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/export.h>
18 #include <linux/spinlock.h>
19 #include <linux/slab.h>
20 #include <linux/sched.h>
23 #include <linux/pagemap.h>
24 #include <linux/kthread.h>
25 #include <linux/freezer.h>
26 #include <linux/writeback.h>
27 #include <linux/blkdev.h>
28 #include <linux/backing-dev.h>
29 #include <linux/tracepoint.h>
33 * 4MB minimal write chunk size
35 #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10))
38 * Passed into wb_writeback(), essentially a subset of writeback_control
40 struct wb_writeback_work {
42 struct super_block *sb;
43 unsigned long *older_than_this;
44 enum writeback_sync_modes sync_mode;
45 unsigned int tagged_writepages:1;
46 unsigned int for_kupdate:1;
47 unsigned int range_cyclic:1;
48 unsigned int for_background:1;
49 enum wb_reason reason; /* why was writeback initiated? */
51 struct list_head list; /* pending work list */
52 struct completion *done; /* set if the caller waits */
56 * writeback_in_progress - determine whether there is writeback in progress
57 * @bdi: the device's backing_dev_info structure.
59 * Determine whether there is writeback waiting to be handled against a
62 int writeback_in_progress(struct backing_dev_info *bdi)
64 return test_bit(BDI_writeback_running, &bdi->state);
66 EXPORT_SYMBOL(writeback_in_progress);
68 static inline struct backing_dev_info *inode_to_bdi(struct inode *inode)
70 struct super_block *sb = inode->i_sb;
72 if (strcmp(sb->s_type->name, "bdev") == 0)
73 return inode->i_mapping->backing_dev_info;
78 static inline struct inode *wb_inode(struct list_head *head)
80 return list_entry(head, struct inode, i_wb_list);
84 * Include the creation of the trace points after defining the
85 * wb_writeback_work structure and inline functions so that the definition
86 * remains local to this file.
88 #define CREATE_TRACE_POINTS
89 #include <trace/events/writeback.h>
91 /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */
92 static void bdi_wakeup_flusher(struct backing_dev_info *bdi)
95 wake_up_process(bdi->wb.task);
98 * The bdi thread isn't there, wake up the forker thread which
99 * will create and run it.
101 wake_up_process(default_backing_dev_info.wb.task);
105 static void bdi_queue_work(struct backing_dev_info *bdi,
106 struct wb_writeback_work *work)
108 trace_writeback_queue(bdi, work);
110 spin_lock_bh(&bdi->wb_lock);
111 list_add_tail(&work->list, &bdi->work_list);
113 trace_writeback_nothread(bdi, work);
114 bdi_wakeup_flusher(bdi);
115 spin_unlock_bh(&bdi->wb_lock);
119 __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
120 bool range_cyclic, enum wb_reason reason)
122 struct wb_writeback_work *work;
125 * This is WB_SYNC_NONE writeback, so if allocation fails just
126 * wakeup the thread for old dirty data writeback
128 work = kzalloc(sizeof(*work), GFP_ATOMIC);
131 trace_writeback_nowork(bdi);
132 wake_up_process(bdi->wb.task);
137 work->sync_mode = WB_SYNC_NONE;
138 work->nr_pages = nr_pages;
139 work->range_cyclic = range_cyclic;
140 work->reason = reason;
142 bdi_queue_work(bdi, work);
146 * bdi_start_writeback - start writeback
147 * @bdi: the backing device to write from
148 * @nr_pages: the number of pages to write
149 * @reason: reason why some writeback work was initiated
152 * This does WB_SYNC_NONE opportunistic writeback. The IO is only
153 * started when this function returns, we make no guarantees on
154 * completion. Caller need not hold sb s_umount semaphore.
157 void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages,
158 enum wb_reason reason)
160 __bdi_start_writeback(bdi, nr_pages, true, reason);
164 * bdi_start_background_writeback - start background writeback
165 * @bdi: the backing device to write from
168 * This makes sure WB_SYNC_NONE background writeback happens. When
169 * this function returns, it is only guaranteed that for given BDI
170 * some IO is happening if we are over background dirty threshold.
171 * Caller need not hold sb s_umount semaphore.
173 void bdi_start_background_writeback(struct backing_dev_info *bdi)
176 * We just wake up the flusher thread. It will perform background
177 * writeback as soon as there is no other work to do.
179 trace_writeback_wake_background(bdi);
180 spin_lock_bh(&bdi->wb_lock);
181 bdi_wakeup_flusher(bdi);
182 spin_unlock_bh(&bdi->wb_lock);
186 * Remove the inode from the writeback list it is on.
188 void inode_wb_list_del(struct inode *inode)
190 struct backing_dev_info *bdi = inode_to_bdi(inode);
192 spin_lock(&bdi->wb.list_lock);
193 list_del_init(&inode->i_wb_list);
194 spin_unlock(&bdi->wb.list_lock);
198 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
199 * furthest end of its superblock's dirty-inode list.
