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>
30 #include <linux/memcontrol.h>
34 * 4MB minimal write chunk size
36 #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
38 struct wb_completion
{
43 * Passed into wb_writeback(), essentially a subset of writeback_control
45 struct wb_writeback_work
{
47 struct super_block
*sb
;
48 unsigned long *older_than_this
;
49 enum writeback_sync_modes sync_mode
;
50 unsigned int tagged_writepages
:1;
51 unsigned int for_kupdate
:1;
52 unsigned int range_cyclic
:1;
53 unsigned int for_background
:1;
54 unsigned int for_sync
:1; /* sync(2) WB_SYNC_ALL writeback */
55 unsigned int auto_free
:1; /* free on completion */
56 unsigned int single_wait
:1;
57 unsigned int single_done
:1;
58 enum wb_reason reason
; /* why was writeback initiated? */
60 struct list_head list
; /* pending work list */
61 struct wb_completion
*done
; /* set if the caller waits */
65 * If one wants to wait for one or more wb_writeback_works, each work's
66 * ->done should be set to a wb_completion defined using the following
67 * macro. Once all work items are issued with wb_queue_work(), the caller
68 * can wait for the completion of all using wb_wait_for_completion(). Work
69 * items which are waited upon aren't freed automatically on completion.
71 #define DEFINE_WB_COMPLETION_ONSTACK(cmpl) \
72 struct wb_completion cmpl = { \
73 .cnt = ATOMIC_INIT(1), \
78 * If an inode is constantly having its pages dirtied, but then the
79 * updates stop dirtytime_expire_interval seconds in the past, it's
80 * possible for the worst case time between when an inode has its
81 * timestamps updated and when they finally get written out to be two
82 * dirtytime_expire_intervals. We set the default to 12 hours (in
83 * seconds), which means most of the time inodes will have their
84 * timestamps written to disk after 12 hours, but in the worst case a
85 * few inodes might not their timestamps updated for 24 hours.
87 unsigned int dirtytime_expire_interval
= 12 * 60 * 60;
89 static inline struct inode
*wb_inode(struct list_head
*head
)
91 return list_entry(head
, struct inode
, i_wb_list
);
95 * Include the creation of the trace points after defining the
96 * wb_writeback_work structure and inline functions so that the definition
97 * remains local to this file.
99 #define CREATE_TRACE_POINTS
100 #include <trace/events/writeback.h>
102 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage
);
104 static bool wb_io_lists_populated(struct bdi_writeback
*wb
)
106 if (wb_has_dirty_io(wb
)) {
109 set_bit(WB_has_dirty_io
, &wb
->state
);
110 WARN_ON_ONCE(!wb
->avg_write_bandwidth
);
111 atomic_long_add(wb
->avg_write_bandwidth
,
112 &wb
->bdi
->tot_write_bandwidth
);
117 static void wb_io_lists_depopulated(struct bdi_writeback
*wb
)
119 if (wb_has_dirty_io(wb
) && list_empty(&wb
->b_dirty
) &&
120 list_empty(&wb
->b_io
) && list_empty(&wb
->b_more_io
)) {
121 clear_bit(WB_has_dirty_io
, &wb
->state
);
122 WARN_ON_ONCE(atomic_long_sub_return(wb
->avg_write_bandwidth
,
123 &wb
->bdi
->tot_write_bandwidth
) < 0);
128 * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list
129 * @inode: inode to be moved
130 * @wb: target bdi_writeback
131 * @head: one of @wb->b_{dirty|io|more_io}
133 * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io.
134 * Returns %true if @inode is the first occupant of the !dirty_time IO
135 * lists; otherwise, %false.
137 static bool inode_wb_list_move_locked(struct inode
*inode
,
138 struct bdi_writeback
*wb
,
139 struct list_head
*head
)
141 assert_spin_locked(&wb
->list_lock
);
143 list_move(&inode
->i_wb_list
, head
);
145 /* dirty_time doesn't count as dirty_io until expiration */
146 if (head
!= &wb
->b_dirty_time
)
147 return wb_io_lists_populated(wb
);
149 wb_io_lists_depopulated(wb
);
154 * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list
155 * @inode: inode to be removed
156 * @wb: bdi_writeback @inode is being removed from
158 * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and
159 * clear %WB_has_dirty_io if all are empty afterwards.
161 static void inode_wb_list_del_locked(struct inode
*inode
,
162 struct bdi_writeback
*wb
)
164 assert_spin_locked(&wb
->list_lock
);
166 list_del_init(&inode
->i_wb_list
);
167 wb_io_lists_depopulated(wb
);
170 static void wb_wakeup(struct bdi_writeback
*wb
)
172 spin_lock_bh(&wb
->work_lock
);
173 if (test_bit(WB_registered
, &wb
->state
))
174 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
175 spin_unlock_bh(&wb
->work_lock
);
178 static void wb_queue_work(struct bdi_writeback
*wb
,
179 struct wb_writeback_work
*work
)
181 trace_writeback_queue(wb
->bdi
, work
);
183 spin_lock_bh(&wb
->work_lock
);
184 if (!test_bit(WB_registered
, &wb
->state
)) {
185 if (work
->single_wait
)
186 work
->single_done
= 1;
190 atomic_inc(&work
->done
->cnt
);
191 list_add_tail(&work
->list
, &wb
->work_list
);
192 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
194 spin_unlock_bh(&wb
->work_lock
);
198 * wb_wait_for_completion - wait for completion of bdi_writeback_works
199 * @bdi: bdi work items were issued to
200 * @done: target wb_completion
202 * Wait for one or more work items issued to @bdi with their ->done field
203 * set to @done, which should have been defined with
204 * DEFINE_WB_COMPLETION_ONSTACK(). This function returns after all such
205 * work items are completed. Work items which are waited upon aren't freed
206 * automatically on completion.
