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/sched.h>
22 #include <linux/kthread.h>
23 #include <linux/freezer.h>
24 #include <linux/writeback.h>
25 #include <linux/blkdev.h>
26 #include <linux/backing-dev.h>
27 #include <linux/buffer_head.h>
30 #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info)
33 * We don't actually have pdflush, but this one is exported though /proc...
35 int nr_pdflush_threads;
38 * Passed into wb_writeback(), essentially a subset of writeback_control
40 struct wb_writeback_args {
42 struct super_block *sb;
43 enum writeback_sync_modes sync_mode;
50 * Work items for the bdi_writeback threads
53 struct list_head list; /* pending work list */
54 struct rcu_head rcu_head; /* for RCU free/clear of work */
56 unsigned long seen; /* threads that have seen this work */
57 atomic_t pending; /* number of threads still to do work */
59 struct wb_writeback_args args; /* writeback arguments */
61 unsigned long state; /* flag bits, see WS_* */
69 #define WS_USED (1 << WS_USED_B)
70 #define WS_ONSTACK (1 << WS_ONSTACK_B)
72 static inline bool bdi_work_on_stack(struct bdi_work *work)
74 return test_bit(WS_ONSTACK_B, &work->state);
77 static inline void bdi_work_init(struct bdi_work *work,
78 struct wb_writeback_args *args)
80 INIT_RCU_HEAD(&work->rcu_head);
82 work->state = WS_USED;
86 * writeback_in_progress - determine whether there is writeback in progress
87 * @bdi: the device's backing_dev_info structure.
89 * Determine whether there is writeback waiting to be handled against a
92 int writeback_in_progress(struct backing_dev_info *bdi)
94 return !list_empty(&bdi->work_list);
97 static void bdi_work_clear(struct bdi_work *work)
99 clear_bit(WS_USED_B, &work->state);
100 smp_mb__after_clear_bit();
102 * work can have disappeared at this point. bit waitq functions
103 * should be able to tolerate this, provided bdi_sched_wait does
104 * not dereference it's pointer argument.
106 wake_up_bit(&work->state, WS_USED_B);
109 static void bdi_work_free(struct rcu_head *head)
111 struct bdi_work *work = container_of(head, struct bdi_work, rcu_head);
113 if (!bdi_work_on_stack(work))
116 bdi_work_clear(work);
119 static void wb_work_complete(struct bdi_work *work)
121 const enum writeback_sync_modes sync_mode = work->args.sync_mode;
122 int onstack = bdi_work_on_stack(work);
125 * For allocated work, we can clear the done/seen bit right here.
126 * For on-stack work, we need to postpone both the clear and free
127 * to after the RCU grace period, since the stack could be invalidated
128 * as soon as bdi_work_clear() has done the wakeup.
131 bdi_work_clear(work);
132 if (sync_mode == WB_SYNC_NONE || onstack)
133 call_rcu(&work->rcu_head, bdi_work_free);
136 static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work)
139 * The caller has retrieved the work arguments from this work,
140 * drop our reference. If this is the last ref, delete and free it
142 if (atomic_dec_and_test(&work->pending)) {
143 struct backing_dev_info *bdi = wb->bdi;
145 spin_lock(&bdi->wb_lock);
146 list_del_rcu(&work->list);
147 spin_unlock(&bdi->wb_lock);
149 wb_work_complete(work);
153 static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work)
155 work->seen = bdi->wb_mask;
157 atomic_set(&work->pending, bdi->wb_cnt);
158 BUG_ON(!bdi->wb_cnt);
161 * list_add_tail_rcu() contains the necessary barriers to
162 * make sure the above stores are seen before the item is
163 * noticed on the list
165 spin_lock(&bdi->wb_lock);
166 list_add_tail_rcu(&work->list, &bdi->work_list);
167 spin_unlock(&bdi->wb_lock);
170 * If the default thread isn't there, make sure we add it. When
171 * it gets created and wakes up, we'll run this work.
173 if (unlikely(list_empty_careful(&bdi->wb_list)))
174 wake_up_process(default_backing_dev_info.wb.task);
176 struct bdi_writeback *wb = &bdi->wb;
179 wake_up_process(wb->task);
184 * Used for on-stack allocated work items. The caller needs to wait until
185 * the wb threads have acked the work before it's safe to continue.
