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>
30 #include <trace/events/vfs.h>
34 * Passed into wb_writeback(), essentially a subset of writeback_control
36 struct wb_writeback_work {
38 struct super_block *sb;
39 enum writeback_sync_modes sync_mode;
40 unsigned int for_kupdate:1;
41 unsigned int range_cyclic:1;
42 unsigned int for_background:1;
44 struct list_head list; /* pending work list */
45 struct completion *done; /* set if the caller waits */
49 * Include the creation of the trace points after defining the
50 * wb_writeback_work structure so that the definition remains local to this
53 #define CREATE_TRACE_POINTS
54 #include <trace/events/writeback.h>
57 * We don't actually have pdflush, but this one is exported though /proc...
59 int nr_pdflush_threads;
62 * writeback_in_progress - determine whether there is writeback in progress
63 * @bdi: the device's backing_dev_info structure.
65 * Determine whether there is writeback waiting to be handled against a
68 int writeback_in_progress(struct backing_dev_info *bdi)
70 return test_bit(BDI_writeback_running, &bdi->state);
73 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
75 struct super_block *sb = inode->i_sb;
77 if (strcmp(sb->s_type->name, "bdev") == 0)
78 return inode->i_mapping->backing_dev_info;
83 static inline struct inode *wb_inode(struct list_head *head)
85 return list_entry(head, struct inode, i_wb_list);
88 /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
89 static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
92 wake_up_process(bdi->wb.task);
95 * The bdi thread isn't there, wake up the forker thread which
96 * will create and run it.
98 wake_up_process(default_backing_dev_info.wb.task);
102 static void bdi_queue_work(struct backing_dev_info *bdi,
103 struct wb_writeback_work *work)
105 trace_writeback_queue(bdi, work);
107 spin_lock_bh(&bdi->wb_lock);
108 list_add_tail(&work->list, &bdi->work_list);
110 trace_writeback_nothread(bdi, work);
111 bdi_wakeup_flusher(bdi);
112 spin_unlock_bh(&bdi->wb_lock);
116 __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
119 struct wb_writeback_work *work;
122 * This is WB_SYNC_NONE writeback, so if allocation fails just
123 * wakeup the thread for old dirty data writeback
125 work = kzalloc(sizeof(*work), GFP_ATOMIC);
128 trace_writeback_nowork(bdi);
129 wake_up_process(bdi->wb.task);
134 work->sync_mode = WB_SYNC_NONE;
135 work->nr_pages = nr_pages;
136 work->range_cyclic = range_cyclic;
138 bdi_queue_work(bdi, work);
142 * bdi_start_writeback - start writeback
143 * @bdi: the backing device to write from
144 * @nr_pages: the number of pages to write
147 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
148 * started when this function returns, we make no guarentees on
149 * completion. Caller need not hold sb s_umount semaphore.
152 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
154 __bdi_start_writeback(bdi, nr_pages, true);
158 * bdi_start_background_writeback - start background writeback
159 * @bdi: the backing device to write from
162 * This makes sure WB_SYNC_NONE background writeback happens. When
163 * this function returns, it is only guaranteed that for given BDI
164 * some IO is happening if we are over background dirty threshold.
165 * Caller need not hold sb s_umount semaphore.
167 void bdi_start_background_writeback(struct backing_dev_info *bdi)
170 * We just wake up the flusher thread. It will perform background
171 * writeback as soon as there is no other work to do.
173 trace_writeback_wake_background(bdi);
174 spin_lock_bh(&bdi->wb_lock);
175 bdi_wakeup_flusher(bdi);
176 spin_unlock_bh(&bdi->wb_lock);
180 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
181 * furthest end of its superblock's dirty-inode list.
183 * Before stamping the inode's ->dirtied_when, we check to see whether it is
184 * already the most-recently-dirtied inode on the b_dirty list. If that is
185 * the case then the inode must have been redirtied while it was being written
186 * out and we don't reset its dirtied_when.