208 static void wb_wait_for_completion(struct backing_dev_info
*bdi
,
209 struct wb_completion
*done
)
211 atomic_dec(&done
->cnt
); /* put down the initial count */
212 wait_event(bdi
->wb_waitq
, !atomic_read(&done
->cnt
));
215 #ifdef CONFIG_CGROUP_WRITEBACK
217 void __inode_attach_wb(struct inode
*inode
, struct page
*page
)
219 struct backing_dev_info
*bdi
= inode_to_bdi(inode
);
220 struct bdi_writeback
*wb
= NULL
;
222 if (inode_cgwb_enabled(inode
)) {
223 struct cgroup_subsys_state
*memcg_css
;
226 memcg_css
= mem_cgroup_css_from_page(page
);
227 wb
= wb_get_create(bdi
, memcg_css
, GFP_ATOMIC
);
229 /* must pin memcg_css, see wb_get_create() */
230 memcg_css
= task_get_css(current
, memory_cgrp_id
);
231 wb
= wb_get_create(bdi
, memcg_css
, GFP_ATOMIC
);
240 * There may be multiple instances of this function racing to
241 * update the same inode. Use cmpxchg() to tell the winner.
243 if (unlikely(cmpxchg(&inode
->i_wb
, NULL
, wb
)))
248 * inode_congested - test whether an inode is congested
249 * @inode: inode to test for congestion
250 * @cong_bits: mask of WB_[a]sync_congested bits to test
252 * Tests whether @inode is congested. @cong_bits is the mask of congestion
253 * bits to test and the return value is the mask of set bits.
255 * If cgroup writeback is enabled for @inode, the congestion state is
256 * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
257 * associated with @inode is congested; otherwise, the root wb's congestion
260 int inode_congested(struct inode
*inode
, int cong_bits
)
263 struct bdi_writeback
*wb
= inode_to_wb(inode
);
265 return wb_congested(wb
, cong_bits
);
268 return wb_congested(&inode_to_bdi(inode
)->wb
, cong_bits
);
270 EXPORT_SYMBOL_GPL(inode_congested
);
273 * wb_wait_for_single_work - wait for completion of a single bdi_writeback_work
274 * @bdi: bdi the work item was issued to
275 * @work: work item to wait for
277 * Wait for the completion of @work which was issued to one of @bdi's
278 * bdi_writeback's. The caller must have set @work->single_wait before
279 * issuing it. This wait operates independently fo
280 * wb_wait_for_completion() and also disables automatic freeing of @work.
282 static void wb_wait_for_single_work(struct backing_dev_info
*bdi
,
283 struct wb_writeback_work
*work
)
285 if (WARN_ON_ONCE(!work
->single_wait
))
288 wait_event(bdi
->wb_waitq
, work
->single_done
);
291 * Paired with smp_wmb() in wb_do_writeback() and ensures that all
292 * modifications to @work prior to assertion of ->single_done is
293 * visible to the caller once this function returns.
299 * wb_split_bdi_pages - split nr_pages to write according to bandwidth
300 * @wb: target bdi_writeback to split @nr_pages to
301 * @nr_pages: number of pages to write for the whole bdi
303 * Split @wb's portion of @nr_pages according to @wb's write bandwidth in
304 * relation to the total write bandwidth of all wb's w/ dirty inodes on
307 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
309 unsigned long this_bw
= wb
->avg_write_bandwidth
;
310 unsigned long tot_bw
= atomic_long_read(&wb
->bdi
->tot_write_bandwidth
);
312 if (nr_pages
== LONG_MAX
)
316 * This may be called on clean wb's and proportional distribution
317 * may not make sense, just use the original @nr_pages in those
318 * cases. In general, we wanna err on the side of writing more.
320 if (!tot_bw
|| this_bw
>= tot_bw
)
323 return DIV_ROUND_UP_ULL((u64
)nr_pages
* this_bw
, tot_bw
);
327 * wb_clone_and_queue_work - clone a wb_writeback_work and issue it to a wb
328 * @wb: target bdi_writeback
329 * @base_work: source wb_writeback_work
331 * Try to make a clone of @base_work and issue it to @wb. If cloning
332 * succeeds, %true is returned; otherwise, @base_work is issued directly
333 * and %false is returned. In the latter case, the caller is required to
334 * wait for @base_work's completion using wb_wait_for_single_work().
336 * A clone is auto-freed on completion. @base_work never is.
338 static bool wb_clone_and_queue_work(struct bdi_writeback
*wb
,
339 struct wb_writeback_work
*base_work
)
341 struct wb_writeback_work
*work
;
343 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
347 work
->single_wait
= 0;
351 work
->single_wait
= 1;
353 work
->single_done
= 0;
354 wb_queue_work(wb
, work
);
355 return work
!= base_work
;
359 * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi
360 * @bdi: target backing_dev_info
361 * @base_work: wb_writeback_work to issue
362 * @skip_if_busy: skip wb's which already have writeback in progress
364 * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which
365 * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's
366 * distributed to the busy wbs according to each wb's proportion in the
367 * total active write bandwidth of @bdi.
369 static void bdi_split_work_to_wbs(struct backing_dev_info
*bdi
,
370 struct wb_writeback_work
*base_work
,
373 long nr_pages
= base_work
->nr_pages
;
374 int next_blkcg_id
= 0;
375 struct bdi_writeback
*wb
;
380 if (!bdi_has_dirty_io(bdi
))
384 bdi_for_each_wb(wb
, bdi
, &iter
, next_blkcg_id
) {
385 if (!wb_has_dirty_io(wb
) ||
386 (skip_if_busy
&& writeback_in_progress(wb
)))
389 base_work
->nr_pages
= wb_split_bdi_pages(wb
, nr_pages
);
390 if (!wb_clone_and_queue_work(wb
, base_work
)) {
391 next_blkcg_id
= wb
->blkcg_css
->id
+ 1;
393 wb_wait_for_single_work(bdi
, base_work
);
400 #else /* CONFIG_CGROUP_WRITEBACK */
402 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
407 static void bdi_split_work_to_wbs(struct backing_dev_info
*bdi
,
408 struct wb_writeback_work
*base_work
,
413 if (bdi_has_dirty_io(bdi
) &&
414 (!skip_if_busy
|| !writeback_in_progress(&bdi
->wb
))) {
415 base_work
->auto_free
= 0;
416 base_work
->single_wait
= 0;
417 base_work
->single_done
= 0;
418 wb_queue_work(&bdi
->wb
, base_work
);
422 #endif /* CONFIG_CGROUP_WRITEBACK */
424 void wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
425 bool range_cyclic
, enum wb_reason reason
)
427 struct wb_writeback_work
*work
;
429 if (!wb_has_dirty_io(wb
))
433 * This is WB_SYNC_NONE writeback, so if allocation fails just
434 * wakeup the thread for old dirty data writeback
436 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
438 trace_writeback_nowork(wb
->bdi
);
443 work
->sync_mode
= WB_SYNC_NONE
;
444 work
->nr_pages
= nr_pages
;
445 work
->range_cyclic
= range_cyclic
;
446 work
->reason
= reason
;
449 wb_queue_work(wb
, work
);
453 * wb_start_background_writeback - start background writeback
454 * @wb: bdi_writback to write from
457 * This makes sure WB_SYNC_NONE background writeback happens. When
458 * this function returns, it is only guaranteed that for given wb
459 * some IO is happening if we are over background dirty threshold.
