]>
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 | ||
c4a77a6c JA |
37 | /* |
38 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
39 | */ | |
83ba7b07 | 40 | struct wb_writeback_work { |
c4a77a6c JA |
41 | long nr_pages; |
42 | struct super_block *sb; | |
0dc83bd3 | 43 | unsigned long *older_than_this; |
c4a77a6c | 44 | enum writeback_sync_modes sync_mode; |
6e6938b6 | 45 | unsigned int tagged_writepages:1; |
52957fe1 HS |
46 | unsigned int for_kupdate:1; |
47 | unsigned int range_cyclic:1; | |
48 | unsigned int for_background:1; | |
7747bd4b | 49 | unsigned int for_sync:1; /* sync(2) WB_SYNC_ALL writeback */ |
0e175a18 | 50 | enum wb_reason reason; /* why was writeback initiated? */ |
c4a77a6c | 51 | |
8010c3b6 | 52 | struct list_head list; /* pending work list */ |
83ba7b07 | 53 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
54 | }; |
55 | ||
a2f48706 TT |
56 | /* |
57 | * If an inode is constantly having its pages dirtied, but then the | |
58 | * updates stop dirtytime_expire_interval seconds in the past, it's | |
59 | * possible for the worst case time between when an inode has its | |
60 | * timestamps updated and when they finally get written out to be two | |
61 | * dirtytime_expire_intervals. We set the default to 12 hours (in | |
62 | * seconds), which means most of the time inodes will have their | |
63 | * timestamps written to disk after 12 hours, but in the worst case a | |
64 | * few inodes might not their timestamps updated for 24 hours. | |
65 | */ | |
66 | unsigned int dirtytime_expire_interval = 12 * 60 * 60; | |
67 | ||
7ccf19a8 NP |
68 | static inline struct inode *wb_inode(struct list_head *head) |
69 | { | |
70 | return list_entry(head, struct inode, i_wb_list); | |
71 | } | |
72 | ||
15eb77a0 WF |
73 | /* |
74 | * Include the creation of the trace points after defining the | |
75 | * wb_writeback_work structure and inline functions so that the definition | |
76 | * remains local to this file. | |
77 | */ | |
78 | #define CREATE_TRACE_POINTS | |
79 | #include <trace/events/writeback.h> | |
80 | ||
774016b2 SW |
81 | EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); |
82 | ||
d6c10f1f TH |
83 | static bool wb_io_lists_populated(struct bdi_writeback *wb) |
84 | { | |
85 | if (wb_has_dirty_io(wb)) { | |
86 | return false; | |
87 | } else { | |
88 | set_bit(WB_has_dirty_io, &wb->state); | |
95a46c65 | 89 | WARN_ON_ONCE(!wb->avg_write_bandwidth); |
766a9d6e TH |
90 | atomic_long_add(wb->avg_write_bandwidth, |
91 | &wb->bdi->tot_write_bandwidth); | |
d6c10f1f TH |
92 | return true; |
93 | } | |
94 | } | |
95 | ||
96 | static void wb_io_lists_depopulated(struct bdi_writeback *wb) | |
97 | { | |
98 | if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && | |
766a9d6e | 99 | list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { |
d6c10f1f | 100 | clear_bit(WB_has_dirty_io, &wb->state); |
95a46c65 TH |
101 | WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth, |
102 | &wb->bdi->tot_write_bandwidth) < 0); | |
766a9d6e | 103 | } |
d6c10f1f TH |
104 | } |
105 | ||
106 | /** | |
107 | * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list | |
108 | * @inode: inode to be moved | |
109 | * @wb: target bdi_writeback | |
110 | * @head: one of @wb->b_{dirty|io|more_io} | |
111 | * | |
112 | * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io. | |
113 | * Returns %true if @inode is the first occupant of the !dirty_time IO | |
114 | * lists; otherwise, %false. | |
115 | */ | |
116 | static bool inode_wb_list_move_locked(struct inode *inode, | |
117 | struct bdi_writeback *wb, | |
118 | struct list_head *head) | |
119 | { | |
120 | assert_spin_locked(&wb->list_lock); | |
121 | ||
122 | list_move(&inode->i_wb_list, head); | |
123 | ||
124 | /* dirty_time doesn't count as dirty_io until expiration */ | |
125 | if (head != &wb->b_dirty_time) | |
126 | return wb_io_lists_populated(wb); | |
127 | ||
128 | wb_io_lists_depopulated(wb); | |
129 | return false; | |
130 | } | |
131 | ||
132 | /** | |
133 | * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list | |
134 | * @inode: inode to be removed | |
135 | * @wb: bdi_writeback @inode is being removed from | |
136 | * | |
137 | * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and | |
138 | * clear %WB_has_dirty_io if all are empty afterwards. | |
139 | */ | |
140 | static void inode_wb_list_del_locked(struct inode *inode, | |
141 | struct bdi_writeback *wb) | |
142 | { | |
143 | assert_spin_locked(&wb->list_lock); | |
144 | ||
145 | list_del_init(&inode->i_wb_list); | |
146 | wb_io_lists_depopulated(wb); | |
147 | } | |
148 | ||
f0054bb1 | 149 | static void wb_wakeup(struct bdi_writeback *wb) |
5acda9d1 | 150 | { |
f0054bb1 TH |
151 | spin_lock_bh(&wb->work_lock); |
152 | if (test_bit(WB_registered, &wb->state)) | |
153 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
154 | spin_unlock_bh(&wb->work_lock); | |
5acda9d1 JK |
155 | } |
156 | ||
f0054bb1 TH |
157 | static void wb_queue_work(struct bdi_writeback *wb, |
158 | struct wb_writeback_work *work) | |
6585027a | 159 | { |
f0054bb1 | 160 | trace_writeback_queue(wb->bdi, work); |
6585027a | 161 | |
f0054bb1 TH |
162 | spin_lock_bh(&wb->work_lock); |
163 | if (!test_bit(WB_registered, &wb->state)) { | |
5acda9d1 JK |
164 | if (work->done) |
165 | complete(work->done); | |
166 | goto out_unlock; | |
167 | } | |
f0054bb1 TH |
168 | list_add_tail(&work->list, &wb->work_list); |
169 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
5acda9d1 | 170 | out_unlock: |
f0054bb1 | 171 | spin_unlock_bh(&wb->work_lock); |
1da177e4 LT |
172 | } |
173 | ||
703c2708 TH |
174 | #ifdef CONFIG_CGROUP_WRITEBACK |
175 | ||
176 | /** | |
177 | * inode_congested - test whether an inode is congested | |
178 | * @inode: inode to test for congestion | |
179 | * @cong_bits: mask of WB_[a]sync_congested bits to test | |
180 | * | |
181 | * Tests whether @inode is congested. @cong_bits is the mask of congestion | |
182 | * bits to test and the return value is the mask of set bits. | |
183 | * | |
184 | * If cgroup writeback is enabled for @inode, the congestion state is | |
185 | * determined by whether the cgwb (cgroup bdi_writeback) for the blkcg | |
186 | * associated with @inode is congested; otherwise, the root wb's congestion | |
187 | * state is used. | |
188 | */ | |
189 | int inode_congested(struct inode *inode, int cong_bits) | |
190 | { | |
191 | if (inode) { | |
192 | struct bdi_writeback *wb = inode_to_wb(inode); | |
193 | if (wb) | |
194 | return wb_congested(wb, cong_bits); | |
195 | } | |
196 | ||
197 | return wb_congested(&inode_to_bdi(inode)->wb, cong_bits); | |
198 | } | |
199 | EXPORT_SYMBOL_GPL(inode_congested); | |
200 | ||
f2b65121 TH |
201 | /** |
202 | * wb_split_bdi_pages - split nr_pages to write according to bandwidth | |
203 | * @wb: target bdi_writeback to split @nr_pages to | |
204 | * @nr_pages: number of pages to write for the whole bdi | |
205 | * | |
206 | * Split @wb's portion of @nr_pages according to @wb's write bandwidth in | |
207 | * relation to the total write bandwidth of all wb's w/ dirty inodes on | |
208 | * @wb->bdi. | |
209 | */ | |
210 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) | |
211 | { | |
212 | unsigned long this_bw = wb->avg_write_bandwidth; | |
213 | unsigned long tot_bw = atomic_long_read(&wb->bdi->tot_write_bandwidth); | |
214 | ||
215 | if (nr_pages == LONG_MAX) | |
216 | return LONG_MAX; | |
217 | ||
218 | /* | |
219 | * This may be called on clean wb's and proportional distribution | |
220 | * may not make sense, just use the original @nr_pages in those | |
221 | * cases. In general, we wanna err on the side of writing more. | |
222 | */ | |
223 | if (!tot_bw || this_bw >= tot_bw) | |
224 | return nr_pages; | |
225 | else | |
226 | return DIV_ROUND_UP_ULL((u64)nr_pages * this_bw, tot_bw); | |
227 | } | |
228 | ||
229 | #else /* CONFIG_CGROUP_WRITEBACK */ | |
230 | ||
231 | static long wb_split_bdi_pages(struct bdi_writeback *wb, long nr_pages) | |
232 | { | |
233 | return nr_pages; | |
234 | } | |
235 | ||
703c2708 TH |
236 | #endif /* CONFIG_CGROUP_WRITEBACK */ |
237 | ||
c00ddad3 TH |
238 | void wb_start_writeback(struct bdi_writeback *wb, long nr_pages, |
239 | bool range_cyclic, enum wb_reason reason) | |
