4 * Copyright (C) 2002, Linus Torvalds.
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the
9 * inode itself is not handled here.
11 * 10Apr2002 Andrew Morton
12 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/spinlock.h>
19 #include <linux/slab.h>
20 #include <linux/sched.h>
23 #include <linux/pagemap.h>
24 #include <linux/kthread.h>
25 #include <linux/writeback.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/tracepoint.h>
29 #include <linux/device.h>
33 * 4MB minimal write chunk size
35 #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
38 * Passed into wb_writeback(), essentially a subset of writeback_control
40 struct wb_writeback_work
{
42 struct super_block
*sb
;
43 unsigned long *older_than_this
;
44 enum writeback_sync_modes sync_mode
;
45 unsigned int tagged_writepages
:1;
46 unsigned int for_kupdate
:1;
47 unsigned int range_cyclic
:1;
48 unsigned int for_background
:1;
49 unsigned int for_sync
:1; /* sync(2) WB_SYNC_ALL writeback */
50 enum wb_reason reason
; /* why was writeback initiated? */
52 struct list_head list
; /* pending work list */
53 struct completion
*done
; /* set if the caller waits */
57 * If an inode is constantly having its pages dirtied, but then the
58 * updates stop dirtytime_expire_interval seconds in the past, it's
59 * possible for the worst case time between when an inode has its
60 * timestamps updated and when they finally get written out to be two
61 * dirtytime_expire_intervals. We set the default to 12 hours (in
62 * seconds), which means most of the time inodes will have their
63 * timestamps written to disk after 12 hours, but in the worst case a
64 * few inodes might not their timestamps updated for 24 hours.
66 unsigned int dirtytime_expire_interval
= 12 * 60 * 60;
69 * writeback_in_progress - determine whether there is writeback in progress
70 * @bdi: the device's backing_dev_info structure.
72 * Determine whether there is writeback waiting to be handled against a
75 int writeback_in_progress(struct backing_dev_info
*bdi
)
77 return test_bit(WB_writeback_running
, &bdi
->wb
.state
);
79 EXPORT_SYMBOL(writeback_in_progress
);
81 static inline struct inode
*wb_inode(struct list_head
*head
)
83 return list_entry(head
, struct inode
, i_wb_list
);
87 * Include the creation of the trace points after defining the
88 * wb_writeback_work structure and inline functions so that the definition
89 * remains local to this file.
91 #define CREATE_TRACE_POINTS
92 #include <trace/events/writeback.h>
94 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage
);
96 static bool wb_io_lists_populated(struct bdi_writeback
*wb
)
98 if (wb_has_dirty_io(wb
)) {
101 set_bit(WB_has_dirty_io
, &wb
->state
);
102 atomic_long_add(wb
->avg_write_bandwidth
,
103 &wb
->bdi
->tot_write_bandwidth
);
108 static void wb_io_lists_depopulated(struct bdi_writeback
*wb
)
110 if (wb_has_dirty_io(wb
) && list_empty(&wb
->b_dirty
) &&
111 list_empty(&wb
->b_io
) && list_empty(&wb
->b_more_io
)) {
112 clear_bit(WB_has_dirty_io
, &wb
->state
);
113 atomic_long_sub(wb
->avg_write_bandwidth
,
114 &wb
->bdi
->tot_write_bandwidth
);
119 * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list
120 * @inode: inode to be moved
121 * @wb: target bdi_writeback
122 * @head: one of @wb->b_{dirty|io|more_io}
124 * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io.
125 * Returns %true if @inode is the first occupant of the !dirty_time IO
126 * lists; otherwise, %false.
128 static bool inode_wb_list_move_locked(struct inode
*inode
,
129 struct bdi_writeback
*wb
,
130 struct list_head
*head
)
132 assert_spin_locked(&wb
->list_lock
);
134 list_move(&inode
->i_wb_list
, head
);
136 /* dirty_time doesn't count as dirty_io until expiration */
137 if (head
!= &wb
->b_dirty_time
)
138 return wb_io_lists_populated(wb
);
140 wb_io_lists_depopulated(wb
);
145 * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list
146 * @inode: inode to be removed
147 * @wb: bdi_writeback @inode is being removed from
149 * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and
150 * clear %WB_has_dirty_io if all are empty afterwards.
152 static void inode_wb_list_del_locked(struct inode
*inode
,
153 struct bdi_writeback
*wb
)
155 assert_spin_locked(&wb
->list_lock
);
157 list_del_init(&inode
->i_wb_list
);
158 wb_io_lists_depopulated(wb
);
161 static void wb_wakeup(struct bdi_writeback
*wb
)
163 spin_lock_bh(&wb
->work_lock
);
164 if (test_bit(WB_registered
, &wb
->state
))
165 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
166 spin_unlock_bh(&wb
->work_lock
);
169 static void wb_queue_work(struct bdi_writeback
*wb
,
170 struct wb_writeback_work
*work
)
172 trace_writeback_queue(wb
->bdi
, work
);
174 spin_lock_bh(&wb
->work_lock
);
175 if (!test_bit(WB_registered
, &wb
->state
)) {
177 complete(work
->done
);
180 list_add_tail(&work
->list
, &wb
->work_list
);
181 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
183 spin_unlock_bh(&wb
->work_lock
);
186 static void __wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
187 bool range_cyclic
, enum wb_reason reason
)
189 struct wb_writeback_work
*work
;
192 * This is WB_SYNC_NONE writeback, so if allocation fails just
193 * wakeup the thread for old dirty data writeback
195 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
197 trace_writeback_nowork(wb
->bdi
);
202 work
->sync_mode
= WB_SYNC_NONE
;
203 work
->nr_pages
= nr_pages
;
204 work
->range_cyclic
= range_cyclic
;
205 work
->reason
= reason
;
207 wb_queue_work(wb
, work
);
210 #ifdef CONFIG_CGROUP_WRITEBACK
213 * inode_congested - test whether an inode is congested
214 * @inode: inode to test for congestion
215 * @cong_bits: mask of WB_[a]sync_congested bits to test
217 * Tests whether @inode is congested. @cong_bits is the mask of congestion
218 * bits to test and the return value is the mask of set bits.
