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 * wbc_attach_and_unlock_inode - associate wbc with target inode and unlock it
249 * @wbc: writeback_control of interest
250 * @inode: target inode
252 * @inode is locked and about to be written back under the control of @wbc.
253 * Record @inode's writeback context into @wbc and unlock the i_lock. On
254 * writeback completion, wbc_detach_inode() should be called. This is used
255 * to track the cgroup writeback context.
257 void wbc_attach_and_unlock_inode(struct writeback_control
*wbc
,
260 wbc
->wb
= inode_to_wb(inode
);
262 spin_unlock(&inode
->i_lock
);
266 * wbc_detach_inode - disassociate wbc from its target inode
267 * @wbc: writeback_control of interest
269 * To be called after a writeback attempt of an inode finishes and undoes
270 * wbc_attach_and_unlock_inode(). Can be called under any context.
272 void wbc_detach_inode(struct writeback_control
*wbc
)
279 * inode_congested - test whether an inode is congested
280 * @inode: inode to test for congestion
281 * @cong_bits: mask of WB_[a]sync_congested bits to test
283 * Tests whether @inode is congested. @cong_bits is the mask of congestion
284 * bits to test and the return value is the mask of set bits.
286 * If cgroup writeback is enabled for @inode, the congestion state is
287 * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
288 * associated with @inode is congested; otherwise, the root wb's congestion
291 int inode_congested(struct inode
*inode
, int cong_bits
)
294 struct bdi_writeback
*wb
= inode_to_wb(inode
);
296 return wb_congested(wb
, cong_bits
);
299 return wb_congested(&inode_to_bdi(inode
)->wb
, cong_bits
);
301 EXPORT_SYMBOL_GPL(inode_congested
);
304 * wb_wait_for_single_work - wait for completion of a single bdi_writeback_work
305 * @bdi: bdi the work item was issued to
306 * @work: work item to wait for
308 * Wait for the completion of @work which was issued to one of @bdi's
309 * bdi_writeback's. The caller must have set @work->single_wait before
310 * issuing it. This wait operates independently fo
311 * wb_wait_for_completion() and also disables automatic freeing of @work.
313 static void wb_wait_for_single_work(struct backing_dev_info
*bdi
,
314 struct wb_writeback_work
*work
)
316 if (WARN_ON_ONCE(!work
->single_wait
))
319 wait_event(bdi
->wb_waitq
, work
->single_done
);
322 * Paired with smp_wmb() in wb_do_writeback() and ensures that all
323 * modifications to @work prior to assertion of ->single_done is
324 * visible to the caller once this function returns.
330 * wb_split_bdi_pages - split nr_pages to write according to bandwidth
331 * @wb: target bdi_writeback to split @nr_pages to
332 * @nr_pages: number of pages to write for the whole bdi
334 * Split @wb's portion of @nr_pages according to @wb's write bandwidth in
335 * relation to the total write bandwidth of all wb's w/ dirty inodes on
338 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
340 unsigned long this_bw
= wb
->avg_write_bandwidth
;
341 unsigned long tot_bw
= atomic_long_read(&wb
->bdi
->tot_write_bandwidth
);
343 if (nr_pages
== LONG_MAX
)
347 * This may be called on clean wb's and proportional distribution
348 * may not make sense, just use the original @nr_pages in those
349 * cases. In general, we wanna err on the side of writing more.
351 if (!tot_bw
|| this_bw
>= tot_bw
)
354 return DIV_ROUND_UP_ULL((u64
)nr_pages
* this_bw
, tot_bw
);
358 * wb_clone_and_queue_work - clone a wb_writeback_work and issue it to a wb
359 * @wb: target bdi_writeback
360 * @base_work: source wb_writeback_work
362 * Try to make a clone of @base_work and issue it to @wb. If cloning
363 * succeeds, %true is returned; otherwise, @base_work is issued directly
364 * and %false is returned. In the latter case, the caller is required to
365 * wait for @base_work's completion using wb_wait_for_single_work().
367 * A clone is auto-freed on completion. @base_work never is.
369 static bool wb_clone_and_queue_work(struct bdi_writeback
*wb
,
370 struct wb_writeback_work
*base_work
)
372 struct wb_writeback_work
*work
;
374 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
378 work
->single_wait
= 0;
382 work
->single_wait
= 1;
384 work
->single_done
= 0;
385 wb_queue_work(wb
, work
);
386 return work
!= base_work
;
390 * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi
391 * @bdi: target backing_dev_info
392 * @base_work: wb_writeback_work to issue
393 * @skip_if_busy: skip wb's which already have writeback in progress
395 * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which
396 * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's
397 * distributed to the busy wbs according to each wb's proportion in the
398 * total active write bandwidth of @bdi.
