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/writeback.h>
26 #include <linux/blkdev.h>
27 #include <linux/backing-dev.h>
28 #include <linux/tracepoint.h>
29 #include <linux/device.h>
30 #include <linux/memcontrol.h>
34 * 4MB minimal write chunk size
36 #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_SHIFT - 10))
38 struct wb_completion
{
43 * Passed into wb_writeback(), essentially a subset of writeback_control
45 struct wb_writeback_work
{
47 struct super_block
*sb
;
48 unsigned long *older_than_this
;
49 enum writeback_sync_modes sync_mode
;
50 unsigned int tagged_writepages
:1;
51 unsigned int for_kupdate
:1;
52 unsigned int range_cyclic
:1;
53 unsigned int for_background
:1;
54 unsigned int for_sync
:1; /* sync(2) WB_SYNC_ALL writeback */
55 unsigned int auto_free
:1; /* free on completion */
56 enum wb_reason reason
; /* why was writeback initiated? */
58 struct list_head list
; /* pending work list */
59 struct wb_completion
*done
; /* set if the caller waits */
63 * If one wants to wait for one or more wb_writeback_works, each work's
64 * ->done should be set to a wb_completion defined using the following
65 * macro. Once all work items are issued with wb_queue_work(), the caller
66 * can wait for the completion of all using wb_wait_for_completion(). Work
67 * items which are waited upon aren't freed automatically on completion.
69 #define DEFINE_WB_COMPLETION_ONSTACK(cmpl) \
70 struct wb_completion cmpl = { \
71 .cnt = ATOMIC_INIT(1), \
76 * If an inode is constantly having its pages dirtied, but then the
77 * updates stop dirtytime_expire_interval seconds in the past, it's
78 * possible for the worst case time between when an inode has its
79 * timestamps updated and when they finally get written out to be two
80 * dirtytime_expire_intervals. We set the default to 12 hours (in
81 * seconds), which means most of the time inodes will have their
82 * timestamps written to disk after 12 hours, but in the worst case a
83 * few inodes might not their timestamps updated for 24 hours.
85 unsigned int dirtytime_expire_interval
= 12 * 60 * 60;
87 static inline struct inode
*wb_inode(struct list_head
*head
)
89 return list_entry(head
, struct inode
, i_io_list
);
93 * Include the creation of the trace points after defining the
94 * wb_writeback_work structure and inline functions so that the definition
95 * remains local to this file.
97 #define CREATE_TRACE_POINTS
98 #include <trace/events/writeback.h>
100 EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage
);
102 static bool wb_io_lists_populated(struct bdi_writeback
*wb
)
104 if (wb_has_dirty_io(wb
)) {
107 set_bit(WB_has_dirty_io
, &wb
->state
);
108 WARN_ON_ONCE(!wb
->avg_write_bandwidth
);
109 atomic_long_add(wb
->avg_write_bandwidth
,
110 &wb
->bdi
->tot_write_bandwidth
);
115 static void wb_io_lists_depopulated(struct bdi_writeback
*wb
)
117 if (wb_has_dirty_io(wb
) && list_empty(&wb
->b_dirty
) &&
118 list_empty(&wb
->b_io
) && list_empty(&wb
->b_more_io
)) {
119 clear_bit(WB_has_dirty_io
, &wb
->state
);
120 WARN_ON_ONCE(atomic_long_sub_return(wb
->avg_write_bandwidth
,
121 &wb
->bdi
->tot_write_bandwidth
) < 0);
126 * inode_io_list_move_locked - move an inode onto a bdi_writeback IO list
127 * @inode: inode to be moved
128 * @wb: target bdi_writeback
129 * @head: one of @wb->b_{dirty|io|more_io}
131 * Move @inode->i_io_list to @list of @wb and set %WB_has_dirty_io.
132 * Returns %true if @inode is the first occupant of the !dirty_time IO
133 * lists; otherwise, %false.
135 static bool inode_io_list_move_locked(struct inode
*inode
,
136 struct bdi_writeback
*wb
,
137 struct list_head
*head
)
139 assert_spin_locked(&wb
->list_lock
);
141 list_move(&inode
->i_io_list
, head
);
143 /* dirty_time doesn't count as dirty_io until expiration */
144 if (head
!= &wb
->b_dirty_time
)
145 return wb_io_lists_populated(wb
);
147 wb_io_lists_depopulated(wb
);
152 * inode_io_list_del_locked - remove an inode from its bdi_writeback IO list
153 * @inode: inode to be removed
154 * @wb: bdi_writeback @inode is being removed from
156 * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and
157 * clear %WB_has_dirty_io if all are empty afterwards.
159 static void inode_io_list_del_locked(struct inode
*inode
,
160 struct bdi_writeback
*wb
)
162 assert_spin_locked(&wb
->list_lock
);
164 list_del_init(&inode
->i_io_list
);
165 wb_io_lists_depopulated(wb
);
168 static void wb_wakeup(struct bdi_writeback
*wb
)
170 spin_lock_bh(&wb
->work_lock
);
171 if (test_bit(WB_registered
, &wb
->state
))
172 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
173 spin_unlock_bh(&wb
->work_lock
);
176 static void wb_queue_work(struct bdi_writeback
*wb
,
177 struct wb_writeback_work
*work
)
179 trace_writeback_queue(wb
, work
);
181 spin_lock_bh(&wb
->work_lock
);
182 if (!test_bit(WB_registered
, &wb
->state
))
185 atomic_inc(&work
->done
->cnt
);
186 list_add_tail(&work
->list
, &wb
->work_list
);
187 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
189 spin_unlock_bh(&wb
->work_lock
);
193 * wb_wait_for_completion - wait for completion of bdi_writeback_works
194 * @bdi: bdi work items were issued to
195 * @done: target wb_completion
197 * Wait for one or more work items issued to @bdi with their ->done field
198 * set to @done, which should have been defined with
199 * DEFINE_WB_COMPLETION_ONSTACK(). This function returns after all such
200 * work items are completed. Work items which are waited upon aren't freed
201 * automatically on completion.
203 static void wb_wait_for_completion(struct backing_dev_info
*bdi
,
204 struct wb_completion
*done
)
206 atomic_dec(&done
->cnt
); /* put down the initial count */
207 wait_event(bdi
->wb_waitq
, !atomic_read(&done
->cnt
));
210 #ifdef CONFIG_CGROUP_WRITEBACK
212 /* parameters for foreign inode detection, see wb_detach_inode() */
213 #define WB_FRN_TIME_SHIFT 13 /* 1s = 2^13, upto 8 secs w/ 16bit */
214 #define WB_FRN_TIME_AVG_SHIFT 3 /* avg = avg * 7/8 + new * 1/8 */
215 #define WB_FRN_TIME_CUT_DIV 2 /* ignore rounds < avg / 2 */
216 #define WB_FRN_TIME_PERIOD (2 * (1 << WB_FRN_TIME_SHIFT)) /* 2s */
218 #define WB_FRN_HIST_SLOTS 16 /* inode->i_wb_frn_history is 16bit */
219 #define WB_FRN_HIST_UNIT (WB_FRN_TIME_PERIOD / WB_FRN_HIST_SLOTS)
220 /* each slot's duration is 2s / 16 */
221 #define WB_FRN_HIST_THR_SLOTS (WB_FRN_HIST_SLOTS / 2)
222 /* if foreign slots >= 8, switch */
223 #define WB_FRN_HIST_MAX_SLOTS (WB_FRN_HIST_THR_SLOTS / 2 + 1)
224 /* one round can affect upto 5 slots */
226 static atomic_t isw_nr_in_flight
= ATOMIC_INIT(0);
227 static struct workqueue_struct
*isw_wq
;
229 void __inode_attach_wb(struct inode
*inode
, struct page
*page
)
231 struct backing_dev_info
*bdi
= inode_to_bdi(inode
);
232 struct bdi_writeback
*wb
= NULL
;
234 if (inode_cgwb_enabled(inode
)) {
235 struct cgroup_subsys_state
*memcg_css
;
238 memcg_css
= mem_cgroup_css_from_page(page
);
239 wb
= wb_get_create(bdi
, memcg_css
, GFP_ATOMIC
);
241 /* must pin memcg_css, see wb_get_create() */
242 memcg_css
= task_get_css(current
, memory_cgrp_id
);
243 wb
= wb_get_create(bdi
, memcg_css
, GFP_ATOMIC
);
252 * There may be multiple instances of this function racing to
253 * update the same inode. Use cmpxchg() to tell the winner.
255 if (unlikely(cmpxchg(&inode
->i_wb
, NULL
, wb
)))
260 * locked_inode_to_wb_and_lock_list - determine a locked inode's wb and lock it
261 * @inode: inode of interest with i_lock held
263 * Returns @inode's wb with its list_lock held. @inode->i_lock must be
264 * held on entry and is released on return. The returned wb is guaranteed
265 * to stay @inode's associated wb until its list_lock is released.
267 static struct bdi_writeback
*
268 locked_inode_to_wb_and_lock_list(struct inode
*inode
)
269 __releases(&inode
->i_lock
)
270 __acquires(&wb
->list_lock
)
273 struct bdi_writeback
*wb
= inode_to_wb(inode
);
276 * inode_to_wb() association is protected by both
277 * @inode->i_lock and @wb->list_lock but list_lock nests
278 * outside i_lock. Drop i_lock and verify that the
279 * association hasn't changed after acquiring list_lock.
