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 tagged_writepages:1;
40 unsigned int for_kupdate:1;
41 unsigned int range_cyclic:1;
42 unsigned int for_background:1;
44 struct list_head list; /* pending work list */
45 struct completion *done; /* set if the caller waits */
49 * Include the creation of the trace points after defining the
50 * wb_writeback_work structure so that the definition remains local to this
53 #define CREATE_TRACE_POINTS
54 #include <trace/events/writeback.h>
57 * We don't actually have pdflush, but this one is exported though /proc...
59 int nr_pdflush_threads;
62 * writeback_in_progress - determine whether there is writeback in progress
63 * @bdi: the device's backing_dev_info structure.
65 * Determine whether there is writeback waiting to be handled against a
68 int writeback_in_progress(struct backing_dev_info *bdi)
70 return test_bit(BDI_writeback_running, &bdi->state);
73 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
75 struct super_block *sb = inode->i_sb;
77 if (strcmp(sb->s_type->name, "bdev") == 0)
78 return inode->i_mapping->backing_dev_info;
83 static inline struct inode *wb_inode(struct list_head *head)
85 return list_entry(head, struct inode, i_wb_list);
88 /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
89 static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
92 wake_up_process(bdi->wb.task);
95 * The bdi thread isn't there, wake up the forker thread which
96 * will create and run it.
98 wake_up_process(default_backing_dev_info.wb.task);
102 static void bdi_queue_work(struct backing_dev_info *bdi,
103 struct wb_writeback_work *work)
105 trace_writeback_queue(bdi, work);
107 spin_lock_bh(&bdi->wb_lock);
108 list_add_tail(&work->list, &bdi->work_list);
110 trace_writeback_nothread(bdi, work);
111 bdi_wakeup_flusher(bdi);
112 spin_unlock_bh(&bdi->wb_lock);
116 __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
119 struct wb_writeback_work *work;
122 * This is WB_SYNC_NONE writeback, so if allocation fails just
123 * wakeup the thread for old dirty data writeback
125 work = kzalloc(sizeof(*work), GFP_ATOMIC);
128 trace_writeback_nowork(bdi);
129 wake_up_process(bdi->wb.task);
134 work->sync_mode = WB_SYNC_NONE;
135 work->nr_pages = nr_pages;
136 work->range_cyclic = range_cyclic;
138 bdi_queue_work(bdi, work);
142 * bdi_start_writeback - start writeback
143 * @bdi: the backing device to write from
144 * @nr_pages: the number of pages to write
147 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
148 * started when this function returns, we make no guarantees on
149 * completion. Caller need not hold sb s_umount semaphore.
152 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
154 __bdi_start_writeback(bdi, nr_pages, true);
158 * bdi_start_background_writeback - start background writeback
159 * @bdi: the backing device to write from
162 * This makes sure WB_SYNC_NONE background writeback happens. When
163 * this function returns, it is only guaranteed that for given BDI
164 * some IO is happening if we are over background dirty threshold.
165 * Caller need not hold sb s_umount semaphore.
167 void bdi_start_background_writeback(struct backing_dev_info *bdi)
170 * We just wake up the flusher thread. It will perform background
171 * writeback as soon as there is no other work to do.
173 trace_writeback_wake_background(bdi);
174 spin_lock_bh(&bdi->wb_lock);
175 bdi_wakeup_flusher(bdi);
176 spin_unlock_bh(&bdi->wb_lock);
180 * Remove the inode from the writeback list it is on.
182 void inode_wb_list_del(struct inode *inode)
184 spin_lock(&inode_wb_list_lock);
185 list_del_init(&inode->i_wb_list);
186 spin_unlock(&inode_wb_list_lock);
191 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
192 * furthest end of its superblock's dirty-inode list.
194 * Before stamping the inode's ->dirtied_when, we check to see whether it is
195 * already the most-recently-dirtied inode on the b_dirty list. If that is
196 * the case then the inode must have been redirtied while it was being written
197 * out and we don't reset its dirtied_when.
