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
);
295 #else /* CONFIG_CGROUP_WRITEBACK */
297 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
302 #endif /* CONFIG_CGROUP_WRITEBACK */
304 void wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
305 bool range_cyclic
, enum wb_reason reason
)
307 struct wb_writeback_work
*work
;
309 if (!wb_has_dirty_io(wb
))
313 * This is WB_SYNC_NONE writeback, so if allocation fails just
314 * wakeup the thread for old dirty data writeback
316 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
318 trace_writeback_nowork(wb
->bdi
);
323 work
->sync_mode
= WB_SYNC_NONE
;
324 work
->nr_pages
= nr_pages
;
325 work
->range_cyclic
= range_cyclic
;
326 work
->reason
= reason
;
329 wb_queue_work(wb
, work
);
333 * wb_start_background_writeback - start background writeback
334 * @wb: bdi_writback to write from
337 * This makes sure WB_SYNC_NONE background writeback happens. When
338 * this function returns, it is only guaranteed that for given wb
339 * some IO is happening if we are over background dirty threshold.
340 * Caller need not hold sb s_umount semaphore.
342 void wb_start_background_writeback(struct bdi_writeback
*wb
)
345 * We just wake up the flusher thread. It will perform background
346 * writeback as soon as there is no other work to do.
348 trace_writeback_wake_background(wb
->bdi
);
353 * Remove the inode from the writeback list it is on.
355 void inode_wb_list_del(struct inode
*inode
)
357 struct bdi_writeback
*wb
= inode_to_wb(inode
);
359 spin_lock(&wb
->list_lock
);
360 inode_wb_list_del_locked(inode
, wb
);
361 spin_unlock(&wb
->list_lock
);
365 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
366 * furthest end of its superblock's dirty-inode list.
368 * Before stamping the inode's ->dirtied_when, we check to see whether it is
369 * already the most-recently-dirtied inode on the b_dirty list. If that is
370 * the case then the inode must have been redirtied while it was being written
371 * out and we don't reset its dirtied_when.
373 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
375 if (!list_empty(&wb
->b_dirty
)) {
378 tail
= wb_inode(wb
->b_dirty
.next
);
379 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
380 inode
->dirtied_when
= jiffies
;
382 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty
);
386 * requeue inode for re-scanning after bdi->b_io list is exhausted.
388 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
390 inode_wb_list_move_locked(inode
, wb
, &wb
->b_more_io
);
393 static void inode_sync_complete(struct inode
*inode
)
395 inode
->i_state
&= ~I_SYNC
;
396 /* If inode is clean an unused, put it into LRU now... */
397 inode_add_lru(inode
);
398 /* Waiters must see I_SYNC cleared before being woken up */
400 wake_up_bit(&inode
->i_state
, __I_SYNC
);
403 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
405 bool ret
= time_after(inode
->dirtied_when
, t
);
408 * For inodes being constantly redirtied, dirtied_when can get stuck.
409 * It _appears_ to be in the future, but is actually in distant past.
410 * This test is necessary to prevent such wrapped-around relative times
411 * from permanently stopping the whole bdi writeback.
413 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
418 #define EXPIRE_DIRTY_ATIME 0x0001
421 * Move expired (dirtied before work->older_than_this) dirty inodes from
422 * @delaying_queue to @dispatch_queue.
424 static int move_expired_inodes(struct list_head
*delaying_queue
,
425 struct list_head
*dispatch_queue
,
427 struct wb_writeback_work
*work
)
429 unsigned long *older_than_this
= NULL
;
430 unsigned long expire_time
;
432 struct list_head
*pos
, *node
;
433 struct super_block
*sb
= NULL
;
438 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
439 older_than_this
= work
->older_than_this
;
440 else if (!work
->for_sync
) {
441 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
442 older_than_this
= &expire_time
;
444 while (!list_empty(delaying_queue
)) {
445 inode
= wb_inode(delaying_queue
->prev
);
446 if (older_than_this
&&
447 inode_dirtied_after(inode
, *older_than_this
))
449 list_move(&inode
->i_wb_list
, &tmp
);
451 if (flags
& EXPIRE_DIRTY_ATIME
)
452 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
453 if (sb_is_blkdev_sb(inode
->i_sb
))
455 if (sb
&& sb
!= inode
->i_sb
)
460 /* just one sb in list, splice to dispatch_queue and we're done */
462 list_splice(&tmp
, dispatch_queue
);
466 /* Move inodes from one superblock together */
467 while (!list_empty(&tmp
)) {
468 sb
= wb_inode(tmp
.prev
)->i_sb
;
469 list_for_each_prev_safe(pos
, node
, &tmp
) {
470 inode
= wb_inode(pos
);
471 if (inode
->i_sb
== sb
)
472 list_move(&inode
->i_wb_list
, dispatch_queue
);
480 * Queue all expired dirty inodes for io, eldest first.
