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
; /* pending work list */
53 struct rcu_head rcu_head
; /* for RCU free/clear of work */
55 unsigned long seen
; /* threads that have seen this work */
56 atomic_t pending
; /* number of threads still to do work */
58 struct wb_writeback_args args
; /* writeback arguments */
60 unsigned long state
; /* flag bits, see WS_* */
68 #define WS_USED (1 << WS_USED_B)
69 #define WS_ONSTACK (1 << WS_ONSTACK_B)
71 static inline bool bdi_work_on_stack(struct bdi_work
*work
)
73 return test_bit(WS_ONSTACK_B
, &work
->state
);
76 static inline void bdi_work_init(struct bdi_work
*work
,
77 struct wb_writeback_args
*args
)
79 INIT_RCU_HEAD(&work
->rcu_head
);
81 work
->state
= WS_USED
;
85 * writeback_in_progress - determine whether there is writeback in progress
86 * @bdi: the device's backing_dev_info structure.
88 * Determine whether there is writeback waiting to be handled against a
91 int writeback_in_progress(struct backing_dev_info
*bdi
)
93 return !list_empty(&bdi
->work_list
);
96 static void bdi_work_clear(struct bdi_work
*work
)
98 clear_bit(WS_USED_B
, &work
->state
);
99 smp_mb__after_clear_bit();
100 wake_up_bit(&work
->state
, WS_USED_B
);
103 static void bdi_work_free(struct rcu_head
*head
)
105 struct bdi_work
*work
= container_of(head
, struct bdi_work
, rcu_head
);
107 if (!bdi_work_on_stack(work
))
110 bdi_work_clear(work
);
113 static void wb_work_complete(struct bdi_work
*work
)
115 const enum writeback_sync_modes sync_mode
= work
->args
.sync_mode
;
118 * For allocated work, we can clear the done/seen bit right here.
119 * For on-stack work, we need to postpone both the clear and free
120 * to after the RCU grace period, since the stack could be invalidated
121 * as soon as bdi_work_clear() has done the wakeup.
123 if (!bdi_work_on_stack(work
))
124 bdi_work_clear(work
);
125 if (sync_mode
== WB_SYNC_NONE
|| bdi_work_on_stack(work
))
126 call_rcu(&work
->rcu_head
, bdi_work_free
);
129 static void wb_clear_pending(struct bdi_writeback
*wb
, struct bdi_work
*work
)
132 * The caller has retrieved the work arguments from this work,
133 * drop our reference. If this is the last ref, delete and free it
135 if (atomic_dec_and_test(&work
->pending
)) {
136 struct backing_dev_info
*bdi
= wb
->bdi
;
138 spin_lock(&bdi
->wb_lock
);
139 list_del_rcu(&work
->list
);
140 spin_unlock(&bdi
->wb_lock
);
142 wb_work_complete(work
);
146 static void bdi_queue_work(struct backing_dev_info
*bdi
, struct bdi_work
*work
)
148 work
->seen
= bdi
->wb_mask
;
150 atomic_set(&work
->pending
, bdi
->wb_cnt
);
151 BUG_ON(!bdi
->wb_cnt
);
154 * Make sure stores are seen before it appears on the list
158 spin_lock(&bdi
->wb_lock
);
159 list_add_tail_rcu(&work
->list
, &bdi
->work_list
);
160 spin_unlock(&bdi
->wb_lock
);
163 * If the default thread isn't there, make sure we add it. When
164 * it gets created and wakes up, we'll run this work.
166 if (unlikely(list_empty_careful(&bdi
->wb_list
)))
167 wake_up_process(default_backing_dev_info
.wb
.task
);
169 struct bdi_writeback
*wb
= &bdi
->wb
;
172 * End work now if this wb has no dirty IO pending. Otherwise
173 * wakeup the handling thread
175 if (!wb_has_dirty_io(wb
))
176 wb_clear_pending(wb
, work
);
178 wake_up_process(wb
->task
);
183 * Used for on-stack allocated work items. The caller needs to wait until
184 * the wb threads have acked the work before it's safe to continue.
