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;
48 unsigned int sb_pinned
:1;
52 * Work items for the bdi_writeback threads
55 struct list_head list
; /* pending work list */
56 struct rcu_head rcu_head
; /* for RCU free/clear of work */
58 unsigned long seen
; /* threads that have seen this work */
59 atomic_t pending
; /* number of threads still to do work */
61 struct wb_writeback_args args
; /* writeback arguments */
63 unsigned long state
; /* flag bits, see WS_* */
71 #define WS_USED (1 << WS_USED_B)
72 #define WS_ONSTACK (1 << WS_ONSTACK_B)
74 static inline bool bdi_work_on_stack(struct bdi_work
*work
)
76 return test_bit(WS_ONSTACK_B
, &work
->state
);
79 static inline void bdi_work_init(struct bdi_work
*work
,
80 struct wb_writeback_args
*args
)
82 INIT_RCU_HEAD(&work
->rcu_head
);
84 work
->state
= WS_USED
;
88 * writeback_in_progress - determine whether there is writeback in progress
89 * @bdi: the device's backing_dev_info structure.
91 * Determine whether there is writeback waiting to be handled against a
94 int writeback_in_progress(struct backing_dev_info
*bdi
)
96 return !list_empty(&bdi
->work_list
);
99 static void bdi_work_clear(struct bdi_work
*work
)
101 clear_bit(WS_USED_B
, &work
->state
);
102 smp_mb__after_clear_bit();
104 * work can have disappeared at this point. bit waitq functions
105 * should be able to tolerate this, provided bdi_sched_wait does
106 * not dereference it's pointer argument.
108 wake_up_bit(&work
->state
, WS_USED_B
);
111 static void bdi_work_free(struct rcu_head
*head
)
113 struct bdi_work
*work
= container_of(head
, struct bdi_work
, rcu_head
);
115 if (!bdi_work_on_stack(work
))
118 bdi_work_clear(work
);
121 static void wb_work_complete(struct bdi_work
*work
)
123 const enum writeback_sync_modes sync_mode
= work
->args
.sync_mode
;
124 int onstack
= bdi_work_on_stack(work
);
127 * For allocated work, we can clear the done/seen bit right here.
128 * For on-stack work, we need to postpone both the clear and free
129 * to after the RCU grace period, since the stack could be invalidated
130 * as soon as bdi_work_clear() has done the wakeup.
133 bdi_work_clear(work
);
134 if (sync_mode
== WB_SYNC_NONE
|| onstack
)
135 call_rcu(&work
->rcu_head
, bdi_work_free
);
138 static void wb_clear_pending(struct bdi_writeback
*wb
, struct bdi_work
*work
)
141 * The caller has retrieved the work arguments from this work,
142 * drop our reference. If this is the last ref, delete and free it
144 if (atomic_dec_and_test(&work
->pending
)) {
145 struct backing_dev_info
*bdi
= wb
->bdi
;
147 spin_lock(&bdi
->wb_lock
);
148 list_del_rcu(&work
->list
);
149 spin_unlock(&bdi
->wb_lock
);
151 wb_work_complete(work
);
155 static void bdi_queue_work(struct backing_dev_info
*bdi
, struct bdi_work
*work
)
157 work
->seen
= bdi
->wb_mask
;
159 atomic_set(&work
->pending
, bdi
->wb_cnt
);
160 BUG_ON(!bdi
->wb_cnt
);
163 * list_add_tail_rcu() contains the necessary barriers to
164 * make sure the above stores are seen before the item is
165 * noticed on the list
167 spin_lock(&bdi
->wb_lock
);
168 list_add_tail_rcu(&work
->list
, &bdi
->work_list
);
169 spin_unlock(&bdi
->wb_lock
);
172 * If the default thread isn't there, make sure we add it. When
173 * it gets created and wakes up, we'll run this work.
175 if (unlikely(list_empty_careful(&bdi
->wb_list
)))
176 wake_up_process(default_backing_dev_info
.wb
.task
);
178 struct bdi_writeback
*wb
= &bdi
->wb
;
181 wake_up_process(wb
->task
);
186 * Used for on-stack allocated work items. The caller needs to wait until
187 * the wb threads have acked the work before it's safe to continue.
