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;
49 * Work items for the bdi_writeback threads
52 struct list_head list;
53 struct rcu_head rcu_head;
58 struct wb_writeback_args args;
68 #define WS_USED (1 << WS_USED_B)
69 #define WS_ONSTACK (1 << WS_ONSTACK_B)
71 static inline bool bdi_work_on_stack(struct bdi_work *work)
73 return test_bit(WS_ONSTACK_B, &work->state);
76 static inline void bdi_work_init(struct bdi_work *work,
77 struct writeback_control *wbc)
79 INIT_RCU_HEAD(&work->rcu_head);
80 work->args.sb = wbc->sb;
81 work->args.nr_pages = wbc->nr_to_write;
82 work->args.sync_mode = wbc->sync_mode;
83 work->args.range_cyclic = wbc->range_cyclic;
84 work->args.for_kupdate = 0;
85 work->state = WS_USED;
89 * writeback_in_progress - determine whether there is writeback in progress
90 * @bdi: the device's backing_dev_info structure.
92 * Determine whether there is writeback waiting to be handled against a
95 int writeback_in_progress(struct backing_dev_info *bdi)
97 return !list_empty(&bdi->work_list);
100 static void bdi_work_clear(struct bdi_work *work)
102 clear_bit(WS_USED_B, &work->state);
103 smp_mb__after_clear_bit();
104 wake_up_bit(&work->state, WS_USED_B);
107 static void bdi_work_free(struct rcu_head *head)
109 struct bdi_work *work = container_of(head, struct bdi_work, rcu_head);
111 if (!bdi_work_on_stack(work))
114 bdi_work_clear(work);
117 static void wb_work_complete(struct bdi_work *work)
119 const enum writeback_sync_modes sync_mode = work->args.sync_mode;
122 * For allocated work, we can clear the done/seen bit right here.
123 * For on-stack work, we need to postpone both the clear and free
124 * to after the RCU grace period, since the stack could be invalidated
125 * as soon as bdi_work_clear() has done the wakeup.
127 if (!bdi_work_on_stack(work))
128 bdi_work_clear(work);
129 if (sync_mode == WB_SYNC_NONE || bdi_work_on_stack(work))
130 call_rcu(&work->rcu_head, bdi_work_free);
133 static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work)
136 * The caller has retrieved the work arguments from this work,
137 * drop our reference. If this is the last ref, delete and free it
139 if (atomic_dec_and_test(&work->pending)) {
140 struct backing_dev_info *bdi = wb->bdi;
142 spin_lock(&bdi->wb_lock);
143 list_del_rcu(&work->list);
144 spin_unlock(&bdi->wb_lock);
146 wb_work_complete(work);
150 static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work)
152 work->seen = bdi->wb_mask;
154 atomic_set(&work->pending, bdi->wb_cnt);
155 BUG_ON(!bdi->wb_cnt);
158 * Make sure stores are seen before it appears on the list
162 spin_lock(&bdi->wb_lock);
163 list_add_tail_rcu(&work->list, &bdi->work_list);
164 spin_unlock(&bdi->wb_lock);
167 * If the default thread isn't there, make sure we add it. When
168 * it gets created and wakes up, we'll run this work.
170 if (unlikely(list_empty_careful(&bdi->wb_list)))
171 wake_up_process(default_backing_dev_info.wb.task);
173 struct bdi_writeback *wb = &bdi->wb;
176 * End work now if this wb has no dirty IO pending. Otherwise
177 * wakeup the handling thread
179 if (!wb_has_dirty_io(wb))
180 wb_clear_pending(wb, work);
182 wake_up_process(wb->task);
187 * Used for on-stack allocated work items. The caller needs to wait until
188 * the wb threads have acked the work before it's safe to continue.