186 static void bdi_wait_on_work_clear(struct bdi_work
*work
)
188 wait_on_bit(&work
->state
, WS_USED_B
, bdi_sched_wait
,
189 TASK_UNINTERRUPTIBLE
);
192 static void bdi_alloc_queue_work(struct backing_dev_info
*bdi
,
193 struct wb_writeback_args
*args
)
195 struct bdi_work
*work
;
198 * This is WB_SYNC_NONE writeback, so if allocation fails just
199 * wakeup the thread for old dirty data writeback
201 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
203 bdi_work_init(work
, args
);
204 bdi_queue_work(bdi
, work
);
206 struct bdi_writeback
*wb
= &bdi
->wb
;
209 wake_up_process(wb
->task
);
214 * bdi_sync_writeback - start and wait for writeback
215 * @bdi: the backing device to write from
216 * @sb: write inodes from this super_block
219 * This does WB_SYNC_ALL data integrity writeback and waits for the
220 * IO to complete. Callers must hold the sb s_umount semaphore for
221 * reading, to avoid having the super disappear before we are done.
223 static void bdi_sync_writeback(struct backing_dev_info
*bdi
,
224 struct super_block
*sb
)
226 struct wb_writeback_args args
= {
228 .sync_mode
= WB_SYNC_ALL
,
229 .nr_pages
= LONG_MAX
,
232 struct bdi_work work
;
234 bdi_work_init(&work
, &args
);
235 work
.state
|= WS_ONSTACK
;
237 bdi_queue_work(bdi
, &work
);
238 bdi_wait_on_work_clear(&work
);
242 * bdi_start_writeback - start writeback
243 * @bdi: the backing device to write from
244 * @nr_pages: the number of pages to write
247 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
248 * started when this function returns, we make no guarentees on
249 * completion. Caller need not hold sb s_umount semaphore.
252 void bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
)
254 struct wb_writeback_args args
= {
255 .sync_mode
= WB_SYNC_NONE
,
256 .nr_pages
= nr_pages
,
260 bdi_alloc_queue_work(bdi
, &args
);
264 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
265 * furthest end of its superblock's dirty-inode list.
267 * Before stamping the inode's ->dirtied_when, we check to see whether it is
268 * already the most-recently-dirtied inode on the b_dirty list. If that is
269 * the case then the inode must have been redirtied while it was being written
270 * out and we don't reset its dirtied_when.
272 static void redirty_tail(struct inode
*inode
)
274 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
276 if (!list_empty(&wb
->b_dirty
)) {
279 tail
= list_entry(wb
->b_dirty
.next
, struct inode
, i_list
);
280 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
281 inode
->dirtied_when
= jiffies
;
283 list_move(&inode
->i_list
, &wb
->b_dirty
);
287 * requeue inode for re-scanning after bdi->b_io list is exhausted.
289 static void requeue_io(struct inode
*inode
)
291 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
293 list_move(&inode
->i_list
, &wb
->b_more_io
);
296 static void inode_sync_complete(struct inode
*inode
)
299 * Prevent speculative execution through spin_unlock(&inode_lock);
302 wake_up_bit(&inode
->i_state
, __I_SYNC
);
305 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
307 bool ret
= time_after(inode
->dirtied_when
, t
);
310 * For inodes being constantly redirtied, dirtied_when can get stuck.
311 * It _appears_ to be in the future, but is actually in distant past.
312 * This test is necessary to prevent such wrapped-around relative times
313 * from permanently stopping the whole pdflush writeback.
315 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
321 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
323 static void move_expired_inodes(struct list_head
*delaying_queue
,
324 struct list_head
*dispatch_queue
,
325 unsigned long *older_than_this
)
327 while (!list_empty(delaying_queue
)) {
328 struct inode
*inode
= list_entry(delaying_queue
->prev
,
329 struct inode
, i_list
);
330 if (older_than_this
&&
331 inode_dirtied_after(inode
, *older_than_this
))
333 list_move(&inode
->i_list
, dispatch_queue
);
338 * Queue all expired dirty inodes for io, eldest first.
