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 guarentees 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 spin_lock_bh(&bdi
->wb_lock
);
173 bdi_wakeup_flusher(bdi
);
174 spin_unlock_bh(&bdi
->wb_lock
);
178 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
179 * furthest end of its superblock's dirty-inode list.
181 * Before stamping the inode's ->dirtied_when, we check to see whether it is
182 * already the most-recently-dirtied inode on the b_dirty list. If that is
183 * the case then the inode must have been redirtied while it was being written
184 * out and we don't reset its dirtied_when.
186 static void redirty_tail(struct inode
*inode
)
188 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
190 if (!list_empty(&wb
->b_dirty
)) {
193 tail
= wb_inode(wb
->b_dirty
.next
);
194 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
195 inode
->dirtied_when
= jiffies
;
197 list_move(&inode
->i_wb_list
, &wb
->b_dirty
);
201 * requeue inode for re-scanning after bdi->b_io list is exhausted.
203 static void requeue_io(struct inode
*inode
)
205 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
207 list_move(&inode
->i_wb_list
, &wb
->b_more_io
);
210 static void inode_sync_complete(struct inode
*inode
)
213 * Prevent speculative execution through spin_unlock(&inode_lock);
216 wake_up_bit(&inode
->i_state
, __I_SYNC
);
219 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
221 bool ret
= time_after(inode
->dirtied_when
, t
);
224 * For inodes being constantly redirtied, dirtied_when can get stuck.
225 * It _appears_ to be in the future, but is actually in distant past.
226 * This test is necessary to prevent such wrapped-around relative times
227 * from permanently stopping the whole bdi writeback.
229 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
235 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
237 static void move_expired_inodes(struct list_head
*delaying_queue
,
238 struct list_head
*dispatch_queue
,
239 unsigned long *older_than_this
)
242 struct list_head
*pos
, *node
;
243 struct super_block
*sb
= NULL
;
247 while (!list_empty(delaying_queue
)) {
248 inode
= wb_inode(delaying_queue
->prev
);
249 if (older_than_this
&&
250 inode_dirtied_after(inode
, *older_than_this
))
252 if (sb
&& sb
!= inode
->i_sb
)
255 list_move(&inode
->i_wb_list
, &tmp
);
258 /* just one sb in list, splice to dispatch_queue and we're done */
260 list_splice(&tmp
, dispatch_queue
);
264 /* Move inodes from one superblock together */
265 while (!list_empty(&tmp
)) {
266 sb
= wb_inode(tmp
.prev
)->i_sb
;
267 list_for_each_prev_safe(pos
, node
, &tmp
) {
268 inode
= wb_inode(pos
);
269 if (inode
->i_sb
== sb
)
270 list_move(&inode
->i_wb_list
, dispatch_queue
);
276 * Queue all expired dirty inodes for io, eldest first.
278 * newly dirtied b_dirty b_io b_more_io
279 * =============> gf edc BA
281 * newly dirtied b_dirty b_io b_more_io
282 * =============> g fBAedc
284 * +--> dequeue for IO
286 static void queue_io(struct bdi_writeback
*wb
, unsigned long *older_than_this
)
288 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
289 move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, older_than_this
);
292 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
294 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
))
295 return inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
300 * Wait for writeback on an inode to complete.
302 static void inode_wait_for_writeback(struct inode
*inode
)
304 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
305 wait_queue_head_t
*wqh
;
307 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
308 while (inode
->i_state
& I_SYNC
) {
309 spin_unlock(&inode_lock
);
310 __wait_on_bit(wqh
, &wq
, inode_wait
, TASK_UNINTERRUPTIBLE
);
311 spin_lock(&inode_lock
);
316 * Write out an inode's dirty pages. Called under inode_lock. Either the
317 * caller has ref on the inode (either via __iget or via syscall against an fd)
318 * or the inode has I_WILL_FREE set (via generic_forget_inode)
320 * If `wait' is set, wait on the writeout.
322 * The whole writeout design is quite complex and fragile. We want to avoid
323 * starvation of particular inodes when others are being redirtied, prevent
326 * Called under inode_lock.