190 static void bdi_wait_on_work_clear(struct bdi_work *work)
192 wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait,
193 TASK_UNINTERRUPTIBLE);
196 static void bdi_alloc_queue_work(struct backing_dev_info *bdi,
197 struct writeback_control *wbc)
199 struct bdi_work *work;
202 * This is WB_SYNC_NONE writeback, so if allocation fails just
203 * wakeup the thread for old dirty data writeback
205 work = kmalloc(sizeof(*work), GFP_ATOMIC);
207 bdi_work_init(work, wbc);
208 bdi_queue_work(bdi, work);
210 struct bdi_writeback *wb = &bdi->wb;
213 wake_up_process(wb->task);
217 void bdi_start_writeback(struct writeback_control *wbc)
220 * WB_SYNC_NONE is opportunistic writeback. If this allocation fails,
221 * bdi_queue_work() will wake up the thread and flush old data. This
222 * should ensure some amount of progress in freeing memory.
224 if (wbc->sync_mode != WB_SYNC_ALL)
225 bdi_alloc_queue_work(wbc->bdi, wbc);
227 struct bdi_work work;
229 bdi_work_init(&work, wbc);
230 work.state |= WS_ONSTACK;
232 bdi_queue_work(wbc->bdi, &work);
233 bdi_wait_on_work_clear(&work);
238 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
239 * furthest end of its superblock's dirty-inode list.
241 * Before stamping the inode's ->dirtied_when, we check to see whether it is
242 * already the most-recently-dirtied inode on the b_dirty list. If that is
243 * the case then the inode must have been redirtied while it was being written
244 * out and we don't reset its dirtied_when.
246 static void redirty_tail(struct inode *inode)
248 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
250 if (!list_empty(&wb->b_dirty)) {
253 tail = list_entry(wb->b_dirty.next, struct inode, i_list);
254 if (time_before(inode->dirtied_when, tail->dirtied_when))
255 inode->dirtied_when = jiffies;
257 list_move(&inode->i_list, &wb->b_dirty);
261 * requeue inode for re-scanning after bdi->b_io list is exhausted.
263 static void requeue_io(struct inode *inode)
265 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
267 list_move(&inode->i_list, &wb->b_more_io);
270 static void inode_sync_complete(struct inode *inode)
273 * Prevent speculative execution through spin_unlock(&inode_lock);
276 wake_up_bit(&inode->i_state, __I_SYNC);
279 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
281 bool ret = time_after(inode->dirtied_when, t);
284 * For inodes being constantly redirtied, dirtied_when can get stuck.
285 * It _appears_ to be in the future, but is actually in distant past.
286 * This test is necessary to prevent such wrapped-around relative times
287 * from permanently stopping the whole pdflush writeback.
289 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
295 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
297 static void move_expired_inodes(struct list_head *delaying_queue,
298 struct list_head *dispatch_queue,
299 unsigned long *older_than_this)
301 while (!list_empty(delaying_queue)) {
302 struct inode *inode = list_entry(delaying_queue->prev,
303 struct inode, i_list);
304 if (older_than_this &&
305 inode_dirtied_after(inode, *older_than_this))
307 list_move(&inode->i_list, dispatch_queue);
312 * Queue all expired dirty inodes for io, eldest first.
314 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
316 list_splice_init(&wb->b_more_io, wb->b_io.prev);
317 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
320 static int write_inode(struct inode *inode, int sync)
322 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
323 return inode->i_sb->s_op->write_inode(inode, sync);
328 * Wait for writeback on an inode to complete.
330 static void inode_wait_for_writeback(struct inode *inode)
332 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
333 wait_queue_head_t *wqh;
335 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
337 spin_unlock(&inode_lock);
338 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
339 spin_lock(&inode_lock);
340 } while (inode->i_state & I_SYNC);
344 * Write out an inode's dirty pages. Called under inode_lock. Either the
345 * caller has ref on the inode (either via __iget or via syscall against an fd)
346 * or the inode has I_WILL_FREE set (via generic_forget_inode)
348 * If `wait' is set, wait on the writeout.
350 * The whole writeout design is quite complex and fragile. We want to avoid
351 * starvation of particular inodes when others are being redirtied, prevent
354 * Called under inode_lock.
357 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
359 struct address_space *mapping = inode->i_mapping;
360 int wait = wbc->sync_mode == WB_SYNC_ALL;
364 if (!atomic_read(&inode->i_count))
365 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
367 WARN_ON(inode->i_state & I_WILL_FREE);
369 if (inode->i_state & I_SYNC) {
371 * If this inode is locked for writeback and we are not doing
372 * writeback-for-data-integrity, move it to b_more_io so that
373 * writeback can proceed with the other inodes on s_io.
