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>
55 #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info)
58 * We don't actually have pdflush, but this one is exported though /proc...
60 int nr_pdflush_threads;
63 * writeback_in_progress - determine whether there is writeback in progress
64 * @bdi: the device's backing_dev_info structure.
66 * Determine whether there is writeback waiting to be handled against a
69 int writeback_in_progress(struct backing_dev_info *bdi)
71 return !list_empty(&bdi->work_list);
74 static void bdi_queue_work(struct backing_dev_info *bdi,
75 struct wb_writeback_work *work)
77 trace_writeback_queue(bdi, work);
79 spin_lock(&bdi->wb_lock);
80 list_add_tail(&work->list, &bdi->work_list);
81 spin_unlock(&bdi->wb_lock);
84 * If the default thread isn't there, make sure we add it. When
85 * it gets created and wakes up, we'll run this work.
87 if (unlikely(!bdi->wb.task)) {
88 trace_writeback_nothread(bdi, work);
89 wake_up_process(default_backing_dev_info.wb.task);
91 struct bdi_writeback *wb = &bdi->wb;
94 wake_up_process(wb->task);
99 __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
100 bool range_cyclic, bool for_background)
102 struct wb_writeback_work *work;
105 * This is WB_SYNC_NONE writeback, so if allocation fails just
106 * wakeup the thread for old dirty data writeback
108 work = kzalloc(sizeof(*work), GFP_ATOMIC);
111 trace_writeback_nowork(bdi);
112 wake_up_process(bdi->wb.task);
117 work->sync_mode = WB_SYNC_NONE;
118 work->nr_pages = nr_pages;
119 work->range_cyclic = range_cyclic;
120 work->for_background = for_background;
122 bdi_queue_work(bdi, work);
126 * bdi_start_writeback - start writeback
127 * @bdi: the backing device to write from
128 * @nr_pages: the number of pages to write
131 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
132 * started when this function returns, we make no guarentees on
133 * completion. Caller need not hold sb s_umount semaphore.
136 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
138 __bdi_start_writeback(bdi, nr_pages, true, false);
142 * bdi_start_background_writeback - start background writeback
143 * @bdi: the backing device to write from
146 * This does WB_SYNC_NONE background 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.
150 void bdi_start_background_writeback(struct backing_dev_info *bdi)
152 __bdi_start_writeback(bdi, LONG_MAX, true, true);
156 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
157 * furthest end of its superblock's dirty-inode list.
159 * Before stamping the inode's ->dirtied_when, we check to see whether it is
160 * already the most-recently-dirtied inode on the b_dirty list. If that is
161 * the case then the inode must have been redirtied while it was being written
162 * out and we don't reset its dirtied_when.
164 static void redirty_tail(struct inode *inode)
166 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
168 if (!list_empty(&wb->b_dirty)) {
171 tail = list_entry(wb->b_dirty.next, struct inode, i_list);
172 if (time_before(inode->dirtied_when, tail->dirtied_when))
173 inode->dirtied_when = jiffies;
175 list_move(&inode->i_list, &wb->b_dirty);
179 * requeue inode for re-scanning after bdi->b_io list is exhausted.
181 static void requeue_io(struct inode *inode)
183 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
185 list_move(&inode->i_list, &wb->b_more_io);
188 static void inode_sync_complete(struct inode *inode)
191 * Prevent speculative execution through spin_unlock(&inode_lock);
194 wake_up_bit(&inode->i_state, __I_SYNC);
197 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
199 bool ret = time_after(inode->dirtied_when, t);
202 * For inodes being constantly redirtied, dirtied_when can get stuck.
203 * It _appears_ to be in the future, but is actually in distant past.
204 * This test is necessary to prevent such wrapped-around relative times
205 * from permanently stopping the whole bdi writeback.