329 writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
331 struct address_space
*mapping
= inode
->i_mapping
;
335 if (!atomic_read(&inode
->i_count
))
336 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
338 WARN_ON(inode
->i_state
& I_WILL_FREE
);
340 if (inode
->i_state
& I_SYNC
) {
342 * If this inode is locked for writeback and we are not doing
343 * writeback-for-data-integrity, move it to b_more_io so that
344 * writeback can proceed with the other inodes on s_io.
346 * We'll have another go at writing back this inode when we
347 * completed a full scan of b_io.
349 if (wbc
->sync_mode
!= WB_SYNC_ALL
) {
355 * It's a data-integrity sync. We must wait.
357 inode_wait_for_writeback(inode
);
360 BUG_ON(inode
->i_state
& I_SYNC
);
362 /* Set I_SYNC, reset I_DIRTY_PAGES */
363 inode
->i_state
|= I_SYNC
;
364 inode
->i_state
&= ~I_DIRTY_PAGES
;
365 spin_unlock(&inode_lock
);
367 ret
= do_writepages(mapping
, wbc
);
370 * Make sure to wait on the data before writing out the metadata.
371 * This is important for filesystems that modify metadata on data
374 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
375 int err
= filemap_fdatawait(mapping
);
381 * Some filesystems may redirty the inode during the writeback
382 * due to delalloc, clear dirty metadata flags right before
385 spin_lock(&inode_lock
);
386 dirty
= inode
->i_state
& I_DIRTY
;
387 inode
->i_state
&= ~(I_DIRTY_SYNC
| I_DIRTY_DATASYNC
);
388 spin_unlock(&inode_lock
);
389 /* Don't write the inode if only I_DIRTY_PAGES was set */
390 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
391 int err
= write_inode(inode
, wbc
);
396 spin_lock(&inode_lock
);
397 inode
->i_state
&= ~I_SYNC
;
398 if (!(inode
->i_state
& I_FREEING
)) {
399 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
401 * We didn't write back all the pages. nfs_writepages()
402 * sometimes bales out without doing anything.
404 inode
->i_state
|= I_DIRTY_PAGES
;
405 if (wbc
->nr_to_write
<= 0) {
407 * slice used up: queue for next turn
412 * Writeback blocked by something other than
413 * congestion. Delay the inode for some time to
414 * avoid spinning on the CPU (100% iowait)
415 * retrying writeback of the dirty page/inode
416 * that cannot be performed immediately.
420 } else if (inode
->i_state
& I_DIRTY
) {
422 * Filesystems can dirty the inode during writeback
423 * operations, such as delayed allocation during
424 * submission or metadata updates after data IO
430 * The inode is clean. At this point we either have
431 * a reference to the inode or it's on it's way out.
432 * No need to add it back to the LRU.
434 list_del_init(&inode
->i_wb_list
);
437 inode_sync_complete(inode
);
442 * For background writeback the caller does not have the sb pinned
443 * before calling writeback. So make sure that we do pin it, so it doesn't
444 * go away while we are writing inodes from it.
446 static bool pin_sb_for_writeback(struct super_block
*sb
)
449 if (list_empty(&sb
->s_instances
)) {
450 spin_unlock(&sb_lock
);
455 spin_unlock(&sb_lock
);
457 if (down_read_trylock(&sb
->s_umount
)) {
460 up_read(&sb
->s_umount
);
468 * Write a portion of b_io inodes which belong to @sb.
470 * If @only_this_sb is true, then find and write all such
471 * inodes. Otherwise write only ones which go sequentially
474 * Return 1, if the caller writeback routine should be
475 * interrupted. Otherwise return 0.
477 static int writeback_sb_inodes(struct super_block
*sb
, struct bdi_writeback
*wb
,
478 struct writeback_control
*wbc
, bool only_this_sb
)
480 while (!list_empty(&wb
->b_io
)) {
482 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
484 if (inode
->i_sb
!= sb
) {
487 * We only want to write back data for this
488 * superblock, move all inodes not belonging
489 * to it back onto the dirty list.
496 * The inode belongs to a different superblock.
497 * Bounce back to the caller to unpin this and
498 * pin the next superblock.
