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>
31 #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info)
34 * We don't actually have pdflush, but this one is exported though /proc...
36 int nr_pdflush_threads;
39 * Passed into wb_writeback(), essentially a subset of writeback_control
41 struct wb_writeback_args {
43 struct super_block *sb;
44 enum writeback_sync_modes sync_mode;
51 * Work items for the bdi_writeback threads
54 struct list_head list; /* pending work list */
55 struct rcu_head rcu_head; /* for RCU free/clear of work */
57 unsigned long seen; /* threads that have seen this work */
58 atomic_t pending; /* number of threads still to do work */
60 struct wb_writeback_args args; /* writeback arguments */
62 unsigned long state; /* flag bits, see WS_* */
70 #define WS_USED (1 << WS_USED_B)
71 #define WS_ONSTACK (1 << WS_ONSTACK_B)
73 static inline bool bdi_work_on_stack(struct bdi_work *work)
75 return test_bit(WS_ONSTACK_B, &work->state);
78 static inline void bdi_work_init(struct bdi_work *work,
79 struct wb_writeback_args *args)
81 INIT_RCU_HEAD(&work->rcu_head);
83 work->state = WS_USED;
87 * writeback_in_progress - determine whether there is writeback in progress
88 * @bdi: the device's backing_dev_info structure.
90 * Determine whether there is writeback waiting to be handled against a
93 int writeback_in_progress(struct backing_dev_info *bdi)
95 return !list_empty(&bdi->work_list);
98 static void bdi_work_clear(struct bdi_work *work)
100 clear_bit(WS_USED_B, &work->state);
101 smp_mb__after_clear_bit();
103 * work can have disappeared at this point. bit waitq functions
104 * should be able to tolerate this, provided bdi_sched_wait does
105 * not dereference it's pointer argument.
107 wake_up_bit(&work->state, WS_USED_B);
110 static void bdi_work_free(struct rcu_head *head)
112 struct bdi_work *work = container_of(head, struct bdi_work, rcu_head);
114 if (!bdi_work_on_stack(work))
117 bdi_work_clear(work);
120 static void wb_work_complete(struct bdi_work *work)
122 const enum writeback_sync_modes sync_mode = work->args.sync_mode;
123 int onstack = bdi_work_on_stack(work);
126 * For allocated work, we can clear the done/seen bit right here.
127 * For on-stack work, we need to postpone both the clear and free
128 * to after the RCU grace period, since the stack could be invalidated
129 * as soon as bdi_work_clear() has done the wakeup.
132 bdi_work_clear(work);
133 if (sync_mode == WB_SYNC_NONE || onstack)
134 call_rcu(&work->rcu_head, bdi_work_free);
137 static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work)
140 * The caller has retrieved the work arguments from this work,
141 * drop our reference. If this is the last ref, delete and free it
143 if (atomic_dec_and_test(&work->pending)) {
144 struct backing_dev_info *bdi = wb->bdi;
146 spin_lock(&bdi->wb_lock);
147 list_del_rcu(&work->list);
148 spin_unlock(&bdi->wb_lock);
150 wb_work_complete(work);
154 static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work)
156 work->seen = bdi->wb_mask;
158 atomic_set(&work->pending, bdi->wb_cnt);
159 BUG_ON(!bdi->wb_cnt);
162 * list_add_tail_rcu() contains the necessary barriers to
163 * make sure the above stores are seen before the item is
164 * noticed on the list
166 spin_lock(&bdi->wb_lock);
167 list_add_tail_rcu(&work->list, &bdi->work_list);
168 spin_unlock(&bdi->wb_lock);
171 * If the default thread isn't there, make sure we add it. When
172 * it gets created and wakes up, we'll run this work.
174 if (unlikely(list_empty_careful(&bdi->wb_list)))
175 wake_up_process(default_backing_dev_info.wb.task);
177 struct bdi_writeback *wb = &bdi->wb;
180 wake_up_process(wb->task);
185 * Used for on-stack allocated work items. The caller needs to wait until
186 * the wb threads have acked the work before it's safe to continue.
