4 * Copyright (C) 2002, Linus Torvalds.
6 * Contains all the functions related to writing back and waiting
7 * upon dirty inodes against superblocks, and writing back dirty
8 * pages against inodes. ie: data writeback. Writeout of the
9 * inode itself is not handled here.
11 * 10Apr2002 Andrew Morton
12 * Split out of fs/inode.c
13 * Additions for address_space-based writeback
16 #include <linux/kernel.h>
17 #include <linux/module.h>
18 #include <linux/spinlock.h>
19 #include <linux/slab.h>
20 #include <linux/sched.h>
23 #include <linux/kthread.h>
24 #include <linux/freezer.h>
25 #include <linux/writeback.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/buffer_head.h>
29 #include <linux/tracepoint.h>
33 * Passed into wb_writeback(), essentially a subset of writeback_control
35 struct wb_writeback_work {
37 struct super_block *sb;
38 enum writeback_sync_modes sync_mode;
39 unsigned int for_kupdate:1;
40 unsigned int range_cyclic:1;
41 unsigned int for_background:1;
43 struct list_head list; /* pending work list */
44 struct completion *done; /* set if the caller waits */
48 * Include the creation of the trace points after defining the
49 * wb_writeback_work structure so that the definition remains local to this
52 #define CREATE_TRACE_POINTS
53 #include <trace/events/writeback.h>
56 * We don't actually have pdflush, but this one is exported though /proc...
58 int nr_pdflush_threads;
61 * writeback_in_progress - determine whether there is writeback in progress
62 * @bdi: the device's backing_dev_info structure.
64 * Determine whether there is writeback waiting to be handled against a
67 int writeback_in_progress(struct backing_dev_info *bdi)
69 return test_bit(BDI_writeback_running, &bdi->state);
72 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
74 struct super_block *sb = inode->i_sb;
76 if (strcmp(sb->s_type->name, "bdev") == 0)
77 return inode->i_mapping->backing_dev_info;
82 static inline struct inode *wb_inode(struct list_head *head)
84 return list_entry(head, struct inode, i_wb_list);
87 /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
88 static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
91 wake_up_process(bdi->wb.task);
94 * The bdi thread isn't there, wake up the forker thread which
95 * will create and run it.
97 wake_up_process(default_backing_dev_info.wb.task);
101 static void bdi_queue_work(struct backing_dev_info *bdi,
102 struct wb_writeback_work *work)
104 trace_writeback_queue(bdi, work);
106 spin_lock_bh(&bdi->wb_lock);
107 list_add_tail(&work->list, &bdi->work_list);
109 trace_writeback_nothread(bdi, work);
110 bdi_wakeup_flusher(bdi);
111 spin_unlock_bh(&bdi->wb_lock);
115 __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
118 struct wb_writeback_work *work;
121 * This is WB_SYNC_NONE writeback, so if allocation fails just
122 * wakeup the thread for old dirty data writeback
124 work = kzalloc(sizeof(*work), GFP_ATOMIC);
127 trace_writeback_nowork(bdi);
128 wake_up_process(bdi->wb.task);
133 work->sync_mode = WB_SYNC_NONE;
134 work->nr_pages = nr_pages;
135 work->range_cyclic = range_cyclic;
137 bdi_queue_work(bdi, work);
141 * bdi_start_writeback - start writeback
142 * @bdi: the backing device to write from
143 * @nr_pages: the number of pages to write
146 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
147 * started when this function returns, we make no guarentees on
148 * completion. Caller need not hold sb s_umount semaphore.
151 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages)
153 __bdi_start_writeback(bdi, nr_pages, true);
157 * bdi_start_background_writeback - start background writeback
158 * @bdi: the backing device to write from
161 * This makes sure WB_SYNC_NONE background writeback happens. When
162 * this function returns, it is only guaranteed that for given BDI
163 * some IO is happening if we are over background dirty threshold.
164 * Caller need not hold sb s_umount semaphore.