482 * newly dirtied b_dirty b_io b_more_io
483 * =============> gf edc BA
485 * newly dirtied b_dirty b_io b_more_io
486 * =============> g fBAedc
488 * +--> dequeue for IO
490 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
494 assert_spin_locked(&wb
->list_lock
);
495 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
496 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
497 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
498 EXPIRE_DIRTY_ATIME
, work
);
500 wb_io_lists_populated(wb
);
501 trace_writeback_queue_io(wb
, work
, moved
);
504 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
508 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
509 trace_writeback_write_inode_start(inode
, wbc
);
510 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
511 trace_writeback_write_inode(inode
, wbc
);
518 * Wait for writeback on an inode to complete. Called with i_lock held.
519 * Caller must make sure inode cannot go away when we drop i_lock.
521 static void __inode_wait_for_writeback(struct inode
*inode
)
522 __releases(inode
->i_lock
)
523 __acquires(inode
->i_lock
)
525 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
526 wait_queue_head_t
*wqh
;
528 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
529 while (inode
->i_state
& I_SYNC
) {
530 spin_unlock(&inode
->i_lock
);
531 __wait_on_bit(wqh
, &wq
, bit_wait
,
532 TASK_UNINTERRUPTIBLE
);
533 spin_lock(&inode
->i_lock
);
538 * Wait for writeback on an inode to complete. Caller must have inode pinned.
540 void inode_wait_for_writeback(struct inode
*inode
)
542 spin_lock(&inode
->i_lock
);
543 __inode_wait_for_writeback(inode
);
544 spin_unlock(&inode
->i_lock
);
548 * Sleep until I_SYNC is cleared. This function must be called with i_lock
549 * held and drops it. It is aimed for callers not holding any inode reference
550 * so once i_lock is dropped, inode can go away.
552 static void inode_sleep_on_writeback(struct inode
*inode
)
553 __releases(inode
->i_lock
)
556 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
559 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
560 sleep
= inode
->i_state
& I_SYNC
;
561 spin_unlock(&inode
->i_lock
);
564 finish_wait(wqh
, &wait
);
568 * Find proper writeback list for the inode depending on its current state and
569 * possibly also change of its state while we were doing writeback. Here we
570 * handle things such as livelock prevention or fairness of writeback among
571 * inodes. This function can be called only by flusher thread - noone else
572 * processes all inodes in writeback lists and requeueing inodes behind flusher
573 * thread's back can have unexpected consequences.
575 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
576 struct writeback_control
*wbc
)
578 if (inode
->i_state
& I_FREEING
)
582 * Sync livelock prevention. Each inode is tagged and synced in one
583 * shot. If still dirty, it will be redirty_tail()'ed below. Update
584 * the dirty time to prevent enqueue and sync it again.
586 if ((inode
->i_state
& I_DIRTY
) &&
587 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
588 inode
->dirtied_when
= jiffies
;
590 if (wbc
->pages_skipped
) {
592 * writeback is not making progress due to locked
593 * buffers. Skip this inode for now.
595 redirty_tail(inode
, wb
);
599 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
601 * We didn't write back all the pages. nfs_writepages()
602 * sometimes bales out without doing anything.
604 if (wbc
->nr_to_write
<= 0) {
605 /* Slice used up. Queue for next turn. */
606 requeue_io(inode
, wb
);
609 * Writeback blocked by something other than
610 * congestion. Delay the inode for some time to
611 * avoid spinning on the CPU (100% iowait)
612 * retrying writeback of the dirty page/inode
613 * that cannot be performed immediately.
615 redirty_tail(inode
, wb
);
617 } else if (inode
->i_state
& I_DIRTY
) {
619 * Filesystems can dirty the inode during writeback operations,
620 * such as delayed allocation during submission or metadata
621 * updates after data IO completion.
623 redirty_tail(inode
, wb
);
624 } else if (inode
->i_state
& I_DIRTY_TIME
) {
625 inode
->dirtied_when
= jiffies
;
626 inode_wb_list_move_locked(inode
, wb
, &wb
->b_dirty_time
);
628 /* The inode is clean. Remove from writeback lists. */
629 inode_wb_list_del_locked(inode
, wb
);
634 * Write out an inode and its dirty pages. Do not update the writeback list
635 * linkage. That is left to the caller. The caller is also responsible for
636 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
639 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
641 struct address_space
*mapping
= inode
->i_mapping
;
642 long nr_to_write
= wbc
->nr_to_write
;
646 WARN_ON(!(inode
->i_state
& I_SYNC
));
648 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
650 ret
= do_writepages(mapping
, wbc
);
653 * Make sure to wait on the data before writing out the metadata.
654 * This is important for filesystems that modify metadata on data
655 * I/O completion. We don't do it for sync(2) writeback because it has a
656 * separate, external IO completion path and ->sync_fs for guaranteeing
657 * inode metadata is written back correctly.