400 static void bdi_split_work_to_wbs(struct backing_dev_info
*bdi
,
401 struct wb_writeback_work
*base_work
,
404 long nr_pages
= base_work
->nr_pages
;
405 int next_blkcg_id
= 0;
406 struct bdi_writeback
*wb
;
411 if (!bdi_has_dirty_io(bdi
))
415 bdi_for_each_wb(wb
, bdi
, &iter
, next_blkcg_id
) {
416 if (!wb_has_dirty_io(wb
) ||
417 (skip_if_busy
&& writeback_in_progress(wb
)))
420 base_work
->nr_pages
= wb_split_bdi_pages(wb
, nr_pages
);
421 if (!wb_clone_and_queue_work(wb
, base_work
)) {
422 next_blkcg_id
= wb
->blkcg_css
->id
+ 1;
424 wb_wait_for_single_work(bdi
, base_work
);
431 #else /* CONFIG_CGROUP_WRITEBACK */
433 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
438 static void bdi_split_work_to_wbs(struct backing_dev_info
*bdi
,
439 struct wb_writeback_work
*base_work
,
444 if (bdi_has_dirty_io(bdi
) &&
445 (!skip_if_busy
|| !writeback_in_progress(&bdi
->wb
))) {
446 base_work
->auto_free
= 0;
447 base_work
->single_wait
= 0;
448 base_work
->single_done
= 0;
449 wb_queue_work(&bdi
->wb
, base_work
);
453 #endif /* CONFIG_CGROUP_WRITEBACK */
455 void wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
456 bool range_cyclic
, enum wb_reason reason
)
458 struct wb_writeback_work
*work
;
460 if (!wb_has_dirty_io(wb
))
464 * This is WB_SYNC_NONE writeback, so if allocation fails just
465 * wakeup the thread for old dirty data writeback
467 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
469 trace_writeback_nowork(wb
->bdi
);
474 work
->sync_mode
= WB_SYNC_NONE
;
475 work
->nr_pages
= nr_pages
;
476 work
->range_cyclic
= range_cyclic
;
477 work
->reason
= reason
;
480 wb_queue_work(wb
, work
);
484 * wb_start_background_writeback - start background writeback
485 * @wb: bdi_writback to write from
488 * This makes sure WB_SYNC_NONE background writeback happens. When
489 * this function returns, it is only guaranteed that for given wb
490 * some IO is happening if we are over background dirty threshold.
491 * Caller need not hold sb s_umount semaphore.
493 void wb_start_background_writeback(struct bdi_writeback
*wb
)
496 * We just wake up the flusher thread. It will perform background
497 * writeback as soon as there is no other work to do.
499 trace_writeback_wake_background(wb
->bdi
);
504 * Remove the inode from the writeback list it is on.
506 void inode_wb_list_del(struct inode
*inode
)
508 struct bdi_writeback
*wb
= inode_to_wb(inode
);
510 spin_lock(&wb
->list_lock
);
511 inode_wb_list_del_locked(inode
, wb
);
512 spin_unlock(&wb
->list_lock
);
516 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
517 * furthest end of its superblock's dirty-inode list.
519 * Before stamping the inode's ->dirtied_when, we check to see whether it is
520 * already the most-recently-dirtied inode on the b_dirty list. If that is
521 * the case then the inode must have been redirtied while it was being written
522 * out and we don't reset its dirtied_when.
524 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
526 if (!list_empty(&wb
->b_dirty
)) {
529 tail
= wb_inode(wb
->b_dirty
.next
);
530 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
531 inode
->dirtied_when
= jiffies
;
533 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty
);
537 * requeue inode for re-scanning after bdi->b_io list is exhausted.
539 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
541 inode_wb_list_move_locked(inode
, wb
, &wb
->b_more_io
);
544 static void inode_sync_complete(struct inode
*inode
)
546 inode
->i_state
&= ~I_SYNC
;
547 /* If inode is clean an unused, put it into LRU now... */
548 inode_add_lru(inode
);
549 /* Waiters must see I_SYNC cleared before being woken up */
551 wake_up_bit(&inode
->i_state
, __I_SYNC
);
554 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
556 bool ret
= time_after(inode
->dirtied_when
, t
);
559 * For inodes being constantly redirtied, dirtied_when can get stuck.
560 * It _appears_ to be in the future, but is actually in distant past.
561 * This test is necessary to prevent such wrapped-around relative times
562 * from permanently stopping the whole bdi writeback.
564 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
569 #define EXPIRE_DIRTY_ATIME 0x0001
572 * Move expired (dirtied before work->older_than_this) dirty inodes from
573 * @delaying_queue to @dispatch_queue.
575 static int move_expired_inodes(struct list_head
*delaying_queue
,
576 struct list_head
*dispatch_queue
,
578 struct wb_writeback_work
*work
)
580 unsigned long *older_than_this
= NULL
;
581 unsigned long expire_time
;
583 struct list_head
*pos
, *node
;
584 struct super_block
*sb
= NULL
;
589 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
590 older_than_this
= work
->older_than_this
;
591 else if (!work
->for_sync
) {
592 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
593 older_than_this
= &expire_time
;
595 while (!list_empty(delaying_queue
)) {
596 inode
= wb_inode(delaying_queue
->prev
);
597 if (older_than_this
&&
598 inode_dirtied_after(inode
, *older_than_this
))
600 list_move(&inode
->i_wb_list
, &tmp
);
602 if (flags
& EXPIRE_DIRTY_ATIME
)
603 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
604 if (sb_is_blkdev_sb(inode
->i_sb
))
606 if (sb
&& sb
!= inode
->i_sb
)
611 /* just one sb in list, splice to dispatch_queue and we're done */
613 list_splice(&tmp
, dispatch_queue
);
617 /* Move inodes from one superblock together */
618 while (!list_empty(&tmp
)) {
619 sb
= wb_inode(tmp
.prev
)->i_sb
;
620 list_for_each_prev_safe(pos
, node
, &tmp
) {
621 inode
= wb_inode(pos
);
622 if (inode
->i_sb
== sb
)
623 list_move(&inode
->i_wb_list
, dispatch_queue
);
631 * Queue all expired dirty inodes for io, eldest first.
