4 * Copyright (C) 2002, Linus Torvalds.
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the
9 * inode itself is not handled here.
11 * 10Apr2002 Andrew Morton
12 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/spinlock.h>
19 #include <linux/slab.h>
20 #include <linux/sched.h>
23 #include <linux/kthread.h>
24 #include <linux/freezer.h>
25 #include <linux/writeback.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/buffer_head.h>
29 #include <linux/tracepoint.h>
33 * Passed into wb_writeback(), essentially a subset of writeback_control
35 struct wb_writeback_work {
37 struct super_block *sb;
38 enum writeback_sync_modes sync_mode;
39 unsigned int for_kupdate:1;
40 unsigned int range_cyclic:1;
41 unsigned int for_background:1;
43 struct list_head list; /* pending work list */
44 struct completion *done; /* set if the caller waits */
48 * Include the creation of the trace points after defining the
49 * wb_writeback_work structure so that the definition remains local to this
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/writeback.h>
56 * We don't actually have pdflush, but this one is exported though /proc...
58 int nr_pdflush_threads;
61 * writeback_in_progress - determine whether there is writeback in progress
62 * @bdi: the device's backing_dev_info structure.
64 * Determine whether there is writeback waiting to be handled against a
67 int writeback_in_progress(struct backing_dev_info *bdi)
69 return test_bit(BDI_writeback_running, &bdi->state);
72 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
74 struct super_block *sb = inode->i_sb;
76 if (strcmp(sb->s_type->name, "bdev") == 0)
77 return inode->i_mapping->backing_dev_info;
82 static inline struct inode *wb_inode(struct list_head *head)
84 return list_entry(head, struct inode, i_wb_list);
87 /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
88 static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
91 wake_up_process(bdi->wb.task);
94 * The bdi thread isn't there, wake up the forker thread which
95 * will create and run it.
97 wake_up_process(default_backing_dev_info.wb.task);
101 static void bdi_queue_work(struct backing_dev_info *bdi,
102 struct wb_writeback_work *work)
104 trace_writeback_queue(bdi, work);
106 spin_lock_bh(&bdi->wb_lock);
107 list_add_tail(&work->list, &bdi->work_list);
109 trace_writeback_nothread(bdi, work);
110 bdi_wakeup_flusher(bdi);
111 spin_unlock_bh(&bdi->wb_lock);
115 __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
118 struct wb_writeback_work *work;
121 * This is WB_SYNC_NONE writeback, so if allocation fails just
122 * wakeup the thread for old dirty data writeback
124 work = kzalloc(sizeof(*work), GFP_ATOMIC);
127 trace_writeback_nowork(bdi);
128 wake_up_process(bdi->wb.task);
133 work->sync_mode = WB_SYNC_NONE;
134 work->nr_pages = nr_pages;
135 work->range_cyclic = range_cyclic;
137 bdi_queue_work(bdi, work);
141 * bdi_start_writeback - start writeback
142 * @bdi: the backing device to write from
143 * @nr_pages: the number of pages to write
146 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
147 * started when this function returns, we make no guarentees on
148 * completion. Caller need not hold sb s_umount semaphore.
151 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
153 __bdi_start_writeback(bdi, nr_pages, true);
157 * bdi_start_background_writeback - start background writeback
158 * @bdi: the backing device to write from
161 * This makes sure WB_SYNC_NONE background writeback happens. When
162 * this function returns, it is only guaranteed that for given BDI
163 * some IO is happening if we are over background dirty threshold.
164 * Caller need not hold sb s_umount semaphore.
166 void bdi_start_background_writeback(struct backing_dev_info *bdi)
169 * We just wake up the flusher thread. It will perform background
170 * writeback as soon as there is no other work to do.
172 trace_writeback_wake_background(bdi);
173 spin_lock_bh(&bdi->wb_lock);
174 bdi_wakeup_flusher(bdi);
175 spin_unlock_bh(&bdi->wb_lock);
179 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
180 * furthest end of its superblock's dirty-inode list.
