]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/fs-writeback.c | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds. | |
5 | * | |
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. | |
10 | * | |
e1f8e874 | 11 | * 10Apr2002 Andrew Morton |
1da177e4 LT |
12 | * Split out of fs/inode.c |
13 | * Additions for address_space-based writeback | |
14 | */ | |
15 | ||
16 | #include <linux/kernel.h> | |
630d9c47 | 17 | #include <linux/export.h> |
1da177e4 | 18 | #include <linux/spinlock.h> |
5a0e3ad6 | 19 | #include <linux/slab.h> |
1da177e4 LT |
20 | #include <linux/sched.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/mm.h> | |
bc31b86a | 23 | #include <linux/pagemap.h> |
03ba3782 | 24 | #include <linux/kthread.h> |
1da177e4 LT |
25 | #include <linux/writeback.h> |
26 | #include <linux/blkdev.h> | |
27 | #include <linux/backing-dev.h> | |
455b2864 | 28 | #include <linux/tracepoint.h> |
719ea2fb | 29 | #include <linux/device.h> |
07f3f05c | 30 | #include "internal.h" |
1da177e4 | 31 | |
bc31b86a WF |
32 | /* |
33 | * 4MB minimal write chunk size | |
34 | */ | |
35 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) | |
36 | ||
cc395d7f TH |
37 | struct wb_completion { |
38 | atomic_t cnt; | |
39 | }; | |
40 | ||
c4a77a6c JA |
41 | /* |
42 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
43 | */ | |
83ba7b07 | 44 | struct wb_writeback_work { |
c4a77a6c JA |
45 | long nr_pages; |
46 | struct super_block *sb; | |
0dc83bd3 | 47 | unsigned long *older_than_this; |
c4a77a6c | 48 | enum writeback_sync_modes sync_mode; |
6e6938b6 | 49 | unsigned int tagged_writepages:1; |
52957fe1 HS |
50 | unsigned int for_kupdate:1; |
51 | unsigned int range_cyclic:1; | |
52 | unsigned int for_background:1; | |
7747bd4b | 53 | unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ |
ac7b19a3 | 54 | unsigned int auto_free:1; /* free on completion */ |
98754bf7 TH |
55 | unsigned int single_wait:1; |
56 | unsigned int single_done:1; | |
0e175a18 | 57 | enum wb_reason reason; /* why was writeback initiated? */ |
c4a77a6c | 58 | |
8010c3b6 | 59 | struct list_head list; /* pending work list */ |
cc395d7f | 60 | struct wb_completion *done; /* set if the caller waits */ |
03ba3782 JA |
61 | }; |
62 | ||
cc395d7f TH |
63 | /* |
64 | * If one wants to wait for one or more wb_writeback_works, each work's | |
65 | * ->done should be set to a wb_completion defined using the following | |
66 | * macro. Once all work items are issued with wb_queue_work(), the caller | |
67 | * can wait for the completion of all using wb_wait_for_completion(). Work | |
68 | * items which are waited upon aren't freed automatically on completion. | |
69 | */ | |
70 | #define DEFINE_WB_COMPLETION_ONSTACK(cmpl) \ | |
71 | struct wb_completion cmpl = { \ | |
72 | .cnt = ATOMIC_INIT(1), \ | |
73 | } | |
74 | ||
75 | ||
a2f48706 TT |
76 | /* |
77 | * If an inode is constantly having its pages dirtied, but then the | |
78 | * updates stop dirtytime_expire_interval seconds in the past, it's | |
79 | * possible for the worst case time between when an inode has its | |
80 | * timestamps updated and when they finally get written out to be two | |
81 | * dirtytime_expire_intervals. We set the default to 12 hours (in | |
82 | * seconds), which means most of the time inodes will have their | |
83 | * timestamps written to disk after 12 hours, but in the worst case a | |
84 | * few inodes might not their timestamps updated for 24 hours. | |
85 | */ | |
86 | unsigned int dirtytime_expire_interval = 12 * 60 * 60; | |
87 | ||
7ccf19a8 NP |
88 | static inline struct inode *wb_inode(struct list_head *head) |
89 | { | |
90 | return list_entry(head, struct inode, i_wb_list); | |
91 | } | |
92 | ||
15eb77a0 WF |
93 | /* |
94 | * Include the creation of the trace points after defining the | |
95 | * wb_writeback_work structure and inline functions so that the definition | |
96 | * remains local to this file. | |
97 | */ | |
98 | #define CREATE_TRACE_POINTS | |
99 | #include <trace/events/writeback.h> | |
100 | ||
774016b2 SW |
101 | EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); |
102 | ||
d6c10f1f TH |
103 | static bool wb_io_lists_populated(struct bdi_writeback *wb) |
104 | { | |
105 | if (wb_has_dirty_io(wb)) { | |
106 | return false; | |
107 | } else { | |
108 | set_bit(WB_has_dirty_io, &wb->state); | |
95a46c65 | 109 | WARN_ON_ONCE(!wb->avg_write_bandwidth); |
766a9d6e TH |
110 | atomic_long_add(wb->avg_write_bandwidth, |
111 | &wb->bdi->tot_write_bandwidth); | |
d6c10f1f TH |
112 | return true; |
113 | } | |
114 | } | |
115 | ||
116 | static void wb_io_lists_depopulated(struct bdi_writeback *wb) | |
117 | { | |
118 | if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && | |
766a9d6e | 119 | list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { |
d6c10f1f | 120 | clear_bit(WB_has_dirty_io, &wb->state); |
95a46c65 TH |
121 | WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth, |
122 | &wb->bdi->tot_write_bandwidth) < 0); | |
766a9d6e | 123 | } |
d6c10f1f TH |
124 | } |
125 | ||
126 | /** | |
127 | * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list | |
128 | * @inode: inode to be moved | |
129 | * @wb: target bdi_writeback | |
130 | * @head: one of @wb->b_{dirty|io|more_io} | |
131 | * | |
132 | * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io. | |
133 | * Returns %true if @inode is the first occupant of the !dirty_time IO | |
134 | * lists; otherwise, %false. | |
135 | */ | |
136 | static bool inode_wb_list_move_locked(struct inode *inode, | |
137 | struct bdi_writeback *wb, | |
138 | struct list_head *head) | |
139 | { | |
140 | assert_spin_locked(&wb->list_lock); | |
141 | ||
142 | list_move(&inode->i_wb_list, head); | |
143 | ||
144 | /* dirty_time doesn't count as dirty_io until expiration */ | |
145 | if (head != &wb->b_dirty_time) | |
146 | return wb_io_lists_populated(wb); | |
147 | ||
148 | wb_io_lists_depopulated(wb); | |
149 | return false; | |
150 | } | |
151 | ||
152 | /** | |
153 | * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list | |
154 | * @inode: inode to be removed | |
155 | * @wb: bdi_writeback @inode is being removed from | |
156 | * | |
157 | * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and | |
158 | * clear %WB_has_dirty_io if all are empty afterwards. | |
159 | */ | |
160 | static void inode_wb_list_del_locked(struct inode *inode, | |
161 | struct bdi_writeback *wb) | |
162 | { | |
163 | assert_spin_locked(&wb->list_lock); | |
164 | ||
165 | list_del_init(&inode->i_wb_list); | |
166 | wb_io_lists_depopulated(wb); | |
167 | } | |
168 | ||
f0054bb1 | 169 | static void wb_wakeup(struct bdi_writeback *wb) |
5acda9d1 | 170 | { |
f0054bb1 TH |
171 | spin_lock_bh(&wb->work_lock); |
172 | if (test_bit(WB_registered, &wb->state)) | |
173 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
174 | spin_unlock_bh(&wb->work_lock); | |
5acda9d1 JK |
175 | } |
176 | ||
f0054bb1 TH |
177 | static void wb_queue_work(struct bdi_writeback *wb, |
178 | struct wb_writeback_work *work) | |
6585027a | 179 | { |
f0054bb1 | 180 | trace_writeback_queue(wb->bdi, work); |
6585027a | 181 | |
f0054bb1 | 182 | spin_lock_bh(&wb->work_lock); |
98754bf7 TH |
183 | if (!test_bit(WB_registered, &wb->state)) { |
184 | if (work->single_wait) | |
185 | work->single_done = 1; | |
5acda9d1 | 186 | goto out_unlock; |
98754bf7 | 187 | } |
cc395d7f TH |
188 | if (work->done) |
189 | atomic_inc(&work->done->cnt); | |
f0054bb1 TH |
190 | list_add_tail(&work->list, &wb->work_list); |
191 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
5acda9d1 | 192 | out_unlock: |
f0054bb1 | 193 | spin_unlock_bh(&wb->work_lock); |
1da177e4 LT |
194 | } |
195 | ||
cc395d7f TH |
196 | /** |
197 | * wb_wait_for_completion - wait for completion of bdi_writeback_works | |
198 | * @bdi: bdi work items were issued to | |
199 | * @done: target wb_completion | |
200 | * | |
201 | * Wait for one or more work items issued to @bdi with their ->done field | |
202 | * set to @done, which should have been defined with | |
203 | * DEFINE_WB_COMPLETION_ONSTACK(). This function returns after all such | |
204 | * work items are completed. Work items which are waited upon aren't freed | |
205 | * automatically on completion. | |
206 | */ | |
207 | static void wb_wait_for_completion(struct backing_dev_info *bdi, | |
208 | struct wb_completion *done) | |
209 | { | |
210 | atomic_dec(&done->cnt); /* put down the initial count */ | |
211 | wait_event(bdi->wb_waitq, !atomic_read(&done->cnt)); | |
212 | } | |
213 | ||
703c2708 TH |
214 | #ifdef CONFIG_CGROUP_WRITEBACK |
215 | ||
216 | /** | |
217 | * inode_congested - test whether an inode is congested | |
218 | * @inode: inode to test for congestion | |
219 | * @cong_bits: mask of WB_[a]sync_congested bits to test | |
220 | * | |
221 | * Tests whether @inode is congested. @cong_bits is the mask of congestion | |
222 | * bits to test and the return value is the mask of set bits. | |
223 | * | |
224 | * If cgroup writeback is enabled for @inode, the congestion state is | |
225 | * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg | |
226 | * associated with @inode is congested; otherwise, the root wb's congestion | |
227 | * state is used. | |
228 | */ | |
229 | int inode_congested(struct inode *inode, int cong_bits) | |
230 | { | |
231 | if (inode) { | |
232 | struct bdi_writeback *wb = inode_to_wb(inode); | |
233 | if (wb) | |
234 | return wb_congested(wb, cong_bits); | |
235 | } | |
236 | ||
237 | return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); | |
238 | } | |
239 | EXPORT_SYMBOL_GPL(inode_congested); | |
240 | ||
98754bf7 TH |
241 | /** |
242 | * wb_wait_for_single_work - wait for completion of a single bdi_writeback_work | |
243 | * @bdi: bdi the work item was issued to | |
244 | * @work: work item to wait for | |
245 | * | |
246 | * Wait for the completion of @work which was issued to one of @bdi's | |
247 | * bdi_writeback's. The caller must have set @work->single_wait before | |
248 | * issuing it. This wait operates independently fo | |
249 | * wb_wait_for_completion() and also disables automatic freeing of @work. | |
250 | */ | |
251 | static void wb_wait_for_single_work(struct backing_dev_info *bdi, | |
252 | struct wb_writeback_work *work) | |
253 | { | |
254 | if (WARN_ON_ONCE(!work->single_wait)) | |
255 | return; | |
256 | ||
257 | wait_event(bdi->wb_waitq, work->single_done); | |
258 | ||
259 | /* | |
260 | * Paired with smp_wmb() in wb_do_writeback() and ensures that all | |
261 | * modifications to @work prior to assertion of ->single_done is | |
262 | * visible to the caller once this function returns. | |
263 | */ | |
264 | smp_rmb(); | |
265 | } | |
266 | ||
f2b65121 TH |
267 | /** |
268 | * wb_split_bdi_pages - split nr_pages to write according to bandwidth | |
269 | * @wb: target bdi_writeback to split @nr_pages to | |
270 | * @nr_pages: number of pages to write for the whole bdi | |
271 | * | |
272 | * Split @wb's portion of @nr_pages according to @wb's write bandwidth in | |
273 | * relation to the total write bandwidth of all wb's w/ dirty inodes on | |
274 | * @wb->bdi. | |
275 | */ | |
276 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) | |
277 | { | |
278 | unsigned long this_bw = wb->avg_write_bandwidth; | |
279 | unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth); | |
280 | ||
281 | if (nr_pages == LONG_MAX) | |
282 | return LONG_MAX; | |
283 | ||
284 | /* | |
285 | * This may be called on clean wb's and proportional distribution | |
286 | * may not make sense, just use the original @nr_pages in those | |
287 | * cases. In general, we wanna err on the side of writing more. | |
288 | */ | |
289 | if (!tot_bw || this_bw >= tot_bw) | |
290 | return nr_pages; | |
291 | else | |
292 | return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw); | |
293 | } | |
294 | ||
db125360 TH |
295 | /** |
296 | * wb_clone_and_queue_work - clone a wb_writeback_work and issue it to a wb | |
297 | * @wb: target bdi_writeback | |
298 | * @base_work: source wb_writeback_work | |
299 | * | |
300 | * Try to make a clone of @base_work and issue it to @wb. If cloning | |
301 | * succeeds, %true is returned; otherwise, @base_work is issued directly | |
302 | * and %false is returned. In the latter case, the caller is required to | |
303 | * wait for @base_work's completion using wb_wait_for_single_work(). | |
304 | * | |
305 | * A clone is auto-freed on completion. @base_work never is. | |
306 | */ | |
307 | static bool wb_clone_and_queue_work(struct bdi_writeback *wb, | |
308 | struct wb_writeback_work *base_work) | |
309 | { | |
310 | struct wb_writeback_work *work; | |
311 | ||
312 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
313 | if (work) { | |
314 | *work = *base_work; | |
315 | work->auto_free = 1; | |
316 | work->single_wait = 0; | |
317 | } else { | |
318 | work = base_work; | |
319 | work->auto_free = 0; | |
320 | work->single_wait = 1; | |
321 | } | |
322 | work->single_done = 0; | |
323 | wb_queue_work(wb, work); | |
324 | return work != base_work; | |
325 | } | |
326 | ||
327 | /** | |
328 | * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi | |
329 | * @bdi: target backing_dev_info | |
330 | * @base_work: wb_writeback_work to issue | |
331 | * @skip_if_busy: skip wb's which already have writeback in progress | |
332 | * | |
333 | * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which | |
334 | * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's | |
335 | * distributed to the busy wbs according to each wb's proportion in the | |
336 | * total active write bandwidth of @bdi. | |
337 | */ | |
338 | static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, | |
339 | struct wb_writeback_work *base_work, | |
340 | bool skip_if_busy) | |
341 | { | |
342 | long nr_pages = base_work->nr_pages; | |
343 | int next_blkcg_id = 0; | |
344 | struct bdi_writeback *wb; | |
345 | struct wb_iter iter; | |
346 | ||
347 | might_sleep(); | |
348 | ||
349 | if (!bdi_has_dirty_io(bdi)) | |
350 | return; | |
351 | restart: | |
352 | rcu_read_lock(); | |
353 | bdi_for_each_wb(wb, bdi, &iter, next_blkcg_id) { | |
354 | if (!wb_has_dirty_io(wb) || | |
355 | (skip_if_busy && writeback_in_progress(wb))) | |
356 | continue; | |
357 | ||
358 | base_work->nr_pages = wb_split_bdi_pages(wb, nr_pages); | |
359 | if (!wb_clone_and_queue_work(wb, base_work)) { | |
360 | next_blkcg_id = wb->blkcg_css->id + 1; | |
361 | rcu_read_unlock(); | |
362 | wb_wait_for_single_work(bdi, base_work); | |
363 | goto restart; | |
364 | } | |
365 | } | |
366 | rcu_read_unlock(); | |
367 | } | |
368 | ||
f2b65121 TH |
369 | #else /* CONFIG_CGROUP_WRITEBACK */ |
370 | ||
371 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) | |
372 | { | |
373 | return nr_pages; | |
374 | } | |
375 | ||
db125360 TH |
376 | static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, |
377 | struct wb_writeback_work *base_work, | |
378 | bool skip_if_busy) | |
379 | { | |
380 | might_sleep(); | |
381 | ||
382 | if (bdi_has_dirty_io(bdi) && | |
383 | (!skip_if_busy || !writeback_in_progress(&bdi->wb))) { | |
384 | base_work->auto_free = 0; | |
385 | base_work->single_wait = 0; | |
386 | base_work->single_done = 0; | |
387 | wb_queue_work(&bdi->wb, base_work); | |
388 | } | |
389 | } | |
390 | ||
703c2708 TH |
391 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
392 | ||
c00ddad3 TH |
393 | void wb_start_writeback(struct bdi_writeback *wb, long nr_pages, |
394 | bool range_cyclic, enum wb_reason reason) | |
b6e51316 | 395 | { |
c00ddad3 TH |
396 | struct wb_writeback_work *work; |
397 | ||
398 | if (!wb_has_dirty_io(wb)) | |
399 | return; | |
400 | ||
401 | /* | |
402 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
403 | * wakeup the thread for old dirty data writeback | |
404 | */ | |
405 | work = kzalloc(sizeof(*work), GFP_ATOMIC); | |
406 | if (!work) { | |
407 | trace_writeback_nowork(wb->bdi); | |
408 | wb_wakeup(wb); | |
409 | return; | |
410 | } | |
411 | ||
412 | work->sync_mode = WB_SYNC_NONE; | |
413 | work->nr_pages = nr_pages; | |
414 | work->range_cyclic = range_cyclic; | |
415 | work->reason = reason; | |
ac7b19a3 | 416 | work->auto_free = 1; |
c00ddad3 TH |
417 | |
418 | wb_queue_work(wb, work); | |
c5444198 | 419 | } |
d3ddec76 | 420 | |
c5444198 | 421 | /** |
9ecf4866 TH |
422 | * wb_start_background_writeback - start background writeback |
423 | * @wb: bdi_writback to write from | |
c5444198 CH |
424 | * |
425 | * Description: | |
6585027a | 426 | * This makes sure WB_SYNC_NONE background writeback happens. When |
9ecf4866 | 427 | * this function returns, it is only guaranteed that for given wb |
6585027a JK |
428 | * some IO is happening if we are over background dirty threshold. |
429 | * Caller need not hold sb s_umount semaphore. | |
c5444198 | 430 | */ |
9ecf4866 | 431 | void wb_start_background_writeback(struct bdi_writeback *wb) |
c5444198 | 432 | { |
6585027a JK |
433 | /* |
434 | * We just wake up the flusher thread. It will perform background | |
435 | * writeback as soon as there is no other work to do. | |
436 | */ | |
9ecf4866 TH |
437 | trace_writeback_wake_background(wb->bdi); |
438 | wb_wakeup(wb); | |
1da177e4 LT |
439 | } |
440 | ||
a66979ab DC |
441 | /* |
442 | * Remove the inode from the writeback list it is on. | |
443 | */ | |
444 | void inode_wb_list_del(struct inode *inode) | |
445 | { | |
52ebea74 | 446 | struct bdi_writeback *wb = inode_to_wb(inode); |
f758eeab | 447 | |
52ebea74 | 448 | spin_lock(&wb->list_lock); |
d6c10f1f | 449 | inode_wb_list_del_locked(inode, wb); |
52ebea74 | 450 | spin_unlock(&wb->list_lock); |
a66979ab DC |
451 | } |
452 | ||
6610a0bc AM |
453 | /* |
454 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
455 | * furthest end of its superblock's dirty-inode list. | |
456 | * | |
457 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 458 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
459 | * the case then the inode must have been redirtied while it was being written |
460 | * out and we don't reset its dirtied_when. | |
461 | */ | |
f758eeab | 462 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 463 | { |
03ba3782 | 464 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 465 | struct inode *tail; |
6610a0bc | 466 | |
