]>
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 | ||
f11b00f3 AB |
68 | /** |
69 | * writeback_in_progress - determine whether there is writeback in progress | |
70 | * @bdi: the device's backing_dev_info structure. | |
71 | * | |
03ba3782 JA |
72 | * Determine whether there is writeback waiting to be handled against a |
73 | * backing device. | |
f11b00f3 AB |
74 | */ |
75 | int writeback_in_progress(struct backing_dev_info *bdi) | |
76 | { | |
4452226e | 77 | return test_bit(WB_writeback_running, &bdi->wb.state); |
f11b00f3 | 78 | } |
00d4e736 | 79 | EXPORT_SYMBOL(writeback_in_progress); |
f11b00f3 | 80 | |
7ccf19a8 NP |
81 | static inline struct inode *wb_inode(struct list_head *head) |
82 | { | |
83 | return list_entry(head, struct inode, i_wb_list); | |
84 | } | |
85 | ||
15eb77a0 WF |
86 | /* |
87 | * Include the creation of the trace points after defining the | |
88 | * wb_writeback_work structure and inline functions so that the definition | |
89 | * remains local to this file. | |
90 | */ | |
91 | #define CREATE_TRACE_POINTS | |
92 | #include <trace/events/writeback.h> | |
93 | ||
774016b2 SW |
94 | EXPORT_TRACEPOINT_SYMBOL_GPL(wbc_writepage); |
95 | ||
d6c10f1f TH |
96 | static bool wb_io_lists_populated(struct bdi_writeback *wb) |
97 | { | |
98 | if (wb_has_dirty_io(wb)) { | |
99 | return false; | |
100 | } else { | |
101 | set_bit(WB_has_dirty_io, &wb->state); | |
95a46c65 | 102 | WARN_ON_ONCE(!wb->avg_write_bandwidth); |
766a9d6e TH |
103 | atomic_long_add(wb->avg_write_bandwidth, |
104 | &wb->bdi->tot_write_bandwidth); | |
d6c10f1f TH |
105 | return true; |
106 | } | |
107 | } | |
108 | ||
109 | static void wb_io_lists_depopulated(struct bdi_writeback *wb) | |
110 | { | |
111 | if (wb_has_dirty_io(wb) && list_empty(&wb->b_dirty) && | |
766a9d6e | 112 | list_empty(&wb->b_io) && list_empty(&wb->b_more_io)) { |
d6c10f1f | 113 | clear_bit(WB_has_dirty_io, &wb->state); |
95a46c65 TH |
114 | WARN_ON_ONCE(atomic_long_sub_return(wb->avg_write_bandwidth, |
115 | &wb->bdi->tot_write_bandwidth) < 0); | |
766a9d6e | 116 | } |
d6c10f1f TH |
117 | } |
118 | ||
119 | /** | |
120 | * inode_wb_list_move_locked - move an inode onto a bdi_writeback IO list | |
121 | * @inode: inode to be moved | |
122 | * @wb: target bdi_writeback | |
123 | * @head: one of @wb->b_{dirty|io|more_io} | |
124 | * | |
125 | * Move @inode->i_wb_list to @list of @wb and set %WB_has_dirty_io. | |
126 | * Returns %true if @inode is the first occupant of the !dirty_time IO | |
127 | * lists; otherwise, %false. | |
128 | */ | |
129 | static bool inode_wb_list_move_locked(struct inode *inode, | |
130 | struct bdi_writeback *wb, | |
131 | struct list_head *head) | |
132 | { | |
133 | assert_spin_locked(&wb->list_lock); | |
134 | ||
135 | list_move(&inode->i_wb_list, head); | |
136 | ||
137 | /* dirty_time doesn't count as dirty_io until expiration */ | |
138 | if (head != &wb->b_dirty_time) | |
139 | return wb_io_lists_populated(wb); | |
140 | ||
141 | wb_io_lists_depopulated(wb); | |
142 | return false; | |
143 | } | |
144 | ||
145 | /** | |
146 | * inode_wb_list_del_locked - remove an inode from its bdi_writeback IO list | |
147 | * @inode: inode to be removed | |
148 | * @wb: bdi_writeback @inode is being removed from | |
149 | * | |
150 | * Remove @inode which may be on one of @wb->b_{dirty|io|more_io} lists and | |
151 | * clear %WB_has_dirty_io if all are empty afterwards. | |
152 | */ | |
153 | static void inode_wb_list_del_locked(struct inode *inode, | |
154 | struct bdi_writeback *wb) | |
155 | { | |
156 | assert_spin_locked(&wb->list_lock); | |
157 | ||
158 | list_del_init(&inode->i_wb_list); | |
159 | wb_io_lists_depopulated(wb); | |
160 | } | |
161 | ||
f0054bb1 | 162 | static void wb_wakeup(struct bdi_writeback *wb) |
5acda9d1 | 163 | { |
f0054bb1 TH |
164 | spin_lock_bh(&wb->work_lock); |
165 | if (test_bit(WB_registered, &wb->state)) | |
166 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
167 | spin_unlock_bh(&wb->work_lock); | |
5acda9d1 JK |
168 | } |
169 | ||
f0054bb1 TH |
170 | static void wb_queue_work(struct bdi_writeback *wb, |
171 | struct wb_writeback_work *work) | |
6585027a | 172 | { |
f0054bb1 | 173 | trace_writeback_queue(wb->bdi, work); |
6585027a | 174 | |
f0054bb1 TH |
175 | spin_lock_bh(&wb->work_lock); |
176 | if (!test_bit(WB_registered, &wb->state)) { | |
5acda9d1 JK |
177 | if (work->done) |
178 | complete(work->done); | |
179 | goto out_unlock; | |
180 | } | |
f0054bb1 TH |
181 | list_add_tail(&work->list, &wb->work_list); |
182 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
5acda9d1 | 183 | out_unlock: |
f0054bb1 | 184 | spin_unlock_bh(&wb->work_lock); |
1da177e4 LT |
185 | } |
186 | ||
f0054bb1 TH |
187 | static void __wb_start_writeback(struct bdi_writeback *wb, long nr_pages, |
188 | bool range_cyclic, enum wb_reason reason) | |
1da177e4 | 189 | { |
83ba7b07 | 190 | struct wb_writeback_work *work; |
03ba3782 | 191 | |
e7972912 TH |
192 | if (!wb_has_dirty_io(wb)) |
193 | return; | |
194 | ||
bcddc3f0 JA |
195 | /* |
196 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
197 | * wakeup the thread for old dirty data writeback | |
198 | */ | |
83ba7b07 CH |
199 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
200 | if (!work) { | |
f0054bb1 TH |
201 | trace_writeback_nowork(wb->bdi); |
202 | wb_wakeup(wb); | |
83ba7b07 | 203 | return; |
bcddc3f0 | 204 | } |
03ba3782 | 205 | |
83ba7b07 CH |
206 | work->sync_mode = WB_SYNC_NONE; |
207 | work->nr_pages = nr_pages; | |
208 | work->range_cyclic = range_cyclic; | |
0e175a18 | 209 | work->reason = reason; |
03ba3782 | 210 | |
f0054bb1 | 211 | wb_queue_work(wb, work); |
b6e51316 JA |
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 | ||
241 | #endif /* CONFIG_CGROUP_WRITEBACK */ | |
242 | ||
b6e51316 JA |
243 | /** |
244 | * bdi_start_writeback - start writeback | |
245 | * @bdi: the backing device to write from | |
246 | * @nr_pages: the number of pages to write | |
786228ab | 247 | * @reason: reason why some writeback work was initiated |
b6e51316 JA |
248 | * |
249 | * Description: | |
250 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
25985edc | 251 | * started when this function returns, we make no guarantees on |
0e3c9a22 | 252 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
253 | * |
254 | */ | |
0e175a18 CW |
255 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
256 | enum wb_reason reason) | |
b6e51316 | 257 | { |
f0054bb1 | 258 | __wb_start_writeback(&bdi->wb, nr_pages, true, reason); |
c5444198 | 259 | } |
d3ddec76 | 260 | |
c5444198 CH |
261 | /** |
262 | * bdi_start_background_writeback - start background writeback | |
263 | * @bdi: the backing device to write from | |
264 | * | |
265 | * Description: | |
6585027a JK |
266 | * This makes sure WB_SYNC_NONE background writeback happens. When |
267 | * this function returns, it is only guaranteed that for given BDI | |
268 | * some IO is happening if we are over background dirty threshold. | |
269 | * Caller need not hold sb s_umount semaphore. | |
c5444198 CH |
270 | */ |
271 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
272 | { | |
6585027a JK |
273 | /* |
274 | * We just wake up the flusher thread. It will perform background | |
275 | * writeback as soon as there is no other work to do. | |
276 | */ | |
71927e84 | 277 | trace_writeback_wake_background(bdi); |
f0054bb1 | 278 | wb_wakeup(&bdi->wb); |
1da177e4 LT |
279 | } |
