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>
29 #include <linux/tracepoint.h>
33 * Passed into wb_writeback(), essentially a subset of writeback_control
35 struct wb_writeback_work
{
37 struct super_block
*sb
;
38 enum writeback_sync_modes sync_mode
;
39 unsigned int for_kupdate
:1;
40 unsigned int range_cyclic
:1;
41 unsigned int for_background
:1;
43 struct list_head list
; /* pending work list */
44 struct completion
*done
; /* set if the caller waits */
48 * Include the creation of the trace points after defining the
49 * wb_writeback_work structure so that the definition remains local to this
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/writeback.h>
56 * We don't actually have pdflush, but this one is exported though /proc...
58 int nr_pdflush_threads
;
61 * writeback_in_progress - determine whether there is writeback in progress
62 * @bdi: the device's backing_dev_info structure.
64 * Determine whether there is writeback waiting to be handled against a
67 int writeback_in_progress(struct backing_dev_info
*bdi
)
69 return test_bit(BDI_writeback_running
, &bdi
->state
);
72 static inline struct backing_dev_info
*inode_to_bdi(struct inode
*inode
)
74 struct super_block
*sb
= inode
->i_sb
;
76 if (strcmp(sb
->s_type
->name
, "bdev") == 0)
77 return inode
->i_mapping
->backing_dev_info
;
82 static inline struct inode
*wb_inode(struct list_head
*head
)
84 return list_entry(head
, struct inode
, i_wb_list
);
87 /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
88 static void bdi_wakeup_flusher(struct backing_dev_info
*bdi
)
91 wake_up_process(bdi
->wb
.task
);
94 * The bdi thread isn't there, wake up the forker thread which
95 * will create and run it.
97 wake_up_process(default_backing_dev_info
.wb
.task
);
101 static void bdi_queue_work(struct backing_dev_info
*bdi
,
102 struct wb_writeback_work
*work
)
104 trace_writeback_queue(bdi
, work
);
106 spin_lock_bh(&bdi
->wb_lock
);
107 list_add_tail(&work
->list
, &bdi
->work_list
);
109 trace_writeback_nothread(bdi
, work
);
110 bdi_wakeup_flusher(bdi
);
111 spin_unlock_bh(&bdi
->wb_lock
);
115 __bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
118 struct wb_writeback_work
*work
;
121 * This is WB_SYNC_NONE writeback, so if allocation fails just
122 * wakeup the thread for old dirty data writeback
124 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
127 trace_writeback_nowork(bdi
);
128 wake_up_process(bdi
->wb
.task
);
133 work
->sync_mode
= WB_SYNC_NONE
;
134 work
->nr_pages
= nr_pages
;
135 work
->range_cyclic
= range_cyclic
;
137 bdi_queue_work(bdi
, work
);
141 * bdi_start_writeback - start writeback
142 * @bdi: the backing device to write from
143 * @nr_pages: the number of pages to write
146 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
147 * started when this function returns, we make no guarantees on
148 * completion. Caller need not hold sb s_umount semaphore.
151 void bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
)
153 __bdi_start_writeback(bdi
, nr_pages
, true);
157 * bdi_start_background_writeback - start background writeback
158 * @bdi: the backing device to write from
161 * This makes sure WB_SYNC_NONE background writeback happens. When
162 * this function returns, it is only guaranteed that for given BDI
163 * some IO is happening if we are over background dirty threshold.
164 * Caller need not hold sb s_umount semaphore.
166 void bdi_start_background_writeback(struct backing_dev_info
*bdi
)
169 * We just wake up the flusher thread. It will perform background
170 * writeback as soon as there is no other work to do.
172 trace_writeback_wake_background(bdi
);
173 spin_lock_bh(&bdi
->wb_lock
);
174 bdi_wakeup_flusher(bdi
);
175 spin_unlock_bh(&bdi
->wb_lock
);
179 * Remove the inode from the writeback list it is on.
181 void inode_wb_list_del(struct inode
*inode
)
183 spin_lock(&inode_wb_list_lock
);
184 list_del_init(&inode
->i_wb_list
);
185 spin_unlock(&inode_wb_list_lock
);
190 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
191 * furthest end of its superblock's dirty-inode list.
