]>
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 | ||
295 | #else /* CONFIG_CGROUP_WRITEBACK */ | |
296 | ||
297 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) | |
298 | { | |
299 | return nr_pages; | |
300 | } | |
301 | ||
703c2708 TH |
302 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
303 | ||
c00ddad3 TH |
304 | void wb_start_writeback(struct bdi_writeback *wb, long nr_pages, |
305 | bool range_cyclic, enum wb_reason reason) | |
b6e51316 | 306 | { |
c00ddad3 TH |
307 | struct wb_writeback_work *work; |
308 | ||
309 | if (!wb_has_dirty_io(wb)) | |
310 | return; | |
311 | ||
312 | /* | |
313 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
314 | * wakeup the thread for old dirty data writeback | |
315 | */ | |
316 | work = kzalloc(sizeof(*work), GFP_ATOMIC); | |
317 | if (!work) { | |
318 | trace_writeback_nowork(wb->bdi); | |
319 | wb_wakeup(wb); | |
320 | return; | |
321 | } | |
322 | ||
323 | work->sync_mode = WB_SYNC_NONE; | |
324 | work->nr_pages = nr_pages; | |
325 | work->range_cyclic = range_cyclic; | |
326 | work->reason = reason; | |
ac7b19a3 | 327 | work->auto_free = 1; |
c00ddad3 TH |
328 | |
329 | wb_queue_work(wb, work); | |
c5444198 | 330 | } |
d3ddec76 | 331 | |
c5444198 | 332 | /** |
9ecf4866 TH |
333 | * wb_start_background_writeback - start background writeback |
334 | * @wb: bdi_writback to write from | |
c5444198 CH |
335 | * |
336 | * Description: | |
6585027a | 337 | * This makes sure WB_SYNC_NONE background writeback happens. When |
9ecf4866 | 338 | * this function returns, it is only guaranteed that for given wb |
6585027a JK |
339 | * some IO is happening if we are over background dirty threshold. |
340 | * Caller need not hold sb s_umount semaphore. | |
c5444198 | 341 | */ |
9ecf4866 | 342 | void wb_start_background_writeback(struct bdi_writeback *wb) |
c5444198 | 343 | { |
6585027a JK |
344 | /* |
345 | * We just wake up the flusher thread. It will perform background | |
346 | * writeback as soon as there is no other work to do. | |
347 | */ | |
9ecf4866 TH |
348 | trace_writeback_wake_background(wb->bdi); |
349 | wb_wakeup(wb); | |
1da177e4 LT |
350 | } |
351 | ||
a66979ab DC |
352 | /* |
353 | * Remove the inode from the writeback list it is on. | |
354 | */ | |
355 | void inode_wb_list_del(struct inode *inode) | |
356 | { | |
52ebea74 | 357 | struct bdi_writeback *wb = inode_to_wb(inode); |
f758eeab | 358 | |
52ebea74 | 359 | spin_lock(&wb->list_lock); |
d6c10f1f | 360 | inode_wb_list_del_locked(inode, wb); |
52ebea74 | 361 | spin_unlock(&wb->list_lock); |
a66979ab DC |
362 | } |
363 | ||
6610a0bc AM |
364 | /* |
365 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
366 | * furthest end of its superblock's dirty-inode list. | |
367 | * | |
368 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 369 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
370 | * the case then the inode must have been redirtied while it was being written |
371 | * out and we don't reset its dirtied_when. | |
372 | */ | |
f758eeab | 373 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 374 | { |
03ba3782 | 375 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 376 | struct inode *tail; |
6610a0bc | 377 | |
7ccf19a8 | 378 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 379 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
380 | inode->dirtied_when = jiffies; |
381 | } | |
d6c10f1f | 382 | inode_wb_list_move_locked(inode, wb, &wb->b_dirty); |
6610a0bc AM |
383 | } |
384 | ||
c986d1e2 | 385 | /* |
66f3b8e2 | 386 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 387 | */ |
f758eeab | 388 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 389 | { |
d6c10f1f | 390 | inode_wb_list_move_locked(inode, wb, &wb->b_more_io); |
c986d1e2 AM |
391 | } |
392 | ||
1c0eeaf5 JE |
393 | static void inode_sync_complete(struct inode *inode) |
394 | { | |
365b94ae | 395 | inode->i_state &= ~I_SYNC; |
4eff96dd JK |
396 | /* If inode is clean an unused, put it into LRU now... */ |
397 | inode_add_lru(inode); | |
365b94ae | 398 | /* Waiters must see I_SYNC cleared before being woken up */ |
1c0eeaf5 JE |
399 | smp_mb(); |
400 | wake_up_bit(&inode->i_state, __I_SYNC); | |
401 | } | |
402 | ||
d2caa3c5 JL |
403 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
404 | { | |
405 | bool ret = time_after(inode->dirtied_when, t); | |
406 | #ifndef CONFIG_64BIT | |
407 | /* | |
408 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
409 | * It _appears_ to be in the future, but is actually in distant past. | |
410 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 411 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
412 | */ |
413 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
414 | #endif | |
415 | return ret; | |
416 | } | |
417 | ||
0ae45f63 TT |
418 | #define EXPIRE_DIRTY_ATIME 0x0001 |
419 | ||
2c136579 | 420 | /* |
0e2f2b23 | 421 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
697e6fed | 422 | * @delaying_queue to @dispatch_queue. |
2c136579 | 423 | */ |
e84d0a4f | 424 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 425 | struct list_head *dispatch_queue, |
0ae45f63 | 426 | int flags, |
ad4e38dd | 427 | struct wb_writeback_work *work) |
2c136579 | 428 | { |
0ae45f63 TT |
429 | unsigned long *older_than_this = NULL; |
430 | unsigned long expire_time; | |
5c03449d SL |
431 | LIST_HEAD(tmp); |
432 | struct list_head *pos, *node; | |
cf137307 | 433 | struct super_block *sb = NULL; |
5c03449d | 434 | struct inode *inode; |
cf137307 | 435 | int do_sb_sort = 0; |
e84d0a4f | 436 | int moved = 0; |
5c03449d | 437 | |
0ae45f63 TT |
438 | if ((flags & EXPIRE_DIRTY_ATIME) == 0) |
439 | older_than_this = work->older_than_this; | |
a2f48706 TT |
440 | else if (!work->for_sync) { |
441 | expire_time = jiffies - (dirtytime_expire_interval * HZ); | |
0ae45f63 TT |
442 | older_than_this = &expire_time; |
443 | } | |
2c136579 | 444 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 445 | inode = wb_inode(delaying_queue->prev); |
0ae45f63 TT |
446 | if (older_than_this && |
447 | inode_dirtied_after(inode, *older_than_this)) | |
2c136579 | 448 | break; |
a8855990 JK |
449 | list_move(&inode->i_wb_list, &tmp); |
450 | moved++; | |
0ae45f63 TT |
451 | if (flags & EXPIRE_DIRTY_ATIME) |
452 | set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); | |
a8855990 JK |
453 | if (sb_is_blkdev_sb(inode->i_sb)) |
454 | continue; | |
cf137307 JA |
455 | if (sb && sb != inode->i_sb) |
456 | do_sb_sort = 1; | |
457 | sb = inode->i_sb; | |
5c03449d SL |
458 | } |
459 | ||
cf137307 JA |
460 | /* just one sb in list, splice to dispatch_queue and we're done */ |
461 | if (!do_sb_sort) { | |
462 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 463 | goto out; |
cf137307 JA |
464 | } |
465 | ||
5c03449d SL |
466 | /* Move inodes from one superblock together */ |
467 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 468 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 469 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 470 | inode = wb_inode(pos); |
5c03449d | 471 | if (inode->i_sb == sb) |
7ccf19a8 | 472 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 473 | } |
2c136579 | 474 | } |
e84d0a4f WF |
475 | out: |
476 | return moved; | |
2c136579 FW |
477 | } |
478 | ||
479 | /* | |
480 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
481 | * Before |
