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>
55 #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info)
58 * We don't actually have pdflush, but this one is exported though /proc...
60 int nr_pdflush_threads
;
63 * writeback_in_progress - determine whether there is writeback in progress
64 * @bdi: the device's backing_dev_info structure.
66 * Determine whether there is writeback waiting to be handled against a
69 int writeback_in_progress(struct backing_dev_info
*bdi
)
71 return !list_empty(&bdi
->work_list
);
74 static void bdi_queue_work(struct backing_dev_info
*bdi
,
75 struct wb_writeback_work
*work
)
77 trace_writeback_queue(bdi
, work
);
79 spin_lock_bh(&bdi
->wb_lock
);
80 list_add_tail(&work
->list
, &bdi
->work_list
);
82 wake_up_process(bdi
->wb
.task
);
85 * The bdi thread isn't there, wake up the forker thread which
86 * will create and run it.
88 trace_writeback_nothread(bdi
, work
);
89 wake_up_process(default_backing_dev_info
.wb
.task
);
91 spin_unlock_bh(&bdi
->wb_lock
);
95 __bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
,
96 bool range_cyclic
, bool for_background
)
98 struct wb_writeback_work
*work
;
101 * This is WB_SYNC_NONE writeback, so if allocation fails just
102 * wakeup the thread for old dirty data writeback
104 work
= kzalloc(sizeof(*work
), GFP_ATOMIC
);
107 trace_writeback_nowork(bdi
);
108 wake_up_process(bdi
->wb
.task
);
113 work
->sync_mode
= WB_SYNC_NONE
;
114 work
->nr_pages
= nr_pages
;
115 work
->range_cyclic
= range_cyclic
;
116 work
->for_background
= for_background
;
118 bdi_queue_work(bdi
, work
);
122 * bdi_start_writeback - start writeback
123 * @bdi: the backing device to write from
124 * @nr_pages: the number of pages to write
127 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
128 * started when this function returns, we make no guarentees on
129 * completion. Caller need not hold sb s_umount semaphore.
132 void bdi_start_writeback(struct backing_dev_info
*bdi
, long nr_pages
)
134 __bdi_start_writeback(bdi
, nr_pages
, true, false);
138 * bdi_start_background_writeback - start background writeback
139 * @bdi: the backing device to write from
142 * This does WB_SYNC_NONE background writeback. The IO is only
143 * started when this function returns, we make no guarentees on
144 * completion. Caller need not hold sb s_umount semaphore.
146 void bdi_start_background_writeback(struct backing_dev_info
*bdi
)
148 __bdi_start_writeback(bdi
, LONG_MAX
, true, true);
152 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
153 * furthest end of its superblock's dirty-inode list.
155 * Before stamping the inode's ->dirtied_when, we check to see whether it is
156 * already the most-recently-dirtied inode on the b_dirty list. If that is
157 * the case then the inode must have been redirtied while it was being written
158 * out and we don't reset its dirtied_when.
160 static void redirty_tail(struct inode
*inode
)
162 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
164 if (!list_empty(&wb
->b_dirty
)) {
167 tail
= list_entry(wb
->b_dirty
.next
, struct inode
, i_list
);
168 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
169 inode
->dirtied_when
= jiffies
;
171 list_move(&inode
->i_list
, &wb
->b_dirty
);
175 * requeue inode for re-scanning after bdi->b_io list is exhausted.
177 static void requeue_io(struct inode
*inode
)
179 struct bdi_writeback
*wb
= &inode_to_bdi(inode
)->wb
;
181 list_move(&inode
->i_list
, &wb
->b_more_io
);
184 static void inode_sync_complete(struct inode
*inode
)
187 * Prevent speculative execution through spin_unlock(&inode_lock);
190 wake_up_bit(&inode
->i_state
, __I_SYNC
);
193 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
195 bool ret
= time_after(inode
->dirtied_when
, t
);
198 * For inodes being constantly redirtied, dirtied_when can get stuck.
199 * It _appears_ to be in the future, but is actually in distant past.
200 * This test is necessary to prevent such wrapped-around relative times
201 * from permanently stopping the whole bdi writeback.