504 * Don't bother with new inodes or inodes beeing freed, first
505 * kind does not need peridic writeout yet, and for the latter
506 * kind writeout is handled by the freer.
508 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
514 * Was this inode dirtied after sync_sb_inodes was called?
515 * This keeps sync from extra jobs and livelock.
517 if (inode_dirtied_after(inode
, wbc
->wb_start
))
521 pages_skipped
= wbc
->pages_skipped
;
522 writeback_single_inode(inode
, wbc
);
523 if (wbc
->pages_skipped
!= pages_skipped
) {
525 * writeback is not making progress due to locked
526 * buffers. Skip this inode for now.
530 spin_unlock(&inode_lock
);
533 spin_lock(&inode_lock
);
534 if (wbc
->nr_to_write
<= 0) {
538 if (!list_empty(&wb
->b_more_io
))
545 void writeback_inodes_wb(struct bdi_writeback
*wb
,
546 struct writeback_control
*wbc
)
551 wbc
->wb_start
= jiffies
; /* livelock avoidance */
552 spin_lock(&inode_lock
);
553 if (!wbc
->for_kupdate
|| list_empty(&wb
->b_io
))
554 queue_io(wb
, wbc
->older_than_this
);
556 while (!list_empty(&wb
->b_io
)) {
557 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
558 struct super_block
*sb
= inode
->i_sb
;
560 if (!pin_sb_for_writeback(sb
)) {
564 ret
= writeback_sb_inodes(sb
, wb
, wbc
, false);
570 spin_unlock(&inode_lock
);
571 /* Leave any unwritten inodes on b_io */
574 static void __writeback_inodes_sb(struct super_block
*sb
,
575 struct bdi_writeback
*wb
, struct writeback_control
*wbc
)
577 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
579 spin_lock(&inode_lock
);
580 if (!wbc
->for_kupdate
|| list_empty(&wb
->b_io
))
581 queue_io(wb
, wbc
->older_than_this
);
582 writeback_sb_inodes(sb
, wb
, wbc
, true);
583 spin_unlock(&inode_lock
);
587 * The maximum number of pages to writeout in a single bdi flush/kupdate
588 * operation. We do this so we don't hold I_SYNC against an inode for
589 * enormous amounts of time, which would block a userspace task which has
590 * been forced to throttle against that inode. Also, the code reevaluates
591 * the dirty each time it has written this many pages.
593 #define MAX_WRITEBACK_PAGES 1024
595 static inline bool over_bground_thresh(void)
597 unsigned long background_thresh
, dirty_thresh
;
599 global_dirty_limits(&background_thresh
, &dirty_thresh
);
601 return (global_page_state(NR_FILE_DIRTY
) +
602 global_page_state(NR_UNSTABLE_NFS
) > background_thresh
);
606 * Explicit flushing or periodic writeback of "old" data.
608 * Define "old": the first time one of an inode's pages is dirtied, we mark the
609 * dirtying-time in the inode's address_space. So this periodic writeback code
610 * just walks the superblock inode list, writing back any inodes which are
611 * older than a specific point in time.
613 * Try to run once per dirty_writeback_interval. But if a writeback event
614 * takes longer than a dirty_writeback_interval interval, then leave a
617 * older_than_this takes precedence over nr_to_write. So we'll only write back
618 * all dirty pages if they are all attached to "old" mappings.