482 | * newly dirtied b_dirty b_io b_more_io | |
483 | * =============> gf edc BA | |
484 | * After | |
485 | * newly dirtied b_dirty b_io b_more_io | |
486 | * =============> g fBAedc | |
487 | * | | |
488 | * +--> dequeue for IO | |
2c136579 | 489 | */ |
ad4e38dd | 490 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
66f3b8e2 | 491 | { |
e84d0a4f | 492 | int moved; |
0ae45f63 | 493 | |
f758eeab | 494 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 495 | list_splice_init(&wb->b_more_io, &wb->b_io); |
0ae45f63 TT |
496 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); |
497 | moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, | |
498 | EXPIRE_DIRTY_ATIME, work); | |
d6c10f1f TH |
499 | if (moved) |
500 | wb_io_lists_populated(wb); | |
ad4e38dd | 501 | trace_writeback_queue_io(wb, work, moved); |
66f3b8e2 JA |
502 | } |
503 | ||
a9185b41 | 504 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 505 | { |
9fb0a7da TH |
506 | int ret; |
507 | ||
508 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
509 | trace_writeback_write_inode_start(inode, wbc); | |
510 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
511 | trace_writeback_write_inode(inode, wbc); | |
512 | return ret; | |
513 | } | |
03ba3782 | 514 | return 0; |
08d8e974 | 515 | } |
08d8e974 | 516 | |
1da177e4 | 517 | /* |
169ebd90 JK |
518 | * Wait for writeback on an inode to complete. Called with i_lock held. |
519 | * Caller must make sure inode cannot go away when we drop i_lock. | |
01c03194 | 520 | */ |
169ebd90 JK |
521 | static void __inode_wait_for_writeback(struct inode *inode) |
522 | __releases(inode->i_lock) | |
523 | __acquires(inode->i_lock) | |
01c03194 CH |
524 | { |
525 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
526 | wait_queue_head_t *wqh; | |
527 | ||
528 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
529 | while (inode->i_state & I_SYNC) { |
530 | spin_unlock(&inode->i_lock); | |
74316201 N |
531 | __wait_on_bit(wqh, &wq, bit_wait, |
532 | TASK_UNINTERRUPTIBLE); | |
250df6ed | 533 | spin_lock(&inode->i_lock); |
58a9d3d8 | 534 | } |
01c03194 CH |
535 | } |
536 | ||
169ebd90 JK |
537 | /* |
538 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
539 | */ | |
540 | void inode_wait_for_writeback(struct inode *inode) | |
541 | { | |
542 | spin_lock(&inode->i_lock); | |
543 | __inode_wait_for_writeback(inode); | |
544 | spin_unlock(&inode->i_lock); | |
545 | } | |
546 | ||
547 | /* | |
548 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
549 | * held and drops it. It is aimed for callers not holding any inode reference | |
550 | * so once i_lock is dropped, inode can go away. | |
551 | */ | |
552 | static void inode_sleep_on_writeback(struct inode *inode) | |
553 | __releases(inode->i_lock) | |
554 | { | |
555 | DEFINE_WAIT(wait); | |
556 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
557 | int sleep; | |
558 | ||
559 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
560 | sleep = inode->i_state & I_SYNC; | |
561 | spin_unlock(&inode->i_lock); | |
562 | if (sleep) | |
563 | schedule(); | |
564 | finish_wait(wqh, &wait); | |
565 | } | |
566 | ||
ccb26b5a JK |
567 | /* |
568 | * Find proper writeback list for the inode depending on its current state and | |
569 | * possibly also change of its state while we were doing writeback. Here we | |
570 | * handle things such as livelock prevention or fairness of writeback among | |
571 | * inodes. This function can be called only by flusher thread - noone else | |
572 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
573 | * thread's back can have unexpected consequences. | |
574 | */ | |
575 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
576 | struct writeback_control *wbc) | |
577 | { | |
578 | if (inode->i_state & I_FREEING) | |
579 | return; | |
580 | ||
581 | /* | |
582 | * Sync livelock prevention. Each inode is tagged and synced in one | |
583 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
584 | * the dirty time to prevent enqueue and sync it again. | |
585 | */ | |
586 | if ((inode->i_state & I_DIRTY) && | |
587 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
588 | inode->dirtied_when = jiffies; | |
589 | ||
4f8ad655 JK |
590 | if (wbc->pages_skipped) { |
591 | /* | |
592 | * writeback is not making progress due to locked | |
593 | * buffers. Skip this inode for now. | |
594 | */ | |
595 | redirty_tail(inode, wb); | |
596 | return; | |
597 | } | |
598 | ||
ccb26b5a JK |
599 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
600 | /* | |
601 | * We didn't write back all the pages. nfs_writepages() | |
602 | * sometimes bales out without doing anything. | |
603 | */ | |
604 | if (wbc->nr_to_write <= 0) { | |
605 | /* Slice used up. Queue for next turn. */ | |
606 | requeue_io(inode, wb); | |
607 | } else { | |
608 | /* | |
609 | * Writeback blocked by something other than | |
610 | * congestion. Delay the inode for some time to | |
611 | * avoid spinning on the CPU (100% iowait) | |
612 | * retrying writeback of the dirty page/inode | |
613 | * that cannot be performed immediately. | |
614 | */ | |
615 | redirty_tail(inode, wb); | |
616 | } | |
617 | } else if (inode->i_state & I_DIRTY) { | |
618 | /* | |
619 | * Filesystems can dirty the inode during writeback operations, | |
620 | * such as delayed allocation during submission or metadata | |
621 | * updates after data IO completion. | |
622 | */ | |
623 | redirty_tail(inode, wb); | |
0ae45f63 | 624 | } else if (inode->i_state & I_DIRTY_TIME) { |
a2f48706 | 625 | inode->dirtied_when = jiffies; |
d6c10f1f | 626 | inode_wb_list_move_locked(inode, wb, &wb->b_dirty_time); |
ccb26b5a JK |
627 | } else { |
628 | /* The inode is clean. Remove from writeback lists. */ | |
d6c10f1f | 629 | inode_wb_list_del_locked(inode, wb); |
ccb26b5a JK |
630 | } |
631 | } | |
632 | ||
01c03194 | 633 | /* |
4f8ad655 JK |
634 | * Write out an inode and its dirty pages. Do not update the writeback list |
635 | * linkage. That is left to the caller. The caller is also responsible for | |
636 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. | |
1da177e4 LT |
637 | */ |
638 | static int | |
cd8ed2a4 | 639 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 640 | { |
1da177e4 | 641 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 642 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 643 | unsigned dirty; |
1da177e4 LT |
644 | int ret; |
645 | ||
4f8ad655 | 646 | WARN_ON(!(inode->i_state & I_SYNC)); |
1da177e4 | 647 | |
9fb0a7da TH |
648 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
649 | ||
1da177e4 LT |
650 | ret = do_writepages(mapping, wbc); |
651 | ||
26821ed4 CH |
652 | /* |
653 | * Make sure to wait on the data before writing out the metadata. | |
654 | * This is important for filesystems that modify metadata on data | |
7747bd4b DC |
655 | * I/O completion. We don't do it for sync(2) writeback because it has a |
656 | * separate, external IO completion path and ->sync_fs for guaranteeing | |
657 | * inode metadata is written back correctly. | |
26821ed4 | 658 | */ |
7747bd4b | 659 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
26821ed4 | 660 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
661 | if (ret == 0) |
662 | ret = err; | |
663 | } | |
664 | ||
5547e8aa DM |
665 | /* |
666 | * Some filesystems may redirty the inode during the writeback | |
