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_bh(&bdi->wb_lock);
80 list_add_tail(&work->list, &bdi->work_list);
82 wake_up_process(bdi->wb.task);
85 * The bdi thread isn't there, wake up the forker thread which
86 * will create and run it.
88 trace_writeback_nothread(bdi, work);
89 wake_up_process(default_backing_dev_info.wb.task);
91 spin_unlock_bh(&bdi->wb_lock);
95 __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
96 bool range_cyclic, bool for_background)
98 struct wb_writeback_work *work;
101 * This is WB_SYNC_NONE writeback, so if allocation fails just
102 * wakeup the thread for old dirty data writeback
104 work = kzalloc(sizeof(*work), GFP_ATOMIC);
107 trace_writeback_nowork(bdi);
108 wake_up_process(bdi->wb.task);
113 work->sync_mode = WB_SYNC_NONE;
114 work->nr_pages = nr_pages;
115 work->range_cyclic = range_cyclic;
116 work->for_background = for_background;
118 bdi_queue_work(bdi, work);
122 * bdi_start_writeback - start writeback
123 * @bdi: the backing device to write from
124 * @nr_pages: the number of pages to write
127 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
128 * started when this function returns, we make no guarentees on
129 * completion. Caller need not hold sb s_umount semaphore.
132 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
134 __bdi_start_writeback(bdi, nr_pages, true, false);
138 * bdi_start_background_writeback - start background writeback
139 * @bdi: the backing device to write from
142 * This does WB_SYNC_NONE background writeback. The IO is only
143 * started when this function returns, we make no guarentees on
144 * completion. Caller need not hold sb s_umount semaphore.
146 void bdi_start_background_writeback(struct backing_dev_info *bdi)
148 __bdi_start_writeback(bdi, LONG_MAX, true, true);
152 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
153 * furthest end of its superblock's dirty-inode list.
155 * Before stamping the inode's ->dirtied_when, we check to see whether it is
156 * already the most-recently-dirtied inode on the b_dirty list. If that is
157 * the case then the inode must have been redirtied while it was being written
158 * out and we don't reset its dirtied_when.
160 static void redirty_tail(struct inode *inode)
162 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
164 if (!list_empty(&wb->b_dirty)) {
167 tail = list_entry(wb->b_dirty.next, struct inode, i_list);
168 if (time_before(inode->dirtied_when, tail->dirtied_when))
169 inode->dirtied_when = jiffies;
171 list_move(&inode->i_list, &wb->b_dirty);
175 * requeue inode for re-scanning after bdi->b_io list is exhausted.
177 static void requeue_io(struct inode *inode)
179 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
181 list_move(&inode->i_list, &wb->b_more_io);
184 static void inode_sync_complete(struct inode *inode)
187 * Prevent speculative execution through spin_unlock(&inode_lock);
190 wake_up_bit(&inode->i_state, __I_SYNC);
193 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
195 bool ret = time_after(inode->dirtied_when, t);
198 * For inodes being constantly redirtied, dirtied_when can get stuck.
199 * It _appears_ to be in the future, but is actually in distant past.
200 * This test is necessary to prevent such wrapped-around relative times
201 * from permanently stopping the whole bdi writeback.
203 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
209 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
211 static void move_expired_inodes(struct list_head *delaying_queue,
212 struct list_head *dispatch_queue,
213 unsigned long *older_than_this)
216 struct list_head *pos, *node;
217 struct super_block *sb = NULL;
221 while (!list_empty(delaying_queue)) {
222 inode = list_entry(delaying_queue->prev, struct inode, i_list);
223 if (older_than_this &&
224 inode_dirtied_after(inode, *older_than_this))
226 if (sb && sb != inode->i_sb)
229 list_move(&inode->i_list, &tmp);
232 /* just one sb in list, splice to dispatch_queue and we're done */
234 list_splice(&tmp, dispatch_queue);
238 /* Move inodes from one superblock together */
239 while (!list_empty(&tmp)) {
240 inode = list_entry(tmp.prev, struct inode, i_list);
242 list_for_each_prev_safe(pos, node, &tmp) {
243 inode = list_entry(pos, struct inode, i_list);
244 if (inode->i_sb == sb)
245 list_move(&inode->i_list, dispatch_queue);
251 * Queue all expired dirty inodes for io, eldest first.
