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))
37 struct wb_completion
{
42 * Passed into wb_writeback(), essentially a subset of writeback_control
44 struct wb_writeback_work
{
46 struct super_block
*sb
;
47 unsigned long *older_than_this
;
48 enum writeback_sync_modes sync_mode
;
49 unsigned int tagged_writepages
:1;
50 unsigned int for_kupdate
:1;
51 unsigned int range_cyclic
:1;
52 unsigned int for_background
:1;
53 unsigned int for_sync
:1; /* sync(2) WB_SYNC_ALL writeback */
54 unsigned int auto_free
:1; /* free on completion */
55 enum wb_reason reason
; /* why was writeback initiated? */
57 struct list_head list
; /* pending work list */
58 struct wb_completion
*done
; /* set if the caller waits */
62 * If one wants to wait for one or more wb_writeback_works, each work's
63 * ->done should be set to a wb_completion defined using the following
64 * macro. Once all work items are issued with wb_queue_work(), the caller
65 * can wait for the completion of all using wb_wait_for_completion(). Work
66 * items which are waited upon aren't freed automatically on completion.
68 #define DEFINE_WB_COMPLETION_ONSTACK(cmpl) \
69 struct wb_completion cmpl = { \
70 .cnt = ATOMIC_INIT(1), \
75 * If an inode is constantly having its pages dirtied, but then the
76 * updates stop dirtytime_expire_interval seconds in the past, it's
77 * possible for the worst case time between when an inode has its
78 * timestamps updated and when they finally get written out to be two
79 * dirtytime_expire_intervals. We set the default to 12 hours (in
80 * seconds), which means most of the time inodes will have their
81 * timestamps written to disk after 12 hours, but in the worst case a
82 * few inodes might not their timestamps updated for 24 hours.
84 unsigned int dirtytime_expire_interval
= 12 * 60 * 60;
86 static inline struct inode
*wb_inode(struct list_head
*head
)
88 return list_entry(head
, struct inode
, i_wb_list
);
92 * Include the creation of the trace points after defining the
93 * wb_writeback_work structure and inline functions so that the definition
94 * remains local to this file.
96 #define CREATE_TRACE_POINTS
97 #include <trace/events/writeback.h>
99 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage
);
101 static bool wb_io_lists_populated(struct bdi_writeback
*wb
)
103 if (wb_has_dirty_io(wb
)) {
106 set_bit(WB_has_dirty_io
, &wb
->state
);
107 WARN_ON_ONCE(!wb
->avg_write_bandwidth
);
108 atomic_long_add(wb
->avg_write_bandwidth
,
109 &wb
->bdi
->tot_write_bandwidth
);
114 static void wb_io_lists_depopulated(struct bdi_writeback
*wb
)
116 if (wb_has_dirty_io(wb
) && list_empty(&wb
->b_dirty
) &&
117 list_empty(&wb
->b_io
) && list_empty(&wb
->b_more_io
)) {
118 clear_bit(WB_has_dirty_io
, &wb
->state
);
119 WARN_ON_ONCE(atomic_long_sub_return(wb
->avg_write_bandwidth
,
120 &wb
->bdi
->tot_write_bandwidth
) < 0);
125 * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list
126 * @inode: inode to be moved
127 * @wb: target bdi_writeback
128 * @head: one of @wb->b_{dirty|io|more_io}
130 * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io.
131 * Returns %true if @inode is the first occupant of the !dirty_time IO
132 * lists; otherwise, %false.
134 static bool inode_wb_list_move_locked(struct inode
*inode
,
135 struct bdi_writeback
*wb
,
136 struct list_head
*head
)
138 assert_spin_locked(&wb
->list_lock
);
140 list_move(&inode
->i_wb_list
, head
);
142 /* dirty_time doesn't count as dirty_io until expiration */
143 if (head
!= &wb
->b_dirty_time
)
144 return wb_io_lists_populated(wb
);
146 wb_io_lists_depopulated(wb
);
151 * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list
152 * @inode: inode to be removed
153 * @wb: bdi_writeback @inode is being removed from
155 * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and
156 * clear %WB_has_dirty_io if all are empty afterwards.
158 static void inode_wb_list_del_locked(struct inode
*inode
,
159 struct bdi_writeback
*wb
)
161 assert_spin_locked(&wb
->list_lock
);
163 list_del_init(&inode
->i_wb_list
);
164 wb_io_lists_depopulated(wb
);
167 static void wb_wakeup(struct bdi_writeback
*wb
)
169 spin_lock_bh(&wb
->work_lock
);
170 if (test_bit(WB_registered
, &wb
->state
))
171 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
172 spin_unlock_bh(&wb
->work_lock
);
175 static void wb_queue_work(struct bdi_writeback
*wb
,
176 struct wb_writeback_work
*work
)
178 trace_writeback_queue(wb
->bdi
, work
);
180 spin_lock_bh(&wb
->work_lock
);
181 if (!test_bit(WB_registered
, &wb
->state
))
184 atomic_inc(&work
->done
->cnt
);
185 list_add_tail(&work
->list
, &wb
->work_list
);
186 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
188 spin_unlock_bh(&wb
->work_lock
);
192 * wb_wait_for_completion - wait for completion of bdi_writeback_works
193 * @bdi: bdi work items were issued to
194 * @done: target wb_completion
196 * Wait for one or more work items issued to @bdi with their ->done field
197 * set to @done, which should have been defined with
198 * DEFINE_WB_COMPLETION_ONSTACK(). This function returns after all such
199 * work items are completed. Work items which are waited upon aren't freed
200 * automatically on completion.