b6e51316 | 240 | { |
c00ddad3 TH |
241 | struct wb_writeback_work *work; |
242 | ||
243 | if (!wb_has_dirty_io(wb)) | |
244 | return; | |
245 | ||
246 | /* | |
247 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
248 | * wakeup the thread for old dirty data writeback | |
249 | */ | |
250 | work = kzalloc(sizeof(*work), GFP_ATOMIC); | |
251 | if (!work) { | |
252 | trace_writeback_nowork(wb->bdi); | |
253 | wb_wakeup(wb); | |
254 | return; | |
255 | } | |
256 | ||
257 | work->sync_mode = WB_SYNC_NONE; | |
258 | work->nr_pages = nr_pages; | |
259 | work->range_cyclic = range_cyclic; | |
260 | work->reason = reason; | |
261 | ||
262 | wb_queue_work(wb, work); | |
c5444198 | 263 | } |
d3ddec76 | 264 | |
c5444198 | 265 | /** |
9ecf4866 TH |
266 | * wb_start_background_writeback - start background writeback |
267 | * @wb: bdi_writback to write from | |
c5444198 CH |
268 | * |
269 | * Description: | |
6585027a | 270 | * This makes sure WB_SYNC_NONE background writeback happens. When |
9ecf4866 | 271 | * this function returns, it is only guaranteed that for given wb |
6585027a JK |
272 | * some IO is happening if we are over background dirty threshold. |
273 | * Caller need not hold sb s_umount semaphore. | |
c5444198 | 274 | */ |
9ecf4866 | 275 | void wb_start_background_writeback(struct bdi_writeback *wb) |
c5444198 | 276 | { |
6585027a JK |
277 | /* |
278 | * We just wake up the flusher thread. It will perform background | |
279 | * writeback as soon as there is no other work to do. | |
280 | */ | |
9ecf4866 TH |
281 | trace_writeback_wake_background(wb->bdi); |
282 | wb_wakeup(wb); | |
1da177e4 LT |
283 | } |
284 | ||
a66979ab DC |
285 | /* |
286 | * Remove the inode from the writeback list it is on. | |
287 | */ | |
288 | void inode_wb_list_del(struct inode *inode) | |
289 | { | |
52ebea74 | 290 | struct bdi_writeback *wb = inode_to_wb(inode); |
f758eeab | 291 | |
52ebea74 | 292 | spin_lock(&wb->list_lock); |
d6c10f1f | 293 | inode_wb_list_del_locked(inode, wb); |
52ebea74 | 294 | spin_unlock(&wb->list_lock); |
a66979ab DC |
295 | } |
296 | ||
6610a0bc AM |
297 | /* |
298 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
299 | * furthest end of its superblock's dirty-inode list. | |
300 | * | |
301 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 302 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
303 | * the case then the inode must have been redirtied while it was being written |
304 | * out and we don't reset its dirtied_when. | |
305 | */ | |
f758eeab | 306 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 307 | { |
03ba3782 | 308 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 309 | struct inode *tail; |
6610a0bc | 310 | |
7ccf19a8 | 311 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 312 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
313 | inode->dirtied_when = jiffies; |
314 | } | |
d6c10f1f | 315 | inode_wb_list_move_locked(inode, wb, &wb->b_dirty); |
6610a0bc AM |
316 | } |
317 | ||
c986d1e2 | 318 | /* |
66f3b8e2 | 319 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 320 | */ |
f758eeab | 321 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 322 | { |
d6c10f1f | 323 | inode_wb_list_move_locked(inode, wb, &wb->b_more_io); |
c986d1e2 AM |
324 | } |
325 | ||
1c0eeaf5 JE |
326 | static void inode_sync_complete(struct inode *inode) |
327 | { | |
365b94ae | 328 | inode->i_state &= ~I_SYNC; |
4eff96dd JK |
329 | /* If inode is clean an unused, put it into LRU now... */ |
330 | inode_add_lru(inode); | |
365b94ae | 331 | /* Waiters must see I_SYNC cleared before being woken up */ |
1c0eeaf5 JE |
332 | smp_mb(); |
333 | wake_up_bit(&inode->i_state, __I_SYNC); | |
334 | } | |
335 | ||
d2caa3c5 JL |
336 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
337 | { | |
338 | bool ret = time_after(inode->dirtied_when, t); | |
339 | #ifndef CONFIG_64BIT | |
340 | /* | |
341 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
342 | * It _appears_ to be in the future, but is actually in distant past. | |
343 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 344 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
345 | */ |
346 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
347 | #endif | |
348 | return ret; | |
349 | } | |
350 | ||
0ae45f63 TT |
351 | #define EXPIRE_DIRTY_ATIME 0x0001 |
352 | ||
2c136579 | 353 | /* |
0e2f2b23 | 354 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
697e6fed | 355 | * @delaying_queue to @dispatch_queue. |
2c136579 | 356 | */ |
e84d0a4f | 357 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 358 | struct list_head *dispatch_queue, |
0ae45f63 | 359 | int flags, |
ad4e38dd | 360 | struct wb_writeback_work *work) |
2c136579 | 361 | { |
0ae45f63 TT |
362 | unsigned long *older_than_this = NULL; |
363 | unsigned long expire_time; | |
5c03449d SL |
364 | LIST_HEAD(tmp); |
365 | struct list_head *pos, *node; | |
cf137307 | 366 | struct super_block *sb = NULL; |
5c03449d | 367 | struct inode *inode; |
cf137307 | 368 | int do_sb_sort = 0; |
e84d0a4f | 369 | int moved = 0; |
5c03449d | 370 | |
0ae45f63 TT |
371 | if ((flags & EXPIRE_DIRTY_ATIME) == 0) |
372 | older_than_this = work->older_than_this; | |
a2f48706 TT |
373 | else if (!work->for_sync) { |
374 | expire_time = jiffies - (dirtytime_expire_interval * HZ); | |
0ae45f63 TT |
375 | older_than_this = &expire_time; |
376 | } | |
2c136579 | 377 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 378 | inode = wb_inode(delaying_queue->prev); |
0ae45f63 TT |
379 | if (older_than_this && |
380 | inode_dirtied_after(inode, *older_than_this)) | |
2c136579 | 381 | break; |
a8855990 JK |
382 | list_move(&inode->i_wb_list, &tmp); |
383 | moved++; | |
0ae45f63 TT |
384 | if (flags & EXPIRE_DIRTY_ATIME) |
385 | set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); | |
a8855990 JK |
386 | if (sb_is_blkdev_sb(inode->i_sb)) |
387 | continue; | |
cf137307 JA |
388 | if (sb && sb != inode->i_sb) |
389 | do_sb_sort = 1; | |
390 | sb = inode->i_sb; | |
5c03449d SL |
391 | } |
392 | ||
cf137307 JA |
393 | /* just one sb in list, splice to dispatch_queue and we're done */ |
394 | if (!do_sb_sort) { | |
395 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 396 | goto out; |
cf137307 JA |
397 | } |
398 | ||
5c03449d SL |
399 | /* Move inodes from one superblock together */ |
400 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 401 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 402 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 403 | inode = wb_inode(pos); |
5c03449d | 404 | if (inode->i_sb == sb) |
7ccf19a8 | 405 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 406 | } |
2c136579 | 407 | } |
e84d0a4f WF |
408 | out: |
409 | return moved; | |
2c136579 FW |
410 | } |
411 | ||
412 | /* | |
413 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
414 | * Before |
415 | * newly dirtied b_dirty b_io b_more_io | |
416 | * =============> gf edc BA | |
417 | * After | |
418 | * newly dirtied b_dirty b_io b_more_io | |
419 | * =============> g fBAedc | |
420 | * | | |
421 | * +--> dequeue for IO | |
2c136579 | 422 | */ |
ad4e38dd | 423 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
66f3b8e2 | 424 | { |
e84d0a4f | 425 | int moved; |
0ae45f63 | 426 | |
f758eeab | 427 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 428 | list_splice_init(&wb->b_more_io, &wb->b_io); |
0ae45f63 TT |
429 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); |
430 | moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, | |
431 | EXPIRE_DIRTY_ATIME, work); | |
d6c10f1f TH |
432 | if (moved) |
433 | wb_io_lists_populated(wb); | |
ad4e38dd | 434 | trace_writeback_queue_io(wb, work, moved); |
66f3b8e2 JA |
435 | } |
436 | ||
a9185b41 | 437 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 438 | { |
9fb0a7da TH |
439 | int ret; |
440 | ||
441 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
442 | trace_writeback_write_inode_start(inode, wbc); | |
443 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
444 | trace_writeback_write_inode(inode, wbc); | |
445 | return ret; | |
446 | } | |
03ba3782 | 447 | return 0; |
08d8e974 | 448 | } |
08d8e974 | 449 | |
1da177e4 | 450 | /* |
169ebd90 JK |
451 | * Wait for writeback on an inode to complete. Called with i_lock held. |