187 static void bdi_wait_on_work_clear(struct bdi_work *work)
189 wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait,
190 TASK_UNINTERRUPTIBLE);
193 static void bdi_alloc_queue_work(struct backing_dev_info *bdi,
194 struct wb_writeback_args *args)
196 struct bdi_work *work;
199 * This is WB_SYNC_NONE writeback, so if allocation fails just
200 * wakeup the thread for old dirty data writeback
202 work = kmalloc(sizeof(*work), GFP_ATOMIC);
204 bdi_work_init(work, args);
205 bdi_queue_work(bdi, work);
207 struct bdi_writeback *wb = &bdi->wb;
210 wake_up_process(wb->task);
215 * bdi_sync_writeback - start and wait for writeback
216 * @bdi: the backing device to write from
217 * @sb: write inodes from this super_block
220 * This does WB_SYNC_ALL data integrity writeback and waits for the
221 * IO to complete. Callers must hold the sb s_umount semaphore for
222 * reading, to avoid having the super disappear before we are done.
224 static void bdi_sync_writeback(struct backing_dev_info *bdi,
225 struct super_block *sb)
227 struct wb_writeback_args args = {
229 .sync_mode = WB_SYNC_ALL,
230 .nr_pages = LONG_MAX,
233 struct bdi_work work;
235 bdi_work_init(&work, &args);
236 work.state |= WS_ONSTACK;
238 bdi_queue_work(bdi, &work);
239 bdi_wait_on_work_clear(&work);
243 * bdi_start_writeback - start writeback
244 * @bdi: the backing device to write from
245 * @nr_pages: the number of pages to write
248 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
249 * started when this function returns, we make no guarentees on
250 * completion. Caller need not hold sb s_umount semaphore.
253 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
255 struct wb_writeback_args args = {
256 .sync_mode = WB_SYNC_NONE,
257 .nr_pages = nr_pages,
262 * We treat @nr_pages=0 as the special case to do background writeback,
263 * ie. to sync pages until the background dirty threshold is reached.
266 args.nr_pages = LONG_MAX;
267 args.for_background = 1;
270 bdi_alloc_queue_work(bdi, &args);
274 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
275 * furthest end of its superblock's dirty-inode list.
277 * Before stamping the inode's ->dirtied_when, we check to see whether it is
278 * already the most-recently-dirtied inode on the b_dirty list. If that is
279 * the case then the inode must have been redirtied while it was being written
280 * out and we don't reset its dirtied_when.
282 static void redirty_tail(struct inode *inode)
284 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
286 if (!list_empty(&wb->b_dirty)) {
289 tail = list_entry(wb->b_dirty.next, struct inode, i_list);
290 if (time_before(inode->dirtied_when, tail->dirtied_when))
291 inode->dirtied_when = jiffies;
293 list_move(&inode->i_list, &wb->b_dirty);
297 * requeue inode for re-scanning after bdi->b_io list is exhausted.
299 static void requeue_io(struct inode *inode)
301 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
303 list_move(&inode->i_list, &wb->b_more_io);
306 static void inode_sync_complete(struct inode *inode)
309 * Prevent speculative execution through spin_unlock(&inode_lock);
312 wake_up_bit(&inode->i_state, __I_SYNC);
315 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
317 bool ret = time_after(inode->dirtied_when, t);
320 * For inodes being constantly redirtied, dirtied_when can get stuck.
321 * It _appears_ to be in the future, but is actually in distant past.
322 * This test is necessary to prevent such wrapped-around relative times
323 * from permanently stopping the whole bdi writeback.
325 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
331 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
333 static void move_expired_inodes(struct list_head *delaying_queue,
334 struct list_head *dispatch_queue,
335 unsigned long *older_than_this)
338 struct list_head *pos, *node;
339 struct super_block *sb = NULL;
343 while (!list_empty(delaying_queue)) {
344 inode = list_entry(delaying_queue->prev, struct inode, i_list);
345 if (older_than_this &&
346 inode_dirtied_after(inode, *older_than_this))
348 if (sb && sb != inode->i_sb)
351 list_move(&inode->i_list, &tmp);
354 /* just one sb in list, splice to dispatch_queue and we're done */
356 list_splice(&tmp, dispatch_queue);
360 /* Move inodes from one superblock together */
361 while (!list_empty(&tmp)) {
362 inode = list_entry(tmp.prev, struct inode, i_list);
364 list_for_each_prev_safe(pos, node, &tmp) {
365 inode = list_entry(pos, struct inode, i_list);
366 if (inode->i_sb == sb)
367 list_move(&inode->i_list, dispatch_queue);
373 * Queue all expired dirty inodes for io, eldest first.