282 spin_unlock(&inode
->i_lock
);
283 spin_lock(&wb
->list_lock
);
285 /* i_wb may have changed inbetween, can't use inode_to_wb() */
286 if (likely(wb
== inode
->i_wb
)) {
287 wb_put(wb
); /* @inode already has ref */
291 spin_unlock(&wb
->list_lock
);
294 spin_lock(&inode
->i_lock
);
299 * inode_to_wb_and_lock_list - determine an inode's wb and lock it
300 * @inode: inode of interest
302 * Same as locked_inode_to_wb_and_lock_list() but @inode->i_lock isn't held
305 static struct bdi_writeback
*inode_to_wb_and_lock_list(struct inode
*inode
)
306 __acquires(&wb
->list_lock
)
308 spin_lock(&inode
->i_lock
);
309 return locked_inode_to_wb_and_lock_list(inode
);
312 struct inode_switch_wbs_context
{
314 struct bdi_writeback
*new_wb
;
316 struct rcu_head rcu_head
;
317 struct work_struct work
;
320 static void inode_switch_wbs_work_fn(struct work_struct
*work
)
322 struct inode_switch_wbs_context
*isw
=
323 container_of(work
, struct inode_switch_wbs_context
, work
);
324 struct inode
*inode
= isw
->inode
;
325 struct address_space
*mapping
= inode
->i_mapping
;
326 struct bdi_writeback
*old_wb
= inode
->i_wb
;
327 struct bdi_writeback
*new_wb
= isw
->new_wb
;
328 struct radix_tree_iter iter
;
329 bool switched
= false;
333 * By the time control reaches here, RCU grace period has passed
334 * since I_WB_SWITCH assertion and all wb stat update transactions
335 * between unlocked_inode_to_wb_begin/end() are guaranteed to be
336 * synchronizing against mapping->tree_lock.
338 * Grabbing old_wb->list_lock, inode->i_lock and mapping->tree_lock
339 * gives us exclusion against all wb related operations on @inode
340 * including IO list manipulations and stat updates.
342 if (old_wb
< new_wb
) {
343 spin_lock(&old_wb
->list_lock
);
344 spin_lock_nested(&new_wb
->list_lock
, SINGLE_DEPTH_NESTING
);
346 spin_lock(&new_wb
->list_lock
);
347 spin_lock_nested(&old_wb
->list_lock
, SINGLE_DEPTH_NESTING
);
349 spin_lock(&inode
->i_lock
);
350 spin_lock_irq(&mapping
->tree_lock
);
353 * Once I_FREEING is visible under i_lock, the eviction path owns
354 * the inode and we shouldn't modify ->i_io_list.
356 if (unlikely(inode
->i_state
& I_FREEING
))
360 * Count and transfer stats. Note that PAGECACHE_TAG_DIRTY points
361 * to possibly dirty pages while PAGECACHE_TAG_WRITEBACK points to
362 * pages actually under underwriteback.
364 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
, 0,
365 PAGECACHE_TAG_DIRTY
) {
366 struct page
*page
= radix_tree_deref_slot_protected(slot
,
367 &mapping
->tree_lock
);
368 if (likely(page
) && PageDirty(page
)) {
369 __dec_wb_stat(old_wb
, WB_RECLAIMABLE
);
370 __inc_wb_stat(new_wb
, WB_RECLAIMABLE
);
374 radix_tree_for_each_tagged(slot
, &mapping
->page_tree
, &iter
, 0,
375 PAGECACHE_TAG_WRITEBACK
) {
376 struct page
*page
= radix_tree_deref_slot_protected(slot
,
377 &mapping
->tree_lock
);
379 WARN_ON_ONCE(!PageWriteback(page
));
380 __dec_wb_stat(old_wb
, WB_WRITEBACK
);
381 __inc_wb_stat(new_wb
, WB_WRITEBACK
);
388 * Transfer to @new_wb's IO list if necessary. The specific list
389 * @inode was on is ignored and the inode is put on ->b_dirty which
390 * is always correct including from ->b_dirty_time. The transfer
391 * preserves @inode->dirtied_when ordering.
393 if (!list_empty(&inode
->i_io_list
)) {
396 inode_io_list_del_locked(inode
, old_wb
);
397 inode
->i_wb
= new_wb
;
398 list_for_each_entry(pos
, &new_wb
->b_dirty
, i_io_list
)
399 if (time_after_eq(inode
->dirtied_when
,
402 inode_io_list_move_locked(inode
, new_wb
, pos
->i_io_list
.prev
);
404 inode
->i_wb
= new_wb
;
407 /* ->i_wb_frn updates may race wbc_detach_inode() but doesn't matter */
408 inode
->i_wb_frn_winner
= 0;
409 inode
->i_wb_frn_avg_time
= 0;
410 inode
->i_wb_frn_history
= 0;
414 * Paired with load_acquire in unlocked_inode_to_wb_begin() and
415 * ensures that the new wb is visible if they see !I_WB_SWITCH.
417 smp_store_release(&inode
->i_state
, inode
->i_state
& ~I_WB_SWITCH
);
419 spin_unlock_irq(&mapping
->tree_lock
);
420 spin_unlock(&inode
->i_lock
);
421 spin_unlock(&new_wb
->list_lock
);
422 spin_unlock(&old_wb
->list_lock
);
433 atomic_dec(&isw_nr_in_flight
);
436 static void inode_switch_wbs_rcu_fn(struct rcu_head
*rcu_head
)
438 struct inode_switch_wbs_context
*isw
= container_of(rcu_head
,
439 struct inode_switch_wbs_context
, rcu_head
);
441 /* needs to grab bh-unsafe locks, bounce to work item */
442 INIT_WORK(&isw
->work
, inode_switch_wbs_work_fn
);
443 queue_work(isw_wq
, &isw
->work
);
447 * inode_switch_wbs - change the wb association of an inode
448 * @inode: target inode
449 * @new_wb_id: ID of the new wb
451 * Switch @inode's wb association to the wb identified by @new_wb_id. The
452 * switching is performed asynchronously and may fail silently.
454 static void inode_switch_wbs(struct inode
*inode
, int new_wb_id
)
456 struct backing_dev_info
*bdi
= inode_to_bdi(inode
);
457 struct cgroup_subsys_state
*memcg_css
;
458 struct inode_switch_wbs_context
*isw
;
460 /* noop if seems to be already in progress */
461 if (inode
->i_state
& I_WB_SWITCH
)
464 isw
= kzalloc(sizeof(*isw
), GFP_ATOMIC
);
468 /* find and pin the new wb */
470 memcg_css
= css_from_id(new_wb_id
, &memory_cgrp_subsys
);
472 isw
->new_wb
= wb_get_create(bdi
, memcg_css
, GFP_ATOMIC
);
477 /* while holding I_WB_SWITCH, no one else can update the association */
478 spin_lock(&inode
->i_lock
);
479 if (!(inode
->i_sb
->s_flags
& MS_ACTIVE
) ||
480 inode
->i_state
& (I_WB_SWITCH
| I_FREEING
) ||
481 inode_to_wb(inode
) == isw
->new_wb
) {
482 spin_unlock(&inode
->i_lock
);
485 inode
->i_state
|= I_WB_SWITCH
;
486 spin_unlock(&inode
->i_lock
);
491 atomic_inc(&isw_nr_in_flight
);
494 * In addition to synchronizing among switchers, I_WB_SWITCH tells
495 * the RCU protected stat update paths to grab the mapping's
496 * tree_lock so that stat transfer can synchronize against them.
497 * Let's continue after I_WB_SWITCH is guaranteed to be visible.
499 call_rcu(&isw
->rcu_head
, inode_switch_wbs_rcu_fn
);
509 * wbc_attach_and_unlock_inode - associate wbc with target inode and unlock it
510 * @wbc: writeback_control of interest
511 * @inode: target inode
513 * @inode is locked and about to be written back under the control of @wbc.
514 * Record @inode's writeback context into @wbc and unlock the i_lock. On
515 * writeback completion, wbc_detach_inode() should be called. This is used
516 * to track the cgroup writeback context.
518 void wbc_attach_and_unlock_inode(struct writeback_control
*wbc
,
521 if (!inode_cgwb_enabled(inode
)) {
522 spin_unlock(&inode
->i_lock
);
526 wbc
->wb
= inode_to_wb(inode
);
529 wbc
->wb_id
= wbc
->wb
->memcg_css
->id
;
530 wbc
->wb_lcand_id
= inode
->i_wb_frn_winner
;
531 wbc
->wb_tcand_id
= 0;
533 wbc
->wb_lcand_bytes
= 0;
534 wbc
->wb_tcand_bytes
= 0;
537 spin_unlock(&inode
->i_lock
);
540 * A dying wb indicates that the memcg-blkcg mapping has changed
541 * and a new wb is already serving the memcg. Switch immediately.
543 if (unlikely(wb_dying(wbc
->wb
)))
544 inode_switch_wbs(inode
, wbc
->wb_id
);
548 * wbc_detach_inode - disassociate wbc from inode and perform foreign detection
549 * @wbc: writeback_control of the just finished writeback
551 * To be called after a writeback attempt of an inode finishes and undoes
552 * wbc_attach_and_unlock_inode(). Can be called under any context.
554 * As concurrent write sharing of an inode is expected to be very rare and
555 * memcg only tracks page ownership on first-use basis severely confining
556 * the usefulness of such sharing, cgroup writeback tracks ownership
557 * per-inode. While the support for concurrent write sharing of an inode
558 * is deemed unnecessary, an inode being written to by different cgroups at
559 * different points in time is a lot more common, and, more importantly,
560 * charging only by first-use can too readily lead to grossly incorrect
561 * behaviors (single foreign page can lead to gigabytes of writeback to be
562 * incorrectly attributed).
564 * To resolve this issue, cgroup writeback detects the majority dirtier of
565 * an inode and transfers the ownership to it. To avoid unnnecessary
566 * oscillation, the detection mechanism keeps track of history and gives
567 * out the switch verdict only if the foreign usage pattern is stable over
568 * a certain amount of time and/or writeback attempts.
570 * On each writeback attempt, @wbc tries to detect the majority writer
571 * using Boyer-Moore majority vote algorithm. In addition to the byte
572 * count from the majority voting, it also counts the bytes written for the
573 * current wb and the last round's winner wb (max of last round's current
574 * wb, the winner from two rounds ago, and the last round's majority
575 * candidate). Keeping track of the historical winner helps the algorithm
576 * to semi-reliably detect the most active writer even when it's not the
579 * Once the winner of the round is determined, whether the winner is
580 * foreign or not and how much IO time the round consumed is recorded in
581 * inode->i_wb_frn_history. If the amount of recorded foreign IO time is
582 * over a certain threshold, the switch verdict is given.