667 | * due to delalloc, clear dirty metadata flags right before | |
668 | * write_inode() | |
669 | */ | |
250df6ed | 670 | spin_lock(&inode->i_lock); |
9c6ac78e | 671 | |
5547e8aa | 672 | dirty = inode->i_state & I_DIRTY; |
a2f48706 TT |
673 | if (inode->i_state & I_DIRTY_TIME) { |
674 | if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
675 | unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || | |
676 | unlikely(time_after(jiffies, | |
677 | (inode->dirtied_time_when + | |
678 | dirtytime_expire_interval * HZ)))) { | |
679 | dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; | |
680 | trace_writeback_lazytime(inode); | |
681 | } | |
682 | } else | |
683 | inode->i_state &= ~I_DIRTY_TIME_EXPIRED; | |
0ae45f63 | 684 | inode->i_state &= ~dirty; |
9c6ac78e TH |
685 | |
686 | /* | |
687 | * Paired with smp_mb() in __mark_inode_dirty(). This allows | |
688 | * __mark_inode_dirty() to test i_state without grabbing i_lock - | |
689 | * either they see the I_DIRTY bits cleared or we see the dirtied | |
690 | * inode. | |
691 | * | |
692 | * I_DIRTY_PAGES is always cleared together above even if @mapping | |
693 | * still has dirty pages. The flag is reinstated after smp_mb() if | |
694 | * necessary. This guarantees that either __mark_inode_dirty() | |
695 | * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. | |
696 | */ | |
697 | smp_mb(); | |
698 | ||
699 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
700 | inode->i_state |= I_DIRTY_PAGES; | |
701 | ||
250df6ed | 702 | spin_unlock(&inode->i_lock); |
9c6ac78e | 703 | |
0ae45f63 TT |
704 | if (dirty & I_DIRTY_TIME) |
705 | mark_inode_dirty_sync(inode); | |
26821ed4 | 706 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
0ae45f63 | 707 | if (dirty & ~I_DIRTY_PAGES) { |
a9185b41 | 708 | int err = write_inode(inode, wbc); |
1da177e4 LT |
709 | if (ret == 0) |
710 | ret = err; | |
711 | } | |
4f8ad655 JK |
712 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
713 | return ret; | |
714 | } | |
715 | ||
716 | /* | |
717 | * Write out an inode's dirty pages. Either the caller has an active reference | |
718 | * on the inode or the inode has I_WILL_FREE set. | |
719 | * | |
720 | * This function is designed to be called for writing back one inode which | |
721 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() | |
722 | * and does more profound writeback list handling in writeback_sb_inodes(). | |
723 | */ | |
724 | static int | |
725 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, | |
726 | struct writeback_control *wbc) | |
727 | { | |
728 | int ret = 0; | |
729 | ||
730 | spin_lock(&inode->i_lock); | |
731 | if (!atomic_read(&inode->i_count)) | |
732 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
733 | else | |
734 | WARN_ON(inode->i_state & I_WILL_FREE); | |
735 | ||
736 | if (inode->i_state & I_SYNC) { | |
737 | if (wbc->sync_mode != WB_SYNC_ALL) | |
738 | goto out; | |
739 | /* | |
169ebd90 JK |
740 | * It's a data-integrity sync. We must wait. Since callers hold |
741 | * inode reference or inode has I_WILL_FREE set, it cannot go | |
742 | * away under us. | |
4f8ad655 | 743 | */ |
169ebd90 | 744 | __inode_wait_for_writeback(inode); |
4f8ad655 JK |
745 | } |
746 | WARN_ON(inode->i_state & I_SYNC); | |
747 | /* | |
f9b0e058 JK |
748 | * Skip inode if it is clean and we have no outstanding writeback in |
749 | * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this | |
750 | * function since flusher thread may be doing for example sync in | |
751 | * parallel and if we move the inode, it could get skipped. So here we | |
752 | * make sure inode is on some writeback list and leave it there unless | |
753 | * we have completely cleaned the inode. | |
4f8ad655 | 754 | */ |
0ae45f63 | 755 | if (!(inode->i_state & I_DIRTY_ALL) && |
f9b0e058 JK |
756 | (wbc->sync_mode != WB_SYNC_ALL || |
757 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) | |
4f8ad655 JK |
758 | goto out; |
759 | inode->i_state |= I_SYNC; | |
760 | spin_unlock(&inode->i_lock); | |
761 | ||
cd8ed2a4 | 762 | ret = __writeback_single_inode(inode, wbc); |
1da177e4 | 763 | |
f758eeab | 764 | spin_lock(&wb->list_lock); |
250df6ed | 765 | spin_lock(&inode->i_lock); |
4f8ad655 JK |
766 | /* |
767 | * If inode is clean, remove it from writeback lists. Otherwise don't | |
768 | * touch it. See comment above for explanation. | |
769 | */ | |
0ae45f63 | 770 | if (!(inode->i_state & I_DIRTY_ALL)) |
d6c10f1f | 771 | inode_wb_list_del_locked(inode, wb); |
4f8ad655 | 772 | spin_unlock(&wb->list_lock); |
1c0eeaf5 | 773 | inode_sync_complete(inode); |
4f8ad655 JK |
774 | out: |
775 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
776 | return ret; |
777 | } | |
778 | ||
a88a341a | 779 | static long writeback_chunk_size(struct bdi_writeback *wb, |
1a12d8bd | 780 | struct wb_writeback_work *work) |
d46db3d5 WF |
781 | { |
782 | long pages; | |
783 | ||
784 | /* | |
785 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
786 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
787 | * here avoids calling into writeback_inodes_wb() more than once. | |
788 | * | |
789 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
790 | * | |
791 | * wb_writeback() | |
792 | * writeback_sb_inodes() <== called only once | |
793 | * write_cache_pages() <== called once for each inode | |
794 | * (quickly) tag currently dirty pages | |
795 | * (maybe slowly) sync all tagged pages | |
796 | */ | |
797 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
798 | pages = LONG_MAX; | |
1a12d8bd | 799 | else { |
a88a341a | 800 | pages = min(wb->avg_write_bandwidth / 2, |
1a12d8bd WF |
801 | global_dirty_limit / DIRTY_SCOPE); |
802 | pages = min(pages, work->nr_pages); | |
803 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
804 | MIN_WRITEBACK_PAGES); | |
805 | } | |
d46db3d5 WF |
806 | |
807 | return pages; | |
808 | } | |
809 | ||
f11c9c5c ES |
810 | /* |
811 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 | 812 | * |
d46db3d5 | 813 | * Return the number of pages and/or inodes written. |
f11c9c5c | 814 | */ |
d46db3d5 WF |
815 | static long writeback_sb_inodes(struct super_block *sb, |
816 | struct bdi_writeback *wb, | |
817 | struct wb_writeback_work *work) | |
1da177e4 | 818 | { |
d46db3d5 WF |
819 | struct writeback_control wbc = { |
820 | .sync_mode = work->sync_mode, | |
821 | .tagged_writepages = work->tagged_writepages, | |
822 | .for_kupdate = work->for_kupdate, | |
823 | .for_background = work->for_background, | |
7747bd4b | 824 | .for_sync = work->for_sync, |
d46db3d5 WF |
825 | .range_cyclic = work->range_cyclic, |
826 | .range_start = 0, | |
827 | .range_end = LLONG_MAX, | |
828 | }; | |
829 | unsigned long start_time = jiffies; | |
830 | long write_chunk; | |
831 | long wrote = 0; /* count both pages and inodes */ | |
832 | ||
03ba3782 | 833 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 834 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
835 | |
836 | if (inode->i_sb != sb) { | |
d46db3d5 | 837 | if (work->sb) { |
edadfb10 CH |
838 | /* |
839 | * We only want to write back data for this | |
840 | * superblock, move all inodes not belonging | |
841 | * to it back onto the dirty list. | |
842 | */ | |
f758eeab | 843 | redirty_tail(inode, wb); |
edadfb10 CH |
844 | continue; |
845 | } | |
846 | ||
847 | /* | |