253 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
255 list_splice_init(&wb->b_more_io, wb->b_io.prev);
256 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
259 static int write_inode(struct inode *inode, struct writeback_control *wbc)
261 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
262 return inode->i_sb->s_op->write_inode(inode, wbc);
267 * Wait for writeback on an inode to complete.
269 static void inode_wait_for_writeback(struct inode *inode)
271 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
272 wait_queue_head_t *wqh;
274 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
275 while (inode->i_state & I_SYNC) {
276 spin_unlock(&inode_lock);
277 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
278 spin_lock(&inode_lock);
283 * Write out an inode's dirty pages. Called under inode_lock. Either the
284 * caller has ref on the inode (either via __iget or via syscall against an fd)
285 * or the inode has I_WILL_FREE set (via generic_forget_inode)
287 * If `wait' is set, wait on the writeout.
289 * The whole writeout design is quite complex and fragile. We want to avoid
290 * starvation of particular inodes when others are being redirtied, prevent
293 * Called under inode_lock.
296 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
298 struct address_space *mapping = inode->i_mapping;
302 if (!atomic_read(&inode->i_count))
303 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
305 WARN_ON(inode->i_state & I_WILL_FREE);
307 if (inode->i_state & I_SYNC) {
309 * If this inode is locked for writeback and we are not doing
310 * writeback-for-data-integrity, move it to b_more_io so that
311 * writeback can proceed with the other inodes on s_io.
313 * We'll have another go at writing back this inode when we
314 * completed a full scan of b_io.
316 if (wbc->sync_mode != WB_SYNC_ALL) {
322 * It's a data-integrity sync. We must wait.
324 inode_wait_for_writeback(inode);
327 BUG_ON(inode->i_state & I_SYNC);
329 /* Set I_SYNC, reset I_DIRTY_PAGES */
330 inode->i_state |= I_SYNC;
331 inode->i_state &= ~I_DIRTY_PAGES;
332 spin_unlock(&inode_lock);
334 ret = do_writepages(mapping, wbc);
337 * Make sure to wait on the data before writing out the metadata.
338 * This is important for filesystems that modify metadata on data
341 if (wbc->sync_mode == WB_SYNC_ALL) {
342 int err = filemap_fdatawait(mapping);
348 * Some filesystems may redirty the inode during the writeback
349 * due to delalloc, clear dirty metadata flags right before
352 spin_lock(&inode_lock);
353 dirty = inode->i_state & I_DIRTY;
354 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
355 spin_unlock(&inode_lock);
356 /* Don't write the inode if only I_DIRTY_PAGES was set */
357 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
358 int err = write_inode(inode, wbc);
363 spin_lock(&inode_lock);
364 inode->i_state &= ~I_SYNC;
365 if (!(inode->i_state & I_FREEING)) {
366 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
368 * We didn't write back all the pages. nfs_writepages()
369 * sometimes bales out without doing anything. Redirty
370 * the inode; Move it from b_io onto b_more_io/b_dirty.
373 * akpm: if the caller was the kupdate function we put
374 * this inode at the head of b_dirty so it gets first
375 * consideration. Otherwise, move it to the tail, for
376 * the reasons described there. I'm not really sure
377 * how much sense this makes. Presumably I had a good
378 * reasons for doing it this way, and I'd rather not
379 * muck with it at present.
381 if (wbc->for_kupdate) {
383 * For the kupdate function we move the inode
384 * to b_more_io so it will get more writeout as
385 * soon as the queue becomes uncongested.
