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 void bdi_queue_work(struct backing_dev_info
*bdi
,
83 struct wb_writeback_work
*work
)
85 trace_writeback_queue(bdi
, work
);
87 spin_lock_bh(&bdi
->wb_lock
);
88 list_add_tail(&work
->list
, &bdi
->work_list
);
90 wake_up_process(bdi
->wb
.task
);
93 * The bdi thread isn't there, wake up the forker thread which
94 * will create and run it.
96 trace_writeback_nothread(bdi
, work
);
97 wake_up_process(default_backing_dev_info
.wb
.task
);
99 spin_unlock_bh(&bdi
->wb_lock
);
103 __bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
104 bool range_cyclic
, bool for_background
)
106 struct wb_writeback_work
*work
;
109 * This is WB_SYNC_NONE writeback, so if allocation fails just
110 * wakeup the thread for old dirty data writeback
112 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
115 trace_writeback_nowork(bdi
);
116 wake_up_process(bdi
->wb
.task
);
121 work
->sync_mode
= WB_SYNC_NONE
;
122 work
->nr_pages
= nr_pages
;
123 work
->range_cyclic
= range_cyclic
;
124 work
->for_background
= for_background
;
126 bdi_queue_work(bdi
, work
);
130 * bdi_start_writeback - start writeback
131 * @bdi: the backing device to write from
132 * @nr_pages: the number of pages to write
135 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
136 * started when this function returns, we make no guarentees on
137 * completion. Caller need not hold sb s_umount semaphore.
140 void bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
)
142 __bdi_start_writeback(bdi
, nr_pages
, true, false);
146 * bdi_start_background_writeback - start background writeback
147 * @bdi: the backing device to write from
150 * This does WB_SYNC_NONE background writeback. The IO is only
151 * started when this function returns, we make no guarentees on
152 * completion. Caller need not hold sb s_umount semaphore.
154 void bdi_start_background_writeback(struct backing_dev_info
*bdi
)
156 __bdi_start_writeback(bdi
, LONG_MAX
, true, true);
160 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
161 * furthest end of its superblock's dirty-inode list.
163 * Before stamping the inode's ->dirtied_when, we check to see whether it is
164 * already the most-recently-dirtied inode on the b_dirty list. If that is
165 * the case then the inode must have been redirtied while it was being written
166 * out and we don't reset its dirtied_when.
168 static void redirty_tail(struct inode
*inode
)
170 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
172 if (!list_empty(&wb
->b_dirty
)) {
175 tail
= list_entry(wb
->b_dirty
.next
, struct inode
, i_list
);
176 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
177 inode
->dirtied_when
= jiffies
;
179 list_move(&inode
->i_list
, &wb
->b_dirty
);
183 * requeue inode for re-scanning after bdi->b_io list is exhausted.
185 static void requeue_io(struct inode
*inode
)
187 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
189 list_move(&inode
->i_list
, &wb
->b_more_io
);
192 static void inode_sync_complete(struct inode
*inode
)
195 * Prevent speculative execution through spin_unlock(&inode_lock);
198 wake_up_bit(&inode
->i_state
, __I_SYNC
);
201 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
203 bool ret
= time_after(inode
->dirtied_when
, t
);
206 * For inodes being constantly redirtied, dirtied_when can get stuck.
207 * It _appears_ to be in the future, but is actually in distant past.
208 * This test is necessary to prevent such wrapped-around relative times
209 * from permanently stopping the whole bdi writeback.
211 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
217 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
219 static void move_expired_inodes(struct list_head
*delaying_queue
,
220 struct list_head
*dispatch_queue
,
221 unsigned long *older_than_this
)
224 struct list_head
*pos
, *node
;
225 struct super_block
*sb
= NULL
;
229 while (!list_empty(delaying_queue
)) {
230 inode
= list_entry(delaying_queue
->prev
, struct inode
, i_list
);
231 if (older_than_this
&&
232 inode_dirtied_after(inode
, *older_than_this
))
234 if (sb
&& sb
!= inode
->i_sb
)
237 list_move(&inode
->i_list
, &tmp
);
240 /* just one sb in list, splice to dispatch_queue and we're done */
242 list_splice(&tmp
, dispatch_queue
);
246 /* Move inodes from one superblock together */
247 while (!list_empty(&tmp
)) {
248 inode
= list_entry(tmp
.prev
, struct inode
, i_list
);
250 list_for_each_prev_safe(pos
, node
, &tmp
) {
251 inode
= list_entry(pos
, struct inode
, i_list
);
252 if (inode
->i_sb
== sb
)
253 list_move(&inode
->i_list
, dispatch_queue
);
259 * Queue all expired dirty inodes for io, eldest first.