201 * Before stamping the inode's ->dirtied_when, we check to see whether it is
202 * already the most-recently-dirtied inode on the b_dirty list. If that is
203 * the case then the inode must have been redirtied while it was being written
204 * out and we don't reset its dirtied_when.
206 static void redirty_tail(struct inode *inode, struct bdi_writeback *wb)
208 assert_spin_locked(&wb->list_lock);
209 if (!list_empty(&wb->b_dirty)) {
212 tail = wb_inode(wb->b_dirty.next);
213 if (time_before(inode->dirtied_when, tail->dirtied_when))
214 inode->dirtied_when = jiffies;
216 list_move(&inode->i_wb_list, &wb->b_dirty);
220 * requeue inode for re-scanning after bdi->b_io list is exhausted.
222 static void requeue_io(struct inode *inode, struct bdi_writeback *wb)
224 assert_spin_locked(&wb->list_lock);
225 list_move(&inode->i_wb_list, &wb->b_more_io);
228 static void inode_sync_complete(struct inode *inode)
230 inode->i_state &= ~I_SYNC;
231 /* Waiters must see I_SYNC cleared before being woken up */
233 wake_up_bit(&inode->i_state, __I_SYNC);
236 static bool inode_dirtied_after(struct inode *inode, unsigned long t)
238 bool ret = time_after(inode->dirtied_when, t);
241 * For inodes being constantly redirtied, dirtied_when can get stuck.
242 * It _appears_ to be in the future, but is actually in distant past.
243 * This test is necessary to prevent such wrapped-around relative times
244 * from permanently stopping the whole bdi writeback.
246 ret = ret && time_before_eq(inode->dirtied_when, jiffies);
252 * Move expired (dirtied after work->older_than_this) dirty inodes from
253 * @delaying_queue to @dispatch_queue.
255 static int move_expired_inodes(struct list_head *delaying_queue,
256 struct list_head *dispatch_queue,
257 struct wb_writeback_work *work)
260 struct list_head *pos, *node;
261 struct super_block *sb = NULL;
266 while (!list_empty(delaying_queue)) {
267 inode = wb_inode(delaying_queue->prev);
268 if (work->older_than_this &&
269 inode_dirtied_after(inode, *work->older_than_this))
271 if (sb && sb != inode->i_sb)
274 list_move(&inode->i_wb_list, &tmp);
278 /* just one sb in list, splice to dispatch_queue and we're done */
280 list_splice(&tmp, dispatch_queue);
284 /* Move inodes from one superblock together */
285 while (!list_empty(&tmp)) {
286 sb = wb_inode(tmp.prev)->i_sb;
287 list_for_each_prev_safe(pos, node, &tmp) {
288 inode = wb_inode(pos);
289 if (inode->i_sb == sb)
290 list_move(&inode->i_wb_list, dispatch_queue);
298 * Queue all expired dirty inodes for io, eldest first.
300 * newly dirtied b_dirty b_io b_more_io
301 * =============> gf edc BA
303 * newly dirtied b_dirty b_io b_more_io
304 * =============> g fBAedc
306 * +--> dequeue for IO
308 static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work)
311 assert_spin_locked(&wb->list_lock);
312 list_splice_init(&wb->b_more_io, &wb->b_io);
313 moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work);
314 trace_writeback_queue_io(wb, work, moved);
317 static int write_inode(struct inode *inode, struct writeback_control *wbc)
319 if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode))
320 return inode->i_sb->s_op->write_inode(inode, wbc);
325 * Wait for writeback on an inode to complete. Called with i_lock held.
326 * Caller must make sure inode cannot go away when we drop i_lock.
328 static void __inode_wait_for_writeback(struct inode *inode)
329 __releases(inode->i_lock)
330 __acquires(inode->i_lock)
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);
336 while (inode->i_state & I_SYNC) {
337 spin_unlock(&inode->i_lock);
338 __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE);
339 spin_lock(&inode->i_lock);
344 * Wait for writeback on an inode to complete. Caller must have inode pinned.
346 void inode_wait_for_writeback(struct inode *inode)
348 spin_lock(&inode->i_lock);
349 __inode_wait_for_writeback(inode);
350 spin_unlock(&inode->i_lock);
354 * Sleep until I_SYNC is cleared. This function must be called with i_lock
355 * held and drops it. It is aimed for callers not holding any inode reference
356 * so once i_lock is dropped, inode can go away.
358 static void inode_sleep_on_writeback(struct inode *inode)
359 __releases(inode->i_lock)
362 wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC);
365 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
366 sleep = inode->i_state & I_SYNC;
367 spin_unlock(&inode->i_lock);
370 finish_wait(wqh, &wait);
374 * Find proper writeback list for the inode depending on its current state and
375 * possibly also change of its state while we were doing writeback. Here we
376 * handle things such as livelock prevention or fairness of writeback among
377 * inodes. This function can be called only by flusher thread - noone else
378 * processes all inodes in writeback lists and requeueing inodes behind flusher
379 * thread's back can have unexpected consequences.