7ccf19a8 | 467 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 468 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
469 | inode->dirtied_when = jiffies; |
470 | } | |
d6c10f1f | 471 | inode_wb_list_move_locked(inode, wb, &wb->b_dirty); |
6610a0bc AM |
472 | } |
473 | ||
c986d1e2 | 474 | /* |
66f3b8e2 | 475 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 476 | */ |
f758eeab | 477 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 478 | { |
d6c10f1f | 479 | inode_wb_list_move_locked(inode, wb, &wb->b_more_io); |
c986d1e2 AM |
480 | } |
481 | ||
1c0eeaf5 JE |
482 | static void inode_sync_complete(struct inode *inode) |
483 | { | |
365b94ae | 484 | inode->i_state &= ~I_SYNC; |
4eff96dd JK |
485 | /* If inode is clean an unused, put it into LRU now... */ |
486 | inode_add_lru(inode); | |
365b94ae | 487 | /* Waiters must see I_SYNC cleared before being woken up */ |
1c0eeaf5 JE |
488 | smp_mb(); |
489 | wake_up_bit(&inode->i_state, __I_SYNC); | |
490 | } | |
491 | ||
d2caa3c5 JL |
492 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
493 | { | |
494 | bool ret = time_after(inode->dirtied_when, t); | |
495 | #ifndef CONFIG_64BIT | |
496 | /* | |
497 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
498 | * It _appears_ to be in the future, but is actually in distant past. | |
499 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 500 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
501 | */ |
502 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
503 | #endif | |
504 | return ret; | |
505 | } | |
506 | ||
0ae45f63 TT |
507 | #define EXPIRE_DIRTY_ATIME 0x0001 |
508 | ||
2c136579 | 509 | /* |
0e2f2b23 | 510 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
697e6fed | 511 | * @delaying_queue to @dispatch_queue. |
2c136579 | 512 | */ |
e84d0a4f | 513 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 514 | struct list_head *dispatch_queue, |
0ae45f63 | 515 | int flags, |
ad4e38dd | 516 | struct wb_writeback_work *work) |
2c136579 | 517 | { |
0ae45f63 TT |
518 | unsigned long *older_than_this = NULL; |
519 | unsigned long expire_time; | |
5c03449d SL |
520 | LIST_HEAD(tmp); |
521 | struct list_head *pos, *node; | |
cf137307 | 522 | struct super_block *sb = NULL; |
5c03449d | 523 | struct inode *inode; |
cf137307 | 524 | int do_sb_sort = 0; |
e84d0a4f | 525 | int moved = 0; |
5c03449d | 526 | |
0ae45f63 TT |
527 | if ((flags & EXPIRE_DIRTY_ATIME) == 0) |
528 | older_than_this = work->older_than_this; | |
a2f48706 TT |
529 | else if (!work->for_sync) { |
530 | expire_time = jiffies - (dirtytime_expire_interval * HZ); | |
0ae45f63 TT |
531 | older_than_this = &expire_time; |
532 | } | |
2c136579 | 533 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 534 | inode = wb_inode(delaying_queue->prev); |
0ae45f63 TT |
535 | if (older_than_this && |
536 | inode_dirtied_after(inode, *older_than_this)) | |
2c136579 | 537 | break; |
a8855990 JK |
538 | list_move(&inode->i_wb_list, &tmp); |
539 | moved++; | |
0ae45f63 TT |
540 | if (flags & EXPIRE_DIRTY_ATIME) |
541 | set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); | |
a8855990 JK |
542 | if (sb_is_blkdev_sb(inode->i_sb)) |
543 | continue; | |
cf137307 JA |
544 | if (sb && sb != inode->i_sb) |
545 | do_sb_sort = 1; | |
546 | sb = inode->i_sb; | |
5c03449d SL |
547 | } |
548 | ||
cf137307 JA |
549 | /* just one sb in list, splice to dispatch_queue and we're done */ |
550 | if (!do_sb_sort) { | |
551 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 552 | goto out; |
cf137307 JA |
553 | } |
554 | ||
5c03449d SL |
555 | /* Move inodes from one superblock together */ |
556 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 557 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 558 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 559 | inode = wb_inode(pos); |
5c03449d | 560 | if (inode->i_sb == sb) |
7ccf19a8 | 561 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 562 | } |
2c136579 | 563 | } |
e84d0a4f WF |
564 | out: |
565 | return moved; | |
2c136579 FW |
566 | } |
567 | ||
568 | /* | |
569 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
570 | * Before |
571 | * newly dirtied b_dirty b_io b_more_io | |
572 | * =============> gf edc BA | |
573 | * After | |
574 | * newly dirtied b_dirty b_io b_more_io | |
575 | * =============> g fBAedc | |
576 | * | | |
577 | * +--> dequeue for IO | |
2c136579 | 578 | */ |
ad4e38dd | 579 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
66f3b8e2 | 580 | { |
e84d0a4f | 581 | int moved; |
0ae45f63 | 582 | |
f758eeab | 583 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 584 | list_splice_init(&wb->b_more_io, &wb->b_io); |
0ae45f63 TT |
585 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); |
586 | moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, | |
587 | EXPIRE_DIRTY_ATIME, work); | |
d6c10f1f TH |
588 | if (moved) |
589 | wb_io_lists_populated(wb); | |
ad4e38dd | 590 | trace_writeback_queue_io(wb, work, moved); |
66f3b8e2 JA |
591 | } |
592 | ||
a9185b41 | 593 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 594 | { |
9fb0a7da TH |
595 | int ret; |
596 | ||
597 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
598 | trace_writeback_write_inode_start(inode, wbc); | |
599 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
600 | trace_writeback_write_inode(inode, wbc); | |
601 | return ret; | |
602 | } | |
03ba3782 | 603 | return 0; |
08d8e974 | 604 | } |
08d8e974 | 605 | |
1da177e4 | 606 | /* |
169ebd90 JK |
607 | * Wait for writeback on an inode to complete. Called with i_lock held. |
608 | * Caller must make sure inode cannot go away when we drop i_lock. | |
01c03194 | 609 | */ |
169ebd90 JK |
610 | static void __inode_wait_for_writeback(struct inode *inode) |
611 | __releases(inode->i_lock) | |
612 | __acquires(inode->i_lock) | |
01c03194 CH |
613 | { |
614 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
615 | wait_queue_head_t *wqh; | |
616 | ||
617 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
618 | while (inode->i_state & I_SYNC) { |
619 | spin_unlock(&inode->i_lock); | |
74316201 N |
620 | __wait_on_bit(wqh, &wq, bit_wait, |
621 | TASK_UNINTERRUPTIBLE); | |
250df6ed | 622 | spin_lock(&inode->i_lock); |
58a9d3d8 | 623 | } |
01c03194 CH |
624 | } |
625 | ||
169ebd90 JK |
626 | /* |
627 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
628 | */ | |
629 | void inode_wait_for_writeback(struct inode *inode) | |
630 | { | |
631 | spin_lock(&inode->i_lock); | |
632 | __inode_wait_for_writeback(inode); | |
633 | spin_unlock(&inode->i_lock); | |
634 | } | |
635 | ||
636 | /* | |
637 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
638 | * held and drops it. It is aimed for callers not holding any inode reference | |
639 | * so once i_lock is dropped, inode can go away. | |
640 | */ | |
641 | static void inode_sleep_on_writeback(struct inode *inode) | |
642 | __releases(inode->i_lock) | |
643 | { | |
644 | DEFINE_WAIT(wait); | |
645 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
646 | int sleep; | |
647 | ||
648 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
649 | sleep = inode->i_state & I_SYNC; | |
650 | spin_unlock(&inode->i_lock); | |
651 | if (sleep) | |
652 | schedule(); | |
653 | finish_wait(wqh, &wait); | |
654 | } | |
655 | ||
ccb26b5a JK |
656 | /* |
657 | * Find proper writeback list for the inode depending on its current state and | |
658 | * possibly also change of its state while we were doing writeback. Here we | |
659 | * handle things such as livelock prevention or fairness of writeback among | |
660 | * inodes. This function can be called only by flusher thread - noone else | |
661 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
662 | * thread's back can have unexpected consequences. | |
663 | */ | |
664 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
665 | struct writeback_control *wbc) | |
666 | { | |
667 | if (inode->i_state & I_FREEING) | |
668 | return; | |
669 | ||
670 | /* | |
671 | * Sync livelock prevention. Each inode is tagged and synced in one | |
672 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
673 | * the dirty time to prevent enqueue and sync it again. | |
674 | */ | |
675 | if ((inode->i_state & I_DIRTY) && | |
676 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
677 | inode->dirtied_when = jiffies; | |
678 | ||
4f8ad655 JK |
679 | if (wbc->pages_skipped) { |
680 | /* | |
681 | * writeback is not making progress due to locked | |
682 | * buffers. Skip this inode for now. | |
683 | */ | |
684 | redirty_tail(inode, wb); | |
685 | return; | |
686 | } | |
687 | ||
ccb26b5a JK |
688 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
689 | /* | |
690 | * We didn't write back all the pages. nfs_writepages() | |
691 | * sometimes bales out without doing anything. | |
692 | */ | |
693 | if (wbc->nr_to_write <= 0) { | |
694 | /* Slice used up. Queue for next turn. */ | |
695 | requeue_io(inode, wb); | |
696 | } else { | |
697 | /* | |
698 | * Writeback blocked by something other than | |
699 | * congestion. Delay the inode for some time to | |
700 | * avoid spinning on the CPU (100% iowait) | |
