4 * Copyright (C) 2002, Linus Torvalds.
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the
9 * inode itself is not handled here.
11 * 10Apr2002 Andrew Morton
12 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/spinlock.h>
19 #include <linux/sched.h>
22 #include <linux/kthread.h>
23 #include <linux/freezer.h>
24 #include <linux/writeback.h>
25 #include <linux/blkdev.h>
26 #include <linux/backing-dev.h>
27 #include <linux/buffer_head.h>
30 #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info)
33 * We don't actually have pdflush, but this one is exported though /proc...
35 int nr_pdflush_threads;
38 * Passed into wb_writeback(), essentially a subset of writeback_control
40 struct wb_writeback_args {
42 struct super_block *sb;
43 enum writeback_sync_modes sync_mode;
50 * Work items for the bdi_writeback threads
53 struct list_head list; /* pending work list */
54 struct rcu_head rcu_head; /* for RCU free/clear of work */
56 unsigned long seen; /* threads that have seen this work */
57 atomic_t pending; /* number of threads still to do work */
59 struct wb_writeback_args args; /* writeback arguments */
61 unsigned long state; /* flag bits, see WS_* */
69 #define WS_USED (1 << WS_USED_B)
70 #define WS_ONSTACK (1 << WS_ONSTACK_B)
72 static inline bool bdi_work_on_stack(struct bdi_work *work)
74 return test_bit(WS_ONSTACK_B, &work->state);
77 static inline void bdi_work_init(struct bdi_work *work,
78 struct wb_writeback_args *args)
80 INIT_RCU_HEAD(&work->rcu_head);
82 work->state = WS_USED;
86 * writeback_in_progress - determine whether there is writeback in progress
87 * @bdi: the device's backing_dev_info structure.
89 * Determine whether there is writeback waiting to be handled against a
92 int writeback_in_progress(struct backing_dev_info *bdi)
94 return !list_empty(&bdi->work_list);
97 static void bdi_work_clear(struct bdi_work *work)
99 clear_bit(WS_USED_B, &work->state);
100 smp_mb__after_clear_bit();
102 * work can have disappeared at this point. bit waitq functions
103 * should be able to tolerate this, provided bdi_sched_wait does
104 * not dereference it's pointer argument.
106 wake_up_bit(&work->state, WS_USED_B);
109 static void bdi_work_free(struct rcu_head *head)
111 struct bdi_work *work = container_of(head, struct bdi_work, rcu_head);
113 if (!bdi_work_on_stack(work))
116 bdi_work_clear(work);
119 static void wb_work_complete(struct bdi_work *work)
121 const enum writeback_sync_modes sync_mode = work->args.sync_mode;
122 int onstack = bdi_work_on_stack(work);
125 * For allocated work, we can clear the done/seen bit right here.
126 * For on-stack work, we need to postpone both the clear and free
127 * to after the RCU grace period, since the stack could be invalidated
128 * as soon as bdi_work_clear() has done the wakeup.
131 bdi_work_clear(work);
132 if (sync_mode == WB_SYNC_NONE || onstack)
133 call_rcu(&work->rcu_head, bdi_work_free);
136 static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work)
139 * The caller has retrieved the work arguments from this work,
140 * drop our reference. If this is the last ref, delete and free it
142 if (atomic_dec_and_test(&work->pending)) {
143 struct backing_dev_info *bdi = wb->bdi;
145 spin_lock(&bdi->wb_lock);
146 list_del_rcu(&work->list);
147 spin_unlock(&bdi->wb_lock);
149 wb_work_complete(work);
153 static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work)
155 work->seen = bdi->wb_mask;
157 atomic_set(&work->pending, bdi->wb_cnt);
158 BUG_ON(!bdi->wb_cnt);
161 * list_add_tail_rcu() contains the necessary barriers to
162 * make sure the above stores are seen before the item is
163 * noticed on the list
165 spin_lock(&bdi->wb_lock);
166 list_add_tail_rcu(&work->list, &bdi->work_list);
167 spin_unlock(&bdi->wb_lock);
170 * If the default thread isn't there, make sure we add it. When
171 * it gets created and wakes up, we'll run this work.