375 * We'll have another go at writing back this inode when we
376 * completed a full scan of b_io.
384 * It's a data-integrity sync. We must wait.
386 inode_wait_for_writeback(inode);
389 BUG_ON(inode->i_state & I_SYNC);
391 /* Set I_SYNC, reset I_DIRTY */
392 dirty = inode->i_state & I_DIRTY;
393 inode->i_state |= I_SYNC;
394 inode->i_state &= ~I_DIRTY;
396 spin_unlock(&inode_lock);
398 ret = do_writepages(mapping, wbc);
400 /* Don't write the inode if only I_DIRTY_PAGES was set */
401 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
402 int err = write_inode(inode, wait);
408 int err = filemap_fdatawait(mapping);
413 spin_lock(&inode_lock);
414 inode->i_state &= ~I_SYNC;
415 if (!(inode->i_state & (I_FREEING | I_CLEAR))) {
416 if (!(inode->i_state & I_DIRTY) &&
417 mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
419 * We didn't write back all the pages. nfs_writepages()
420 * sometimes bales out without doing anything. Redirty
421 * the inode; Move it from b_io onto b_more_io/b_dirty.
424 * akpm: if the caller was the kupdate function we put
425 * this inode at the head of b_dirty so it gets first
426 * consideration. Otherwise, move it to the tail, for
427 * the reasons described there. I'm not really sure
428 * how much sense this makes. Presumably I had a good
429 * reasons for doing it this way, and I'd rather not
430 * muck with it at present.
432 if (wbc->for_kupdate) {
434 * For the kupdate function we move the inode
435 * to b_more_io so it will get more writeout as
436 * soon as the queue becomes uncongested.
438 inode->i_state |= I_DIRTY_PAGES;
439 if (wbc->nr_to_write <= 0) {
441 * slice used up: queue for next turn
446 * somehow blocked: retry later
452 * Otherwise fully redirty the inode so that
453 * other inodes on this superblock will get some
454 * writeout. Otherwise heavy writing to one
455 * file would indefinitely suspend writeout of
456 * all the other files.
458 inode->i_state |= I_DIRTY_PAGES;
461 } else if (inode->i_state & I_DIRTY) {
463 * Someone redirtied the inode while were writing back
467 } else if (atomic_read(&inode->i_count)) {
469 * The inode is clean, inuse
471 list_move(&inode->i_list, &inode_in_use);
474 * The inode is clean, unused
476 list_move(&inode->i_list, &inode_unused);
479 inode_sync_complete(inode);
484 * For WB_SYNC_NONE writeback, the caller does not have the sb pinned
485 * before calling writeback. So make sure that we do pin it, so it doesn't
486 * go away while we are writing inodes from it.
488 * Returns 0 if the super was successfully pinned (or pinning wasn't needed),
491 static int pin_sb_for_writeback(struct writeback_control *wbc,
494 struct super_block *sb = inode->i_sb;
497 * Caller must already hold the ref for this
499 if (wbc->sync_mode == WB_SYNC_ALL) {
500 WARN_ON(!rwsem_is_locked(&sb->s_umount));
506 if (down_read_trylock(&sb->s_umount)) {
508 spin_unlock(&sb_lock);
512 * umounted, drop rwsem again and fall through to failure
514 up_read(&sb->s_umount);
518 spin_unlock(&sb_lock);
522 static void unpin_sb_for_writeback(struct writeback_control *wbc,
525 struct super_block *sb = inode->i_sb;
527 if (wbc->sync_mode == WB_SYNC_ALL)
530 up_read(&sb->s_umount);
534 static void writeback_inodes_wb(struct bdi_writeback *wb,
535 struct writeback_control *wbc)
537 struct super_block *sb = wbc->sb;
538 const int is_blkdev_sb = sb_is_blkdev_sb(sb);
539 const unsigned long start = jiffies; /* livelock avoidance */
541 spin_lock(&inode_lock);
543 if (!wbc->for_kupdate || list_empty(&wb->b_io))
544 queue_io(wb, wbc->older_than_this);
546 while (!list_empty(&wb->b_io)) {
547 struct inode *inode = list_entry(wb->b_io.prev,
548 struct inode, i_list);
552 * super block given and doesn't match, skip this inode
554 if (sb && sb != inode->i_sb) {
559 if (!bdi_cap_writeback_dirty(wb->bdi)) {
563 * Dirty memory-backed blockdev: the ramdisk
564 * driver does this. Skip just this inode
569 * Dirty memory-backed inode against a filesystem other
570 * than the kernel-internal bdev filesystem. Skip the
576 if (inode->i_state & (I_NEW | I_WILL_FREE)) {
581 if (wbc->nonblocking && bdi_write_congested(wb->bdi)) {
582 wbc->encountered_congestion = 1;
584 break; /* Skip a congested fs */
586 continue; /* Skip a congested blockdev */
590 * Was this inode dirtied after sync_sb_inodes was called?