620 static long wb_writeback(struct bdi_writeback
*wb
,
621 struct wb_writeback_work
*work
)
623 struct writeback_control wbc
= {
624 .sync_mode
= work
->sync_mode
,
625 .older_than_this
= NULL
,
626 .for_kupdate
= work
->for_kupdate
,
627 .for_background
= work
->for_background
,
628 .range_cyclic
= work
->range_cyclic
,
630 unsigned long oldest_jif
;
634 if (wbc
.for_kupdate
) {
635 wbc
.older_than_this
= &oldest_jif
;
636 oldest_jif
= jiffies
-
637 msecs_to_jiffies(dirty_expire_interval
* 10);
639 if (!wbc
.range_cyclic
) {
641 wbc
.range_end
= LLONG_MAX
;
644 wbc
.wb_start
= jiffies
; /* livelock avoidance */
647 * Stop writeback when nr_pages has been consumed
649 if (work
->nr_pages
<= 0)
653 * For background writeout, stop when we are below the
654 * background dirty threshold
656 if (work
->for_background
&& !over_bground_thresh())
660 wbc
.nr_to_write
= MAX_WRITEBACK_PAGES
;
661 wbc
.pages_skipped
= 0;
663 trace_wbc_writeback_start(&wbc
, wb
->bdi
);
665 __writeback_inodes_sb(work
->sb
, wb
, &wbc
);
667 writeback_inodes_wb(wb
, &wbc
);
668 trace_wbc_writeback_written(&wbc
, wb
->bdi
);
670 work
->nr_pages
-= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
671 wrote
+= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
674 * If we consumed everything, see if we have more
676 if (wbc
.nr_to_write
<= 0)
679 * Didn't write everything and we don't have more IO, bail
684 * Did we write something? Try for more
686 if (wbc
.nr_to_write
< MAX_WRITEBACK_PAGES
)
689 * Nothing written. Wait for some inode to
690 * become available for writeback. Otherwise
691 * we'll just busyloop.
693 spin_lock(&inode_lock
);
694 if (!list_empty(&wb
->b_more_io
)) {
695 inode
= wb_inode(wb
->b_more_io
.prev
);
696 trace_wbc_writeback_wait(&wbc
, wb
->bdi
);
697 inode_wait_for_writeback(inode
);
699 spin_unlock(&inode_lock
);
706 * Return the next wb_writeback_work struct that hasn't been processed yet.
708 static struct wb_writeback_work
*
709 get_next_work_item(struct backing_dev_info
*bdi
)
711 struct wb_writeback_work
*work
= NULL
;
713 spin_lock_bh(&bdi
->wb_lock
);
714 if (!list_empty(&bdi
->work_list
)) {
715 work
= list_entry(bdi
->work_list
.next
,
716 struct wb_writeback_work
, list
);
717 list_del_init(&work
->list
);
719 spin_unlock_bh(&bdi
->wb_lock
);
724 * Add in the number of potentially dirty inodes, because each inode
725 * write can dirty pagecache in the underlying blockdev.
727 static unsigned long get_nr_dirty_pages(void)
729 return global_page_state(NR_FILE_DIRTY
) +
730 global_page_state(NR_UNSTABLE_NFS
) +
731 get_nr_dirty_inodes();
734 static long wb_check_background_flush(struct bdi_writeback
*wb
)
736 if (over_bground_thresh()) {
738 struct wb_writeback_work work
= {
739 .nr_pages
= LONG_MAX
,
740 .sync_mode
= WB_SYNC_NONE
,
745 return wb_writeback(wb
, &work
);
751 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
753 unsigned long expired
;
757 * When set to zero, disable periodic writeback
759 if (!dirty_writeback_interval
)
762 expired
= wb
->last_old_flush
+
763 msecs_to_jiffies(dirty_writeback_interval
* 10);
764 if (time_before(jiffies
, expired
))
767 wb
->last_old_flush
= jiffies
;
768 nr_pages
= get_nr_dirty_pages();
771 struct wb_writeback_work work
= {
772 .nr_pages
= nr_pages
,
773 .sync_mode
= WB_SYNC_NONE
,
778 return wb_writeback(wb
, &work
);
785 * Retrieve work items and do the writeback they describe
787 long wb_do_writeback(struct bdi_writeback
*wb
, int force_wait
)
789 struct backing_dev_info
*bdi
= wb
->bdi
;
790 struct wb_writeback_work
*work
;
793 set_bit(BDI_writeback_running
, &wb
->bdi
->state
);
794 while ((work
= get_next_work_item(bdi
)) != NULL
) {
796 * Override sync mode, in case we must wait for completion
797 * because this thread is exiting now.
800 work
->sync_mode
= WB_SYNC_ALL
;
802 trace_writeback_exec(bdi
, work
);
804 wrote
+= wb_writeback(wb
, work
);
807 * Notify the caller of completion if this is a synchronous
808 * work item, otherwise just free it.