387 inode->i_state |= I_DIRTY_PAGES;
388 if (wbc->nr_to_write <= 0) {
390 * slice used up: queue for next turn
395 * somehow blocked: retry later
401 * Otherwise fully redirty the inode so that
402 * other inodes on this superblock will get some
403 * writeout. Otherwise heavy writing to one
404 * file would indefinitely suspend writeout of
405 * all the other files.
407 inode->i_state |= I_DIRTY_PAGES;
410 } else if (inode->i_state & I_DIRTY) {
412 * Filesystems can dirty the inode during writeback
413 * operations, such as delayed allocation during
414 * submission or metadata updates after data IO
418 } else if (atomic_read(&inode->i_count)) {
420 * The inode is clean, inuse
422 list_move(&inode->i_list, &inode_in_use);
425 * The inode is clean, unused
427 list_move(&inode->i_list, &inode_unused);
430 inode_sync_complete(inode);
435 * For background writeback the caller does not have the sb pinned
436 * before calling writeback. So make sure that we do pin it, so it doesn't
437 * go away while we are writing inodes from it.
439 static bool pin_sb_for_writeback(struct super_block *sb)
442 if (list_empty(&sb->s_instances)) {
443 spin_unlock(&sb_lock);
448 spin_unlock(&sb_lock);
450 if (down_read_trylock(&sb->s_umount)) {
453 up_read(&sb->s_umount);
461 * Write a portion of b_io inodes which belong to @sb.
463 * If @only_this_sb is true, then find and write all such
464 * inodes. Otherwise write only ones which go sequentially
467 * Return 1, if the caller writeback routine should be
468 * interrupted. Otherwise return 0.
470 static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
471 struct writeback_control *wbc, bool only_this_sb)
473 while (!list_empty(&wb->b_io)) {
475 struct inode *inode = list_entry(wb->b_io.prev,
476 struct inode, i_list);
478 if (inode->i_sb != sb) {
481 * We only want to write back data for this
482 * superblock, move all inodes not belonging
483 * to it back onto the dirty list.
490 * The inode belongs to a different superblock.
491 * Bounce back to the caller to unpin this and
492 * pin the next superblock.
497 if (inode->i_state & (I_NEW | I_WILL_FREE)) {
502 * Was this inode dirtied after sync_sb_inodes was called?
503 * This keeps sync from extra jobs and livelock.
505 if (inode_dirtied_after(inode, wbc->wb_start))
508 BUG_ON(inode->i_state & I_FREEING);
510 pages_skipped = wbc->pages_skipped;
511 writeback_single_inode(inode, wbc);
512 if (wbc->pages_skipped != pages_skipped) {
514 * writeback is not making progress due to locked
515 * buffers. Skip this inode for now.
519 spin_unlock(&inode_lock);
522 spin_lock(&inode_lock);
523 if (wbc->nr_to_write <= 0) {
527 if (!list_empty(&wb->b_more_io))
534 void writeback_inodes_wb(struct bdi_writeback *wb,
535 struct writeback_control *wbc)
540 wbc->wb_start = jiffies; /* livelock avoidance */
541 spin_lock(&inode_lock);
542 if (!wbc->for_kupdate || list_empty(&wb->b_io))
543 queue_io(wb, wbc->older_than_this);
545 while (!list_empty(&wb->b_io)) {
546 struct inode *inode = list_entry(wb->b_io.prev,
547 struct inode, i_list);
548 struct super_block *sb = inode->i_sb;
550 if (!pin_sb_for_writeback(sb)) {
554 ret = writeback_sb_inodes(sb, wb, wbc, false);
560 spin_unlock(&inode_lock);
561 /* Leave any unwritten inodes on b_io */
564 static void __writeback_inodes_sb(struct super_block *sb,
565 struct bdi_writeback *wb, struct writeback_control *wbc)
567 WARN_ON(!rwsem_is_locked(&sb->s_umount));
569 spin_lock(&inode_lock);
570 if (!wbc->for_kupdate || list_empty(&wb->b_io))
571 queue_io(wb, wbc->older_than_this);
572 writeback_sb_inodes(sb, wb, wbc, true);
573 spin_unlock(&inode_lock);
577 * The maximum number of pages to writeout in a single bdi flush/kupdate
578 * operation. We do this so we don't hold I_SYNC against an inode for
579 * enormous amounts of time, which would block a userspace task which has
580 * been forced to throttle against that inode. Also, the code reevaluates
581 * the dirty each time it has written this many pages.
