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> |
21c6321f | 30 | #include <linux/memcontrol.h> |
07f3f05c | 31 | #include "internal.h" |
1da177e4 | 32 | |
bc31b86a WF |
33 | /* |
34 | * 4MB minimal write chunk size | |
35 | */ | |
36 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) | |
37 | ||
cc395d7f TH |
38 | struct wb_completion { |
39 | atomic_t cnt; | |
40 | }; | |
41 | ||
c4a77a6c JA |
42 | /* |
43 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
44 | */ | |
83ba7b07 | 45 | struct wb_writeback_work { |
c4a77a6c JA |
46 | long nr_pages; |
47 | struct super_block *sb; | |
0dc83bd3 | 48 | unsigned long *older_than_this; |
c4a77a6c | 49 | enum writeback_sync_modes sync_mode; |
6e6938b6 | 50 | unsigned int tagged_writepages:1; |
52957fe1 HS |
51 | unsigned int for_kupdate:1; |
52 | unsigned int range_cyclic:1; | |
53 | unsigned int for_background:1; | |
7747bd4b | 54 | unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ |
ac7b19a3 | 55 | unsigned int auto_free:1; /* free on completion */ |
98754bf7 TH |
56 | unsigned int single_wait:1; |
57 | unsigned int single_done:1; | |
0e175a18 | 58 | enum wb_reason reason; /* why was writeback initiated? */ |
c4a77a6c | 59 | |
8010c3b6 | 60 | struct list_head list; /* pending work list */ |
cc395d7f | 61 | struct wb_completion *done; /* set if the caller waits */ |
03ba3782 JA |
62 | }; |
63 | ||
cc395d7f TH |
64 | /* |
65 | * If one wants to wait for one or more wb_writeback_works, each work's | |
66 | * ->done should be set to a wb_completion defined using the following | |
67 | * macro. Once all work items are issued with wb_queue_work(), the caller | |
68 | * can wait for the completion of all using wb_wait_for_completion(). Work | |
69 | * items which are waited upon aren't freed automatically on completion. | |
70 | */ | |
71 | #define DEFINE_WB_COMPLETION_ONSTACK(cmpl) \ | |
72 | struct wb_completion cmpl = { \ | |
73 | .cnt = ATOMIC_INIT(1), \ | |
74 | } | |
75 | ||
76 | ||
a2f48706 TT |
77 | /* |
78 | * If an inode is constantly having its pages dirtied, but then the | |
79 | * updates stop dirtytime_expire_interval seconds in the past, it's | |
80 | * possible for the worst case time between when an inode has its | |
81 | * timestamps updated and when they finally get written out to be two | |
82 | * dirtytime_expire_intervals. We set the default to 12 hours (in | |
83 | * seconds), which means most of the time inodes will have their | |
84 | * timestamps written to disk after 12 hours, but in the worst case a | |
85 | * few inodes might not their timestamps updated for 24 hours. | |
86 | */ | |
87 | unsigned int dirtytime_expire_interval = 12 * 60 * 60; | |
88 | ||
7ccf19a8 NP |
89 | static inline struct inode *wb_inode(struct list_head *head) |
90 | { | |
91 | return list_entry(head, struct inode, i_wb_list); | |
92 | } | |
93 | ||
15eb77a0 WF |
94 | /* |
95 | * Include the creation of the trace points after defining the | |
96 | * wb_writeback_work structure and inline functions so that the definition | |
97 | * remains local to this file. | |
98 | */ | |
99 | #define CREATE_TRACE_POINTS | |
100 | #include <trace/events/writeback.h> | |
101 | ||
774016b2 SW |
102 | EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); |
103 | ||
d6c10f1f TH |
104 | static bool wb_io_lists_populated(struct bdi_writeback *wb) |
105 | { | |
106 | if (wb_has_dirty_io(wb)) { | |
107 | return false; | |
108 | } else { | |
109 | set_bit(WB_has_dirty_io, &wb->state); | |
95a46c65 | 110 | WARN_ON_ONCE(!wb->avg_write_bandwidth); |
766a9d6e TH |
111 | atomic_long_add(wb->avg_write_bandwidth, |
112 | &wb->bdi->tot_write_bandwidth); | |
d6c10f1f TH |
113 | return true; |
114 | } | |
115 | } | |
116 | ||
117 | static void wb_io_lists_depopulated(struct bdi_writeback *wb) | |
118 | { | |
119 | if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && | |
766a9d6e | 120 | list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { |
d6c10f1f | 121 | clear_bit(WB_has_dirty_io, &wb->state); |
95a46c65 TH |
122 | WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth, |
123 | &wb->bdi->tot_write_bandwidth) < 0); | |
766a9d6e | 124 | } |
d6c10f1f TH |
125 | } |
126 | ||
127 | /** | |
128 | * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list | |
129 | * @inode: inode to be moved | |
130 | * @wb: target bdi_writeback | |
131 | * @head: one of @wb->b_{dirty|io|more_io} | |
132 | * | |
133 | * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io. | |
134 | * Returns %true if @inode is the first occupant of the !dirty_time IO | |
135 | * lists; otherwise, %false. | |
136 | */ | |
137 | static bool inode_wb_list_move_locked(struct inode *inode, | |
138 | struct bdi_writeback *wb, | |
139 | struct list_head *head) | |
140 | { | |
141 | assert_spin_locked(&wb->list_lock); | |
142 | ||
143 | list_move(&inode->i_wb_list, head); | |
144 | ||
145 | /* dirty_time doesn't count as dirty_io until expiration */ | |
146 | if (head != &wb->b_dirty_time) | |
147 | return wb_io_lists_populated(wb); | |
148 | ||
149 | wb_io_lists_depopulated(wb); | |
150 | return false; | |
151 | } | |
152 | ||
153 | /** | |
154 | * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list | |
155 | * @inode: inode to be removed | |
156 | * @wb: bdi_writeback @inode is being removed from | |
157 | * | |
158 | * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and | |
159 | * clear %WB_has_dirty_io if all are empty afterwards. | |
160 | */ | |
161 | static void inode_wb_list_del_locked(struct inode *inode, | |
162 | struct bdi_writeback *wb) | |
163 | { | |
164 | assert_spin_locked(&wb->list_lock); | |
165 | ||
166 | list_del_init(&inode->i_wb_list); | |
167 | wb_io_lists_depopulated(wb); | |
168 | } | |
169 | ||
f0054bb1 | 170 | static void wb_wakeup(struct bdi_writeback *wb) |
5acda9d1 | 171 | { |
f0054bb1 TH |
172 | spin_lock_bh(&wb->work_lock); |
173 | if (test_bit(WB_registered, &wb->state)) | |
174 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
175 | spin_unlock_bh(&wb->work_lock); | |
5acda9d1 JK |
176 | } |
177 | ||
f0054bb1 TH |
178 | static void wb_queue_work(struct bdi_writeback *wb, |
179 | struct wb_writeback_work *work) | |
6585027a | 180 | { |
f0054bb1 | 181 | trace_writeback_queue(wb->bdi, work); |
6585027a | 182 | |
f0054bb1 | 183 | spin_lock_bh(&wb->work_lock); |
98754bf7 TH |
184 | if (!test_bit(WB_registered, &wb->state)) { |
185 | if (work->single_wait) | |
186 | work->single_done = 1; | |
5acda9d1 | 187 | goto out_unlock; |
98754bf7 | 188 | } |
cc395d7f TH |
189 | if (work->done) |
190 | atomic_inc(&work->done->cnt); | |
f0054bb1 TH |
191 | list_add_tail(&work->list, &wb->work_list); |
192 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
5acda9d1 | 193 | out_unlock: |
f0054bb1 | 194 | spin_unlock_bh(&wb->work_lock); |
1da177e4 LT |
195 | } |
196 | ||
cc395d7f TH |
197 | /** |
198 | * wb_wait_for_completion - wait for completion of bdi_writeback_works | |
199 | * @bdi: bdi work items were issued to | |
200 | * @done: target wb_completion | |
201 | * | |
202 | * Wait for one or more work items issued to @bdi with their ->done field | |
203 | * set to @done, which should have been defined with | |
204 | * DEFINE_WB_COMPLETION_ONSTACK(). This function returns after all such | |
205 | * work items are completed. Work items which are waited upon aren't freed | |
206 | * automatically on completion. | |
207 | */ | |
208 | static void wb_wait_for_completion(struct backing_dev_info *bdi, | |
209 | struct wb_completion *done) | |
210 | { | |
211 | atomic_dec(&done->cnt); /* put down the initial count */ | |
212 | wait_event(bdi->wb_waitq, !atomic_read(&done->cnt)); | |
213 | } | |
214 | ||
703c2708 TH |
215 | #ifdef CONFIG_CGROUP_WRITEBACK |
216 | ||
21c6321f TH |
217 | void __inode_attach_wb(struct inode *inode, struct page *page) |
218 | { | |
219 | struct backing_dev_info *bdi = inode_to_bdi(inode); | |
220 | struct bdi_writeback *wb = NULL; | |
221 | ||
222 | if (inode_cgwb_enabled(inode)) { | |
223 | struct cgroup_subsys_state *memcg_css; | |
224 | ||
225 | if (page) { | |
226 | memcg_css = mem_cgroup_css_from_page(page); | |
227 | wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); | |
228 | } else { | |
229 | /* must pin memcg_css, see wb_get_create() */ | |
230 | memcg_css = task_get_css(current, memory_cgrp_id); | |
231 | wb = wb_get_create(bdi, memcg_css, GFP_ATOMIC); | |
232 | css_put(memcg_css); | |
233 | } | |
234 | } | |
235 | ||
236 | if (!wb) | |
237 | wb = &bdi->wb; | |
238 | ||
239 | /* | |
240 | * There may be multiple instances of this function racing to | |
241 | * update the same inode. Use cmpxchg() to tell the winner. | |
242 | */ | |
243 | if (unlikely(cmpxchg(&inode->i_wb, NULL, wb))) | |
244 | wb_put(wb); | |
245 | } | |
246 | ||
703c2708 TH |
247 | /** |
248 | * inode_congested - test whether an inode is congested | |
249 | * @inode: inode to test for congestion | |
250 | * @cong_bits: mask of WB_[a]sync_congested bits to test | |
251 | * | |
252 | * Tests whether @inode is congested. @cong_bits is the mask of congestion | |
253 | * bits to test and the return value is the mask of set bits. | |
254 | * | |