173 if (unlikely(list_empty_careful(&bdi->wb_list)))
174 wake_up_process(default_backing_dev_info.wb.task);
176 struct bdi_writeback *wb = &bdi->wb;
179 wake_up_process(wb->task);
184 * Used for on-stack allocated work items. The caller needs to wait until
185 * the wb threads have acked the work before it's safe to continue.
187 static void bdi_wait_on_work_clear(struct bdi_work *work)
189 wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait,
190 TASK_UNINTERRUPTIBLE);
193 static void bdi_alloc_queue_work(struct backing_dev_info *bdi,
194 struct wb_writeback_args *args)
196 struct bdi_work *work;
199 * This is WB_SYNC_NONE writeback, so if allocation fails just
200 * wakeup the thread for old dirty data writeback
202 work = kmalloc(sizeof(*work), GFP_ATOMIC);
204 bdi_work_init(work, args);
205 bdi_queue_work(bdi, work);
207 struct bdi_writeback *wb = &bdi->wb;
210 wake_up_process(wb->task);
215 * bdi_sync_writeback - start and wait for writeback
216 * @bdi: the backing device to write from
217 * @sb: write inodes from this super_block
220 * This does WB_SYNC_ALL data integrity writeback and waits for the
221 * IO to complete. Callers must hold the sb s_umount semaphore for
222 * reading, to avoid having the super disappear before we are done.
224 static void bdi_sync_writeback(struct backing_dev_info *bdi,
225 struct super_block *sb)
227 struct wb_writeback_args args = {
229 .sync_mode = WB_SYNC_ALL,
230 .nr_pages = LONG_MAX,
233 struct bdi_work work;
235 bdi_work_init(&work, &args);
236 work.state |= WS_ONSTACK;
238 bdi_queue_work(bdi, &work);
239 bdi_wait_on_work_clear(&work);
243 * bdi_start_writeback - start writeback
244 * @bdi: the backing device to write from
245 * @sb: write inodes from this super_block
246 * @nr_pages: the number of pages to write
249 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
250 * started when this function returns, we make no guarentees on
251 * completion. Caller need not hold sb s_umount semaphore.
254 void bdi_start_writeback(struct backing_dev_info *bdi, struct super_block *sb,
257 struct wb_writeback_args args = {
259 .sync_mode = WB_SYNC_NONE,
260 .nr_pages = nr_pages,
265 * We treat @nr_pages=0 as the special case to do background writeback,
266 * ie. to sync pages until the background dirty threshold is reached.
269 args.nr_pages = LONG_MAX;
270 args.for_background = 1;
273 bdi_alloc_queue_work(bdi, &args);
277 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
278 * furthest end of its superblock's dirty-inode list.
280 * Before stamping the inode's ->dirtied_when, we check to see whether it is
281 * already the most-recently-dirtied inode on the b_dirty list. If that is
282 * the case then the inode must have been redirtied while it was being written
283 * out and we don't reset its dirtied_when.
285 static void redirty_tail(struct inode *inode)
287 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
289 if (!list_empty(&wb->b_dirty)) {
292 tail = list_entry(wb->b_dirty.next, struct inode, i_list);
293 if (time_before(inode->dirtied_when, tail->dirtied_when))
294 inode->dirtied_when = jiffies;
296 list_move(&inode->i_list, &wb->b_dirty);
300 * requeue inode for re-scanning after bdi->b_io list is exhausted.
302 static void requeue_io(struct inode *inode)
304 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
306 list_move(&inode->i_list, &wb->b_more_io);
309 static void inode_sync_complete(struct inode *inode)
312 * Prevent speculative execution through spin_unlock(&inode_lock);
315 wake_up_bit(&inode->i_state, __I_SYNC);
318 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
320 bool ret = time_after(inode->dirtied_when, t);
323 * For inodes being constantly redirtied, dirtied_when can get stuck.