340 static void queue_io(struct bdi_writeback
*wb
, unsigned long *older_than_this
)
342 list_splice_init(&wb
->b_more_io
, wb
->b_io
.prev
);
343 move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, older_than_this
);
346 static int write_inode(struct inode
*inode
, int sync
)
348 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
))
349 return inode
->i_sb
->s_op
->write_inode(inode
, sync
);
354 * Wait for writeback on an inode to complete.
356 static void inode_wait_for_writeback(struct inode
*inode
)
358 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
359 wait_queue_head_t
*wqh
;
361 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
363 spin_unlock(&inode_lock
);
364 __wait_on_bit(wqh
, &wq
, inode_wait
, TASK_UNINTERRUPTIBLE
);
365 spin_lock(&inode_lock
);
366 } while (inode
->i_state
& I_SYNC
);
370 * Write out an inode's dirty pages. Called under inode_lock. Either the
371 * caller has ref on the inode (either via __iget or via syscall against an fd)
372 * or the inode has I_WILL_FREE set (via generic_forget_inode)
374 * If `wait' is set, wait on the writeout.
376 * The whole writeout design is quite complex and fragile. We want to avoid
377 * starvation of particular inodes when others are being redirtied, prevent
380 * Called under inode_lock.
383 writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
385 struct address_space
*mapping
= inode
->i_mapping
;
386 int wait
= wbc
->sync_mode
== WB_SYNC_ALL
;
390 if (!atomic_read(&inode
->i_count
))
391 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
393 WARN_ON(inode
->i_state
& I_WILL_FREE
);
395 if (inode
->i_state
& I_SYNC
) {
397 * If this inode is locked for writeback and we are not doing
398 * writeback-for-data-integrity, move it to b_more_io so that
399 * writeback can proceed with the other inodes on s_io.
401 * We'll have another go at writing back this inode when we
402 * completed a full scan of b_io.
410 * It's a data-integrity sync. We must wait.
412 inode_wait_for_writeback(inode
);
415 BUG_ON(inode
->i_state
& I_SYNC
);
417 /* Set I_SYNC, reset I_DIRTY */
418 dirty
= inode
->i_state
& I_DIRTY
;
419 inode
->i_state
|= I_SYNC
;
420 inode
->i_state
&= ~I_DIRTY
;
422 spin_unlock(&inode_lock
);
424 ret
= do_writepages(mapping
, wbc
);
426 /* Don't write the inode if only I_DIRTY_PAGES was set */
427 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
428 int err
= write_inode(inode
, wait
);
434 int err
= filemap_fdatawait(mapping
);
439 spin_lock(&inode_lock
);
440 inode
->i_state
&= ~I_SYNC
;
441 if (!(inode
->i_state
& (I_FREEING
| I_CLEAR
))) {
442 if (!(inode
->i_state
& I_DIRTY
) &&
443 mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
445 * We didn't write back all the pages. nfs_writepages()
446 * sometimes bales out without doing anything. Redirty
447 * the inode; Move it from b_io onto b_more_io/b_dirty.
450 * akpm: if the caller was the kupdate function we put
451 * this inode at the head of b_dirty so it gets first
452 * consideration. Otherwise, move it to the tail, for
453 * the reasons described there. I'm not really sure
454 * how much sense this makes. Presumably I had a good
455 * reasons for doing it this way, and I'd rather not
456 * muck with it at present.
458 if (wbc
->for_kupdate
) {
460 * For the kupdate function we move the inode
461 * to b_more_io so it will get more writeout as
462 * soon as the queue becomes uncongested.
