chunk
= writeback_chunk_size(wb
, work
);
747 wbc
.nr_to_write
= write_chunk
;
748 wbc
.pages_skipped
= 0;
751 * We use I_SYNC to pin the inode in memory. While it is set
752 * evict_inode() will wait so the inode cannot be freed.
754 __writeback_single_inode(inode
, &wbc
);
756 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
757 wrote
+= write_chunk
- wbc
.nr_to_write
;
758 spin_lock(&wb
->list_lock
);
759 spin_lock(&inode
->i_lock
);
760 if (!(inode
->i_state
& I_DIRTY_ALL
))
762 requeue_inode(inode
, wb
, &wbc
);
763 inode_sync_complete(inode
);
764 spin_unlock(&inode
->i_lock
);
765 cond_resched_lock(&wb
->list_lock
);
767 * bail out to wb_writeback() often enough to check
768 * background threshold and other termination conditions.
771 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
773 if (work
->nr_pages
<= 0)
780 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
781 struct wb_writeback_work
*work
)
783 unsigned long start_time
= jiffies
;
786 while (!list_empty(&wb
->b_io
)) {
787 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
788 struct super_block
*sb
= inode
->i_sb
;
790 if (!trylock_super(sb
)) {
792 * trylock_super() may fail consistently due to
793 * s_umount being grabbed by someone else. Don't use
794 * requeue_io() to avoid busy retrying the inode/sb.
796 redirty_tail(inode
, wb
);
799 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
800 up_read(&sb
->s_umount
);
802 /* refer to the same tests at the end of writeback_sb_inodes */
804 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
806 if (work
->nr_pages
<= 0)
810 /* Leave any unwritten inodes on b_io */
814 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
815 enum wb_reason reason
)
817 struct wb_writeback_work work
= {
818 .nr_pages
= nr_pages
,
819 .sync_mode
= WB_SYNC_NONE
,
824 spin_lock(&wb
->list_lock
);
825 if (list_empty(&wb
->b_io
))
827 __writeback_inodes_wb(wb
, &work
);
828 spin_unlock(&wb
->list_lock
);
830 return nr_pages
- work
.nr_pages
;
833 static bool over_bground_thresh(struct bdi_writeback
*wb
)
835 unsigned long background_thresh
, dirty_thresh
;
837 global_dirty_limits(&background_thresh
, &dirty_thresh
);
839 if (global_page_state(NR_FILE_DIRTY
) +
840 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
)
843 if (wb_stat(wb
, WB_RECLAIMABLE
) > wb_dirty_limit(wb
, background_thresh
))
850 * Called under wb->list_lock. If there are multiple wb per bdi,
851 * only the flusher working on the first wb should do it.
853 static void wb_update_bandwidth(struct bdi_writeback
*wb
,
854 unsigned long start_time
)
856 __wb_update_bandwidth(wb
, 0, 0, 0, 0, 0, start_time
);
860 * Explicit flushing or periodic writeback of "old" data.
862 * Define "old": the first time one of an inode's pages is dirtied, we mark the
863 * dirtying-time in the inode's address_space. So this periodic writeback code
864 * just walks the superblock inode list, writing back any inodes which are
865 * older than a specific point in time.
867 * Try to run once per dirty_writeback_interval. But if a writeback event
868 * takes longer than a dirty_writeback_interval interval, then leave a
871 * older_than_this takes precedence over nr_to_write. So we'll only write back
872 * all dirty pages if they are all attached to "old" mappings.
874 static long wb_writeback(struct bdi_writeback
*wb
,
875 struct wb_writeback_work
*work
)
877 unsigned long wb_start
= jiffies
;
878 long nr_pages
= work
->nr_pages
;
879 unsigned long oldest_jif
;
883 oldest_jif
= jiffies
;
884 work
->older_than_this
= &oldest_jif
;
886 spin_lock(&wb
->list_lock
);
889 * Stop writeback when nr_pages has been consumed
891 if (work
->nr_pages
<= 0)
895 * Background writeout and kupdate-style writeback may
896 * run forever. Stop them if there is other work to do
897 * so that e.g. sync can proceed. They'll be restarted
898 * after the other works are all done.