166 void bdi_start_background_writeback(struct backing_dev_info *bdi)
169 * We just wake up the flusher thread. It will perform background
170 * writeback as soon as there is no other work to do.
172 trace_writeback_wake_background(bdi);
173 spin_lock_bh(&bdi->wb_lock);
174 bdi_wakeup_flusher(bdi);
175 spin_unlock_bh(&bdi->wb_lock);
179 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
180 * furthest end of its superblock's dirty-inode list.
182 * Before stamping the inode's ->dirtied_when, we check to see whether it is
183 * already the most-recently-dirtied inode on the b_dirty list. If that is
184 * the case then the inode must have been redirtied while it was being written
185 * out and we don't reset its dirtied_when.
187 static void redirty_tail(struct inode *inode)
189 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
191 if (!list_empty(&wb->b_dirty)) {
194 tail = wb_inode(wb->b_dirty.next);
195 if (time_before(inode->dirtied_when, tail->dirtied_when))
196 inode->dirtied_when = jiffies;
198 list_move(&inode->i_wb_list, &wb->b_dirty);
202 * requeue inode for re-scanning after bdi->b_io list is exhausted.
204 static void requeue_io(struct inode *inode)
206 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
208 list_move(&inode->i_wb_list, &wb->b_more_io);
211 static void inode_sync_complete(struct inode *inode)
214 * Prevent speculative execution through spin_unlock(&inode_lock);
217 wake_up_bit(&inode->i_state, __I_SYNC);
220 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
222 bool ret = time_after(inode->dirtied_when, t);
225 * For inodes being constantly redirtied, dirtied_when can get stuck.
226 * It _appears_ to be in the future, but is actually in distant past.
227 * This test is necessary to prevent such wrapped-around relative times
228 * from permanently stopping the whole bdi writeback.
230 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
236 * Move expired dirty inodes from @delaying_queue to @dispatch_queue.
238 static void move_expired_inodes(struct list_head *delaying_queue,
239 struct list_head *dispatch_queue,
240 unsigned long *older_than_this)
243 struct list_head *pos, *node;
244 struct super_block *sb = NULL;
248 while (!list_empty(delaying_queue)) {
249 inode = wb_inode(delaying_queue->prev);
250 if (older_than_this &&
251 inode_dirtied_after(inode, *older_than_this))
253 if (sb && sb != inode->i_sb)
256 list_move(&inode->i_wb_list, &tmp);
259 /* just one sb in list, splice to dispatch_queue and we're done */
261 list_splice(&tmp, dispatch_queue);
265 /* Move inodes from one superblock together */
266 while (!list_empty(&tmp)) {
267 sb = wb_inode(tmp.prev)->i_sb;
268 list_for_each_prev_safe(pos, node, &tmp) {
269 inode = wb_inode(pos);
270 if (inode->i_sb == sb)
271 list_move(&inode->i_wb_list, dispatch_queue);
277 * Queue all expired dirty inodes for io, eldest first.
279 * newly dirtied b_dirty b_io b_more_io
280 * =============> gf edc BA
282 * newly dirtied b_dirty b_io b_more_io
283 * =============> g fBAedc
285 * +--> dequeue for IO
287 static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this)
289 list_splice_init(&wb->b_more_io, &wb->b_io);
290 move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this);
293 static int write_inode(struct inode *inode, struct writeback_control *wbc)
295 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
296 return inode->i_sb->s_op->write_inode(inode, wbc);
301 * Wait for writeback on an inode to complete.
303 static void inode_wait_for_writeback(struct inode *inode)
305 DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC);
306 wait_queue_head_t *wqh;
308 wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
309 while (inode->i_state & I_SYNC) {
310 spin_unlock(&inode_lock);
311 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
312 spin_lock(&inode_lock);
317 * Write out an inode's dirty pages. Called under inode_lock. Either the
318 * caller has ref on the inode (either via __iget or via syscall against an fd)
319 * or the inode has I_WILL_FREE set (via generic_forget_inode)
321 * If `wait' is set, wait on the writeout.
