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 static void bdi_queue_work(struct backing_dev_info *bdi,
88 struct wb_writeback_work *work)
90 trace_writeback_queue(bdi, work);
92 spin_lock_bh(&bdi->wb_lock);
93 list_add_tail(&work->list, &bdi->work_list);
95 wake_up_process(bdi->wb.task);
98 * The bdi thread isn't there, wake up the forker thread which
99 * will create and run it.
101 trace_writeback_nothread(bdi, work);
102 wake_up_process(default_backing_dev_info.wb.task);
104 spin_unlock_bh(&bdi->wb_lock);
108 __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
109 bool range_cyclic, bool for_background)
111 struct wb_writeback_work *work;
114 * This is WB_SYNC_NONE writeback, so if allocation fails just
115 * wakeup the thread for old dirty data writeback
117 work = kzalloc(sizeof(*work), GFP_ATOMIC);
120 trace_writeback_nowork(bdi);
121 wake_up_process(bdi->wb.task);
126 work->sync_mode = WB_SYNC_NONE;
127 work->nr_pages = nr_pages;
128 work->range_cyclic = range_cyclic;
129 work->for_background = for_background;
131 bdi_queue_work(bdi, work);
135 * bdi_start_writeback - start writeback
136 * @bdi: the backing device to write from
137 * @nr_pages: the number of pages to write
140 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
141 * started when this function returns, we make no guarentees on
142 * completion. Caller need not hold sb s_umount semaphore.
145 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
147 __bdi_start_writeback(bdi, nr_pages, true, false);
151 * bdi_start_background_writeback - start background writeback
152 * @bdi: the backing device to write from
155 * This does WB_SYNC_NONE background writeback. The IO is only
156 * started when this function returns, we make no guarentees on
157 * completion. Caller need not hold sb s_umount semaphore.
159 void bdi_start_background_writeback(struct backing_dev_info *bdi)
161 __bdi_start_writeback(bdi, LONG_MAX, true, true);
165 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
166 * furthest end of its superblock's dirty-inode list.
168 * Before stamping the inode's ->dirtied_when, we check to see whether it is
169 * already the most-recently-dirtied inode on the b_dirty list. If that is
170 * the case then the inode must have been redirtied while it was being written
171 * out and we don't reset its dirtied_when.
173 static void redirty_tail(struct inode *inode)
175 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
177 if (!list_empty(&wb->b_dirty)) {
180 tail = wb_inode(wb->b_dirty.next);
181 if (time_before(inode->dirtied_when, tail->dirtied_when))
182 inode->dirtied_when = jiffies;
184 list_move(&inode->i_wb_list, &wb->b_dirty);
188 * requeue inode for re-scanning after bdi->b_io list is exhausted.
190 static void requeue_io(struct inode *inode)
192 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
194 list_move(&inode->i_wb_list, &wb->b_more_io);
197 static void inode_sync_complete(struct inode *inode)
200 * Prevent speculative execution through spin_unlock(&inode_lock);
203 wake_up_bit(&inode->i_state, __I_SYNC);
206 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
208 bool ret = time_after(inode->dirtied_when, t);
211 * For inodes being constantly redirtied, dirtied_when can get stuck.
212 * It _appears_ to be in the future, but is actually in distant past.
213 * This test is necessary to prevent such wrapped-around relative times
214 * from permanently stopping the whole bdi writeback.
216 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
222 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
224 static void move_expired_inodes(struct list_head *delaying_queue,
225 struct list_head *dispatch_queue,
226 unsigned long *older_than_this)
229 struct list_head *pos, *node;
230 struct super_block *sb = NULL;
234 while (!list_empty(delaying_queue)) {
235 inode = wb_inode(delaying_queue->prev);
236 if (older_than_this &&
237 inode_dirtied_after(inode, *older_than_this))
239 if (sb && sb != inode->i_sb)
242 list_move(&inode->i_wb_list, &tmp);
245 /* just one sb in list, splice to dispatch_queue and we're done */
247 list_splice(&tmp, dispatch_queue);
251 /* Move inodes from one superblock together */
252 while (!list_empty(&tmp)) {
253 sb = wb_inode(tmp.prev)->i_sb;
254 list_for_each_prev_safe(pos, node, &tmp) {
255 inode = wb_inode(pos);
256 if (inode->i_sb == sb)
257 list_move(&inode->i_wb_list, dispatch_queue);
263 * Queue all expired dirty inodes for io, eldest first.