900 if ((work
->for_background
|| work
->for_kupdate
) &&
901 !list_empty(&wb
->bdi
->work_list
))
905 * For background writeout, stop when we are below the
906 * background dirty threshold
908 if (work
->for_background
&& !over_bground_thresh(wb
))
912 * Kupdate and background works are special and we want to
913 * include all inodes that need writing. Livelock avoidance is
914 * handled by these works yielding to any other work so we are
917 if (work
->for_kupdate
) {
918 oldest_jif
= jiffies
-
919 msecs_to_jiffies(dirty_expire_interval
* 10);
920 } else if (work
->for_background
)
921 oldest_jif
= jiffies
;
923 trace_writeback_start(wb
->bdi
, work
);
924 if (list_empty(&wb
->b_io
))
927 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
929 progress
= __writeback_inodes_wb(wb
, work
);
930 trace_writeback_written(wb
->bdi
, work
);
932 wb_update_bandwidth(wb
, wb_start
);
935 * Did we write something? Try for more
937 * Dirty inodes are moved to b_io for writeback in batches.
938 * The completion of the current batch does not necessarily
939 * mean the overall work is done. So we keep looping as long
940 * as made some progress on cleaning pages or inodes.
945 * No more inodes for IO, bail
947 if (list_empty(&wb
->b_more_io
))
950 * Nothing written. Wait for some inode to
951 * become available for writeback. Otherwise
952 * we'll just busyloop.
954 if (!list_empty(&wb
->b_more_io
)) {
955 trace_writeback_wait(wb
->bdi
, work
);
956 inode
= wb_inode(wb
->b_more_io
.prev
);
957 spin_lock(&inode
->i_lock
);
958 spin_unlock(&wb
->list_lock
);
959 /* This function drops i_lock... */
960 inode_sleep_on_writeback(inode
);
961 spin_lock(&wb
->list_lock
);
964 spin_unlock(&wb
->list_lock
);
966 return nr_pages
- work
->nr_pages
;
970 * Return the next wb_writeback_work struct that hasn't been processed yet.
972 static struct wb_writeback_work
*
973 get_next_work_item(struct backing_dev_info
*bdi
)
975 struct wb_writeback_work
*work
= NULL
;
977 spin_lock_bh(&bdi
->wb_lock
);
978 if (!list_empty(&bdi
->work_list
)) {
979 work
= list_entry(bdi
->work_list
.next
,
980 struct wb_writeback_work
, list
);
981 list_del_init(&work
->list
);
983 spin_unlock_bh(&bdi
->wb_lock
);
988 * Add in the number of potentially dirty inodes, because each inode
989 * write can dirty pagecache in the underlying blockdev.
991 static unsigned long get_nr_dirty_pages(void)
993 return global_page_state(NR_FILE_DIRTY
) +
994 global_page_state(NR_UNSTABLE_NFS
) +
995 get_nr_dirty_inodes();
998 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1000 if (over_bground_thresh(wb
)) {
1002 struct wb_writeback_work work
= {
1003 .nr_pages
= LONG_MAX
,
1004 .sync_mode
= WB_SYNC_NONE
,
1005 .for_background
= 1,
1007 .reason
= WB_REASON_BACKGROUND
,
1010 return wb_writeback(wb
, &work
);
1016 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1018 unsigned long expired
;
1022 * When set to zero, disable periodic writeback
1024 if (!dirty_writeback_interval
)
1027 expired
= wb
->last_old_flush
+
1028 msecs_to_jiffies(dirty_writeback_interval
* 10);
1029 if (time_before(jiffies
, expired
))
1032 wb
->last_old_flush
= jiffies
;
1033 nr_pages
= get_nr_dirty_pages();
1036 struct wb_writeback_work work
= {
1037 .nr_pages
= nr_pages
,
1038 .sync_mode
= WB_SYNC_NONE
,
1041 .reason
= WB_REASON_PERIODIC
,
1044 return wb_writeback(wb
, &work
);
1051 * Retrieve work items and do the writeback they describe
1053 static long wb_do_writeback(struct bdi_writeback
*wb
)
1055 struct backing_dev_info
*bdi
= wb
->bdi
;
1056 struct wb_writeback_work
*work
;
1059 set_bit(WB_writeback_running
, &wb
->state
);
1060 while ((work
= get_next_work_item(bdi
)) != NULL
) {
1062 trace_writeback_exec(bdi
, work
);
1064 wrote
+= wb_writeback(wb
, work
);
1067 * Notify the caller of completion if this is a synchronous
1068 * work item, otherwise just free it.
1071 complete(work
->done
);
1077 * Check for periodic writeback, kupdated() style
1079 wrote
+= wb_check_old_data_flush(wb
);
1080 wrote
+= wb_check_background_flush(wb
);
1081 clear_bit(WB_writeback_running
, &wb
->state
);
1087 * Handle writeback of dirty data for the device backed by this bdi. Also
1088 * reschedules periodically and does kupdated style flushing.
