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/sched.h>
22 #include <linux/kthread.h>
23 #include <linux/freezer.h>
24 #include <linux/writeback.h>
25 #include <linux/blkdev.h>
26 #include <linux/backing-dev.h>
27 #include <linux/buffer_head.h>
30 #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info)
33 * We don't actually have pdflush, but this one is exported though /proc...
35 int nr_pdflush_threads
;
38 * Passed into wb_writeback(), essentially a subset of writeback_control
40 struct wb_writeback_args
{
42 struct super_block
*sb
;
43 enum writeback_sync_modes sync_mode
;
49 * Work items for the bdi_writeback threads
52 struct list_head list
;
53 struct list_head wait_list
;
54 struct rcu_head rcu_head
;
59 struct wb_writeback_args args
;
69 #define WS_USED (1 << WS_USED_B)
70 #define WS_ONSTACK (1 << WS_ONSTACK_B)
72 static inline bool bdi_work_on_stack(struct bdi_work
*work
)
74 return test_bit(WS_ONSTACK_B
, &work
->state
);
77 static inline void bdi_work_init(struct bdi_work
*work
,
78 struct writeback_control
*wbc
)
80 INIT_RCU_HEAD(&work
->rcu_head
);
81 work
->args
.sb
= wbc
->sb
;
82 work
->args
.nr_pages
= wbc
->nr_to_write
;
83 work
->args
.sync_mode
= wbc
->sync_mode
;
84 work
->args
.range_cyclic
= wbc
->range_cyclic
;
85 work
->args
.for_kupdate
= 0;
86 work
->state
= WS_USED
;
90 * writeback_in_progress - determine whether there is writeback in progress
91 * @bdi: the device's backing_dev_info structure.
93 * Determine whether there is writeback waiting to be handled against a
96 int writeback_in_progress(struct backing_dev_info
*bdi
)
98 return !list_empty(&bdi
->work_list
);
101 static void bdi_work_clear(struct bdi_work
*work
)
103 clear_bit(WS_USED_B
, &work
->state
);
104 smp_mb__after_clear_bit();
105 wake_up_bit(&work
->state
, WS_USED_B
);
108 static void bdi_work_free(struct rcu_head
*head
)
110 struct bdi_work
*work
= container_of(head
, struct bdi_work
, rcu_head
);
112 if (!bdi_work_on_stack(work
))
115 bdi_work_clear(work
);
118 static void wb_work_complete(struct bdi_work
*work
)
120 const enum writeback_sync_modes sync_mode
= work
->args
.sync_mode
;
123 * For allocated work, we can clear the done/seen bit right here.
124 * For on-stack work, we need to postpone both the clear and free
125 * to after the RCU grace period, since the stack could be invalidated
126 * as soon as bdi_work_clear() has done the wakeup.
128 if (!bdi_work_on_stack(work
))
129 bdi_work_clear(work
);
130 if (sync_mode
== WB_SYNC_NONE
|| bdi_work_on_stack(work
))
131 call_rcu(&work
->rcu_head
, bdi_work_free
);
134 static void wb_clear_pending(struct bdi_writeback
*wb
, struct bdi_work
*work
)
137 * The caller has retrieved the work arguments from this work,
138 * drop our reference. If this is the last ref, delete and free it
140 if (atomic_dec_and_test(&work
->pending
)) {
141 struct backing_dev_info
*bdi
= wb
->bdi
;
143 spin_lock(&bdi
->wb_lock
);
144 list_del_rcu(&work
->list
);
145 spin_unlock(&bdi
->wb_lock
);
147 wb_work_complete(work
);
151 static void bdi_queue_work(struct backing_dev_info
*bdi
, struct bdi_work
*work
)
154 work
->seen
= bdi
->wb_mask
;
156 atomic_set(&work
->pending
, bdi
->wb_cnt
);
157 BUG_ON(!bdi
->wb_cnt
);
160 * Make sure stores are seen before it appears on the list
164 spin_lock(&bdi
->wb_lock
);
165 list_add_tail_rcu(&work
->list
, &bdi
->work_list
);
166 spin_unlock(&bdi
->wb_lock
);
170 * If the default thread isn't there, make sure we add it. When
171 * it gets created and wakes up, we'll run this work.