188 static void bdi_wait_on_work_clear(struct bdi_work *work)
190 wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait,
191 TASK_UNINTERRUPTIBLE);
194 static void bdi_alloc_queue_work(struct backing_dev_info *bdi,
195 struct wb_writeback_args *args)
197 struct bdi_work *work;
200 * This is WB_SYNC_NONE writeback, so if allocation fails just
201 * wakeup the thread for old dirty data writeback
203 work = kmalloc(sizeof(*work), GFP_ATOMIC);
205 bdi_work_init(work, args);
206 bdi_queue_work(bdi, work);
208 struct bdi_writeback *wb = &bdi->wb;
211 wake_up_process(wb->task);
216 * bdi_sync_writeback - start and wait for writeback
217 * @bdi: the backing device to write from
218 * @sb: write inodes from this super_block
221 * This does WB_SYNC_ALL data integrity writeback and waits for the
222 * IO to complete. Callers must hold the sb s_umount semaphore for
223 * reading, to avoid having the super disappear before we are done.
225 static void bdi_sync_writeback(struct backing_dev_info *bdi,
226 struct super_block *sb)
228 struct wb_writeback_args args = {
230 .sync_mode = WB_SYNC_ALL,
231 .nr_pages = LONG_MAX,
234 struct bdi_work work;
236 bdi_work_init(&work, &args);
237 work.state |= WS_ONSTACK;
239 bdi_queue_work(bdi, &work);
240 bdi_wait_on_work_clear(&work);
244 * bdi_start_writeback - start writeback
245 * @bdi: the backing device to write from
246 * @sb: write inodes from this super_block
247 * @nr_pages: the number of pages to write
250 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
251 * started when this function returns, we make no guarentees on
252 * completion. Caller need not hold sb s_umount semaphore.
255 void bdi_start_writeback(struct backing_dev_info *bdi, struct super_block *sb,
258 struct wb_writeback_args args = {
260 .sync_mode = WB_SYNC_NONE,
261 .nr_pages = nr_pages,
266 * We treat @nr_pages=0 as the special case to do background writeback,
267 * ie. to sync pages until the background dirty threshold is reached.
270 args.nr_pages = LONG_MAX;
271 args.for_background = 1;
274 bdi_alloc_queue_work(bdi, &args);
278 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
279 * furthest end of its superblock's dirty-inode list.
281 * Before stamping the inode's ->dirtied_when, we check to see whether it is
282 * already the most-recently-dirtied inode on the b_dirty list. If that is
283 * the case then the inode must have been redirtied while it was being written
284 * out and we don't reset its dirtied_when.
286 static void redirty_tail(struct inode *inode)
288 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
290 if (!list_empty(&wb->b_dirty)) {
293 tail = list_entry(wb->b_dirty.next, struct inode, i_list);
294 if (time_before(inode->dirtied_when, tail->dirtied_when))
295 inode->dirtied_when = jiffies;
297 list_move(&inode->i_list, &wb->b_dirty);
301 * requeue inode for re-scanning after bdi->b_io list is exhausted.
303 static void requeue_io(struct inode *inode)
305 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
307 list_move(&inode->i_list, &wb->b_more_io);
310 static void inode_sync_complete(struct inode *inode)
313 * Prevent speculative execution through spin_unlock(&inode_lock);
316 wake_up_bit(&inode->i_state, __I_SYNC);
319 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
321 bool ret = time_after(inode->dirtied_when, t);
324 * For inodes being constantly redirtied, dirtied_when can get stuck.
325 * It _appears_ to be in the future, but is actually in distant past.
326 * This test is necessary to prevent such wrapped-around relative times
327 * from permanently stopping the whole bdi writeback.
