4 * Copyright (C) 2002, Linus Torvalds.
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the
9 * inode itself is not handled here.
11 * 10Apr2002 Andrew Morton
12 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
16 #include <linux/kernel.h>
17 #include <linux/export.h>
18 #include <linux/spinlock.h>
19 #include <linux/slab.h>
20 #include <linux/sched.h>
23 #include <linux/pagemap.h>
24 #include <linux/kthread.h>
25 #include <linux/writeback.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/tracepoint.h>
29 #include <linux/device.h>
33 * 4MB minimal write chunk size
35 #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
37 struct wb_completion
{
42 * Passed into wb_writeback(), essentially a subset of writeback_control
44 struct wb_writeback_work
{
46 struct super_block
*sb
;
47 unsigned long *older_than_this
;
48 enum writeback_sync_modes sync_mode
;
49 unsigned int tagged_writepages
:1;
50 unsigned int for_kupdate
:1;
51 unsigned int range_cyclic
:1;
52 unsigned int for_background
:1;
53 unsigned int for_sync
:1; /* sync(2) WB_SYNC_ALL writeback */
54 unsigned int auto_free
:1; /* free on completion */
55 unsigned int single_wait
:1;
56 unsigned int single_done
:1;
57 enum wb_reason reason
; /* why was writeback initiated? */
59 struct list_head list
; /* pending work list */
60 struct wb_completion
*done
; /* set if the caller waits */
64 * If one wants to wait for one or more wb_writeback_works, each work's
65 * ->done should be set to a wb_completion defined using the following
66 * macro. Once all work items are issued with wb_queue_work(), the caller
67 * can wait for the completion of all using wb_wait_for_completion(). Work
68 * items which are waited upon aren't freed automatically on completion.
70 #define DEFINE_WB_COMPLETION_ONSTACK(cmpl) \
71 struct wb_completion cmpl = { \
72 .cnt = ATOMIC_INIT(1), \
77 * If an inode is constantly having its pages dirtied, but then the
78 * updates stop dirtytime_expire_interval seconds in the past, it's
79 * possible for the worst case time between when an inode has its
80 * timestamps updated and when they finally get written out to be two
81 * dirtytime_expire_intervals. We set the default to 12 hours (in
82 * seconds), which means most of the time inodes will have their
83 * timestamps written to disk after 12 hours, but in the worst case a
84 * few inodes might not their timestamps updated for 24 hours.
86 unsigned int dirtytime_expire_interval
= 12 * 60 * 60;
88 static inline struct inode
*wb_inode(struct list_head
*head
)
90 return list_entry(head
, struct inode
, i_wb_list
);
94 * Include the creation of the trace points after defining the
95 * wb_writeback_work structure and inline functions so that the definition
96 * remains local to this file.
98 #define CREATE_TRACE_POINTS
99 #include <trace/events/writeback.h>
101 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage
);
103 static bool wb_io_lists_populated(struct bdi_writeback
*wb
)
105 if (wb_has_dirty_io(wb
)) {
108 set_bit(WB_has_dirty_io
, &wb
->state
);
109 WARN_ON_ONCE(!wb
->avg_write_bandwidth
);
110 atomic_long_add(wb
->avg_write_bandwidth
,
111 &wb
->bdi
->tot_write_bandwidth
);
116 static void wb_io_lists_depopulated(struct bdi_writeback
*wb
)
118 if (wb_has_dirty_io(wb
) && list_empty(&wb
->b_dirty
) &&
119 list_empty(&wb
->b_io
) && list_empty(&wb
->b_more_io
)) {
120 clear_bit(WB_has_dirty_io
, &wb
->state
);
121 WARN_ON_ONCE(atomic_long_sub_return(wb
->avg_write_bandwidth
,
122 &wb
->bdi
->tot_write_bandwidth
) < 0);
127 * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list
128 * @inode: inode to be moved
129 * @wb: target bdi_writeback
130 * @head: one of @wb->b_{dirty|io|more_io}
132 * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io.
133 * Returns %true if @inode is the first occupant of the !dirty_time IO
134 * lists; otherwise, %false.
136 static bool inode_wb_list_move_locked(struct inode
*inode
,
137 struct bdi_writeback
*wb
,
138 struct list_head
*head
)
140 assert_spin_locked(&wb
->list_lock
);
142 list_move(&inode
->i_wb_list
, head
);
144 /* dirty_time doesn't count as dirty_io until expiration */
145 if (head
!= &wb
->b_dirty_time
)
146 return wb_io_lists_populated(wb
);
148 wb_io_lists_depopulated(wb
);
153 * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list
154 * @inode: inode to be removed
155 * @wb: bdi_writeback @inode is being removed from
157 * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and
158 * clear %WB_has_dirty_io if all are empty afterwards.
160 static void inode_wb_list_del_locked(struct inode
*inode
,
161 struct bdi_writeback
*wb
)
163 assert_spin_locked(&wb
->list_lock
);
165 list_del_init(&inode
->i_wb_list
);
166 wb_io_lists_depopulated(wb
);
169 static void wb_wakeup(struct bdi_writeback
*wb
)
171 spin_lock_bh(&wb
->work_lock
);
172 if (test_bit(WB_registered
, &wb
->state
))
173 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
174 spin_unlock_bh(&wb
->work_lock
);
177 static void wb_queue_work(struct bdi_writeback
*wb
,
178 struct wb_writeback_work
*work
)
180 trace_writeback_queue(wb
->bdi
, work
);
182 spin_lock_bh(&wb
->work_lock
);
183 if (!test_bit(WB_registered
, &wb
->state
)) {
184 if (work
->single_wait
)
185 work
->single_done
= 1;
189 atomic_inc(&work
->done
->cnt
);
190 list_add_tail(&work
->list
, &wb
->work_list
);
191 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
193 spin_unlock_bh(&wb
->work_lock
);
197 * wb_wait_for_completion - wait for completion of bdi_writeback_works
198 * @bdi: bdi work items were issued to
199 * @done: target wb_completion
201 * Wait for one or more work items issued to @bdi with their ->done field
202 * set to @done, which should have been defined with
203 * DEFINE_WB_COMPLETION_ONSTACK(). This function returns after all such
204 * work items are completed. Work items which are waited upon aren't freed
205 * automatically on completion.