173 if (unlikely(list_empty_careful(&bdi
->wb_list
)))
174 wake_up_process(default_backing_dev_info
.wb
.task
);
176 struct bdi_writeback
*wb
= &bdi
->wb
;
179 * If we failed allocating the bdi work item, wake up the wb
180 * thread always. As a safety precaution, it'll flush out
183 if (!wb_has_dirty_io(wb
)) {
185 wb_clear_pending(wb
, work
);
187 wake_up_process(wb
->task
);
192 * Used for on-stack allocated work items. The caller needs to wait until
193 * the wb threads have acked the work before it's safe to continue.
195 static void bdi_wait_on_work_clear(struct bdi_work
*work
)
197 wait_on_bit(&work
->state
, WS_USED_B
, bdi_sched_wait
,
198 TASK_UNINTERRUPTIBLE
);
201 static struct bdi_work
*bdi_alloc_work(struct writeback_control
*wbc
)
203 struct bdi_work
*work
;
205 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
207 bdi_work_init(work
, wbc
);
212 void bdi_start_writeback(struct writeback_control
*wbc
)
215 * WB_SYNC_NONE is opportunistic writeback. If this allocation fails,
216 * bdi_queue_work() will wake up the thread and flush old data. This
217 * should ensure some amount of progress in freeing memory.
219 if (wbc
->sync_mode
!= WB_SYNC_ALL
) {
220 struct bdi_work
*w
= bdi_alloc_work(wbc
);
222 bdi_queue_work(wbc
->bdi
, w
);
224 struct bdi_work work
;
226 bdi_work_init(&work
, wbc
);
227 work
.state
|= WS_ONSTACK
;
229 bdi_queue_work(wbc
->bdi
, &work
);
230 bdi_wait_on_work_clear(&work
);
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 pdflush 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
)
298 while (!list_empty(delaying_queue
)) {
299 struct inode
*inode
= list_entry(delaying_queue
->prev
,
300 struct inode
, i_list
);
301 if (older_than_this
&&
302 inode_dirtied_after(inode
, *older_than_this
))
304 list_move(&inode
->i_list
, dispatch_queue
);
309 * Queue all expired dirty inodes for io, eldest first.
311 static void queue_io(struct bdi_writeback
*wb
, unsigned long *older_than_this
)
313 list_splice_init(&wb
->b_more_io
, wb
->b_io
.prev
);
314 move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, older_than_this
);
317 static int write_inode(struct inode
*inode
, int sync
)
319 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
))
320 return inode
->i_sb
->s_op
->write_inode(inode
, sync
);
325 * Wait for writeback on an inode to complete.
327 static void inode_wait_for_writeback(struct inode
*inode
)
329 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
330 wait_queue_head_t
*wqh
;
332 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
334 spin_unlock(&inode_lock
);
335 __wait_on_bit(wqh
, &wq
, inode_wait
, TASK_UNINTERRUPTIBLE
);
336 spin_lock(&inode_lock
);
337 } while (inode
->i_state
& I_SYNC
);
341 * Write out an inode's dirty pages. Called under inode_lock. Either the
342 * caller has ref on the inode (either via __iget or via syscall against an fd)
343 * or the inode has I_WILL_FREE set (via generic_forget_inode)
345 * If `wait' is set, wait on the writeout.
347 * The whole writeout design is quite complex and fragile. We want to avoid
348 * starvation of particular inodes when others are being redirtied, prevent
351 * Called under inode_lock.