182 * Before stamping the inode's ->dirtied_when, we check to see whether it is
183 * already the most-recently-dirtied inode on the b_dirty list. If that is
184 * the case then the inode must have been redirtied while it was being written
185 * out and we don't reset its dirtied_when.
187 static void redirty_tail(struct inode *inode)
189 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
191 if (!list_empty(&wb->b_dirty)) {
194 tail = wb_inode(wb->b_dirty.next);
195 if (time_before(inode->dirtied_when, tail->dirtied_when))
196 inode->dirtied_when = jiffies;
198 list_move(&inode->i_wb_list, &wb->b_dirty);
202 * requeue inode for re-scanning after bdi->b_io list is exhausted.
204 static void requeue_io(struct inode *inode)
206 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
208 list_move(&inode->i_wb_list, &wb->b_more_io);
211 static void inode_sync_complete(struct inode *inode)
214 * Prevent speculative execution through spin_unlock(&inode_lock);
217 wake_up_bit(&inode->i_state, __I_SYNC);
220 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
222 bool ret = time_after(inode->dirtied_when, t);
225 * For inodes being constantly redirtied, dirtied_when can get stuck.
226 * It _appears_ to be in the future, but is actually in distant past.
227 * This test is necessary to prevent such wrapped-around relative times
228 * from permanently stopping the whole bdi writeback.
230 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
236 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
238 static void move_expired_inodes(struct list_head *delaying_queue,
239 struct list_head *dispatch_queue,
240 unsigned long *older_than_this)
243 struct list_head *pos, *node;
244 struct super_block *sb = NULL;
248 while (!list_empty(delaying_queue)) {
249 inode = wb_inode(delaying_queue->prev);
250 if (older_than_this &&
251 inode_dirtied_after(inode, *older_than_this))
253 if (sb && sb != inode->i_sb)
256 list_move(&inode->i_wb_list, &tmp);
259 /* just one sb in list, splice to dispatch_queue and we're done */
261 list_splice(&tmp, dispatch_queue);
265 /* Move inodes from one superblock together */
266 while (!list_empty(&tmp)) {
267 sb = wb_inode(tmp.prev)->i_sb;
268 list_for_each_prev_safe(pos, node, &tmp) {
269 inode = wb_inode(pos);
270 if (inode->i_sb == sb)
271 list_move(&inode->i_wb_list, dispatch_queue);
277 * Queue all expired dirty inodes for io, eldest first.
279 * newly dirtied b_dirty b_io b_more_io
280 * =============> gf edc BA
282 * newly dirtied b_dirty b_io b_more_io
283 * =============> g fBAedc
285 * +--> dequeue for IO
287 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
289 list_splice_init(&wb->b_more_io, &wb->b_io);
290 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
293 static int write_inode(struct inode *inode, struct writeback_control *wbc)
295 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
296 return inode->i_sb->s_op->write_inode(inode, wbc);
301 * Wait for writeback on an inode to complete.
303 static void inode_wait_for_writeback(struct inode *inode)
305 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
306 wait_queue_head_t *wqh;
308 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
309 while (inode->i_state & I_SYNC) {
310 spin_unlock(&inode->i_lock);
311 spin_unlock(&inode_lock);
312 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
313 spin_lock(&inode_lock);
314 spin_lock(&inode->i_lock);
319 * Write out an inode's dirty pages. Called under inode_lock. Either the
320 * caller has ref on the inode (either via __iget or via syscall against an fd)
321 * or the inode has I_WILL_FREE set (via generic_forget_inode)
323 * If `wait' is set, wait on the writeout.
325 * The whole writeout design is quite complex and fragile. We want to avoid
326 * starvation of particular inodes when others are being redirtied, prevent
329 * Called under inode_lock.
332 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
334 struct address_space *mapping = inode->i_mapping;
338 spin_lock(&inode->i_lock);
339 if (!atomic_read(&inode->i_count))
340 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
342 WARN_ON(inode->i_state & I_WILL_FREE);
344 if (inode->i_state & I_SYNC) {
346 * If this inode is locked for writeback and we are not doing
347 * writeback-for-data-integrity, move it to b_more_io so that
348 * writeback can proceed with the other inodes on s_io.