584 void wbc_detach_inode(struct writeback_control
*wbc
)
586 struct bdi_writeback
*wb
= wbc
->wb
;
587 struct inode
*inode
= wbc
->inode
;
588 unsigned long avg_time
, max_bytes
, max_time
;
595 history
= inode
->i_wb_frn_history
;
596 avg_time
= inode
->i_wb_frn_avg_time
;
598 /* pick the winner of this round */
599 if (wbc
->wb_bytes
>= wbc
->wb_lcand_bytes
&&
600 wbc
->wb_bytes
>= wbc
->wb_tcand_bytes
) {
602 max_bytes
= wbc
->wb_bytes
;
603 } else if (wbc
->wb_lcand_bytes
>= wbc
->wb_tcand_bytes
) {
604 max_id
= wbc
->wb_lcand_id
;
605 max_bytes
= wbc
->wb_lcand_bytes
;
607 max_id
= wbc
->wb_tcand_id
;
608 max_bytes
= wbc
->wb_tcand_bytes
;
612 * Calculate the amount of IO time the winner consumed and fold it
613 * into the running average kept per inode. If the consumed IO
614 * time is lower than avag / WB_FRN_TIME_CUT_DIV, ignore it for
615 * deciding whether to switch or not. This is to prevent one-off
616 * small dirtiers from skewing the verdict.
618 max_time
= DIV_ROUND_UP((max_bytes
>> PAGE_SHIFT
) << WB_FRN_TIME_SHIFT
,
619 wb
->avg_write_bandwidth
);
621 avg_time
+= (max_time
>> WB_FRN_TIME_AVG_SHIFT
) -
622 (avg_time
>> WB_FRN_TIME_AVG_SHIFT
);
624 avg_time
= max_time
; /* immediate catch up on first run */
626 if (max_time
>= avg_time
/ WB_FRN_TIME_CUT_DIV
) {
630 * The switch verdict is reached if foreign wb's consume
631 * more than a certain proportion of IO time in a
632 * WB_FRN_TIME_PERIOD. This is loosely tracked by 16 slot
633 * history mask where each bit represents one sixteenth of
634 * the period. Determine the number of slots to shift into
635 * history from @max_time.
637 slots
= min(DIV_ROUND_UP(max_time
, WB_FRN_HIST_UNIT
),
638 (unsigned long)WB_FRN_HIST_MAX_SLOTS
);
640 if (wbc
->wb_id
!= max_id
)
641 history
|= (1U << slots
) - 1;
644 * Switch if the current wb isn't the consistent winner.
645 * If there are multiple closely competing dirtiers, the
646 * inode may switch across them repeatedly over time, which
647 * is okay. The main goal is avoiding keeping an inode on
648 * the wrong wb for an extended period of time.
650 if (hweight32(history
) > WB_FRN_HIST_THR_SLOTS
)
651 inode_switch_wbs(inode
, max_id
);
655 * Multiple instances of this function may race to update the
656 * following fields but we don't mind occassional inaccuracies.
658 inode
->i_wb_frn_winner
= max_id
;
659 inode
->i_wb_frn_avg_time
= min(avg_time
, (unsigned long)U16_MAX
);
660 inode
->i_wb_frn_history
= history
;
667 * wbc_account_io - account IO issued during writeback
668 * @wbc: writeback_control of the writeback in progress
669 * @page: page being written out
670 * @bytes: number of bytes being written out
672 * @bytes from @page are about to written out during the writeback
673 * controlled by @wbc. Keep the book for foreign inode detection. See
674 * wbc_detach_inode().
676 void wbc_account_io(struct writeback_control
*wbc
, struct page
*page
,
682 * pageout() path doesn't attach @wbc to the inode being written
683 * out. This is intentional as we don't want the function to block
684 * behind a slow cgroup. Ultimately, we want pageout() to kick off
685 * regular writeback instead of writing things out itself.
690 id
= mem_cgroup_css_from_page(page
)->id
;
692 if (id
== wbc
->wb_id
) {
693 wbc
->wb_bytes
+= bytes
;
697 if (id
== wbc
->wb_lcand_id
)
698 wbc
->wb_lcand_bytes
+= bytes
;
700 /* Boyer-Moore majority vote algorithm */
701 if (!wbc
->wb_tcand_bytes
)
702 wbc
->wb_tcand_id
= id
;
703 if (id
== wbc
->wb_tcand_id
)
704 wbc
->wb_tcand_bytes
+= bytes
;
706 wbc
->wb_tcand_bytes
-= min(bytes
, wbc
->wb_tcand_bytes
);
708 EXPORT_SYMBOL_GPL(wbc_account_io
);
711 * inode_congested - test whether an inode is congested
712 * @inode: inode to test for congestion (may be NULL)
713 * @cong_bits: mask of WB_[a]sync_congested bits to test
715 * Tests whether @inode is congested. @cong_bits is the mask of congestion
716 * bits to test and the return value is the mask of set bits.
718 * If cgroup writeback is enabled for @inode, the congestion state is
719 * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg
720 * associated with @inode is congested; otherwise, the root wb's congestion
723 * @inode is allowed to be NULL as this function is often called on
724 * mapping->host which is NULL for the swapper space.
726 int inode_congested(struct inode
*inode
, int cong_bits
)
729 * Once set, ->i_wb never becomes NULL while the inode is alive.
730 * Start transaction iff ->i_wb is visible.
732 if (inode
&& inode_to_wb_is_valid(inode
)) {
733 struct bdi_writeback
*wb
;
734 bool locked
, congested
;
736 wb
= unlocked_inode_to_wb_begin(inode
, &locked
);
737 congested
= wb_congested(wb
, cong_bits
);
738 unlocked_inode_to_wb_end(inode
, locked
);
742 return wb_congested(&inode_to_bdi(inode
)->wb
, cong_bits
);
744 EXPORT_SYMBOL_GPL(inode_congested
);
747 * wb_split_bdi_pages - split nr_pages to write according to bandwidth
748 * @wb: target bdi_writeback to split @nr_pages to
749 * @nr_pages: number of pages to write for the whole bdi
751 * Split @wb's portion of @nr_pages according to @wb's write bandwidth in
752 * relation to the total write bandwidth of all wb's w/ dirty inodes on
755 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
757 unsigned long this_bw
= wb
->avg_write_bandwidth
;
758 unsigned long tot_bw
= atomic_long_read(&wb
->bdi
->tot_write_bandwidth
);
760 if (nr_pages
== LONG_MAX
)
764 * This may be called on clean wb's and proportional distribution
765 * may not make sense, just use the original @nr_pages in those
766 * cases. In general, we wanna err on the side of writing more.
768 if (!tot_bw
|| this_bw
>= tot_bw
)
771 return DIV_ROUND_UP_ULL((u64
)nr_pages
* this_bw
, tot_bw
);
775 * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi
776 * @bdi: target backing_dev_info
777 * @base_work: wb_writeback_work to issue
778 * @skip_if_busy: skip wb's which already have writeback in progress
780 * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which
781 * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's
782 * distributed to the busy wbs according to each wb's proportion in the
783 * total active write bandwidth of @bdi.
785 static void bdi_split_work_to_wbs(struct backing_dev_info
*bdi
,
786 struct wb_writeback_work
*base_work
,
789 struct bdi_writeback
*last_wb
= NULL
;
790 struct bdi_writeback
*wb
= list_entry(&bdi
->wb_list
,
791 struct bdi_writeback
, bdi_node
);
796 list_for_each_entry_continue_rcu(wb
, &bdi
->wb_list
, bdi_node
) {
797 DEFINE_WB_COMPLETION_ONSTACK(fallback_work_done
);
798 struct wb_writeback_work fallback_work
;
799 struct wb_writeback_work
*work
;
807 /* SYNC_ALL writes out I_DIRTY_TIME too */
808 if (!wb_has_dirty_io(wb
) &&
809 (base_work
->sync_mode
== WB_SYNC_NONE
||
810 list_empty(&wb
->b_dirty_time
)))
812 if (skip_if_busy
&& writeback_in_progress(wb
))
815 nr_pages
= wb_split_bdi_pages(wb
, base_work
->nr_pages
);
817 work
= kmalloc(sizeof(*work
), GFP_ATOMIC
);
820 work
->nr_pages
= nr_pages
;
822 wb_queue_work(wb
, work
);
826 /* alloc failed, execute synchronously using on-stack fallback */
827 work
= &fallback_work
;
829 work
->nr_pages
= nr_pages
;
831 work
->done
= &fallback_work_done
;
833 wb_queue_work(wb
, work
);
836 * Pin @wb so that it stays on @bdi->wb_list. This allows
837 * continuing iteration from @wb after dropping and
838 * regrabbing rcu read lock.
844 wb_wait_for_completion(bdi
, &fallback_work_done
);
854 * cgroup_writeback_umount - flush inode wb switches for umount
856 * This function is called when a super_block is about to be destroyed and
857 * flushes in-flight inode wb switches. An inode wb switch goes through
858 * RCU and then workqueue, so the two need to be flushed in order to ensure
859 * that all previously scheduled switches are finished. As wb switches are
860 * rare occurrences and synchronize_rcu() can take a while, perform
861 * flushing iff wb switches are in flight.