255 | * If cgroup writeback is enabled for @inode, the congestion state is | |
256 | * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg | |
257 | * associated with @inode is congested; otherwise, the root wb's congestion | |
258 | * state is used. | |
259 | */ | |
260 | int inode_congested(struct inode *inode, int cong_bits) | |
261 | { | |
262 | if (inode) { | |
263 | struct bdi_writeback *wb = inode_to_wb(inode); | |
264 | if (wb) | |
265 | return wb_congested(wb, cong_bits); | |
266 | } | |
267 | ||
268 | return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); | |
269 | } | |
270 | EXPORT_SYMBOL_GPL(inode_congested); | |
271 | ||
98754bf7 TH |
272 | /** |
273 | * wb_wait_for_single_work - wait for completion of a single bdi_writeback_work | |
274 | * @bdi: bdi the work item was issued to | |
275 | * @work: work item to wait for | |
276 | * | |
277 | * Wait for the completion of @work which was issued to one of @bdi's | |
278 | * bdi_writeback's. The caller must have set @work->single_wait before | |
279 | * issuing it. This wait operates independently fo | |
280 | * wb_wait_for_completion() and also disables automatic freeing of @work. | |
281 | */ | |
282 | static void wb_wait_for_single_work(struct backing_dev_info *bdi, | |
283 | struct wb_writeback_work *work) | |
284 | { | |
285 | if (WARN_ON_ONCE(!work->single_wait)) | |
286 | return; | |
287 | ||
288 | wait_event(bdi->wb_waitq, work->single_done); | |
289 | ||
290 | /* | |
291 | * Paired with smp_wmb() in wb_do_writeback() and ensures that all | |
292 | * modifications to @work prior to assertion of ->single_done is | |
293 | * visible to the caller once this function returns. | |
294 | */ | |
295 | smp_rmb(); | |
296 | } | |
297 | ||
f2b65121 TH |
298 | /** |
299 | * wb_split_bdi_pages - split nr_pages to write according to bandwidth | |
300 | * @wb: target bdi_writeback to split @nr_pages to | |
301 | * @nr_pages: number of pages to write for the whole bdi | |
302 | * | |
303 | * Split @wb's portion of @nr_pages according to @wb's write bandwidth in | |
304 | * relation to the total write bandwidth of all wb's w/ dirty inodes on | |
305 | * @wb->bdi. | |
306 | */ | |
307 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) | |
308 | { | |
309 | unsigned long this_bw = wb->avg_write_bandwidth; | |
310 | unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth); | |
311 | ||
312 | if (nr_pages == LONG_MAX) | |
313 | return LONG_MAX; | |
314 | ||
315 | /* | |
316 | * This may be called on clean wb's and proportional distribution | |
317 | * may not make sense, just use the original @nr_pages in those | |
318 | * cases. In general, we wanna err on the side of writing more. | |
319 | */ | |
320 | if (!tot_bw || this_bw >= tot_bw) | |
321 | return nr_pages; | |
322 | else | |
323 | return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw); | |
324 | } | |
325 | ||
db125360 TH |
326 | /** |
327 | * wb_clone_and_queue_work - clone a wb_writeback_work and issue it to a wb | |
328 | * @wb: target bdi_writeback | |
329 | * @base_work: source wb_writeback_work | |
330 | * | |
331 | * Try to make a clone of @base_work and issue it to @wb. If cloning | |
332 | * succeeds, %true is returned; otherwise, @base_work is issued directly | |
333 | * and %false is returned. In the latter case, the caller is required to | |
334 | * wait for @base_work's completion using wb_wait_for_single_work(). | |
335 | * | |
336 | * A clone is auto-freed on completion. @base_work never is. | |
337 | */ | |
338 | static bool wb_clone_and_queue_work(struct bdi_writeback *wb, | |
339 | struct wb_writeback_work *base_work) | |
340 | { | |
341 | struct wb_writeback_work *work; | |
342 | ||
343 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
344 | if (work) { | |
345 | *work = *base_work; | |
346 | work->auto_free = 1; | |
347 | work->single_wait = 0; | |
348 | } else { | |
349 | work = base_work; | |
350 | work->auto_free = 0; | |
351 | work->single_wait = 1; | |
352 | } | |
353 | work->single_done = 0; | |
354 | wb_queue_work(wb, work); | |
355 | return work != base_work; | |
356 | } | |
357 | ||
358 | /** | |
359 | * bdi_split_work_to_wbs - split a wb_writeback_work to all wb's of a bdi | |
360 | * @bdi: target backing_dev_info | |
361 | * @base_work: wb_writeback_work to issue | |
362 | * @skip_if_busy: skip wb's which already have writeback in progress | |
363 | * | |
364 | * Split and issue @base_work to all wb's (bdi_writeback's) of @bdi which | |
365 | * have dirty inodes. If @base_work->nr_page isn't %LONG_MAX, it's | |
366 | * distributed to the busy wbs according to each wb's proportion in the | |
367 | * total active write bandwidth of @bdi. | |
368 | */ | |
369 | static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, | |
370 | struct wb_writeback_work *base_work, | |
371 | bool skip_if_busy) | |
372 | { | |
373 | long nr_pages = base_work->nr_pages; | |
374 | int next_blkcg_id = 0; | |
375 | struct bdi_writeback *wb; | |
376 | struct wb_iter iter; | |
377 | ||
378 | might_sleep(); | |
379 | ||
380 | if (!bdi_has_dirty_io(bdi)) | |
381 | return; | |
382 | restart: | |
383 | rcu_read_lock(); | |
384 | bdi_for_each_wb(wb, bdi, &iter, next_blkcg_id) { | |
385 | if (!wb_has_dirty_io(wb) || | |
386 | (skip_if_busy && writeback_in_progress(wb))) | |
387 | continue; | |
388 | ||
389 | base_work->nr_pages = wb_split_bdi_pages(wb, nr_pages); | |
390 | if (!wb_clone_and_queue_work(wb, base_work)) { | |
391 | next_blkcg_id = wb->blkcg_css->id + 1; | |
392 | rcu_read_unlock(); | |
393 | wb_wait_for_single_work(bdi, base_work); | |
394 | goto restart; | |
395 | } | |
396 | } | |
397 | rcu_read_unlock(); | |
398 | } | |
399 | ||
f2b65121 TH |
400 | #else /* CONFIG_CGROUP_WRITEBACK */ |
401 | ||
402 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) | |
403 | { | |
404 | return nr_pages; | |
405 | } | |
406 | ||
db125360 TH |
407 | static void bdi_split_work_to_wbs(struct backing_dev_info *bdi, |
408 | struct wb_writeback_work *base_work, | |
409 | bool skip_if_busy) | |
410 | { | |
411 | might_sleep(); | |
412 | ||
413 | if (bdi_has_dirty_io(bdi) && | |
414 | (!skip_if_busy || !writeback_in_progress(&bdi->wb))) { | |
415 | base_work->auto_free = 0; | |
416 | base_work->single_wait = 0; | |
417 | base_work->single_done = 0; | |
418 | wb_queue_work(&bdi->wb, base_work); | |
419 | } | |
420 | } | |
421 | ||
703c2708 TH |
422 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
423 | ||
c00ddad3 TH |
424 | void wb_start_writeback(struct bdi_writeback *wb, long nr_pages, |
425 | bool range_cyclic, enum wb_reason reason) | |
b6e51316 | 426 | { |
c00ddad3 TH |
427 | struct wb_writeback_work *work; |
428 | ||
429 | if (!wb_has_dirty_io(wb)) | |
430 | return; | |
431 | ||
432 | /* | |
433 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
434 | * wakeup the thread for old dirty data writeback | |
435 | */ | |
436 | work = kzalloc(sizeof(*work), GFP_ATOMIC); | |
437 | if (!work) { | |
438 | trace_writeback_nowork(wb->bdi); | |
439 | wb_wakeup(wb); | |
440 | return; | |
441 | } | |
442 | ||
443 | work->sync_mode = WB_SYNC_NONE; | |
444 | work->nr_pages = nr_pages; | |
445 | work->range_cyclic = range_cyclic; | |
446 | work->reason = reason; | |
ac7b19a3 | 447 | work->auto_free = 1; |
c00ddad3 TH |
448 | |
449 | wb_queue_work(wb, work); | |
c5444198 | 450 | } |
d3ddec76 | 451 | |
c5444198 | 452 | /** |
9ecf4866 TH |
453 | * wb_start_background_writeback - start background writeback |
454 | * @wb: bdi_writback to write from | |
c5444198 CH |
455 | * |
456 | * Description: | |
6585027a | 457 | * This makes sure WB_SYNC_NONE background writeback happens. When |
9ecf4866 | 458 | * this function returns, it is only guaranteed that for given wb |
6585027a JK |
459 | * some IO is happening if we are over background dirty threshold. |
460 | * Caller need not hold sb s_umount semaphore. | |
c5444198 | 461 | */ |
9ecf4866 | 462 | void wb_start_background_writeback(struct bdi_writeback *wb) |
c5444198 | 463 | { |
6585027a JK |
464 | /* |
465 | * We just wake up the flusher thread. It will perform background | |
466 | * writeback as soon as there is no other work to do. | |
467 | */ | |
9ecf4866 TH |
468 | trace_writeback_wake_background(wb->bdi); |
469 | wb_wakeup(wb); | |
1da177e4 LT |
470 | } |
471 | ||
a66979ab DC |
472 | /* |
473 | * Remove the inode from the writeback list it is on. | |
474 | */ | |
475 | void inode_wb_list_del(struct inode *inode) | |
476 | { | |
52ebea74 | 477 | struct bdi_writeback *wb = inode_to_wb(inode); |
f758eeab | 478 | |
52ebea74 | 479 | spin_lock(&wb->list_lock); |
d6c10f1f | 480 | inode_wb_list_del_locked(inode, wb); |
52ebea74 | 481 | spin_unlock(&wb->list_lock); |
a66979ab DC |
482 | } |
483 | ||
6610a0bc AM |
484 | /* |
485 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
486 | * furthest end of its superblock's dirty-inode list. | |
487 | * | |
488 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 489 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
490 | * the case then the inode must have been redirtied while it was being written |
491 | * out and we don't reset its dirtied_when. | |
492 | */ | |
