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/module.h>
18 #include <linux/spinlock.h>
19 #include <linux/slab.h>
20 #include <linux/sched.h>
23 #include <linux/kthread.h>
24 #include <linux/freezer.h>
25 #include <linux/writeback.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/buffer_head.h>
31 #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info)
34 * We don't actually have pdflush, but this one is exported though /proc...
36 int nr_pdflush_threads
;
39 * Passed into wb_writeback(), essentially a subset of writeback_control
41 struct wb_writeback_args
{
43 struct super_block
*sb
;
44 enum writeback_sync_modes sync_mode
;
45 unsigned int for_kupdate
:1;
46 unsigned int range_cyclic
:1;
47 unsigned int for_background
:1;
51 * Work items for the bdi_writeback threads
54 struct list_head list
; /* pending work list */
55 struct rcu_head rcu_head
; /* for RCU free/clear of work */
57 unsigned long seen
; /* threads that have seen this work */
58 atomic_t pending
; /* number of threads still to do work */
60 struct wb_writeback_args args
; /* writeback arguments */
62 unsigned long state
; /* flag bits, see WS_* */
70 static inline void bdi_work_init(struct bdi_work
*work
,
71 struct wb_writeback_args
*args
)
73 INIT_RCU_HEAD(&work
->rcu_head
);
75 __set_bit(WS_INPROGRESS
, &work
->state
);
79 * writeback_in_progress - determine whether there is writeback in progress
80 * @bdi: the device's backing_dev_info structure.
82 * Determine whether there is writeback waiting to be handled against a
85 int writeback_in_progress(struct backing_dev_info
*bdi
)
87 return !list_empty(&bdi
->work_list
);
90 static void bdi_work_free(struct rcu_head
*head
)
92 struct bdi_work
*work
= container_of(head
, struct bdi_work
, rcu_head
);
94 clear_bit(WS_INPROGRESS
, &work
->state
);
95 smp_mb__after_clear_bit();
96 wake_up_bit(&work
->state
, WS_INPROGRESS
);
98 if (!test_bit(WS_ONSTACK
, &work
->state
))
102 static void wb_clear_pending(struct bdi_writeback
*wb
, struct bdi_work
*work
)
105 * The caller has retrieved the work arguments from this work,
106 * drop our reference. If this is the last ref, delete and free it
108 if (atomic_dec_and_test(&work
->pending
)) {
109 struct backing_dev_info
*bdi
= wb
->bdi
;
111 spin_lock(&bdi
->wb_lock
);
112 list_del_rcu(&work
->list
);
113 spin_unlock(&bdi
->wb_lock
);
115 call_rcu(&work
->rcu_head
, bdi_work_free
);
119 static void bdi_queue_work(struct backing_dev_info
*bdi
, struct bdi_work
*work
)
121 work
->seen
= bdi
->wb_mask
;
123 atomic_set(&work
->pending
, bdi
->wb_cnt
);
124 BUG_ON(!bdi
->wb_cnt
);
127 * list_add_tail_rcu() contains the necessary barriers to
128 * make sure the above stores are seen before the item is
129 * noticed on the list
131 spin_lock(&bdi
->wb_lock
);
132 list_add_tail_rcu(&work
->list
, &bdi
->work_list
);
133 spin_unlock(&bdi
->wb_lock
);
136 * If the default thread isn't there, make sure we add it. When
137 * it gets created and wakes up, we'll run this work.
139 if (unlikely(list_empty_careful(&bdi
->wb_list
)))
140 wake_up_process(default_backing_dev_info
.wb
.task
);
142 struct bdi_writeback
*wb
= &bdi
->wb
;
145 wake_up_process(wb
->task
);
150 * Used for on-stack allocated work items. The caller needs to wait until
151 * the wb threads have acked the work before it's safe to continue.