863 void cgroup_writeback_umount(void)
865 if (atomic_read(&isw_nr_in_flight
)) {
867 flush_workqueue(isw_wq
);
871 static int __init
cgroup_writeback_init(void)
873 isw_wq
= alloc_workqueue("inode_switch_wbs", 0, 0);
878 fs_initcall(cgroup_writeback_init
);
880 #else /* CONFIG_CGROUP_WRITEBACK */
882 static struct bdi_writeback
*
883 locked_inode_to_wb_and_lock_list(struct inode
*inode
)
884 __releases(&inode
->i_lock
)
885 __acquires(&wb
->list_lock
)
887 struct bdi_writeback
*wb
= inode_to_wb(inode
);
889 spin_unlock(&inode
->i_lock
);
890 spin_lock(&wb
->list_lock
);
894 static struct bdi_writeback
*inode_to_wb_and_lock_list(struct inode
*inode
)
895 __acquires(&wb
->list_lock
)
897 struct bdi_writeback
*wb
= inode_to_wb(inode
);
899 spin_lock(&wb
->list_lock
);
903 static long wb_split_bdi_pages(struct bdi_writeback
*wb
, long nr_pages
)
908 static void bdi_split_work_to_wbs(struct backing_dev_info
*bdi
,
909 struct wb_writeback_work
*base_work
,
914 if (!skip_if_busy
|| !writeback_in_progress(&bdi
->wb
)) {
915 base_work
->auto_free
= 0;
916 wb_queue_work(&bdi
->wb
, base_work
);
920 #endif /* CONFIG_CGROUP_WRITEBACK */
922 void wb_start_writeback(struct bdi_writeback
*wb
, long nr_pages
,
923 bool range_cyclic
, enum wb_reason reason
)
925 struct wb_writeback_work
*work
;
927 if (!wb_has_dirty_io(wb
))
931 * This is WB_SYNC_NONE writeback, so if allocation fails just
932 * wakeup the thread for old dirty data writeback
934 work
= kzalloc(sizeof(*work
),
935 GFP_NOWAIT
| __GFP_NOMEMALLOC
| __GFP_NOWARN
);
937 trace_writeback_nowork(wb
);
942 work
->sync_mode
= WB_SYNC_NONE
;
943 work
->nr_pages
= nr_pages
;
944 work
->range_cyclic
= range_cyclic
;
945 work
->reason
= reason
;
948 wb_queue_work(wb
, work
);
952 * wb_start_background_writeback - start background writeback
953 * @wb: bdi_writback to write from
956 * This makes sure WB_SYNC_NONE background writeback happens. When
957 * this function returns, it is only guaranteed that for given wb
958 * some IO is happening if we are over background dirty threshold.
959 * Caller need not hold sb s_umount semaphore.
961 void wb_start_background_writeback(struct bdi_writeback
*wb
)
964 * We just wake up the flusher thread. It will perform background
965 * writeback as soon as there is no other work to do.
967 trace_writeback_wake_background(wb
);
972 * Remove the inode from the writeback list it is on.
974 void inode_io_list_del(struct inode
*inode
)
976 struct bdi_writeback
*wb
;
978 wb
= inode_to_wb_and_lock_list(inode
);
979 inode_io_list_del_locked(inode
, wb
);
980 spin_unlock(&wb
->list_lock
);
984 * Redirty an inode: set its when-it-was dirtied timestamp and move it to the
985 * furthest end of its superblock's dirty-inode list.
987 * Before stamping the inode's ->dirtied_when, we check to see whether it is
988 * already the most-recently-dirtied inode on the b_dirty list. If that is
989 * the case then the inode must have been redirtied while it was being written
990 * out and we don't reset its dirtied_when.
992 static void redirty_tail(struct inode
*inode
, struct bdi_writeback
*wb
)
994 if (!list_empty(&wb
->b_dirty
)) {
997 tail
= wb_inode(wb
->b_dirty
.next
);
998 if (time_before(inode
->dirtied_when
, tail
->dirtied_when
))
999 inode
->dirtied_when
= jiffies
;
1001 inode_io_list_move_locked(inode
, wb
, &wb
->b_dirty
);
1005 * requeue inode for re-scanning after bdi->b_io list is exhausted.
1007 static void requeue_io(struct inode
*inode
, struct bdi_writeback
*wb
)
1009 inode_io_list_move_locked(inode
, wb
, &wb
->b_more_io
);
1012 static void inode_sync_complete(struct inode
*inode
)
1014 inode
->i_state
&= ~I_SYNC
;
1015 /* If inode is clean an unused, put it into LRU now... */
1016 inode_add_lru(inode
);
1017 /* Waiters must see I_SYNC cleared before being woken up */
1019 wake_up_bit(&inode
->i_state
, __I_SYNC
);
1022 static bool inode_dirtied_after(struct inode
*inode
, unsigned long t
)
1024 bool ret
= time_after(inode
->dirtied_when
, t
);
1025 #ifndef CONFIG_64BIT
1027 * For inodes being constantly redirtied, dirtied_when can get stuck.
1028 * It _appears_ to be in the future, but is actually in distant past.
1029 * This test is necessary to prevent such wrapped-around relative times
1030 * from permanently stopping the whole bdi writeback.
1032 ret
= ret
&& time_before_eq(inode
->dirtied_when
, jiffies
);
1037 #define EXPIRE_DIRTY_ATIME 0x0001
1040 * Move expired (dirtied before work->older_than_this) dirty inodes from
1041 * @delaying_queue to @dispatch_queue.
1043 static int move_expired_inodes(struct list_head
*delaying_queue
,
1044 struct list_head
*dispatch_queue
,
1046 struct wb_writeback_work
*work
)
1048 unsigned long *older_than_this
= NULL
;
1049 unsigned long expire_time
;
1051 struct list_head
*pos
, *node
;
1052 struct super_block
*sb
= NULL
;
1053 struct inode
*inode
;
1057 if ((flags
& EXPIRE_DIRTY_ATIME
) == 0)
1058 older_than_this
= work
->older_than_this
;
1059 else if (!work
->for_sync
) {
1060 expire_time
= jiffies
- (dirtytime_expire_interval
* HZ
);
1061 older_than_this
= &expire_time
;
1063 while (!list_empty(delaying_queue
)) {
1064 inode
= wb_inode(delaying_queue
->prev
);
1065 if (older_than_this
&&
1066 inode_dirtied_after(inode
, *older_than_this
))
1068 list_move(&inode
->i_io_list
, &tmp
);
1070 if (flags
& EXPIRE_DIRTY_ATIME
)
1071 set_bit(__I_DIRTY_TIME_EXPIRED
, &inode
->i_state
);
1072 if (sb_is_blkdev_sb(inode
->i_sb
))
1074 if (sb
&& sb
!= inode
->i_sb
)
1079 /* just one sb in list, splice to dispatch_queue and we're done */
1081 list_splice(&tmp
, dispatch_queue
);
1085 /* Move inodes from one superblock together */
1086 while (!list_empty(&tmp
)) {
1087 sb
= wb_inode(tmp
.prev
)->i_sb
;
1088 list_for_each_prev_safe(pos
, node
, &tmp
) {
1089 inode
= wb_inode(pos
);
1090 if (inode
->i_sb
== sb
)
1091 list_move(&inode
->i_io_list
, dispatch_queue
);
1099 * Queue all expired dirty inodes for io, eldest first.
1101 * newly dirtied b_dirty b_io b_more_io
1102 * =============> gf edc BA
1104 * newly dirtied b_dirty b_io b_more_io
1105 * =============> g fBAedc
1107 * +--> dequeue for IO
1109 static void queue_io(struct bdi_writeback
*wb
, struct wb_writeback_work
*work
)
1113 assert_spin_locked(&wb
->list_lock
);
1114 list_splice_init(&wb
->b_more_io
, &wb
->b_io
);
1115 moved
= move_expired_inodes(&wb
->b_dirty
, &wb
->b_io
, 0, work
);
1116 moved
+= move_expired_inodes(&wb
->b_dirty_time
, &wb
->b_io
,
1117 EXPIRE_DIRTY_ATIME
, work
);
1119 wb_io_lists_populated(wb
);
1120 trace_writeback_queue_io(wb
, work
, moved
);
1123 static int write_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1127 if (inode
->i_sb
->s_op
->write_inode
&& !is_bad_inode(inode
)) {
1128 trace_writeback_write_inode_start(inode
, wbc
);
1129 ret
= inode
->i_sb
->s_op
->write_inode(inode
, wbc
);
1130 trace_writeback_write_inode(inode
, wbc
);
1137 * Wait for writeback on an inode to complete. Called with i_lock held.
1138 * Caller must make sure inode cannot go away when we drop i_lock.
1140 static void __inode_wait_for_writeback(struct inode
*inode
)
1141 __releases(inode
->i_lock
)
1142 __acquires(inode
->i_lock
)
1144 DEFINE_WAIT_BIT(wq
, &inode
->i_state
, __I_SYNC
);
1145 wait_queue_head_t
*wqh
;
1147 wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
1148 while (inode
->i_state
& I_SYNC
) {
1149 spin_unlock(&inode
->i_lock
);
1150 __wait_on_bit(wqh
, &wq
, bit_wait
,
1151 TASK_UNINTERRUPTIBLE
);
1152 spin_lock(&inode
->i_lock
);
1157 * Wait for writeback on an inode to complete. Caller must have inode pinned.
1159 void inode_wait_for_writeback(struct inode
*inode
)
1161 spin_lock(&inode
->i_lock
);
1162 __inode_wait_for_writeback(inode
);
1163 spin_unlock(&inode
->i_lock
);
1167 * Sleep until I_SYNC is cleared. This function must be called with i_lock
1168 * held and drops it. It is aimed for callers not holding any inode reference
1169 * so once i_lock is dropped, inode can go away.
1171 static void inode_sleep_on_writeback(struct inode
*inode
)
1172 __releases(inode
->i_lock
)
1175 wait_queue_head_t
*wqh
= bit_waitqueue(&inode
->i_state
, __I_SYNC
);
1178 prepare_to_wait(wqh
, &wait
, TASK_UNINTERRUPTIBLE
);
1179 sleep
= inode
->i_state
& I_SYNC
;
1180 spin_unlock(&inode
->i_lock
);
1183 finish_wait(wqh
, &wait
);
1187 * Find proper writeback list for the inode depending on its current state and
1188 * possibly also change of its state while we were doing writeback. Here we
1189 * handle things such as livelock prevention or fairness of writeback among
1190 * inodes. This function can be called only by flusher thread - noone else
1191 * processes all inodes in writeback lists and requeueing inodes behind flusher
1192 * thread's back can have unexpected consequences.