354 writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
356 struct address_space
*mapping
= inode
->i_mapping
;
357 int wait
= wbc
->sync_mode
== WB_SYNC_ALL
;
361 if (!atomic_read(&inode
->i_count
))
362 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
364 WARN_ON(inode
->i_state
& I_WILL_FREE
);
366 if (inode
->i_state
& I_SYNC
) {
368 * If this inode is locked for writeback and we are not doing
369 * writeback-for-data-integrity, move it to b_more_io so that
370 * writeback can proceed with the other inodes on s_io.
372 * We'll have another go at writing back this inode when we
373 * completed a full scan of b_io.
381 * It's a data-integrity sync. We must wait.
383 inode_wait_for_writeback(inode
);
386 BUG_ON(inode
->i_state
& I_SYNC
);
388 /* Set I_SYNC, reset I_DIRTY */
389 dirty
= inode
->i_state
& I_DIRTY
;
390 inode
->i_state
|= I_SYNC
;
391 inode
->i_state
&= ~I_DIRTY
;
393 spin_unlock(&inode_lock
);
395 ret
= do_writepages(mapping
, wbc
);
397 /* Don't write the inode if only I_DIRTY_PAGES was set */
398 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
399 int err
= write_inode(inode
, wait
);
405 int err
= filemap_fdatawait(mapping
);
410 spin_lock(&inode_lock
);
411 inode
->i_state
&= ~I_SYNC
;
412 if (!(inode
->i_state
& (I_FREEING
| I_CLEAR
))) {
413 if (!(inode
->i_state
& I_DIRTY
) &&
414 mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
416 * We didn't write back all the pages. nfs_writepages()
417 * sometimes bales out without doing anything. Redirty
418 * the inode; Move it from b_io onto b_more_io/b_dirty.
421 * akpm: if the caller was the kupdate function we put
422 * this inode at the head of b_dirty so it gets first
423 * consideration. Otherwise, move it to the tail, for
424 * the reasons described there. I'm not really sure
425 * how much sense this makes. Presumably I had a good
426 * reasons for doing it this way, and I'd rather not
427 * muck with it at present.
429 if (wbc
->for_kupdate
) {
431 * For the kupdate function we move the inode
432 * to b_more_io so it will get more writeout as
433 * soon as the queue becomes uncongested.
435 inode
->i_state
|= I_DIRTY_PAGES
;
436 if (wbc
->nr_to_write
<= 0) {
438 * slice used up: queue for next turn
443 * somehow blocked: retry later
449 * Otherwise fully redirty the inode so that
450 * other inodes on this superblock will get some
451 * writeout. Otherwise heavy writing to one
452 * file would indefinitely suspend writeout of
453 * all the other files.
455 inode
->i_state
|= I_DIRTY_PAGES
;
458 } else if (inode
->i_state
& I_DIRTY
) {
460 * Someone redirtied the inode while were writing back
464 } else if (atomic_read(&inode
->i_count
)) {
466 * The inode is clean, inuse
468 list_move(&inode
->i_list
, &inode_in_use
);
471 * The inode is clean, unused
473 list_move(&inode
->i_list
, &inode_unused
);
476 inode_sync_complete(inode
);
481 * For WB_SYNC_NONE writeback, the caller does not have the sb pinned
482 * before calling writeback. So make sure that we do pin it, so it doesn't
483 * go away while we are writing inodes from it.