811 complete(work
->done
);
817 * Check for periodic writeback, kupdated() style
819 wrote
+= wb_check_old_data_flush(wb
);
820 wrote
+= wb_check_background_flush(wb
);
821 clear_bit(BDI_writeback_running
, &wb
->bdi
->state
);
827 * Handle writeback of dirty data for the device backed by this bdi. Also
828 * wakes up periodically and does kupdated style flushing.
830 int bdi_writeback_thread(void *data
)
832 struct bdi_writeback
*wb
= data
;
833 struct backing_dev_info
*bdi
= wb
->bdi
;
836 current
->flags
|= PF_SWAPWRITE
;
838 wb
->last_active
= jiffies
;
841 * Our parent may run at a different priority, just set us to normal
843 set_user_nice(current
, 0);
845 trace_writeback_thread_start(bdi
);
847 while (!kthread_should_stop()) {
849 * Remove own delayed wake-up timer, since we are already awake
850 * and we'll take care of the preriodic write-back.
852 del_timer(&wb
->wakeup_timer
);
854 pages_written
= wb_do_writeback(wb
, 0);
856 trace_writeback_pages_written(pages_written
);
859 wb
->last_active
= jiffies
;
861 set_current_state(TASK_INTERRUPTIBLE
);
862 if (!list_empty(&bdi
->work_list
) || kthread_should_stop()) {
863 __set_current_state(TASK_RUNNING
);
867 if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
868 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval
* 10));
871 * We have nothing to do, so can go sleep without any
872 * timeout and save power. When a work is queued or
873 * something is made dirty - we will be woken up.
881 /* Flush any work that raced with us exiting */
882 if (!list_empty(&bdi
->work_list
))
883 wb_do_writeback(wb
, 1);
885 trace_writeback_thread_stop(bdi
);
891 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
894 void wakeup_flusher_threads(long nr_pages
)
896 struct backing_dev_info
*bdi
;
899 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
900 global_page_state(NR_UNSTABLE_NFS
);
904 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
905 if (!bdi_has_dirty_io(bdi
))
907 __bdi_start_writeback(bdi
, nr_pages
, false);
912 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
914 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
915 struct dentry
*dentry
;
916 const char *name
= "?";
918 dentry
= d_find_alias(inode
);
920 spin_lock(&dentry
->d_lock
);
921 name
= (const char *) dentry
->d_name
.name
;
924 "%s(%d): dirtied inode %lu (%s) on %s\n",
925 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
926 name
, inode
->i_sb
->s_id
);
928 spin_unlock(&dentry
->d_lock
);
935 * __mark_inode_dirty - internal function
936 * @inode: inode to mark
937 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
938 * Mark an inode as dirty. Callers should use mark_inode_dirty or
939 * mark_inode_dirty_sync.
941 * Put the inode on the super block's dirty list.
943 * CAREFUL! We mark it dirty unconditionally, but move it onto the
944 * dirty list only if it is hashed or if it refers to a blockdev.
945 * If it was not hashed, it will never be added to the dirty list
946 * even if it is later hashed, as it will have been marked dirty already.
948 * In short, make sure you hash any inodes _before_ you start marking
951 * This function *must* be atomic for the I_DIRTY_PAGES case -
952 * set_page_dirty() is called under spinlock in several places.
954 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
955 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
956 * the kernel-internal blockdev inode represents the dirtying time of the
957 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
958 * page->mapping->host, so the page-dirtying time is recorded in the internal
961 void __mark_inode_dirty(struct inode
*inode
, int flags
)
963 struct super_block
*sb
= inode
->i_sb
;
964 struct backing_dev_info
*bdi
= NULL
;
965 bool wakeup_bdi
= false;
968 * Don't do this for I_DIRTY_PAGES - that doesn't actually
969 * dirty the inode itself
971 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
972 if (sb
->s_op
->dirty_inode
)
973 sb
->s_op
->dirty_inode(inode
);
977 * make sure that changes are seen by all cpus before we test i_state
982 /* avoid the locking if we can */
983 if ((inode
->i_state
& flags
) == flags
)
986 if (unlikely(block_dump
))
987 block_dump___mark_inode_dirty(inode
);
989 spin_lock(&inode_lock
);
990 if ((inode
->i_state
& flags
) != flags
) {
991 const int was_dirty
= inode
->i_state
& I_DIRTY
;
993 inode
->i_state
|= flags
;
996 * If the inode is being synced, just update its dirty state.