350 * We'll have another go at writing back this inode when we
351 * completed a full scan of b_io.
353 if (wbc->sync_mode != WB_SYNC_ALL) {
354 spin_unlock(&inode->i_lock);
360 * It's a data-integrity sync. We must wait.
362 inode_wait_for_writeback(inode);
365 BUG_ON(inode->i_state & I_SYNC);
367 /* Set I_SYNC, reset I_DIRTY_PAGES */
368 inode->i_state |= I_SYNC;
369 inode->i_state &= ~I_DIRTY_PAGES;
370 spin_unlock(&inode->i_lock);
371 spin_unlock(&inode_lock);
373 ret = do_writepages(mapping, wbc);
376 * Make sure to wait on the data before writing out the metadata.
377 * This is important for filesystems that modify metadata on data
380 if (wbc->sync_mode == WB_SYNC_ALL) {
381 int err = filemap_fdatawait(mapping);
387 * Some filesystems may redirty the inode during the writeback
388 * due to delalloc, clear dirty metadata flags right before
391 spin_lock(&inode_lock);
392 spin_lock(&inode->i_lock);
393 dirty = inode->i_state & I_DIRTY;
394 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
395 spin_unlock(&inode->i_lock);
396 spin_unlock(&inode_lock);
397 /* Don't write the inode if only I_DIRTY_PAGES was set */
398 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
399 int err = write_inode(inode, wbc);
404 spin_lock(&inode_lock);
405 spin_lock(&inode->i_lock);
406 inode->i_state &= ~I_SYNC;
407 if (!(inode->i_state & I_FREEING)) {
408 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
410 * We didn't write back all the pages. nfs_writepages()
411 * sometimes bales out without doing anything.
413 inode->i_state |= I_DIRTY_PAGES;
414 if (wbc->nr_to_write <= 0) {
416 * slice used up: queue for next turn
421 * Writeback blocked by something other than
422 * congestion. Delay the inode for some time to
423 * avoid spinning on the CPU (100% iowait)
424 * retrying writeback of the dirty page/inode
425 * that cannot be performed immediately.
429 } else if (inode->i_state & I_DIRTY) {
431 * Filesystems can dirty the inode during writeback
432 * operations, such as delayed allocation during
433 * submission or metadata updates after data IO
439 * The inode is clean. At this point we either have
440 * a reference to the inode or it's on it's way out.
441 * No need to add it back to the LRU.
443 list_del_init(&inode->i_wb_list);
446 inode_sync_complete(inode);
447 spin_unlock(&inode->i_lock);
452 * For background writeback the caller does not have the sb pinned
453 * before calling writeback. So make sure that we do pin it, so it doesn't
454 * go away while we are writing inodes from it.
456 static bool pin_sb_for_writeback(struct super_block *sb)
459 if (list_empty(&sb->s_instances)) {
460 spin_unlock(&sb_lock);
465 spin_unlock(&sb_lock);
467 if (down_read_trylock(&sb->s_umount)) {
470 up_read(&sb->s_umount);
478 * Write a portion of b_io inodes which belong to @sb.
480 * If @only_this_sb is true, then find and write all such
481 * inodes. Otherwise write only ones which go sequentially
484 * Return 1, if the caller writeback routine should be
485 * interrupted. Otherwise return 0.
487 static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
488 struct writeback_control *wbc, bool only_this_sb)
490 while (!list_empty(&wb->b_io)) {
492 struct inode *inode = wb_inode(wb->b_io.prev);
494 if (inode->i_sb != sb) {
497 * We only want to write back data for this
498 * superblock, move all inodes not belonging
499 * to it back onto the dirty list.
506 * The inode belongs to a different superblock.
507 * Bounce back to the caller to unpin this and
508 * pin the next superblock.
514 * Don't bother with new inodes or inodes beeing freed, first
515 * kind does not need peridic writeout yet, and for the latter
516 * kind writeout is handled by the freer.