323 * The whole writeout design is quite complex and fragile. We want to avoid
324 * starvation of particular inodes when others are being redirtied, prevent
327 * Called under inode_lock.
330 writeback_single_inode(struct inode *inode, struct writeback_control *wbc)
332 struct address_space *mapping = inode->i_mapping;
336 if (!atomic_read(&inode->i_count))
337 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
339 WARN_ON(inode->i_state & I_WILL_FREE);
341 if (inode->i_state & I_SYNC) {
343 * If this inode is locked for writeback and we are not doing
344 * writeback-for-data-integrity, move it to b_more_io so that
345 * writeback can proceed with the other inodes on s_io.
347 * We'll have another go at writing back this inode when we
348 * completed a full scan of b_io.
350 if (wbc->sync_mode != WB_SYNC_ALL) {
356 * It's a data-integrity sync. We must wait.
358 inode_wait_for_writeback(inode);
361 BUG_ON(inode->i_state & I_SYNC);
363 /* Set I_SYNC, reset I_DIRTY_PAGES */
364 inode->i_state |= I_SYNC;
365 inode->i_state &= ~I_DIRTY_PAGES;
366 spin_unlock(&inode_lock);
368 ret = do_writepages(mapping, wbc);
371 * Make sure to wait on the data before writing out the metadata.
372 * This is important for filesystems that modify metadata on data
375 if (wbc->sync_mode == WB_SYNC_ALL) {
376 int err = filemap_fdatawait(mapping);
382 * Some filesystems may redirty the inode during the writeback
383 * due to delalloc, clear dirty metadata flags right before
386 spin_lock(&inode_lock);
387 dirty = inode->i_state & I_DIRTY;
388 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
389 spin_unlock(&inode_lock);
390 /* Don't write the inode if only I_DIRTY_PAGES was set */
391 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
392 int err = write_inode(inode, wbc);
397 spin_lock(&inode_lock);
398 inode->i_state &= ~I_SYNC;
399 if (!(inode->i_state & I_FREEING)) {
400 if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) {
402 * We didn't write back all the pages. nfs_writepages()
403 * sometimes bales out without doing anything.
405 inode->i_state |= I_DIRTY_PAGES;
406 if (wbc->nr_to_write <= 0) {
408 * slice used up: queue for next turn
413 * Writeback blocked by something other than
414 * congestion. Delay the inode for some time to
415 * avoid spinning on the CPU (100% iowait)
416 * retrying writeback of the dirty page/inode
417 * that cannot be performed immediately.
421 } else if (inode->i_state & I_DIRTY) {
423 * Filesystems can dirty the inode during writeback
424 * operations, such as delayed allocation during
425 * submission or metadata updates after data IO
431 * The inode is clean. At this point we either have
432 * a reference to the inode or it's on it's way out.
433 * No need to add it back to the LRU.
435 list_del_init(&inode->i_wb_list);
438 inode_sync_complete(inode);
443 * For background writeback the caller does not have the sb pinned
444 * before calling writeback. So make sure that we do pin it, so it doesn't
445 * go away while we are writing inodes from it.
447 static bool pin_sb_for_writeback(struct super_block *sb)
450 if (list_empty(&sb->s_instances)) {
451 spin_unlock(&sb_lock);
456 spin_unlock(&sb_lock);
458 if (down_read_trylock(&sb->s_umount)) {
461 up_read(&sb->s_umount);
469 * Write a portion of b_io inodes which belong to @sb.
471 * If @only_this_sb is true, then find and write all such
472 * inodes. Otherwise write only ones which go sequentially
475 * Return 1, if the caller writeback routine should be
476 * interrupted. Otherwise return 0.
478 static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb,
479 struct writeback_control *wbc, bool only_this_sb)
481 while (!list_empty(&wb->b_io)) {
483 struct inode *inode = wb_inode(wb->b_io.prev);
485 if (inode->i_sb != sb) {
488 * We only want to write back data for this
489 * superblock, move all inodes not belonging
490 * to it back onto the dirty list.
