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 struct backing_dev_info *bdi = inode_to_bdi(inode);
186 spin_lock(&bdi->wb.list_lock);
187 list_del_init(&inode->i_wb_list);
188 spin_unlock(&bdi->wb.list_lock);
192 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
193 * furthest end of its superblock's dirty-inode list.
195 * Before stamping the inode's ->dirtied_when, we check to see whether it is
196 * already the most-recently-dirtied inode on the b_dirty list. If that is
197 * the case then the inode must have been redirtied while it was being written
198 * out and we don't reset its dirtied_when.
200 static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
202 assert_spin_locked(&wb->list_lock);
203 if (!list_empty(&wb->b_dirty)) {
206 tail = wb_inode(wb->b_dirty.next);
207 if (time_before(inode->dirtied_when, tail->dirtied_when))
208 inode->dirtied_when = jiffies;
210 list_move(&inode->i_wb_list, &wb->b_dirty);
214 * requeue inode for re-scanning after bdi->b_io list is exhausted.
216 static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
218 assert_spin_locked(&wb->list_lock);
219 list_move(&inode->i_wb_list, &wb->b_more_io);
222 static void inode_sync_complete(struct inode *inode)
225 * Prevent speculative execution through
226 * spin_unlock(&wb->list_lock);
230 wake_up_bit(&inode->i_state, __I_SYNC);
233 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
235 bool ret = time_after(inode->dirtied_when, t);
238 * For inodes being constantly redirtied, dirtied_when can get stuck.
239 * It _appears_ to be in the future, but is actually in distant past.
240 * This test is necessary to prevent such wrapped-around relative times
241 * from permanently stopping the whole bdi writeback.
243 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
249 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
251 static void move_expired_inodes(struct list_head *delaying_queue,
252 struct list_head *dispatch_queue,
253 unsigned long *older_than_this)
256 struct list_head *pos, *node;
257 struct super_block *sb = NULL;
261 while (!list_empty(delaying_queue)) {
262 inode = wb_inode(delaying_queue->prev);
263 if (older_than_this &&
264 inode_dirtied_after(inode, *older_than_this))
266 if (sb && sb != inode->i_sb)
269 list_move(&inode->i_wb_list, &tmp);
272 /* just one sb in list, splice to dispatch_queue and we're done */
274 list_splice(&tmp, dispatch_queue);
278 /* Move inodes from one superblock together */
279 while (!list_empty(&tmp)) {
280 sb = wb_inode(tmp.prev)->i_sb;
281 list_for_each_prev_safe(pos, node, &tmp) {
282 inode = wb_inode(pos);
283 if (inode->i_sb == sb)
284 list_move(&inode->i_wb_list, dispatch_queue);
290 * Queue all expired dirty inodes for io, eldest first.
292 * newly dirtied b_dirty b_io b_more_io
293 * =============> gf edc BA
295 * newly dirtied b_dirty b_io b_more_io
296 * =============> g fBAedc
298 * +--> dequeue for IO
300 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
302 assert_spin_locked(&wb->list_lock);
303 list_splice_init(&wb->b_more_io, &wb->b_io);
304 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
307 static int write_inode(struct inode *inode, struct writeback_control *wbc)
309 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
310 return inode->i_sb->s_op->write_inode(inode, wbc);
315 * Wait for writeback on an inode to complete.
317 static void inode_wait_for_writeback(struct inode *inode,
318 struct bdi_writeback *wb)
320 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
321 wait_queue_head_t *wqh;
323 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
324 while (inode->i_state & I_SYNC) {
325 spin_unlock(&inode->i_lock);
326 spin_unlock(&wb->list_lock);
327 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
328 spin_lock(&wb->list_lock);
329 spin_lock(&inode->i_lock);
334 * Write out an inode's dirty pages. Called under wb->list_lock and
335 * inode->i_lock. Either the caller has an active reference on the inode or
336 * the inode has I_WILL_FREE set.
338 * If `wait' is set, wait on the writeout.
340 * The whole writeout design is quite complex and fragile. We want to avoid
341 * starvation of particular inodes when others are being redirtied, prevent
345 writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
346 struct writeback_control *wbc)
348 struct address_space *mapping = inode->i_mapping;
349 long nr_to_write = wbc->nr_to_write;
353 assert_spin_locked(&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) {
371 requeue_io(inode, wb);
372 trace_writeback_single_inode_requeue(inode, wbc,
378 * It's a data-integrity sync. We must wait.