997 * The unlocker will place the inode on the appropriate
998 * superblock list, based upon its state.
1000 if (inode
->i_state
& I_SYNC
)
1004 * Only add valid (hashed) inodes to the superblock's
1005 * dirty list. Add blockdev inodes as well.
1007 if (!S_ISBLK(inode
->i_mode
)) {
1008 if (inode_unhashed(inode
))
1011 if (inode
->i_state
& I_FREEING
)
1015 * If the inode was already on b_dirty/b_io/b_more_io, don't
1016 * reposition it (that would break b_dirty time-ordering).
1019 bdi
= inode_to_bdi(inode
);
1021 if (bdi_cap_writeback_dirty(bdi
)) {
1022 WARN(!test_bit(BDI_registered
, &bdi
->state
),
1023 "bdi-%s not registered\n", bdi
->name
);
1026 * If this is the first dirty inode for this
1027 * bdi, we have to wake-up the corresponding
1028 * bdi thread to make sure background
1029 * write-back happens later.
1031 if (!wb_has_dirty_io(&bdi
->wb
))
1035 inode
->dirtied_when
= jiffies
;
1036 list_move(&inode
->i_wb_list
, &bdi
->wb
.b_dirty
);
1040 spin_unlock(&inode_lock
);
1043 bdi_wakeup_thread_delayed(bdi
);
1045 EXPORT_SYMBOL(__mark_inode_dirty
);
1048 * Write out a superblock's list of dirty inodes. A wait will be performed
1049 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1051 * If older_than_this is non-NULL, then only write out inodes which
1052 * had their first dirtying at a time earlier than *older_than_this.
1054 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1055 * This function assumes that the blockdev superblock's inodes are backed by
1056 * a variety of queues, so all inodes are searched. For other superblocks,
1057 * assume that all inodes are backed by the same queue.
1059 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1060 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1061 * on the writer throttling path, and we get decent balancing between many
1062 * throttled threads: we don't want them all piling up on inode_sync_wait.
1064 static void wait_sb_inodes(struct super_block
*sb
)
1066 struct inode
*inode
, *old_inode
= NULL
;
1069 * We need to be protected against the filesystem going from
1070 * r/o to r/w or vice versa.
1072 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1074 spin_lock(&inode_lock
);
1077 * Data integrity sync. Must wait for all pages under writeback,
1078 * because there may have been pages dirtied before our sync
1079 * call, but which had writeout started before we write it out.
1080 * In which case, the inode may not be on the dirty list, but
1081 * we still have to wait for that writeout.
1083 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1084 struct address_space
*mapping
;
1086 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
))
1088 mapping
= inode
->i_mapping
;
1089 if (mapping
->nrpages
== 0)
1092 spin_unlock(&inode_lock
);
1094 * We hold a reference to 'inode' so it couldn't have
1095 * been removed from s_inodes list while we dropped the
1096 * inode_lock. We cannot iput the inode now as we can
1097 * be holding the last reference and we cannot iput it
1098 * under inode_lock. So we keep the reference and iput
1104 filemap_fdatawait(mapping
);
1108 spin_lock(&inode_lock
);
1110 spin_unlock(&inode_lock
);
1115 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1116 * @sb: the superblock
1117 * @nr: the number of pages to write
1119 * Start writeback on some inodes on this super_block. No guarantees are made
1120 * on how many (if any) will be written, and this function does not wait
1121 * for IO completion of submitted IO.
1123 void writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
)
1125 DECLARE_COMPLETION_ONSTACK(done
);
1126 struct wb_writeback_work work
= {
1128 .sync_mode
= WB_SYNC_NONE
,
1133 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1134 bdi_queue_work(sb
->s_bdi
, &work
);
1135 wait_for_completion(&done
);
1137 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
1140 * writeback_inodes_sb - writeback dirty inodes from given super_block
1141 * @sb: the superblock
1143 * Start writeback on some inodes on this super_block. No guarantees are made
1144 * on how many (if any) will be written, and this function does not wait
1145 * for IO completion of submitted IO.