1090 void bdi_writeback_workfn(struct work_struct
*work
)
1092 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1093 struct bdi_writeback
, dwork
);
1094 struct backing_dev_info
*bdi
= wb
->bdi
;
1097 set_worker_desc("flush-%s", dev_name(bdi
->dev
));
1098 current
->flags
|= PF_SWAPWRITE
;
1100 if (likely(!current_is_workqueue_rescuer() ||
1101 !test_bit(WB_registered
, &wb
->state
))) {
1103 * The normal path. Keep writing back @bdi until its
1104 * work_list is empty. Note that this path is also taken
1105 * if @bdi is shutting down even when we're running off the
1106 * rescuer as work_list needs to be drained.
1109 pages_written
= wb_do_writeback(wb
);
1110 trace_writeback_pages_written(pages_written
);
1111 } while (!list_empty(&bdi
->work_list
));
1114 * bdi_wq can't get enough workers and we're running off
1115 * the emergency worker. Don't hog it. Hopefully, 1024 is
1116 * enough for efficient IO.
1118 pages_written
= writeback_inodes_wb(&bdi
->wb
, 1024,
1119 WB_REASON_FORKER_THREAD
);
1120 trace_writeback_pages_written(pages_written
);
1123 if (!list_empty(&bdi
->work_list
))
1124 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1125 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1126 bdi_wakeup_thread_delayed(bdi
);
1128 current
->flags
&= ~PF_SWAPWRITE
;
1132 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1135 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1137 struct backing_dev_info
*bdi
;
1140 nr_pages
= get_nr_dirty_pages();
1143 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1144 if (!bdi_has_dirty_io(bdi
))
1146 __bdi_start_writeback(bdi
, nr_pages
, false, reason
);
1152 * Wake up bdi's periodically to make sure dirtytime inodes gets
1153 * written back periodically. We deliberately do *not* check the
1154 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1155 * kernel to be constantly waking up once there are any dirtytime
1156 * inodes on the system. So instead we define a separate delayed work
1157 * function which gets called much more rarely. (By default, only
1158 * once every 12 hours.)
1160 * If there is any other write activity going on in the file system,
1161 * this function won't be necessary. But if the only thing that has
1162 * happened on the file system is a dirtytime inode caused by an atime
1163 * update, we need this infrastructure below to make sure that inode
1164 * eventually gets pushed out to disk.
1166 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1167 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1169 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1171 struct backing_dev_info
*bdi
;
1174 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1175 if (list_empty(&bdi
->wb
.b_dirty_time
))
1177 bdi_wakeup_thread(bdi
);
1180 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1183 static int __init
start_dirtytime_writeback(void)
1185 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1188 __initcall(start_dirtytime_writeback
);
1190 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1191 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1195 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1196 if (ret
== 0 && write
)
1197 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1201 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1203 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1204 struct dentry
*dentry
;
1205 const char *name
= "?";
1207 dentry
= d_find_alias(inode
);
1209 spin_lock(&dentry
->d_lock
);
1210 name
= (const char *) dentry
->d_name
.name
;
1213 "%s(%d): dirtied inode %lu (%s) on %s\n",
1214 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1215 name
, inode
->i_sb
->s_id
);
1217 spin_unlock(&dentry
->d_lock
);
1224 * __mark_inode_dirty - internal function
1225 * @inode: inode to mark
1226 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1227 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1228 * mark_inode_dirty_sync.
1230 * Put the inode on the super block's dirty list.
1232 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1233 * dirty list only if it is hashed or if it refers to a blockdev.
1234 * If it was not hashed, it will never be added to the dirty list
1235 * even if it is later hashed, as it will have been marked dirty already.
1237 * In short, make sure you hash any inodes _before_ you start marking
1240 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1241 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1242 * the kernel-internal blockdev inode represents the dirtying time of the
1243 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1244 * page->mapping->host, so the page-dirtying time is recorded in the internal
1247 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1248 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1250 struct super_block
*sb
= inode
->i_sb
;
1251 struct backing_dev_info
*bdi
= NULL
;
1254 trace_writeback_mark_inode_dirty(inode
, flags
);
1257 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1258 * dirty the inode itself
1260 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
1261 trace_writeback_dirty_inode_start(inode
, flags
);
1263 if (sb
->s_op
->dirty_inode
)
1264 sb
->s_op
->dirty_inode(inode
, flags
);
1266 trace_writeback_dirty_inode(inode
, flags
);
1268 if (flags
& I_DIRTY_INODE
)
1269 flags
&= ~I_DIRTY_TIME
;
1270 dirtytime
= flags
& I_DIRTY_TIME
;
1273 * Paired with smp_mb() in __writeback_single_inode() for the
1274 * following lockless i_state test. See there for details.