380 inode_wait_for_writeback(inode, wb);
383 BUG_ON(inode->i_state & I_SYNC);
385 /* Set I_SYNC, reset I_DIRTY_PAGES */
386 inode->i_state |= I_SYNC;
387 inode->i_state &= ~I_DIRTY_PAGES;
388 spin_unlock(&inode->i_lock);
389 spin_unlock(&wb->list_lock);
391 ret = do_writepages(mapping, wbc);
394 * Make sure to wait on the data before writing out the metadata.
395 * This is important for filesystems that modify metadata on data
398 if (wbc->sync_mode == WB_SYNC_ALL) {
399 int err = filemap_fdatawait(mapping);
405 * Some filesystems may redirty the inode during the writeback
406 * due to delalloc, clear dirty metadata flags right before
409 spin_lock(&inode->i_lock);
410 dirty = inode->i_state & I_DIRTY;
411 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
412 spin_unlock(&inode->i_lock);
413 /* Don't write the inode if only I_DIRTY_PAGES was set */
414 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
415 int err = write_inode(inode, wbc);
420 spin_lock(&wb->list_lock);
421 spin_lock(&inode->i_lock);
422 inode->i_state &= ~I_SYNC;
423 if (!(inode->i_state & I_FREEING)) {
425 * Sync livelock prevention. Each inode is tagged and synced in
426 * one shot. If still dirty, it will be redirty_tail()'ed below.
427 * Update the dirty time to prevent enqueue and sync it again.
429 if ((inode->i_state & I_DIRTY) &&
430 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
431 inode->dirtied_when = jiffies;
433 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
435 * We didn't write back all the pages. nfs_writepages()
436 * sometimes bales out without doing anything.
438 inode->i_state |= I_DIRTY_PAGES;
439 if (wbc->nr_to_write <= 0) {
441 * slice used up: queue for next turn
443 requeue_io(inode, wb);
446 * Writeback blocked by something other than
447 * congestion. Delay the inode for some time to
448 * avoid spinning on the CPU (100% iowait)
449 * retrying writeback of the dirty page/inode
450 * that cannot be performed immediately.
452 redirty_tail(inode, wb);
454 } else if (inode->i_state & I_DIRTY) {
456 * Filesystems can dirty the inode during writeback
457 * operations, such as delayed allocation during
458 * submission or metadata updates after data IO
461 redirty_tail(inode, wb);
464 * The inode is clean. At this point we either have
465 * a reference to the inode or it's on it's way out.
466 * No need to add it back to the LRU.
468 list_del_init(&inode->i_wb_list);
469 wbc->inodes_written++;
472 inode_sync_complete(inode);
473 trace_writeback_single_inode(inode, wbc, nr_to_write);
478 * For background writeback the caller does not have the sb pinned
479 * before calling writeback. So make sure that we do pin it, so it doesn't
480 * go away while we are writing inodes from it.
482 static bool pin_sb_for_writeback(struct super_block *sb)
485 if (list_empty(&sb->s_instances)) {
486 spin_unlock(&sb_lock);
491 spin_unlock(&sb_lock);
493 if (down_read_trylock(&sb->s_umount)) {
496 up_read(&sb->s_umount);
504 * Write a portion of b_io inodes which belong to @sb.
506 * If @only_this_sb is true, then find and write all such
507 * inodes. Otherwise write only ones which go sequentially
510 * Return 1, if the caller writeback routine should be
511 * interrupted. Otherwise return 0.
513 static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
514 struct writeback_control *wbc, bool only_this_sb)
516 while (!list_empty(&wb->b_io)) {
518 struct inode *inode = wb_inode(wb->b_io.prev);
520 if (inode->i_sb != sb) {
523 * We only want to write back data for this
524 * superblock, move all inodes not belonging
525 * to it back onto the dirty list.
527 redirty_tail(inode, wb);
532 * The inode belongs to a different superblock.
533 * Bounce back to the caller to unpin this and
534 * pin the next superblock.