207 static void wb_wait_for_completion(struct backing_dev_info
*bdi
,
208 struct wb_completion
*done
)
210 atomic_dec(&done
->cnt
); /* put down the initial count */
211 wait_event(bdi
->wb_waitq
, !atomic_read(&done
->cnt
));
214 #ifdef CONFIG_CGROUP_WRITEBACK
217 * inode_congested - test whether an inode is congested
218 * @inode: inode to test for congestion
219 * @cong_bits: mask of WB_[a]sync_congested bits to test
221 * Tests whether @inode is congested. @cong_bits is the mask of congestion
222 * bits to test and the return value is the mask of set bits.
224 * If cgroup writeback is enabled for @inode, the congestion state is
225 * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
226 * associated with @inode is congested; otherwise, the root wb's congestion
229 int inode_congested(struct inode
*inode
, int cong_bits
)
232 struct bdi_writeback
*wb
= inode_to_wb(inode
);
234 return wb_congested(wb
, cong_bits
);
237 return wb_congested(&inode_to_bdi(inode
)->wb
, cong_bits
);
239 EXPORT_SYMBOL_GPL(inode_congested
);
242 * wb_wait_for_single_work - wait for completion of a single bdi_writeback_work
243 * @bdi: bdi the work item was issued to
244 * @work: work item to wait for
246 * Wait for the completion of @work which was issued to one of @bdi's
247 * bdi_writeback's. The caller must have set @work->single_wait before
248 * issuing it. This wait operates independently fo
249 * wb_wait_for_completion() and also disables automatic freeing of @work.
251 static void wb_wait_for_single_work(struct backing_dev_info
*bdi
,
252 struct wb_writeback_work
*work
)
254 if (WARN_ON_ONCE(!work
->single_wait
))
257 wait_event(bdi
->wb_waitq
, work
->single_done
);
260 * Paired with smp_wmb() in wb_do_writeback() and ensures that all
261 * modifications to @work prior to assertion of ->single_done is
262 * visible to the caller once this function returns.
268 * wb_split_bdi_pages - split nr_pages to write according to bandwidth
269 * @wb: target bdi_writeback to split @nr_pages to
270 * @nr_pages: number of pages to write for the whole bdi
272 * Split @wb's portion of @nr_pages according to @wb's write bandwidth in
273 * relation to the total write bandwidth of all wb's w/ dirty inodes on
276 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
278 unsigned long this_bw
= wb
->avg_write_bandwidth
;
279 unsigned long tot_bw
= atomic_long_read(&wb
->bdi
->tot_write_bandwidth
);
281 if (nr_pages
== LONG_MAX
)
285 * This may be called on clean wb's and proportional distribution
286 * may not make sense, just use the original @nr_pages in those
287 * cases. In general, we wanna err on the side of writing more.
289 if (!tot_bw
|| this_bw
>= tot_bw
)
292 return DIV_ROUND_UP_ULL((u64
)nr_pages
* this_bw
, tot_bw
);
296 * wb_clone_and_queue_work - clone a wb_writeback_work and issue it to a wb
297 * @wb: target bdi_writeback
298 * @base_work: source wb_writeback_work
300 * Try to make a clone of @base_work and issue it to @wb. If cloning
301 * succeeds, %true is returned; otherwise, @base_work is issued directly
302 * and %false is returned. In the latter case, the caller is required to
303 * wait for @base_work's completion using wb_wait_for_single_work().
305 * A clone is auto-freed on completion. @base_work never is.
307 static bool wb_clone_and_queue_work(struct bdi_writeback
*wb
,
308 struct wb_writeback_work
*base_work
)
310 struct wb_writeback_work
*work
;
312 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
316 work
->single_wait
= 0;
320 work
->single_wait
= 1;
322 work
->single_done
= 0;
323 wb_queue_work(wb
, work
);
324 return work
!= base_work
;
328 * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi
329 * @bdi: target backing_dev_info
330 * @base_work: wb_writeback_work to issue
331 * @skip_if_busy: skip wb's which already have writeback in progress
333 * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which
334 * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's
335 * distributed to the busy wbs according to each wb's proportion in the
336 * total active write bandwidth of @bdi.
338 static void bdi_split_work_to_wbs(struct backing_dev_info
*bdi
,
339 struct wb_writeback_work
*base_work
,
342 long nr_pages
= base_work
->nr_pages
;
343 int next_blkcg_id
= 0;
344 struct bdi_writeback
*wb
;
349 if (!bdi_has_dirty_io(bdi
))
353 bdi_for_each_wb(wb
, bdi
, &iter
, next_blkcg_id
) {
354 if (!wb_has_dirty_io(wb
) ||
355 (skip_if_busy
&& writeback_in_progress(wb
)))
358 base_work
->nr_pages
= wb_split_bdi_pages(wb
, nr_pages
);
359 if (!wb_clone_and_queue_work(wb
, base_work
)) {
360 next_blkcg_id
= wb
->blkcg_css
->id
+ 1;
362 wb_wait_for_single_work(bdi
, base_work
);
369 #else /* CONFIG_CGROUP_WRITEBACK */
371 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
376 static void bdi_split_work_to_wbs(struct backing_dev_info
*bdi
,
377 struct wb_writeback_work
*base_work
,
382 if (bdi_has_dirty_io(bdi
) &&
383 (!skip_if_busy
|| !writeback_in_progress(&bdi
->wb
))) {
384 base_work
->auto_free
= 0;
385 base_work
->single_wait
= 0;
386 base_work
->single_done
= 0;
387 wb_queue_work(&bdi
->wb
, base_work
);
391 #endif /* CONFIG_CGROUP_WRITEBACK */
393 void wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
394 bool range_cyclic
, enum wb_reason reason
)
396 struct wb_writeback_work
*work
;
398 if (!wb_has_dirty_io(wb
))
402 * This is WB_SYNC_NONE writeback, so if allocation fails just
403 * wakeup the thread for old dirty data writeback
405 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
407 trace_writeback_nowork(wb
->bdi
);
412 work
->sync_mode
= WB_SYNC_NONE
;
413 work
->nr_pages
= nr_pages
;
414 work
->range_cyclic
= range_cyclic
;
415 work
->reason
= reason
;
418 wb_queue_work(wb
, work
);
422 * wb_start_background_writeback - start background writeback
423 * @wb: bdi_writback to write from
426 * This makes sure WB_SYNC_NONE background writeback happens. When
427 * this function returns, it is only guaranteed that for given wb
428 * some IO is happening if we are over background dirty threshold.