207 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
213 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
215 static void move_expired_inodes(struct list_head *delaying_queue,
216 struct list_head *dispatch_queue,
217 unsigned long *older_than_this)
220 struct list_head *pos, *node;
221 struct super_block *sb = NULL;
225 while (!list_empty(delaying_queue)) {
226 inode = list_entry(delaying_queue->prev, struct inode, i_list);
227 if (older_than_this &&
228 inode_dirtied_after(inode, *older_than_this))
230 if (sb && sb != inode->i_sb)
233 list_move(&inode->i_list, &tmp);
236 /* just one sb in list, splice to dispatch_queue and we're done */
238 list_splice(&tmp, dispatch_queue);
242 /* Move inodes from one superblock together */
243 while (!list_empty(&tmp)) {
244 inode = list_entry(tmp.prev, struct inode, i_list);
246 list_for_each_prev_safe(pos, node, &tmp) {
247 inode = list_entry(pos, struct inode, i_list);
248 if (inode->i_sb == sb)
249 list_move(&inode->i_list, dispatch_queue);
255 * Queue all expired dirty inodes for io, eldest first.
257 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
259 list_splice_init(&wb->b_more_io, wb->b_io.prev);
260 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
263 static int write_inode(struct inode *inode, struct writeback_control *wbc)
265 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
266 return inode->i_sb->s_op->write_inode(inode, wbc);
271 * Wait for writeback on an inode to complete.
273 static void inode_wait_for_writeback(struct inode *inode)
275 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
276 wait_queue_head_t *wqh;
278 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
279 while (inode->i_state & I_SYNC) {
280 spin_unlock(&inode_lock);
281 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
282 spin_lock(&inode_lock);
287 * Write out an inode's dirty pages. Called under inode_lock. Either the
288 * caller has ref on the inode (either via __iget or via syscall against an fd)
289 * or the inode has I_WILL_FREE set (via generic_forget_inode)
291 * If `wait' is set, wait on the writeout.
293 * The whole writeout design is quite complex and fragile. We want to avoid
294 * starvation of particular inodes when others are being redirtied, prevent
297 * Called under inode_lock.
300 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
302 struct address_space *mapping = inode->i_mapping;
306 if (!atomic_read(&inode->i_count))
307 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
309 WARN_ON(inode->i_state & I_WILL_FREE);
311 if (inode->i_state & I_SYNC) {
313 * If this inode is locked for writeback and we are not doing
314 * writeback-for-data-integrity, move it to b_more_io so that
315 * writeback can proceed with the other inodes on s_io.
317 * We'll have another go at writing back this inode when we
318 * completed a full scan of b_io.
320 if (wbc->sync_mode != WB_SYNC_ALL) {
326 * It's a data-integrity sync. We must wait.
328 inode_wait_for_writeback(inode);
331 BUG_ON(inode->i_state & I_SYNC);
333 /* Set I_SYNC, reset I_DIRTY_PAGES */
334 inode->i_state |= I_SYNC;
335 inode->i_state &= ~I_DIRTY_PAGES;
336 spin_unlock(&inode_lock);
338 ret = do_writepages(mapping, wbc);
341 * Make sure to wait on the data before writing out the metadata.
342 * This is important for filesystems that modify metadata on data
345 if (wbc->sync_mode == WB_SYNC_ALL) {
346 int err = filemap_fdatawait(mapping);
352 * Some filesystems may redirty the inode during the writeback
353 * due to delalloc, clear dirty metadata flags right before
356 spin_lock(&inode_lock);
357 dirty = inode->i_state & I_DIRTY;
358 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
359 spin_unlock(&inode_lock);
360 /* Don't write the inode if only I_DIRTY_PAGES was set */
361 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
362 int err = write_inode(inode, wbc);
367 spin_lock(&inode_lock);
368 inode->i_state &= ~I_SYNC;
369 if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
370 if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) {
372 * More pages get dirtied by a fast dirtier.
375 } else if (inode->i_state & I_DIRTY) {
377 * At least XFS will redirty the inode during the
378 * writeback (delalloc) and on io completion (isize).
381 } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
383 * We didn't write back all the pages. nfs_writepages()
384 * sometimes bales out without doing anything. Redirty
385 * the inode; Move it from b_io onto b_more_io/b_dirty.