464 inode
->i_state
|= I_DIRTY_PAGES
;
465 if (wbc
->nr_to_write
<= 0) {
467 * slice used up: queue for next turn
472 * somehow blocked: retry later
478 * Otherwise fully redirty the inode so that
479 * other inodes on this superblock will get some
480 * writeout. Otherwise heavy writing to one
481 * file would indefinitely suspend writeout of
482 * all the other files.
484 inode
->i_state
|= I_DIRTY_PAGES
;
487 } else if (inode
->i_state
& I_DIRTY
) {
489 * Someone redirtied the inode while were writing back
493 } else if (atomic_read(&inode
->i_count
)) {
495 * The inode is clean, inuse
497 list_move(&inode
->i_list
, &inode_in_use
);
500 * The inode is clean, unused
502 list_move(&inode
->i_list
, &inode_unused
);
505 inode_sync_complete(inode
);
510 * For WB_SYNC_NONE writeback, the caller does not have the sb pinned
511 * before calling writeback. So make sure that we do pin it, so it doesn't
512 * go away while we are writing inodes from it.
514 * Returns 0 if the super was successfully pinned (or pinning wasn't needed),
517 static int pin_sb_for_writeback(struct writeback_control
*wbc
,
520 struct super_block
*sb
= inode
->i_sb
;
523 * Caller must already hold the ref for this
525 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
526 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
532 if (down_read_trylock(&sb
->s_umount
)) {
534 spin_unlock(&sb_lock
);
538 * umounted, drop rwsem again and fall through to failure
540 up_read(&sb
->s_umount
);
544 spin_unlock(&sb_lock
);
548 static void unpin_sb_for_writeback(struct writeback_control
*wbc
,
551 struct super_block
*sb
= inode
->i_sb
;
553 if (wbc
->sync_mode
== WB_SYNC_ALL
)
556 up_read(&sb
->s_umount
);
560 static void writeback_inodes_wb(struct bdi_writeback
*wb
,
561 struct writeback_control
*wbc
)
563 struct super_block
*sb
= wbc
->sb
;
564 const int is_blkdev_sb
= sb_is_blkdev_sb(sb
);
565 const unsigned long start
= jiffies
; /* livelock avoidance */
567 spin_lock(&inode_lock
);
569 if (!wbc
->for_kupdate
|| list_empty(&wb
->b_io
))
570 queue_io(wb
, wbc
->older_than_this
);
572 while (!list_empty(&wb
->b_io
)) {
573 struct inode
*inode
= list_entry(wb
->b_io
.prev
,
574 struct inode
, i_list
);
578 * super block given and doesn't match, skip this inode
580 if (sb
&& sb
!= inode
->i_sb
) {
585 if (!bdi_cap_writeback_dirty(wb
->bdi
)) {
589 * Dirty memory-backed blockdev: the ramdisk
590 * driver does this. Skip just this inode
595 * Dirty memory-backed inode against a filesystem other
596 * than the kernel-internal bdev filesystem. Skip the
602 if (inode
->i_state
& (I_NEW
| I_WILL_FREE
)) {
607 if (wbc
->nonblocking
&& bdi_write_congested(wb
->bdi
)) {
608 wbc
->encountered_congestion
= 1;
610 break; /* Skip a congested fs */
612 continue; /* Skip a congested blockdev */
616 * Was this inode dirtied after sync_sb_inodes was called?
617 * This keeps sync from extra jobs and livelock.
619 if (inode_dirtied_after(inode
, start
))
622 if (pin_sb_for_writeback(wbc
, inode
)) {
627 BUG_ON(inode
->i_state
& (I_FREEING
| I_CLEAR
));
629 pages_skipped
= wbc
->pages_skipped
;
630 writeback_single_inode(inode
, wbc
);
631 unpin_sb_for_writeback(wbc
, inode
);
632 if (wbc
->pages_skipped
!= pages_skipped
) {
634 * writeback is not making progress due to locked
635 * buffers. Skip this inode for now.