429 * Caller need not hold sb s_umount semaphore.
431 void wb_start_background_writeback(struct bdi_writeback
*wb
)
434 * We just wake up the flusher thread. It will perform background
435 * writeback as soon as there is no other work to do.
437 trace_writeback_wake_background(wb
->bdi
);
442 * Remove the inode from the writeback list it is on.
444 void inode_wb_list_del(struct inode
*inode
)
446 struct bdi_writeback
*wb
= inode_to_wb(inode
);
448 spin_lock(&wb
->list_lock
);
449 inode_wb_list_del_locked(inode
, wb
);
450 spin_unlock(&wb
->list_lock
);
454 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
455 * furthest end of its superblock's dirty-inode list.
457 * Before stamping the inode's ->dirtied_when, we check to see whether it is
458 * already the most-recently-dirtied inode on the b_dirty list. If that is
459 * the case then the inode must have been redirtied while it was being written
460 * out and we don't reset its dirtied_when.
462 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
464 if (!list_empty(&wb
->b_dirty
)) {
467 tail
= wb_inode(wb
->b_dirty
.next
);
468 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
469 inode
->dirtied_when
= jiffies
;
471 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty
);
475 * requeue inode for re-scanning after bdi->b_io list is exhausted.
477 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
479 inode_wb_list_move_locked(inode
, wb
, &wb
->b_more_io
);
482 static void inode_sync_complete(struct inode
*inode
)
484 inode
->i_state
&= ~I_SYNC
;
485 /* If inode is clean an unused, put it into LRU now... */
486 inode_add_lru(inode
);
487 /* Waiters must see I_SYNC cleared before being woken up */
489 wake_up_bit(&inode
->i_state
, __I_SYNC
);
492 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
494 bool ret
= time_after(inode
->dirtied_when
, t
);
497 * For inodes being constantly redirtied, dirtied_when can get stuck.
498 * It _appears_ to be in the future, but is actually in distant past.
499 * This test is necessary to prevent such wrapped-around relative times
500 * from permanently stopping the whole bdi writeback.
502 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
507 #define EXPIRE_DIRTY_ATIME 0x0001
510 * Move expired (dirtied before work->older_than_this) dirty inodes from
511 * @delaying_queue to @dispatch_queue.
513 static int move_expired_inodes(struct list_head
*delaying_queue
,
514 struct list_head
*dispatch_queue
,
516 struct wb_writeback_work
*work
)
518 unsigned long *older_than_this
= NULL
;
519 unsigned long expire_time
;
521 struct list_head
*pos
, *node
;
522 struct super_block
*sb
= NULL
;
527 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
528 older_than_this
= work
->older_than_this
;
529 else if (!work
->for_sync
) {
530 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
531 older_than_this
= &expire_time
;
533 while (!list_empty(delaying_queue
)) {
534 inode
= wb_inode(delaying_queue
->prev
);
535 if (older_than_this
&&
536 inode_dirtied_after(inode
, *older_than_this
))
538 list_move(&inode
->i_wb_list
, &tmp
);
540 if (flags
& EXPIRE_DIRTY_ATIME
)
541 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
542 if (sb_is_blkdev_sb(inode
->i_sb
))
544 if (sb
&& sb
!= inode
->i_sb
)
549 /* just one sb in list, splice to dispatch_queue and we're done */
551 list_splice(&tmp
, dispatch_queue
);
555 /* Move inodes from one superblock together */
556 while (!list_empty(&tmp
)) {
557 sb
= wb_inode(tmp
.prev
)->i_sb
;
558 list_for_each_prev_safe(pos
, node
, &tmp
) {
559 inode
= wb_inode(pos
);
560 if (inode
->i_sb
== sb
)
561 list_move(&inode
->i_wb_list
, dispatch_queue
);
569 * Queue all expired dirty inodes for io, eldest first.
571 * newly dirtied b_dirty b_io b_more_io
572 * =============> gf edc BA
574 * newly dirtied b_dirty b_io b_more_io
575 * =============> g fBAedc
577 * +--> dequeue for IO
579 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
583 assert_spin_locked(&wb
->list_lock
);
584 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
585 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
586 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
587 EXPIRE_DIRTY_ATIME
, work
);
589 wb_io_lists_populated(wb
);
590 trace_writeback_queue_io(wb
, work
, moved
);
593 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
597 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
598 trace_writeback_write_inode_start(inode
, wbc
);
599 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
600 trace_writeback_write_inode(inode
, wbc
);
607 * Wait for writeback on an inode to complete. Called with i_lock held.
608 * Caller must make sure inode cannot go away when we drop i_lock.
610 static void __inode_wait_for_writeback(struct inode
*inode
)
611 __releases(inode
->i_lock
)
612 __acquires(inode
->i_lock
)
614 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
615 wait_queue_head_t
*wqh
;
617 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
618 while (inode
->i_state
& I_SYNC
) {
619 spin_unlock(&inode
->i_lock
);
620 __wait_on_bit(wqh
, &wq
, bit_wait
,
621 TASK_UNINTERRUPTIBLE
);
622 spin_lock(&inode
->i_lock
);
627 * Wait for writeback on an inode to complete. Caller must have inode pinned.
629 void inode_wait_for_writeback(struct inode
*inode
)
631 spin_lock(&inode
->i_lock
);
632 __inode_wait_for_writeback(inode
);
633 spin_unlock(&inode
->i_lock
);
637 * Sleep until I_SYNC is cleared. This function must be called with i_lock
638 * held and drops it. It is aimed for callers not holding any inode reference
639 * so once i_lock is dropped, inode can go away.
641 static void inode_sleep_on_writeback(struct inode
*inode
)
642 __releases(inode
->i_lock
)
645 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
648 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
649 sleep
= inode
->i_state
& I_SYNC
;
650 spin_unlock(&inode
->i_lock
);
653 finish_wait(wqh
, &wait
);
657 * Find proper writeback list for the inode depending on its current state and
658 * possibly also change of its state while we were doing writeback. Here we
659 * handle things such as livelock prevention or fairness of writeback among
660 * inodes. This function can be called only by flusher thread - noone else
661 * processes all inodes in writeback lists and requeueing inodes behind flusher
662 * thread's back can have unexpected consequences.