329 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
335 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
337 static void move_expired_inodes(struct list_head *delaying_queue,
338 struct list_head *dispatch_queue,
339 unsigned long *older_than_this)
342 struct list_head *pos, *node;
343 struct super_block *sb = NULL;
347 while (!list_empty(delaying_queue)) {
348 inode = list_entry(delaying_queue->prev, struct inode, i_list);
349 if (older_than_this &&
350 inode_dirtied_after(inode, *older_than_this))
352 if (sb && sb != inode->i_sb)
355 list_move(&inode->i_list, &tmp);
358 /* just one sb in list, splice to dispatch_queue and we're done */
360 list_splice(&tmp, dispatch_queue);
364 /* Move inodes from one superblock together */
365 while (!list_empty(&tmp)) {
366 inode = list_entry(tmp.prev, struct inode, i_list);
368 list_for_each_prev_safe(pos, node, &tmp) {
369 inode = list_entry(pos, struct inode, i_list);
370 if (inode->i_sb == sb)
371 list_move(&inode->i_list, dispatch_queue);
377 * Queue all expired dirty inodes for io, eldest first.
379 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
381 list_splice_init(&wb->b_more_io, wb->b_io.prev);
382 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
385 static int write_inode(struct inode *inode, struct writeback_control *wbc)
387 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
388 return inode->i_sb->s_op->write_inode(inode, wbc);
393 * Wait for writeback on an inode to complete.
395 static void inode_wait_for_writeback(struct inode *inode)
397 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
398 wait_queue_head_t *wqh;
400 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
402 spin_unlock(&inode_lock);
403 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
404 spin_lock(&inode_lock);
405 } while (inode->i_state & I_SYNC);
409 * Write out an inode's dirty pages. Called under inode_lock. Either the
410 * caller has ref on the inode (either via __iget or via syscall against an fd)
411 * or the inode has I_WILL_FREE set (via generic_forget_inode)
413 * If `wait' is set, wait on the writeout.
415 * The whole writeout design is quite complex and fragile. We want to avoid
416 * starvation of particular inodes when others are being redirtied, prevent
419 * Called under inode_lock.
422 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
424 struct address_space *mapping = inode->i_mapping;
428 if (!atomic_read(&inode->i_count))
429 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
431 WARN_ON(inode->i_state & I_WILL_FREE);
433 if (inode->i_state & I_SYNC) {
435 * If this inode is locked for writeback and we are not doing
436 * writeback-for-data-integrity, move it to b_more_io so that
437 * writeback can proceed with the other inodes on s_io.
439 * We'll have another go at writing back this inode when we
440 * completed a full scan of b_io.
442 if (wbc->sync_mode != WB_SYNC_ALL) {
448 * It's a data-integrity sync. We must wait.
450 inode_wait_for_writeback(inode);
453 BUG_ON(inode->i_state & I_SYNC);
455 /* Set I_SYNC, reset I_DIRTY */
456 dirty = inode->i_state & I_DIRTY;
457 inode->i_state |= I_SYNC;
458 inode->i_state &= ~I_DIRTY;
460 spin_unlock(&inode_lock);
462 ret = do_writepages(mapping, wbc);
465 * Make sure to wait on the data before writing out the metadata.
466 * This is important for filesystems that modify metadata on data
469 if (wbc->sync_mode == WB_SYNC_ALL) {
470 int err = filemap_fdatawait(mapping);
475 /* Don't write the inode if only I_DIRTY_PAGES was set */
476 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
477 int err = write_inode(inode, wbc);
482 spin_lock(&inode_lock);
483 inode->i_state &= ~I_SYNC;
484 if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
485 if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) {
487 * More pages get dirtied by a fast dirtier.
490 } else if (inode->i_state & I_DIRTY) {
492 * At least XFS will redirty the inode during the
493 * writeback (delalloc) and on io completion (isize).