f758eeab | 493 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 494 | { |
03ba3782 | 495 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 496 | struct inode *tail; |
6610a0bc | 497 | |
7ccf19a8 | 498 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 499 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
500 | inode->dirtied_when = jiffies; |
501 | } | |
d6c10f1f | 502 | inode_wb_list_move_locked(inode, wb, &wb->b_dirty); |
6610a0bc AM |
503 | } |
504 | ||
c986d1e2 | 505 | /* |
66f3b8e2 | 506 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 507 | */ |
f758eeab | 508 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 509 | { |
d6c10f1f | 510 | inode_wb_list_move_locked(inode, wb, &wb->b_more_io); |
c986d1e2 AM |
511 | } |
512 | ||
1c0eeaf5 JE |
513 | static void inode_sync_complete(struct inode *inode) |
514 | { | |
365b94ae | 515 | inode->i_state &= ~I_SYNC; |
4eff96dd JK |
516 | /* If inode is clean an unused, put it into LRU now... */ |
517 | inode_add_lru(inode); | |
365b94ae | 518 | /* Waiters must see I_SYNC cleared before being woken up */ |
1c0eeaf5 JE |
519 | smp_mb(); |
520 | wake_up_bit(&inode->i_state, __I_SYNC); | |
521 | } | |
522 | ||
d2caa3c5 JL |
523 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
524 | { | |
525 | bool ret = time_after(inode->dirtied_when, t); | |
526 | #ifndef CONFIG_64BIT | |
527 | /* | |
528 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
529 | * It _appears_ to be in the future, but is actually in distant past. | |
530 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 531 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
532 | */ |
533 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
534 | #endif | |
535 | return ret; | |
536 | } | |
537 | ||
0ae45f63 TT |
538 | #define EXPIRE_DIRTY_ATIME 0x0001 |
539 | ||
2c136579 | 540 | /* |
0e2f2b23 | 541 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
697e6fed | 542 | * @delaying_queue to @dispatch_queue. |
2c136579 | 543 | */ |
e84d0a4f | 544 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 545 | struct list_head *dispatch_queue, |
0ae45f63 | 546 | int flags, |
ad4e38dd | 547 | struct wb_writeback_work *work) |
2c136579 | 548 | { |
0ae45f63 TT |
549 | unsigned long *older_than_this = NULL; |
550 | unsigned long expire_time; | |
5c03449d SL |
551 | LIST_HEAD(tmp); |
552 | struct list_head *pos, *node; | |
cf137307 | 553 | struct super_block *sb = NULL; |
5c03449d | 554 | struct inode *inode; |
cf137307 | 555 | int do_sb_sort = 0; |
e84d0a4f | 556 | int moved = 0; |
5c03449d | 557 | |
0ae45f63 TT |
558 | if ((flags & EXPIRE_DIRTY_ATIME) == 0) |
559 | older_than_this = work->older_than_this; | |
a2f48706 TT |
560 | else if (!work->for_sync) { |
561 | expire_time = jiffies - (dirtytime_expire_interval * HZ); | |
0ae45f63 TT |
562 | older_than_this = &expire_time; |
563 | } | |
2c136579 | 564 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 565 | inode = wb_inode(delaying_queue->prev); |
0ae45f63 TT |
566 | if (older_than_this && |
567 | inode_dirtied_after(inode, *older_than_this)) | |
2c136579 | 568 | break; |
a8855990 JK |
569 | list_move(&inode->i_wb_list, &tmp); |
570 | moved++; | |
0ae45f63 TT |
571 | if (flags & EXPIRE_DIRTY_ATIME) |
572 | set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); | |
a8855990 JK |
573 | if (sb_is_blkdev_sb(inode->i_sb)) |
574 | continue; | |
cf137307 JA |
575 | if (sb && sb != inode->i_sb) |
576 | do_sb_sort = 1; | |
577 | sb = inode->i_sb; | |
5c03449d SL |
578 | } |
579 | ||
cf137307 JA |
580 | /* just one sb in list, splice to dispatch_queue and we're done */ |
581 | if (!do_sb_sort) { | |
582 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 583 | goto out; |
cf137307 JA |
584 | } |
585 | ||
5c03449d SL |
586 | /* Move inodes from one superblock together */ |
587 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 588 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 589 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 590 | inode = wb_inode(pos); |
5c03449d | 591 | if (inode->i_sb == sb) |
7ccf19a8 | 592 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 593 | } |
2c136579 | 594 | } |
e84d0a4f WF |
595 | out: |
596 | return moved; | |
2c136579 FW |
597 | } |
598 | ||
599 | /* | |
600 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
601 | * Before |
602 | * newly dirtied b_dirty b_io b_more_io | |
603 | * =============> gf edc BA | |
604 | * After | |
605 | * newly dirtied b_dirty b_io b_more_io | |
606 | * =============> g fBAedc | |
607 | * | | |
608 | * +--> dequeue for IO | |
2c136579 | 609 | */ |
ad4e38dd | 610 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
66f3b8e2 | 611 | { |
e84d0a4f | 612 | int moved; |
0ae45f63 | 613 | |
f758eeab | 614 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 615 | list_splice_init(&wb->b_more_io, &wb->b_io); |
0ae45f63 TT |
616 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); |
617 | moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, | |
618 | EXPIRE_DIRTY_ATIME, work); | |
d6c10f1f TH |
619 | if (moved) |
620 | wb_io_lists_populated(wb); | |
ad4e38dd | 621 | trace_writeback_queue_io(wb, work, moved); |
66f3b8e2 JA |
622 | } |
623 | ||
a9185b41 | 624 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 625 | { |
9fb0a7da TH |
626 | int ret; |
627 | ||
628 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
629 | trace_writeback_write_inode_start(inode, wbc); | |
630 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
631 | trace_writeback_write_inode(inode, wbc); | |
632 | return ret; | |
633 | } | |
03ba3782 | 634 | return 0; |
08d8e974 | 635 | } |
08d8e974 | 636 | |
1da177e4 | 637 | /* |
169ebd90 JK |
638 | * Wait for writeback on an inode to complete. Called with i_lock held. |
639 | * Caller must make sure inode cannot go away when we drop i_lock. | |
01c03194 | 640 | */ |
169ebd90 JK |
641 | static void __inode_wait_for_writeback(struct inode *inode) |
642 | __releases(inode->i_lock) | |
643 | __acquires(inode->i_lock) | |
01c03194 CH |
644 | { |
645 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
646 | wait_queue_head_t *wqh; | |
647 | ||
648 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
649 | while (inode->i_state & I_SYNC) { |
650 | spin_unlock(&inode->i_lock); | |
74316201 N |
651 | __wait_on_bit(wqh, &wq, bit_wait, |
652 | TASK_UNINTERRUPTIBLE); | |
250df6ed | 653 | spin_lock(&inode->i_lock); |
58a9d3d8 | 654 | } |
01c03194 CH |
655 | } |
656 | ||
169ebd90 JK |
657 | /* |
658 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
659 | */ | |
660 | void inode_wait_for_writeback(struct inode *inode) | |
661 | { | |
662 | spin_lock(&inode->i_lock); | |
663 | __inode_wait_for_writeback(inode); | |
664 | spin_unlock(&inode->i_lock); | |
665 | } | |
666 | ||
667 | /* | |
668 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
669 | * held and drops it. It is aimed for callers not holding any inode reference | |
670 | * so once i_lock is dropped, inode can go away. | |
671 | */ | |
672 | static void inode_sleep_on_writeback(struct inode *inode) | |
673 | __releases(inode->i_lock) | |
674 | { | |
675 | DEFINE_WAIT(wait); | |
676 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
677 | int sleep; | |
678 | ||
679 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
680 | sleep = inode->i_state & I_SYNC; | |
681 | spin_unlock(&inode->i_lock); | |
682 | if (sleep) | |
683 | schedule(); | |
684 | finish_wait(wqh, &wait); | |
685 | } | |
686 | ||
ccb26b5a JK |
687 | /* |
688 | * Find proper writeback list for the inode depending on its current state and | |
689 | * possibly also change of its state while we were doing writeback. Here we | |
690 | * handle things such as livelock prevention or fairness of writeback among | |
691 | * inodes. This function can be called only by flusher thread - noone else | |
692 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
693 | * thread's back can have unexpected consequences. | |
694 | */ | |
695 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
696 | struct writeback_control *wbc) | |
697 | { | |
698 | if (inode->i_state & I_FREEING) | |
699 | return; | |
700 | ||
701 | /* | |
702 | * Sync livelock prevention. Each inode is tagged and synced in one | |
703 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
704 | * the dirty time to prevent enqueue and sync it again. | |
705 | */ | |
706 | if ((inode->i_state & I_DIRTY) && | |
707 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
708 | inode->dirtied_when = jiffies; | |
709 | ||
4f8ad655 JK |
710 | if (wbc->pages_skipped) { |
711 | /* | |
712 | * writeback is not making progress due to locked | |
713 | * buffers. Skip this inode for now. | |
714 | */ | |
715 | redirty_tail(inode, wb); | |
716 | return; | |
717 | } | |
718 | ||
ccb26b5a JK |
719 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
720 | /* | |
721 | * We didn't write back all the pages. nfs_writepages() | |
722 | * sometimes bales out without doing anything. | |
723 | */ | |
724 | if (wbc->nr_to_write <= 0) { | |
725 | /* Slice used up. Queue for next turn. */ | |
726 | requeue_io(inode, wb); | |
727 | } else { | |
728 | /* | |
729 | * Writeback blocked by something other than | |