485 * Returns 0 if the super was successfully pinned (or pinning wasn't needed),
488 static int pin_sb_for_writeback(struct writeback_control
*wbc
,
491 struct super_block
*sb
= inode
->i_sb
;
494 * Caller must already hold the ref for this
496 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
497 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
503 if (down_read_trylock(&sb
->s_umount
)) {
505 spin_unlock(&sb_lock
);
509 * umounted, drop rwsem again and fall through to failure
511 up_read(&sb
->s_umount
);
515 spin_unlock(&sb_lock
);
519 static void unpin_sb_for_writeback(struct writeback_control
*wbc
,
522 struct super_block
*sb
= inode
->i_sb
;
524 if (wbc
->sync_mode
== WB_SYNC_ALL
)
527 up_read(&sb
->s_umount
);
531 static void writeback_inodes_wb(struct bdi_writeback
*wb
,
532 struct writeback_control
*wbc
)
534 struct super_block
*sb
= wbc
->sb
;
535 const int is_blkdev_sb
= sb_is_blkdev_sb(sb
);
536 const unsigned long start
= jiffies
; /* livelock avoidance */
538 spin_lock(&inode_lock
);
540 if (!wbc
->for_kupdate
|| list_empty(&wb
->b_io
))
541 queue_io(wb
, wbc
->older_than_this
);
543 while (!list_empty(&wb
->b_io
)) {
544 struct inode
*inode
= list_entry(wb
->b_io
.prev
,
545 struct inode
, i_list
);
549 * super block given and doesn't match, skip this inode
551 if (sb
&& sb
!= inode
->i_sb
) {
556 if (!bdi_cap_writeback_dirty(wb
->bdi
)) {
560 * Dirty memory-backed blockdev: the ramdisk
561 * driver does this. Skip just this inode
566 * Dirty memory-backed inode against a filesystem other
567 * than the kernel-internal bdev filesystem. Skip the
573 if (inode
->i_state
& (I_NEW
| I_WILL_FREE
)) {
578 if (wbc
->nonblocking
&& bdi_write_congested(wb
->bdi
)) {
579 wbc
->encountered_congestion
= 1;
581 break; /* Skip a congested fs */
583 continue; /* Skip a congested blockdev */
587 * Was this inode dirtied after sync_sb_inodes was called?
588 * This keeps sync from extra jobs and livelock.
590 if (inode_dirtied_after(inode
, start
))
593 if (pin_sb_for_writeback(wbc
, inode
)) {
598 BUG_ON(inode
->i_state
& (I_FREEING
| I_CLEAR
));
600 pages_skipped
= wbc
->pages_skipped
;
601 writeback_single_inode(inode
, wbc
);
602 unpin_sb_for_writeback(wbc
, inode
);
603 if (wbc
->pages_skipped
!= pages_skipped
) {
605 * writeback is not making progress due to locked
606 * buffers. Skip this inode for now.
610 spin_unlock(&inode_lock
);
613 spin_lock(&inode_lock
);
614 if (wbc
->nr_to_write
<= 0) {
618 if (!list_empty(&wb
->b_more_io
))
622 spin_unlock(&inode_lock
);
623 /* Leave any unwritten inodes on b_io */
626 void writeback_inodes_wbc(struct writeback_control
*wbc
)
628 struct backing_dev_info
*bdi
= wbc
->bdi
;
630 writeback_inodes_wb(&bdi
->wb
, wbc
);
634 * The maximum number of pages to writeout in a single bdi flush/kupdate
635 * operation. We do this so we don't hold I_SYNC against an inode for
636 * enormous amounts of time, which would block a userspace task which has
637 * been forced to throttle against that inode. Also, the code reevaluates
638 * the dirty each time it has written this many pages.
640 #define MAX_WRITEBACK_PAGES 1024
642 static inline bool over_bground_thresh(void)
644 unsigned long background_thresh
, dirty_thresh
;
646 get_dirty_limits(&background_thresh
, &dirty_thresh
, NULL
, NULL
);
648 return (global_page_state(NR_FILE_DIRTY
) +
649 global_page_state(NR_UNSTABLE_NFS
) >= background_thresh
);
653 * Explicit flushing or periodic writeback of "old" data.
655 * Define "old": the first time one of an inode's pages is dirtied, we mark the
656 * dirtying-time in the inode's address_space. So this periodic writeback code
657 * just walks the superblock inode list, writing back any inodes which are
658 * older than a specific point in time.
660 * Try to run once per dirty_writeback_interval. But if a writeback event
661 * takes longer than a dirty_writeback_interval interval, then leave a
664 * older_than_this takes precedence over nr_to_write. So we'll only write back
665 * all dirty pages if they are all attached to "old" mappings.