518 spin_lock(&inode->i_lock);
519 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
520 spin_unlock(&inode->i_lock);
526 * Was this inode dirtied after sync_sb_inodes was called?
527 * This keeps sync from extra jobs and livelock.
529 if (inode_dirtied_after(inode, wbc->wb_start)) {
530 spin_unlock(&inode->i_lock);
535 spin_unlock(&inode->i_lock);
537 pages_skipped = wbc->pages_skipped;
538 writeback_single_inode(inode, wbc);
539 if (wbc->pages_skipped != pages_skipped) {
541 * writeback is not making progress due to locked
542 * buffers. Skip this inode for now.
546 spin_unlock(&inode_lock);
549 spin_lock(&inode_lock);
550 if (wbc->nr_to_write <= 0) {
554 if (!list_empty(&wb->b_more_io))
561 void writeback_inodes_wb(struct bdi_writeback *wb,
562 struct writeback_control *wbc)
567 wbc->wb_start = jiffies; /* livelock avoidance */
568 spin_lock(&inode_lock);
569 if (!wbc->for_kupdate || list_empty(&wb->b_io))
570 queue_io(wb, wbc->older_than_this);
572 while (!list_empty(&wb->b_io)) {
573 struct inode *inode = wb_inode(wb->b_io.prev);
574 struct super_block *sb = inode->i_sb;
576 if (!pin_sb_for_writeback(sb)) {
580 ret = writeback_sb_inodes(sb, wb, wbc, false);
586 spin_unlock(&inode_lock);
587 /* Leave any unwritten inodes on b_io */
590 static void __writeback_inodes_sb(struct super_block *sb,
591 struct bdi_writeback *wb, struct writeback_control *wbc)
593 WARN_ON(!rwsem_is_locked(&sb->s_umount));
595 spin_lock(&inode_lock);
596 if (!wbc->for_kupdate || list_empty(&wb->b_io))
597 queue_io(wb, wbc->older_than_this);
598 writeback_sb_inodes(sb, wb, wbc, true);
599 spin_unlock(&inode_lock);
603 * The maximum number of pages to writeout in a single bdi flush/kupdate
604 * operation. We do this so we don't hold I_SYNC against an inode for
605 * enormous amounts of time, which would block a userspace task which has
606 * been forced to throttle against that inode. Also, the code reevaluates
607 * the dirty each time it has written this many pages.
609 #define MAX_WRITEBACK_PAGES 1024
611 static inline bool over_bground_thresh(void)
613 unsigned long background_thresh, dirty_thresh;
615 global_dirty_limits(&background_thresh, &dirty_thresh);
617 return (global_page_state(NR_FILE_DIRTY) +
618 global_page_state(NR_UNSTABLE_NFS) > background_thresh);
622 * Explicit flushing or periodic writeback of "old" data.
624 * Define "old": the first time one of an inode's pages is dirtied, we mark the
625 * dirtying-time in the inode's address_space. So this periodic writeback code
626 * just walks the superblock inode list, writing back any inodes which are
627 * older than a specific point in time.
629 * Try to run once per dirty_writeback_interval. But if a writeback event
630 * takes longer than a dirty_writeback_interval interval, then leave a
633 * older_than_this takes precedence over nr_to_write. So we'll only write back
634 * all dirty pages if they are all attached to "old" mappings.
636 static long wb_writeback(struct bdi_writeback *wb,
637 struct wb_writeback_work *work)
639 struct writeback_control wbc = {
640 .sync_mode = work->sync_mode,
641 .older_than_this = NULL,
642 .for_kupdate = work->for_kupdate,
643 .for_background = work->for_background,
644 .range_cyclic = work->range_cyclic,
646 unsigned long oldest_jif;
651 if (wbc.for_kupdate) {
652 wbc.older_than_this = &oldest_jif;
653 oldest_jif = jiffies -
654 msecs_to_jiffies(dirty_expire_interval * 10);
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_inodes_sb() <== 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_NONE)
675 write_chunk = MAX_WRITEBACK_PAGES;
677 write_chunk = LONG_MAX;
679 wbc.wb_start = jiffies; /* livelock avoidance */
682 * Stop writeback when nr_pages has been consumed
684 if (work->nr_pages <= 0)
688 * Background writeout and kupdate-style writeback may
689 * run forever. Stop them if there is other work to do
690 * so that e.g. sync can proceed. They'll be restarted
691 * after the other works are all done.