261 * newly dirtied b_dirty b_io b_more_io
262 * =============> gf edc BA
264 * newly dirtied b_dirty b_io b_more_io
265 * =============> g fBAedc
267 * +--> dequeue for IO
269 static void queue_io(struct bdi_writeback
*wb
, unsigned long *older_than_this
)
271 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
272 move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, older_than_this
);
275 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
277 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
))
278 return inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
283 * Wait for writeback on an inode to complete.
285 static void inode_wait_for_writeback(struct inode
*inode
)
287 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
288 wait_queue_head_t
*wqh
;
290 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
291 while (inode
->i_state
& I_SYNC
) {
292 spin_unlock(&inode_lock
);
293 __wait_on_bit(wqh
, &wq
, inode_wait
, TASK_UNINTERRUPTIBLE
);
294 spin_lock(&inode_lock
);
299 * Write out an inode's dirty pages. Called under inode_lock. Either the
300 * caller has ref on the inode (either via __iget or via syscall against an fd)
301 * or the inode has I_WILL_FREE set (via generic_forget_inode)
303 * If `wait' is set, wait on the writeout.
305 * The whole writeout design is quite complex and fragile. We want to avoid
306 * starvation of particular inodes when others are being redirtied, prevent
309 * Called under inode_lock.
312 writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
314 struct address_space
*mapping
= inode
->i_mapping
;
318 if (!atomic_read(&inode
->i_count
))
319 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
321 WARN_ON(inode
->i_state
& I_WILL_FREE
);
323 if (inode
->i_state
& I_SYNC
) {
325 * If this inode is locked for writeback and we are not doing
326 * writeback-for-data-integrity, move it to b_more_io so that
327 * writeback can proceed with the other inodes on s_io.
329 * We'll have another go at writing back this inode when we
330 * completed a full scan of b_io.
332 if (wbc
->sync_mode
!= WB_SYNC_ALL
) {
338 * It's a data-integrity sync. We must wait.
340 inode_wait_for_writeback(inode
);
343 BUG_ON(inode
->i_state
& I_SYNC
);
345 /* Set I_SYNC, reset I_DIRTY_PAGES */
346 inode
->i_state
|= I_SYNC
;
347 inode
->i_state
&= ~I_DIRTY_PAGES
;
348 spin_unlock(&inode_lock
);
350 ret
= do_writepages(mapping
, wbc
);
353 * Make sure to wait on the data before writing out the metadata.
354 * This is important for filesystems that modify metadata on data
357 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
358 int err
= filemap_fdatawait(mapping
);
364 * Some filesystems may redirty the inode during the writeback
365 * due to delalloc, clear dirty metadata flags right before
368 spin_lock(&inode_lock
);
369 dirty
= inode
->i_state
& I_DIRTY
;
370 inode
->i_state
&= ~(I_DIRTY_SYNC
| I_DIRTY_DATASYNC
);
371 spin_unlock(&inode_lock
);
372 /* Don't write the inode if only I_DIRTY_PAGES was set */
373 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
374 int err
= write_inode(inode
, wbc
);
379 spin_lock(&inode_lock
);
380 inode
->i_state
&= ~I_SYNC
;
381 if (!(inode
->i_state
& I_FREEING
)) {
382 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
384 * We didn't write back all the pages. nfs_writepages()
385 * sometimes bales out without doing anything.
387 inode
->i_state
|= I_DIRTY_PAGES
;
388 if (wbc
->nr_to_write
<= 0) {
390 * slice used up: queue for next turn
395 * Writeback blocked by something other than
396 * congestion. Delay the inode for some time to
397 * avoid spinning on the CPU (100% iowait)
398 * retrying writeback of the dirty page/inode
399 * that cannot be performed immediately.