1147 void writeback_inodes_sb(struct super_block
*sb
)
1149 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages());
1151 EXPORT_SYMBOL(writeback_inodes_sb
);
1154 * writeback_inodes_sb_if_idle - start writeback if none underway
1155 * @sb: the superblock
1157 * Invoke writeback_inodes_sb if no writeback is currently underway.
1158 * Returns 1 if writeback was started, 0 if not.
1160 int writeback_inodes_sb_if_idle(struct super_block
*sb
)
1162 if (!writeback_in_progress(sb
->s_bdi
)) {
1163 down_read(&sb
->s_umount
);
1164 writeback_inodes_sb(sb
);
1165 up_read(&sb
->s_umount
);
1170 EXPORT_SYMBOL(writeback_inodes_sb_if_idle
);
1173 * writeback_inodes_sb_if_idle - start writeback if none underway
1174 * @sb: the superblock
1175 * @nr: the number of pages to write
1177 * Invoke writeback_inodes_sb if no writeback is currently underway.
1178 * Returns 1 if writeback was started, 0 if not.
1180 int writeback_inodes_sb_nr_if_idle(struct super_block
*sb
,
1183 if (!writeback_in_progress(sb
->s_bdi
)) {
1184 down_read(&sb
->s_umount
);
1185 writeback_inodes_sb_nr(sb
, nr
);
1186 up_read(&sb
->s_umount
);
1191 EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle
);
1194 * sync_inodes_sb - sync sb inode pages
1195 * @sb: the superblock
1197 * This function writes and waits on any dirty inode belonging to this
1198 * super_block. The number of pages synced is returned.
1200 void sync_inodes_sb(struct super_block
*sb
)
1202 DECLARE_COMPLETION_ONSTACK(done
);
1203 struct wb_writeback_work work
= {
1205 .sync_mode
= WB_SYNC_ALL
,
1206 .nr_pages
= LONG_MAX
,
1211 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1213 bdi_queue_work(sb
->s_bdi
, &work
);
1214 wait_for_completion(&done
);
1218 EXPORT_SYMBOL(sync_inodes_sb
);
1221 * write_inode_now - write an inode to disk
1222 * @inode: inode to write to disk
1223 * @sync: whether the write should be synchronous or not
1225 * This function commits an inode to disk immediately if it is dirty. This is
1226 * primarily needed by knfsd.
1228 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1230 int write_inode_now(struct inode
*inode
, int sync
)
1233 struct writeback_control wbc
= {
1234 .nr_to_write
= LONG_MAX
,
1235 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1237 .range_end
= LLONG_MAX
,
1240 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1241 wbc
.nr_to_write
= 0;
1244 spin_lock(&inode_lock
);
1245 ret
= writeback_single_inode(inode
, &wbc
);
1246 spin_unlock(&inode_lock
);
1248 inode_sync_wait(inode
);
1251 EXPORT_SYMBOL(write_inode_now
);
1254 * sync_inode - write an inode and its pages to disk.
1255 * @inode: the inode to sync
1256 * @wbc: controls the writeback mode
1258 * sync_inode() will write an inode and its pages to disk. It will also
1259 * correctly update the inode on its superblock's dirty inode lists and will
1260 * update inode->i_state.
1262 * The caller must have a ref on the inode.
1264 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1268 spin_lock(&inode_lock
);
1269 ret
= writeback_single_inode(inode
, wbc
);
1270 spin_unlock(&inode_lock
);
1273 EXPORT_SYMBOL(sync_inode
);
1276 * sync_inode - write an inode to disk
1277 * @inode: the inode to sync
1278 * @wait: wait for I/O to complete.
1280 * Write an inode to disk and adjust it's dirty state after completion.
1282 * Note: only writes the actual inode, no associated data or other metadata.
1284 int sync_inode_metadata(struct inode
*inode
, int wait
)
1286 struct writeback_control wbc
= {
1287 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1288 .nr_to_write
= 0, /* metadata-only */
1291 return sync_inode(inode
, &wbc
);
1293 EXPORT_SYMBOL(sync_inode_metadata
);