667 static long wb_writeback(struct bdi_writeback
*wb
,
668 struct wb_writeback_args
*args
)
670 struct writeback_control wbc
= {
673 .sync_mode
= args
->sync_mode
,
674 .older_than_this
= NULL
,
675 .for_kupdate
= args
->for_kupdate
,
676 .range_cyclic
= args
->range_cyclic
,
678 unsigned long oldest_jif
;
681 if (wbc
.for_kupdate
) {
682 wbc
.older_than_this
= &oldest_jif
;
683 oldest_jif
= jiffies
-
684 msecs_to_jiffies(dirty_expire_interval
* 10);
686 if (!wbc
.range_cyclic
) {
688 wbc
.range_end
= LLONG_MAX
;
693 * Don't flush anything for non-integrity writeback where
694 * no nr_pages was given
696 if (!args
->for_kupdate
&& args
->nr_pages
<= 0 &&
697 args
->sync_mode
== WB_SYNC_NONE
)
701 * If no specific pages were given and this is just a
702 * periodic background writeout and we are below the
703 * background dirty threshold, don't do anything
705 if (args
->for_kupdate
&& args
->nr_pages
<= 0 &&
706 !over_bground_thresh())
710 wbc
.encountered_congestion
= 0;
711 wbc
.nr_to_write
= MAX_WRITEBACK_PAGES
;
712 wbc
.pages_skipped
= 0;
713 writeback_inodes_wb(wb
, &wbc
);
714 args
->nr_pages
-= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
715 wrote
+= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
718 * If we ran out of stuff to write, bail unless more_io got set
720 if (wbc
.nr_to_write
> 0 || wbc
.pages_skipped
> 0) {
721 if (wbc
.more_io
&& !wbc
.for_kupdate
)
731 * Return the next bdi_work struct that hasn't been processed by this
734 static struct bdi_work
*get_next_work_item(struct backing_dev_info
*bdi
,
735 struct bdi_writeback
*wb
)
737 struct bdi_work
*work
, *ret
= NULL
;
741 list_for_each_entry_rcu(work
, &bdi
->work_list
, list
) {
742 if (!test_and_clear_bit(wb
->nr
, &work
->seen
))
753 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
755 unsigned long expired
;
758 expired
= wb
->last_old_flush
+
759 msecs_to_jiffies(dirty_writeback_interval
* 10);
760 if (time_before(jiffies
, expired
))
763 wb
->last_old_flush
= jiffies
;
764 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
765 global_page_state(NR_UNSTABLE_NFS
) +
766 (inodes_stat
.nr_inodes
- inodes_stat
.nr_unused
);
769 struct wb_writeback_args args
= {
770 .nr_pages
= nr_pages
,
771 .sync_mode
= WB_SYNC_NONE
,
776 return wb_writeback(wb
, &args
);
783 * Retrieve work items and do the writeback they describe
785 long wb_do_writeback(struct bdi_writeback
*wb
, int force_wait
)
787 struct backing_dev_info
*bdi
= wb
->bdi
;
788 struct bdi_work
*work
;
791 while ((work
= get_next_work_item(bdi
, wb
)) != NULL
) {
792 struct wb_writeback_args args
= work
->args
;
795 * Override sync mode, in case we must wait for completion
798 work
->args
.sync_mode
= args
.sync_mode
= WB_SYNC_ALL
;
801 * If this isn't a data integrity operation, just notify
802 * that we have seen this work and we are now starting it.
804 if (args
.sync_mode
== WB_SYNC_NONE
)
805 wb_clear_pending(wb
, work
);
807 wrote
+= wb_writeback(wb
, &args
);
810 * This is a data integrity writeback, so only do the
811 * notification when we have completed the work.