153 static void bdi_wait_on_work_done(struct bdi_work
*work
)
155 wait_on_bit(&work
->state
, WS_INPROGRESS
, bdi_sched_wait
,
156 TASK_UNINTERRUPTIBLE
);
159 static void bdi_alloc_queue_work(struct backing_dev_info
*bdi
,
160 struct wb_writeback_args
*args
)
162 struct bdi_work
*work
;
165 * This is WB_SYNC_NONE writeback, so if allocation fails just
166 * wakeup the thread for old dirty data writeback
168 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
170 bdi_work_init(work
, args
);
171 bdi_queue_work(bdi
, work
);
173 struct bdi_writeback
*wb
= &bdi
->wb
;
176 wake_up_process(wb
->task
);
181 * bdi_queue_work_onstack - start and wait for writeback
182 * @sb: write inodes from this super_block
185 * This function initiates writeback and waits for the operation to
186 * complete. Callers must hold the sb s_umount semaphore for
187 * reading, to avoid having the super disappear before we are done.
189 static void bdi_queue_work_onstack(struct wb_writeback_args
*args
)
191 struct bdi_work work
;
193 bdi_work_init(&work
, args
);
194 __set_bit(WS_ONSTACK
, &work
.state
);
196 bdi_queue_work(args
->sb
->s_bdi
, &work
);
197 bdi_wait_on_work_done(&work
);
201 * bdi_start_writeback - start writeback
202 * @bdi: the backing device to write from
203 * @sb: write inodes from this super_block
204 * @nr_pages: the number of pages to write
207 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
208 * started when this function returns, we make no guarentees on
209 * completion. Caller need not hold sb s_umount semaphore.
212 void bdi_start_writeback(struct backing_dev_info
*bdi
, struct super_block
*sb
,
215 struct wb_writeback_args args
= {
217 .sync_mode
= WB_SYNC_NONE
,
218 .nr_pages
= nr_pages
,
223 * We treat @nr_pages=0 as the special case to do background writeback,
224 * ie. to sync pages until the background dirty threshold is reached.
227 args
.nr_pages
= LONG_MAX
;
228 args
.for_background
= 1;
231 bdi_alloc_queue_work(bdi
, &args
);
235 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
236 * furthest end of its superblock's dirty-inode list.
238 * Before stamping the inode's ->dirtied_when, we check to see whether it is
239 * already the most-recently-dirtied inode on the b_dirty list. If that is
240 * the case then the inode must have been redirtied while it was being written
241 * out and we don't reset its dirtied_when.
243 static void redirty_tail(struct inode
*inode
)
245 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
247 if (!list_empty(&wb
->b_dirty
)) {
250 tail
= list_entry(wb
->b_dirty
.next
, struct inode
, i_list
);
251 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
252 inode
->dirtied_when
= jiffies
;
254 list_move(&inode
->i_list
, &wb
->b_dirty
);
258 * requeue inode for re-scanning after bdi->b_io list is exhausted.
260 static void requeue_io(struct inode
*inode
)
262 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
264 list_move(&inode
->i_list
, &wb
->b_more_io
);
267 static void inode_sync_complete(struct inode
*inode
)
270 * Prevent speculative execution through spin_unlock(&inode_lock);
273 wake_up_bit(&inode
->i_state
, __I_SYNC
);
276 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
278 bool ret
= time_after(inode
->dirtied_when
, t
);
281 * For inodes being constantly redirtied, dirtied_when can get stuck.
282 * It _appears_ to be in the future, but is actually in distant past.
283 * This test is necessary to prevent such wrapped-around relative times
284 * from permanently stopping the whole bdi writeback.
286 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
292 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
294 static void move_expired_inodes(struct list_head
*delaying_queue
,
295 struct list_head
*dispatch_queue
,
296 unsigned long *older_than_this
)
299 struct list_head
*pos
, *node
;
300 struct super_block
*sb
= NULL
;
304 while (!list_empty(delaying_queue
)) {
305 inode
= list_entry(delaying_queue
->prev
, struct inode
, i_list
);
306 if (older_than_this
&&
307 inode_dirtied_after(inode
, *older_than_this
))
309 if (sb
&& sb
!= inode
->i_sb
)
312 list_move(&inode
->i_list
, &tmp
);
315 /* just one sb in list, splice to dispatch_queue and we're done */
317 list_splice(&tmp
, dispatch_queue
);
321 /* Move inodes from one superblock together */
322 while (!list_empty(&tmp
)) {
323 inode
= list_entry(tmp
.prev
, struct inode
, i_list
);
325 list_for_each_prev_safe(pos
, node
, &tmp
) {
326 inode
= list_entry(pos
, struct inode
, i_list
);
327 if (inode
->i_sb
== sb
)
328 list_move(&inode
->i_list
, dispatch_queue
);
334 * Queue all expired dirty inodes for io, eldest first.