202 static void wb_wait_for_completion(struct backing_dev_info
*bdi
,
203 struct wb_completion
*done
)
205 atomic_dec(&done
->cnt
); /* put down the initial count */
206 wait_event(bdi
->wb_waitq
, !atomic_read(&done
->cnt
));
209 #ifdef CONFIG_CGROUP_WRITEBACK
212 * inode_congested - test whether an inode is congested
213 * @inode: inode to test for congestion
214 * @cong_bits: mask of WB_[a]sync_congested bits to test
216 * Tests whether @inode is congested. @cong_bits is the mask of congestion
217 * bits to test and the return value is the mask of set bits.
219 * If cgroup writeback is enabled for @inode, the congestion state is
220 * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
221 * associated with @inode is congested; otherwise, the root wb's congestion
224 int inode_congested(struct inode
*inode
, int cong_bits
)
227 struct bdi_writeback
*wb
= inode_to_wb(inode
);
229 return wb_congested(wb
, cong_bits
);
232 return wb_congested(&inode_to_bdi(inode
)->wb
, cong_bits
);
234 EXPORT_SYMBOL_GPL(inode_congested
);
237 * wb_split_bdi_pages - split nr_pages to write according to bandwidth
238 * @wb: target bdi_writeback to split @nr_pages to
239 * @nr_pages: number of pages to write for the whole bdi
241 * Split @wb's portion of @nr_pages according to @wb's write bandwidth in
242 * relation to the total write bandwidth of all wb's w/ dirty inodes on
245 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
247 unsigned long this_bw
= wb
->avg_write_bandwidth
;
248 unsigned long tot_bw
= atomic_long_read(&wb
->bdi
->tot_write_bandwidth
);
250 if (nr_pages
== LONG_MAX
)
254 * This may be called on clean wb's and proportional distribution
255 * may not make sense, just use the original @nr_pages in those
256 * cases. In general, we wanna err on the side of writing more.
258 if (!tot_bw
|| this_bw
>= tot_bw
)
261 return DIV_ROUND_UP_ULL((u64
)nr_pages
* this_bw
, tot_bw
);
264 #else /* CONFIG_CGROUP_WRITEBACK */
266 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
271 #endif /* CONFIG_CGROUP_WRITEBACK */
273 void wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
274 bool range_cyclic
, enum wb_reason reason
)
276 struct wb_writeback_work
*work
;
278 if (!wb_has_dirty_io(wb
))
282 * This is WB_SYNC_NONE writeback, so if allocation fails just
283 * wakeup the thread for old dirty data writeback
285 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
287 trace_writeback_nowork(wb
->bdi
);
292 work
->sync_mode
= WB_SYNC_NONE
;
293 work
->nr_pages
= nr_pages
;
294 work
->range_cyclic
= range_cyclic
;
295 work
->reason
= reason
;
298 wb_queue_work(wb
, work
);
302 * wb_start_background_writeback - start background writeback
303 * @wb: bdi_writback to write from
306 * This makes sure WB_SYNC_NONE background writeback happens. When
307 * this function returns, it is only guaranteed that for given wb
308 * some IO is happening if we are over background dirty threshold.
309 * Caller need not hold sb s_umount semaphore.
311 void wb_start_background_writeback(struct bdi_writeback
*wb
)
314 * We just wake up the flusher thread. It will perform background
315 * writeback as soon as there is no other work to do.
317 trace_writeback_wake_background(wb
->bdi
);
322 * Remove the inode from the writeback list it is on.
324 void inode_wb_list_del(struct inode
*inode
)
326 struct bdi_writeback
*wb
= inode_to_wb(inode
);
328 spin_lock(&wb
->list_lock
);
329 inode_wb_list_del_locked(inode
, wb
);
330 spin_unlock(&wb
->list_lock
);
334 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
335 * furthest end of its superblock's dirty-inode list.
337 * Before stamping the inode's ->dirtied_when, we check to see whether it is
338 * already the most-recently-dirtied inode on the b_dirty list. If that is
339 * the case then the inode must have been redirtied while it was being written
340 * out and we don't reset its dirtied_when.
342 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
344 if (!list_empty(&wb
->b_dirty
)) {
347 tail
= wb_inode(wb
->b_dirty
.next
);
348 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
349 inode
->dirtied_when
= jiffies
;
351 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty
);
355 * requeue inode for re-scanning after bdi->b_io list is exhausted.
357 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
359 inode_wb_list_move_locked(inode
, wb
, &wb
->b_more_io
);
362 static void inode_sync_complete(struct inode
*inode
)
364 inode
->i_state
&= ~I_SYNC
;
365 /* If inode is clean an unused, put it into LRU now... */
366 inode_add_lru(inode
);
367 /* Waiters must see I_SYNC cleared before being woken up */
369 wake_up_bit(&inode
->i_state
, __I_SYNC
);
372 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
374 bool ret
= time_after(inode
->dirtied_when
, t
);
377 * For inodes being constantly redirtied, dirtied_when can get stuck.
378 * It _appears_ to be in the future, but is actually in distant past.
379 * This test is necessary to prevent such wrapped-around relative times
380 * from permanently stopping the whole bdi writeback.
382 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
387 #define EXPIRE_DIRTY_ATIME 0x0001
390 * Move expired (dirtied before work->older_than_this) dirty inodes from
391 * @delaying_queue to @dispatch_queue.