193 * Before stamping the inode's ->dirtied_when, we check to see whether it is
194 * already the most-recently-dirtied inode on the b_dirty list. If that is
195 * the case then the inode must have been redirtied while it was being written
196 * out and we don't reset its dirtied_when.
198 static void redirty_tail(struct inode
*inode
)
200 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
202 assert_spin_locked(&inode_wb_list_lock
);
203 if (!list_empty(&wb
->b_dirty
)) {
206 tail
= wb_inode(wb
->b_dirty
.next
);
207 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
208 inode
->dirtied_when
= jiffies
;
210 list_move(&inode
->i_wb_list
, &wb
->b_dirty
);
214 * requeue inode for re-scanning after bdi->b_io list is exhausted.
216 static void requeue_io(struct inode
*inode
)
218 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
220 assert_spin_locked(&inode_wb_list_lock
);
221 list_move(&inode
->i_wb_list
, &wb
->b_more_io
);
224 static void inode_sync_complete(struct inode
*inode
)
227 * Prevent speculative execution through
228 * spin_unlock(&inode_wb_list_lock);
232 wake_up_bit(&inode
->i_state
, __I_SYNC
);
235 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
237 bool ret
= time_after(inode
->dirtied_when
, t
);
240 * For inodes being constantly redirtied, dirtied_when can get stuck.
241 * It _appears_ to be in the future, but is actually in distant past.
242 * This test is necessary to prevent such wrapped-around relative times
243 * from permanently stopping the whole bdi writeback.
245 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
251 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
253 static void move_expired_inodes(struct list_head
*delaying_queue
,
254 struct list_head
*dispatch_queue
,
255 unsigned long *older_than_this
)
258 struct list_head
*pos
, *node
;
259 struct super_block
*sb
= NULL
;
263 while (!list_empty(delaying_queue
)) {
264 inode
= wb_inode(delaying_queue
->prev
);
265 if (older_than_this
&&
266 inode_dirtied_after(inode
, *older_than_this
))
268 if (sb
&& sb
!= inode
->i_sb
)
271 list_move(&inode
->i_wb_list
, &tmp
);
274 /* just one sb in list, splice to dispatch_queue and we're done */
276 list_splice(&tmp
, dispatch_queue
);
280 /* Move inodes from one superblock together */
281 while (!list_empty(&tmp
)) {
282 sb
= wb_inode(tmp
.prev
)->i_sb
;
283 list_for_each_prev_safe(pos
, node
, &tmp
) {
284 inode
= wb_inode(pos
);
285 if (inode
->i_sb
== sb
)
286 list_move(&inode
->i_wb_list
, dispatch_queue
);
292 * Queue all expired dirty inodes for io, eldest first.
294 * newly dirtied b_dirty b_io b_more_io
295 * =============> gf edc BA
297 * newly dirtied b_dirty b_io b_more_io
298 * =============> g fBAedc
300 * +--> dequeue for IO
302 static void queue_io(struct bdi_writeback
*wb
, unsigned long *older_than_this
)
304 assert_spin_locked(&inode_wb_list_lock
);
305 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
306 move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, older_than_this
);
309 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
311 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
))
312 return inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
317 * Wait for writeback on an inode to complete.
319 static void inode_wait_for_writeback(struct inode
*inode
)
321 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
322 wait_queue_head_t
*wqh
;
324 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
325 while (inode
->i_state
& I_SYNC
) {
326 spin_unlock(&inode
->i_lock
);
327 spin_unlock(&inode_wb_list_lock
);
328 __wait_on_bit(wqh
, &wq
, inode_wait
, TASK_UNINTERRUPTIBLE
);
329 spin_lock(&inode_wb_list_lock
);
330 spin_lock(&inode
->i_lock
);
335 * Write out an inode's dirty pages. Called under inode_wb_list_lock and
336 * inode->i_lock. Either the caller has an active reference on the inode or
337 * the inode has I_WILL_FREE set.
339 * If `wait' is set, wait on the writeout.