280 | ||
a66979ab DC |
281 | /* |
282 | * Remove the inode from the writeback list it is on. | |
283 | */ | |
284 | void inode_wb_list_del(struct inode *inode) | |
285 | { | |
52ebea74 | 286 | struct bdi_writeback *wb = inode_to_wb(inode); |
f758eeab | 287 | |
52ebea74 | 288 | spin_lock(&wb->list_lock); |
d6c10f1f | 289 | inode_wb_list_del_locked(inode, wb); |
52ebea74 | 290 | spin_unlock(&wb->list_lock); |
a66979ab DC |
291 | } |
292 | ||
6610a0bc AM |
293 | /* |
294 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
295 | * furthest end of its superblock's dirty-inode list. | |
296 | * | |
297 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 298 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
299 | * the case then the inode must have been redirtied while it was being written |
300 | * out and we don't reset its dirtied_when. | |
301 | */ | |
f758eeab | 302 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 303 | { |
03ba3782 | 304 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 305 | struct inode *tail; |
6610a0bc | 306 | |
7ccf19a8 | 307 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 308 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
309 | inode->dirtied_when = jiffies; |
310 | } | |
d6c10f1f | 311 | inode_wb_list_move_locked(inode, wb, &wb->b_dirty); |
6610a0bc AM |
312 | } |
313 | ||
c986d1e2 | 314 | /* |
66f3b8e2 | 315 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 316 | */ |
f758eeab | 317 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 318 | { |
d6c10f1f | 319 | inode_wb_list_move_locked(inode, wb, &wb->b_more_io); |
c986d1e2 AM |
320 | } |
321 | ||
1c0eeaf5 JE |
322 | static void inode_sync_complete(struct inode *inode) |
323 | { | |
365b94ae | 324 | inode->i_state &= ~I_SYNC; |
4eff96dd JK |
325 | /* If inode is clean an unused, put it into LRU now... */ |
326 | inode_add_lru(inode); | |
365b94ae | 327 | /* Waiters must see I_SYNC cleared before being woken up */ |
1c0eeaf5 JE |
328 | smp_mb(); |
329 | wake_up_bit(&inode->i_state, __I_SYNC); | |
330 | } | |
331 | ||
d2caa3c5 JL |
332 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
333 | { | |
334 | bool ret = time_after(inode->dirtied_when, t); | |
335 | #ifndef CONFIG_64BIT | |
336 | /* | |
337 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
338 | * It _appears_ to be in the future, but is actually in distant past. | |
339 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 340 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
341 | */ |
342 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
343 | #endif | |
344 | return ret; | |
345 | } | |
346 | ||
0ae45f63 TT |
347 | #define EXPIRE_DIRTY_ATIME 0x0001 |
348 | ||
2c136579 | 349 | /* |
0e2f2b23 | 350 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
697e6fed | 351 | * @delaying_queue to @dispatch_queue. |
2c136579 | 352 | */ |
e84d0a4f | 353 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 354 | struct list_head *dispatch_queue, |
0ae45f63 | 355 | int flags, |
ad4e38dd | 356 | struct wb_writeback_work *work) |
2c136579 | 357 | { |
0ae45f63 TT |
358 | unsigned long *older_than_this = NULL; |
359 | unsigned long expire_time; | |
5c03449d SL |
360 | LIST_HEAD(tmp); |
361 | struct list_head *pos, *node; | |
cf137307 | 362 | struct super_block *sb = NULL; |
5c03449d | 363 | struct inode *inode; |
cf137307 | 364 | int do_sb_sort = 0; |
e84d0a4f | 365 | int moved = 0; |
5c03449d | 366 | |
0ae45f63 TT |
367 | if ((flags & EXPIRE_DIRTY_ATIME) == 0) |
368 | older_than_this = work->older_than_this; | |
a2f48706 TT |
369 | else if (!work->for_sync) { |
370 | expire_time = jiffies - (dirtytime_expire_interval * HZ); | |
0ae45f63 TT |
371 | older_than_this = &expire_time; |
372 | } | |
2c136579 | 373 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 374 | inode = wb_inode(delaying_queue->prev); |
0ae45f63 TT |
375 | if (older_than_this && |
376 | inode_dirtied_after(inode, *older_than_this)) | |
2c136579 | 377 | break; |
a8855990 JK |
378 | list_move(&inode->i_wb_list, &tmp); |
379 | moved++; | |
0ae45f63 TT |
380 | if (flags & EXPIRE_DIRTY_ATIME) |
381 | set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); | |
a8855990 JK |
382 | if (sb_is_blkdev_sb(inode->i_sb)) |
383 | continue; | |
cf137307 JA |
384 | if (sb && sb != inode->i_sb) |
385 | do_sb_sort = 1; | |
386 | sb = inode->i_sb; | |
5c03449d SL |
387 | } |
388 | ||
cf137307 JA |
389 | /* just one sb in list, splice to dispatch_queue and we're done */ |
390 | if (!do_sb_sort) { | |
391 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 392 | goto out; |
cf137307 JA |
393 | } |
394 | ||
5c03449d SL |
395 | /* Move inodes from one superblock together */ |
396 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 397 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 398 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 399 | inode = wb_inode(pos); |
5c03449d | 400 | if (inode->i_sb == sb) |
7ccf19a8 | 401 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 402 | } |
2c136579 | 403 | } |
e84d0a4f WF |
404 | out: |
405 | return moved; | |
2c136579 FW |
406 | } |
407 | ||
408 | /* | |
409 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
410 | * Before |
411 | * newly dirtied b_dirty b_io b_more_io | |
412 | * =============> gf edc BA | |
413 | * After | |
414 | * newly dirtied b_dirty b_io b_more_io | |
415 | * =============> g fBAedc | |
416 | * | | |
417 | * +--> dequeue for IO | |
2c136579 | 418 | */ |
ad4e38dd | 419 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
66f3b8e2 | 420 | { |
e84d0a4f | 421 | int moved; |
0ae45f63 | 422 | |
f758eeab | 423 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 424 | list_splice_init(&wb->b_more_io, &wb->b_io); |
0ae45f63 TT |
425 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); |
426 | moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, | |
427 | EXPIRE_DIRTY_ATIME, work); | |
d6c10f1f TH |
428 | if (moved) |
429 | wb_io_lists_populated(wb); | |
ad4e38dd | 430 | trace_writeback_queue_io(wb, work, moved); |
66f3b8e2 JA |
431 | } |
432 | ||
a9185b41 | 433 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 434 | { |
9fb0a7da TH |
435 | int ret; |
436 | ||
437 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
438 | trace_writeback_write_inode_start(inode, wbc); | |
439 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
440 | trace_writeback_write_inode(inode, wbc); | |
441 | return ret; | |
442 | } | |
03ba3782 | 443 | return 0; |
08d8e974 | 444 | } |
08d8e974 | 445 | |
1da177e4 | 446 | /* |
169ebd90 JK |
447 | * Wait for writeback on an inode to complete. Called with i_lock held. |
448 | * Caller must make sure inode cannot go away when we drop i_lock. | |
01c03194 | 449 | */ |
169ebd90 JK |
450 | static void __inode_wait_for_writeback(struct inode *inode) |
451 | __releases(inode->i_lock) | |
452 | __acquires(inode->i_lock) | |
01c03194 CH |
453 | { |
454 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
455 | wait_queue_head_t *wqh; | |
456 | ||
457 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
458 | while (inode->i_state & I_SYNC) { |
459 | spin_unlock(&inode->i_lock); | |
74316201 N |
460 | __wait_on_bit(wqh, &wq, bit_wait, |
461 | TASK_UNINTERRUPTIBLE); | |
250df6ed | 462 | spin_lock(&inode->i_lock); |
58a9d3d8 | 463 | } |
01c03194 CH |
464 | } |
465 | ||
169ebd90 JK |
466 | /* |
467 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
468 | */ | |