199 static void redirty_tail(struct inode *inode)
201 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
203 assert_spin_locked(&inode_wb_list_lock);
204 if (!list_empty(&wb->b_dirty)) {
207 tail = wb_inode(wb->b_dirty.next);
208 if (time_before(inode->dirtied_when, tail->dirtied_when))
209 inode->dirtied_when = jiffies;
211 list_move(&inode->i_wb_list, &wb->b_dirty);
215 * requeue inode for re-scanning after bdi->b_io list is exhausted.
217 static void requeue_io(struct inode *inode)
219 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
221 assert_spin_locked(&inode_wb_list_lock);
222 list_move(&inode->i_wb_list, &wb->b_more_io);
225 static void inode_sync_complete(struct inode *inode)
228 * Prevent speculative execution through
229 * spin_unlock(&inode_wb_list_lock);
233 wake_up_bit(&inode->i_state, __I_SYNC);
236 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
238 bool ret = time_after(inode->dirtied_when, t);
241 * For inodes being constantly redirtied, dirtied_when can get stuck.
242 * It _appears_ to be in the future, but is actually in distant past.
243 * This test is necessary to prevent such wrapped-around relative times
244 * from permanently stopping the whole bdi writeback.
246 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
252 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
254 static void move_expired_inodes(struct list_head *delaying_queue,
255 struct list_head *dispatch_queue,
256 unsigned long *older_than_this)
259 struct list_head *pos, *node;
260 struct super_block *sb = NULL;
264 while (!list_empty(delaying_queue)) {
265 inode = wb_inode(delaying_queue->prev);
266 if (older_than_this &&
267 inode_dirtied_after(inode, *older_than_this))
269 if (sb && sb != inode->i_sb)
272 list_move(&inode->i_wb_list, &tmp);
275 /* just one sb in list, splice to dispatch_queue and we're done */
277 list_splice(&tmp, dispatch_queue);
281 /* Move inodes from one superblock together */
282 while (!list_empty(&tmp)) {
283 sb = wb_inode(tmp.prev)->i_sb;
284 list_for_each_prev_safe(pos, node, &tmp) {
285 inode = wb_inode(pos);
286 if (inode->i_sb == sb)
287 list_move(&inode->i_wb_list, dispatch_queue);
293 * Queue all expired dirty inodes for io, eldest first.
295 * newly dirtied b_dirty b_io b_more_io
296 * =============> gf edc BA
298 * newly dirtied b_dirty b_io b_more_io
299 * =============> g fBAedc
301 * +--> dequeue for IO
303 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
305 assert_spin_locked(&inode_wb_list_lock);
306 list_splice_init(&wb->b_more_io, &wb->b_io);
307 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
310 static int write_inode(struct inode *inode, struct writeback_control *wbc)
312 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
313 return inode->i_sb->s_op->write_inode(inode, wbc);
318 * Wait for writeback on an inode to complete.
320 static void inode_wait_for_writeback(struct inode *inode)
322 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
323 wait_queue_head_t *wqh;
325 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
326 while (inode->i_state & I_SYNC) {
327 spin_unlock(&inode->i_lock);
328 spin_unlock(&inode_wb_list_lock);
329 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
330 spin_lock(&inode_wb_list_lock);
331 spin_lock(&inode->i_lock);
336 * Write out an inode's dirty pages. Called under inode_wb_list_lock and
337 * inode->i_lock. Either the caller has an active reference on the inode or
338 * the inode has I_WILL_FREE set.
340 * If `wait' is set, wait on the writeout.
342 * The whole writeout design is quite complex and fragile. We want to avoid
343 * starvation of particular inodes when others are being redirtied, prevent
347 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
349 struct address_space *mapping = inode->i_mapping;
353 assert_spin_locked(&inode_wb_list_lock);
354 assert_spin_locked(&inode->i_lock);
356 if (!atomic_read(&inode->i_count))
357 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
359 WARN_ON(inode->i_state & I_WILL_FREE);
361 if (inode->i_state & I_SYNC) {
363 * If this inode is locked for writeback and we are not doing
364 * writeback-for-data-integrity, move it to b_more_io so that
365 * writeback can proceed with the other inodes on s_io.
367 * We'll have another go at writing back this inode when we
368 * completed a full scan of b_io.
370 if (wbc->sync_mode != WB_SYNC_ALL) {
376 * It's a data-integrity sync. We must wait.