701 | * retrying writeback of the dirty page/inode | |
702 | * that cannot be performed immediately. | |
703 | */ | |
704 | redirty_tail(inode, wb); | |
705 | } | |
706 | } else if (inode->i_state & I_DIRTY) { | |
707 | /* | |
708 | * Filesystems can dirty the inode during writeback operations, | |
709 | * such as delayed allocation during submission or metadata | |
710 | * updates after data IO completion. | |
711 | */ | |
712 | redirty_tail(inode, wb); | |
0ae45f63 | 713 | } else if (inode->i_state & I_DIRTY_TIME) { |
a2f48706 | 714 | inode->dirtied_when = jiffies; |
d6c10f1f | 715 | inode_wb_list_move_locked(inode, wb, &wb->b_dirty_time); |
ccb26b5a JK |
716 | } else { |
717 | /* The inode is clean. Remove from writeback lists. */ | |
d6c10f1f | 718 | inode_wb_list_del_locked(inode, wb); |
ccb26b5a JK |
719 | } |
720 | } | |
721 | ||
01c03194 | 722 | /* |
4f8ad655 JK |
723 | * Write out an inode and its dirty pages. Do not update the writeback list |
724 | * linkage. That is left to the caller. The caller is also responsible for | |
725 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. | |
1da177e4 LT |
726 | */ |
727 | static int | |
cd8ed2a4 | 728 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 729 | { |
1da177e4 | 730 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 731 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 732 | unsigned dirty; |
1da177e4 LT |
733 | int ret; |
734 | ||
4f8ad655 | 735 | WARN_ON(!(inode->i_state & I_SYNC)); |
1da177e4 | 736 | |
9fb0a7da TH |
737 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
738 | ||
1da177e4 LT |
739 | ret = do_writepages(mapping, wbc); |
740 | ||
26821ed4 CH |
741 | /* |
742 | * Make sure to wait on the data before writing out the metadata. | |
743 | * This is important for filesystems that modify metadata on data | |
7747bd4b DC |
744 | * I/O completion. We don't do it for sync(2) writeback because it has a |
745 | * separate, external IO completion path and ->sync_fs for guaranteeing | |
746 | * inode metadata is written back correctly. | |
26821ed4 | 747 | */ |
7747bd4b | 748 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
26821ed4 | 749 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
750 | if (ret == 0) |
751 | ret = err; | |
752 | } | |
753 | ||
5547e8aa DM |
754 | /* |
755 | * Some filesystems may redirty the inode during the writeback | |
756 | * due to delalloc, clear dirty metadata flags right before | |
757 | * write_inode() | |
758 | */ | |
250df6ed | 759 | spin_lock(&inode->i_lock); |
9c6ac78e | 760 | |
5547e8aa | 761 | dirty = inode->i_state & I_DIRTY; |
a2f48706 TT |
762 | if (inode->i_state & I_DIRTY_TIME) { |
763 | if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
764 | unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || | |
765 | unlikely(time_after(jiffies, | |
766 | (inode->dirtied_time_when + | |
767 | dirtytime_expire_interval * HZ)))) { | |
768 | dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; | |
769 | trace_writeback_lazytime(inode); | |
770 | } | |
771 | } else | |
772 | inode->i_state &= ~I_DIRTY_TIME_EXPIRED; | |
0ae45f63 | 773 | inode->i_state &= ~dirty; |
9c6ac78e TH |
774 | |
775 | /* | |
776 | * Paired with smp_mb() in __mark_inode_dirty(). This allows | |
777 | * __mark_inode_dirty() to test i_state without grabbing i_lock - | |
778 | * either they see the I_DIRTY bits cleared or we see the dirtied | |
779 | * inode. | |
780 | * | |
781 | * I_DIRTY_PAGES is always cleared together above even if @mapping | |
782 | * still has dirty pages. The flag is reinstated after smp_mb() if | |
783 | * necessary. This guarantees that either __mark_inode_dirty() | |
784 | * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. | |
785 | */ | |
786 | smp_mb(); | |
787 | ||
788 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
789 | inode->i_state |= I_DIRTY_PAGES; | |
790 | ||
250df6ed | 791 | spin_unlock(&inode->i_lock); |
9c6ac78e | 792 | |
0ae45f63 TT |
793 | if (dirty & I_DIRTY_TIME) |
794 | mark_inode_dirty_sync(inode); | |
26821ed4 | 795 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
0ae45f63 | 796 | if (dirty & ~I_DIRTY_PAGES) { |
a9185b41 | 797 | int err = write_inode(inode, wbc); |
1da177e4 LT |
798 | if (ret == 0) |
799 | ret = err; | |
800 | } | |
4f8ad655 JK |
801 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
802 | return ret; | |
803 | } | |
804 | ||
805 | /* | |
806 | * Write out an inode's dirty pages. Either the caller has an active reference | |
807 | * on the inode or the inode has I_WILL_FREE set. | |
808 | * | |
809 | * This function is designed to be called for writing back one inode which | |
810 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() | |
811 | * and does more profound writeback list handling in writeback_sb_inodes(). | |
812 | */ | |
813 | static int | |
814 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, | |
815 | struct writeback_control *wbc) | |
816 | { | |
817 | int ret = 0; | |
818 | ||
819 | spin_lock(&inode->i_lock); | |
820 | if (!atomic_read(&inode->i_count)) | |
821 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
822 | else | |
823 | WARN_ON(inode->i_state & I_WILL_FREE); | |
824 | ||
825 | if (inode->i_state & I_SYNC) { | |
826 | if (wbc->sync_mode != WB_SYNC_ALL) | |
827 | goto out; | |
828 | /* | |
169ebd90 JK |
829 | * It's a data-integrity sync. We must wait. Since callers hold |
830 | * inode reference or inode has I_WILL_FREE set, it cannot go | |
831 | * away under us. | |
4f8ad655 | 832 | */ |
169ebd90 | 833 | __inode_wait_for_writeback(inode); |
4f8ad655 JK |
834 | } |
835 | WARN_ON(inode->i_state & I_SYNC); | |
836 | /* | |
f9b0e058 JK |
837 | * Skip inode if it is clean and we have no outstanding writeback in |
838 | * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this | |
839 | * function since flusher thread may be doing for example sync in | |
840 | * parallel and if we move the inode, it could get skipped. So here we | |
841 | * make sure inode is on some writeback list and leave it there unless | |
842 | * we have completely cleaned the inode. | |
4f8ad655 | 843 | */ |
0ae45f63 | 844 | if (!(inode->i_state & I_DIRTY_ALL) && |
f9b0e058 JK |
845 | (wbc->sync_mode != WB_SYNC_ALL || |
846 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) | |
4f8ad655 JK |
847 | goto out; |
848 | inode->i_state |= I_SYNC; | |
849 | spin_unlock(&inode->i_lock); | |
850 | ||
cd8ed2a4 | 851 | ret = __writeback_single_inode(inode, wbc); |
1da177e4 | 852 | |
f758eeab | 853 | spin_lock(&wb->list_lock); |
250df6ed | 854 | spin_lock(&inode->i_lock); |
4f8ad655 JK |
855 | /* |
856 | * If inode is clean, remove it from writeback lists. Otherwise don't | |
857 | * touch it. See comment above for explanation. | |
858 | */ | |
0ae45f63 | 859 | if (!(inode->i_state & I_DIRTY_ALL)) |
d6c10f1f | 860 | inode_wb_list_del_locked(inode, wb); |
4f8ad655 | 861 | spin_unlock(&wb->list_lock); |
1c0eeaf5 | 862 | inode_sync_complete(inode); |
4f8ad655 JK |
863 | out: |
864 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
865 | return ret; |
866 | } | |
867 | ||
a88a341a | 868 | static long writeback_chunk_size(struct bdi_writeback *wb, |
1a12d8bd | 869 | struct wb_writeback_work *work) |
d46db3d5 WF |
870 | { |
871 | long pages; | |
872 | ||
873 | /* | |
874 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
875 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
876 | * here avoids calling into writeback_inodes_wb() more than once. | |
877 | * | |
878 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
879 | * | |
880 | * wb_writeback() | |
881 | * writeback_sb_inodes() <== called only once | |
882 | * write_cache_pages() <== called once for each inode | |
883 | * (quickly) tag currently dirty pages | |
884 | * (maybe slowly) sync all tagged pages | |
885 | */ | |
886 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
887 | pages = LONG_MAX; | |
1a12d8bd | 888 | else { |
a88a341a | 889 | pages = min(wb->avg_write_bandwidth / 2, |
dcc25ae7 | 890 | global_wb_domain.dirty_limit / DIRTY_SCOPE); |
1a12d8bd WF |
891 | pages = min(pages, work->nr_pages); |
892 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
893 | MIN_WRITEBACK_PAGES); | |
894 | } | |
d46db3d5 WF |
895 | |
896 | return pages; | |
897 | } | |
898 | ||
f11c9c5c ES |
899 | /* |
900 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 | 901 | * |
d46db3d5 | 902 | * Return the number of pages and/or inodes written. |
f11c9c5c | 903 | */ |
d46db3d5 WF |
904 | static long writeback_sb_inodes(struct super_block *sb, |
905 | struct bdi_writeback *wb, | |
906 | struct wb_writeback_work *work) | |
1da177e4 | 907 | { |
d46db3d5 WF |
908 | struct writeback_control wbc = { |
909 | .sync_mode = work->sync_mode, | |
910 | .tagged_writepages = work->tagged_writepages, | |
911 | .for_kupdate = work->for_kupdate, | |
912 | .for_background = work->for_background, | |
7747bd4b | 913 | .for_sync = work->for_sync, |
d46db3d5 WF |
914 | .range_cyclic = work->range_cyclic, |
915 | .range_start = 0, | |
916 | .range_end = LLONG_MAX, | |
917 | }; | |
918 | unsigned long start_time = jiffies; | |
919 | long write_chunk; | |