469 | void inode_wait_for_writeback(struct inode *inode) | |
470 | { | |
471 | spin_lock(&inode->i_lock); | |
472 | __inode_wait_for_writeback(inode); | |
473 | spin_unlock(&inode->i_lock); | |
474 | } | |
475 | ||
476 | /* | |
477 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
478 | * held and drops it. It is aimed for callers not holding any inode reference | |
479 | * so once i_lock is dropped, inode can go away. | |
480 | */ | |
481 | static void inode_sleep_on_writeback(struct inode *inode) | |
482 | __releases(inode->i_lock) | |
483 | { | |
484 | DEFINE_WAIT(wait); | |
485 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
486 | int sleep; | |
487 | ||
488 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
489 | sleep = inode->i_state & I_SYNC; | |
490 | spin_unlock(&inode->i_lock); | |
491 | if (sleep) | |
492 | schedule(); | |
493 | finish_wait(wqh, &wait); | |
494 | } | |
495 | ||
ccb26b5a JK |
496 | /* |
497 | * Find proper writeback list for the inode depending on its current state and | |
498 | * possibly also change of its state while we were doing writeback. Here we | |
499 | * handle things such as livelock prevention or fairness of writeback among | |
500 | * inodes. This function can be called only by flusher thread - noone else | |
501 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
502 | * thread's back can have unexpected consequences. | |
503 | */ | |
504 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
505 | struct writeback_control *wbc) | |
506 | { | |
507 | if (inode->i_state & I_FREEING) | |
508 | return; | |
509 | ||
510 | /* | |
511 | * Sync livelock prevention. Each inode is tagged and synced in one | |
512 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
513 | * the dirty time to prevent enqueue and sync it again. | |
514 | */ | |
515 | if ((inode->i_state & I_DIRTY) && | |
516 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
517 | inode->dirtied_when = jiffies; | |
518 | ||
4f8ad655 JK |
519 | if (wbc->pages_skipped) { |
520 | /* | |
521 | * writeback is not making progress due to locked | |
522 | * buffers. Skip this inode for now. | |
523 | */ | |
524 | redirty_tail(inode, wb); | |
525 | return; | |
526 | } | |
527 | ||
ccb26b5a JK |
528 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
529 | /* | |
530 | * We didn't write back all the pages. nfs_writepages() | |
531 | * sometimes bales out without doing anything. | |
532 | */ | |
533 | if (wbc->nr_to_write <= 0) { | |
534 | /* Slice used up. Queue for next turn. */ | |
535 | requeue_io(inode, wb); | |
536 | } else { | |
537 | /* | |
538 | * Writeback blocked by something other than | |
539 | * congestion. Delay the inode for some time to | |
540 | * avoid spinning on the CPU (100% iowait) | |
541 | * retrying writeback of the dirty page/inode | |
542 | * that cannot be performed immediately. | |
543 | */ | |
544 | redirty_tail(inode, wb); | |
545 | } | |
546 | } else if (inode->i_state & I_DIRTY) { | |
547 | /* | |
548 | * Filesystems can dirty the inode during writeback operations, | |
549 | * such as delayed allocation during submission or metadata | |
550 | * updates after data IO completion. | |
551 | */ | |
552 | redirty_tail(inode, wb); | |
0ae45f63 | 553 | } else if (inode->i_state & I_DIRTY_TIME) { |
a2f48706 | 554 | inode->dirtied_when = jiffies; |
d6c10f1f | 555 | inode_wb_list_move_locked(inode, wb, &wb->b_dirty_time); |
ccb26b5a JK |
556 | } else { |
557 | /* The inode is clean. Remove from writeback lists. */ | |
d6c10f1f | 558 | inode_wb_list_del_locked(inode, wb); |
ccb26b5a JK |
559 | } |
560 | } | |
561 | ||
01c03194 | 562 | /* |
4f8ad655 JK |
563 | * Write out an inode and its dirty pages. Do not update the writeback list |
564 | * linkage. That is left to the caller. The caller is also responsible for | |
565 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. | |
1da177e4 LT |
566 | */ |
567 | static int | |
cd8ed2a4 | 568 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 569 | { |
1da177e4 | 570 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 571 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 572 | unsigned dirty; |
1da177e4 LT |
573 | int ret; |
574 | ||
4f8ad655 | 575 | WARN_ON(!(inode->i_state & I_SYNC)); |
1da177e4 | 576 | |
9fb0a7da TH |
577 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
578 | ||
1da177e4 LT |
579 | ret = do_writepages(mapping, wbc); |
580 | ||
26821ed4 CH |
581 | /* |
582 | * Make sure to wait on the data before writing out the metadata. | |
583 | * This is important for filesystems that modify metadata on data | |
7747bd4b DC |
584 | * I/O completion. We don't do it for sync(2) writeback because it has a |
585 | * separate, external IO completion path and ->sync_fs for guaranteeing | |
586 | * inode metadata is written back correctly. | |
26821ed4 | 587 | */ |
7747bd4b | 588 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
26821ed4 | 589 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
590 | if (ret == 0) |
591 | ret = err; | |
592 | } | |
593 | ||
5547e8aa DM |
594 | /* |
595 | * Some filesystems may redirty the inode during the writeback | |
596 | * due to delalloc, clear dirty metadata flags right before | |
597 | * write_inode() | |
598 | */ | |
250df6ed | 599 | spin_lock(&inode->i_lock); |
9c6ac78e | 600 | |
5547e8aa | 601 | dirty = inode->i_state & I_DIRTY; |
a2f48706 TT |
602 | if (inode->i_state & I_DIRTY_TIME) { |
603 | if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
604 | unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || | |
605 | unlikely(time_after(jiffies, | |
606 | (inode->dirtied_time_when + | |
607 | dirtytime_expire_interval * HZ)))) { | |
608 | dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; | |
609 | trace_writeback_lazytime(inode); | |
610 | } | |
611 | } else | |
612 | inode->i_state &= ~I_DIRTY_TIME_EXPIRED; | |
0ae45f63 | 613 | inode->i_state &= ~dirty; |
9c6ac78e TH |
614 | |
615 | /* | |
616 | * Paired with smp_mb() in __mark_inode_dirty(). This allows | |
617 | * __mark_inode_dirty() to test i_state without grabbing i_lock - | |
618 | * either they see the I_DIRTY bits cleared or we see the dirtied | |
619 | * inode. | |
620 | * | |
621 | * I_DIRTY_PAGES is always cleared together above even if @mapping | |
622 | * still has dirty pages. The flag is reinstated after smp_mb() if | |
623 | * necessary. This guarantees that either __mark_inode_dirty() | |
624 | * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. | |
625 | */ | |
626 | smp_mb(); | |
627 | ||
628 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
629 | inode->i_state |= I_DIRTY_PAGES; | |
630 | ||
250df6ed | 631 | spin_unlock(&inode->i_lock); |
9c6ac78e | 632 | |
0ae45f63 TT |
633 | if (dirty & I_DIRTY_TIME) |
634 | mark_inode_dirty_sync(inode); | |
26821ed4 | 635 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
0ae45f63 | 636 | if (dirty & ~I_DIRTY_PAGES) { |
a9185b41 | 637 | int err = write_inode(inode, wbc); |
1da177e4 LT |
638 | if (ret == 0) |
639 | ret = err; | |
640 | } | |
4f8ad655 JK |
641 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
642 | return ret; | |
643 | } | |
644 | ||
645 | /* | |
646 | * Write out an inode's dirty pages. Either the caller has an active reference | |
647 | * on the inode or the inode has I_WILL_FREE set. | |
648 | * | |
649 | * This function is designed to be called for writing back one inode which | |