730 | * congestion. Delay the inode for some time to | |
731 | * avoid spinning on the CPU (100% iowait) | |
732 | * retrying writeback of the dirty page/inode | |
733 | * that cannot be performed immediately. | |
734 | */ | |
735 | redirty_tail(inode, wb); | |
736 | } | |
737 | } else if (inode->i_state & I_DIRTY) { | |
738 | /* | |
739 | * Filesystems can dirty the inode during writeback operations, | |
740 | * such as delayed allocation during submission or metadata | |
741 | * updates after data IO completion. | |
742 | */ | |
743 | redirty_tail(inode, wb); | |
0ae45f63 | 744 | } else if (inode->i_state & I_DIRTY_TIME) { |
a2f48706 | 745 | inode->dirtied_when = jiffies; |
d6c10f1f | 746 | inode_wb_list_move_locked(inode, wb, &wb->b_dirty_time); |
ccb26b5a JK |
747 | } else { |
748 | /* The inode is clean. Remove from writeback lists. */ | |
d6c10f1f | 749 | inode_wb_list_del_locked(inode, wb); |
ccb26b5a JK |
750 | } |
751 | } | |
752 | ||
01c03194 | 753 | /* |
4f8ad655 JK |
754 | * Write out an inode and its dirty pages. Do not update the writeback list |
755 | * linkage. That is left to the caller. The caller is also responsible for | |
756 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. | |
1da177e4 LT |
757 | */ |
758 | static int | |
cd8ed2a4 | 759 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 760 | { |
1da177e4 | 761 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 762 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 763 | unsigned dirty; |
1da177e4 LT |
764 | int ret; |
765 | ||
4f8ad655 | 766 | WARN_ON(!(inode->i_state & I_SYNC)); |
1da177e4 | 767 | |
9fb0a7da TH |
768 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
769 | ||
1da177e4 LT |
770 | ret = do_writepages(mapping, wbc); |
771 | ||
26821ed4 CH |
772 | /* |
773 | * Make sure to wait on the data before writing out the metadata. | |
774 | * This is important for filesystems that modify metadata on data | |
7747bd4b DC |
775 | * I/O completion. We don't do it for sync(2) writeback because it has a |
776 | * separate, external IO completion path and ->sync_fs for guaranteeing | |
777 | * inode metadata is written back correctly. | |
26821ed4 | 778 | */ |
7747bd4b | 779 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
26821ed4 | 780 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
781 | if (ret == 0) |
782 | ret = err; | |
783 | } | |
784 | ||
5547e8aa DM |
785 | /* |
786 | * Some filesystems may redirty the inode during the writeback | |
787 | * due to delalloc, clear dirty metadata flags right before | |
788 | * write_inode() | |
789 | */ | |
250df6ed | 790 | spin_lock(&inode->i_lock); |
9c6ac78e | 791 | |
5547e8aa | 792 | dirty = inode->i_state & I_DIRTY; |
a2f48706 TT |
793 | if (inode->i_state & I_DIRTY_TIME) { |
794 | if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
795 | unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || | |
796 | unlikely(time_after(jiffies, | |
797 | (inode->dirtied_time_when + | |
798 | dirtytime_expire_interval * HZ)))) { | |
799 | dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; | |
800 | trace_writeback_lazytime(inode); | |
801 | } | |
802 | } else | |
803 | inode->i_state &= ~I_DIRTY_TIME_EXPIRED; | |
0ae45f63 | 804 | inode->i_state &= ~dirty; |
9c6ac78e TH |
805 | |
806 | /* | |
807 | * Paired with smp_mb() in __mark_inode_dirty(). This allows | |
808 | * __mark_inode_dirty() to test i_state without grabbing i_lock - | |
809 | * either they see the I_DIRTY bits cleared or we see the dirtied | |
810 | * inode. | |
811 | * | |
812 | * I_DIRTY_PAGES is always cleared together above even if @mapping | |
813 | * still has dirty pages. The flag is reinstated after smp_mb() if | |
814 | * necessary. This guarantees that either __mark_inode_dirty() | |
815 | * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. | |
816 | */ | |
817 | smp_mb(); | |
818 | ||
819 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
820 | inode->i_state |= I_DIRTY_PAGES; | |
821 | ||
250df6ed | 822 | spin_unlock(&inode->i_lock); |
9c6ac78e | 823 | |
0ae45f63 TT |
824 | if (dirty & I_DIRTY_TIME) |
825 | mark_inode_dirty_sync(inode); | |
26821ed4 | 826 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
0ae45f63 | 827 | if (dirty & ~I_DIRTY_PAGES) { |
a9185b41 | 828 | int err = write_inode(inode, wbc); |
1da177e4 LT |
829 | if (ret == 0) |
830 | ret = err; | |
831 | } | |
4f8ad655 JK |
832 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
833 | return ret; | |
834 | } | |
835 | ||
836 | /* | |
837 | * Write out an inode's dirty pages. Either the caller has an active reference | |
838 | * on the inode or the inode has I_WILL_FREE set. | |
839 | * | |
840 | * This function is designed to be called for writing back one inode which | |
841 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() | |
842 | * and does more profound writeback list handling in writeback_sb_inodes(). | |
843 | */ | |
844 | static int | |
845 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, | |
846 | struct writeback_control *wbc) | |
847 | { | |
848 | int ret = 0; | |
849 | ||
850 | spin_lock(&inode->i_lock); | |
851 | if (!atomic_read(&inode->i_count)) | |
852 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
853 | else | |
854 | WARN_ON(inode->i_state & I_WILL_FREE); | |
855 | ||
856 | if (inode->i_state & I_SYNC) { | |
857 | if (wbc->sync_mode != WB_SYNC_ALL) | |
858 | goto out; | |
859 | /* | |
169ebd90 JK |
860 | * It's a data-integrity sync. We must wait. Since callers hold |
861 | * inode reference or inode has I_WILL_FREE set, it cannot go | |
862 | * away under us. | |
4f8ad655 | 863 | */ |
169ebd90 | 864 | __inode_wait_for_writeback(inode); |
4f8ad655 JK |
865 | } |
866 | WARN_ON(inode->i_state & I_SYNC); | |
867 | /* | |
f9b0e058 JK |
868 | * Skip inode if it is clean and we have no outstanding writeback in |
869 | * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this | |
870 | * function since flusher thread may be doing for example sync in | |
871 | * parallel and if we move the inode, it could get skipped. So here we | |
872 | * make sure inode is on some writeback list and leave it there unless | |
873 | * we have completely cleaned the inode. | |
4f8ad655 | 874 | */ |
0ae45f63 | 875 | if (!(inode->i_state & I_DIRTY_ALL) && |
f9b0e058 JK |
876 | (wbc->sync_mode != WB_SYNC_ALL || |
877 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) | |
4f8ad655 JK |
878 | goto out; |
879 | inode->i_state |= I_SYNC; | |
880 | spin_unlock(&inode->i_lock); | |
881 | ||
cd8ed2a4 | 882 | ret = __writeback_single_inode(inode, wbc); |
1da177e4 | 883 | |
f758eeab | 884 | spin_lock(&wb->list_lock); |
250df6ed | 885 | spin_lock(&inode->i_lock); |
4f8ad655 JK |
886 | /* |
887 | * If inode is clean, remove it from writeback lists. Otherwise don't | |
888 | * touch it. See comment above for explanation. | |
889 | */ | |
0ae45f63 | 890 | if (!(inode->i_state & I_DIRTY_ALL)) |
d6c10f1f | 891 | inode_wb_list_del_locked(inode, wb); |
4f8ad655 | 892 | spin_unlock(&wb->list_lock); |
1c0eeaf5 | 893 | inode_sync_complete(inode); |
4f8ad655 JK |
894 | out: |
895 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
896 | return ret; |
897 | } | |
898 | ||
a88a341a | 899 | static long writeback_chunk_size(struct bdi_writeback *wb, |
1a12d8bd | 900 | struct wb_writeback_work *work) |
d46db3d5 WF |
901 | { |
902 | long pages; | |
903 | ||
904 | /* | |
905 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
906 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
907 | * here avoids calling into writeback_inodes_wb() more than once. | |
908 | * | |
909 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
910 | * | |
911 | * wb_writeback() | |
912 | * writeback_sb_inodes() <== called only once | |
913 | * write_cache_pages() <== called once for each inode | |
914 | * (quickly) tag currently dirty pages | |
915 | * (maybe slowly) sync all tagged pages | |
916 | */ | |
917 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
918 | pages = LONG_MAX; | |
1a12d8bd | 919 | else { |
a88a341a | 920 | pages = min(wb->avg_write_bandwidth / 2, |
dcc25ae7 | 921 | global_wb_domain.dirty_limit / DIRTY_SCOPE); |
1a12d8bd WF |
922 | pages = min(pages, work->nr_pages); |
923 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
924 | MIN_WRITEBACK_PAGES); | |
925 | } | |
d46db3d5 WF |
926 | |
927 | return pages; | |
928 | } | |
929 | ||
f11c9c5c ES |
930 | /* |
931 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 | 932 | * |
d46db3d5 | 933 | * Return the number of pages and/or inodes written. |
f11c9c5c | 934 | */ |
d46db3d5 WF |
935 | static long writeback_sb_inodes(struct super_block *sb, |
936 | struct bdi_writeback *wb, | |
937 | struct wb_writeback_work *work) | |
1da177e4 | 938 | { |
d46db3d5 WF |
939 | struct writeback_control wbc = { |
940 | .sync_mode = work->sync_mode, | |
941 | .tagged_writepages = work->tagged_writepages, | |
942 | .for_kupdate = work->for_kupdate, | |
943 | .for_background = work->for_background, | |
7747bd4b | 944 | .for_sync = work->for_sync, |
d46db3d5 WF |
945 | .range_cyclic = work->range_cyclic, |
946 | .range_start = 0, | |
947 | .range_end = LLONG_MAX, | |
948 | }; | |