393 static int move_expired_inodes(struct list_head
*delaying_queue
,
394 struct list_head
*dispatch_queue
,
396 struct wb_writeback_work
*work
)
398 unsigned long *older_than_this
= NULL
;
399 unsigned long expire_time
;
401 struct list_head
*pos
, *node
;
402 struct super_block
*sb
= NULL
;
407 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
408 older_than_this
= work
->older_than_this
;
409 else if (!work
->for_sync
) {
410 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
411 older_than_this
= &expire_time
;
413 while (!list_empty(delaying_queue
)) {
414 inode
= wb_inode(delaying_queue
->prev
);
415 if (older_than_this
&&
416 inode_dirtied_after(inode
, *older_than_this
))
418 list_move(&inode
->i_wb_list
, &tmp
);
420 if (flags
& EXPIRE_DIRTY_ATIME
)
421 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
422 if (sb_is_blkdev_sb(inode
->i_sb
))
424 if (sb
&& sb
!= inode
->i_sb
)
429 /* just one sb in list, splice to dispatch_queue and we're done */
431 list_splice(&tmp
, dispatch_queue
);
435 /* Move inodes from one superblock together */
436 while (!list_empty(&tmp
)) {
437 sb
= wb_inode(tmp
.prev
)->i_sb
;
438 list_for_each_prev_safe(pos
, node
, &tmp
) {
439 inode
= wb_inode(pos
);
440 if (inode
->i_sb
== sb
)
441 list_move(&inode
->i_wb_list
, dispatch_queue
);
449 * Queue all expired dirty inodes for io, eldest first.
451 * newly dirtied b_dirty b_io b_more_io
452 * =============> gf edc BA
454 * newly dirtied b_dirty b_io b_more_io
455 * =============> g fBAedc
457 * +--> dequeue for IO
459 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
463 assert_spin_locked(&wb
->list_lock
);
464 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
465 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
466 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
467 EXPIRE_DIRTY_ATIME
, work
);
469 wb_io_lists_populated(wb
);
470 trace_writeback_queue_io(wb
, work
, moved
);
473 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
477 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
478 trace_writeback_write_inode_start(inode
, wbc
);
479 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
480 trace_writeback_write_inode(inode
, wbc
);
487 * Wait for writeback on an inode to complete. Called with i_lock held.
488 * Caller must make sure inode cannot go away when we drop i_lock.
490 static void __inode_wait_for_writeback(struct inode
*inode
)
491 __releases(inode
->i_lock
)
492 __acquires(inode
->i_lock
)
494 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
495 wait_queue_head_t
*wqh
;
497 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
498 while (inode
->i_state
& I_SYNC
) {
499 spin_unlock(&inode
->i_lock
);
500 __wait_on_bit(wqh
, &wq
, bit_wait
,
501 TASK_UNINTERRUPTIBLE
);
502 spin_lock(&inode
->i_lock
);
507 * Wait for writeback on an inode to complete. Caller must have inode pinned.
509 void inode_wait_for_writeback(struct inode
*inode
)
511 spin_lock(&inode
->i_lock
);
512 __inode_wait_for_writeback(inode
);
513 spin_unlock(&inode
->i_lock
);
517 * Sleep until I_SYNC is cleared. This function must be called with i_lock
518 * held and drops it. It is aimed for callers not holding any inode reference
519 * so once i_lock is dropped, inode can go away.
521 static void inode_sleep_on_writeback(struct inode
*inode
)
522 __releases(inode
->i_lock
)
525 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
528 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
529 sleep
= inode
->i_state
& I_SYNC
;
530 spin_unlock(&inode
->i_lock
);
533 finish_wait(wqh
, &wait
);
537 * Find proper writeback list for the inode depending on its current state and
538 * possibly also change of its state while we were doing writeback. Here we
539 * handle things such as livelock prevention or fairness of writeback among
540 * inodes. This function can be called only by flusher thread - noone else
541 * processes all inodes in writeback lists and requeueing inodes behind flusher
542 * thread's back can have unexpected consequences.
544 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
545 struct writeback_control
*wbc
)
547 if (inode
->i_state
& I_FREEING
)
551 * Sync livelock prevention. Each inode is tagged and synced in one
552 * shot. If still dirty, it will be redirty_tail()'ed below. Update
553 * the dirty time to prevent enqueue and sync it again.
555 if ((inode
->i_state
& I_DIRTY
) &&
556 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
557 inode
->dirtied_when
= jiffies
;
559 if (wbc
->pages_skipped
) {
561 * writeback is not making progress due to locked
562 * buffers. Skip this inode for now.
564 redirty_tail(inode
, wb
);
568 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
570 * We didn't write back all the pages. nfs_writepages()
571 * sometimes bales out without doing anything.
573 if (wbc
->nr_to_write
<= 0) {
574 /* Slice used up. Queue for next turn. */
575 requeue_io(inode
, wb
);
578 * Writeback blocked by something other than
579 * congestion. Delay the inode for some time to
580 * avoid spinning on the CPU (100% iowait)
581 * retrying writeback of the dirty page/inode
582 * that cannot be performed immediately.
584 redirty_tail(inode
, wb
);
586 } else if (inode
->i_state
& I_DIRTY
) {
588 * Filesystems can dirty the inode during writeback operations,
589 * such as delayed allocation during submission or metadata
590 * updates after data IO completion.
592 redirty_tail(inode
, wb
);
593 } else if (inode
->i_state
& I_DIRTY_TIME
) {
594 inode
->dirtied_when
= jiffies
;
595 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty_time
);
597 /* The inode is clean. Remove from writeback lists. */
598 inode_wb_list_del_locked(inode
, wb
);
603 * Write out an inode and its dirty pages. Do not update the writeback list
604 * linkage. That is left to the caller. The caller is also responsible for
605 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
608 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
610 struct address_space
*mapping
= inode
->i_mapping
;
611 long nr_to_write
= wbc
->nr_to_write
;
615 WARN_ON(!(inode
->i_state
& I_SYNC
));
617 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
619 ret
= do_writepages(mapping
, wbc
);
622 * Make sure to wait on the data before writing out the metadata.
623 * This is important for filesystems that modify metadata on data
624 * I/O completion. We don't do it for sync(2) writeback because it has a
625 * separate, external IO completion path and ->sync_fs for guaranteeing
626 * inode metadata is written back correctly.