650 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() | |
651 | * and does more profound writeback list handling in writeback_sb_inodes(). | |
652 | */ | |
653 | static int | |
654 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, | |
655 | struct writeback_control *wbc) | |
656 | { | |
657 | int ret = 0; | |
658 | ||
659 | spin_lock(&inode->i_lock); | |
660 | if (!atomic_read(&inode->i_count)) | |
661 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
662 | else | |
663 | WARN_ON(inode->i_state & I_WILL_FREE); | |
664 | ||
665 | if (inode->i_state & I_SYNC) { | |
666 | if (wbc->sync_mode != WB_SYNC_ALL) | |
667 | goto out; | |
668 | /* | |
169ebd90 JK |
669 | * It's a data-integrity sync. We must wait. Since callers hold |
670 | * inode reference or inode has I_WILL_FREE set, it cannot go | |
671 | * away under us. | |
4f8ad655 | 672 | */ |
169ebd90 | 673 | __inode_wait_for_writeback(inode); |
4f8ad655 JK |
674 | } |
675 | WARN_ON(inode->i_state & I_SYNC); | |
676 | /* | |
f9b0e058 JK |
677 | * Skip inode if it is clean and we have no outstanding writeback in |
678 | * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this | |
679 | * function since flusher thread may be doing for example sync in | |
680 | * parallel and if we move the inode, it could get skipped. So here we | |
681 | * make sure inode is on some writeback list and leave it there unless | |
682 | * we have completely cleaned the inode. | |
4f8ad655 | 683 | */ |
0ae45f63 | 684 | if (!(inode->i_state & I_DIRTY_ALL) && |
f9b0e058 JK |
685 | (wbc->sync_mode != WB_SYNC_ALL || |
686 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) | |
4f8ad655 JK |
687 | goto out; |
688 | inode->i_state |= I_SYNC; | |
689 | spin_unlock(&inode->i_lock); | |
690 | ||
cd8ed2a4 | 691 | ret = __writeback_single_inode(inode, wbc); |
1da177e4 | 692 | |
f758eeab | 693 | spin_lock(&wb->list_lock); |
250df6ed | 694 | spin_lock(&inode->i_lock); |
4f8ad655 JK |
695 | /* |
696 | * If inode is clean, remove it from writeback lists. Otherwise don't | |
697 | * touch it. See comment above for explanation. | |
698 | */ | |
0ae45f63 | 699 | if (!(inode->i_state & I_DIRTY_ALL)) |
d6c10f1f | 700 | inode_wb_list_del_locked(inode, wb); |
4f8ad655 | 701 | spin_unlock(&wb->list_lock); |
1c0eeaf5 | 702 | inode_sync_complete(inode); |
4f8ad655 JK |
703 | out: |
704 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
705 | return ret; |
706 | } | |
707 | ||
a88a341a | 708 | static long writeback_chunk_size(struct bdi_writeback *wb, |
1a12d8bd | 709 | struct wb_writeback_work *work) |
d46db3d5 WF |
710 | { |
711 | long pages; | |
712 | ||
713 | /* | |
714 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
715 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
716 | * here avoids calling into writeback_inodes_wb() more than once. | |
717 | * | |
718 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
719 | * | |
720 | * wb_writeback() | |
721 | * writeback_sb_inodes() <== called only once | |
722 | * write_cache_pages() <== called once for each inode | |
723 | * (quickly) tag currently dirty pages | |
724 | * (maybe slowly) sync all tagged pages | |
725 | */ | |
726 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
727 | pages = LONG_MAX; | |
1a12d8bd | 728 | else { |
a88a341a | 729 | pages = min(wb->avg_write_bandwidth / 2, |
1a12d8bd WF |
730 | global_dirty_limit / DIRTY_SCOPE); |
731 | pages = min(pages, work->nr_pages); | |
732 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
733 | MIN_WRITEBACK_PAGES); | |
734 | } | |
d46db3d5 WF |
735 | |
736 | return pages; | |
737 | } | |
738 | ||
f11c9c5c ES |
739 | /* |
740 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 | 741 | * |
d46db3d5 | 742 | * Return the number of pages and/or inodes written. |
f11c9c5c | 743 | */ |
d46db3d5 WF |
744 | static long writeback_sb_inodes(struct super_block *sb, |
745 | struct bdi_writeback *wb, | |
746 | struct wb_writeback_work *work) | |
1da177e4 | 747 | { |
d46db3d5 WF |
748 | struct writeback_control wbc = { |
749 | .sync_mode = work->sync_mode, | |
750 | .tagged_writepages = work->tagged_writepages, | |
751 | .for_kupdate = work->for_kupdate, | |
752 | .for_background = work->for_background, | |
7747bd4b | 753 | .for_sync = work->for_sync, |
d46db3d5 WF |
754 | .range_cyclic = work->range_cyclic, |
755 | .range_start = 0, | |
756 | .range_end = LLONG_MAX, | |
757 | }; | |
758 | unsigned long start_time = jiffies; | |
759 | long write_chunk; | |
760 | long wrote = 0; /* count both pages and inodes */ | |
761 | ||
03ba3782 | 762 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 763 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
764 | |
765 | if (inode->i_sb != sb) { | |
d46db3d5 | 766 | if (work->sb) { |
edadfb10 CH |
767 | /* |
768 | * We only want to write back data for this | |
769 | * superblock, move all inodes not belonging | |
770 | * to it back onto the dirty list. | |
771 | */ | |
f758eeab | 772 | redirty_tail(inode, wb); |
edadfb10 CH |
773 | continue; |
774 | } | |
775 | ||
776 | /* | |
777 | * The inode belongs to a different superblock. | |
778 | * Bounce back to the caller to unpin this and | |
779 | * pin the next superblock. | |
780 | */ | |
d46db3d5 | 781 | break; |
edadfb10 CH |
782 | } |
783 | ||
9843b76a | 784 | /* |
331cbdee WL |
785 | * Don't bother with new inodes or inodes being freed, first |
786 | * kind does not need periodic writeout yet, and for the latter | |
9843b76a CH |
787 | * kind writeout is handled by the freer. |
788 | */ | |
250df6ed | 789 | spin_lock(&inode->i_lock); |
9843b76a | 790 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 791 | spin_unlock(&inode->i_lock); |
fcc5c222 | 792 | redirty_tail(inode, wb); |
7ef0d737 NP |
793 | continue; |
794 | } | |
cc1676d9 JK |
795 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
796 | /* | |
797 | * If this inode is locked for writeback and we are not | |
798 | * doing writeback-for-data-integrity, move it to | |
799 | * b_more_io so that writeback can proceed with the | |
800 | * other inodes on s_io. | |
801 | * | |
802 | * We'll have another go at writing back this inode | |
803 | * when we completed a full scan of b_io. | |
804 | */ | |
805 | spin_unlock(&inode->i_lock); | |
806 | requeue_io(inode, wb); | |
807 | trace_writeback_sb_inodes_requeue(inode); | |
808 | continue; | |
809 | } | |
f0d07b7f JK |
810 | spin_unlock(&wb->list_lock); |
811 | ||
4f8ad655 JK |
812 | /* |
813 | * We already requeued the inode if it had I_SYNC set and we | |
814 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
815 | * WB_SYNC_ALL case. | |
816 | */ | |
169ebd90 JK |
817 | if (inode->i_state & I_SYNC) { |
818 | /* Wait for I_SYNC. This function drops i_lock... */ | |
819 | inode_sleep_on_writeback(inode); | |
820 | /* Inode may be gone, start again */ | |
ead188f9 | 821 | spin_lock(&wb->list_lock); |
169ebd90 JK |
822 | continue; |
823 | } | |
4f8ad655 JK |
824 | inode->i_state |= I_SYNC; |
825 | spin_unlock(&inode->i_lock); | |
169ebd90 | 826 | |
a88a341a | 827 | write_chunk = writeback_chunk_size(wb, work); |
d46db3d5 WF |
828 | wbc.nr_to_write = write_chunk; |
829 | wbc.pages_skipped = 0; | |
250df6ed | 830 | |
169ebd90 JK |
831 | /* |
832 | * We use I_SYNC to pin the inode in memory. While it is set | |
833 | * evict_inode() will wait so the inode cannot be freed. | |
834 | */ | |
cd8ed2a4 | 835 | __writeback_single_inode(inode, &wbc); |
250df6ed | 836 | |