220 * If cgroup writeback is enabled for @inode, the congestion state is
221 * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
222 * associated with @inode is congested; otherwise, the root wb's congestion
225 int inode_congested(struct inode
*inode
, int cong_bits
)
228 struct bdi_writeback
*wb
= inode_to_wb(inode
);
230 return wb_congested(wb
, cong_bits
);
233 return wb_congested(&inode_to_bdi(inode
)->wb
, cong_bits
);
235 EXPORT_SYMBOL_GPL(inode_congested
);
237 #endif /* CONFIG_CGROUP_WRITEBACK */
240 * bdi_start_writeback - start writeback
241 * @bdi: the backing device to write from
242 * @nr_pages: the number of pages to write
243 * @reason: reason why some writeback work was initiated
246 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
247 * started when this function returns, we make no guarantees on
248 * completion. Caller need not hold sb s_umount semaphore.
251 void bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
252 enum wb_reason reason
)
254 __wb_start_writeback(&bdi
->wb
, nr_pages
, true, reason
);
258 * bdi_start_background_writeback - start background writeback
259 * @bdi: the backing device to write from
262 * This makes sure WB_SYNC_NONE background writeback happens. When
263 * this function returns, it is only guaranteed that for given BDI
264 * some IO is happening if we are over background dirty threshold.
265 * Caller need not hold sb s_umount semaphore.
267 void bdi_start_background_writeback(struct backing_dev_info
*bdi
)
270 * We just wake up the flusher thread. It will perform background
271 * writeback as soon as there is no other work to do.
273 trace_writeback_wake_background(bdi
);
278 * Remove the inode from the writeback list it is on.
280 void inode_wb_list_del(struct inode
*inode
)
282 struct bdi_writeback
*wb
= inode_to_wb(inode
);
284 spin_lock(&wb
->list_lock
);
285 inode_wb_list_del_locked(inode
, wb
);
286 spin_unlock(&wb
->list_lock
);
290 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
291 * furthest end of its superblock's dirty-inode list.
293 * Before stamping the inode's ->dirtied_when, we check to see whether it is
294 * already the most-recently-dirtied inode on the b_dirty list. If that is
295 * the case then the inode must have been redirtied while it was being written
296 * out and we don't reset its dirtied_when.
298 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
300 if (!list_empty(&wb
->b_dirty
)) {
303 tail
= wb_inode(wb
->b_dirty
.next
);
304 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
305 inode
->dirtied_when
= jiffies
;
307 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty
);
311 * requeue inode for re-scanning after bdi->b_io list is exhausted.
313 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
315 inode_wb_list_move_locked(inode
, wb
, &wb
->b_more_io
);
318 static void inode_sync_complete(struct inode
*inode
)
320 inode
->i_state
&= ~I_SYNC
;
321 /* If inode is clean an unused, put it into LRU now... */
322 inode_add_lru(inode
);
323 /* Waiters must see I_SYNC cleared before being woken up */
325 wake_up_bit(&inode
->i_state
, __I_SYNC
);
328 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
330 bool ret
= time_after(inode
->dirtied_when
, t
);
333 * For inodes being constantly redirtied, dirtied_when can get stuck.
334 * It _appears_ to be in the future, but is actually in distant past.
335 * This test is necessary to prevent such wrapped-around relative times
336 * from permanently stopping the whole bdi writeback.
338 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
343 #define EXPIRE_DIRTY_ATIME 0x0001
346 * Move expired (dirtied before work->older_than_this) dirty inodes from
347 * @delaying_queue to @dispatch_queue.
349 static int move_expired_inodes(struct list_head
*delaying_queue
,
350 struct list_head
*dispatch_queue
,
352 struct wb_writeback_work
*work
)
354 unsigned long *older_than_this
= NULL
;
355 unsigned long expire_time
;
357 struct list_head
*pos
, *node
;
358 struct super_block
*sb
= NULL
;
363 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
364 older_than_this
= work
->older_than_this
;
365 else if (!work
->for_sync
) {
366 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
367 older_than_this
= &expire_time
;
369 while (!list_empty(delaying_queue
)) {
370 inode
= wb_inode(delaying_queue
->prev
);
371 if (older_than_this
&&
372 inode_dirtied_after(inode
, *older_than_this
))
374 list_move(&inode
->i_wb_list
, &tmp
);
376 if (flags
& EXPIRE_DIRTY_ATIME
)
377 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
378 if (sb_is_blkdev_sb(inode
->i_sb
))
380 if (sb
&& sb
!= inode
->i_sb
)
385 /* just one sb in list, splice to dispatch_queue and we're done */
387 list_splice(&tmp
, dispatch_queue
);
391 /* Move inodes from one superblock together */
392 while (!list_empty(&tmp
)) {
393 sb
= wb_inode(tmp
.prev
)->i_sb
;
394 list_for_each_prev_safe(pos
, node
, &tmp
) {
395 inode
= wb_inode(pos
);
396 if (inode
->i_sb
== sb
)
397 list_move(&inode
->i_wb_list
, dispatch_queue
);
405 * Queue all expired dirty inodes for io, eldest first.