659 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
660 int err
= filemap_fdatawait(mapping
);
666 * Some filesystems may redirty the inode during the writeback
667 * due to delalloc, clear dirty metadata flags right before
670 spin_lock(&inode
->i_lock
);
672 dirty
= inode
->i_state
& I_DIRTY
;
673 if (inode
->i_state
& I_DIRTY_TIME
) {
674 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
675 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
676 unlikely(time_after(jiffies
,
677 (inode
->dirtied_time_when
+
678 dirtytime_expire_interval
* HZ
)))) {
679 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
680 trace_writeback_lazytime(inode
);
683 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
684 inode
->i_state
&= ~dirty
;
687 * Paired with smp_mb() in __mark_inode_dirty(). This allows
688 * __mark_inode_dirty() to test i_state without grabbing i_lock -
689 * either they see the I_DIRTY bits cleared or we see the dirtied
692 * I_DIRTY_PAGES is always cleared together above even if @mapping
693 * still has dirty pages. The flag is reinstated after smp_mb() if
694 * necessary. This guarantees that either __mark_inode_dirty()
695 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
699 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
700 inode
->i_state
|= I_DIRTY_PAGES
;
702 spin_unlock(&inode
->i_lock
);
704 if (dirty
& I_DIRTY_TIME
)
705 mark_inode_dirty_sync(inode
);
706 /* Don't write the inode if only I_DIRTY_PAGES was set */
707 if (dirty
& ~I_DIRTY_PAGES
) {
708 int err
= write_inode(inode
, wbc
);
712 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
717 * Write out an inode's dirty pages. Either the caller has an active reference
718 * on the inode or the inode has I_WILL_FREE set.
720 * This function is designed to be called for writing back one inode which
721 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
722 * and does more profound writeback list handling in writeback_sb_inodes().
725 writeback_single_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
726 struct writeback_control
*wbc
)
730 spin_lock(&inode
->i_lock
);
731 if (!atomic_read(&inode
->i_count
))
732 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
734 WARN_ON(inode
->i_state
& I_WILL_FREE
);
736 if (inode
->i_state
& I_SYNC
) {
737 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
740 * It's a data-integrity sync. We must wait. Since callers hold
741 * inode reference or inode has I_WILL_FREE set, it cannot go
744 __inode_wait_for_writeback(inode
);
746 WARN_ON(inode
->i_state
& I_SYNC
);
748 * Skip inode if it is clean and we have no outstanding writeback in
749 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
750 * function since flusher thread may be doing for example sync in
751 * parallel and if we move the inode, it could get skipped. So here we
752 * make sure inode is on some writeback list and leave it there unless
753 * we have completely cleaned the inode.
755 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
756 (wbc
->sync_mode
!= WB_SYNC_ALL
||
757 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
759 inode
->i_state
|= I_SYNC
;
760 spin_unlock(&inode
->i_lock
);
762 ret
= __writeback_single_inode(inode
, wbc
);
764 spin_lock(&wb
->list_lock
);
765 spin_lock(&inode
->i_lock
);
767 * If inode is clean, remove it from writeback lists. Otherwise don't
768 * touch it. See comment above for explanation.
770 if (!(inode
->i_state
& I_DIRTY_ALL
))
771 inode_wb_list_del_locked(inode
, wb
);
772 spin_unlock(&wb
->list_lock
);
773 inode_sync_complete(inode
);
775 spin_unlock(&inode
->i_lock
);
779 static long writeback_chunk_size(struct bdi_writeback
*wb
,
780 struct wb_writeback_work
*work
)
785 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
786 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
787 * here avoids calling into writeback_inodes_wb() more than once.
789 * The intended call sequence for WB_SYNC_ALL writeback is:
792 * writeback_sb_inodes() <== called only once
793 * write_cache_pages() <== called once for each inode
794 * (quickly) tag currently dirty pages
795 * (maybe slowly) sync all tagged pages
797 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
800 pages
= min(wb
->avg_write_bandwidth
/ 2,
801 global_dirty_limit
/ DIRTY_SCOPE
);
802 pages
= min(pages
, work
->nr_pages
);
803 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
804 MIN_WRITEBACK_PAGES
);
811 * Write a portion of b_io inodes which belong to @sb.
813 * Return the number of pages and/or inodes written.
815 static long writeback_sb_inodes(struct super_block
*sb
,
816 struct bdi_writeback
*wb
,
817 struct wb_writeback_work
*work
)
819 struct writeback_control wbc
= {
820 .sync_mode
= work
->sync_mode
,
821 .tagged_writepages
= work
->tagged_writepages
,
822 .for_kupdate
= work
->for_kupdate
,
823 .for_background
= work
->for_background
,
824 .for_sync
= work
->for_sync
,
825 .range_cyclic
= work
->range_cyclic
,
827 .range_end
= LLONG_MAX
,
829 unsigned long start_time
= jiffies
;
831 long wrote
= 0; /* count both pages and inodes */
833 while (!list_empty(&wb
->b_io
)) {
834 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
836 if (inode
->i_sb
!= sb
) {
839 * We only want to write back data for this
840 * superblock, move all inodes not belonging
841 * to it back onto the dirty list.