664 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
665 struct writeback_control
*wbc
)
667 if (inode
->i_state
& I_FREEING
)
671 * Sync livelock prevention. Each inode is tagged and synced in one
672 * shot. If still dirty, it will be redirty_tail()'ed below. Update
673 * the dirty time to prevent enqueue and sync it again.
675 if ((inode
->i_state
& I_DIRTY
) &&
676 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
677 inode
->dirtied_when
= jiffies
;
679 if (wbc
->pages_skipped
) {
681 * writeback is not making progress due to locked
682 * buffers. Skip this inode for now.
684 redirty_tail(inode
, wb
);
688 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
690 * We didn't write back all the pages. nfs_writepages()
691 * sometimes bales out without doing anything.
693 if (wbc
->nr_to_write
<= 0) {
694 /* Slice used up. Queue for next turn. */
695 requeue_io(inode
, wb
);
698 * Writeback blocked by something other than
699 * congestion. Delay the inode for some time to
700 * avoid spinning on the CPU (100% iowait)
701 * retrying writeback of the dirty page/inode
702 * that cannot be performed immediately.
704 redirty_tail(inode
, wb
);
706 } else if (inode
->i_state
& I_DIRTY
) {
708 * Filesystems can dirty the inode during writeback operations,
709 * such as delayed allocation during submission or metadata
710 * updates after data IO completion.
712 redirty_tail(inode
, wb
);
713 } else if (inode
->i_state
& I_DIRTY_TIME
) {
714 inode
->dirtied_when
= jiffies
;
715 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty_time
);
717 /* The inode is clean. Remove from writeback lists. */
718 inode_wb_list_del_locked(inode
, wb
);
723 * Write out an inode and its dirty pages. Do not update the writeback list
724 * linkage. That is left to the caller. The caller is also responsible for
725 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
728 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
730 struct address_space
*mapping
= inode
->i_mapping
;
731 long nr_to_write
= wbc
->nr_to_write
;
735 WARN_ON(!(inode
->i_state
& I_SYNC
));
737 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
739 ret
= do_writepages(mapping
, wbc
);
742 * Make sure to wait on the data before writing out the metadata.
743 * This is important for filesystems that modify metadata on data
744 * I/O completion. We don't do it for sync(2) writeback because it has a
745 * separate, external IO completion path and ->sync_fs for guaranteeing
746 * inode metadata is written back correctly.
748 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
749 int err
= filemap_fdatawait(mapping
);
755 * Some filesystems may redirty the inode during the writeback
756 * due to delalloc, clear dirty metadata flags right before
759 spin_lock(&inode
->i_lock
);
761 dirty
= inode
->i_state
& I_DIRTY
;
762 if (inode
->i_state
& I_DIRTY_TIME
) {
763 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
764 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
765 unlikely(time_after(jiffies
,
766 (inode
->dirtied_time_when
+
767 dirtytime_expire_interval
* HZ
)))) {
768 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
769 trace_writeback_lazytime(inode
);
772 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
773 inode
->i_state
&= ~dirty
;
776 * Paired with smp_mb() in __mark_inode_dirty(). This allows
777 * __mark_inode_dirty() to test i_state without grabbing i_lock -
778 * either they see the I_DIRTY bits cleared or we see the dirtied
781 * I_DIRTY_PAGES is always cleared together above even if @mapping
782 * still has dirty pages. The flag is reinstated after smp_mb() if
783 * necessary. This guarantees that either __mark_inode_dirty()
784 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
788 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
789 inode
->i_state
|= I_DIRTY_PAGES
;
791 spin_unlock(&inode
->i_lock
);
793 if (dirty
& I_DIRTY_TIME
)
794 mark_inode_dirty_sync(inode
);
795 /* Don't write the inode if only I_DIRTY_PAGES was set */
796 if (dirty
& ~I_DIRTY_PAGES
) {
797 int err
= write_inode(inode
, wbc
);
801 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
806 * Write out an inode's dirty pages. Either the caller has an active reference
807 * on the inode or the inode has I_WILL_FREE set.
809 * This function is designed to be called for writing back one inode which
810 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
811 * and does more profound writeback list handling in writeback_sb_inodes().
814 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
815 struct writeback_control
*wbc
)
819 spin_lock(&inode
->i_lock
);
820 if (!atomic_read(&inode
->i_count
))
821 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
823 WARN_ON(inode
->i_state
& I_WILL_FREE
);
825 if (inode
->i_state
& I_SYNC
) {
826 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
829 * It's a data-integrity sync. We must wait. Since callers hold
830 * inode reference or inode has I_WILL_FREE set, it cannot go
833 __inode_wait_for_writeback(inode
);
835 WARN_ON(inode
->i_state
& I_SYNC
);
837 * Skip inode if it is clean and we have no outstanding writeback in
838 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
839 * function since flusher thread may be doing for example sync in
840 * parallel and if we move the inode, it could get skipped. So here we
841 * make sure inode is on some writeback list and leave it there unless
842 * we have completely cleaned the inode.
844 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
845 (wbc
->sync_mode
!= WB_SYNC_ALL
||
846 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
848 inode
->i_state
|= I_SYNC
;
849 spin_unlock(&inode
->i_lock
);
851 ret
= __writeback_single_inode(inode
, wbc
);
853 spin_lock(&wb
->list_lock
);
854 spin_lock(&inode
->i_lock
);
856 * If inode is clean, remove it from writeback lists. Otherwise don't
857 * touch it. See comment above for explanation.
859 if (!(inode
->i_state
& I_DIRTY_ALL
))
860 inode_wb_list_del_locked(inode
, wb
);
861 spin_unlock(&wb
->list_lock
);
862 inode_sync_complete(inode
);
864 spin_unlock(&inode
->i_lock
);
868 static long writeback_chunk_size(struct bdi_writeback
*wb
,
869 struct wb_writeback_work
*work
)
874 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
875 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
876 * here avoids calling into writeback_inodes_wb() more than once.
878 * The intended call sequence for WB_SYNC_ALL writeback is:
881 * writeback_sb_inodes() <== called only once
882 * write_cache_pages() <== called once for each inode
883 * (quickly) tag currently dirty pages
884 * (maybe slowly) sync all tagged pages
886 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
889 pages
= min(wb
->avg_write_bandwidth
/ 2,
890 global_wb_domain
.dirty_limit
/ DIRTY_SCOPE
);
891 pages
= min(pages
, work
->nr_pages
);
892 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
893 MIN_WRITEBACK_PAGES
);
900 * Write a portion of b_io inodes which belong to @sb.