949 | unsigned long start_time = jiffies; | |
950 | long write_chunk; | |
951 | long wrote = 0; /* count both pages and inodes */ | |
952 | ||
03ba3782 | 953 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 954 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
955 | |
956 | if (inode->i_sb != sb) { | |
d46db3d5 | 957 | if (work->sb) { |
edadfb10 CH |
958 | /* |
959 | * We only want to write back data for this | |
960 | * superblock, move all inodes not belonging | |
961 | * to it back onto the dirty list. | |
962 | */ | |
f758eeab | 963 | redirty_tail(inode, wb); |
edadfb10 CH |
964 | continue; |
965 | } | |
966 | ||
967 | /* | |
968 | * The inode belongs to a different superblock. | |
969 | * Bounce back to the caller to unpin this and | |
970 | * pin the next superblock. | |
971 | */ | |
d46db3d5 | 972 | break; |
edadfb10 CH |
973 | } |
974 | ||
9843b76a | 975 | /* |
331cbdee WL |
976 | * Don't bother with new inodes or inodes being freed, first |
977 | * kind does not need periodic writeout yet, and for the latter | |
9843b76a CH |
978 | * kind writeout is handled by the freer. |
979 | */ | |
250df6ed | 980 | spin_lock(&inode->i_lock); |
9843b76a | 981 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 982 | spin_unlock(&inode->i_lock); |
fcc5c222 | 983 | redirty_tail(inode, wb); |
7ef0d737 NP |
984 | continue; |
985 | } | |
cc1676d9 JK |
986 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
987 | /* | |
988 | * If this inode is locked for writeback and we are not | |
989 | * doing writeback-for-data-integrity, move it to | |
990 | * b_more_io so that writeback can proceed with the | |
991 | * other inodes on s_io. | |
992 | * | |
993 | * We'll have another go at writing back this inode | |
994 | * when we completed a full scan of b_io. | |
995 | */ | |
996 | spin_unlock(&inode->i_lock); | |
997 | requeue_io(inode, wb); | |
998 | trace_writeback_sb_inodes_requeue(inode); | |
999 | continue; | |
1000 | } | |
f0d07b7f JK |
1001 | spin_unlock(&wb->list_lock); |
1002 | ||
4f8ad655 JK |
1003 | /* |
1004 | * We already requeued the inode if it had I_SYNC set and we | |
1005 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
1006 | * WB_SYNC_ALL case. | |
1007 | */ | |
169ebd90 JK |
1008 | if (inode->i_state & I_SYNC) { |
1009 | /* Wait for I_SYNC. This function drops i_lock... */ | |
1010 | inode_sleep_on_writeback(inode); | |
1011 | /* Inode may be gone, start again */ | |
ead188f9 | 1012 | spin_lock(&wb->list_lock); |
169ebd90 JK |
1013 | continue; |
1014 | } | |
4f8ad655 JK |
1015 | inode->i_state |= I_SYNC; |
1016 | spin_unlock(&inode->i_lock); | |
169ebd90 | 1017 | |
a88a341a | 1018 | write_chunk = writeback_chunk_size(wb, work); |
d46db3d5 WF |
1019 | wbc.nr_to_write = write_chunk; |
1020 | wbc.pages_skipped = 0; | |
250df6ed | 1021 | |
169ebd90 JK |
1022 | /* |
1023 | * We use I_SYNC to pin the inode in memory. While it is set | |
1024 | * evict_inode() will wait so the inode cannot be freed. | |
1025 | */ | |
cd8ed2a4 | 1026 | __writeback_single_inode(inode, &wbc); |
250df6ed | 1027 | |
d46db3d5 WF |
1028 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
1029 | wrote += write_chunk - wbc.nr_to_write; | |
4f8ad655 JK |
1030 | spin_lock(&wb->list_lock); |
1031 | spin_lock(&inode->i_lock); | |
0ae45f63 | 1032 | if (!(inode->i_state & I_DIRTY_ALL)) |
d46db3d5 | 1033 | wrote++; |
4f8ad655 JK |
1034 | requeue_inode(inode, wb, &wbc); |
1035 | inode_sync_complete(inode); | |
0f1b1fd8 | 1036 | spin_unlock(&inode->i_lock); |
169ebd90 | 1037 | cond_resched_lock(&wb->list_lock); |
d46db3d5 WF |
1038 | /* |
1039 | * bail out to wb_writeback() often enough to check | |
1040 | * background threshold and other termination conditions. | |
1041 | */ | |
1042 | if (wrote) { | |
1043 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
1044 | break; | |
1045 | if (work->nr_pages <= 0) | |
1046 | break; | |
8bc3be27 | 1047 | } |
1da177e4 | 1048 | } |
d46db3d5 | 1049 | return wrote; |
f11c9c5c ES |
1050 | } |
1051 | ||
d46db3d5 WF |
1052 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
1053 | struct wb_writeback_work *work) | |
f11c9c5c | 1054 | { |
d46db3d5 WF |
1055 | unsigned long start_time = jiffies; |
1056 | long wrote = 0; | |
38f21977 | 1057 | |
f11c9c5c | 1058 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 1059 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 1060 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 1061 | |
eb6ef3df | 1062 | if (!trylock_super(sb)) { |
0e995816 | 1063 | /* |
eb6ef3df | 1064 | * trylock_super() may fail consistently due to |
0e995816 WF |
1065 | * s_umount being grabbed by someone else. Don't use |
1066 | * requeue_io() to avoid busy retrying the inode/sb. | |
1067 | */ | |
1068 | redirty_tail(inode, wb); | |
edadfb10 | 1069 | continue; |
f11c9c5c | 1070 | } |
d46db3d5 | 1071 | wrote += writeback_sb_inodes(sb, wb, work); |
eb6ef3df | 1072 | up_read(&sb->s_umount); |
f11c9c5c | 1073 | |
d46db3d5 WF |
1074 | /* refer to the same tests at the end of writeback_sb_inodes */ |
1075 | if (wrote) { | |
1076 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
1077 | break; | |
1078 | if (work->nr_pages <= 0) | |
1079 | break; | |
1080 | } | |
f11c9c5c | 1081 | } |
66f3b8e2 | 1082 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 1083 | return wrote; |
66f3b8e2 JA |
1084 | } |
1085 | ||
7d9f073b | 1086 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
0e175a18 | 1087 | enum wb_reason reason) |
edadfb10 | 1088 | { |
d46db3d5 WF |
1089 | struct wb_writeback_work work = { |
1090 | .nr_pages = nr_pages, | |
1091 | .sync_mode = WB_SYNC_NONE, | |
1092 | .range_cyclic = 1, | |
0e175a18 | 1093 | .reason = reason, |
d46db3d5 | 1094 | }; |
edadfb10 | 1095 | |
f758eeab | 1096 | spin_lock(&wb->list_lock); |
424b351f | 1097 | if (list_empty(&wb->b_io)) |
ad4e38dd | 1098 | queue_io(wb, &work); |
d46db3d5 | 1099 | __writeback_inodes_wb(wb, &work); |
f758eeab | 1100 | spin_unlock(&wb->list_lock); |
edadfb10 | 1101 | |
d46db3d5 WF |
1102 | return nr_pages - work.nr_pages; |
1103 | } | |
03ba3782 | 1104 | |
03ba3782 JA |
1105 | /* |
1106 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 1107 | * |
03ba3782 JA |
1108 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
1109 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
1110 | * just walks the superblock inode list, writing back any inodes which are | |
1111 | * older than a specific point in time. | |
66f3b8e2 | 1112 | * |
03ba3782 JA |
1113 | * Try to run once per dirty_writeback_interval. But if a writeback event |
1114 | * takes longer than a dirty_writeback_interval interval, then leave a | |
1115 | * one-second gap. | |
66f3b8e2 | 1116 | * |
03ba3782 JA |
1117 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
1118 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 1119 | */ |
c4a77a6c | 1120 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 1121 | struct wb_writeback_work *work) |
66f3b8e2 | 1122 | { |
e98be2d5 | 1123 | unsigned long wb_start = jiffies; |
d46db3d5 | 1124 | long nr_pages = work->nr_pages; |
0dc83bd3 | 1125 | unsigned long oldest_jif; |
a5989bdc | 1126 | struct inode *inode; |
d46db3d5 | 1127 | long progress; |
66f3b8e2 | 1128 | |
0dc83bd3 JK |
1129 | oldest_jif = jiffies; |
1130 | work->older_than_this = &oldest_jif; | |
38f21977 | 1131 | |
e8dfc305 | 1132 | spin_lock(&wb->list_lock); |
03ba3782 JA |
1133 | for (;;) { |
1134 | /* | |
d3ddec76 | 1135 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 1136 | */ |
83ba7b07 | 1137 | if (work->nr_pages <= 0) |
03ba3782 | 1138 | break; |
66f3b8e2 | 1139 | |
aa373cf5 JK |
1140 | /* |
1141 | * Background writeout and kupdate-style writeback may | |
1142 | * run forever. Stop them if there is other work to do | |
1143 | * so that e.g. sync can proceed. They'll be restarted | |
1144 | * after the other works are all done. | |
1145 | */ | |
1146 | if ((work->for_background || work->for_kupdate) && | |
f0054bb1 | 1147 | !list_empty(&wb->work_list)) |
aa373cf5 JK |
1148 | break; |
1149 | ||
38f21977 | 1150 | /* |
d3ddec76 WF |
1151 | * For background writeout, stop when we are below the |
1152 | * background dirty threshold | |
38f21977 | 1153 | */ |
aa661bbe | 1154 | if (work->for_background && !wb_over_bg_thresh(wb)) |
03ba3782 | 1155 | break; |
38f21977 | 1156 | |
1bc36b64 JK |
1157 | /* |
1158 | * Kupdate and background works are special and we want to | |
1159 | * include all inodes that need writing. Livelock avoidance is | |
1160 | * handled by these works yielding to any other work so we are | |
1161 | * safe. | |
1162 | */ | |
ba9aa839 | 1163 | if (work->for_kupdate) { |
0dc83bd3 | 1164 | oldest_jif = jiffies - |
ba9aa839 | 1165 | msecs_to_jiffies(dirty_expire_interval * 10); |
1bc36b64 | 1166 | } else if (work->for_background) |
0dc83bd3 | 1167 | oldest_jif = jiffies; |
028c2dd1 | 1168 | |
d46db3d5 | 1169 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 1170 | if (list_empty(&wb->b_io)) |
ad4e38dd | 1171 | queue_io(wb, work); |
83ba7b07 | 1172 | if (work->sb) |
d46db3d5 | 1173 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 1174 | else |
d46db3d5 WF |
1175 | progress = __writeback_inodes_wb(wb, work); |