633 * newly dirtied b_dirty b_io b_more_io
634 * =============> gf edc BA
636 * newly dirtied b_dirty b_io b_more_io
637 * =============> g fBAedc
639 * +--> dequeue for IO
641 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
645 assert_spin_locked(&wb
->list_lock
);
646 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
647 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
648 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
649 EXPIRE_DIRTY_ATIME
, work
);
651 wb_io_lists_populated(wb
);
652 trace_writeback_queue_io(wb
, work
, moved
);
655 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
659 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
660 trace_writeback_write_inode_start(inode
, wbc
);
661 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
662 trace_writeback_write_inode(inode
, wbc
);
669 * Wait for writeback on an inode to complete. Called with i_lock held.
670 * Caller must make sure inode cannot go away when we drop i_lock.
672 static void __inode_wait_for_writeback(struct inode
*inode
)
673 __releases(inode
->i_lock
)
674 __acquires(inode
->i_lock
)
676 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
677 wait_queue_head_t
*wqh
;
679 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
680 while (inode
->i_state
& I_SYNC
) {
681 spin_unlock(&inode
->i_lock
);
682 __wait_on_bit(wqh
, &wq
, bit_wait
,
683 TASK_UNINTERRUPTIBLE
);
684 spin_lock(&inode
->i_lock
);
689 * Wait for writeback on an inode to complete. Caller must have inode pinned.
691 void inode_wait_for_writeback(struct inode
*inode
)
693 spin_lock(&inode
->i_lock
);
694 __inode_wait_for_writeback(inode
);
695 spin_unlock(&inode
->i_lock
);
699 * Sleep until I_SYNC is cleared. This function must be called with i_lock
700 * held and drops it. It is aimed for callers not holding any inode reference
701 * so once i_lock is dropped, inode can go away.
703 static void inode_sleep_on_writeback(struct inode
*inode
)
704 __releases(inode
->i_lock
)
707 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
710 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
711 sleep
= inode
->i_state
& I_SYNC
;
712 spin_unlock(&inode
->i_lock
);
715 finish_wait(wqh
, &wait
);
719 * Find proper writeback list for the inode depending on its current state and
720 * possibly also change of its state while we were doing writeback. Here we
721 * handle things such as livelock prevention or fairness of writeback among
722 * inodes. This function can be called only by flusher thread - noone else
723 * processes all inodes in writeback lists and requeueing inodes behind flusher
724 * thread's back can have unexpected consequences.
726 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
727 struct writeback_control
*wbc
)
729 if (inode
->i_state
& I_FREEING
)
733 * Sync livelock prevention. Each inode is tagged and synced in one
734 * shot. If still dirty, it will be redirty_tail()'ed below. Update
735 * the dirty time to prevent enqueue and sync it again.
737 if ((inode
->i_state
& I_DIRTY
) &&
738 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
739 inode
->dirtied_when
= jiffies
;
741 if (wbc
->pages_skipped
) {
743 * writeback is not making progress due to locked
744 * buffers. Skip this inode for now.
746 redirty_tail(inode
, wb
);
750 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
752 * We didn't write back all the pages. nfs_writepages()
753 * sometimes bales out without doing anything.
755 if (wbc
->nr_to_write
<= 0) {
756 /* Slice used up. Queue for next turn. */
757 requeue_io(inode
, wb
);
760 * Writeback blocked by something other than
761 * congestion. Delay the inode for some time to
762 * avoid spinning on the CPU (100% iowait)
763 * retrying writeback of the dirty page/inode
764 * that cannot be performed immediately.
766 redirty_tail(inode
, wb
);
768 } else if (inode
->i_state
& I_DIRTY
) {
770 * Filesystems can dirty the inode during writeback operations,
771 * such as delayed allocation during submission or metadata
772 * updates after data IO completion.
774 redirty_tail(inode
, wb
);
775 } else if (inode
->i_state
& I_DIRTY_TIME
) {
776 inode
->dirtied_when
= jiffies
;
777 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty_time
);
779 /* The inode is clean. Remove from writeback lists. */
780 inode_wb_list_del_locked(inode
, wb
);
785 * Write out an inode and its dirty pages. Do not update the writeback list
786 * linkage. That is left to the caller. The caller is also responsible for
787 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
790 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
792 struct address_space
*mapping
= inode
->i_mapping
;
793 long nr_to_write
= wbc
->nr_to_write
;
797 WARN_ON(!(inode
->i_state
& I_SYNC
));
799 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
801 ret
= do_writepages(mapping
, wbc
);
804 * Make sure to wait on the data before writing out the metadata.
805 * This is important for filesystems that modify metadata on data
806 * I/O completion. We don't do it for sync(2) writeback because it has a
807 * separate, external IO completion path and ->sync_fs for guaranteeing
808 * inode metadata is written back correctly.
810 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
811 int err
= filemap_fdatawait(mapping
);
817 * Some filesystems may redirty the inode during the writeback
818 * due to delalloc, clear dirty metadata flags right before
821 spin_lock(&inode
->i_lock
);
823 dirty
= inode
->i_state
& I_DIRTY
;
824 if (inode
->i_state
& I_DIRTY_TIME
) {
825 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
826 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
827 unlikely(time_after(jiffies
,
828 (inode
->dirtied_time_when
+
829 dirtytime_expire_interval
* HZ
)))) {
830 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
831 trace_writeback_lazytime(inode
);
834 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
835 inode
->i_state
&= ~dirty
;
838 * Paired with smp_mb() in __mark_inode_dirty(). This allows
839 * __mark_inode_dirty() to test i_state without grabbing i_lock -
840 * either they see the I_DIRTY bits cleared or we see the dirtied
843 * I_DIRTY_PAGES is always cleared together above even if @mapping
844 * still has dirty pages. The flag is reinstated after smp_mb() if
845 * necessary. This guarantees that either __mark_inode_dirty()
846 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
850 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
851 inode
->i_state
|= I_DIRTY_PAGES
;
853 spin_unlock(&inode
->i_lock
);
855 if (dirty
& I_DIRTY_TIME
)
856 mark_inode_dirty_sync(inode
);
857 /* Don't write the inode if only I_DIRTY_PAGES was set */
858 if (dirty
& ~I_DIRTY_PAGES
) {
859 int err
= write_inode(inode
, wbc
);
863 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
868 * Write out an inode's dirty pages. Either the caller has an active reference
869 * on the inode or the inode has I_WILL_FREE set.