336 static void queue_io(struct bdi_writeback
*wb
, unsigned long *older_than_this
)
338 list_splice_init(&wb
->b_more_io
, wb
->b_io
.prev
);
339 move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, older_than_this
);
342 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
344 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
))
345 return inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
350 * Wait for writeback on an inode to complete.
352 static void inode_wait_for_writeback(struct inode
*inode
)
354 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
355 wait_queue_head_t
*wqh
;
357 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
358 while (inode
->i_state
& I_SYNC
) {
359 spin_unlock(&inode_lock
);
360 __wait_on_bit(wqh
, &wq
, inode_wait
, TASK_UNINTERRUPTIBLE
);
361 spin_lock(&inode_lock
);
366 * Write out an inode's dirty pages. Called under inode_lock. Either the
367 * caller has ref on the inode (either via __iget or via syscall against an fd)
368 * or the inode has I_WILL_FREE set (via generic_forget_inode)
370 * If `wait' is set, wait on the writeout.
372 * The whole writeout design is quite complex and fragile. We want to avoid
373 * starvation of particular inodes when others are being redirtied, prevent
376 * Called under inode_lock.
379 writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
381 struct address_space
*mapping
= inode
->i_mapping
;
385 if (!atomic_read(&inode
->i_count
))
386 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
388 WARN_ON(inode
->i_state
& I_WILL_FREE
);
390 if (inode
->i_state
& I_SYNC
) {
392 * If this inode is locked for writeback and we are not doing
393 * writeback-for-data-integrity, move it to b_more_io so that
394 * writeback can proceed with the other inodes on s_io.
396 * We'll have another go at writing back this inode when we
397 * completed a full scan of b_io.
399 if (wbc
->sync_mode
!= WB_SYNC_ALL
) {
405 * It's a data-integrity sync. We must wait.
407 inode_wait_for_writeback(inode
);
410 BUG_ON(inode
->i_state
& I_SYNC
);
412 /* Set I_SYNC, reset I_DIRTY_PAGES */
413 inode
->i_state
|= I_SYNC
;
414 inode
->i_state
&= ~I_DIRTY_PAGES
;
415 spin_unlock(&inode_lock
);
417 ret
= do_writepages(mapping
, wbc
);
420 * Make sure to wait on the data before writing out the metadata.
421 * This is important for filesystems that modify metadata on data
424 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
425 int err
= filemap_fdatawait(mapping
);
431 * Some filesystems may redirty the inode during the writeback
432 * due to delalloc, clear dirty metadata flags right before
435 spin_lock(&inode_lock
);
436 dirty
= inode
->i_state
& I_DIRTY
;
437 inode
->i_state
&= ~(I_DIRTY_SYNC
| I_DIRTY_DATASYNC
);
438 spin_unlock(&inode_lock
);
439 /* Don't write the inode if only I_DIRTY_PAGES was set */
440 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
441 int err
= write_inode(inode
, wbc
);
446 spin_lock(&inode_lock
);
447 inode
->i_state
&= ~I_SYNC
;
448 if (!(inode
->i_state
& (I_FREEING
| I_CLEAR
))) {
449 if ((inode
->i_state
& I_DIRTY_PAGES
) && wbc
->for_kupdate
) {
451 * More pages get dirtied by a fast dirtier.
454 } else if (inode
->i_state
& I_DIRTY
) {
456 * At least XFS will redirty the inode during the
457 * writeback (delalloc) and on io completion (isize).
460 } else if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
462 * We didn't write back all the pages. nfs_writepages()
463 * sometimes bales out without doing anything. Redirty
464 * the inode; Move it from b_io onto b_more_io/b_dirty.
467 * akpm: if the caller was the kupdate function we put
468 * this inode at the head of b_dirty so it gets first
469 * consideration. Otherwise, move it to the tail, for
470 * the reasons described there. I'm not really sure
471 * how much sense this makes. Presumably I had a good
472 * reasons for doing it this way, and I'd rather not
473 * muck with it at present.