628 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
629 int err
= filemap_fdatawait(mapping
);
635 * Some filesystems may redirty the inode during the writeback
636 * due to delalloc, clear dirty metadata flags right before
639 spin_lock(&inode
->i_lock
);
641 dirty
= inode
->i_state
& I_DIRTY
;
642 if (inode
->i_state
& I_DIRTY_TIME
) {
643 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
644 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
645 unlikely(time_after(jiffies
,
646 (inode
->dirtied_time_when
+
647 dirtytime_expire_interval
* HZ
)))) {
648 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
649 trace_writeback_lazytime(inode
);
652 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
653 inode
->i_state
&= ~dirty
;
656 * Paired with smp_mb() in __mark_inode_dirty(). This allows
657 * __mark_inode_dirty() to test i_state without grabbing i_lock -
658 * either they see the I_DIRTY bits cleared or we see the dirtied
661 * I_DIRTY_PAGES is always cleared together above even if @mapping
662 * still has dirty pages. The flag is reinstated after smp_mb() if
663 * necessary. This guarantees that either __mark_inode_dirty()
664 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
668 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
669 inode
->i_state
|= I_DIRTY_PAGES
;
671 spin_unlock(&inode
->i_lock
);
673 if (dirty
& I_DIRTY_TIME
)
674 mark_inode_dirty_sync(inode
);
675 /* Don't write the inode if only I_DIRTY_PAGES was set */
676 if (dirty
& ~I_DIRTY_PAGES
) {
677 int err
= write_inode(inode
, wbc
);
681 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
686 * Write out an inode's dirty pages. Either the caller has an active reference
687 * on the inode or the inode has I_WILL_FREE set.
689 * This function is designed to be called for writing back one inode which
690 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
691 * and does more profound writeback list handling in writeback_sb_inodes().
694 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
695 struct writeback_control
*wbc
)
699 spin_lock(&inode
->i_lock
);
700 if (!atomic_read(&inode
->i_count
))
701 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
703 WARN_ON(inode
->i_state
& I_WILL_FREE
);
705 if (inode
->i_state
& I_SYNC
) {
706 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
709 * It's a data-integrity sync. We must wait. Since callers hold
710 * inode reference or inode has I_WILL_FREE set, it cannot go
713 __inode_wait_for_writeback(inode
);
715 WARN_ON(inode
->i_state
& I_SYNC
);
717 * Skip inode if it is clean and we have no outstanding writeback in
718 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
719 * function since flusher thread may be doing for example sync in
720 * parallel and if we move the inode, it could get skipped. So here we
721 * make sure inode is on some writeback list and leave it there unless
722 * we have completely cleaned the inode.
724 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
725 (wbc
->sync_mode
!= WB_SYNC_ALL
||
726 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
728 inode
->i_state
|= I_SYNC
;
729 spin_unlock(&inode
->i_lock
);
731 ret
= __writeback_single_inode(inode
, wbc
);
733 spin_lock(&wb
->list_lock
);
734 spin_lock(&inode
->i_lock
);
736 * If inode is clean, remove it from writeback lists. Otherwise don't
737 * touch it. See comment above for explanation.
739 if (!(inode
->i_state
& I_DIRTY_ALL
))
740 inode_wb_list_del_locked(inode
, wb
);
741 spin_unlock(&wb
->list_lock
);
742 inode_sync_complete(inode
);
744 spin_unlock(&inode
->i_lock
);
748 static long writeback_chunk_size(struct bdi_writeback
*wb
,
749 struct wb_writeback_work
*work
)
754 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
755 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
756 * here avoids calling into writeback_inodes_wb() more than once.
758 * The intended call sequence for WB_SYNC_ALL writeback is:
761 * writeback_sb_inodes() <== called only once
762 * write_cache_pages() <== called once for each inode
763 * (quickly) tag currently dirty pages
764 * (maybe slowly) sync all tagged pages
766 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
769 pages
= min(wb
->avg_write_bandwidth
/ 2,
770 global_dirty_limit
/ DIRTY_SCOPE
);
771 pages
= min(pages
, work
->nr_pages
);
772 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
773 MIN_WRITEBACK_PAGES
);
780 * Write a portion of b_io inodes which belong to @sb.
782 * Return the number of pages and/or inodes written.
784 static long writeback_sb_inodes(struct super_block
*sb
,
785 struct bdi_writeback
*wb
,
786 struct wb_writeback_work
*work
)
788 struct writeback_control wbc
= {
789 .sync_mode
= work
->sync_mode
,
790 .tagged_writepages
= work
->tagged_writepages
,
791 .for_kupdate
= work
->for_kupdate
,
792 .for_background
= work
->for_background
,
793 .for_sync
= work
->for_sync
,
794 .range_cyclic
= work
->range_cyclic
,
796 .range_end
= LLONG_MAX
,
798 unsigned long start_time
= jiffies
;
800 long wrote
= 0; /* count both pages and inodes */
802 while (!list_empty(&wb
->b_io
)) {
803 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
805 if (inode
->i_sb
!= sb
) {
808 * We only want to write back data for this
809 * superblock, move all inodes not belonging
810 * to it back onto the dirty list.
812 redirty_tail(inode
, wb
);
817 * The inode belongs to a different superblock.
818 * Bounce back to the caller to unpin this and
819 * pin the next superblock.
825 * Don't bother with new inodes or inodes being freed, first
826 * kind does not need periodic writeout yet, and for the latter
827 * kind writeout is handled by the freer.
829 spin_lock(&inode
->i_lock
);
830 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
831 spin_unlock(&inode
->i_lock
);
832 redirty_tail(inode
, wb
);
835 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
837 * If this inode is locked for writeback and we are not
838 * doing writeback-for-data-integrity, move it to
839 * b_more_io so that writeback can proceed with the
840 * other inodes on s_io.
842 * We'll have another go at writing back this inode
843 * when we completed a full scan of b_io.