920 | long wrote = 0; /* count both pages and inodes */ | |
921 | ||
03ba3782 | 922 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 923 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
924 | |
925 | if (inode->i_sb != sb) { | |
d46db3d5 | 926 | if (work->sb) { |
edadfb10 CH |
927 | /* |
928 | * We only want to write back data for this | |
929 | * superblock, move all inodes not belonging | |
930 | * to it back onto the dirty list. | |
931 | */ | |
f758eeab | 932 | redirty_tail(inode, wb); |
edadfb10 CH |
933 | continue; |
934 | } | |
935 | ||
936 | /* | |
937 | * The inode belongs to a different superblock. | |
938 | * Bounce back to the caller to unpin this and | |
939 | * pin the next superblock. | |
940 | */ | |
d46db3d5 | 941 | break; |
edadfb10 CH |
942 | } |
943 | ||
9843b76a | 944 | /* |
331cbdee WL |
945 | * Don't bother with new inodes or inodes being freed, first |
946 | * kind does not need periodic writeout yet, and for the latter | |
9843b76a CH |
947 | * kind writeout is handled by the freer. |
948 | */ | |
250df6ed | 949 | spin_lock(&inode->i_lock); |
9843b76a | 950 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 951 | spin_unlock(&inode->i_lock); |
fcc5c222 | 952 | redirty_tail(inode, wb); |
7ef0d737 NP |
953 | continue; |
954 | } | |
cc1676d9 JK |
955 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
956 | /* | |
957 | * If this inode is locked for writeback and we are not | |
958 | * doing writeback-for-data-integrity, move it to | |
959 | * b_more_io so that writeback can proceed with the | |
960 | * other inodes on s_io. | |
961 | * | |
962 | * We'll have another go at writing back this inode | |
963 | * when we completed a full scan of b_io. | |
964 | */ | |
965 | spin_unlock(&inode->i_lock); | |
966 | requeue_io(inode, wb); | |
967 | trace_writeback_sb_inodes_requeue(inode); | |
968 | continue; | |
969 | } | |
f0d07b7f JK |
970 | spin_unlock(&wb->list_lock); |
971 | ||
4f8ad655 JK |
972 | /* |
973 | * We already requeued the inode if it had I_SYNC set and we | |
974 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
975 | * WB_SYNC_ALL case. | |
976 | */ | |
169ebd90 JK |
977 | if (inode->i_state & I_SYNC) { |
978 | /* Wait for I_SYNC. This function drops i_lock... */ | |
979 | inode_sleep_on_writeback(inode); | |
980 | /* Inode may be gone, start again */ | |
ead188f9 | 981 | spin_lock(&wb->list_lock); |
169ebd90 JK |
982 | continue; |
983 | } | |
4f8ad655 JK |
984 | inode->i_state |= I_SYNC; |
985 | spin_unlock(&inode->i_lock); | |
169ebd90 | 986 | |
a88a341a | 987 | write_chunk = writeback_chunk_size(wb, work); |
d46db3d5 WF |
988 | wbc.nr_to_write = write_chunk; |
989 | wbc.pages_skipped = 0; | |
250df6ed | 990 | |
169ebd90 JK |
991 | /* |
992 | * We use I_SYNC to pin the inode in memory. While it is set | |
993 | * evict_inode() will wait so the inode cannot be freed. | |
994 | */ | |
cd8ed2a4 | 995 | __writeback_single_inode(inode, &wbc); |
250df6ed | 996 | |
d46db3d5 WF |
997 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
998 | wrote += write_chunk - wbc.nr_to_write; | |
4f8ad655 JK |
999 | spin_lock(&wb->list_lock); |
1000 | spin_lock(&inode->i_lock); | |
0ae45f63 | 1001 | if (!(inode->i_state & I_DIRTY_ALL)) |
d46db3d5 | 1002 | wrote++; |
4f8ad655 JK |
1003 | requeue_inode(inode, wb, &wbc); |
1004 | inode_sync_complete(inode); | |
0f1b1fd8 | 1005 | spin_unlock(&inode->i_lock); |
169ebd90 | 1006 | cond_resched_lock(&wb->list_lock); |
d46db3d5 WF |
1007 | /* |
1008 | * bail out to wb_writeback() often enough to check | |
1009 | * background threshold and other termination conditions. | |
1010 | */ | |
1011 | if (wrote) { | |
1012 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
1013 | break; | |
1014 | if (work->nr_pages <= 0) | |
1015 | break; | |
8bc3be27 | 1016 | } |
1da177e4 | 1017 | } |
d46db3d5 | 1018 | return wrote; |
f11c9c5c ES |
1019 | } |
1020 | ||
d46db3d5 WF |
1021 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
1022 | struct wb_writeback_work *work) | |
f11c9c5c | 1023 | { |
d46db3d5 WF |
1024 | unsigned long start_time = jiffies; |
1025 | long wrote = 0; | |
38f21977 | 1026 | |
f11c9c5c | 1027 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 1028 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 1029 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 1030 | |
eb6ef3df | 1031 | if (!trylock_super(sb)) { |
0e995816 | 1032 | /* |
eb6ef3df | 1033 | * trylock_super() may fail consistently due to |
0e995816 WF |
1034 | * s_umount being grabbed by someone else. Don't use |
1035 | * requeue_io() to avoid busy retrying the inode/sb. | |
1036 | */ | |
1037 | redirty_tail(inode, wb); | |
edadfb10 | 1038 | continue; |
f11c9c5c | 1039 | } |
d46db3d5 | 1040 | wrote += writeback_sb_inodes(sb, wb, work); |
eb6ef3df | 1041 | up_read(&sb->s_umount); |
f11c9c5c | 1042 | |
d46db3d5 WF |
1043 | /* refer to the same tests at the end of writeback_sb_inodes */ |
1044 | if (wrote) { | |
1045 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
1046 | break; | |
1047 | if (work->nr_pages <= 0) | |
1048 | break; | |
1049 | } | |
f11c9c5c | 1050 | } |
66f3b8e2 | 1051 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 1052 | return wrote; |
66f3b8e2 JA |
1053 | } |
1054 | ||
7d9f073b | 1055 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
0e175a18 | 1056 | enum wb_reason reason) |
edadfb10 | 1057 | { |
d46db3d5 WF |
1058 | struct wb_writeback_work work = { |
1059 | .nr_pages = nr_pages, | |
1060 | .sync_mode = WB_SYNC_NONE, | |
1061 | .range_cyclic = 1, | |
0e175a18 | 1062 | .reason = reason, |
d46db3d5 | 1063 | }; |
edadfb10 | 1064 | |
f758eeab | 1065 | spin_lock(&wb->list_lock); |
424b351f | 1066 | if (list_empty(&wb->b_io)) |
ad4e38dd | 1067 | queue_io(wb, &work); |
d46db3d5 | 1068 | __writeback_inodes_wb(wb, &work); |
f758eeab | 1069 | spin_unlock(&wb->list_lock); |
edadfb10 | 1070 | |
d46db3d5 WF |
1071 | return nr_pages - work.nr_pages; |
1072 | } | |
03ba3782 | 1073 | |
03ba3782 JA |
1074 | /* |
1075 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 1076 | * |
03ba3782 JA |
1077 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
1078 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
1079 | * just walks the superblock inode list, writing back any inodes which are | |
1080 | * older than a specific point in time. | |
66f3b8e2 | 1081 | * |
03ba3782 JA |
1082 | * Try to run once per dirty_writeback_interval. But if a writeback event |
1083 | * takes longer than a dirty_writeback_interval interval, then leave a | |
1084 | * one-second gap. | |
66f3b8e2 | 1085 | * |
03ba3782 JA |
1086 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
1087 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 1088 | */ |
c4a77a6c | 1089 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 1090 | struct wb_writeback_work *work) |
66f3b8e2 | 1091 | { |
e98be2d5 | 1092 | unsigned long wb_start = jiffies; |
d46db3d5 | 1093 | long nr_pages = work->nr_pages; |
0dc83bd3 | 1094 | unsigned long oldest_jif; |
a5989bdc | 1095 | struct inode *inode; |
d46db3d5 | 1096 | long progress; |
66f3b8e2 | 1097 | |
0dc83bd3 JK |
1098 | oldest_jif = jiffies; |
1099 | work->older_than_this = &oldest_jif; | |
38f21977 | 1100 | |
e8dfc305 | 1101 | spin_lock(&wb->list_lock); |
03ba3782 JA |
1102 | for (;;) { |
1103 | /* | |
d3ddec76 | 1104 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 1105 | */ |
83ba7b07 | 1106 | if (work->nr_pages <= 0) |
03ba3782 | 1107 | break; |
66f3b8e2 | 1108 | |
aa373cf5 JK |
1109 | /* |
1110 | * Background writeout and kupdate-style writeback may | |
1111 | * run forever. Stop them if there is other work to do | |
1112 | * so that e.g. sync can proceed. They'll be restarted | |
1113 | * after the other works are all done. | |
1114 | */ | |
1115 | if ((work->for_background || work->for_kupdate) && | |
f0054bb1 | 1116 | !list_empty(&wb->work_list)) |
aa373cf5 JK |
1117 | break; |
1118 | ||
38f21977 | 1119 | /* |
d3ddec76 WF |
1120 | * For background writeout, stop when we are below the |
1121 | * background dirty threshold | |
38f21977 | 1122 | */ |
aa661bbe | 1123 | if (work->for_background && !wb_over_bg_thresh(wb)) |
03ba3782 | 1124 | break; |
38f21977 | 1125 | |
1bc36b64 JK |
1126 | /* |
1127 | * Kupdate and background works are special and we want to | |
1128 | * include all inodes that need writing. Livelock avoidance is | |
1129 | * handled by these works yielding to any other work so we are | |
1130 | * safe. | |
1131 | */ | |
ba9aa839 | 1132 | if (work->for_kupdate) { |
0dc83bd3 | 1133 | oldest_jif = jiffies - |
ba9aa839 | 1134 | msecs_to_jiffies(dirty_expire_interval * 10); |
1bc36b64 | 1135 | } else if (work->for_background) |
0dc83bd3 | 1136 | oldest_jif = jiffies; |
028c2dd1 | 1137 | |
d46db3d5 | 1138 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 1139 | if (list_empty(&wb->b_io)) |
ad4e38dd | 1140 | queue_io(wb, work); |
83ba7b07 | 1141 | if (work->sb) |
d46db3d5 | 1142 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 1143 | else |
d46db3d5 WF |
1144 | progress = __writeback_inodes_wb(wb, work); |
1145 | trace_writeback_written(wb->bdi, work); | |
028c2dd1 | 1146 | |