639 spin_unlock(&inode_lock
);
642 spin_lock(&inode_lock
);
643 if (wbc
->nr_to_write
<= 0) {
647 if (!list_empty(&wb
->b_more_io
))
651 spin_unlock(&inode_lock
);
652 /* Leave any unwritten inodes on b_io */
655 void writeback_inodes_wbc(struct writeback_control
*wbc
)
657 struct backing_dev_info
*bdi
= wbc
->bdi
;
659 writeback_inodes_wb(&bdi
->wb
, wbc
);
663 * The maximum number of pages to writeout in a single bdi flush/kupdate
664 * operation. We do this so we don't hold I_SYNC against an inode for
665 * enormous amounts of time, which would block a userspace task which has
666 * been forced to throttle against that inode. Also, the code reevaluates
667 * the dirty each time it has written this many pages.
669 #define MAX_WRITEBACK_PAGES 1024
671 static inline bool over_bground_thresh(void)
673 unsigned long background_thresh
, dirty_thresh
;
675 get_dirty_limits(&background_thresh
, &dirty_thresh
, NULL
, NULL
);
677 return (global_page_state(NR_FILE_DIRTY
) +
678 global_page_state(NR_UNSTABLE_NFS
) >= background_thresh
);
682 * Explicit flushing or periodic writeback of "old" data.
684 * Define "old": the first time one of an inode's pages is dirtied, we mark the
685 * dirtying-time in the inode's address_space. So this periodic writeback code
686 * just walks the superblock inode list, writing back any inodes which are
687 * older than a specific point in time.
689 * Try to run once per dirty_writeback_interval. But if a writeback event
690 * takes longer than a dirty_writeback_interval interval, then leave a
693 * older_than_this takes precedence over nr_to_write. So we'll only write back
694 * all dirty pages if they are all attached to "old" mappings.
696 static long wb_writeback(struct bdi_writeback
*wb
,
697 struct wb_writeback_args
*args
)
699 struct writeback_control wbc
= {
702 .sync_mode
= args
->sync_mode
,
703 .older_than_this
= NULL
,
704 .for_kupdate
= args
->for_kupdate
,
705 .range_cyclic
= args
->range_cyclic
,
707 unsigned long oldest_jif
;
710 if (wbc
.for_kupdate
) {
711 wbc
.older_than_this
= &oldest_jif
;
712 oldest_jif
= jiffies
-
713 msecs_to_jiffies(dirty_expire_interval
* 10);
715 if (!wbc
.range_cyclic
) {
717 wbc
.range_end
= LLONG_MAX
;
722 * Don't flush anything for non-integrity writeback where
723 * no nr_pages was given
725 if (!args
->for_kupdate
&& args
->nr_pages
<= 0 &&
726 args
->sync_mode
== WB_SYNC_NONE
)
730 * If no specific pages were given and this is just a
731 * periodic background writeout and we are below the
732 * background dirty threshold, don't do anything
734 if (args
->for_kupdate
&& args
->nr_pages
<= 0 &&
735 !over_bground_thresh())
739 wbc
.encountered_congestion
= 0;
740 wbc
.nr_to_write
= MAX_WRITEBACK_PAGES
;
741 wbc
.pages_skipped
= 0;
742 writeback_inodes_wb(wb
, &wbc
);
743 args
->nr_pages
-= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
744 wrote
+= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
747 * If we ran out of stuff to write, bail unless more_io got set
749 if (wbc
.nr_to_write
> 0 || wbc
.pages_skipped
> 0) {
750 if (wbc
.more_io
&& !wbc
.for_kupdate
)
760 * Return the next bdi_work struct that hasn't been processed by this
761 * wb thread yet. ->seen is initially set for each thread that exists
762 * for this device, when a thread first notices a piece of work it
763 * clears its bit. Depending on writeback type, the thread will notify
764 * completion on either receiving the work (WB_SYNC_NONE) or after
765 * it is done (WB_SYNC_ALL).