265 * newly dirtied b_dirty b_io b_more_io
266 * =============> gf edc BA
268 * newly dirtied b_dirty b_io b_more_io
269 * =============> g fBAedc
271 * +--> dequeue for IO
273 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
275 list_splice_init(&wb->b_more_io, &wb->b_io);
276 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
279 static int write_inode(struct inode *inode, struct writeback_control *wbc)
281 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
282 return inode->i_sb->s_op->write_inode(inode, wbc);
287 * Wait for writeback on an inode to complete.
289 static void inode_wait_for_writeback(struct inode *inode)
291 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
292 wait_queue_head_t *wqh;
294 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
295 while (inode->i_state & I_SYNC) {
296 spin_unlock(&inode_lock);
297 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
298 spin_lock(&inode_lock);
303 * Write out an inode's dirty pages. Called under inode_lock. Either the
304 * caller has ref on the inode (either via __iget or via syscall against an fd)
305 * or the inode has I_WILL_FREE set (via generic_forget_inode)
307 * If `wait' is set, wait on the writeout.
309 * The whole writeout design is quite complex and fragile. We want to avoid
310 * starvation of particular inodes when others are being redirtied, prevent
313 * Called under inode_lock.
316 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
318 struct address_space *mapping = inode->i_mapping;
322 if (!atomic_read(&inode->i_count))
323 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
325 WARN_ON(inode->i_state & I_WILL_FREE);
327 if (inode->i_state & I_SYNC) {
329 * If this inode is locked for writeback and we are not doing
330 * writeback-for-data-integrity, move it to b_more_io so that
331 * writeback can proceed with the other inodes on s_io.
333 * We'll have another go at writing back this inode when we
334 * completed a full scan of b_io.
336 if (wbc->sync_mode != WB_SYNC_ALL) {
342 * It's a data-integrity sync. We must wait.
344 inode_wait_for_writeback(inode);
347 BUG_ON(inode->i_state & I_SYNC);
349 /* Set I_SYNC, reset I_DIRTY_PAGES */
350 inode->i_state |= I_SYNC;
351 inode->i_state &= ~I_DIRTY_PAGES;
352 spin_unlock(&inode_lock);
354 ret = do_writepages(mapping, wbc);
357 * Make sure to wait on the data before writing out the metadata.
358 * This is important for filesystems that modify metadata on data
361 if (wbc->sync_mode == WB_SYNC_ALL) {
362 int err = filemap_fdatawait(mapping);
368 * Some filesystems may redirty the inode during the writeback
369 * due to delalloc, clear dirty metadata flags right before
372 spin_lock(&inode_lock);
373 dirty = inode->i_state & I_DIRTY;
374 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
375 spin_unlock(&inode_lock);
376 /* Don't write the inode if only I_DIRTY_PAGES was set */
377 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
378 int err = write_inode(inode, wbc);
383 spin_lock(&inode_lock);
384 inode->i_state &= ~I_SYNC;
385 if (!(inode->i_state & I_FREEING)) {
386 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
388 * We didn't write back all the pages. nfs_writepages()
389 * sometimes bales out without doing anything.
391 inode->i_state |= I_DIRTY_PAGES;
392 if (wbc->nr_to_write <= 0) {
394 * slice used up: queue for next turn
399 * Writeback blocked by something other than
400 * congestion. Delay the inode for some time to
401 * avoid spinning on the CPU (100% iowait)
402 * retrying writeback of the dirty page/inode
403 * that cannot be performed immediately.