460 * Caller need not hold sb s_umount semaphore.
462 void wb_start_background_writeback(struct bdi_writeback
*wb
)
465 * We just wake up the flusher thread. It will perform background
466 * writeback as soon as there is no other work to do.
468 trace_writeback_wake_background(wb
->bdi
);
473 * Remove the inode from the writeback list it is on.
475 void inode_wb_list_del(struct inode
*inode
)
477 struct bdi_writeback
*wb
= inode_to_wb(inode
);
479 spin_lock(&wb
->list_lock
);
480 inode_wb_list_del_locked(inode
, wb
);
481 spin_unlock(&wb
->list_lock
);
485 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
486 * furthest end of its superblock's dirty-inode list.
488 * Before stamping the inode's ->dirtied_when, we check to see whether it is
489 * already the most-recently-dirtied inode on the b_dirty list. If that is
490 * the case then the inode must have been redirtied while it was being written
491 * out and we don't reset its dirtied_when.
493 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
495 if (!list_empty(&wb
->b_dirty
)) {
498 tail
= wb_inode(wb
->b_dirty
.next
);
499 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
500 inode
->dirtied_when
= jiffies
;
502 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty
);
506 * requeue inode for re-scanning after bdi->b_io list is exhausted.
508 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
510 inode_wb_list_move_locked(inode
, wb
, &wb
->b_more_io
);
513 static void inode_sync_complete(struct inode
*inode
)
515 inode
->i_state
&= ~I_SYNC
;
516 /* If inode is clean an unused, put it into LRU now... */
517 inode_add_lru(inode
);
518 /* Waiters must see I_SYNC cleared before being woken up */
520 wake_up_bit(&inode
->i_state
, __I_SYNC
);
523 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
525 bool ret
= time_after(inode
->dirtied_when
, t
);
528 * For inodes being constantly redirtied, dirtied_when can get stuck.
529 * It _appears_ to be in the future, but is actually in distant past.
530 * This test is necessary to prevent such wrapped-around relative times
531 * from permanently stopping the whole bdi writeback.
533 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
538 #define EXPIRE_DIRTY_ATIME 0x0001
541 * Move expired (dirtied before work->older_than_this) dirty inodes from
542 * @delaying_queue to @dispatch_queue.
544 static int move_expired_inodes(struct list_head
*delaying_queue
,
545 struct list_head
*dispatch_queue
,
547 struct wb_writeback_work
*work
)
549 unsigned long *older_than_this
= NULL
;
550 unsigned long expire_time
;
552 struct list_head
*pos
, *node
;
553 struct super_block
*sb
= NULL
;
558 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
559 older_than_this
= work
->older_than_this
;
560 else if (!work
->for_sync
) {
561 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
562 older_than_this
= &expire_time
;
564 while (!list_empty(delaying_queue
)) {
565 inode
= wb_inode(delaying_queue
->prev
);
566 if (older_than_this
&&
567 inode_dirtied_after(inode
, *older_than_this
))
569 list_move(&inode
->i_wb_list
, &tmp
);
571 if (flags
& EXPIRE_DIRTY_ATIME
)
572 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
573 if (sb_is_blkdev_sb(inode
->i_sb
))
575 if (sb
&& sb
!= inode
->i_sb
)
580 /* just one sb in list, splice to dispatch_queue and we're done */
582 list_splice(&tmp
, dispatch_queue
);
586 /* Move inodes from one superblock together */
587 while (!list_empty(&tmp
)) {
588 sb
= wb_inode(tmp
.prev
)->i_sb
;
589 list_for_each_prev_safe(pos
, node
, &tmp
) {
590 inode
= wb_inode(pos
);
591 if (inode
->i_sb
== sb
)
592 list_move(&inode
->i_wb_list
, dispatch_queue
);
600 * Queue all expired dirty inodes for io, eldest first.
602 * newly dirtied b_dirty b_io b_more_io
603 * =============> gf edc BA
605 * newly dirtied b_dirty b_io b_more_io
606 * =============> g fBAedc
608 * +--> dequeue for IO
610 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
614 assert_spin_locked(&wb
->list_lock
);
615 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
616 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
617 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
618 EXPIRE_DIRTY_ATIME
, work
);
620 wb_io_lists_populated(wb
);
621 trace_writeback_queue_io(wb
, work
, moved
);
624 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
628 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
629 trace_writeback_write_inode_start(inode
, wbc
);
630 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
631 trace_writeback_write_inode(inode
, wbc
);
638 * Wait for writeback on an inode to complete. Called with i_lock held.
639 * Caller must make sure inode cannot go away when we drop i_lock.
641 static void __inode_wait_for_writeback(struct inode
*inode
)
642 __releases(inode
->i_lock
)
643 __acquires(inode
->i_lock
)
645 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
646 wait_queue_head_t
*wqh
;
648 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
649 while (inode
->i_state
& I_SYNC
) {
650 spin_unlock(&inode
->i_lock
);
651 __wait_on_bit(wqh
, &wq
, bit_wait
,
652 TASK_UNINTERRUPTIBLE
);
653 spin_lock(&inode
->i_lock
);
658 * Wait for writeback on an inode to complete. Caller must have inode pinned.
660 void inode_wait_for_writeback(struct inode
*inode
)
662 spin_lock(&inode
->i_lock
);
663 __inode_wait_for_writeback(inode
);
664 spin_unlock(&inode
->i_lock
);
668 * Sleep until I_SYNC is cleared. This function must be called with i_lock
669 * held and drops it. It is aimed for callers not holding any inode reference
670 * so once i_lock is dropped, inode can go away.