378 inode_wait_for_writeback(inode);
381 BUG_ON(inode->i_state & I_SYNC);
383 /* Set I_SYNC, reset I_DIRTY_PAGES */
384 inode->i_state |= I_SYNC;
385 inode->i_state &= ~I_DIRTY_PAGES;
386 spin_unlock(&inode->i_lock);
387 spin_unlock(&inode_wb_list_lock);
389 ret = do_writepages(mapping, wbc);
392 * Make sure to wait on the data before writing out the metadata.
393 * This is important for filesystems that modify metadata on data
396 if (wbc->sync_mode == WB_SYNC_ALL) {
397 int err = filemap_fdatawait(mapping);
403 * Some filesystems may redirty the inode during the writeback
404 * due to delalloc, clear dirty metadata flags right before
407 spin_lock(&inode->i_lock);
408 dirty = inode->i_state & I_DIRTY;
409 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
410 spin_unlock(&inode->i_lock);
411 /* Don't write the inode if only I_DIRTY_PAGES was set */
412 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
413 int err = write_inode(inode, wbc);
418 spin_lock(&inode_wb_list_lock);
419 spin_lock(&inode->i_lock);
420 inode->i_state &= ~I_SYNC;
421 if (!(inode->i_state & I_FREEING)) {
423 * Sync livelock prevention. Each inode is tagged and synced in
424 * one shot. If still dirty, it will be redirty_tail()'ed below.
425 * Update the dirty time to prevent enqueue and sync it again.
427 if ((inode->i_state & I_DIRTY) &&
428 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
429 inode->dirtied_when = jiffies;
431 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
433 * We didn't write back all the pages. nfs_writepages()
434 * sometimes bales out without doing anything.
436 inode->i_state |= I_DIRTY_PAGES;
437 if (wbc->nr_to_write <= 0) {
439 * slice used up: queue for next turn
444 * Writeback blocked by something other than
445 * congestion. Delay the inode for some time to
446 * avoid spinning on the CPU (100% iowait)
447 * retrying writeback of the dirty page/inode
448 * that cannot be performed immediately.
452 } else if (inode->i_state & I_DIRTY) {
454 * Filesystems can dirty the inode during writeback
455 * operations, such as delayed allocation during
456 * submission or metadata updates after data IO
462 * The inode is clean. At this point we either have
463 * a reference to the inode or it's on it's way out.
464 * No need to add it back to the LRU.
466 list_del_init(&inode->i_wb_list);
467 wbc->inodes_written++;
470 inode_sync_complete(inode);
475 * For background writeback the caller does not have the sb pinned
476 * before calling writeback. So make sure that we do pin it, so it doesn't
477 * go away while we are writing inodes from it.
479 static bool pin_sb_for_writeback(struct super_block *sb)
482 if (list_empty(&sb->s_instances)) {
483 spin_unlock(&sb_lock);
488 spin_unlock(&sb_lock);
490 if (down_read_trylock(&sb->s_umount)) {
493 up_read(&sb->s_umount);
501 * Write a portion of b_io inodes which belong to @sb.
503 * If @only_this_sb is true, then find and write all such
504 * inodes. Otherwise write only ones which go sequentially
507 * Return 1, if the caller writeback routine should be
508 * interrupted. Otherwise return 0.
510 static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
511 struct writeback_control *wbc, bool only_this_sb)
513 while (!list_empty(&wb->b_io)) {
515 struct inode *inode = wb_inode(wb->b_io.prev);
517 if (inode->i_sb != sb) {
520 * We only want to write back data for this
521 * superblock, move all inodes not belonging
522 * to it back onto the dirty list.
529 * The inode belongs to a different superblock.
530 * Bounce back to the caller to unpin this and
531 * pin the next superblock.
537 * Don't bother with new inodes or inodes beeing freed, first
538 * kind does not need peridic writeout yet, and for the latter
539 * kind writeout is handled by the freer.
541 spin_lock(&inode->i_lock);
542 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
543 spin_unlock(&inode->i_lock);
549 * Was this inode dirtied after sync_sb_inodes was called?
550 * This keeps sync from extra jobs and livelock.