189 static void bdi_wait_on_work_clear(struct bdi_work
*work
)
191 wait_on_bit(&work
->state
, WS_USED_B
, bdi_sched_wait
,
192 TASK_UNINTERRUPTIBLE
);
195 static void bdi_alloc_queue_work(struct backing_dev_info
*bdi
,
196 struct wb_writeback_args
*args
)
198 struct bdi_work
*work
;
201 * This is WB_SYNC_NONE writeback, so if allocation fails just
202 * wakeup the thread for old dirty data writeback
204 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
206 bdi_work_init(work
, args
);
207 bdi_queue_work(bdi
, work
);
209 struct bdi_writeback
*wb
= &bdi
->wb
;
212 wake_up_process(wb
->task
);
217 * bdi_sync_writeback - start and wait for writeback
218 * @bdi: the backing device to write from
219 * @sb: write inodes from this super_block
222 * This does WB_SYNC_ALL data integrity writeback and waits for the
223 * IO to complete. Callers must hold the sb s_umount semaphore for
224 * reading, to avoid having the super disappear before we are done.
226 static void bdi_sync_writeback(struct backing_dev_info
*bdi
,
227 struct super_block
*sb
)
229 struct wb_writeback_args args
= {
231 .sync_mode
= WB_SYNC_ALL
,
232 .nr_pages
= LONG_MAX
,
235 * Setting sb_pinned is not necessary for WB_SYNC_ALL, but
236 * lets make it explicitly clear.
240 struct bdi_work work
;
242 bdi_work_init(&work
, &args
);
243 work
.state
|= WS_ONSTACK
;
245 bdi_queue_work(bdi
, &work
);
246 bdi_wait_on_work_clear(&work
);
250 * bdi_start_writeback - start writeback
251 * @bdi: the backing device to write from
252 * @sb: write inodes from this super_block
253 * @nr_pages: the number of pages to write
254 * @sb_locked: caller already holds sb umount sem.
257 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
258 * started when this function returns, we make no guarentees on
259 * completion. Caller specifies whether sb umount sem is held already or not.
262 void bdi_start_writeback(struct backing_dev_info
*bdi
, struct super_block
*sb
,
263 long nr_pages
, int sb_locked
)
265 struct wb_writeback_args args
= {
267 .sync_mode
= WB_SYNC_NONE
,
268 .nr_pages
= nr_pages
,
270 .sb_pinned
= sb_locked
,
274 * We treat @nr_pages=0 as the special case to do background writeback,
275 * ie. to sync pages until the background dirty threshold is reached.
278 args
.nr_pages
= LONG_MAX
;
279 args
.for_background
= 1;
282 bdi_alloc_queue_work(bdi
, &args
);
286 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
287 * furthest end of its superblock's dirty-inode list.
289 * Before stamping the inode's ->dirtied_when, we check to see whether it is
290 * already the most-recently-dirtied inode on the b_dirty list. If that is
291 * the case then the inode must have been redirtied while it was being written
292 * out and we don't reset its dirtied_when.
294 static void redirty_tail(struct inode
*inode
)
296 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
298 if (!list_empty(&wb
->b_dirty
)) {
301 tail
= list_entry(wb
->b_dirty
.next
, struct inode
, i_list
);
302 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
303 inode
->dirtied_when
= jiffies
;
305 list_move(&inode
->i_list
, &wb
->b_dirty
);
309 * requeue inode for re-scanning after bdi->b_io list is exhausted.
311 static void requeue_io(struct inode
*inode
)
313 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
315 list_move(&inode
->i_list
, &wb
->b_more_io
);
318 static void inode_sync_complete(struct inode
*inode
)
321 * Prevent speculative execution through spin_unlock(&inode_lock);
324 wake_up_bit(&inode
->i_state
, __I_SYNC
);
327 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
329 bool ret
= time_after(inode
->dirtied_when
, t
);
332 * For inodes being constantly redirtied, dirtied_when can get stuck.
333 * It _appears_ to be in the future, but is actually in distant past.
334 * This test is necessary to prevent such wrapped-around relative times
335 * from permanently stopping the whole bdi writeback.