188 static void redirty_tail(struct inode *inode)
190 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
192 if (!list_empty(&wb->b_dirty)) {
195 tail = wb_inode(wb->b_dirty.next);
196 if (time_before(inode->dirtied_when, tail->dirtied_when))
197 inode->dirtied_when = jiffies;
199 list_move(&inode->i_wb_list, &wb->b_dirty);
203 * requeue inode for re-scanning after bdi->b_io list is exhausted.
205 static void requeue_io(struct inode *inode)
207 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
209 list_move(&inode->i_wb_list, &wb->b_more_io);
212 static void inode_sync_complete(struct inode *inode)
215 * Prevent speculative execution through spin_unlock(&inode_lock);
218 wake_up_bit(&inode->i_state, __I_SYNC);
221 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
223 bool ret = time_after(inode->dirtied_when, t);
226 * For inodes being constantly redirtied, dirtied_when can get stuck.
227 * It _appears_ to be in the future, but is actually in distant past.
228 * This test is necessary to prevent such wrapped-around relative times
229 * from permanently stopping the whole bdi writeback.
231 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
237 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
239 static void move_expired_inodes(struct list_head *delaying_queue,
240 struct list_head *dispatch_queue,
241 unsigned long *older_than_this)
244 struct list_head *pos, *node;
245 struct super_block *sb = NULL;
249 while (!list_empty(delaying_queue)) {
250 inode = wb_inode(delaying_queue->prev);
251 if (older_than_this &&
252 inode_dirtied_after(inode, *older_than_this))
254 if (sb && sb != inode->i_sb)
257 list_move(&inode->i_wb_list, &tmp);
260 /* just one sb in list, splice to dispatch_queue and we're done */
262 list_splice(&tmp, dispatch_queue);
266 /* Move inodes from one superblock together */
267 while (!list_empty(&tmp)) {
268 sb = wb_inode(tmp.prev)->i_sb;
269 list_for_each_prev_safe(pos, node, &tmp) {
270 inode = wb_inode(pos);
271 if (inode->i_sb == sb)
272 list_move(&inode->i_wb_list, dispatch_queue);
278 * Queue all expired dirty inodes for io, eldest first.
280 * newly dirtied b_dirty b_io b_more_io
281 * =============> gf edc BA
283 * newly dirtied b_dirty b_io b_more_io
284 * =============> g fBAedc
286 * +--> dequeue for IO
288 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
290 list_splice_init(&wb->b_more_io, &wb->b_io);
291 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
294 static int write_inode(struct inode *inode, struct writeback_control *wbc)
296 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
297 return inode->i_sb->s_op->write_inode(inode, wbc);
302 * Wait for writeback on an inode to complete.
304 static void inode_wait_for_writeback(struct inode *inode)
306 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
307 wait_queue_head_t *wqh;
309 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
310 while (inode->i_state & I_SYNC) {
311 spin_unlock(&inode_lock);
312 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
313 spin_lock(&inode_lock);
318 * Write out an inode's dirty pages. Called under inode_lock. Either the
319 * caller has ref on the inode (either via __iget or via syscall against an fd)
320 * or the inode has I_WILL_FREE set (via generic_forget_inode)
322 * If `wait' is set, wait on the writeout.
324 * The whole writeout design is quite complex and fragile. We want to avoid
325 * starvation of particular inodes when others are being redirtied, prevent
328 * Called under inode_lock.
331 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
333 struct address_space *mapping = inode->i_mapping;
337 if (!atomic_read(&inode->i_count))
338 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
340 WARN_ON(inode->i_state & I_WILL_FREE);
342 if (inode->i_state & I_SYNC) {
344 * If this inode is locked for writeback and we are not doing
345 * writeback-for-data-integrity, move it to b_more_io so that
346 * writeback can proceed with the other inodes on s_io.
348 * We'll have another go at writing back this inode when we
349 * completed a full scan of b_io.
351 if (wbc->sync_mode != WB_SYNC_ALL) {
357 * It's a data-integrity sync. We must wait.