375 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
377 list_splice_init(&wb->b_more_io, wb->b_io.prev);
378 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
381 static int write_inode(struct inode *inode, int sync)
383 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
384 return inode->i_sb->s_op->write_inode(inode, sync);
389 * Wait for writeback on an inode to complete.
391 static void inode_wait_for_writeback(struct inode *inode)
393 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
394 wait_queue_head_t *wqh;
396 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
398 spin_unlock(&inode_lock);
399 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
400 spin_lock(&inode_lock);
401 } while (inode->i_state & I_SYNC);
405 * Write out an inode's dirty pages. Called under inode_lock. Either the
406 * caller has ref on the inode (either via __iget or via syscall against an fd)
407 * or the inode has I_WILL_FREE set (via generic_forget_inode)
409 * If `wait' is set, wait on the writeout.
411 * The whole writeout design is quite complex and fragile. We want to avoid
412 * starvation of particular inodes when others are being redirtied, prevent
415 * Called under inode_lock.
418 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
420 struct address_space *mapping = inode->i_mapping;
421 int wait = wbc->sync_mode == WB_SYNC_ALL;
425 if (!atomic_read(&inode->i_count))
426 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
428 WARN_ON(inode->i_state & I_WILL_FREE);
430 if (inode->i_state & I_SYNC) {
432 * If this inode is locked for writeback and we are not doing
433 * writeback-for-data-integrity, move it to b_more_io so that
434 * writeback can proceed with the other inodes on s_io.
436 * We'll have another go at writing back this inode when we
437 * completed a full scan of b_io.
445 * It's a data-integrity sync. We must wait.
447 inode_wait_for_writeback(inode);
450 BUG_ON(inode->i_state & I_SYNC);
452 /* Set I_SYNC, reset I_DIRTY */
453 dirty = inode->i_state & I_DIRTY;
454 inode->i_state |= I_SYNC;
455 inode->i_state &= ~I_DIRTY;
457 spin_unlock(&inode_lock);
459 ret = do_writepages(mapping, wbc);
461 /* Don't write the inode if only I_DIRTY_PAGES was set */
462 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
463 int err = write_inode(inode, wait);
469 int err = filemap_fdatawait(mapping);
474 spin_lock(&inode_lock);
475 inode->i_state &= ~I_SYNC;
476 if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
477 if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) {
479 * More pages get dirtied by a fast dirtier.
482 } else if (inode->i_state & I_DIRTY) {
484 * At least XFS will redirty the inode during the
485 * writeback (delalloc) and on io completion (isize).
488 } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
490 * We didn't write back all the pages. nfs_writepages()
491 * sometimes bales out without doing anything. Redirty
492 * the inode; Move it from b_io onto b_more_io/b_dirty.
495 * akpm: if the caller was the kupdate function we put
496 * this inode at the head of b_dirty so it gets first
497 * consideration. Otherwise, move it to the tail, for
498 * the reasons described there. I'm not really sure
499 * how much sense this makes. Presumably I had a good
500 * reasons for doing it this way, and I'd rather not
501 * muck with it at present.
503 if (wbc->for_kupdate) {
505 * For the kupdate function we move the inode
506 * to b_more_io so it will get more writeout as
507 * soon as the queue becomes uncongested.
509 inode->i_state |= I_DIRTY_PAGES;
511 if (wbc->nr_to_write <= 0) {
513 * slice used up: queue for next turn
518 * somehow blocked: retry later
524 * Otherwise fully redirty the inode so that
525 * other inodes on this superblock will get some
526 * writeout. Otherwise heavy writing to one
527 * file would indefinitely suspend writeout of
528 * all the other files.
530 inode->i_state |= I_DIRTY_PAGES;
533 } else if (atomic_read(&inode->i_count)) {
535 * The inode is clean, inuse
537 list_move(&inode->i_list, &inode_in_use);
540 * The inode is clean, unused
542 list_move(&inode->i_list, &inode_unused);
545 inode_sync_complete(inode);
549 static void unpin_sb_for_writeback(struct super_block **psb)
551 struct super_block *sb = *psb;
554 up_read(&sb->s_umount);
561 * For WB_SYNC_NONE writeback, the caller does not have the sb pinned
562 * before calling writeback. So make sure that we do pin it, so it doesn't
563 * go away while we are writing inodes from it.