324 * It _appears_ to be in the future, but is actually in distant past.
325 * This test is necessary to prevent such wrapped-around relative times
326 * from permanently stopping the whole bdi writeback.
328 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
334 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
336 static void move_expired_inodes(struct list_head *delaying_queue,
337 struct list_head *dispatch_queue,
338 unsigned long *older_than_this)
341 struct list_head *pos, *node;
342 struct super_block *sb = NULL;
346 while (!list_empty(delaying_queue)) {
347 inode = list_entry(delaying_queue->prev, struct inode, i_list);
348 if (older_than_this &&
349 inode_dirtied_after(inode, *older_than_this))
351 if (sb && sb != inode->i_sb)
354 list_move(&inode->i_list, &tmp);
357 /* just one sb in list, splice to dispatch_queue and we're done */
359 list_splice(&tmp, dispatch_queue);
363 /* Move inodes from one superblock together */
364 while (!list_empty(&tmp)) {
365 inode = list_entry(tmp.prev, struct inode, i_list);
367 list_for_each_prev_safe(pos, node, &tmp) {
368 inode = list_entry(pos, struct inode, i_list);
369 if (inode->i_sb == sb)
370 list_move(&inode->i_list, dispatch_queue);
376 * Queue all expired dirty inodes for io, eldest first.
378 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
380 list_splice_init(&wb->b_more_io, wb->b_io.prev);
381 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
384 static int write_inode(struct inode *inode, struct writeback_control *wbc)
386 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
387 return inode->i_sb->s_op->write_inode(inode, wbc);
392 * Wait for writeback on an inode to complete.
394 static void inode_wait_for_writeback(struct inode *inode)
396 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
397 wait_queue_head_t *wqh;
399 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
401 spin_unlock(&inode_lock);
402 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
403 spin_lock(&inode_lock);
404 } while (inode->i_state & I_SYNC);
408 * Write out an inode's dirty pages. Called under inode_lock. Either the
409 * caller has ref on the inode (either via __iget or via syscall against an fd)
410 * or the inode has I_WILL_FREE set (via generic_forget_inode)
412 * If `wait' is set, wait on the writeout.
414 * The whole writeout design is quite complex and fragile. We want to avoid
415 * starvation of particular inodes when others are being redirtied, prevent
418 * Called under inode_lock.
421 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
423 struct address_space *mapping = inode->i_mapping;
427 if (!atomic_read(&inode->i_count))
428 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
430 WARN_ON(inode->i_state & I_WILL_FREE);
432 if (inode->i_state & I_SYNC) {
434 * If this inode is locked for writeback and we are not doing
435 * writeback-for-data-integrity, move it to b_more_io so that
436 * writeback can proceed with the other inodes on s_io.
438 * We'll have another go at writing back this inode when we
439 * completed a full scan of b_io.
441 if (wbc->sync_mode != WB_SYNC_ALL) {
447 * It's a data-integrity sync. We must wait.
449 inode_wait_for_writeback(inode);
452 BUG_ON(inode->i_state & I_SYNC);
454 /* Set I_SYNC, reset I_DIRTY */
455 dirty = inode->i_state & I_DIRTY;
456 inode->i_state |= I_SYNC;
457 inode->i_state &= ~I_DIRTY;
459 spin_unlock(&inode_lock);
461 ret = do_writepages(mapping, wbc);
464 * Make sure to wait on the data before writing out the metadata.
465 * This is important for filesystems that modify metadata on data
468 if (wbc->sync_mode == WB_SYNC_ALL) {
469 int err = filemap_fdatawait(mapping);
474 /* Don't write the inode if only I_DIRTY_PAGES was set */
475 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
476 int err = write_inode(inode, wbc);
481 spin_lock(&inode_lock);
482 inode->i_state &= ~I_SYNC;
483 if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
484 if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) {
486 * More pages get dirtied by a fast dirtier.