203 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
209 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
211 static void move_expired_inodes(struct list_head
*delaying_queue
,
212 struct list_head
*dispatch_queue
,
213 unsigned long *older_than_this
)
216 struct list_head
*pos
, *node
;
217 struct super_block
*sb
= NULL
;
221 while (!list_empty(delaying_queue
)) {
222 inode
= list_entry(delaying_queue
->prev
, struct inode
, i_list
);
223 if (older_than_this
&&
224 inode_dirtied_after(inode
, *older_than_this
))
226 if (sb
&& sb
!= inode
->i_sb
)
229 list_move(&inode
->i_list
, &tmp
);
232 /* just one sb in list, splice to dispatch_queue and we're done */
234 list_splice(&tmp
, dispatch_queue
);
238 /* Move inodes from one superblock together */
239 while (!list_empty(&tmp
)) {
240 inode
= list_entry(tmp
.prev
, struct inode
, i_list
);
242 list_for_each_prev_safe(pos
, node
, &tmp
) {
243 inode
= list_entry(pos
, struct inode
, i_list
);
244 if (inode
->i_sb
== sb
)
245 list_move(&inode
->i_list
, dispatch_queue
);
251 * Queue all expired dirty inodes for io, eldest first.
253 * newly dirtied b_dirty b_io b_more_io
254 * =============> gf edc BA
256 * newly dirtied b_dirty b_io b_more_io
257 * =============> g fBAedc
259 * +--> dequeue for IO
261 static void queue_io(struct bdi_writeback
*wb
, unsigned long *older_than_this
)
263 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
264 move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, older_than_this
);
267 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
269 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
))
270 return inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
275 * Wait for writeback on an inode to complete.
277 static void inode_wait_for_writeback(struct inode
*inode
)
279 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
280 wait_queue_head_t
*wqh
;
282 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
283 while (inode
->i_state
& I_SYNC
) {
284 spin_unlock(&inode_lock
);
285 __wait_on_bit(wqh
, &wq
, inode_wait
, TASK_UNINTERRUPTIBLE
);
286 spin_lock(&inode_lock
);
291 * Write out an inode's dirty pages. Called under inode_lock. Either the
292 * caller has ref on the inode (either via __iget or via syscall against an fd)
293 * or the inode has I_WILL_FREE set (via generic_forget_inode)
295 * If `wait' is set, wait on the writeout.
297 * The whole writeout design is quite complex and fragile. We want to avoid
298 * starvation of particular inodes when others are being redirtied, prevent
301 * Called under inode_lock.
304 writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
306 struct address_space
*mapping
= inode
->i_mapping
;
310 if (!atomic_read(&inode
->i_count
))
311 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
313 WARN_ON(inode
->i_state
& I_WILL_FREE
);
315 if (inode
->i_state
& I_SYNC
) {
317 * If this inode is locked for writeback and we are not doing
318 * writeback-for-data-integrity, move it to b_more_io so that
319 * writeback can proceed with the other inodes on s_io.
321 * We'll have another go at writing back this inode when we
322 * completed a full scan of b_io.
324 if (wbc
->sync_mode
!= WB_SYNC_ALL
) {
330 * It's a data-integrity sync. We must wait.
332 inode_wait_for_writeback(inode
);
335 BUG_ON(inode
->i_state
& I_SYNC
);
337 /* Set I_SYNC, reset I_DIRTY_PAGES */
338 inode
->i_state
|= I_SYNC
;
339 inode
->i_state
&= ~I_DIRTY_PAGES
;
340 spin_unlock(&inode_lock
);
342 ret
= do_writepages(mapping
, wbc
);
345 * Make sure to wait on the data before writing out the metadata.
346 * This is important for filesystems that modify metadata on data
349 if (wbc
->sync_mode
== WB_SYNC_ALL
) {
350 int err
= filemap_fdatawait(mapping
);
356 * Some filesystems may redirty the inode during the writeback
357 * due to delalloc, clear dirty metadata flags right before
360 spin_lock(&inode_lock
);
361 dirty
= inode
->i_state
& I_DIRTY
;
362 inode
->i_state
&= ~(I_DIRTY_SYNC
| I_DIRTY_DATASYNC
);
363 spin_unlock(&inode_lock
);
364 /* Don't write the inode if only I_DIRTY_PAGES was set */
365 if (dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
366 int err
= write_inode(inode
, wbc
);
371 spin_lock(&inode_lock
);
372 inode
->i_state
&= ~I_SYNC
;
373 if (!(inode
->i_state
& I_FREEING
)) {
374 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
)) {
376 * We didn't write back all the pages. nfs_writepages()
377 * sometimes bales out without doing anything. Redirty
378 * the inode; Move it from b_io onto b_more_io/b_dirty.