341 * The whole writeout design is quite complex and fragile. We want to avoid
342 * starvation of particular inodes when others are being redirtied, prevent
346 writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
348 struct address_space
*mapping
= inode
->i_mapping
;
352 assert_spin_locked(&inode_wb_list_lock
);
353 assert_spin_locked(&inode
->i_lock
);
355 if (!atomic_read(&inode
->i_count
))
356 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
358 WARN_ON(inode
->i_state
& I_WILL_FREE
);
360 if (inode
->i_state
& I_SYNC
) {
362 * If this inode is locked for writeback and we are not doing
363 * writeback-for-data-integrity, move it to b_more_io so that
364 * writeback can proceed with the other inodes on s_io.
366 * We'll have another go at writing back this inode when we
367 * completed a full scan of b_io.
369 if (wbc
->sync_mode
!= WB_SYNC_ALL
) {
375 * It's a data-integrity sync. We must wait.
377 inode_wait_for_writeback(inode
);
380 BUG_ON(inode
->i_state
& I_SYNC
);
382 /* Set I_SYNC, reset I_DIRTY_PAGES */
383 inode
->i_state
|= I_SYNC
;
384 inode
->i_state
&= ~I_DIRTY_PAGES
;
385 spin_unlock(&inode
->i_lock
);
386 spin_unlock(&inode_wb_list_lock
);
388 ret
= do_writepages(mapping
, wbc
);
391 * Make sure to wait on the data before writing out the metadata.
392 * This is important for filesystems that modify metadata on data
395 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
396 int err
= filemap_fdatawait(mapping
);
402 * Some filesystems may redirty the inode during the writeback
403 * due to delalloc, clear dirty metadata flags right before
406 spin_lock(&inode
->i_lock
);
407 dirty
= inode
->i_state
& I_DIRTY
;
408 inode
->i_state
&= ~(I_DIRTY_SYNC
| I_DIRTY_DATASYNC
);
409 spin_unlock(&inode
->i_lock
);
410 /* Don't write the inode if only I_DIRTY_PAGES was set */
411 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
412 int err
= write_inode(inode
, wbc
);
417 spin_lock(&inode_wb_list_lock
);
418 spin_lock(&inode
->i_lock
);
419 inode
->i_state
&= ~I_SYNC
;
420 if (!(inode
->i_state
& I_FREEING
)) {
421 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
423 * We didn't write back all the pages. nfs_writepages()
424 * sometimes bales out without doing anything.
426 inode
->i_state
|= I_DIRTY_PAGES
;
427 if (wbc
->nr_to_write
<= 0) {
429 * slice used up: queue for next turn
434 * Writeback blocked by something other than
435 * congestion. Delay the inode for some time to
436 * avoid spinning on the CPU (100% iowait)
437 * retrying writeback of the dirty page/inode
438 * that cannot be performed immediately.
442 } else if (inode
->i_state
& I_DIRTY
) {
444 * Filesystems can dirty the inode during writeback
445 * operations, such as delayed allocation during
446 * submission or metadata updates after data IO
452 * The inode is clean. At this point we either have
453 * a reference to the inode or it's on it's way out.
454 * No need to add it back to the LRU.
456 list_del_init(&inode
->i_wb_list
);
459 inode_sync_complete(inode
);
464 * Write a portion of b_io inodes which belong to @sb.
466 * If @only_this_sb is true, then find and write all such
467 * inodes. Otherwise write only ones which go sequentially
470 * Return 1, if the caller writeback routine should be
471 * interrupted. Otherwise return 0.
473 static int writeback_sb_inodes(struct super_block
*sb
, struct bdi_writeback
*wb
,
474 struct writeback_control
*wbc
, bool only_this_sb
)
476 while (!list_empty(&wb
->b_io
)) {
478 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
480 if (inode
->i_sb
!= sb
) {
483 * We only want to write back data for this
484 * superblock, move all inodes not belonging
485 * to it back onto the dirty list.
492 * The inode belongs to a different superblock.
493 * Bounce back to the caller to unpin this and
494 * pin the next superblock.
500 * Don't bother with new inodes or inodes beeing freed, first
501 * kind does not need peridic writeout yet, and for the latter
502 * kind writeout is handled by the freer.
504 spin_lock(&inode
->i_lock
);
505 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
506 spin_unlock(&inode
->i_lock
);
512 * Was this inode dirtied after sync_sb_inodes was called?
513 * This keeps sync from extra jobs and livelock.
