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>
31 #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info)
34 * We don't actually have pdflush, but this one is exported though /proc...
36 int nr_pdflush_threads;
39 * Passed into wb_writeback(), essentially a subset of writeback_control
41 struct wb_writeback_args {
43 struct super_block *sb;
44 enum writeback_sync_modes sync_mode;
45 unsigned int for_kupdate:1;
46 unsigned int range_cyclic:1;
47 unsigned int for_background:1;
51 * Work items for the bdi_writeback threads
54 struct list_head list; /* pending work list */
55 struct rcu_head rcu_head; /* for RCU free/clear of work */
57 unsigned long seen; /* threads that have seen this work */
58 atomic_t pending; /* number of threads still to do work */
60 struct wb_writeback_args args; /* writeback arguments */
62 unsigned long state; /* flag bits, see WS_* */
70 static inline void bdi_work_init(struct bdi_work *work,
71 struct wb_writeback_args *args)
73 INIT_RCU_HEAD(&work->rcu_head);
75 __set_bit(WS_INPROGRESS, &work->state);
79 * writeback_in_progress - determine whether there is writeback in progress
80 * @bdi: the device's backing_dev_info structure.
82 * Determine whether there is writeback waiting to be handled against a
85 int writeback_in_progress(struct backing_dev_info *bdi)
87 return !list_empty(&bdi->work_list);
90 static void bdi_work_free(struct rcu_head *head)
92 struct bdi_work *work = container_of(head, struct bdi_work, rcu_head);
94 clear_bit(WS_INPROGRESS, &work->state);
95 smp_mb__after_clear_bit();
96 wake_up_bit(&work->state, WS_INPROGRESS);
98 if (!test_bit(WS_ONSTACK, &work->state))
102 static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work)
105 * The caller has retrieved the work arguments from this work,
106 * drop our reference. If this is the last ref, delete and free it
108 if (atomic_dec_and_test(&work->pending)) {
109 struct backing_dev_info *bdi = wb->bdi;
111 spin_lock(&bdi->wb_lock);
112 list_del_rcu(&work->list);
113 spin_unlock(&bdi->wb_lock);
115 call_rcu(&work->rcu_head, bdi_work_free);
119 static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work)
121 work->seen = bdi->wb_mask;
123 atomic_set(&work->pending, bdi->wb_cnt);
124 BUG_ON(!bdi->wb_cnt);
127 * list_add_tail_rcu() contains the necessary barriers to
128 * make sure the above stores are seen before the item is
129 * noticed on the list
131 spin_lock(&bdi->wb_lock);
132 list_add_tail_rcu(&work->list, &bdi->work_list);
133 spin_unlock(&bdi->wb_lock);
136 * If the default thread isn't there, make sure we add it. When
137 * it gets created and wakes up, we'll run this work.
139 if (unlikely(list_empty_careful(&bdi->wb_list)))
140 wake_up_process(default_backing_dev_info.wb.task);
142 struct bdi_writeback *wb = &bdi->wb;
145 wake_up_process(wb->task);
150 * Used for on-stack allocated work items. The caller needs to wait until
151 * the wb threads have acked the work before it's safe to continue.
153 static void bdi_wait_on_work_done(struct bdi_work *work)
155 wait_on_bit(&work->state, WS_INPROGRESS, bdi_sched_wait,
156 TASK_UNINTERRUPTIBLE);
159 static void bdi_alloc_queue_work(struct backing_dev_info *bdi,
160 struct wb_writeback_args *args)
162 struct bdi_work *work;
165 * This is WB_SYNC_NONE writeback, so if allocation fails just
166 * wakeup the thread for old dirty data writeback
168 work = kmalloc(sizeof(*work), GFP_ATOMIC);
170 bdi_work_init(work, args);
171 bdi_queue_work(bdi, work);
173 struct bdi_writeback *wb = &bdi->wb;
176 wake_up_process(wb->task);
181 * bdi_queue_work_onstack - start and wait for writeback
182 * @sb: write inodes from this super_block
185 * This function initiates writeback and waits for the operation to
186 * complete. Callers must hold the sb s_umount semaphore for
187 * reading, to avoid having the super disappear before we are done.