388 * akpm: if the caller was the kupdate function we put
389 * this inode at the head of b_dirty so it gets first
390 * consideration. Otherwise, move it to the tail, for
391 * the reasons described there. I'm not really sure
392 * how much sense this makes. Presumably I had a good
393 * reasons for doing it this way, and I'd rather not
394 * muck with it at present.
396 if (wbc->for_kupdate) {
398 * For the kupdate function we move the inode
399 * to b_more_io so it will get more writeout as
400 * soon as the queue becomes uncongested.
402 inode->i_state |= I_DIRTY_PAGES;
404 if (wbc->nr_to_write <= 0) {
406 * slice used up: queue for next turn
411 * somehow blocked: retry later
417 * Otherwise fully redirty the inode so that
418 * other inodes on this superblock will get some
419 * writeout. Otherwise heavy writing to one
420 * file would indefinitely suspend writeout of
421 * all the other files.
423 inode->i_state |= I_DIRTY_PAGES;
426 } else if (atomic_read(&inode->i_count)) {
428 * The inode is clean, inuse
430 list_move(&inode->i_list, &inode_in_use);
433 * The inode is clean, unused
435 list_move(&inode->i_list, &inode_unused);
438 inode_sync_complete(inode);
443 * For background writeback the caller does not have the sb pinned
444 * before calling writeback. So make sure that we do pin it, so it doesn't
445 * go away while we are writing inodes from it.
447 static bool pin_sb_for_writeback(struct super_block *sb)
450 if (list_empty(&sb->s_instances)) {
451 spin_unlock(&sb_lock);
456 spin_unlock(&sb_lock);
458 if (down_read_trylock(&sb->s_umount)) {
461 up_read(&sb->s_umount);
469 * Write a portion of b_io inodes which belong to @sb.
471 * If @only_this_sb is true, then find and write all such
472 * inodes. Otherwise write only ones which go sequentially
475 * Return 1, if the caller writeback routine should be
476 * interrupted. Otherwise return 0.
478 static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
479 struct writeback_control *wbc, bool only_this_sb)
481 while (!list_empty(&wb->b_io)) {
483 struct inode *inode = list_entry(wb->b_io.prev,
484 struct inode, i_list);
486 if (inode->i_sb != sb) {
489 * We only want to write back data for this
490 * superblock, move all inodes not belonging
491 * to it back onto the dirty list.
498 * The inode belongs to a different superblock.
499 * Bounce back to the caller to unpin this and
500 * pin the next superblock.
505 if (inode->i_state & (I_NEW | I_WILL_FREE)) {
510 * Was this inode dirtied after sync_sb_inodes was called?
511 * This keeps sync from extra jobs and livelock.
513 if (inode_dirtied_after(inode, wbc->wb_start))
516 BUG_ON(inode->i_state & (I_FREEING | I_CLEAR));
518 pages_skipped = wbc->pages_skipped;
519 writeback_single_inode(inode, wbc);
520 if (wbc->pages_skipped != pages_skipped) {
522 * writeback is not making progress due to locked
523 * buffers. Skip this inode for now.
527 spin_unlock(&inode_lock);
530 spin_lock(&inode_lock);
531 if (wbc->nr_to_write <= 0) {
535 if (!list_empty(&wb->b_more_io))
542 void writeback_inodes_wb(struct bdi_writeback *wb,
543 struct writeback_control *wbc)
547 wbc->wb_start = jiffies; /* livelock avoidance */
548 spin_lock(&inode_lock);
549 if (!wbc->for_kupdate || list_empty(&wb->b_io))
550 queue_io(wb, wbc->older_than_this);
552 while (!list_empty(&wb->b_io)) {
553 struct inode *inode = list_entry(wb->b_io.prev,
554 struct inode, i_list);
555 struct super_block *sb = inode->i_sb;
557 if (!pin_sb_for_writeback(sb)) {
561 ret = writeback_sb_inodes(sb, wb, wbc, false);
567 spin_unlock(&inode_lock);
568 /* Leave any unwritten inodes on b_io */
571 static void __writeback_inodes_sb(struct super_block *sb,
572 struct bdi_writeback *wb, struct writeback_control *wbc)
574 WARN_ON(!rwsem_is_locked(&sb->s_umount));
576 wbc->wb_start = jiffies; /* livelock avoidance */
577 spin_lock(&inode_lock);
578 if (!wbc->for_kupdate || list_empty(&wb->b_io))
579 queue_io(wb, wbc->older_than_this);
580 writeback_sb_inodes(sb, wb, wbc, true);
581 spin_unlock(&inode_lock);
585 * The maximum number of pages to writeout in a single bdi flush/kupdate
586 * operation. We do this so we don't hold I_SYNC against an inode for
587 * enormous amounts of time, which would block a userspace task which has
588 * been forced to throttle against that inode. Also, the code reevaluates
589 * the dirty each time it has written this many pages.