407 * newly dirtied b_dirty b_io b_more_io
408 * =============> gf edc BA
410 * newly dirtied b_dirty b_io b_more_io
411 * =============> g fBAedc
413 * +--> dequeue for IO
415 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
419 assert_spin_locked(&wb
->list_lock
);
420 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
421 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
422 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
423 EXPIRE_DIRTY_ATIME
, work
);
425 wb_io_lists_populated(wb
);
426 trace_writeback_queue_io(wb
, work
, moved
);
429 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
433 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
434 trace_writeback_write_inode_start(inode
, wbc
);
435 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
436 trace_writeback_write_inode(inode
, wbc
);
443 * Wait for writeback on an inode to complete. Called with i_lock held.
444 * Caller must make sure inode cannot go away when we drop i_lock.
446 static void __inode_wait_for_writeback(struct inode
*inode
)
447 __releases(inode
->i_lock
)
448 __acquires(inode
->i_lock
)
450 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
451 wait_queue_head_t
*wqh
;
453 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
454 while (inode
->i_state
& I_SYNC
) {
455 spin_unlock(&inode
->i_lock
);
456 __wait_on_bit(wqh
, &wq
, bit_wait
,
457 TASK_UNINTERRUPTIBLE
);
458 spin_lock(&inode
->i_lock
);
463 * Wait for writeback on an inode to complete. Caller must have inode pinned.
465 void inode_wait_for_writeback(struct inode
*inode
)
467 spin_lock(&inode
->i_lock
);
468 __inode_wait_for_writeback(inode
);
469 spin_unlock(&inode
->i_lock
);
473 * Sleep until I_SYNC is cleared. This function must be called with i_lock
474 * held and drops it. It is aimed for callers not holding any inode reference
475 * so once i_lock is dropped, inode can go away.
477 static void inode_sleep_on_writeback(struct inode
*inode
)
478 __releases(inode
->i_lock
)
481 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
484 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
485 sleep
= inode
->i_state
& I_SYNC
;
486 spin_unlock(&inode
->i_lock
);
489 finish_wait(wqh
, &wait
);
493 * Find proper writeback list for the inode depending on its current state and
494 * possibly also change of its state while we were doing writeback. Here we
495 * handle things such as livelock prevention or fairness of writeback among
496 * inodes. This function can be called only by flusher thread - noone else
497 * processes all inodes in writeback lists and requeueing inodes behind flusher
498 * thread's back can have unexpected consequences.
500 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
501 struct writeback_control
*wbc
)
503 if (inode
->i_state
& I_FREEING
)
507 * Sync livelock prevention. Each inode is tagged and synced in one
508 * shot. If still dirty, it will be redirty_tail()'ed below. Update
509 * the dirty time to prevent enqueue and sync it again.
511 if ((inode
->i_state
& I_DIRTY
) &&
512 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
513 inode
->dirtied_when
= jiffies
;
515 if (wbc
->pages_skipped
) {
517 * writeback is not making progress due to locked
518 * buffers. Skip this inode for now.
520 redirty_tail(inode
, wb
);
524 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
526 * We didn't write back all the pages. nfs_writepages()
527 * sometimes bales out without doing anything.
529 if (wbc
->nr_to_write
<= 0) {
530 /* Slice used up. Queue for next turn. */
531 requeue_io(inode
, wb
);
534 * Writeback blocked by something other than
535 * congestion. Delay the inode for some time to
536 * avoid spinning on the CPU (100% iowait)
537 * retrying writeback of the dirty page/inode
538 * that cannot be performed immediately.
540 redirty_tail(inode
, wb
);
542 } else if (inode
->i_state
& I_DIRTY
) {
544 * Filesystems can dirty the inode during writeback operations,
545 * such as delayed allocation during submission or metadata
546 * updates after data IO completion.
548 redirty_tail(inode
, wb
);
549 } else if (inode
->i_state
& I_DIRTY_TIME
) {
550 inode
->dirtied_when
= jiffies
;
551 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty_time
);
553 /* The inode is clean. Remove from writeback lists. */
554 inode_wb_list_del_locked(inode
, wb
);
559 * Write out an inode and its dirty pages. Do not update the writeback list
560 * linkage. That is left to the caller. The caller is also responsible for
561 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
564 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
566 struct address_space
*mapping
= inode
->i_mapping
;
567 long nr_to_write
= wbc
->nr_to_write
;
571 WARN_ON(!(inode
->i_state
& I_SYNC
));
573 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
575 ret
= do_writepages(mapping
, wbc
);
578 * Make sure to wait on the data before writing out the metadata.
579 * This is important for filesystems that modify metadata on data
580 * I/O completion. We don't do it for sync(2) writeback because it has a
581 * separate, external IO completion path and ->sync_fs for guaranteeing
582 * inode metadata is written back correctly.