496 } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
498 * We didn't write back all the pages. nfs_writepages()
499 * sometimes bales out without doing anything. Redirty
500 * the inode; Move it from b_io onto b_more_io/b_dirty.
503 * akpm: if the caller was the kupdate function we put
504 * this inode at the head of b_dirty so it gets first
505 * consideration. Otherwise, move it to the tail, for
506 * the reasons described there. I'm not really sure
507 * how much sense this makes. Presumably I had a good
508 * reasons for doing it this way, and I'd rather not
509 * muck with it at present.
511 if (wbc->for_kupdate) {
513 * For the kupdate function we move the inode
514 * to b_more_io so it will get more writeout as
515 * soon as the queue becomes uncongested.
517 inode->i_state |= I_DIRTY_PAGES;
519 if (wbc->nr_to_write <= 0) {
521 * slice used up: queue for next turn
526 * somehow blocked: retry later
532 * Otherwise fully redirty the inode so that
533 * other inodes on this superblock will get some
534 * writeout. Otherwise heavy writing to one
535 * file would indefinitely suspend writeout of
536 * all the other files.
538 inode->i_state |= I_DIRTY_PAGES;
541 } else if (atomic_read(&inode->i_count)) {
543 * The inode is clean, inuse
545 list_move(&inode->i_list, &inode_in_use);
548 * The inode is clean, unused
550 list_move(&inode->i_list, &inode_unused);
553 inode_sync_complete(inode);
557 static void unpin_sb_for_writeback(struct super_block *sb)
559 up_read(&sb->s_umount);
570 * For WB_SYNC_NONE writeback, the caller does not have the sb pinned
571 * before calling writeback. So make sure that we do pin it, so it doesn't
572 * go away while we are writing inodes from it.
574 static enum sb_pin_state pin_sb_for_writeback(struct writeback_control *wbc,
575 struct super_block *sb)
578 * Caller must already hold the ref for this
580 if (wbc->sync_mode == WB_SYNC_ALL) {
581 WARN_ON(!rwsem_is_locked(&sb->s_umount));
582 return SB_NOT_PINNED;
586 if (down_read_trylock(&sb->s_umount)) {
588 spin_unlock(&sb_lock);
592 * umounted, drop rwsem again and fall through to failure
594 up_read(&sb->s_umount);
597 spin_unlock(&sb_lock);
598 return SB_PIN_FAILED;
602 * Write a portion of b_io inodes which belong to @sb.
603 * If @wbc->sb != NULL, then find and write all such
604 * inodes. Otherwise write only ones which go sequentially
606 * Return 1, if the caller writeback routine should be
607 * interrupted. Otherwise return 0.
609 static int writeback_sb_inodes(struct super_block *sb,
610 struct bdi_writeback *wb,
611 struct writeback_control *wbc)
613 while (!list_empty(&wb->b_io)) {
615 struct inode *inode = list_entry(wb->b_io.prev,
616 struct inode, i_list);
617 if (wbc->sb && sb != inode->i_sb) {
618 /* super block given and doesn't
619 match, skip this inode */
623 if (sb != inode->i_sb)
624 /* finish with this superblock */
626 if (inode->i_state & (I_NEW | I_WILL_FREE)) {
631 * Was this inode dirtied after sync_sb_inodes was called?
632 * This keeps sync from extra jobs and livelock.
634 if (inode_dirtied_after(inode, wbc->wb_start))
637 BUG_ON(inode->i_state & (I_FREEING | I_CLEAR));
639 pages_skipped = wbc->pages_skipped;
640 writeback_single_inode(inode, wbc);
641 if (wbc->pages_skipped != pages_skipped) {
643 * writeback is not making progress due to locked
644 * buffers. Skip this inode for now.