359 inode_wait_for_writeback(inode);
362 BUG_ON(inode->i_state & I_SYNC);
364 /* Set I_SYNC, reset I_DIRTY_PAGES */
365 inode->i_state |= I_SYNC;
366 inode->i_state &= ~I_DIRTY_PAGES;
367 spin_unlock(&inode_lock);
369 ret = do_writepages(mapping, wbc);
372 * Make sure to wait on the data before writing out the metadata.
373 * This is important for filesystems that modify metadata on data
376 if (wbc->sync_mode == WB_SYNC_ALL) {
377 int err = filemap_fdatawait(mapping);
383 * Some filesystems may redirty the inode during the writeback
384 * due to delalloc, clear dirty metadata flags right before
387 spin_lock(&inode_lock);
388 dirty = inode->i_state & I_DIRTY;
389 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
390 spin_unlock(&inode_lock);
391 /* Don't write the inode if only I_DIRTY_PAGES was set */
392 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
393 int err = write_inode(inode, wbc);
398 spin_lock(&inode_lock);
399 inode->i_state &= ~I_SYNC;
400 if (!(inode->i_state & I_FREEING)) {
401 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
403 * We didn't write back all the pages. nfs_writepages()
404 * sometimes bales out without doing anything.
406 inode->i_state |= I_DIRTY_PAGES;
407 if (wbc->nr_to_write <= 0) {
409 * slice used up: queue for next turn
414 * Writeback blocked by something other than
415 * congestion. Delay the inode for some time to
416 * avoid spinning on the CPU (100% iowait)
417 * retrying writeback of the dirty page/inode
418 * that cannot be performed immediately.
422 } else if (inode->i_state & I_DIRTY) {
424 * Filesystems can dirty the inode during writeback
425 * operations, such as delayed allocation during
426 * submission or metadata updates after data IO
432 * The inode is clean. At this point we either have
433 * a reference to the inode or it's on it's way out.
434 * No need to add it back to the LRU.
436 list_del_init(&inode->i_wb_list);
439 inode_sync_complete(inode);
444 * For background writeback the caller does not have the sb pinned
445 * before calling writeback. So make sure that we do pin it, so it doesn't
446 * go away while we are writing inodes from it.
448 static bool pin_sb_for_writeback(struct super_block *sb)
451 if (list_empty(&sb->s_instances)) {
452 spin_unlock(&sb_lock);
457 spin_unlock(&sb_lock);
459 if (down_read_trylock(&sb->s_umount)) {
462 up_read(&sb->s_umount);
470 * Write a portion of b_io inodes which belong to @sb.
472 * If @only_this_sb is true, then find and write all such
473 * inodes. Otherwise write only ones which go sequentially
476 * Return 1, if the caller writeback routine should be
477 * interrupted. Otherwise return 0.
479 static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
480 struct writeback_control *wbc, bool only_this_sb)
482 while (!list_empty(&wb->b_io)) {
484 struct inode *inode = wb_inode(wb->b_io.prev);
486 if (inode->i_sb != sb) {
489 * We only want to write back data for this
490 * superblock, move all inodes not belonging
491 * to it back onto the dirty list.
498 * The inode belongs to a different superblock.
499 * Bounce back to the caller to unpin this and
500 * pin the next superblock.
506 * Don't bother with new inodes or inodes beeing freed, first
507 * kind does not need peridic writeout yet, and for the latter
508 * kind writeout is handled by the freer.
510 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
516 * Was this inode dirtied after sync_sb_inodes was called?
517 * This keeps sync from extra jobs and livelock.
519 if (inode_dirtied_after(inode, wbc->wb_start))
523 pages_skipped = wbc->pages_skipped;
524 writeback_single_inode(inode, wbc);
525 if (wbc->pages_skipped != pages_skipped) {
527 * writeback is not making progress due to locked
528 * buffers. Skip this inode for now.