848 | * The inode belongs to a different superblock. | |
849 | * Bounce back to the caller to unpin this and | |
850 | * pin the next superblock. | |
851 | */ | |
d46db3d5 | 852 | break; |
edadfb10 CH |
853 | } |
854 | ||
9843b76a | 855 | /* |
331cbdee WL |
856 | * Don't bother with new inodes or inodes being freed, first |
857 | * kind does not need periodic writeout yet, and for the latter | |
9843b76a CH |
858 | * kind writeout is handled by the freer. |
859 | */ | |
250df6ed | 860 | spin_lock(&inode->i_lock); |
9843b76a | 861 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 862 | spin_unlock(&inode->i_lock); |
fcc5c222 | 863 | redirty_tail(inode, wb); |
7ef0d737 NP |
864 | continue; |
865 | } | |
cc1676d9 JK |
866 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
867 | /* | |
868 | * If this inode is locked for writeback and we are not | |
869 | * doing writeback-for-data-integrity, move it to | |
870 | * b_more_io so that writeback can proceed with the | |
871 | * other inodes on s_io. | |
872 | * | |
873 | * We'll have another go at writing back this inode | |
874 | * when we completed a full scan of b_io. | |
875 | */ | |
876 | spin_unlock(&inode->i_lock); | |
877 | requeue_io(inode, wb); | |
878 | trace_writeback_sb_inodes_requeue(inode); | |
879 | continue; | |
880 | } | |
f0d07b7f JK |
881 | spin_unlock(&wb->list_lock); |
882 | ||
4f8ad655 JK |
883 | /* |
884 | * We already requeued the inode if it had I_SYNC set and we | |
885 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
886 | * WB_SYNC_ALL case. | |
887 | */ | |
169ebd90 JK |
888 | if (inode->i_state & I_SYNC) { |
889 | /* Wait for I_SYNC. This function drops i_lock... */ | |
890 | inode_sleep_on_writeback(inode); | |
891 | /* Inode may be gone, start again */ | |
ead188f9 | 892 | spin_lock(&wb->list_lock); |
169ebd90 JK |
893 | continue; |
894 | } | |
4f8ad655 JK |
895 | inode->i_state |= I_SYNC; |
896 | spin_unlock(&inode->i_lock); | |
169ebd90 | 897 | |
a88a341a | 898 | write_chunk = writeback_chunk_size(wb, work); |
d46db3d5 WF |
899 | wbc.nr_to_write = write_chunk; |
900 | wbc.pages_skipped = 0; | |
250df6ed | 901 | |
169ebd90 JK |
902 | /* |
903 | * We use I_SYNC to pin the inode in memory. While it is set | |
904 | * evict_inode() will wait so the inode cannot be freed. | |
905 | */ | |
cd8ed2a4 | 906 | __writeback_single_inode(inode, &wbc); |
250df6ed | 907 | |
d46db3d5 WF |
908 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
909 | wrote += write_chunk - wbc.nr_to_write; | |
4f8ad655 JK |
910 | spin_lock(&wb->list_lock); |
911 | spin_lock(&inode->i_lock); | |
0ae45f63 | 912 | if (!(inode->i_state & I_DIRTY_ALL)) |
d46db3d5 | 913 | wrote++; |
4f8ad655 JK |
914 | requeue_inode(inode, wb, &wbc); |
915 | inode_sync_complete(inode); | |
0f1b1fd8 | 916 | spin_unlock(&inode->i_lock); |
169ebd90 | 917 | cond_resched_lock(&wb->list_lock); |
d46db3d5 WF |
918 | /* |
919 | * bail out to wb_writeback() often enough to check | |
920 | * background threshold and other termination conditions. | |
921 | */ | |
922 | if (wrote) { | |
923 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
924 | break; | |
925 | if (work->nr_pages <= 0) | |
926 | break; | |
8bc3be27 | 927 | } |
1da177e4 | 928 | } |
d46db3d5 | 929 | return wrote; |
f11c9c5c ES |
930 | } |
931 | ||
d46db3d5 WF |
932 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
933 | struct wb_writeback_work *work) | |
f11c9c5c | 934 | { |
d46db3d5 WF |
935 | unsigned long start_time = jiffies; |
936 | long wrote = 0; | |
38f21977 | 937 | |
f11c9c5c | 938 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 939 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 940 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 941 | |
eb6ef3df | 942 | if (!trylock_super(sb)) { |
0e995816 | 943 | /* |
eb6ef3df | 944 | * trylock_super() may fail consistently due to |
0e995816 WF |
945 | * s_umount being grabbed by someone else. Don't use |
946 | * requeue_io() to avoid busy retrying the inode/sb. | |
947 | */ | |
948 | redirty_tail(inode, wb); | |
edadfb10 | 949 | continue; |
f11c9c5c | 950 | } |
d46db3d5 | 951 | wrote += writeback_sb_inodes(sb, wb, work); |
eb6ef3df | 952 | up_read(&sb->s_umount); |
f11c9c5c | 953 | |
d46db3d5 WF |
954 | /* refer to the same tests at the end of writeback_sb_inodes */ |
955 | if (wrote) { | |
956 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
957 | break; | |
958 | if (work->nr_pages <= 0) | |
959 | break; | |
960 | } | |
f11c9c5c | 961 | } |
66f3b8e2 | 962 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 963 | return wrote; |
66f3b8e2 JA |
964 | } |
965 | ||
7d9f073b | 966 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
0e175a18 | 967 | enum wb_reason reason) |
edadfb10 | 968 | { |
d46db3d5 WF |
969 | struct wb_writeback_work work = { |
970 | .nr_pages = nr_pages, | |
971 | .sync_mode = WB_SYNC_NONE, | |
972 | .range_cyclic = 1, | |
0e175a18 | 973 | .reason = reason, |
d46db3d5 | 974 | }; |
edadfb10 | 975 | |
f758eeab | 976 | spin_lock(&wb->list_lock); |
424b351f | 977 | if (list_empty(&wb->b_io)) |
ad4e38dd | 978 | queue_io(wb, &work); |
d46db3d5 | 979 | __writeback_inodes_wb(wb, &work); |
f758eeab | 980 | spin_unlock(&wb->list_lock); |
edadfb10 | 981 | |
d46db3d5 WF |
982 | return nr_pages - work.nr_pages; |
983 | } | |
03ba3782 | 984 | |
a88a341a | 985 | static bool over_bground_thresh(struct bdi_writeback *wb) |
03ba3782 JA |
986 | { |
987 | unsigned long background_thresh, dirty_thresh; | |
988 | ||
16c4042f | 989 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 | 990 | |
b00949aa WF |
991 | if (global_page_state(NR_FILE_DIRTY) + |
992 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) | |
993 | return true; | |
994 | ||
a88a341a | 995 | if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh)) |
b00949aa WF |
996 | return true; |
997 | ||
998 | return false; | |
03ba3782 JA |
999 | } |
1000 | ||
e98be2d5 WF |
1001 | /* |
1002 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
1003 | * only the flusher working on the first wb should do it. | |
1004 | */ | |
1005 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
1006 | unsigned long start_time) | |
1007 | { | |
a88a341a | 1008 | __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time); |
e98be2d5 WF |
1009 | } |
1010 | ||
03ba3782 JA |
1011 | /* |
1012 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 1013 | * |
03ba3782 JA |
1014 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
1015 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
1016 | * just walks the superblock inode list, writing back any inodes which are | |
1017 | * older than a specific point in time. | |
66f3b8e2 | 1018 | * |
03ba3782 JA |
1019 | * Try to run once per dirty_writeback_interval. But if a writeback event |
1020 | * takes longer than a dirty_writeback_interval interval, then leave a | |
1021 | * one-second gap. | |
66f3b8e2 | 1022 | * |
03ba3782 JA |
1023 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
1024 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 1025 | */ |
c4a77a6c | 1026 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 1027 | struct wb_writeback_work *work) |
66f3b8e2 | 1028 | { |
e98be2d5 | 1029 | unsigned long wb_start = jiffies; |
d46db3d5 | 1030 | long nr_pages = work->nr_pages; |
0dc83bd3 | 1031 | unsigned long oldest_jif; |
a5989bdc | 1032 | struct inode *inode; |