591 * This keeps sync from extra jobs and livelock.
593 if (inode_dirtied_after(inode, start))
596 if (pin_sb_for_writeback(wbc, inode)) {
601 BUG_ON(inode->i_state & (I_FREEING | I_CLEAR));
603 pages_skipped = wbc->pages_skipped;
604 writeback_single_inode(inode, wbc);
605 unpin_sb_for_writeback(wbc, inode);
606 if (wbc->pages_skipped != pages_skipped) {
608 * writeback is not making progress due to locked
609 * buffers. Skip this inode for now.
613 spin_unlock(&inode_lock);
616 spin_lock(&inode_lock);
617 if (wbc->nr_to_write <= 0) {
621 if (!list_empty(&wb->b_more_io))
625 spin_unlock(&inode_lock);
626 /* Leave any unwritten inodes on b_io */
629 void writeback_inodes_wbc(struct writeback_control *wbc)
631 struct backing_dev_info *bdi = wbc->bdi;
633 writeback_inodes_wb(&bdi->wb, wbc);
637 * The maximum number of pages to writeout in a single bdi flush/kupdate
638 * operation. We do this so we don't hold I_SYNC against an inode for
639 * enormous amounts of time, which would block a userspace task which has
640 * been forced to throttle against that inode. Also, the code reevaluates
641 * the dirty each time it has written this many pages.
643 #define MAX_WRITEBACK_PAGES 1024
645 static inline bool over_bground_thresh(void)
647 unsigned long background_thresh, dirty_thresh;
649 get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL);
651 return (global_page_state(NR_FILE_DIRTY) +
652 global_page_state(NR_UNSTABLE_NFS) >= background_thresh);
656 * Explicit flushing or periodic writeback of "old" data.
658 * Define "old": the first time one of an inode's pages is dirtied, we mark the
659 * dirtying-time in the inode's address_space. So this periodic writeback code
660 * just walks the superblock inode list, writing back any inodes which are
661 * older than a specific point in time.
663 * Try to run once per dirty_writeback_interval. But if a writeback event
664 * takes longer than a dirty_writeback_interval interval, then leave a
667 * older_than_this takes precedence over nr_to_write. So we'll only write back
668 * all dirty pages if they are all attached to "old" mappings.