189 static void bdi_queue_work_onstack(struct wb_writeback_args *args)
191 struct bdi_work work;
193 bdi_work_init(&work, args);
194 __set_bit(WS_ONSTACK, &work.state);
196 bdi_queue_work(args->sb->s_bdi, &work);
197 bdi_wait_on_work_done(&work);
201 * bdi_start_writeback - start writeback
202 * @bdi: the backing device to write from
203 * @nr_pages: the number of pages to write
206 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
207 * started when this function returns, we make no guarentees on
208 * completion. Caller need not hold sb s_umount semaphore.
211 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
213 struct wb_writeback_args args = {
214 .sync_mode = WB_SYNC_NONE,
215 .nr_pages = nr_pages,
219 bdi_alloc_queue_work(bdi, &args);
223 * bdi_start_background_writeback - start background writeback
224 * @bdi: the backing device to write from
227 * This does WB_SYNC_NONE background writeback. The IO is only
228 * started when this function returns, we make no guarentees on
229 * completion. Caller need not hold sb s_umount semaphore.
231 void bdi_start_background_writeback(struct backing_dev_info *bdi)
233 struct wb_writeback_args args = {
234 .sync_mode = WB_SYNC_NONE,
235 .nr_pages = LONG_MAX,
239 bdi_alloc_queue_work(bdi, &args);
243 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
244 * furthest end of its superblock's dirty-inode list.
246 * Before stamping the inode's ->dirtied_when, we check to see whether it is
247 * already the most-recently-dirtied inode on the b_dirty list. If that is
248 * the case then the inode must have been redirtied while it was being written
249 * out and we don't reset its dirtied_when.
251 static void redirty_tail(struct inode *inode)
253 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
255 if (!list_empty(&wb->b_dirty)) {
258 tail = list_entry(wb->b_dirty.next, struct inode, i_list);
259 if (time_before(inode->dirtied_when, tail->dirtied_when))
260 inode->dirtied_when = jiffies;
262 list_move(&inode->i_list, &wb->b_dirty);
266 * requeue inode for re-scanning after bdi->b_io list is exhausted.
268 static void requeue_io(struct inode *inode)
270 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
272 list_move(&inode->i_list, &wb->b_more_io);
275 static void inode_sync_complete(struct inode *inode)
278 * Prevent speculative execution through spin_unlock(&inode_lock);
281 wake_up_bit(&inode->i_state, __I_SYNC);
284 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
286 bool ret = time_after(inode->dirtied_when, t);
289 * For inodes being constantly redirtied, dirtied_when can get stuck.
290 * It _appears_ to be in the future, but is actually in distant past.
291 * This test is necessary to prevent such wrapped-around relative times
292 * from permanently stopping the whole bdi writeback.
294 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
300 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
302 static void move_expired_inodes(struct list_head *delaying_queue,
303 struct list_head *dispatch_queue,
304 unsigned long *older_than_this)
307 struct list_head *pos, *node;
308 struct super_block *sb = NULL;
312 while (!list_empty(delaying_queue)) {
313 inode = list_entry(delaying_queue->prev, struct inode, i_list);
314 if (older_than_this &&
315 inode_dirtied_after(inode, *older_than_this))
317 if (sb && sb != inode->i_sb)
320 list_move(&inode->i_list, &tmp);
323 /* just one sb in list, splice to dispatch_queue and we're done */
325 list_splice(&tmp, dispatch_queue);
329 /* Move inodes from one superblock together */
330 while (!list_empty(&tmp)) {
331 inode = list_entry(tmp.prev, struct inode, i_list);
333 list_for_each_prev_safe(pos, node, &tmp) {
334 inode = list_entry(pos, struct inode, i_list);
335 if (inode->i_sb == sb)
336 list_move(&inode->i_list, dispatch_queue);
342 * Queue all expired dirty inodes for io, eldest first.
344 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
346 list_splice_init(&wb->b_more_io, wb->b_io.prev);
347 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
350 static int write_inode(struct inode *inode, struct writeback_control *wbc)
352 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
353 return inode->i_sb->s_op->write_inode(inode, wbc);
358 * Wait for writeback on an inode to complete.