583 #define MAX_WRITEBACK_PAGES 1024
585 static inline bool over_bground_thresh(void)
587 unsigned long background_thresh, dirty_thresh;
589 global_dirty_limits(&background_thresh, &dirty_thresh);
591 return (global_page_state(NR_FILE_DIRTY) +
592 global_page_state(NR_UNSTABLE_NFS) >= background_thresh);
596 * Explicit flushing or periodic writeback of "old" data.
598 * Define "old": the first time one of an inode's pages is dirtied, we mark the
599 * dirtying-time in the inode's address_space. So this periodic writeback code
600 * just walks the superblock inode list, writing back any inodes which are
601 * older than a specific point in time.
603 * Try to run once per dirty_writeback_interval. But if a writeback event
604 * takes longer than a dirty_writeback_interval interval, then leave a
607 * older_than_this takes precedence over nr_to_write. So we'll only write back
608 * all dirty pages if they are all attached to "old" mappings.
610 static long wb_writeback(struct bdi_writeback *wb,
611 struct wb_writeback_work *work)
613 struct writeback_control wbc = {
614 .sync_mode = work->sync_mode,
615 .older_than_this = NULL,
616 .for_kupdate = work->for_kupdate,
617 .for_background = work->for_background,
618 .range_cyclic = work->range_cyclic,
620 unsigned long oldest_jif;
624 if (wbc.for_kupdate) {
625 wbc.older_than_this = &oldest_jif;
626 oldest_jif = jiffies -
627 msecs_to_jiffies(dirty_expire_interval * 10);
629 if (!wbc.range_cyclic) {
631 wbc.range_end = LLONG_MAX;
634 wbc.wb_start = jiffies; /* livelock avoidance */
637 * Stop writeback when nr_pages has been consumed
639 if (work->nr_pages <= 0)
643 * For background writeout, stop when we are below the
644 * background dirty threshold
646 if (work->for_background && !over_bground_thresh())
650 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
651 wbc.pages_skipped = 0;
653 trace_wbc_writeback_start(&wbc, wb->bdi);
655 __writeback_inodes_sb(work->sb, wb, &wbc);
657 writeback_inodes_wb(wb, &wbc);
658 trace_wbc_writeback_written(&wbc, wb->bdi);
660 work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
661 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
664 * If we consumed everything, see if we have more
666 if (wbc.nr_to_write <= 0)
669 * Didn't write everything and we don't have more IO, bail
674 * Did we write something? Try for more
676 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
679 * Nothing written. Wait for some inode to
680 * become available for writeback. Otherwise
681 * we'll just busyloop.
683 spin_lock(&inode_lock);
684 if (!list_empty(&wb->b_more_io)) {
685 inode = list_entry(wb->b_more_io.prev,
686 struct inode, i_list);
687 trace_wbc_writeback_wait(&wbc, wb->bdi);
688 inode_wait_for_writeback(inode);
690 spin_unlock(&inode_lock);
697 * Return the next wb_writeback_work struct that hasn't been processed yet.