407 } else if (inode->i_state & I_DIRTY) {
409 * Filesystems can dirty the inode during writeback
410 * operations, such as delayed allocation during
411 * submission or metadata updates after data IO
417 * The inode is clean. At this point we either have
418 * a reference to the inode or it's on it's way out.
419 * No need to add it back to the LRU.
421 list_del_init(&inode->i_wb_list);
424 inode_sync_complete(inode);
429 * For background writeback the caller does not have the sb pinned
430 * before calling writeback. So make sure that we do pin it, so it doesn't
431 * go away while we are writing inodes from it.
433 static bool pin_sb_for_writeback(struct super_block *sb)
436 if (list_empty(&sb->s_instances)) {
437 spin_unlock(&sb_lock);
442 spin_unlock(&sb_lock);
444 if (down_read_trylock(&sb->s_umount)) {
447 up_read(&sb->s_umount);
455 * Write a portion of b_io inodes which belong to @sb.
457 * If @only_this_sb is true, then find and write all such
458 * inodes. Otherwise write only ones which go sequentially
461 * Return 1, if the caller writeback routine should be
462 * interrupted. Otherwise return 0.
464 static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
465 struct writeback_control *wbc, bool only_this_sb)
467 while (!list_empty(&wb->b_io)) {
469 struct inode *inode = wb_inode(wb->b_io.prev);
471 if (inode->i_sb != sb) {
474 * We only want to write back data for this
475 * superblock, move all inodes not belonging
476 * to it back onto the dirty list.
483 * The inode belongs to a different superblock.
484 * Bounce back to the caller to unpin this and
485 * pin the next superblock.
490 if (inode->i_state & (I_NEW | I_WILL_FREE)) {
495 * Was this inode dirtied after sync_sb_inodes was called?
496 * This keeps sync from extra jobs and livelock.
498 if (inode_dirtied_after(inode, wbc->wb_start))
501 BUG_ON(inode->i_state & I_FREEING);
503 pages_skipped = wbc->pages_skipped;
504 writeback_single_inode(inode, wbc);
505 if (wbc->pages_skipped != pages_skipped) {
507 * writeback is not making progress due to locked
508 * buffers. Skip this inode for now.
512 spin_unlock(&inode_lock);
515 spin_lock(&inode_lock);
516 if (wbc->nr_to_write <= 0) {
520 if (!list_empty(&wb->b_more_io))
527 void writeback_inodes_wb(struct bdi_writeback *wb,
528 struct writeback_control *wbc)
533 wbc->wb_start = jiffies; /* livelock avoidance */
534 spin_lock(&inode_lock);
535 if (!wbc->for_kupdate || list_empty(&wb->b_io))
536 queue_io(wb, wbc->older_than_this);
538 while (!list_empty(&wb->b_io)) {
539 struct inode *inode = wb_inode(wb->b_io.prev);
540 struct super_block *sb = inode->i_sb;
542 if (!pin_sb_for_writeback(sb)) {
546 ret = writeback_sb_inodes(sb, wb, wbc, false);
552 spin_unlock(&inode_lock);
553 /* Leave any unwritten inodes on b_io */
556 static void __writeback_inodes_sb(struct super_block *sb,
557 struct bdi_writeback *wb, struct writeback_control *wbc)
559 WARN_ON(!rwsem_is_locked(&sb->s_umount));
561 spin_lock(&inode_lock);
562 if (!wbc->for_kupdate || list_empty(&wb->b_io))
563 queue_io(wb, wbc->older_than_this);
564 writeback_sb_inodes(sb, wb, wbc, true);
565 spin_unlock(&inode_lock);
569 * The maximum number of pages to writeout in a single bdi flush/kupdate
570 * operation. We do this so we don't hold I_SYNC against an inode for
571 * enormous amounts of time, which would block a userspace task which has
572 * been forced to throttle against that inode. Also, the code reevaluates
573 * the dirty each time it has written this many pages.