475 if (wbc
->for_kupdate
) {
477 * For the kupdate function we move the inode
478 * to b_more_io so it will get more writeout as
479 * soon as the queue becomes uncongested.
481 inode
->i_state
|= I_DIRTY_PAGES
;
483 if (wbc
->nr_to_write
<= 0) {
485 * slice used up: queue for next turn
490 * somehow blocked: retry later
496 * Otherwise fully redirty the inode so that
497 * other inodes on this superblock will get some
498 * writeout. Otherwise heavy writing to one
499 * file would indefinitely suspend writeout of
500 * all the other files.
502 inode
->i_state
|= I_DIRTY_PAGES
;
505 } else if (atomic_read(&inode
->i_count
)) {
507 * The inode is clean, inuse
509 list_move(&inode
->i_list
, &inode_in_use
);
512 * The inode is clean, unused
514 list_move(&inode
->i_list
, &inode_unused
);
517 inode_sync_complete(inode
);
521 static void unpin_sb_for_writeback(struct super_block
*sb
)
523 up_read(&sb
->s_umount
);
534 * For WB_SYNC_NONE writeback, the caller does not have the sb pinned
535 * before calling writeback. So make sure that we do pin it, so it doesn't
536 * go away while we are writing inodes from it.
538 static enum sb_pin_state
pin_sb_for_writeback(struct writeback_control
*wbc
,
539 struct super_block
*sb
)
542 * Caller must already hold the ref for this
544 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
545 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
546 return SB_NOT_PINNED
;
550 if (down_read_trylock(&sb
->s_umount
)) {
552 spin_unlock(&sb_lock
);
556 * umounted, drop rwsem again and fall through to failure
558 up_read(&sb
->s_umount
);
561 spin_unlock(&sb_lock
);
562 return SB_PIN_FAILED
;
566 * Write a portion of b_io inodes which belong to @sb.
567 * If @wbc->sb != NULL, then find and write all such
568 * inodes. Otherwise write only ones which go sequentially
570 * Return 1, if the caller writeback routine should be
571 * interrupted. Otherwise return 0.
573 static int writeback_sb_inodes(struct super_block
*sb
,
574 struct bdi_writeback
*wb
,
575 struct writeback_control
*wbc
)
577 while (!list_empty(&wb
->b_io
)) {
579 struct inode
*inode
= list_entry(wb
->b_io
.prev
,
580 struct inode
, i_list
);
581 if (wbc
->sb
&& sb
!= inode
->i_sb
) {
582 /* super block given and doesn't
583 match, skip this inode */
587 if (sb
!= inode
->i_sb
)
588 /* finish with this superblock */
590 if (inode
->i_state
& (I_NEW
| I_WILL_FREE
)) {
595 * Was this inode dirtied after sync_sb_inodes was called?
596 * This keeps sync from extra jobs and livelock.
598 if (inode_dirtied_after(inode
, wbc
->wb_start
))
601 BUG_ON(inode
->i_state
& (I_FREEING
| I_CLEAR
));
603 pages_skipped
= wbc
->pages_skipped
;
604 writeback_single_inode(inode
, wbc
);
605 if (wbc
->pages_skipped
!= pages_skipped
) {
607 * writeback is not making progress due to locked
608 * buffers. Skip this inode for now.
612 spin_unlock(&inode_lock
);
615 spin_lock(&inode_lock
);
616 if (wbc
->nr_to_write
<= 0) {
620 if (!list_empty(&wb
->b_more_io
))
627 static void writeback_inodes_wb(struct bdi_writeback
*wb
,
628 struct writeback_control
*wbc
)
632 wbc
->wb_start
= jiffies
; /* livelock avoidance */
633 spin_lock(&inode_lock
);
634 if (!wbc
->for_kupdate
|| list_empty(&wb
->b_io
))
635 queue_io(wb
, wbc
->older_than_this
);
637 while (!list_empty(&wb
->b_io
)) {
638 struct inode
*inode
= list_entry(wb
->b_io
.prev
,
639 struct inode
, i_list
);
640 struct super_block
*sb
= inode
->i_sb
;
641 enum sb_pin_state state
;
643 if (wbc
->sb
&& sb
!= wbc
->sb
) {
644 /* super block given and doesn't
645 match, skip this inode */
649 state
= pin_sb_for_writeback(wbc
, sb
);
651 if (state
== SB_PIN_FAILED
) {
655 ret
= writeback_sb_inodes(sb
, wb
, wbc
);
657 if (state
== SB_PINNED
)
658 unpin_sb_for_writeback(sb
);
662 spin_unlock(&inode_lock
);
663 /* Leave any unwritten inodes on b_io */
666 void writeback_inodes_wbc(struct writeback_control
*wbc
)
668 struct backing_dev_info
*bdi
= wbc
->bdi
;
670 writeback_inodes_wb(&bdi
->wb
, wbc
);
674 * The maximum number of pages to writeout in a single bdi flush/kupdate
675 * operation. We do this so we don't hold I_SYNC against an inode for
676 * enormous amounts of time, which would block a userspace task which has
677 * been forced to throttle against that inode. Also, the code reevaluates
678 * the dirty each time it has written this many pages.