552 if (inode_dirtied_after(inode, wbc->wb_start)) {
553 spin_unlock(&inode->i_lock);
559 pages_skipped = wbc->pages_skipped;
560 writeback_single_inode(inode, wbc);
561 if (wbc->pages_skipped != pages_skipped) {
563 * writeback is not making progress due to locked
564 * buffers. Skip this inode for now.
568 spin_unlock(&inode->i_lock);
569 spin_unlock(&inode_wb_list_lock);
572 spin_lock(&inode_wb_list_lock);
573 if (wbc->nr_to_write <= 0) {
577 if (!list_empty(&wb->b_more_io))
584 void writeback_inodes_wb(struct bdi_writeback *wb,
585 struct writeback_control *wbc)
590 wbc->wb_start = jiffies; /* livelock avoidance */
591 spin_lock(&inode_wb_list_lock);
592 if (!wbc->for_kupdate || list_empty(&wb->b_io))
593 queue_io(wb, wbc->older_than_this);
595 while (!list_empty(&wb->b_io)) {
596 struct inode *inode = wb_inode(wb->b_io.prev);
597 struct super_block *sb = inode->i_sb;
599 if (!pin_sb_for_writeback(sb)) {
603 ret = writeback_sb_inodes(sb, wb, wbc, false);
609 spin_unlock(&inode_wb_list_lock);
610 /* Leave any unwritten inodes on b_io */
613 static void __writeback_inodes_sb(struct super_block *sb,
614 struct bdi_writeback *wb, struct writeback_control *wbc)
616 WARN_ON(!rwsem_is_locked(&sb->s_umount));
618 spin_lock(&inode_wb_list_lock);
619 if (!wbc->for_kupdate || list_empty(&wb->b_io))
620 queue_io(wb, wbc->older_than_this);
621 writeback_sb_inodes(sb, wb, wbc, true);
622 spin_unlock(&inode_wb_list_lock);
626 * The maximum number of pages to writeout in a single bdi flush/kupdate
627 * operation. We do this so we don't hold I_SYNC against an inode for
628 * enormous amounts of time, which would block a userspace task which has
629 * been forced to throttle against that inode. Also, the code reevaluates
630 * the dirty each time it has written this many pages.
632 #define MAX_WRITEBACK_PAGES 1024
634 static inline bool over_bground_thresh(void)
636 unsigned long background_thresh, dirty_thresh;
638 global_dirty_limits(&background_thresh, &dirty_thresh);
640 return (global_page_state(NR_FILE_DIRTY) +
641 global_page_state(NR_UNSTABLE_NFS) > background_thresh);
645 * Explicit flushing or periodic writeback of "old" data.
647 * Define "old": the first time one of an inode's pages is dirtied, we mark the
648 * dirtying-time in the inode's address_space. So this periodic writeback code
649 * just walks the superblock inode list, writing back any inodes which are
650 * older than a specific point in time.
652 * Try to run once per dirty_writeback_interval. But if a writeback event
653 * takes longer than a dirty_writeback_interval interval, then leave a
656 * older_than_this takes precedence over nr_to_write. So we'll only write back
657 * all dirty pages if they are all attached to "old" mappings.
659 static long wb_writeback(struct bdi_writeback *wb,
660 struct wb_writeback_work *work)
662 struct writeback_control wbc = {
663 .sync_mode = work->sync_mode,
664 .tagged_writepages = work->tagged_writepages,
665 .older_than_this = NULL,
666 .for_kupdate = work->for_kupdate,
667 .for_background = work->for_background,
668 .range_cyclic = work->range_cyclic,
670 unsigned long oldest_jif;
672 long write_chunk = MAX_WRITEBACK_PAGES;
675 if (!wbc.range_cyclic) {
677 wbc.range_end = LLONG_MAX;
681 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
682 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
683 * here avoids calling into writeback_inodes_wb() more than once.
685 * The intended call sequence for WB_SYNC_ALL writeback is:
688 * __writeback_inodes_sb() <== called only once
689 * write_cache_pages() <== called once for each inode
690 * (quickly) tag currently dirty pages
691 * (maybe slowly) sync all tagged pages
693 if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_writepages)
694 write_chunk = LONG_MAX;
696 wbc.wb_start = jiffies; /* livelock avoidance */
699 * Stop writeback when nr_pages has been consumed
701 if (work->nr_pages <= 0)
705 * Background writeout and kupdate-style writeback may
706 * run forever. Stop them if there is other work to do
707 * so that e.g. sync can proceed. They'll be restarted
708 * after the other works are all done.