381 static void requeue_inode(struct inode *inode, struct bdi_writeback *wb,
382 struct writeback_control *wbc)
384 if (inode->i_state & I_FREEING)
388 * Sync livelock prevention. Each inode is tagged and synced in one
389 * shot. If still dirty, it will be redirty_tail()'ed below. Update
390 * the dirty time to prevent enqueue and sync it again.
392 if ((inode->i_state & I_DIRTY) &&
393 (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages))
394 inode->dirtied_when = jiffies;
396 if (wbc->pages_skipped) {
398 * writeback is not making progress due to locked
399 * buffers. Skip this inode for now.
401 redirty_tail(inode, wb);
405 if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) {
407 * We didn't write back all the pages. nfs_writepages()
408 * sometimes bales out without doing anything.
410 if (wbc->nr_to_write <= 0) {
411 /* Slice used up. Queue for next turn. */
412 requeue_io(inode, wb);
415 * Writeback blocked by something other than
416 * congestion. Delay the inode for some time to
417 * avoid spinning on the CPU (100% iowait)
418 * retrying writeback of the dirty page/inode
419 * that cannot be performed immediately.
421 redirty_tail(inode, wb);
423 } else if (inode->i_state & I_DIRTY) {
425 * Filesystems can dirty the inode during writeback operations,
426 * such as delayed allocation during submission or metadata
427 * updates after data IO completion.
429 redirty_tail(inode, wb);
431 /* The inode is clean. Remove from writeback lists. */
432 list_del_init(&inode->i_wb_list);
437 * Write out an inode and its dirty pages. Do not update the writeback list
438 * linkage. That is left to the caller. The caller is also responsible for
439 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
442 __writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
443 struct writeback_control *wbc)
445 struct address_space *mapping = inode->i_mapping;
446 long nr_to_write = wbc->nr_to_write;
450 WARN_ON(!(inode->i_state & I_SYNC));
452 ret = do_writepages(mapping, wbc);
455 * Make sure to wait on the data before writing out the metadata.
456 * This is important for filesystems that modify metadata on data
459 if (wbc->sync_mode == WB_SYNC_ALL) {
460 int err = filemap_fdatawait(mapping);
466 * Some filesystems may redirty the inode during the writeback
467 * due to delalloc, clear dirty metadata flags right before
470 spin_lock(&inode->i_lock);
471 /* Clear I_DIRTY_PAGES if we've written out all dirty pages */
472 if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY))
473 inode->i_state &= ~I_DIRTY_PAGES;
474 dirty = inode->i_state & I_DIRTY;
475 inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC);
476 spin_unlock(&inode->i_lock);
477 /* Don't write the inode if only I_DIRTY_PAGES was set */
478 if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
479 int err = write_inode(inode, wbc);
483 trace_writeback_single_inode(inode, wbc, nr_to_write);
488 * Write out an inode's dirty pages. Either the caller has an active reference
489 * on the inode or the inode has I_WILL_FREE set.
491 * This function is designed to be called for writing back one inode which
492 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
493 * and does more profound writeback list handling in writeback_sb_inodes().
496 writeback_single_inode(struct inode *inode, struct bdi_writeback *wb,
497 struct writeback_control *wbc)
501 spin_lock(&inode->i_lock);
502 if (!atomic_read(&inode->i_count))
503 WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING)));
505 WARN_ON(inode->i_state & I_WILL_FREE);
507 if (inode->i_state & I_SYNC) {
508 if (wbc->sync_mode != WB_SYNC_ALL)
511 * It's a data-integrity sync. We must wait. Since callers hold
512 * inode reference or inode has I_WILL_FREE set, it cannot go
515 __inode_wait_for_writeback(inode);
517 WARN_ON(inode->i_state & I_SYNC);
519 * Skip inode if it is clean. We don't want to mess with writeback
520 * lists in this function since flusher thread may be doing for example
521 * sync in parallel and if we move the inode, it could get skipped. So
522 * here we make sure inode is on some writeback list and leave it there
523 * unless we have completely cleaned the inode.
525 if (!(inode->i_state & I_DIRTY))
527 inode->i_state |= I_SYNC;
528 spin_unlock(&inode->i_lock);
530 ret = __writeback_single_inode(inode, wb, wbc);
532 spin_lock(&wb->list_lock);
533 spin_lock(&inode->i_lock);
535 * If inode is clean, remove it from writeback lists. Otherwise don't
536 * touch it. See comment above for explanation.
538 if (!(inode->i_state & I_DIRTY))
539 list_del_init(&inode->i_wb_list);
540 spin_unlock(&wb->list_lock);
541 inode_sync_complete(inode);
543 spin_unlock(&inode->i_lock);
547 static long writeback_chunk_size(struct backing_dev_info *bdi,
548 struct wb_writeback_work *work)
553 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
554 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
555 * here avoids calling into writeback_inodes_wb() more than once.