381 * akpm: if the caller was the kupdate function we put
382 * this inode at the head of b_dirty so it gets first
383 * consideration. Otherwise, move it to the tail, for
384 * the reasons described there. I'm not really sure
385 * how much sense this makes. Presumably I had a good
386 * reasons for doing it this way, and I'd rather not
387 * muck with it at present.
389 if (wbc
->for_kupdate
) {
391 * For the kupdate function we move the inode
392 * to b_more_io so it will get more writeout as
393 * soon as the queue becomes uncongested.
395 inode
->i_state
|= I_DIRTY_PAGES
;
396 if (wbc
->nr_to_write
<= 0) {
398 * slice used up: queue for next turn
403 * somehow blocked: retry later
409 * Otherwise fully redirty the inode so that
410 * other inodes on this superblock will get some
411 * writeout. Otherwise heavy writing to one
412 * file would indefinitely suspend writeout of
413 * all the other files.
415 inode
->i_state
|= I_DIRTY_PAGES
;
418 } else if (inode
->i_state
& I_DIRTY
) {
420 * Filesystems can dirty the inode during writeback
421 * operations, such as delayed allocation during
422 * submission or metadata updates after data IO
426 } else if (atomic_read(&inode
->i_count
)) {
428 * The inode is clean, inuse
430 list_move(&inode
->i_list
, &inode_in_use
);
433 * The inode is clean, unused
435 list_move(&inode
->i_list
, &inode_unused
);
438 inode_sync_complete(inode
);
443 * For background writeback the caller does not have the sb pinned
444 * before calling writeback. So make sure that we do pin it, so it doesn't
445 * go away while we are writing inodes from it.
447 static bool pin_sb_for_writeback(struct super_block
*sb
)
450 if (list_empty(&sb
->s_instances
)) {
451 spin_unlock(&sb_lock
);
456 spin_unlock(&sb_lock
);
458 if (down_read_trylock(&sb
->s_umount
)) {
461 up_read(&sb
->s_umount
);
469 * Write a portion of b_io inodes which belong to @sb.
471 * If @only_this_sb is true, then find and write all such
472 * inodes. Otherwise write only ones which go sequentially
475 * Return 1, if the caller writeback routine should be
476 * interrupted. Otherwise return 0.
478 static int writeback_sb_inodes(struct super_block
*sb
, struct bdi_writeback
*wb
,
479 struct writeback_control
*wbc
, bool only_this_sb
)
481 while (!list_empty(&wb
->b_io
)) {
483 struct inode
*inode
= list_entry(wb
->b_io
.prev
,
484 struct inode
, i_list
);
486 if (inode
->i_sb
!= sb
) {
489 * We only want to write back data for this
490 * superblock, move all inodes not belonging
491 * to it back onto the dirty list.
498 * The inode belongs to a different superblock.
499 * Bounce back to the caller to unpin this and
500 * pin the next superblock.
505 if (inode
->i_state
& (I_NEW
| I_WILL_FREE
)) {
510 * Was this inode dirtied after sync_sb_inodes was called?
511 * This keeps sync from extra jobs and livelock.
513 if (inode_dirtied_after(inode
, wbc
->wb_start
))
516 BUG_ON(inode
->i_state
& I_FREEING
);
518 pages_skipped
= wbc
->pages_skipped
;
519 writeback_single_inode(inode
, wbc
);
520 if (wbc
->pages_skipped
!= pages_skipped
) {
522 * writeback is not making progress due to locked
523 * buffers. Skip this inode for now.