710 if ((work->for_background || work->for_kupdate) &&
711 !list_empty(&wb->bdi->work_list))
715 * For background writeout, stop when we are below the
716 * background dirty threshold
718 if (work->for_background && !over_bground_thresh())
721 if (work->for_kupdate) {
722 oldest_jif = jiffies -
723 msecs_to_jiffies(dirty_expire_interval * 10);
724 wbc.older_than_this = &oldest_jif;
728 wbc.nr_to_write = write_chunk;
729 wbc.pages_skipped = 0;
730 wbc.inodes_written = 0;
732 trace_wbc_writeback_start(&wbc, wb->bdi);
734 __writeback_inodes_sb(work->sb, wb, &wbc);
736 writeback_inodes_wb(wb, &wbc);
737 trace_wbc_writeback_written(&wbc, wb->bdi);
739 work->nr_pages -= write_chunk - wbc.nr_to_write;
740 wrote += write_chunk - wbc.nr_to_write;
743 * Did we write something? Try for more
745 * Dirty inodes are moved to b_io for writeback in batches.
746 * The completion of the current batch does not necessarily
747 * mean the overall work is done. So we keep looping as long
748 * as made some progress on cleaning pages or inodes.
750 if (wbc.nr_to_write < write_chunk)
752 if (wbc.inodes_written)
755 * No more inodes for IO, bail
760 * Nothing written. Wait for some inode to
761 * become available for writeback. Otherwise
762 * we'll just busyloop.
764 spin_lock(&inode_wb_list_lock);
765 if (!list_empty(&wb->b_more_io)) {
766 inode = wb_inode(wb->b_more_io.prev);
767 trace_wbc_writeback_wait(&wbc, wb->bdi);
768 spin_lock(&inode->i_lock);
769 inode_wait_for_writeback(inode);
770 spin_unlock(&inode->i_lock);
772 spin_unlock(&inode_wb_list_lock);
779 * Return the next wb_writeback_work struct that hasn't been processed yet.
781 static struct wb_writeback_work *
782 get_next_work_item(struct backing_dev_info *bdi)
784 struct wb_writeback_work *work = NULL;
786 spin_lock_bh(&bdi->wb_lock);
787 if (!list_empty(&bdi->work_list)) {
788 work = list_entry(bdi->work_list.next,
789 struct wb_writeback_work, list);
790 list_del_init(&work->list);
792 spin_unlock_bh(&bdi->wb_lock);
797 * Add in the number of potentially dirty inodes, because each inode
798 * write can dirty pagecache in the underlying blockdev.
800 static unsigned long get_nr_dirty_pages(void)
802 return global_page_state(NR_FILE_DIRTY) +
803 global_page_state(NR_UNSTABLE_NFS) +
804 get_nr_dirty_inodes();
807 static long wb_check_background_flush(struct bdi_writeback *wb)
809 if (over_bground_thresh()) {
811 struct wb_writeback_work work = {
812 .nr_pages = LONG_MAX,
813 .sync_mode = WB_SYNC_NONE,
818 return wb_writeback(wb, &work);
824 static long wb_check_old_data_flush(struct bdi_writeback *wb)
826 unsigned long expired;
830 * When set to zero, disable periodic writeback
832 if (!dirty_writeback_interval)
835 expired = wb->last_old_flush +
836 msecs_to_jiffies(dirty_writeback_interval * 10);
837 if (time_before(jiffies, expired))
840 wb->last_old_flush = jiffies;
841 nr_pages = get_nr_dirty_pages();
844 struct wb_writeback_work work = {
845 .nr_pages = nr_pages,
846 .sync_mode = WB_SYNC_NONE,
851 return wb_writeback(wb, &work);
858 * Retrieve work items and do the writeback they describe
860 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
862 struct backing_dev_info *bdi = wb->bdi;
863 struct wb_writeback_work *work;
866 set_bit(BDI_writeback_running, &wb->bdi->state);
867 while ((work = get_next_work_item(bdi)) != NULL) {
869 * Override sync mode, in case we must wait for completion
870 * because this thread is exiting now.