497 * The inode belongs to a different superblock.
498 * Bounce back to the caller to unpin this and
499 * pin the next superblock.
505 * Don't bother with new inodes or inodes beeing freed, first
506 * kind does not need peridic writeout yet, and for the latter
507 * kind writeout is handled by the freer.
509 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
515 * Was this inode dirtied after sync_sb_inodes was called?
516 * This keeps sync from extra jobs and livelock.
518 if (inode_dirtied_after(inode, wbc->wb_start))
522 pages_skipped = wbc->pages_skipped;
523 writeback_single_inode(inode, wbc);
524 if (wbc->pages_skipped != pages_skipped) {
526 * writeback is not making progress due to locked
527 * buffers. Skip this inode for now.
531 spin_unlock(&inode_lock);
534 spin_lock(&inode_lock);
535 if (wbc->nr_to_write <= 0) {
539 if (!list_empty(&wb->b_more_io))
546 void writeback_inodes_wb(struct bdi_writeback *wb,
547 struct writeback_control *wbc)
552 wbc->wb_start = jiffies; /* livelock avoidance */
553 spin_lock(&inode_lock);
554 if (!wbc->for_kupdate || list_empty(&wb->b_io))
555 queue_io(wb, wbc->older_than_this);
557 while (!list_empty(&wb->b_io)) {
558 struct inode *inode = wb_inode(wb->b_io.prev);
559 struct super_block *sb = inode->i_sb;
561 if (!pin_sb_for_writeback(sb)) {
565 ret = writeback_sb_inodes(sb, wb, wbc, false);
571 spin_unlock(&inode_lock);
572 /* Leave any unwritten inodes on b_io */
575 static void __writeback_inodes_sb(struct super_block *sb,
576 struct bdi_writeback *wb, struct writeback_control *wbc)
578 WARN_ON(!rwsem_is_locked(&sb->s_umount));
580 spin_lock(&inode_lock);
581 if (!wbc->for_kupdate || list_empty(&wb->b_io))
582 queue_io(wb, wbc->older_than_this);
583 writeback_sb_inodes(sb, wb, wbc, true);
584 spin_unlock(&inode_lock);
588 * The maximum number of pages to writeout in a single bdi flush/kupdate
589 * operation. We do this so we don't hold I_SYNC against an inode for
590 * enormous amounts of time, which would block a userspace task which has
591 * been forced to throttle against that inode. Also, the code reevaluates
592 * the dirty each time it has written this many pages.
594 #define MAX_WRITEBACK_PAGES 1024
596 static inline bool over_bground_thresh(void)
598 unsigned long background_thresh, dirty_thresh;
600 global_dirty_limits(&background_thresh, &dirty_thresh);
602 return (global_page_state(NR_FILE_DIRTY) +
603 global_page_state(NR_UNSTABLE_NFS) > background_thresh);
607 * Explicit flushing or periodic writeback of "old" data.
609 * Define "old": the first time one of an inode's pages is dirtied, we mark the
610 * dirtying-time in the inode's address_space. So this periodic writeback code
611 * just walks the superblock inode list, writing back any inodes which are
612 * older than a specific point in time.
614 * Try to run once per dirty_writeback_interval. But if a writeback event
615 * takes longer than a dirty_writeback_interval interval, then leave a
618 * older_than_this takes precedence over nr_to_write. So we'll only write back
619 * all dirty pages if they are all attached to "old" mappings.