670 static long wb_writeback(struct bdi_writeback *wb,
671 struct wb_writeback_args *args)
673 struct writeback_control wbc = {
676 .sync_mode = args->sync_mode,
677 .older_than_this = NULL,
678 .for_kupdate = args->for_kupdate,
679 .range_cyclic = args->range_cyclic,
681 unsigned long oldest_jif;
684 if (wbc.for_kupdate) {
685 wbc.older_than_this = &oldest_jif;
686 oldest_jif = jiffies -
687 msecs_to_jiffies(dirty_expire_interval * 10);
689 if (!wbc.range_cyclic) {
691 wbc.range_end = LLONG_MAX;
696 * Don't flush anything for non-integrity writeback where
697 * no nr_pages was given
699 if (!args->for_kupdate && args->nr_pages <= 0 &&
700 args->sync_mode == WB_SYNC_NONE)
704 * If no specific pages were given and this is just a
705 * periodic background writeout and we are below the
706 * background dirty threshold, don't do anything
708 if (args->for_kupdate && args->nr_pages <= 0 &&
709 !over_bground_thresh())
713 wbc.encountered_congestion = 0;
714 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
715 wbc.pages_skipped = 0;
716 writeback_inodes_wb(wb, &wbc);
717 args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
718 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
721 * If we ran out of stuff to write, bail unless more_io got set
723 if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) {
724 if (wbc.more_io && !wbc.for_kupdate)
734 * Return the next bdi_work struct that hasn't been processed by this
737 static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi,
738 struct bdi_writeback *wb)
740 struct bdi_work *work, *ret = NULL;
744 list_for_each_entry_rcu(work, &bdi->work_list, list) {
745 if (!test_and_clear_bit(wb->nr, &work->seen))
756 static long wb_check_old_data_flush(struct bdi_writeback *wb)
758 unsigned long expired;
761 expired = wb->last_old_flush +
762 msecs_to_jiffies(dirty_writeback_interval * 10);
763 if (time_before(jiffies, expired))
766 wb->last_old_flush = jiffies;
767 nr_pages = global_page_state(NR_FILE_DIRTY) +
768 global_page_state(NR_UNSTABLE_NFS) +
769 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
772 struct wb_writeback_args args = {
773 .nr_pages = nr_pages,
774 .sync_mode = WB_SYNC_NONE,
779 return wb_writeback(wb, &args);
786 * Retrieve work items and do the writeback they describe
788 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
790 struct backing_dev_info *bdi = wb->bdi;
791 struct bdi_work *work;
794 while ((work = get_next_work_item(bdi, wb)) != NULL) {
795 struct wb_writeback_args args = work->args;
798 * Override sync mode, in case we must wait for completion
801 work->args.sync_mode = args.sync_mode = WB_SYNC_ALL;
804 * If this isn't a data integrity operation, just notify
805 * that we have seen this work and we are now starting it.
807 if (args.sync_mode == WB_SYNC_NONE)
808 wb_clear_pending(wb, work);
810 wrote += wb_writeback(wb, &args);
813 * This is a data integrity writeback, so only do the
814 * notification when we have completed the work.
816 if (args.sync_mode == WB_SYNC_ALL)
817 wb_clear_pending(wb, work);
821 * Check for periodic writeback, kupdated() style
823 wrote += wb_check_old_data_flush(wb);
829 * Handle writeback of dirty data for the device backed by this bdi. Also
830 * wakes up periodically and does kupdated style flushing.
832 int bdi_writeback_task(struct bdi_writeback *wb)
834 unsigned long last_active = jiffies;
835 unsigned long wait_jiffies = -1UL;
838 while (!kthread_should_stop()) {
839 pages_written = wb_do_writeback(wb, 0);
842 last_active = jiffies;
843 else if (wait_jiffies != -1UL) {
844 unsigned long max_idle;
847 * Longest period of inactivity that we tolerate. If we
848 * see dirty data again later, the task will get
849 * recreated automatically.
851 max_idle = max(5UL * 60 * HZ, wait_jiffies);
852 if (time_after(jiffies, max_idle + last_active))
856 wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10);
857 set_current_state(TASK_INTERRUPTIBLE);
858 schedule_timeout(wait_jiffies);
866 * Schedule writeback for all backing devices. Can only be used for
867 * WB_SYNC_NONE writeback, WB_SYNC_ALL should use bdi_start_writeback()
868 * and pass in the superblock.