532 spin_unlock(&inode_lock);
535 spin_lock(&inode_lock);
536 if (wbc->nr_to_write <= 0) {
540 if (!list_empty(&wb->b_more_io))
547 void writeback_inodes_wb(struct bdi_writeback *wb,
548 struct writeback_control *wbc)
553 wbc->wb_start = jiffies; /* livelock avoidance */
554 spin_lock(&inode_lock);
555 if (!wbc->for_kupdate || list_empty(&wb->b_io))
556 queue_io(wb, wbc->older_than_this);
558 while (!list_empty(&wb->b_io)) {
559 struct inode *inode = wb_inode(wb->b_io.prev);
560 struct super_block *sb = inode->i_sb;
562 if (!pin_sb_for_writeback(sb)) {
566 ret = writeback_sb_inodes(sb, wb, wbc, false);
572 spin_unlock(&inode_lock);
573 /* Leave any unwritten inodes on b_io */
576 static void __writeback_inodes_sb(struct super_block *sb,
577 struct bdi_writeback *wb, struct writeback_control *wbc)
579 WARN_ON(!rwsem_is_locked(&sb->s_umount));
581 spin_lock(&inode_lock);
582 if (!wbc->for_kupdate || list_empty(&wb->b_io))
583 queue_io(wb, wbc->older_than_this);
584 writeback_sb_inodes(sb, wb, wbc, true);
585 spin_unlock(&inode_lock);
589 * The maximum number of pages to writeout in a single bdi flush/kupdate
590 * operation. We do this so we don't hold I_SYNC against an inode for
591 * enormous amounts of time, which would block a userspace task which has
592 * been forced to throttle against that inode. Also, the code reevaluates
593 * the dirty each time it has written this many pages.
595 #define MAX_WRITEBACK_PAGES 1024
597 static inline bool over_bground_thresh(void)
599 unsigned long background_thresh, dirty_thresh;
601 global_dirty_limits(&background_thresh, &dirty_thresh);
603 return (global_page_state(NR_FILE_DIRTY) +
604 global_page_state(NR_UNSTABLE_NFS) > background_thresh);
608 * Explicit flushing or periodic writeback of "old" data.
610 * Define "old": the first time one of an inode's pages is dirtied, we mark the
611 * dirtying-time in the inode's address_space. So this periodic writeback code
612 * just walks the superblock inode list, writing back any inodes which are
613 * older than a specific point in time.
615 * Try to run once per dirty_writeback_interval. But if a writeback event
616 * takes longer than a dirty_writeback_interval interval, then leave a
619 * older_than_this takes precedence over nr_to_write. So we'll only write back
620 * all dirty pages if they are all attached to "old" mappings.
622 static long wb_writeback(struct bdi_writeback *wb,
623 struct wb_writeback_work *work)
625 struct writeback_control wbc = {
626 .sync_mode = work->sync_mode,
627 .older_than_this = NULL,
628 .for_kupdate = work->for_kupdate,
629 .for_background = work->for_background,
630 .range_cyclic = work->range_cyclic,
632 unsigned long oldest_jif;
637 if (wbc.for_kupdate) {
638 wbc.older_than_this = &oldest_jif;
639 oldest_jif = jiffies -
640 msecs_to_jiffies(dirty_expire_interval * 10);
642 if (!wbc.range_cyclic) {
644 wbc.range_end = LLONG_MAX;
648 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
649 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
650 * here avoids calling into writeback_inodes_wb() more than once.
652 * The intended call sequence for WB_SYNC_ALL writeback is:
655 * __writeback_inodes_sb() <== called only once
656 * write_cache_pages() <== called once for each inode
657 * (quickly) tag currently dirty pages
658 * (maybe slowly) sync all tagged pages
660 if (wbc.sync_mode == WB_SYNC_NONE)
661 write_chunk = MAX_WRITEBACK_PAGES;
663 write_chunk = LONG_MAX;
665 wbc.wb_start = jiffies; /* livelock avoidance */
668 * Stop writeback when nr_pages has been consumed
670 if (work->nr_pages <= 0)
674 * Background writeout and kupdate-style writeback may
675 * run forever. Stop them if there is other work to do
676 * so that e.g. sync can proceed. They'll be restarted
677 * after the other works are all done.