d46db3d5 WF |
837 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
838 | wrote += write_chunk - wbc.nr_to_write; | |
4f8ad655 JK |
839 | spin_lock(&wb->list_lock); |
840 | spin_lock(&inode->i_lock); | |
0ae45f63 | 841 | if (!(inode->i_state & I_DIRTY_ALL)) |
d46db3d5 | 842 | wrote++; |
4f8ad655 JK |
843 | requeue_inode(inode, wb, &wbc); |
844 | inode_sync_complete(inode); | |
0f1b1fd8 | 845 | spin_unlock(&inode->i_lock); |
169ebd90 | 846 | cond_resched_lock(&wb->list_lock); |
d46db3d5 WF |
847 | /* |
848 | * bail out to wb_writeback() often enough to check | |
849 | * background threshold and other termination conditions. | |
850 | */ | |
851 | if (wrote) { | |
852 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
853 | break; | |
854 | if (work->nr_pages <= 0) | |
855 | break; | |
8bc3be27 | 856 | } |
1da177e4 | 857 | } |
d46db3d5 | 858 | return wrote; |
f11c9c5c ES |
859 | } |
860 | ||
d46db3d5 WF |
861 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
862 | struct wb_writeback_work *work) | |
f11c9c5c | 863 | { |
d46db3d5 WF |
864 | unsigned long start_time = jiffies; |
865 | long wrote = 0; | |
38f21977 | 866 | |
f11c9c5c | 867 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 868 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 869 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 870 | |
eb6ef3df | 871 | if (!trylock_super(sb)) { |
0e995816 | 872 | /* |
eb6ef3df | 873 | * trylock_super() may fail consistently due to |
0e995816 WF |
874 | * s_umount being grabbed by someone else. Don't use |
875 | * requeue_io() to avoid busy retrying the inode/sb. | |
876 | */ | |
877 | redirty_tail(inode, wb); | |
edadfb10 | 878 | continue; |
f11c9c5c | 879 | } |
d46db3d5 | 880 | wrote += writeback_sb_inodes(sb, wb, work); |
eb6ef3df | 881 | up_read(&sb->s_umount); |
f11c9c5c | 882 | |
d46db3d5 WF |
883 | /* refer to the same tests at the end of writeback_sb_inodes */ |
884 | if (wrote) { | |
885 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
886 | break; | |
887 | if (work->nr_pages <= 0) | |
888 | break; | |
889 | } | |
f11c9c5c | 890 | } |
66f3b8e2 | 891 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 892 | return wrote; |
66f3b8e2 JA |
893 | } |
894 | ||
7d9f073b | 895 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
0e175a18 | 896 | enum wb_reason reason) |
edadfb10 | 897 | { |
d46db3d5 WF |
898 | struct wb_writeback_work work = { |
899 | .nr_pages = nr_pages, | |
900 | .sync_mode = WB_SYNC_NONE, | |
901 | .range_cyclic = 1, | |
0e175a18 | 902 | .reason = reason, |
d46db3d5 | 903 | }; |
edadfb10 | 904 | |
f758eeab | 905 | spin_lock(&wb->list_lock); |
424b351f | 906 | if (list_empty(&wb->b_io)) |
ad4e38dd | 907 | queue_io(wb, &work); |
d46db3d5 | 908 | __writeback_inodes_wb(wb, &work); |
f758eeab | 909 | spin_unlock(&wb->list_lock); |
edadfb10 | 910 | |
d46db3d5 WF |
911 | return nr_pages - work.nr_pages; |
912 | } | |
03ba3782 | 913 | |
a88a341a | 914 | static bool over_bground_thresh(struct bdi_writeback *wb) |
03ba3782 JA |
915 | { |
916 | unsigned long background_thresh, dirty_thresh; | |
917 | ||
16c4042f | 918 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 | 919 | |
b00949aa WF |
920 | if (global_page_state(NR_FILE_DIRTY) + |
921 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) | |
922 | return true; | |
923 | ||
a88a341a | 924 | if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh)) |
b00949aa WF |
925 | return true; |
926 | ||
927 | return false; | |
03ba3782 JA |
928 | } |
929 | ||
e98be2d5 WF |
930 | /* |
931 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
932 | * only the flusher working on the first wb should do it. | |
933 | */ | |
934 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
935 | unsigned long start_time) | |
936 | { | |
a88a341a | 937 | __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time); |
e98be2d5 WF |
938 | } |
939 | ||
03ba3782 JA |
940 | /* |
941 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 942 | * |
03ba3782 JA |
943 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
944 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
945 | * just walks the superblock inode list, writing back any inodes which are | |
946 | * older than a specific point in time. | |
66f3b8e2 | 947 | * |
03ba3782 JA |
948 | * Try to run once per dirty_writeback_interval. But if a writeback event |
949 | * takes longer than a dirty_writeback_interval interval, then leave a | |
950 | * one-second gap. | |
66f3b8e2 | 951 | * |
03ba3782 JA |
952 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
953 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 954 | */ |
c4a77a6c | 955 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 956 | struct wb_writeback_work *work) |
66f3b8e2 | 957 | { |
e98be2d5 | 958 | unsigned long wb_start = jiffies; |
d46db3d5 | 959 | long nr_pages = work->nr_pages; |
0dc83bd3 | 960 | unsigned long oldest_jif; |
a5989bdc | 961 | struct inode *inode; |
d46db3d5 | 962 | long progress; |
66f3b8e2 | 963 | |
0dc83bd3 JK |
964 | oldest_jif = jiffies; |
965 | work->older_than_this = &oldest_jif; | |
38f21977 | 966 | |
e8dfc305 | 967 | spin_lock(&wb->list_lock); |
03ba3782 JA |
968 | for (;;) { |
969 | /* | |
d3ddec76 | 970 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 971 | */ |
83ba7b07 | 972 | if (work->nr_pages <= 0) |
03ba3782 | 973 | break; |
66f3b8e2 | 974 | |
aa373cf5 JK |
975 | /* |
976 | * Background writeout and kupdate-style writeback may | |
977 | * run forever. Stop them if there is other work to do | |
978 | * so that e.g. sync can proceed. They'll be restarted | |
979 | * after the other works are all done. | |
980 | */ | |
981 | if ((work->for_background || work->for_kupdate) && | |
f0054bb1 | 982 | !list_empty(&wb->work_list)) |
aa373cf5 JK |
983 | break; |
984 | ||
38f21977 | 985 | /* |
d3ddec76 WF |
986 | * For background writeout, stop when we are below the |
987 | * background dirty threshold | |
38f21977 | 988 | */ |
a88a341a | 989 | if (work->for_background && !over_bground_thresh(wb)) |
03ba3782 | 990 | break; |
38f21977 | 991 | |
1bc36b64 JK |
992 | /* |
993 | * Kupdate and background works are special and we want to | |
994 | * include all inodes that need writing. Livelock avoidance is | |
995 | * handled by these works yielding to any other work so we are | |
996 | * safe. | |
997 | */ | |
ba9aa839 | 998 | if (work->for_kupdate) { |
0dc83bd3 | 999 | oldest_jif = jiffies - |
ba9aa839 | 1000 | msecs_to_jiffies(dirty_expire_interval * 10); |
1bc36b64 | 1001 | } else if (work->for_background) |
0dc83bd3 | 1002 | oldest_jif = jiffies; |
028c2dd1 | 1003 | |
d46db3d5 | 1004 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 1005 | if (list_empty(&wb->b_io)) |
ad4e38dd | 1006 | queue_io(wb, work); |
83ba7b07 | 1007 | if (work->sb) |
d46db3d5 | 1008 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 1009 | else |
d46db3d5 WF |
1010 | progress = __writeback_inodes_wb(wb, work); |
1011 | trace_writeback_written(wb->bdi, work); | |
028c2dd1 | 1012 | |
e98be2d5 | 1013 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
1014 | |
1015 | /* | |
e6fb6da2 WF |
1016 | * Did we write something? Try for more |
1017 | * | |
1018 | * Dirty inodes are moved to b_io for writeback in batches. | |
1019 | * The completion of the current batch does not necessarily | |
1020 | * mean the overall work is done. So we keep looping as long | |