1194 static void requeue_inode(struct inode
*inode
, struct bdi_writeback
*wb
,
1195 struct writeback_control
*wbc
)
1197 if (inode
->i_state
& I_FREEING
)
1201 * Sync livelock prevention. Each inode is tagged and synced in one
1202 * shot. If still dirty, it will be redirty_tail()'ed below. Update
1203 * the dirty time to prevent enqueue and sync it again.
1205 if ((inode
->i_state
& I_DIRTY
) &&
1206 (wbc
->sync_mode
== WB_SYNC_ALL
|| wbc
->tagged_writepages
))
1207 inode
->dirtied_when
= jiffies
;
1209 if (wbc
->pages_skipped
) {
1211 * writeback is not making progress due to locked
1212 * buffers. Skip this inode for now.
1214 redirty_tail(inode
, wb
);
1218 if (mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_DIRTY
)) {
1220 * We didn't write back all the pages. nfs_writepages()
1221 * sometimes bales out without doing anything.
1223 if (wbc
->nr_to_write
<= 0) {
1224 /* Slice used up. Queue for next turn. */
1225 requeue_io(inode
, wb
);
1228 * Writeback blocked by something other than
1229 * congestion. Delay the inode for some time to
1230 * avoid spinning on the CPU (100% iowait)
1231 * retrying writeback of the dirty page/inode
1232 * that cannot be performed immediately.
1234 redirty_tail(inode
, wb
);
1236 } else if (inode
->i_state
& I_DIRTY
) {
1238 * Filesystems can dirty the inode during writeback operations,
1239 * such as delayed allocation during submission or metadata
1240 * updates after data IO completion.
1242 redirty_tail(inode
, wb
);
1243 } else if (inode
->i_state
& I_DIRTY_TIME
) {
1244 inode
->dirtied_when
= jiffies
;
1245 inode_io_list_move_locked(inode
, wb
, &wb
->b_dirty_time
);
1247 /* The inode is clean. Remove from writeback lists. */
1248 inode_io_list_del_locked(inode
, wb
);
1253 * Write out an inode and its dirty pages. Do not update the writeback list
1254 * linkage. That is left to the caller. The caller is also responsible for
1255 * setting I_SYNC flag and calling inode_sync_complete() to clear it.
1258 __writeback_single_inode(struct inode
*inode
, struct writeback_control
*wbc
)
1260 struct address_space
*mapping
= inode
->i_mapping
;
1261 long nr_to_write
= wbc
->nr_to_write
;
1265 WARN_ON(!(inode
->i_state
& I_SYNC
));
1267 trace_writeback_single_inode_start(inode
, wbc
, nr_to_write
);
1269 ret
= do_writepages(mapping
, wbc
);
1272 * Make sure to wait on the data before writing out the metadata.
1273 * This is important for filesystems that modify metadata on data
1274 * I/O completion. We don't do it for sync(2) writeback because it has a
1275 * separate, external IO completion path and ->sync_fs for guaranteeing
1276 * inode metadata is written back correctly.
1278 if (wbc
->sync_mode
== WB_SYNC_ALL
&& !wbc
->for_sync
) {
1279 int err
= filemap_fdatawait(mapping
);
1285 * Some filesystems may redirty the inode during the writeback
1286 * due to delalloc, clear dirty metadata flags right before
1289 spin_lock(&inode
->i_lock
);
1291 dirty
= inode
->i_state
& I_DIRTY
;
1292 if (inode
->i_state
& I_DIRTY_TIME
) {
1293 if ((dirty
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
)) ||
1294 unlikely(inode
->i_state
& I_DIRTY_TIME_EXPIRED
) ||
1295 unlikely(time_after(jiffies
,
1296 (inode
->dirtied_time_when
+
1297 dirtytime_expire_interval
* HZ
)))) {
1298 dirty
|= I_DIRTY_TIME
| I_DIRTY_TIME_EXPIRED
;
1299 trace_writeback_lazytime(inode
);
1302 inode
->i_state
&= ~I_DIRTY_TIME_EXPIRED
;
1303 inode
->i_state
&= ~dirty
;
1306 * Paired with smp_mb() in __mark_inode_dirty(). This allows
1307 * __mark_inode_dirty() to test i_state without grabbing i_lock -
1308 * either they see the I_DIRTY bits cleared or we see the dirtied
1311 * I_DIRTY_PAGES is always cleared together above even if @mapping
1312 * still has dirty pages. The flag is reinstated after smp_mb() if
1313 * necessary. This guarantees that either __mark_inode_dirty()
1314 * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY.
1318 if (mapping_tagged(mapping
, PAGECACHE_TAG_DIRTY
))
1319 inode
->i_state
|= I_DIRTY_PAGES
;
1321 spin_unlock(&inode
->i_lock
);
1323 if (dirty
& I_DIRTY_TIME
)
1324 mark_inode_dirty_sync(inode
);
1325 /* Don't write the inode if only I_DIRTY_PAGES was set */
1326 if (dirty
& ~I_DIRTY_PAGES
) {
1327 int err
= write_inode(inode
, wbc
);
1331 trace_writeback_single_inode(inode
, wbc
, nr_to_write
);
1336 * Write out an inode's dirty pages. Either the caller has an active reference
1337 * on the inode or the inode has I_WILL_FREE set.
1339 * This function is designed to be called for writing back one inode which
1340 * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode()
1341 * and does more profound writeback list handling in writeback_sb_inodes().
1343 static int writeback_single_inode(struct inode
*inode
,
1344 struct writeback_control
*wbc
)
1346 struct bdi_writeback
*wb
;
1349 spin_lock(&inode
->i_lock
);
1350 if (!atomic_read(&inode
->i_count
))
1351 WARN_ON(!(inode
->i_state
& (I_WILL_FREE
|I_FREEING
)));
1353 WARN_ON(inode
->i_state
& I_WILL_FREE
);
1355 if (inode
->i_state
& I_SYNC
) {
1356 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
1359 * It's a data-integrity sync. We must wait. Since callers hold
1360 * inode reference or inode has I_WILL_FREE set, it cannot go
1363 __inode_wait_for_writeback(inode
);
1365 WARN_ON(inode
->i_state
& I_SYNC
);
1367 * Skip inode if it is clean and we have no outstanding writeback in
1368 * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this
1369 * function since flusher thread may be doing for example sync in
1370 * parallel and if we move the inode, it could get skipped. So here we
1371 * make sure inode is on some writeback list and leave it there unless
1372 * we have completely cleaned the inode.
1374 if (!(inode
->i_state
& I_DIRTY_ALL
) &&
1375 (wbc
->sync_mode
!= WB_SYNC_ALL
||
1376 !mapping_tagged(inode
->i_mapping
, PAGECACHE_TAG_WRITEBACK
)))
1378 inode
->i_state
|= I_SYNC
;
1379 wbc_attach_and_unlock_inode(wbc
, inode
);
1381 ret
= __writeback_single_inode(inode
, wbc
);
1383 wbc_detach_inode(wbc
);
1385 wb
= inode_to_wb_and_lock_list(inode
);
1386 spin_lock(&inode
->i_lock
);
1388 * If inode is clean, remove it from writeback lists. Otherwise don't
1389 * touch it. See comment above for explanation.
1391 if (!(inode
->i_state
& I_DIRTY_ALL
))
1392 inode_io_list_del_locked(inode
, wb
);
1393 spin_unlock(&wb
->list_lock
);
1394 inode_sync_complete(inode
);
1396 spin_unlock(&inode
->i_lock
);
1400 static long writeback_chunk_size(struct bdi_writeback
*wb
,
1401 struct wb_writeback_work
*work
)
1406 * WB_SYNC_ALL mode does livelock avoidance by syncing dirty
1407 * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX
1408 * here avoids calling into writeback_inodes_wb() more than once.
1410 * The intended call sequence for WB_SYNC_ALL writeback is:
1413 * writeback_sb_inodes() <== called only once
1414 * write_cache_pages() <== called once for each inode
1415 * (quickly) tag currently dirty pages
1416 * (maybe slowly) sync all tagged pages
1418 if (work
->sync_mode
== WB_SYNC_ALL
|| work
->tagged_writepages
)
1421 pages
= min(wb
->avg_write_bandwidth
/ 2,
1422 global_wb_domain
.dirty_limit
/ DIRTY_SCOPE
);
1423 pages
= min(pages
, work
->nr_pages
);
1424 pages
= round_down(pages
+ MIN_WRITEBACK_PAGES
,
1425 MIN_WRITEBACK_PAGES
);
1432 * Write a portion of b_io inodes which belong to @sb.
1434 * Return the number of pages and/or inodes written.
1436 * NOTE! This is called with wb->list_lock held, and will
1437 * unlock and relock that for each inode it ends up doing
1440 static long writeback_sb_inodes(struct super_block
*sb
,
1441 struct bdi_writeback
*wb
,
1442 struct wb_writeback_work
*work
)
1444 struct writeback_control wbc
= {
1445 .sync_mode
= work
->sync_mode
,
1446 .tagged_writepages
= work
->tagged_writepages
,
1447 .for_kupdate
= work
->for_kupdate
,
1448 .for_background
= work
->for_background
,
1449 .for_sync
= work
->for_sync
,
1450 .range_cyclic
= work
->range_cyclic
,
1452 .range_end
= LLONG_MAX
,
1454 unsigned long start_time
= jiffies
;
1456 long wrote
= 0; /* count both pages and inodes */
1458 while (!list_empty(&wb
->b_io
)) {
1459 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
1460 struct bdi_writeback
*tmp_wb
;
1462 if (inode
->i_sb
!= sb
) {
1465 * We only want to write back data for this
1466 * superblock, move all inodes not belonging
1467 * to it back onto the dirty list.
1469 redirty_tail(inode
, wb
);
1474 * The inode belongs to a different superblock.
1475 * Bounce back to the caller to unpin this and
1476 * pin the next superblock.