843 redirty_tail(inode
, wb
);
848 * The inode belongs to a different superblock.
849 * Bounce back to the caller to unpin this and
850 * pin the next superblock.
856 * Don't bother with new inodes or inodes being freed, first
857 * kind does not need periodic writeout yet, and for the latter
858 * kind writeout is handled by the freer.
860 spin_lock(&inode
->i_lock
);
861 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
862 spin_unlock(&inode
->i_lock
);
863 redirty_tail(inode
, wb
);
866 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
868 * If this inode is locked for writeback and we are not
869 * doing writeback-for-data-integrity, move it to
870 * b_more_io so that writeback can proceed with the
871 * other inodes on s_io.
873 * We'll have another go at writing back this inode
874 * when we completed a full scan of b_io.
876 spin_unlock(&inode
->i_lock
);
877 requeue_io(inode
, wb
);
878 trace_writeback_sb_inodes_requeue(inode
);
881 spin_unlock(&wb
->list_lock
);
884 * We already requeued the inode if it had I_SYNC set and we
885 * are doing WB_SYNC_NONE writeback. So this catches only the
888 if (inode
->i_state
& I_SYNC
) {
889 /* Wait for I_SYNC. This function drops i_lock... */
890 inode_sleep_on_writeback(inode
);
891 /* Inode may be gone, start again */
892 spin_lock(&wb
->list_lock
);
895 inode
->i_state
|= I_SYNC
;
896 spin_unlock(&inode
->i_lock
);
898 write_chunk
= writeback_chunk_size(wb
, work
);
899 wbc
.nr_to_write
= write_chunk
;
900 wbc
.pages_skipped
= 0;
903 * We use I_SYNC to pin the inode in memory. While it is set
904 * evict_inode() will wait so the inode cannot be freed.
906 __writeback_single_inode(inode
, &wbc
);
908 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
909 wrote
+= write_chunk
- wbc
.nr_to_write
;
910 spin_lock(&wb
->list_lock
);
911 spin_lock(&inode
->i_lock
);
912 if (!(inode
->i_state
& I_DIRTY_ALL
))
914 requeue_inode(inode
, wb
, &wbc
);
915 inode_sync_complete(inode
);
916 spin_unlock(&inode
->i_lock
);
917 cond_resched_lock(&wb
->list_lock
);
919 * bail out to wb_writeback() often enough to check
920 * background threshold and other termination conditions.
923 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
925 if (work
->nr_pages
<= 0)
932 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
933 struct wb_writeback_work
*work
)
935 unsigned long start_time
= jiffies
;
938 while (!list_empty(&wb
->b_io
)) {
939 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
940 struct super_block
*sb
= inode
->i_sb
;
942 if (!trylock_super(sb
)) {
944 * trylock_super() may fail consistently due to
945 * s_umount being grabbed by someone else. Don't use
946 * requeue_io() to avoid busy retrying the inode/sb.
948 redirty_tail(inode
, wb
);
951 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
952 up_read(&sb
->s_umount
);
954 /* refer to the same tests at the end of writeback_sb_inodes */
956 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
958 if (work
->nr_pages
<= 0)
962 /* Leave any unwritten inodes on b_io */
966 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
967 enum wb_reason reason
)
969 struct wb_writeback_work work
= {
970 .nr_pages
= nr_pages
,
971 .sync_mode
= WB_SYNC_NONE
,
976 spin_lock(&wb
->list_lock
);
977 if (list_empty(&wb
->b_io
))
979 __writeback_inodes_wb(wb
, &work
);
980 spin_unlock(&wb
->list_lock
);
982 return nr_pages
- work
.nr_pages
;
985 static bool over_bground_thresh(struct bdi_writeback
*wb
)
987 unsigned long background_thresh
, dirty_thresh
;
989 global_dirty_limits(&background_thresh
, &dirty_thresh
);
991 if (global_page_state(NR_FILE_DIRTY
) +
992 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
)
995 if (wb_stat(wb
, WB_RECLAIMABLE
) > wb_dirty_limit(wb
, background_thresh
))
1002 * Called under wb->list_lock. If there are multiple wb per bdi,
1003 * only the flusher working on the first wb should do it.
1005 static void wb_update_bandwidth(struct bdi_writeback
*wb
,
1006 unsigned long start_time
)
1008 __wb_update_bandwidth(wb
, 0, 0, 0, 0, 0, start_time
);
1012 * Explicit flushing or periodic writeback of "old" data.
1014 * Define "old": the first time one of an inode's pages is dirtied, we mark the
1015 * dirtying-time in the inode's address_space. So this periodic writeback code
1016 * just walks the superblock inode list, writing back any inodes which are
1017 * older than a specific point in time.
1019 * Try to run once per dirty_writeback_interval. But if a writeback event
1020 * takes longer than a dirty_writeback_interval interval, then leave a
1023 * older_than_this takes precedence over nr_to_write. So we'll only write back
1024 * all dirty pages if they are all attached to "old" mappings.