871 * This function is designed to be called for writing back one inode which
872 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
873 * and does more profound writeback list handling in writeback_sb_inodes().
876 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
877 struct writeback_control
*wbc
)
881 spin_lock(&inode
->i_lock
);
882 if (!atomic_read(&inode
->i_count
))
883 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
885 WARN_ON(inode
->i_state
& I_WILL_FREE
);
887 if (inode
->i_state
& I_SYNC
) {
888 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
891 * It's a data-integrity sync. We must wait. Since callers hold
892 * inode reference or inode has I_WILL_FREE set, it cannot go
895 __inode_wait_for_writeback(inode
);
897 WARN_ON(inode
->i_state
& I_SYNC
);
899 * Skip inode if it is clean and we have no outstanding writeback in
900 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
901 * function since flusher thread may be doing for example sync in
902 * parallel and if we move the inode, it could get skipped. So here we
903 * make sure inode is on some writeback list and leave it there unless
904 * we have completely cleaned the inode.
906 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
907 (wbc
->sync_mode
!= WB_SYNC_ALL
||
908 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
910 inode
->i_state
|= I_SYNC
;
911 wbc_attach_and_unlock_inode(wbc
, inode
);
913 ret
= __writeback_single_inode(inode
, wbc
);
915 wbc_detach_inode(wbc
);
916 spin_lock(&wb
->list_lock
);
917 spin_lock(&inode
->i_lock
);
919 * If inode is clean, remove it from writeback lists. Otherwise don't
920 * touch it. See comment above for explanation.
922 if (!(inode
->i_state
& I_DIRTY_ALL
))
923 inode_wb_list_del_locked(inode
, wb
);
924 spin_unlock(&wb
->list_lock
);
925 inode_sync_complete(inode
);
927 spin_unlock(&inode
->i_lock
);
931 static long writeback_chunk_size(struct bdi_writeback
*wb
,
932 struct wb_writeback_work
*work
)
937 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
938 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
939 * here avoids calling into writeback_inodes_wb() more than once.
941 * The intended call sequence for WB_SYNC_ALL writeback is:
944 * writeback_sb_inodes() <== called only once
945 * write_cache_pages() <== called once for each inode
946 * (quickly) tag currently dirty pages
947 * (maybe slowly) sync all tagged pages
949 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
952 pages
= min(wb
->avg_write_bandwidth
/ 2,
953 global_wb_domain
.dirty_limit
/ DIRTY_SCOPE
);
954 pages
= min(pages
, work
->nr_pages
);
955 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
956 MIN_WRITEBACK_PAGES
);
963 * Write a portion of b_io inodes which belong to @sb.
965 * Return the number of pages and/or inodes written.
967 static long writeback_sb_inodes(struct super_block
*sb
,
968 struct bdi_writeback
*wb
,
969 struct wb_writeback_work
*work
)
971 struct writeback_control wbc
= {
972 .sync_mode
= work
->sync_mode
,
973 .tagged_writepages
= work
->tagged_writepages
,
974 .for_kupdate
= work
->for_kupdate
,
975 .for_background
= work
->for_background
,
976 .for_sync
= work
->for_sync
,
977 .range_cyclic
= work
->range_cyclic
,
979 .range_end
= LLONG_MAX
,
981 unsigned long start_time
= jiffies
;
983 long wrote
= 0; /* count both pages and inodes */
985 while (!list_empty(&wb
->b_io
)) {
986 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
988 if (inode
->i_sb
!= sb
) {
991 * We only want to write back data for this
992 * superblock, move all inodes not belonging
993 * to it back onto the dirty list.
995 redirty_tail(inode
, wb
);
1000 * The inode belongs to a different superblock.
1001 * Bounce back to the caller to unpin this and
1002 * pin the next superblock.
1008 * Don't bother with new inodes or inodes being freed, first
1009 * kind does not need periodic writeout yet, and for the latter
1010 * kind writeout is handled by the freer.
1012 spin_lock(&inode
->i_lock
);
1013 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
1014 spin_unlock(&inode
->i_lock
);
1015 redirty_tail(inode
, wb
);
1018 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
1020 * If this inode is locked for writeback and we are not
1021 * doing writeback-for-data-integrity, move it to
1022 * b_more_io so that writeback can proceed with the
1023 * other inodes on s_io.
1025 * We'll have another go at writing back this inode
1026 * when we completed a full scan of b_io.
1028 spin_unlock(&inode
->i_lock
);
1029 requeue_io(inode
, wb
);
1030 trace_writeback_sb_inodes_requeue(inode
);
1033 spin_unlock(&wb
->list_lock
);
1036 * We already requeued the inode if it had I_SYNC set and we
1037 * are doing WB_SYNC_NONE writeback. So this catches only the
1040 if (inode
->i_state
& I_SYNC
) {
1041 /* Wait for I_SYNC. This function drops i_lock... */
1042 inode_sleep_on_writeback(inode
);
1043 /* Inode may be gone, start again */
1044 spin_lock(&wb
->list_lock
);
1047 inode
->i_state
|= I_SYNC
;
1048 wbc_attach_and_unlock_inode(&wbc
, inode
);
1050 write_chunk
= writeback_chunk_size(wb
, work
);
1051 wbc
.nr_to_write
= write_chunk
;
1052 wbc
.pages_skipped
= 0;
1055 * We use I_SYNC to pin the inode in memory. While it is set
1056 * evict_inode() will wait so the inode cannot be freed.