540 * Don't bother with new inodes or inodes beeing freed, first
541 * kind does not need peridic writeout yet, and for the latter
542 * kind writeout is handled by the freer.
544 spin_lock(&inode->i_lock);
545 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
546 spin_unlock(&inode->i_lock);
547 requeue_io(inode, wb);
553 pages_skipped = wbc->pages_skipped;
554 writeback_single_inode(inode, wb, wbc);
555 if (wbc->pages_skipped != pages_skipped) {
557 * writeback is not making progress due to locked
558 * buffers. Skip this inode for now.
560 redirty_tail(inode, wb);
562 spin_unlock(&inode->i_lock);
563 spin_unlock(&wb->list_lock);
566 spin_lock(&wb->list_lock);
567 if (wbc->nr_to_write <= 0)
574 static void __writeback_inodes_wb(struct bdi_writeback *wb,
575 struct writeback_control *wbc)
579 while (!list_empty(&wb->b_io)) {
580 struct inode *inode = wb_inode(wb->b_io.prev);
581 struct super_block *sb = inode->i_sb;
583 if (!pin_sb_for_writeback(sb)) {
584 requeue_io(inode, wb);
587 ret = writeback_sb_inodes(sb, wb, wbc, false);
593 /* Leave any unwritten inodes on b_io */
596 void writeback_inodes_wb(struct bdi_writeback *wb,
597 struct writeback_control *wbc)
599 spin_lock(&wb->list_lock);
600 if (list_empty(&wb->b_io))
601 queue_io(wb, wbc->older_than_this);
602 __writeback_inodes_wb(wb, wbc);
603 spin_unlock(&wb->list_lock);
607 * The maximum number of pages to writeout in a single bdi flush/kupdate
608 * operation. We do this so we don't hold I_SYNC against an inode for
609 * enormous amounts of time, which would block a userspace task which has
610 * been forced to throttle against that inode. Also, the code reevaluates
611 * the dirty each time it has written this many pages.
613 #define MAX_WRITEBACK_PAGES 1024
615 static inline bool over_bground_thresh(void)
617 unsigned long background_thresh, dirty_thresh;
619 global_dirty_limits(&background_thresh, &dirty_thresh);
621 return (global_page_state(NR_FILE_DIRTY) +
622 global_page_state(NR_UNSTABLE_NFS) > background_thresh);
626 * Explicit flushing or periodic writeback of "old" data.
628 * Define "old": the first time one of an inode's pages is dirtied, we mark the
629 * dirtying-time in the inode's address_space. So this periodic writeback code
630 * just walks the superblock inode list, writing back any inodes which are
631 * older than a specific point in time.
633 * Try to run once per dirty_writeback_interval. But if a writeback event
634 * takes longer than a dirty_writeback_interval interval, then leave a
637 * older_than_this takes precedence over nr_to_write. So we'll only write back
638 * all dirty pages if they are all attached to "old" mappings.
640 static long wb_writeback(struct bdi_writeback *wb,
641 struct wb_writeback_work *work)
643 struct writeback_control wbc = {
644 .sync_mode = work->sync_mode,
645 .tagged_writepages = work->tagged_writepages,
646 .older_than_this = NULL,
647 .for_kupdate = work->for_kupdate,
648 .for_background = work->for_background,
649 .range_cyclic = work->range_cyclic,
651 unsigned long oldest_jif;
653 long write_chunk = MAX_WRITEBACK_PAGES;
656 if (!wbc.range_cyclic) {
658 wbc.range_end = LLONG_MAX;
662 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
663 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
664 * here avoids calling into writeback_inodes_wb() more than once.
666 * The intended call sequence for WB_SYNC_ALL writeback is:
669 * writeback_sb_inodes() <== called only once
670 * write_cache_pages() <== called once for each inode
671 * (quickly) tag currently dirty pages
672 * (maybe slowly) sync all tagged pages
674 if (wbc.sync_mode == WB_SYNC_ALL || wbc.tagged_writepages)
675 write_chunk = LONG_MAX;
677 oldest_jif = jiffies;
678 wbc.older_than_this = &oldest_jif;
680 spin_lock(&wb->list_lock);
683 * Stop writeback when nr_pages has been consumed
685 if (work->nr_pages <= 0)
689 * Background writeout and kupdate-style writeback may
690 * run forever. Stop them if there is other work to do
691 * so that e.g. sync can proceed. They'll be restarted
692 * after the other works are all done.