360 static void inode_wait_for_writeback(struct inode *inode)
362 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
363 wait_queue_head_t *wqh;
365 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
366 while (inode->i_state & I_SYNC) {
367 spin_unlock(&inode_lock);
368 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
369 spin_lock(&inode_lock);
374 * Write out an inode's dirty pages. Called under inode_lock. Either the
375 * caller has ref on the inode (either via __iget or via syscall against an fd)
376 * or the inode has I_WILL_FREE set (via generic_forget_inode)
378 * If `wait' is set, wait on the writeout.
380 * The whole writeout design is quite complex and fragile. We want to avoid
381 * starvation of particular inodes when others are being redirtied, prevent
384 * Called under inode_lock.
387 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
389 struct address_space *mapping = inode->i_mapping;
393 if (!atomic_read(&inode->i_count))
394 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
396 WARN_ON(inode->i_state & I_WILL_FREE);
398 if (inode->i_state & I_SYNC) {
400 * If this inode is locked for writeback and we are not doing
401 * writeback-for-data-integrity, move it to b_more_io so that
402 * writeback can proceed with the other inodes on s_io.
404 * We'll have another go at writing back this inode when we
405 * completed a full scan of b_io.
407 if (wbc->sync_mode != WB_SYNC_ALL) {
413 * It's a data-integrity sync. We must wait.
415 inode_wait_for_writeback(inode);
418 BUG_ON(inode->i_state & I_SYNC);
420 /* Set I_SYNC, reset I_DIRTY_PAGES */
421 inode->i_state |= I_SYNC;
422 inode->i_state &= ~I_DIRTY_PAGES;
423 spin_unlock(&inode_lock);
425 ret = do_writepages(mapping, wbc);
428 * Make sure to wait on the data before writing out the metadata.
429 * This is important for filesystems that modify metadata on data
432 if (wbc->sync_mode == WB_SYNC_ALL) {
433 int err = filemap_fdatawait(mapping);
439 * Some filesystems may redirty the inode during the writeback
440 * due to delalloc, clear dirty metadata flags right before
443 spin_lock(&inode_lock);
444 dirty = inode->i_state & I_DIRTY;
445 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
446 spin_unlock(&inode_lock);
447 /* Don't write the inode if only I_DIRTY_PAGES was set */
448 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
449 int err = write_inode(inode, wbc);
454 spin_lock(&inode_lock);
455 inode->i_state &= ~I_SYNC;
456 if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
457 if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) {
459 * More pages get dirtied by a fast dirtier.
462 } else if (inode->i_state & I_DIRTY) {
464 * At least XFS will redirty the inode during the
465 * writeback (delalloc) and on io completion (isize).
468 } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
470 * We didn't write back all the pages. nfs_writepages()
471 * sometimes bales out without doing anything. Redirty
472 * the inode; Move it from b_io onto b_more_io/b_dirty.
475 * akpm: if the caller was the kupdate function we put
476 * this inode at the head of b_dirty so it gets first
477 * consideration. Otherwise, move it to the tail, for
478 * the reasons described there. I'm not really sure
479 * how much sense this makes. Presumably I had a good
480 * reasons for doing it this way, and I'd rather not
481 * muck with it at present.
483 if (wbc->for_kupdate) {
485 * For the kupdate function we move the inode
486 * to b_more_io so it will get more writeout as
487 * soon as the queue becomes uncongested.
489 inode->i_state |= I_DIRTY_PAGES;
491 if (wbc->nr_to_write <= 0) {
493 * slice used up: queue for next turn
498 * somehow blocked: retry later
504 * Otherwise fully redirty the inode so that
505 * other inodes on this superblock will get some
506 * writeout. Otherwise heavy writing to one
507 * file would indefinitely suspend writeout of
508 * all the other files.
510 inode->i_state |= I_DIRTY_PAGES;
513 } else if (atomic_read(&inode->i_count)) {
515 * The inode is clean, inuse
517 list_move(&inode->i_list, &inode_in_use);
520 * The inode is clean, unused
522 list_move(&inode->i_list, &inode_unused);
525 inode_sync_complete(inode);
530 * For background writeback the caller does not have the sb pinned
531 * before calling writeback. So make sure that we do pin it, so it doesn't
532 * go away while we are writing inodes from it.