902 * Return the number of pages and/or inodes written.
904 static long writeback_sb_inodes(struct super_block
*sb
,
905 struct bdi_writeback
*wb
,
906 struct wb_writeback_work
*work
)
908 struct writeback_control wbc
= {
909 .sync_mode
= work
->sync_mode
,
910 .tagged_writepages
= work
->tagged_writepages
,
911 .for_kupdate
= work
->for_kupdate
,
912 .for_background
= work
->for_background
,
913 .for_sync
= work
->for_sync
,
914 .range_cyclic
= work
->range_cyclic
,
916 .range_end
= LLONG_MAX
,
918 unsigned long start_time
= jiffies
;
920 long wrote
= 0; /* count both pages and inodes */
922 while (!list_empty(&wb
->b_io
)) {
923 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
925 if (inode
->i_sb
!= sb
) {
928 * We only want to write back data for this
929 * superblock, move all inodes not belonging
930 * to it back onto the dirty list.
932 redirty_tail(inode
, wb
);
937 * The inode belongs to a different superblock.
938 * Bounce back to the caller to unpin this and
939 * pin the next superblock.
945 * Don't bother with new inodes or inodes being freed, first
946 * kind does not need periodic writeout yet, and for the latter
947 * kind writeout is handled by the freer.
949 spin_lock(&inode
->i_lock
);
950 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
951 spin_unlock(&inode
->i_lock
);
952 redirty_tail(inode
, wb
);
955 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
957 * If this inode is locked for writeback and we are not
958 * doing writeback-for-data-integrity, move it to
959 * b_more_io so that writeback can proceed with the
960 * other inodes on s_io.
962 * We'll have another go at writing back this inode
963 * when we completed a full scan of b_io.
965 spin_unlock(&inode
->i_lock
);
966 requeue_io(inode
, wb
);
967 trace_writeback_sb_inodes_requeue(inode
);
970 spin_unlock(&wb
->list_lock
);
973 * We already requeued the inode if it had I_SYNC set and we
974 * are doing WB_SYNC_NONE writeback. So this catches only the
977 if (inode
->i_state
& I_SYNC
) {
978 /* Wait for I_SYNC. This function drops i_lock... */
979 inode_sleep_on_writeback(inode
);
980 /* Inode may be gone, start again */
981 spin_lock(&wb
->list_lock
);
984 inode
->i_state
|= I_SYNC
;
985 spin_unlock(&inode
->i_lock
);
987 write_chunk
= writeback_chunk_size(wb
, work
);
988 wbc
.nr_to_write
= write_chunk
;
989 wbc
.pages_skipped
= 0;
992 * We use I_SYNC to pin the inode in memory. While it is set
993 * evict_inode() will wait so the inode cannot be freed.
995 __writeback_single_inode(inode
, &wbc
);
997 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
998 wrote
+= write_chunk
- wbc
.nr_to_write
;
999 spin_lock(&wb
->list_lock
);
1000 spin_lock(&inode
->i_lock
);
1001 if (!(inode
->i_state
& I_DIRTY_ALL
))
1003 requeue_inode(inode
, wb
, &wbc
);
1004 inode_sync_complete(inode
);
1005 spin_unlock(&inode
->i_lock
);
1006 cond_resched_lock(&wb
->list_lock
);
1008 * bail out to wb_writeback() often enough to check
1009 * background threshold and other termination conditions.
1012 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
1014 if (work
->nr_pages
<= 0)
1021 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
1022 struct wb_writeback_work
*work
)
1024 unsigned long start_time
= jiffies
;
1027 while (!list_empty(&wb
->b_io
)) {
1028 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
1029 struct super_block
*sb
= inode
->i_sb
;
1031 if (!trylock_super(sb
)) {
1033 * trylock_super() may fail consistently due to
1034 * s_umount being grabbed by someone else. Don't use
1035 * requeue_io() to avoid busy retrying the inode/sb.
1037 redirty_tail(inode
, wb
);
1040 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
1041 up_read(&sb
->s_umount
);
1043 /* refer to the same tests at the end of writeback_sb_inodes */
1045 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
1047 if (work
->nr_pages
<= 0)
1051 /* Leave any unwritten inodes on b_io */
1055 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
1056 enum wb_reason reason
)
1058 struct wb_writeback_work work
= {
1059 .nr_pages
= nr_pages
,
1060 .sync_mode
= WB_SYNC_NONE
,
1065 spin_lock(&wb
->list_lock
);
1066 if (list_empty(&wb
->b_io
))
1067 queue_io(wb
, &work
);
1068 __writeback_inodes_wb(wb
, &work
);
1069 spin_unlock(&wb
->list_lock
);
1071 return nr_pages
- work
.nr_pages
;
1074 static bool over_bground_thresh(struct bdi_writeback
*wb
)
1076 unsigned long background_thresh
, dirty_thresh
;
1078 global_dirty_limits(&background_thresh
, &dirty_thresh
);
1080 if (global_page_state(NR_FILE_DIRTY
) +
1081 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
)
1084 if (wb_stat(wb
, WB_RECLAIMABLE
) > wb_calc_thresh(wb
, background_thresh
))
1091 * Explicit flushing or periodic writeback of "old" data.
1093 * Define "old": the first time one of an inode's pages is dirtied, we mark the
1094 * dirtying-time in the inode's address_space. So this periodic writeback code
1095 * just walks the superblock inode list, writing back any inodes which are
1096 * older than a specific point in time.
1098 * Try to run once per dirty_writeback_interval. But if a writeback event
1099 * takes longer than a dirty_writeback_interval interval, then leave a
1102 * older_than_this takes precedence over nr_to_write. So we'll only write back
1103 * all dirty pages if they are all attached to "old" mappings.
1105 static long wb_writeback(struct bdi_writeback
*wb
,
1106 struct wb_writeback_work
*work
)
1108 unsigned long wb_start
= jiffies
;
1109 long nr_pages
= work
->nr_pages
;
1110 unsigned long oldest_jif
;
1111 struct inode
*inode
;
1114 oldest_jif
= jiffies
;
1115 work
->older_than_this
= &oldest_jif
;
1117 spin_lock(&wb
->list_lock
);
1120 * Stop writeback when nr_pages has been consumed
1122 if (work
->nr_pages
<= 0)
1126 * Background writeout and kupdate-style writeback may
1127 * run forever. Stop them if there is other work to do
1128 * so that e.g. sync can proceed. They'll be restarted
1129 * after the other works are all done.