584 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
585 int err
= filemap_fdatawait(mapping
);
591 * Some filesystems may redirty the inode during the writeback
592 * due to delalloc, clear dirty metadata flags right before
595 spin_lock(&inode
->i_lock
);
597 dirty
= inode
->i_state
& I_DIRTY
;
598 if (inode
->i_state
& I_DIRTY_TIME
) {
599 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
600 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
601 unlikely(time_after(jiffies
,
602 (inode
->dirtied_time_when
+
603 dirtytime_expire_interval
* HZ
)))) {
604 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
605 trace_writeback_lazytime(inode
);
608 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
609 inode
->i_state
&= ~dirty
;
612 * Paired with smp_mb() in __mark_inode_dirty(). This allows
613 * __mark_inode_dirty() to test i_state without grabbing i_lock -
614 * either they see the I_DIRTY bits cleared or we see the dirtied
617 * I_DIRTY_PAGES is always cleared together above even if @mapping
618 * still has dirty pages. The flag is reinstated after smp_mb() if
619 * necessary. This guarantees that either __mark_inode_dirty()
620 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
624 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
625 inode
->i_state
|= I_DIRTY_PAGES
;
627 spin_unlock(&inode
->i_lock
);
629 if (dirty
& I_DIRTY_TIME
)
630 mark_inode_dirty_sync(inode
);
631 /* Don't write the inode if only I_DIRTY_PAGES was set */
632 if (dirty
& ~I_DIRTY_PAGES
) {
633 int err
= write_inode(inode
, wbc
);
637 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
642 * Write out an inode's dirty pages. Either the caller has an active reference
643 * on the inode or the inode has I_WILL_FREE set.
645 * This function is designed to be called for writing back one inode which
646 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
647 * and does more profound writeback list handling in writeback_sb_inodes().
650 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
651 struct writeback_control
*wbc
)
655 spin_lock(&inode
->i_lock
);
656 if (!atomic_read(&inode
->i_count
))
657 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
659 WARN_ON(inode
->i_state
& I_WILL_FREE
);
661 if (inode
->i_state
& I_SYNC
) {
662 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
665 * It's a data-integrity sync. We must wait. Since callers hold
666 * inode reference or inode has I_WILL_FREE set, it cannot go
669 __inode_wait_for_writeback(inode
);
671 WARN_ON(inode
->i_state
& I_SYNC
);
673 * Skip inode if it is clean and we have no outstanding writeback in
674 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
675 * function since flusher thread may be doing for example sync in
676 * parallel and if we move the inode, it could get skipped. So here we
677 * make sure inode is on some writeback list and leave it there unless
678 * we have completely cleaned the inode.
680 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
681 (wbc
->sync_mode
!= WB_SYNC_ALL
||
682 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
684 inode
->i_state
|= I_SYNC
;
685 spin_unlock(&inode
->i_lock
);
687 ret
= __writeback_single_inode(inode
, wbc
);
689 spin_lock(&wb
->list_lock
);
690 spin_lock(&inode
->i_lock
);
692 * If inode is clean, remove it from writeback lists. Otherwise don't
693 * touch it. See comment above for explanation.
695 if (!(inode
->i_state
& I_DIRTY_ALL
))
696 inode_wb_list_del_locked(inode
, wb
);
697 spin_unlock(&wb
->list_lock
);
698 inode_sync_complete(inode
);
700 spin_unlock(&inode
->i_lock
);
704 static long writeback_chunk_size(struct bdi_writeback
*wb
,
705 struct wb_writeback_work
*work
)
710 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
711 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
712 * here avoids calling into writeback_inodes_wb() more than once.
714 * The intended call sequence for WB_SYNC_ALL writeback is:
717 * writeback_sb_inodes() <== called only once
718 * write_cache_pages() <== called once for each inode
719 * (quickly) tag currently dirty pages
720 * (maybe slowly) sync all tagged pages
722 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
725 pages
= min(wb
->avg_write_bandwidth
/ 2,
726 global_dirty_limit
/ DIRTY_SCOPE
);
727 pages
= min(pages
, work
->nr_pages
);
728 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
729 MIN_WRITEBACK_PAGES
);
736 * Write a portion of b_io inodes which belong to @sb.
738 * Return the number of pages and/or inodes written.
740 static long writeback_sb_inodes(struct super_block
*sb
,
741 struct bdi_writeback
*wb
,
742 struct wb_writeback_work
*work
)
744 struct writeback_control wbc
= {
745 .sync_mode
= work
->sync_mode
,
746 .tagged_writepages
= work
->tagged_writepages
,
747 .for_kupdate
= work
->for_kupdate
,
748 .for_background
= work
->for_background
,
749 .for_sync
= work
->for_sync
,
750 .range_cyclic
= work
->range_cyclic
,
752 .range_end
= LLONG_MAX
,
754 unsigned long start_time
= jiffies
;
756 long wrote
= 0; /* count both pages and inodes */
758 while (!list_empty(&wb
->b_io
)) {
759 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
761 if (inode
->i_sb
!= sb
) {
764 * We only want to write back data for this
765 * superblock, move all inodes not belonging
766 * to it back onto the dirty list.
768 redirty_tail(inode
, wb
);
773 * The inode belongs to a different superblock.
774 * Bounce back to the caller to unpin this and
775 * pin the next superblock.
781 * Don't bother with new inodes or inodes being freed, first
782 * kind does not need periodic writeout yet, and for the latter
783 * kind writeout is handled by the freer.
785 spin_lock(&inode
->i_lock
);
786 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
787 spin_unlock(&inode
->i_lock
);
788 redirty_tail(inode
, wb
);
791 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
793 * If this inode is locked for writeback and we are not
794 * doing writeback-for-data-integrity, move it to
795 * b_more_io so that writeback can proceed with the
796 * other inodes on s_io.
798 * We'll have another go at writing back this inode
799 * when we completed a full scan of b_io.
801 spin_unlock(&inode
->i_lock
);
802 requeue_io(inode
, wb
);
803 trace_writeback_sb_inodes_requeue(inode
);
806 spin_unlock(&wb
->list_lock
);
809 * We already requeued the inode if it had I_SYNC set and we
810 * are doing WB_SYNC_NONE writeback. So this catches only the
813 if (inode
->i_state
& I_SYNC
) {
814 /* Wait for I_SYNC. This function drops i_lock... */
815 inode_sleep_on_writeback(inode
);
816 /* Inode may be gone, start again */
817 spin_lock(&wb
->list_lock
);
820 inode
->i_state
|= I_SYNC
;
821 spin_unlock(&inode
->i_lock
);
823 write_chunk
= writeback_chunk_size(wb
, work
);
824 wbc
.nr_to_write
= write_chunk
;
825 wbc
.pages_skipped
= 0;
828 * We use I_SYNC to pin the inode in memory. While it is set
829 * evict_inode() will wait so the inode cannot be freed.