767 static struct bdi_work
*get_next_work_item(struct backing_dev_info
*bdi
,
768 struct bdi_writeback
*wb
)
770 struct bdi_work
*work
, *ret
= NULL
;
774 list_for_each_entry_rcu(work
, &bdi
->work_list
, list
) {
775 if (!test_and_clear_bit(wb
->nr
, &work
->seen
))
786 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
788 unsigned long expired
;
791 expired
= wb
->last_old_flush
+
792 msecs_to_jiffies(dirty_writeback_interval
* 10);
793 if (time_before(jiffies
, expired
))
796 wb
->last_old_flush
= jiffies
;
797 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
798 global_page_state(NR_UNSTABLE_NFS
) +
799 (inodes_stat
.nr_inodes
- inodes_stat
.nr_unused
);
802 struct wb_writeback_args args
= {
803 .nr_pages
= nr_pages
,
804 .sync_mode
= WB_SYNC_NONE
,
809 return wb_writeback(wb
, &args
);
816 * Retrieve work items and do the writeback they describe
818 long wb_do_writeback(struct bdi_writeback
*wb
, int force_wait
)
820 struct backing_dev_info
*bdi
= wb
->bdi
;
821 struct bdi_work
*work
;
824 while ((work
= get_next_work_item(bdi
, wb
)) != NULL
) {
825 struct wb_writeback_args args
= work
->args
;
828 * Override sync mode, in case we must wait for completion
831 work
->args
.sync_mode
= args
.sync_mode
= WB_SYNC_ALL
;
834 * If this isn't a data integrity operation, just notify
835 * that we have seen this work and we are now starting it.
837 if (args
.sync_mode
== WB_SYNC_NONE
)
838 wb_clear_pending(wb
, work
);
840 wrote
+= wb_writeback(wb
, &args
);
843 * This is a data integrity writeback, so only do the
844 * notification when we have completed the work.
846 if (args
.sync_mode
== WB_SYNC_ALL
)
847 wb_clear_pending(wb
, work
);
851 * Check for periodic writeback, kupdated() style
853 wrote
+= wb_check_old_data_flush(wb
);
859 * Handle writeback of dirty data for the device backed by this bdi. Also
860 * wakes up periodically and does kupdated style flushing.
862 int bdi_writeback_task(struct bdi_writeback
*wb
)
864 unsigned long last_active
= jiffies
;
865 unsigned long wait_jiffies
= -1UL;
868 while (!kthread_should_stop()) {
869 pages_written
= wb_do_writeback(wb
, 0);
872 last_active
= jiffies
;
873 else if (wait_jiffies
!= -1UL) {
874 unsigned long max_idle
;
877 * Longest period of inactivity that we tolerate. If we
878 * see dirty data again later, the task will get
879 * recreated automatically.
881 max_idle
= max(5UL * 60 * HZ
, wait_jiffies
);
882 if (time_after(jiffies
, max_idle
+ last_active
))
886 wait_jiffies
= msecs_to_jiffies(dirty_writeback_interval
* 10);
887 schedule_timeout_interruptible(wait_jiffies
);
895 * Schedule writeback for all backing devices. This does WB_SYNC_NONE
896 * writeback, for integrity writeback see bdi_sync_writeback().
898 static void bdi_writeback_all(struct super_block
*sb
, long nr_pages
)
900 struct wb_writeback_args args
= {
902 .nr_pages
= nr_pages
,
903 .sync_mode
= WB_SYNC_NONE
,
905 struct backing_dev_info
*bdi
;
909 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
910 if (!bdi_has_dirty_io(bdi
))
913 bdi_alloc_queue_work(bdi
, &args
);
920 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
923 void wakeup_flusher_threads(long nr_pages
)
926 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
927 global_page_state(NR_UNSTABLE_NFS
);
928 bdi_writeback_all(NULL
, nr_pages
);
931 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
933 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
934 struct dentry
*dentry
;
935 const char *name
= "?";
937 dentry
= d_find_alias(inode
);
939 spin_lock(&dentry
->d_lock
);
940 name
= (const char *) dentry
->d_name
.name
;
943 "%s(%d): dirtied inode %lu (%s) on %s\n",
944 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
945 name
, inode
->i_sb
->s_id
);
947 spin_unlock(&dentry
->d_lock
);
954 * __mark_inode_dirty - internal function
955 * @inode: inode to mark
956 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
957 * Mark an inode as dirty. Callers should use mark_inode_dirty or
958 * mark_inode_dirty_sync.