1131 if ((work
->for_background
|| work
->for_kupdate
) &&
1132 !list_empty(&wb
->work_list
))
1136 * For background writeout, stop when we are below the
1137 * background dirty threshold
1139 if (work
->for_background
&& !over_bground_thresh(wb
))
1143 * Kupdate and background works are special and we want to
1144 * include all inodes that need writing. Livelock avoidance is
1145 * handled by these works yielding to any other work so we are
1148 if (work
->for_kupdate
) {
1149 oldest_jif
= jiffies
-
1150 msecs_to_jiffies(dirty_expire_interval
* 10);
1151 } else if (work
->for_background
)
1152 oldest_jif
= jiffies
;
1154 trace_writeback_start(wb
->bdi
, work
);
1155 if (list_empty(&wb
->b_io
))
1158 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
1160 progress
= __writeback_inodes_wb(wb
, work
);
1161 trace_writeback_written(wb
->bdi
, work
);
1163 wb_update_bandwidth(wb
, wb_start
);
1166 * Did we write something? Try for more
1168 * Dirty inodes are moved to b_io for writeback in batches.
1169 * The completion of the current batch does not necessarily
1170 * mean the overall work is done. So we keep looping as long
1171 * as made some progress on cleaning pages or inodes.
1176 * No more inodes for IO, bail
1178 if (list_empty(&wb
->b_more_io
))
1181 * Nothing written. Wait for some inode to
1182 * become available for writeback. Otherwise
1183 * we'll just busyloop.
1185 if (!list_empty(&wb
->b_more_io
)) {
1186 trace_writeback_wait(wb
->bdi
, work
);
1187 inode
= wb_inode(wb
->b_more_io
.prev
);
1188 spin_lock(&inode
->i_lock
);
1189 spin_unlock(&wb
->list_lock
);
1190 /* This function drops i_lock... */
1191 inode_sleep_on_writeback(inode
);
1192 spin_lock(&wb
->list_lock
);
1195 spin_unlock(&wb
->list_lock
);
1197 return nr_pages
- work
->nr_pages
;
1201 * Return the next wb_writeback_work struct that hasn't been processed yet.
1203 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
1205 struct wb_writeback_work
*work
= NULL
;
1207 spin_lock_bh(&wb
->work_lock
);
1208 if (!list_empty(&wb
->work_list
)) {
1209 work
= list_entry(wb
->work_list
.next
,
1210 struct wb_writeback_work
, list
);
1211 list_del_init(&work
->list
);
1213 spin_unlock_bh(&wb
->work_lock
);
1218 * Add in the number of potentially dirty inodes, because each inode
1219 * write can dirty pagecache in the underlying blockdev.
1221 static unsigned long get_nr_dirty_pages(void)
1223 return global_page_state(NR_FILE_DIRTY
) +
1224 global_page_state(NR_UNSTABLE_NFS
) +
1225 get_nr_dirty_inodes();
1228 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1230 if (over_bground_thresh(wb
)) {
1232 struct wb_writeback_work work
= {
1233 .nr_pages
= LONG_MAX
,
1234 .sync_mode
= WB_SYNC_NONE
,
1235 .for_background
= 1,
1237 .reason
= WB_REASON_BACKGROUND
,
1240 return wb_writeback(wb
, &work
);
1246 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1248 unsigned long expired
;
1252 * When set to zero, disable periodic writeback
1254 if (!dirty_writeback_interval
)
1257 expired
= wb
->last_old_flush
+
1258 msecs_to_jiffies(dirty_writeback_interval
* 10);
1259 if (time_before(jiffies
, expired
))
1262 wb
->last_old_flush
= jiffies
;
1263 nr_pages
= get_nr_dirty_pages();
1266 struct wb_writeback_work work
= {
1267 .nr_pages
= nr_pages
,
1268 .sync_mode
= WB_SYNC_NONE
,
1271 .reason
= WB_REASON_PERIODIC
,
1274 return wb_writeback(wb
, &work
);
1281 * Retrieve work items and do the writeback they describe
1283 static long wb_do_writeback(struct bdi_writeback
*wb
)
1285 struct wb_writeback_work
*work
;
1288 set_bit(WB_writeback_running
, &wb
->state
);
1289 while ((work
= get_next_work_item(wb
)) != NULL
) {
1290 struct wb_completion
*done
= work
->done
;
1291 bool need_wake_up
= false;
1293 trace_writeback_exec(wb
->bdi
, work
);
1295 wrote
+= wb_writeback(wb
, work
);
1297 if (work
->single_wait
) {
1298 WARN_ON_ONCE(work
->auto_free
);
1299 /* paired w/ rmb in wb_wait_for_single_work() */
1301 work
->single_done
= 1;
1302 need_wake_up
= true;
1303 } else if (work
->auto_free
) {
1307 if (done
&& atomic_dec_and_test(&done
->cnt
))
1308 need_wake_up
= true;
1311 wake_up_all(&wb
->bdi
->wb_waitq
);
1315 * Check for periodic writeback, kupdated() style
1317 wrote
+= wb_check_old_data_flush(wb
);
1318 wrote
+= wb_check_background_flush(wb
);
1319 clear_bit(WB_writeback_running
, &wb
->state
);
1325 * Handle writeback of dirty data for the device backed by this bdi. Also
1326 * reschedules periodically and does kupdated style flushing.
1328 void wb_workfn(struct work_struct
*work
)
1330 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1331 struct bdi_writeback
, dwork
);
1334 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1335 current
->flags
|= PF_SWAPWRITE
;
1337 if (likely(!current_is_workqueue_rescuer() ||
1338 !test_bit(WB_registered
, &wb
->state
))) {
1340 * The normal path. Keep writing back @wb until its
1341 * work_list is empty. Note that this path is also taken
1342 * if @wb is shutting down even when we're running off the
1343 * rescuer as work_list needs to be drained.
1346 pages_written
= wb_do_writeback(wb
);
1347 trace_writeback_pages_written(pages_written
);
1348 } while (!list_empty(&wb
->work_list
));
1351 * bdi_wq can't get enough workers and we're running off
1352 * the emergency worker. Don't hog it. Hopefully, 1024 is
1353 * enough for efficient IO.