813 if (args
.sync_mode
== WB_SYNC_ALL
)
814 wb_clear_pending(wb
, work
);
818 * Check for periodic writeback, kupdated() style
820 wrote
+= wb_check_old_data_flush(wb
);
826 * Handle writeback of dirty data for the device backed by this bdi. Also
827 * wakes up periodically and does kupdated style flushing.
829 int bdi_writeback_task(struct bdi_writeback
*wb
)
831 unsigned long last_active
= jiffies
;
832 unsigned long wait_jiffies
= -1UL;
835 while (!kthread_should_stop()) {
836 pages_written
= wb_do_writeback(wb
, 0);
839 last_active
= jiffies
;
840 else if (wait_jiffies
!= -1UL) {
841 unsigned long max_idle
;
844 * Longest period of inactivity that we tolerate. If we
845 * see dirty data again later, the task will get
846 * recreated automatically.
848 max_idle
= max(5UL * 60 * HZ
, wait_jiffies
);
849 if (time_after(jiffies
, max_idle
+ last_active
))
853 wait_jiffies
= msecs_to_jiffies(dirty_writeback_interval
* 10);
854 set_current_state(TASK_INTERRUPTIBLE
);
855 schedule_timeout(wait_jiffies
);
863 * Schedule writeback for all backing devices. Expensive! If this is a data
864 * integrity operation, writeback will be complete when this returns. If
865 * we are simply called for WB_SYNC_NONE, then writeback will merely be
868 static void bdi_writeback_all(struct writeback_control
*wbc
)
870 const bool must_wait
= wbc
->sync_mode
== WB_SYNC_ALL
;
871 struct backing_dev_info
*bdi
;
872 struct bdi_work
*work
;
876 spin_lock(&bdi_lock
);
878 list_for_each_entry(bdi
, &bdi_list
, bdi_list
) {
879 struct bdi_work
*work
;
881 if (!bdi_has_dirty_io(bdi
))
885 * If work allocation fails, do the writes inline. We drop
886 * the lock and restart the list writeout. This should be OK,
887 * since this happens rarely and because the writeout should
888 * eventually make more free memory available.
890 work
= bdi_alloc_work(wbc
);
892 struct writeback_control __wbc
;
895 * Not a data integrity writeout, just continue
900 spin_unlock(&bdi_lock
);
903 writeback_inodes_wbc(&__wbc
);
907 list_add_tail(&work
->wait_list
, &list
);
909 bdi_queue_work(bdi
, work
);
912 spin_unlock(&bdi_lock
);
915 * If this is for WB_SYNC_ALL, wait for pending work to complete
918 while (!list_empty(&list
)) {
919 work
= list_entry(list
.next
, struct bdi_work
, wait_list
);
920 list_del(&work
->wait_list
);
921 bdi_wait_on_work_clear(work
);
922 call_rcu(&work
->rcu_head
, bdi_work_free
);
927 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
930 void wakeup_flusher_threads(long nr_pages
)
932 struct writeback_control wbc
= {
933 .sync_mode
= WB_SYNC_NONE
,
934 .older_than_this
= NULL
,
939 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
940 global_page_state(NR_UNSTABLE_NFS
);
941 wbc
.nr_to_write
= nr_pages
;
942 bdi_writeback_all(&wbc
);
945 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
947 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
948 struct dentry
*dentry
;
949 const char *name
= "?";
951 dentry
= d_find_alias(inode
);
953 spin_lock(&dentry
->d_lock
);
954 name
= (const char *) dentry
->d_name
.name
;
957 "%s(%d): dirtied inode %lu (%s) on %s\n",
958 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
959 name
, inode
->i_sb
->s_id
);
961 spin_unlock(&dentry
->d_lock
);
968 * __mark_inode_dirty - internal function
969 * @inode: inode to mark
970 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
971 * Mark an inode as dirty. Callers should use mark_inode_dirty or
972 * mark_inode_dirty_sync.
974 * Put the inode on the super block's dirty list.
976 * CAREFUL! We mark it dirty unconditionally, but move it onto the
977 * dirty list only if it is hashed or if it refers to a blockdev.