960 * Put the inode on the super block's dirty list.
962 * CAREFUL! We mark it dirty unconditionally, but move it onto the
963 * dirty list only if it is hashed or if it refers to a blockdev.
964 * If it was not hashed, it will never be added to the dirty list
965 * even if it is later hashed, as it will have been marked dirty already.
967 * In short, make sure you hash any inodes _before_ you start marking
970 * This function *must* be atomic for the I_DIRTY_PAGES case -
971 * set_page_dirty() is called under spinlock in several places.
973 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
974 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
975 * the kernel-internal blockdev inode represents the dirtying time of the
976 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
977 * page->mapping->host, so the page-dirtying time is recorded in the internal
980 void __mark_inode_dirty(struct inode
*inode
, int flags
)
982 struct super_block
*sb
= inode
->i_sb
;
985 * Don't do this for I_DIRTY_PAGES - that doesn't actually
986 * dirty the inode itself
988 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
989 if (sb
->s_op
->dirty_inode
)
990 sb
->s_op
->dirty_inode(inode
);
994 * make sure that changes are seen by all cpus before we test i_state
999 /* avoid the locking if we can */
1000 if ((inode
->i_state
& flags
) == flags
)
1003 if (unlikely(block_dump
))
1004 block_dump___mark_inode_dirty(inode
);
1006 spin_lock(&inode_lock
);
1007 if ((inode
->i_state
& flags
) != flags
) {
1008 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1010 inode
->i_state
|= flags
;
1013 * If the inode is being synced, just update its dirty state.
1014 * The unlocker will place the inode on the appropriate
1015 * superblock list, based upon its state.
1017 if (inode
->i_state
& I_SYNC
)
1021 * Only add valid (hashed) inodes to the superblock's
1022 * dirty list. Add blockdev inodes as well.
1024 if (!S_ISBLK(inode
->i_mode
)) {
1025 if (hlist_unhashed(&inode
->i_hash
))
1028 if (inode
->i_state
& (I_FREEING
|I_CLEAR
))
1032 * If the inode was already on b_dirty/b_io/b_more_io, don't
1033 * reposition it (that would break b_dirty time-ordering).
1036 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
1037 struct backing_dev_info
*bdi
= wb
->bdi
;
1039 if (bdi_cap_writeback_dirty(bdi
) &&
1040 !test_bit(BDI_registered
, &bdi
->state
)) {
1042 printk(KERN_ERR
"bdi-%s not registered\n",
1046 inode
->dirtied_when
= jiffies
;
1047 list_move(&inode
->i_list
, &wb
->b_dirty
);
1051 spin_unlock(&inode_lock
);
1053 EXPORT_SYMBOL(__mark_inode_dirty
);
1056 * Write out a superblock's list of dirty inodes. A wait will be performed
1057 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1059 * If older_than_this is non-NULL, then only write out inodes which
1060 * had their first dirtying at a time earlier than *older_than_this.
1062 * If we're a pdlfush thread, then implement pdflush collision avoidance
1063 * against the entire list.
1065 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1066 * This function assumes that the blockdev superblock's inodes are backed by
1067 * a variety of queues, so all inodes are searched. For other superblocks,
1068 * assume that all inodes are backed by the same queue.
1070 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1071 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1072 * on the writer throttling path, and we get decent balancing between many
1073 * throttled threads: we don't want them all piling up on inode_sync_wait.