1176 | trace_writeback_written(wb->bdi, work); | |
028c2dd1 | 1177 | |
e98be2d5 | 1178 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
1179 | |
1180 | /* | |
e6fb6da2 WF |
1181 | * Did we write something? Try for more |
1182 | * | |
1183 | * Dirty inodes are moved to b_io for writeback in batches. | |
1184 | * The completion of the current batch does not necessarily | |
1185 | * mean the overall work is done. So we keep looping as long | |
1186 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 1187 | */ |
d46db3d5 | 1188 | if (progress) |
71fd05a8 JA |
1189 | continue; |
1190 | /* | |
e6fb6da2 | 1191 | * No more inodes for IO, bail |
71fd05a8 | 1192 | */ |
b7a2441f | 1193 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 1194 | break; |
71fd05a8 JA |
1195 | /* |
1196 | * Nothing written. Wait for some inode to | |
1197 | * become available for writeback. Otherwise | |
1198 | * we'll just busyloop. | |
1199 | */ | |
71fd05a8 | 1200 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 1201 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 1202 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 1203 | spin_lock(&inode->i_lock); |
f0d07b7f | 1204 | spin_unlock(&wb->list_lock); |
169ebd90 JK |
1205 | /* This function drops i_lock... */ |
1206 | inode_sleep_on_writeback(inode); | |
f0d07b7f | 1207 | spin_lock(&wb->list_lock); |
03ba3782 JA |
1208 | } |
1209 | } | |
e8dfc305 | 1210 | spin_unlock(&wb->list_lock); |
03ba3782 | 1211 | |
d46db3d5 | 1212 | return nr_pages - work->nr_pages; |
03ba3782 JA |
1213 | } |
1214 | ||
1215 | /* | |
83ba7b07 | 1216 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 1217 | */ |
f0054bb1 | 1218 | static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) |
03ba3782 | 1219 | { |
83ba7b07 | 1220 | struct wb_writeback_work *work = NULL; |
03ba3782 | 1221 | |
f0054bb1 TH |
1222 | spin_lock_bh(&wb->work_lock); |
1223 | if (!list_empty(&wb->work_list)) { | |
1224 | work = list_entry(wb->work_list.next, | |
83ba7b07 CH |
1225 | struct wb_writeback_work, list); |
1226 | list_del_init(&work->list); | |
03ba3782 | 1227 | } |
f0054bb1 | 1228 | spin_unlock_bh(&wb->work_lock); |
83ba7b07 | 1229 | return work; |
03ba3782 JA |
1230 | } |
1231 | ||
cdf01dd5 LT |
1232 | /* |
1233 | * Add in the number of potentially dirty inodes, because each inode | |
1234 | * write can dirty pagecache in the underlying blockdev. | |
1235 | */ | |
1236 | static unsigned long get_nr_dirty_pages(void) | |
1237 | { | |
1238 | return global_page_state(NR_FILE_DIRTY) + | |
1239 | global_page_state(NR_UNSTABLE_NFS) + | |
1240 | get_nr_dirty_inodes(); | |
1241 | } | |
1242 | ||
6585027a JK |
1243 | static long wb_check_background_flush(struct bdi_writeback *wb) |
1244 | { | |
aa661bbe | 1245 | if (wb_over_bg_thresh(wb)) { |
6585027a JK |
1246 | |
1247 | struct wb_writeback_work work = { | |
1248 | .nr_pages = LONG_MAX, | |
1249 | .sync_mode = WB_SYNC_NONE, | |
1250 | .for_background = 1, | |
1251 | .range_cyclic = 1, | |
0e175a18 | 1252 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
1253 | }; |
1254 | ||
1255 | return wb_writeback(wb, &work); | |
1256 | } | |
1257 | ||
1258 | return 0; | |
1259 | } | |
1260 | ||
03ba3782 JA |
1261 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
1262 | { | |
1263 | unsigned long expired; | |
1264 | long nr_pages; | |
1265 | ||
69b62d01 JA |
1266 | /* |
1267 | * When set to zero, disable periodic writeback | |
1268 | */ | |
1269 | if (!dirty_writeback_interval) | |
1270 | return 0; | |
1271 | ||
03ba3782 JA |
1272 | expired = wb->last_old_flush + |
1273 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
1274 | if (time_before(jiffies, expired)) | |
1275 | return 0; | |
1276 | ||
1277 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 1278 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 1279 | |
c4a77a6c | 1280 | if (nr_pages) { |
83ba7b07 | 1281 | struct wb_writeback_work work = { |
c4a77a6c JA |
1282 | .nr_pages = nr_pages, |
1283 | .sync_mode = WB_SYNC_NONE, | |
1284 | .for_kupdate = 1, | |
1285 | .range_cyclic = 1, | |
0e175a18 | 1286 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
1287 | }; |
1288 | ||
83ba7b07 | 1289 | return wb_writeback(wb, &work); |
c4a77a6c | 1290 | } |
03ba3782 JA |
1291 | |
1292 | return 0; | |
1293 | } | |
1294 | ||
1295 | /* | |
1296 | * Retrieve work items and do the writeback they describe | |
1297 | */ | |
25d130ba | 1298 | static long wb_do_writeback(struct bdi_writeback *wb) |
03ba3782 | 1299 | { |
83ba7b07 | 1300 | struct wb_writeback_work *work; |
c4a77a6c | 1301 | long wrote = 0; |
03ba3782 | 1302 | |
4452226e | 1303 | set_bit(WB_writeback_running, &wb->state); |
f0054bb1 | 1304 | while ((work = get_next_work_item(wb)) != NULL) { |
cc395d7f | 1305 | struct wb_completion *done = work->done; |
98754bf7 | 1306 | bool need_wake_up = false; |
03ba3782 | 1307 | |
f0054bb1 | 1308 | trace_writeback_exec(wb->bdi, work); |
455b2864 | 1309 | |
83ba7b07 | 1310 | wrote += wb_writeback(wb, work); |
03ba3782 | 1311 | |
98754bf7 TH |
1312 | if (work->single_wait) { |
1313 | WARN_ON_ONCE(work->auto_free); | |
1314 | /* paired w/ rmb in wb_wait_for_single_work() */ | |
1315 | smp_wmb(); | |
1316 | work->single_done = 1; | |
1317 | need_wake_up = true; | |
1318 | } else if (work->auto_free) { | |
83ba7b07 | 1319 | kfree(work); |
98754bf7 TH |
1320 | } |
1321 | ||
cc395d7f | 1322 | if (done && atomic_dec_and_test(&done->cnt)) |
98754bf7 TH |
1323 | need_wake_up = true; |
1324 | ||
1325 | if (need_wake_up) | |
cc395d7f | 1326 | wake_up_all(&wb->bdi->wb_waitq); |
03ba3782 JA |
1327 | } |
1328 | ||
1329 | /* | |
1330 | * Check for periodic writeback, kupdated() style | |
1331 | */ | |
1332 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 1333 | wrote += wb_check_background_flush(wb); |
4452226e | 1334 | clear_bit(WB_writeback_running, &wb->state); |
03ba3782 JA |
1335 | |
1336 | return wrote; | |
1337 | } | |
1338 | ||
1339 | /* | |
1340 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
839a8e86 | 1341 | * reschedules periodically and does kupdated style flushing. |
03ba3782 | 1342 | */ |
f0054bb1 | 1343 | void wb_workfn(struct work_struct *work) |
03ba3782 | 1344 | { |
839a8e86 TH |
1345 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
1346 | struct bdi_writeback, dwork); | |
03ba3782 JA |
1347 | long pages_written; |
1348 | ||
f0054bb1 | 1349 | set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); |
766f9164 | 1350 | current->flags |= PF_SWAPWRITE; |
455b2864 | 1351 | |
839a8e86 | 1352 | if (likely(!current_is_workqueue_rescuer() || |
4452226e | 1353 | !test_bit(WB_registered, &wb->state))) { |
6467716a | 1354 | /* |
f0054bb1 | 1355 | * The normal path. Keep writing back @wb until its |
839a8e86 | 1356 | * work_list is empty. Note that this path is also taken |
f0054bb1 | 1357 | * if @wb is shutting down even when we're running off the |
839a8e86 | 1358 | * rescuer as work_list needs to be drained. |
6467716a | 1359 | */ |
839a8e86 | 1360 | do { |
25d130ba | 1361 | pages_written = wb_do_writeback(wb); |
839a8e86 | 1362 | trace_writeback_pages_written(pages_written); |
f0054bb1 | 1363 | } while (!list_empty(&wb->work_list)); |
839a8e86 TH |
1364 | } else { |
1365 | /* | |
1366 | * bdi_wq can't get enough workers and we're running off | |
1367 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
1368 | * enough for efficient IO. | |
1369 | */ | |
f0054bb1 | 1370 | pages_written = writeback_inodes_wb(wb, 1024, |
839a8e86 | 1371 | WB_REASON_FORKER_THREAD); |
455b2864 | 1372 | trace_writeback_pages_written(pages_written); |
03ba3782 JA |
1373 | } |
1374 | ||
f0054bb1 | 1375 | if (!list_empty(&wb->work_list)) |
6ca738d6 DB |
1376 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
1377 | else if (wb_has_dirty_io(wb) && dirty_writeback_interval) | |
f0054bb1 | 1378 | wb_wakeup_delayed(wb); |
455b2864 | 1379 | |
839a8e86 | 1380 | current->flags &= ~PF_SWAPWRITE; |
03ba3782 JA |
1381 | } |
1382 | ||
1383 | /* | |
b8c2f347 CH |
1384 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
1385 | * the whole world. | |
03ba3782 | 1386 | */ |
0e175a18 | 1387 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
03ba3782 | 1388 | { |
b8c2f347 | 1389 | struct backing_dev_info *bdi; |
03ba3782 | 1390 | |
47df3dde JK |
1391 | if (!nr_pages) |
1392 | nr_pages = get_nr_dirty_pages(); | |
03ba3782 | 1393 | |
b8c2f347 | 1394 | rcu_read_lock(); |
f2b65121 TH |
1395 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
1396 | struct bdi_writeback *wb; | |
1397 | struct wb_iter iter; | |
1398 | ||
1399 | if (!bdi_has_dirty_io(bdi)) | |
1400 | continue; | |
1401 | ||
1402 | bdi_for_each_wb(wb, bdi, &iter, 0) | |
1403 | wb_start_writeback(wb, wb_split_bdi_pages(wb, nr_pages), | |
1404 | false, reason); | |
1405 | } | |
cfc4ba53 | 1406 | rcu_read_unlock(); |
1da177e4 LT |
1407 | } |
1408 | ||
a2f48706 TT |
1409 | /* |
1410 | * Wake up bdi's periodically to make sure dirtytime inodes gets | |
1411 | * written back periodically. We deliberately do *not* check the | |
1412 | * b_dirtytime list in wb_has_dirty_io(), since this would cause the | |
1413 | * kernel to be constantly waking up once there are any dirtytime | |