1026 static long wb_writeback(struct bdi_writeback
*wb
,
1027 struct wb_writeback_work
*work
)
1029 unsigned long wb_start
= jiffies
;
1030 long nr_pages
= work
->nr_pages
;
1031 unsigned long oldest_jif
;
1032 struct inode
*inode
;
1035 oldest_jif
= jiffies
;
1036 work
->older_than_this
= &oldest_jif
;
1038 spin_lock(&wb
->list_lock
);
1041 * Stop writeback when nr_pages has been consumed
1043 if (work
->nr_pages
<= 0)
1047 * Background writeout and kupdate-style writeback may
1048 * run forever. Stop them if there is other work to do
1049 * so that e.g. sync can proceed. They'll be restarted
1050 * after the other works are all done.
1052 if ((work
->for_background
|| work
->for_kupdate
) &&
1053 !list_empty(&wb
->work_list
))
1057 * For background writeout, stop when we are below the
1058 * background dirty threshold
1060 if (work
->for_background
&& !over_bground_thresh(wb
))
1064 * Kupdate and background works are special and we want to
1065 * include all inodes that need writing. Livelock avoidance is
1066 * handled by these works yielding to any other work so we are
1069 if (work
->for_kupdate
) {
1070 oldest_jif
= jiffies
-
1071 msecs_to_jiffies(dirty_expire_interval
* 10);
1072 } else if (work
->for_background
)
1073 oldest_jif
= jiffies
;
1075 trace_writeback_start(wb
->bdi
, work
);
1076 if (list_empty(&wb
->b_io
))
1079 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
1081 progress
= __writeback_inodes_wb(wb
, work
);
1082 trace_writeback_written(wb
->bdi
, work
);
1084 wb_update_bandwidth(wb
, wb_start
);
1087 * Did we write something? Try for more
1089 * Dirty inodes are moved to b_io for writeback in batches.
1090 * The completion of the current batch does not necessarily
1091 * mean the overall work is done. So we keep looping as long
1092 * as made some progress on cleaning pages or inodes.
1097 * No more inodes for IO, bail
1099 if (list_empty(&wb
->b_more_io
))
1102 * Nothing written. Wait for some inode to
1103 * become available for writeback. Otherwise
1104 * we'll just busyloop.
1106 if (!list_empty(&wb
->b_more_io
)) {
1107 trace_writeback_wait(wb
->bdi
, work
);
1108 inode
= wb_inode(wb
->b_more_io
.prev
);
1109 spin_lock(&inode
->i_lock
);
1110 spin_unlock(&wb
->list_lock
);
1111 /* This function drops i_lock... */
1112 inode_sleep_on_writeback(inode
);
1113 spin_lock(&wb
->list_lock
);
1116 spin_unlock(&wb
->list_lock
);
1118 return nr_pages
- work
->nr_pages
;
1122 * Return the next wb_writeback_work struct that hasn't been processed yet.
1124 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
1126 struct wb_writeback_work
*work
= NULL
;
1128 spin_lock_bh(&wb
->work_lock
);
1129 if (!list_empty(&wb
->work_list
)) {
1130 work
= list_entry(wb
->work_list
.next
,
1131 struct wb_writeback_work
, list
);
1132 list_del_init(&work
->list
);
1134 spin_unlock_bh(&wb
->work_lock
);
1139 * Add in the number of potentially dirty inodes, because each inode
1140 * write can dirty pagecache in the underlying blockdev.
1142 static unsigned long get_nr_dirty_pages(void)
1144 return global_page_state(NR_FILE_DIRTY
) +
1145 global_page_state(NR_UNSTABLE_NFS
) +
1146 get_nr_dirty_inodes();
1149 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1151 if (over_bground_thresh(wb
)) {
1153 struct wb_writeback_work work
= {
1154 .nr_pages
= LONG_MAX
,
1155 .sync_mode
= WB_SYNC_NONE
,
1156 .for_background
= 1,
1158 .reason
= WB_REASON_BACKGROUND
,
1161 return wb_writeback(wb
, &work
);
1167 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1169 unsigned long expired
;
1173 * When set to zero, disable periodic writeback
1175 if (!dirty_writeback_interval
)
1178 expired
= wb
->last_old_flush
+
1179 msecs_to_jiffies(dirty_writeback_interval
* 10);
1180 if (time_before(jiffies
, expired
))
1183 wb
->last_old_flush
= jiffies
;
1184 nr_pages
= get_nr_dirty_pages();
1187 struct wb_writeback_work work
= {
1188 .nr_pages
= nr_pages
,
1189 .sync_mode
= WB_SYNC_NONE
,
1192 .reason
= WB_REASON_PERIODIC
,
1195 return wb_writeback(wb
, &work
);
1202 * Retrieve work items and do the writeback they describe
1204 static long wb_do_writeback(struct bdi_writeback
*wb
)
1206 struct wb_writeback_work
*work
;
1209 set_bit(WB_writeback_running
, &wb
->state
);
1210 while ((work
= get_next_work_item(wb
)) != NULL
) {
1211 struct wb_completion
*done
= work
->done
;
1212 bool need_wake_up
= false;
1214 trace_writeback_exec(wb
->bdi
, work
);
1216 wrote
+= wb_writeback(wb
, work
);
1218 if (work
->single_wait
) {
1219 WARN_ON_ONCE(work
->auto_free
);
1220 /* paired w/ rmb in wb_wait_for_single_work() */
1222 work
->single_done
= 1;
1223 need_wake_up
= true;
1224 } else if (work
->auto_free
) {
1228 if (done
&& atomic_dec_and_test(&done
->cnt
))
1229 need_wake_up
= true;
1232 wake_up_all(&wb
->bdi
->wb_waitq
);
1236 * Check for periodic writeback, kupdated() style
1238 wrote
+= wb_check_old_data_flush(wb
);
1239 wrote
+= wb_check_background_flush(wb
);
1240 clear_bit(WB_writeback_running
, &wb
->state
);
1246 * Handle writeback of dirty data for the device backed by this bdi. Also
1247 * reschedules periodically and does kupdated style flushing.