452 | * Caller must make sure inode cannot go away when we drop i_lock. | |
01c03194 | 453 | */ |
169ebd90 JK |
454 | static void __inode_wait_for_writeback(struct inode *inode) |
455 | __releases(inode->i_lock) | |
456 | __acquires(inode->i_lock) | |
01c03194 CH |
457 | { |
458 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
459 | wait_queue_head_t *wqh; | |
460 | ||
461 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
462 | while (inode->i_state & I_SYNC) { |
463 | spin_unlock(&inode->i_lock); | |
74316201 N |
464 | __wait_on_bit(wqh, &wq, bit_wait, |
465 | TASK_UNINTERRUPTIBLE); | |
250df6ed | 466 | spin_lock(&inode->i_lock); |
58a9d3d8 | 467 | } |
01c03194 CH |
468 | } |
469 | ||
169ebd90 JK |
470 | /* |
471 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
472 | */ | |
473 | void inode_wait_for_writeback(struct inode *inode) | |
474 | { | |
475 | spin_lock(&inode->i_lock); | |
476 | __inode_wait_for_writeback(inode); | |
477 | spin_unlock(&inode->i_lock); | |
478 | } | |
479 | ||
480 | /* | |
481 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
482 | * held and drops it. It is aimed for callers not holding any inode reference | |
483 | * so once i_lock is dropped, inode can go away. | |
484 | */ | |
485 | static void inode_sleep_on_writeback(struct inode *inode) | |
486 | __releases(inode->i_lock) | |
487 | { | |
488 | DEFINE_WAIT(wait); | |
489 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
490 | int sleep; | |
491 | ||
492 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
493 | sleep = inode->i_state & I_SYNC; | |
494 | spin_unlock(&inode->i_lock); | |
495 | if (sleep) | |
496 | schedule(); | |
497 | finish_wait(wqh, &wait); | |
498 | } | |
499 | ||
ccb26b5a JK |
500 | /* |
501 | * Find proper writeback list for the inode depending on its current state and | |
502 | * possibly also change of its state while we were doing writeback. Here we | |
503 | * handle things such as livelock prevention or fairness of writeback among | |
504 | * inodes. This function can be called only by flusher thread - noone else | |
505 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
506 | * thread's back can have unexpected consequences. | |
507 | */ | |
508 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
509 | struct writeback_control *wbc) | |
510 | { | |
511 | if (inode->i_state & I_FREEING) | |
512 | return; | |
513 | ||
514 | /* | |
515 | * Sync livelock prevention. Each inode is tagged and synced in one | |
516 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
517 | * the dirty time to prevent enqueue and sync it again. | |
518 | */ | |
519 | if ((inode->i_state & I_DIRTY) && | |
520 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
521 | inode->dirtied_when = jiffies; | |
522 | ||
4f8ad655 JK |
523 | if (wbc->pages_skipped) { |
524 | /* | |
525 | * writeback is not making progress due to locked | |
526 | * buffers. Skip this inode for now. | |
527 | */ | |
528 | redirty_tail(inode, wb); | |
529 | return; | |
530 | } | |
531 | ||
ccb26b5a JK |
532 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
533 | /* | |
534 | * We didn't write back all the pages. nfs_writepages() | |
535 | * sometimes bales out without doing anything. | |
536 | */ | |
537 | if (wbc->nr_to_write <= 0) { | |
538 | /* Slice used up. Queue for next turn. */ | |
539 | requeue_io(inode, wb); | |
540 | } else { | |
541 | /* | |
542 | * Writeback blocked by something other than | |
543 | * congestion. Delay the inode for some time to | |
544 | * avoid spinning on the CPU (100% iowait) | |
545 | * retrying writeback of the dirty page/inode | |
546 | * that cannot be performed immediately. | |
547 | */ | |
548 | redirty_tail(inode, wb); | |
549 | } | |
550 | } else if (inode->i_state & I_DIRTY) { | |
551 | /* | |
552 | * Filesystems can dirty the inode during writeback operations, | |
553 | * such as delayed allocation during submission or metadata | |
554 | * updates after data IO completion. | |
555 | */ | |
556 | redirty_tail(inode, wb); | |
0ae45f63 | 557 | } else if (inode->i_state & I_DIRTY_TIME) { |
a2f48706 | 558 | inode->dirtied_when = jiffies; |
d6c10f1f | 559 | inode_wb_list_move_locked(inode, wb, &wb->b_dirty_time); |
ccb26b5a JK |
560 | } else { |
561 | /* The inode is clean. Remove from writeback lists. */ | |
d6c10f1f | 562 | inode_wb_list_del_locked(inode, wb); |
ccb26b5a JK |
563 | } |
564 | } | |
565 | ||
01c03194 | 566 | /* |
4f8ad655 JK |
567 | * Write out an inode and its dirty pages. Do not update the writeback list |
568 | * linkage. That is left to the caller. The caller is also responsible for | |
569 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. | |
1da177e4 LT |
570 | */ |
571 | static int | |
cd8ed2a4 | 572 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 573 | { |
1da177e4 | 574 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 575 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 576 | unsigned dirty; |
1da177e4 LT |
577 | int ret; |
578 | ||
4f8ad655 | 579 | WARN_ON(!(inode->i_state & I_SYNC)); |
1da177e4 | 580 | |
9fb0a7da TH |
581 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
582 | ||
1da177e4 LT |
583 | ret = do_writepages(mapping, wbc); |
584 | ||
26821ed4 CH |
585 | /* |
586 | * Make sure to wait on the data before writing out the metadata. | |
587 | * This is important for filesystems that modify metadata on data | |
7747bd4b DC |
588 | * I/O completion. We don't do it for sync(2) writeback because it has a |
589 | * separate, external IO completion path and ->sync_fs for guaranteeing | |
590 | * inode metadata is written back correctly. | |
26821ed4 | 591 | */ |
7747bd4b | 592 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
26821ed4 | 593 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
594 | if (ret == 0) |
595 | ret = err; | |
596 | } | |
597 | ||
5547e8aa DM |
598 | /* |
599 | * Some filesystems may redirty the inode during the writeback | |
600 | * due to delalloc, clear dirty metadata flags right before | |
601 | * write_inode() | |
602 | */ | |
250df6ed | 603 | spin_lock(&inode->i_lock); |
9c6ac78e | 604 | |
5547e8aa | 605 | dirty = inode->i_state & I_DIRTY; |
a2f48706 TT |
606 | if (inode->i_state & I_DIRTY_TIME) { |
607 | if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
608 | unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || | |
609 | unlikely(time_after(jiffies, | |
610 | (inode->dirtied_time_when + | |
611 | dirtytime_expire_interval * HZ)))) { | |
612 | dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; | |
613 | trace_writeback_lazytime(inode); | |
614 | } | |
615 | } else | |
616 | inode->i_state &= ~I_DIRTY_TIME_EXPIRED; | |
0ae45f63 | 617 | inode->i_state &= ~dirty; |
9c6ac78e TH |
618 | |
619 | /* | |
620 | * Paired with smp_mb() in __mark_inode_dirty(). This allows | |
621 | * __mark_inode_dirty() to test i_state without grabbing i_lock - | |
622 | * either they see the I_DIRTY bits cleared or we see the dirtied | |
623 | * inode. | |
624 | * | |
625 | * I_DIRTY_PAGES is always cleared together above even if @mapping | |
626 | * still has dirty pages. The flag is reinstated after smp_mb() if | |
627 | * necessary. This guarantees that either __mark_inode_dirty() | |
628 | * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. | |
629 | */ | |
630 | smp_mb(); | |
631 | ||
632 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
633 | inode->i_state |= I_DIRTY_PAGES; | |
634 | ||
250df6ed | 635 | spin_unlock(&inode->i_lock); |
9c6ac78e | 636 | |
0ae45f63 TT |
637 | if (dirty & I_DIRTY_TIME) |
638 | mark_inode_dirty_sync(inode); | |
26821ed4 | 639 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
0ae45f63 | 640 | if (dirty & ~I_DIRTY_PAGES) { |
a9185b41 | 641 | int err = write_inode(inode, wbc); |
1da177e4 LT |
642 | if (ret == 0) |
643 | ret = err; | |
644 | } | |
4f8ad655 JK |
645 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
646 | return ret; | |
647 | } | |
648 | ||
649 | /* | |
650 | * Write out an inode's dirty pages. Either the caller has an active reference | |
651 | * on the inode or the inode has I_WILL_FREE set. | |
652 | * | |
653 | * This function is designed to be called for writing back one inode which | |