699 static struct wb_writeback_work *
700 get_next_work_item(struct backing_dev_info *bdi)
702 struct wb_writeback_work *work = NULL;
704 spin_lock_bh(&bdi->wb_lock);
705 if (!list_empty(&bdi->work_list)) {
706 work = list_entry(bdi->work_list.next,
707 struct wb_writeback_work, list);
708 list_del_init(&work->list);
710 spin_unlock_bh(&bdi->wb_lock);
714 static long wb_check_old_data_flush(struct bdi_writeback *wb)
716 unsigned long expired;
720 * When set to zero, disable periodic writeback
722 if (!dirty_writeback_interval)
725 expired = wb->last_old_flush +
726 msecs_to_jiffies(dirty_writeback_interval * 10);
727 if (time_before(jiffies, expired))
730 wb->last_old_flush = jiffies;
731 nr_pages = global_page_state(NR_FILE_DIRTY) +
732 global_page_state(NR_UNSTABLE_NFS) +
733 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
736 struct wb_writeback_work work = {
737 .nr_pages = nr_pages,
738 .sync_mode = WB_SYNC_NONE,
743 return wb_writeback(wb, &work);
750 * Retrieve work items and do the writeback they describe
752 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
754 struct backing_dev_info *bdi = wb->bdi;
755 struct wb_writeback_work *work;
758 while ((work = get_next_work_item(bdi)) != NULL) {
760 * Override sync mode, in case we must wait for completion
761 * because this thread is exiting now.
764 work->sync_mode = WB_SYNC_ALL;
766 trace_writeback_exec(bdi, work);
768 wrote += wb_writeback(wb, work);
771 * Notify the caller of completion if this is a synchronous
772 * work item, otherwise just free it.
775 complete(work->done);
781 * Check for periodic writeback, kupdated() style
783 wrote += wb_check_old_data_flush(wb);
789 * Handle writeback of dirty data for the device backed by this bdi. Also
790 * wakes up periodically and does kupdated style flushing.
792 int bdi_writeback_thread(void *data)
794 struct bdi_writeback *wb = data;
795 struct backing_dev_info *bdi = wb->bdi;
798 current->flags |= PF_FLUSHER | PF_SWAPWRITE;
800 wb->last_active = jiffies;
803 * Our parent may run at a different priority, just set us to normal
805 set_user_nice(current, 0);
807 trace_writeback_thread_start(bdi);
809 while (!kthread_should_stop()) {
811 * Remove own delayed wake-up timer, since we are already awake
812 * and we'll take care of the preriodic write-back.
814 del_timer(&wb->wakeup_timer);
816 pages_written = wb_do_writeback(wb, 0);
818 trace_writeback_pages_written(pages_written);
821 wb->last_active = jiffies;
823 set_current_state(TASK_INTERRUPTIBLE);
824 if (!list_empty(&bdi->work_list)) {
825 __set_current_state(TASK_RUNNING);
829 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
830 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
833 * We have nothing to do, so can go sleep without any
834 * timeout and save power. When a work is queued or
835 * something is made dirty - we will be woken up.
843 /* Flush any work that raced with us exiting */
844 if (!list_empty(&bdi->work_list))
845 wb_do_writeback(wb, 1);
847 trace_writeback_thread_stop(bdi);
853 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
856 void wakeup_flusher_threads(long nr_pages)
858 struct backing_dev_info *bdi;
861 nr_pages = global_page_state(NR_FILE_DIRTY) +
862 global_page_state(NR_UNSTABLE_NFS);
866 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
867 if (!bdi_has_dirty_io(bdi))
869 __bdi_start_writeback(bdi, nr_pages, false, false);
874 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
876 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
877 struct dentry *dentry;
878 const char *name = "?";
880 dentry = d_find_alias(inode);
882 spin_lock(&dentry->d_lock);
883 name = (const char *) dentry->d_name.name;
886 "%s(%d): dirtied inode %lu (%s) on %s\n",
887 current->comm, task_pid_nr(current), inode->i_ino,
888 name, inode->i_sb->s_id);
890 spin_unlock(&dentry->d_lock);
897 * __mark_inode_dirty - internal function
898 * @inode: inode to mark
899 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
900 * Mark an inode as dirty. Callers should use mark_inode_dirty or
901 * mark_inode_dirty_sync.
903 * Put the inode on the super block's dirty list.