679 if ((work->for_background || work->for_kupdate) &&
680 !list_empty(&wb->bdi->work_list))
684 * For background writeout, stop when we are below the
685 * background dirty threshold
687 if (work->for_background && !over_bground_thresh())
691 wbc.nr_to_write = write_chunk;
692 wbc.pages_skipped = 0;
694 trace_wbc_writeback_start(&wbc, wb->bdi);
696 __writeback_inodes_sb(work->sb, wb, &wbc);
698 writeback_inodes_wb(wb, &wbc);
699 trace_wbc_writeback_written(&wbc, wb->bdi);
701 work->nr_pages -= write_chunk - wbc.nr_to_write;
702 wrote += write_chunk - wbc.nr_to_write;
705 * If we consumed everything, see if we have more
707 if (wbc.nr_to_write <= 0)
710 * Didn't write everything and we don't have more IO, bail
715 * Did we write something? Try for more
717 if (wbc.nr_to_write < write_chunk)
720 * Nothing written. Wait for some inode to
721 * become available for writeback. Otherwise
722 * we'll just busyloop.
724 spin_lock(&inode_lock);
725 if (!list_empty(&wb->b_more_io)) {
726 inode = wb_inode(wb->b_more_io.prev);
727 trace_wbc_writeback_wait(&wbc, wb->bdi);
728 inode_wait_for_writeback(inode);
730 spin_unlock(&inode_lock);
737 * Return the next wb_writeback_work struct that hasn't been processed yet.
739 static struct wb_writeback_work *
740 get_next_work_item(struct backing_dev_info *bdi)
742 struct wb_writeback_work *work = NULL;
744 spin_lock_bh(&bdi->wb_lock);
745 if (!list_empty(&bdi->work_list)) {
746 work = list_entry(bdi->work_list.next,
747 struct wb_writeback_work, list);
748 list_del_init(&work->list);
750 spin_unlock_bh(&bdi->wb_lock);
755 * Add in the number of potentially dirty inodes, because each inode
756 * write can dirty pagecache in the underlying blockdev.
758 static unsigned long get_nr_dirty_pages(void)
760 return global_page_state(NR_FILE_DIRTY) +
761 global_page_state(NR_UNSTABLE_NFS) +
762 get_nr_dirty_inodes();
765 static long wb_check_background_flush(struct bdi_writeback *wb)
767 if (over_bground_thresh()) {
769 struct wb_writeback_work work = {
770 .nr_pages = LONG_MAX,
771 .sync_mode = WB_SYNC_NONE,
776 return wb_writeback(wb, &work);
782 static long wb_check_old_data_flush(struct bdi_writeback *wb)
784 unsigned long expired;
788 * When set to zero, disable periodic writeback
790 if (!dirty_writeback_interval)
793 expired = wb->last_old_flush +
794 msecs_to_jiffies(dirty_writeback_interval * 10);
795 if (time_before(jiffies, expired))
798 wb->last_old_flush = jiffies;
799 nr_pages = get_nr_dirty_pages();
802 struct wb_writeback_work work = {
803 .nr_pages = nr_pages,
804 .sync_mode = WB_SYNC_NONE,
809 return wb_writeback(wb, &work);
816 * Retrieve work items and do the writeback they describe
818 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
820 struct backing_dev_info *bdi = wb->bdi;
821 struct wb_writeback_work *work;
824 set_bit(BDI_writeback_running, &wb->bdi->state);
825 while ((work = get_next_work_item(bdi)) != NULL) {
827 * Override sync mode, in case we must wait for completion
828 * because this thread is exiting now.
831 work->sync_mode = WB_SYNC_ALL;
833 trace_writeback_exec(bdi, work);
835 wrote += wb_writeback(wb, work);
838 * Notify the caller of completion if this is a synchronous
839 * work item, otherwise just free it.