654 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() | |
655 | * and does more profound writeback list handling in writeback_sb_inodes(). | |
656 | */ | |
657 | static int | |
658 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, | |
659 | struct writeback_control *wbc) | |
660 | { | |
661 | int ret = 0; | |
662 | ||
663 | spin_lock(&inode->i_lock); | |
664 | if (!atomic_read(&inode->i_count)) | |
665 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
666 | else | |
667 | WARN_ON(inode->i_state & I_WILL_FREE); | |
668 | ||
669 | if (inode->i_state & I_SYNC) { | |
670 | if (wbc->sync_mode != WB_SYNC_ALL) | |
671 | goto out; | |
672 | /* | |
169ebd90 JK |
673 | * It's a data-integrity sync. We must wait. Since callers hold |
674 | * inode reference or inode has I_WILL_FREE set, it cannot go | |
675 | * away under us. | |
4f8ad655 | 676 | */ |
169ebd90 | 677 | __inode_wait_for_writeback(inode); |
4f8ad655 JK |
678 | } |
679 | WARN_ON(inode->i_state & I_SYNC); | |
680 | /* | |
f9b0e058 JK |
681 | * Skip inode if it is clean and we have no outstanding writeback in |
682 | * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this | |
683 | * function since flusher thread may be doing for example sync in | |
684 | * parallel and if we move the inode, it could get skipped. So here we | |
685 | * make sure inode is on some writeback list and leave it there unless | |
686 | * we have completely cleaned the inode. | |
4f8ad655 | 687 | */ |
0ae45f63 | 688 | if (!(inode->i_state & I_DIRTY_ALL) && |
f9b0e058 JK |
689 | (wbc->sync_mode != WB_SYNC_ALL || |
690 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) | |
4f8ad655 JK |
691 | goto out; |
692 | inode->i_state |= I_SYNC; | |
693 | spin_unlock(&inode->i_lock); | |
694 | ||
cd8ed2a4 | 695 | ret = __writeback_single_inode(inode, wbc); |
1da177e4 | 696 | |
f758eeab | 697 | spin_lock(&wb->list_lock); |
250df6ed | 698 | spin_lock(&inode->i_lock); |
4f8ad655 JK |
699 | /* |
700 | * If inode is clean, remove it from writeback lists. Otherwise don't | |
701 | * touch it. See comment above for explanation. | |
702 | */ | |
0ae45f63 | 703 | if (!(inode->i_state & I_DIRTY_ALL)) |
d6c10f1f | 704 | inode_wb_list_del_locked(inode, wb); |
4f8ad655 | 705 | spin_unlock(&wb->list_lock); |
1c0eeaf5 | 706 | inode_sync_complete(inode); |
4f8ad655 JK |
707 | out: |
708 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
709 | return ret; |
710 | } | |
711 | ||
a88a341a | 712 | static long writeback_chunk_size(struct bdi_writeback *wb, |
1a12d8bd | 713 | struct wb_writeback_work *work) |
d46db3d5 WF |
714 | { |
715 | long pages; | |
716 | ||
717 | /* | |
718 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
719 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
720 | * here avoids calling into writeback_inodes_wb() more than once. | |
721 | * | |
722 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
723 | * | |
724 | * wb_writeback() | |
725 | * writeback_sb_inodes() <== called only once | |
726 | * write_cache_pages() <== called once for each inode | |
727 | * (quickly) tag currently dirty pages | |
728 | * (maybe slowly) sync all tagged pages | |
729 | */ | |
730 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
731 | pages = LONG_MAX; | |
1a12d8bd | 732 | else { |
a88a341a | 733 | pages = min(wb->avg_write_bandwidth / 2, |
1a12d8bd WF |
734 | global_dirty_limit / DIRTY_SCOPE); |
735 | pages = min(pages, work->nr_pages); | |
736 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
737 | MIN_WRITEBACK_PAGES); | |
738 | } | |
d46db3d5 WF |
739 | |
740 | return pages; | |
741 | } | |
742 | ||
f11c9c5c ES |
743 | /* |
744 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 | 745 | * |
d46db3d5 | 746 | * Return the number of pages and/or inodes written. |
f11c9c5c | 747 | */ |
d46db3d5 WF |
748 | static long writeback_sb_inodes(struct super_block *sb, |
749 | struct bdi_writeback *wb, | |
750 | struct wb_writeback_work *work) | |
1da177e4 | 751 | { |
d46db3d5 WF |
752 | struct writeback_control wbc = { |
753 | .sync_mode = work->sync_mode, | |
754 | .tagged_writepages = work->tagged_writepages, | |
755 | .for_kupdate = work->for_kupdate, | |
756 | .for_background = work->for_background, | |
7747bd4b | 757 | .for_sync = work->for_sync, |
d46db3d5 WF |
758 | .range_cyclic = work->range_cyclic, |
759 | .range_start = 0, | |
760 | .range_end = LLONG_MAX, | |
761 | }; | |
762 | unsigned long start_time = jiffies; | |
763 | long write_chunk; | |
764 | long wrote = 0; /* count both pages and inodes */ | |
765 | ||
03ba3782 | 766 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 767 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
768 | |
769 | if (inode->i_sb != sb) { | |
d46db3d5 | 770 | if (work->sb) { |
edadfb10 CH |
771 | /* |
772 | * We only want to write back data for this | |
773 | * superblock, move all inodes not belonging | |
774 | * to it back onto the dirty list. | |
775 | */ | |
f758eeab | 776 | redirty_tail(inode, wb); |
edadfb10 CH |
777 | continue; |
778 | } | |
779 | ||
780 | /* | |
781 | * The inode belongs to a different superblock. | |
782 | * Bounce back to the caller to unpin this and | |
783 | * pin the next superblock. | |
784 | */ | |
d46db3d5 | 785 | break; |
edadfb10 CH |
786 | } |
787 | ||
9843b76a | 788 | /* |
331cbdee WL |
789 | * Don't bother with new inodes or inodes being freed, first |
790 | * kind does not need periodic writeout yet, and for the latter | |
9843b76a CH |
791 | * kind writeout is handled by the freer. |
792 | */ | |
250df6ed | 793 | spin_lock(&inode->i_lock); |
9843b76a | 794 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 795 | spin_unlock(&inode->i_lock); |
fcc5c222 | 796 | redirty_tail(inode, wb); |
7ef0d737 NP |
797 | continue; |
798 | } | |
cc1676d9 JK |
799 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
800 | /* | |
801 | * If this inode is locked for writeback and we are not | |
802 | * doing writeback-for-data-integrity, move it to | |
803 | * b_more_io so that writeback can proceed with the | |
804 | * other inodes on s_io. | |
805 | * | |
806 | * We'll have another go at writing back this inode | |
807 | * when we completed a full scan of b_io. | |
808 | */ | |
809 | spin_unlock(&inode->i_lock); | |
810 | requeue_io(inode, wb); | |
811 | trace_writeback_sb_inodes_requeue(inode); | |
812 | continue; | |
813 | } | |
f0d07b7f JK |
814 | spin_unlock(&wb->list_lock); |
815 | ||
4f8ad655 JK |
816 | /* |
817 | * We already requeued the inode if it had I_SYNC set and we | |
818 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
819 | * WB_SYNC_ALL case. | |
820 | */ | |
169ebd90 JK |
821 | if (inode->i_state & I_SYNC) { |
822 | /* Wait for I_SYNC. This function drops i_lock... */ | |
823 | inode_sleep_on_writeback(inode); | |
824 | /* Inode may be gone, start again */ | |
ead188f9 | 825 | spin_lock(&wb->list_lock); |
169ebd90 JK |
826 | continue; |
827 | } | |
4f8ad655 JK |
828 | inode->i_state |= I_SYNC; |
829 | spin_unlock(&inode->i_lock); | |
169ebd90 | 830 | |
a88a341a | 831 | write_chunk = writeback_chunk_size(wb, work); |
d46db3d5 WF |
832 | wbc.nr_to_write = write_chunk; |
833 | wbc.pages_skipped = 0; | |
250df6ed | 834 | |
169ebd90 JK |
835 | /* |
836 | * We use I_SYNC to pin the inode in memory. While it is set | |
837 | * evict_inode() will wait so the inode cannot be freed. | |
838 | */ | |
cd8ed2a4 | 839 | __writeback_single_inode(inode, &wbc); |
250df6ed | 840 | |
d46db3d5 WF |
841 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
842 | wrote += write_chunk - wbc.nr_to_write; | |
4f8ad655 JK |
843 | spin_lock(&wb->list_lock); |
844 | spin_lock(&inode->i_lock); | |
0ae45f63 | 845 | if (!(inode->i_state & I_DIRTY_ALL)) |
d46db3d5 | 846 | wrote++; |
4f8ad655 JK |
847 | requeue_inode(inode, wb, &wbc); |
848 | inode_sync_complete(inode); | |
0f1b1fd8 | 849 | spin_unlock(&inode->i_lock); |
169ebd90 | 850 | cond_resched_lock(&wb->list_lock); |
d46db3d5 WF |
851 | /* |
852 | * bail out to wb_writeback() often enough to check | |
853 | * background threshold and other termination conditions. | |
854 | */ | |
855 | if (wrote) { | |
856 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
857 | break; | |
858 | if (work->nr_pages <= 0) | |
859 | break; | |