648 spin_unlock(&inode_lock);
651 spin_lock(&inode_lock);
652 if (wbc->nr_to_write <= 0) {
656 if (!list_empty(&wb->b_more_io))
663 static void writeback_inodes_wb(struct bdi_writeback *wb,
664 struct writeback_control *wbc)
668 wbc->wb_start = jiffies; /* livelock avoidance */
669 spin_lock(&inode_lock);
670 if (!wbc->for_kupdate || list_empty(&wb->b_io))
671 queue_io(wb, wbc->older_than_this);
673 while (!list_empty(&wb->b_io)) {
674 struct inode *inode = list_entry(wb->b_io.prev,
675 struct inode, i_list);
676 struct super_block *sb = inode->i_sb;
677 enum sb_pin_state state;
679 if (wbc->sb && sb != wbc->sb) {
680 /* super block given and doesn't
681 match, skip this inode */
685 state = pin_sb_for_writeback(wbc, sb);
687 if (state == SB_PIN_FAILED) {
691 ret = writeback_sb_inodes(sb, wb, wbc);
693 if (state == SB_PINNED)
694 unpin_sb_for_writeback(sb);
698 spin_unlock(&inode_lock);
699 /* Leave any unwritten inodes on b_io */
702 void writeback_inodes_wbc(struct writeback_control *wbc)
704 struct backing_dev_info *bdi = wbc->bdi;
706 writeback_inodes_wb(&bdi->wb, wbc);
710 * The maximum number of pages to writeout in a single bdi flush/kupdate
711 * operation. We do this so we don't hold I_SYNC against an inode for
712 * enormous amounts of time, which would block a userspace task which has
713 * been forced to throttle against that inode. Also, the code reevaluates
714 * the dirty each time it has written this many pages.
716 #define MAX_WRITEBACK_PAGES 1024
718 static inline bool over_bground_thresh(void)
720 unsigned long background_thresh, dirty_thresh;
722 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
724 return (global_page_state(NR_FILE_DIRTY) +
725 global_page_state(NR_UNSTABLE_NFS) >= background_thresh);
729 * Explicit flushing or periodic writeback of "old" data.
731 * Define "old": the first time one of an inode's pages is dirtied, we mark the
732 * dirtying-time in the inode's address_space. So this periodic writeback code
733 * just walks the superblock inode list, writing back any inodes which are
734 * older than a specific point in time.
736 * Try to run once per dirty_writeback_interval. But if a writeback event
737 * takes longer than a dirty_writeback_interval interval, then leave a
740 * older_than_this takes precedence over nr_to_write. So we'll only write back
741 * all dirty pages if they are all attached to "old" mappings.
743 static long wb_writeback(struct bdi_writeback *wb,
744 struct wb_writeback_args *args)
746 struct writeback_control wbc = {
749 .sync_mode = args->sync_mode,
750 .older_than_this = NULL,
751 .for_kupdate = args->for_kupdate,
752 .for_background = args->for_background,
753 .range_cyclic = args->range_cyclic,
755 unsigned long oldest_jif;
759 if (wbc.for_kupdate) {
760 wbc.older_than_this = &oldest_jif;
761 oldest_jif = jiffies -
762 msecs_to_jiffies(dirty_expire_interval * 10);
764 if (!wbc.range_cyclic) {
766 wbc.range_end = LLONG_MAX;
771 * Stop writeback when nr_pages has been consumed
773 if (args->nr_pages <= 0)
777 * For background writeout, stop when we are below the
778 * background dirty threshold
780 if (args->for_background && !over_bground_thresh())
784 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
785 wbc.pages_skipped = 0;
786 writeback_inodes_wb(wb, &wbc);
787 args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
788 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
791 * If we consumed everything, see if we have more
793 if (wbc.nr_to_write <= 0)
796 * Didn't write everything and we don't have more IO, bail
801 * Did we write something? Try for more
803 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
806 * Nothing written. Wait for some inode to
807 * become available for writeback. Otherwise
808 * we'll just busyloop.