1249 void wb_workfn(struct work_struct
*work
)
1251 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1252 struct bdi_writeback
, dwork
);
1255 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1256 current
->flags
|= PF_SWAPWRITE
;
1258 if (likely(!current_is_workqueue_rescuer() ||
1259 !test_bit(WB_registered
, &wb
->state
))) {
1261 * The normal path. Keep writing back @wb until its
1262 * work_list is empty. Note that this path is also taken
1263 * if @wb is shutting down even when we're running off the
1264 * rescuer as work_list needs to be drained.
1267 pages_written
= wb_do_writeback(wb
);
1268 trace_writeback_pages_written(pages_written
);
1269 } while (!list_empty(&wb
->work_list
));
1272 * bdi_wq can't get enough workers and we're running off
1273 * the emergency worker. Don't hog it. Hopefully, 1024 is
1274 * enough for efficient IO.
1276 pages_written
= writeback_inodes_wb(wb
, 1024,
1277 WB_REASON_FORKER_THREAD
);
1278 trace_writeback_pages_written(pages_written
);
1281 if (!list_empty(&wb
->work_list
))
1282 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1283 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1284 wb_wakeup_delayed(wb
);
1286 current
->flags
&= ~PF_SWAPWRITE
;
1290 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1293 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1295 struct backing_dev_info
*bdi
;
1298 nr_pages
= get_nr_dirty_pages();
1301 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1302 struct bdi_writeback
*wb
;
1303 struct wb_iter iter
;
1305 if (!bdi_has_dirty_io(bdi
))
1308 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1309 wb_start_writeback(wb
, wb_split_bdi_pages(wb
, nr_pages
),
1316 * Wake up bdi's periodically to make sure dirtytime inodes gets
1317 * written back periodically. We deliberately do *not* check the
1318 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1319 * kernel to be constantly waking up once there are any dirtytime
1320 * inodes on the system. So instead we define a separate delayed work
1321 * function which gets called much more rarely. (By default, only
1322 * once every 12 hours.)
1324 * If there is any other write activity going on in the file system,
1325 * this function won't be necessary. But if the only thing that has
1326 * happened on the file system is a dirtytime inode caused by an atime
1327 * update, we need this infrastructure below to make sure that inode
1328 * eventually gets pushed out to disk.
1330 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1331 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1333 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1335 struct backing_dev_info
*bdi
;
1338 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1339 struct bdi_writeback
*wb
;
1340 struct wb_iter iter
;
1342 bdi_for_each_wb(wb
, bdi
, &iter
, 0)
1343 if (!list_empty(&bdi
->wb
.b_dirty_time
))
1344 wb_wakeup(&bdi
->wb
);
1347 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1350 static int __init
start_dirtytime_writeback(void)
1352 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1355 __initcall(start_dirtytime_writeback
);
1357 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1358 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1362 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1363 if (ret
== 0 && write
)
1364 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1368 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1370 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1371 struct dentry
*dentry
;
1372 const char *name
= "?";
1374 dentry
= d_find_alias(inode
);
1376 spin_lock(&dentry
->d_lock
);
1377 name
= (const char *) dentry
->d_name
.name
;
1380 "%s(%d): dirtied inode %lu (%s) on %s\n",
1381 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1382 name
, inode
->i_sb
->s_id
);
1384 spin_unlock(&dentry
->d_lock
);
1391 * __mark_inode_dirty - internal function
1392 * @inode: inode to mark
1393 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1394 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1395 * mark_inode_dirty_sync.
1397 * Put the inode on the super block's dirty list.
1399 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1400 * dirty list only if it is hashed or if it refers to a blockdev.
1401 * If it was not hashed, it will never be added to the dirty list
1402 * even if it is later hashed, as it will have been marked dirty already.