680 #define MAX_WRITEBACK_PAGES 1024
682 static inline bool over_bground_thresh(void)
684 unsigned long background_thresh
, dirty_thresh
;
686 get_dirty_limits(&background_thresh
, &dirty_thresh
, NULL
, NULL
);
688 return (global_page_state(NR_FILE_DIRTY
) +
689 global_page_state(NR_UNSTABLE_NFS
) >= background_thresh
);
693 * Explicit flushing or periodic writeback of "old" data.
695 * Define "old": the first time one of an inode's pages is dirtied, we mark the
696 * dirtying-time in the inode's address_space. So this periodic writeback code
697 * just walks the superblock inode list, writing back any inodes which are
698 * older than a specific point in time.
700 * Try to run once per dirty_writeback_interval. But if a writeback event
701 * takes longer than a dirty_writeback_interval interval, then leave a
704 * older_than_this takes precedence over nr_to_write. So we'll only write back
705 * all dirty pages if they are all attached to "old" mappings.
707 static long wb_writeback(struct bdi_writeback
*wb
,
708 struct wb_writeback_args
*args
)
710 struct writeback_control wbc
= {
713 .sync_mode
= args
->sync_mode
,
714 .older_than_this
= NULL
,
715 .for_kupdate
= args
->for_kupdate
,
716 .for_background
= args
->for_background
,
717 .range_cyclic
= args
->range_cyclic
,
719 unsigned long oldest_jif
;
723 if (wbc
.for_kupdate
) {
724 wbc
.older_than_this
= &oldest_jif
;
725 oldest_jif
= jiffies
-
726 msecs_to_jiffies(dirty_expire_interval
* 10);
728 if (!wbc
.range_cyclic
) {
730 wbc
.range_end
= LLONG_MAX
;
735 * Stop writeback when nr_pages has been consumed
737 if (args
->nr_pages
<= 0)
741 * For background writeout, stop when we are below the
742 * background dirty threshold
744 if (args
->for_background
&& !over_bground_thresh())
748 wbc
.nr_to_write
= MAX_WRITEBACK_PAGES
;
749 wbc
.pages_skipped
= 0;
750 writeback_inodes_wb(wb
, &wbc
);
751 args
->nr_pages
-= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
752 wrote
+= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
755 * If we consumed everything, see if we have more
757 if (wbc
.nr_to_write
<= 0)
760 * Didn't write everything and we don't have more IO, bail
765 * Did we write something? Try for more
767 if (wbc
.nr_to_write
< MAX_WRITEBACK_PAGES
)
770 * Nothing written. Wait for some inode to
771 * become available for writeback. Otherwise
772 * we'll just busyloop.
774 spin_lock(&inode_lock
);
775 if (!list_empty(&wb
->b_more_io
)) {
776 inode
= list_entry(wb
->b_more_io
.prev
,
777 struct inode
, i_list
);
778 inode_wait_for_writeback(inode
);
780 spin_unlock(&inode_lock
);
787 * Return the next bdi_work struct that hasn't been processed by this
788 * wb thread yet. ->seen is initially set for each thread that exists
789 * for this device, when a thread first notices a piece of work it
790 * clears its bit. Depending on writeback type, the thread will notify
791 * completion on either receiving the work (WB_SYNC_NONE) or after
792 * it is done (WB_SYNC_ALL).