d46db3d5 | 1033 | long progress; |
66f3b8e2 | 1034 | |
0dc83bd3 JK |
1035 | oldest_jif = jiffies; |
1036 | work->older_than_this = &oldest_jif; | |
38f21977 | 1037 | |
e8dfc305 | 1038 | spin_lock(&wb->list_lock); |
03ba3782 JA |
1039 | for (;;) { |
1040 | /* | |
d3ddec76 | 1041 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 1042 | */ |
83ba7b07 | 1043 | if (work->nr_pages <= 0) |
03ba3782 | 1044 | break; |
66f3b8e2 | 1045 | |
aa373cf5 JK |
1046 | /* |
1047 | * Background writeout and kupdate-style writeback may | |
1048 | * run forever. Stop them if there is other work to do | |
1049 | * so that e.g. sync can proceed. They'll be restarted | |
1050 | * after the other works are all done. | |
1051 | */ | |
1052 | if ((work->for_background || work->for_kupdate) && | |
f0054bb1 | 1053 | !list_empty(&wb->work_list)) |
aa373cf5 JK |
1054 | break; |
1055 | ||
38f21977 | 1056 | /* |
d3ddec76 WF |
1057 | * For background writeout, stop when we are below the |
1058 | * background dirty threshold | |
38f21977 | 1059 | */ |
a88a341a | 1060 | if (work->for_background && !over_bground_thresh(wb)) |
03ba3782 | 1061 | break; |
38f21977 | 1062 | |
1bc36b64 JK |
1063 | /* |
1064 | * Kupdate and background works are special and we want to | |
1065 | * include all inodes that need writing. Livelock avoidance is | |
1066 | * handled by these works yielding to any other work so we are | |
1067 | * safe. | |
1068 | */ | |
ba9aa839 | 1069 | if (work->for_kupdate) { |
0dc83bd3 | 1070 | oldest_jif = jiffies - |
ba9aa839 | 1071 | msecs_to_jiffies(dirty_expire_interval * 10); |
1bc36b64 | 1072 | } else if (work->for_background) |
0dc83bd3 | 1073 | oldest_jif = jiffies; |
028c2dd1 | 1074 | |
d46db3d5 | 1075 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 1076 | if (list_empty(&wb->b_io)) |
ad4e38dd | 1077 | queue_io(wb, work); |
83ba7b07 | 1078 | if (work->sb) |
d46db3d5 | 1079 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 1080 | else |
d46db3d5 WF |
1081 | progress = __writeback_inodes_wb(wb, work); |
1082 | trace_writeback_written(wb->bdi, work); | |
028c2dd1 | 1083 | |
e98be2d5 | 1084 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
1085 | |
1086 | /* | |
e6fb6da2 WF |
1087 | * Did we write something? Try for more |
1088 | * | |
1089 | * Dirty inodes are moved to b_io for writeback in batches. | |
1090 | * The completion of the current batch does not necessarily | |
1091 | * mean the overall work is done. So we keep looping as long | |
1092 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 1093 | */ |
d46db3d5 | 1094 | if (progress) |
71fd05a8 JA |
1095 | continue; |
1096 | /* | |
e6fb6da2 | 1097 | * No more inodes for IO, bail |
71fd05a8 | 1098 | */ |
b7a2441f | 1099 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 1100 | break; |
71fd05a8 JA |
1101 | /* |
1102 | * Nothing written. Wait for some inode to | |
1103 | * become available for writeback. Otherwise | |
1104 | * we'll just busyloop. | |
1105 | */ | |
71fd05a8 | 1106 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 1107 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 1108 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 1109 | spin_lock(&inode->i_lock); |
f0d07b7f | 1110 | spin_unlock(&wb->list_lock); |
169ebd90 JK |
1111 | /* This function drops i_lock... */ |
1112 | inode_sleep_on_writeback(inode); | |
f0d07b7f | 1113 | spin_lock(&wb->list_lock); |
03ba3782 JA |
1114 | } |
1115 | } | |
e8dfc305 | 1116 | spin_unlock(&wb->list_lock); |
03ba3782 | 1117 | |
d46db3d5 | 1118 | return nr_pages - work->nr_pages; |
03ba3782 JA |
1119 | } |
1120 | ||
1121 | /* | |
83ba7b07 | 1122 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 1123 | */ |
f0054bb1 | 1124 | static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) |
03ba3782 | 1125 | { |
83ba7b07 | 1126 | struct wb_writeback_work *work = NULL; |
03ba3782 | 1127 | |
f0054bb1 TH |
1128 | spin_lock_bh(&wb->work_lock); |
1129 | if (!list_empty(&wb->work_list)) { | |
1130 | work = list_entry(wb->work_list.next, | |
83ba7b07 CH |
1131 | struct wb_writeback_work, list); |
1132 | list_del_init(&work->list); | |
03ba3782 | 1133 | } |
f0054bb1 | 1134 | spin_unlock_bh(&wb->work_lock); |
83ba7b07 | 1135 | return work; |
03ba3782 JA |
1136 | } |
1137 | ||
cdf01dd5 LT |
1138 | /* |
1139 | * Add in the number of potentially dirty inodes, because each inode | |
1140 | * write can dirty pagecache in the underlying blockdev. | |
1141 | */ | |
1142 | static unsigned long get_nr_dirty_pages(void) | |
1143 | { | |
1144 | return global_page_state(NR_FILE_DIRTY) + | |
1145 | global_page_state(NR_UNSTABLE_NFS) + | |
1146 | get_nr_dirty_inodes(); | |
1147 | } | |
1148 | ||
6585027a JK |
1149 | static long wb_check_background_flush(struct bdi_writeback *wb) |
1150 | { | |
a88a341a | 1151 | if (over_bground_thresh(wb)) { |
6585027a JK |
1152 | |
1153 | struct wb_writeback_work work = { | |
1154 | .nr_pages = LONG_MAX, | |
1155 | .sync_mode = WB_SYNC_NONE, | |
1156 | .for_background = 1, | |
1157 | .range_cyclic = 1, | |
0e175a18 | 1158 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
1159 | }; |
1160 | ||
1161 | return wb_writeback(wb, &work); | |
1162 | } | |
1163 | ||
1164 | return 0; | |
1165 | } | |
1166 | ||
03ba3782 JA |
1167 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
1168 | { | |
1169 | unsigned long expired; | |
1170 | long nr_pages; | |
1171 | ||
69b62d01 JA |
1172 | /* |
1173 | * When set to zero, disable periodic writeback | |
1174 | */ | |
1175 | if (!dirty_writeback_interval) | |
1176 | return 0; | |
1177 | ||
03ba3782 JA |
1178 | expired = wb->last_old_flush + |
1179 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
1180 | if (time_before(jiffies, expired)) | |
1181 | return 0; | |
1182 | ||
1183 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 1184 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 1185 | |
c4a77a6c | 1186 | if (nr_pages) { |
83ba7b07 | 1187 | struct wb_writeback_work work = { |
c4a77a6c JA |
1188 | .nr_pages = nr_pages, |
1189 | .sync_mode = WB_SYNC_NONE, | |
1190 | .for_kupdate = 1, | |
1191 | .range_cyclic = 1, | |
0e175a18 | 1192 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
1193 | }; |
1194 | ||
83ba7b07 | 1195 | return wb_writeback(wb, &work); |
c4a77a6c | 1196 | } |
03ba3782 JA |
1197 | |
1198 | return 0; | |
1199 | } | |
1200 | ||
1201 | /* | |
1202 | * Retrieve work items and do the writeback they describe | |
1203 | */ | |
25d130ba | 1204 | static long wb_do_writeback(struct bdi_writeback *wb) |
03ba3782 | 1205 | { |
83ba7b07 | 1206 | struct wb_writeback_work *work; |
c4a77a6c | 1207 | long wrote = 0; |
03ba3782 | 1208 | |
4452226e | 1209 | set_bit(WB_writeback_running, &wb->state); |
f0054bb1 | 1210 | while ((work = get_next_work_item(wb)) != NULL) { |
cc395d7f | 1211 | struct wb_completion *done = work->done; |
98754bf7 | 1212 | bool need_wake_up = false; |
03ba3782 | 1213 | |
f0054bb1 | 1214 | trace_writeback_exec(wb->bdi, work); |
455b2864 | 1215 | |
83ba7b07 | 1216 | wrote += wb_writeback(wb, work); |
03ba3782 | 1217 | |
98754bf7 TH |
1218 | if (work->single_wait) { |
1219 | WARN_ON_ONCE(work->auto_free); | |
1220 | /* paired w/ rmb in wb_wait_for_single_work() */ | |
1221 | smp_wmb(); | |
1222 | work->single_done = 1; | |
1223 | need_wake_up = true; | |
1224 | } else if (work->auto_free) { | |
83ba7b07 | 1225 | kfree(work); |