534 static bool pin_sb_for_writeback(struct super_block *sb)
538 if (down_read_trylock(&sb->s_umount)) {
540 spin_unlock(&sb_lock);
544 * umounted, drop rwsem again and fall through to failure
546 up_read(&sb->s_umount);
549 spin_unlock(&sb_lock);
554 * Write a portion of b_io inodes which belong to @sb.
555 * If @wbc->sb != NULL, then find and write all such
556 * inodes. Otherwise write only ones which go sequentially
558 * Return 1, if the caller writeback routine should be
559 * interrupted. Otherwise return 0.
561 static int writeback_sb_inodes(struct super_block *sb,
562 struct bdi_writeback *wb,
563 struct writeback_control *wbc)
565 while (!list_empty(&wb->b_io)) {
567 struct inode *inode = list_entry(wb->b_io.prev,
568 struct inode, i_list);
569 if (wbc->sb && sb != inode->i_sb) {
570 /* super block given and doesn't
571 match, skip this inode */
575 if (sb != inode->i_sb)
576 /* finish with this superblock */
578 if (inode->i_state & (I_NEW | I_WILL_FREE)) {
583 * Was this inode dirtied after sync_sb_inodes was called?
584 * This keeps sync from extra jobs and livelock.
586 if (inode_dirtied_after(inode, wbc->wb_start))
589 BUG_ON(inode->i_state & (I_FREEING | I_CLEAR));
591 pages_skipped = wbc->pages_skipped;
592 writeback_single_inode(inode, wbc);
593 if (wbc->pages_skipped != pages_skipped) {
595 * writeback is not making progress due to locked
596 * buffers. Skip this inode for now.
600 spin_unlock(&inode_lock);
603 spin_lock(&inode_lock);
604 if (wbc->nr_to_write <= 0) {
608 if (!list_empty(&wb->b_more_io))
615 static void writeback_inodes_wb(struct bdi_writeback *wb,
616 struct writeback_control *wbc)
620 wbc->wb_start = jiffies; /* livelock avoidance */
621 spin_lock(&inode_lock);
622 if (!wbc->for_kupdate || list_empty(&wb->b_io))
623 queue_io(wb, wbc->older_than_this);
625 while (!list_empty(&wb->b_io)) {
626 struct inode *inode = list_entry(wb->b_io.prev,
627 struct inode, i_list);
628 struct super_block *sb = inode->i_sb;
632 * We are requested to write out inodes for a specific
633 * superblock. This means we already have s_umount
634 * taken by the caller which also waits for us to
635 * complete the writeout.
642 WARN_ON(!rwsem_is_locked(&sb->s_umount));
644 ret = writeback_sb_inodes(sb, wb, wbc);
646 if (!pin_sb_for_writeback(sb)) {
650 ret = writeback_sb_inodes(sb, wb, wbc);
657 spin_unlock(&inode_lock);
658 /* Leave any unwritten inodes on b_io */
661 void writeback_inodes_wbc(struct writeback_control *wbc)
663 struct backing_dev_info *bdi = wbc->bdi;
665 writeback_inodes_wb(&bdi->wb, wbc);
669 * The maximum number of pages to writeout in a single bdi flush/kupdate
670 * operation. We do this so we don't hold I_SYNC against an inode for
671 * enormous amounts of time, which would block a userspace task which has
672 * been forced to throttle against that inode. Also, the code reevaluates
673 * the dirty each time it has written this many pages.
675 #define MAX_WRITEBACK_PAGES 1024
677 static inline bool over_bground_thresh(void)
679 unsigned long background_thresh, dirty_thresh;
681 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
683 return (global_page_state(NR_FILE_DIRTY) +
684 global_page_state(NR_UNSTABLE_NFS) >= background_thresh);
688 * Explicit flushing or periodic writeback of "old" data.