1021 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 1022 | */ |
d46db3d5 | 1023 | if (progress) |
71fd05a8 JA |
1024 | continue; |
1025 | /* | |
e6fb6da2 | 1026 | * No more inodes for IO, bail |
71fd05a8 | 1027 | */ |
b7a2441f | 1028 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 1029 | break; |
71fd05a8 JA |
1030 | /* |
1031 | * Nothing written. Wait for some inode to | |
1032 | * become available for writeback. Otherwise | |
1033 | * we'll just busyloop. | |
1034 | */ | |
71fd05a8 | 1035 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 1036 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 1037 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 1038 | spin_lock(&inode->i_lock); |
f0d07b7f | 1039 | spin_unlock(&wb->list_lock); |
169ebd90 JK |
1040 | /* This function drops i_lock... */ |
1041 | inode_sleep_on_writeback(inode); | |
f0d07b7f | 1042 | spin_lock(&wb->list_lock); |
03ba3782 JA |
1043 | } |
1044 | } | |
e8dfc305 | 1045 | spin_unlock(&wb->list_lock); |
03ba3782 | 1046 | |
d46db3d5 | 1047 | return nr_pages - work->nr_pages; |
03ba3782 JA |
1048 | } |
1049 | ||
1050 | /* | |
83ba7b07 | 1051 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 1052 | */ |
f0054bb1 | 1053 | static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) |
03ba3782 | 1054 | { |
83ba7b07 | 1055 | struct wb_writeback_work *work = NULL; |
03ba3782 | 1056 | |
f0054bb1 TH |
1057 | spin_lock_bh(&wb->work_lock); |
1058 | if (!list_empty(&wb->work_list)) { | |
1059 | work = list_entry(wb->work_list.next, | |
83ba7b07 CH |
1060 | struct wb_writeback_work, list); |
1061 | list_del_init(&work->list); | |
03ba3782 | 1062 | } |
f0054bb1 | 1063 | spin_unlock_bh(&wb->work_lock); |
83ba7b07 | 1064 | return work; |
03ba3782 JA |
1065 | } |
1066 | ||
cdf01dd5 LT |
1067 | /* |
1068 | * Add in the number of potentially dirty inodes, because each inode | |
1069 | * write can dirty pagecache in the underlying blockdev. | |
1070 | */ | |
1071 | static unsigned long get_nr_dirty_pages(void) | |
1072 | { | |
1073 | return global_page_state(NR_FILE_DIRTY) + | |
1074 | global_page_state(NR_UNSTABLE_NFS) + | |
1075 | get_nr_dirty_inodes(); | |
1076 | } | |
1077 | ||
6585027a JK |
1078 | static long wb_check_background_flush(struct bdi_writeback *wb) |
1079 | { | |
a88a341a | 1080 | if (over_bground_thresh(wb)) { |
6585027a JK |
1081 | |
1082 | struct wb_writeback_work work = { | |
1083 | .nr_pages = LONG_MAX, | |
1084 | .sync_mode = WB_SYNC_NONE, | |
1085 | .for_background = 1, | |
1086 | .range_cyclic = 1, | |
0e175a18 | 1087 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
1088 | }; |
1089 | ||
1090 | return wb_writeback(wb, &work); | |
1091 | } | |
1092 | ||
1093 | return 0; | |
1094 | } | |
1095 | ||
03ba3782 JA |
1096 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
1097 | { | |
1098 | unsigned long expired; | |
1099 | long nr_pages; | |
1100 | ||
69b62d01 JA |
1101 | /* |
1102 | * When set to zero, disable periodic writeback | |
1103 | */ | |
1104 | if (!dirty_writeback_interval) | |
1105 | return 0; | |
1106 | ||
03ba3782 JA |
1107 | expired = wb->last_old_flush + |
1108 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
1109 | if (time_before(jiffies, expired)) | |
1110 | return 0; | |
1111 | ||
1112 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 1113 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 1114 | |
c4a77a6c | 1115 | if (nr_pages) { |
83ba7b07 | 1116 | struct wb_writeback_work work = { |
c4a77a6c JA |
1117 | .nr_pages = nr_pages, |
1118 | .sync_mode = WB_SYNC_NONE, | |
1119 | .for_kupdate = 1, | |
1120 | .range_cyclic = 1, | |
0e175a18 | 1121 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
1122 | }; |
1123 | ||
83ba7b07 | 1124 | return wb_writeback(wb, &work); |
c4a77a6c | 1125 | } |
03ba3782 JA |
1126 | |
1127 | return 0; | |
1128 | } | |
1129 | ||
1130 | /* | |
1131 | * Retrieve work items and do the writeback they describe | |
1132 | */ | |
25d130ba | 1133 | static long wb_do_writeback(struct bdi_writeback *wb) |
03ba3782 | 1134 | { |
83ba7b07 | 1135 | struct wb_writeback_work *work; |
c4a77a6c | 1136 | long wrote = 0; |
03ba3782 | 1137 | |
4452226e | 1138 | set_bit(WB_writeback_running, &wb->state); |
f0054bb1 | 1139 | while ((work = get_next_work_item(wb)) != NULL) { |
03ba3782 | 1140 | |
f0054bb1 | 1141 | trace_writeback_exec(wb->bdi, work); |
455b2864 | 1142 | |
83ba7b07 | 1143 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
1144 | |
1145 | /* | |
83ba7b07 CH |
1146 | * Notify the caller of completion if this is a synchronous |
1147 | * work item, otherwise just free it. | |
03ba3782 | 1148 | */ |
83ba7b07 CH |
1149 | if (work->done) |
1150 | complete(work->done); | |
1151 | else | |
1152 | kfree(work); | |
03ba3782 JA |
1153 | } |
1154 | ||
1155 | /* | |
1156 | * Check for periodic writeback, kupdated() style | |
1157 | */ | |
1158 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 1159 | wrote += wb_check_background_flush(wb); |
4452226e | 1160 | clear_bit(WB_writeback_running, &wb->state); |
03ba3782 JA |
1161 | |
1162 | return wrote; | |
1163 | } | |
1164 | ||
1165 | /* | |
1166 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
839a8e86 | 1167 | * reschedules periodically and does kupdated style flushing. |
03ba3782 | 1168 | */ |
f0054bb1 | 1169 | void wb_workfn(struct work_struct *work) |
03ba3782 | 1170 | { |
839a8e86 TH |
1171 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
1172 | struct bdi_writeback, dwork); | |
03ba3782 JA |
1173 | long pages_written; |
1174 | ||
f0054bb1 | 1175 | set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); |
766f9164 | 1176 | current->flags |= PF_SWAPWRITE; |
455b2864 | 1177 | |
839a8e86 | 1178 | if (likely(!current_is_workqueue_rescuer() || |
4452226e | 1179 | !test_bit(WB_registered, &wb->state))) { |
6467716a | 1180 | /* |
f0054bb1 | 1181 | * The normal path. Keep writing back @wb until its |
839a8e86 | 1182 | * work_list is empty. Note that this path is also taken |
f0054bb1 | 1183 | * if @wb is shutting down even when we're running off the |
839a8e86 | 1184 | * rescuer as work_list needs to be drained. |
6467716a | 1185 | */ |
839a8e86 | 1186 | do { |
25d130ba | 1187 | pages_written = wb_do_writeback(wb); |
839a8e86 | 1188 | trace_writeback_pages_written(pages_written); |
f0054bb1 | 1189 | } while (!list_empty(&wb->work_list)); |
839a8e86 TH |
1190 | } else { |
1191 | /* | |
1192 | * bdi_wq can't get enough workers and we're running off | |
1193 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
1194 | * enough for efficient IO. | |
1195 | */ | |
f0054bb1 | 1196 | pages_written = writeback_inodes_wb(wb, 1024, |
839a8e86 | 1197 | WB_REASON_FORKER_THREAD); |
455b2864 | 1198 | trace_writeback_pages_written(pages_written); |
03ba3782 JA |
1199 | } |
1200 | ||
f0054bb1 | 1201 | if (!list_empty(&wb->work_list)) |
6ca738d6 DB |
1202 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
1203 | else if (wb_has_dirty_io(wb) && dirty_writeback_interval) | |
f0054bb1 | 1204 | wb_wakeup_delayed(wb); |
455b2864 | 1205 | |
839a8e86 | 1206 | current->flags &= ~PF_SWAPWRITE; |
03ba3782 JA |
1207 | } |
1208 | ||
1209 | /* | |
b8c2f347 CH |
1210 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
1211 | * the whole world. | |
03ba3782 | 1212 | */ |
0e175a18 | 1213 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
03ba3782 | 1214 | { |
b8c2f347 | 1215 | struct backing_dev_info *bdi; |
03ba3782 | 1216 | |
47df3dde JK |