8bc3be27 | 860 | } |
1da177e4 | 861 | } |
d46db3d5 | 862 | return wrote; |
f11c9c5c ES |
863 | } |
864 | ||
d46db3d5 WF |
865 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
866 | struct wb_writeback_work *work) | |
f11c9c5c | 867 | { |
d46db3d5 WF |
868 | unsigned long start_time = jiffies; |
869 | long wrote = 0; | |
38f21977 | 870 | |
f11c9c5c | 871 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 872 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 873 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 874 | |
eb6ef3df | 875 | if (!trylock_super(sb)) { |
0e995816 | 876 | /* |
eb6ef3df | 877 | * trylock_super() may fail consistently due to |
0e995816 WF |
878 | * s_umount being grabbed by someone else. Don't use |
879 | * requeue_io() to avoid busy retrying the inode/sb. | |
880 | */ | |
881 | redirty_tail(inode, wb); | |
edadfb10 | 882 | continue; |
f11c9c5c | 883 | } |
d46db3d5 | 884 | wrote += writeback_sb_inodes(sb, wb, work); |
eb6ef3df | 885 | up_read(&sb->s_umount); |
f11c9c5c | 886 | |
d46db3d5 WF |
887 | /* refer to the same tests at the end of writeback_sb_inodes */ |
888 | if (wrote) { | |
889 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
890 | break; | |
891 | if (work->nr_pages <= 0) | |
892 | break; | |
893 | } | |
f11c9c5c | 894 | } |
66f3b8e2 | 895 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 896 | return wrote; |
66f3b8e2 JA |
897 | } |
898 | ||
7d9f073b | 899 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
0e175a18 | 900 | enum wb_reason reason) |
edadfb10 | 901 | { |
d46db3d5 WF |
902 | struct wb_writeback_work work = { |
903 | .nr_pages = nr_pages, | |
904 | .sync_mode = WB_SYNC_NONE, | |
905 | .range_cyclic = 1, | |
0e175a18 | 906 | .reason = reason, |
d46db3d5 | 907 | }; |
edadfb10 | 908 | |
f758eeab | 909 | spin_lock(&wb->list_lock); |
424b351f | 910 | if (list_empty(&wb->b_io)) |
ad4e38dd | 911 | queue_io(wb, &work); |
d46db3d5 | 912 | __writeback_inodes_wb(wb, &work); |
f758eeab | 913 | spin_unlock(&wb->list_lock); |
edadfb10 | 914 | |
d46db3d5 WF |
915 | return nr_pages - work.nr_pages; |
916 | } | |
03ba3782 | 917 | |
a88a341a | 918 | static bool over_bground_thresh(struct bdi_writeback *wb) |
03ba3782 JA |
919 | { |
920 | unsigned long background_thresh, dirty_thresh; | |
921 | ||
16c4042f | 922 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 | 923 | |
b00949aa WF |
924 | if (global_page_state(NR_FILE_DIRTY) + |
925 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) | |
926 | return true; | |
927 | ||
a88a341a | 928 | if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh)) |
b00949aa WF |
929 | return true; |
930 | ||
931 | return false; | |
03ba3782 JA |
932 | } |
933 | ||
e98be2d5 WF |
934 | /* |
935 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
936 | * only the flusher working on the first wb should do it. | |
937 | */ | |
938 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
939 | unsigned long start_time) | |
940 | { | |
a88a341a | 941 | __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time); |
e98be2d5 WF |
942 | } |
943 | ||
03ba3782 JA |
944 | /* |
945 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 946 | * |
03ba3782 JA |
947 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
948 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
949 | * just walks the superblock inode list, writing back any inodes which are | |
950 | * older than a specific point in time. | |
66f3b8e2 | 951 | * |
03ba3782 JA |
952 | * Try to run once per dirty_writeback_interval. But if a writeback event |
953 | * takes longer than a dirty_writeback_interval interval, then leave a | |
954 | * one-second gap. | |
66f3b8e2 | 955 | * |
03ba3782 JA |
956 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
957 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 958 | */ |
c4a77a6c | 959 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 960 | struct wb_writeback_work *work) |
66f3b8e2 | 961 | { |
e98be2d5 | 962 | unsigned long wb_start = jiffies; |
d46db3d5 | 963 | long nr_pages = work->nr_pages; |
0dc83bd3 | 964 | unsigned long oldest_jif; |
a5989bdc | 965 | struct inode *inode; |
d46db3d5 | 966 | long progress; |
66f3b8e2 | 967 | |
0dc83bd3 JK |
968 | oldest_jif = jiffies; |
969 | work->older_than_this = &oldest_jif; | |
38f21977 | 970 | |
e8dfc305 | 971 | spin_lock(&wb->list_lock); |
03ba3782 JA |
972 | for (;;) { |
973 | /* | |
d3ddec76 | 974 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 975 | */ |
83ba7b07 | 976 | if (work->nr_pages <= 0) |
03ba3782 | 977 | break; |
66f3b8e2 | 978 | |
aa373cf5 JK |
979 | /* |
980 | * Background writeout and kupdate-style writeback may | |
981 | * run forever. Stop them if there is other work to do | |
982 | * so that e.g. sync can proceed. They'll be restarted | |
983 | * after the other works are all done. | |
984 | */ | |
985 | if ((work->for_background || work->for_kupdate) && | |
f0054bb1 | 986 | !list_empty(&wb->work_list)) |
aa373cf5 JK |
987 | break; |
988 | ||
38f21977 | 989 | /* |
d3ddec76 WF |
990 | * For background writeout, stop when we are below the |
991 | * background dirty threshold | |
38f21977 | 992 | */ |
a88a341a | 993 | if (work->for_background && !over_bground_thresh(wb)) |
03ba3782 | 994 | break; |
38f21977 | 995 | |
1bc36b64 JK |
996 | /* |
997 | * Kupdate and background works are special and we want to | |
998 | * include all inodes that need writing. Livelock avoidance is | |
999 | * handled by these works yielding to any other work so we are | |
1000 | * safe. | |
1001 | */ | |
ba9aa839 | 1002 | if (work->for_kupdate) { |
0dc83bd3 | 1003 | oldest_jif = jiffies - |
ba9aa839 | 1004 | msecs_to_jiffies(dirty_expire_interval * 10); |
1bc36b64 | 1005 | } else if (work->for_background) |
0dc83bd3 | 1006 | oldest_jif = jiffies; |
028c2dd1 | 1007 | |
d46db3d5 | 1008 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 1009 | if (list_empty(&wb->b_io)) |
ad4e38dd | 1010 | queue_io(wb, work); |
83ba7b07 | 1011 | if (work->sb) |
d46db3d5 | 1012 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 1013 | else |
d46db3d5 WF |
1014 | progress = __writeback_inodes_wb(wb, work); |
1015 | trace_writeback_written(wb->bdi, work); | |
028c2dd1 | 1016 | |
e98be2d5 | 1017 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
1018 | |
1019 | /* | |
e6fb6da2 WF |
1020 | * Did we write something? Try for more |
1021 | * | |
1022 | * Dirty inodes are moved to b_io for writeback in batches. | |
1023 | * The completion of the current batch does not necessarily | |
1024 | * mean the overall work is done. So we keep looping as long | |
1025 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 1026 | */ |
d46db3d5 | 1027 | if (progress) |
71fd05a8 JA |
1028 | continue; |
1029 | /* | |
e6fb6da2 | 1030 | * No more inodes for IO, bail |
71fd05a8 | 1031 | */ |
b7a2441f | 1032 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 1033 | break; |
71fd05a8 JA |
1034 | /* |
1035 | * Nothing written. Wait for some inode to | |
1036 | * become available for writeback. Otherwise | |
1037 | * we'll just busyloop. | |
1038 | */ | |
71fd05a8 | 1039 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 1040 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 1041 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 1042 | spin_lock(&inode->i_lock); |
f0d07b7f | 1043 | spin_unlock(&wb->list_lock); |
169ebd90 JK |
1044 | /* This function drops i_lock... */ |
1045 | inode_sleep_on_writeback(inode); | |
f0d07b7f | 1046 | spin_lock(&wb->list_lock); |
03ba3782 JA |
1047 | } |
1048 | } | |
e8dfc305 | 1049 | spin_unlock(&wb->list_lock); |
03ba3782 | 1050 | |
d46db3d5 | 1051 | return nr_pages - work->nr_pages; |
03ba3782 JA |
1052 | } |
1053 | ||
1054 | /* | |
83ba7b07 | 1055 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 1056 | */ |
f0054bb1 | 1057 | static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) |
03ba3782 | 1058 | { |
83ba7b07 | 1059 | struct wb_writeback_work *work = NULL; |
03ba3782 | 1060 | |
f0054bb1 TH |
1061 | spin_lock_bh(&wb->work_lock); |
1062 | if (!list_empty(&wb->work_list)) { | |
1063 | work = list_entry(wb->work_list.next, | |
83ba7b07 CH |
1064 | struct wb_writeback_work, list); |
1065 | list_del_init(&work->list); | |
03ba3782 | 1066 | } |