403 } else if (inode
->i_state
& I_DIRTY
) {
405 * Filesystems can dirty the inode during writeback
406 * operations, such as delayed allocation during
407 * submission or metadata updates after data IO
413 * The inode is clean. At this point we either have
414 * a reference to the inode or it's on it's way out.
415 * No need to add it back to the LRU.
417 list_del_init(&inode
->i_list
);
420 inode_sync_complete(inode
);
425 * For background writeback the caller does not have the sb pinned
426 * before calling writeback. So make sure that we do pin it, so it doesn't
427 * go away while we are writing inodes from it.
429 static bool pin_sb_for_writeback(struct super_block
*sb
)
432 if (list_empty(&sb
->s_instances
)) {
433 spin_unlock(&sb_lock
);
438 spin_unlock(&sb_lock
);
440 if (down_read_trylock(&sb
->s_umount
)) {
443 up_read(&sb
->s_umount
);
451 * Write a portion of b_io inodes which belong to @sb.
453 * If @only_this_sb is true, then find and write all such
454 * inodes. Otherwise write only ones which go sequentially
457 * Return 1, if the caller writeback routine should be
458 * interrupted. Otherwise return 0.
460 static int writeback_sb_inodes(struct super_block
*sb
, struct bdi_writeback
*wb
,
461 struct writeback_control
*wbc
, bool only_this_sb
)
463 while (!list_empty(&wb
->b_io
)) {
465 struct inode
*inode
= list_entry(wb
->b_io
.prev
,
466 struct inode
, i_list
);
468 if (inode
->i_sb
!= sb
) {
471 * We only want to write back data for this
472 * superblock, move all inodes not belonging
473 * to it back onto the dirty list.
480 * The inode belongs to a different superblock.
481 * Bounce back to the caller to unpin this and
482 * pin the next superblock.
487 if (inode
->i_state
& (I_NEW
| I_WILL_FREE
)) {
492 * Was this inode dirtied after sync_sb_inodes was called?
493 * This keeps sync from extra jobs and livelock.
495 if (inode_dirtied_after(inode
, wbc
->wb_start
))
498 BUG_ON(inode
->i_state
& I_FREEING
);
500 pages_skipped
= wbc
->pages_skipped
;
501 writeback_single_inode(inode
, wbc
);
502 if (wbc
->pages_skipped
!= pages_skipped
) {
504 * writeback is not making progress due to locked
505 * buffers. Skip this inode for now.
509 spin_unlock(&inode_lock
);
512 spin_lock(&inode_lock
);
513 if (wbc
->nr_to_write
<= 0) {
517 if (!list_empty(&wb
->b_more_io
))
524 void writeback_inodes_wb(struct bdi_writeback
*wb
,
525 struct writeback_control
*wbc
)
530 wbc
->wb_start
= jiffies
; /* livelock avoidance */
531 spin_lock(&inode_lock
);
532 if (!wbc
->for_kupdate
|| list_empty(&wb
->b_io
))
533 queue_io(wb
, wbc
->older_than_this
);
535 while (!list_empty(&wb
->b_io
)) {
536 struct inode
*inode
= list_entry(wb
->b_io
.prev
,
537 struct inode
, i_list
);
538 struct super_block
*sb
= inode
->i_sb
;
540 if (!pin_sb_for_writeback(sb
)) {
544 ret
= writeback_sb_inodes(sb
, wb
, wbc
, false);
550 spin_unlock(&inode_lock
);
551 /* Leave any unwritten inodes on b_io */
554 static void __writeback_inodes_sb(struct super_block
*sb
,
555 struct bdi_writeback
*wb
, struct writeback_control
*wbc
)
557 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
559 spin_lock(&inode_lock
);
560 if (!wbc
->for_kupdate
|| list_empty(&wb
->b_io
))
561 queue_io(wb
, wbc
->older_than_this
);
562 writeback_sb_inodes(sb
, wb
, wbc
, true);
563 spin_unlock(&inode_lock
);
567 * The maximum number of pages to writeout in a single bdi flush/kupdate
568 * operation. We do this so we don't hold I_SYNC against an inode for
569 * enormous amounts of time, which would block a userspace task which has
570 * been forced to throttle against that inode. Also, the code reevaluates
571 * the dirty each time it has written this many pages.