672 static void inode_sleep_on_writeback(struct inode
*inode
)
673 __releases(inode
->i_lock
)
676 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
679 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
680 sleep
= inode
->i_state
& I_SYNC
;
681 spin_unlock(&inode
->i_lock
);
684 finish_wait(wqh
, &wait
);
688 * Find proper writeback list for the inode depending on its current state and
689 * possibly also change of its state while we were doing writeback. Here we
690 * handle things such as livelock prevention or fairness of writeback among
691 * inodes. This function can be called only by flusher thread - noone else
692 * processes all inodes in writeback lists and requeueing inodes behind flusher
693 * thread's back can have unexpected consequences.
695 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
696 struct writeback_control
*wbc
)
698 if (inode
->i_state
& I_FREEING
)
702 * Sync livelock prevention. Each inode is tagged and synced in one
703 * shot. If still dirty, it will be redirty_tail()'ed below. Update
704 * the dirty time to prevent enqueue and sync it again.
706 if ((inode
->i_state
& I_DIRTY
) &&
707 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
708 inode
->dirtied_when
= jiffies
;
710 if (wbc
->pages_skipped
) {
712 * writeback is not making progress due to locked
713 * buffers. Skip this inode for now.
715 redirty_tail(inode
, wb
);
719 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
721 * We didn't write back all the pages. nfs_writepages()
722 * sometimes bales out without doing anything.
724 if (wbc
->nr_to_write
<= 0) {
725 /* Slice used up. Queue for next turn. */
726 requeue_io(inode
, wb
);
729 * Writeback blocked by something other than
730 * congestion. Delay the inode for some time to
731 * avoid spinning on the CPU (100% iowait)
732 * retrying writeback of the dirty page/inode
733 * that cannot be performed immediately.
735 redirty_tail(inode
, wb
);
737 } else if (inode
->i_state
& I_DIRTY
) {
739 * Filesystems can dirty the inode during writeback operations,
740 * such as delayed allocation during submission or metadata
741 * updates after data IO completion.
743 redirty_tail(inode
, wb
);
744 } else if (inode
->i_state
& I_DIRTY_TIME
) {
745 inode
->dirtied_when
= jiffies
;
746 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty_time
);
748 /* The inode is clean. Remove from writeback lists. */
749 inode_wb_list_del_locked(inode
, wb
);
754 * Write out an inode and its dirty pages. Do not update the writeback list
755 * linkage. That is left to the caller. The caller is also responsible for
756 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
759 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
761 struct address_space
*mapping
= inode
->i_mapping
;
762 long nr_to_write
= wbc
->nr_to_write
;
766 WARN_ON(!(inode
->i_state
& I_SYNC
));
768 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
770 ret
= do_writepages(mapping
, wbc
);
773 * Make sure to wait on the data before writing out the metadata.
774 * This is important for filesystems that modify metadata on data
775 * I/O completion. We don't do it for sync(2) writeback because it has a
776 * separate, external IO completion path and ->sync_fs for guaranteeing
777 * inode metadata is written back correctly.
779 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
780 int err
= filemap_fdatawait(mapping
);
786 * Some filesystems may redirty the inode during the writeback
787 * due to delalloc, clear dirty metadata flags right before
790 spin_lock(&inode
->i_lock
);
792 dirty
= inode
->i_state
& I_DIRTY
;
793 if (inode
->i_state
& I_DIRTY_TIME
) {
794 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
795 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
796 unlikely(time_after(jiffies
,
797 (inode
->dirtied_time_when
+
798 dirtytime_expire_interval
* HZ
)))) {
799 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
800 trace_writeback_lazytime(inode
);
803 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
804 inode
->i_state
&= ~dirty
;
807 * Paired with smp_mb() in __mark_inode_dirty(). This allows
808 * __mark_inode_dirty() to test i_state without grabbing i_lock -
809 * either they see the I_DIRTY bits cleared or we see the dirtied
812 * I_DIRTY_PAGES is always cleared together above even if @mapping
813 * still has dirty pages. The flag is reinstated after smp_mb() if
814 * necessary. This guarantees that either __mark_inode_dirty()
815 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
819 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
820 inode
->i_state
|= I_DIRTY_PAGES
;
822 spin_unlock(&inode
->i_lock
);
824 if (dirty
& I_DIRTY_TIME
)
825 mark_inode_dirty_sync(inode
);
826 /* Don't write the inode if only I_DIRTY_PAGES was set */
827 if (dirty
& ~I_DIRTY_PAGES
) {
828 int err
= write_inode(inode
, wbc
);
832 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
837 * Write out an inode's dirty pages. Either the caller has an active reference
838 * on the inode or the inode has I_WILL_FREE set.
840 * This function is designed to be called for writing back one inode which
841 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
842 * and does more profound writeback list handling in writeback_sb_inodes().
845 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
846 struct writeback_control
*wbc
)
850 spin_lock(&inode
->i_lock
);
851 if (!atomic_read(&inode
->i_count
))
852 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
854 WARN_ON(inode
->i_state
& I_WILL_FREE
);
856 if (inode
->i_state
& I_SYNC
) {
857 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
860 * It's a data-integrity sync. We must wait. Since callers hold
861 * inode reference or inode has I_WILL_FREE set, it cannot go
864 __inode_wait_for_writeback(inode
);
866 WARN_ON(inode
->i_state
& I_SYNC
);
868 * Skip inode if it is clean and we have no outstanding writeback in
869 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
870 * function since flusher thread may be doing for example sync in
871 * parallel and if we move the inode, it could get skipped. So here we
872 * make sure inode is on some writeback list and leave it there unless
873 * we have completely cleaned the inode.
875 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
876 (wbc
->sync_mode
!= WB_SYNC_ALL
||
877 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
879 inode
->i_state
|= I_SYNC
;
880 spin_unlock(&inode
->i_lock
);
882 ret
= __writeback_single_inode(inode
, wbc
);
884 spin_lock(&wb
->list_lock
);
885 spin_lock(&inode
->i_lock
);
887 * If inode is clean, remove it from writeback lists. Otherwise don't
888 * touch it. See comment above for explanation.
890 if (!(inode
->i_state
& I_DIRTY_ALL
))
891 inode_wb_list_del_locked(inode
, wb
);
892 spin_unlock(&wb
->list_lock
);
893 inode_sync_complete(inode
);
895 spin_unlock(&inode
->i_lock
);
899 static long writeback_chunk_size(struct bdi_writeback
*wb
,
900 struct wb_writeback_work
*work
)
905 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
906 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
907 * here avoids calling into writeback_inodes_wb() more than once.