842 complete(work->done);
848 * Check for periodic writeback, kupdated() style
850 wrote += wb_check_old_data_flush(wb);
851 wrote += wb_check_background_flush(wb);
852 clear_bit(BDI_writeback_running, &wb->bdi->state);
858 * Handle writeback of dirty data for the device backed by this bdi. Also
859 * wakes up periodically and does kupdated style flushing.
861 int bdi_writeback_thread(void *data)
863 struct bdi_writeback *wb = data;
864 struct backing_dev_info *bdi = wb->bdi;
867 current->flags |= PF_SWAPWRITE;
869 wb->last_active = jiffies;
872 * Our parent may run at a different priority, just set us to normal
874 set_user_nice(current, 0);
876 trace_writeback_thread_start(bdi);
878 while (!kthread_should_stop()) {
880 * Remove own delayed wake-up timer, since we are already awake
881 * and we'll take care of the preriodic write-back.
883 del_timer(&wb->wakeup_timer);
885 pages_written = wb_do_writeback(wb, 0);
887 trace_writeback_pages_written(pages_written);
890 wb->last_active = jiffies;
892 set_current_state(TASK_INTERRUPTIBLE);
893 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
894 __set_current_state(TASK_RUNNING);
898 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
899 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
902 * We have nothing to do, so can go sleep without any
903 * timeout and save power. When a work is queued or
904 * something is made dirty - we will be woken up.
912 /* Flush any work that raced with us exiting */
913 if (!list_empty(&bdi->work_list))
914 wb_do_writeback(wb, 1);
916 trace_writeback_thread_stop(bdi);
922 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
925 void wakeup_flusher_threads(long nr_pages)
927 struct backing_dev_info *bdi;
930 nr_pages = global_page_state(NR_FILE_DIRTY) +
931 global_page_state(NR_UNSTABLE_NFS);
935 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
936 if (!bdi_has_dirty_io(bdi))
938 __bdi_start_writeback(bdi, nr_pages, false);
943 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
945 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
946 struct dentry *dentry;
947 const char *name = "?";
949 dentry = d_find_alias(inode);
951 spin_lock(&dentry->d_lock);
952 name = (const char *) dentry->d_name.name;
955 "%s(%d): dirtied inode %lu (%s) on %s\n",
956 current->comm, task_pid_nr(current), inode->i_ino,
957 name, inode->i_sb->s_id);
959 spin_unlock(&dentry->d_lock);
966 * __mark_inode_dirty - internal function
967 * @inode: inode to mark
968 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
969 * Mark an inode as dirty. Callers should use mark_inode_dirty or
970 * mark_inode_dirty_sync.
972 * Put the inode on the super block's dirty list.
974 * CAREFUL! We mark it dirty unconditionally, but move it onto the
975 * dirty list only if it is hashed or if it refers to a blockdev.
976 * If it was not hashed, it will never be added to the dirty list
977 * even if it is later hashed, as it will have been marked dirty already.
979 * In short, make sure you hash any inodes _before_ you start marking
982 * This function *must* be atomic for the I_DIRTY_PAGES case -
983 * set_page_dirty() is called under spinlock in several places.
985 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
986 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
987 * the kernel-internal blockdev inode represents the dirtying time of the
988 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
989 * page->mapping->host, so the page-dirtying time is recorded in the internal
992 void __mark_inode_dirty(struct inode *inode, int flags)
994 struct super_block *sb = inode->i_sb;
995 struct backing_dev_info *bdi = NULL;
996 bool wakeup_bdi = false;
999 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1000 * dirty the inode itself
1002 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1003 if (sb->s_op->dirty_inode)
1004 sb->s_op->dirty_inode(inode);
1008 * make sure that changes are seen by all cpus before we test i_state
1013 /* avoid the locking if we can */
1014 if ((inode->i_state & flags) == flags)
1017 trace_dirty_inode(inode, current);
1019 if (unlikely(block_dump))
1020 block_dump___mark_inode_dirty(inode);
1022 spin_lock(&inode_lock);
1023 if ((inode->i_state & flags) != flags) {
1024 const int was_dirty = inode->i_state & I_DIRTY;
1026 inode->i_state |= flags;
1029 * If the inode is being synced, just update its dirty state.