565 * Returns 0 if the super was successfully pinned (or pinning wasn't needed),
568 static int pin_sb_for_writeback(struct writeback_control *wbc,
569 struct inode *inode, struct super_block **psb)
571 struct super_block *sb = inode->i_sb;
574 * If this sb is already pinned, nothing more to do. If not and
575 * *psb is non-NULL, unpin the old one first
580 unpin_sb_for_writeback(psb);
583 * Caller must already hold the ref for this
585 if (wbc->sync_mode == WB_SYNC_ALL) {
586 WARN_ON(!rwsem_is_locked(&sb->s_umount));
592 if (down_read_trylock(&sb->s_umount)) {
594 spin_unlock(&sb_lock);
598 * umounted, drop rwsem again and fall through to failure
600 up_read(&sb->s_umount);
604 spin_unlock(&sb_lock);
611 static void writeback_inodes_wb(struct bdi_writeback *wb,
612 struct writeback_control *wbc)
614 struct super_block *sb = wbc->sb, *pin_sb = NULL;
615 const int is_blkdev_sb = sb_is_blkdev_sb(sb);
616 const unsigned long start = jiffies; /* livelock avoidance */
618 spin_lock(&inode_lock);
620 if (!wbc->for_kupdate || list_empty(&wb->b_io))
621 queue_io(wb, wbc->older_than_this);
623 while (!list_empty(&wb->b_io)) {
624 struct inode *inode = list_entry(wb->b_io.prev,
625 struct inode, i_list);
629 * super block given and doesn't match, skip this inode
631 if (sb && sb != inode->i_sb) {
636 if (!bdi_cap_writeback_dirty(wb->bdi)) {
640 * Dirty memory-backed blockdev: the ramdisk
641 * driver does this. Skip just this inode
646 * Dirty memory-backed inode against a filesystem other
647 * than the kernel-internal bdev filesystem. Skip the
653 if (inode->i_state & (I_NEW | I_WILL_FREE)) {
658 if (wbc->nonblocking && bdi_write_congested(wb->bdi)) {
659 wbc->encountered_congestion = 1;
661 break; /* Skip a congested fs */
663 continue; /* Skip a congested blockdev */
667 * Was this inode dirtied after sync_sb_inodes was called?
668 * This keeps sync from extra jobs and livelock.
670 if (inode_dirtied_after(inode, start))
673 if (pin_sb_for_writeback(wbc, inode, &pin_sb)) {
678 BUG_ON(inode->i_state & (I_FREEING | I_CLEAR));
680 pages_skipped = wbc->pages_skipped;
681 writeback_single_inode(inode, wbc);
682 if (wbc->pages_skipped != pages_skipped) {
684 * writeback is not making progress due to locked
685 * buffers. Skip this inode for now.
689 spin_unlock(&inode_lock);
692 spin_lock(&inode_lock);
693 if (wbc->nr_to_write <= 0) {
697 if (!list_empty(&wb->b_more_io))
701 unpin_sb_for_writeback(&pin_sb);
703 spin_unlock(&inode_lock);
704 /* Leave any unwritten inodes on b_io */
707 void writeback_inodes_wbc(struct writeback_control *wbc)
709 struct backing_dev_info *bdi = wbc->bdi;
711 writeback_inodes_wb(&bdi->wb, wbc);
715 * The maximum number of pages to writeout in a single bdi flush/kupdate
716 * operation. We do this so we don't hold I_SYNC against an inode for
717 * enormous amounts of time, which would block a userspace task which has
718 * been forced to throttle against that inode. Also, the code reevaluates
719 * the dirty each time it has written this many pages.
721 #define MAX_WRITEBACK_PAGES 1024
723 static inline bool over_bground_thresh(void)
725 unsigned long background_thresh, dirty_thresh;
727 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
729 return (global_page_state(NR_FILE_DIRTY) +
730 global_page_state(NR_UNSTABLE_NFS) >= background_thresh);
734 * Explicit flushing or periodic writeback of "old" data.
736 * Define "old": the first time one of an inode's pages is dirtied, we mark the
737 * dirtying-time in the inode's address_space. So this periodic writeback code
738 * just walks the superblock inode list, writing back any inodes which are
739 * older than a specific point in time.