591 #define MAX_WRITEBACK_PAGES 1024
593 static inline bool over_bground_thresh(void)
595 unsigned long background_thresh, dirty_thresh;
597 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
599 return (global_page_state(NR_FILE_DIRTY) +
600 global_page_state(NR_UNSTABLE_NFS) >= background_thresh);
604 * Explicit flushing or periodic writeback of "old" data.
606 * Define "old": the first time one of an inode's pages is dirtied, we mark the
607 * dirtying-time in the inode's address_space. So this periodic writeback code
608 * just walks the superblock inode list, writing back any inodes which are
609 * older than a specific point in time.
611 * Try to run once per dirty_writeback_interval. But if a writeback event
612 * takes longer than a dirty_writeback_interval interval, then leave a
615 * older_than_this takes precedence over nr_to_write. So we'll only write back
616 * all dirty pages if they are all attached to "old" mappings.
618 static long wb_writeback(struct bdi_writeback *wb,
619 struct wb_writeback_work *work)
621 struct writeback_control wbc = {
622 .sync_mode = work->sync_mode,
623 .older_than_this = NULL,
624 .for_kupdate = work->for_kupdate,
625 .for_background = work->for_background,
626 .range_cyclic = work->range_cyclic,
628 unsigned long oldest_jif;
632 if (wbc.for_kupdate) {
633 wbc.older_than_this = &oldest_jif;
634 oldest_jif = jiffies -
635 msecs_to_jiffies(dirty_expire_interval * 10);
637 if (!wbc.range_cyclic) {
639 wbc.range_end = LLONG_MAX;
644 * Stop writeback when nr_pages has been consumed
646 if (work->nr_pages <= 0)
650 * For background writeout, stop when we are below the
651 * background dirty threshold
653 if (work->for_background && !over_bground_thresh())
657 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
658 wbc.pages_skipped = 0;
660 trace_wbc_writeback_start(&wbc, wb->bdi);
662 __writeback_inodes_sb(work->sb, wb, &wbc);
664 writeback_inodes_wb(wb, &wbc);
665 trace_wbc_writeback_written(&wbc, wb->bdi);
667 work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
668 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
671 * If we consumed everything, see if we have more
673 if (wbc.nr_to_write <= 0)
676 * Didn't write everything and we don't have more IO, bail
681 * Did we write something? Try for more
683 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
686 * Nothing written. Wait for some inode to
687 * become available for writeback. Otherwise
688 * we'll just busyloop.
690 spin_lock(&inode_lock);
691 if (!list_empty(&wb->b_more_io)) {
692 inode = list_entry(wb->b_more_io.prev,
693 struct inode, i_list);
694 trace_wbc_writeback_wait(&wbc, wb->bdi);
695 inode_wait_for_writeback(inode);
697 spin_unlock(&inode_lock);
704 * Return the next wb_writeback_work struct that hasn't been processed yet.