905 * CAREFUL! We mark it dirty unconditionally, but move it onto the
906 * dirty list only if it is hashed or if it refers to a blockdev.
907 * If it was not hashed, it will never be added to the dirty list
908 * even if it is later hashed, as it will have been marked dirty already.
910 * In short, make sure you hash any inodes _before_ you start marking
913 * This function *must* be atomic for the I_DIRTY_PAGES case -
914 * set_page_dirty() is called under spinlock in several places.
916 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
917 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
918 * the kernel-internal blockdev inode represents the dirtying time of the
919 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
920 * page->mapping->host, so the page-dirtying time is recorded in the internal
923 void __mark_inode_dirty(struct inode *inode, int flags)
925 struct super_block *sb = inode->i_sb;
926 struct backing_dev_info *bdi = NULL;
927 bool wakeup_bdi = false;
930 * Don't do this for I_DIRTY_PAGES - that doesn't actually
931 * dirty the inode itself
933 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
934 if (sb->s_op->dirty_inode)
935 sb->s_op->dirty_inode(inode);
939 * make sure that changes are seen by all cpus before we test i_state
944 /* avoid the locking if we can */
945 if ((inode->i_state & flags) == flags)
948 if (unlikely(block_dump))
949 block_dump___mark_inode_dirty(inode);
951 spin_lock(&inode_lock);
952 if ((inode->i_state & flags) != flags) {
953 const int was_dirty = inode->i_state & I_DIRTY;
955 inode->i_state |= flags;
958 * If the inode is being synced, just update its dirty state.
959 * The unlocker will place the inode on the appropriate
960 * superblock list, based upon its state.
962 if (inode->i_state & I_SYNC)
966 * Only add valid (hashed) inodes to the superblock's
967 * dirty list. Add blockdev inodes as well.
969 if (!S_ISBLK(inode->i_mode)) {
970 if (hlist_unhashed(&inode->i_hash))
973 if (inode->i_state & I_FREEING)
977 * If the inode was already on b_dirty/b_io/b_more_io, don't
978 * reposition it (that would break b_dirty time-ordering).
981 bdi = inode_to_bdi(inode);
983 if (bdi_cap_writeback_dirty(bdi)) {
984 WARN(!test_bit(BDI_registered, &bdi->state),
985 "bdi-%s not registered\n", bdi->name);
988 * If this is the first dirty inode for this
989 * bdi, we have to wake-up the corresponding
990 * bdi thread to make sure background
991 * write-back happens later.
993 if (!wb_has_dirty_io(&bdi->wb))
997 inode->dirtied_when = jiffies;
998 list_move(&inode->i_list, &bdi->wb.b_dirty);
1002 spin_unlock(&inode_lock);
1005 bdi_wakeup_thread_delayed(bdi);
1007 EXPORT_SYMBOL(__mark_inode_dirty);
1010 * Write out a superblock's list of dirty inodes. A wait will be performed
1011 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1013 * If older_than_this is non-NULL, then only write out inodes which
1014 * had their first dirtying at a time earlier than *older_than_this.
1016 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1017 * This function assumes that the blockdev superblock's inodes are backed by
1018 * a variety of queues, so all inodes are searched. For other superblocks,
1019 * assume that all inodes are backed by the same queue.
1021 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1022 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1023 * on the writer throttling path, and we get decent balancing between many
1024 * throttled threads: we don't want them all piling up on inode_sync_wait.
1026 static void wait_sb_inodes(struct super_block *sb)
1028 struct inode *inode, *old_inode = NULL;
1031 * We need to be protected against the filesystem going from
1032 * r/o to r/w or vice versa.
1034 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1036 spin_lock(&inode_lock);
1039 * Data integrity sync. Must wait for all pages under writeback,
1040 * because there may have been pages dirtied before our sync
1041 * call, but which had writeout started before we write it out.
1042 * In which case, the inode may not be on the dirty list, but
1043 * we still have to wait for that writeout.