845 spin_unlock(&inode
->i_lock
);
846 requeue_io(inode
, wb
);
847 trace_writeback_sb_inodes_requeue(inode
);
850 spin_unlock(&wb
->list_lock
);
853 * We already requeued the inode if it had I_SYNC set and we
854 * are doing WB_SYNC_NONE writeback. So this catches only the
857 if (inode
->i_state
& I_SYNC
) {
858 /* Wait for I_SYNC. This function drops i_lock... */
859 inode_sleep_on_writeback(inode
);
860 /* Inode may be gone, start again */
861 spin_lock(&wb
->list_lock
);
864 inode
->i_state
|= I_SYNC
;
865 spin_unlock(&inode
->i_lock
);
867 write_chunk
= writeback_chunk_size(wb
, work
);
868 wbc
.nr_to_write
= write_chunk
;
869 wbc
.pages_skipped
= 0;
872 * We use I_SYNC to pin the inode in memory. While it is set
873 * evict_inode() will wait so the inode cannot be freed.
875 __writeback_single_inode(inode
, &wbc
);
877 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
878 wrote
+= write_chunk
- wbc
.nr_to_write
;
879 spin_lock(&wb
->list_lock
);
880 spin_lock(&inode
->i_lock
);
881 if (!(inode
->i_state
& I_DIRTY_ALL
))
883 requeue_inode(inode
, wb
, &wbc
);
884 inode_sync_complete(inode
);
885 spin_unlock(&inode
->i_lock
);
886 cond_resched_lock(&wb
->list_lock
);
888 * bail out to wb_writeback() often enough to check
889 * background threshold and other termination conditions.
892 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
894 if (work
->nr_pages
<= 0)
901 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
902 struct wb_writeback_work
*work
)
904 unsigned long start_time
= jiffies
;
907 while (!list_empty(&wb
->b_io
)) {
908 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
909 struct super_block
*sb
= inode
->i_sb
;
911 if (!trylock_super(sb
)) {
913 * trylock_super() may fail consistently due to
914 * s_umount being grabbed by someone else. Don't use
915 * requeue_io() to avoid busy retrying the inode/sb.
917 redirty_tail(inode
, wb
);
920 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
921 up_read(&sb
->s_umount
);
923 /* refer to the same tests at the end of writeback_sb_inodes */
925 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
927 if (work
->nr_pages
<= 0)
931 /* Leave any unwritten inodes on b_io */
935 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
936 enum wb_reason reason
)
938 struct wb_writeback_work work
= {
939 .nr_pages
= nr_pages
,
940 .sync_mode
= WB_SYNC_NONE
,
945 spin_lock(&wb
->list_lock
);
946 if (list_empty(&wb
->b_io
))
948 __writeback_inodes_wb(wb
, &work
);
949 spin_unlock(&wb
->list_lock
);
951 return nr_pages
- work
.nr_pages
;
954 static bool over_bground_thresh(struct bdi_writeback
*wb
)
956 unsigned long background_thresh
, dirty_thresh
;
958 global_dirty_limits(&background_thresh
, &dirty_thresh
);
960 if (global_page_state(NR_FILE_DIRTY
) +
961 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
)
964 if (wb_stat(wb
, WB_RECLAIMABLE
) > wb_dirty_limit(wb
, background_thresh
))
971 * Called under wb->list_lock. If there are multiple wb per bdi,
972 * only the flusher working on the first wb should do it.
974 static void wb_update_bandwidth(struct bdi_writeback
*wb
,
975 unsigned long start_time
)
977 __wb_update_bandwidth(wb
, 0, 0, 0, 0, 0, start_time
);
981 * Explicit flushing or periodic writeback of "old" data.
983 * Define "old": the first time one of an inode's pages is dirtied, we mark the
984 * dirtying-time in the inode's address_space. So this periodic writeback code
985 * just walks the superblock inode list, writing back any inodes which are
986 * older than a specific point in time.
988 * Try to run once per dirty_writeback_interval. But if a writeback event
989 * takes longer than a dirty_writeback_interval interval, then leave a
992 * older_than_this takes precedence over nr_to_write. So we'll only write back
993 * all dirty pages if they are all attached to "old" mappings.
995 static long wb_writeback(struct bdi_writeback
*wb
,
996 struct wb_writeback_work
*work
)
998 unsigned long wb_start
= jiffies
;
999 long nr_pages
= work
->nr_pages
;
1000 unsigned long oldest_jif
;
1001 struct inode
*inode
;
1004 oldest_jif
= jiffies
;
1005 work
->older_than_this
= &oldest_jif
;
1007 spin_lock(&wb
->list_lock
);
1010 * Stop writeback when nr_pages has been consumed
1012 if (work
->nr_pages
<= 0)
1016 * Background writeout and kupdate-style writeback may
1017 * run forever. Stop them if there is other work to do
1018 * so that e.g. sync can proceed. They'll be restarted
1019 * after the other works are all done.
1021 if ((work
->for_background
|| work
->for_kupdate
) &&
1022 !list_empty(&wb
->work_list
))
1026 * For background writeout, stop when we are below the
1027 * background dirty threshold
1029 if (work
->for_background
&& !over_bground_thresh(wb
))
1033 * Kupdate and background works are special and we want to
1034 * include all inodes that need writing. Livelock avoidance is
1035 * handled by these works yielding to any other work so we are
1038 if (work
->for_kupdate
) {
1039 oldest_jif
= jiffies
-
1040 msecs_to_jiffies(dirty_expire_interval
* 10);
1041 } else if (work
->for_background
)
1042 oldest_jif
= jiffies
;
1044 trace_writeback_start(wb
->bdi
, work
);
1045 if (list_empty(&wb
->b_io
))
1048 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
1050 progress
= __writeback_inodes_wb(wb
, work
);
1051 trace_writeback_written(wb
->bdi
, work
);
1053 wb_update_bandwidth(wb
, wb_start
);
1056 * Did we write something? Try for more
1058 * Dirty inodes are moved to b_io for writeback in batches.