1355 pages_written
= writeback_inodes_wb(wb
, 1024,
1356 WB_REASON_FORKER_THREAD
);
1357 trace_writeback_pages_written(pages_written
);
1360 if (!list_empty(&wb
->work_list
))
1361 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1362 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1363 wb_wakeup_delayed(wb
);
1365 current
->flags
&= ~PF_SWAPWRITE
;
1369 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1372 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1374 struct backing_dev_info
*bdi
;
1377 nr_pages
= get_nr_dirty_pages();
1380 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1381 struct bdi_writeback
*wb
;
1382 struct wb_iter iter
;
1384 if (!bdi_has_dirty_io(bdi
))
1387 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1388 wb_start_writeback(wb
, wb_split_bdi_pages(wb
, nr_pages
),
1395 * Wake up bdi's periodically to make sure dirtytime inodes gets
1396 * written back periodically. We deliberately do *not* check the
1397 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1398 * kernel to be constantly waking up once there are any dirtytime
1399 * inodes on the system. So instead we define a separate delayed work
1400 * function which gets called much more rarely. (By default, only
1401 * once every 12 hours.)
1403 * If there is any other write activity going on in the file system,
1404 * this function won't be necessary. But if the only thing that has
1405 * happened on the file system is a dirtytime inode caused by an atime
1406 * update, we need this infrastructure below to make sure that inode
1407 * eventually gets pushed out to disk.
1409 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1410 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1412 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1414 struct backing_dev_info
*bdi
;
1417 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1418 struct bdi_writeback
*wb
;
1419 struct wb_iter iter
;
1421 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1422 if (!list_empty(&bdi
->wb
.b_dirty_time
))
1423 wb_wakeup(&bdi
->wb
);
1426 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1429 static int __init
start_dirtytime_writeback(void)
1431 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1434 __initcall(start_dirtytime_writeback
);
1436 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1437 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1441 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1442 if (ret
== 0 && write
)
1443 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1447 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1449 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1450 struct dentry
*dentry
;
1451 const char *name
= "?";
1453 dentry
= d_find_alias(inode
);
1455 spin_lock(&dentry
->d_lock
);
1456 name
= (const char *) dentry
->d_name
.name
;
1459 "%s(%d): dirtied inode %lu (%s) on %s\n",
1460 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1461 name
, inode
->i_sb
->s_id
);
1463 spin_unlock(&dentry
->d_lock
);
1470 * __mark_inode_dirty - internal function
1471 * @inode: inode to mark
1472 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1473 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1474 * mark_inode_dirty_sync.
1476 * Put the inode on the super block's dirty list.
1478 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1479 * dirty list only if it is hashed or if it refers to a blockdev.
1480 * If it was not hashed, it will never be added to the dirty list
1481 * even if it is later hashed, as it will have been marked dirty already.
1483 * In short, make sure you hash any inodes _before_ you start marking
1486 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1487 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1488 * the kernel-internal blockdev inode represents the dirtying time of the
1489 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1490 * page->mapping->host, so the page-dirtying time is recorded in the internal
1493 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1494 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1496 struct super_block
*sb
= inode
->i_sb
;
1499 trace_writeback_mark_inode_dirty(inode
, flags
);
1502 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1503 * dirty the inode itself
1505 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
1506 trace_writeback_dirty_inode_start(inode
, flags
);
1508 if (sb
->s_op
->dirty_inode
)
1509 sb
->s_op
->dirty_inode(inode
, flags
);
1511 trace_writeback_dirty_inode(inode
, flags
);
1513 if (flags
& I_DIRTY_INODE
)
1514 flags
&= ~I_DIRTY_TIME
;
1515 dirtytime
= flags
& I_DIRTY_TIME
;
1518 * Paired with smp_mb() in __writeback_single_inode() for the
1519 * following lockless i_state test. See there for details.
1523 if (((inode
->i_state
& flags
) == flags
) ||
1524 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
1527 if (unlikely(block_dump
))
1528 block_dump___mark_inode_dirty(inode
);
1530 spin_lock(&inode
->i_lock
);
1531 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
1532 goto out_unlock_inode
;
1533 if ((inode
->i_state
& flags
) != flags
) {
1534 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1536 inode_attach_wb(inode
, NULL
);
1538 if (flags
& I_DIRTY_INODE
)
1539 inode
->i_state
&= ~I_DIRTY_TIME
;
1540 inode
->i_state
|= flags
;
1543 * If the inode is being synced, just update its dirty state.
1544 * The unlocker will place the inode on the appropriate
1545 * superblock list, based upon its state.
1547 if (inode
->i_state
& I_SYNC
)
1548 goto out_unlock_inode
;
1551 * Only add valid (hashed) inodes to the superblock's
1552 * dirty list. Add blockdev inodes as well.
1554 if (!S_ISBLK(inode
->i_mode
)) {
1555 if (inode_unhashed(inode
))
1556 goto out_unlock_inode
;
1558 if (inode
->i_state
& I_FREEING
)
1559 goto out_unlock_inode
;
1562 * If the inode was already on b_dirty/b_io/b_more_io, don't
1563 * reposition it (that would break b_dirty time-ordering).
1566 struct bdi_writeback
*wb
= inode_to_wb(inode
);
1567 struct list_head
*dirty_list
;
1568 bool wakeup_bdi
= false;
1570 spin_unlock(&inode
->i_lock
);
1571 spin_lock(&wb
->list_lock
);
1573 WARN(bdi_cap_writeback_dirty(wb
->bdi
) &&
1574 !test_bit(WB_registered
, &wb
->state
),
1575 "bdi-%s not registered\n", wb
->bdi
->name
);
1577 inode
->dirtied_when
= jiffies
;
1579 inode
->dirtied_time_when
= jiffies
;
1581 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
1582 dirty_list
= &wb
->b_dirty
;
1584 dirty_list
= &wb
->b_dirty_time
;
1586 wakeup_bdi
= inode_wb_list_move_locked(inode
, wb
,
1589 spin_unlock(&wb
->list_lock
);
1590 trace_writeback_dirty_inode_enqueue(inode
);
1593 * If this is the first dirty inode for this bdi,
1594 * we have to wake-up the corresponding bdi thread
1595 * to make sure background write-back happens
1598 if (bdi_cap_writeback_dirty(wb
->bdi
) && wakeup_bdi
)
1599 wb_wakeup_delayed(wb
);
1604 spin_unlock(&inode
->i_lock
);
1607 EXPORT_SYMBOL(__mark_inode_dirty
);
1609 static void wait_sb_inodes(struct super_block
*sb
)
1611 struct inode
*inode
, *old_inode
= NULL
;
1614 * We need to be protected against the filesystem going from
1615 * r/o to r/w or vice versa.