573 #define MAX_WRITEBACK_PAGES 1024
575 static inline bool over_bground_thresh(void)
577 unsigned long background_thresh
, dirty_thresh
;
579 global_dirty_limits(&background_thresh
, &dirty_thresh
);
581 return (global_page_state(NR_FILE_DIRTY
) +
582 global_page_state(NR_UNSTABLE_NFS
) >= background_thresh
);
586 * Explicit flushing or periodic writeback of "old" data.
588 * Define "old": the first time one of an inode's pages is dirtied, we mark the
589 * dirtying-time in the inode's address_space. So this periodic writeback code
590 * just walks the superblock inode list, writing back any inodes which are
591 * older than a specific point in time.
593 * Try to run once per dirty_writeback_interval. But if a writeback event
594 * takes longer than a dirty_writeback_interval interval, then leave a
597 * older_than_this takes precedence over nr_to_write. So we'll only write back
598 * all dirty pages if they are all attached to "old" mappings.
600 static long wb_writeback(struct bdi_writeback
*wb
,
601 struct wb_writeback_work
*work
)
603 struct writeback_control wbc
= {
604 .sync_mode
= work
->sync_mode
,
605 .older_than_this
= NULL
,
606 .for_kupdate
= work
->for_kupdate
,
607 .for_background
= work
->for_background
,
608 .range_cyclic
= work
->range_cyclic
,
610 unsigned long oldest_jif
;
614 if (wbc
.for_kupdate
) {
615 wbc
.older_than_this
= &oldest_jif
;
616 oldest_jif
= jiffies
-
617 msecs_to_jiffies(dirty_expire_interval
* 10);
619 if (!wbc
.range_cyclic
) {
621 wbc
.range_end
= LLONG_MAX
;
624 wbc
.wb_start
= jiffies
; /* livelock avoidance */
627 * Stop writeback when nr_pages has been consumed
629 if (work
->nr_pages
<= 0)
633 * For background writeout, stop when we are below the
634 * background dirty threshold
636 if (work
->for_background
&& !over_bground_thresh())
640 wbc
.nr_to_write
= MAX_WRITEBACK_PAGES
;
641 wbc
.pages_skipped
= 0;
643 trace_wbc_writeback_start(&wbc
, wb
->bdi
);
645 __writeback_inodes_sb(work
->sb
, wb
, &wbc
);
647 writeback_inodes_wb(wb
, &wbc
);
648 trace_wbc_writeback_written(&wbc
, wb
->bdi
);
650 work
->nr_pages
-= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
651 wrote
+= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
654 * If we consumed everything, see if we have more
656 if (wbc
.nr_to_write
<= 0)
659 * Didn't write everything and we don't have more IO, bail
664 * Did we write something? Try for more
666 if (wbc
.nr_to_write
< MAX_WRITEBACK_PAGES
)
669 * Nothing written. Wait for some inode to
670 * become available for writeback. Otherwise
671 * we'll just busyloop.
673 spin_lock(&inode_lock
);
674 if (!list_empty(&wb
->b_more_io
)) {
675 inode
= list_entry(wb
->b_more_io
.prev
,
676 struct inode
, i_list
);
677 trace_wbc_writeback_wait(&wbc
, wb
->bdi
);
678 inode_wait_for_writeback(inode
);
680 spin_unlock(&inode_lock
);
687 * Return the next wb_writeback_work struct that hasn't been processed yet.
689 static struct wb_writeback_work
*
690 get_next_work_item(struct backing_dev_info
*bdi
)
692 struct wb_writeback_work
*work
= NULL
;
694 spin_lock_bh(&bdi
->wb_lock
);
695 if (!list_empty(&bdi
->work_list
)) {
696 work
= list_entry(bdi
->work_list
.next
,
697 struct wb_writeback_work
, list
);
698 list_del_init(&work
->list
);
700 spin_unlock_bh(&bdi
->wb_lock
);
704 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
706 unsigned long expired
;
710 * When set to zero, disable periodic writeback
712 if (!dirty_writeback_interval
)
715 expired
= wb
->last_old_flush
+
716 msecs_to_jiffies(dirty_writeback_interval
* 10);
717 if (time_before(jiffies
, expired
))
720 wb
->last_old_flush
= jiffies
;
721 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
722 global_page_state(NR_UNSTABLE_NFS
) +
723 get_nr_dirty_inodes();
726 struct wb_writeback_work work
= {
727 .nr_pages
= nr_pages
,
728 .sync_mode
= WB_SYNC_NONE
,
733 return wb_writeback(wb
, &work
);
740 * Retrieve work items and do the writeback they describe
742 long wb_do_writeback(struct bdi_writeback
*wb
, int force_wait
)
744 struct backing_dev_info
*bdi
= wb
->bdi
;
745 struct wb_writeback_work
*work
;
748 set_bit(BDI_writeback_running
, &wb
->bdi
->state
);
749 while ((work
= get_next_work_item(bdi
)) != NULL
) {
751 * Override sync mode, in case we must wait for completion
752 * because this thread is exiting now.