575 #define MAX_WRITEBACK_PAGES 1024
577 static inline bool over_bground_thresh(void)
579 unsigned long background_thresh, dirty_thresh;
581 global_dirty_limits(&background_thresh, &dirty_thresh);
583 return (global_page_state(NR_FILE_DIRTY) +
584 global_page_state(NR_UNSTABLE_NFS) >= background_thresh);
588 * Explicit flushing or periodic writeback of "old" data.
590 * Define "old": the first time one of an inode's pages is dirtied, we mark the
591 * dirtying-time in the inode's address_space. So this periodic writeback code
592 * just walks the superblock inode list, writing back any inodes which are
593 * older than a specific point in time.
595 * Try to run once per dirty_writeback_interval. But if a writeback event
596 * takes longer than a dirty_writeback_interval interval, then leave a
599 * older_than_this takes precedence over nr_to_write. So we'll only write back
600 * all dirty pages if they are all attached to "old" mappings.
602 static long wb_writeback(struct bdi_writeback *wb,
603 struct wb_writeback_work *work)
605 struct writeback_control wbc = {
606 .sync_mode = work->sync_mode,
607 .older_than_this = NULL,
608 .for_kupdate = work->for_kupdate,
609 .for_background = work->for_background,
610 .range_cyclic = work->range_cyclic,
612 unsigned long oldest_jif;
616 if (wbc.for_kupdate) {
617 wbc.older_than_this = &oldest_jif;
618 oldest_jif = jiffies -
619 msecs_to_jiffies(dirty_expire_interval * 10);
621 if (!wbc.range_cyclic) {
623 wbc.range_end = LLONG_MAX;
626 wbc.wb_start = jiffies; /* livelock avoidance */
629 * Stop writeback when nr_pages has been consumed
631 if (work->nr_pages <= 0)
635 * For background writeout, stop when we are below the
636 * background dirty threshold
638 if (work->for_background && !over_bground_thresh())
642 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
643 wbc.pages_skipped = 0;
645 trace_wbc_writeback_start(&wbc, wb->bdi);
647 __writeback_inodes_sb(work->sb, wb, &wbc);
649 writeback_inodes_wb(wb, &wbc);
650 trace_wbc_writeback_written(&wbc, wb->bdi);
652 work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
653 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
656 * If we consumed everything, see if we have more
658 if (wbc.nr_to_write <= 0)
661 * Didn't write everything and we don't have more IO, bail
666 * Did we write something? Try for more
668 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
671 * Nothing written. Wait for some inode to
672 * become available for writeback. Otherwise
673 * we'll just busyloop.
675 spin_lock(&inode_lock);
676 if (!list_empty(&wb->b_more_io)) {
677 inode = wb_inode(wb->b_more_io.prev);
678 trace_wbc_writeback_wait(&wbc, wb->bdi);
679 inode_wait_for_writeback(inode);
681 spin_unlock(&inode_lock);
688 * Return the next wb_writeback_work struct that hasn't been processed yet.
690 static struct wb_writeback_work *
691 get_next_work_item(struct backing_dev_info *bdi)
693 struct wb_writeback_work *work = NULL;
695 spin_lock_bh(&bdi->wb_lock);
696 if (!list_empty(&bdi->work_list)) {
697 work = list_entry(bdi->work_list.next,
698 struct wb_writeback_work, list);
699 list_del_init(&work->list);
701 spin_unlock_bh(&bdi->wb_lock);
705 static long wb_check_old_data_flush(struct bdi_writeback *wb)
707 unsigned long expired;
711 * When set to zero, disable periodic writeback
713 if (!dirty_writeback_interval)
716 expired = wb->last_old_flush +
717 msecs_to_jiffies(dirty_writeback_interval * 10);
718 if (time_before(jiffies, expired))
721 wb->last_old_flush = jiffies;
722 nr_pages = global_page_state(NR_FILE_DIRTY) +
723 global_page_state(NR_UNSTABLE_NFS) +
724 get_nr_dirty_inodes();
727 struct wb_writeback_work work = {
728 .nr_pages = nr_pages,
729 .sync_mode = WB_SYNC_NONE,
734 return wb_writeback(wb, &work);
741 * Retrieve work items and do the writeback they describe
743 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
745 struct backing_dev_info *bdi = wb->bdi;
746 struct wb_writeback_work *work;
749 set_bit(BDI_writeback_running, &wb->bdi->state);
750 while ((work = get_next_work_item(bdi)) != NULL) {
752 * Override sync mode, in case we must wait for completion
753 * because this thread is exiting now.