489 } else if (inode->i_state & I_DIRTY) {
491 * At least XFS will redirty the inode during the
492 * writeback (delalloc) and on io completion (isize).
495 } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
497 * We didn't write back all the pages. nfs_writepages()
498 * sometimes bales out without doing anything. Redirty
499 * the inode; Move it from b_io onto b_more_io/b_dirty.
502 * akpm: if the caller was the kupdate function we put
503 * this inode at the head of b_dirty so it gets first
504 * consideration. Otherwise, move it to the tail, for
505 * the reasons described there. I'm not really sure
506 * how much sense this makes. Presumably I had a good
507 * reasons for doing it this way, and I'd rather not
508 * muck with it at present.
510 if (wbc->for_kupdate) {
512 * For the kupdate function we move the inode
513 * to b_more_io so it will get more writeout as
514 * soon as the queue becomes uncongested.
516 inode->i_state |= I_DIRTY_PAGES;
518 if (wbc->nr_to_write <= 0) {
520 * slice used up: queue for next turn
525 * somehow blocked: retry later
531 * Otherwise fully redirty the inode so that
532 * other inodes on this superblock will get some
533 * writeout. Otherwise heavy writing to one
534 * file would indefinitely suspend writeout of
535 * all the other files.
537 inode->i_state |= I_DIRTY_PAGES;
540 } else if (atomic_read(&inode->i_count)) {
542 * The inode is clean, inuse
544 list_move(&inode->i_list, &inode_in_use);
547 * The inode is clean, unused
549 list_move(&inode->i_list, &inode_unused);
552 inode_sync_complete(inode);
556 static void unpin_sb_for_writeback(struct super_block *sb)
558 up_read(&sb->s_umount);
569 * For WB_SYNC_NONE writeback, the caller does not have the sb pinned
570 * before calling writeback. So make sure that we do pin it, so it doesn't
571 * go away while we are writing inodes from it.
573 static enum sb_pin_state pin_sb_for_writeback(struct writeback_control *wbc,
574 struct super_block *sb)
577 * Caller must already hold the ref for this
579 if (wbc->sync_mode == WB_SYNC_ALL) {
580 WARN_ON(!rwsem_is_locked(&sb->s_umount));
581 return SB_NOT_PINNED;
585 if (down_read_trylock(&sb->s_umount)) {
587 spin_unlock(&sb_lock);
591 * umounted, drop rwsem again and fall through to failure
593 up_read(&sb->s_umount);
596 spin_unlock(&sb_lock);
597 return SB_PIN_FAILED;
601 * Write a portion of b_io inodes which belong to @sb.
602 * If @wbc->sb != NULL, then find and write all such
603 * inodes. Otherwise write only ones which go sequentially
605 * Return 1, if the caller writeback routine should be
606 * interrupted. Otherwise return 0.
608 static int writeback_sb_inodes(struct super_block *sb,
609 struct bdi_writeback *wb,
610 struct writeback_control *wbc)
612 while (!list_empty(&wb->b_io)) {
614 struct inode *inode = list_entry(wb->b_io.prev,
615 struct inode, i_list);
616 if (wbc->sb && sb != inode->i_sb) {
617 /* super block given and doesn't
618 match, skip this inode */
622 if (sb != inode->i_sb)
623 /* finish with this superblock */
625 if (inode->i_state & (I_NEW | I_WILL_FREE)) {
630 * Was this inode dirtied after sync_sb_inodes was called?
631 * This keeps sync from extra jobs and livelock.
633 if (inode_dirtied_after(inode, wbc->wb_start))
636 BUG_ON(inode->i_state & (I_FREEING | I_CLEAR));
638 pages_skipped = wbc->pages_skipped;
639 writeback_single_inode(inode, wbc);
640 if (wbc->pages_skipped != pages_skipped) {
642 * writeback is not making progress due to locked
643 * buffers. Skip this inode for now.