1404 * In short, make sure you hash any inodes _before_ you start marking
1407 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1408 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1409 * the kernel-internal blockdev inode represents the dirtying time of the
1410 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1411 * page->mapping->host, so the page-dirtying time is recorded in the internal
1414 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
1415 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1417 struct super_block
*sb
= inode
->i_sb
;
1418 struct backing_dev_info
*bdi
= NULL
;
1421 trace_writeback_mark_inode_dirty(inode
, flags
);
1424 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1425 * dirty the inode itself
1427 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
1428 trace_writeback_dirty_inode_start(inode
, flags
);
1430 if (sb
->s_op
->dirty_inode
)
1431 sb
->s_op
->dirty_inode(inode
, flags
);
1433 trace_writeback_dirty_inode(inode
, flags
);
1435 if (flags
& I_DIRTY_INODE
)
1436 flags
&= ~I_DIRTY_TIME
;
1437 dirtytime
= flags
& I_DIRTY_TIME
;
1440 * Paired with smp_mb() in __writeback_single_inode() for the
1441 * following lockless i_state test. See there for details.
1445 if (((inode
->i_state
& flags
) == flags
) ||
1446 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
1449 if (unlikely(block_dump
))
1450 block_dump___mark_inode_dirty(inode
);
1452 spin_lock(&inode
->i_lock
);
1453 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
1454 goto out_unlock_inode
;
1455 if ((inode
->i_state
& flags
) != flags
) {
1456 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1458 inode_attach_wb(inode
, NULL
);
1460 if (flags
& I_DIRTY_INODE
)
1461 inode
->i_state
&= ~I_DIRTY_TIME
;
1462 inode
->i_state
|= flags
;
1465 * If the inode is being synced, just update its dirty state.
1466 * The unlocker will place the inode on the appropriate
1467 * superblock list, based upon its state.
1469 if (inode
->i_state
& I_SYNC
)
1470 goto out_unlock_inode
;
1473 * Only add valid (hashed) inodes to the superblock's
1474 * dirty list. Add blockdev inodes as well.
1476 if (!S_ISBLK(inode
->i_mode
)) {
1477 if (inode_unhashed(inode
))
1478 goto out_unlock_inode
;
1480 if (inode
->i_state
& I_FREEING
)
1481 goto out_unlock_inode
;
1484 * If the inode was already on b_dirty/b_io/b_more_io, don't
1485 * reposition it (that would break b_dirty time-ordering).
1488 struct list_head
*dirty_list
;
1489 bool wakeup_bdi
= false;
1490 bdi
= inode_to_bdi(inode
);
1492 spin_unlock(&inode
->i_lock
);
1493 spin_lock(&bdi
->wb
.list_lock
);
1495 WARN(bdi_cap_writeback_dirty(bdi
) &&
1496 !test_bit(WB_registered
, &bdi
->wb
.state
),
1497 "bdi-%s not registered\n", bdi
->name
);
1499 inode
->dirtied_when
= jiffies
;
1501 inode
->dirtied_time_when
= jiffies
;
1503 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
1504 dirty_list
= &bdi
->wb
.b_dirty
;
1506 dirty_list
= &bdi
->wb
.b_dirty_time
;
1508 wakeup_bdi
= inode_wb_list_move_locked(inode
, &bdi
->wb
,
1511 spin_unlock(&bdi
->wb
.list_lock
);
1512 trace_writeback_dirty_inode_enqueue(inode
);
1515 * If this is the first dirty inode for this bdi,
1516 * we have to wake-up the corresponding bdi thread
1517 * to make sure background write-back happens
1520 if (bdi_cap_writeback_dirty(bdi
) && wakeup_bdi
)
1521 wb_wakeup_delayed(&bdi
->wb
);
1526 spin_unlock(&inode
->i_lock
);
1529 EXPORT_SYMBOL(__mark_inode_dirty
);
1531 static void wait_sb_inodes(struct super_block
*sb
)
1533 struct inode
*inode
, *old_inode
= NULL
;
1536 * We need to be protected against the filesystem going from
1537 * r/o to r/w or vice versa.
1539 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1541 spin_lock(&inode_sb_list_lock
);
1544 * Data integrity sync. Must wait for all pages under writeback,
1545 * because there may have been pages dirtied before our sync
1546 * call, but which had writeout started before we write it out.
1547 * In which case, the inode may not be on the dirty list, but
1548 * we still have to wait for that writeout.
1550 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1551 struct address_space
*mapping
= inode
->i_mapping
;
1553 spin_lock(&inode
->i_lock
);
1554 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1555 (mapping
->nrpages
== 0)) {
1556 spin_unlock(&inode
->i_lock
);
1560 spin_unlock(&inode
->i_lock
);
1561 spin_unlock(&inode_sb_list_lock
);
1564 * We hold a reference to 'inode' so it couldn't have been
1565 * removed from s_inodes list while we dropped the
1566 * inode_sb_list_lock. We cannot iput the inode now as we can
1567 * be holding the last reference and we cannot iput it under
1568 * inode_sb_list_lock. So we keep the reference and iput it
1574 filemap_fdatawait(mapping
);
1578 spin_lock(&inode_sb_list_lock
);
1580 spin_unlock(&inode_sb_list_lock
);
1585 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1586 * @sb: the superblock
1587 * @nr: the number of pages to write
1588 * @reason: reason why some writeback work initiated
1590 * Start writeback on some inodes on this super_block. No guarantees are made
1591 * on how many (if any) will be written, and this function does not wait
1592 * for IO completion of submitted IO.