756 work->sync_mode = WB_SYNC_ALL;
758 trace_writeback_exec(bdi, work);
760 wrote += wb_writeback(wb, work);
763 * Notify the caller of completion if this is a synchronous
764 * work item, otherwise just free it.
767 complete(work->done);
773 * Check for periodic writeback, kupdated() style
775 wrote += wb_check_old_data_flush(wb);
776 clear_bit(BDI_writeback_running, &wb->bdi->state);
782 * Handle writeback of dirty data for the device backed by this bdi. Also
783 * wakes up periodically and does kupdated style flushing.
785 int bdi_writeback_thread(void *data)
787 struct bdi_writeback *wb = data;
788 struct backing_dev_info *bdi = wb->bdi;
791 current->flags |= PF_FLUSHER | PF_SWAPWRITE;
793 wb->last_active = jiffies;
796 * Our parent may run at a different priority, just set us to normal
798 set_user_nice(current, 0);
800 trace_writeback_thread_start(bdi);
802 while (!kthread_should_stop()) {
804 * Remove own delayed wake-up timer, since we are already awake
805 * and we'll take care of the preriodic write-back.
807 del_timer(&wb->wakeup_timer);
809 pages_written = wb_do_writeback(wb, 0);
811 trace_writeback_pages_written(pages_written);
814 wb->last_active = jiffies;
816 set_current_state(TASK_INTERRUPTIBLE);
817 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
818 __set_current_state(TASK_RUNNING);
822 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
823 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
826 * We have nothing to do, so can go sleep without any
827 * timeout and save power. When a work is queued or
828 * something is made dirty - we will be woken up.
836 /* Flush any work that raced with us exiting */
837 if (!list_empty(&bdi->work_list))
838 wb_do_writeback(wb, 1);
840 trace_writeback_thread_stop(bdi);
846 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
849 void wakeup_flusher_threads(long nr_pages)
851 struct backing_dev_info *bdi;
854 nr_pages = global_page_state(NR_FILE_DIRTY) +
855 global_page_state(NR_UNSTABLE_NFS);
859 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
860 if (!bdi_has_dirty_io(bdi))
862 __bdi_start_writeback(bdi, nr_pages, false, false);
867 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
869 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
870 struct dentry *dentry;
871 const char *name = "?";
873 dentry = d_find_alias(inode);
875 spin_lock(&dentry->d_lock);
876 name = (const char *) dentry->d_name.name;
879 "%s(%d): dirtied inode %lu (%s) on %s\n",
880 current->comm, task_pid_nr(current), inode->i_ino,
881 name, inode->i_sb->s_id);
883 spin_unlock(&dentry->d_lock);
890 * __mark_inode_dirty - internal function
891 * @inode: inode to mark
892 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
893 * Mark an inode as dirty. Callers should use mark_inode_dirty or
894 * mark_inode_dirty_sync.
896 * Put the inode on the super block's dirty list.
898 * CAREFUL! We mark it dirty unconditionally, but move it onto the
899 * dirty list only if it is hashed or if it refers to a blockdev.
900 * If it was not hashed, it will never be added to the dirty list
901 * even if it is later hashed, as it will have been marked dirty already.
903 * In short, make sure you hash any inodes _before_ you start marking
906 * This function *must* be atomic for the I_DIRTY_PAGES case -
907 * set_page_dirty() is called under spinlock in several places.
