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 WARN_ON_ONCE(!wb
->avg_write_bandwidth
);
103 atomic_long_add(wb
->avg_write_bandwidth
,
104 &wb
->bdi
->tot_write_bandwidth
);
109 static void wb_io_lists_depopulated(struct bdi_writeback
*wb
)
111 if (wb_has_dirty_io(wb
) && list_empty(&wb
->b_dirty
) &&
112 list_empty(&wb
->b_io
) && list_empty(&wb
->b_more_io
)) {
113 clear_bit(WB_has_dirty_io
, &wb
->state
);
114 WARN_ON_ONCE(atomic_long_sub_return(wb
->avg_write_bandwidth
,
115 &wb
->bdi
->tot_write_bandwidth
) < 0);
120 * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list
121 * @inode: inode to be moved
122 * @wb: target bdi_writeback
123 * @head: one of @wb->b_{dirty|io|more_io}
125 * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io.
126 * Returns %true if @inode is the first occupant of the !dirty_time IO
127 * lists; otherwise, %false.
129 static bool inode_wb_list_move_locked(struct inode
*inode
,
130 struct bdi_writeback
*wb
,
131 struct list_head
*head
)
133 assert_spin_locked(&wb
->list_lock
);
135 list_move(&inode
->i_wb_list
, head
);
137 /* dirty_time doesn't count as dirty_io until expiration */
138 if (head
!= &wb
->b_dirty_time
)
139 return wb_io_lists_populated(wb
);
141 wb_io_lists_depopulated(wb
);
146 * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list
147 * @inode: inode to be removed
148 * @wb: bdi_writeback @inode is being removed from
150 * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and
151 * clear %WB_has_dirty_io if all are empty afterwards.
153 static void inode_wb_list_del_locked(struct inode
*inode
,
154 struct bdi_writeback
*wb
)
156 assert_spin_locked(&wb
->list_lock
);
158 list_del_init(&inode
->i_wb_list
);
159 wb_io_lists_depopulated(wb
);
162 static void wb_wakeup(struct bdi_writeback
*wb
)
164 spin_lock_bh(&wb
->work_lock
);
165 if (test_bit(WB_registered
, &wb
->state
))
166 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
167 spin_unlock_bh(&wb
->work_lock
);
170 static void wb_queue_work(struct bdi_writeback
*wb
,
171 struct wb_writeback_work
*work
)
173 trace_writeback_queue(wb
->bdi
, work
);
175 spin_lock_bh(&wb
->work_lock
);
176 if (!test_bit(WB_registered
, &wb
->state
)) {
178 complete(work
->done
);
181 list_add_tail(&work
->list
, &wb
->work_list
);
182 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
184 spin_unlock_bh(&wb
->work_lock
);
187 static void __wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
188 bool range_cyclic
, enum wb_reason reason
)
190 struct wb_writeback_work
*work
;
192 if (!wb_has_dirty_io(wb
))
196 * This is WB_SYNC_NONE writeback, so if allocation fails just
197 * wakeup the thread for old dirty data writeback
199 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
201 trace_writeback_nowork(wb
->bdi
);
206 work
->sync_mode
= WB_SYNC_NONE
;
207 work
->nr_pages
= nr_pages
;
208 work
->range_cyclic
= range_cyclic
;
209 work
->reason
= reason
;
211 wb_queue_work(wb
, work
);
214 #ifdef CONFIG_CGROUP_WRITEBACK
217 * inode_congested - test whether an inode is congested
218 * @inode: inode to test for congestion
219 * @cong_bits: mask of WB_[a]sync_congested bits to test
221 * Tests whether @inode is congested. @cong_bits is the mask of congestion
222 * bits to test and the return value is the mask of set bits.
224 * If cgroup writeback is enabled for @inode, the congestion state is
225 * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
226 * associated with @inode is congested; otherwise, the root wb's congestion
229 int inode_congested(struct inode
*inode
, int cong_bits
)
232 struct bdi_writeback
*wb
= inode_to_wb(inode
);
234 return wb_congested(wb
, cong_bits
);
237 return wb_congested(&inode_to_bdi(inode
)->wb
, cong_bits
);
239 EXPORT_SYMBOL_GPL(inode_congested
);
241 #endif /* CONFIG_CGROUP_WRITEBACK */
244 * bdi_start_writeback - start writeback
245 * @bdi: the backing device to write from
246 * @nr_pages: the number of pages to write
247 * @reason: reason why some writeback work was initiated
250 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
251 * started when this function returns, we make no guarantees on
252 * completion. Caller need not hold sb s_umount semaphore.
255 void bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
256 enum wb_reason reason
)
258 __wb_start_writeback(&bdi
->wb
, nr_pages
, true, reason
);
262 * bdi_start_background_writeback - start background writeback
263 * @bdi: the backing device to write from
266 * This makes sure WB_SYNC_NONE background writeback happens. When
267 * this function returns, it is only guaranteed that for given BDI
268 * some IO is happening if we are over background dirty threshold.
269 * Caller need not hold sb s_umount semaphore.
271 void bdi_start_background_writeback(struct backing_dev_info
*bdi
)
274 * We just wake up the flusher thread. It will perform background
275 * writeback as soon as there is no other work to do.
277 trace_writeback_wake_background(bdi
);
282 * Remove the inode from the writeback list it is on.