909 * The intended call sequence for WB_SYNC_ALL writeback is:
912 * writeback_sb_inodes() <== called only once
913 * write_cache_pages() <== called once for each inode
914 * (quickly) tag currently dirty pages
915 * (maybe slowly) sync all tagged pages
917 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
920 pages
= min(wb
->avg_write_bandwidth
/ 2,
921 global_wb_domain
.dirty_limit
/ DIRTY_SCOPE
);
922 pages
= min(pages
, work
->nr_pages
);
923 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
924 MIN_WRITEBACK_PAGES
);
931 * Write a portion of b_io inodes which belong to @sb.
933 * Return the number of pages and/or inodes written.
935 static long writeback_sb_inodes(struct super_block
*sb
,
936 struct bdi_writeback
*wb
,
937 struct wb_writeback_work
*work
)
939 struct writeback_control wbc
= {
940 .sync_mode
= work
->sync_mode
,
941 .tagged_writepages
= work
->tagged_writepages
,
942 .for_kupdate
= work
->for_kupdate
,
943 .for_background
= work
->for_background
,
944 .for_sync
= work
->for_sync
,
945 .range_cyclic
= work
->range_cyclic
,
947 .range_end
= LLONG_MAX
,
949 unsigned long start_time
= jiffies
;
951 long wrote
= 0; /* count both pages and inodes */
953 while (!list_empty(&wb
->b_io
)) {
954 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
956 if (inode
->i_sb
!= sb
) {
959 * We only want to write back data for this
960 * superblock, move all inodes not belonging
961 * to it back onto the dirty list.
963 redirty_tail(inode
, wb
);
968 * The inode belongs to a different superblock.
969 * Bounce back to the caller to unpin this and
970 * pin the next superblock.
976 * Don't bother with new inodes or inodes being freed, first
977 * kind does not need periodic writeout yet, and for the latter
978 * kind writeout is handled by the freer.
980 spin_lock(&inode
->i_lock
);
981 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
982 spin_unlock(&inode
->i_lock
);
983 redirty_tail(inode
, wb
);
986 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
988 * If this inode is locked for writeback and we are not
989 * doing writeback-for-data-integrity, move it to
990 * b_more_io so that writeback can proceed with the
991 * other inodes on s_io.
993 * We'll have another go at writing back this inode
994 * when we completed a full scan of b_io.
996 spin_unlock(&inode
->i_lock
);
997 requeue_io(inode
, wb
);
998 trace_writeback_sb_inodes_requeue(inode
);
1001 spin_unlock(&wb
->list_lock
);
1004 * We already requeued the inode if it had I_SYNC set and we
1005 * are doing WB_SYNC_NONE writeback. So this catches only the
1008 if (inode
->i_state
& I_SYNC
) {
1009 /* Wait for I_SYNC. This function drops i_lock... */
1010 inode_sleep_on_writeback(inode
);
1011 /* Inode may be gone, start again */
1012 spin_lock(&wb
->list_lock
);
1015 inode
->i_state
|= I_SYNC
;
1016 spin_unlock(&inode
->i_lock
);
1018 write_chunk
= writeback_chunk_size(wb
, work
);
1019 wbc
.nr_to_write
= write_chunk
;
1020 wbc
.pages_skipped
= 0;
1023 * We use I_SYNC to pin the inode in memory. While it is set
1024 * evict_inode() will wait so the inode cannot be freed.
1026 __writeback_single_inode(inode
, &wbc
);
1028 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
1029 wrote
+= write_chunk
- wbc
.nr_to_write
;
1030 spin_lock(&wb
->list_lock
);
1031 spin_lock(&inode
->i_lock
);
1032 if (!(inode
->i_state
& I_DIRTY_ALL
))
1034 requeue_inode(inode
, wb
, &wbc
);
1035 inode_sync_complete(inode
);
1036 spin_unlock(&inode
->i_lock
);
1037 cond_resched_lock(&wb
->list_lock
);
1039 * bail out to wb_writeback() often enough to check
1040 * background threshold and other termination conditions.
1043 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
1045 if (work
->nr_pages
<= 0)
1052 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
1053 struct wb_writeback_work
*work
)
1055 unsigned long start_time
= jiffies
;
1058 while (!list_empty(&wb
->b_io
)) {
1059 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
1060 struct super_block
*sb
= inode
->i_sb
;
1062 if (!trylock_super(sb
)) {
1064 * trylock_super() may fail consistently due to
1065 * s_umount being grabbed by someone else. Don't use
1066 * requeue_io() to avoid busy retrying the inode/sb.
1068 redirty_tail(inode
, wb
);
1071 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
1072 up_read(&sb
->s_umount
);
1074 /* refer to the same tests at the end of writeback_sb_inodes */
1076 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
1078 if (work
->nr_pages
<= 0)
1082 /* Leave any unwritten inodes on b_io */
1086 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
1087 enum wb_reason reason
)
1089 struct wb_writeback_work work
= {
1090 .nr_pages
= nr_pages
,
1091 .sync_mode
= WB_SYNC_NONE
,
1096 spin_lock(&wb
->list_lock
);
1097 if (list_empty(&wb
->b_io
))
1098 queue_io(wb
, &work
);
1099 __writeback_inodes_wb(wb
, &work
);
1100 spin_unlock(&wb
->list_lock
);
1102 return nr_pages
- work
.nr_pages
;
1106 * Explicit flushing or periodic writeback of "old" data.
1108 * Define "old": the first time one of an inode's pages is dirtied, we mark the
1109 * dirtying-time in the inode's address_space. So this periodic writeback code
1110 * just walks the superblock inode list, writing back any inodes which are
1111 * older than a specific point in time.
1113 * Try to run once per dirty_writeback_interval. But if a writeback event
1114 * takes longer than a dirty_writeback_interval interval, then leave a
1117 * older_than_this takes precedence over nr_to_write. So we'll only write back
1118 * all dirty pages if they are all attached to "old" mappings.
1120 static long wb_writeback(struct bdi_writeback
*wb
,
1121 struct wb_writeback_work
*work
)
1123 unsigned long wb_start
= jiffies
;
1124 long nr_pages
= work
->nr_pages
;
1125 unsigned long oldest_jif
;
1126 struct inode
*inode
;
1129 oldest_jif
= jiffies
;
1130 work
->older_than_this
= &oldest_jif
;
1132 spin_lock(&wb
->list_lock
);
1135 * Stop writeback when nr_pages has been consumed
1137 if (work
->nr_pages
<= 0)
1141 * Background writeout and kupdate-style writeback may
1142 * run forever. Stop them if there is other work to do
1143 * so that e.g. sync can proceed. They'll be restarted
1144 * after the other works are all done.