909 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
910 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
911 * the kernel-internal blockdev inode represents the dirtying time of the
912 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
913 * page->mapping->host, so the page-dirtying time is recorded in the internal
916 void __mark_inode_dirty(struct inode *inode, int flags)
918 struct super_block *sb = inode->i_sb;
919 struct backing_dev_info *bdi = NULL;
920 bool wakeup_bdi = false;
923 * Don't do this for I_DIRTY_PAGES - that doesn't actually
924 * dirty the inode itself
926 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
927 if (sb->s_op->dirty_inode)
928 sb->s_op->dirty_inode(inode);
932 * make sure that changes are seen by all cpus before we test i_state
937 /* avoid the locking if we can */
938 if ((inode->i_state & flags) == flags)
941 if (unlikely(block_dump))
942 block_dump___mark_inode_dirty(inode);
944 spin_lock(&inode_lock);
945 if ((inode->i_state & flags) != flags) {
946 const int was_dirty = inode->i_state & I_DIRTY;
948 inode->i_state |= flags;
951 * If the inode is being synced, just update its dirty state.
952 * The unlocker will place the inode on the appropriate
953 * superblock list, based upon its state.
955 if (inode->i_state & I_SYNC)
959 * Only add valid (hashed) inodes to the superblock's
960 * dirty list. Add blockdev inodes as well.
962 if (!S_ISBLK(inode->i_mode)) {
963 if (inode_unhashed(inode))
966 if (inode->i_state & I_FREEING)
970 * If the inode was already on b_dirty/b_io/b_more_io, don't
971 * reposition it (that would break b_dirty time-ordering).
974 bdi = inode_to_bdi(inode);
976 if (bdi_cap_writeback_dirty(bdi)) {
977 WARN(!test_bit(BDI_registered, &bdi->state),
978 "bdi-%s not registered\n", bdi->name);
981 * If this is the first dirty inode for this
982 * bdi, we have to wake-up the corresponding
983 * bdi thread to make sure background
984 * write-back happens later.
986 if (!wb_has_dirty_io(&bdi->wb))
990 inode->dirtied_when = jiffies;
991 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
995 spin_unlock(&inode_lock);
998 bdi_wakeup_thread_delayed(bdi);
1000 EXPORT_SYMBOL(__mark_inode_dirty);
1003 * Write out a superblock's list of dirty inodes. A wait will be performed
1004 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1006 * If older_than_this is non-NULL, then only write out inodes which
1007 * had their first dirtying at a time earlier than *older_than_this.
1009 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1010 * This function assumes that the blockdev superblock's inodes are backed by
1011 * a variety of queues, so all inodes are searched. For other superblocks,
1012 * assume that all inodes are backed by the same queue.
1014 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1015 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1016 * on the writer throttling path, and we get decent balancing between many
1017 * throttled threads: we don't want them all piling up on inode_sync_wait.
1019 static void wait_sb_inodes(struct super_block *sb)
1021 struct inode *inode, *old_inode = NULL;
1024 * We need to be protected against the filesystem going from
1025 * r/o to r/w or vice versa.
1027 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1029 spin_lock(&inode_lock);
1032 * Data integrity sync. Must wait for all pages under writeback,
1033 * because there may have been pages dirtied before our sync
1034 * call, but which had writeout started before we write it out.
1035 * In which case, the inode may not be on the dirty list, but
1036 * we still have to wait for that writeout.