527 spin_unlock(&inode_lock
);
530 spin_lock(&inode_lock
);
531 if (wbc
->nr_to_write
<= 0) {
535 if (!list_empty(&wb
->b_more_io
))
542 void writeback_inodes_wb(struct bdi_writeback
*wb
,
543 struct writeback_control
*wbc
)
548 wbc
->wb_start
= jiffies
; /* livelock avoidance */
549 spin_lock(&inode_lock
);
550 if (!wbc
->for_kupdate
|| list_empty(&wb
->b_io
))
551 queue_io(wb
, wbc
->older_than_this
);
553 while (!list_empty(&wb
->b_io
)) {
554 struct inode
*inode
= list_entry(wb
->b_io
.prev
,
555 struct inode
, i_list
);
556 struct super_block
*sb
= inode
->i_sb
;
558 if (!pin_sb_for_writeback(sb
)) {
562 ret
= writeback_sb_inodes(sb
, wb
, wbc
, false);
568 spin_unlock(&inode_lock
);
569 /* Leave any unwritten inodes on b_io */
572 static void __writeback_inodes_sb(struct super_block
*sb
,
573 struct bdi_writeback
*wb
, struct writeback_control
*wbc
)
575 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
577 spin_lock(&inode_lock
);
578 if (!wbc
->for_kupdate
|| list_empty(&wb
->b_io
))
579 queue_io(wb
, wbc
->older_than_this
);
580 writeback_sb_inodes(sb
, wb
, wbc
, true);
581 spin_unlock(&inode_lock
);
585 * The maximum number of pages to writeout in a single bdi flush/kupdate
586 * operation. We do this so we don't hold I_SYNC against an inode for
587 * enormous amounts of time, which would block a userspace task which has
588 * been forced to throttle against that inode. Also, the code reevaluates
589 * the dirty each time it has written this many pages.
591 #define MAX_WRITEBACK_PAGES 1024
593 static inline bool over_bground_thresh(void)
595 unsigned long background_thresh
, dirty_thresh
;
597 global_dirty_limits(&background_thresh
, &dirty_thresh
);
599 return (global_page_state(NR_FILE_DIRTY
) +
600 global_page_state(NR_UNSTABLE_NFS
) >= background_thresh
);
604 * Explicit flushing or periodic writeback of "old" data.
606 * Define "old": the first time one of an inode's pages is dirtied, we mark the
607 * dirtying-time in the inode's address_space. So this periodic writeback code
608 * just walks the superblock inode list, writing back any inodes which are
609 * older than a specific point in time.
611 * Try to run once per dirty_writeback_interval. But if a writeback event
612 * takes longer than a dirty_writeback_interval interval, then leave a
615 * older_than_this takes precedence over nr_to_write. So we'll only write back
616 * all dirty pages if they are all attached to "old" mappings.
618 static long wb_writeback(struct bdi_writeback
*wb
,
619 struct wb_writeback_work
*work
)
621 struct writeback_control wbc
= {
622 .sync_mode
= work
->sync_mode
,
623 .older_than_this
= NULL
,
624 .for_kupdate
= work
->for_kupdate
,
625 .for_background
= work
->for_background
,
626 .range_cyclic
= work
->range_cyclic
,
628 unsigned long oldest_jif
;
632 if (wbc
.for_kupdate
) {
633 wbc
.older_than_this
= &oldest_jif
;
634 oldest_jif
= jiffies
-
635 msecs_to_jiffies(dirty_expire_interval
* 10);
637 if (!wbc
.range_cyclic
) {
639 wbc
.range_end
= LLONG_MAX
;
642 wbc
.wb_start
= jiffies
; /* livelock avoidance */
645 * Stop writeback when nr_pages has been consumed
647 if (work
->nr_pages
<= 0)
651 * For background writeout, stop when we are below the
652 * background dirty threshold
654 if (work
->for_background
&& !over_bground_thresh())
658 wbc
.nr_to_write
= MAX_WRITEBACK_PAGES
;
659 wbc
.pages_skipped
= 0;
661 trace_wbc_writeback_start(&wbc
, wb
->bdi
);
663 __writeback_inodes_sb(work
->sb
, wb
, &wbc
);
665 writeback_inodes_wb(wb
, &wbc
);
666 trace_wbc_writeback_written(&wbc
, wb
->bdi
);
668 work
->nr_pages
-= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
669 wrote
+= MAX_WRITEBACK_PAGES
- wbc
.nr_to_write
;
672 * If we consumed everything, see if we have more
674 if (wbc
.nr_to_write
<= 0)
677 * Didn't write everything and we don't have more IO, bail
682 * Did we write something? Try for more
684 if (wbc
.nr_to_write
< MAX_WRITEBACK_PAGES
)
687 * Nothing written. Wait for some inode to
688 * become available for writeback. Otherwise
689 * we'll just busyloop.
691 spin_lock(&inode_lock
);
692 if (!list_empty(&wb
->b_more_io
)) {
693 inode
= list_entry(wb
->b_more_io
.prev
,
694 struct inode
, i_list
);
695 trace_wbc_writeback_wait(&wbc
, wb
->bdi
);
696 inode_wait_for_writeback(inode
);
698 spin_unlock(&inode_lock
);
705 * Return the next wb_writeback_work struct that hasn't been processed yet.