690 * Define "old": the first time one of an inode's pages is dirtied, we mark the
691 * dirtying-time in the inode's address_space. So this periodic writeback code
692 * just walks the superblock inode list, writing back any inodes which are
693 * older than a specific point in time.
695 * Try to run once per dirty_writeback_interval. But if a writeback event
696 * takes longer than a dirty_writeback_interval interval, then leave a
699 * older_than_this takes precedence over nr_to_write. So we'll only write back
700 * all dirty pages if they are all attached to "old" mappings.
702 static long wb_writeback(struct bdi_writeback *wb,
703 struct wb_writeback_args *args)
705 struct writeback_control wbc = {
708 .sync_mode = args->sync_mode,
709 .older_than_this = NULL,
710 .for_kupdate = args->for_kupdate,
711 .for_background = args->for_background,
712 .range_cyclic = args->range_cyclic,
714 unsigned long oldest_jif;
718 if (wbc.for_kupdate) {
719 wbc.older_than_this = &oldest_jif;
720 oldest_jif = jiffies -
721 msecs_to_jiffies(dirty_expire_interval * 10);
723 if (!wbc.range_cyclic) {
725 wbc.range_end = LLONG_MAX;
730 * Stop writeback when nr_pages has been consumed
732 if (args->nr_pages <= 0)
736 * For background writeout, stop when we are below the
737 * background dirty threshold
739 if (args->for_background && !over_bground_thresh())
743 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
744 wbc.pages_skipped = 0;
745 writeback_inodes_wb(wb, &wbc);
746 args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
747 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
750 * If we consumed everything, see if we have more
752 if (wbc.nr_to_write <= 0)
755 * Didn't write everything and we don't have more IO, bail
760 * Did we write something? Try for more
762 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
765 * Nothing written. Wait for some inode to
766 * become available for writeback. Otherwise
767 * we'll just busyloop.
769 spin_lock(&inode_lock);
770 if (!list_empty(&wb->b_more_io)) {
771 inode = list_entry(wb->b_more_io.prev,
772 struct inode, i_list);
773 inode_wait_for_writeback(inode);
775 spin_unlock(&inode_lock);
782 * Return the next bdi_work struct that hasn't been processed by this
783 * wb thread yet. ->seen is initially set for each thread that exists
784 * for this device, when a thread first notices a piece of work it
785 * clears its bit. Depending on writeback type, the thread will notify
786 * completion on either receiving the work (WB_SYNC_NONE) or after
787 * it is done (WB_SYNC_ALL).
789 static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi,
790 struct bdi_writeback *wb)
792 struct bdi_work *work, *ret = NULL;
796 list_for_each_entry_rcu(work, &bdi->work_list, list) {
797 if (!test_bit(wb->nr, &work->seen))
799 clear_bit(wb->nr, &work->seen);
809 static long wb_check_old_data_flush(struct bdi_writeback *wb)
811 unsigned long expired;
815 * When set to zero, disable periodic writeback
817 if (!dirty_writeback_interval)
820 expired = wb->last_old_flush +
821 msecs_to_jiffies(dirty_writeback_interval * 10);
822 if (time_before(jiffies, expired))
825 wb->last_old_flush = jiffies;
826 nr_pages = global_page_state(NR_FILE_DIRTY) +
827 global_page_state(NR_UNSTABLE_NFS) +
828 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
831 struct wb_writeback_args args = {
832 .nr_pages = nr_pages,
833 .sync_mode = WB_SYNC_NONE,
838 return wb_writeback(wb, &args);
845 * Retrieve work items and do the writeback they describe
847 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
849 struct backing_dev_info *bdi = wb->bdi;
850 struct bdi_work *work;
853 while ((work = get_next_work_item(bdi, wb)) != NULL) {
854 struct wb_writeback_args args = work->args;
857 * Override sync mode, in case we must wait for completion
860 work->args.sync_mode = args.sync_mode = WB_SYNC_ALL;
863 * If this isn't a data integrity operation, just notify
864 * that we have seen this work and we are now starting it.
866 if (!test_bit(WS_ONSTACK, &work->state))
867 wb_clear_pending(wb, work);
869 wrote += wb_writeback(wb, &args);
872 * This is a data integrity writeback, so only do the
873 * notification when we have completed the work.