693 if ((work->for_background || work->for_kupdate) &&
694 !list_empty(&wb->bdi->work_list))
698 * For background writeout, stop when we are below the
699 * background dirty threshold
701 if (work->for_background && !over_bground_thresh())
705 wbc.nr_to_write = write_chunk;
706 wbc.pages_skipped = 0;
708 trace_wbc_writeback_start(&wbc, wb->bdi);
710 __writeback_inodes_sb(work->sb, wb, &wbc);
712 writeback_inodes_wb(wb, &wbc);
713 trace_wbc_writeback_written(&wbc, wb->bdi);
715 work->nr_pages -= write_chunk - wbc.nr_to_write;
716 wrote += write_chunk - wbc.nr_to_write;
719 * If we consumed everything, see if we have more
721 if (wbc.nr_to_write <= 0)
724 * Didn't write everything and we don't have more IO, bail
729 * Did we write something? Try for more
731 if (wbc.nr_to_write < write_chunk)
734 * Nothing written. Wait for some inode to
735 * become available for writeback. Otherwise
736 * we'll just busyloop.
738 spin_lock(&inode_lock);
739 if (!list_empty(&wb->b_more_io)) {
740 inode = wb_inode(wb->b_more_io.prev);
741 trace_wbc_writeback_wait(&wbc, wb->bdi);
742 spin_lock(&inode->i_lock);
743 inode_wait_for_writeback(inode);
744 spin_unlock(&inode->i_lock);
746 spin_unlock(&inode_lock);
753 * Return the next wb_writeback_work struct that hasn't been processed yet.
755 static struct wb_writeback_work *
756 get_next_work_item(struct backing_dev_info *bdi)
758 struct wb_writeback_work *work = NULL;
760 spin_lock_bh(&bdi->wb_lock);
761 if (!list_empty(&bdi->work_list)) {
762 work = list_entry(bdi->work_list.next,
763 struct wb_writeback_work, list);
764 list_del_init(&work->list);
766 spin_unlock_bh(&bdi->wb_lock);
771 * Add in the number of potentially dirty inodes, because each inode
772 * write can dirty pagecache in the underlying blockdev.
774 static unsigned long get_nr_dirty_pages(void)
776 return global_page_state(NR_FILE_DIRTY) +
777 global_page_state(NR_UNSTABLE_NFS) +
778 get_nr_dirty_inodes();
781 static long wb_check_background_flush(struct bdi_writeback *wb)
783 if (over_bground_thresh()) {
785 struct wb_writeback_work work = {
786 .nr_pages = LONG_MAX,
787 .sync_mode = WB_SYNC_NONE,
792 return wb_writeback(wb, &work);
798 static long wb_check_old_data_flush(struct bdi_writeback *wb)
800 unsigned long expired;
804 * When set to zero, disable periodic writeback
806 if (!dirty_writeback_interval)
809 expired = wb->last_old_flush +
810 msecs_to_jiffies(dirty_writeback_interval * 10);
811 if (time_before(jiffies, expired))
814 wb->last_old_flush = jiffies;
815 nr_pages = get_nr_dirty_pages();
818 struct wb_writeback_work work = {
819 .nr_pages = nr_pages,
820 .sync_mode = WB_SYNC_NONE,
825 return wb_writeback(wb, &work);
832 * Retrieve work items and do the writeback they describe
834 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
836 struct backing_dev_info *bdi = wb->bdi;
837 struct wb_writeback_work *work;
840 set_bit(BDI_writeback_running, &wb->bdi->state);
841 while ((work = get_next_work_item(bdi)) != NULL) {
843 * Override sync mode, in case we must wait for completion
844 * because this thread is exiting now.