1617 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1619 spin_lock(&inode_sb_list_lock
);
1622 * Data integrity sync. Must wait for all pages under writeback,
1623 * because there may have been pages dirtied before our sync
1624 * call, but which had writeout started before we write it out.
1625 * In which case, the inode may not be on the dirty list, but
1626 * we still have to wait for that writeout.
1628 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1629 struct address_space
*mapping
= inode
->i_mapping
;
1631 spin_lock(&inode
->i_lock
);
1632 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1633 (mapping
->nrpages
== 0)) {
1634 spin_unlock(&inode
->i_lock
);
1638 spin_unlock(&inode
->i_lock
);
1639 spin_unlock(&inode_sb_list_lock
);
1642 * We hold a reference to 'inode' so it couldn't have been
1643 * removed from s_inodes list while we dropped the
1644 * inode_sb_list_lock. We cannot iput the inode now as we can
1645 * be holding the last reference and we cannot iput it under
1646 * inode_sb_list_lock. So we keep the reference and iput it
1652 filemap_fdatawait(mapping
);
1656 spin_lock(&inode_sb_list_lock
);
1658 spin_unlock(&inode_sb_list_lock
);
1662 static void __writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
,
1663 enum wb_reason reason
, bool skip_if_busy
)
1665 DEFINE_WB_COMPLETION_ONSTACK(done
);
1666 struct wb_writeback_work work
= {
1668 .sync_mode
= WB_SYNC_NONE
,
1669 .tagged_writepages
= 1,
1674 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1676 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1678 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1680 bdi_split_work_to_wbs(sb
->s_bdi
, &work
, skip_if_busy
);
1681 wb_wait_for_completion(bdi
, &done
);
1685 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1686 * @sb: the superblock
1687 * @nr: the number of pages to write
1688 * @reason: reason why some writeback work initiated
1690 * Start writeback on some inodes on this super_block. No guarantees are made
1691 * on how many (if any) will be written, and this function does not wait
1692 * for IO completion of submitted IO.
1694 void writeback_inodes_sb_nr(struct super_block
*sb
,
1696 enum wb_reason reason
)
1698 __writeback_inodes_sb_nr(sb
, nr
, reason
, false);
1700 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1703 * writeback_inodes_sb - writeback dirty inodes from given super_block
1704 * @sb: the superblock
1705 * @reason: reason why some writeback work was initiated
1707 * Start writeback on some inodes on this super_block. No guarantees are made
1708 * on how many (if any) will be written, and this function does not wait
1709 * for IO completion of submitted IO.
1711 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1713 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1715 EXPORT_SYMBOL(writeback_inodes_sb
);
1718 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1719 * @sb: the superblock
1720 * @nr: the number of pages to write
1721 * @reason: the reason of writeback
1723 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1724 * Returns 1 if writeback was started, 0 if not.
1726 bool try_to_writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
,
1727 enum wb_reason reason
)
1729 if (!down_read_trylock(&sb
->s_umount
))
1732 __writeback_inodes_sb_nr(sb
, nr
, reason
, true);
1733 up_read(&sb
->s_umount
);
1736 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1739 * try_to_writeback_inodes_sb - try to start writeback if none underway
1740 * @sb: the superblock
1741 * @reason: reason why some writeback work was initiated
1743 * Implement by try_to_writeback_inodes_sb_nr()
1744 * Returns 1 if writeback was started, 0 if not.
1746 bool try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1748 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1750 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1753 * sync_inodes_sb - sync sb inode pages
1754 * @sb: the superblock
1756 * This function writes and waits on any dirty inode belonging to this
1759 void sync_inodes_sb(struct super_block
*sb
)
1761 DEFINE_WB_COMPLETION_ONSTACK(done
);
1762 struct wb_writeback_work work
= {
1764 .sync_mode
= WB_SYNC_ALL
,
1765 .nr_pages
= LONG_MAX
,
1768 .reason
= WB_REASON_SYNC
,
1771 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1773 /* Nothing to do? */
1774 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1776 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1778 bdi_split_work_to_wbs(bdi
, &work
, false);
1779 wb_wait_for_completion(bdi
, &done
);
1783 EXPORT_SYMBOL(sync_inodes_sb
);
1786 * write_inode_now - write an inode to disk
1787 * @inode: inode to write to disk
1788 * @sync: whether the write should be synchronous or not
1790 * This function commits an inode to disk immediately if it is dirty. This is
1791 * primarily needed by knfsd.
1793 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1795 int write_inode_now(struct inode
*inode
, int sync
)
1797 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1798 struct writeback_control wbc
= {
1799 .nr_to_write
= LONG_MAX
,
1800 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1802 .range_end
= LLONG_MAX
,
1805 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1806 wbc
.nr_to_write
= 0;
1809 return writeback_single_inode(inode
, wb
, &wbc
);
1811 EXPORT_SYMBOL(write_inode_now
);
1814 * sync_inode - write an inode and its pages to disk.
1815 * @inode: the inode to sync
1816 * @wbc: controls the writeback mode
1818 * sync_inode() will write an inode and its pages to disk. It will also
1819 * correctly update the inode on its superblock's dirty inode lists and will
1820 * update inode->i_state.
1822 * The caller must have a ref on the inode.
1824 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1826 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1828 EXPORT_SYMBOL(sync_inode
);
1831 * sync_inode_metadata - write an inode to disk
1832 * @inode: the inode to sync
1833 * @wait: wait for I/O to complete.
1835 * Write an inode to disk and adjust its dirty state after completion.
1837 * Note: only writes the actual inode, no associated data or other metadata.
1839 int sync_inode_metadata(struct inode
*inode
, int wait
)
1841 struct writeback_control wbc
= {
1842 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1843 .nr_to_write
= 0, /* metadata-only */
1846 return sync_inode(inode
, &wbc
);
1848 EXPORT_SYMBOL(sync_inode_metadata
);