1217 | if (!nr_pages) |
1218 | nr_pages = get_nr_dirty_pages(); | |
03ba3782 | 1219 | |
b8c2f347 | 1220 | rcu_read_lock(); |
e7972912 | 1221 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) |
f0054bb1 | 1222 | __wb_start_writeback(&bdi->wb, nr_pages, false, reason); |
cfc4ba53 | 1223 | rcu_read_unlock(); |
1da177e4 LT |
1224 | } |
1225 | ||
a2f48706 TT |
1226 | /* |
1227 | * Wake up bdi's periodically to make sure dirtytime inodes gets | |
1228 | * written back periodically. We deliberately do *not* check the | |
1229 | * b_dirtytime list in wb_has_dirty_io(), since this would cause the | |
1230 | * kernel to be constantly waking up once there are any dirtytime | |
1231 | * inodes on the system. So instead we define a separate delayed work | |
1232 | * function which gets called much more rarely. (By default, only | |
1233 | * once every 12 hours.) | |
1234 | * | |
1235 | * If there is any other write activity going on in the file system, | |
1236 | * this function won't be necessary. But if the only thing that has | |
1237 | * happened on the file system is a dirtytime inode caused by an atime | |
1238 | * update, we need this infrastructure below to make sure that inode | |
1239 | * eventually gets pushed out to disk. | |
1240 | */ | |
1241 | static void wakeup_dirtytime_writeback(struct work_struct *w); | |
1242 | static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); | |
1243 | ||
1244 | static void wakeup_dirtytime_writeback(struct work_struct *w) | |
1245 | { | |
1246 | struct backing_dev_info *bdi; | |
1247 | ||
1248 | rcu_read_lock(); | |
1249 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { | |
1250 | if (list_empty(&bdi->wb.b_dirty_time)) | |
1251 | continue; | |
f0054bb1 | 1252 | wb_wakeup(&bdi->wb); |
a2f48706 TT |
1253 | } |
1254 | rcu_read_unlock(); | |
1255 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
1256 | } | |
1257 | ||
1258 | static int __init start_dirtytime_writeback(void) | |
1259 | { | |
1260 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
1261 | return 0; | |
1262 | } | |
1263 | __initcall(start_dirtytime_writeback); | |
1264 | ||
1efff914 TT |
1265 | int dirtytime_interval_handler(struct ctl_table *table, int write, |
1266 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1267 | { | |
1268 | int ret; | |
1269 | ||
1270 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
1271 | if (ret == 0 && write) | |
1272 | mod_delayed_work(system_wq, &dirtytime_work, 0); | |
1273 | return ret; | |
1274 | } | |
1275 | ||
03ba3782 JA |
1276 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1277 | { | |
1278 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1279 | struct dentry *dentry; | |
1280 | const char *name = "?"; | |
1281 | ||
1282 | dentry = d_find_alias(inode); | |
1283 | if (dentry) { | |
1284 | spin_lock(&dentry->d_lock); | |
1285 | name = (const char *) dentry->d_name.name; | |
1286 | } | |
1287 | printk(KERN_DEBUG | |
1288 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1289 | current->comm, task_pid_nr(current), inode->i_ino, | |
1290 | name, inode->i_sb->s_id); | |
1291 | if (dentry) { | |
1292 | spin_unlock(&dentry->d_lock); | |
1293 | dput(dentry); | |
1294 | } | |
1295 | } | |
1296 | } | |
1297 | ||
1298 | /** | |
1299 | * __mark_inode_dirty - internal function | |
1300 | * @inode: inode to mark | |
1301 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1302 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1303 | * mark_inode_dirty_sync. | |
1da177e4 | 1304 | * |
03ba3782 JA |
1305 | * Put the inode on the super block's dirty list. |
1306 | * | |
1307 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1308 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1309 | * If it was not hashed, it will never be added to the dirty list | |
1310 | * even if it is later hashed, as it will have been marked dirty already. | |
1311 | * | |
1312 | * In short, make sure you hash any inodes _before_ you start marking | |
1313 | * them dirty. | |
1da177e4 | 1314 | * |
03ba3782 JA |
1315 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1316 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1317 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1318 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1319 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1320 | * blockdev inode. | |
1da177e4 | 1321 | */ |
0ae45f63 | 1322 | #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) |
03ba3782 | 1323 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1324 | { |
03ba3782 | 1325 | struct super_block *sb = inode->i_sb; |
253c34e9 | 1326 | struct backing_dev_info *bdi = NULL; |
0ae45f63 TT |
1327 | int dirtytime; |
1328 | ||
1329 | trace_writeback_mark_inode_dirty(inode, flags); | |
1da177e4 | 1330 | |
03ba3782 JA |
1331 | /* |
1332 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1333 | * dirty the inode itself | |
1334 | */ | |
0ae45f63 | 1335 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { |
9fb0a7da TH |
1336 | trace_writeback_dirty_inode_start(inode, flags); |
1337 | ||
03ba3782 | 1338 | if (sb->s_op->dirty_inode) |
aa385729 | 1339 | sb->s_op->dirty_inode(inode, flags); |
9fb0a7da TH |
1340 | |
1341 | trace_writeback_dirty_inode(inode, flags); | |
03ba3782 | 1342 | } |
0ae45f63 TT |
1343 | if (flags & I_DIRTY_INODE) |
1344 | flags &= ~I_DIRTY_TIME; | |
1345 | dirtytime = flags & I_DIRTY_TIME; | |
03ba3782 JA |
1346 | |
1347 | /* | |
9c6ac78e TH |
1348 | * Paired with smp_mb() in __writeback_single_inode() for the |
1349 | * following lockless i_state test. See there for details. | |
03ba3782 JA |
1350 | */ |
1351 | smp_mb(); | |
1352 | ||
0ae45f63 TT |
1353 | if (((inode->i_state & flags) == flags) || |
1354 | (dirtytime && (inode->i_state & I_DIRTY_INODE))) | |
03ba3782 JA |
1355 | return; |
1356 | ||
1357 | if (unlikely(block_dump)) | |
1358 | block_dump___mark_inode_dirty(inode); | |
1359 | ||
250df6ed | 1360 | spin_lock(&inode->i_lock); |
0ae45f63 TT |
1361 | if (dirtytime && (inode->i_state & I_DIRTY_INODE)) |
1362 | goto out_unlock_inode; | |
03ba3782 JA |
1363 | if ((inode->i_state & flags) != flags) { |
1364 | const int was_dirty = inode->i_state & I_DIRTY; | |
1365 | ||
52ebea74 TH |
1366 | inode_attach_wb(inode, NULL); |
1367 | ||
0ae45f63 TT |
1368 | if (flags & I_DIRTY_INODE) |
1369 | inode->i_state &= ~I_DIRTY_TIME; | |
03ba3782 JA |
1370 | inode->i_state |= flags; |
1371 | ||
1372 | /* | |
1373 | * If the inode is being synced, just update its dirty state. | |
1374 | * The unlocker will place the inode on the appropriate | |
1375 | * superblock list, based upon its state. | |
1376 | */ | |
1377 | if (inode->i_state & I_SYNC) | |
250df6ed | 1378 | goto out_unlock_inode; |
03ba3782 JA |
1379 | |
1380 | /* | |
1381 | * Only add valid (hashed) inodes to the superblock's | |
1382 | * dirty list. Add blockdev inodes as well. | |
1383 | */ | |
1384 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1385 | if (inode_unhashed(inode)) |
250df6ed | 1386 | goto out_unlock_inode; |
03ba3782 | 1387 | } |
a4ffdde6 | 1388 | if (inode->i_state & I_FREEING) |
250df6ed | 1389 | goto out_unlock_inode; |
03ba3782 JA |
1390 | |
1391 | /* | |
1392 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1393 | * reposition it (that would break b_dirty time-ordering). | |
1394 | */ | |
1395 | if (!was_dirty) { | |
d6c10f1f | 1396 | struct list_head *dirty_list; |
a66979ab | 1397 | bool wakeup_bdi = false; |