621 static long wb_writeback(struct bdi_writeback *wb,
622 struct wb_writeback_work *work)
624 struct writeback_control wbc = {
625 .sync_mode = work->sync_mode,
626 .older_than_this = NULL,
627 .for_kupdate = work->for_kupdate,
628 .for_background = work->for_background,
629 .range_cyclic = work->range_cyclic,
631 unsigned long oldest_jif;
635 if (wbc.for_kupdate) {
636 wbc.older_than_this = &oldest_jif;
637 oldest_jif = jiffies -
638 msecs_to_jiffies(dirty_expire_interval * 10);
640 if (!wbc.range_cyclic) {
642 wbc.range_end = LLONG_MAX;
645 wbc.wb_start = jiffies; /* livelock avoidance */
648 * Stop writeback when nr_pages has been consumed
650 if (work->nr_pages <= 0)
654 * For background writeout, stop when we are below the
655 * background dirty threshold
657 if (work->for_background && !over_bground_thresh())
661 wbc.nr_to_write = MAX_WRITEBACK_PAGES;
662 wbc.pages_skipped = 0;
664 trace_wbc_writeback_start(&wbc, wb->bdi);
666 __writeback_inodes_sb(work->sb, wb, &wbc);
668 writeback_inodes_wb(wb, &wbc);
669 trace_wbc_writeback_written(&wbc, wb->bdi);
671 work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write;
672 wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write;
675 * If we consumed everything, see if we have more
677 if (wbc.nr_to_write <= 0)
680 * Didn't write everything and we don't have more IO, bail
685 * Did we write something? Try for more
687 if (wbc.nr_to_write < MAX_WRITEBACK_PAGES)
690 * Nothing written. Wait for some inode to
691 * become available for writeback. Otherwise
692 * we'll just busyloop.
694 spin_lock(&inode_lock);
695 if (!list_empty(&wb->b_more_io)) {
696 inode = wb_inode(wb->b_more_io.prev);
697 trace_wbc_writeback_wait(&wbc, wb->bdi);
698 inode_wait_for_writeback(inode);
700 spin_unlock(&inode_lock);
707 * Return the next wb_writeback_work struct that hasn't been processed yet.
709 static struct wb_writeback_work *
710 get_next_work_item(struct backing_dev_info *bdi)
712 struct wb_writeback_work *work = NULL;
714 spin_lock_bh(&bdi->wb_lock);
715 if (!list_empty(&bdi->work_list)) {
716 work = list_entry(bdi->work_list.next,
717 struct wb_writeback_work, list);
718 list_del_init(&work->list);
720 spin_unlock_bh(&bdi->wb_lock);
725 * Add in the number of potentially dirty inodes, because each inode
726 * write can dirty pagecache in the underlying blockdev.
728 static unsigned long get_nr_dirty_pages(void)
730 return global_page_state(NR_FILE_DIRTY) +
731 global_page_state(NR_UNSTABLE_NFS) +
732 get_nr_dirty_inodes();
735 static long wb_check_background_flush(struct bdi_writeback *wb)
737 if (over_bground_thresh()) {
739 struct wb_writeback_work work = {
740 .nr_pages = LONG_MAX,
741 .sync_mode = WB_SYNC_NONE,
746 return wb_writeback(wb, &work);
752 static long wb_check_old_data_flush(struct bdi_writeback *wb)
754 unsigned long expired;
758 * When set to zero, disable periodic writeback
760 if (!dirty_writeback_interval)
763 expired = wb->last_old_flush +
764 msecs_to_jiffies(dirty_writeback_interval * 10);
765 if (time_before(jiffies, expired))
768 wb->last_old_flush = jiffies;
769 nr_pages = get_nr_dirty_pages();
772 struct wb_writeback_work work = {
773 .nr_pages = nr_pages,
774 .sync_mode = WB_SYNC_NONE,
779 return wb_writeback(wb, &work);
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 wb_writeback_work *work;
794 set_bit(BDI_writeback_running, &wb->bdi->state);
795 while ((work = get_next_work_item(bdi)) != NULL) {
797 * Override sync mode, in case we must wait for completion
798 * because this thread is exiting now.
801 work->sync_mode = WB_SYNC_ALL;
803 trace_writeback_exec(bdi, work);
805 wrote += wb_writeback(wb, work);
808 * Notify the caller of completion if this is a synchronous
809 * work item, otherwise just free it.