515 if (inode_dirtied_after(inode
, wbc
->wb_start
)) {
516 spin_unlock(&inode
->i_lock
);
522 pages_skipped
= wbc
->pages_skipped
;
523 writeback_single_inode(inode
, wbc
);
524 if (wbc
->pages_skipped
!= pages_skipped
) {
526 * writeback is not making progress due to locked
527 * buffers. Skip this inode for now.
531 spin_unlock(&inode
->i_lock
);
532 spin_unlock(&inode_wb_list_lock
);
535 spin_lock(&inode_wb_list_lock
);
536 if (wbc
->nr_to_write
<= 0) {
540 if (!list_empty(&wb
->b_more_io
))
547 void writeback_inodes_wb(struct bdi_writeback
*wb
,
548 struct writeback_control
*wbc
)
553 wbc
->wb_start
= jiffies
; /* livelock avoidance */
554 spin_lock(&inode_wb_list_lock
);
555 if (!wbc
->for_kupdate
|| list_empty(&wb
->b_io
))
556 queue_io(wb
, wbc
->older_than_this
);
558 while (!list_empty(&wb
->b_io
)) {
559 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
560 struct super_block
*sb
= inode
->i_sb
;
562 if (!grab_super_passive(sb
)) {
566 ret
= writeback_sb_inodes(sb
, wb
, wbc
, false);
572 spin_unlock(&inode_wb_list_lock
);
573 /* Leave any unwritten inodes on b_io */
576 static void __writeback_inodes_sb(struct super_block
*sb
,
577 struct bdi_writeback
*wb
, struct writeback_control
*wbc
)
579 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
581 spin_lock(&inode_wb_list_lock
);
582 if (!wbc
->for_kupdate
|| list_empty(&wb
->b_io
))
583 queue_io(wb
, wbc
->older_than_this
);
584 writeback_sb_inodes(sb
, wb
, wbc
, true);
585 spin_unlock(&inode_wb_list_lock
);
589 * The maximum number of pages to writeout in a single bdi flush/kupdate
590 * operation. We do this so we don't hold I_SYNC against an inode for
591 * enormous amounts of time, which would block a userspace task which has
592 * been forced to throttle against that inode. Also, the code reevaluates
593 * the dirty each time it has written this many pages.
595 #define MAX_WRITEBACK_PAGES 1024
597 static inline bool over_bground_thresh(void)
599 unsigned long background_thresh
, dirty_thresh
;
601 global_dirty_limits(&background_thresh
, &dirty_thresh
);
603 return (global_page_state(NR_FILE_DIRTY
) +
604 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
);
608 * Explicit flushing or periodic writeback of "old" data.
610 * Define "old": the first time one of an inode's pages is dirtied, we mark the
611 * dirtying-time in the inode's address_space. So this periodic writeback code
612 * just walks the superblock inode list, writing back any inodes which are
613 * older than a specific point in time.
615 * Try to run once per dirty_writeback_interval. But if a writeback event
616 * takes longer than a dirty_writeback_interval interval, then leave a
619 * older_than_this takes precedence over nr_to_write. So we'll only write back
620 * all dirty pages if they are all attached to "old" mappings.
622 static long wb_writeback(struct bdi_writeback
*wb
,
623 struct wb_writeback_work
*work
)
625 struct writeback_control wbc
= {
626 .sync_mode
= work
->sync_mode
,
627 .older_than_this
= NULL
,
628 .for_kupdate
= work
->for_kupdate
,
629 .for_background
= work
->for_background
,
630 .range_cyclic
= work
->range_cyclic
,
632 unsigned long oldest_jif
;
637 if (wbc
.for_kupdate
) {
638 wbc
.older_than_this
= &oldest_jif
;
639 oldest_jif
= jiffies
-
640 msecs_to_jiffies(dirty_expire_interval
* 10);
642 if (!wbc
.range_cyclic
) {
644 wbc
.range_end
= LLONG_MAX
;
648 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
649 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
650 * here avoids calling into writeback_inodes_wb() more than once.