847 work->sync_mode = WB_SYNC_ALL;
849 trace_writeback_exec(bdi, work);
851 wrote += wb_writeback(wb, work);
854 * Notify the caller of completion if this is a synchronous
855 * work item, otherwise just free it.
858 complete(work->done);
864 * Check for periodic writeback, kupdated() style
866 wrote += wb_check_old_data_flush(wb);
867 wrote += wb_check_background_flush(wb);
868 clear_bit(BDI_writeback_running, &wb->bdi->state);
874 * Handle writeback of dirty data for the device backed by this bdi. Also
875 * wakes up periodically and does kupdated style flushing.
877 int bdi_writeback_thread(void *data)
879 struct bdi_writeback *wb = data;
880 struct backing_dev_info *bdi = wb->bdi;
883 current->flags |= PF_SWAPWRITE;
885 wb->last_active = jiffies;
888 * Our parent may run at a different priority, just set us to normal
890 set_user_nice(current, 0);
892 trace_writeback_thread_start(bdi);
894 while (!kthread_should_stop()) {
896 * Remove own delayed wake-up timer, since we are already awake
897 * and we'll take care of the preriodic write-back.
899 del_timer(&wb->wakeup_timer);
901 pages_written = wb_do_writeback(wb, 0);
903 trace_writeback_pages_written(pages_written);
906 wb->last_active = jiffies;
908 set_current_state(TASK_INTERRUPTIBLE);
909 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
910 __set_current_state(TASK_RUNNING);
914 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
915 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
918 * We have nothing to do, so can go sleep without any
919 * timeout and save power. When a work is queued or
920 * something is made dirty - we will be woken up.
928 /* Flush any work that raced with us exiting */
929 if (!list_empty(&bdi->work_list))
930 wb_do_writeback(wb, 1);
932 trace_writeback_thread_stop(bdi);
938 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
941 void wakeup_flusher_threads(long nr_pages)
943 struct backing_dev_info *bdi;
946 nr_pages = global_page_state(NR_FILE_DIRTY) +
947 global_page_state(NR_UNSTABLE_NFS);
951 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
952 if (!bdi_has_dirty_io(bdi))
954 __bdi_start_writeback(bdi, nr_pages, false);
959 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
961 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
962 struct dentry *dentry;
963 const char *name = "?";
965 dentry = d_find_alias(inode);
967 spin_lock(&dentry->d_lock);
968 name = (const char *) dentry->d_name.name;
971 "%s(%d): dirtied inode %lu (%s) on %s\n",
972 current->comm, task_pid_nr(current), inode->i_ino,
973 name, inode->i_sb->s_id);
975 spin_unlock(&dentry->d_lock);
982 * __mark_inode_dirty - internal function
983 * @inode: inode to mark
984 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
985 * Mark an inode as dirty. Callers should use mark_inode_dirty or
986 * mark_inode_dirty_sync.
988 * Put the inode on the super block's dirty list.
990 * CAREFUL! We mark it dirty unconditionally, but move it onto the
991 * dirty list only if it is hashed or if it refers to a blockdev.
992 * If it was not hashed, it will never be added to the dirty list
993 * even if it is later hashed, as it will have been marked dirty already.
995 * In short, make sure you hash any inodes _before_ you start marking
998 * This function *must* be atomic for the I_DIRTY_PAGES case -
999 * set_page_dirty() is called under spinlock in several places.
1001 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1002 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1003 * the kernel-internal blockdev inode represents the dirtying time of the
1004 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1005 * page->mapping->host, so the page-dirtying time is recorded in the internal
1008 void __mark_inode_dirty(struct inode *inode, int flags)
1010 struct super_block *sb = inode->i_sb;
1011 struct backing_dev_info *bdi = NULL;
1012 bool wakeup_bdi = false;
1015 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1016 * dirty the inode itself
1018 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1019 if (sb->s_op->dirty_inode)
1020 sb->s_op->dirty_inode(inode);
1024 * make sure that changes are seen by all cpus before we test i_state
1029 /* avoid the locking if we can */
1030 if ((inode->i_state & flags) == flags)
1033 if (unlikely(block_dump))
1034 block_dump___mark_inode_dirty(inode);
1036 spin_lock(&inode_lock);
1037 spin_lock(&inode->i_lock);
1038 if ((inode->i_state & flags) != flags) {
1039 const int was_dirty = inode->i_state & I_DIRTY;
1041 inode->i_state |= flags;
1044 * If the inode is being synced, just update its dirty state.