1045 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1046 struct address_space *mapping;
1048 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW))
1050 mapping = inode->i_mapping;
1051 if (mapping->nrpages == 0)
1054 spin_unlock(&inode_lock);
1056 * We hold a reference to 'inode' so it couldn't have
1057 * been removed from s_inodes list while we dropped the
1058 * inode_lock. We cannot iput the inode now as we can
1059 * be holding the last reference and we cannot iput it
1060 * under inode_lock. So we keep the reference and iput
1066 filemap_fdatawait(mapping);
1070 spin_lock(&inode_lock);
1072 spin_unlock(&inode_lock);
1077 * writeback_inodes_sb - writeback dirty inodes from given super_block
1078 * @sb: the superblock
1080 * Start writeback on some inodes on this super_block. No guarantees are made
1081 * on how many (if any) will be written, and this function does not wait
1082 * for IO completion of submitted IO. The number of pages submitted is
1085 void writeback_inodes_sb(struct super_block *sb)
1087 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
1088 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
1089 DECLARE_COMPLETION_ONSTACK(done);
1090 struct wb_writeback_work work = {
1092 .sync_mode = WB_SYNC_NONE,
1096 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1098 work.nr_pages = nr_dirty + nr_unstable +
1099 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
1101 bdi_queue_work(sb->s_bdi, &work);
1102 wait_for_completion(&done);
1104 EXPORT_SYMBOL(writeback_inodes_sb);
1107 * writeback_inodes_sb_if_idle - start writeback if none underway
1108 * @sb: the superblock
1110 * Invoke writeback_inodes_sb if no writeback is currently underway.
1111 * Returns 1 if writeback was started, 0 if not.
1113 int writeback_inodes_sb_if_idle(struct super_block *sb)
1115 if (!writeback_in_progress(sb->s_bdi)) {
1116 down_read(&sb->s_umount);
1117 writeback_inodes_sb(sb);
1118 up_read(&sb->s_umount);
1123 EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1126 * sync_inodes_sb - sync sb inode pages
1127 * @sb: the superblock
1129 * This function writes and waits on any dirty inode belonging to this
1130 * super_block. The number of pages synced is returned.
1132 void sync_inodes_sb(struct super_block *sb)
1134 DECLARE_COMPLETION_ONSTACK(done);
1135 struct wb_writeback_work work = {
1137 .sync_mode = WB_SYNC_ALL,
1138 .nr_pages = LONG_MAX,
1143 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1145 bdi_queue_work(sb->s_bdi, &work);
1146 wait_for_completion(&done);
1150 EXPORT_SYMBOL(sync_inodes_sb);
1153 * write_inode_now - write an inode to disk
1154 * @inode: inode to write to disk
1155 * @sync: whether the write should be synchronous or not
1157 * This function commits an inode to disk immediately if it is dirty. This is
1158 * primarily needed by knfsd.
1160 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1162 int write_inode_now(struct inode *inode, int sync)
1165 struct writeback_control wbc = {
1166 .nr_to_write = LONG_MAX,
1167 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1169 .range_end = LLONG_MAX,
1172 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1173 wbc.nr_to_write = 0;
1176 spin_lock(&inode_lock);
1177 ret = writeback_single_inode(inode, &wbc);
1178 spin_unlock(&inode_lock);
1180 inode_sync_wait(inode);
1183 EXPORT_SYMBOL(write_inode_now);
1186 * sync_inode - write an inode and its pages to disk.
1187 * @inode: the inode to sync
1188 * @wbc: controls the writeback mode
1190 * sync_inode() will write an inode and its pages to disk. It will also
1191 * correctly update the inode on its superblock's dirty inode lists and will
1192 * update inode->i_state.
1194 * The caller must have a ref on the inode.
1196 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1200 spin_lock(&inode_lock);
1201 ret = writeback_single_inode(inode, wbc);
1202 spin_unlock(&inode_lock);
1205 EXPORT_SYMBOL(sync_inode);