755 work
->sync_mode
= WB_SYNC_ALL
;
757 trace_writeback_exec(bdi
, work
);
759 wrote
+= wb_writeback(wb
, work
);
762 * Notify the caller of completion if this is a synchronous
763 * work item, otherwise just free it.
766 complete(work
->done
);
772 * Check for periodic writeback, kupdated() style
774 wrote
+= wb_check_old_data_flush(wb
);
775 clear_bit(BDI_writeback_running
, &wb
->bdi
->state
);
781 * Handle writeback of dirty data for the device backed by this bdi. Also
782 * wakes up periodically and does kupdated style flushing.
784 int bdi_writeback_thread(void *data
)
786 struct bdi_writeback
*wb
= data
;
787 struct backing_dev_info
*bdi
= wb
->bdi
;
790 current
->flags
|= PF_FLUSHER
| PF_SWAPWRITE
;
792 wb
->last_active
= jiffies
;
795 * Our parent may run at a different priority, just set us to normal
797 set_user_nice(current
, 0);
799 trace_writeback_thread_start(bdi
);
801 while (!kthread_should_stop()) {
803 * Remove own delayed wake-up timer, since we are already awake
804 * and we'll take care of the preriodic write-back.
806 del_timer(&wb
->wakeup_timer
);
808 pages_written
= wb_do_writeback(wb
, 0);
810 trace_writeback_pages_written(pages_written
);
813 wb
->last_active
= jiffies
;
815 set_current_state(TASK_INTERRUPTIBLE
);
816 if (!list_empty(&bdi
->work_list
) || kthread_should_stop()) {
817 __set_current_state(TASK_RUNNING
);
821 if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
822 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval
* 10));
825 * We have nothing to do, so can go sleep without any
826 * timeout and save power. When a work is queued or
827 * something is made dirty - we will be woken up.
835 /* Flush any work that raced with us exiting */
836 if (!list_empty(&bdi
->work_list
))
837 wb_do_writeback(wb
, 1);
839 trace_writeback_thread_stop(bdi
);
845 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
848 void wakeup_flusher_threads(long nr_pages
)
850 struct backing_dev_info
*bdi
;
853 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
854 global_page_state(NR_UNSTABLE_NFS
);
858 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
859 if (!bdi_has_dirty_io(bdi
))
861 __bdi_start_writeback(bdi
, nr_pages
, false, false);
866 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
868 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
869 struct dentry
*dentry
;
870 const char *name
= "?";
872 dentry
= d_find_alias(inode
);
874 spin_lock(&dentry
->d_lock
);
875 name
= (const char *) dentry
->d_name
.name
;
878 "%s(%d): dirtied inode %lu (%s) on %s\n",
879 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
880 name
, inode
->i_sb
->s_id
);
882 spin_unlock(&dentry
->d_lock
);
889 * __mark_inode_dirty - internal function
890 * @inode: inode to mark
891 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
892 * Mark an inode as dirty. Callers should use mark_inode_dirty or
893 * mark_inode_dirty_sync.
895 * Put the inode on the super block's dirty list.
897 * CAREFUL! We mark it dirty unconditionally, but move it onto the
898 * dirty list only if it is hashed or if it refers to a blockdev.
899 * If it was not hashed, it will never be added to the dirty list
900 * even if it is later hashed, as it will have been marked dirty already.
902 * In short, make sure you hash any inodes _before_ you start marking
905 * This function *must* be atomic for the I_DIRTY_PAGES case -
906 * set_page_dirty() is called under spinlock in several places.