f0054bb1 | 1067 | spin_unlock_bh(&wb->work_lock); |
83ba7b07 | 1068 | return work; |
03ba3782 JA |
1069 | } |
1070 | ||
cdf01dd5 LT |
1071 | /* |
1072 | * Add in the number of potentially dirty inodes, because each inode | |
1073 | * write can dirty pagecache in the underlying blockdev. | |
1074 | */ | |
1075 | static unsigned long get_nr_dirty_pages(void) | |
1076 | { | |
1077 | return global_page_state(NR_FILE_DIRTY) + | |
1078 | global_page_state(NR_UNSTABLE_NFS) + | |
1079 | get_nr_dirty_inodes(); | |
1080 | } | |
1081 | ||
6585027a JK |
1082 | static long wb_check_background_flush(struct bdi_writeback *wb) |
1083 | { | |
a88a341a | 1084 | if (over_bground_thresh(wb)) { |
6585027a JK |
1085 | |
1086 | struct wb_writeback_work work = { | |
1087 | .nr_pages = LONG_MAX, | |
1088 | .sync_mode = WB_SYNC_NONE, | |
1089 | .for_background = 1, | |
1090 | .range_cyclic = 1, | |
0e175a18 | 1091 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
1092 | }; |
1093 | ||
1094 | return wb_writeback(wb, &work); | |
1095 | } | |
1096 | ||
1097 | return 0; | |
1098 | } | |
1099 | ||
03ba3782 JA |
1100 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
1101 | { | |
1102 | unsigned long expired; | |
1103 | long nr_pages; | |
1104 | ||
69b62d01 JA |
1105 | /* |
1106 | * When set to zero, disable periodic writeback | |
1107 | */ | |
1108 | if (!dirty_writeback_interval) | |
1109 | return 0; | |
1110 | ||
03ba3782 JA |
1111 | expired = wb->last_old_flush + |
1112 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
1113 | if (time_before(jiffies, expired)) | |
1114 | return 0; | |
1115 | ||
1116 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 1117 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 1118 | |
c4a77a6c | 1119 | if (nr_pages) { |
83ba7b07 | 1120 | struct wb_writeback_work work = { |
c4a77a6c JA |
1121 | .nr_pages = nr_pages, |
1122 | .sync_mode = WB_SYNC_NONE, | |
1123 | .for_kupdate = 1, | |
1124 | .range_cyclic = 1, | |
0e175a18 | 1125 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
1126 | }; |
1127 | ||
83ba7b07 | 1128 | return wb_writeback(wb, &work); |
c4a77a6c | 1129 | } |
03ba3782 JA |
1130 | |
1131 | return 0; | |
1132 | } | |
1133 | ||
1134 | /* | |
1135 | * Retrieve work items and do the writeback they describe | |
1136 | */ | |
25d130ba | 1137 | static long wb_do_writeback(struct bdi_writeback *wb) |
03ba3782 | 1138 | { |
83ba7b07 | 1139 | struct wb_writeback_work *work; |
c4a77a6c | 1140 | long wrote = 0; |
03ba3782 | 1141 | |
4452226e | 1142 | set_bit(WB_writeback_running, &wb->state); |
f0054bb1 | 1143 | while ((work = get_next_work_item(wb)) != NULL) { |
03ba3782 | 1144 | |
f0054bb1 | 1145 | trace_writeback_exec(wb->bdi, work); |
455b2864 | 1146 | |
83ba7b07 | 1147 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
1148 | |
1149 | /* | |
83ba7b07 CH |
1150 | * Notify the caller of completion if this is a synchronous |
1151 | * work item, otherwise just free it. | |
03ba3782 | 1152 | */ |
83ba7b07 CH |
1153 | if (work->done) |
1154 | complete(work->done); | |
1155 | else | |
1156 | kfree(work); | |
03ba3782 JA |
1157 | } |
1158 | ||
1159 | /* | |
1160 | * Check for periodic writeback, kupdated() style | |
1161 | */ | |
1162 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 1163 | wrote += wb_check_background_flush(wb); |
4452226e | 1164 | clear_bit(WB_writeback_running, &wb->state); |
03ba3782 JA |
1165 | |
1166 | return wrote; | |
1167 | } | |
1168 | ||
1169 | /* | |
1170 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
839a8e86 | 1171 | * reschedules periodically and does kupdated style flushing. |
03ba3782 | 1172 | */ |
f0054bb1 | 1173 | void wb_workfn(struct work_struct *work) |
03ba3782 | 1174 | { |
839a8e86 TH |
1175 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
1176 | struct bdi_writeback, dwork); | |
03ba3782 JA |
1177 | long pages_written; |
1178 | ||
f0054bb1 | 1179 | set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); |
766f9164 | 1180 | current->flags |= PF_SWAPWRITE; |
455b2864 | 1181 | |
839a8e86 | 1182 | if (likely(!current_is_workqueue_rescuer() || |
4452226e | 1183 | !test_bit(WB_registered, &wb->state))) { |
6467716a | 1184 | /* |
f0054bb1 | 1185 | * The normal path. Keep writing back @wb until its |
839a8e86 | 1186 | * work_list is empty. Note that this path is also taken |
f0054bb1 | 1187 | * if @wb is shutting down even when we're running off the |
839a8e86 | 1188 | * rescuer as work_list needs to be drained. |
6467716a | 1189 | */ |
839a8e86 | 1190 | do { |
25d130ba | 1191 | pages_written = wb_do_writeback(wb); |
839a8e86 | 1192 | trace_writeback_pages_written(pages_written); |
f0054bb1 | 1193 | } while (!list_empty(&wb->work_list)); |
839a8e86 TH |
1194 | } else { |
1195 | /* | |
1196 | * bdi_wq can't get enough workers and we're running off | |
1197 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
1198 | * enough for efficient IO. | |
1199 | */ | |
f0054bb1 | 1200 | pages_written = writeback_inodes_wb(wb, 1024, |
839a8e86 | 1201 | WB_REASON_FORKER_THREAD); |
455b2864 | 1202 | trace_writeback_pages_written(pages_written); |
03ba3782 JA |
1203 | } |
1204 | ||
f0054bb1 | 1205 | if (!list_empty(&wb->work_list)) |
6ca738d6 DB |
1206 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
1207 | else if (wb_has_dirty_io(wb) && dirty_writeback_interval) | |
f0054bb1 | 1208 | wb_wakeup_delayed(wb); |
455b2864 | 1209 | |
839a8e86 | 1210 | current->flags &= ~PF_SWAPWRITE; |
03ba3782 JA |
1211 | } |
1212 | ||
1213 | /* | |
b8c2f347 CH |
1214 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
1215 | * the whole world. | |
03ba3782 | 1216 | */ |
0e175a18 | 1217 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
03ba3782 | 1218 | { |
b8c2f347 | 1219 | struct backing_dev_info *bdi; |
03ba3782 | 1220 | |
47df3dde JK |
1221 | if (!nr_pages) |
1222 | nr_pages = get_nr_dirty_pages(); | |
03ba3782 | 1223 | |
b8c2f347 | 1224 | rcu_read_lock(); |
f2b65121 TH |
1225 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
1226 | struct bdi_writeback *wb; | |
1227 | struct wb_iter iter; | |
1228 | ||
1229 | if (!bdi_has_dirty_io(bdi)) | |
1230 | continue; | |
1231 | ||
1232 | bdi_for_each_wb(wb, bdi, &iter, 0) | |
1233 | wb_start_writeback(wb, wb_split_bdi_pages(wb, nr_pages), | |
1234 | false, reason); | |
1235 | } | |
cfc4ba53 | 1236 | rcu_read_unlock(); |
1da177e4 LT |
1237 | } |
1238 | ||
a2f48706 TT |
1239 | /* |
1240 | * Wake up bdi's periodically to make sure dirtytime inodes gets | |
1241 | * written back periodically. We deliberately do *not* check the | |
1242 | * b_dirtytime list in wb_has_dirty_io(), since this would cause the | |
1243 | * kernel to be constantly waking up once there are any dirtytime | |
1244 | * inodes on the system. So instead we define a separate delayed work | |
1245 | * function which gets called much more rarely. (By default, only | |
1246 | * once every 12 hours.) | |
1247 | * | |
1248 | * If there is any other write activity going on in the file system, | |
1249 | * this function won't be necessary. But if the only thing that has | |
1250 | * happened on the file system is a dirtytime inode caused by an atime | |
1251 | * update, we need this infrastructure below to make sure that inode | |
1252 | * eventually gets pushed out to disk. | |
1253 | */ | |
1254 | static void wakeup_dirtytime_writeback(struct work_struct *w); | |
1255 | static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); | |
1256 | ||
1257 | static void wakeup_dirtytime_writeback(struct work_struct *w) | |
1258 | { | |
1259 | struct backing_dev_info *bdi; | |
1260 | ||
1261 | rcu_read_lock(); | |
1262 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { | |
1263 | if (list_empty(&bdi->wb.b_dirty_time)) | |
1264 | continue; | |
f0054bb1 | 1265 | wb_wakeup(&bdi->wb); |
a2f48706 TT |
1266 | } |
1267 | rcu_read_unlock(); | |
1268 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
1269 | } | |
1270 | ||
1271 | static int __init start_dirtytime_writeback(void) | |
1272 | { | |
1273 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
1274 | return 0; | |
1275 | } | |
1276 | __initcall(start_dirtytime_writeback); | |
1277 | ||
1efff914 TT |
1278 | int dirtytime_interval_handler(struct ctl_table *table, int write, |