1059 * The completion of the current batch does not necessarily
1060 * mean the overall work is done. So we keep looping as long
1061 * as made some progress on cleaning pages or inodes.
1066 * No more inodes for IO, bail
1068 if (list_empty(&wb
->b_more_io
))
1071 * Nothing written. Wait for some inode to
1072 * become available for writeback. Otherwise
1073 * we'll just busyloop.
1075 if (!list_empty(&wb
->b_more_io
)) {
1076 trace_writeback_wait(wb
->bdi
, work
);
1077 inode
= wb_inode(wb
->b_more_io
.prev
);
1078 spin_lock(&inode
->i_lock
);
1079 spin_unlock(&wb
->list_lock
);
1080 /* This function drops i_lock... */
1081 inode_sleep_on_writeback(inode
);
1082 spin_lock(&wb
->list_lock
);
1085 spin_unlock(&wb
->list_lock
);
1087 return nr_pages
- work
->nr_pages
;
1091 * Return the next wb_writeback_work struct that hasn't been processed yet.
1093 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
1095 struct wb_writeback_work
*work
= NULL
;
1097 spin_lock_bh(&wb
->work_lock
);
1098 if (!list_empty(&wb
->work_list
)) {
1099 work
= list_entry(wb
->work_list
.next
,
1100 struct wb_writeback_work
, list
);
1101 list_del_init(&work
->list
);
1103 spin_unlock_bh(&wb
->work_lock
);
1108 * Add in the number of potentially dirty inodes, because each inode
1109 * write can dirty pagecache in the underlying blockdev.
1111 static unsigned long get_nr_dirty_pages(void)
1113 return global_page_state(NR_FILE_DIRTY
) +
1114 global_page_state(NR_UNSTABLE_NFS
) +
1115 get_nr_dirty_inodes();
1118 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1120 if (over_bground_thresh(wb
)) {
1122 struct wb_writeback_work work
= {
1123 .nr_pages
= LONG_MAX
,
1124 .sync_mode
= WB_SYNC_NONE
,
1125 .for_background
= 1,
1127 .reason
= WB_REASON_BACKGROUND
,
1130 return wb_writeback(wb
, &work
);
1136 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1138 unsigned long expired
;
1142 * When set to zero, disable periodic writeback
1144 if (!dirty_writeback_interval
)
1147 expired
= wb
->last_old_flush
+
1148 msecs_to_jiffies(dirty_writeback_interval
* 10);
1149 if (time_before(jiffies
, expired
))
1152 wb
->last_old_flush
= jiffies
;
1153 nr_pages
= get_nr_dirty_pages();
1156 struct wb_writeback_work work
= {
1157 .nr_pages
= nr_pages
,
1158 .sync_mode
= WB_SYNC_NONE
,
1161 .reason
= WB_REASON_PERIODIC
,
1164 return wb_writeback(wb
, &work
);
1171 * Retrieve work items and do the writeback they describe
1173 static long wb_do_writeback(struct bdi_writeback
*wb
)
1175 struct wb_writeback_work
*work
;
1178 set_bit(WB_writeback_running
, &wb
->state
);
1179 while ((work
= get_next_work_item(wb
)) != NULL
) {
1180 struct wb_completion
*done
= work
->done
;
1182 trace_writeback_exec(wb
->bdi
, work
);
1184 wrote
+= wb_writeback(wb
, work
);
1186 if (work
->auto_free
)
1188 if (done
&& atomic_dec_and_test(&done
->cnt
))
1189 wake_up_all(&wb
->bdi
->wb_waitq
);
1193 * Check for periodic writeback, kupdated() style
1195 wrote
+= wb_check_old_data_flush(wb
);
1196 wrote
+= wb_check_background_flush(wb
);
1197 clear_bit(WB_writeback_running
, &wb
->state
);
1203 * Handle writeback of dirty data for the device backed by this bdi. Also
1204 * reschedules periodically and does kupdated style flushing.
1206 void wb_workfn(struct work_struct
*work
)
1208 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1209 struct bdi_writeback
, dwork
);
1212 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1213 current
->flags
|= PF_SWAPWRITE
;
1215 if (likely(!current_is_workqueue_rescuer() ||
1216 !test_bit(WB_registered
, &wb
->state
))) {
1218 * The normal path. Keep writing back @wb until its
1219 * work_list is empty. Note that this path is also taken
1220 * if @wb is shutting down even when we're running off the
1221 * rescuer as work_list needs to be drained.
1224 pages_written
= wb_do_writeback(wb
);
1225 trace_writeback_pages_written(pages_written
);
1226 } while (!list_empty(&wb
->work_list
));
1229 * bdi_wq can't get enough workers and we're running off
1230 * the emergency worker. Don't hog it. Hopefully, 1024 is
1231 * enough for efficient IO.
1233 pages_written
= writeback_inodes_wb(wb
, 1024,
1234 WB_REASON_FORKER_THREAD
);
1235 trace_writeback_pages_written(pages_written
);
1238 if (!list_empty(&wb
->work_list
))
1239 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1240 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1241 wb_wakeup_delayed(wb
);
1243 current
->flags
&= ~PF_SWAPWRITE
;
1247 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1250 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1252 struct backing_dev_info
*bdi
;
1255 nr_pages
= get_nr_dirty_pages();
1258 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1259 struct bdi_writeback
*wb
;
1260 struct wb_iter iter
;
1262 if (!bdi_has_dirty_io(bdi
))
1265 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1266 wb_start_writeback(wb
, wb_split_bdi_pages(wb
, nr_pages
),
1273 * Wake up bdi's periodically to make sure dirtytime inodes gets
1274 * written back periodically. We deliberately do *not* check the
1275 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1276 * kernel to be constantly waking up once there are any dirtytime
1277 * inodes on the system. So instead we define a separate delayed work
1278 * function which gets called much more rarely. (By default, only
1279 * once every 12 hours.)