337 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
343 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
345 static void move_expired_inodes(struct list_head
*delaying_queue
,
346 struct list_head
*dispatch_queue
,
347 unsigned long *older_than_this
)
350 struct list_head
*pos
, *node
;
351 struct super_block
*sb
= NULL
;
355 while (!list_empty(delaying_queue
)) {
356 inode
= list_entry(delaying_queue
->prev
, struct inode
, i_list
);
357 if (older_than_this
&&
358 inode_dirtied_after(inode
, *older_than_this
))
360 if (sb
&& sb
!= inode
->i_sb
)
363 list_move(&inode
->i_list
, &tmp
);
366 /* just one sb in list, splice to dispatch_queue and we're done */
368 list_splice(&tmp
, dispatch_queue
);
372 /* Move inodes from one superblock together */
373 while (!list_empty(&tmp
)) {
374 inode
= list_entry(tmp
.prev
, struct inode
, i_list
);
376 list_for_each_prev_safe(pos
, node
, &tmp
) {
377 inode
= list_entry(pos
, struct inode
, i_list
);
378 if (inode
->i_sb
== sb
)
379 list_move(&inode
->i_list
, dispatch_queue
);
385 * Queue all expired dirty inodes for io, eldest first.
387 static void queue_io(struct bdi_writeback
*wb
, unsigned long *older_than_this
)
389 list_splice_init(&wb
->b_more_io
, wb
->b_io
.prev
);
390 move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, older_than_this
);
393 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
395 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
))
396 return inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
401 * Wait for writeback on an inode to complete.
403 static void inode_wait_for_writeback(struct inode
*inode
)
405 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
406 wait_queue_head_t
*wqh
;
408 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
410 spin_unlock(&inode_lock
);
411 __wait_on_bit(wqh
, &wq
, inode_wait
, TASK_UNINTERRUPTIBLE
);
412 spin_lock(&inode_lock
);
413 } while (inode
->i_state
& I_SYNC
);
417 * Write out an inode's dirty pages. Called under inode_lock. Either the
418 * caller has ref on the inode (either via __iget or via syscall against an fd)
419 * or the inode has I_WILL_FREE set (via generic_forget_inode)
421 * If `wait' is set, wait on the writeout.
423 * The whole writeout design is quite complex and fragile. We want to avoid
424 * starvation of particular inodes when others are being redirtied, prevent
427 * Called under inode_lock.
430 writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
432 struct address_space
*mapping
= inode
->i_mapping
;
436 if (!atomic_read(&inode
->i_count
))
437 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
439 WARN_ON(inode
->i_state
& I_WILL_FREE
);
441 if (inode
->i_state
& I_SYNC
) {
443 * If this inode is locked for writeback and we are not doing
444 * writeback-for-data-integrity, move it to b_more_io so that
445 * writeback can proceed with the other inodes on s_io.
447 * We'll have another go at writing back this inode when we
448 * completed a full scan of b_io.
450 if (wbc
->sync_mode
!= WB_SYNC_ALL
) {
456 * It's a data-integrity sync. We must wait.
458 inode_wait_for_writeback(inode
);
461 BUG_ON(inode
->i_state
& I_SYNC
);
463 /* Set I_SYNC, reset I_DIRTY_PAGES */
464 inode
->i_state
|= I_SYNC
;
465 inode
->i_state
&= ~I_DIRTY_PAGES
;
466 spin_unlock(&inode_lock
);
468 ret
= do_writepages(mapping
, wbc
);
471 * Make sure to wait on the data before writing out the metadata.
472 * This is important for filesystems that modify metadata on data
475 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
476 int err
= filemap_fdatawait(mapping
);
482 * Some filesystems may redirty the inode during the writeback
483 * due to delalloc, clear dirty metadata flags right before
486 spin_lock(&inode_lock
);
487 dirty
= inode
->i_state
& I_DIRTY
;
488 inode
->i_state
&= ~(I_DIRTY_SYNC
| I_DIRTY_DATASYNC
);
489 spin_unlock(&inode_lock
);
490 /* Don't write the inode if only I_DIRTY_PAGES was set */
491 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
492 int err
= write_inode(inode
, wbc
);
497 spin_lock(&inode_lock
);
498 inode
->i_state
&= ~I_SYNC
;
499 if (!(inode
->i_state
& (I_FREEING
| I_CLEAR
))) {
500 if ((inode
->i_state
& I_DIRTY_PAGES
) && wbc
->for_kupdate
) {
502 * More pages get dirtied by a fast dirtier.
505 } else if (inode
->i_state
& I_DIRTY
) {
507 * At least XFS will redirty the inode during the
508 * writeback (delalloc) and on io completion (isize).
511 } else if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
513 * We didn't write back all the pages. nfs_writepages()
514 * sometimes bales out without doing anything. Redirty
515 * the inode; Move it from b_io onto b_more_io/b_dirty.