1279 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1280 | { | |
1281 | int ret; | |
1282 | ||
1283 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
1284 | if (ret == 0 && write) | |
1285 | mod_delayed_work(system_wq, &dirtytime_work, 0); | |
1286 | return ret; | |
1287 | } | |
1288 | ||
03ba3782 JA |
1289 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1290 | { | |
1291 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1292 | struct dentry *dentry; | |
1293 | const char *name = "?"; | |
1294 | ||
1295 | dentry = d_find_alias(inode); | |
1296 | if (dentry) { | |
1297 | spin_lock(&dentry->d_lock); | |
1298 | name = (const char *) dentry->d_name.name; | |
1299 | } | |
1300 | printk(KERN_DEBUG | |
1301 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1302 | current->comm, task_pid_nr(current), inode->i_ino, | |
1303 | name, inode->i_sb->s_id); | |
1304 | if (dentry) { | |
1305 | spin_unlock(&dentry->d_lock); | |
1306 | dput(dentry); | |
1307 | } | |
1308 | } | |
1309 | } | |
1310 | ||
1311 | /** | |
1312 | * __mark_inode_dirty - internal function | |
1313 | * @inode: inode to mark | |
1314 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1315 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1316 | * mark_inode_dirty_sync. | |
1da177e4 | 1317 | * |
03ba3782 JA |
1318 | * Put the inode on the super block's dirty list. |
1319 | * | |
1320 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1321 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1322 | * If it was not hashed, it will never be added to the dirty list | |
1323 | * even if it is later hashed, as it will have been marked dirty already. | |
1324 | * | |
1325 | * In short, make sure you hash any inodes _before_ you start marking | |
1326 | * them dirty. | |
1da177e4 | 1327 | * |
03ba3782 JA |
1328 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1329 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1330 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1331 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1332 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1333 | * blockdev inode. | |
1da177e4 | 1334 | */ |
0ae45f63 | 1335 | #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) |
03ba3782 | 1336 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1337 | { |
03ba3782 | 1338 | struct super_block *sb = inode->i_sb; |
253c34e9 | 1339 | struct backing_dev_info *bdi = NULL; |
0ae45f63 TT |
1340 | int dirtytime; |
1341 | ||
1342 | trace_writeback_mark_inode_dirty(inode, flags); | |
1da177e4 | 1343 | |
03ba3782 JA |
1344 | /* |
1345 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1346 | * dirty the inode itself | |
1347 | */ | |
0ae45f63 | 1348 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { |
9fb0a7da TH |
1349 | trace_writeback_dirty_inode_start(inode, flags); |
1350 | ||
03ba3782 | 1351 | if (sb->s_op->dirty_inode) |
aa385729 | 1352 | sb->s_op->dirty_inode(inode, flags); |
9fb0a7da TH |
1353 | |
1354 | trace_writeback_dirty_inode(inode, flags); | |
03ba3782 | 1355 | } |
0ae45f63 TT |
1356 | if (flags & I_DIRTY_INODE) |
1357 | flags &= ~I_DIRTY_TIME; | |
1358 | dirtytime = flags & I_DIRTY_TIME; | |
03ba3782 JA |
1359 | |
1360 | /* | |
9c6ac78e TH |
1361 | * Paired with smp_mb() in __writeback_single_inode() for the |
1362 | * following lockless i_state test. See there for details. | |
03ba3782 JA |
1363 | */ |
1364 | smp_mb(); | |
1365 | ||
0ae45f63 TT |
1366 | if (((inode->i_state & flags) == flags) || |
1367 | (dirtytime && (inode->i_state & I_DIRTY_INODE))) | |
03ba3782 JA |
1368 | return; |
1369 | ||
1370 | if (unlikely(block_dump)) | |
1371 | block_dump___mark_inode_dirty(inode); | |
1372 | ||
250df6ed | 1373 | spin_lock(&inode->i_lock); |
0ae45f63 TT |
1374 | if (dirtytime && (inode->i_state & I_DIRTY_INODE)) |
1375 | goto out_unlock_inode; | |
03ba3782 JA |
1376 | if ((inode->i_state & flags) != flags) { |
1377 | const int was_dirty = inode->i_state & I_DIRTY; | |
1378 | ||
52ebea74 TH |
1379 | inode_attach_wb(inode, NULL); |
1380 | ||
0ae45f63 TT |
1381 | if (flags & I_DIRTY_INODE) |
1382 | inode->i_state &= ~I_DIRTY_TIME; | |
03ba3782 JA |
1383 | inode->i_state |= flags; |
1384 | ||
1385 | /* | |
1386 | * If the inode is being synced, just update its dirty state. | |
1387 | * The unlocker will place the inode on the appropriate | |
1388 | * superblock list, based upon its state. | |
1389 | */ | |
1390 | if (inode->i_state & I_SYNC) | |
250df6ed | 1391 | goto out_unlock_inode; |
03ba3782 JA |
1392 | |
1393 | /* | |
1394 | * Only add valid (hashed) inodes to the superblock's | |
1395 | * dirty list. Add blockdev inodes as well. | |
1396 | */ | |
1397 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1398 | if (inode_unhashed(inode)) |
250df6ed | 1399 | goto out_unlock_inode; |
03ba3782 | 1400 | } |
a4ffdde6 | 1401 | if (inode->i_state & I_FREEING) |
250df6ed | 1402 | goto out_unlock_inode; |
03ba3782 JA |
1403 | |
1404 | /* | |
1405 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1406 | * reposition it (that would break b_dirty time-ordering). | |
1407 | */ | |
1408 | if (!was_dirty) { | |
d6c10f1f | 1409 | struct list_head *dirty_list; |
a66979ab | 1410 | bool wakeup_bdi = false; |
253c34e9 AB |
1411 | bdi = inode_to_bdi(inode); |
1412 | ||
146d7009 JB |
1413 | spin_unlock(&inode->i_lock); |
1414 | spin_lock(&bdi->wb.list_lock); | |
253c34e9 | 1415 | |
d6c10f1f TH |
1416 | WARN(bdi_cap_writeback_dirty(bdi) && |
1417 | !test_bit(WB_registered, &bdi->wb.state), | |
1418 | "bdi-%s not registered\n", bdi->name); | |
03ba3782 JA |
1419 | |
1420 | inode->dirtied_when = jiffies; | |
a2f48706 TT |
1421 | if (dirtytime) |
1422 | inode->dirtied_time_when = jiffies; | |
d6c10f1f | 1423 | |
a2f48706 | 1424 | if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) |
d6c10f1f | 1425 | dirty_list = &bdi->wb.b_dirty; |
a2f48706 | 1426 | else |
d6c10f1f TH |
1427 | dirty_list = &bdi->wb.b_dirty_time; |
1428 | ||
1429 | wakeup_bdi = inode_wb_list_move_locked(inode, &bdi->wb, | |
1430 | dirty_list); | |
1431 | ||
f758eeab | 1432 | spin_unlock(&bdi->wb.list_lock); |
0ae45f63 | 1433 | trace_writeback_dirty_inode_enqueue(inode); |
a66979ab | 1434 | |
d6c10f1f TH |
1435 | /* |
1436 | * If this is the first dirty inode for this bdi, | |
1437 | * we have to wake-up the corresponding bdi thread | |
1438 | * to make sure background write-back happens | |
1439 | * later. | |
1440 | */ | |
1441 | if (bdi_cap_writeback_dirty(bdi) && wakeup_bdi) | |
f0054bb1 | 1442 | wb_wakeup_delayed(&bdi->wb); |
a66979ab | 1443 | return; |
1da177e4 | 1444 | } |
1da177e4 | 1445 | } |
250df6ed DC |
1446 | out_unlock_inode: |
1447 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1448 | |
03ba3782 JA |
1449 | } |
1450 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1451 | ||
b6e51316 | 1452 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1453 | { |
1454 | struct inode *inode, *old_inode = NULL; | |
1455 | ||
1456 | /* | |
1457 | * We need to be protected against the filesystem going from | |
1458 | * r/o to r/w or vice versa. | |
1459 | */ | |
b6e51316 | 1460 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1461 | |
55fa6091 | 1462 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1463 | |
1464 | /* | |
1465 | * Data integrity sync. Must wait for all pages under writeback, | |
1466 | * because there may have been pages dirtied before our sync | |
1467 | * call, but which had writeout started before we write it out. | |
1468 | * In which case, the inode may not be on the dirty list, but | |
1469 | * we still have to wait for that writeout. | |
1470 | */ | |
b6e51316 | 1471 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1472 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1473 | |
250df6ed DC |
1474 | spin_lock(&inode->i_lock); |
1475 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1476 | (mapping->nrpages == 0)) { | |
1477 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1478 | continue; |
250df6ed | 1479 | } |
03ba3782 | 1480 | __iget(inode); |
250df6ed | 1481 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1482 | spin_unlock(&inode_sb_list_lock); |
1483 | ||
03ba3782 | 1484 | /* |
55fa6091 DC |
1485 | * We hold a reference to 'inode' so it couldn't have been |