706 static struct wb_writeback_work *
707 get_next_work_item(struct backing_dev_info *bdi)
709 struct wb_writeback_work *work = NULL;
711 spin_lock(&bdi->wb_lock);
712 if (!list_empty(&bdi->work_list)) {
713 work = list_entry(bdi->work_list.next,
714 struct wb_writeback_work, list);
715 list_del_init(&work->list);
717 spin_unlock(&bdi->wb_lock);
721 static long wb_check_old_data_flush(struct bdi_writeback *wb)
723 unsigned long expired;
727 * When set to zero, disable periodic writeback
729 if (!dirty_writeback_interval)
732 expired = wb->last_old_flush +
733 msecs_to_jiffies(dirty_writeback_interval * 10);
734 if (time_before(jiffies, expired))
737 wb->last_old_flush = jiffies;
738 nr_pages = global_page_state(NR_FILE_DIRTY) +
739 global_page_state(NR_UNSTABLE_NFS) +
740 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
743 struct wb_writeback_work work = {
744 .nr_pages = nr_pages,
745 .sync_mode = WB_SYNC_NONE,
750 return wb_writeback(wb, &work);
757 * Retrieve work items and do the writeback they describe
759 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
761 struct backing_dev_info *bdi = wb->bdi;
762 struct wb_writeback_work *work;
765 while ((work = get_next_work_item(bdi)) != NULL) {
767 * Override sync mode, in case we must wait for completion
768 * because this thread is exiting now.
771 work->sync_mode = WB_SYNC_ALL;
773 trace_writeback_exec(bdi, work);
775 wrote += wb_writeback(wb, work);
778 * Notify the caller of completion if this is a synchronous
779 * work item, otherwise just free it.
782 complete(work->done);
788 * Check for periodic writeback, kupdated() style
790 wrote += wb_check_old_data_flush(wb);
796 * Handle writeback of dirty data for the device backed by this bdi. Also
797 * wakes up periodically and does kupdated style flushing.
799 int bdi_writeback_thread(void *data)
801 struct bdi_writeback *wb = data;
802 struct backing_dev_info *bdi = wb->bdi;
803 unsigned long last_active = jiffies;
804 unsigned long wait_jiffies = -1UL;
807 current->flags |= PF_FLUSHER | PF_SWAPWRITE;
811 * Our parent may run at a different priority, just set us to normal
813 set_user_nice(current, 0);
816 * Clear pending bit and wakeup anybody waiting to tear us down
818 clear_bit(BDI_pending, &bdi->state);
819 smp_mb__after_clear_bit();
820 wake_up_bit(&bdi->state, BDI_pending);
822 trace_writeback_thread_start(bdi);
824 while (!kthread_should_stop()) {
825 pages_written = wb_do_writeback(wb, 0);
827 trace_writeback_pages_written(pages_written);
830 last_active = jiffies;
831 else if (wait_jiffies != -1UL) {
832 unsigned long max_idle;
835 * Longest period of inactivity that we tolerate. If we
836 * see dirty data again later, the thread will get
837 * recreated automatically.
839 max_idle = max(5UL * 60 * HZ, wait_jiffies);
840 if (time_after(jiffies, max_idle + last_active))
844 set_current_state(TASK_INTERRUPTIBLE);
845 if (!list_empty(&bdi->work_list)) {
846 __set_current_state(TASK_RUNNING);
850 if (dirty_writeback_interval) {
851 wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10);
852 schedule_timeout(wait_jiffies);
862 * Flush any work that raced with us exiting. No new work
863 * will be added, since this bdi isn't discoverable anymore.
865 if (!list_empty(&bdi->work_list))
866 wb_do_writeback(wb, 1);
868 trace_writeback_thread_stop(bdi);
874 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
877 void wakeup_flusher_threads(long nr_pages)
879 struct backing_dev_info *bdi;
882 nr_pages = global_page_state(NR_FILE_DIRTY) +
883 global_page_state(NR_UNSTABLE_NFS);
887 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
888 if (!bdi_has_dirty_io(bdi))
890 __bdi_start_writeback(bdi, nr_pages, false, false);
895 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
897 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
898 struct dentry *dentry;
899 const char *name = "?";
901 dentry = d_find_alias(inode);
903 spin_lock(&dentry->d_lock);
904 name = (const char *) dentry->d_name.name;
907 "%s(%d): dirtied inode %lu (%s) on %s\n",
908 current->comm, task_pid_nr(current), inode->i_ino,
909 name, inode->i_sb->s_id);
911 spin_unlock(&dentry->d_lock);
918 * __mark_inode_dirty - internal function
919 * @inode: inode to mark
920 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
921 * Mark an inode as dirty. Callers should use mark_inode_dirty or
922 * mark_inode_dirty_sync.