284 void inode_wb_list_del(struct inode
*inode
)
286 struct bdi_writeback
*wb
= inode_to_wb(inode
);
288 spin_lock(&wb
->list_lock
);
289 inode_wb_list_del_locked(inode
, wb
);
290 spin_unlock(&wb
->list_lock
);
294 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
295 * furthest end of its superblock's dirty-inode list.
297 * Before stamping the inode's ->dirtied_when, we check to see whether it is
298 * already the most-recently-dirtied inode on the b_dirty list. If that is
299 * the case then the inode must have been redirtied while it was being written
300 * out and we don't reset its dirtied_when.
302 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
304 if (!list_empty(&wb
->b_dirty
)) {
307 tail
= wb_inode(wb
->b_dirty
.next
);
308 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
309 inode
->dirtied_when
= jiffies
;
311 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty
);
315 * requeue inode for re-scanning after bdi->b_io list is exhausted.
317 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
319 inode_wb_list_move_locked(inode
, wb
, &wb
->b_more_io
);
322 static void inode_sync_complete(struct inode
*inode
)
324 inode
->i_state
&= ~I_SYNC
;
325 /* If inode is clean an unused, put it into LRU now... */
326 inode_add_lru(inode
);
327 /* Waiters must see I_SYNC cleared before being woken up */
329 wake_up_bit(&inode
->i_state
, __I_SYNC
);
332 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
334 bool ret
= time_after(inode
->dirtied_when
, t
);
337 * For inodes being constantly redirtied, dirtied_when can get stuck.
338 * It _appears_ to be in the future, but is actually in distant past.
339 * This test is necessary to prevent such wrapped-around relative times
340 * from permanently stopping the whole bdi writeback.
342 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
347 #define EXPIRE_DIRTY_ATIME 0x0001
350 * Move expired (dirtied before work->older_than_this) dirty inodes from
351 * @delaying_queue to @dispatch_queue.
353 static int move_expired_inodes(struct list_head
*delaying_queue
,
354 struct list_head
*dispatch_queue
,
356 struct wb_writeback_work
*work
)
358 unsigned long *older_than_this
= NULL
;
359 unsigned long expire_time
;
361 struct list_head
*pos
, *node
;
362 struct super_block
*sb
= NULL
;
367 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
368 older_than_this
= work
->older_than_this
;
369 else if (!work
->for_sync
) {
370 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
371 older_than_this
= &expire_time
;
373 while (!list_empty(delaying_queue
)) {
374 inode
= wb_inode(delaying_queue
->prev
);
375 if (older_than_this
&&
376 inode_dirtied_after(inode
, *older_than_this
))
378 list_move(&inode
->i_wb_list
, &tmp
);
380 if (flags
& EXPIRE_DIRTY_ATIME
)
381 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
382 if (sb_is_blkdev_sb(inode
->i_sb
))
384 if (sb
&& sb
!= inode
->i_sb
)
389 /* just one sb in list, splice to dispatch_queue and we're done */
391 list_splice(&tmp
, dispatch_queue
);
395 /* Move inodes from one superblock together */
396 while (!list_empty(&tmp
)) {
397 sb
= wb_inode(tmp
.prev
)->i_sb
;
398 list_for_each_prev_safe(pos
, node
, &tmp
) {
399 inode
= wb_inode(pos
);
400 if (inode
->i_sb
== sb
)
401 list_move(&inode
->i_wb_list
, dispatch_queue
);
409 * Queue all expired dirty inodes for io, eldest first.
411 * newly dirtied b_dirty b_io b_more_io
412 * =============> gf edc BA
414 * newly dirtied b_dirty b_io b_more_io
415 * =============> g fBAedc
417 * +--> dequeue for IO
419 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
423 assert_spin_locked(&wb
->list_lock
);
424 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
425 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
426 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
427 EXPIRE_DIRTY_ATIME
, work
);
429 wb_io_lists_populated(wb
);
430 trace_writeback_queue_io(wb
, work
, moved
);
433 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
437 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
438 trace_writeback_write_inode_start(inode
, wbc
);
439 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
440 trace_writeback_write_inode(inode
, wbc
);
447 * Wait for writeback on an inode to complete. Called with i_lock held.
448 * Caller must make sure inode cannot go away when we drop i_lock.
450 static void __inode_wait_for_writeback(struct inode
*inode
)
451 __releases(inode
->i_lock
)
452 __acquires(inode
->i_lock
)
454 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
455 wait_queue_head_t
*wqh
;
457 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
458 while (inode
->i_state
& I_SYNC
) {
459 spin_unlock(&inode
->i_lock
);
460 __wait_on_bit(wqh
, &wq
, bit_wait
,
461 TASK_UNINTERRUPTIBLE
);
462 spin_lock(&inode
->i_lock
);
467 * Wait for writeback on an inode to complete. Caller must have inode pinned.
469 void inode_wait_for_writeback(struct inode
*inode
)
471 spin_lock(&inode
->i_lock
);
472 __inode_wait_for_writeback(inode
);
473 spin_unlock(&inode
->i_lock
);
477 * Sleep until I_SYNC is cleared. This function must be called with i_lock
478 * held and drops it. It is aimed for callers not holding any inode reference
479 * so once i_lock is dropped, inode can go away.