1045 * The unlocker will place the inode on the appropriate
1046 * superblock list, based upon its state.
1048 if (inode->i_state & I_SYNC)
1049 goto out_unlock_inode;
1052 * Only add valid (hashed) inodes to the superblock's
1053 * dirty list. Add blockdev inodes as well.
1055 if (!S_ISBLK(inode->i_mode)) {
1056 if (inode_unhashed(inode))
1057 goto out_unlock_inode;
1059 if (inode->i_state & I_FREEING)
1060 goto out_unlock_inode;
1062 spin_unlock(&inode->i_lock);
1064 * If the inode was already on b_dirty/b_io/b_more_io, don't
1065 * reposition it (that would break b_dirty time-ordering).
1068 bdi = inode_to_bdi(inode);
1070 if (bdi_cap_writeback_dirty(bdi)) {
1071 WARN(!test_bit(BDI_registered, &bdi->state),
1072 "bdi-%s not registered\n", bdi->name);
1075 * If this is the first dirty inode for this
1076 * bdi, we have to wake-up the corresponding
1077 * bdi thread to make sure background
1078 * write-back happens later.
1080 if (!wb_has_dirty_io(&bdi->wb))
1084 inode->dirtied_when = jiffies;
1085 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1090 spin_unlock(&inode->i_lock);
1092 spin_unlock(&inode_lock);
1095 bdi_wakeup_thread_delayed(bdi);
1097 EXPORT_SYMBOL(__mark_inode_dirty);
1100 * Write out a superblock's list of dirty inodes. A wait will be performed
1101 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1103 * If older_than_this is non-NULL, then only write out inodes which
1104 * had their first dirtying at a time earlier than *older_than_this.
1106 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1107 * This function assumes that the blockdev superblock's inodes are backed by
1108 * a variety of queues, so all inodes are searched. For other superblocks,
1109 * assume that all inodes are backed by the same queue.
1111 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1112 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1113 * on the writer throttling path, and we get decent balancing between many
1114 * throttled threads: we don't want them all piling up on inode_sync_wait.
1116 static void wait_sb_inodes(struct super_block *sb)
1118 struct inode *inode, *old_inode = NULL;
1121 * We need to be protected against the filesystem going from
1122 * r/o to r/w or vice versa.
1124 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1126 spin_lock(&inode_sb_list_lock);
1129 * Data integrity sync. Must wait for all pages under writeback,
1130 * because there may have been pages dirtied before our sync
1131 * call, but which had writeout started before we write it out.
1132 * In which case, the inode may not be on the dirty list, but
1133 * we still have to wait for that writeout.
1135 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1136 struct address_space *mapping = inode->i_mapping;
1138 spin_lock(&inode->i_lock);
1139 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1140 (mapping->nrpages == 0)) {
1141 spin_unlock(&inode->i_lock);
1145 spin_unlock(&inode->i_lock);
1146 spin_unlock(&inode_sb_list_lock);
1149 * We hold a reference to 'inode' so it couldn't have been
1150 * removed from s_inodes list while we dropped the
1151 * inode_sb_list_lock. We cannot iput the inode now as we can
1152 * be holding the last reference and we cannot iput it under
1153 * inode_sb_list_lock. So we keep the reference and iput it
1159 filemap_fdatawait(mapping);
1163 spin_lock(&inode_sb_list_lock);
1165 spin_unlock(&inode_sb_list_lock);
1170 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1171 * @sb: the superblock
1172 * @nr: the number of pages to write
1174 * Start writeback on some inodes on this super_block. No guarantees are made
1175 * on how many (if any) will be written, and this function does not wait
1176 * for IO completion of submitted IO.