875 if (test_bit(WS_ONSTACK, &work->state))
876 wb_clear_pending(wb, work);
880 * Check for periodic writeback, kupdated() style
882 wrote += wb_check_old_data_flush(wb);
888 * Handle writeback of dirty data for the device backed by this bdi. Also
889 * wakes up periodically and does kupdated style flushing.
891 int bdi_writeback_task(struct bdi_writeback *wb)
893 unsigned long last_active = jiffies;
894 unsigned long wait_jiffies = -1UL;
897 while (!kthread_should_stop()) {
898 pages_written = wb_do_writeback(wb, 0);
901 last_active = jiffies;
902 else if (wait_jiffies != -1UL) {
903 unsigned long max_idle;
906 * Longest period of inactivity that we tolerate. If we
907 * see dirty data again later, the task will get
908 * recreated automatically.
910 max_idle = max(5UL * 60 * HZ, wait_jiffies);
911 if (time_after(jiffies, max_idle + last_active))
915 if (dirty_writeback_interval) {
916 wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10);
917 schedule_timeout_interruptible(wait_jiffies);
919 set_current_state(TASK_INTERRUPTIBLE);
920 if (list_empty_careful(&wb->bdi->work_list) &&
921 !kthread_should_stop())
923 __set_current_state(TASK_RUNNING);
933 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
936 void wakeup_flusher_threads(long nr_pages)
938 struct backing_dev_info *bdi;
939 struct wb_writeback_args args = {
940 .sync_mode = WB_SYNC_NONE,
944 args.nr_pages = nr_pages;
946 args.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_alloc_queue_work(bdi, &args);
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;
1013 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1014 * dirty the inode itself
1016 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1017 if (sb->s_op->dirty_inode)
1018 sb->s_op->dirty_inode(inode);
1022 * make sure that changes are seen by all cpus before we test i_state
1027 /* avoid the locking if we can */
1028 if ((inode->i_state & flags) == flags)
1031 if (unlikely(block_dump))
1032 block_dump___mark_inode_dirty(inode);
1034 spin_lock(&inode_lock);
1035 if ((inode->i_state & flags) != flags) {
1036 const int was_dirty = inode->i_state & I_DIRTY;
1038 inode->i_state |= flags;
1041 * If the inode is being synced, just update its dirty state.
1042 * The unlocker will place the inode on the appropriate
1043 * superblock list, based upon its state.
1045 if (inode->i_state & I_SYNC)
1049 * Only add valid (hashed) inodes to the superblock's
1050 * dirty list. Add blockdev inodes as well.
1052 if (!S_ISBLK(inode->i_mode)) {
1053 if (hlist_unhashed(&inode->i_hash))
1056 if (inode->i_state & (I_FREEING|I_CLEAR))
1060 * If the inode was already on b_dirty/b_io/b_more_io, don't
1061 * reposition it (that would break b_dirty time-ordering).
1064 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1065 struct backing_dev_info *bdi = wb->bdi;
1067 if (bdi_cap_writeback_dirty(bdi) &&
1068 !test_bit(BDI_registered, &bdi->state)) {
1070 printk(KERN_ERR "bdi-%s not registered\n",
1074 inode->dirtied_when = jiffies;
1075 list_move(&inode->i_list, &wb->b_dirty);
1079 spin_unlock(&inode_lock);
1081 EXPORT_SYMBOL(__mark_inode_dirty);
1084 * Write out a superblock's list of dirty inodes. A wait will be performed
1085 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1087 * If older_than_this is non-NULL, then only write out inodes which
1088 * had their first dirtying at a time earlier than *older_than_this.
1090 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1091 * This function assumes that the blockdev superblock's inodes are backed by
1092 * a variety of queues, so all inodes are searched. For other superblocks,
1093 * assume that all inodes are backed by the same queue.
1095 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1096 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1097 * on the writer throttling path, and we get decent balancing between many
1098 * throttled threads: we don't want them all piling up on inode_sync_wait.
1100 static void wait_sb_inodes(struct super_block *sb)
1102 struct inode *inode, *old_inode = NULL;
1105 * We need to be protected against the filesystem going from
1106 * r/o to r/w or vice versa.