557 * The intended call sequence for WB_SYNC_ALL writeback is:
560 * writeback_sb_inodes() <== called only once
561 * write_cache_pages() <== called once for each inode
562 * (quickly) tag currently dirty pages
563 * (maybe slowly) sync all tagged pages
565 if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages)
568 pages = min(bdi->avg_write_bandwidth / 2,
569 global_dirty_limit / DIRTY_SCOPE);
570 pages = min(pages, work->nr_pages);
571 pages = round_down(pages + MIN_WRITEBACK_PAGES,
572 MIN_WRITEBACK_PAGES);
579 * Write a portion of b_io inodes which belong to @sb.
581 * If @only_this_sb is true, then find and write all such
582 * inodes. Otherwise write only ones which go sequentially
585 * Return the number of pages and/or inodes written.
587 static long writeback_sb_inodes(struct super_block *sb,
588 struct bdi_writeback *wb,
589 struct wb_writeback_work *work)
591 struct writeback_control wbc = {
592 .sync_mode = work->sync_mode,
593 .tagged_writepages = work->tagged_writepages,
594 .for_kupdate = work->for_kupdate,
595 .for_background = work->for_background,
596 .range_cyclic = work->range_cyclic,
598 .range_end = LLONG_MAX,
600 unsigned long start_time = jiffies;
602 long wrote = 0; /* count both pages and inodes */
604 while (!list_empty(&wb->b_io)) {
605 struct inode *inode = wb_inode(wb->b_io.prev);
607 if (inode->i_sb != sb) {
610 * We only want to write back data for this
611 * superblock, move all inodes not belonging
612 * to it back onto the dirty list.
614 redirty_tail(inode, wb);
619 * The inode belongs to a different superblock.
620 * Bounce back to the caller to unpin this and
621 * pin the next superblock.
627 * Don't bother with new inodes or inodes being freed, first
628 * kind does not need periodic writeout yet, and for the latter
629 * kind writeout is handled by the freer.
631 spin_lock(&inode->i_lock);
632 if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) {
633 spin_unlock(&inode->i_lock);
634 redirty_tail(inode, wb);
637 if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) {
639 * If this inode is locked for writeback and we are not
640 * doing writeback-for-data-integrity, move it to
641 * b_more_io so that writeback can proceed with the
642 * other inodes on s_io.
644 * We'll have another go at writing back this inode
645 * when we completed a full scan of b_io.
647 spin_unlock(&inode->i_lock);
648 requeue_io(inode, wb);
649 trace_writeback_sb_inodes_requeue(inode);
652 spin_unlock(&wb->list_lock);
655 * We already requeued the inode if it had I_SYNC set and we
656 * are doing WB_SYNC_NONE writeback. So this catches only the
659 if (inode->i_state & I_SYNC) {
660 /* Wait for I_SYNC. This function drops i_lock... */
661 inode_sleep_on_writeback(inode);
662 /* Inode may be gone, start again */
663 spin_lock(&wb->list_lock);
666 inode->i_state |= I_SYNC;
667 spin_unlock(&inode->i_lock);
669 write_chunk = writeback_chunk_size(wb->bdi, work);
670 wbc.nr_to_write = write_chunk;
671 wbc.pages_skipped = 0;
674 * We use I_SYNC to pin the inode in memory. While it is set
675 * evict_inode() will wait so the inode cannot be freed.
677 __writeback_single_inode(inode, wb, &wbc);
679 work->nr_pages -= write_chunk - wbc.nr_to_write;
680 wrote += write_chunk - wbc.nr_to_write;
681 spin_lock(&wb->list_lock);
682 spin_lock(&inode->i_lock);
683 if (!(inode->i_state & I_DIRTY))
685 requeue_inode(inode, wb, &wbc);
686 inode_sync_complete(inode);
687 spin_unlock(&inode->i_lock);
688 cond_resched_lock(&wb->list_lock);
690 * bail out to wb_writeback() often enough to check
691 * background threshold and other termination conditions.
694 if (time_is_before_jiffies(start_time + HZ / 10UL))
696 if (work->nr_pages <= 0)
703 static long __writeback_inodes_wb(struct bdi_writeback *wb,
704 struct wb_writeback_work *work)
706 unsigned long start_time = jiffies;
709 while (!list_empty(&wb->b_io)) {
710 struct inode *inode = wb_inode(wb->b_io.prev);
711 struct super_block *sb = inode->i_sb;
713 if (!grab_super_passive(sb)) {
715 * grab_super_passive() may fail consistently due to
716 * s_umount being grabbed by someone else. Don't use
717 * requeue_io() to avoid busy retrying the inode/sb.