652 * The intended call sequence for WB_SYNC_ALL writeback is:
655 * __writeback_inodes_sb() <== called only once
656 * write_cache_pages() <== called once for each inode
657 * (quickly) tag currently dirty pages
658 * (maybe slowly) sync all tagged pages
660 if (wbc
.sync_mode
== WB_SYNC_NONE
)
661 write_chunk
= MAX_WRITEBACK_PAGES
;
663 write_chunk
= LONG_MAX
;
665 wbc
.wb_start
= jiffies
; /* livelock avoidance */
668 * Stop writeback when nr_pages has been consumed
670 if (work
->nr_pages
<= 0)
674 * Background writeout and kupdate-style writeback may
675 * run forever. Stop them if there is other work to do
676 * so that e.g. sync can proceed. They'll be restarted
677 * after the other works are all done.
679 if ((work
->for_background
|| work
->for_kupdate
) &&
680 !list_empty(&wb
->bdi
->work_list
))
684 * For background writeout, stop when we are below the
685 * background dirty threshold
687 if (work
->for_background
&& !over_bground_thresh())
691 wbc
.nr_to_write
= write_chunk
;
692 wbc
.pages_skipped
= 0;
694 trace_wbc_writeback_start(&wbc
, wb
->bdi
);
696 __writeback_inodes_sb(work
->sb
, wb
, &wbc
);
698 writeback_inodes_wb(wb
, &wbc
);
699 trace_wbc_writeback_written(&wbc
, wb
->bdi
);
701 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
702 wrote
+= write_chunk
- wbc
.nr_to_write
;
705 * If we consumed everything, see if we have more
707 if (wbc
.nr_to_write
<= 0)
710 * Didn't write everything and we don't have more IO, bail
715 * Did we write something? Try for more
717 if (wbc
.nr_to_write
< write_chunk
)
720 * Nothing written. Wait for some inode to
721 * become available for writeback. Otherwise
722 * we'll just busyloop.
724 spin_lock(&inode_wb_list_lock
);
725 if (!list_empty(&wb
->b_more_io
)) {
726 inode
= wb_inode(wb
->b_more_io
.prev
);
727 trace_wbc_writeback_wait(&wbc
, wb
->bdi
);
728 spin_lock(&inode
->i_lock
);
729 inode_wait_for_writeback(inode
);
730 spin_unlock(&inode
->i_lock
);
732 spin_unlock(&inode_wb_list_lock
);
739 * Return the next wb_writeback_work struct that hasn't been processed yet.
741 static struct wb_writeback_work
*
742 get_next_work_item(struct backing_dev_info
*bdi
)
744 struct wb_writeback_work
*work
= NULL
;
746 spin_lock_bh(&bdi
->wb_lock
);
747 if (!list_empty(&bdi
->work_list
)) {
748 work
= list_entry(bdi
->work_list
.next
,
749 struct wb_writeback_work
, list
);
750 list_del_init(&work
->list
);
752 spin_unlock_bh(&bdi
->wb_lock
);
757 * Add in the number of potentially dirty inodes, because each inode
758 * write can dirty pagecache in the underlying blockdev.
760 static unsigned long get_nr_dirty_pages(void)
762 return global_page_state(NR_FILE_DIRTY
) +
763 global_page_state(NR_UNSTABLE_NFS
) +
764 get_nr_dirty_inodes();
767 static long wb_check_background_flush(struct bdi_writeback
*wb
)
769 if (over_bground_thresh()) {
771 struct wb_writeback_work work
= {
772 .nr_pages
= LONG_MAX
,
773 .sync_mode
= WB_SYNC_NONE
,
778 return wb_writeback(wb
, &work
);
784 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
786 unsigned long expired
;
790 * When set to zero, disable periodic writeback
792 if (!dirty_writeback_interval
)
795 expired
= wb
->last_old_flush
+
796 msecs_to_jiffies(dirty_writeback_interval
* 10);
797 if (time_before(jiffies
, expired
))
800 wb
->last_old_flush
= jiffies
;
801 nr_pages
= get_nr_dirty_pages();
804 struct wb_writeback_work work
= {
805 .nr_pages
= nr_pages
,
806 .sync_mode
= WB_SYNC_NONE
,
811 return wb_writeback(wb
, &work
);
818 * Retrieve work items and do the writeback they describe
820 long wb_do_writeback(struct bdi_writeback
*wb
, int force_wait
)
822 struct backing_dev_info
*bdi
= wb
->bdi
;
823 struct wb_writeback_work
*work
;
826 set_bit(BDI_writeback_running
, &wb
->bdi
->state
);
827 while ((work
= get_next_work_item(bdi
)) != NULL
) {
829 * Override sync mode, in case we must wait for completion
830 * because this thread is exiting now.