1030 * The unlocker will place the inode on the appropriate
1031 * superblock list, based upon its state.
1033 if (inode->i_state & I_SYNC)
1037 * Only add valid (hashed) inodes to the superblock's
1038 * dirty list. Add blockdev inodes as well.
1040 if (!S_ISBLK(inode->i_mode)) {
1041 if (inode_unhashed(inode))
1044 if (inode->i_state & I_FREEING)
1048 * If the inode was already on b_dirty/b_io/b_more_io, don't
1049 * reposition it (that would break b_dirty time-ordering).
1052 bdi = inode_to_bdi(inode);
1054 if (bdi_cap_writeback_dirty(bdi)) {
1055 WARN(!test_bit(BDI_registered, &bdi->state),
1056 "bdi-%s not registered\n", bdi->name);
1059 * If this is the first dirty inode for this
1060 * bdi, we have to wake-up the corresponding
1061 * bdi thread to make sure background
1062 * write-back happens later.
1064 if (!wb_has_dirty_io(&bdi->wb))
1068 inode->dirtied_when = jiffies;
1069 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1073 spin_unlock(&inode_lock);
1076 bdi_wakeup_thread_delayed(bdi);
1078 EXPORT_SYMBOL(__mark_inode_dirty);
1081 * Write out a superblock's list of dirty inodes. A wait will be performed
1082 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1084 * If older_than_this is non-NULL, then only write out inodes which
1085 * had their first dirtying at a time earlier than *older_than_this.
1087 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1088 * This function assumes that the blockdev superblock's inodes are backed by
1089 * a variety of queues, so all inodes are searched. For other superblocks,
1090 * assume that all inodes are backed by the same queue.
1092 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1093 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1094 * on the writer throttling path, and we get decent balancing between many
1095 * throttled threads: we don't want them all piling up on inode_sync_wait.
1097 static void wait_sb_inodes(struct super_block *sb)
1099 struct inode *inode, *old_inode = NULL;
1102 * We need to be protected against the filesystem going from
1103 * r/o to r/w or vice versa.
1105 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1107 spin_lock(&inode_lock);
1110 * Data integrity sync. Must wait for all pages under writeback,
1111 * because there may have been pages dirtied before our sync
1112 * call, but which had writeout started before we write it out.
1113 * In which case, the inode may not be on the dirty list, but
1114 * we still have to wait for that writeout.
1116 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1117 struct address_space *mapping;
1119 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW))
1121 mapping = inode->i_mapping;
1122 if (mapping->nrpages == 0)
1125 spin_unlock(&inode_lock);
1127 * We hold a reference to 'inode' so it couldn't have
1128 * been removed from s_inodes list while we dropped the
1129 * inode_lock. We cannot iput the inode now as we can
1130 * be holding the last reference and we cannot iput it
1131 * under inode_lock. So we keep the reference and iput
1137 filemap_fdatawait(mapping);
1141 spin_lock(&inode_lock);
1143 spin_unlock(&inode_lock);
1148 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1149 * @sb: the superblock
1150 * @nr: the number of pages to write
1152 * Start writeback on some inodes on this super_block. No guarantees are made
1153 * on how many (if any) will be written, and this function does not wait
1154 * for IO completion of submitted IO.
1156 void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr)
1158 DECLARE_COMPLETION_ONSTACK(done);
1159 struct wb_writeback_work work = {
1161 .sync_mode = WB_SYNC_NONE,
1166 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1167 bdi_queue_work(sb->s_bdi, &work);
1168 wait_for_completion(&done);
1170 EXPORT_SYMBOL(writeback_inodes_sb_nr);
1173 * writeback_inodes_sb - writeback dirty inodes from given super_block
1174 * @sb: the superblock
1176 * Start writeback on some inodes on this super_block. No guarantees are made
1177 * on how many (if any) will be written, and this function does not wait
1178 * for IO completion of submitted IO.