719 redirty_tail(inode, wb);
722 wrote += writeback_sb_inodes(sb, wb, work);
725 /* refer to the same tests at the end of writeback_sb_inodes */
727 if (time_is_before_jiffies(start_time + HZ / 10UL))
729 if (work->nr_pages <= 0)
733 /* Leave any unwritten inodes on b_io */
737 long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages,
738 enum wb_reason reason)
740 struct wb_writeback_work work = {
741 .nr_pages = nr_pages,
742 .sync_mode = WB_SYNC_NONE,
747 spin_lock(&wb->list_lock);
748 if (list_empty(&wb->b_io))
750 __writeback_inodes_wb(wb, &work);
751 spin_unlock(&wb->list_lock);
753 return nr_pages - work.nr_pages;
756 static bool over_bground_thresh(struct backing_dev_info *bdi)
758 unsigned long background_thresh, dirty_thresh;
760 global_dirty_limits(&background_thresh, &dirty_thresh);
762 if (global_page_state(NR_FILE_DIRTY) +
763 global_page_state(NR_UNSTABLE_NFS) > background_thresh)
766 if (bdi_stat(bdi, BDI_RECLAIMABLE) >
767 bdi_dirty_limit(bdi, background_thresh))
774 * Called under wb->list_lock. If there are multiple wb per bdi,
775 * only the flusher working on the first wb should do it.
777 static void wb_update_bandwidth(struct bdi_writeback *wb,
778 unsigned long start_time)
780 __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time);
784 * Explicit flushing or periodic writeback of "old" data.
786 * Define "old": the first time one of an inode's pages is dirtied, we mark the
787 * dirtying-time in the inode's address_space. So this periodic writeback code
788 * just walks the superblock inode list, writing back any inodes which are
789 * older than a specific point in time.
791 * Try to run once per dirty_writeback_interval. But if a writeback event
792 * takes longer than a dirty_writeback_interval interval, then leave a
795 * older_than_this takes precedence over nr_to_write. So we'll only write back
796 * all dirty pages if they are all attached to "old" mappings.
798 static long wb_writeback(struct bdi_writeback *wb,
799 struct wb_writeback_work *work)
801 unsigned long wb_start = jiffies;
802 long nr_pages = work->nr_pages;
803 unsigned long oldest_jif;
807 oldest_jif = jiffies;
808 work->older_than_this = &oldest_jif;
810 spin_lock(&wb->list_lock);
813 * Stop writeback when nr_pages has been consumed
815 if (work->nr_pages <= 0)
819 * Background writeout and kupdate-style writeback may
820 * run forever. Stop them if there is other work to do
821 * so that e.g. sync can proceed. They'll be restarted
822 * after the other works are all done.
824 if ((work->for_background || work->for_kupdate) &&
825 !list_empty(&wb->bdi->work_list))
829 * For background writeout, stop when we are below the
830 * background dirty threshold
832 if (work->for_background && !over_bground_thresh(wb->bdi))
836 * Kupdate and background works are special and we want to
837 * include all inodes that need writing. Livelock avoidance is
838 * handled by these works yielding to any other work so we are
841 if (work->for_kupdate) {
842 oldest_jif = jiffies -
843 msecs_to_jiffies(dirty_expire_interval * 10);
844 } else if (work->for_background)
845 oldest_jif = jiffies;
847 trace_writeback_start(wb->bdi, work);
848 if (list_empty(&wb->b_io))
851 progress = writeback_sb_inodes(work->sb, wb, work);
853 progress = __writeback_inodes_wb(wb, work);
854 trace_writeback_written(wb->bdi, work);
856 wb_update_bandwidth(wb, wb_start);
859 * Did we write something? Try for more
861 * Dirty inodes are moved to b_io for writeback in batches.
862 * The completion of the current batch does not necessarily
863 * mean the overall work is done. So we keep looping as long
864 * as made some progress on cleaning pages or inodes.
869 * No more inodes for IO, bail
871 if (list_empty(&wb->b_more_io))
874 * Nothing written. Wait for some inode to
875 * become available for writeback. Otherwise
876 * we'll just busyloop.
878 if (!list_empty(&wb->b_more_io)) {
879 trace_writeback_wait(wb->bdi, work);
880 inode = wb_inode(wb->b_more_io.prev);
881 spin_lock(&inode->i_lock);
882 spin_unlock(&wb->list_lock);
883 /* This function drops i_lock... */
884 inode_sleep_on_writeback(inode);
885 spin_lock(&wb->list_lock);
888 spin_unlock(&wb->list_lock);
890 return nr_pages - work->nr_pages;
894 * Return the next wb_writeback_work struct that hasn't been processed yet.
896 static struct wb_writeback_work *
897 get_next_work_item(struct backing_dev_info *bdi)
899 struct wb_writeback_work *work = NULL;
901 spin_lock_bh(&bdi->wb_lock);
902 if (!list_empty(&bdi->work_list)) {
903 work = list_entry(bdi->work_list.next,
904 struct wb_writeback_work, list);
905 list_del_init(&work->list);
907 spin_unlock_bh(&bdi->wb_lock);
912 * Add in the number of potentially dirty inodes, because each inode
913 * write can dirty pagecache in the underlying blockdev.