1058 __writeback_single_inode(inode
, &wbc
);
1060 wbc_detach_inode(&wbc
);
1061 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
1062 wrote
+= write_chunk
- wbc
.nr_to_write
;
1063 spin_lock(&wb
->list_lock
);
1064 spin_lock(&inode
->i_lock
);
1065 if (!(inode
->i_state
& I_DIRTY_ALL
))
1067 requeue_inode(inode
, wb
, &wbc
);
1068 inode_sync_complete(inode
);
1069 spin_unlock(&inode
->i_lock
);
1070 cond_resched_lock(&wb
->list_lock
);
1072 * bail out to wb_writeback() often enough to check
1073 * background threshold and other termination conditions.
1076 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
1078 if (work
->nr_pages
<= 0)
1085 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
1086 struct wb_writeback_work
*work
)
1088 unsigned long start_time
= jiffies
;
1091 while (!list_empty(&wb
->b_io
)) {
1092 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
1093 struct super_block
*sb
= inode
->i_sb
;
1095 if (!trylock_super(sb
)) {
1097 * trylock_super() may fail consistently due to
1098 * s_umount being grabbed by someone else. Don't use
1099 * requeue_io() to avoid busy retrying the inode/sb.
1101 redirty_tail(inode
, wb
);
1104 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
1105 up_read(&sb
->s_umount
);
1107 /* refer to the same tests at the end of writeback_sb_inodes */
1109 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
1111 if (work
->nr_pages
<= 0)
1115 /* Leave any unwritten inodes on b_io */
1119 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
1120 enum wb_reason reason
)
1122 struct wb_writeback_work work
= {
1123 .nr_pages
= nr_pages
,
1124 .sync_mode
= WB_SYNC_NONE
,
1129 spin_lock(&wb
->list_lock
);
1130 if (list_empty(&wb
->b_io
))
1131 queue_io(wb
, &work
);
1132 __writeback_inodes_wb(wb
, &work
);
1133 spin_unlock(&wb
->list_lock
);
1135 return nr_pages
- work
.nr_pages
;
1139 * Explicit flushing or periodic writeback of "old" data.
1141 * Define "old": the first time one of an inode's pages is dirtied, we mark the
1142 * dirtying-time in the inode's address_space. So this periodic writeback code
1143 * just walks the superblock inode list, writing back any inodes which are
1144 * older than a specific point in time.
1146 * Try to run once per dirty_writeback_interval. But if a writeback event
1147 * takes longer than a dirty_writeback_interval interval, then leave a
1150 * older_than_this takes precedence over nr_to_write. So we'll only write back
1151 * all dirty pages if they are all attached to "old" mappings.
1153 static long wb_writeback(struct bdi_writeback
*wb
,
1154 struct wb_writeback_work
*work
)
1156 unsigned long wb_start
= jiffies
;
1157 long nr_pages
= work
->nr_pages
;
1158 unsigned long oldest_jif
;
1159 struct inode
*inode
;
1162 oldest_jif
= jiffies
;
1163 work
->older_than_this
= &oldest_jif
;
1165 spin_lock(&wb
->list_lock
);
1168 * Stop writeback when nr_pages has been consumed
1170 if (work
->nr_pages
<= 0)
1174 * Background writeout and kupdate-style writeback may
1175 * run forever. Stop them if there is other work to do
1176 * so that e.g. sync can proceed. They'll be restarted
1177 * after the other works are all done.
1179 if ((work
->for_background
|| work
->for_kupdate
) &&
1180 !list_empty(&wb
->work_list
))
1184 * For background writeout, stop when we are below the
1185 * background dirty threshold
1187 if (work
->for_background
&& !wb_over_bg_thresh(wb
))
1191 * Kupdate and background works are special and we want to
1192 * include all inodes that need writing. Livelock avoidance is
1193 * handled by these works yielding to any other work so we are
1196 if (work
->for_kupdate
) {
1197 oldest_jif
= jiffies
-
1198 msecs_to_jiffies(dirty_expire_interval
* 10);
1199 } else if (work
->for_background
)
1200 oldest_jif
= jiffies
;
1202 trace_writeback_start(wb
->bdi
, work
);
1203 if (list_empty(&wb
->b_io
))
1206 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
1208 progress
= __writeback_inodes_wb(wb
, work
);
1209 trace_writeback_written(wb
->bdi
, work
);
1211 wb_update_bandwidth(wb
, wb_start
);
1214 * Did we write something? Try for more
1216 * Dirty inodes are moved to b_io for writeback in batches.
1217 * The completion of the current batch does not necessarily
1218 * mean the overall work is done. So we keep looping as long
1219 * as made some progress on cleaning pages or inodes.
1224 * No more inodes for IO, bail
1226 if (list_empty(&wb
->b_more_io
))
1229 * Nothing written. Wait for some inode to
1230 * become available for writeback. Otherwise
1231 * we'll just busyloop.
1233 if (!list_empty(&wb
->b_more_io
)) {
1234 trace_writeback_wait(wb
->bdi
, work
);
1235 inode
= wb_inode(wb
->b_more_io
.prev
);
1236 spin_lock(&inode
->i_lock
);
1237 spin_unlock(&wb
->list_lock
);
1238 /* This function drops i_lock... */
1239 inode_sleep_on_writeback(inode
);
1240 spin_lock(&wb
->list_lock
);
1243 spin_unlock(&wb
->list_lock
);
1245 return nr_pages
- work
->nr_pages
;
1249 * Return the next wb_writeback_work struct that hasn't been processed yet.