1278 if (((inode
->i_state
& flags
) == flags
) ||
1279 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
1282 if (unlikely(block_dump
))
1283 block_dump___mark_inode_dirty(inode
);
1285 spin_lock(&inode
->i_lock
);
1286 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
1287 goto out_unlock_inode
;
1288 if ((inode
->i_state
& flags
) != flags
) {
1289 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1291 if (flags
& I_DIRTY_INODE
)
1292 inode
->i_state
&= ~I_DIRTY_TIME
;
1293 inode
->i_state
|= flags
;
1296 * If the inode is being synced, just update its dirty state.
1297 * The unlocker will place the inode on the appropriate
1298 * superblock list, based upon its state.
1300 if (inode
->i_state
& I_SYNC
)
1301 goto out_unlock_inode
;
1304 * Only add valid (hashed) inodes to the superblock's
1305 * dirty list. Add blockdev inodes as well.
1307 if (!S_ISBLK(inode
->i_mode
)) {
1308 if (inode_unhashed(inode
))
1309 goto out_unlock_inode
;
1311 if (inode
->i_state
& I_FREEING
)
1312 goto out_unlock_inode
;
1315 * If the inode was already on b_dirty/b_io/b_more_io, don't
1316 * reposition it (that would break b_dirty time-ordering).
1319 bool wakeup_bdi
= false;
1320 bdi
= inode_to_bdi(inode
);
1322 spin_unlock(&inode
->i_lock
);
1323 spin_lock(&bdi
->wb
.list_lock
);
1324 if (bdi_cap_writeback_dirty(bdi
)) {
1325 WARN(!test_bit(WB_registered
, &bdi
->wb
.state
),
1326 "bdi-%s not registered\n", bdi
->name
);
1329 * If this is the first dirty inode for this
1330 * bdi, we have to wake-up the corresponding
1331 * bdi thread to make sure background
1332 * write-back happens later.
1334 if (!wb_has_dirty_io(&bdi
->wb
))
1338 inode
->dirtied_when
= jiffies
;
1340 inode
->dirtied_time_when
= jiffies
;
1341 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
1342 list_move(&inode
->i_wb_list
, &bdi
->wb
.b_dirty
);
1344 list_move(&inode
->i_wb_list
,
1345 &bdi
->wb
.b_dirty_time
);
1346 spin_unlock(&bdi
->wb
.list_lock
);
1347 trace_writeback_dirty_inode_enqueue(inode
);
1350 bdi_wakeup_thread_delayed(bdi
);
1355 spin_unlock(&inode
->i_lock
);
1358 EXPORT_SYMBOL(__mark_inode_dirty
);
1360 static void wait_sb_inodes(struct super_block
*sb
)
1362 struct inode
*inode
, *old_inode
= NULL
;
1365 * We need to be protected against the filesystem going from
1366 * r/o to r/w or vice versa.
1368 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1370 spin_lock(&inode_sb_list_lock
);
1373 * Data integrity sync. Must wait for all pages under writeback,
1374 * because there may have been pages dirtied before our sync
1375 * call, but which had writeout started before we write it out.
1376 * In which case, the inode may not be on the dirty list, but
1377 * we still have to wait for that writeout.
1379 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1380 struct address_space
*mapping
= inode
->i_mapping
;
1382 spin_lock(&inode
->i_lock
);
1383 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1384 (mapping
->nrpages
== 0)) {
1385 spin_unlock(&inode
->i_lock
);
1389 spin_unlock(&inode
->i_lock
);
1390 spin_unlock(&inode_sb_list_lock
);
1393 * We hold a reference to 'inode' so it couldn't have been
1394 * removed from s_inodes list while we dropped the
1395 * inode_sb_list_lock. We cannot iput the inode now as we can
1396 * be holding the last reference and we cannot iput it under
1397 * inode_sb_list_lock. So we keep the reference and iput it
1403 filemap_fdatawait(mapping
);
1407 spin_lock(&inode_sb_list_lock
);
1409 spin_unlock(&inode_sb_list_lock
);
1414 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1415 * @sb: the superblock
1416 * @nr: the number of pages to write
1417 * @reason: reason why some writeback work initiated
1419 * Start writeback on some inodes on this super_block. No guarantees are made
1420 * on how many (if any) will be written, and this function does not wait
1421 * for IO completion of submitted IO.