831 __writeback_single_inode(inode
, &wbc
);
833 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
834 wrote
+= write_chunk
- wbc
.nr_to_write
;
835 spin_lock(&wb
->list_lock
);
836 spin_lock(&inode
->i_lock
);
837 if (!(inode
->i_state
& I_DIRTY_ALL
))
839 requeue_inode(inode
, wb
, &wbc
);
840 inode_sync_complete(inode
);
841 spin_unlock(&inode
->i_lock
);
842 cond_resched_lock(&wb
->list_lock
);
844 * bail out to wb_writeback() often enough to check
845 * background threshold and other termination conditions.
848 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
850 if (work
->nr_pages
<= 0)
857 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
858 struct wb_writeback_work
*work
)
860 unsigned long start_time
= jiffies
;
863 while (!list_empty(&wb
->b_io
)) {
864 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
865 struct super_block
*sb
= inode
->i_sb
;
867 if (!trylock_super(sb
)) {
869 * trylock_super() may fail consistently due to
870 * s_umount being grabbed by someone else. Don't use
871 * requeue_io() to avoid busy retrying the inode/sb.
873 redirty_tail(inode
, wb
);
876 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
877 up_read(&sb
->s_umount
);
879 /* refer to the same tests at the end of writeback_sb_inodes */
881 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
883 if (work
->nr_pages
<= 0)
887 /* Leave any unwritten inodes on b_io */
891 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
892 enum wb_reason reason
)
894 struct wb_writeback_work work
= {
895 .nr_pages
= nr_pages
,
896 .sync_mode
= WB_SYNC_NONE
,
901 spin_lock(&wb
->list_lock
);
902 if (list_empty(&wb
->b_io
))
904 __writeback_inodes_wb(wb
, &work
);
905 spin_unlock(&wb
->list_lock
);
907 return nr_pages
- work
.nr_pages
;
910 static bool over_bground_thresh(struct bdi_writeback
*wb
)
912 unsigned long background_thresh
, dirty_thresh
;
914 global_dirty_limits(&background_thresh
, &dirty_thresh
);
916 if (global_page_state(NR_FILE_DIRTY
) +
917 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
)
920 if (wb_stat(wb
, WB_RECLAIMABLE
) > wb_dirty_limit(wb
, background_thresh
))
927 * Called under wb->list_lock. If there are multiple wb per bdi,
928 * only the flusher working on the first wb should do it.
930 static void wb_update_bandwidth(struct bdi_writeback
*wb
,
931 unsigned long start_time
)
933 __wb_update_bandwidth(wb
, 0, 0, 0, 0, 0, start_time
);
937 * Explicit flushing or periodic writeback of "old" data.
939 * Define "old": the first time one of an inode's pages is dirtied, we mark the
940 * dirtying-time in the inode's address_space. So this periodic writeback code
941 * just walks the superblock inode list, writing back any inodes which are
942 * older than a specific point in time.
944 * Try to run once per dirty_writeback_interval. But if a writeback event
945 * takes longer than a dirty_writeback_interval interval, then leave a
948 * older_than_this takes precedence over nr_to_write. So we'll only write back
949 * all dirty pages if they are all attached to "old" mappings.
951 static long wb_writeback(struct bdi_writeback
*wb
,
952 struct wb_writeback_work
*work
)
954 unsigned long wb_start
= jiffies
;
955 long nr_pages
= work
->nr_pages
;
956 unsigned long oldest_jif
;
960 oldest_jif
= jiffies
;
961 work
->older_than_this
= &oldest_jif
;
963 spin_lock(&wb
->list_lock
);
966 * Stop writeback when nr_pages has been consumed
968 if (work
->nr_pages
<= 0)
972 * Background writeout and kupdate-style writeback may
973 * run forever. Stop them if there is other work to do
974 * so that e.g. sync can proceed. They'll be restarted
975 * after the other works are all done.
977 if ((work
->for_background
|| work
->for_kupdate
) &&
978 !list_empty(&wb
->work_list
))
982 * For background writeout, stop when we are below the
983 * background dirty threshold
985 if (work
->for_background
&& !over_bground_thresh(wb
))
989 * Kupdate and background works are special and we want to
990 * include all inodes that need writing. Livelock avoidance is
991 * handled by these works yielding to any other work so we are
994 if (work
->for_kupdate
) {
995 oldest_jif
= jiffies
-
996 msecs_to_jiffies(dirty_expire_interval
* 10);
997 } else if (work
->for_background
)
998 oldest_jif
= jiffies
;
1000 trace_writeback_start(wb
->bdi
, work
);
1001 if (list_empty(&wb
->b_io
))
1004 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
1006 progress
= __writeback_inodes_wb(wb
, work
);
1007 trace_writeback_written(wb
->bdi
, work
);
1009 wb_update_bandwidth(wb
, wb_start
);
1012 * Did we write something? Try for more
1014 * Dirty inodes are moved to b_io for writeback in batches.
1015 * The completion of the current batch does not necessarily
1016 * mean the overall work is done. So we keep looping as long
1017 * as made some progress on cleaning pages or inodes.