694 if ((work->for_background || work->for_kupdate) &&
695 !list_empty(&wb->bdi->work_list))
699 * For background writeout, stop when we are below the
700 * background dirty threshold
702 if (work->for_background && !over_bground_thresh())
705 if (work->for_kupdate) {
706 oldest_jif = jiffies -
707 msecs_to_jiffies(dirty_expire_interval * 10);
708 wbc.older_than_this = &oldest_jif;
711 wbc.nr_to_write = write_chunk;
712 wbc.pages_skipped = 0;
713 wbc.inodes_written = 0;
715 trace_wbc_writeback_start(&wbc, wb->bdi);
716 if (list_empty(&wb->b_io))
717 queue_io(wb, wbc.older_than_this);
719 writeback_sb_inodes(work->sb, wb, &wbc, true);
721 __writeback_inodes_wb(wb, &wbc);
722 trace_wbc_writeback_written(&wbc, wb->bdi);
724 work->nr_pages -= write_chunk - wbc.nr_to_write;
725 wrote += write_chunk - wbc.nr_to_write;
728 * Did we write something? Try for more
730 * Dirty inodes are moved to b_io for writeback in batches.
731 * The completion of the current batch does not necessarily
732 * mean the overall work is done. So we keep looping as long
733 * as made some progress on cleaning pages or inodes.
735 if (wbc.nr_to_write < write_chunk)
737 if (wbc.inodes_written)
740 * No more inodes for IO, bail
742 if (list_empty(&wb->b_more_io))
745 * Nothing written. Wait for some inode to
746 * become available for writeback. Otherwise
747 * we'll just busyloop.
749 if (!list_empty(&wb->b_more_io)) {
750 inode = wb_inode(wb->b_more_io.prev);
751 trace_wbc_writeback_wait(&wbc, wb->bdi);
752 spin_lock(&inode->i_lock);
753 inode_wait_for_writeback(inode, wb);
754 spin_unlock(&inode->i_lock);
757 spin_unlock(&wb->list_lock);
763 * Return the next wb_writeback_work struct that hasn't been processed yet.
765 static struct wb_writeback_work *
766 get_next_work_item(struct backing_dev_info *bdi)
768 struct wb_writeback_work *work = NULL;
770 spin_lock_bh(&bdi->wb_lock);
771 if (!list_empty(&bdi->work_list)) {
772 work = list_entry(bdi->work_list.next,
773 struct wb_writeback_work, list);
774 list_del_init(&work->list);
776 spin_unlock_bh(&bdi->wb_lock);
781 * Add in the number of potentially dirty inodes, because each inode
782 * write can dirty pagecache in the underlying blockdev.
784 static unsigned long get_nr_dirty_pages(void)
786 return global_page_state(NR_FILE_DIRTY) +
787 global_page_state(NR_UNSTABLE_NFS) +
788 get_nr_dirty_inodes();
791 static long wb_check_background_flush(struct bdi_writeback *wb)
793 if (over_bground_thresh()) {
795 struct wb_writeback_work work = {
796 .nr_pages = LONG_MAX,
797 .sync_mode = WB_SYNC_NONE,
802 return wb_writeback(wb, &work);
808 static long wb_check_old_data_flush(struct bdi_writeback *wb)
810 unsigned long expired;
814 * When set to zero, disable periodic writeback
816 if (!dirty_writeback_interval)
819 expired = wb->last_old_flush +
820 msecs_to_jiffies(dirty_writeback_interval * 10);
821 if (time_before(jiffies, expired))
824 wb->last_old_flush = jiffies;
825 nr_pages = get_nr_dirty_pages();
828 struct wb_writeback_work work = {
829 .nr_pages = nr_pages,
830 .sync_mode = WB_SYNC_NONE,
835 return wb_writeback(wb, &work);
842 * Retrieve work items and do the writeback they describe
844 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
846 struct backing_dev_info *bdi = wb->bdi;
847 struct wb_writeback_work *work;
850 set_bit(BDI_writeback_running, &wb->bdi->state);
851 while ((work = get_next_work_item(bdi)) != NULL) {
853 * Override sync mode, in case we must wait for completion
854 * because this thread is exiting now.