707 static struct wb_writeback_work
*
708 get_next_work_item(struct backing_dev_info
*bdi
)
710 struct wb_writeback_work
*work
= NULL
;
712 spin_lock_bh(&bdi
->wb_lock
);
713 if (!list_empty(&bdi
->work_list
)) {
714 work
= list_entry(bdi
->work_list
.next
,
715 struct wb_writeback_work
, list
);
716 list_del_init(&work
->list
);
718 spin_unlock_bh(&bdi
->wb_lock
);
722 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
724 unsigned long expired
;
728 * When set to zero, disable periodic writeback
730 if (!dirty_writeback_interval
)
733 expired
= wb
->last_old_flush
+
734 msecs_to_jiffies(dirty_writeback_interval
* 10);
735 if (time_before(jiffies
, expired
))
738 wb
->last_old_flush
= jiffies
;
739 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
740 global_page_state(NR_UNSTABLE_NFS
) +
741 (inodes_stat
.nr_inodes
- inodes_stat
.nr_unused
);
744 struct wb_writeback_work work
= {
745 .nr_pages
= nr_pages
,
746 .sync_mode
= WB_SYNC_NONE
,
751 return wb_writeback(wb
, &work
);
758 * Retrieve work items and do the writeback they describe
760 long wb_do_writeback(struct bdi_writeback
*wb
, int force_wait
)
762 struct backing_dev_info
*bdi
= wb
->bdi
;
763 struct wb_writeback_work
*work
;
766 while ((work
= get_next_work_item(bdi
)) != NULL
) {
768 * Override sync mode, in case we must wait for completion
769 * because this thread is exiting now.
772 work
->sync_mode
= WB_SYNC_ALL
;
774 trace_writeback_exec(bdi
, work
);
776 wrote
+= wb_writeback(wb
, work
);
779 * Notify the caller of completion if this is a synchronous
780 * work item, otherwise just free it.
783 complete(work
->done
);
789 * Check for periodic writeback, kupdated() style
791 wrote
+= wb_check_old_data_flush(wb
);
797 * Handle writeback of dirty data for the device backed by this bdi. Also
798 * wakes up periodically and does kupdated style flushing.
800 int bdi_writeback_thread(void *data
)
802 struct bdi_writeback
*wb
= data
;
803 struct backing_dev_info
*bdi
= wb
->bdi
;
806 current
->flags
|= PF_FLUSHER
| PF_SWAPWRITE
;
808 wb
->last_active
= jiffies
;
811 * Our parent may run at a different priority, just set us to normal
813 set_user_nice(current
, 0);
815 trace_writeback_thread_start(bdi
);
817 while (!kthread_should_stop()) {
819 * Remove own delayed wake-up timer, since we are already awake
820 * and we'll take care of the preriodic write-back.
822 del_timer(&wb
->wakeup_timer
);
824 pages_written
= wb_do_writeback(wb
, 0);
826 trace_writeback_pages_written(pages_written
);
829 wb
->last_active
= jiffies
;
831 set_current_state(TASK_INTERRUPTIBLE
);
832 if (!list_empty(&bdi
->work_list
)) {
833 __set_current_state(TASK_RUNNING
);
837 if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
838 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval
* 10));
841 * We have nothing to do, so can go sleep without any
842 * timeout and save power. When a work is queued or
843 * something is made dirty - we will be woken up.
851 /* Flush any work that raced with us exiting */
852 if (!list_empty(&bdi
->work_list
))
853 wb_do_writeback(wb
, 1);
855 trace_writeback_thread_stop(bdi
);
861 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
864 void wakeup_flusher_threads(long nr_pages
)
866 struct backing_dev_info
*bdi
;
869 nr_pages
= global_page_state(NR_FILE_DIRTY
) +
870 global_page_state(NR_UNSTABLE_NFS
);
874 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
875 if (!bdi_has_dirty_io(bdi
))
877 __bdi_start_writeback(bdi
, nr_pages
, false, false);
882 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
884 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
885 struct dentry
*dentry
;
886 const char *name
= "?";
888 dentry
= d_find_alias(inode
);
890 spin_lock(&dentry
->d_lock
);
891 name
= (const char *) dentry
->d_name
.name
;
894 "%s(%d): dirtied inode %lu (%s) on %s\n",
895 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
896 name
, inode
->i_sb
->s_id
);
898 spin_unlock(&dentry
->d_lock
);
905 * __mark_inode_dirty - internal function
906 * @inode: inode to mark
907 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
908 * Mark an inode as dirty. Callers should use mark_inode_dirty or
909 * mark_inode_dirty_sync.