1038 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1039 struct address_space *mapping;
1041 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW))
1043 mapping = inode->i_mapping;
1044 if (mapping->nrpages == 0)
1047 spin_unlock(&inode_lock);
1049 * We hold a reference to 'inode' so it couldn't have
1050 * been removed from s_inodes list while we dropped the
1051 * inode_lock. We cannot iput the inode now as we can
1052 * be holding the last reference and we cannot iput it
1053 * under inode_lock. So we keep the reference and iput
1059 filemap_fdatawait(mapping);
1063 spin_lock(&inode_lock);
1065 spin_unlock(&inode_lock);
1070 * writeback_inodes_sb - writeback dirty inodes from given super_block
1071 * @sb: the superblock
1073 * Start writeback on some inodes on this super_block. No guarantees are made
1074 * on how many (if any) will be written, and this function does not wait
1075 * for IO completion of submitted IO. The number of pages submitted is
1078 void writeback_inodes_sb(struct super_block *sb)
1080 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
1081 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
1082 DECLARE_COMPLETION_ONSTACK(done);
1083 struct wb_writeback_work work = {
1085 .sync_mode = WB_SYNC_NONE,
1089 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1091 work.nr_pages = nr_dirty + nr_unstable + get_nr_dirty_inodes();
1093 bdi_queue_work(sb->s_bdi, &work);
1094 wait_for_completion(&done);
1096 EXPORT_SYMBOL(writeback_inodes_sb);
1099 * writeback_inodes_sb_if_idle - start writeback if none underway
1100 * @sb: the superblock
1102 * Invoke writeback_inodes_sb if no writeback is currently underway.
1103 * Returns 1 if writeback was started, 0 if not.
1105 int writeback_inodes_sb_if_idle(struct super_block *sb)
1107 if (!writeback_in_progress(sb->s_bdi)) {
1108 down_read(&sb->s_umount);
1109 writeback_inodes_sb(sb);
1110 up_read(&sb->s_umount);
1115 EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1118 * sync_inodes_sb - sync sb inode pages
1119 * @sb: the superblock
1121 * This function writes and waits on any dirty inode belonging to this
1122 * super_block. The number of pages synced is returned.
1124 void sync_inodes_sb(struct super_block *sb)
1126 DECLARE_COMPLETION_ONSTACK(done);
1127 struct wb_writeback_work work = {
1129 .sync_mode = WB_SYNC_ALL,
1130 .nr_pages = LONG_MAX,
1135 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1137 bdi_queue_work(sb->s_bdi, &work);
1138 wait_for_completion(&done);
1142 EXPORT_SYMBOL(sync_inodes_sb);
1145 * write_inode_now - write an inode to disk
1146 * @inode: inode to write to disk
1147 * @sync: whether the write should be synchronous or not
1149 * This function commits an inode to disk immediately if it is dirty. This is
1150 * primarily needed by knfsd.
1152 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1154 int write_inode_now(struct inode *inode, int sync)
1157 struct writeback_control wbc = {
1158 .nr_to_write = LONG_MAX,
1159 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1161 .range_end = LLONG_MAX,
1164 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1165 wbc.nr_to_write = 0;
1168 spin_lock(&inode_lock);
1169 ret = writeback_single_inode(inode, &wbc);
1170 spin_unlock(&inode_lock);
1172 inode_sync_wait(inode);
1175 EXPORT_SYMBOL(write_inode_now);
1178 * sync_inode - write an inode and its pages to disk.
1179 * @inode: the inode to sync
1180 * @wbc: controls the writeback mode
1182 * sync_inode() will write an inode and its pages to disk. It will also
1183 * correctly update the inode on its superblock's dirty inode lists and will
1184 * update inode->i_state.
1186 * The caller must have a ref on the inode.
1188 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1192 spin_lock(&inode_lock);
1193 ret = writeback_single_inode(inode, wbc);
1194 spin_unlock(&inode_lock);
1197 EXPORT_SYMBOL(sync_inode);
1200 * sync_inode - write an inode to disk
1201 * @inode: the inode to sync
1202 * @wait: wait for I/O to complete.
1204 * Write an inode to disk and adjust it's dirty state after completion.
1206 * Note: only writes the actual inode, no associated data or other metadata.
1208 int sync_inode_metadata(struct inode *inode, int wait)
1210 struct writeback_control wbc = {
1211 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1212 .nr_to_write = 0, /* metadata-only */
1215 return sync_inode(inode, &wbc);
1217 EXPORT_SYMBOL(sync_inode_metadata);