481 static void inode_sleep_on_writeback(struct inode
*inode
)
482 __releases(inode
->i_lock
)
485 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
488 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
489 sleep
= inode
->i_state
& I_SYNC
;
490 spin_unlock(&inode
->i_lock
);
493 finish_wait(wqh
, &wait
);
497 * Find proper writeback list for the inode depending on its current state and
498 * possibly also change of its state while we were doing writeback. Here we
499 * handle things such as livelock prevention or fairness of writeback among
500 * inodes. This function can be called only by flusher thread - noone else
501 * processes all inodes in writeback lists and requeueing inodes behind flusher
502 * thread's back can have unexpected consequences.
504 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
505 struct writeback_control
*wbc
)
507 if (inode
->i_state
& I_FREEING
)
511 * Sync livelock prevention. Each inode is tagged and synced in one
512 * shot. If still dirty, it will be redirty_tail()'ed below. Update
513 * the dirty time to prevent enqueue and sync it again.
515 if ((inode
->i_state
& I_DIRTY
) &&
516 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
517 inode
->dirtied_when
= jiffies
;
519 if (wbc
->pages_skipped
) {
521 * writeback is not making progress due to locked
522 * buffers. Skip this inode for now.
524 redirty_tail(inode
, wb
);
528 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
530 * We didn't write back all the pages. nfs_writepages()
531 * sometimes bales out without doing anything.
533 if (wbc
->nr_to_write
<= 0) {
534 /* Slice used up. Queue for next turn. */
535 requeue_io(inode
, wb
);
538 * Writeback blocked by something other than
539 * congestion. Delay the inode for some time to
540 * avoid spinning on the CPU (100% iowait)
541 * retrying writeback of the dirty page/inode
542 * that cannot be performed immediately.
544 redirty_tail(inode
, wb
);
546 } else if (inode
->i_state
& I_DIRTY
) {
548 * Filesystems can dirty the inode during writeback operations,
549 * such as delayed allocation during submission or metadata
550 * updates after data IO completion.
552 redirty_tail(inode
, wb
);
553 } else if (inode
->i_state
& I_DIRTY_TIME
) {
554 inode
->dirtied_when
= jiffies
;
555 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty_time
);
557 /* The inode is clean. Remove from writeback lists. */
558 inode_wb_list_del_locked(inode
, wb
);
563 * Write out an inode and its dirty pages. Do not update the writeback list
564 * linkage. That is left to the caller. The caller is also responsible for
565 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
568 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
570 struct address_space
*mapping
= inode
->i_mapping
;
571 long nr_to_write
= wbc
->nr_to_write
;
575 WARN_ON(!(inode
->i_state
& I_SYNC
));
577 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
579 ret
= do_writepages(mapping
, wbc
);
582 * Make sure to wait on the data before writing out the metadata.
583 * This is important for filesystems that modify metadata on data
584 * I/O completion. We don't do it for sync(2) writeback because it has a
585 * separate, external IO completion path and ->sync_fs for guaranteeing
586 * inode metadata is written back correctly.
588 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
589 int err
= filemap_fdatawait(mapping
);
595 * Some filesystems may redirty the inode during the writeback
596 * due to delalloc, clear dirty metadata flags right before
599 spin_lock(&inode
->i_lock
);
601 dirty
= inode
->i_state
& I_DIRTY
;
602 if (inode
->i_state
& I_DIRTY_TIME
) {
603 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
604 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
605 unlikely(time_after(jiffies
,
606 (inode
->dirtied_time_when
+
607 dirtytime_expire_interval
* HZ
)))) {
608 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
609 trace_writeback_lazytime(inode
);
612 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
613 inode
->i_state
&= ~dirty
;
616 * Paired with smp_mb() in __mark_inode_dirty(). This allows
617 * __mark_inode_dirty() to test i_state without grabbing i_lock -
618 * either they see the I_DIRTY bits cleared or we see the dirtied
621 * I_DIRTY_PAGES is always cleared together above even if @mapping
622 * still has dirty pages. The flag is reinstated after smp_mb() if
623 * necessary. This guarantees that either __mark_inode_dirty()
624 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
628 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
629 inode
->i_state
|= I_DIRTY_PAGES
;
631 spin_unlock(&inode
->i_lock
);
633 if (dirty
& I_DIRTY_TIME
)
634 mark_inode_dirty_sync(inode
);
635 /* Don't write the inode if only I_DIRTY_PAGES was set */
636 if (dirty
& ~I_DIRTY_PAGES
) {
637 int err
= write_inode(inode
, wbc
);
641 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
646 * Write out an inode's dirty pages. Either the caller has an active reference
647 * on the inode or the inode has I_WILL_FREE set.
649 * This function is designed to be called for writing back one inode which
650 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
651 * and does more profound writeback list handling in writeback_sb_inodes().
654 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
655 struct writeback_control
*wbc
)
659 spin_lock(&inode
->i_lock
);
660 if (!atomic_read(&inode
->i_count
))
661 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
663 WARN_ON(inode
->i_state
& I_WILL_FREE
);
665 if (inode
->i_state
& I_SYNC
) {
666 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
669 * It's a data-integrity sync. We must wait. Since callers hold
670 * inode reference or inode has I_WILL_FREE set, it cannot go
673 __inode_wait_for_writeback(inode
);
675 WARN_ON(inode
->i_state
& I_SYNC
);
677 * Skip inode if it is clean and we have no outstanding writeback in
678 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
679 * function since flusher thread may be doing for example sync in
680 * parallel and if we move the inode, it could get skipped. So here we
681 * make sure inode is on some writeback list and leave it there unless
682 * we have completely cleaned the inode.