1482 * Don't bother with new inodes or inodes being freed, first
1483 * kind does not need periodic writeout yet, and for the latter
1484 * kind writeout is handled by the freer.
1486 spin_lock(&inode
->i_lock
);
1487 if (inode
->i_state
& (I_NEW
| I_FREEING
| I_WILL_FREE
)) {
1488 spin_unlock(&inode
->i_lock
);
1489 redirty_tail(inode
, wb
);
1492 if ((inode
->i_state
& I_SYNC
) && wbc
.sync_mode
!= WB_SYNC_ALL
) {
1494 * If this inode is locked for writeback and we are not
1495 * doing writeback-for-data-integrity, move it to
1496 * b_more_io so that writeback can proceed with the
1497 * other inodes on s_io.
1499 * We'll have another go at writing back this inode
1500 * when we completed a full scan of b_io.
1502 spin_unlock(&inode
->i_lock
);
1503 requeue_io(inode
, wb
);
1504 trace_writeback_sb_inodes_requeue(inode
);
1507 spin_unlock(&wb
->list_lock
);
1510 * We already requeued the inode if it had I_SYNC set and we
1511 * are doing WB_SYNC_NONE writeback. So this catches only the
1514 if (inode
->i_state
& I_SYNC
) {
1515 /* Wait for I_SYNC. This function drops i_lock... */
1516 inode_sleep_on_writeback(inode
);
1517 /* Inode may be gone, start again */
1518 spin_lock(&wb
->list_lock
);
1521 inode
->i_state
|= I_SYNC
;
1522 wbc_attach_and_unlock_inode(&wbc
, inode
);
1524 write_chunk
= writeback_chunk_size(wb
, work
);
1525 wbc
.nr_to_write
= write_chunk
;
1526 wbc
.pages_skipped
= 0;
1529 * We use I_SYNC to pin the inode in memory. While it is set
1530 * evict_inode() will wait so the inode cannot be freed.
1532 __writeback_single_inode(inode
, &wbc
);
1534 wbc_detach_inode(&wbc
);
1535 work
->nr_pages
-= write_chunk
- wbc
.nr_to_write
;
1536 wrote
+= write_chunk
- wbc
.nr_to_write
;
1538 if (need_resched()) {
1540 * We're trying to balance between building up a nice
1541 * long list of IOs to improve our merge rate, and
1542 * getting those IOs out quickly for anyone throttling
1543 * in balance_dirty_pages(). cond_resched() doesn't
1544 * unplug, so get our IOs out the door before we
1547 blk_flush_plug(current
);
1552 * Requeue @inode if still dirty. Be careful as @inode may
1553 * have been switched to another wb in the meantime.
1555 tmp_wb
= inode_to_wb_and_lock_list(inode
);
1556 spin_lock(&inode
->i_lock
);
1557 if (!(inode
->i_state
& I_DIRTY_ALL
))
1559 requeue_inode(inode
, tmp_wb
, &wbc
);
1560 inode_sync_complete(inode
);
1561 spin_unlock(&inode
->i_lock
);
1563 if (unlikely(tmp_wb
!= wb
)) {
1564 spin_unlock(&tmp_wb
->list_lock
);
1565 spin_lock(&wb
->list_lock
);
1569 * bail out to wb_writeback() often enough to check
1570 * background threshold and other termination conditions.
1573 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
1575 if (work
->nr_pages
<= 0)
1582 static long __writeback_inodes_wb(struct bdi_writeback
*wb
,
1583 struct wb_writeback_work
*work
)
1585 unsigned long start_time
= jiffies
;
1588 while (!list_empty(&wb
->b_io
)) {
1589 struct inode
*inode
= wb_inode(wb
->b_io
.prev
);
1590 struct super_block
*sb
= inode
->i_sb
;
1592 if (!trylock_super(sb
)) {
1594 * trylock_super() may fail consistently due to
1595 * s_umount being grabbed by someone else. Don't use
1596 * requeue_io() to avoid busy retrying the inode/sb.
1598 redirty_tail(inode
, wb
);
1601 wrote
+= writeback_sb_inodes(sb
, wb
, work
);
1602 up_read(&sb
->s_umount
);
1604 /* refer to the same tests at the end of writeback_sb_inodes */
1606 if (time_is_before_jiffies(start_time
+ HZ
/ 10UL))
1608 if (work
->nr_pages
<= 0)
1612 /* Leave any unwritten inodes on b_io */
1616 static long writeback_inodes_wb(struct bdi_writeback
*wb
, long nr_pages
,
1617 enum wb_reason reason
)
1619 struct wb_writeback_work work
= {
1620 .nr_pages
= nr_pages
,
1621 .sync_mode
= WB_SYNC_NONE
,
1625 struct blk_plug plug
;
1627 blk_start_plug(&plug
);
1628 spin_lock(&wb
->list_lock
);
1629 if (list_empty(&wb
->b_io
))
1630 queue_io(wb
, &work
);
1631 __writeback_inodes_wb(wb
, &work
);
1632 spin_unlock(&wb
->list_lock
);
1633 blk_finish_plug(&plug
);
1635 return nr_pages
- work
.nr_pages
;
1639 * Explicit flushing or periodic writeback of "old" data.
1641 * Define "old": the first time one of an inode's pages is dirtied, we mark the
1642 * dirtying-time in the inode's address_space. So this periodic writeback code
1643 * just walks the superblock inode list, writing back any inodes which are
1644 * older than a specific point in time.
1646 * Try to run once per dirty_writeback_interval. But if a writeback event
1647 * takes longer than a dirty_writeback_interval interval, then leave a
1650 * older_than_this takes precedence over nr_to_write. So we'll only write back
1651 * all dirty pages if they are all attached to "old" mappings.
1653 static long wb_writeback(struct bdi_writeback
*wb
,
1654 struct wb_writeback_work
*work
)
1656 unsigned long wb_start
= jiffies
;
1657 long nr_pages
= work
->nr_pages
;
1658 unsigned long oldest_jif
;
1659 struct inode
*inode
;
1661 struct blk_plug plug
;
1663 oldest_jif
= jiffies
;
1664 work
->older_than_this
= &oldest_jif
;
1666 blk_start_plug(&plug
);
1667 spin_lock(&wb
->list_lock
);
1670 * Stop writeback when nr_pages has been consumed
1672 if (work
->nr_pages
<= 0)
1676 * Background writeout and kupdate-style writeback may
1677 * run forever. Stop them if there is other work to do
1678 * so that e.g. sync can proceed. They'll be restarted
1679 * after the other works are all done.
1681 if ((work
->for_background
|| work
->for_kupdate
) &&
1682 !list_empty(&wb
->work_list
))
1686 * For background writeout, stop when we are below the
1687 * background dirty threshold
1689 if (work
->for_background
&& !wb_over_bg_thresh(wb
))
1693 * Kupdate and background works are special and we want to
1694 * include all inodes that need writing. Livelock avoidance is
1695 * handled by these works yielding to any other work so we are
1698 if (work
->for_kupdate
) {
1699 oldest_jif
= jiffies
-
1700 msecs_to_jiffies(dirty_expire_interval
* 10);
1701 } else if (work
->for_background
)
1702 oldest_jif
= jiffies
;
1704 trace_writeback_start(wb
, work
);
1705 if (list_empty(&wb
->b_io
))
1708 progress
= writeback_sb_inodes(work
->sb
, wb
, work
);
1710 progress
= __writeback_inodes_wb(wb
, work
);
1711 trace_writeback_written(wb
, work
);
1713 wb_update_bandwidth(wb
, wb_start
);
1716 * Did we write something? Try for more
1718 * Dirty inodes are moved to b_io for writeback in batches.
1719 * The completion of the current batch does not necessarily
1720 * mean the overall work is done. So we keep looping as long
1721 * as made some progress on cleaning pages or inodes.
1726 * No more inodes for IO, bail
1728 if (list_empty(&wb
->b_more_io
))
1731 * Nothing written. Wait for some inode to
1732 * become available for writeback. Otherwise
1733 * we'll just busyloop.
1735 if (!list_empty(&wb
->b_more_io
)) {
1736 trace_writeback_wait(wb
, work
);
1737 inode
= wb_inode(wb
->b_more_io
.prev
);
1738 spin_lock(&inode
->i_lock
);
1739 spin_unlock(&wb
->list_lock
);
1740 /* This function drops i_lock... */
1741 inode_sleep_on_writeback(inode
);
1742 spin_lock(&wb
->list_lock
);
1745 spin_unlock(&wb
->list_lock
);
1746 blk_finish_plug(&plug
);
1748 return nr_pages
- work
->nr_pages
;
1752 * Return the next wb_writeback_work struct that hasn't been processed yet.
1754 static struct wb_writeback_work
*get_next_work_item(struct bdi_writeback
*wb
)
1756 struct wb_writeback_work
*work
= NULL
;
1758 spin_lock_bh(&wb
->work_lock
);
1759 if (!list_empty(&wb
->work_list
)) {
1760 work
= list_entry(wb
->work_list
.next
,
1761 struct wb_writeback_work
, list
);
1762 list_del_init(&work
->list
);
1764 spin_unlock_bh(&wb
->work_lock
);
1769 * Add in the number of potentially dirty inodes, because each inode
1770 * write can dirty pagecache in the underlying blockdev.