924 * Put the inode on the super block's dirty list.
926 * CAREFUL! We mark it dirty unconditionally, but move it onto the
927 * dirty list only if it is hashed or if it refers to a blockdev.
928 * If it was not hashed, it will never be added to the dirty list
929 * even if it is later hashed, as it will have been marked dirty already.
931 * In short, make sure you hash any inodes _before_ you start marking
934 * This function *must* be atomic for the I_DIRTY_PAGES case -
935 * set_page_dirty() is called under spinlock in several places.
937 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
938 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
939 * the kernel-internal blockdev inode represents the dirtying time of the
940 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
941 * page->mapping->host, so the page-dirtying time is recorded in the internal
944 void __mark_inode_dirty(struct inode *inode, int flags)
946 struct super_block *sb = inode->i_sb;
949 * Don't do this for I_DIRTY_PAGES - that doesn't actually
950 * dirty the inode itself
952 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
953 if (sb->s_op->dirty_inode)
954 sb->s_op->dirty_inode(inode);
958 * make sure that changes are seen by all cpus before we test i_state
963 /* avoid the locking if we can */
964 if ((inode->i_state & flags) == flags)
967 if (unlikely(block_dump))
968 block_dump___mark_inode_dirty(inode);
970 spin_lock(&inode_lock);
971 if ((inode->i_state & flags) != flags) {
972 const int was_dirty = inode->i_state & I_DIRTY;
974 inode->i_state |= flags;
977 * If the inode is being synced, just update its dirty state.
978 * The unlocker will place the inode on the appropriate
979 * superblock list, based upon its state.
981 if (inode->i_state & I_SYNC)
985 * Only add valid (hashed) inodes to the superblock's
986 * dirty list. Add blockdev inodes as well.
988 if (!S_ISBLK(inode->i_mode)) {
989 if (hlist_unhashed(&inode->i_hash))
992 if (inode->i_state & (I_FREEING|I_CLEAR))
996 * If the inode was already on b_dirty/b_io/b_more_io, don't
997 * reposition it (that would break b_dirty time-ordering).
1000 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1001 struct backing_dev_info *bdi = wb->bdi;
1003 if (bdi_cap_writeback_dirty(bdi) &&
1004 !test_bit(BDI_registered, &bdi->state)) {
1006 printk(KERN_ERR "bdi-%s not registered\n",
1010 inode->dirtied_when = jiffies;
1011 list_move(&inode->i_list, &wb->b_dirty);
1015 spin_unlock(&inode_lock);
1017 EXPORT_SYMBOL(__mark_inode_dirty);
1020 * Write out a superblock's list of dirty inodes. A wait will be performed
1021 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1023 * If older_than_this is non-NULL, then only write out inodes which
1024 * had their first dirtying at a time earlier than *older_than_this.
1026 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1027 * This function assumes that the blockdev superblock's inodes are backed by
1028 * a variety of queues, so all inodes are searched. For other superblocks,
1029 * assume that all inodes are backed by the same queue.
1031 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1032 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1033 * on the writer throttling path, and we get decent balancing between many
1034 * throttled threads: we don't want them all piling up on inode_sync_wait.
1036 static void wait_sb_inodes(struct super_block *sb)
1038 struct inode *inode, *old_inode = NULL;
1041 * We need to be protected against the filesystem going from
1042 * r/o to r/w or vice versa.
1044 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1046 spin_lock(&inode_lock);
1049 * Data integrity sync. Must wait for all pages under writeback,
1050 * because there may have been pages dirtied before our sync
1051 * call, but which had writeout started before we write it out.