1022 * No more inodes for IO, bail
1024 if (list_empty(&wb
->b_more_io
))
1027 * Nothing written. Wait for some inode to
1028 * become available for writeback. Otherwise
1029 * we'll just busyloop.
1031 if (!list_empty(&wb
->b_more_io
)) {
1032 trace_writeback_wait(wb
->bdi
, work
);
1033 inode
= wb_inode(wb
->b_more_io
.prev
);
1034 spin_lock(&inode
->i_lock
);
1035 spin_unlock(&wb
->list_lock
);
1036 /* This function drops i_lock... */
1037 inode_sleep_on_writeback(inode
);
1038 spin_lock(&wb
->list_lock
);
1041 spin_unlock(&wb
->list_lock
);
1043 return nr_pages
- work
->nr_pages
;
1047 * Return the next wb_writeback_work struct that hasn't been processed yet.
1049 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
1051 struct wb_writeback_work
*work
= NULL
;
1053 spin_lock_bh(&wb
->work_lock
);
1054 if (!list_empty(&wb
->work_list
)) {
1055 work
= list_entry(wb
->work_list
.next
,
1056 struct wb_writeback_work
, list
);
1057 list_del_init(&work
->list
);
1059 spin_unlock_bh(&wb
->work_lock
);
1064 * Add in the number of potentially dirty inodes, because each inode
1065 * write can dirty pagecache in the underlying blockdev.
1067 static unsigned long get_nr_dirty_pages(void)
1069 return global_page_state(NR_FILE_DIRTY
) +
1070 global_page_state(NR_UNSTABLE_NFS
) +
1071 get_nr_dirty_inodes();
1074 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1076 if (over_bground_thresh(wb
)) {
1078 struct wb_writeback_work work
= {
1079 .nr_pages
= LONG_MAX
,
1080 .sync_mode
= WB_SYNC_NONE
,
1081 .for_background
= 1,
1083 .reason
= WB_REASON_BACKGROUND
,
1086 return wb_writeback(wb
, &work
);
1092 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1094 unsigned long expired
;
1098 * When set to zero, disable periodic writeback
1100 if (!dirty_writeback_interval
)
1103 expired
= wb
->last_old_flush
+
1104 msecs_to_jiffies(dirty_writeback_interval
* 10);
1105 if (time_before(jiffies
, expired
))
1108 wb
->last_old_flush
= jiffies
;
1109 nr_pages
= get_nr_dirty_pages();
1112 struct wb_writeback_work work
= {
1113 .nr_pages
= nr_pages
,
1114 .sync_mode
= WB_SYNC_NONE
,
1117 .reason
= WB_REASON_PERIODIC
,
1120 return wb_writeback(wb
, &work
);
1127 * Retrieve work items and do the writeback they describe
1129 static long wb_do_writeback(struct bdi_writeback
*wb
)
1131 struct wb_writeback_work
*work
;
1134 set_bit(WB_writeback_running
, &wb
->state
);
1135 while ((work
= get_next_work_item(wb
)) != NULL
) {
1137 trace_writeback_exec(wb
->bdi
, work
);
1139 wrote
+= wb_writeback(wb
, work
);
1142 * Notify the caller of completion if this is a synchronous
1143 * work item, otherwise just free it.
1146 complete(work
->done
);
1152 * Check for periodic writeback, kupdated() style
1154 wrote
+= wb_check_old_data_flush(wb
);
1155 wrote
+= wb_check_background_flush(wb
);
1156 clear_bit(WB_writeback_running
, &wb
->state
);
1162 * Handle writeback of dirty data for the device backed by this bdi. Also
1163 * reschedules periodically and does kupdated style flushing.
1165 void wb_workfn(struct work_struct
*work
)
1167 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1168 struct bdi_writeback
, dwork
);
1171 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1172 current
->flags
|= PF_SWAPWRITE
;
1174 if (likely(!current_is_workqueue_rescuer() ||
1175 !test_bit(WB_registered
, &wb
->state
))) {
1177 * The normal path. Keep writing back @wb until its
1178 * work_list is empty. Note that this path is also taken
1179 * if @wb is shutting down even when we're running off the
1180 * rescuer as work_list needs to be drained.
1183 pages_written
= wb_do_writeback(wb
);
1184 trace_writeback_pages_written(pages_written
);
1185 } while (!list_empty(&wb
->work_list
));
1188 * bdi_wq can't get enough workers and we're running off
1189 * the emergency worker. Don't hog it. Hopefully, 1024 is
1190 * enough for efficient IO.
1192 pages_written
= writeback_inodes_wb(wb
, 1024,
1193 WB_REASON_FORKER_THREAD
);
1194 trace_writeback_pages_written(pages_written
);
1197 if (!list_empty(&wb
->work_list
))
1198 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1199 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1200 wb_wakeup_delayed(wb
);
1202 current
->flags
&= ~PF_SWAPWRITE
;
1206 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1209 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1211 struct backing_dev_info
*bdi
;
1214 nr_pages
= get_nr_dirty_pages();
1217 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1218 if (!bdi_has_dirty_io(bdi
))
1220 __wb_start_writeback(&bdi
->wb
, nr_pages
, false, reason
);
1226 * Wake up bdi's periodically to make sure dirtytime inodes gets
1227 * written back periodically. We deliberately do *not* check the
1228 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1229 * kernel to be constantly waking up once there are any dirtytime
1230 * inodes on the system. So instead we define a separate delayed work
1231 * function which gets called much more rarely. (By default, only
1232 * once every 12 hours.)
1234 * If there is any other write activity going on in the file system,
1235 * this function won't be necessary. But if the only thing that has
1236 * happened on the file system is a dirtytime inode caused by an atime
1237 * update, we need this infrastructure below to make sure that inode
1238 * eventually gets pushed out to disk.