915 static unsigned long get_nr_dirty_pages(void)
917 return global_page_state(NR_FILE_DIRTY) +
918 global_page_state(NR_UNSTABLE_NFS) +
919 get_nr_dirty_inodes();
922 static long wb_check_background_flush(struct bdi_writeback *wb)
924 if (over_bground_thresh(wb->bdi)) {
926 struct wb_writeback_work work = {
927 .nr_pages = LONG_MAX,
928 .sync_mode = WB_SYNC_NONE,
931 .reason = WB_REASON_BACKGROUND,
934 return wb_writeback(wb, &work);
940 static long wb_check_old_data_flush(struct bdi_writeback *wb)
942 unsigned long expired;
946 * When set to zero, disable periodic writeback
948 if (!dirty_writeback_interval)
951 expired = wb->last_old_flush +
952 msecs_to_jiffies(dirty_writeback_interval * 10);
953 if (time_before(jiffies, expired))
956 wb->last_old_flush = jiffies;
957 nr_pages = get_nr_dirty_pages();
960 struct wb_writeback_work work = {
961 .nr_pages = nr_pages,
962 .sync_mode = WB_SYNC_NONE,
965 .reason = WB_REASON_PERIODIC,
968 return wb_writeback(wb, &work);
975 * Retrieve work items and do the writeback they describe
977 long wb_do_writeback(struct bdi_writeback *wb, int force_wait)
979 struct backing_dev_info *bdi = wb->bdi;
980 struct wb_writeback_work *work;
983 set_bit(BDI_writeback_running, &wb->bdi->state);
984 while ((work = get_next_work_item(bdi)) != NULL) {
986 * Override sync mode, in case we must wait for completion
987 * because this thread is exiting now.
990 work->sync_mode = WB_SYNC_ALL;
992 trace_writeback_exec(bdi, work);
994 wrote += wb_writeback(wb, work);
997 * Notify the caller of completion if this is a synchronous
998 * work item, otherwise just free it.
1001 complete(work->done);
1007 * Check for periodic writeback, kupdated() style
1009 wrote += wb_check_old_data_flush(wb);
1010 wrote += wb_check_background_flush(wb);
1011 clear_bit(BDI_writeback_running, &wb->bdi->state);
1017 * Handle writeback of dirty data for the device backed by this bdi. Also
1018 * wakes up periodically and does kupdated style flushing.
1020 int bdi_writeback_thread(void *data)
1022 struct bdi_writeback *wb = data;
1023 struct backing_dev_info *bdi = wb->bdi;
1026 current->flags |= PF_SWAPWRITE;
1028 wb->last_active = jiffies;
1031 * Our parent may run at a different priority, just set us to normal
1033 set_user_nice(current, 0);
1035 trace_writeback_thread_start(bdi);
1037 while (!kthread_freezable_should_stop(NULL)) {
1039 * Remove own delayed wake-up timer, since we are already awake
1040 * and we'll take care of the preriodic write-back.
1042 del_timer(&wb->wakeup_timer);
1044 pages_written = wb_do_writeback(wb, 0);
1046 trace_writeback_pages_written(pages_written);
1049 wb->last_active = jiffies;
1051 set_current_state(TASK_INTERRUPTIBLE);
1052 if (!list_empty(&bdi->work_list) || kthread_should_stop()) {
1053 __set_current_state(TASK_RUNNING);
1057 if (wb_has_dirty_io(wb) && dirty_writeback_interval)
1058 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
1061 * We have nothing to do, so can go sleep without any
1062 * timeout and save power. When a work is queued or
1063 * something is made dirty - we will be woken up.
1069 /* Flush any work that raced with us exiting */
1070 if (!list_empty(&bdi->work_list))
1071 wb_do_writeback(wb, 1);
1073 trace_writeback_thread_stop(bdi);
1079 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1082 void wakeup_flusher_threads(long nr_pages, enum wb_reason reason)
1084 struct backing_dev_info *bdi;
1087 nr_pages = global_page_state(NR_FILE_DIRTY) +
1088 global_page_state(NR_UNSTABLE_NFS);
1092 list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) {
1093 if (!bdi_has_dirty_io(bdi))
1095 __bdi_start_writeback(bdi, nr_pages, false, reason);
1100 static noinline void block_dump___mark_inode_dirty(struct inode *inode)
1102 if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) {
1103 struct dentry *dentry;
1104 const char *name = "?";
1106 dentry = d_find_alias(inode);
1108 spin_lock(&dentry->d_lock);
1109 name = (const char *) dentry->d_name.name;
1112 "%s(%d): dirtied inode %lu (%s) on %s\n",
1113 current->comm, task_pid_nr(current), inode->i_ino,
1114 name, inode->i_sb->s_id);
1116 spin_unlock(&dentry->d_lock);
1123 * __mark_inode_dirty - internal function
1124 * @inode: inode to mark
1125 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
1126 * Mark an inode as dirty. Callers should use mark_inode_dirty or
1127 * mark_inode_dirty_sync.
1129 * Put the inode on the super block's dirty list.