518 * akpm: if the caller was the kupdate function we put
519 * this inode at the head of b_dirty so it gets first
520 * consideration. Otherwise, move it to the tail, for
521 * the reasons described there. I'm not really sure
522 * how much sense this makes. Presumably I had a good
523 * reasons for doing it this way, and I'd rather not
524 * muck with it at present.
526 if (wbc
->for_kupdate
) {
528 * For the kupdate function we move the inode
529 * to b_more_io so it will get more writeout as
530 * soon as the queue becomes uncongested.
532 inode
->i_state
|= I_DIRTY_PAGES
;
534 if (wbc
->nr_to_write
<= 0) {
536 * slice used up: queue for next turn
541 * somehow blocked: retry later
547 * Otherwise fully redirty the inode so that
548 * other inodes on this superblock will get some
549 * writeout. Otherwise heavy writing to one
550 * file would indefinitely suspend writeout of
551 * all the other files.
553 inode
->i_state
|= I_DIRTY_PAGES
;
556 } else if (atomic_read(&inode
->i_count
)) {
558 * The inode is clean, inuse
560 list_move(&inode
->i_list
, &inode_in_use
);
563 * The inode is clean, unused
565 list_move(&inode
->i_list
, &inode_unused
);
568 inode_sync_complete(inode
);
572 static void unpin_sb_for_writeback(struct super_block
*sb
)
574 up_read(&sb
->s_umount
);
585 * For WB_SYNC_NONE writeback, the caller does not have the sb pinned
586 * before calling writeback. So make sure that we do pin it, so it doesn't
587 * go away while we are writing inodes from it.
589 static enum sb_pin_state
pin_sb_for_writeback(struct writeback_control
*wbc
,
590 struct super_block
*sb
)
593 * Caller must already hold the ref for this
595 if (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->sb_pinned
) {
596 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
597 return SB_NOT_PINNED
;
601 if (down_read_trylock(&sb
->s_umount
)) {
603 spin_unlock(&sb_lock
);
607 * umounted, drop rwsem again and fall through to failure
609 up_read(&sb
->s_umount
);
612 spin_unlock(&sb_lock
);
613 return SB_PIN_FAILED
;
617 * Write a portion of b_io inodes which belong to @sb.
618 * If @wbc->sb != NULL, then find and write all such
619 * inodes. Otherwise write only ones which go sequentially
621 * Return 1, if the caller writeback routine should be
622 * interrupted. Otherwise return 0.
624 static int writeback_sb_inodes(struct super_block
*sb
,
625 struct bdi_writeback
*wb
,
626 struct writeback_control
*wbc
)
628 while (!list_empty(&wb
->b_io
)) {
630 struct inode
*inode
= list_entry(wb
->b_io
.prev
,
631 struct inode
, i_list
);
632 if (wbc
->sb
&& sb
!= inode
->i_sb
) {
633 /* super block given and doesn't
634 match, skip this inode */
638 if (sb
!= inode
->i_sb
)
639 /* finish with this superblock */
641 if (inode
->i_state
& (I_NEW
| I_WILL_FREE
)) {
646 * Was this inode dirtied after sync_sb_inodes was called?
647 * This keeps sync from extra jobs and livelock.
649 if (inode_dirtied_after(inode
, wbc
->wb_start
))
652 BUG_ON(inode
->i_state
& (I_FREEING
| I_CLEAR
));
654 pages_skipped
= wbc
->pages_skipped
;
655 writeback_single_inode(inode
, wbc
);
656 if (wbc
->pages_skipped
!= pages_skipped
) {
658 * writeback is not making progress due to locked
659 * buffers. Skip this inode for now.