1240 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1241 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1243 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1245 struct backing_dev_info
*bdi
;
1248 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1249 if (list_empty(&bdi
->wb
.b_dirty_time
))
1251 wb_wakeup(&bdi
->wb
);
1254 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1257 static int __init
start_dirtytime_writeback(void)
1259 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1262 __initcall(start_dirtytime_writeback
);
1264 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1265 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1269 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1270 if (ret
== 0 && write
)
1271 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1275 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1277 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1278 struct dentry
*dentry
;
1279 const char *name
= "?";
1281 dentry
= d_find_alias(inode
);
1283 spin_lock(&dentry
->d_lock
);
1284 name
= (const char *) dentry
->d_name
.name
;
1287 "%s(%d): dirtied inode %lu (%s) on %s\n",
1288 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1289 name
, inode
->i_sb
->s_id
);
1291 spin_unlock(&dentry
->d_lock
);
1298 * __mark_inode_dirty - internal function
1299 * @inode: inode to mark
1300 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1301 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1302 * mark_inode_dirty_sync.
1304 * Put the inode on the super block's dirty list.
1306 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1307 * dirty list only if it is hashed or if it refers to a blockdev.
1308 * If it was not hashed, it will never be added to the dirty list
1309 * even if it is later hashed, as it will have been marked dirty already.
1311 * In short, make sure you hash any inodes _before_ you start marking
1314 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1315 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1316 * the kernel-internal blockdev inode represents the dirtying time of the
1317 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1318 * page->mapping->host, so the page-dirtying time is recorded in the internal
1321 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1322 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1324 struct super_block
*sb
= inode
->i_sb
;
1325 struct backing_dev_info
*bdi
= NULL
;
1328 trace_writeback_mark_inode_dirty(inode
, flags
);
1331 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1332 * dirty the inode itself
1334 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
1335 trace_writeback_dirty_inode_start(inode
, flags
);
1337 if (sb
->s_op
->dirty_inode
)
1338 sb
->s_op
->dirty_inode(inode
, flags
);
1340 trace_writeback_dirty_inode(inode
, flags
);
1342 if (flags
& I_DIRTY_INODE
)
1343 flags
&= ~I_DIRTY_TIME
;
1344 dirtytime
= flags
& I_DIRTY_TIME
;
1347 * Paired with smp_mb() in __writeback_single_inode() for the
1348 * following lockless i_state test. See there for details.
1352 if (((inode
->i_state
& flags
) == flags
) ||
1353 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
1356 if (unlikely(block_dump
))
1357 block_dump___mark_inode_dirty(inode
);
1359 spin_lock(&inode
->i_lock
);
1360 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
1361 goto out_unlock_inode
;
1362 if ((inode
->i_state
& flags
) != flags
) {
1363 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1365 inode_attach_wb(inode
, NULL
);
1367 if (flags
& I_DIRTY_INODE
)
1368 inode
->i_state
&= ~I_DIRTY_TIME
;
1369 inode
->i_state
|= flags
;
1372 * If the inode is being synced, just update its dirty state.
1373 * The unlocker will place the inode on the appropriate
1374 * superblock list, based upon its state.
1376 if (inode
->i_state
& I_SYNC
)
1377 goto out_unlock_inode
;
1380 * Only add valid (hashed) inodes to the superblock's
1381 * dirty list. Add blockdev inodes as well.
1383 if (!S_ISBLK(inode
->i_mode
)) {
1384 if (inode_unhashed(inode
))
1385 goto out_unlock_inode
;
1387 if (inode
->i_state
& I_FREEING
)
1388 goto out_unlock_inode
;
1391 * If the inode was already on b_dirty/b_io/b_more_io, don't
1392 * reposition it (that would break b_dirty time-ordering).
1395 struct list_head
*dirty_list
;
1396 bool wakeup_bdi
= false;
1397 bdi
= inode_to_bdi(inode
);
1399 spin_unlock(&inode
->i_lock
);
1400 spin_lock(&bdi
->wb
.list_lock
);
1402 WARN(bdi_cap_writeback_dirty(bdi
) &&
1403 !test_bit(WB_registered
, &bdi
->wb
.state
),
1404 "bdi-%s not registered\n", bdi
->name
);
1406 inode
->dirtied_when
= jiffies
;
1408 inode
->dirtied_time_when
= jiffies
;
1410 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
1411 dirty_list
= &bdi
->wb
.b_dirty
;
1413 dirty_list
= &bdi
->wb
.b_dirty_time
;
1415 wakeup_bdi
= inode_wb_list_move_locked(inode
, &bdi
->wb
,
1418 spin_unlock(&bdi
->wb
.list_lock
);
1419 trace_writeback_dirty_inode_enqueue(inode
);
1422 * If this is the first dirty inode for this bdi,
1423 * we have to wake-up the corresponding bdi thread
1424 * to make sure background write-back happens
1427 if (bdi_cap_writeback_dirty(bdi
) && wakeup_bdi
)
1428 wb_wakeup_delayed(&bdi
->wb
);
1433 spin_unlock(&inode
->i_lock
);
1436 EXPORT_SYMBOL(__mark_inode_dirty
);
1438 static void wait_sb_inodes(struct super_block
*sb
)
1440 struct inode
*inode
, *old_inode
= NULL
;
1443 * We need to be protected against the filesystem going from
1444 * r/o to r/w or vice versa.
1446 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1448 spin_lock(&inode_sb_list_lock
);
1451 * Data integrity sync. Must wait for all pages under writeback,
1452 * because there may have been pages dirtied before our sync
1453 * call, but which had writeout started before we write it out.