833 work
->sync_mode
= WB_SYNC_ALL
;
835 trace_writeback_exec(bdi
, work
);
837 wrote
+= wb_writeback(wb
, work
);
840 * Notify the caller of completion if this is a synchronous
841 * work item, otherwise just free it.
844 complete(work
->done
);
850 * Check for periodic writeback, kupdated() style
852 wrote
+= wb_check_old_data_flush(wb
);
853 wrote
+= wb_check_background_flush(wb
);
854 clear_bit(BDI_writeback_running
, &wb
->bdi
->state
);
860 * Handle writeback of dirty data for the device backed by this bdi. Also
861 * wakes up periodically and does kupdated style flushing.
863 int bdi_writeback_thread(void *data
)
865 struct bdi_writeback
*wb
= data
;
866 struct backing_dev_info
*bdi
= wb
->bdi
;
869 current
->flags
|= PF_SWAPWRITE
;
871 wb
->last_active
= jiffies
;
874 * Our parent may run at a different priority, just set us to normal
876 set_user_nice(current
, 0);
878 trace_writeback_thread_start(bdi
);
880 while (!kthread_should_stop()) {
882 * Remove own delayed wake-up timer, since we are already awake
883 * and we'll take care of the preriodic write-back.
885 del_timer(&wb
->wakeup_timer
);
887 pages_written
= wb_do_writeback(wb
, 0);
889 trace_writeback_pages_written(pages_written
);
892 wb
->last_active
= jiffies
;
894 set_current_state(TASK_INTERRUPTIBLE
);
895 if (!list_empty(&bdi
->work_list
) || kthread_should_stop()) {
896 __set_current_state(TASK_RUNNING
);
900 if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
901 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval
* 10));
904 * We have nothing to do, so can go sleep without any
905 * timeout and save power. When a work is queued or
906 * something is made dirty - we will be woken up.
914 /* Flush any work that raced with us exiting */
915 if (!list_empty(&bdi
->work_list
))
916 wb_do_writeback(wb
, 1);
918 trace_writeback_thread_stop(bdi
);
924 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
927 void wakeup_flusher_threads(long nr_pages
)
929 struct backing_dev_info
*bdi
;
932 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
933 global_page_state(NR_UNSTABLE_NFS
);
937 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
938 if (!bdi_has_dirty_io(bdi
))
940 __bdi_start_writeback(bdi
, nr_pages
, false);
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 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
985 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
986 * the kernel-internal blockdev inode represents the dirtying time of the
987 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
988 * page->mapping->host, so the page-dirtying time is recorded in the internal
991 void __mark_inode_dirty(struct inode
*inode
, int flags
)
993 struct super_block
*sb
= inode
->i_sb
;
994 struct backing_dev_info
*bdi
= NULL
;
997 * Don't do this for I_DIRTY_PAGES - that doesn't actually
998 * dirty the inode itself
1000 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
1001 if (sb
->s_op
->dirty_inode
)
1002 sb
->s_op
->dirty_inode(inode
, flags
);
1006 * make sure that changes are seen by all cpus before we test i_state
1011 /* avoid the locking if we can */
1012 if ((inode
->i_state
& flags
) == flags
)
1015 if (unlikely(block_dump
))
1016 block_dump___mark_inode_dirty(inode
);
1018 spin_lock(&inode
->i_lock
);
1019 if ((inode
->i_state
& flags
) != flags
) {
1020 const int was_dirty
= inode
->i_state
& I_DIRTY
;
1022 inode
->i_state
|= flags
;
1025 * If the inode is being synced, just update its dirty state.
1026 * The unlocker will place the inode on the appropriate
1027 * superblock list, based upon its state.
1029 if (inode
->i_state
& I_SYNC
)
1030 goto out_unlock_inode
;
1033 * Only add valid (hashed) inodes to the superblock's
1034 * dirty list. Add blockdev inodes as well.