647 spin_unlock(&inode_lock);
650 spin_lock(&inode_lock);
651 if (wbc->nr_to_write <= 0) {
655 if (!list_empty(&wb->b_more_io))
662 static void writeback_inodes_wb(struct bdi_writeback *wb,
663 struct writeback_control *wbc)
667 wbc->wb_start = jiffies; /* livelock avoidance */
668 spin_lock(&inode_lock);
669 if (!wbc->for_kupdate || list_empty(&wb->b_io))
670 queue_io(wb, wbc->older_than_this);
672 while (!list_empty(&wb->b_io)) {
673 struct inode *inode = list_entry(wb->b_io.prev,
674 struct inode, i_list);
675 struct super_block *sb = inode->i_sb;
676 enum sb_pin_state state;
678 if (wbc->sb && sb != wbc->sb) {
679 /* super block given and doesn't
680 match, skip this inode */
684 state = pin_sb_for_writeback(wbc, sb);
686 if (state == SB_PIN_FAILED) {
690 ret = writeback_sb_inodes(sb, wb, wbc);
692 if (state == SB_PINNED)
693 unpin_sb_for_writeback(sb);
697 spin_unlock(&inode_lock);
698 /* Leave any unwritten inodes on b_io */
701 void writeback_inodes_wbc(struct writeback_control *wbc)
703 struct backing_dev_info *bdi = wbc->bdi;
705 writeback_inodes_wb(&bdi->wb, wbc);
709 * The maximum number of pages to writeout in a single bdi flush/kupdate
710 * operation. We do this so we don't hold I_SYNC against an inode for
711 * enormous amounts of time, which would block a userspace task which has
712 * been forced to throttle against that inode. Also, the code reevaluates
713 * the dirty each time it has written this many pages.
715 #define MAX_WRITEBACK_PAGES 1024
717 static inline bool over_bground_thresh(void)
719 unsigned long background_thresh, dirty_thresh;
721 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
723 return (global_page_state(NR_FILE_DIRTY) +
724 global_page_state(NR_UNSTABLE_NFS) >= background_thresh);
728 * Explicit flushing or periodic writeback of "old" data.
730 * Define "old": the first time one of an inode's pages is dirtied, we mark the
731 * dirtying-time in the inode's address_space. So this periodic writeback code
732 * just walks the superblock inode list, writing back any inodes which are
733 * older than a specific point in time.
735 * Try to run once per dirty_writeback_interval. But if a writeback event
736 * takes longer than a dirty_writeback_interval interval, then leave a
739 * older_than_this takes precedence over nr_to_write. So we'll only write back
740 * all dirty pages if they are all attached to "old" mappings.
742 static long wb_writeback(struct bdi_writeback *wb,
743 struct wb_writeback_args *args)
745 struct writeback_control wbc = {
748 .sync_mode = args->sync_mode,
749 .older_than_this = NULL,
750 .for_kupdate = args->for_kupdate,
751 .for_background = args->for_background,
752 .range_cyclic = args->range_cyclic,
754 unsigned long oldest_jif;
758 if (wbc.for_kupdate) {
759 wbc.older_than_this = &oldest_jif;
760 oldest_jif = jiffies -
761 msecs_to_jiffies(dirty_expire_interval * 10);
763 if (!wbc.range_cyclic) {
765 wbc.range_end = LLONG_MAX;
770 * Stop writeback when nr_pages has been consumed
772 if (args->nr_pages <= 0)
776 * For background writeout, stop when we are below the
777 * background dirty threshold
779 if (args->for_background && !over_bground_thresh())
783 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
784 wbc.pages_skipped = 0;
785 writeback_inodes_wb(wb, &wbc);
786 args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
787 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
790 * If we consumed everything, see if we have more
792 if (wbc.nr_to_write <= 0)
795 * Didn't write everything and we don't have more IO, bail
800 * Did we write something? Try for more
802 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
805 * Nothing written. Wait for some inode to
806 * become available for writeback. Otherwise
807 * we'll just busyloop.