1454 * In which case, the inode may not be on the dirty list, but
1455 * we still have to wait for that writeout.
1457 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1458 struct address_space
*mapping
= inode
->i_mapping
;
1460 spin_lock(&inode
->i_lock
);
1461 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1462 (mapping
->nrpages
== 0)) {
1463 spin_unlock(&inode
->i_lock
);
1467 spin_unlock(&inode
->i_lock
);
1468 spin_unlock(&inode_sb_list_lock
);
1471 * We hold a reference to 'inode' so it couldn't have been
1472 * removed from s_inodes list while we dropped the
1473 * inode_sb_list_lock. We cannot iput the inode now as we can
1474 * be holding the last reference and we cannot iput it under
1475 * inode_sb_list_lock. So we keep the reference and iput it
1481 filemap_fdatawait(mapping
);
1485 spin_lock(&inode_sb_list_lock
);
1487 spin_unlock(&inode_sb_list_lock
);
1492 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1493 * @sb: the superblock
1494 * @nr: the number of pages to write
1495 * @reason: reason why some writeback work initiated
1497 * Start writeback on some inodes on this super_block. No guarantees are made
1498 * on how many (if any) will be written, and this function does not wait
1499 * for IO completion of submitted IO.
1501 void writeback_inodes_sb_nr(struct super_block
*sb
,
1503 enum wb_reason reason
)
1505 DECLARE_COMPLETION_ONSTACK(done
);
1506 struct wb_writeback_work work
= {
1508 .sync_mode
= WB_SYNC_NONE
,
1509 .tagged_writepages
= 1,
1515 if (sb
->s_bdi
== &noop_backing_dev_info
)
1517 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1518 wb_queue_work(&sb
->s_bdi
->wb
, &work
);
1519 wait_for_completion(&done
);
1521 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1524 * writeback_inodes_sb - writeback dirty inodes from given super_block
1525 * @sb: the superblock
1526 * @reason: reason why some writeback work was initiated
1528 * Start writeback on some inodes on this super_block. No guarantees are made
1529 * on how many (if any) will be written, and this function does not wait
1530 * for IO completion of submitted IO.
1532 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1534 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1536 EXPORT_SYMBOL(writeback_inodes_sb
);
1539 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1540 * @sb: the superblock
1541 * @nr: the number of pages to write
1542 * @reason: the reason of writeback
1544 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1545 * Returns 1 if writeback was started, 0 if not.
1547 int try_to_writeback_inodes_sb_nr(struct super_block
*sb
,
1549 enum wb_reason reason
)
1551 if (writeback_in_progress(sb
->s_bdi
))
1554 if (!down_read_trylock(&sb
->s_umount
))
1557 writeback_inodes_sb_nr(sb
, nr
, reason
);
1558 up_read(&sb
->s_umount
);
1561 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1564 * try_to_writeback_inodes_sb - try to start writeback if none underway
1565 * @sb: the superblock
1566 * @reason: reason why some writeback work was initiated
1568 * Implement by try_to_writeback_inodes_sb_nr()
1569 * Returns 1 if writeback was started, 0 if not.
1571 int try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1573 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1575 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1578 * sync_inodes_sb - sync sb inode pages
1579 * @sb: the superblock
1581 * This function writes and waits on any dirty inode belonging to this
1584 void sync_inodes_sb(struct super_block
*sb
)
1586 DECLARE_COMPLETION_ONSTACK(done
);
1587 struct wb_writeback_work work
= {
1589 .sync_mode
= WB_SYNC_ALL
,
1590 .nr_pages
= LONG_MAX
,
1593 .reason
= WB_REASON_SYNC
,
1597 /* Nothing to do? */
1598 if (sb
->s_bdi
== &noop_backing_dev_info
)
1600 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1602 wb_queue_work(&sb
->s_bdi
->wb
, &work
);
1603 wait_for_completion(&done
);
1607 EXPORT_SYMBOL(sync_inodes_sb
);
1610 * write_inode_now - write an inode to disk
1611 * @inode: inode to write to disk
1612 * @sync: whether the write should be synchronous or not
1614 * This function commits an inode to disk immediately if it is dirty. This is
1615 * primarily needed by knfsd.
1617 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1619 int write_inode_now(struct inode
*inode
, int sync
)
1621 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1622 struct writeback_control wbc
= {
1623 .nr_to_write
= LONG_MAX
,
1624 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1626 .range_end
= LLONG_MAX
,
1629 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1630 wbc
.nr_to_write
= 0;
1633 return writeback_single_inode(inode
, wb
, &wbc
);
1635 EXPORT_SYMBOL(write_inode_now
);
1638 * sync_inode - write an inode and its pages to disk.
1639 * @inode: the inode to sync
1640 * @wbc: controls the writeback mode
1642 * sync_inode() will write an inode and its pages to disk. It will also
1643 * correctly update the inode on its superblock's dirty inode lists and will
1644 * update inode->i_state.
1646 * The caller must have a ref on the inode.
1648 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1650 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1652 EXPORT_SYMBOL(sync_inode
);
1655 * sync_inode_metadata - write an inode to disk
1656 * @inode: the inode to sync
1657 * @wait: wait for I/O to complete.
1659 * Write an inode to disk and adjust its dirty state after completion.
1661 * Note: only writes the actual inode, no associated data or other metadata.
1663 int sync_inode_metadata(struct inode
*inode
, int wait
)
1665 struct writeback_control wbc
= {
1666 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1667 .nr_to_write
= 0, /* metadata-only */
1670 return sync_inode(inode
, &wbc
);
1672 EXPORT_SYMBOL(sync_inode_metadata
);