663 spin_unlock(&inode_lock
);
666 spin_lock(&inode_lock
);
667 if (wbc
->nr_to_write
<= 0) {
671 if (!list_empty(&wb
->b_more_io
))
678 static void writeback_inodes_wb(struct bdi_writeback
*wb
,
679 struct writeback_control
*wbc
)
683 wbc
->wb_start
= jiffies
; /* livelock avoidance */
684 spin_lock(&inode_lock
);
685 if (!wbc
->for_kupdate
|| list_empty(&wb
->b_io
))
686 queue_io(wb
, wbc
->older_than_this
);
688 while (!list_empty(&wb
->b_io
)) {
689 struct inode
*inode
= list_entry(wb
->b_io
.prev
,
690 struct inode
, i_list
);
691 struct super_block
*sb
= inode
->i_sb
;
692 enum sb_pin_state state
;
694 if (wbc
->sb
&& sb
!= wbc
->sb
) {
695 /* super block given and doesn't
696 match, skip this inode */
700 state
= pin_sb_for_writeback(wbc
, sb
);
702 if (state
== SB_PIN_FAILED
) {
706 ret
= writeback_sb_inodes(sb
, wb
, wbc
);
708 if (state
== SB_PINNED
)
709 unpin_sb_for_writeback(sb
);
713 spin_unlock(&inode_lock
);
714 /* Leave any unwritten inodes on b_io */
717 void writeback_inodes_wbc(struct writeback_control
*wbc
)
719 struct backing_dev_info
*bdi
= wbc
->bdi
;
721 writeback_inodes_wb(&bdi
->wb
, wbc
);
725 * The maximum number of pages to writeout in a single bdi flush/kupdate
726 * operation. We do this so we don't hold I_SYNC against an inode for
727 * enormous amounts of time, which would block a userspace task which has
728 * been forced to throttle against that inode. Also, the code reevaluates
729 * the dirty each time it has written this many pages.
731 #define MAX_WRITEBACK_PAGES 1024
733 static inline bool over_bground_thresh(void)
735 unsigned long background_thresh
, dirty_thresh
;
737 get_dirty_limits(&background_thresh
, &dirty_thresh
, NULL
, NULL
);
739 return (global_page_state(NR_FILE_DIRTY
) +
740 global_page_state(NR_UNSTABLE_NFS
) >= background_thresh
);
744 * Explicit flushing or periodic writeback of "old" data.
746 * Define "old": the first time one of an inode's pages is dirtied, we mark the
747 * dirtying-time in the inode's address_space. So this periodic writeback code
748 * just walks the superblock inode list, writing back any inodes which are
749 * older than a specific point in time.
751 * Try to run once per dirty_writeback_interval. But if a writeback event
752 * takes longer than a dirty_writeback_interval interval, then leave a
755 * older_than_this takes precedence over nr_to_write. So we'll only write back
756 * all dirty pages if they are all attached to "old" mappings.
758 static long wb_writeback(struct bdi_writeback
*wb
,
759 struct wb_writeback_args
*args
)
761 struct writeback_control wbc
= {
764 .sync_mode
= args
->sync_mode
,
765 .older_than_this
= NULL
,
766 .for_kupdate
= args
->for_kupdate
,
767 .for_background
= args
->for_background
,
768 .range_cyclic
= args
->range_cyclic
,
769 .sb_pinned
= args
->sb_pinned
,
771 unsigned long oldest_jif
;
775 if (wbc
.for_kupdate
) {
776 wbc
.older_than_this
= &oldest_jif
;
777 oldest_jif
= jiffies
-
778 msecs_to_jiffies(dirty_expire_interval
* 10);
780 if (!wbc
.range_cyclic
) {
782 wbc
.range_end
= LLONG_MAX
;
787 * Stop writeback when nr_pages has been consumed
789 if (args
->nr_pages
<= 0)
793 * For background writeout, stop when we are below the
794 * background dirty threshold
796 if (args
->for_background
&& !over_bground_thresh())
800 wbc
.nr_to_write
= MAX_WRITEBACK_PAGES
;
801 wbc
.pages_skipped
= 0;
802 writeback_inodes_wb(wb
, &wbc
);
803 args
->nr_pages
-= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
804 wrote
+= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
807 * If we consumed everything, see if we have more
809 if (wbc
.nr_to_write
<= 0)
812 * Didn't write everything and we don't have more IO, bail
817 * Did we write something? Try for more
819 if (wbc
.nr_to_write
< MAX_WRITEBACK_PAGES
)
822 * Nothing written. Wait for some inode to
823 * become available for writeback. Otherwise
824 * we'll just busyloop.
826 spin_lock(&inode_lock
);
827 if (!list_empty(&wb
->b_more_io
)) {
828 inode
= list_entry(wb
->b_more_io
.prev
,
829 struct inode
, i_list
);
830 inode_wait_for_writeback(inode
);
832 spin_unlock(&inode_lock
);
839 * Return the next bdi_work struct that hasn't been processed by this
840 * wb thread yet. ->seen is initially set for each thread that exists
841 * for this device, when a thread first notices a piece of work it
842 * clears its bit. Depending on writeback type, the thread will notify
843 * completion on either receiving the work (WB_SYNC_NONE) or after
844 * it is done (WB_SYNC_ALL).