794 static struct bdi_work
*get_next_work_item(struct backing_dev_info
*bdi
,
795 struct bdi_writeback
*wb
)
797 struct bdi_work
*work
, *ret
= NULL
;
801 list_for_each_entry_rcu(work
, &bdi
->work_list
, list
) {
802 if (!test_bit(wb
->nr
, &work
->seen
))
804 clear_bit(wb
->nr
, &work
->seen
);
814 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
816 unsigned long expired
;
820 * When set to zero, disable periodic writeback
822 if (!dirty_writeback_interval
)
825 expired
= wb
->last_old_flush
+
826 msecs_to_jiffies(dirty_writeback_interval
* 10);
827 if (time_before(jiffies
, expired
))
830 wb
->last_old_flush
= jiffies
;
831 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
832 global_page_state(NR_UNSTABLE_NFS
) +
833 (inodes_stat
.nr_inodes
- inodes_stat
.nr_unused
);
836 struct wb_writeback_args args
= {
837 .nr_pages
= nr_pages
,
838 .sync_mode
= WB_SYNC_NONE
,
843 return wb_writeback(wb
, &args
);
850 * Retrieve work items and do the writeback they describe
852 long wb_do_writeback(struct bdi_writeback
*wb
, int force_wait
)
854 struct backing_dev_info
*bdi
= wb
->bdi
;
855 struct bdi_work
*work
;
858 while ((work
= get_next_work_item(bdi
, wb
)) != NULL
) {
859 struct wb_writeback_args args
= work
->args
;
862 * Override sync mode, in case we must wait for completion
865 work
->args
.sync_mode
= args
.sync_mode
= WB_SYNC_ALL
;
868 * If this isn't a data integrity operation, just notify
869 * that we have seen this work and we are now starting it.
871 if (!test_bit(WS_ONSTACK
, &work
->state
))
872 wb_clear_pending(wb
, work
);
874 wrote
+= wb_writeback(wb
, &args
);
877 * This is a data integrity writeback, so only do the
878 * notification when we have completed the work.
880 if (test_bit(WS_ONSTACK
, &work
->state
))
881 wb_clear_pending(wb
, work
);
885 * Check for periodic writeback, kupdated() style
887 wrote
+= wb_check_old_data_flush(wb
);
893 * Handle writeback of dirty data for the device backed by this bdi. Also
894 * wakes up periodically and does kupdated style flushing.
896 int bdi_writeback_task(struct bdi_writeback
*wb
)
898 unsigned long last_active
= jiffies
;
899 unsigned long wait_jiffies
= -1UL;
902 while (!kthread_should_stop()) {
903 pages_written
= wb_do_writeback(wb
, 0);
906 last_active
= jiffies
;
907 else if (wait_jiffies
!= -1UL) {
908 unsigned long max_idle
;
911 * Longest period of inactivity that we tolerate. If we
912 * see dirty data again later, the task will get
913 * recreated automatically.
915 max_idle
= max(5UL * 60 * HZ
, wait_jiffies
);
916 if (time_after(jiffies
, max_idle
+ last_active
))
920 if (dirty_writeback_interval
) {
921 wait_jiffies
= msecs_to_jiffies(dirty_writeback_interval
* 10);
922 schedule_timeout_interruptible(wait_jiffies
);
924 set_current_state(TASK_INTERRUPTIBLE
);
925 if (list_empty_careful(&wb
->bdi
->work_list
) &&
926 !kthread_should_stop())
928 __set_current_state(TASK_RUNNING
);
938 * Schedule writeback for all backing devices. This does WB_SYNC_NONE
939 * writeback, for integrity writeback see bdi_queue_work_onstack().