1131 * CAREFUL! We mark it dirty unconditionally, but move it onto the
1132 * dirty list only if it is hashed or if it refers to a blockdev.
1133 * If it was not hashed, it will never be added to the dirty list
1134 * even if it is later hashed, as it will have been marked dirty already.
1136 * In short, make sure you hash any inodes _before_ you start marking
1139 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
1140 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
1141 * the kernel-internal blockdev inode represents the dirtying time of the
1142 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
1143 * page->mapping->host, so the page-dirtying time is recorded in the internal
1146 void __mark_inode_dirty(struct inode *inode, int flags)
1148 struct super_block *sb = inode->i_sb;
1149 struct backing_dev_info *bdi = NULL;
1152 * Don't do this for I_DIRTY_PAGES - that doesn't actually
1153 * dirty the inode itself
1155 if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) {
1156 if (sb->s_op->dirty_inode)
1157 sb->s_op->dirty_inode(inode, flags);
1161 * make sure that changes are seen by all cpus before we test i_state
1166 /* avoid the locking if we can */
1167 if ((inode->i_state & flags) == flags)
1170 if (unlikely(block_dump))
1171 block_dump___mark_inode_dirty(inode);
1173 spin_lock(&inode->i_lock);
1174 if ((inode->i_state & flags) != flags) {
1175 const int was_dirty = inode->i_state & I_DIRTY;
1177 inode->i_state |= flags;
1180 * If the inode is being synced, just update its dirty state.
1181 * The unlocker will place the inode on the appropriate
1182 * superblock list, based upon its state.
1184 if (inode->i_state & I_SYNC)
1185 goto out_unlock_inode;
1188 * Only add valid (hashed) inodes to the superblock's
1189 * dirty list. Add blockdev inodes as well.
1191 if (!S_ISBLK(inode->i_mode)) {
1192 if (inode_unhashed(inode))
1193 goto out_unlock_inode;
1195 if (inode->i_state & I_FREEING)
1196 goto out_unlock_inode;
1199 * If the inode was already on b_dirty/b_io/b_more_io, don't
1200 * reposition it (that would break b_dirty time-ordering).
1203 bool wakeup_bdi = false;
1204 bdi = inode_to_bdi(inode);
1206 if (bdi_cap_writeback_dirty(bdi)) {
1207 WARN(!test_bit(BDI_registered, &bdi->state),
1208 "bdi-%s not registered\n", bdi->name);
1211 * If this is the first dirty inode for this
1212 * bdi, we have to wake-up the corresponding
1213 * bdi thread to make sure background
1214 * write-back happens later.
1216 if (!wb_has_dirty_io(&bdi->wb))
1220 spin_unlock(&inode->i_lock);
1221 spin_lock(&bdi->wb.list_lock);
1222 inode->dirtied_when = jiffies;
1223 list_move(&inode->i_wb_list, &bdi->wb.b_dirty);
1224 spin_unlock(&bdi->wb.list_lock);
1227 bdi_wakeup_thread_delayed(bdi);
1232 spin_unlock(&inode->i_lock);
1235 EXPORT_SYMBOL(__mark_inode_dirty);
1237 static void wait_sb_inodes(struct super_block *sb)
1239 struct inode *inode, *old_inode = NULL;
1242 * We need to be protected against the filesystem going from
1243 * r/o to r/w or vice versa.
1245 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1247 spin_lock(&inode_sb_list_lock);
1250 * Data integrity sync. Must wait for all pages under writeback,
1251 * because there may have been pages dirtied before our sync
1252 * call, but which had writeout started before we write it out.
1253 * In which case, the inode may not be on the dirty list, but
1254 * we still have to wait for that writeout.
1256 list_for_each_entry(inode, &sb->s_inodes, i_sb_list) {
1257 struct address_space *mapping = inode->i_mapping;
1259 spin_lock(&inode->i_lock);
1260 if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) ||
1261 (mapping->nrpages == 0)) {
1262 spin_unlock(&inode->i_lock);
1266 spin_unlock(&inode->i_lock);
1267 spin_unlock(&inode_sb_list_lock);
1270 * We hold a reference to 'inode' so it couldn't have been
1271 * removed from s_inodes list while we dropped the
1272 * inode_sb_list_lock. We cannot iput the inode now as we can
1273 * be holding the last reference and we cannot iput it under
1274 * inode_sb_list_lock. So we keep the reference and iput it
1280 filemap_fdatawait(mapping);
1284 spin_lock(&inode_sb_list_lock);
1286 spin_unlock(&inode_sb_list_lock);
1291 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
1292 * @sb: the superblock
1293 * @nr: the number of pages to write
1294 * @reason: reason why some writeback work initiated
1296 * Start writeback on some inodes on this super_block. No guarantees are made
1297 * on how many (if any) will be written, and this function does not wait
1298 * for IO completion of submitted IO.