846 static struct bdi_work
*get_next_work_item(struct backing_dev_info
*bdi
,
847 struct bdi_writeback
*wb
)
849 struct bdi_work
*work
, *ret
= NULL
;
853 list_for_each_entry_rcu(work
, &bdi
->work_list
, list
) {
854 if (!test_bit(wb
->nr
, &work
->seen
))
856 clear_bit(wb
->nr
, &work
->seen
);
866 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
868 unsigned long expired
;
872 * When set to zero, disable periodic writeback
874 if (!dirty_writeback_interval
)
877 expired
= wb
->last_old_flush
+
878 msecs_to_jiffies(dirty_writeback_interval
* 10);
879 if (time_before(jiffies
, expired
))
882 wb
->last_old_flush
= jiffies
;
883 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
884 global_page_state(NR_UNSTABLE_NFS
) +
885 (inodes_stat
.nr_inodes
- inodes_stat
.nr_unused
);
888 struct wb_writeback_args args
= {
889 .nr_pages
= nr_pages
,
890 .sync_mode
= WB_SYNC_NONE
,
895 return wb_writeback(wb
, &args
);
902 * Retrieve work items and do the writeback they describe
904 long wb_do_writeback(struct bdi_writeback
*wb
, int force_wait
)
906 struct backing_dev_info
*bdi
= wb
->bdi
;
907 struct bdi_work
*work
;
910 while ((work
= get_next_work_item(bdi
, wb
)) != NULL
) {
911 struct wb_writeback_args args
= work
->args
;
914 * Override sync mode, in case we must wait for completion
917 work
->args
.sync_mode
= args
.sync_mode
= WB_SYNC_ALL
;
920 * If this isn't a data integrity operation, just notify
921 * that we have seen this work and we are now starting it.
923 if (args
.sync_mode
== WB_SYNC_NONE
)
924 wb_clear_pending(wb
, work
);
926 wrote
+= wb_writeback(wb
, &args
);
929 * This is a data integrity writeback, so only do the
930 * notification when we have completed the work.
932 if (args
.sync_mode
== WB_SYNC_ALL
)
933 wb_clear_pending(wb
, work
);
937 * Check for periodic writeback, kupdated() style
939 wrote
+= wb_check_old_data_flush(wb
);
945 * Handle writeback of dirty data for the device backed by this bdi. Also
946 * wakes up periodically and does kupdated style flushing.
948 int bdi_writeback_task(struct bdi_writeback
*wb
)
950 unsigned long last_active
= jiffies
;
951 unsigned long wait_jiffies
= -1UL;
954 while (!kthread_should_stop()) {
955 pages_written
= wb_do_writeback(wb
, 0);
958 last_active
= jiffies
;
959 else if (wait_jiffies
!= -1UL) {
960 unsigned long max_idle
;
963 * Longest period of inactivity that we tolerate. If we
964 * see dirty data again later, the task will get
965 * recreated automatically.
967 max_idle
= max(5UL * 60 * HZ
, wait_jiffies
);
968 if (time_after(jiffies
, max_idle
+ last_active
))
972 if (dirty_writeback_interval
) {
973 wait_jiffies
= msecs_to_jiffies(dirty_writeback_interval
* 10);
974 schedule_timeout_interruptible(wait_jiffies
);
976 set_current_state(TASK_INTERRUPTIBLE
);
977 if (list_empty_careful(&wb
->bdi
->work_list
) &&
978 !kthread_should_stop())
980 __set_current_state(TASK_RUNNING
);
990 * Schedule writeback for all backing devices. This does WB_SYNC_NONE
991 * writeback, for integrity writeback see bdi_sync_writeback().