684 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
685 (wbc
->sync_mode
!= WB_SYNC_ALL
||
686 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
688 inode
->i_state
|= I_SYNC
;
689 spin_unlock(&inode
->i_lock
);
691 ret
= __writeback_single_inode(inode
, wbc
);
693 spin_lock(&wb
->list_lock
);
694 spin_lock(&inode
->i_lock
);
696 * If inode is clean, remove it from writeback lists. Otherwise don't
697 * touch it. See comment above for explanation.
699 if (!(inode
->i_state
& I_DIRTY_ALL
))
700 inode_wb_list_del_locked(inode
, wb
);
701 spin_unlock(&wb
->list_lock
);
702 inode_sync_complete(inode
);
704 spin_unlock(&inode
->i_lock
);
708 static long writeback_chunk_size(struct bdi_writeback
*wb
,
709 struct wb_writeback_work
*work
)
714 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
715 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
716 * here avoids calling into writeback_inodes_wb() more than once.
718 * The intended call sequence for WB_SYNC_ALL writeback is:
721 * writeback_sb_inodes() <== called only once
722 * write_cache_pages() <== called once for each inode
723 * (quickly) tag currently dirty pages
724 * (maybe slowly) sync all tagged pages
726 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
729 pages
= min(wb
->avg_write_bandwidth
/ 2,
730 global_dirty_limit
/ DIRTY_SCOPE
);
731 pages
= min(pages
, work
->nr_pages
);
732 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
733 MIN_WRITEBACK_PAGES
);
740 * Write a portion of b_io inodes which belong to @sb.
742 * Return the number of pages and/or inodes written.
744 static long writeback_sb_inodes(struct super_block
*sb
,
745 struct bdi_writeback
*wb
,
746 struct wb_writeback_work
*work
)
748 struct writeback_control wbc
= {
749 .sync_mode
= work
->sync_mode
,
750 .tagged_writepages
= work
->tagged_writepages
,
751 .for_kupdate
= work
->for_kupdate
,
752 .for_background
= work
->for_background
,
753 .for_sync
= work
->for_sync
,
754 .range_cyclic
= work
->range_cyclic
,
756 .range_end
= LLONG_MAX
,
758 unsigned long start_time
= jiffies
;
760 long wrote
= 0; /* count both pages and inodes */
762 while (!list_empty(&wb
->b_io
)) {
763 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
765 if (inode
->i_sb
!= sb
) {
768 * We only want to write back data for this
769 * superblock, move all inodes not belonging
770 * to it back onto the dirty list.
772 redirty_tail(inode
, wb
);
777 * The inode belongs to a different superblock.
778 * Bounce back to the caller to unpin this and
779 * pin the next superblock.
785 * Don't bother with new inodes or inodes being freed, first
786 * kind does not need periodic writeout yet, and for the latter
787 * kind writeout is handled by the freer.
789 spin_lock(&inode
->i_lock
);
790 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
791 spin_unlock(&inode
->i_lock
);
792 redirty_tail(inode
, wb
);
795 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
797 * If this inode is locked for writeback and we are not
798 * doing writeback-for-data-integrity, move it to
799 * b_more_io so that writeback can proceed with the
800 * other inodes on s_io.
802 * We'll have another go at writing back this inode
803 * when we completed a full scan of b_io.
805 spin_unlock(&inode
->i_lock
);
806 requeue_io(inode
, wb
);
807 trace_writeback_sb_inodes_requeue(inode
);
810 spin_unlock(&wb
->list_lock
);
813 * We already requeued the inode if it had I_SYNC set and we
814 * are doing WB_SYNC_NONE writeback. So this catches only the
817 if (inode
->i_state
& I_SYNC
) {
818 /* Wait for I_SYNC. This function drops i_lock... */
819 inode_sleep_on_writeback(inode
);
820 /* Inode may be gone, start again */
821 spin_lock(&wb
->list_lock
);
824 inode
->i_state
|= I_SYNC
;
825 spin_unlock(&inode
->i_lock
);
827 write_chunk
= writeback_chunk_size(wb
, work
);
828 wbc
.nr_to_write
= write_chunk
;
829 wbc
.pages_skipped
= 0;
832 * We use I_SYNC to pin the inode in memory. While it is set
833 * evict_inode() will wait so the inode cannot be freed.
835 __writeback_single_inode(inode
, &wbc
);
837 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
838 wrote
+= write_chunk
- wbc
.nr_to_write
;
839 spin_lock(&wb
->list_lock
);
840 spin_lock(&inode
->i_lock
);
841 if (!(inode
->i_state
& I_DIRTY_ALL
))
843 requeue_inode(inode
, wb
, &wbc
);
844 inode_sync_complete(inode
);
845 spin_unlock(&inode
->i_lock
);
846 cond_resched_lock(&wb
->list_lock
);
848 * bail out to wb_writeback() often enough to check
849 * background threshold and other termination conditions.