908 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
909 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
910 * the kernel-internal blockdev inode represents the dirtying time of the
911 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
912 * page->mapping->host, so the page-dirtying time is recorded in the internal
915 void __mark_inode_dirty(struct inode
*inode
, int flags
)
917 struct super_block
*sb
= inode
->i_sb
;
918 struct backing_dev_info
*bdi
= NULL
;
919 bool wakeup_bdi
= false;
922 * Don't do this for I_DIRTY_PAGES - that doesn't actually
923 * dirty the inode itself
925 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
926 if (sb
->s_op
->dirty_inode
)
927 sb
->s_op
->dirty_inode(inode
);
931 * make sure that changes are seen by all cpus before we test i_state
936 /* avoid the locking if we can */
937 if ((inode
->i_state
& flags
) == flags
)
940 if (unlikely(block_dump
))
941 block_dump___mark_inode_dirty(inode
);
943 spin_lock(&inode_lock
);
944 if ((inode
->i_state
& flags
) != flags
) {
945 const int was_dirty
= inode
->i_state
& I_DIRTY
;
947 inode
->i_state
|= flags
;
950 * If the inode is being synced, just update its dirty state.
951 * The unlocker will place the inode on the appropriate
952 * superblock list, based upon its state.
954 if (inode
->i_state
& I_SYNC
)
958 * Only add valid (hashed) inodes to the superblock's
959 * dirty list. Add blockdev inodes as well.
961 if (!S_ISBLK(inode
->i_mode
)) {
962 if (inode_unhashed(inode
))
965 if (inode
->i_state
& I_FREEING
)
969 * If the inode was already on b_dirty/b_io/b_more_io, don't
970 * reposition it (that would break b_dirty time-ordering).
973 bdi
= inode_to_bdi(inode
);
975 if (bdi_cap_writeback_dirty(bdi
)) {
976 WARN(!test_bit(BDI_registered
, &bdi
->state
),
977 "bdi-%s not registered\n", bdi
->name
);
980 * If this is the first dirty inode for this
981 * bdi, we have to wake-up the corresponding
982 * bdi thread to make sure background
983 * write-back happens later.
985 if (!wb_has_dirty_io(&bdi
->wb
))
989 inode
->dirtied_when
= jiffies
;
990 list_move(&inode
->i_list
, &bdi
->wb
.b_dirty
);
994 spin_unlock(&inode_lock
);
997 bdi_wakeup_thread_delayed(bdi
);
999 EXPORT_SYMBOL(__mark_inode_dirty
);
1002 * Write out a superblock's list of dirty inodes. A wait will be performed
1003 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1005 * If older_than_this is non-NULL, then only write out inodes which
1006 * had their first dirtying at a time earlier than *older_than_this.
1008 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1009 * This function assumes that the blockdev superblock's inodes are backed by
1010 * a variety of queues, so all inodes are searched. For other superblocks,
1011 * assume that all inodes are backed by the same queue.
1013 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1014 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1015 * on the writer throttling path, and we get decent balancing between many
1016 * throttled threads: we don't want them all piling up on inode_sync_wait.
1018 static void wait_sb_inodes(struct super_block
*sb
)
1020 struct inode
*inode
, *old_inode
= NULL
;
1023 * We need to be protected against the filesystem going from
1024 * r/o to r/w or vice versa.
1026 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1028 spin_lock(&inode_lock
);
1031 * Data integrity sync. Must wait for all pages under writeback,
1032 * because there may have been pages dirtied before our sync
1033 * call, but which had writeout started before we write it out.
1034 * In which case, the inode may not be on the dirty list, but
1035 * we still have to wait for that writeout.