870 static void bdi_writeback_all(struct writeback_control *wbc)
872 struct backing_dev_info *bdi;
874 WARN_ON(wbc->sync_mode == WB_SYNC_ALL);
878 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
879 if (!bdi_has_dirty_io(bdi))
882 bdi_alloc_queue_work(bdi, wbc);
889 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
892 void wakeup_flusher_threads(long nr_pages)
894 struct writeback_control wbc = {
895 .sync_mode = WB_SYNC_NONE,
896 .older_than_this = NULL,
901 nr_pages = global_page_state(NR_FILE_DIRTY) +
902 global_page_state(NR_UNSTABLE_NFS);
903 wbc.nr_to_write = nr_pages;
904 bdi_writeback_all(&wbc);
907 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
909 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
910 struct dentry *dentry;
911 const char *name = "?";
913 dentry = d_find_alias(inode);
915 spin_lock(&dentry->d_lock);
916 name = (const char *) dentry->d_name.name;
919 "%s(%d): dirtied inode %lu (%s) on %s\n",
920 current->comm, task_pid_nr(current), inode->i_ino,
921 name, inode->i_sb->s_id);
923 spin_unlock(&dentry->d_lock);
930 * __mark_inode_dirty - internal function
931 * @inode: inode to mark
932 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
933 * Mark an inode as dirty. Callers should use mark_inode_dirty or
934 * mark_inode_dirty_sync.
936 * Put the inode on the super block's dirty list.
938 * CAREFUL! We mark it dirty unconditionally, but move it onto the
939 * dirty list only if it is hashed or if it refers to a blockdev.
940 * If it was not hashed, it will never be added to the dirty list
941 * even if it is later hashed, as it will have been marked dirty already.
943 * In short, make sure you hash any inodes _before_ you start marking
946 * This function *must* be atomic for the I_DIRTY_PAGES case -
947 * set_page_dirty() is called under spinlock in several places.
949 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
950 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
951 * the kernel-internal blockdev inode represents the dirtying time of the
952 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
953 * page->mapping->host, so the page-dirtying time is recorded in the internal
956 void __mark_inode_dirty(struct inode *inode, int flags)
958 struct super_block *sb = inode->i_sb;
961 * Don't do this for I_DIRTY_PAGES - that doesn't actually
962 * dirty the inode itself
964 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
965 if (sb->s_op->dirty_inode)
966 sb->s_op->dirty_inode(inode);
970 * make sure that changes are seen by all cpus before we test i_state
975 /* avoid the locking if we can */
976 if ((inode->i_state & flags) == flags)
979 if (unlikely(block_dump))
980 block_dump___mark_inode_dirty(inode);
982 spin_lock(&inode_lock);
983 if ((inode->i_state & flags) != flags) {
984 const int was_dirty = inode->i_state & I_DIRTY;
986 inode->i_state |= flags;
989 * If the inode is being synced, just update its dirty state.
990 * The unlocker will place the inode on the appropriate
991 * superblock list, based upon its state.
993 if (inode->i_state & I_SYNC)
997 * Only add valid (hashed) inodes to the superblock's
998 * dirty list. Add blockdev inodes as well.
1000 if (!S_ISBLK(inode->i_mode)) {
1001 if (hlist_unhashed(&inode->i_hash))
1004 if (inode->i_state & (I_FREEING|I_CLEAR))
1008 * If the inode was already on b_dirty/b_io/b_more_io, don't
1009 * reposition it (that would break b_dirty time-ordering).
1012 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1013 struct backing_dev_info *bdi = wb->bdi;
1015 if (bdi_cap_writeback_dirty(bdi) &&
1016 !test_bit(BDI_registered, &bdi->state)) {
1018 printk(KERN_ERR "bdi-%s not registered\n",
1022 inode->dirtied_when = jiffies;
1023 list_move(&inode->i_list, &wb->b_dirty);
1027 spin_unlock(&inode_lock);
1029 EXPORT_SYMBOL(__mark_inode_dirty);
1032 * Write out a superblock's list of dirty inodes. A wait will be performed
1033 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1035 * If older_than_this is non-NULL, then only write out inodes which
1036 * had their first dirtying at a time earlier than *older_than_this.
1038 * If we're a pdlfush thread, then implement pdflush collision avoidance
1039 * against the entire list.
1041 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1042 * This function assumes that the blockdev superblock's inodes are backed by
1043 * a variety of queues, so all inodes are searched. For other superblocks,
1044 * assume that all inodes are backed by the same queue.
1046 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1047 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1048 * on the writer throttling path, and we get decent balancing between many
1049 * throttled threads: we don't want them all piling up on inode_sync_wait.
1051 static void wait_sb_inodes(struct writeback_control *wbc)
1053 struct inode *inode, *old_inode = NULL;
1056 * We need to be protected against the filesystem going from
1057 * r/o to r/w or vice versa.