1251 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
1253 struct wb_writeback_work
*work
= NULL
;
1255 spin_lock_bh(&wb
->work_lock
);
1256 if (!list_empty(&wb
->work_list
)) {
1257 work
= list_entry(wb
->work_list
.next
,
1258 struct wb_writeback_work
, list
);
1259 list_del_init(&work
->list
);
1261 spin_unlock_bh(&wb
->work_lock
);
1266 * Add in the number of potentially dirty inodes, because each inode
1267 * write can dirty pagecache in the underlying blockdev.
1269 static unsigned long get_nr_dirty_pages(void)
1271 return global_page_state(NR_FILE_DIRTY
) +
1272 global_page_state(NR_UNSTABLE_NFS
) +
1273 get_nr_dirty_inodes();
1276 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1278 if (wb_over_bg_thresh(wb
)) {
1280 struct wb_writeback_work work
= {
1281 .nr_pages
= LONG_MAX
,
1282 .sync_mode
= WB_SYNC_NONE
,
1283 .for_background
= 1,
1285 .reason
= WB_REASON_BACKGROUND
,
1288 return wb_writeback(wb
, &work
);
1294 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1296 unsigned long expired
;
1300 * When set to zero, disable periodic writeback
1302 if (!dirty_writeback_interval
)
1305 expired
= wb
->last_old_flush
+
1306 msecs_to_jiffies(dirty_writeback_interval
* 10);
1307 if (time_before(jiffies
, expired
))
1310 wb
->last_old_flush
= jiffies
;
1311 nr_pages
= get_nr_dirty_pages();
1314 struct wb_writeback_work work
= {
1315 .nr_pages
= nr_pages
,
1316 .sync_mode
= WB_SYNC_NONE
,
1319 .reason
= WB_REASON_PERIODIC
,
1322 return wb_writeback(wb
, &work
);
1329 * Retrieve work items and do the writeback they describe
1331 static long wb_do_writeback(struct bdi_writeback
*wb
)
1333 struct wb_writeback_work
*work
;
1336 set_bit(WB_writeback_running
, &wb
->state
);
1337 while ((work
= get_next_work_item(wb
)) != NULL
) {
1338 struct wb_completion
*done
= work
->done
;
1339 bool need_wake_up
= false;
1341 trace_writeback_exec(wb
->bdi
, work
);
1343 wrote
+= wb_writeback(wb
, work
);
1345 if (work
->single_wait
) {
1346 WARN_ON_ONCE(work
->auto_free
);
1347 /* paired w/ rmb in wb_wait_for_single_work() */
1349 work
->single_done
= 1;
1350 need_wake_up
= true;
1351 } else if (work
->auto_free
) {
1355 if (done
&& atomic_dec_and_test(&done
->cnt
))
1356 need_wake_up
= true;
1359 wake_up_all(&wb
->bdi
->wb_waitq
);
1363 * Check for periodic writeback, kupdated() style
1365 wrote
+= wb_check_old_data_flush(wb
);
1366 wrote
+= wb_check_background_flush(wb
);
1367 clear_bit(WB_writeback_running
, &wb
->state
);
1373 * Handle writeback of dirty data for the device backed by this bdi. Also
1374 * reschedules periodically and does kupdated style flushing.
1376 void wb_workfn(struct work_struct
*work
)
1378 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1379 struct bdi_writeback
, dwork
);
1382 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1383 current
->flags
|= PF_SWAPWRITE
;
1385 if (likely(!current_is_workqueue_rescuer() ||
1386 !test_bit(WB_registered
, &wb
->state
))) {
1388 * The normal path. Keep writing back @wb until its
1389 * work_list is empty. Note that this path is also taken
1390 * if @wb is shutting down even when we're running off the
1391 * rescuer as work_list needs to be drained.
1394 pages_written
= wb_do_writeback(wb
);
1395 trace_writeback_pages_written(pages_written
);
1396 } while (!list_empty(&wb
->work_list
));
1399 * bdi_wq can't get enough workers and we're running off
1400 * the emergency worker. Don't hog it. Hopefully, 1024 is
1401 * enough for efficient IO.
1403 pages_written
= writeback_inodes_wb(wb
, 1024,
1404 WB_REASON_FORKER_THREAD
);
1405 trace_writeback_pages_written(pages_written
);
1408 if (!list_empty(&wb
->work_list
))
1409 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1410 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1411 wb_wakeup_delayed(wb
);
1413 current
->flags
&= ~PF_SWAPWRITE
;
1417 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1420 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1422 struct backing_dev_info
*bdi
;
1425 nr_pages
= get_nr_dirty_pages();
1428 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1429 struct bdi_writeback
*wb
;
1430 struct wb_iter iter
;
1432 if (!bdi_has_dirty_io(bdi
))
1435 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1436 wb_start_writeback(wb
, wb_split_bdi_pages(wb
, nr_pages
),
1443 * Wake up bdi's periodically to make sure dirtytime inodes gets
1444 * written back periodically. We deliberately do *not* check the
1445 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1446 * kernel to be constantly waking up once there are any dirtytime
1447 * inodes on the system. So instead we define a separate delayed work
1448 * function which gets called much more rarely. (By default, only
1449 * once every 12 hours.)
1451 * If there is any other write activity going on in the file system,
1452 * this function won't be necessary. But if the only thing that has
1453 * happened on the file system is a dirtytime inode caused by an atime
1454 * update, we need this infrastructure below to make sure that inode
1455 * eventually gets pushed out to disk.