1146 if ((work
->for_background
|| work
->for_kupdate
) &&
1147 !list_empty(&wb
->work_list
))
1151 * For background writeout, stop when we are below the
1152 * background dirty threshold
1154 if (work
->for_background
&& !wb_over_bg_thresh(wb
))
1158 * Kupdate and background works are special and we want to
1159 * include all inodes that need writing. Livelock avoidance is
1160 * handled by these works yielding to any other work so we are
1163 if (work
->for_kupdate
) {
1164 oldest_jif
= jiffies
-
1165 msecs_to_jiffies(dirty_expire_interval
* 10);
1166 } else if (work
->for_background
)
1167 oldest_jif
= jiffies
;
1169 trace_writeback_start(wb
->bdi
, work
);
1170 if (list_empty(&wb
->b_io
))
1173 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
1175 progress
= __writeback_inodes_wb(wb
, work
);
1176 trace_writeback_written(wb
->bdi
, work
);
1178 wb_update_bandwidth(wb
, wb_start
);
1181 * Did we write something? Try for more
1183 * Dirty inodes are moved to b_io for writeback in batches.
1184 * The completion of the current batch does not necessarily
1185 * mean the overall work is done. So we keep looping as long
1186 * as made some progress on cleaning pages or inodes.
1191 * No more inodes for IO, bail
1193 if (list_empty(&wb
->b_more_io
))
1196 * Nothing written. Wait for some inode to
1197 * become available for writeback. Otherwise
1198 * we'll just busyloop.
1200 if (!list_empty(&wb
->b_more_io
)) {
1201 trace_writeback_wait(wb
->bdi
, work
);
1202 inode
= wb_inode(wb
->b_more_io
.prev
);
1203 spin_lock(&inode
->i_lock
);
1204 spin_unlock(&wb
->list_lock
);
1205 /* This function drops i_lock... */
1206 inode_sleep_on_writeback(inode
);
1207 spin_lock(&wb
->list_lock
);
1210 spin_unlock(&wb
->list_lock
);
1212 return nr_pages
- work
->nr_pages
;
1216 * Return the next wb_writeback_work struct that hasn't been processed yet.
1218 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
1220 struct wb_writeback_work
*work
= NULL
;
1222 spin_lock_bh(&wb
->work_lock
);
1223 if (!list_empty(&wb
->work_list
)) {
1224 work
= list_entry(wb
->work_list
.next
,
1225 struct wb_writeback_work
, list
);
1226 list_del_init(&work
->list
);
1228 spin_unlock_bh(&wb
->work_lock
);
1233 * Add in the number of potentially dirty inodes, because each inode
1234 * write can dirty pagecache in the underlying blockdev.
1236 static unsigned long get_nr_dirty_pages(void)
1238 return global_page_state(NR_FILE_DIRTY
) +
1239 global_page_state(NR_UNSTABLE_NFS
) +
1240 get_nr_dirty_inodes();
1243 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1245 if (wb_over_bg_thresh(wb
)) {
1247 struct wb_writeback_work work
= {
1248 .nr_pages
= LONG_MAX
,
1249 .sync_mode
= WB_SYNC_NONE
,
1250 .for_background
= 1,
1252 .reason
= WB_REASON_BACKGROUND
,
1255 return wb_writeback(wb
, &work
);
1261 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1263 unsigned long expired
;
1267 * When set to zero, disable periodic writeback
1269 if (!dirty_writeback_interval
)
1272 expired
= wb
->last_old_flush
+
1273 msecs_to_jiffies(dirty_writeback_interval
* 10);
1274 if (time_before(jiffies
, expired
))
1277 wb
->last_old_flush
= jiffies
;
1278 nr_pages
= get_nr_dirty_pages();
1281 struct wb_writeback_work work
= {
1282 .nr_pages
= nr_pages
,
1283 .sync_mode
= WB_SYNC_NONE
,
1286 .reason
= WB_REASON_PERIODIC
,
1289 return wb_writeback(wb
, &work
);
1296 * Retrieve work items and do the writeback they describe
1298 static long wb_do_writeback(struct bdi_writeback
*wb
)
1300 struct wb_writeback_work
*work
;
1303 set_bit(WB_writeback_running
, &wb
->state
);
1304 while ((work
= get_next_work_item(wb
)) != NULL
) {
1305 struct wb_completion
*done
= work
->done
;
1306 bool need_wake_up
= false;
1308 trace_writeback_exec(wb
->bdi
, work
);
1310 wrote
+= wb_writeback(wb
, work
);
1312 if (work
->single_wait
) {
1313 WARN_ON_ONCE(work
->auto_free
);
1314 /* paired w/ rmb in wb_wait_for_single_work() */
1316 work
->single_done
= 1;
1317 need_wake_up
= true;
1318 } else if (work
->auto_free
) {
1322 if (done
&& atomic_dec_and_test(&done
->cnt
))
1323 need_wake_up
= true;
1326 wake_up_all(&wb
->bdi
->wb_waitq
);
1330 * Check for periodic writeback, kupdated() style
1332 wrote
+= wb_check_old_data_flush(wb
);
1333 wrote
+= wb_check_background_flush(wb
);
1334 clear_bit(WB_writeback_running
, &wb
->state
);
1340 * Handle writeback of dirty data for the device backed by this bdi. Also
1341 * reschedules periodically and does kupdated style flushing.
1343 void wb_workfn(struct work_struct
*work
)
1345 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1346 struct bdi_writeback
, dwork
);
1349 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1350 current
->flags
|= PF_SWAPWRITE
;
1352 if (likely(!current_is_workqueue_rescuer() ||
1353 !test_bit(WB_registered
, &wb
->state
))) {
1355 * The normal path. Keep writing back @wb until its
1356 * work_list is empty. Note that this path is also taken
1357 * if @wb is shutting down even when we're running off the
1358 * rescuer as work_list needs to be drained.
1361 pages_written
= wb_do_writeback(wb
);
1362 trace_writeback_pages_written(pages_written
);
1363 } while (!list_empty(&wb
->work_list
));
1366 * bdi_wq can't get enough workers and we're running off
1367 * the emergency worker. Don't hog it. Hopefully, 1024 is
1368 * enough for efficient IO.