978 * If it was not hashed, it will never be added to the dirty list
979 * even if it is later hashed, as it will have been marked dirty already.
981 * In short, make sure you hash any inodes _before_ you start marking
984 * This function *must* be atomic for the I_DIRTY_PAGES case -
985 * set_page_dirty() is called under spinlock in several places.
987 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
988 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
989 * the kernel-internal blockdev inode represents the dirtying time of the
990 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
991 * page->mapping->host, so the page-dirtying time is recorded in the internal
994 void __mark_inode_dirty(struct inode
*inode
, int flags
)
996 struct super_block
*sb
= inode
->i_sb
;
999 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1000 * dirty the inode itself
1002 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
1003 if (sb
->s_op
->dirty_inode
)
1004 sb
->s_op
->dirty_inode(inode
);
1008 * make sure that changes are seen by all cpus before we test i_state
1013 /* avoid the locking if we can */
1014 if ((inode
->i_state
& flags
) == flags
)
1017 if (unlikely(block_dump
))
1018 block_dump___mark_inode_dirty(inode
);
1020 spin_lock(&inode_lock
);
1021 if ((inode
->i_state
& flags
) != flags
) {
1022 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1024 inode
->i_state
|= flags
;
1027 * If the inode is being synced, just update its dirty state.
1028 * The unlocker will place the inode on the appropriate
1029 * superblock list, based upon its state.
1031 if (inode
->i_state
& I_SYNC
)
1035 * Only add valid (hashed) inodes to the superblock's
1036 * dirty list. Add blockdev inodes as well.
1038 if (!S_ISBLK(inode
->i_mode
)) {
1039 if (hlist_unhashed(&inode
->i_hash
))
1042 if (inode
->i_state
& (I_FREEING
|I_CLEAR
))
1046 * If the inode was already on b_dirty/b_io/b_more_io, don't
1047 * reposition it (that would break b_dirty time-ordering).
1050 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1051 struct backing_dev_info
*bdi
= wb
->bdi
;
1053 if (bdi_cap_writeback_dirty(bdi
) &&
1054 !test_bit(BDI_registered
, &bdi
->state
)) {
1056 printk(KERN_ERR
"bdi-%s not registered\n",
1060 inode
->dirtied_when
= jiffies
;
1061 list_move(&inode
->i_list
, &wb
->b_dirty
);
1065 spin_unlock(&inode_lock
);
1067 EXPORT_SYMBOL(__mark_inode_dirty
);
1070 * Write out a superblock's list of dirty inodes. A wait will be performed
1071 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1073 * If older_than_this is non-NULL, then only write out inodes which
1074 * had their first dirtying at a time earlier than *older_than_this.
1076 * If we're a pdlfush thread, then implement pdflush collision avoidance
1077 * against the entire list.
1079 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1080 * This function assumes that the blockdev superblock's inodes are backed by
1081 * a variety of queues, so all inodes are searched. For other superblocks,
1082 * assume that all inodes are backed by the same queue.
1084 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1085 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1086 * on the writer throttling path, and we get decent balancing between many
1087 * throttled threads: we don't want them all piling up on inode_sync_wait.
1089 static void wait_sb_inodes(struct writeback_control
*wbc
)
1091 struct inode
*inode
, *old_inode
= NULL
;
1094 * We need to be protected against the filesystem going from
1095 * r/o to r/w or vice versa.
1097 WARN_ON(!rwsem_is_locked(&wbc
->sb
->s_umount
));
1099 spin_lock(&inode_lock
);
1102 * Data integrity sync. Must wait for all pages under writeback,
1103 * because there may have been pages dirtied before our sync
1104 * call, but which had writeout started before we write it out.
1105 * In which case, the inode may not be on the dirty list, but
1106 * we still have to wait for that writeout.