1457 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1458 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1460 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1462 struct backing_dev_info
*bdi
;
1465 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1466 struct bdi_writeback
*wb
;
1467 struct wb_iter iter
;
1469 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1470 if (!list_empty(&bdi
->wb
.b_dirty_time
))
1471 wb_wakeup(&bdi
->wb
);
1474 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1477 static int __init
start_dirtytime_writeback(void)
1479 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1482 __initcall(start_dirtytime_writeback
);
1484 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1485 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1489 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1490 if (ret
== 0 && write
)
1491 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1495 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1497 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1498 struct dentry
*dentry
;
1499 const char *name
= "?";
1501 dentry
= d_find_alias(inode
);
1503 spin_lock(&dentry
->d_lock
);
1504 name
= (const char *) dentry
->d_name
.name
;
1507 "%s(%d): dirtied inode %lu (%s) on %s\n",
1508 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1509 name
, inode
->i_sb
->s_id
);
1511 spin_unlock(&dentry
->d_lock
);
1518 * __mark_inode_dirty - internal function
1519 * @inode: inode to mark
1520 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1521 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1522 * mark_inode_dirty_sync.
1524 * Put the inode on the super block's dirty list.
1526 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1527 * dirty list only if it is hashed or if it refers to a blockdev.
1528 * If it was not hashed, it will never be added to the dirty list
1529 * even if it is later hashed, as it will have been marked dirty already.
1531 * In short, make sure you hash any inodes _before_ you start marking
1534 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1535 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1536 * the kernel-internal blockdev inode represents the dirtying time of the
1537 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1538 * page->mapping->host, so the page-dirtying time is recorded in the internal
1541 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1542 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1544 struct super_block
*sb
= inode
->i_sb
;
1547 trace_writeback_mark_inode_dirty(inode
, flags
);
1550 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1551 * dirty the inode itself
1553 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
1554 trace_writeback_dirty_inode_start(inode
, flags
);
1556 if (sb
->s_op
->dirty_inode
)
1557 sb
->s_op
->dirty_inode(inode
, flags
);
1559 trace_writeback_dirty_inode(inode
, flags
);
1561 if (flags
& I_DIRTY_INODE
)
1562 flags
&= ~I_DIRTY_TIME
;
1563 dirtytime
= flags
& I_DIRTY_TIME
;
1566 * Paired with smp_mb() in __writeback_single_inode() for the
1567 * following lockless i_state test. See there for details.
1571 if (((inode
->i_state
& flags
) == flags
) ||
1572 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
1575 if (unlikely(block_dump
))
1576 block_dump___mark_inode_dirty(inode
);
1578 spin_lock(&inode
->i_lock
);
1579 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
1580 goto out_unlock_inode
;
1581 if ((inode
->i_state
& flags
) != flags
) {
1582 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1584 inode_attach_wb(inode
, NULL
);
1586 if (flags
& I_DIRTY_INODE
)
1587 inode
->i_state
&= ~I_DIRTY_TIME
;
1588 inode
->i_state
|= flags
;
1591 * If the inode is being synced, just update its dirty state.
1592 * The unlocker will place the inode on the appropriate
1593 * superblock list, based upon its state.
1595 if (inode
->i_state
& I_SYNC
)
1596 goto out_unlock_inode
;
1599 * Only add valid (hashed) inodes to the superblock's
1600 * dirty list. Add blockdev inodes as well.
1602 if (!S_ISBLK(inode
->i_mode
)) {
1603 if (inode_unhashed(inode
))
1604 goto out_unlock_inode
;
1606 if (inode
->i_state
& I_FREEING
)
1607 goto out_unlock_inode
;
1610 * If the inode was already on b_dirty/b_io/b_more_io, don't
1611 * reposition it (that would break b_dirty time-ordering).
1614 struct bdi_writeback
*wb
= inode_to_wb(inode
);
1615 struct list_head
*dirty_list
;
1616 bool wakeup_bdi
= false;
1618 spin_unlock(&inode
->i_lock
);
1619 spin_lock(&wb
->list_lock
);
1621 WARN(bdi_cap_writeback_dirty(wb
->bdi
) &&
1622 !test_bit(WB_registered
, &wb
->state
),
1623 "bdi-%s not registered\n", wb
->bdi
->name
);
1625 inode
->dirtied_when
= jiffies
;
1627 inode
->dirtied_time_when
= jiffies
;
1629 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
1630 dirty_list
= &wb
->b_dirty
;
1632 dirty_list
= &wb
->b_dirty_time
;
1634 wakeup_bdi
= inode_wb_list_move_locked(inode
, wb
,
1637 spin_unlock(&wb
->list_lock
);
1638 trace_writeback_dirty_inode_enqueue(inode
);
1641 * If this is the first dirty inode for this bdi,
1642 * we have to wake-up the corresponding bdi thread
1643 * to make sure background write-back happens
1646 if (bdi_cap_writeback_dirty(wb
->bdi
) && wakeup_bdi
)
1647 wb_wakeup_delayed(wb
);
1652 spin_unlock(&inode
->i_lock
);
1655 EXPORT_SYMBOL(__mark_inode_dirty
);
1657 static void wait_sb_inodes(struct super_block
*sb
)
1659 struct inode
*inode
, *old_inode
= NULL
;
1662 * We need to be protected against the filesystem going from
1663 * r/o to r/w or vice versa.
1665 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1667 spin_lock(&inode_sb_list_lock
);
1670 * Data integrity sync. Must wait for all pages under writeback,
1671 * because there may have been pages dirtied before our sync
1672 * call, but which had writeout started before we write it out.