911 * Put the inode on the super block's dirty list.
913 * CAREFUL! We mark it dirty unconditionally, but move it onto the
914 * dirty list only if it is hashed or if it refers to a blockdev.
915 * If it was not hashed, it will never be added to the dirty list
916 * even if it is later hashed, as it will have been marked dirty already.
918 * In short, make sure you hash any inodes _before_ you start marking
921 * This function *must* be atomic for the I_DIRTY_PAGES case -
922 * set_page_dirty() is called under spinlock in several places.
924 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
925 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
926 * the kernel-internal blockdev inode represents the dirtying time of the
927 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
928 * page->mapping->host, so the page-dirtying time is recorded in the internal
931 void __mark_inode_dirty(struct inode
*inode
, int flags
)
933 struct super_block
*sb
= inode
->i_sb
;
934 struct backing_dev_info
*bdi
= NULL
;
935 bool wakeup_bdi
= false;
938 * Don't do this for I_DIRTY_PAGES - that doesn't actually
939 * dirty the inode itself
941 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) {
942 if (sb
->s_op
->dirty_inode
)
943 sb
->s_op
->dirty_inode(inode
);
947 * make sure that changes are seen by all cpus before we test i_state
952 /* avoid the locking if we can */
953 if ((inode
->i_state
& flags
) == flags
)
956 if (unlikely(block_dump
))
957 block_dump___mark_inode_dirty(inode
);
959 spin_lock(&inode_lock
);
960 if ((inode
->i_state
& flags
) != flags
) {
961 const int was_dirty
= inode
->i_state
& I_DIRTY
;
963 inode
->i_state
|= flags
;
966 * If the inode is being synced, just update its dirty state.
967 * The unlocker will place the inode on the appropriate
968 * superblock list, based upon its state.
970 if (inode
->i_state
& I_SYNC
)
974 * Only add valid (hashed) inodes to the superblock's
975 * dirty list. Add blockdev inodes as well.
977 if (!S_ISBLK(inode
->i_mode
)) {
978 if (hlist_unhashed(&inode
->i_hash
))
981 if (inode
->i_state
& I_FREEING
)
985 * If the inode was already on b_dirty/b_io/b_more_io, don't
986 * reposition it (that would break b_dirty time-ordering).
989 bdi
= inode_to_bdi(inode
);
991 if (bdi_cap_writeback_dirty(bdi
)) {
992 WARN(!test_bit(BDI_registered
, &bdi
->state
),
993 "bdi-%s not registered\n", bdi
->name
);
996 * If this is the first dirty inode for this
997 * bdi, we have to wake-up the corresponding
998 * bdi thread to make sure background
999 * write-back happens later.
1001 if (!wb_has_dirty_io(&bdi
->wb
))
1005 inode
->dirtied_when
= jiffies
;
1006 list_move(&inode
->i_list
, &bdi
->wb
.b_dirty
);
1010 spin_unlock(&inode_lock
);
1013 bdi_wakeup_thread_delayed(bdi
);
1015 EXPORT_SYMBOL(__mark_inode_dirty
);
1018 * Write out a superblock's list of dirty inodes. A wait will be performed
1019 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1021 * If older_than_this is non-NULL, then only write out inodes which
1022 * had their first dirtying at a time earlier than *older_than_this.
1024 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1025 * This function assumes that the blockdev superblock's inodes are backed by
1026 * a variety of queues, so all inodes are searched. For other superblocks,
1027 * assume that all inodes are backed by the same queue.
1029 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1030 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1031 * on the writer throttling path, and we get decent balancing between many
1032 * throttled threads: we don't want them all piling up on inode_sync_wait.
1034 static void wait_sb_inodes(struct super_block
*sb
)
1036 struct inode
*inode
, *old_inode
= NULL
;
1039 * We need to be protected against the filesystem going from
1040 * r/o to r/w or vice versa.
1042 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1044 spin_lock(&inode_lock
);
1047 * Data integrity sync. Must wait for all pages under writeback,
1048 * because there may have been pages dirtied before our sync
1049 * call, but which had writeout started before we write it out.
1050 * In which case, the inode may not be on the dirty list, but
1051 * we still have to wait for that writeout.