1486 | * removed from s_inodes list while we dropped the | |
1487 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1488 | * be holding the last reference and we cannot iput it under | |
1489 | * inode_sb_list_lock. So we keep the reference and iput it | |
1490 | * later. | |
03ba3782 JA |
1491 | */ |
1492 | iput(old_inode); | |
1493 | old_inode = inode; | |
1494 | ||
1495 | filemap_fdatawait(mapping); | |
1496 | ||
1497 | cond_resched(); | |
1498 | ||
55fa6091 | 1499 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1500 | } |
55fa6091 | 1501 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1502 | iput(old_inode); |
1da177e4 LT |
1503 | } |
1504 | ||
d8a8559c | 1505 | /** |
3259f8be | 1506 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1507 | * @sb: the superblock |
3259f8be | 1508 | * @nr: the number of pages to write |
786228ab | 1509 | * @reason: reason why some writeback work initiated |
1da177e4 | 1510 | * |
d8a8559c JA |
1511 | * Start writeback on some inodes on this super_block. No guarantees are made |
1512 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1513 | * for IO completion of submitted IO. |
1da177e4 | 1514 | */ |
0e175a18 CW |
1515 | void writeback_inodes_sb_nr(struct super_block *sb, |
1516 | unsigned long nr, | |
1517 | enum wb_reason reason) | |
1da177e4 | 1518 | { |
83ba7b07 CH |
1519 | DECLARE_COMPLETION_ONSTACK(done); |
1520 | struct wb_writeback_work work = { | |
6e6938b6 WF |
1521 | .sb = sb, |
1522 | .sync_mode = WB_SYNC_NONE, | |
1523 | .tagged_writepages = 1, | |
1524 | .done = &done, | |
1525 | .nr_pages = nr, | |
0e175a18 | 1526 | .reason = reason, |
3c4d7165 | 1527 | }; |
e7972912 | 1528 | struct backing_dev_info *bdi = sb->s_bdi; |
d8a8559c | 1529 | |
e7972912 | 1530 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
6eedc701 | 1531 | return; |
cf37e972 | 1532 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
e7972912 | 1533 | wb_queue_work(&bdi->wb, &work); |
83ba7b07 | 1534 | wait_for_completion(&done); |
e913fc82 | 1535 | } |
3259f8be CM |
1536 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1537 | ||
1538 | /** | |
1539 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1540 | * @sb: the superblock | |
786228ab | 1541 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1542 | * |
1543 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1544 | * on how many (if any) will be written, and this function does not wait | |
1545 | * for IO completion of submitted IO. | |
1546 | */ | |
0e175a18 | 1547 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1548 | { |
0e175a18 | 1549 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1550 | } |
0e3c9a22 | 1551 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1552 | |
17bd55d0 | 1553 | /** |
10ee27a0 | 1554 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway |
17bd55d0 | 1555 | * @sb: the superblock |
10ee27a0 MX |
1556 | * @nr: the number of pages to write |
1557 | * @reason: the reason of writeback | |
17bd55d0 | 1558 | * |
10ee27a0 | 1559 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. |
17bd55d0 ES |
1560 | * Returns 1 if writeback was started, 0 if not. |
1561 | */ | |
10ee27a0 MX |
1562 | int try_to_writeback_inodes_sb_nr(struct super_block *sb, |
1563 | unsigned long nr, | |
1564 | enum wb_reason reason) | |
17bd55d0 | 1565 | { |
bc05873d | 1566 | if (writeback_in_progress(&sb->s_bdi->wb)) |
17bd55d0 | 1567 | return 1; |
10ee27a0 MX |
1568 | |
1569 | if (!down_read_trylock(&sb->s_umount)) | |
17bd55d0 | 1570 | return 0; |
10ee27a0 MX |
1571 | |
1572 | writeback_inodes_sb_nr(sb, nr, reason); | |
1573 | up_read(&sb->s_umount); | |
1574 | return 1; | |
17bd55d0 | 1575 | } |
10ee27a0 | 1576 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); |
17bd55d0 | 1577 | |
3259f8be | 1578 | /** |
10ee27a0 | 1579 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
3259f8be | 1580 | * @sb: the superblock |
786228ab | 1581 | * @reason: reason why some writeback work was initiated |
3259f8be | 1582 | * |
10ee27a0 | 1583 | * Implement by try_to_writeback_inodes_sb_nr() |
3259f8be CM |
1584 | * Returns 1 if writeback was started, 0 if not. |
1585 | */ | |
10ee27a0 | 1586 | int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1587 | { |
10ee27a0 | 1588 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1589 | } |
10ee27a0 | 1590 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
3259f8be | 1591 | |
d8a8559c JA |
1592 | /** |
1593 | * sync_inodes_sb - sync sb inode pages | |
0dc83bd3 | 1594 | * @sb: the superblock |
d8a8559c JA |
1595 | * |
1596 | * This function writes and waits on any dirty inode belonging to this | |
0dc83bd3 | 1597 | * super_block. |
d8a8559c | 1598 | */ |
0dc83bd3 | 1599 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1600 | { |
83ba7b07 CH |
1601 | DECLARE_COMPLETION_ONSTACK(done); |
1602 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1603 | .sb = sb, |
1604 | .sync_mode = WB_SYNC_ALL, | |
1605 | .nr_pages = LONG_MAX, | |
1606 | .range_cyclic = 0, | |
83ba7b07 | 1607 | .done = &done, |
0e175a18 | 1608 | .reason = WB_REASON_SYNC, |
7747bd4b | 1609 | .for_sync = 1, |
3c4d7165 | 1610 | }; |
e7972912 | 1611 | struct backing_dev_info *bdi = sb->s_bdi; |
3c4d7165 | 1612 | |
6eedc701 | 1613 | /* Nothing to do? */ |
e7972912 | 1614 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
6eedc701 | 1615 | return; |
cf37e972 CH |
1616 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1617 | ||
e7972912 | 1618 | wb_queue_work(&bdi->wb, &work); |
83ba7b07 CH |
1619 | wait_for_completion(&done); |
1620 | ||
b6e51316 | 1621 | wait_sb_inodes(sb); |
1da177e4 | 1622 | } |
d8a8559c | 1623 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1624 | |
1da177e4 | 1625 | /** |
7f04c26d AA |
1626 | * write_inode_now - write an inode to disk |
1627 | * @inode: inode to write to disk | |
1628 | * @sync: whether the write should be synchronous or not | |
1629 | * | |
1630 | * This function commits an inode to disk immediately if it is dirty. This is | |
1631 | * primarily needed by knfsd. | |
1da177e4 | 1632 | * |
7f04c26d | 1633 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1634 | */ |
1da177e4 LT |
1635 | int write_inode_now(struct inode *inode, int sync) |
1636 | { | |
f758eeab | 1637 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1638 | struct writeback_control wbc = { |
1639 | .nr_to_write = LONG_MAX, | |
18914b18 | 1640 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1641 | .range_start = 0, |
1642 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1643 | }; |
1644 | ||
1645 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1646 | wbc.nr_to_write = 0; |
1da177e4 LT |
1647 | |
1648 | might_sleep(); | |
4f8ad655 | 1649 | return writeback_single_inode(inode, wb, &wbc); |
1da177e4 LT |
1650 | } |
1651 | EXPORT_SYMBOL(write_inode_now); | |
1652 | ||
1653 | /** | |
1654 | * sync_inode - write an inode and its pages to disk. | |
1655 | * @inode: the inode to sync | |
1656 | * @wbc: controls the writeback mode | |
1657 | * | |
1658 | * sync_inode() will write an inode and its pages to disk. It will also | |
1659 | * correctly update the inode on its superblock's dirty inode lists and will | |
1660 | * update inode->i_state. | |
1661 | * | |
1662 | * The caller must have a ref on the inode. | |
1663 | */ | |
1664 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1665 | { | |
4f8ad655 | 1666 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); |
1da177e4 LT |
1667 | } |
1668 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1669 | |
1670 | /** | |
c691b9d9 | 1671 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1672 | * @inode: the inode to sync |
1673 | * @wait: wait for I/O to complete. | |
1674 | * | |
c691b9d9 | 1675 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1676 | * |
1677 | * Note: only writes the actual inode, no associated data or other metadata. | |
1678 | */ | |
1679 | int sync_inode_metadata(struct inode *inode, int wait) | |
1680 | { | |
1681 | struct writeback_control wbc = { | |
1682 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1683 | .nr_to_write = 0, /* metadata-only */ | |
1684 | }; | |
1685 | ||
1686 | return sync_inode(inode, &wbc); | |
1687 | } | |
1688 | EXPORT_SYMBOL(sync_inode_metadata); |