810 spin_lock(&inode_lock);
811 if (!list_empty(&wb->b_more_io)) {
812 inode = list_entry(wb->b_more_io.prev,
813 struct inode, i_list);
814 inode_wait_for_writeback(inode);
816 spin_unlock(&inode_lock);
823 * Return the next bdi_work struct that hasn't been processed by this
824 * wb thread yet. ->seen is initially set for each thread that exists
825 * for this device, when a thread first notices a piece of work it
826 * clears its bit. Depending on writeback type, the thread will notify
827 * completion on either receiving the work (WB_SYNC_NONE) or after
828 * it is done (WB_SYNC_ALL).
830 static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi,
831 struct bdi_writeback *wb)
833 struct bdi_work *work, *ret = NULL;
837 list_for_each_entry_rcu(work, &bdi->work_list, list) {
838 if (!test_bit(wb->nr, &work->seen))
840 clear_bit(wb->nr, &work->seen);
850 static long wb_check_old_data_flush(struct bdi_writeback *wb)
852 unsigned long expired;
855 expired = wb->last_old_flush +
856 msecs_to_jiffies(dirty_writeback_interval * 10);
857 if (time_before(jiffies, expired))
860 wb->last_old_flush = jiffies;
861 nr_pages = global_page_state(NR_FILE_DIRTY) +
862 global_page_state(NR_UNSTABLE_NFS) +
863 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
866 struct wb_writeback_args args = {
867 .nr_pages = nr_pages,
868 .sync_mode = WB_SYNC_NONE,
873 return wb_writeback(wb, &args);
880 * Retrieve work items and do the writeback they describe
882 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
884 struct backing_dev_info *bdi = wb->bdi;
885 struct bdi_work *work;
888 while ((work = get_next_work_item(bdi, wb)) != NULL) {
889 struct wb_writeback_args args = work->args;
892 * Override sync mode, in case we must wait for completion
895 work->args.sync_mode = args.sync_mode = WB_SYNC_ALL;
898 * If this isn't a data integrity operation, just notify
899 * that we have seen this work and we are now starting it.
901 if (args.sync_mode == WB_SYNC_NONE)
902 wb_clear_pending(wb, work);
904 wrote += wb_writeback(wb, &args);
907 * This is a data integrity writeback, so only do the
908 * notification when we have completed the work.
910 if (args.sync_mode == WB_SYNC_ALL)
911 wb_clear_pending(wb, work);
915 * Check for periodic writeback, kupdated() style
917 wrote += wb_check_old_data_flush(wb);
923 * Handle writeback of dirty data for the device backed by this bdi. Also
924 * wakes up periodically and does kupdated style flushing.
926 int bdi_writeback_task(struct bdi_writeback *wb)
928 unsigned long last_active = jiffies;
929 unsigned long wait_jiffies = -1UL;
932 while (!kthread_should_stop()) {
933 pages_written = wb_do_writeback(wb, 0);
936 last_active = jiffies;
937 else if (wait_jiffies != -1UL) {
938 unsigned long max_idle;
941 * Longest period of inactivity that we tolerate. If we
942 * see dirty data again later, the task will get
943 * recreated automatically.
945 max_idle = max(5UL * 60 * HZ, wait_jiffies);
946 if (time_after(jiffies, max_idle + last_active))
950 wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10);
951 schedule_timeout_interruptible(wait_jiffies);
959 * Schedule writeback for all backing devices. This does WB_SYNC_NONE
960 * writeback, for integrity writeback see bdi_sync_writeback().