1370 pages_written
= writeback_inodes_wb(wb
, 1024,
1371 WB_REASON_FORKER_THREAD
);
1372 trace_writeback_pages_written(pages_written
);
1375 if (!list_empty(&wb
->work_list
))
1376 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1377 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1378 wb_wakeup_delayed(wb
);
1380 current
->flags
&= ~PF_SWAPWRITE
;
1384 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1387 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1389 struct backing_dev_info
*bdi
;
1392 nr_pages
= get_nr_dirty_pages();
1395 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1396 struct bdi_writeback
*wb
;
1397 struct wb_iter iter
;
1399 if (!bdi_has_dirty_io(bdi
))
1402 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1403 wb_start_writeback(wb
, wb_split_bdi_pages(wb
, nr_pages
),
1410 * Wake up bdi's periodically to make sure dirtytime inodes gets
1411 * written back periodically. We deliberately do *not* check the
1412 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1413 * kernel to be constantly waking up once there are any dirtytime
1414 * inodes on the system. So instead we define a separate delayed work
1415 * function which gets called much more rarely. (By default, only
1416 * once every 12 hours.)
1418 * If there is any other write activity going on in the file system,
1419 * this function won't be necessary. But if the only thing that has
1420 * happened on the file system is a dirtytime inode caused by an atime
1421 * update, we need this infrastructure below to make sure that inode
1422 * eventually gets pushed out to disk.
1424 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1425 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1427 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1429 struct backing_dev_info
*bdi
;
1432 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1433 struct bdi_writeback
*wb
;
1434 struct wb_iter iter
;
1436 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1437 if (!list_empty(&bdi
->wb
.b_dirty_time
))
1438 wb_wakeup(&bdi
->wb
);
1441 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1444 static int __init
start_dirtytime_writeback(void)
1446 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1449 __initcall(start_dirtytime_writeback
);
1451 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1452 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1456 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1457 if (ret
== 0 && write
)
1458 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1462 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1464 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1465 struct dentry
*dentry
;
1466 const char *name
= "?";
1468 dentry
= d_find_alias(inode
);
1470 spin_lock(&dentry
->d_lock
);
1471 name
= (const char *) dentry
->d_name
.name
;
1474 "%s(%d): dirtied inode %lu (%s) on %s\n",
1475 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1476 name
, inode
->i_sb
->s_id
);
1478 spin_unlock(&dentry
->d_lock
);
1485 * __mark_inode_dirty - internal function
1486 * @inode: inode to mark
1487 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1488 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1489 * mark_inode_dirty_sync.
1491 * Put the inode on the super block's dirty list.
1493 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1494 * dirty list only if it is hashed or if it refers to a blockdev.
1495 * If it was not hashed, it will never be added to the dirty list
1496 * even if it is later hashed, as it will have been marked dirty already.
1498 * In short, make sure you hash any inodes _before_ you start marking
1501 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1502 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1503 * the kernel-internal blockdev inode represents the dirtying time of the
1504 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1505 * page->mapping->host, so the page-dirtying time is recorded in the internal
1508 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1509 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1511 struct super_block
*sb
= inode
->i_sb
;
1514 trace_writeback_mark_inode_dirty(inode
, flags
);
1517 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1518 * dirty the inode itself
1520 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
1521 trace_writeback_dirty_inode_start(inode
, flags
);
1523 if (sb
->s_op
->dirty_inode
)
1524 sb
->s_op
->dirty_inode(inode
, flags
);
1526 trace_writeback_dirty_inode(inode
, flags
);
1528 if (flags
& I_DIRTY_INODE
)
1529 flags
&= ~I_DIRTY_TIME
;
1530 dirtytime
= flags
& I_DIRTY_TIME
;
1533 * Paired with smp_mb() in __writeback_single_inode() for the
1534 * following lockless i_state test. See there for details.
1538 if (((inode
->i_state
& flags
) == flags
) ||
1539 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
1542 if (unlikely(block_dump
))
1543 block_dump___mark_inode_dirty(inode
);
1545 spin_lock(&inode
->i_lock
);
1546 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
1547 goto out_unlock_inode
;
1548 if ((inode
->i_state
& flags
) != flags
) {
1549 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1551 inode_attach_wb(inode
, NULL
);
1553 if (flags
& I_DIRTY_INODE
)
1554 inode
->i_state
&= ~I_DIRTY_TIME
;
1555 inode
->i_state
|= flags
;
1558 * If the inode is being synced, just update its dirty state.
1559 * The unlocker will place the inode on the appropriate
1560 * superblock list, based upon its state.
1562 if (inode
->i_state
& I_SYNC
)
1563 goto out_unlock_inode
;
1566 * Only add valid (hashed) inodes to the superblock's
1567 * dirty list. Add blockdev inodes as well.
1569 if (!S_ISBLK(inode
->i_mode
)) {
1570 if (inode_unhashed(inode
))
1571 goto out_unlock_inode
;
1573 if (inode
->i_state
& I_FREEING
)
1574 goto out_unlock_inode
;
1577 * If the inode was already on b_dirty/b_io/b_more_io, don't
1578 * reposition it (that would break b_dirty time-ordering).
1581 struct bdi_writeback
*wb
= inode_to_wb(inode
);
1582 struct list_head
*dirty_list
;
1583 bool wakeup_bdi
= false;
1585 spin_unlock(&inode
->i_lock
);
1586 spin_lock(&wb
->list_lock
);
1588 WARN(bdi_cap_writeback_dirty(wb
->bdi
) &&
1589 !test_bit(WB_registered
, &wb
->state
),
1590 "bdi-%s not registered\n", wb
->bdi
->name
);
1592 inode
->dirtied_when
= jiffies
;
1594 inode
->dirtied_time_when
= jiffies
;
1596 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
1597 dirty_list
= &wb
->b_dirty
;
1599 dirty_list
= &wb
->b_dirty_time
;
1601 wakeup_bdi
= inode_wb_list_move_locked(inode
, wb
,
1604 spin_unlock(&wb
->list_lock
);
1605 trace_writeback_dirty_inode_enqueue(inode
);
1608 * If this is the first dirty inode for this bdi,
1609 * we have to wake-up the corresponding bdi thread
1610 * to make sure background write-back happens
1613 if (bdi_cap_writeback_dirty(wb
->bdi
) && wakeup_bdi
)
1614 wb_wakeup_delayed(wb
);
1619 spin_unlock(&inode
->i_lock
);
1622 EXPORT_SYMBOL(__mark_inode_dirty
);
1624 static void wait_sb_inodes(struct super_block
*sb
)
1626 struct inode
*inode
, *old_inode
= NULL
;
1629 * We need to be protected against the filesystem going from
1630 * r/o to r/w or vice versa.