852 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
854 if (work
->nr_pages
<= 0)
861 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
862 struct wb_writeback_work
*work
)
864 unsigned long start_time
= jiffies
;
867 while (!list_empty(&wb
->b_io
)) {
868 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
869 struct super_block
*sb
= inode
->i_sb
;
871 if (!trylock_super(sb
)) {
873 * trylock_super() may fail consistently due to
874 * s_umount being grabbed by someone else. Don't use
875 * requeue_io() to avoid busy retrying the inode/sb.
877 redirty_tail(inode
, wb
);
880 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
881 up_read(&sb
->s_umount
);
883 /* refer to the same tests at the end of writeback_sb_inodes */
885 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
887 if (work
->nr_pages
<= 0)
891 /* Leave any unwritten inodes on b_io */
895 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
896 enum wb_reason reason
)
898 struct wb_writeback_work work
= {
899 .nr_pages
= nr_pages
,
900 .sync_mode
= WB_SYNC_NONE
,
905 spin_lock(&wb
->list_lock
);
906 if (list_empty(&wb
->b_io
))
908 __writeback_inodes_wb(wb
, &work
);
909 spin_unlock(&wb
->list_lock
);
911 return nr_pages
- work
.nr_pages
;
914 static bool over_bground_thresh(struct bdi_writeback
*wb
)
916 unsigned long background_thresh
, dirty_thresh
;
918 global_dirty_limits(&background_thresh
, &dirty_thresh
);
920 if (global_page_state(NR_FILE_DIRTY
) +
921 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
)
924 if (wb_stat(wb
, WB_RECLAIMABLE
) > wb_dirty_limit(wb
, background_thresh
))
931 * Called under wb->list_lock. If there are multiple wb per bdi,
932 * only the flusher working on the first wb should do it.
934 static void wb_update_bandwidth(struct bdi_writeback
*wb
,
935 unsigned long start_time
)
937 __wb_update_bandwidth(wb
, 0, 0, 0, 0, 0, start_time
);
941 * Explicit flushing or periodic writeback of "old" data.
943 * Define "old": the first time one of an inode's pages is dirtied, we mark the
944 * dirtying-time in the inode's address_space. So this periodic writeback code
945 * just walks the superblock inode list, writing back any inodes which are
946 * older than a specific point in time.
948 * Try to run once per dirty_writeback_interval. But if a writeback event
949 * takes longer than a dirty_writeback_interval interval, then leave a
952 * older_than_this takes precedence over nr_to_write. So we'll only write back
953 * all dirty pages if they are all attached to "old" mappings.
955 static long wb_writeback(struct bdi_writeback
*wb
,
956 struct wb_writeback_work
*work
)
958 unsigned long wb_start
= jiffies
;
959 long nr_pages
= work
->nr_pages
;
960 unsigned long oldest_jif
;
964 oldest_jif
= jiffies
;
965 work
->older_than_this
= &oldest_jif
;
967 spin_lock(&wb
->list_lock
);
970 * Stop writeback when nr_pages has been consumed
972 if (work
->nr_pages
<= 0)
976 * Background writeout and kupdate-style writeback may
977 * run forever. Stop them if there is other work to do
978 * so that e.g. sync can proceed. They'll be restarted
979 * after the other works are all done.
981 if ((work
->for_background
|| work
->for_kupdate
) &&
982 !list_empty(&wb
->work_list
))
986 * For background writeout, stop when we are below the
987 * background dirty threshold
989 if (work
->for_background
&& !over_bground_thresh(wb
))
993 * Kupdate and background works are special and we want to
994 * include all inodes that need writing. Livelock avoidance is
995 * handled by these works yielding to any other work so we are
998 if (work
->for_kupdate
) {
999 oldest_jif
= jiffies
-
1000 msecs_to_jiffies(dirty_expire_interval
* 10);
1001 } else if (work
->for_background
)
1002 oldest_jif
= jiffies
;
1004 trace_writeback_start(wb
->bdi
, work
);
1005 if (list_empty(&wb
->b_io
))
1008 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
1010 progress
= __writeback_inodes_wb(wb
, work
);
1011 trace_writeback_written(wb
->bdi
, work
);
1013 wb_update_bandwidth(wb
, wb_start
);
1016 * Did we write something? Try for more
1018 * Dirty inodes are moved to b_io for writeback in batches.
1019 * The completion of the current batch does not necessarily
1020 * mean the overall work is done. So we keep looping as long
1021 * as made some progress on cleaning pages or inodes.
1026 * No more inodes for IO, bail
1028 if (list_empty(&wb
->b_more_io
))
1031 * Nothing written. Wait for some inode to
1032 * become available for writeback. Otherwise
1033 * we'll just busyloop.
1035 if (!list_empty(&wb
->b_more_io
)) {
1036 trace_writeback_wait(wb
->bdi
, work
);
1037 inode
= wb_inode(wb
->b_more_io
.prev
);
1038 spin_lock(&inode
->i_lock
);
1039 spin_unlock(&wb
->list_lock
);
1040 /* This function drops i_lock... */
1041 inode_sleep_on_writeback(inode
);
1042 spin_lock(&wb
->list_lock
);
1045 spin_unlock(&wb
->list_lock
);
1047 return nr_pages
- work
->nr_pages
;
1051 * Return the next wb_writeback_work struct that hasn't been processed yet.