1036 if (!S_ISBLK(inode
->i_mode
)) {
1037 if (inode_unhashed(inode
))
1038 goto out_unlock_inode
;
1040 if (inode
->i_state
& I_FREEING
)
1041 goto out_unlock_inode
;
1044 * If the inode was already on b_dirty/b_io/b_more_io, don't
1045 * reposition it (that would break b_dirty time-ordering).
1048 bool wakeup_bdi
= false;
1049 bdi
= inode_to_bdi(inode
);
1051 if (bdi_cap_writeback_dirty(bdi
)) {
1052 WARN(!test_bit(BDI_registered
, &bdi
->state
),
1053 "bdi-%s not registered\n", bdi
->name
);
1056 * If this is the first dirty inode for this
1057 * bdi, we have to wake-up the corresponding
1058 * bdi thread to make sure background
1059 * write-back happens later.
1061 if (!wb_has_dirty_io(&bdi
->wb
))
1065 spin_unlock(&inode
->i_lock
);
1066 spin_lock(&inode_wb_list_lock
);
1067 inode
->dirtied_when
= jiffies
;
1068 list_move(&inode
->i_wb_list
, &bdi
->wb
.b_dirty
);
1069 spin_unlock(&inode_wb_list_lock
);
1072 bdi_wakeup_thread_delayed(bdi
);
1077 spin_unlock(&inode
->i_lock
);
1080 EXPORT_SYMBOL(__mark_inode_dirty
);
1083 * Write out a superblock's list of dirty inodes. A wait will be performed
1084 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1086 * If older_than_this is non-NULL, then only write out inodes which
1087 * had their first dirtying at a time earlier than *older_than_this.
1089 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1090 * This function assumes that the blockdev superblock's inodes are backed by
1091 * a variety of queues, so all inodes are searched. For other superblocks,
1092 * assume that all inodes are backed by the same queue.
1094 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1095 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1096 * on the writer throttling path, and we get decent balancing between many
1097 * throttled threads: we don't want them all piling up on inode_sync_wait.
1099 static void wait_sb_inodes(struct super_block
*sb
)
1101 struct inode
*inode
, *old_inode
= NULL
;
1104 * We need to be protected against the filesystem going from
1105 * r/o to r/w or vice versa.
1107 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1109 spin_lock(&inode_sb_list_lock
);
1112 * Data integrity sync. Must wait for all pages under writeback,
1113 * because there may have been pages dirtied before our sync
1114 * call, but which had writeout started before we write it out.
1115 * In which case, the inode may not be on the dirty list, but
1116 * we still have to wait for that writeout.
1118 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1119 struct address_space
*mapping
= inode
->i_mapping
;
1121 spin_lock(&inode
->i_lock
);
1122 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
1123 (mapping
->nrpages
== 0)) {
1124 spin_unlock(&inode
->i_lock
);
1128 spin_unlock(&inode
->i_lock
);
1129 spin_unlock(&inode_sb_list_lock
);
1132 * We hold a reference to 'inode' so it couldn't have been
1133 * removed from s_inodes list while we dropped the
1134 * inode_sb_list_lock. We cannot iput the inode now as we can
1135 * be holding the last reference and we cannot iput it under
1136 * inode_sb_list_lock. So we keep the reference and iput it
1142 filemap_fdatawait(mapping
);
1146 spin_lock(&inode_sb_list_lock
);
1148 spin_unlock(&inode_sb_list_lock
);
1153 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1154 * @sb: the superblock
1155 * @nr: the number of pages to write
1157 * Start writeback on some inodes on this super_block. No guarantees are made
1158 * on how many (if any) will be written, and this function does not wait
1159 * for IO completion of submitted IO.
1161 void writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
)
1163 DECLARE_COMPLETION_ONSTACK(done
);
1164 struct wb_writeback_work work
= {
1166 .sync_mode
= WB_SYNC_NONE
,
1171 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1172 bdi_queue_work(sb
->s_bdi
, &work
);
1173 wait_for_completion(&done
);
1175 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1178 * writeback_inodes_sb - writeback dirty inodes from given super_block
1179 * @sb: the superblock
1181 * Start writeback on some inodes on this super_block. No guarantees are made
1182 * on how many (if any) will be written, and this function does not wait
1183 * for IO completion of submitted IO.