941 static void bdi_writeback_all(struct super_block
*sb
, long nr_pages
)
943 struct wb_writeback_args args
= {
945 .nr_pages
= nr_pages
,
946 .sync_mode
= WB_SYNC_NONE
,
948 struct backing_dev_info
*bdi
;
952 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
953 if (!bdi_has_dirty_io(bdi
))
956 bdi_alloc_queue_work(bdi
, &args
);
963 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
966 void wakeup_flusher_threads(long nr_pages
)
969 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
970 global_page_state(NR_UNSTABLE_NFS
);
971 bdi_writeback_all(NULL
, nr_pages
);
974 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
976 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
977 struct dentry
*dentry
;
978 const char *name
= "?";
980 dentry
= d_find_alias(inode
);
982 spin_lock(&dentry
->d_lock
);
983 name
= (const char *) dentry
->d_name
.name
;
986 "%s(%d): dirtied inode %lu (%s) on %s\n",
987 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
988 name
, inode
->i_sb
->s_id
);
990 spin_unlock(&dentry
->d_lock
);
997 * __mark_inode_dirty - internal function
998 * @inode: inode to mark
999 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1000 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1001 * mark_inode_dirty_sync.
1003 * Put the inode on the super block's dirty list.
1005 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1006 * dirty list only if it is hashed or if it refers to a blockdev.
1007 * If it was not hashed, it will never be added to the dirty list
1008 * even if it is later hashed, as it will have been marked dirty already.
1010 * In short, make sure you hash any inodes _before_ you start marking
1013 * This function *must* be atomic for the I_DIRTY_PAGES case -
1014 * set_page_dirty() is called under spinlock in several places.
1016 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1017 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1018 * the kernel-internal blockdev inode represents the dirtying time of the
1019 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1020 * page->mapping->host, so the page-dirtying time is recorded in the internal
1023 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1025 struct super_block
*sb
= inode
->i_sb
;
1028 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1029 * dirty the inode itself
1031 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
1032 if (sb
->s_op
->dirty_inode
)
1033 sb
->s_op
->dirty_inode(inode
);
1037 * make sure that changes are seen by all cpus before we test i_state
1042 /* avoid the locking if we can */
1043 if ((inode
->i_state
& flags
) == flags
)
1046 if (unlikely(block_dump
))
1047 block_dump___mark_inode_dirty(inode
);
1049 spin_lock(&inode_lock
);
1050 if ((inode
->i_state
& flags
) != flags
) {
1051 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1053 inode
->i_state
|= flags
;
1056 * If the inode is being synced, just update its dirty state.
1057 * The unlocker will place the inode on the appropriate
1058 * superblock list, based upon its state.
1060 if (inode
->i_state
& I_SYNC
)
1064 * Only add valid (hashed) inodes to the superblock's
1065 * dirty list. Add blockdev inodes as well.
1067 if (!S_ISBLK(inode
->i_mode
)) {
1068 if (hlist_unhashed(&inode
->i_hash
))
1071 if (inode
->i_state
& (I_FREEING
|I_CLEAR
))
1075 * If the inode was already on b_dirty/b_io/b_more_io, don't
1076 * reposition it (that would break b_dirty time-ordering).
1079 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1080 struct backing_dev_info
*bdi
= wb
->bdi
;
1082 if (bdi_cap_writeback_dirty(bdi
) &&
1083 !test_bit(BDI_registered
, &bdi
->state
)) {
1085 printk(KERN_ERR
"bdi-%s not registered\n",
1089 inode
->dirtied_when
= jiffies
;
1090 list_move(&inode
->i_list
, &wb
->b_dirty
);
1094 spin_unlock(&inode_lock
);
1096 EXPORT_SYMBOL(__mark_inode_dirty
);
1099 * Write out a superblock's list of dirty inodes. A wait will be performed
1100 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1102 * If older_than_this is non-NULL, then only write out inodes which
1103 * had their first dirtying at a time earlier than *older_than_this.
1105 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1106 * This function assumes that the blockdev superblock's inodes are backed by
1107 * a variety of queues, so all inodes are searched. For other superblocks,
1108 * assume that all inodes are backed by the same queue.
1110 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1111 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1112 * on the writer throttling path, and we get decent balancing between many
1113 * throttled threads: we don't want them all piling up on inode_sync_wait.
1115 static void wait_sb_inodes(struct super_block
*sb
)
1117 struct inode
*inode
, *old_inode
= NULL
;
1120 * We need to be protected against the filesystem going from
1121 * r/o to r/w or vice versa.
1123 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1125 spin_lock(&inode_lock
);
1128 * Data integrity sync. Must wait for all pages under writeback,
1129 * because there may have been pages dirtied before our sync
1130 * call, but which had writeout started before we write it out.