873 work->sync_mode = WB_SYNC_ALL;
875 trace_writeback_exec(bdi, work);
877 wrote += wb_writeback(wb, work);
880 * Notify the caller of completion if this is a synchronous
881 * work item, otherwise just free it.
884 complete(work->done);
890 * Check for periodic writeback, kupdated() style
892 wrote += wb_check_old_data_flush(wb);
893 wrote += wb_check_background_flush(wb);
894 clear_bit(BDI_writeback_running, &wb->bdi->state);
900 * Handle writeback of dirty data for the device backed by this bdi. Also
901 * wakes up periodically and does kupdated style flushing.
903 int bdi_writeback_thread(void *data)
905 struct bdi_writeback *wb = data;
906 struct backing_dev_info *bdi = wb->bdi;
909 current->flags |= PF_SWAPWRITE;
911 wb->last_active = jiffies;
914 * Our parent may run at a different priority, just set us to normal
916 set_user_nice(current, 0);
918 trace_writeback_thread_start(bdi);
920 while (!kthread_should_stop()) {
922 * Remove own delayed wake-up timer, since we are already awake
923 * and we'll take care of the preriodic write-back.
925 del_timer(&wb->wakeup_timer);
927 pages_written = wb_do_writeback(wb, 0);
929 trace_writeback_pages_written(pages_written);
932 wb->last_active = jiffies;
934 set_current_state(TASK_INTERRUPTIBLE);
935 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
936 __set_current_state(TASK_RUNNING);
940 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
941 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
944 * We have nothing to do, so can go sleep without any
945 * timeout and save power. When a work is queued or
946 * something is made dirty - we will be woken up.
954 /* Flush any work that raced with us exiting */
955 if (!list_empty(&bdi->work_list))
956 wb_do_writeback(wb, 1);
958 trace_writeback_thread_stop(bdi);
964 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
967 void wakeup_flusher_threads(long nr_pages)
969 struct backing_dev_info *bdi;
972 nr_pages = global_page_state(NR_FILE_DIRTY) +
973 global_page_state(NR_UNSTABLE_NFS);
977 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
978 if (!bdi_has_dirty_io(bdi))
980 __bdi_start_writeback(bdi, nr_pages, false);
985 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
987 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
988 struct dentry *dentry;
989 const char *name = "?";
991 dentry = d_find_alias(inode);
993 spin_lock(&dentry->d_lock);
994 name = (const char *) dentry->d_name.name;
997 "%s(%d): dirtied inode %lu (%s) on %s\n",
998 current->comm, task_pid_nr(current), inode->i_ino,
999 name, inode->i_sb->s_id);
1001 spin_unlock(&dentry->d_lock);
1008 * __mark_inode_dirty - internal function
1009 * @inode: inode to mark
1010 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1011 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1012 * mark_inode_dirty_sync.
1014 * Put the inode on the super block's dirty list.
1016 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1017 * dirty list only if it is hashed or if it refers to a blockdev.
1018 * If it was not hashed, it will never be added to the dirty list
1019 * even if it is later hashed, as it will have been marked dirty already.
1021 * In short, make sure you hash any inodes _before_ you start marking
1024 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1025 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1026 * the kernel-internal blockdev inode represents the dirtying time of the
1027 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1028 * page->mapping->host, so the page-dirtying time is recorded in the internal
1031 void __mark_inode_dirty(struct inode *inode, int flags)
1033 struct super_block *sb = inode->i_sb;
1034 struct backing_dev_info *bdi = NULL;
1037 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1038 * dirty the inode itself
1040 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1041 if (sb->s_op->dirty_inode)
1042 sb->s_op->dirty_inode(inode, flags);
1046 * make sure that changes are seen by all cpus before we test i_state
1051 /* avoid the locking if we can */
1052 if ((inode->i_state & flags) == flags)
1055 if (unlikely(block_dump))
1056 block_dump___mark_inode_dirty(inode);
1058 spin_lock(&inode->i_lock);
1059 if ((inode->i_state & flags) != flags) {
1060 const int was_dirty = inode->i_state & I_DIRTY;
1062 inode->i_state |= flags;
1065 * If the inode is being synced, just update its dirty state.