98754bf7 TH |
1226 | } |
1227 | ||
cc395d7f | 1228 | if (done && atomic_dec_and_test(&done->cnt)) |
98754bf7 TH |
1229 | need_wake_up = true; |
1230 | ||
1231 | if (need_wake_up) | |
cc395d7f | 1232 | wake_up_all(&wb->bdi->wb_waitq); |
03ba3782 JA |
1233 | } |
1234 | ||
1235 | /* | |
1236 | * Check for periodic writeback, kupdated() style | |
1237 | */ | |
1238 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 1239 | wrote += wb_check_background_flush(wb); |
4452226e | 1240 | clear_bit(WB_writeback_running, &wb->state); |
03ba3782 JA |
1241 | |
1242 | return wrote; | |
1243 | } | |
1244 | ||
1245 | /* | |
1246 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
839a8e86 | 1247 | * reschedules periodically and does kupdated style flushing. |
03ba3782 | 1248 | */ |
f0054bb1 | 1249 | void wb_workfn(struct work_struct *work) |
03ba3782 | 1250 | { |
839a8e86 TH |
1251 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
1252 | struct bdi_writeback, dwork); | |
03ba3782 JA |
1253 | long pages_written; |
1254 | ||
f0054bb1 | 1255 | set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); |
766f9164 | 1256 | current->flags |= PF_SWAPWRITE; |
455b2864 | 1257 | |
839a8e86 | 1258 | if (likely(!current_is_workqueue_rescuer() || |
4452226e | 1259 | !test_bit(WB_registered, &wb->state))) { |
6467716a | 1260 | /* |
f0054bb1 | 1261 | * The normal path. Keep writing back @wb until its |
839a8e86 | 1262 | * work_list is empty. Note that this path is also taken |
f0054bb1 | 1263 | * if @wb is shutting down even when we're running off the |
839a8e86 | 1264 | * rescuer as work_list needs to be drained. |
6467716a | 1265 | */ |
839a8e86 | 1266 | do { |
25d130ba | 1267 | pages_written = wb_do_writeback(wb); |
839a8e86 | 1268 | trace_writeback_pages_written(pages_written); |
f0054bb1 | 1269 | } while (!list_empty(&wb->work_list)); |
839a8e86 TH |
1270 | } else { |
1271 | /* | |
1272 | * bdi_wq can't get enough workers and we're running off | |
1273 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
1274 | * enough for efficient IO. | |
1275 | */ | |
f0054bb1 | 1276 | pages_written = writeback_inodes_wb(wb, 1024, |
839a8e86 | 1277 | WB_REASON_FORKER_THREAD); |
455b2864 | 1278 | trace_writeback_pages_written(pages_written); |
03ba3782 JA |
1279 | } |
1280 | ||
f0054bb1 | 1281 | if (!list_empty(&wb->work_list)) |
6ca738d6 DB |
1282 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
1283 | else if (wb_has_dirty_io(wb) && dirty_writeback_interval) | |
f0054bb1 | 1284 | wb_wakeup_delayed(wb); |
455b2864 | 1285 | |
839a8e86 | 1286 | current->flags &= ~PF_SWAPWRITE; |
03ba3782 JA |
1287 | } |
1288 | ||
1289 | /* | |
b8c2f347 CH |
1290 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
1291 | * the whole world. | |
03ba3782 | 1292 | */ |
0e175a18 | 1293 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
03ba3782 | 1294 | { |
b8c2f347 | 1295 | struct backing_dev_info *bdi; |
03ba3782 | 1296 | |
47df3dde JK |
1297 | if (!nr_pages) |
1298 | nr_pages = get_nr_dirty_pages(); | |
03ba3782 | 1299 | |
b8c2f347 | 1300 | rcu_read_lock(); |
f2b65121 TH |
1301 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
1302 | struct bdi_writeback *wb; | |
1303 | struct wb_iter iter; | |
1304 | ||
1305 | if (!bdi_has_dirty_io(bdi)) | |
1306 | continue; | |
1307 | ||
1308 | bdi_for_each_wb(wb, bdi, &iter, 0) | |
1309 | wb_start_writeback(wb, wb_split_bdi_pages(wb, nr_pages), | |
1310 | false, reason); | |
1311 | } | |
cfc4ba53 | 1312 | rcu_read_unlock(); |
1da177e4 LT |
1313 | } |
1314 | ||
a2f48706 TT |
1315 | /* |
1316 | * Wake up bdi's periodically to make sure dirtytime inodes gets | |
1317 | * written back periodically. We deliberately do *not* check the | |
1318 | * b_dirtytime list in wb_has_dirty_io(), since this would cause the | |
1319 | * kernel to be constantly waking up once there are any dirtytime | |
1320 | * inodes on the system. So instead we define a separate delayed work | |
1321 | * function which gets called much more rarely. (By default, only | |
1322 | * once every 12 hours.) | |
1323 | * | |
1324 | * If there is any other write activity going on in the file system, | |
1325 | * this function won't be necessary. But if the only thing that has | |
1326 | * happened on the file system is a dirtytime inode caused by an atime | |
1327 | * update, we need this infrastructure below to make sure that inode | |
1328 | * eventually gets pushed out to disk. | |
1329 | */ | |
1330 | static void wakeup_dirtytime_writeback(struct work_struct *w); | |
1331 | static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); | |
1332 | ||
1333 | static void wakeup_dirtytime_writeback(struct work_struct *w) | |
1334 | { | |
1335 | struct backing_dev_info *bdi; | |
1336 | ||
1337 | rcu_read_lock(); | |
1338 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { | |
001fe6f6 TH |
1339 | struct bdi_writeback *wb; |
1340 | struct wb_iter iter; | |
1341 | ||
1342 | bdi_for_each_wb(wb, bdi, &iter, 0) | |
1343 | if (!list_empty(&bdi->wb.b_dirty_time)) | |
1344 | wb_wakeup(&bdi->wb); | |
a2f48706 TT |
1345 | } |
1346 | rcu_read_unlock(); | |
1347 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
1348 | } | |
1349 | ||
1350 | static int __init start_dirtytime_writeback(void) | |
1351 | { | |
1352 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
1353 | return 0; | |
1354 | } | |
1355 | __initcall(start_dirtytime_writeback); | |
1356 | ||
1efff914 TT |
1357 | int dirtytime_interval_handler(struct ctl_table *table, int write, |
1358 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1359 | { | |
1360 | int ret; | |
1361 | ||
1362 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
1363 | if (ret == 0 && write) | |
1364 | mod_delayed_work(system_wq, &dirtytime_work, 0); | |
1365 | return ret; | |
1366 | } | |
1367 | ||
03ba3782 JA |
1368 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1369 | { | |
1370 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1371 | struct dentry *dentry; | |
1372 | const char *name = "?"; | |
1373 | ||
1374 | dentry = d_find_alias(inode); | |
1375 | if (dentry) { | |
1376 | spin_lock(&dentry->d_lock); | |
1377 | name = (const char *) dentry->d_name.name; | |
1378 | } | |
1379 | printk(KERN_DEBUG | |
1380 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1381 | current->comm, task_pid_nr(current), inode->i_ino, | |
1382 | name, inode->i_sb->s_id); | |
1383 | if (dentry) { | |
1384 | spin_unlock(&dentry->d_lock); | |
1385 | dput(dentry); | |
1386 | } | |
1387 | } | |
1388 | } | |
1389 | ||
1390 | /** | |
1391 | * __mark_inode_dirty - internal function | |
1392 | * @inode: inode to mark | |
1393 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1394 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1395 | * mark_inode_dirty_sync. | |
1da177e4 | 1396 | * |
03ba3782 JA |
1397 | * Put the inode on the super block's dirty list. |
1398 | * | |
1399 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1400 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1401 | * If it was not hashed, it will never be added to the dirty list | |
1402 | * even if it is later hashed, as it will have been marked dirty already. | |
1403 | * | |
1404 | * In short, make sure you hash any inodes _before_ you start marking | |
1405 | * them dirty. | |
1da177e4 | 1406 | * |
03ba3782 JA |