1059 WARN_ON(!rwsem_is_locked(&wbc->sb->s_umount));
1061 spin_lock(&inode_lock);
1064 * Data integrity sync. Must wait for all pages under writeback,
1065 * because there may have been pages dirtied before our sync
1066 * call, but which had writeout started before we write it out.
1067 * In which case, the inode may not be on the dirty list, but
1068 * we still have to wait for that writeout.
1070 list_for_each_entry(inode, &wbc->sb->s_inodes, i_sb_list) {
1071 struct address_space *mapping;
1073 if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW))
1075 mapping = inode->i_mapping;
1076 if (mapping->nrpages == 0)
1079 spin_unlock(&inode_lock);
1081 * We hold a reference to 'inode' so it couldn't have
1082 * been removed from s_inodes list while we dropped the
1083 * inode_lock. We cannot iput the inode now as we can
1084 * be holding the last reference and we cannot iput it
1085 * under inode_lock. So we keep the reference and iput
1091 filemap_fdatawait(mapping);
1095 spin_lock(&inode_lock);
1097 spin_unlock(&inode_lock);
1102 * writeback_inodes_sb - writeback dirty inodes from given super_block
1103 * @sb: the superblock
1105 * Start writeback on some inodes on this super_block. No guarantees are made
1106 * on how many (if any) will be written, and this function does not wait
1107 * for IO completion of submitted IO. The number of pages submitted is
1110 long writeback_inodes_sb(struct super_block *sb)
1112 struct writeback_control wbc = {
1114 .sync_mode = WB_SYNC_NONE,
1116 .range_end = LLONG_MAX,
1118 unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY);
1119 unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS);
1122 nr_to_write = nr_dirty + nr_unstable +
1123 (inodes_stat.nr_inodes - inodes_stat.nr_unused);
1125 wbc.nr_to_write = nr_to_write;
1126 bdi_writeback_all(&wbc);
1127 return nr_to_write - wbc.nr_to_write;
1129 EXPORT_SYMBOL(writeback_inodes_sb);
1132 * sync_inodes_sb - sync sb inode pages
1133 * @sb: the superblock
1135 * This function writes and waits on any dirty inode belonging to this
1136 * super_block. The number of pages synced is returned.
1138 long sync_inodes_sb(struct super_block *sb)
1140 struct writeback_control wbc = {
1143 .sync_mode = WB_SYNC_ALL,
1145 .range_end = LLONG_MAX,
1147 long nr_to_write = LONG_MAX; /* doesn't actually matter */
1149 wbc.nr_to_write = nr_to_write;
1150 bdi_start_writeback(&wbc);
1151 wait_sb_inodes(&wbc);
1152 return nr_to_write - wbc.nr_to_write;
1154 EXPORT_SYMBOL(sync_inodes_sb);
1157 * write_inode_now - write an inode to disk
1158 * @inode: inode to write to disk
1159 * @sync: whether the write should be synchronous or not
1161 * This function commits an inode to disk immediately if it is dirty. This is
1162 * primarily needed by knfsd.
1164 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1166 int write_inode_now(struct inode *inode, int sync)
1169 struct writeback_control wbc = {
1170 .nr_to_write = LONG_MAX,
1171 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1173 .range_end = LLONG_MAX,
1176 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1177 wbc.nr_to_write = 0;
1180 spin_lock(&inode_lock);
1181 ret = writeback_single_inode(inode, &wbc);
1182 spin_unlock(&inode_lock);
1184 inode_sync_wait(inode);
1187 EXPORT_SYMBOL(write_inode_now);
1190 * sync_inode - write an inode and its pages to disk.
1191 * @inode: the inode to sync
1192 * @wbc: controls the writeback mode
1194 * sync_inode() will write an inode and its pages to disk. It will also
1195 * correctly update the inode on its superblock's dirty inode lists and will
1196 * update inode->i_state.
1198 * The caller must have a ref on the inode.
1200 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1204 spin_lock(&inode_lock);
1205 ret = writeback_single_inode(inode, wbc);
1206 spin_unlock(&inode_lock);
1209 EXPORT_SYMBOL(sync_inode);