1066 * The unlocker will place the inode on the appropriate
1067 * superblock list, based upon its state.
1069 if (inode->i_state & I_SYNC)
1070 goto out_unlock_inode;
1073 * Only add valid (hashed) inodes to the superblock's
1074 * dirty list. Add blockdev inodes as well.
1076 if (!S_ISBLK(inode->i_mode)) {
1077 if (inode_unhashed(inode))
1078 goto out_unlock_inode;
1080 if (inode->i_state & I_FREEING)
1081 goto out_unlock_inode;
1084 * If the inode was already on b_dirty/b_io/b_more_io, don't
1085 * reposition it (that would break b_dirty time-ordering).
1088 bool wakeup_bdi = false;
1089 bdi = inode_to_bdi(inode);
1091 if (bdi_cap_writeback_dirty(bdi)) {
1092 WARN(!test_bit(BDI_registered, &bdi->state),
1093 "bdi-%s not registered\n", bdi->name);
1096 * If this is the first dirty inode for this
1097 * bdi, we have to wake-up the corresponding
1098 * bdi thread to make sure background
1099 * write-back happens later.
1101 if (!wb_has_dirty_io(&bdi->wb))
1105 spin_unlock(&inode->i_lock);
1106 spin_lock(&inode_wb_list_lock);
1107 inode->dirtied_when = jiffies;
1108 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1109 spin_unlock(&inode_wb_list_lock);
1112 bdi_wakeup_thread_delayed(bdi);
1117 spin_unlock(&inode->i_lock);
1120 EXPORT_SYMBOL(__mark_inode_dirty);
1123 * Write out a superblock's list of dirty inodes. A wait will be performed
1124 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1126 * If older_than_this is non-NULL, then only write out inodes which
1127 * had their first dirtying at a time earlier than *older_than_this.
1129 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1130 * This function assumes that the blockdev superblock's inodes are backed by
1131 * a variety of queues, so all inodes are searched. For other superblocks,
1132 * assume that all inodes are backed by the same queue.
1134 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1135 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1136 * on the writer throttling path, and we get decent balancing between many
1137 * throttled threads: we don't want them all piling up on inode_sync_wait.
1139 static void wait_sb_inodes(struct super_block *sb)
1141 struct inode *inode, *old_inode = NULL;
1144 * We need to be protected against the filesystem going from
1145 * r/o to r/w or vice versa.
1147 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1149 spin_lock(&inode_sb_list_lock);
1152 * Data integrity sync. Must wait for all pages under writeback,
1153 * because there may have been pages dirtied before our sync
1154 * call, but which had writeout started before we write it out.
1155 * In which case, the inode may not be on the dirty list, but
1156 * we still have to wait for that writeout.
1158 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1159 struct address_space *mapping = inode->i_mapping;
1161 spin_lock(&inode->i_lock);
1162 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1163 (mapping->nrpages == 0)) {
1164 spin_unlock(&inode->i_lock);
1168 spin_unlock(&inode->i_lock);
1169 spin_unlock(&inode_sb_list_lock);
1172 * We hold a reference to 'inode' so it couldn't have been
1173 * removed from s_inodes list while we dropped the
1174 * inode_sb_list_lock. We cannot iput the inode now as we can
1175 * be holding the last reference and we cannot iput it under
1176 * inode_sb_list_lock. So we keep the reference and iput it
1182 filemap_fdatawait(mapping);
1186 spin_lock(&inode_sb_list_lock);
1188 spin_unlock(&inode_sb_list_lock);
1193 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1194 * @sb: the superblock
1195 * @nr: the number of pages to write
1197 * Start writeback on some inodes on this super_block. No guarantees are made
1198 * on how many (if any) will be written, and this function does not wait
1199 * for IO completion of submitted IO.
1201 void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr)
1203 DECLARE_COMPLETION_ONSTACK(done);
1204 struct wb_writeback_work work = {
1206 .sync_mode = WB_SYNC_NONE,
1207 .tagged_writepages = 1,
1212 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1213 bdi_queue_work(sb->s_bdi, &work);
1214 wait_for_completion(&done);
1216 EXPORT_SYMBOL(writeback_inodes_sb_nr);
1219 * writeback_inodes_sb - writeback dirty inodes from given super_block
1220 * @sb: the superblock
1222 * Start writeback on some inodes on this super_block. No guarantees are made
1223 * on how many (if any) will be written, and this function does not wait
1224 * for IO completion of submitted IO.