857 work->sync_mode = WB_SYNC_ALL;
859 trace_writeback_exec(bdi, work);
861 wrote += wb_writeback(wb, work);
864 * Notify the caller of completion if this is a synchronous
865 * work item, otherwise just free it.
868 complete(work->done);
874 * Check for periodic writeback, kupdated() style
876 wrote += wb_check_old_data_flush(wb);
877 wrote += wb_check_background_flush(wb);
878 clear_bit(BDI_writeback_running, &wb->bdi->state);
884 * Handle writeback of dirty data for the device backed by this bdi. Also
885 * wakes up periodically and does kupdated style flushing.
887 int bdi_writeback_thread(void *data)
889 struct bdi_writeback *wb = data;
890 struct backing_dev_info *bdi = wb->bdi;
893 current->flags |= PF_SWAPWRITE;
895 wb->last_active = jiffies;
898 * Our parent may run at a different priority, just set us to normal
900 set_user_nice(current, 0);
902 trace_writeback_thread_start(bdi);
904 while (!kthread_should_stop()) {
906 * Remove own delayed wake-up timer, since we are already awake
907 * and we'll take care of the preriodic write-back.
909 del_timer(&wb->wakeup_timer);
911 pages_written = wb_do_writeback(wb, 0);
913 trace_writeback_pages_written(pages_written);
916 wb->last_active = jiffies;
918 set_current_state(TASK_INTERRUPTIBLE);
919 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
920 __set_current_state(TASK_RUNNING);
924 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
925 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
928 * We have nothing to do, so can go sleep without any
929 * timeout and save power. When a work is queued or
930 * something is made dirty - we will be woken up.
938 /* Flush any work that raced with us exiting */
939 if (!list_empty(&bdi->work_list))
940 wb_do_writeback(wb, 1);
942 trace_writeback_thread_stop(bdi);
948 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
951 void wakeup_flusher_threads(long nr_pages)
953 struct backing_dev_info *bdi;
956 nr_pages = global_page_state(NR_FILE_DIRTY) +
957 global_page_state(NR_UNSTABLE_NFS);
961 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
962 if (!bdi_has_dirty_io(bdi))
964 __bdi_start_writeback(bdi, nr_pages, false);
969 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
971 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
972 struct dentry *dentry;
973 const char *name = "?";
975 dentry = d_find_alias(inode);
977 spin_lock(&dentry->d_lock);
978 name = (const char *) dentry->d_name.name;
981 "%s(%d): dirtied inode %lu (%s) on %s\n",
982 current->comm, task_pid_nr(current), inode->i_ino,
983 name, inode->i_sb->s_id);
985 spin_unlock(&dentry->d_lock);
992 * __mark_inode_dirty - internal function
993 * @inode: inode to mark
994 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
995 * Mark an inode as dirty. Callers should use mark_inode_dirty or
996 * mark_inode_dirty_sync.
998 * Put the inode on the super block's dirty list.
1000 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1001 * dirty list only if it is hashed or if it refers to a blockdev.
1002 * If it was not hashed, it will never be added to the dirty list
1003 * even if it is later hashed, as it will have been marked dirty already.
1005 * In short, make sure you hash any inodes _before_ you start marking
1008 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1009 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1010 * the kernel-internal blockdev inode represents the dirtying time of the
1011 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1012 * page->mapping->host, so the page-dirtying time is recorded in the internal
1015 void __mark_inode_dirty(struct inode *inode, int flags)
1017 struct super_block *sb = inode->i_sb;
1018 struct backing_dev_info *bdi = NULL;
1021 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1022 * dirty the inode itself
1024 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1025 if (sb->s_op->dirty_inode)
1026 sb->s_op->dirty_inode(inode, flags);
1030 * make sure that changes are seen by all cpus before we test i_state
1035 /* avoid the locking if we can */
1036 if ((inode->i_state & flags) == flags)
1039 if (unlikely(block_dump))
1040 block_dump___mark_inode_dirty(inode);
1042 spin_lock(&inode->i_lock);
1043 if ((inode->i_state & flags) != flags) {
1044 const int was_dirty = inode->i_state & I_DIRTY;
1046 inode->i_state |= flags;
1049 * If the inode is being synced, just update its dirty state.