1414 | * inodes on the system. So instead we define a separate delayed work | |
1415 | * function which gets called much more rarely. (By default, only | |
1416 | * once every 12 hours.) | |
1417 | * | |
1418 | * If there is any other write activity going on in the file system, | |
1419 | * this function won't be necessary. But if the only thing that has | |
1420 | * happened on the file system is a dirtytime inode caused by an atime | |
1421 | * update, we need this infrastructure below to make sure that inode | |
1422 | * eventually gets pushed out to disk. | |
1423 | */ | |
1424 | static void wakeup_dirtytime_writeback(struct work_struct *w); | |
1425 | static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); | |
1426 | ||
1427 | static void wakeup_dirtytime_writeback(struct work_struct *w) | |
1428 | { | |
1429 | struct backing_dev_info *bdi; | |
1430 | ||
1431 | rcu_read_lock(); | |
1432 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { | |
001fe6f6 TH |
1433 | struct bdi_writeback *wb; |
1434 | struct wb_iter iter; | |
1435 | ||
1436 | bdi_for_each_wb(wb, bdi, &iter, 0) | |
1437 | if (!list_empty(&bdi->wb.b_dirty_time)) | |
1438 | wb_wakeup(&bdi->wb); | |
a2f48706 TT |
1439 | } |
1440 | rcu_read_unlock(); | |
1441 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
1442 | } | |
1443 | ||
1444 | static int __init start_dirtytime_writeback(void) | |
1445 | { | |
1446 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
1447 | return 0; | |
1448 | } | |
1449 | __initcall(start_dirtytime_writeback); | |
1450 | ||
1efff914 TT |
1451 | int dirtytime_interval_handler(struct ctl_table *table, int write, |
1452 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1453 | { | |
1454 | int ret; | |
1455 | ||
1456 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
1457 | if (ret == 0 && write) | |
1458 | mod_delayed_work(system_wq, &dirtytime_work, 0); | |
1459 | return ret; | |
1460 | } | |
1461 | ||
03ba3782 JA |
1462 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1463 | { | |
1464 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1465 | struct dentry *dentry; | |
1466 | const char *name = "?"; | |
1467 | ||
1468 | dentry = d_find_alias(inode); | |
1469 | if (dentry) { | |
1470 | spin_lock(&dentry->d_lock); | |
1471 | name = (const char *) dentry->d_name.name; | |
1472 | } | |
1473 | printk(KERN_DEBUG | |
1474 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1475 | current->comm, task_pid_nr(current), inode->i_ino, | |
1476 | name, inode->i_sb->s_id); | |
1477 | if (dentry) { | |
1478 | spin_unlock(&dentry->d_lock); | |
1479 | dput(dentry); | |
1480 | } | |
1481 | } | |
1482 | } | |
1483 | ||
1484 | /** | |
1485 | * __mark_inode_dirty - internal function | |
1486 | * @inode: inode to mark | |
1487 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1488 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1489 | * mark_inode_dirty_sync. | |
1da177e4 | 1490 | * |
03ba3782 JA |
1491 | * Put the inode on the super block's dirty list. |
1492 | * | |
1493 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1494 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1495 | * If it was not hashed, it will never be added to the dirty list | |
1496 | * even if it is later hashed, as it will have been marked dirty already. | |
1497 | * | |
1498 | * In short, make sure you hash any inodes _before_ you start marking | |
1499 | * them dirty. | |
1da177e4 | 1500 | * |
03ba3782 JA |
1501 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1502 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1503 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1504 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1505 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1506 | * blockdev inode. | |
1da177e4 | 1507 | */ |
0ae45f63 | 1508 | #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) |
03ba3782 | 1509 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1510 | { |
03ba3782 | 1511 | struct super_block *sb = inode->i_sb; |
0ae45f63 TT |
1512 | int dirtytime; |
1513 | ||
1514 | trace_writeback_mark_inode_dirty(inode, flags); | |
1da177e4 | 1515 | |
03ba3782 JA |
1516 | /* |
1517 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1518 | * dirty the inode itself | |
1519 | */ | |
0ae45f63 | 1520 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { |
9fb0a7da TH |
1521 | trace_writeback_dirty_inode_start(inode, flags); |
1522 | ||
03ba3782 | 1523 | if (sb->s_op->dirty_inode) |
aa385729 | 1524 | sb->s_op->dirty_inode(inode, flags); |
9fb0a7da TH |
1525 | |
1526 | trace_writeback_dirty_inode(inode, flags); | |
03ba3782 | 1527 | } |
0ae45f63 TT |
1528 | if (flags & I_DIRTY_INODE) |
1529 | flags &= ~I_DIRTY_TIME; | |
1530 | dirtytime = flags & I_DIRTY_TIME; | |
03ba3782 JA |
1531 | |
1532 | /* | |
9c6ac78e TH |
1533 | * Paired with smp_mb() in __writeback_single_inode() for the |
1534 | * following lockless i_state test. See there for details. | |
03ba3782 JA |
1535 | */ |
1536 | smp_mb(); | |
1537 | ||
0ae45f63 TT |
1538 | if (((inode->i_state & flags) == flags) || |
1539 | (dirtytime && (inode->i_state & I_DIRTY_INODE))) | |
03ba3782 JA |
1540 | return; |
1541 | ||
1542 | if (unlikely(block_dump)) | |
1543 | block_dump___mark_inode_dirty(inode); | |
1544 | ||
250df6ed | 1545 | spin_lock(&inode->i_lock); |
0ae45f63 TT |
1546 | if (dirtytime && (inode->i_state & I_DIRTY_INODE)) |
1547 | goto out_unlock_inode; | |
03ba3782 JA |
1548 | if ((inode->i_state & flags) != flags) { |
1549 | const int was_dirty = inode->i_state & I_DIRTY; | |
1550 | ||
52ebea74 TH |
1551 | inode_attach_wb(inode, NULL); |
1552 | ||
0ae45f63 TT |
1553 | if (flags & I_DIRTY_INODE) |
1554 | inode->i_state &= ~I_DIRTY_TIME; | |
03ba3782 JA |
1555 | inode->i_state |= flags; |
1556 | ||
1557 | /* | |
1558 | * If the inode is being synced, just update its dirty state. | |
1559 | * The unlocker will place the inode on the appropriate | |
1560 | * superblock list, based upon its state. | |
1561 | */ | |
1562 | if (inode->i_state & I_SYNC) | |
250df6ed | 1563 | goto out_unlock_inode; |
03ba3782 JA |
1564 | |
1565 | /* | |
1566 | * Only add valid (hashed) inodes to the superblock's | |
1567 | * dirty list. Add blockdev inodes as well. | |
1568 | */ | |
1569 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1570 | if (inode_unhashed(inode)) |
250df6ed | 1571 | goto out_unlock_inode; |
03ba3782 | 1572 | } |
a4ffdde6 | 1573 | if (inode->i_state & I_FREEING) |
250df6ed | 1574 | goto out_unlock_inode; |
03ba3782 JA |
1575 | |
1576 | /* | |
1577 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1578 | * reposition it (that would break b_dirty time-ordering). | |
1579 | */ | |
1580 | if (!was_dirty) { | |
0747259d | 1581 | struct bdi_writeback *wb = inode_to_wb(inode); |
d6c10f1f | 1582 | struct list_head *dirty_list; |
a66979ab | 1583 | bool wakeup_bdi = false; |
253c34e9 | 1584 | |
146d7009 | 1585 | spin_unlock(&inode->i_lock); |
0747259d | 1586 | spin_lock(&wb->list_lock); |
253c34e9 | 1587 | |
0747259d TH |
1588 | WARN(bdi_cap_writeback_dirty(wb->bdi) && |
1589 | !test_bit(WB_registered, &wb->state), | |
1590 | "bdi-%s not registered\n", wb->bdi->name); | |
03ba3782 JA |
1591 | |
1592 | inode->dirtied_when = jiffies; | |
a2f48706 TT |
1593 | if (dirtytime) |
1594 | inode->dirtied_time_when = jiffies; | |
d6c10f1f | 1595 | |
a2f48706 | 1596 | if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) |
0747259d | 1597 | dirty_list = &wb->b_dirty; |
a2f48706 | 1598 | else |
0747259d | 1599 | dirty_list = &wb->b_dirty_time; |
d6c10f1f | 1600 | |
0747259d | 1601 | wakeup_bdi = inode_wb_list_move_locked(inode, wb, |
d6c10f1f TH |
1602 | dirty_list); |
1603 | ||
0747259d | 1604 | spin_unlock(&wb->list_lock); |
0ae45f63 | 1605 | trace_writeback_dirty_inode_enqueue(inode); |
a66979ab | 1606 | |
d6c10f1f TH |
1607 | /* |
1608 | * If this is the first dirty inode for this bdi, | |
1609 | * we have to wake-up the corresponding bdi thread | |
1610 | * to make sure background write-back happens | |
1611 | * later. | |
1612 | */ | |
0747259d TH |
1613 | if (bdi_cap_writeback_dirty(wb->bdi) && wakeup_bdi) |
1614 | wb_wakeup_delayed(wb); | |
a66979ab | 1615 | return; |
1da177e4 | 1616 | } |
1da177e4 | 1617 | } |
250df6ed DC |
1618 | out_unlock_inode: |
1619 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1620 | |
03ba3782 JA |
1621 | } |
1622 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1623 | ||
b6e51316 | 1624 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1625 | { |
1626 | struct inode *inode, *old_inode = NULL; | |
1627 | ||
1628 | /* | |
1629 | * We need to be protected against the filesystem going from | |
1630 | * r/o to r/w or vice versa. | |
1631 | */ | |
b6e51316 | 1632 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1633 | |
55fa6091 | 1634 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1635 | |
1636 | /* | |
1637 | * Data integrity sync. Must wait for all pages under writeback, | |
1638 | * because there may have been pages dirtied before our sync | |
1639 | * call, but which had writeout started before we write it out. | |
1640 | * In which case, the inode may not be on the dirty list, but | |