1053 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
1055 struct wb_writeback_work
*work
= NULL
;
1057 spin_lock_bh(&wb
->work_lock
);
1058 if (!list_empty(&wb
->work_list
)) {
1059 work
= list_entry(wb
->work_list
.next
,
1060 struct wb_writeback_work
, list
);
1061 list_del_init(&work
->list
);
1063 spin_unlock_bh(&wb
->work_lock
);
1068 * Add in the number of potentially dirty inodes, because each inode
1069 * write can dirty pagecache in the underlying blockdev.
1071 static unsigned long get_nr_dirty_pages(void)
1073 return global_page_state(NR_FILE_DIRTY
) +
1074 global_page_state(NR_UNSTABLE_NFS
) +
1075 get_nr_dirty_inodes();
1078 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1080 if (over_bground_thresh(wb
)) {
1082 struct wb_writeback_work work
= {
1083 .nr_pages
= LONG_MAX
,
1084 .sync_mode
= WB_SYNC_NONE
,
1085 .for_background
= 1,
1087 .reason
= WB_REASON_BACKGROUND
,
1090 return wb_writeback(wb
, &work
);
1096 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1098 unsigned long expired
;
1102 * When set to zero, disable periodic writeback
1104 if (!dirty_writeback_interval
)
1107 expired
= wb
->last_old_flush
+
1108 msecs_to_jiffies(dirty_writeback_interval
* 10);
1109 if (time_before(jiffies
, expired
))
1112 wb
->last_old_flush
= jiffies
;
1113 nr_pages
= get_nr_dirty_pages();
1116 struct wb_writeback_work work
= {
1117 .nr_pages
= nr_pages
,
1118 .sync_mode
= WB_SYNC_NONE
,
1121 .reason
= WB_REASON_PERIODIC
,
1124 return wb_writeback(wb
, &work
);
1131 * Retrieve work items and do the writeback they describe
1133 static long wb_do_writeback(struct bdi_writeback
*wb
)
1135 struct wb_writeback_work
*work
;
1138 set_bit(WB_writeback_running
, &wb
->state
);
1139 while ((work
= get_next_work_item(wb
)) != NULL
) {
1141 trace_writeback_exec(wb
->bdi
, work
);
1143 wrote
+= wb_writeback(wb
, work
);
1146 * Notify the caller of completion if this is a synchronous
1147 * work item, otherwise just free it.
1150 complete(work
->done
);
1156 * Check for periodic writeback, kupdated() style
1158 wrote
+= wb_check_old_data_flush(wb
);
1159 wrote
+= wb_check_background_flush(wb
);
1160 clear_bit(WB_writeback_running
, &wb
->state
);
1166 * Handle writeback of dirty data for the device backed by this bdi. Also
1167 * reschedules periodically and does kupdated style flushing.
1169 void wb_workfn(struct work_struct
*work
)
1171 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1172 struct bdi_writeback
, dwork
);
1175 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1176 current
->flags
|= PF_SWAPWRITE
;
1178 if (likely(!current_is_workqueue_rescuer() ||
1179 !test_bit(WB_registered
, &wb
->state
))) {
1181 * The normal path. Keep writing back @wb until its
1182 * work_list is empty. Note that this path is also taken
1183 * if @wb is shutting down even when we're running off the
1184 * rescuer as work_list needs to be drained.
1187 pages_written
= wb_do_writeback(wb
);
1188 trace_writeback_pages_written(pages_written
);
1189 } while (!list_empty(&wb
->work_list
));
1192 * bdi_wq can't get enough workers and we're running off
1193 * the emergency worker. Don't hog it. Hopefully, 1024 is
1194 * enough for efficient IO.
1196 pages_written
= writeback_inodes_wb(wb
, 1024,
1197 WB_REASON_FORKER_THREAD
);
1198 trace_writeback_pages_written(pages_written
);
1201 if (!list_empty(&wb
->work_list
))
1202 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1203 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1204 wb_wakeup_delayed(wb
);
1206 current
->flags
&= ~PF_SWAPWRITE
;
1210 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1213 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1215 struct backing_dev_info
*bdi
;
1218 nr_pages
= get_nr_dirty_pages();
1221 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
)
1222 __wb_start_writeback(&bdi
->wb
, nr_pages
, false, reason
);
1227 * Wake up bdi's periodically to make sure dirtytime inodes gets
1228 * written back periodically. We deliberately do *not* check the
1229 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1230 * kernel to be constantly waking up once there are any dirtytime
1231 * inodes on the system. So instead we define a separate delayed work
1232 * function which gets called much more rarely. (By default, only
1233 * once every 12 hours.)
1235 * If there is any other write activity going on in the file system,
1236 * this function won't be necessary. But if the only thing that has
1237 * happened on the file system is a dirtytime inode caused by an atime
1238 * update, we need this infrastructure below to make sure that inode
1239 * eventually gets pushed out to disk.