741 * Try to run once per dirty_writeback_interval. But if a writeback event
742 * takes longer than a dirty_writeback_interval interval, then leave a
745 * older_than_this takes precedence over nr_to_write. So we'll only write back
746 * all dirty pages if they are all attached to "old" mappings.
748 static long wb_writeback(struct bdi_writeback *wb,
749 struct wb_writeback_args *args)
751 struct writeback_control wbc = {
754 .sync_mode = args->sync_mode,
755 .older_than_this = NULL,
756 .for_kupdate = args->for_kupdate,
757 .range_cyclic = args->range_cyclic,
759 unsigned long oldest_jif;
763 if (wbc.for_kupdate) {
764 wbc.older_than_this = &oldest_jif;
765 oldest_jif = jiffies -
766 msecs_to_jiffies(dirty_expire_interval * 10);
768 if (!wbc.range_cyclic) {
770 wbc.range_end = LLONG_MAX;
775 * Stop writeback when nr_pages has been consumed
777 if (args->nr_pages <= 0)
781 * For background writeout, stop when we are below the
782 * background dirty threshold
784 if (args->for_background && !over_bground_thresh())
788 wbc.encountered_congestion = 0;
789 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
790 wbc.pages_skipped = 0;
791 writeback_inodes_wb(wb, &wbc);
792 args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
793 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
796 * If we consumed everything, see if we have more
798 if (wbc.nr_to_write <= 0)
801 * Didn't write everything and we don't have more IO, bail
806 * Did we write something? Try for more
808 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
811 * Nothing written. Wait for some inode to
812 * become available for writeback. Otherwise
813 * we'll just busyloop.
815 spin_lock(&inode_lock);
816 if (!list_empty(&wb->b_more_io)) {
817 inode = list_entry(wb->b_more_io.prev,
818 struct inode, i_list);
819 inode_wait_for_writeback(inode);
821 spin_unlock(&inode_lock);
828 * Return the next bdi_work struct that hasn't been processed by this
829 * wb thread yet. ->seen is initially set for each thread that exists
830 * for this device, when a thread first notices a piece of work it
831 * clears its bit. Depending on writeback type, the thread will notify
832 * completion on either receiving the work (WB_SYNC_NONE) or after
833 * it is done (WB_SYNC_ALL).
835 static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi,
836 struct bdi_writeback *wb)
838 struct bdi_work *work, *ret = NULL;
842 list_for_each_entry_rcu(work, &bdi->work_list, list) {
843 if (!test_bit(wb->nr, &work->seen))
845 clear_bit(wb->nr, &work->seen);
855 static long wb_check_old_data_flush(struct bdi_writeback *wb)
857 unsigned long expired;
860 expired = wb->last_old_flush +
861 msecs_to_jiffies(dirty_writeback_interval * 10);
862 if (time_before(jiffies, expired))
865 wb->last_old_flush = jiffies;
866 nr_pages = global_page_state(NR_FILE_DIRTY) +
867 global_page_state(NR_UNSTABLE_NFS) +
868 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
871 struct wb_writeback_args args = {
872 .nr_pages = nr_pages,
873 .sync_mode = WB_SYNC_NONE,
878 return wb_writeback(wb, &args);
885 * Retrieve work items and do the writeback they describe
887 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
889 struct backing_dev_info *bdi = wb->bdi;
890 struct bdi_work *work;
893 while ((work = get_next_work_item(bdi, wb)) != NULL) {
894 struct wb_writeback_args args = work->args;
897 * Override sync mode, in case we must wait for completion
900 work->args.sync_mode = args.sync_mode = WB_SYNC_ALL;
903 * If this isn't a data integrity operation, just notify
904 * that we have seen this work and we are now starting it.
906 if (args.sync_mode == WB_SYNC_NONE)
907 wb_clear_pending(wb, work);
909 wrote += wb_writeback(wb, &args);
912 * This is a data integrity writeback, so only do the
913 * notification when we have completed the work.
915 if (args.sync_mode == WB_SYNC_ALL)
916 wb_clear_pending(wb, work);
920 * Check for periodic writeback, kupdated() style
922 wrote += wb_check_old_data_flush(wb);
928 * Handle writeback of dirty data for the device backed by this bdi. Also
929 * wakes up periodically and does kupdated style flushing.