812 complete(work->done);
818 * Check for periodic writeback, kupdated() style
820 wrote += wb_check_old_data_flush(wb);
821 wrote += wb_check_background_flush(wb);
822 clear_bit(BDI_writeback_running, &wb->bdi->state);
828 * Handle writeback of dirty data for the device backed by this bdi. Also
829 * wakes up periodically and does kupdated style flushing.
831 int bdi_writeback_thread(void *data)
833 struct bdi_writeback *wb = data;
834 struct backing_dev_info *bdi = wb->bdi;
837 current->flags |= PF_SWAPWRITE;
839 wb->last_active = jiffies;
842 * Our parent may run at a different priority, just set us to normal
844 set_user_nice(current, 0);
846 trace_writeback_thread_start(bdi);
848 while (!kthread_should_stop()) {
850 * Remove own delayed wake-up timer, since we are already awake
851 * and we'll take care of the preriodic write-back.
853 del_timer(&wb->wakeup_timer);
855 pages_written = wb_do_writeback(wb, 0);
857 trace_writeback_pages_written(pages_written);
860 wb->last_active = jiffies;
862 set_current_state(TASK_INTERRUPTIBLE);
863 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
864 __set_current_state(TASK_RUNNING);
868 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
869 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
872 * We have nothing to do, so can go sleep without any
873 * timeout and save power. When a work is queued or
874 * something is made dirty - we will be woken up.
882 /* Flush any work that raced with us exiting */
883 if (!list_empty(&bdi->work_list))
884 wb_do_writeback(wb, 1);
886 trace_writeback_thread_stop(bdi);
892 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
895 void wakeup_flusher_threads(long nr_pages)
897 struct backing_dev_info *bdi;
900 nr_pages = global_page_state(NR_FILE_DIRTY) +
901 global_page_state(NR_UNSTABLE_NFS);
905 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
906 if (!bdi_has_dirty_io(bdi))
908 __bdi_start_writeback(bdi, nr_pages, false);
913 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
915 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
916 struct dentry *dentry;
917 const char *name = "?";
919 dentry = d_find_alias(inode);
921 spin_lock(&dentry->d_lock);
922 name = (const char *) dentry->d_name.name;
925 "%s(%d): dirtied inode %lu (%s) on %s\n",
926 current->comm, task_pid_nr(current), inode->i_ino,
927 name, inode->i_sb->s_id);
929 spin_unlock(&dentry->d_lock);
936 * __mark_inode_dirty - internal function
937 * @inode: inode to mark
938 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
939 * Mark an inode as dirty. Callers should use mark_inode_dirty or
940 * mark_inode_dirty_sync.
942 * Put the inode on the super block's dirty list.
944 * CAREFUL! We mark it dirty unconditionally, but move it onto the
945 * dirty list only if it is hashed or if it refers to a blockdev.
946 * If it was not hashed, it will never be added to the dirty list
947 * even if it is later hashed, as it will have been marked dirty already.
949 * In short, make sure you hash any inodes _before_ you start marking
952 * This function *must* be atomic for the I_DIRTY_PAGES case -
953 * set_page_dirty() is called under spinlock in several places.
955 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
956 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
957 * the kernel-internal blockdev inode represents the dirtying time of the
958 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
959 * page->mapping->host, so the page-dirtying time is recorded in the internal
962 void __mark_inode_dirty(struct inode *inode, int flags)
964 struct super_block *sb = inode->i_sb;
965 struct backing_dev_info *bdi = NULL;
966 bool wakeup_bdi = false;
969 * Don't do this for I_DIRTY_PAGES - that doesn't actually
970 * dirty the inode itself
972 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
973 if (sb->s_op->dirty_inode)
974 sb->s_op->dirty_inode(inode);
978 * make sure that changes are seen by all cpus before we test i_state
983 /* avoid the locking if we can */
984 if ((inode->i_state & flags) == flags)
987 if (unlikely(block_dump))
988 block_dump___mark_inode_dirty(inode);
990 spin_lock(&inode_lock);
991 if ((inode->i_state & flags) != flags) {
992 const int was_dirty = inode->i_state & I_DIRTY;
994 inode->i_state |= flags;
997 * If the inode is being synced, just update its dirty state.