1772 static unsigned long get_nr_dirty_pages(void)
1774 return global_page_state(NR_FILE_DIRTY
) +
1775 global_page_state(NR_UNSTABLE_NFS
) +
1776 get_nr_dirty_inodes();
1779 static long wb_check_background_flush(struct bdi_writeback
*wb
)
1781 if (wb_over_bg_thresh(wb
)) {
1783 struct wb_writeback_work work
= {
1784 .nr_pages
= LONG_MAX
,
1785 .sync_mode
= WB_SYNC_NONE
,
1786 .for_background
= 1,
1788 .reason
= WB_REASON_BACKGROUND
,
1791 return wb_writeback(wb
, &work
);
1797 static long wb_check_old_data_flush(struct bdi_writeback
*wb
)
1799 unsigned long expired
;
1803 * When set to zero, disable periodic writeback
1805 if (!dirty_writeback_interval
)
1808 expired
= wb
->last_old_flush
+
1809 msecs_to_jiffies(dirty_writeback_interval
* 10);
1810 if (time_before(jiffies
, expired
))
1813 wb
->last_old_flush
= jiffies
;
1814 nr_pages
= get_nr_dirty_pages();
1817 struct wb_writeback_work work
= {
1818 .nr_pages
= nr_pages
,
1819 .sync_mode
= WB_SYNC_NONE
,
1822 .reason
= WB_REASON_PERIODIC
,
1825 return wb_writeback(wb
, &work
);
1832 * Retrieve work items and do the writeback they describe
1834 static long wb_do_writeback(struct bdi_writeback
*wb
)
1836 struct wb_writeback_work
*work
;
1839 set_bit(WB_writeback_running
, &wb
->state
);
1840 while ((work
= get_next_work_item(wb
)) != NULL
) {
1841 struct wb_completion
*done
= work
->done
;
1843 trace_writeback_exec(wb
, work
);
1845 wrote
+= wb_writeback(wb
, work
);
1847 if (work
->auto_free
)
1849 if (done
&& atomic_dec_and_test(&done
->cnt
))
1850 wake_up_all(&wb
->bdi
->wb_waitq
);
1854 * Check for periodic writeback, kupdated() style
1856 wrote
+= wb_check_old_data_flush(wb
);
1857 wrote
+= wb_check_background_flush(wb
);
1858 clear_bit(WB_writeback_running
, &wb
->state
);
1864 * Handle writeback of dirty data for the device backed by this bdi. Also
1865 * reschedules periodically and does kupdated style flushing.
1867 void wb_workfn(struct work_struct
*work
)
1869 struct bdi_writeback
*wb
= container_of(to_delayed_work(work
),
1870 struct bdi_writeback
, dwork
);
1873 set_worker_desc("flush-%s", dev_name(wb
->bdi
->dev
));
1874 current
->flags
|= PF_SWAPWRITE
;
1876 if (likely(!current_is_workqueue_rescuer() ||
1877 !test_bit(WB_registered
, &wb
->state
))) {
1879 * The normal path. Keep writing back @wb until its
1880 * work_list is empty. Note that this path is also taken
1881 * if @wb is shutting down even when we're running off the
1882 * rescuer as work_list needs to be drained.
1885 pages_written
= wb_do_writeback(wb
);
1886 trace_writeback_pages_written(pages_written
);
1887 } while (!list_empty(&wb
->work_list
));
1890 * bdi_wq can't get enough workers and we're running off
1891 * the emergency worker. Don't hog it. Hopefully, 1024 is
1892 * enough for efficient IO.
1894 pages_written
= writeback_inodes_wb(wb
, 1024,
1895 WB_REASON_FORKER_THREAD
);
1896 trace_writeback_pages_written(pages_written
);
1899 if (!list_empty(&wb
->work_list
))
1900 mod_delayed_work(bdi_wq
, &wb
->dwork
, 0);
1901 else if (wb_has_dirty_io(wb
) && dirty_writeback_interval
)
1902 wb_wakeup_delayed(wb
);
1904 current
->flags
&= ~PF_SWAPWRITE
;
1908 * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back
1911 void wakeup_flusher_threads(long nr_pages
, enum wb_reason reason
)
1913 struct backing_dev_info
*bdi
;
1916 nr_pages
= get_nr_dirty_pages();
1919 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1920 struct bdi_writeback
*wb
;
1922 if (!bdi_has_dirty_io(bdi
))
1925 list_for_each_entry_rcu(wb
, &bdi
->wb_list
, bdi_node
)
1926 wb_start_writeback(wb
, wb_split_bdi_pages(wb
, nr_pages
),
1933 * Wake up bdi's periodically to make sure dirtytime inodes gets
1934 * written back periodically. We deliberately do *not* check the
1935 * b_dirtytime list in wb_has_dirty_io(), since this would cause the
1936 * kernel to be constantly waking up once there are any dirtytime
1937 * inodes on the system. So instead we define a separate delayed work
1938 * function which gets called much more rarely. (By default, only
1939 * once every 12 hours.)
1941 * If there is any other write activity going on in the file system,
1942 * this function won't be necessary. But if the only thing that has
1943 * happened on the file system is a dirtytime inode caused by an atime
1944 * update, we need this infrastructure below to make sure that inode
1945 * eventually gets pushed out to disk.
1947 static void wakeup_dirtytime_writeback(struct work_struct
*w
);
1948 static DECLARE_DELAYED_WORK(dirtytime_work
, wakeup_dirtytime_writeback
);
1950 static void wakeup_dirtytime_writeback(struct work_struct
*w
)
1952 struct backing_dev_info
*bdi
;
1955 list_for_each_entry_rcu(bdi
, &bdi_list
, bdi_list
) {
1956 struct bdi_writeback
*wb
;
1958 list_for_each_entry_rcu(wb
, &bdi
->wb_list
, bdi_node
)
1959 if (!list_empty(&wb
->b_dirty_time
))
1963 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1966 static int __init
start_dirtytime_writeback(void)
1968 schedule_delayed_work(&dirtytime_work
, dirtytime_expire_interval
* HZ
);
1971 __initcall(start_dirtytime_writeback
);
1973 int dirtytime_interval_handler(struct ctl_table
*table
, int write
,
1974 void __user
*buffer
, size_t *lenp
, loff_t
*ppos
)
1978 ret
= proc_dointvec_minmax(table
, write
, buffer
, lenp
, ppos
);
1979 if (ret
== 0 && write
)
1980 mod_delayed_work(system_wq
, &dirtytime_work
, 0);
1984 static noinline
void block_dump___mark_inode_dirty(struct inode
*inode
)
1986 if (inode
->i_ino
|| strcmp(inode
->i_sb
->s_id
, "bdev")) {
1987 struct dentry
*dentry
;
1988 const char *name
= "?";
1990 dentry
= d_find_alias(inode
);
1992 spin_lock(&dentry
->d_lock
);
1993 name
= (const char *) dentry
->d_name
.name
;
1996 "%s(%d): dirtied inode %lu (%s) on %s\n",
1997 current
->comm
, task_pid_nr(current
), inode
->i_ino
,
1998 name
, inode
->i_sb
->s_id
);
2000 spin_unlock(&dentry
->d_lock
);
2007 * __mark_inode_dirty - internal function
2008 * @inode: inode to mark
2009 * @flags: what kind of dirty (i.e. I_DIRTY_SYNC)
2010 * Mark an inode as dirty. Callers should use mark_inode_dirty or
2011 * mark_inode_dirty_sync.
2013 * Put the inode on the super block's dirty list.
2015 * CAREFUL! We mark it dirty unconditionally, but move it onto the
2016 * dirty list only if it is hashed or if it refers to a blockdev.
2017 * If it was not hashed, it will never be added to the dirty list
2018 * even if it is later hashed, as it will have been marked dirty already.
2020 * In short, make sure you hash any inodes _before_ you start marking
2023 * Note that for blockdevs, inode->dirtied_when represents the dirtying time of
2024 * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of
2025 * the kernel-internal blockdev inode represents the dirtying time of the
2026 * blockdev's pages. This is why for I_DIRTY_PAGES we always use
2027 * page->mapping->host, so the page-dirtying time is recorded in the internal
2030 void __mark_inode_dirty(struct inode
*inode
, int flags
)
2032 #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC)
2033 struct super_block
*sb
= inode
->i_sb
;
2036 trace_writeback_mark_inode_dirty(inode
, flags
);
2039 * Don't do this for I_DIRTY_PAGES - that doesn't actually
2040 * dirty the inode itself
2042 if (flags
& (I_DIRTY_SYNC
| I_DIRTY_DATASYNC
| I_DIRTY_TIME
)) {
2043 trace_writeback_dirty_inode_start(inode
, flags
);
2045 if (sb
->s_op
->dirty_inode
)
2046 sb
->s_op
->dirty_inode(inode
, flags
);
2048 trace_writeback_dirty_inode(inode
, flags
);
2050 if (flags
& I_DIRTY_INODE
)
2051 flags
&= ~I_DIRTY_TIME
;
2052 dirtytime
= flags
& I_DIRTY_TIME
;
2055 * Paired with smp_mb() in __writeback_single_inode() for the
2056 * following lockless i_state test. See there for details.
2060 if (((inode
->i_state
& flags
) == flags
) ||
2061 (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
)))
2064 if (unlikely(block_dump
))
2065 block_dump___mark_inode_dirty(inode
);
2067 spin_lock(&inode
->i_lock
);
2068 if (dirtytime
&& (inode
->i_state
& I_DIRTY_INODE
))
2069 goto out_unlock_inode
;
2070 if ((inode
->i_state
& flags
) != flags
) {
2071 const int was_dirty
= inode
->i_state
& I_DIRTY
;
2073 inode_attach_wb(inode
, NULL
);
2075 if (flags
& I_DIRTY_INODE
)
2076 inode
->i_state
&= ~I_DIRTY_TIME
;
2077 inode
->i_state
|= flags
;
2080 * If the inode is being synced, just update its dirty state.
2081 * The unlocker will place the inode on the appropriate
2082 * superblock list, based upon its state.
2084 if (inode
->i_state
& I_SYNC
)
2085 goto out_unlock_inode
;
2088 * Only add valid (hashed) inodes to the superblock's
2089 * dirty list. Add blockdev inodes as well.
2091 if (!S_ISBLK(inode
->i_mode
)) {
2092 if (inode_unhashed(inode
))
2093 goto out_unlock_inode
;
2095 if (inode
->i_state
& I_FREEING
)
2096 goto out_unlock_inode
;
2099 * If the inode was already on b_dirty/b_io/b_more_io, don't
2100 * reposition it (that would break b_dirty time-ordering).