1131 * In which case, the inode may not be on the dirty list, but
1132 * we still have to wait for that writeout.
1134 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1135 struct address_space
*mapping
;
1137 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
|I_NEW
))
1139 mapping
= inode
->i_mapping
;
1140 if (mapping
->nrpages
== 0)
1143 spin_unlock(&inode_lock
);
1145 * We hold a reference to 'inode' so it couldn't have
1146 * been removed from s_inodes list while we dropped the
1147 * inode_lock. We cannot iput the inode now as we can
1148 * be holding the last reference and we cannot iput it
1149 * under inode_lock. So we keep the reference and iput
1155 filemap_fdatawait(mapping
);
1159 spin_lock(&inode_lock
);
1161 spin_unlock(&inode_lock
);
1166 * writeback_inodes_sb - writeback dirty inodes from given super_block
1167 * @sb: the superblock
1169 * Start writeback on some inodes on this super_block. No guarantees are made
1170 * on how many (if any) will be written, and this function does not wait
1171 * for IO completion of submitted IO. The number of pages submitted is
1174 void writeback_inodes_sb(struct super_block
*sb
)
1176 unsigned long nr_dirty
= global_page_state(NR_FILE_DIRTY
);
1177 unsigned long nr_unstable
= global_page_state(NR_UNSTABLE_NFS
);
1178 struct wb_writeback_args args
= {
1180 .sync_mode
= WB_SYNC_NONE
,
1183 args
.nr_pages
= nr_dirty
+ nr_unstable
+
1184 (inodes_stat
.nr_inodes
- inodes_stat
.nr_unused
);
1186 bdi_queue_work_onstack(&args
);
1188 EXPORT_SYMBOL(writeback_inodes_sb
);
1191 * writeback_inodes_sb_if_idle - start writeback if none underway
1192 * @sb: the superblock
1194 * Invoke writeback_inodes_sb if no writeback is currently underway.
1195 * Returns 1 if writeback was started, 0 if not.
1197 int writeback_inodes_sb_if_idle(struct super_block
*sb
)
1199 if (!writeback_in_progress(sb
->s_bdi
)) {
1200 writeback_inodes_sb(sb
);
1205 EXPORT_SYMBOL(writeback_inodes_sb_if_idle
);
1208 * sync_inodes_sb - sync sb inode pages
1209 * @sb: the superblock
1211 * This function writes and waits on any dirty inode belonging to this
1212 * super_block. The number of pages synced is returned.
1214 void sync_inodes_sb(struct super_block
*sb
)
1216 struct wb_writeback_args args
= {
1218 .sync_mode
= WB_SYNC_ALL
,
1219 .nr_pages
= LONG_MAX
,
1223 bdi_queue_work_onstack(&args
);
1226 EXPORT_SYMBOL(sync_inodes_sb
);
1229 * write_inode_now - write an inode to disk
1230 * @inode: inode to write to disk
1231 * @sync: whether the write should be synchronous or not
1233 * This function commits an inode to disk immediately if it is dirty. This is
1234 * primarily needed by knfsd.
1236 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1238 int write_inode_now(struct inode
*inode
, int sync
)
1241 struct writeback_control wbc
= {
1242 .nr_to_write
= LONG_MAX
,
1243 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1245 .range_end
= LLONG_MAX
,
1248 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1249 wbc
.nr_to_write
= 0;
1252 spin_lock(&inode_lock
);
1253 ret
= writeback_single_inode(inode
, &wbc
);
1254 spin_unlock(&inode_lock
);
1256 inode_sync_wait(inode
);
1259 EXPORT_SYMBOL(write_inode_now
);
1262 * sync_inode - write an inode and its pages to disk.
1263 * @inode: the inode to sync
1264 * @wbc: controls the writeback mode
1266 * sync_inode() will write an inode and its pages to disk. It will also
1267 * correctly update the inode on its superblock's dirty inode lists and will
1268 * update inode->i_state.
1270 * The caller must have a ref on the inode.
1272 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1276 spin_lock(&inode_lock
);
1277 ret
= writeback_single_inode(inode
, wbc
);
1278 spin_unlock(&inode_lock
);
1281 EXPORT_SYMBOL(sync_inode
);