1178 void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr)
1180 DECLARE_COMPLETION_ONSTACK(done);
1181 struct wb_writeback_work work = {
1183 .sync_mode = WB_SYNC_NONE,
1188 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1189 bdi_queue_work(sb->s_bdi, &work);
1190 wait_for_completion(&done);
1192 EXPORT_SYMBOL(writeback_inodes_sb_nr);
1195 * writeback_inodes_sb - writeback dirty inodes from given super_block
1196 * @sb: the superblock
1198 * Start writeback on some inodes on this super_block. No guarantees are made
1199 * on how many (if any) will be written, and this function does not wait
1200 * for IO completion of submitted IO.
1202 void writeback_inodes_sb(struct super_block *sb)
1204 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages());
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 down_read(&sb->s_umount);
1219 writeback_inodes_sb(sb);
1220 up_read(&sb->s_umount);
1225 EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1228 * writeback_inodes_sb_if_idle - start writeback if none underway
1229 * @sb: the superblock
1230 * @nr: the number of pages to write
1232 * Invoke writeback_inodes_sb if no writeback is currently underway.
1233 * Returns 1 if writeback was started, 0 if not.
1235 int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
1238 if (!writeback_in_progress(sb->s_bdi)) {
1239 down_read(&sb->s_umount);
1240 writeback_inodes_sb_nr(sb, nr);
1241 up_read(&sb->s_umount);
1246 EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1249 * sync_inodes_sb - sync sb inode pages
1250 * @sb: the superblock
1252 * This function writes and waits on any dirty inode belonging to this
1255 void sync_inodes_sb(struct super_block *sb)
1257 DECLARE_COMPLETION_ONSTACK(done);
1258 struct wb_writeback_work work = {
1260 .sync_mode = WB_SYNC_ALL,
1261 .nr_pages = LONG_MAX,
1266 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1268 bdi_queue_work(sb->s_bdi, &work);
1269 wait_for_completion(&done);
1273 EXPORT_SYMBOL(sync_inodes_sb);
1276 * write_inode_now - write an inode to disk
1277 * @inode: inode to write to disk
1278 * @sync: whether the write should be synchronous or not
1280 * This function commits an inode to disk immediately if it is dirty. This is
1281 * primarily needed by knfsd.
1283 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1285 int write_inode_now(struct inode *inode, int sync)
1288 struct writeback_control wbc = {
1289 .nr_to_write = LONG_MAX,
1290 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1292 .range_end = LLONG_MAX,
1295 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1296 wbc.nr_to_write = 0;
1299 spin_lock(&inode_lock);
1300 ret = writeback_single_inode(inode, &wbc);
1301 spin_unlock(&inode_lock);
1303 inode_sync_wait(inode);
1306 EXPORT_SYMBOL(write_inode_now);
1309 * sync_inode - write an inode and its pages to disk.
1310 * @inode: the inode to sync
1311 * @wbc: controls the writeback mode
1313 * sync_inode() will write an inode and its pages to disk. It will also
1314 * correctly update the inode on its superblock's dirty inode lists and will
1315 * update inode->i_state.
1317 * The caller must have a ref on the inode.
1319 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1323 spin_lock(&inode_lock);
1324 ret = writeback_single_inode(inode, wbc);
1325 spin_unlock(&inode_lock);
1328 EXPORT_SYMBOL(sync_inode);
1331 * sync_inode_metadata - write an inode to disk
1332 * @inode: the inode to sync
1333 * @wait: wait for I/O to complete.
1335 * Write an inode to disk and adjust its dirty state after completion.
1337 * Note: only writes the actual inode, no associated data or other metadata.
1339 int sync_inode_metadata(struct inode *inode, int wait)
1341 struct writeback_control wbc = {
1342 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1343 .nr_to_write = 0, /* metadata-only */
1346 return sync_inode(inode, &wbc);
1348 EXPORT_SYMBOL(sync_inode_metadata);