1632 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1634 spin_lock(&inode_sb_list_lock
);
1637 * Data integrity sync. Must wait for all pages under writeback,
1638 * because there may have been pages dirtied before our sync
1639 * call, but which had writeout started before we write it out.
1640 * In which case, the inode may not be on the dirty list, but
1641 * we still have to wait for that writeout.
1643 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1644 struct address_space
*mapping
= inode
->i_mapping
;
1646 spin_lock(&inode
->i_lock
);
1647 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1648 (mapping
->nrpages
== 0)) {
1649 spin_unlock(&inode
->i_lock
);
1653 spin_unlock(&inode
->i_lock
);
1654 spin_unlock(&inode_sb_list_lock
);
1657 * We hold a reference to 'inode' so it couldn't have been
1658 * removed from s_inodes list while we dropped the
1659 * inode_sb_list_lock. We cannot iput the inode now as we can
1660 * be holding the last reference and we cannot iput it under
1661 * inode_sb_list_lock. So we keep the reference and iput it
1667 filemap_fdatawait(mapping
);
1671 spin_lock(&inode_sb_list_lock
);
1673 spin_unlock(&inode_sb_list_lock
);
1677 static void __writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
,
1678 enum wb_reason reason
, bool skip_if_busy
)
1680 DEFINE_WB_COMPLETION_ONSTACK(done
);
1681 struct wb_writeback_work work
= {
1683 .sync_mode
= WB_SYNC_NONE
,
1684 .tagged_writepages
= 1,
1689 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1691 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1693 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1695 bdi_split_work_to_wbs(sb
->s_bdi
, &work
, skip_if_busy
);
1696 wb_wait_for_completion(bdi
, &done
);
1700 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1701 * @sb: the superblock
1702 * @nr: the number of pages to write
1703 * @reason: reason why some writeback work initiated
1705 * Start writeback on some inodes on this super_block. No guarantees are made
1706 * on how many (if any) will be written, and this function does not wait
1707 * for IO completion of submitted IO.
1709 void writeback_inodes_sb_nr(struct super_block
*sb
,
1711 enum wb_reason reason
)
1713 __writeback_inodes_sb_nr(sb
, nr
, reason
, false);
1715 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1718 * writeback_inodes_sb - writeback dirty inodes from given super_block
1719 * @sb: the superblock
1720 * @reason: reason why some writeback work was initiated
1722 * Start writeback on some inodes on this super_block. No guarantees are made
1723 * on how many (if any) will be written, and this function does not wait
1724 * for IO completion of submitted IO.
1726 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1728 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1730 EXPORT_SYMBOL(writeback_inodes_sb
);
1733 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1734 * @sb: the superblock
1735 * @nr: the number of pages to write
1736 * @reason: the reason of writeback
1738 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1739 * Returns 1 if writeback was started, 0 if not.
1741 bool try_to_writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
,
1742 enum wb_reason reason
)
1744 if (!down_read_trylock(&sb
->s_umount
))
1747 __writeback_inodes_sb_nr(sb
, nr
, reason
, true);
1748 up_read(&sb
->s_umount
);
1751 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1754 * try_to_writeback_inodes_sb - try to start writeback if none underway
1755 * @sb: the superblock
1756 * @reason: reason why some writeback work was initiated
1758 * Implement by try_to_writeback_inodes_sb_nr()
1759 * Returns 1 if writeback was started, 0 if not.
1761 bool try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1763 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1765 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1768 * sync_inodes_sb - sync sb inode pages
1769 * @sb: the superblock
1771 * This function writes and waits on any dirty inode belonging to this
1774 void sync_inodes_sb(struct super_block
*sb
)
1776 DEFINE_WB_COMPLETION_ONSTACK(done
);
1777 struct wb_writeback_work work
= {
1779 .sync_mode
= WB_SYNC_ALL
,
1780 .nr_pages
= LONG_MAX
,
1783 .reason
= WB_REASON_SYNC
,
1786 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1788 /* Nothing to do? */
1789 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1791 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1793 bdi_split_work_to_wbs(bdi
, &work
, false);
1794 wb_wait_for_completion(bdi
, &done
);
1798 EXPORT_SYMBOL(sync_inodes_sb
);
1801 * write_inode_now - write an inode to disk
1802 * @inode: inode to write to disk
1803 * @sync: whether the write should be synchronous or not
1805 * This function commits an inode to disk immediately if it is dirty. This is
1806 * primarily needed by knfsd.
1808 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1810 int write_inode_now(struct inode
*inode
, int sync
)
1812 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1813 struct writeback_control wbc
= {
1814 .nr_to_write
= LONG_MAX
,
1815 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1817 .range_end
= LLONG_MAX
,
1820 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1821 wbc
.nr_to_write
= 0;
1824 return writeback_single_inode(inode
, wb
, &wbc
);
1826 EXPORT_SYMBOL(write_inode_now
);
1829 * sync_inode - write an inode and its pages to disk.
1830 * @inode: the inode to sync
1831 * @wbc: controls the writeback mode
1833 * sync_inode() will write an inode and its pages to disk. It will also
1834 * correctly update the inode on its superblock's dirty inode lists and will
1835 * update inode->i_state.
1837 * The caller must have a ref on the inode.
1839 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1841 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1843 EXPORT_SYMBOL(sync_inode
);
1846 * sync_inode_metadata - write an inode to disk
1847 * @inode: the inode to sync
1848 * @wait: wait for I/O to complete.
1850 * Write an inode to disk and adjust its dirty state after completion.
1852 * Note: only writes the actual inode, no associated data or other metadata.
1854 int sync_inode_metadata(struct inode
*inode
, int wait
)
1856 struct writeback_control wbc
= {
1857 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1858 .nr_to_write
= 0, /* metadata-only */
1861 return sync_inode(inode
, &wbc
);
1863 EXPORT_SYMBOL(sync_inode_metadata
);