931 int bdi_writeback_task(struct bdi_writeback *wb)
933 unsigned long last_active = jiffies;
934 unsigned long wait_jiffies = -1UL;
937 while (!kthread_should_stop()) {
938 pages_written = wb_do_writeback(wb, 0);
941 last_active = jiffies;
942 else if (wait_jiffies != -1UL) {
943 unsigned long max_idle;
946 * Longest period of inactivity that we tolerate. If we
947 * see dirty data again later, the task will get
948 * recreated automatically.
950 max_idle = max(5UL * 60 * HZ, wait_jiffies);
951 if (time_after(jiffies, max_idle + last_active))
955 wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10);
956 schedule_timeout_interruptible(wait_jiffies);
964 * Schedule writeback for all backing devices. This does WB_SYNC_NONE
965 * writeback, for integrity writeback see bdi_sync_writeback().
967 static void bdi_writeback_all(struct super_block *sb, long nr_pages)
969 struct wb_writeback_args args = {
971 .nr_pages = nr_pages,
972 .sync_mode = WB_SYNC_NONE,
974 struct backing_dev_info *bdi;
978 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
979 if (!bdi_has_dirty_io(bdi))
982 bdi_alloc_queue_work(bdi, &args);
989 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
992 void wakeup_flusher_threads(long nr_pages)
995 nr_pages = global_page_state(NR_FILE_DIRTY) +
996 global_page_state(NR_UNSTABLE_NFS);
997 bdi_writeback_all(NULL, nr_pages);
1000 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1002 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1003 struct dentry *dentry;
1004 const char *name = "?";
1006 dentry = d_find_alias(inode);
1008 spin_lock(&dentry->d_lock);
1009 name = (const char *) dentry->d_name.name;
1012 "%s(%d): dirtied inode %lu (%s) on %s\n",
1013 current->comm, task_pid_nr(current), inode->i_ino,
1014 name, inode->i_sb->s_id);
1016 spin_unlock(&dentry->d_lock);
1023 * __mark_inode_dirty - internal function
1024 * @inode: inode to mark
1025 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1026 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1027 * mark_inode_dirty_sync.
1029 * Put the inode on the super block's dirty list.
1031 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1032 * dirty list only if it is hashed or if it refers to a blockdev.
1033 * If it was not hashed, it will never be added to the dirty list
1034 * even if it is later hashed, as it will have been marked dirty already.
1036 * In short, make sure you hash any inodes _before_ you start marking
1039 * This function *must* be atomic for the I_DIRTY_PAGES case -
1040 * set_page_dirty() is called under spinlock in several places.
1042 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1043 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1044 * the kernel-internal blockdev inode represents the dirtying time of the
1045 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1046 * page->mapping->host, so the page-dirtying time is recorded in the internal
1049 void __mark_inode_dirty(struct inode *inode, int flags)
1051 struct super_block *sb = inode->i_sb;
1054 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1055 * dirty the inode itself
1057 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1058 if (sb->s_op->dirty_inode)
1059 sb->s_op->dirty_inode(inode);
1063 * make sure that changes are seen by all cpus before we test i_state
1068 /* avoid the locking if we can */
1069 if ((inode->i_state & flags) == flags)
1072 if (unlikely(block_dump))
1073 block_dump___mark_inode_dirty(inode);
1075 spin_lock(&inode_lock);
1076 if ((inode->i_state & flags) != flags) {
1077 const int was_dirty = inode->i_state & I_DIRTY;
1079 inode->i_state |= flags;
1082 * If the inode is being synced, just update its dirty state.
1083 * The unlocker will place the inode on the appropriate
1084 * superblock list, based upon its state.
1086 if (inode->i_state & I_SYNC)
1090 * Only add valid (hashed) inodes to the superblock's
1091 * dirty list. Add blockdev inodes as well.
1093 if (!S_ISBLK(inode->i_mode)) {
1094 if (hlist_unhashed(&inode->i_hash))
1097 if (inode->i_state & (I_FREEING|I_CLEAR))
1101 * If the inode was already on b_dirty/b_io/b_more_io, don't
1102 * reposition it (that would break b_dirty time-ordering).