1052 * In which case, the inode may not be on the dirty list, but
1053 * we still have to wait for that writeout.
1055 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1056 struct address_space *mapping;
1058 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
1060 mapping = inode->i_mapping;
1061 if (mapping->nrpages == 0)
1064 spin_unlock(&inode_lock);
1066 * We hold a reference to 'inode' so it couldn't have
1067 * been removed from s_inodes list while we dropped the
1068 * inode_lock. We cannot iput the inode now as we can
1069 * be holding the last reference and we cannot iput it
1070 * under inode_lock. So we keep the reference and iput
1076 filemap_fdatawait(mapping);
1080 spin_lock(&inode_lock);
1082 spin_unlock(&inode_lock);
1087 * writeback_inodes_sb - writeback dirty inodes from given super_block
1088 * @sb: the superblock
1090 * Start writeback on some inodes on this super_block. No guarantees are made
1091 * on how many (if any) will be written, and this function does not wait
1092 * for IO completion of submitted IO. The number of pages submitted is
1095 void writeback_inodes_sb(struct super_block *sb)
1097 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
1098 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
1099 DECLARE_COMPLETION_ONSTACK(done);
1100 struct wb_writeback_work work = {
1102 .sync_mode = WB_SYNC_NONE,
1106 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1108 work.nr_pages = nr_dirty + nr_unstable +
1109 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
1111 bdi_queue_work(sb->s_bdi, &work);
1112 wait_for_completion(&done);
1114 EXPORT_SYMBOL(writeback_inodes_sb);
1117 * writeback_inodes_sb_if_idle - start writeback if none underway
1118 * @sb: the superblock
1120 * Invoke writeback_inodes_sb if no writeback is currently underway.
1121 * Returns 1 if writeback was started, 0 if not.
1123 int writeback_inodes_sb_if_idle(struct super_block *sb)
1125 if (!writeback_in_progress(sb->s_bdi)) {
1126 down_read(&sb->s_umount);
1127 writeback_inodes_sb(sb);
1128 up_read(&sb->s_umount);
1133 EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1136 * sync_inodes_sb - sync sb inode pages
1137 * @sb: the superblock
1139 * This function writes and waits on any dirty inode belonging to this
1140 * super_block. The number of pages synced is returned.
1142 void sync_inodes_sb(struct super_block *sb)
1144 DECLARE_COMPLETION_ONSTACK(done);
1145 struct wb_writeback_work work = {
1147 .sync_mode = WB_SYNC_ALL,
1148 .nr_pages = LONG_MAX,
1153 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1155 bdi_queue_work(sb->s_bdi, &work);
1156 wait_for_completion(&done);
1160 EXPORT_SYMBOL(sync_inodes_sb);
1163 * write_inode_now - write an inode to disk
1164 * @inode: inode to write to disk
1165 * @sync: whether the write should be synchronous or not
1167 * This function commits an inode to disk immediately if it is dirty. This is
1168 * primarily needed by knfsd.
1170 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1172 int write_inode_now(struct inode *inode, int sync)
1175 struct writeback_control wbc = {
1176 .nr_to_write = LONG_MAX,
1177 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1179 .range_end = LLONG_MAX,
1182 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1183 wbc.nr_to_write = 0;
1186 spin_lock(&inode_lock);
1187 ret = writeback_single_inode(inode, &wbc);
1188 spin_unlock(&inode_lock);
1190 inode_sync_wait(inode);
1193 EXPORT_SYMBOL(write_inode_now);
1196 * sync_inode - write an inode and its pages to disk.
1197 * @inode: the inode to sync
1198 * @wbc: controls the writeback mode
1200 * sync_inode() will write an inode and its pages to disk. It will also
1201 * correctly update the inode on its superblock's dirty inode lists and will
1202 * update inode->i_state.
1204 * The caller must have a ref on the inode.
1206 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1210 spin_lock(&inode_lock);
1211 ret = writeback_single_inode(inode, wbc);
1212 spin_unlock(&inode_lock);
1215 EXPORT_SYMBOL(sync_inode);