1281 * If there is any other write activity going on in the file system,
1282 * this function won't be necessary. But if the only thing that has
1283 * happened on the file system is a dirtytime inode caused by an atime
1284 * update, we need this infrastructure below to make sure that inode
1285 * eventually gets pushed out to disk.
1287 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1288 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1290 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1292 struct backing_dev_info
*bdi
;
1295 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1296 struct bdi_writeback
*wb
;
1297 struct wb_iter iter
;
1299 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1300 if (!list_empty(&bdi
->wb
.b_dirty_time
))
1301 wb_wakeup(&bdi
->wb
);
1304 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1307 static int __init
start_dirtytime_writeback(void)
1309 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1312 __initcall(start_dirtytime_writeback
);
1314 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1315 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1319 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1320 if (ret
== 0 && write
)
1321 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1325 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1327 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1328 struct dentry
*dentry
;
1329 const char *name
= "?";
1331 dentry
= d_find_alias(inode
);
1333 spin_lock(&dentry
->d_lock
);
1334 name
= (const char *) dentry
->d_name
.name
;
1337 "%s(%d): dirtied inode %lu (%s) on %s\n",
1338 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1339 name
, inode
->i_sb
->s_id
);
1341 spin_unlock(&dentry
->d_lock
);
1348 * __mark_inode_dirty - internal function
1349 * @inode: inode to mark
1350 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1351 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1352 * mark_inode_dirty_sync.
1354 * Put the inode on the super block's dirty list.
1356 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1357 * dirty list only if it is hashed or if it refers to a blockdev.
1358 * If it was not hashed, it will never be added to the dirty list
1359 * even if it is later hashed, as it will have been marked dirty already.
1361 * In short, make sure you hash any inodes _before_ you start marking
1364 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1365 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1366 * the kernel-internal blockdev inode represents the dirtying time of the
1367 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1368 * page->mapping->host, so the page-dirtying time is recorded in the internal
1371 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1372 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1374 struct super_block
*sb
= inode
->i_sb
;
1375 struct backing_dev_info
*bdi
= NULL
;
1378 trace_writeback_mark_inode_dirty(inode
, flags
);
1381 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1382 * dirty the inode itself
1384 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
1385 trace_writeback_dirty_inode_start(inode
, flags
);
1387 if (sb
->s_op
->dirty_inode
)
1388 sb
->s_op
->dirty_inode(inode
, flags
);
1390 trace_writeback_dirty_inode(inode
, flags
);
1392 if (flags
& I_DIRTY_INODE
)
1393 flags
&= ~I_DIRTY_TIME
;
1394 dirtytime
= flags
& I_DIRTY_TIME
;
1397 * Paired with smp_mb() in __writeback_single_inode() for the
1398 * following lockless i_state test. See there for details.
1402 if (((inode
->i_state
& flags
) == flags
) ||
1403 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
1406 if (unlikely(block_dump
))
1407 block_dump___mark_inode_dirty(inode
);
1409 spin_lock(&inode
->i_lock
);
1410 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
1411 goto out_unlock_inode
;
1412 if ((inode
->i_state
& flags
) != flags
) {
1413 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1415 inode_attach_wb(inode
, NULL
);
1417 if (flags
& I_DIRTY_INODE
)
1418 inode
->i_state
&= ~I_DIRTY_TIME
;
1419 inode
->i_state
|= flags
;
1422 * If the inode is being synced, just update its dirty state.
1423 * The unlocker will place the inode on the appropriate
1424 * superblock list, based upon its state.
1426 if (inode
->i_state
& I_SYNC
)
1427 goto out_unlock_inode
;
1430 * Only add valid (hashed) inodes to the superblock's
1431 * dirty list. Add blockdev inodes as well.
1433 if (!S_ISBLK(inode
->i_mode
)) {
1434 if (inode_unhashed(inode
))
1435 goto out_unlock_inode
;
1437 if (inode
->i_state
& I_FREEING
)
1438 goto out_unlock_inode
;
1441 * If the inode was already on b_dirty/b_io/b_more_io, don't
1442 * reposition it (that would break b_dirty time-ordering).
1445 struct list_head
*dirty_list
;
1446 bool wakeup_bdi
= false;
1447 bdi
= inode_to_bdi(inode
);
1449 spin_unlock(&inode
->i_lock
);
1450 spin_lock(&bdi
->wb
.list_lock
);
1452 WARN(bdi_cap_writeback_dirty(bdi
) &&
1453 !test_bit(WB_registered
, &bdi
->wb
.state
),
1454 "bdi-%s not registered\n", bdi
->name
);
1456 inode
->dirtied_when
= jiffies
;
1458 inode
->dirtied_time_when
= jiffies
;
1460 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
1461 dirty_list
= &bdi
->wb
.b_dirty
;
1463 dirty_list
= &bdi
->wb
.b_dirty_time
;
1465 wakeup_bdi
= inode_wb_list_move_locked(inode
, &bdi
->wb
,
1468 spin_unlock(&bdi
->wb
.list_lock
);
1469 trace_writeback_dirty_inode_enqueue(inode
);
1472 * If this is the first dirty inode for this bdi,
1473 * we have to wake-up the corresponding bdi thread
1474 * to make sure background write-back happens
1477 if (bdi_cap_writeback_dirty(bdi
) && wakeup_bdi
)
1478 wb_wakeup_delayed(&bdi
->wb
);
1483 spin_unlock(&inode
->i_lock
);
1486 EXPORT_SYMBOL(__mark_inode_dirty
);
1488 static void wait_sb_inodes(struct super_block
*sb
)
1490 struct inode
*inode
, *old_inode
= NULL
;
1493 * We need to be protected against the filesystem going from
1494 * r/o to r/w or vice versa.