993 static void bdi_writeback_all(struct super_block
*sb
, long nr_pages
)
995 struct wb_writeback_args args
= {
997 .nr_pages
= nr_pages
,
998 .sync_mode
= WB_SYNC_NONE
,
1000 struct backing_dev_info
*bdi
;
1004 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1005 if (!bdi_has_dirty_io(bdi
))
1008 bdi_alloc_queue_work(bdi
, &args
);
1015 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1018 void wakeup_flusher_threads(long nr_pages
)
1021 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
1022 global_page_state(NR_UNSTABLE_NFS
);
1023 bdi_writeback_all(NULL
, nr_pages
);
1026 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1028 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1029 struct dentry
*dentry
;
1030 const char *name
= "?";
1032 dentry
= d_find_alias(inode
);
1034 spin_lock(&dentry
->d_lock
);
1035 name
= (const char *) dentry
->d_name
.name
;
1038 "%s(%d): dirtied inode %lu (%s) on %s\n",
1039 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1040 name
, inode
->i_sb
->s_id
);
1042 spin_unlock(&dentry
->d_lock
);
1049 * __mark_inode_dirty - internal function
1050 * @inode: inode to mark
1051 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1052 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1053 * mark_inode_dirty_sync.
1055 * Put the inode on the super block's dirty list.
1057 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1058 * dirty list only if it is hashed or if it refers to a blockdev.
1059 * If it was not hashed, it will never be added to the dirty list
1060 * even if it is later hashed, as it will have been marked dirty already.
1062 * In short, make sure you hash any inodes _before_ you start marking
1065 * This function *must* be atomic for the I_DIRTY_PAGES case -
1066 * set_page_dirty() is called under spinlock in several places.
1068 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1069 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1070 * the kernel-internal blockdev inode represents the dirtying time of the
1071 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1072 * page->mapping->host, so the page-dirtying time is recorded in the internal
1075 void __mark_inode_dirty(struct inode
*inode
, int flags
)
1077 struct super_block
*sb
= inode
->i_sb
;
1080 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1081 * dirty the inode itself
1083 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
1084 if (sb
->s_op
->dirty_inode
)
1085 sb
->s_op
->dirty_inode(inode
);
1089 * make sure that changes are seen by all cpus before we test i_state
1094 /* avoid the locking if we can */
1095 if ((inode
->i_state
& flags
) == flags
)
1098 if (unlikely(block_dump
))
1099 block_dump___mark_inode_dirty(inode
);
1101 spin_lock(&inode_lock
);
1102 if ((inode
->i_state
& flags
) != flags
) {
1103 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1105 inode
->i_state
|= flags
;
1108 * If the inode is being synced, just update its dirty state.
1109 * The unlocker will place the inode on the appropriate
1110 * superblock list, based upon its state.
1112 if (inode
->i_state
& I_SYNC
)
1116 * Only add valid (hashed) inodes to the superblock's
1117 * dirty list. Add blockdev inodes as well.
1119 if (!S_ISBLK(inode
->i_mode
)) {
1120 if (hlist_unhashed(&inode
->i_hash
))
1123 if (inode
->i_state
& (I_FREEING
|I_CLEAR
))
1127 * If the inode was already on b_dirty/b_io/b_more_io, don't
1128 * reposition it (that would break b_dirty time-ordering).
1131 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1132 struct backing_dev_info
*bdi
= wb
->bdi
;
1134 if (bdi_cap_writeback_dirty(bdi
) &&
1135 !test_bit(BDI_registered
, &bdi
->state
)) {
1137 printk(KERN_ERR
"bdi-%s not registered\n",
1141 inode
->dirtied_when
= jiffies
;
1142 list_move(&inode
->i_list
, &wb
->b_dirty
);
1146 spin_unlock(&inode_lock
);
1148 EXPORT_SYMBOL(__mark_inode_dirty
);
1151 * Write out a superblock's list of dirty inodes. A wait will be performed
1152 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1154 * If older_than_this is non-NULL, then only write out inodes which
1155 * had their first dirtying at a time earlier than *older_than_this.
1157 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1158 * This function assumes that the blockdev superblock's inodes are backed by
1159 * a variety of queues, so all inodes are searched. For other superblocks,
1160 * assume that all inodes are backed by the same queue.
1162 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1163 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1164 * on the writer throttling path, and we get decent balancing between many
1165 * throttled threads: we don't want them all piling up on inode_sync_wait.
1167 static void wait_sb_inodes(struct super_block
*sb
)
1169 struct inode
*inode
, *old_inode
= NULL
;
1172 * We need to be protected against the filesystem going from
1173 * r/o to r/w or vice versa.
1175 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1177 spin_lock(&inode_lock
);
1180 * Data integrity sync. Must wait for all pages under writeback,
1181 * because there may have been pages dirtied before our sync
1182 * call, but which had writeout started before we write it out.
1183 * In which case, the inode may not be on the dirty list, but
1184 * we still have to wait for that writeout.