1105 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1106 struct backing_dev_info *bdi = wb->bdi;
1108 if (bdi_cap_writeback_dirty(bdi) &&
1109 !test_bit(BDI_registered, &bdi->state)) {
1111 printk(KERN_ERR "bdi-%s not registered\n",
1115 inode->dirtied_when = jiffies;
1116 list_move(&inode->i_list, &wb->b_dirty);
1120 spin_unlock(&inode_lock);
1122 EXPORT_SYMBOL(__mark_inode_dirty);
1125 * Write out a superblock's list of dirty inodes. A wait will be performed
1126 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1128 * If older_than_this is non-NULL, then only write out inodes which
1129 * had their first dirtying at a time earlier than *older_than_this.
1131 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1132 * This function assumes that the blockdev superblock's inodes are backed by
1133 * a variety of queues, so all inodes are searched. For other superblocks,
1134 * assume that all inodes are backed by the same queue.
1136 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1137 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1138 * on the writer throttling path, and we get decent balancing between many
1139 * throttled threads: we don't want them all piling up on inode_sync_wait.
1141 static void wait_sb_inodes(struct super_block *sb)
1143 struct inode *inode, *old_inode = NULL;
1146 * We need to be protected against the filesystem going from
1147 * r/o to r/w or vice versa.
1149 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1151 spin_lock(&inode_lock);
1154 * Data integrity sync. Must wait for all pages under writeback,
1155 * because there may have been pages dirtied before our sync
1156 * call, but which had writeout started before we write it out.
1157 * In which case, the inode may not be on the dirty list, but
1158 * we still have to wait for that writeout.
1160 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1161 struct address_space *mapping;
1163 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
1165 mapping = inode->i_mapping;
1166 if (mapping->nrpages == 0)
1169 spin_unlock(&inode_lock);
1171 * We hold a reference to 'inode' so it couldn't have
1172 * been removed from s_inodes list while we dropped the
1173 * inode_lock. We cannot iput the inode now as we can
1174 * be holding the last reference and we cannot iput it
1175 * under inode_lock. So we keep the reference and iput
1181 filemap_fdatawait(mapping);
1185 spin_lock(&inode_lock);
1187 spin_unlock(&inode_lock);
1192 * writeback_inodes_sb - writeback dirty inodes from given super_block
1193 * @sb: the superblock
1195 * Start writeback on some inodes on this super_block. No guarantees are made
1196 * on how many (if any) will be written, and this function does not wait
1197 * for IO completion of submitted IO. The number of pages submitted is
1200 void writeback_inodes_sb(struct super_block *sb)
1202 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
1203 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
1206 nr_to_write = nr_dirty + nr_unstable +
1207 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
1209 bdi_start_writeback(sb->s_bdi, nr_to_write);
1211 EXPORT_SYMBOL(writeback_inodes_sb);
1214 * sync_inodes_sb - sync sb inode pages
1215 * @sb: the superblock
1217 * This function writes and waits on any dirty inode belonging to this
1218 * super_block. The number of pages synced is returned.
1220 void sync_inodes_sb(struct super_block *sb)
1222 bdi_sync_writeback(sb->s_bdi, sb);
1225 EXPORT_SYMBOL(sync_inodes_sb);
1228 * write_inode_now - write an inode to disk
1229 * @inode: inode to write to disk
1230 * @sync: whether the write should be synchronous or not
1232 * This function commits an inode to disk immediately if it is dirty. This is
1233 * primarily needed by knfsd.
1235 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1237 int write_inode_now(struct inode *inode, int sync)
1240 struct writeback_control wbc = {
1241 .nr_to_write = LONG_MAX,
1242 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1244 .range_end = LLONG_MAX,
1247 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1248 wbc.nr_to_write = 0;
1251 spin_lock(&inode_lock);
1252 ret = writeback_single_inode(inode, &wbc);
1253 spin_unlock(&inode_lock);
1255 inode_sync_wait(inode);
1258 EXPORT_SYMBOL(write_inode_now);
1261 * sync_inode - write an inode and its pages to disk.
1262 * @inode: the inode to sync
1263 * @wbc: controls the writeback mode
1265 * sync_inode() will write an inode and its pages to disk. It will also
1266 * correctly update the inode on its superblock's dirty inode lists and will
1267 * update inode->i_state.
1269 * The caller must have a ref on the inode.
1271 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1275 spin_lock(&inode_lock);
1276 ret = writeback_single_inode(inode, wbc);
1277 spin_unlock(&inode_lock);
1280 EXPORT_SYMBOL(sync_inode);