962 static void bdi_writeback_all(struct super_block *sb, long nr_pages)
964 struct wb_writeback_args args = {
966 .nr_pages = nr_pages,
967 .sync_mode = WB_SYNC_NONE,
969 struct backing_dev_info *bdi;
973 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
974 if (!bdi_has_dirty_io(bdi))
977 bdi_alloc_queue_work(bdi, &args);
984 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
987 void wakeup_flusher_threads(long nr_pages)
990 nr_pages = global_page_state(NR_FILE_DIRTY) +
991 global_page_state(NR_UNSTABLE_NFS);
992 bdi_writeback_all(NULL, nr_pages);
995 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
997 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
998 struct dentry *dentry;
999 const char *name = "?";
1001 dentry = d_find_alias(inode);
1003 spin_lock(&dentry->d_lock);
1004 name = (const char *) dentry->d_name.name;
1007 "%s(%d): dirtied inode %lu (%s) on %s\n",
1008 current->comm, task_pid_nr(current), inode->i_ino,
1009 name, inode->i_sb->s_id);
1011 spin_unlock(&dentry->d_lock);
1018 * __mark_inode_dirty - internal function
1019 * @inode: inode to mark
1020 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1021 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1022 * mark_inode_dirty_sync.
1024 * Put the inode on the super block's dirty list.
1026 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1027 * dirty list only if it is hashed or if it refers to a blockdev.
1028 * If it was not hashed, it will never be added to the dirty list
1029 * even if it is later hashed, as it will have been marked dirty already.
1031 * In short, make sure you hash any inodes _before_ you start marking
1034 * This function *must* be atomic for the I_DIRTY_PAGES case -
1035 * set_page_dirty() is called under spinlock in several places.
1037 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1038 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1039 * the kernel-internal blockdev inode represents the dirtying time of the
1040 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1041 * page->mapping->host, so the page-dirtying time is recorded in the internal
1044 void __mark_inode_dirty(struct inode *inode, int flags)
1046 struct super_block *sb = inode->i_sb;
1049 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1050 * dirty the inode itself
1052 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1053 if (sb->s_op->dirty_inode)
1054 sb->s_op->dirty_inode(inode);
1058 * make sure that changes are seen by all cpus before we test i_state
1063 /* avoid the locking if we can */
1064 if ((inode->i_state & flags) == flags)
1067 if (unlikely(block_dump))
1068 block_dump___mark_inode_dirty(inode);
1070 spin_lock(&inode_lock);
1071 if ((inode->i_state & flags) != flags) {
1072 const int was_dirty = inode->i_state & I_DIRTY;
1074 inode->i_state |= flags;
1077 * If the inode is being synced, just update its dirty state.
1078 * The unlocker will place the inode on the appropriate
1079 * superblock list, based upon its state.
1081 if (inode->i_state & I_SYNC)
1085 * Only add valid (hashed) inodes to the superblock's
1086 * dirty list. Add blockdev inodes as well.
1088 if (!S_ISBLK(inode->i_mode)) {
1089 if (hlist_unhashed(&inode->i_hash))
1092 if (inode->i_state & (I_FREEING|I_CLEAR))
1096 * If the inode was already on b_dirty/b_io/b_more_io, don't
1097 * reposition it (that would break b_dirty time-ordering).
1100 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1101 struct backing_dev_info *bdi = wb->bdi;
1103 if (bdi_cap_writeback_dirty(bdi) &&
1104 !test_bit(BDI_registered, &bdi->state)) {
1106 printk(KERN_ERR "bdi-%s not registered\n",
1110 inode->dirtied_when = jiffies;
1111 list_move(&inode->i_list, &wb->b_dirty);
1115 spin_unlock(&inode_lock);
1117 EXPORT_SYMBOL(__mark_inode_dirty);
1120 * Write out a superblock's list of dirty inodes. A wait will be performed
1121 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1123 * If older_than_this is non-NULL, then only write out inodes which
1124 * had their first dirtying at a time earlier than *older_than_this.
1126 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1127 * This function assumes that the blockdev superblock's inodes are backed by
1128 * a variety of queues, so all inodes are searched. For other superblocks,
1129 * assume that all inodes are backed by the same queue.
1131 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1132 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1133 * on the writer throttling path, and we get decent balancing between many
1134 * throttled threads: we don't want them all piling up on inode_sync_wait.