1075 static void wait_sb_inodes(struct super_block
*sb
)
1077 struct inode
*inode
, *old_inode
= NULL
;
1080 * We need to be protected against the filesystem going from
1081 * r/o to r/w or vice versa.
1083 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1085 spin_lock(&inode_lock
);
1088 * Data integrity sync. Must wait for all pages under writeback,
1089 * because there may have been pages dirtied before our sync
1090 * call, but which had writeout started before we write it out.
1091 * In which case, the inode may not be on the dirty list, but
1092 * we still have to wait for that writeout.
1094 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1095 struct address_space
*mapping
;
1097 if (inode
->i_state
& (I_FREEING
|I_CLEAR
|I_WILL_FREE
|I_NEW
))
1099 mapping
= inode
->i_mapping
;
1100 if (mapping
->nrpages
== 0)
1103 spin_unlock(&inode_lock
);
1105 * We hold a reference to 'inode' so it couldn't have
1106 * been removed from s_inodes list while we dropped the
1107 * inode_lock. We cannot iput the inode now as we can
1108 * be holding the last reference and we cannot iput it
1109 * under inode_lock. So we keep the reference and iput
1115 filemap_fdatawait(mapping
);
1119 spin_lock(&inode_lock
);
1121 spin_unlock(&inode_lock
);
1126 * writeback_inodes_sb - writeback dirty inodes from given super_block
1127 * @sb: the superblock
1129 * Start writeback on some inodes on this super_block. No guarantees are made
1130 * on how many (if any) will be written, and this function does not wait
1131 * for IO completion of submitted IO. The number of pages submitted is
1134 void writeback_inodes_sb(struct super_block
*sb
)
1136 unsigned long nr_dirty
= global_page_state(NR_FILE_DIRTY
);
1137 unsigned long nr_unstable
= global_page_state(NR_UNSTABLE_NFS
);
1140 nr_to_write
= nr_dirty
+ nr_unstable
+
1141 (inodes_stat
.nr_inodes
- inodes_stat
.nr_unused
);
1143 bdi_writeback_all(sb
, nr_to_write
);
1145 EXPORT_SYMBOL(writeback_inodes_sb
);
1148 * sync_inodes_sb - sync sb inode pages
1149 * @sb: the superblock
1151 * This function writes and waits on any dirty inode belonging to this
1152 * super_block. The number of pages synced is returned.
1154 void sync_inodes_sb(struct super_block
*sb
)
1156 bdi_sync_writeback(sb
->s_bdi
, sb
);
1159 EXPORT_SYMBOL(sync_inodes_sb
);
1162 * write_inode_now - write an inode to disk
1163 * @inode: inode to write to disk
1164 * @sync: whether the write should be synchronous or not
1166 * This function commits an inode to disk immediately if it is dirty. This is
1167 * primarily needed by knfsd.
1169 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1171 int write_inode_now(struct inode
*inode
, int sync
)
1174 struct writeback_control wbc
= {
1175 .nr_to_write
= LONG_MAX
,
1176 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1178 .range_end
= LLONG_MAX
,
1181 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1182 wbc
.nr_to_write
= 0;
1185 spin_lock(&inode_lock
);
1186 ret
= writeback_single_inode(inode
, &wbc
);
1187 spin_unlock(&inode_lock
);
1189 inode_sync_wait(inode
);
1192 EXPORT_SYMBOL(write_inode_now
);
1195 * sync_inode - write an inode and its pages to disk.
1196 * @inode: the inode to sync
1197 * @wbc: controls the writeback mode
1199 * sync_inode() will write an inode and its pages to disk. It will also
1200 * correctly update the inode on its superblock's dirty inode lists and will
1201 * update inode->i_state.
1203 * The caller must have a ref on the inode.
1205 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1209 spin_lock(&inode_lock
);
1210 ret
= writeback_single_inode(inode
, wbc
);
1211 spin_unlock(&inode_lock
);
1214 EXPORT_SYMBOL(sync_inode
);