1185 void writeback_inodes_sb(struct super_block
*sb
)
1187 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages());
1189 EXPORT_SYMBOL(writeback_inodes_sb
);
1192 * writeback_inodes_sb_if_idle - start writeback if none underway
1193 * @sb: the superblock
1195 * Invoke writeback_inodes_sb if no writeback is currently underway.
1196 * Returns 1 if writeback was started, 0 if not.
1198 int writeback_inodes_sb_if_idle(struct super_block
*sb
)
1200 if (!writeback_in_progress(sb
->s_bdi
)) {
1201 down_read(&sb
->s_umount
);
1202 writeback_inodes_sb(sb
);
1203 up_read(&sb
->s_umount
);
1208 EXPORT_SYMBOL(writeback_inodes_sb_if_idle
);
1211 * writeback_inodes_sb_if_idle - start writeback if none underway
1212 * @sb: the superblock
1213 * @nr: the number of pages to write
1215 * Invoke writeback_inodes_sb if no writeback is currently underway.
1216 * Returns 1 if writeback was started, 0 if not.
1218 int writeback_inodes_sb_nr_if_idle(struct super_block
*sb
,
1221 if (!writeback_in_progress(sb
->s_bdi
)) {
1222 down_read(&sb
->s_umount
);
1223 writeback_inodes_sb_nr(sb
, nr
);
1224 up_read(&sb
->s_umount
);
1229 EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle
);
1232 * sync_inodes_sb - sync sb inode pages
1233 * @sb: the superblock
1235 * This function writes and waits on any dirty inode belonging to this
1238 void sync_inodes_sb(struct super_block
*sb
)
1240 DECLARE_COMPLETION_ONSTACK(done
);
1241 struct wb_writeback_work work
= {
1243 .sync_mode
= WB_SYNC_ALL
,
1244 .nr_pages
= LONG_MAX
,
1249 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1251 bdi_queue_work(sb
->s_bdi
, &work
);
1252 wait_for_completion(&done
);
1256 EXPORT_SYMBOL(sync_inodes_sb
);
1259 * write_inode_now - write an inode to disk
1260 * @inode: inode to write to disk
1261 * @sync: whether the write should be synchronous or not
1263 * This function commits an inode to disk immediately if it is dirty. This is
1264 * primarily needed by knfsd.
1266 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1268 int write_inode_now(struct inode
*inode
, int sync
)
1271 struct writeback_control wbc
= {
1272 .nr_to_write
= LONG_MAX
,
1273 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1275 .range_end
= LLONG_MAX
,
1278 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1279 wbc
.nr_to_write
= 0;
1282 spin_lock(&inode_wb_list_lock
);
1283 spin_lock(&inode
->i_lock
);
1284 ret
= writeback_single_inode(inode
, &wbc
);
1285 spin_unlock(&inode
->i_lock
);
1286 spin_unlock(&inode_wb_list_lock
);
1288 inode_sync_wait(inode
);
1291 EXPORT_SYMBOL(write_inode_now
);
1294 * sync_inode - write an inode and its pages to disk.
1295 * @inode: the inode to sync
1296 * @wbc: controls the writeback mode
1298 * sync_inode() will write an inode and its pages to disk. It will also
1299 * correctly update the inode on its superblock's dirty inode lists and will
1300 * update inode->i_state.
1302 * The caller must have a ref on the inode.
1304 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1308 spin_lock(&inode_wb_list_lock
);
1309 spin_lock(&inode
->i_lock
);
1310 ret
= writeback_single_inode(inode
, wbc
);
1311 spin_unlock(&inode
->i_lock
);
1312 spin_unlock(&inode_wb_list_lock
);
1315 EXPORT_SYMBOL(sync_inode
);
1318 * sync_inode_metadata - write an inode to disk
1319 * @inode: the inode to sync
1320 * @wait: wait for I/O to complete.
1322 * Write an inode to disk and adjust its dirty state after completion.
1324 * Note: only writes the actual inode, no associated data or other metadata.
1326 int sync_inode_metadata(struct inode
*inode
, int wait
)
1328 struct writeback_control wbc
= {
1329 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1330 .nr_to_write
= 0, /* metadata-only */
1333 return sync_inode(inode
, &wbc
);
1335 EXPORT_SYMBOL(sync_inode_metadata
);