809 spin_lock(&inode_lock);
810 if (!list_empty(&wb->b_more_io)) {
811 inode = list_entry(wb->b_more_io.prev,
812 struct inode, i_list);
813 inode_wait_for_writeback(inode);
815 spin_unlock(&inode_lock);
822 * Return the next bdi_work struct that hasn't been processed by this
823 * wb thread yet. ->seen is initially set for each thread that exists
824 * for this device, when a thread first notices a piece of work it
825 * clears its bit. Depending on writeback type, the thread will notify
826 * completion on either receiving the work (WB_SYNC_NONE) or after
827 * it is done (WB_SYNC_ALL).
829 static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi,
830 struct bdi_writeback *wb)
832 struct bdi_work *work, *ret = NULL;
836 list_for_each_entry_rcu(work, &bdi->work_list, list) {
837 if (!test_bit(wb->nr, &work->seen))
839 clear_bit(wb->nr, &work->seen);
849 static long wb_check_old_data_flush(struct bdi_writeback *wb)
851 unsigned long expired;
854 expired = wb->last_old_flush +
855 msecs_to_jiffies(dirty_writeback_interval * 10);
856 if (time_before(jiffies, expired))
859 wb->last_old_flush = jiffies;
860 nr_pages = global_page_state(NR_FILE_DIRTY) +
861 global_page_state(NR_UNSTABLE_NFS) +
862 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
865 struct wb_writeback_args args = {
866 .nr_pages = nr_pages,
867 .sync_mode = WB_SYNC_NONE,
872 return wb_writeback(wb, &args);
879 * Retrieve work items and do the writeback they describe
881 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
883 struct backing_dev_info *bdi = wb->bdi;
884 struct bdi_work *work;
887 while ((work = get_next_work_item(bdi, wb)) != NULL) {
888 struct wb_writeback_args args = work->args;
891 * Override sync mode, in case we must wait for completion
894 work->args.sync_mode = args.sync_mode = WB_SYNC_ALL;
897 * If this isn't a data integrity operation, just notify
898 * that we have seen this work and we are now starting it.
900 if (args.sync_mode == WB_SYNC_NONE)
901 wb_clear_pending(wb, work);
903 wrote += wb_writeback(wb, &args);
906 * This is a data integrity writeback, so only do the
907 * notification when we have completed the work.
909 if (args.sync_mode == WB_SYNC_ALL)
910 wb_clear_pending(wb, work);
914 * Check for periodic writeback, kupdated() style
916 wrote += wb_check_old_data_flush(wb);
922 * Handle writeback of dirty data for the device backed by this bdi. Also
923 * wakes up periodically and does kupdated style flushing.
925 int bdi_writeback_task(struct bdi_writeback *wb)
927 unsigned long last_active = jiffies;
928 unsigned long wait_jiffies = -1UL;
931 while (!kthread_should_stop()) {
932 pages_written = wb_do_writeback(wb, 0);
935 last_active = jiffies;
936 else if (wait_jiffies != -1UL) {
937 unsigned long max_idle;
940 * Longest period of inactivity that we tolerate. If we
941 * see dirty data again later, the task will get
942 * recreated automatically.
944 max_idle = max(5UL * 60 * HZ, wait_jiffies);
945 if (time_after(jiffies, max_idle + last_active))
949 wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10);
950 schedule_timeout_interruptible(wait_jiffies);
958 * Schedule writeback for all backing devices. This does WB_SYNC_NONE
959 * writeback, for integrity writeback see bdi_sync_writeback().