1673 * In which case, the inode may not be on the dirty list, but
1674 * we still have to wait for that writeout.
1676 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1677 struct address_space
*mapping
= inode
->i_mapping
;
1679 spin_lock(&inode
->i_lock
);
1680 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1681 (mapping
->nrpages
== 0)) {
1682 spin_unlock(&inode
->i_lock
);
1686 spin_unlock(&inode
->i_lock
);
1687 spin_unlock(&inode_sb_list_lock
);
1690 * We hold a reference to 'inode' so it couldn't have been
1691 * removed from s_inodes list while we dropped the
1692 * inode_sb_list_lock. We cannot iput the inode now as we can
1693 * be holding the last reference and we cannot iput it under
1694 * inode_sb_list_lock. So we keep the reference and iput it
1700 filemap_fdatawait(mapping
);
1704 spin_lock(&inode_sb_list_lock
);
1706 spin_unlock(&inode_sb_list_lock
);
1710 static void __writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
,
1711 enum wb_reason reason
, bool skip_if_busy
)
1713 DEFINE_WB_COMPLETION_ONSTACK(done
);
1714 struct wb_writeback_work work
= {
1716 .sync_mode
= WB_SYNC_NONE
,
1717 .tagged_writepages
= 1,
1722 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1724 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1726 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1728 bdi_split_work_to_wbs(sb
->s_bdi
, &work
, skip_if_busy
);
1729 wb_wait_for_completion(bdi
, &done
);
1733 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1734 * @sb: the superblock
1735 * @nr: the number of pages to write
1736 * @reason: reason why some writeback work initiated
1738 * Start writeback on some inodes on this super_block. No guarantees are made
1739 * on how many (if any) will be written, and this function does not wait
1740 * for IO completion of submitted IO.
1742 void writeback_inodes_sb_nr(struct super_block
*sb
,
1744 enum wb_reason reason
)
1746 __writeback_inodes_sb_nr(sb
, nr
, reason
, false);
1748 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1751 * writeback_inodes_sb - writeback dirty inodes from given super_block
1752 * @sb: the superblock
1753 * @reason: reason why some writeback work was initiated
1755 * Start writeback on some inodes on this super_block. No guarantees are made
1756 * on how many (if any) will be written, and this function does not wait
1757 * for IO completion of submitted IO.
1759 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1761 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1763 EXPORT_SYMBOL(writeback_inodes_sb
);
1766 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1767 * @sb: the superblock
1768 * @nr: the number of pages to write
1769 * @reason: the reason of writeback
1771 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1772 * Returns 1 if writeback was started, 0 if not.
1774 bool try_to_writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
,
1775 enum wb_reason reason
)
1777 if (!down_read_trylock(&sb
->s_umount
))
1780 __writeback_inodes_sb_nr(sb
, nr
, reason
, true);
1781 up_read(&sb
->s_umount
);
1784 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1787 * try_to_writeback_inodes_sb - try to start writeback if none underway
1788 * @sb: the superblock
1789 * @reason: reason why some writeback work was initiated
1791 * Implement by try_to_writeback_inodes_sb_nr()
1792 * Returns 1 if writeback was started, 0 if not.
1794 bool try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1796 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1798 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1801 * sync_inodes_sb - sync sb inode pages
1802 * @sb: the superblock
1804 * This function writes and waits on any dirty inode belonging to this
1807 void sync_inodes_sb(struct super_block
*sb
)
1809 DEFINE_WB_COMPLETION_ONSTACK(done
);
1810 struct wb_writeback_work work
= {
1812 .sync_mode
= WB_SYNC_ALL
,
1813 .nr_pages
= LONG_MAX
,
1816 .reason
= WB_REASON_SYNC
,
1819 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1821 /* Nothing to do? */
1822 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1824 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1826 bdi_split_work_to_wbs(bdi
, &work
, false);
1827 wb_wait_for_completion(bdi
, &done
);
1831 EXPORT_SYMBOL(sync_inodes_sb
);
1834 * write_inode_now - write an inode to disk
1835 * @inode: inode to write to disk
1836 * @sync: whether the write should be synchronous or not
1838 * This function commits an inode to disk immediately if it is dirty. This is
1839 * primarily needed by knfsd.
1841 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1843 int write_inode_now(struct inode
*inode
, int sync
)
1845 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1846 struct writeback_control wbc
= {
1847 .nr_to_write
= LONG_MAX
,
1848 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1850 .range_end
= LLONG_MAX
,
1853 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1854 wbc
.nr_to_write
= 0;
1857 return writeback_single_inode(inode
, wb
, &wbc
);
1859 EXPORT_SYMBOL(write_inode_now
);
1862 * sync_inode - write an inode and its pages to disk.
1863 * @inode: the inode to sync
1864 * @wbc: controls the writeback mode
1866 * sync_inode() will write an inode and its pages to disk. It will also
1867 * correctly update the inode on its superblock's dirty inode lists and will
1868 * update inode->i_state.
1870 * The caller must have a ref on the inode.
1872 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1874 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1876 EXPORT_SYMBOL(sync_inode
);
1879 * sync_inode_metadata - write an inode to disk
1880 * @inode: the inode to sync
1881 * @wait: wait for I/O to complete.
1883 * Write an inode to disk and adjust its dirty state after completion.
1885 * Note: only writes the actual inode, no associated data or other metadata.
1887 int sync_inode_metadata(struct inode
*inode
, int wait
)
1889 struct writeback_control wbc
= {
1890 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1891 .nr_to_write
= 0, /* metadata-only */
1894 return sync_inode(inode
, &wbc
);
1896 EXPORT_SYMBOL(sync_inode_metadata
);