e98be2d5 | 1147 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
1148 | |
1149 | /* | |
e6fb6da2 WF |
1150 | * Did we write something? Try for more |
1151 | * | |
1152 | * Dirty inodes are moved to b_io for writeback in batches. | |
1153 | * The completion of the current batch does not necessarily | |
1154 | * mean the overall work is done. So we keep looping as long | |
1155 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 1156 | */ |
d46db3d5 | 1157 | if (progress) |
71fd05a8 JA |
1158 | continue; |
1159 | /* | |
e6fb6da2 | 1160 | * No more inodes for IO, bail |
71fd05a8 | 1161 | */ |
b7a2441f | 1162 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 1163 | break; |
71fd05a8 JA |
1164 | /* |
1165 | * Nothing written. Wait for some inode to | |
1166 | * become available for writeback. Otherwise | |
1167 | * we'll just busyloop. | |
1168 | */ | |
71fd05a8 | 1169 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 1170 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 1171 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 1172 | spin_lock(&inode->i_lock); |
f0d07b7f | 1173 | spin_unlock(&wb->list_lock); |
169ebd90 JK |
1174 | /* This function drops i_lock... */ |
1175 | inode_sleep_on_writeback(inode); | |
f0d07b7f | 1176 | spin_lock(&wb->list_lock); |
03ba3782 JA |
1177 | } |
1178 | } | |
e8dfc305 | 1179 | spin_unlock(&wb->list_lock); |
03ba3782 | 1180 | |
d46db3d5 | 1181 | return nr_pages - work->nr_pages; |
03ba3782 JA |
1182 | } |
1183 | ||
1184 | /* | |
83ba7b07 | 1185 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 1186 | */ |
f0054bb1 | 1187 | static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) |
03ba3782 | 1188 | { |
83ba7b07 | 1189 | struct wb_writeback_work *work = NULL; |
03ba3782 | 1190 | |
f0054bb1 TH |
1191 | spin_lock_bh(&wb->work_lock); |
1192 | if (!list_empty(&wb->work_list)) { | |
1193 | work = list_entry(wb->work_list.next, | |
83ba7b07 CH |
1194 | struct wb_writeback_work, list); |
1195 | list_del_init(&work->list); | |
03ba3782 | 1196 | } |
f0054bb1 | 1197 | spin_unlock_bh(&wb->work_lock); |
83ba7b07 | 1198 | return work; |
03ba3782 JA |
1199 | } |
1200 | ||
cdf01dd5 LT |
1201 | /* |
1202 | * Add in the number of potentially dirty inodes, because each inode | |
1203 | * write can dirty pagecache in the underlying blockdev. | |
1204 | */ | |
1205 | static unsigned long get_nr_dirty_pages(void) | |
1206 | { | |
1207 | return global_page_state(NR_FILE_DIRTY) + | |
1208 | global_page_state(NR_UNSTABLE_NFS) + | |
1209 | get_nr_dirty_inodes(); | |
1210 | } | |
1211 | ||
6585027a JK |
1212 | static long wb_check_background_flush(struct bdi_writeback *wb) |
1213 | { | |
aa661bbe | 1214 | if (wb_over_bg_thresh(wb)) { |
6585027a JK |
1215 | |
1216 | struct wb_writeback_work work = { | |
1217 | .nr_pages = LONG_MAX, | |
1218 | .sync_mode = WB_SYNC_NONE, | |
1219 | .for_background = 1, | |
1220 | .range_cyclic = 1, | |
0e175a18 | 1221 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
1222 | }; |
1223 | ||
1224 | return wb_writeback(wb, &work); | |
1225 | } | |
1226 | ||
1227 | return 0; | |
1228 | } | |
1229 | ||
03ba3782 JA |
1230 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
1231 | { | |
1232 | unsigned long expired; | |
1233 | long nr_pages; | |
1234 | ||
69b62d01 JA |
1235 | /* |
1236 | * When set to zero, disable periodic writeback | |
1237 | */ | |
1238 | if (!dirty_writeback_interval) | |
1239 | return 0; | |
1240 | ||
03ba3782 JA |
1241 | expired = wb->last_old_flush + |
1242 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
1243 | if (time_before(jiffies, expired)) | |
1244 | return 0; | |
1245 | ||
1246 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 1247 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 1248 | |
c4a77a6c | 1249 | if (nr_pages) { |
83ba7b07 | 1250 | struct wb_writeback_work work = { |
c4a77a6c JA |
1251 | .nr_pages = nr_pages, |
1252 | .sync_mode = WB_SYNC_NONE, | |
1253 | .for_kupdate = 1, | |
1254 | .range_cyclic = 1, | |
0e175a18 | 1255 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
1256 | }; |
1257 | ||
83ba7b07 | 1258 | return wb_writeback(wb, &work); |
c4a77a6c | 1259 | } |
03ba3782 JA |
1260 | |
1261 | return 0; | |
1262 | } | |
1263 | ||
1264 | /* | |
1265 | * Retrieve work items and do the writeback they describe | |
1266 | */ | |
25d130ba | 1267 | static long wb_do_writeback(struct bdi_writeback *wb) |
03ba3782 | 1268 | { |
83ba7b07 | 1269 | struct wb_writeback_work *work; |
c4a77a6c | 1270 | long wrote = 0; |
03ba3782 | 1271 | |
4452226e | 1272 | set_bit(WB_writeback_running, &wb->state); |
f0054bb1 | 1273 | while ((work = get_next_work_item(wb)) != NULL) { |
cc395d7f | 1274 | struct wb_completion *done = work->done; |
98754bf7 | 1275 | bool need_wake_up = false; |
03ba3782 | 1276 | |
f0054bb1 | 1277 | trace_writeback_exec(wb->bdi, work); |
455b2864 | 1278 | |
83ba7b07 | 1279 | wrote += wb_writeback(wb, work); |
03ba3782 | 1280 | |
98754bf7 TH |
1281 | if (work->single_wait) { |
1282 | WARN_ON_ONCE(work->auto_free); | |
1283 | /* paired w/ rmb in wb_wait_for_single_work() */ | |
1284 | smp_wmb(); | |
1285 | work->single_done = 1; | |
1286 | need_wake_up = true; | |
1287 | } else if (work->auto_free) { | |
83ba7b07 | 1288 | kfree(work); |
98754bf7 TH |
1289 | } |
1290 | ||
cc395d7f | 1291 | if (done && atomic_dec_and_test(&done->cnt)) |
98754bf7 TH |
1292 | need_wake_up = true; |
1293 | ||
1294 | if (need_wake_up) | |
cc395d7f | 1295 | wake_up_all(&wb->bdi->wb_waitq); |
03ba3782 JA |
1296 | } |
1297 | ||
1298 | /* | |
1299 | * Check for periodic writeback, kupdated() style | |
1300 | */ | |
1301 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 1302 | wrote += wb_check_background_flush(wb); |
4452226e | 1303 | clear_bit(WB_writeback_running, &wb->state); |
03ba3782 JA |
1304 | |
1305 | return wrote; | |
1306 | } | |
1307 | ||
1308 | /* | |
1309 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
839a8e86 | 1310 | * reschedules periodically and does kupdated style flushing. |
03ba3782 | 1311 | */ |
f0054bb1 | 1312 | void wb_workfn(struct work_struct *work) |
03ba3782 | 1313 | { |
839a8e86 TH |
1314 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
1315 | struct bdi_writeback, dwork); | |
03ba3782 JA |
1316 | long pages_written; |
1317 | ||
f0054bb1 | 1318 | set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); |
766f9164 | 1319 | current->flags |= PF_SWAPWRITE; |
455b2864 | 1320 | |
839a8e86 | 1321 | if (likely(!current_is_workqueue_rescuer() || |
4452226e | 1322 | !test_bit(WB_registered, &wb->state))) { |
6467716a | 1323 | /* |
f0054bb1 | 1324 | * The normal path. Keep writing back @wb until its |
839a8e86 | 1325 | * work_list is empty. Note that this path is also taken |
f0054bb1 | 1326 | * if @wb is shutting down even when we're running off the |
839a8e86 | 1327 | * rescuer as work_list needs to be drained. |
6467716a | 1328 | */ |
839a8e86 | 1329 | do { |
25d130ba | 1330 | pages_written = wb_do_writeback(wb); |
839a8e86 | 1331 | trace_writeback_pages_written(pages_written); |
f0054bb1 | 1332 | } while (!list_empty(&wb->work_list)); |
839a8e86 TH |
1333 | } else { |
1334 | /* | |
1335 | * bdi_wq can't get enough workers and we're running off | |
1336 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
1337 | * enough for efficient IO. | |
1338 | */ | |
f0054bb1 | 1339 | pages_written = writeback_inodes_wb(wb, 1024, |
839a8e86 | 1340 | WB_REASON_FORKER_THREAD); |
455b2864 | 1341 | trace_writeback_pages_written(pages_written); |
03ba3782 JA |
1342 | } |
1343 | ||
f0054bb1 | 1344 | if (!list_empty(&wb->work_list)) |
6ca738d6 DB |
1345 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
1346 | else if (wb_has_dirty_io(wb) && dirty_writeback_interval) | |
f0054bb1 | 1347 | wb_wakeup_delayed(wb); |
455b2864 | 1348 | |
839a8e86 | 1349 | current->flags &= ~PF_SWAPWRITE; |
03ba3782 JA |
1350 | } |
1351 | ||
1352 | /* | |
b8c2f347 CH |
1353 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
1354 | * the whole world. | |
03ba3782 | 1355 | */ |
0e175a18 | 1356 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
03ba3782 | 1357 | { |
b8c2f347 | 1358 | struct backing_dev_info *bdi; |
03ba3782 | 1359 | |
47df3dde JK |
1360 | if (!nr_pages) |
1361 | nr_pages = get_nr_dirty_pages(); | |
03ba3782 | 1362 | |
b8c2f347 | 1363 | rcu_read_lock(); |
f2b65121 TH |
1364 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
1365 | struct bdi_writeback *wb; | |
1366 | struct wb_iter iter; | |
1367 | ||
1368 | if (!bdi_has_dirty_io(bdi)) | |
1369 | continue; | |
1370 | ||
1371 | bdi_for_each_wb(wb, bdi, &iter, 0) | |
1372 | wb_start_writeback(wb, wb_split_bdi_pages(wb, nr_pages), | |
1373 | false, reason); | |
1374 | } | |
cfc4ba53 | 1375 | rcu_read_unlock(); |
1da177e4 LT |
1376 | } |
1377 | ||
a2f48706 TT |
1378 | /* |
1379 | * Wake up bdi's periodically to make sure dirtytime inodes gets | |
1380 | * written back periodically. We deliberately do *not* check the | |
1381 | * b_dirtytime list in wb_has_dirty_io(), since this would cause the | |
1382 | * kernel to be constantly waking up once there are any dirtytime | |
1383 | * inodes on the system. So instead we define a separate delayed work | |