1496 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1498 spin_lock(&inode_sb_list_lock
);
1501 * Data integrity sync. Must wait for all pages under writeback,
1502 * because there may have been pages dirtied before our sync
1503 * call, but which had writeout started before we write it out.
1504 * In which case, the inode may not be on the dirty list, but
1505 * we still have to wait for that writeout.
1507 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1508 struct address_space
*mapping
= inode
->i_mapping
;
1510 spin_lock(&inode
->i_lock
);
1511 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1512 (mapping
->nrpages
== 0)) {
1513 spin_unlock(&inode
->i_lock
);
1517 spin_unlock(&inode
->i_lock
);
1518 spin_unlock(&inode_sb_list_lock
);
1521 * We hold a reference to 'inode' so it couldn't have been
1522 * removed from s_inodes list while we dropped the
1523 * inode_sb_list_lock. We cannot iput the inode now as we can
1524 * be holding the last reference and we cannot iput it under
1525 * inode_sb_list_lock. So we keep the reference and iput it
1531 filemap_fdatawait(mapping
);
1535 spin_lock(&inode_sb_list_lock
);
1537 spin_unlock(&inode_sb_list_lock
);
1542 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1543 * @sb: the superblock
1544 * @nr: the number of pages to write
1545 * @reason: reason why some writeback work initiated
1547 * Start writeback on some inodes on this super_block. No guarantees are made
1548 * on how many (if any) will be written, and this function does not wait
1549 * for IO completion of submitted IO.
1551 void writeback_inodes_sb_nr(struct super_block
*sb
,
1553 enum wb_reason reason
)
1555 DEFINE_WB_COMPLETION_ONSTACK(done
);
1556 struct wb_writeback_work work
= {
1558 .sync_mode
= WB_SYNC_NONE
,
1559 .tagged_writepages
= 1,
1564 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1566 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1568 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1569 wb_queue_work(&bdi
->wb
, &work
);
1570 wb_wait_for_completion(bdi
, &done
);
1572 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1575 * writeback_inodes_sb - writeback dirty inodes from given super_block
1576 * @sb: the superblock
1577 * @reason: reason why some writeback work was initiated
1579 * Start writeback on some inodes on this super_block. No guarantees are made
1580 * on how many (if any) will be written, and this function does not wait
1581 * for IO completion of submitted IO.
1583 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1585 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1587 EXPORT_SYMBOL(writeback_inodes_sb
);
1590 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1591 * @sb: the superblock
1592 * @nr: the number of pages to write
1593 * @reason: the reason of writeback
1595 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1596 * Returns 1 if writeback was started, 0 if not.
1598 int try_to_writeback_inodes_sb_nr(struct super_block
*sb
,
1600 enum wb_reason reason
)
1602 if (writeback_in_progress(&sb
->s_bdi
->wb
))
1605 if (!down_read_trylock(&sb
->s_umount
))
1608 writeback_inodes_sb_nr(sb
, nr
, reason
);
1609 up_read(&sb
->s_umount
);
1612 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1615 * try_to_writeback_inodes_sb - try to start writeback if none underway
1616 * @sb: the superblock
1617 * @reason: reason why some writeback work was initiated
1619 * Implement by try_to_writeback_inodes_sb_nr()
1620 * Returns 1 if writeback was started, 0 if not.
1622 int try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1624 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1626 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1629 * sync_inodes_sb - sync sb inode pages
1630 * @sb: the superblock
1632 * This function writes and waits on any dirty inode belonging to this
1635 void sync_inodes_sb(struct super_block
*sb
)
1637 DEFINE_WB_COMPLETION_ONSTACK(done
);
1638 struct wb_writeback_work work
= {
1640 .sync_mode
= WB_SYNC_ALL
,
1641 .nr_pages
= LONG_MAX
,
1644 .reason
= WB_REASON_SYNC
,
1647 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1649 /* Nothing to do? */
1650 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1652 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1654 wb_queue_work(&bdi
->wb
, &work
);
1655 wb_wait_for_completion(bdi
, &done
);
1659 EXPORT_SYMBOL(sync_inodes_sb
);
1662 * write_inode_now - write an inode to disk
1663 * @inode: inode to write to disk
1664 * @sync: whether the write should be synchronous or not
1666 * This function commits an inode to disk immediately if it is dirty. This is
1667 * primarily needed by knfsd.
1669 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1671 int write_inode_now(struct inode
*inode
, int sync
)
1673 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1674 struct writeback_control wbc
= {
1675 .nr_to_write
= LONG_MAX
,
1676 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1678 .range_end
= LLONG_MAX
,
1681 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1682 wbc
.nr_to_write
= 0;
1685 return writeback_single_inode(inode
, wb
, &wbc
);
1687 EXPORT_SYMBOL(write_inode_now
);
1690 * sync_inode - write an inode and its pages to disk.
1691 * @inode: the inode to sync
1692 * @wbc: controls the writeback mode
1694 * sync_inode() will write an inode and its pages to disk. It will also
1695 * correctly update the inode on its superblock's dirty inode lists and will
1696 * update inode->i_state.
1698 * The caller must have a ref on the inode.
1700 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1702 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1704 EXPORT_SYMBOL(sync_inode
);
1707 * sync_inode_metadata - write an inode to disk
1708 * @inode: the inode to sync
1709 * @wait: wait for I/O to complete.
1711 * Write an inode to disk and adjust its dirty state after completion.
1713 * Note: only writes the actual inode, no associated data or other metadata.
1715 int sync_inode_metadata(struct inode
*inode
, int wait
)
1717 struct writeback_control wbc
= {
1718 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1719 .nr_to_write
= 0, /* metadata-only */
1722 return sync_inode(inode
, &wbc
);
1724 EXPORT_SYMBOL(sync_inode_metadata
);