1180 void writeback_inodes_sb(struct super_block *sb)
1182 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages());
1184 EXPORT_SYMBOL(writeback_inodes_sb);
1187 * writeback_inodes_sb_if_idle - start writeback if none underway
1188 * @sb: the superblock
1190 * Invoke writeback_inodes_sb if no writeback is currently underway.
1191 * Returns 1 if writeback was started, 0 if not.
1193 int writeback_inodes_sb_if_idle(struct super_block *sb)
1195 if (!writeback_in_progress(sb->s_bdi)) {
1196 down_read(&sb->s_umount);
1197 writeback_inodes_sb(sb);
1198 up_read(&sb->s_umount);
1203 EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1206 * writeback_inodes_sb_if_idle - start writeback if none underway
1207 * @sb: the superblock
1208 * @nr: the number of pages to write
1210 * Invoke writeback_inodes_sb if no writeback is currently underway.
1211 * Returns 1 if writeback was started, 0 if not.
1213 int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
1216 if (!writeback_in_progress(sb->s_bdi)) {
1217 down_read(&sb->s_umount);
1218 writeback_inodes_sb_nr(sb, nr);
1219 up_read(&sb->s_umount);
1224 EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1227 * sync_inodes_sb - sync sb inode pages
1228 * @sb: the superblock
1230 * This function writes and waits on any dirty inode belonging to this
1233 void sync_inodes_sb(struct super_block *sb)
1235 DECLARE_COMPLETION_ONSTACK(done);
1236 struct wb_writeback_work work = {
1238 .sync_mode = WB_SYNC_ALL,
1239 .nr_pages = LONG_MAX,
1244 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1246 bdi_queue_work(sb->s_bdi, &work);
1247 wait_for_completion(&done);
1251 EXPORT_SYMBOL(sync_inodes_sb);
1254 * write_inode_now - write an inode to disk
1255 * @inode: inode to write to disk
1256 * @sync: whether the write should be synchronous or not
1258 * This function commits an inode to disk immediately if it is dirty. This is
1259 * primarily needed by knfsd.
1261 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1263 int write_inode_now(struct inode *inode, int sync)
1266 struct writeback_control wbc = {
1267 .nr_to_write = LONG_MAX,
1268 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1270 .range_end = LLONG_MAX,
1273 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1274 wbc.nr_to_write = 0;
1277 spin_lock(&inode_lock);
1278 ret = writeback_single_inode(inode, &wbc);
1279 spin_unlock(&inode_lock);
1281 inode_sync_wait(inode);
1284 EXPORT_SYMBOL(write_inode_now);
1287 * sync_inode - write an inode and its pages to disk.
1288 * @inode: the inode to sync
1289 * @wbc: controls the writeback mode
1291 * sync_inode() will write an inode and its pages to disk. It will also
1292 * correctly update the inode on its superblock's dirty inode lists and will
1293 * update inode->i_state.
1295 * The caller must have a ref on the inode.
1297 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1301 spin_lock(&inode_lock);
1302 ret = writeback_single_inode(inode, wbc);
1303 spin_unlock(&inode_lock);
1306 EXPORT_SYMBOL(sync_inode);
1309 * sync_inode_metadata - write an inode to disk
1310 * @inode: the inode to sync
1311 * @wait: wait for I/O to complete.
1313 * Write an inode to disk and adjust its dirty state after completion.
1315 * Note: only writes the actual inode, no associated data or other metadata.
1317 int sync_inode_metadata(struct inode *inode, int wait)
1319 struct writeback_control wbc = {
1320 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1321 .nr_to_write = 0, /* metadata-only */
1324 return sync_inode(inode, &wbc);
1326 EXPORT_SYMBOL(sync_inode_metadata);