1037 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1038 struct address_space
*mapping
;
1040 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
))
1042 mapping
= inode
->i_mapping
;
1043 if (mapping
->nrpages
== 0)
1046 spin_unlock(&inode_lock
);
1048 * We hold a reference to 'inode' so it couldn't have
1049 * been removed from s_inodes list while we dropped the
1050 * inode_lock. We cannot iput the inode now as we can
1051 * be holding the last reference and we cannot iput it
1052 * under inode_lock. So we keep the reference and iput
1058 filemap_fdatawait(mapping
);
1062 spin_lock(&inode_lock
);
1064 spin_unlock(&inode_lock
);
1069 * writeback_inodes_sb - writeback dirty inodes from given super_block
1070 * @sb: the superblock
1072 * Start writeback on some inodes on this super_block. No guarantees are made
1073 * on how many (if any) will be written, and this function does not wait
1074 * for IO completion of submitted IO. The number of pages submitted is
1077 void writeback_inodes_sb(struct super_block
*sb
)
1079 unsigned long nr_dirty
= global_page_state(NR_FILE_DIRTY
);
1080 unsigned long nr_unstable
= global_page_state(NR_UNSTABLE_NFS
);
1081 DECLARE_COMPLETION_ONSTACK(done
);
1082 struct wb_writeback_work work
= {
1084 .sync_mode
= WB_SYNC_NONE
,
1088 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1090 work
.nr_pages
= nr_dirty
+ nr_unstable
+ get_nr_dirty_inodes();
1092 bdi_queue_work(sb
->s_bdi
, &work
);
1093 wait_for_completion(&done
);
1095 EXPORT_SYMBOL(writeback_inodes_sb
);
1098 * writeback_inodes_sb_if_idle - start writeback if none underway
1099 * @sb: the superblock
1101 * Invoke writeback_inodes_sb if no writeback is currently underway.
1102 * Returns 1 if writeback was started, 0 if not.
1104 int writeback_inodes_sb_if_idle(struct super_block
*sb
)
1106 if (!writeback_in_progress(sb
->s_bdi
)) {
1107 down_read(&sb
->s_umount
);
1108 writeback_inodes_sb(sb
);
1109 up_read(&sb
->s_umount
);
1114 EXPORT_SYMBOL(writeback_inodes_sb_if_idle
);
1117 * sync_inodes_sb - sync sb inode pages
1118 * @sb: the superblock
1120 * This function writes and waits on any dirty inode belonging to this
1121 * super_block. The number of pages synced is returned.
1123 void sync_inodes_sb(struct super_block
*sb
)
1125 DECLARE_COMPLETION_ONSTACK(done
);
1126 struct wb_writeback_work work
= {
1128 .sync_mode
= WB_SYNC_ALL
,
1129 .nr_pages
= LONG_MAX
,
1134 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1136 bdi_queue_work(sb
->s_bdi
, &work
);
1137 wait_for_completion(&done
);
1141 EXPORT_SYMBOL(sync_inodes_sb
);
1144 * write_inode_now - write an inode to disk
1145 * @inode: inode to write to disk
1146 * @sync: whether the write should be synchronous or not
1148 * This function commits an inode to disk immediately if it is dirty. This is
1149 * primarily needed by knfsd.
1151 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1153 int write_inode_now(struct inode
*inode
, int sync
)
1156 struct writeback_control wbc
= {
1157 .nr_to_write
= LONG_MAX
,
1158 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1160 .range_end
= LLONG_MAX
,
1163 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1164 wbc
.nr_to_write
= 0;
1167 spin_lock(&inode_lock
);
1168 ret
= writeback_single_inode(inode
, &wbc
);
1169 spin_unlock(&inode_lock
);
1171 inode_sync_wait(inode
);
1174 EXPORT_SYMBOL(write_inode_now
);
1177 * sync_inode - write an inode and its pages to disk.
1178 * @inode: the inode to sync
1179 * @wbc: controls the writeback mode
1181 * sync_inode() will write an inode and its pages to disk. It will also
1182 * correctly update the inode on its superblock's dirty inode lists and will
1183 * update inode->i_state.
1185 * The caller must have a ref on the inode.
1187 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1191 spin_lock(&inode_lock
);
1192 ret
= writeback_single_inode(inode
, wbc
);
1193 spin_unlock(&inode_lock
);
1196 EXPORT_SYMBOL(sync_inode
);
1199 * sync_inode - write an inode to disk
1200 * @inode: the inode to sync
1201 * @wait: wait for I/O to complete.
1203 * Write an inode to disk and adjust it's dirty state after completion.
1205 * Note: only writes the actual inode, no associated data or other metadata.
1207 int sync_inode_metadata(struct inode
*inode
, int wait
)
1209 struct writeback_control wbc
= {
1210 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1211 .nr_to_write
= 0, /* metadata-only */
1214 return sync_inode(inode
, &wbc
);
1216 EXPORT_SYMBOL(sync_inode_metadata
);