1300 void writeback_inodes_sb_nr(struct super_block *sb,
1302 enum wb_reason reason)
1304 DECLARE_COMPLETION_ONSTACK(done);
1305 struct wb_writeback_work work = {
1307 .sync_mode = WB_SYNC_NONE,
1308 .tagged_writepages = 1,
1314 if (sb->s_bdi == &noop_backing_dev_info)
1316 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1317 bdi_queue_work(sb->s_bdi, &work);
1318 wait_for_completion(&done);
1320 EXPORT_SYMBOL(writeback_inodes_sb_nr);
1323 * writeback_inodes_sb - writeback dirty inodes from given super_block
1324 * @sb: the superblock
1325 * @reason: reason why some writeback work was initiated
1327 * Start writeback on some inodes on this super_block. No guarantees are made
1328 * on how many (if any) will be written, and this function does not wait
1329 * for IO completion of submitted IO.
1331 void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason)
1333 return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason);
1335 EXPORT_SYMBOL(writeback_inodes_sb);
1338 * writeback_inodes_sb_if_idle - start writeback if none underway
1339 * @sb: the superblock
1340 * @reason: reason why some writeback work was initiated
1342 * Invoke writeback_inodes_sb if no writeback is currently underway.
1343 * Returns 1 if writeback was started, 0 if not.
1345 int writeback_inodes_sb_if_idle(struct super_block *sb, enum wb_reason reason)
1347 if (!writeback_in_progress(sb->s_bdi)) {
1348 down_read(&sb->s_umount);
1349 writeback_inodes_sb(sb, reason);
1350 up_read(&sb->s_umount);
1355 EXPORT_SYMBOL(writeback_inodes_sb_if_idle);
1358 * writeback_inodes_sb_nr_if_idle - start writeback if none underway
1359 * @sb: the superblock
1360 * @nr: the number of pages to write
1361 * @reason: reason why some writeback work was initiated
1363 * Invoke writeback_inodes_sb if no writeback is currently underway.
1364 * Returns 1 if writeback was started, 0 if not.
1366 int writeback_inodes_sb_nr_if_idle(struct super_block *sb,
1368 enum wb_reason reason)
1370 if (!writeback_in_progress(sb->s_bdi)) {
1371 down_read(&sb->s_umount);
1372 writeback_inodes_sb_nr(sb, nr, reason);
1373 up_read(&sb->s_umount);
1378 EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle);
1381 * sync_inodes_sb - sync sb inode pages
1382 * @sb: the superblock
1384 * This function writes and waits on any dirty inode belonging to this
1387 void sync_inodes_sb(struct super_block *sb)
1389 DECLARE_COMPLETION_ONSTACK(done);
1390 struct wb_writeback_work work = {
1392 .sync_mode = WB_SYNC_ALL,
1393 .nr_pages = LONG_MAX,
1396 .reason = WB_REASON_SYNC,
1399 /* Nothing to do? */
1400 if (sb->s_bdi == &noop_backing_dev_info)
1402 WARN_ON(!rwsem_is_locked(&sb->s_umount));
1404 bdi_queue_work(sb->s_bdi, &work);
1405 wait_for_completion(&done);
1409 EXPORT_SYMBOL(sync_inodes_sb);
1412 * write_inode_now - write an inode to disk
1413 * @inode: inode to write to disk
1414 * @sync: whether the write should be synchronous or not
1416 * This function commits an inode to disk immediately if it is dirty. This is
1417 * primarily needed by knfsd.
1419 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
1421 int write_inode_now(struct inode *inode, int sync)
1423 struct bdi_writeback *wb = &inode_to_bdi(inode)->wb;
1424 struct writeback_control wbc = {
1425 .nr_to_write = LONG_MAX,
1426 .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE,
1428 .range_end = LLONG_MAX,
1431 if (!mapping_cap_writeback_dirty(inode->i_mapping))
1432 wbc.nr_to_write = 0;
1435 return writeback_single_inode(inode, wb, &wbc);
1437 EXPORT_SYMBOL(write_inode_now);
1440 * sync_inode - write an inode and its pages to disk.
1441 * @inode: the inode to sync
1442 * @wbc: controls the writeback mode
1444 * sync_inode() will write an inode and its pages to disk. It will also
1445 * correctly update the inode on its superblock's dirty inode lists and will
1446 * update inode->i_state.
1448 * The caller must have a ref on the inode.
1450 int sync_inode(struct inode *inode, struct writeback_control *wbc)
1452 return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc);
1454 EXPORT_SYMBOL(sync_inode);
1457 * sync_inode_metadata - write an inode to disk
1458 * @inode: the inode to sync
1459 * @wait: wait for I/O to complete.
1461 * Write an inode to disk and adjust its dirty state after completion.
1463 * Note: only writes the actual inode, no associated data or other metadata.
1465 int sync_inode_metadata(struct inode *inode, int wait)
1467 struct writeback_control wbc = {
1468 .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE,
1469 .nr_to_write = 0, /* metadata-only */
1472 return sync_inode(inode, &wbc);
1474 EXPORT_SYMBOL(sync_inode_metadata);