1053 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
1054 struct address_space
*mapping
;
1056 if (inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
))
1058 mapping
= inode
->i_mapping
;
1059 if (mapping
->nrpages
== 0)
1062 spin_unlock(&inode_lock
);
1064 * We hold a reference to 'inode' so it couldn't have
1065 * been removed from s_inodes list while we dropped the
1066 * inode_lock. We cannot iput the inode now as we can
1067 * be holding the last reference and we cannot iput it
1068 * under inode_lock. So we keep the reference and iput
1074 filemap_fdatawait(mapping
);
1078 spin_lock(&inode_lock
);
1080 spin_unlock(&inode_lock
);
1085 * writeback_inodes_sb - writeback dirty inodes from given super_block
1086 * @sb: the superblock
1088 * Start writeback on some inodes on this super_block. No guarantees are made
1089 * on how many (if any) will be written, and this function does not wait
1090 * for IO completion of submitted IO. The number of pages submitted is
1093 void writeback_inodes_sb(struct super_block
*sb
)
1095 unsigned long nr_dirty
= global_page_state(NR_FILE_DIRTY
);
1096 unsigned long nr_unstable
= global_page_state(NR_UNSTABLE_NFS
);
1097 DECLARE_COMPLETION_ONSTACK(done
);
1098 struct wb_writeback_work work
= {
1100 .sync_mode
= WB_SYNC_NONE
,
1104 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1106 work
.nr_pages
= nr_dirty
+ nr_unstable
+
1107 (inodes_stat
.nr_inodes
- inodes_stat
.nr_unused
);
1109 bdi_queue_work(sb
->s_bdi
, &work
);
1110 wait_for_completion(&done
);
1112 EXPORT_SYMBOL(writeback_inodes_sb
);
1115 * writeback_inodes_sb_if_idle - start writeback if none underway
1116 * @sb: the superblock
1118 * Invoke writeback_inodes_sb if no writeback is currently underway.
1119 * Returns 1 if writeback was started, 0 if not.
1121 int writeback_inodes_sb_if_idle(struct super_block
*sb
)
1123 if (!writeback_in_progress(sb
->s_bdi
)) {
1124 down_read(&sb
->s_umount
);
1125 writeback_inodes_sb(sb
);
1126 up_read(&sb
->s_umount
);
1131 EXPORT_SYMBOL(writeback_inodes_sb_if_idle
);
1134 * sync_inodes_sb - sync sb inode pages
1135 * @sb: the superblock
1137 * This function writes and waits on any dirty inode belonging to this
1138 * super_block. The number of pages synced is returned.
1140 void sync_inodes_sb(struct super_block
*sb
)
1142 DECLARE_COMPLETION_ONSTACK(done
);
1143 struct wb_writeback_work work
= {
1145 .sync_mode
= WB_SYNC_ALL
,
1146 .nr_pages
= LONG_MAX
,
1151 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
1153 bdi_queue_work(sb
->s_bdi
, &work
);
1154 wait_for_completion(&done
);
1158 EXPORT_SYMBOL(sync_inodes_sb
);
1161 * write_inode_now - write an inode to disk
1162 * @inode: inode to write to disk
1163 * @sync: whether the write should be synchronous or not
1165 * This function commits an inode to disk immediately if it is dirty. This is
1166 * primarily needed by knfsd.
1168 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1170 int write_inode_now(struct inode
*inode
, int sync
)
1173 struct writeback_control wbc
= {
1174 .nr_to_write
= LONG_MAX
,
1175 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
1177 .range_end
= LLONG_MAX
,
1180 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
1181 wbc
.nr_to_write
= 0;
1184 spin_lock(&inode_lock
);
1185 ret
= writeback_single_inode(inode
, &wbc
);
1186 spin_unlock(&inode_lock
);
1188 inode_sync_wait(inode
);
1191 EXPORT_SYMBOL(write_inode_now
);
1194 * sync_inode - write an inode and its pages to disk.
1195 * @inode: the inode to sync
1196 * @wbc: controls the writeback mode
1198 * sync_inode() will write an inode and its pages to disk. It will also
1199 * correctly update the inode on its superblock's dirty inode lists and will
1200 * update inode->i_state.
1202 * The caller must have a ref on the inode.
1204 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1208 spin_lock(&inode_lock
);
1209 ret
= writeback_single_inode(inode
, wbc
);
1210 spin_unlock(&inode_lock
);
1213 EXPORT_SYMBOL(sync_inode
);