1594 void writeback_inodes_sb_nr(struct super_block
*sb
,
1596 enum wb_reason reason
)
1598 DEFINE_WB_COMPLETION_ONSTACK(done
);
1599 struct wb_writeback_work work
= {
1601 .sync_mode
= WB_SYNC_NONE
,
1602 .tagged_writepages
= 1,
1607 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1609 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1611 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1612 wb_queue_work(&bdi
->wb
, &work
);
1613 wb_wait_for_completion(bdi
, &done
);
1615 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1618 * writeback_inodes_sb - writeback dirty inodes from given super_block
1619 * @sb: the superblock
1620 * @reason: reason why some writeback work was initiated
1622 * Start writeback on some inodes on this super_block. No guarantees are made
1623 * on how many (if any) will be written, and this function does not wait
1624 * for IO completion of submitted IO.
1626 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1628 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1630 EXPORT_SYMBOL(writeback_inodes_sb
);
1633 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
1634 * @sb: the superblock
1635 * @nr: the number of pages to write
1636 * @reason: the reason of writeback
1638 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
1639 * Returns 1 if writeback was started, 0 if not.
1641 int try_to_writeback_inodes_sb_nr(struct super_block
*sb
,
1643 enum wb_reason reason
)
1645 if (writeback_in_progress(&sb
->s_bdi
->wb
))
1648 if (!down_read_trylock(&sb
->s_umount
))
1651 writeback_inodes_sb_nr(sb
, nr
, reason
);
1652 up_read(&sb
->s_umount
);
1655 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
1658 * try_to_writeback_inodes_sb - try to start writeback if none underway
1659 * @sb: the superblock
1660 * @reason: reason why some writeback work was initiated
1662 * Implement by try_to_writeback_inodes_sb_nr()
1663 * Returns 1 if writeback was started, 0 if not.
1665 int try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
1667 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
1669 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
1672 * sync_inodes_sb - sync sb inode pages
1673 * @sb: the superblock
1675 * This function writes and waits on any dirty inode belonging to this
1678 void sync_inodes_sb(struct super_block
*sb
)
1680 DEFINE_WB_COMPLETION_ONSTACK(done
);
1681 struct wb_writeback_work work
= {
1683 .sync_mode
= WB_SYNC_ALL
,
1684 .nr_pages
= LONG_MAX
,
1687 .reason
= WB_REASON_SYNC
,
1690 struct backing_dev_info
*bdi
= sb
->s_bdi
;
1692 /* Nothing to do? */
1693 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
1695 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1697 wb_queue_work(&bdi
->wb
, &work
);
1698 wb_wait_for_completion(bdi
, &done
);
1702 EXPORT_SYMBOL(sync_inodes_sb
);
1705 * write_inode_now - write an inode to disk
1706 * @inode: inode to write to disk
1707 * @sync: whether the write should be synchronous or not
1709 * This function commits an inode to disk immediately if it is dirty. This is
1710 * primarily needed by knfsd.
1712 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1714 int write_inode_now(struct inode
*inode
, int sync
)
1716 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1717 struct writeback_control wbc
= {
1718 .nr_to_write
= LONG_MAX
,
1719 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1721 .range_end
= LLONG_MAX
,
1724 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1725 wbc
.nr_to_write
= 0;
1728 return writeback_single_inode(inode
, wb
, &wbc
);
1730 EXPORT_SYMBOL(write_inode_now
);
1733 * sync_inode - write an inode and its pages to disk.
1734 * @inode: the inode to sync
1735 * @wbc: controls the writeback mode
1737 * sync_inode() will write an inode and its pages to disk. It will also
1738 * correctly update the inode on its superblock's dirty inode lists and will
1739 * update inode->i_state.
1741 * The caller must have a ref on the inode.
1743 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1745 return writeback_single_inode(inode
, &inode_to_bdi(inode
)->wb
, wbc
);
1747 EXPORT_SYMBOL(sync_inode
);
1750 * sync_inode_metadata - write an inode to disk
1751 * @inode: the inode to sync
1752 * @wait: wait for I/O to complete.
1754 * Write an inode to disk and adjust its dirty state after completion.
1756 * Note: only writes the actual inode, no associated data or other metadata.
1758 int sync_inode_metadata(struct inode
*inode
, int wait
)
1760 struct writeback_control wbc
= {
1761 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1762 .nr_to_write
= 0, /* metadata-only */
1765 return sync_inode(inode
, &wbc
);
1767 EXPORT_SYMBOL(sync_inode_metadata
);