1186 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1187 struct address_space
*mapping
;
1189 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
|I_NEW
))
1191 mapping
= inode
->i_mapping
;
1192 if (mapping
->nrpages
== 0)
1195 spin_unlock(&inode_lock
);
1197 * We hold a reference to 'inode' so it couldn't have
1198 * been removed from s_inodes list while we dropped the
1199 * inode_lock. We cannot iput the inode now as we can
1200 * be holding the last reference and we cannot iput it
1201 * under inode_lock. So we keep the reference and iput
1207 filemap_fdatawait(mapping
);
1211 spin_lock(&inode_lock
);
1213 spin_unlock(&inode_lock
);
1217 static void __writeback_inodes_sb(struct super_block
*sb
, int sb_locked
)
1219 unsigned long nr_dirty
= global_page_state(NR_FILE_DIRTY
);
1220 unsigned long nr_unstable
= global_page_state(NR_UNSTABLE_NFS
);
1223 nr_to_write
= nr_dirty
+ nr_unstable
+
1224 (inodes_stat
.nr_inodes
- inodes_stat
.nr_unused
);
1226 bdi_start_writeback(sb
->s_bdi
, sb
, nr_to_write
, sb_locked
);
1230 * writeback_inodes_sb - writeback dirty inodes from given super_block
1231 * @sb: the superblock
1233 * Start writeback on some inodes on this super_block. No guarantees are made
1234 * on how many (if any) will be written, and this function does not wait
1235 * for IO completion of submitted IO. The number of pages submitted is
1238 void writeback_inodes_sb(struct super_block
*sb
)
1240 __writeback_inodes_sb(sb
, 0);
1242 EXPORT_SYMBOL(writeback_inodes_sb
);
1245 * writeback_inodes_sb_locked - writeback dirty inodes from given super_block
1246 * @sb: the superblock
1248 * Like writeback_inodes_sb(), except the caller already holds the
1251 void writeback_inodes_sb_locked(struct super_block
*sb
)
1253 __writeback_inodes_sb(sb
, 1);
1257 * writeback_inodes_sb_if_idle - start writeback if none underway
1258 * @sb: the superblock
1260 * Invoke writeback_inodes_sb if no writeback is currently underway.
1261 * Returns 1 if writeback was started, 0 if not.
1263 int writeback_inodes_sb_if_idle(struct super_block
*sb
)
1265 if (!writeback_in_progress(sb
->s_bdi
)) {
1266 writeback_inodes_sb(sb
);
1271 EXPORT_SYMBOL(writeback_inodes_sb_if_idle
);
1274 * sync_inodes_sb - sync sb inode pages
1275 * @sb: the superblock
1277 * This function writes and waits on any dirty inode belonging to this
1278 * super_block. The number of pages synced is returned.
1280 void sync_inodes_sb(struct super_block
*sb
)
1282 bdi_sync_writeback(sb
->s_bdi
, sb
);
1285 EXPORT_SYMBOL(sync_inodes_sb
);
1288 * write_inode_now - write an inode to disk
1289 * @inode: inode to write to disk
1290 * @sync: whether the write should be synchronous or not
1292 * This function commits an inode to disk immediately if it is dirty. This is
1293 * primarily needed by knfsd.
1295 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1297 int write_inode_now(struct inode
*inode
, int sync
)
1300 struct writeback_control wbc
= {
1301 .nr_to_write
= LONG_MAX
,
1302 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1304 .range_end
= LLONG_MAX
,
1307 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1308 wbc
.nr_to_write
= 0;
1311 spin_lock(&inode_lock
);
1312 ret
= writeback_single_inode(inode
, &wbc
);
1313 spin_unlock(&inode_lock
);
1315 inode_sync_wait(inode
);
1318 EXPORT_SYMBOL(write_inode_now
);
1321 * sync_inode - write an inode and its pages to disk.
1322 * @inode: the inode to sync
1323 * @wbc: controls the writeback mode
1325 * sync_inode() will write an inode and its pages to disk. It will also
1326 * correctly update the inode on its superblock's dirty inode lists and will
1327 * update inode->i_state.
1329 * The caller must have a ref on the inode.
1331 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1335 spin_lock(&inode_lock
);
1336 ret
= writeback_single_inode(inode
, wbc
);
1337 spin_unlock(&inode_lock
);
1340 EXPORT_SYMBOL(sync_inode
);