1384 | * function which gets called much more rarely. (By default, only | |
1385 | * once every 12 hours.) | |
1386 | * | |
1387 | * If there is any other write activity going on in the file system, | |
1388 | * this function won't be necessary. But if the only thing that has | |
1389 | * happened on the file system is a dirtytime inode caused by an atime | |
1390 | * update, we need this infrastructure below to make sure that inode | |
1391 | * eventually gets pushed out to disk. | |
1392 | */ | |
1393 | static void wakeup_dirtytime_writeback(struct work_struct *w); | |
1394 | static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); | |
1395 | ||
1396 | static void wakeup_dirtytime_writeback(struct work_struct *w) | |
1397 | { | |
1398 | struct backing_dev_info *bdi; | |
1399 | ||
1400 | rcu_read_lock(); | |
1401 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { | |
001fe6f6 TH |
1402 | struct bdi_writeback *wb; |
1403 | struct wb_iter iter; | |
1404 | ||
1405 | bdi_for_each_wb(wb, bdi, &iter, 0) | |
1406 | if (!list_empty(&bdi->wb.b_dirty_time)) | |
1407 | wb_wakeup(&bdi->wb); | |
a2f48706 TT |
1408 | } |
1409 | rcu_read_unlock(); | |
1410 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
1411 | } | |
1412 | ||
1413 | static int __init start_dirtytime_writeback(void) | |
1414 | { | |
1415 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
1416 | return 0; | |
1417 | } | |
1418 | __initcall(start_dirtytime_writeback); | |
1419 | ||
1efff914 TT |
1420 | int dirtytime_interval_handler(struct ctl_table *table, int write, |
1421 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1422 | { | |
1423 | int ret; | |
1424 | ||
1425 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
1426 | if (ret == 0 && write) | |
1427 | mod_delayed_work(system_wq, &dirtytime_work, 0); | |
1428 | return ret; | |
1429 | } | |
1430 | ||
03ba3782 JA |
1431 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1432 | { | |
1433 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1434 | struct dentry *dentry; | |
1435 | const char *name = "?"; | |
1436 | ||
1437 | dentry = d_find_alias(inode); | |
1438 | if (dentry) { | |
1439 | spin_lock(&dentry->d_lock); | |
1440 | name = (const char *) dentry->d_name.name; | |
1441 | } | |
1442 | printk(KERN_DEBUG | |
1443 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1444 | current->comm, task_pid_nr(current), inode->i_ino, | |
1445 | name, inode->i_sb->s_id); | |
1446 | if (dentry) { | |
1447 | spin_unlock(&dentry->d_lock); | |
1448 | dput(dentry); | |
1449 | } | |
1450 | } | |
1451 | } | |
1452 | ||
1453 | /** | |
1454 | * __mark_inode_dirty - internal function | |
1455 | * @inode: inode to mark | |
1456 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1457 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1458 | * mark_inode_dirty_sync. | |
1da177e4 | 1459 | * |
03ba3782 JA |
1460 | * Put the inode on the super block's dirty list. |
1461 | * | |
1462 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1463 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1464 | * If it was not hashed, it will never be added to the dirty list | |
1465 | * even if it is later hashed, as it will have been marked dirty already. | |
1466 | * | |
1467 | * In short, make sure you hash any inodes _before_ you start marking | |
1468 | * them dirty. | |
1da177e4 | 1469 | * |
03ba3782 JA |
1470 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1471 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1472 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1473 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1474 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1475 | * blockdev inode. | |
1da177e4 | 1476 | */ |
0ae45f63 | 1477 | #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) |
03ba3782 | 1478 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1479 | { |
03ba3782 | 1480 | struct super_block *sb = inode->i_sb; |
0ae45f63 TT |
1481 | int dirtytime; |
1482 | ||
1483 | trace_writeback_mark_inode_dirty(inode, flags); | |
1da177e4 | 1484 | |
03ba3782 JA |
1485 | /* |
1486 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1487 | * dirty the inode itself | |
1488 | */ | |
0ae45f63 | 1489 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { |
9fb0a7da TH |
1490 | trace_writeback_dirty_inode_start(inode, flags); |
1491 | ||
03ba3782 | 1492 | if (sb->s_op->dirty_inode) |
aa385729 | 1493 | sb->s_op->dirty_inode(inode, flags); |
9fb0a7da TH |
1494 | |
1495 | trace_writeback_dirty_inode(inode, flags); | |
03ba3782 | 1496 | } |
0ae45f63 TT |
1497 | if (flags & I_DIRTY_INODE) |
1498 | flags &= ~I_DIRTY_TIME; | |
1499 | dirtytime = flags & I_DIRTY_TIME; | |
03ba3782 JA |
1500 | |
1501 | /* | |
9c6ac78e TH |
1502 | * Paired with smp_mb() in __writeback_single_inode() for the |
1503 | * following lockless i_state test. See there for details. | |
03ba3782 JA |
1504 | */ |
1505 | smp_mb(); | |
1506 | ||
0ae45f63 TT |
1507 | if (((inode->i_state & flags) == flags) || |
1508 | (dirtytime && (inode->i_state & I_DIRTY_INODE))) | |
03ba3782 JA |
1509 | return; |
1510 | ||
1511 | if (unlikely(block_dump)) | |
1512 | block_dump___mark_inode_dirty(inode); | |
1513 | ||
250df6ed | 1514 | spin_lock(&inode->i_lock); |
0ae45f63 TT |
1515 | if (dirtytime && (inode->i_state & I_DIRTY_INODE)) |
1516 | goto out_unlock_inode; | |
03ba3782 JA |
1517 | if ((inode->i_state & flags) != flags) { |
1518 | const int was_dirty = inode->i_state & I_DIRTY; | |
1519 | ||
52ebea74 TH |
1520 | inode_attach_wb(inode, NULL); |
1521 | ||
0ae45f63 TT |
1522 | if (flags & I_DIRTY_INODE) |
1523 | inode->i_state &= ~I_DIRTY_TIME; | |
03ba3782 JA |
1524 | inode->i_state |= flags; |
1525 | ||
1526 | /* | |
1527 | * If the inode is being synced, just update its dirty state. | |
1528 | * The unlocker will place the inode on the appropriate | |
1529 | * superblock list, based upon its state. | |
1530 | */ | |
1531 | if (inode->i_state & I_SYNC) | |
250df6ed | 1532 | goto out_unlock_inode; |
03ba3782 JA |
1533 | |
1534 | /* | |
1535 | * Only add valid (hashed) inodes to the superblock's | |
1536 | * dirty list. Add blockdev inodes as well. | |
1537 | */ | |
1538 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1539 | if (inode_unhashed(inode)) |
250df6ed | 1540 | goto out_unlock_inode; |
03ba3782 | 1541 | } |
a4ffdde6 | 1542 | if (inode->i_state & I_FREEING) |
250df6ed | 1543 | goto out_unlock_inode; |
03ba3782 JA |
1544 | |
1545 | /* | |
1546 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1547 | * reposition it (that would break b_dirty time-ordering). | |
1548 | */ | |
1549 | if (!was_dirty) { | |
0747259d | 1550 | struct bdi_writeback *wb = inode_to_wb(inode); |
d6c10f1f | 1551 | struct list_head *dirty_list; |
a66979ab | 1552 | bool wakeup_bdi = false; |
253c34e9 | 1553 | |
146d7009 | 1554 | spin_unlock(&inode->i_lock); |
0747259d | 1555 | spin_lock(&wb->list_lock); |
253c34e9 | 1556 | |
0747259d TH |
1557 | WARN(bdi_cap_writeback_dirty(wb->bdi) && |
1558 | !test_bit(WB_registered, &wb->state), | |
1559 | "bdi-%s not registered\n", wb->bdi->name); | |
03ba3782 JA |
1560 | |
1561 | inode->dirtied_when = jiffies; | |
a2f48706 TT |
1562 | if (dirtytime) |
1563 | inode->dirtied_time_when = jiffies; | |
d6c10f1f | 1564 | |
a2f48706 | 1565 | if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) |
0747259d | 1566 | dirty_list = &wb->b_dirty; |
a2f48706 | 1567 | else |
0747259d | 1568 | dirty_list = &wb->b_dirty_time; |
d6c10f1f | 1569 | |
0747259d | 1570 | wakeup_bdi = inode_wb_list_move_locked(inode, wb, |
d6c10f1f TH |
1571 | dirty_list); |
1572 | ||
0747259d | 1573 | spin_unlock(&wb->list_lock); |
0ae45f63 | 1574 | trace_writeback_dirty_inode_enqueue(inode); |
a66979ab | 1575 | |
d6c10f1f TH |
1576 | /* |
1577 | * If this is the first dirty inode for this bdi, | |
1578 | * we have to wake-up the corresponding bdi thread | |
1579 | * to make sure background write-back happens | |
1580 | * later. | |
1581 | */ | |
0747259d TH |
1582 | if (bdi_cap_writeback_dirty(wb->bdi) && wakeup_bdi) |
1583 | wb_wakeup_delayed(wb); | |
a66979ab | 1584 | return; |
1da177e4 | 1585 | } |
1da177e4 | 1586 | } |
250df6ed DC |
1587 | out_unlock_inode: |
1588 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1589 | |
03ba3782 JA |
1590 | } |
1591 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1592 | ||
b6e51316 | 1593 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1594 | { |
1595 | struct inode *inode, *old_inode = NULL; | |
1596 | ||
1597 | /* | |
1598 | * We need to be protected against the filesystem going from | |
1599 | * r/o to r/w or vice versa. | |
1600 | */ | |
b6e51316 | 1601 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1602 | |
55fa6091 | 1603 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1604 | |
1605 | /* | |
1606 | * Data integrity sync. Must wait for all pages under writeback, | |
1607 | * because there may have been pages dirtied before our sync | |
1608 | * call, but which had writeout started before we write it out. | |
1609 | * In which case, the inode may not be on the dirty list, but | |