998 * The unlocker will place the inode on the appropriate
999 * superblock list, based upon its state.
1001 if (inode->i_state & I_SYNC)
1005 * Only add valid (hashed) inodes to the superblock's
1006 * dirty list. Add blockdev inodes as well.
1008 if (!S_ISBLK(inode->i_mode)) {
1009 if (inode_unhashed(inode))
1012 if (inode->i_state & I_FREEING)
1016 * If the inode was already on b_dirty/b_io/b_more_io, don't
1017 * reposition it (that would break b_dirty time-ordering).
1020 bdi = inode_to_bdi(inode);
1022 if (bdi_cap_writeback_dirty(bdi)) {
1023 WARN(!test_bit(BDI_registered, &bdi->state),
1024 "bdi-%s not registered\n", bdi->name);
1027 * If this is the first dirty inode for this
1028 * bdi, we have to wake-up the corresponding
1029 * bdi thread to make sure background
1030 * write-back happens later.
1032 if (!wb_has_dirty_io(&bdi->wb))
1036 inode->dirtied_when = jiffies;
1037 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1041 spin_unlock(&inode_lock);
1044 bdi_wakeup_thread_delayed(bdi);
1046 EXPORT_SYMBOL(__mark_inode_dirty);
1049 * Write out a superblock's list of dirty inodes. A wait will be performed
1050 * upon no inodes, all inodes or the final one, depending upon sync_mode.
1052 * If older_than_this is non-NULL, then only write out inodes which
1053 * had their first dirtying at a time earlier than *older_than_this.
1055 * If `bdi' is non-zero then we're being asked to writeback a specific queue.
1056 * This function assumes that the blockdev superblock's inodes are backed by
1057 * a variety of queues, so all inodes are searched. For other superblocks,
1058 * assume that all inodes are backed by the same queue.
1060 * The inodes to be written are parked on bdi->b_io. They are moved back onto
1061 * bdi->b_dirty as they are selected for writing. This way, none can be missed
1062 * on the writer throttling path, and we get decent balancing between many
1063 * throttled threads: we don't want them all piling up on inode_sync_wait.
1065 static void wait_sb_inodes(struct super_block *sb)
1067 struct inode *inode, *old_inode = NULL;
1070 * We need to be protected against the filesystem going from
1071 * r/o to r/w or vice versa.
1073 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1075 spin_lock(&inode_lock);
1078 * Data integrity sync. Must wait for all pages under writeback,
1079 * because there may have been pages dirtied before our sync
1080 * call, but which had writeout started before we write it out.
1081 * In which case, the inode may not be on the dirty list, but
1082 * we still have to wait for that writeout.
1084 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1085 struct address_space *mapping;
1087 if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW))
1089 mapping = inode->i_mapping;
1090 if (mapping->nrpages == 0)
1093 spin_unlock(&inode_lock);
1095 * We hold a reference to 'inode' so it couldn't have
1096 * been removed from s_inodes list while we dropped the
1097 * inode_lock. We cannot iput the inode now as we can
1098 * be holding the last reference and we cannot iput it
1099 * under inode_lock. So we keep the reference and iput
1105 filemap_fdatawait(mapping);
1109 spin_lock(&inode_lock);
1111 spin_unlock(&inode_lock);
1116 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1117 * @sb: the superblock
1118 * @nr: the number of pages to write
1120 * Start writeback on some inodes on this super_block. No guarantees are made
1121 * on how many (if any) will be written, and this function does not wait
1122 * for IO completion of submitted IO.
1124 void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr)
1126 DECLARE_COMPLETION_ONSTACK(done);
1127 struct wb_writeback_work work = {
1129 .sync_mode = WB_SYNC_NONE,
1134 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1135 bdi_queue_work(sb->s_bdi, &work);
1136 wait_for_completion(&done);
1138 EXPORT_SYMBOL(writeback_inodes_sb_nr);
1141 * writeback_inodes_sb - writeback dirty inodes from given super_block
1142 * @sb: the superblock
1144 * Start writeback on some inodes on this super_block. No guarantees are made
1145 * on how many (if any) will be written, and this function does not wait
1146 * for IO completion of submitted IO.