961 static void bdi_writeback_all(struct super_block *sb, long nr_pages)
963 struct wb_writeback_args args = {
965 .nr_pages = nr_pages,
966 .sync_mode = WB_SYNC_NONE,
968 struct backing_dev_info *bdi;
972 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
973 if (!bdi_has_dirty_io(bdi))
976 bdi_alloc_queue_work(bdi, &args);
983 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
986 void wakeup_flusher_threads(long nr_pages)
989 nr_pages = global_page_state(NR_FILE_DIRTY) +
990 global_page_state(NR_UNSTABLE_NFS);
991 bdi_writeback_all(NULL, nr_pages);
994 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
996 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
997 struct dentry *dentry;
998 const char *name = "?";
1000 dentry = d_find_alias(inode);
1002 spin_lock(&dentry->d_lock);
1003 name = (const char *) dentry->d_name.name;
1006 "%s(%d): dirtied inode %lu (%s) on %s\n",
1007 current->comm, task_pid_nr(current), inode->i_ino,
1008 name, inode->i_sb->s_id);
1010 spin_unlock(&dentry->d_lock);
1017 * __mark_inode_dirty - internal function
1018 * @inode: inode to mark
1019 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1020 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1021 * mark_inode_dirty_sync.
1023 * Put the inode on the super block's dirty list.
1025 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1026 * dirty list only if it is hashed or if it refers to a blockdev.
1027 * If it was not hashed, it will never be added to the dirty list
1028 * even if it is later hashed, as it will have been marked dirty already.
1030 * In short, make sure you hash any inodes _before_ you start marking
1033 * This function *must* be atomic for the I_DIRTY_PAGES case -
1034 * set_page_dirty() is called under spinlock in several places.
1036 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1037 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1038 * the kernel-internal blockdev inode represents the dirtying time of the
1039 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1040 * page->mapping->host, so the page-dirtying time is recorded in the internal
1043 void __mark_inode_dirty(struct inode *inode, int flags)
1045 struct super_block *sb = inode->i_sb;
1048 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1049 * dirty the inode itself
1051 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1052 if (sb->s_op->dirty_inode)
1053 sb->s_op->dirty_inode(inode);
1057 * make sure that changes are seen by all cpus before we test i_state
1062 /* avoid the locking if we can */
1063 if ((inode->i_state & flags) == flags)
1066 if (unlikely(block_dump))
1067 block_dump___mark_inode_dirty(inode);
1069 spin_lock(&inode_lock);
1070 if ((inode->i_state & flags) != flags) {
1071 const int was_dirty = inode->i_state & I_DIRTY;
1073 inode->i_state |= flags;
1076 * If the inode is being synced, just update its dirty state.
1077 * The unlocker will place the inode on the appropriate
1078 * superblock list, based upon its state.
1080 if (inode->i_state & I_SYNC)
1084 * Only add valid (hashed) inodes to the superblock's
1085 * dirty list. Add blockdev inodes as well.
1087 if (!S_ISBLK(inode->i_mode)) {
1088 if (hlist_unhashed(&inode->i_hash))
1091 if (inode->i_state & (I_FREEING|I_CLEAR))
1095 * If the inode was already on b_dirty/b_io/b_more_io, don't
1096 * reposition it (that would break b_dirty time-ordering).
1099 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1100 struct backing_dev_info *bdi = wb->bdi;
1102 if (bdi_cap_writeback_dirty(bdi) &&
1103 !test_bit(BDI_registered, &bdi->state)) {
1105 printk(KERN_ERR "bdi-%s not registered\n",
1109 inode->dirtied_when = jiffies;
1110 list_move(&inode->i_list, &wb->b_dirty);
1114 spin_unlock(&inode_lock);
1116 EXPORT_SYMBOL(__mark_inode_dirty);
1119 * Write out a superblock's list of dirty inodes. A wait will be performed
1120 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1122 * If older_than_this is non-NULL, then only write out inodes which
1123 * had their first dirtying at a time earlier than *older_than_this.
1125 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1126 * This function assumes that the blockdev superblock's inodes are backed by
1127 * a variety of queues, so all inodes are searched. For other superblocks,
1128 * assume that all inodes are backed by the same queue.
1130 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1131 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1132 * on the writer throttling path, and we get decent balancing between many
1133 * throttled threads: we don't want them all piling up on inode_sync_wait.