1108 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1110 spin_lock(&inode_lock);
1113 * Data integrity sync. Must wait for all pages under writeback,
1114 * because there may have been pages dirtied before our sync
1115 * call, but which had writeout started before we write it out.
1116 * In which case, the inode may not be on the dirty list, but
1117 * we still have to wait for that writeout.
1119 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1120 struct address_space *mapping;
1122 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
1124 mapping = inode->i_mapping;
1125 if (mapping->nrpages == 0)
1128 spin_unlock(&inode_lock);
1130 * We hold a reference to 'inode' so it couldn't have
1131 * been removed from s_inodes list while we dropped the
1132 * inode_lock. We cannot iput the inode now as we can
1133 * be holding the last reference and we cannot iput it
1134 * under inode_lock. So we keep the reference and iput
1140 filemap_fdatawait(mapping);
1144 spin_lock(&inode_lock);
1146 spin_unlock(&inode_lock);
1151 * writeback_inodes_sb - writeback dirty inodes from given super_block
1152 * @sb: the superblock
1154 * Start writeback on some inodes on this super_block. No guarantees are made
1155 * on how many (if any) will be written, and this function does not wait
1156 * for IO completion of submitted IO. The number of pages submitted is
1159 void writeback_inodes_sb(struct super_block *sb)
1161 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
1162 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
1163 struct wb_writeback_args args = {
1165 .sync_mode = WB_SYNC_NONE,
1168 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1170 args.nr_pages = nr_dirty + nr_unstable +
1171 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
1173 bdi_queue_work_onstack(&args);
1175 EXPORT_SYMBOL(writeback_inodes_sb);
1178 * writeback_inodes_sb_if_idle - start writeback if none underway
1179 * @sb: the superblock
1181 * Invoke writeback_inodes_sb if no writeback is currently underway.
1182 * Returns 1 if writeback was started, 0 if not.
1184 int writeback_inodes_sb_if_idle(struct super_block *sb)
1186 if (!writeback_in_progress(sb->s_bdi)) {
1187 down_read(&sb->s_umount);
1188 writeback_inodes_sb(sb);
1189 up_read(&sb->s_umount);
1194 EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1197 * sync_inodes_sb - sync sb inode pages
1198 * @sb: the superblock
1200 * This function writes and waits on any dirty inode belonging to this
1201 * super_block. The number of pages synced is returned.
1203 void sync_inodes_sb(struct super_block *sb)
1205 struct wb_writeback_args args = {
1207 .sync_mode = WB_SYNC_ALL,
1208 .nr_pages = LONG_MAX,
1212 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1214 bdi_queue_work_onstack(&args);
1217 EXPORT_SYMBOL(sync_inodes_sb);
1220 * write_inode_now - write an inode to disk
1221 * @inode: inode to write to disk
1222 * @sync: whether the write should be synchronous or not
1224 * This function commits an inode to disk immediately if it is dirty. This is
1225 * primarily needed by knfsd.
1227 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1229 int write_inode_now(struct inode *inode, int sync)
1232 struct writeback_control wbc = {
1233 .nr_to_write = LONG_MAX,
1234 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1236 .range_end = LLONG_MAX,
1239 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1240 wbc.nr_to_write = 0;
1243 spin_lock(&inode_lock);
1244 ret = writeback_single_inode(inode, &wbc);
1245 spin_unlock(&inode_lock);
1247 inode_sync_wait(inode);
1250 EXPORT_SYMBOL(write_inode_now);
1253 * sync_inode - write an inode and its pages to disk.
1254 * @inode: the inode to sync
1255 * @wbc: controls the writeback mode
1257 * sync_inode() will write an inode and its pages to disk. It will also
1258 * correctly update the inode on its superblock's dirty inode lists and will
1259 * update inode->i_state.
1261 * The caller must have a ref on the inode.
1263 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1267 spin_lock(&inode_lock);
1268 ret = writeback_single_inode(inode, wbc);
1269 spin_unlock(&inode_lock);
1272 EXPORT_SYMBOL(sync_inode);