1148 void writeback_inodes_sb(struct super_block *sb)
1150 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages());
1152 EXPORT_SYMBOL(writeback_inodes_sb);
1155 * writeback_inodes_sb_if_idle - start writeback if none underway
1156 * @sb: the superblock
1158 * Invoke writeback_inodes_sb if no writeback is currently underway.
1159 * Returns 1 if writeback was started, 0 if not.
1161 int writeback_inodes_sb_if_idle(struct super_block *sb)
1163 if (!writeback_in_progress(sb->s_bdi)) {
1164 down_read(&sb->s_umount);
1165 writeback_inodes_sb(sb);
1166 up_read(&sb->s_umount);
1171 EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1174 * writeback_inodes_sb_if_idle - start writeback if none underway
1175 * @sb: the superblock
1176 * @nr: the number of pages to write
1178 * Invoke writeback_inodes_sb if no writeback is currently underway.
1179 * Returns 1 if writeback was started, 0 if not.
1181 int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
1184 if (!writeback_in_progress(sb->s_bdi)) {
1185 down_read(&sb->s_umount);
1186 writeback_inodes_sb_nr(sb, nr);
1187 up_read(&sb->s_umount);
1192 EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1195 * sync_inodes_sb - sync sb inode pages
1196 * @sb: the superblock
1198 * This function writes and waits on any dirty inode belonging to this
1199 * super_block. The number of pages synced is returned.
1201 void sync_inodes_sb(struct super_block *sb)
1203 DECLARE_COMPLETION_ONSTACK(done);
1204 struct wb_writeback_work work = {
1206 .sync_mode = WB_SYNC_ALL,
1207 .nr_pages = LONG_MAX,
1212 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1214 bdi_queue_work(sb->s_bdi, &work);
1215 wait_for_completion(&done);
1219 EXPORT_SYMBOL(sync_inodes_sb);
1222 * write_inode_now - write an inode to disk
1223 * @inode: inode to write to disk
1224 * @sync: whether the write should be synchronous or not
1226 * This function commits an inode to disk immediately if it is dirty. This is
1227 * primarily needed by knfsd.
1229 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1231 int write_inode_now(struct inode *inode, int sync)
1234 struct writeback_control wbc = {
1235 .nr_to_write = LONG_MAX,
1236 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1238 .range_end = LLONG_MAX,
1241 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1242 wbc.nr_to_write = 0;
1245 spin_lock(&inode_lock);
1246 ret = writeback_single_inode(inode, &wbc);
1247 spin_unlock(&inode_lock);
1249 inode_sync_wait(inode);
1252 EXPORT_SYMBOL(write_inode_now);
1255 * sync_inode - write an inode and its pages to disk.
1256 * @inode: the inode to sync
1257 * @wbc: controls the writeback mode
1259 * sync_inode() will write an inode and its pages to disk. It will also
1260 * correctly update the inode on its superblock's dirty inode lists and will
1261 * update inode->i_state.
1263 * The caller must have a ref on the inode.
1265 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1269 spin_lock(&inode_lock);
1270 ret = writeback_single_inode(inode, wbc);
1271 spin_unlock(&inode_lock);
1274 EXPORT_SYMBOL(sync_inode);
1277 * sync_inode - write an inode to disk
1278 * @inode: the inode to sync
1279 * @wait: wait for I/O to complete.
1281 * Write an inode to disk and adjust it's dirty state after completion.
1283 * Note: only writes the actual inode, no associated data or other metadata.
1285 int sync_inode_metadata(struct inode *inode, int wait)
1287 struct writeback_control wbc = {
1288 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1289 .nr_to_write = 0, /* metadata-only */
1292 return sync_inode(inode, &wbc);
1294 EXPORT_SYMBOL(sync_inode_metadata);