1136 static void wait_sb_inodes(struct super_block *sb)
1138 struct inode *inode, *old_inode = NULL;
1141 * We need to be protected against the filesystem going from
1142 * r/o to r/w or vice versa.
1144 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1146 spin_lock(&inode_lock);
1149 * Data integrity sync. Must wait for all pages under writeback,
1150 * because there may have been pages dirtied before our sync
1151 * call, but which had writeout started before we write it out.
1152 * In which case, the inode may not be on the dirty list, but
1153 * we still have to wait for that writeout.
1155 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1156 struct address_space *mapping;
1158 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
1160 mapping = inode->i_mapping;
1161 if (mapping->nrpages == 0)
1164 spin_unlock(&inode_lock);
1166 * We hold a reference to 'inode' so it couldn't have
1167 * been removed from s_inodes list while we dropped the
1168 * inode_lock. We cannot iput the inode now as we can
1169 * be holding the last reference and we cannot iput it
1170 * under inode_lock. So we keep the reference and iput
1176 filemap_fdatawait(mapping);
1180 spin_lock(&inode_lock);
1182 spin_unlock(&inode_lock);
1187 * writeback_inodes_sb - writeback dirty inodes from given super_block
1188 * @sb: the superblock
1190 * Start writeback on some inodes on this super_block. No guarantees are made
1191 * on how many (if any) will be written, and this function does not wait
1192 * for IO completion of submitted IO. The number of pages submitted is
1195 void writeback_inodes_sb(struct super_block *sb)
1197 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
1198 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
1201 nr_to_write = nr_dirty + nr_unstable +
1202 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
1204 bdi_start_writeback(sb->s_bdi, sb, nr_to_write);
1206 EXPORT_SYMBOL(writeback_inodes_sb);
1209 * writeback_inodes_sb_if_idle - start writeback if none underway
1210 * @sb: the superblock
1212 * Invoke writeback_inodes_sb if no writeback is currently underway.
1213 * Returns 1 if writeback was started, 0 if not.
1215 int writeback_inodes_sb_if_idle(struct super_block *sb)
1217 if (!writeback_in_progress(sb->s_bdi)) {
1218 writeback_inodes_sb(sb);
1223 EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1226 * sync_inodes_sb - sync sb inode pages
1227 * @sb: the superblock
1229 * This function writes and waits on any dirty inode belonging to this
1230 * super_block. The number of pages synced is returned.
1232 void sync_inodes_sb(struct super_block *sb)
1234 bdi_sync_writeback(sb->s_bdi, sb);
1237 EXPORT_SYMBOL(sync_inodes_sb);
1240 * write_inode_now - write an inode to disk
1241 * @inode: inode to write to disk
1242 * @sync: whether the write should be synchronous or not
1244 * This function commits an inode to disk immediately if it is dirty. This is
1245 * primarily needed by knfsd.
1247 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1249 int write_inode_now(struct inode *inode, int sync)
1252 struct writeback_control wbc = {
1253 .nr_to_write = LONG_MAX,
1254 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1256 .range_end = LLONG_MAX,
1259 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1260 wbc.nr_to_write = 0;
1263 spin_lock(&inode_lock);
1264 ret = writeback_single_inode(inode, &wbc);
1265 spin_unlock(&inode_lock);
1267 inode_sync_wait(inode);
1270 EXPORT_SYMBOL(write_inode_now);
1273 * sync_inode - write an inode and its pages to disk.
1274 * @inode: the inode to sync
1275 * @wbc: controls the writeback mode
1277 * sync_inode() will write an inode and its pages to disk. It will also
1278 * correctly update the inode on its superblock's dirty inode lists and will
1279 * update inode->i_state.
1281 * The caller must have a ref on the inode.
1283 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1287 spin_lock(&inode_lock);
1288 ret = writeback_single_inode(inode, wbc);
1289 spin_unlock(&inode_lock);
1292 EXPORT_SYMBOL(sync_inode);