1610 | * we still have to wait for that writeout. | |
1611 | */ | |
b6e51316 | 1612 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1613 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1614 | |
250df6ed DC |
1615 | spin_lock(&inode->i_lock); |
1616 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1617 | (mapping->nrpages == 0)) { | |
1618 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1619 | continue; |
250df6ed | 1620 | } |
03ba3782 | 1621 | __iget(inode); |
250df6ed | 1622 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1623 | spin_unlock(&inode_sb_list_lock); |
1624 | ||
03ba3782 | 1625 | /* |
55fa6091 DC |
1626 | * We hold a reference to 'inode' so it couldn't have been |
1627 | * removed from s_inodes list while we dropped the | |
1628 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1629 | * be holding the last reference and we cannot iput it under | |
1630 | * inode_sb_list_lock. So we keep the reference and iput it | |
1631 | * later. | |
03ba3782 JA |
1632 | */ |
1633 | iput(old_inode); | |
1634 | old_inode = inode; | |
1635 | ||
1636 | filemap_fdatawait(mapping); | |
1637 | ||
1638 | cond_resched(); | |
1639 | ||
55fa6091 | 1640 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1641 | } |
55fa6091 | 1642 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1643 | iput(old_inode); |
1da177e4 LT |
1644 | } |
1645 | ||
f30a7d0c TH |
1646 | static void __writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr, |
1647 | enum wb_reason reason, bool skip_if_busy) | |
1da177e4 | 1648 | { |
cc395d7f | 1649 | DEFINE_WB_COMPLETION_ONSTACK(done); |
83ba7b07 | 1650 | struct wb_writeback_work work = { |
6e6938b6 WF |
1651 | .sb = sb, |
1652 | .sync_mode = WB_SYNC_NONE, | |
1653 | .tagged_writepages = 1, | |
1654 | .done = &done, | |
1655 | .nr_pages = nr, | |
0e175a18 | 1656 | .reason = reason, |
3c4d7165 | 1657 | }; |
e7972912 | 1658 | struct backing_dev_info *bdi = sb->s_bdi; |
d8a8559c | 1659 | |
e7972912 | 1660 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
6eedc701 | 1661 | return; |
cf37e972 | 1662 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
f30a7d0c | 1663 | |
db125360 | 1664 | bdi_split_work_to_wbs(sb->s_bdi, &work, skip_if_busy); |
cc395d7f | 1665 | wb_wait_for_completion(bdi, &done); |
e913fc82 | 1666 | } |
f30a7d0c TH |
1667 | |
1668 | /** | |
1669 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block | |
1670 | * @sb: the superblock | |
1671 | * @nr: the number of pages to write | |
1672 | * @reason: reason why some writeback work initiated | |
1673 | * | |
1674 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1675 | * on how many (if any) will be written, and this function does not wait | |
1676 | * for IO completion of submitted IO. | |
1677 | */ | |
1678 | void writeback_inodes_sb_nr(struct super_block *sb, | |
1679 | unsigned long nr, | |
1680 | enum wb_reason reason) | |
1681 | { | |
1682 | __writeback_inodes_sb_nr(sb, nr, reason, false); | |
1683 | } | |
3259f8be CM |
1684 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1685 | ||
1686 | /** | |
1687 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1688 | * @sb: the superblock | |
786228ab | 1689 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1690 | * |
1691 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1692 | * on how many (if any) will be written, and this function does not wait | |
1693 | * for IO completion of submitted IO. | |
1694 | */ | |
0e175a18 | 1695 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1696 | { |
0e175a18 | 1697 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1698 | } |
0e3c9a22 | 1699 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1700 | |
17bd55d0 | 1701 | /** |
10ee27a0 | 1702 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway |
17bd55d0 | 1703 | * @sb: the superblock |
10ee27a0 MX |
1704 | * @nr: the number of pages to write |
1705 | * @reason: the reason of writeback | |
17bd55d0 | 1706 | * |
10ee27a0 | 1707 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. |
17bd55d0 ES |
1708 | * Returns 1 if writeback was started, 0 if not. |
1709 | */ | |
f30a7d0c TH |
1710 | bool try_to_writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr, |
1711 | enum wb_reason reason) | |
17bd55d0 | 1712 | { |
10ee27a0 | 1713 | if (!down_read_trylock(&sb->s_umount)) |
f30a7d0c | 1714 | return false; |
10ee27a0 | 1715 | |
f30a7d0c | 1716 | __writeback_inodes_sb_nr(sb, nr, reason, true); |
10ee27a0 | 1717 | up_read(&sb->s_umount); |
f30a7d0c | 1718 | return true; |
17bd55d0 | 1719 | } |
10ee27a0 | 1720 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); |
17bd55d0 | 1721 | |
3259f8be | 1722 | /** |
10ee27a0 | 1723 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
3259f8be | 1724 | * @sb: the superblock |
786228ab | 1725 | * @reason: reason why some writeback work was initiated |
3259f8be | 1726 | * |
10ee27a0 | 1727 | * Implement by try_to_writeback_inodes_sb_nr() |
3259f8be CM |
1728 | * Returns 1 if writeback was started, 0 if not. |
1729 | */ | |
f30a7d0c | 1730 | bool try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1731 | { |
10ee27a0 | 1732 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1733 | } |
10ee27a0 | 1734 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
3259f8be | 1735 | |
d8a8559c JA |
1736 | /** |
1737 | * sync_inodes_sb - sync sb inode pages | |
0dc83bd3 | 1738 | * @sb: the superblock |
d8a8559c JA |
1739 | * |
1740 | * This function writes and waits on any dirty inode belonging to this | |
0dc83bd3 | 1741 | * super_block. |
d8a8559c | 1742 | */ |
0dc83bd3 | 1743 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1744 | { |
cc395d7f | 1745 | DEFINE_WB_COMPLETION_ONSTACK(done); |
83ba7b07 | 1746 | struct wb_writeback_work work = { |
3c4d7165 CH |
1747 | .sb = sb, |
1748 | .sync_mode = WB_SYNC_ALL, | |
1749 | .nr_pages = LONG_MAX, | |
1750 | .range_cyclic = 0, | |
83ba7b07 | 1751 | .done = &done, |
0e175a18 | 1752 | .reason = WB_REASON_SYNC, |
7747bd4b | 1753 | .for_sync = 1, |
3c4d7165 | 1754 | }; |
e7972912 | 1755 | struct backing_dev_info *bdi = sb->s_bdi; |
3c4d7165 | 1756 | |
6eedc701 | 1757 | /* Nothing to do? */ |
e7972912 | 1758 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
6eedc701 | 1759 | return; |
cf37e972 CH |
1760 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1761 | ||
db125360 | 1762 | bdi_split_work_to_wbs(bdi, &work, false); |
cc395d7f | 1763 | wb_wait_for_completion(bdi, &done); |
83ba7b07 | 1764 | |
b6e51316 | 1765 | wait_sb_inodes(sb); |
1da177e4 | 1766 | } |
d8a8559c | 1767 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1768 | |
1da177e4 | 1769 | /** |
7f04c26d AA |
1770 | * write_inode_now - write an inode to disk |
1771 | * @inode: inode to write to disk | |
1772 | * @sync: whether the write should be synchronous or not | |
1773 | * | |
1774 | * This function commits an inode to disk immediately if it is dirty. This is | |
1775 | * primarily needed by knfsd. | |
1da177e4 | 1776 | * |
7f04c26d | 1777 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1778 | */ |
1da177e4 LT |
1779 | int write_inode_now(struct inode *inode, int sync) |
1780 | { | |
f758eeab | 1781 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1782 | struct writeback_control wbc = { |
1783 | .nr_to_write = LONG_MAX, | |
18914b18 | 1784 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1785 | .range_start = 0, |
1786 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1787 | }; |
1788 | ||
1789 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1790 | wbc.nr_to_write = 0; |
1da177e4 LT |
1791 | |
1792 | might_sleep(); | |
4f8ad655 | 1793 | return writeback_single_inode(inode, wb, &wbc); |
1da177e4 LT |
1794 | } |
1795 | EXPORT_SYMBOL(write_inode_now); | |
1796 | ||
1797 | /** | |
1798 | * sync_inode - write an inode and its pages to disk. | |
1799 | * @inode: the inode to sync | |
1800 | * @wbc: controls the writeback mode | |
1801 | * | |
1802 | * sync_inode() will write an inode and its pages to disk. It will also | |
1803 | * correctly update the inode on its superblock's dirty inode lists and will | |
1804 | * update inode->i_state. | |
1805 | * | |
1806 | * The caller must have a ref on the inode. | |
1807 | */ | |
1808 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1809 | { | |
4f8ad655 | 1810 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); |
1da177e4 LT |
1811 | } |
1812 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1813 | |
1814 | /** | |
c691b9d9 | 1815 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1816 | * @inode: the inode to sync |
1817 | * @wait: wait for I/O to complete. | |
1818 | * | |
c691b9d9 | 1819 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1820 | * |
1821 | * Note: only writes the actual inode, no associated data or other metadata. | |
1822 | */ | |
1823 | int sync_inode_metadata(struct inode *inode, int wait) | |
1824 | { | |
1825 | struct writeback_control wbc = { | |
1826 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1827 | .nr_to_write = 0, /* metadata-only */ | |
1828 | }; | |
1829 | ||
1830 | return sync_inode(inode, &wbc); | |
1831 | } | |
1832 | EXPORT_SYMBOL(sync_inode_metadata); |