1241 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1242 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1244 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1246 struct backing_dev_info
*bdi
;
1249 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1250 if (list_empty(&bdi
->wb
.b_dirty_time
))
1252 wb_wakeup(&bdi
->wb
);
1255 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1258 static int __init
start_dirtytime_writeback(void)
1260 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1263 __initcall(start_dirtytime_writeback
);
1265 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1266 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1270 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1271 if (ret
== 0 && write
)
1272 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1276 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1278 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1279 struct dentry
*dentry
;
1280 const char *name
= "?";
1282 dentry
= d_find_alias(inode
);
1284 spin_lock(&dentry
->d_lock
);
1285 name
= (const char *) dentry
->d_name
.name
;
1288 "%s(%d): dirtied inode %lu (%s) on %s\n",
1289 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1290 name
, inode
->i_sb
->s_id
);
1292 spin_unlock(&dentry
->d_lock
);
1299 * __mark_inode_dirty - internal function
1300 * @inode: inode to mark
1301 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1302 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1303 * mark_inode_dirty_sync.
1305 * Put the inode on the super block's dirty list.
1307 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1308 * dirty list only if it is hashed or if it refers to a blockdev.
1309 * If it was not hashed, it will never be added to the dirty list
1310 * even if it is later hashed, as it will have been marked dirty already.
1312 * In short, make sure you hash any inodes _before_ you start marking
1315 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1316 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1317 * the kernel-internal blockdev inode represents the dirtying time of the
1318 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1319 * page->mapping->host, so the page-dirtying time is recorded in the internal
1322 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1323 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1325 struct super_block
*sb
= inode
->i_sb
;
1326 struct backing_dev_info
*bdi
= NULL
;
1329 trace_writeback_mark_inode_dirty(inode
, flags
);
1332 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1333 * dirty the inode itself
1335 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
1336 trace_writeback_dirty_inode_start(inode
, flags
);
1338 if (sb
->s_op
->dirty_inode
)
1339 sb
->s_op
->dirty_inode(inode
, flags
);
1341 trace_writeback_dirty_inode(inode
, flags
);
1343 if (flags
& I_DIRTY_INODE
)
1344 flags
&= ~I_DIRTY_TIME
;
1345 dirtytime
= flags
& I_DIRTY_TIME
;
1348 * Paired with smp_mb() in __writeback_single_inode() for the
1349 * following lockless i_state test. See there for details.
1353 if (((inode
->i_state
& flags
) == flags
) ||
1354 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
1357 if (unlikely(block_dump
))
1358 block_dump___mark_inode_dirty(inode
);
1360 spin_lock(&inode
->i_lock
);
1361 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
1362 goto out_unlock_inode
;
1363 if ((inode
->i_state
& flags
) != flags
) {
1364 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1366 inode_attach_wb(inode
, NULL
);
1368 if (flags
& I_DIRTY_INODE
)
1369 inode
->i_state
&= ~I_DIRTY_TIME
;
1370 inode
->i_state
|= flags
;
1373 * If the inode is being synced, just update its dirty state.
1374 * The unlocker will place the inode on the appropriate
1375 * superblock list, based upon its state.
1377 if (inode
->i_state
& I_SYNC
)
1378 goto out_unlock_inode
;
1381 * Only add valid (hashed) inodes to the superblock's
1382 * dirty list. Add blockdev inodes as well.
1384 if (!S_ISBLK(inode
->i_mode
)) {
1385 if (inode_unhashed(inode
))
1386 goto out_unlock_inode
;
1388 if (inode
->i_state
& I_FREEING
)
1389 goto out_unlock_inode
;
1392 * If the inode was already on b_dirty/b_io/b_more_io, don't
1393 * reposition it (that would break b_dirty time-ordering).
1396 struct list_head
*dirty_list
;
1397 bool wakeup_bdi
= false;
1398 bdi
= inode_to_bdi(inode
);
1400 spin_unlock(&inode
->i_lock
);
1401 spin_lock(&bdi
->wb
.list_lock
);
1403 WARN(bdi_cap_writeback_dirty(bdi
) &&
1404 !test_bit(WB_registered
, &bdi
->wb
.state
),
1405 "bdi-%s not registered\n", bdi
->name
);
1407 inode
->dirtied_when
= jiffies
;
1409 inode
->dirtied_time_when
= jiffies
;
1411 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
1412 dirty_list
= &bdi
->wb
.b_dirty
;
1414 dirty_list
= &bdi
->wb
.b_dirty_time
;
1416 wakeup_bdi
= inode_wb_list_move_locked(inode
, &bdi
->wb
,
1419 spin_unlock(&bdi
->wb
.list_lock
);
1420 trace_writeback_dirty_inode_enqueue(inode
);
1423 * If this is the first dirty inode for this bdi,
1424 * we have to wake-up the corresponding bdi thread
1425 * to make sure background write-back happens
1428 if (bdi_cap_writeback_dirty(bdi
) && wakeup_bdi
)
1429 wb_wakeup_delayed(&bdi
->wb
);
1434 spin_unlock(&inode
->i_lock
);
1437 EXPORT_SYMBOL(__mark_inode_dirty
);
1439 static void wait_sb_inodes(struct super_block
*sb
)
1441 struct inode
*inode
, *old_inode
= NULL
;
1444 * We need to be protected against the filesystem going from
1445 * r/o to r/w or vice versa.
1447 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1449 spin_lock(&inode_sb_list_lock
);
1452 * Data integrity sync. Must wait for all pages under writeback,
1453 * because there may have been pages dirtied before our sync
1454 * call, but which had writeout started before we write it out.
1455 * In which case, the inode may not be on the dirty list, but
1456 * we still have to wait for that writeout.