2103 struct bdi_writeback
*wb
;
2104 struct list_head
*dirty_list
;
2105 bool wakeup_bdi
= false;
2107 wb
= locked_inode_to_wb_and_lock_list(inode
);
2109 WARN(bdi_cap_writeback_dirty(wb
->bdi
) &&
2110 !test_bit(WB_registered
, &wb
->state
),
2111 "bdi-%s not registered\n", wb
->bdi
->name
);
2113 inode
->dirtied_when
= jiffies
;
2115 inode
->dirtied_time_when
= jiffies
;
2117 if (inode
->i_state
& (I_DIRTY_INODE
| I_DIRTY_PAGES
))
2118 dirty_list
= &wb
->b_dirty
;
2120 dirty_list
= &wb
->b_dirty_time
;
2122 wakeup_bdi
= inode_io_list_move_locked(inode
, wb
,
2125 spin_unlock(&wb
->list_lock
);
2126 trace_writeback_dirty_inode_enqueue(inode
);
2129 * If this is the first dirty inode for this bdi,
2130 * we have to wake-up the corresponding bdi thread
2131 * to make sure background write-back happens
2134 if (bdi_cap_writeback_dirty(wb
->bdi
) && wakeup_bdi
)
2135 wb_wakeup_delayed(wb
);
2140 spin_unlock(&inode
->i_lock
);
2142 #undef I_DIRTY_INODE
2144 EXPORT_SYMBOL(__mark_inode_dirty
);
2147 * The @s_sync_lock is used to serialise concurrent sync operations
2148 * to avoid lock contention problems with concurrent wait_sb_inodes() calls.
2149 * Concurrent callers will block on the s_sync_lock rather than doing contending
2150 * walks. The queueing maintains sync(2) required behaviour as all the IO that
2151 * has been issued up to the time this function is enter is guaranteed to be
2152 * completed by the time we have gained the lock and waited for all IO that is
2153 * in progress regardless of the order callers are granted the lock.
2155 static void wait_sb_inodes(struct super_block
*sb
)
2157 struct inode
*inode
, *old_inode
= NULL
;
2160 * We need to be protected against the filesystem going from
2161 * r/o to r/w or vice versa.
2163 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
2165 mutex_lock(&sb
->s_sync_lock
);
2166 spin_lock(&sb
->s_inode_list_lock
);
2169 * Data integrity sync. Must wait for all pages under writeback,
2170 * because there may have been pages dirtied before our sync
2171 * call, but which had writeout started before we write it out.
2172 * In which case, the inode may not be on the dirty list, but
2173 * we still have to wait for that writeout.
2175 list_for_each_entry(inode
, &sb
->s_inodes
, i_sb_list
) {
2176 struct address_space
*mapping
= inode
->i_mapping
;
2178 spin_lock(&inode
->i_lock
);
2179 if ((inode
->i_state
& (I_FREEING
|I_WILL_FREE
|I_NEW
)) ||
2180 (mapping
->nrpages
== 0)) {
2181 spin_unlock(&inode
->i_lock
);
2185 spin_unlock(&inode
->i_lock
);
2186 spin_unlock(&sb
->s_inode_list_lock
);
2189 * We hold a reference to 'inode' so it couldn't have been
2190 * removed from s_inodes list while we dropped the
2191 * s_inode_list_lock. We cannot iput the inode now as we can
2192 * be holding the last reference and we cannot iput it under
2193 * s_inode_list_lock. So we keep the reference and iput it
2200 * We keep the error status of individual mapping so that
2201 * applications can catch the writeback error using fsync(2).
2202 * See filemap_fdatawait_keep_errors() for details.
2204 filemap_fdatawait_keep_errors(mapping
);
2208 spin_lock(&sb
->s_inode_list_lock
);
2210 spin_unlock(&sb
->s_inode_list_lock
);
2212 mutex_unlock(&sb
->s_sync_lock
);
2215 static void __writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
,
2216 enum wb_reason reason
, bool skip_if_busy
)
2218 DEFINE_WB_COMPLETION_ONSTACK(done
);
2219 struct wb_writeback_work work
= {
2221 .sync_mode
= WB_SYNC_NONE
,
2222 .tagged_writepages
= 1,
2227 struct backing_dev_info
*bdi
= sb
->s_bdi
;
2229 if (!bdi_has_dirty_io(bdi
) || bdi
== &noop_backing_dev_info
)
2231 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
2233 bdi_split_work_to_wbs(sb
->s_bdi
, &work
, skip_if_busy
);
2234 wb_wait_for_completion(bdi
, &done
);
2238 * writeback_inodes_sb_nr - writeback dirty inodes from given super_block
2239 * @sb: the superblock
2240 * @nr: the number of pages to write
2241 * @reason: reason why some writeback work initiated
2243 * Start writeback on some inodes on this super_block. No guarantees are made
2244 * on how many (if any) will be written, and this function does not wait
2245 * for IO completion of submitted IO.
2247 void writeback_inodes_sb_nr(struct super_block
*sb
,
2249 enum wb_reason reason
)
2251 __writeback_inodes_sb_nr(sb
, nr
, reason
, false);
2253 EXPORT_SYMBOL(writeback_inodes_sb_nr
);
2256 * writeback_inodes_sb - writeback dirty inodes from given super_block
2257 * @sb: the superblock
2258 * @reason: reason why some writeback work was initiated
2260 * Start writeback on some inodes on this super_block. No guarantees are made
2261 * on how many (if any) will be written, and this function does not wait
2262 * for IO completion of submitted IO.
2264 void writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
2266 return writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
2268 EXPORT_SYMBOL(writeback_inodes_sb
);
2271 * try_to_writeback_inodes_sb_nr - try to start writeback if none underway
2272 * @sb: the superblock
2273 * @nr: the number of pages to write
2274 * @reason: the reason of writeback
2276 * Invoke writeback_inodes_sb_nr if no writeback is currently underway.
2277 * Returns 1 if writeback was started, 0 if not.
2279 bool try_to_writeback_inodes_sb_nr(struct super_block
*sb
, unsigned long nr
,
2280 enum wb_reason reason
)
2282 if (!down_read_trylock(&sb
->s_umount
))
2285 __writeback_inodes_sb_nr(sb
, nr
, reason
, true);
2286 up_read(&sb
->s_umount
);
2289 EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr
);
2292 * try_to_writeback_inodes_sb - try to start writeback if none underway
2293 * @sb: the superblock
2294 * @reason: reason why some writeback work was initiated
2296 * Implement by try_to_writeback_inodes_sb_nr()
2297 * Returns 1 if writeback was started, 0 if not.
2299 bool try_to_writeback_inodes_sb(struct super_block
*sb
, enum wb_reason reason
)
2301 return try_to_writeback_inodes_sb_nr(sb
, get_nr_dirty_pages(), reason
);
2303 EXPORT_SYMBOL(try_to_writeback_inodes_sb
);
2306 * sync_inodes_sb - sync sb inode pages
2307 * @sb: the superblock
2309 * This function writes and waits on any dirty inode belonging to this
2312 void sync_inodes_sb(struct super_block
*sb
)
2314 DEFINE_WB_COMPLETION_ONSTACK(done
);
2315 struct wb_writeback_work work
= {
2317 .sync_mode
= WB_SYNC_ALL
,
2318 .nr_pages
= LONG_MAX
,
2321 .reason
= WB_REASON_SYNC
,
2324 struct backing_dev_info
*bdi
= sb
->s_bdi
;
2327 * Can't skip on !bdi_has_dirty() because we should wait for !dirty
2328 * inodes under writeback and I_DIRTY_TIME inodes ignored by
2329 * bdi_has_dirty() need to be written out too.
2331 if (bdi
== &noop_backing_dev_info
)
2333 WARN_ON(!rwsem_is_locked(&sb
->s_umount
));
2335 bdi_split_work_to_wbs(bdi
, &work
, false);
2336 wb_wait_for_completion(bdi
, &done
);
2340 EXPORT_SYMBOL(sync_inodes_sb
);
2343 * write_inode_now - write an inode to disk
2344 * @inode: inode to write to disk
2345 * @sync: whether the write should be synchronous or not
2347 * This function commits an inode to disk immediately if it is dirty. This is
2348 * primarily needed by knfsd.
2350 * The caller must either have a ref on the inode or must have set I_WILL_FREE.
2352 int write_inode_now(struct inode
*inode
, int sync
)
2354 struct writeback_control wbc
= {
2355 .nr_to_write
= LONG_MAX
,
2356 .sync_mode
= sync
? WB_SYNC_ALL
: WB_SYNC_NONE
,
2358 .range_end
= LLONG_MAX
,
2361 if (!mapping_cap_writeback_dirty(inode
->i_mapping
))
2362 wbc
.nr_to_write
= 0;
2365 return writeback_single_inode(inode
, &wbc
);
2367 EXPORT_SYMBOL(write_inode_now
);
2370 * sync_inode - write an inode and its pages to disk.
2371 * @inode: the inode to sync
2372 * @wbc: controls the writeback mode
2374 * sync_inode() will write an inode and its pages to disk. It will also
2375 * correctly update the inode on its superblock's dirty inode lists and will
2376 * update inode->i_state.
2378 * The caller must have a ref on the inode.
2380 int sync_inode(struct inode
*inode
, struct writeback_control
*wbc
)
2382 return writeback_single_inode(inode
, wbc
);
2384 EXPORT_SYMBOL(sync_inode
);
2387 * sync_inode_metadata - write an inode to disk
2388 * @inode: the inode to sync
2389 * @wait: wait for I/O to complete.
2391 * Write an inode to disk and adjust its dirty state after completion.
2393 * Note: only writes the actual inode, no associated data or other metadata.
2395 int sync_inode_metadata(struct inode
*inode
, int wait
)
2397 struct writeback_control wbc
= {
2398 .sync_mode
= wait
? WB_SYNC_ALL
: WB_SYNC_NONE
,
2399 .nr_to_write
= 0, /* metadata-only */
2402 return sync_inode(inode
, &wbc
);
2404 EXPORT_SYMBOL(sync_inode_metadata
);