1050 * The unlocker will place the inode on the appropriate
1051 * superblock list, based upon its state.
1053 if (inode->i_state & I_SYNC)
1054 goto out_unlock_inode;
1057 * Only add valid (hashed) inodes to the superblock's
1058 * dirty list. Add blockdev inodes as well.
1060 if (!S_ISBLK(inode->i_mode)) {
1061 if (inode_unhashed(inode))
1062 goto out_unlock_inode;
1064 if (inode->i_state & I_FREEING)
1065 goto out_unlock_inode;
1068 * If the inode was already on b_dirty/b_io/b_more_io, don't
1069 * reposition it (that would break b_dirty time-ordering).
1072 bool wakeup_bdi = false;
1073 bdi = inode_to_bdi(inode);
1075 if (bdi_cap_writeback_dirty(bdi)) {
1076 WARN(!test_bit(BDI_registered, &bdi->state),
1077 "bdi-%s not registered\n", bdi->name);
1080 * If this is the first dirty inode for this
1081 * bdi, we have to wake-up the corresponding
1082 * bdi thread to make sure background
1083 * write-back happens later.
1085 if (!wb_has_dirty_io(&bdi->wb))
1089 spin_unlock(&inode->i_lock);
1090 spin_lock(&bdi->wb.list_lock);
1091 inode->dirtied_when = jiffies;
1092 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1093 spin_unlock(&bdi->wb.list_lock);
1096 bdi_wakeup_thread_delayed(bdi);
1101 spin_unlock(&inode->i_lock);
1104 EXPORT_SYMBOL(__mark_inode_dirty);
1107 * Write out a superblock's list of dirty inodes. A wait will be performed
1108 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1110 * If older_than_this is non-NULL, then only write out inodes which
1111 * had their first dirtying at a time earlier than *older_than_this.
1113 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1114 * This function assumes that the blockdev superblock's inodes are backed by
1115 * a variety of queues, so all inodes are searched. For other superblocks,
1116 * assume that all inodes are backed by the same queue.
1118 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1119 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1120 * on the writer throttling path, and we get decent balancing between many
1121 * throttled threads: we don't want them all piling up on inode_sync_wait.
1123 static void wait_sb_inodes(struct super_block *sb)
1125 struct inode *inode, *old_inode = NULL;
1128 * We need to be protected against the filesystem going from
1129 * r/o to r/w or vice versa.
1131 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1133 spin_lock(&inode_sb_list_lock);
1136 * Data integrity sync. Must wait for all pages under writeback,
1137 * because there may have been pages dirtied before our sync
1138 * call, but which had writeout started before we write it out.
1139 * In which case, the inode may not be on the dirty list, but
1140 * we still have to wait for that writeout.
1142 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1143 struct address_space *mapping = inode->i_mapping;
1145 spin_lock(&inode->i_lock);
1146 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1147 (mapping->nrpages == 0)) {
1148 spin_unlock(&inode->i_lock);
1152 spin_unlock(&inode->i_lock);
1153 spin_unlock(&inode_sb_list_lock);
1156 * We hold a reference to 'inode' so it couldn't have been
1157 * removed from s_inodes list while we dropped the
1158 * inode_sb_list_lock. We cannot iput the inode now as we can
1159 * be holding the last reference and we cannot iput it under
1160 * inode_sb_list_lock. So we keep the reference and iput it
1166 filemap_fdatawait(mapping);
1170 spin_lock(&inode_sb_list_lock);
1172 spin_unlock(&inode_sb_list_lock);
1177 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1178 * @sb: the superblock
1179 * @nr: the number of pages to write
1181 * Start writeback on some inodes on this super_block. No guarantees are made
1182 * on how many (if any) will be written, and this function does not wait
1183 * for IO completion of submitted IO.
1185 void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr)
1187 DECLARE_COMPLETION_ONSTACK(done);
1188 struct wb_writeback_work work = {
1190 .sync_mode = WB_SYNC_NONE,
1191 .tagged_writepages = 1,
1196 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1197 bdi_queue_work(sb->s_bdi, &work);
1198 wait_for_completion(&done);
1200 EXPORT_SYMBOL(writeback_inodes_sb_nr);
1203 * writeback_inodes_sb - writeback dirty inodes from given super_block
1204 * @sb: the superblock
1206 * Start writeback on some inodes on this super_block. No guarantees are made
1207 * on how many (if any) will be written, and this function does not wait
1208 * for IO completion of submitted IO.