1458 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1459 struct address_space
*mapping
= inode
->i_mapping
;
1461 spin_lock(&inode
->i_lock
);
1462 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1463 (mapping
->nrpages
== 0)) {
1464 spin_unlock(&inode
->i_lock
);
1468 spin_unlock(&inode
->i_lock
);
1469 spin_unlock(&inode_sb_list_lock
);
1472 * We hold a reference to 'inode' so it couldn't have been
1473 * removed from s_inodes list while we dropped the
1474 * inode_sb_list_lock. We cannot iput the inode now as we can
1475 * be holding the last reference and we cannot iput it under
1476 * inode_sb_list_lock. So we keep the reference and iput it
1482 filemap_fdatawait(mapping
);
1486 spin_lock(&inode_sb_list_lock
);
1488 spin_unlock(&inode_sb_list_lock
);
1493 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1494 * @sb: the superblock
1495 * @nr: the number of pages to write
1496 * @reason: reason why some writeback work initiated
1498 * Start writeback on some inodes on this super_block. No guarantees are made
1499 * on how many (if any) will be written, and this function does not wait
1500 * for IO completion of submitted IO.
1502 void writeback_inodes_sb_nr(struct super_block
*sb
,
1504 enum wb_reason reason
)
1506 DECLARE_COMPLETION_ONSTACK(done
);
1507 struct wb_writeback_work work
= {
1509 .sync_mode
= WB_SYNC_NONE
,
1510 .tagged_writepages
= 1,
1515 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1517 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1519 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1520 wb_queue_work(&bdi
->wb
, &work
);
1521 wait_for_completion(&done
);
1523 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1526 * writeback_inodes_sb - writeback dirty inodes from given super_block
1527 * @sb: the superblock
1528 * @reason: reason why some writeback work was initiated
1530 * Start writeback on some inodes on this super_block. No guarantees are made
1531 * on how many (if any) will be written, and this function does not wait
1532 * for IO completion of submitted IO.
1534 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1536 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1538 EXPORT_SYMBOL(writeback_inodes_sb
);
1541 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1542 * @sb: the superblock
1543 * @nr: the number of pages to write
1544 * @reason: the reason of writeback
1546 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1547 * Returns 1 if writeback was started, 0 if not.
1549 int try_to_writeback_inodes_sb_nr(struct super_block
*sb
,
1551 enum wb_reason reason
)
1553 if (writeback_in_progress(sb
->s_bdi
))
1556 if (!down_read_trylock(&sb
->s_umount
))
1559 writeback_inodes_sb_nr(sb
, nr
, reason
);
1560 up_read(&sb
->s_umount
);
1563 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1566 * try_to_writeback_inodes_sb - try to start writeback if none underway
1567 * @sb: the superblock
1568 * @reason: reason why some writeback work was initiated
1570 * Implement by try_to_writeback_inodes_sb_nr()
1571 * Returns 1 if writeback was started, 0 if not.
1573 int try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1575 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1577 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1580 * sync_inodes_sb - sync sb inode pages
1581 * @sb: the superblock
1583 * This function writes and waits on any dirty inode belonging to this
1586 void sync_inodes_sb(struct super_block
*sb
)
1588 DECLARE_COMPLETION_ONSTACK(done
);
1589 struct wb_writeback_work work
= {
1591 .sync_mode
= WB_SYNC_ALL
,
1592 .nr_pages
= LONG_MAX
,
1595 .reason
= WB_REASON_SYNC
,
1598 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1600 /* Nothing to do? */
1601 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1603 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1605 wb_queue_work(&bdi
->wb
, &work
);
1606 wait_for_completion(&done
);
1610 EXPORT_SYMBOL(sync_inodes_sb
);
1613 * write_inode_now - write an inode to disk
1614 * @inode: inode to write to disk
1615 * @sync: whether the write should be synchronous or not
1617 * This function commits an inode to disk immediately if it is dirty. This is
1618 * primarily needed by knfsd.
1620 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1622 int write_inode_now(struct inode
*inode
, int sync
)
1624 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1625 struct writeback_control wbc
= {
1626 .nr_to_write
= LONG_MAX
,
1627 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1629 .range_end
= LLONG_MAX
,
1632 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1633 wbc
.nr_to_write
= 0;
1636 return writeback_single_inode(inode
, wb
, &wbc
);
1638 EXPORT_SYMBOL(write_inode_now
);
1641 * sync_inode - write an inode and its pages to disk.
1642 * @inode: the inode to sync
1643 * @wbc: controls the writeback mode
1645 * sync_inode() will write an inode and its pages to disk. It will also
1646 * correctly update the inode on its superblock's dirty inode lists and will
1647 * update inode->i_state.
1649 * The caller must have a ref on the inode.
1651 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1653 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1655 EXPORT_SYMBOL(sync_inode
);
1658 * sync_inode_metadata - write an inode to disk
1659 * @inode: the inode to sync
1660 * @wait: wait for I/O to complete.
1662 * Write an inode to disk and adjust its dirty state after completion.
1664 * Note: only writes the actual inode, no associated data or other metadata.
1666 int sync_inode_metadata(struct inode
*inode
, int wait
)
1668 struct writeback_control wbc
= {
1669 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1670 .nr_to_write
= 0, /* metadata-only */
1673 return sync_inode(inode
, &wbc
);
1675 EXPORT_SYMBOL(sync_inode_metadata
);