1108 list_for_each_entry(inode
, &wbc
->sb
->s_inodes
, i_sb_list
) {
1109 struct address_space
*mapping
;
1111 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
|I_NEW
))
1113 mapping
= inode
->i_mapping
;
1114 if (mapping
->nrpages
== 0)
1117 spin_unlock(&inode_lock
);
1119 * We hold a reference to 'inode' so it couldn't have
1120 * been removed from s_inodes list while we dropped the
1121 * inode_lock. We cannot iput the inode now as we can
1122 * be holding the last reference and we cannot iput it
1123 * under inode_lock. So we keep the reference and iput
1129 filemap_fdatawait(mapping
);
1133 spin_lock(&inode_lock
);
1135 spin_unlock(&inode_lock
);
1140 * writeback_inodes_sb - writeback dirty inodes from given super_block
1141 * @sb: the superblock
1143 * Start writeback on some inodes on this super_block. No guarantees are made
1144 * on how many (if any) will be written, and this function does not wait
1145 * for IO completion of submitted IO. The number of pages submitted is
1148 long writeback_inodes_sb(struct super_block
*sb
)
1150 struct writeback_control wbc
= {
1152 .sync_mode
= WB_SYNC_NONE
,
1154 .range_end
= LLONG_MAX
,
1156 unsigned long nr_dirty
= global_page_state(NR_FILE_DIRTY
);
1157 unsigned long nr_unstable
= global_page_state(NR_UNSTABLE_NFS
);
1160 nr_to_write
= nr_dirty
+ nr_unstable
+
1161 (inodes_stat
.nr_inodes
- inodes_stat
.nr_unused
);
1163 wbc
.nr_to_write
= nr_to_write
;
1164 bdi_writeback_all(&wbc
);
1165 return nr_to_write
- wbc
.nr_to_write
;
1167 EXPORT_SYMBOL(writeback_inodes_sb
);
1170 * sync_inodes_sb - sync sb inode pages
1171 * @sb: the superblock
1173 * This function writes and waits on any dirty inode belonging to this
1174 * super_block. The number of pages synced is returned.
1176 long sync_inodes_sb(struct super_block
*sb
)
1178 struct writeback_control wbc
= {
1180 .sync_mode
= WB_SYNC_ALL
,
1182 .range_end
= LLONG_MAX
,
1184 long nr_to_write
= LONG_MAX
; /* doesn't actually matter */
1186 wbc
.nr_to_write
= nr_to_write
;
1187 bdi_writeback_all(&wbc
);
1188 wait_sb_inodes(&wbc
);
1189 return nr_to_write
- wbc
.nr_to_write
;
1191 EXPORT_SYMBOL(sync_inodes_sb
);
1194 * write_inode_now - write an inode to disk
1195 * @inode: inode to write to disk
1196 * @sync: whether the write should be synchronous or not
1198 * This function commits an inode to disk immediately if it is dirty. This is
1199 * primarily needed by knfsd.
1201 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1203 int write_inode_now(struct inode
*inode
, int sync
)
1206 struct writeback_control wbc
= {
1207 .nr_to_write
= LONG_MAX
,
1208 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1210 .range_end
= LLONG_MAX
,
1213 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1214 wbc
.nr_to_write
= 0;
1217 spin_lock(&inode_lock
);
1218 ret
= writeback_single_inode(inode
, &wbc
);
1219 spin_unlock(&inode_lock
);
1221 inode_sync_wait(inode
);
1224 EXPORT_SYMBOL(write_inode_now
);
1227 * sync_inode - write an inode and its pages to disk.
1228 * @inode: the inode to sync
1229 * @wbc: controls the writeback mode
1231 * sync_inode() will write an inode and its pages to disk. It will also
1232 * correctly update the inode on its superblock's dirty inode lists and will
1233 * update inode->i_state.
1235 * The caller must have a ref on the inode.
1237 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1241 spin_lock(&inode_lock
);
1242 ret
= writeback_single_inode(inode
, wbc
);
1243 spin_unlock(&inode_lock
);
1246 EXPORT_SYMBOL(sync_inode
);