1210 void writeback_inodes_sb(struct super_block *sb)
1212 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages());
1214 EXPORT_SYMBOL(writeback_inodes_sb);
1217 * writeback_inodes_sb_if_idle - start writeback if none underway
1218 * @sb: the superblock
1220 * Invoke writeback_inodes_sb if no writeback is currently underway.
1221 * Returns 1 if writeback was started, 0 if not.
1223 int writeback_inodes_sb_if_idle(struct super_block *sb)
1225 if (!writeback_in_progress(sb->s_bdi)) {
1226 down_read(&sb->s_umount);
1227 writeback_inodes_sb(sb);
1228 up_read(&sb->s_umount);
1233 EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1236 * writeback_inodes_sb_if_idle - start writeback if none underway
1237 * @sb: the superblock
1238 * @nr: the number of pages to write
1240 * Invoke writeback_inodes_sb if no writeback is currently underway.
1241 * Returns 1 if writeback was started, 0 if not.
1243 int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
1246 if (!writeback_in_progress(sb->s_bdi)) {
1247 down_read(&sb->s_umount);
1248 writeback_inodes_sb_nr(sb, nr);
1249 up_read(&sb->s_umount);
1254 EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1257 * sync_inodes_sb - sync sb inode pages
1258 * @sb: the superblock
1260 * This function writes and waits on any dirty inode belonging to this
1263 void sync_inodes_sb(struct super_block *sb)
1265 DECLARE_COMPLETION_ONSTACK(done);
1266 struct wb_writeback_work work = {
1268 .sync_mode = WB_SYNC_ALL,
1269 .nr_pages = LONG_MAX,
1274 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1276 bdi_queue_work(sb->s_bdi, &work);
1277 wait_for_completion(&done);
1281 EXPORT_SYMBOL(sync_inodes_sb);
1284 * write_inode_now - write an inode to disk
1285 * @inode: inode to write to disk
1286 * @sync: whether the write should be synchronous or not
1288 * This function commits an inode to disk immediately if it is dirty. This is
1289 * primarily needed by knfsd.
1291 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1293 int write_inode_now(struct inode *inode, int sync)
1295 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1297 struct writeback_control wbc = {
1298 .nr_to_write = LONG_MAX,
1299 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1301 .range_end = LLONG_MAX,
1304 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1305 wbc.nr_to_write = 0;
1308 spin_lock(&wb->list_lock);
1309 spin_lock(&inode->i_lock);
1310 ret = writeback_single_inode(inode, wb, &wbc);
1311 spin_unlock(&inode->i_lock);
1312 spin_unlock(&wb->list_lock);
1314 inode_sync_wait(inode);
1317 EXPORT_SYMBOL(write_inode_now);
1320 * sync_inode - write an inode and its pages to disk.
1321 * @inode: the inode to sync
1322 * @wbc: controls the writeback mode
1324 * sync_inode() will write an inode and its pages to disk. It will also
1325 * correctly update the inode on its superblock's dirty inode lists and will
1326 * update inode->i_state.
1328 * The caller must have a ref on the inode.
1330 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1332 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1335 spin_lock(&wb->list_lock);
1336 spin_lock(&inode->i_lock);
1337 ret = writeback_single_inode(inode, wb, wbc);
1338 spin_unlock(&inode->i_lock);
1339 spin_unlock(&wb->list_lock);
1342 EXPORT_SYMBOL(sync_inode);
1345 * sync_inode_metadata - write an inode to disk
1346 * @inode: the inode to sync
1347 * @wait: wait for I/O to complete.
1349 * Write an inode to disk and adjust its dirty state after completion.
1351 * Note: only writes the actual inode, no associated data or other metadata.
1353 int sync_inode_metadata(struct inode *inode, int wait)
1355 struct writeback_control wbc = {
1356 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1357 .nr_to_write = 0, /* metadata-only */
1360 return sync_inode(inode, &wbc);
1362 EXPORT_SYMBOL(sync_inode_metadata);