1407 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1408 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1409 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1410 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1411 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1412 | * blockdev inode. | |
1da177e4 | 1413 | */ |
0ae45f63 | 1414 | #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) |
03ba3782 | 1415 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1416 | { |
03ba3782 | 1417 | struct super_block *sb = inode->i_sb; |
253c34e9 | 1418 | struct backing_dev_info *bdi = NULL; |
0ae45f63 TT |
1419 | int dirtytime; |
1420 | ||
1421 | trace_writeback_mark_inode_dirty(inode, flags); | |
1da177e4 | 1422 | |
03ba3782 JA |
1423 | /* |
1424 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1425 | * dirty the inode itself | |
1426 | */ | |
0ae45f63 | 1427 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { |
9fb0a7da TH |
1428 | trace_writeback_dirty_inode_start(inode, flags); |
1429 | ||
03ba3782 | 1430 | if (sb->s_op->dirty_inode) |
aa385729 | 1431 | sb->s_op->dirty_inode(inode, flags); |
9fb0a7da TH |
1432 | |
1433 | trace_writeback_dirty_inode(inode, flags); | |
03ba3782 | 1434 | } |
0ae45f63 TT |
1435 | if (flags & I_DIRTY_INODE) |
1436 | flags &= ~I_DIRTY_TIME; | |
1437 | dirtytime = flags & I_DIRTY_TIME; | |
03ba3782 JA |
1438 | |
1439 | /* | |
9c6ac78e TH |
1440 | * Paired with smp_mb() in __writeback_single_inode() for the |
1441 | * following lockless i_state test. See there for details. | |
03ba3782 JA |
1442 | */ |
1443 | smp_mb(); | |
1444 | ||
0ae45f63 TT |
1445 | if (((inode->i_state & flags) == flags) || |
1446 | (dirtytime && (inode->i_state & I_DIRTY_INODE))) | |
03ba3782 JA |
1447 | return; |
1448 | ||
1449 | if (unlikely(block_dump)) | |
1450 | block_dump___mark_inode_dirty(inode); | |
1451 | ||
250df6ed | 1452 | spin_lock(&inode->i_lock); |
0ae45f63 TT |
1453 | if (dirtytime && (inode->i_state & I_DIRTY_INODE)) |
1454 | goto out_unlock_inode; | |
03ba3782 JA |
1455 | if ((inode->i_state & flags) != flags) { |
1456 | const int was_dirty = inode->i_state & I_DIRTY; | |
1457 | ||
52ebea74 TH |
1458 | inode_attach_wb(inode, NULL); |
1459 | ||
0ae45f63 TT |
1460 | if (flags & I_DIRTY_INODE) |
1461 | inode->i_state &= ~I_DIRTY_TIME; | |
03ba3782 JA |
1462 | inode->i_state |= flags; |
1463 | ||
1464 | /* | |
1465 | * If the inode is being synced, just update its dirty state. | |
1466 | * The unlocker will place the inode on the appropriate | |
1467 | * superblock list, based upon its state. | |
1468 | */ | |
1469 | if (inode->i_state & I_SYNC) | |
250df6ed | 1470 | goto out_unlock_inode; |
03ba3782 JA |
1471 | |
1472 | /* | |
1473 | * Only add valid (hashed) inodes to the superblock's | |
1474 | * dirty list. Add blockdev inodes as well. | |
1475 | */ | |
1476 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1477 | if (inode_unhashed(inode)) |
250df6ed | 1478 | goto out_unlock_inode; |
03ba3782 | 1479 | } |
a4ffdde6 | 1480 | if (inode->i_state & I_FREEING) |
250df6ed | 1481 | goto out_unlock_inode; |
03ba3782 JA |
1482 | |
1483 | /* | |
1484 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1485 | * reposition it (that would break b_dirty time-ordering). | |
1486 | */ | |
1487 | if (!was_dirty) { | |
d6c10f1f | 1488 | struct list_head *dirty_list; |
a66979ab | 1489 | bool wakeup_bdi = false; |
253c34e9 AB |
1490 | bdi = inode_to_bdi(inode); |
1491 | ||
146d7009 JB |
1492 | spin_unlock(&inode->i_lock); |
1493 | spin_lock(&bdi->wb.list_lock); | |
253c34e9 | 1494 | |
d6c10f1f TH |
1495 | WARN(bdi_cap_writeback_dirty(bdi) && |
1496 | !test_bit(WB_registered, &bdi->wb.state), | |
1497 | "bdi-%s not registered\n", bdi->name); | |
03ba3782 JA |
1498 | |
1499 | inode->dirtied_when = jiffies; | |
a2f48706 TT |
1500 | if (dirtytime) |
1501 | inode->dirtied_time_when = jiffies; | |
d6c10f1f | 1502 | |
a2f48706 | 1503 | if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) |
d6c10f1f | 1504 | dirty_list = &bdi->wb.b_dirty; |
a2f48706 | 1505 | else |
d6c10f1f TH |
1506 | dirty_list = &bdi->wb.b_dirty_time; |
1507 | ||
1508 | wakeup_bdi = inode_wb_list_move_locked(inode, &bdi->wb, | |
1509 | dirty_list); | |
1510 | ||
f758eeab | 1511 | spin_unlock(&bdi->wb.list_lock); |
0ae45f63 | 1512 | trace_writeback_dirty_inode_enqueue(inode); |
a66979ab | 1513 | |
d6c10f1f TH |
1514 | /* |
1515 | * If this is the first dirty inode for this bdi, | |
1516 | * we have to wake-up the corresponding bdi thread | |
1517 | * to make sure background write-back happens | |
1518 | * later. | |
1519 | */ | |
1520 | if (bdi_cap_writeback_dirty(bdi) && wakeup_bdi) | |
f0054bb1 | 1521 | wb_wakeup_delayed(&bdi->wb); |
a66979ab | 1522 | return; |
1da177e4 | 1523 | } |
1da177e4 | 1524 | } |
250df6ed DC |
1525 | out_unlock_inode: |
1526 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1527 | |
03ba3782 JA |
1528 | } |
1529 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1530 | ||
b6e51316 | 1531 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1532 | { |
1533 | struct inode *inode, *old_inode = NULL; | |
1534 | ||
1535 | /* | |
1536 | * We need to be protected against the filesystem going from | |
1537 | * r/o to r/w or vice versa. | |
1538 | */ | |
b6e51316 | 1539 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1540 | |
55fa6091 | 1541 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1542 | |
1543 | /* | |
1544 | * Data integrity sync. Must wait for all pages under writeback, | |
1545 | * because there may have been pages dirtied before our sync | |
1546 | * call, but which had writeout started before we write it out. | |
1547 | * In which case, the inode may not be on the dirty list, but | |
1548 | * we still have to wait for that writeout. | |
1549 | */ | |
b6e51316 | 1550 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1551 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1552 | |
250df6ed DC |
1553 | spin_lock(&inode->i_lock); |
1554 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1555 | (mapping->nrpages == 0)) { | |
1556 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1557 | continue; |
250df6ed | 1558 | } |
03ba3782 | 1559 | __iget(inode); |
250df6ed | 1560 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1561 | spin_unlock(&inode_sb_list_lock); |
1562 | ||
03ba3782 | 1563 | /* |
55fa6091 DC |
1564 | * We hold a reference to 'inode' so it couldn't have been |
1565 | * removed from s_inodes list while we dropped the | |
1566 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1567 | * be holding the last reference and we cannot iput it under | |
1568 | * inode_sb_list_lock. So we keep the reference and iput it | |
1569 | * later. | |
03ba3782 JA |
1570 | */ |
1571 | iput(old_inode); | |
1572 | old_inode = inode; | |
1573 | ||
1574 | filemap_fdatawait(mapping); | |
1575 | ||
1576 | cond_resched(); | |
1577 | ||
55fa6091 | 1578 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1579 | } |
55fa6091 | 1580 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1581 | iput(old_inode); |
1da177e4 LT |
1582 | } |
1583 | ||
d8a8559c | 1584 | /** |
3259f8be | 1585 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1586 | * @sb: the superblock |
3259f8be | 1587 | * @nr: the number of pages to write |
786228ab | 1588 | * @reason: reason why some writeback work initiated |
1da177e4 | 1589 | * |