1226 void writeback_inodes_sb(struct super_block *sb)
1228 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages());
1230 EXPORT_SYMBOL(writeback_inodes_sb);
1233 * writeback_inodes_sb_if_idle - start writeback if none underway
1234 * @sb: the superblock
1236 * Invoke writeback_inodes_sb if no writeback is currently underway.
1237 * Returns 1 if writeback was started, 0 if not.
1239 int writeback_inodes_sb_if_idle(struct super_block *sb)
1241 if (!writeback_in_progress(sb->s_bdi)) {
1242 down_read(&sb->s_umount);
1243 writeback_inodes_sb(sb);
1244 up_read(&sb->s_umount);
1249 EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1252 * writeback_inodes_sb_if_idle - start writeback if none underway
1253 * @sb: the superblock
1254 * @nr: the number of pages to write
1256 * Invoke writeback_inodes_sb if no writeback is currently underway.
1257 * Returns 1 if writeback was started, 0 if not.
1259 int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
1262 if (!writeback_in_progress(sb->s_bdi)) {
1263 down_read(&sb->s_umount);
1264 writeback_inodes_sb_nr(sb, nr);
1265 up_read(&sb->s_umount);
1270 EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1273 * sync_inodes_sb - sync sb inode pages
1274 * @sb: the superblock
1276 * This function writes and waits on any dirty inode belonging to this
1279 void sync_inodes_sb(struct super_block *sb)
1281 DECLARE_COMPLETION_ONSTACK(done);
1282 struct wb_writeback_work work = {
1284 .sync_mode = WB_SYNC_ALL,
1285 .nr_pages = LONG_MAX,
1290 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1292 bdi_queue_work(sb->s_bdi, &work);
1293 wait_for_completion(&done);
1297 EXPORT_SYMBOL(sync_inodes_sb);
1300 * write_inode_now - write an inode to disk
1301 * @inode: inode to write to disk
1302 * @sync: whether the write should be synchronous or not
1304 * This function commits an inode to disk immediately if it is dirty. This is
1305 * primarily needed by knfsd.
1307 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1309 int write_inode_now(struct inode *inode, int sync)
1312 struct writeback_control wbc = {
1313 .nr_to_write = LONG_MAX,
1314 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1316 .range_end = LLONG_MAX,
1319 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1320 wbc.nr_to_write = 0;
1323 spin_lock(&inode_wb_list_lock);
1324 spin_lock(&inode->i_lock);
1325 ret = writeback_single_inode(inode, &wbc);
1326 spin_unlock(&inode->i_lock);
1327 spin_unlock(&inode_wb_list_lock);
1329 inode_sync_wait(inode);
1332 EXPORT_SYMBOL(write_inode_now);
1335 * sync_inode - write an inode and its pages to disk.
1336 * @inode: the inode to sync
1337 * @wbc: controls the writeback mode
1339 * sync_inode() will write an inode and its pages to disk. It will also
1340 * correctly update the inode on its superblock's dirty inode lists and will
1341 * update inode->i_state.
1343 * The caller must have a ref on the inode.
1345 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1349 spin_lock(&inode_wb_list_lock);
1350 spin_lock(&inode->i_lock);
1351 ret = writeback_single_inode(inode, wbc);
1352 spin_unlock(&inode->i_lock);
1353 spin_unlock(&inode_wb_list_lock);
1356 EXPORT_SYMBOL(sync_inode);
1359 * sync_inode_metadata - write an inode to disk
1360 * @inode: the inode to sync
1361 * @wait: wait for I/O to complete.
1363 * Write an inode to disk and adjust its dirty state after completion.
1365 * Note: only writes the actual inode, no associated data or other metadata.
1367 int sync_inode_metadata(struct inode *inode, int wait)
1369 struct writeback_control wbc = {
1370 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1371 .nr_to_write = 0, /* metadata-only */
1374 return sync_inode(inode, &wbc);
1376 EXPORT_SYMBOL(sync_inode_metadata);