1135 static void wait_sb_inodes(struct super_block *sb)
1137 struct inode *inode, *old_inode = NULL;
1140 * We need to be protected against the filesystem going from
1141 * r/o to r/w or vice versa.
1143 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1145 spin_lock(&inode_lock);
1148 * Data integrity sync. Must wait for all pages under writeback,
1149 * because there may have been pages dirtied before our sync
1150 * call, but which had writeout started before we write it out.
1151 * In which case, the inode may not be on the dirty list, but
1152 * we still have to wait for that writeout.
1154 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1155 struct address_space *mapping;
1157 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
1159 mapping = inode->i_mapping;
1160 if (mapping->nrpages == 0)
1163 spin_unlock(&inode_lock);
1165 * We hold a reference to 'inode' so it couldn't have
1166 * been removed from s_inodes list while we dropped the
1167 * inode_lock. We cannot iput the inode now as we can
1168 * be holding the last reference and we cannot iput it
1169 * under inode_lock. So we keep the reference and iput
1175 filemap_fdatawait(mapping);
1179 spin_lock(&inode_lock);
1181 spin_unlock(&inode_lock);
1186 * writeback_inodes_sb - writeback dirty inodes from given super_block
1187 * @sb: the superblock
1189 * Start writeback on some inodes on this super_block. No guarantees are made
1190 * on how many (if any) will be written, and this function does not wait
1191 * for IO completion of submitted IO. The number of pages submitted is
1194 void writeback_inodes_sb(struct super_block *sb)
1196 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
1197 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
1200 nr_to_write = nr_dirty + nr_unstable +
1201 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
1203 bdi_start_writeback(sb->s_bdi, sb, nr_to_write);
1205 EXPORT_SYMBOL(writeback_inodes_sb);
1208 * writeback_inodes_sb_if_idle - start writeback if none underway
1209 * @sb: the superblock
1211 * Invoke writeback_inodes_sb if no writeback is currently underway.
1212 * Returns 1 if writeback was started, 0 if not.
1214 int writeback_inodes_sb_if_idle(struct super_block *sb)
1216 if (!writeback_in_progress(sb->s_bdi)) {
1217 writeback_inodes_sb(sb);
1222 EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1225 * sync_inodes_sb - sync sb inode pages
1226 * @sb: the superblock
1228 * This function writes and waits on any dirty inode belonging to this
1229 * super_block. The number of pages synced is returned.
1231 void sync_inodes_sb(struct super_block *sb)
1233 bdi_sync_writeback(sb->s_bdi, sb);
1236 EXPORT_SYMBOL(sync_inodes_sb);
1239 * write_inode_now - write an inode to disk
1240 * @inode: inode to write to disk
1241 * @sync: whether the write should be synchronous or not
1243 * This function commits an inode to disk immediately if it is dirty. This is
1244 * primarily needed by knfsd.
1246 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1248 int write_inode_now(struct inode *inode, int sync)
1251 struct writeback_control wbc = {
1252 .nr_to_write = LONG_MAX,
1253 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1255 .range_end = LLONG_MAX,
1258 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1259 wbc.nr_to_write = 0;
1262 spin_lock(&inode_lock);
1263 ret = writeback_single_inode(inode, &wbc);
1264 spin_unlock(&inode_lock);
1266 inode_sync_wait(inode);
1269 EXPORT_SYMBOL(write_inode_now);
1272 * sync_inode - write an inode and its pages to disk.
1273 * @inode: the inode to sync
1274 * @wbc: controls the writeback mode
1276 * sync_inode() will write an inode and its pages to disk. It will also
1277 * correctly update the inode on its superblock's dirty inode lists and will
1278 * update inode->i_state.
1280 * The caller must have a ref on the inode.
1282 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1286 spin_lock(&inode_lock);
1287 ret = writeback_single_inode(inode, wbc);
1288 spin_unlock(&inode_lock);
1291 EXPORT_SYMBOL(sync_inode);