d8a8559c JA |
1590 | * Start writeback on some inodes on this super_block. No guarantees are made |
1591 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1592 | * for IO completion of submitted IO. |
1da177e4 | 1593 | */ |
0e175a18 CW |
1594 | void writeback_inodes_sb_nr(struct super_block *sb, |
1595 | unsigned long nr, | |
1596 | enum wb_reason reason) | |
1da177e4 | 1597 | { |
cc395d7f | 1598 | DEFINE_WB_COMPLETION_ONSTACK(done); |
83ba7b07 | 1599 | struct wb_writeback_work work = { |
6e6938b6 WF |
1600 | .sb = sb, |
1601 | .sync_mode = WB_SYNC_NONE, | |
1602 | .tagged_writepages = 1, | |
1603 | .done = &done, | |
1604 | .nr_pages = nr, | |
0e175a18 | 1605 | .reason = reason, |
3c4d7165 | 1606 | }; |
e7972912 | 1607 | struct backing_dev_info *bdi = sb->s_bdi; |
d8a8559c | 1608 | |
e7972912 | 1609 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
6eedc701 | 1610 | return; |
cf37e972 | 1611 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
e7972912 | 1612 | wb_queue_work(&bdi->wb, &work); |
cc395d7f | 1613 | wb_wait_for_completion(bdi, &done); |
e913fc82 | 1614 | } |
3259f8be CM |
1615 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1616 | ||
1617 | /** | |
1618 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1619 | * @sb: the superblock | |
786228ab | 1620 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1621 | * |
1622 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1623 | * on how many (if any) will be written, and this function does not wait | |
1624 | * for IO completion of submitted IO. | |
1625 | */ | |
0e175a18 | 1626 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1627 | { |
0e175a18 | 1628 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1629 | } |
0e3c9a22 | 1630 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1631 | |
17bd55d0 | 1632 | /** |
10ee27a0 | 1633 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway |
17bd55d0 | 1634 | * @sb: the superblock |
10ee27a0 MX |
1635 | * @nr: the number of pages to write |
1636 | * @reason: the reason of writeback | |
17bd55d0 | 1637 | * |
10ee27a0 | 1638 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. |
17bd55d0 ES |
1639 | * Returns 1 if writeback was started, 0 if not. |
1640 | */ | |
10ee27a0 MX |
1641 | int try_to_writeback_inodes_sb_nr(struct super_block *sb, |
1642 | unsigned long nr, | |
1643 | enum wb_reason reason) | |
17bd55d0 | 1644 | { |
bc05873d | 1645 | if (writeback_in_progress(&sb->s_bdi->wb)) |
17bd55d0 | 1646 | return 1; |
10ee27a0 MX |
1647 | |
1648 | if (!down_read_trylock(&sb->s_umount)) | |
17bd55d0 | 1649 | return 0; |
10ee27a0 MX |
1650 | |
1651 | writeback_inodes_sb_nr(sb, nr, reason); | |
1652 | up_read(&sb->s_umount); | |
1653 | return 1; | |
17bd55d0 | 1654 | } |
10ee27a0 | 1655 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); |
17bd55d0 | 1656 | |
3259f8be | 1657 | /** |
10ee27a0 | 1658 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
3259f8be | 1659 | * @sb: the superblock |
786228ab | 1660 | * @reason: reason why some writeback work was initiated |
3259f8be | 1661 | * |
10ee27a0 | 1662 | * Implement by try_to_writeback_inodes_sb_nr() |
3259f8be CM |
1663 | * Returns 1 if writeback was started, 0 if not. |
1664 | */ | |
10ee27a0 | 1665 | int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1666 | { |
10ee27a0 | 1667 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1668 | } |
10ee27a0 | 1669 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
3259f8be | 1670 | |
d8a8559c JA |
1671 | /** |
1672 | * sync_inodes_sb - sync sb inode pages | |
0dc83bd3 | 1673 | * @sb: the superblock |
d8a8559c JA |
1674 | * |
1675 | * This function writes and waits on any dirty inode belonging to this | |
0dc83bd3 | 1676 | * super_block. |
d8a8559c | 1677 | */ |
0dc83bd3 | 1678 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1679 | { |
cc395d7f | 1680 | DEFINE_WB_COMPLETION_ONSTACK(done); |
83ba7b07 | 1681 | struct wb_writeback_work work = { |
3c4d7165 CH |
1682 | .sb = sb, |
1683 | .sync_mode = WB_SYNC_ALL, | |
1684 | .nr_pages = LONG_MAX, | |
1685 | .range_cyclic = 0, | |
83ba7b07 | 1686 | .done = &done, |
0e175a18 | 1687 | .reason = WB_REASON_SYNC, |
7747bd4b | 1688 | .for_sync = 1, |
3c4d7165 | 1689 | }; |
e7972912 | 1690 | struct backing_dev_info *bdi = sb->s_bdi; |
3c4d7165 | 1691 | |
6eedc701 | 1692 | /* Nothing to do? */ |
e7972912 | 1693 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
6eedc701 | 1694 | return; |
cf37e972 CH |
1695 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1696 | ||
e7972912 | 1697 | wb_queue_work(&bdi->wb, &work); |
cc395d7f | 1698 | wb_wait_for_completion(bdi, &done); |
83ba7b07 | 1699 | |
b6e51316 | 1700 | wait_sb_inodes(sb); |
1da177e4 | 1701 | } |
d8a8559c | 1702 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1703 | |
1da177e4 | 1704 | /** |
7f04c26d AA |
1705 | * write_inode_now - write an inode to disk |
1706 | * @inode: inode to write to disk | |
1707 | * @sync: whether the write should be synchronous or not | |
1708 | * | |
1709 | * This function commits an inode to disk immediately if it is dirty. This is | |
1710 | * primarily needed by knfsd. | |
1da177e4 | 1711 | * |
7f04c26d | 1712 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1713 | */ |
1da177e4 LT |
1714 | int write_inode_now(struct inode *inode, int sync) |
1715 | { | |
f758eeab | 1716 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1717 | struct writeback_control wbc = { |
1718 | .nr_to_write = LONG_MAX, | |
18914b18 | 1719 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1720 | .range_start = 0, |
1721 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1722 | }; |
1723 | ||
1724 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1725 | wbc.nr_to_write = 0; |
1da177e4 LT |
1726 | |
1727 | might_sleep(); | |
4f8ad655 | 1728 | return writeback_single_inode(inode, wb, &wbc); |
1da177e4 LT |
1729 | } |
1730 | EXPORT_SYMBOL(write_inode_now); | |
1731 | ||
1732 | /** | |
1733 | * sync_inode - write an inode and its pages to disk. | |
1734 | * @inode: the inode to sync | |
1735 | * @wbc: controls the writeback mode | |
1736 | * | |
1737 | * sync_inode() will write an inode and its pages to disk. It will also | |
1738 | * correctly update the inode on its superblock's dirty inode lists and will | |
1739 | * update inode->i_state. | |
1740 | * | |
1741 | * The caller must have a ref on the inode. | |
1742 | */ | |
1743 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1744 | { | |
4f8ad655 | 1745 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); |
1da177e4 LT |
1746 | } |
1747 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1748 | |
1749 | /** | |
c691b9d9 | 1750 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1751 | * @inode: the inode to sync |
1752 | * @wait: wait for I/O to complete. | |
1753 | * | |
c691b9d9 | 1754 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1755 | * |
1756 | * Note: only writes the actual inode, no associated data or other metadata. | |
1757 | */ | |
1758 | int sync_inode_metadata(struct inode *inode, int wait) | |
1759 | { | |
1760 | struct writeback_control wbc = { | |
1761 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1762 | .nr_to_write = 0, /* metadata-only */ | |
1763 | }; | |
1764 | ||
1765 | return sync_inode(inode, &wbc); | |
1766 | } | |
1767 | EXPORT_SYMBOL(sync_inode_metadata); |