1423 void writeback_inodes_sb_nr(struct super_block
*sb
,
1425 enum wb_reason reason
)
1427 DECLARE_COMPLETION_ONSTACK(done
);
1428 struct wb_writeback_work work
= {
1430 .sync_mode
= WB_SYNC_NONE
,
1431 .tagged_writepages
= 1,
1437 if (sb
->s_bdi
== &noop_backing_dev_info
)
1439 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1440 bdi_queue_work(sb
->s_bdi
, &work
);
1441 wait_for_completion(&done
);
1443 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1446 * writeback_inodes_sb - writeback dirty inodes from given super_block
1447 * @sb: the superblock
1448 * @reason: reason why some writeback work was initiated
1450 * Start writeback on some inodes on this super_block. No guarantees are made
1451 * on how many (if any) will be written, and this function does not wait
1452 * for IO completion of submitted IO.
1454 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1456 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1458 EXPORT_SYMBOL(writeback_inodes_sb
);
1461 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1462 * @sb: the superblock
1463 * @nr: the number of pages to write
1464 * @reason: the reason of writeback
1466 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1467 * Returns 1 if writeback was started, 0 if not.
1469 int try_to_writeback_inodes_sb_nr(struct super_block
*sb
,
1471 enum wb_reason reason
)
1473 if (writeback_in_progress(sb
->s_bdi
))
1476 if (!down_read_trylock(&sb
->s_umount
))
1479 writeback_inodes_sb_nr(sb
, nr
, reason
);
1480 up_read(&sb
->s_umount
);
1483 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1486 * try_to_writeback_inodes_sb - try to start writeback if none underway
1487 * @sb: the superblock
1488 * @reason: reason why some writeback work was initiated
1490 * Implement by try_to_writeback_inodes_sb_nr()
1491 * Returns 1 if writeback was started, 0 if not.
1493 int try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1495 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1497 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1500 * sync_inodes_sb - sync sb inode pages
1501 * @sb: the superblock
1503 * This function writes and waits on any dirty inode belonging to this
1506 void sync_inodes_sb(struct super_block
*sb
)
1508 DECLARE_COMPLETION_ONSTACK(done
);
1509 struct wb_writeback_work work
= {
1511 .sync_mode
= WB_SYNC_ALL
,
1512 .nr_pages
= LONG_MAX
,
1515 .reason
= WB_REASON_SYNC
,
1519 /* Nothing to do? */
1520 if (sb
->s_bdi
== &noop_backing_dev_info
)
1522 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1524 bdi_queue_work(sb
->s_bdi
, &work
);
1525 wait_for_completion(&done
);
1529 EXPORT_SYMBOL(sync_inodes_sb
);
1532 * write_inode_now - write an inode to disk
1533 * @inode: inode to write to disk
1534 * @sync: whether the write should be synchronous or not
1536 * This function commits an inode to disk immediately if it is dirty. This is
1537 * primarily needed by knfsd.
1539 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1541 int write_inode_now(struct inode
*inode
, int sync
)
1543 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1544 struct writeback_control wbc
= {
1545 .nr_to_write
= LONG_MAX
,
1546 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1548 .range_end
= LLONG_MAX
,
1551 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1552 wbc
.nr_to_write
= 0;
1555 return writeback_single_inode(inode
, wb
, &wbc
);
1557 EXPORT_SYMBOL(write_inode_now
);
1560 * sync_inode - write an inode and its pages to disk.
1561 * @inode: the inode to sync
1562 * @wbc: controls the writeback mode
1564 * sync_inode() will write an inode and its pages to disk. It will also
1565 * correctly update the inode on its superblock's dirty inode lists and will
1566 * update inode->i_state.
1568 * The caller must have a ref on the inode.
1570 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1572 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1574 EXPORT_SYMBOL(sync_inode
);
1577 * sync_inode_metadata - write an inode to disk
1578 * @inode: the inode to sync
1579 * @wait: wait for I/O to complete.
1581 * Write an inode to disk and adjust its dirty state after completion.
1583 * Note: only writes the actual inode, no associated data or other metadata.
1585 int sync_inode_metadata(struct inode
*inode
, int wait
)
1587 struct writeback_control wbc
= {
1588 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1589 .nr_to_write
= 0, /* metadata-only */
1592 return sync_inode(inode
, &wbc
);
1594 EXPORT_SYMBOL(sync_inode_metadata
);