1641 | * we still have to wait for that writeout. | |
1642 | */ | |
b6e51316 | 1643 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1644 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1645 | |
250df6ed DC |
1646 | spin_lock(&inode->i_lock); |
1647 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1648 | (mapping->nrpages == 0)) { | |
1649 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1650 | continue; |
250df6ed | 1651 | } |
03ba3782 | 1652 | __iget(inode); |
250df6ed | 1653 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1654 | spin_unlock(&inode_sb_list_lock); |
1655 | ||
03ba3782 | 1656 | /* |
55fa6091 DC |
1657 | * We hold a reference to 'inode' so it couldn't have been |
1658 | * removed from s_inodes list while we dropped the | |
1659 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1660 | * be holding the last reference and we cannot iput it under | |
1661 | * inode_sb_list_lock. So we keep the reference and iput it | |
1662 | * later. | |
03ba3782 JA |
1663 | */ |
1664 | iput(old_inode); | |
1665 | old_inode = inode; | |
1666 | ||
1667 | filemap_fdatawait(mapping); | |
1668 | ||
1669 | cond_resched(); | |
1670 | ||
55fa6091 | 1671 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1672 | } |
55fa6091 | 1673 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1674 | iput(old_inode); |
1da177e4 LT |
1675 | } |
1676 | ||
f30a7d0c TH |
1677 | static void __writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr, |
1678 | enum wb_reason reason, bool skip_if_busy) | |
1da177e4 | 1679 | { |
cc395d7f | 1680 | DEFINE_WB_COMPLETION_ONSTACK(done); |
83ba7b07 | 1681 | struct wb_writeback_work work = { |
6e6938b6 WF |
1682 | .sb = sb, |
1683 | .sync_mode = WB_SYNC_NONE, | |
1684 | .tagged_writepages = 1, | |
1685 | .done = &done, | |
1686 | .nr_pages = nr, | |
0e175a18 | 1687 | .reason = reason, |
3c4d7165 | 1688 | }; |
e7972912 | 1689 | struct backing_dev_info *bdi = sb->s_bdi; |
d8a8559c | 1690 | |
e7972912 | 1691 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
6eedc701 | 1692 | return; |
cf37e972 | 1693 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
f30a7d0c | 1694 | |
db125360 | 1695 | bdi_split_work_to_wbs(sb->s_bdi, &work, skip_if_busy); |
cc395d7f | 1696 | wb_wait_for_completion(bdi, &done); |
e913fc82 | 1697 | } |
f30a7d0c TH |
1698 | |
1699 | /** | |
1700 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block | |
1701 | * @sb: the superblock | |
1702 | * @nr: the number of pages to write | |
1703 | * @reason: reason why some writeback work initiated | |
1704 | * | |
1705 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1706 | * on how many (if any) will be written, and this function does not wait | |
1707 | * for IO completion of submitted IO. | |
1708 | */ | |
1709 | void writeback_inodes_sb_nr(struct super_block *sb, | |
1710 | unsigned long nr, | |
1711 | enum wb_reason reason) | |
1712 | { | |
1713 | __writeback_inodes_sb_nr(sb, nr, reason, false); | |
1714 | } | |
3259f8be CM |
1715 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1716 | ||
1717 | /** | |
1718 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1719 | * @sb: the superblock | |
786228ab | 1720 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1721 | * |
1722 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1723 | * on how many (if any) will be written, and this function does not wait | |
1724 | * for IO completion of submitted IO. | |
1725 | */ | |
0e175a18 | 1726 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1727 | { |
0e175a18 | 1728 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1729 | } |
0e3c9a22 | 1730 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1731 | |
17bd55d0 | 1732 | /** |
10ee27a0 | 1733 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway |
17bd55d0 | 1734 | * @sb: the superblock |
10ee27a0 MX |
1735 | * @nr: the number of pages to write |
1736 | * @reason: the reason of writeback | |
17bd55d0 | 1737 | * |
10ee27a0 | 1738 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. |
17bd55d0 ES |
1739 | * Returns 1 if writeback was started, 0 if not. |
1740 | */ | |
f30a7d0c TH |
1741 | bool try_to_writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr, |
1742 | enum wb_reason reason) | |
17bd55d0 | 1743 | { |
10ee27a0 | 1744 | if (!down_read_trylock(&sb->s_umount)) |
f30a7d0c | 1745 | return false; |
10ee27a0 | 1746 | |
f30a7d0c | 1747 | __writeback_inodes_sb_nr(sb, nr, reason, true); |
10ee27a0 | 1748 | up_read(&sb->s_umount); |
f30a7d0c | 1749 | return true; |
17bd55d0 | 1750 | } |
10ee27a0 | 1751 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); |
17bd55d0 | 1752 | |
3259f8be | 1753 | /** |
10ee27a0 | 1754 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
3259f8be | 1755 | * @sb: the superblock |
786228ab | 1756 | * @reason: reason why some writeback work was initiated |
3259f8be | 1757 | * |
10ee27a0 | 1758 | * Implement by try_to_writeback_inodes_sb_nr() |
3259f8be CM |
1759 | * Returns 1 if writeback was started, 0 if not. |
1760 | */ | |
f30a7d0c | 1761 | bool try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1762 | { |
10ee27a0 | 1763 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1764 | } |
10ee27a0 | 1765 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
3259f8be | 1766 | |
d8a8559c JA |
1767 | /** |
1768 | * sync_inodes_sb - sync sb inode pages | |
0dc83bd3 | 1769 | * @sb: the superblock |
d8a8559c JA |
1770 | * |
1771 | * This function writes and waits on any dirty inode belonging to this | |
0dc83bd3 | 1772 | * super_block. |
d8a8559c | 1773 | */ |
0dc83bd3 | 1774 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1775 | { |
cc395d7f | 1776 | DEFINE_WB_COMPLETION_ONSTACK(done); |
83ba7b07 | 1777 | struct wb_writeback_work work = { |
3c4d7165 CH |
1778 | .sb = sb, |
1779 | .sync_mode = WB_SYNC_ALL, | |
1780 | .nr_pages = LONG_MAX, | |
1781 | .range_cyclic = 0, | |
83ba7b07 | 1782 | .done = &done, |
0e175a18 | 1783 | .reason = WB_REASON_SYNC, |
7747bd4b | 1784 | .for_sync = 1, |
3c4d7165 | 1785 | }; |
e7972912 | 1786 | struct backing_dev_info *bdi = sb->s_bdi; |
3c4d7165 | 1787 | |
6eedc701 | 1788 | /* Nothing to do? */ |
e7972912 | 1789 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
6eedc701 | 1790 | return; |
cf37e972 CH |
1791 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1792 | ||
db125360 | 1793 | bdi_split_work_to_wbs(bdi, &work, false); |
cc395d7f | 1794 | wb_wait_for_completion(bdi, &done); |
83ba7b07 | 1795 | |
b6e51316 | 1796 | wait_sb_inodes(sb); |
1da177e4 | 1797 | } |
d8a8559c | 1798 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1799 | |
1da177e4 | 1800 | /** |
7f04c26d AA |
1801 | * write_inode_now - write an inode to disk |
1802 | * @inode: inode to write to disk | |
1803 | * @sync: whether the write should be synchronous or not | |
1804 | * | |
1805 | * This function commits an inode to disk immediately if it is dirty. This is | |
1806 | * primarily needed by knfsd. | |
1da177e4 | 1807 | * |
7f04c26d | 1808 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1809 | */ |
1da177e4 LT |
1810 | int write_inode_now(struct inode *inode, int sync) |
1811 | { | |
f758eeab | 1812 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1813 | struct writeback_control wbc = { |
1814 | .nr_to_write = LONG_MAX, | |
18914b18 | 1815 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1816 | .range_start = 0, |
1817 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1818 | }; |
1819 | ||
1820 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1821 | wbc.nr_to_write = 0; |
1da177e4 LT |
1822 | |
1823 | might_sleep(); | |
4f8ad655 | 1824 | return writeback_single_inode(inode, wb, &wbc); |
1da177e4 LT |
1825 | } |
1826 | EXPORT_SYMBOL(write_inode_now); | |
1827 | ||
1828 | /** | |
1829 | * sync_inode - write an inode and its pages to disk. | |
1830 | * @inode: the inode to sync | |
1831 | * @wbc: controls the writeback mode | |
1832 | * | |
1833 | * sync_inode() will write an inode and its pages to disk. It will also | |
1834 | * correctly update the inode on its superblock's dirty inode lists and will | |
1835 | * update inode->i_state. | |
1836 | * | |
1837 | * The caller must have a ref on the inode. | |
1838 | */ | |
1839 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1840 | { | |
4f8ad655 | 1841 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); |
1da177e4 LT |
1842 | } |
1843 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1844 | |
1845 | /** | |
c691b9d9 | 1846 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1847 | * @inode: the inode to sync |
1848 | * @wait: wait for I/O to complete. | |
1849 | * | |
c691b9d9 | 1850 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1851 | * |
1852 | * Note: only writes the actual inode, no associated data or other metadata. | |
1853 | */ | |
1854 | int sync_inode_metadata(struct inode *inode, int wait) | |
1855 | { | |
1856 | struct writeback_control wbc = { | |
1857 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1858 | .nr_to_write = 0, /* metadata-only */ | |
1859 | }; | |
1860 | ||
1861 | return sync_inode(inode, &wbc); | |
1862 | } | |
1863 | EXPORT_SYMBOL(sync_inode_metadata); |