253c34e9 AB |
1398 | bdi = inode_to_bdi(inode); |
1399 | ||
146d7009 JB |
1400 | spin_unlock(&inode->i_lock); |
1401 | spin_lock(&bdi->wb.list_lock); | |
253c34e9 | 1402 | |
d6c10f1f TH |
1403 | WARN(bdi_cap_writeback_dirty(bdi) && |
1404 | !test_bit(WB_registered, &bdi->wb.state), | |
1405 | "bdi-%s not registered\n", bdi->name); | |
03ba3782 JA |
1406 | |
1407 | inode->dirtied_when = jiffies; | |
a2f48706 TT |
1408 | if (dirtytime) |
1409 | inode->dirtied_time_when = jiffies; | |
d6c10f1f | 1410 | |
a2f48706 | 1411 | if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) |
d6c10f1f | 1412 | dirty_list = &bdi->wb.b_dirty; |
a2f48706 | 1413 | else |
d6c10f1f TH |
1414 | dirty_list = &bdi->wb.b_dirty_time; |
1415 | ||
1416 | wakeup_bdi = inode_wb_list_move_locked(inode, &bdi->wb, | |
1417 | dirty_list); | |
1418 | ||
f758eeab | 1419 | spin_unlock(&bdi->wb.list_lock); |
0ae45f63 | 1420 | trace_writeback_dirty_inode_enqueue(inode); |
a66979ab | 1421 | |
d6c10f1f TH |
1422 | /* |
1423 | * If this is the first dirty inode for this bdi, | |
1424 | * we have to wake-up the corresponding bdi thread | |
1425 | * to make sure background write-back happens | |
1426 | * later. | |
1427 | */ | |
1428 | if (bdi_cap_writeback_dirty(bdi) && wakeup_bdi) | |
f0054bb1 | 1429 | wb_wakeup_delayed(&bdi->wb); |
a66979ab | 1430 | return; |
1da177e4 | 1431 | } |
1da177e4 | 1432 | } |
250df6ed DC |
1433 | out_unlock_inode: |
1434 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1435 | |
03ba3782 JA |
1436 | } |
1437 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1438 | ||
b6e51316 | 1439 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1440 | { |
1441 | struct inode *inode, *old_inode = NULL; | |
1442 | ||
1443 | /* | |
1444 | * We need to be protected against the filesystem going from | |
1445 | * r/o to r/w or vice versa. | |
1446 | */ | |
b6e51316 | 1447 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1448 | |
55fa6091 | 1449 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1450 | |
1451 | /* | |
1452 | * Data integrity sync. Must wait for all pages under writeback, | |
1453 | * because there may have been pages dirtied before our sync | |
1454 | * call, but which had writeout started before we write it out. | |
1455 | * In which case, the inode may not be on the dirty list, but | |
1456 | * we still have to wait for that writeout. | |
1457 | */ | |
b6e51316 | 1458 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1459 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1460 | |
250df6ed DC |
1461 | spin_lock(&inode->i_lock); |
1462 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1463 | (mapping->nrpages == 0)) { | |
1464 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1465 | continue; |
250df6ed | 1466 | } |
03ba3782 | 1467 | __iget(inode); |
250df6ed | 1468 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1469 | spin_unlock(&inode_sb_list_lock); |
1470 | ||
03ba3782 | 1471 | /* |
55fa6091 DC |
1472 | * We hold a reference to 'inode' so it couldn't have been |
1473 | * removed from s_inodes list while we dropped the | |
1474 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1475 | * be holding the last reference and we cannot iput it under | |
1476 | * inode_sb_list_lock. So we keep the reference and iput it | |
1477 | * later. | |
03ba3782 JA |
1478 | */ |
1479 | iput(old_inode); | |
1480 | old_inode = inode; | |
1481 | ||
1482 | filemap_fdatawait(mapping); | |
1483 | ||
1484 | cond_resched(); | |
1485 | ||
55fa6091 | 1486 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1487 | } |
55fa6091 | 1488 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1489 | iput(old_inode); |
1da177e4 LT |
1490 | } |
1491 | ||
d8a8559c | 1492 | /** |
3259f8be | 1493 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1494 | * @sb: the superblock |
3259f8be | 1495 | * @nr: the number of pages to write |
786228ab | 1496 | * @reason: reason why some writeback work initiated |
1da177e4 | 1497 | * |
d8a8559c JA |
1498 | * Start writeback on some inodes on this super_block. No guarantees are made |
1499 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1500 | * for IO completion of submitted IO. |
1da177e4 | 1501 | */ |
0e175a18 CW |
1502 | void writeback_inodes_sb_nr(struct super_block *sb, |
1503 | unsigned long nr, | |
1504 | enum wb_reason reason) | |
1da177e4 | 1505 | { |
83ba7b07 CH |
1506 | DECLARE_COMPLETION_ONSTACK(done); |
1507 | struct wb_writeback_work work = { | |
6e6938b6 WF |
1508 | .sb = sb, |
1509 | .sync_mode = WB_SYNC_NONE, | |
1510 | .tagged_writepages = 1, | |
1511 | .done = &done, | |
1512 | .nr_pages = nr, | |
0e175a18 | 1513 | .reason = reason, |
3c4d7165 | 1514 | }; |
e7972912 | 1515 | struct backing_dev_info *bdi = sb->s_bdi; |
d8a8559c | 1516 | |
e7972912 | 1517 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
6eedc701 | 1518 | return; |
cf37e972 | 1519 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
e7972912 | 1520 | wb_queue_work(&bdi->wb, &work); |
83ba7b07 | 1521 | wait_for_completion(&done); |
e913fc82 | 1522 | } |
3259f8be CM |
1523 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1524 | ||
1525 | /** | |
1526 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1527 | * @sb: the superblock | |
786228ab | 1528 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1529 | * |
1530 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1531 | * on how many (if any) will be written, and this function does not wait | |
1532 | * for IO completion of submitted IO. | |
1533 | */ | |
0e175a18 | 1534 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1535 | { |
0e175a18 | 1536 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1537 | } |
0e3c9a22 | 1538 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1539 | |
17bd55d0 | 1540 | /** |
10ee27a0 | 1541 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway |
17bd55d0 | 1542 | * @sb: the superblock |
10ee27a0 MX |
1543 | * @nr: the number of pages to write |
1544 | * @reason: the reason of writeback | |
17bd55d0 | 1545 | * |
10ee27a0 | 1546 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. |
17bd55d0 ES |
1547 | * Returns 1 if writeback was started, 0 if not. |
1548 | */ | |
10ee27a0 MX |
1549 | int try_to_writeback_inodes_sb_nr(struct super_block *sb, |
1550 | unsigned long nr, | |
1551 | enum wb_reason reason) | |
17bd55d0 | 1552 | { |
10ee27a0 | 1553 | if (writeback_in_progress(sb->s_bdi)) |
17bd55d0 | 1554 | return 1; |
10ee27a0 MX |
1555 | |
1556 | if (!down_read_trylock(&sb->s_umount)) | |
17bd55d0 | 1557 | return 0; |
10ee27a0 MX |
1558 | |
1559 | writeback_inodes_sb_nr(sb, nr, reason); | |
1560 | up_read(&sb->s_umount); | |
1561 | return 1; | |
17bd55d0 | 1562 | } |
10ee27a0 | 1563 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); |
17bd55d0 | 1564 | |
3259f8be | 1565 | /** |
10ee27a0 | 1566 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
3259f8be | 1567 | * @sb: the superblock |
786228ab | 1568 | * @reason: reason why some writeback work was initiated |
3259f8be | 1569 | * |
10ee27a0 | 1570 | * Implement by try_to_writeback_inodes_sb_nr() |
3259f8be CM |
1571 | * Returns 1 if writeback was started, 0 if not. |
1572 | */ | |
10ee27a0 | 1573 | int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1574 | { |
10ee27a0 | 1575 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1576 | } |
10ee27a0 | 1577 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
3259f8be | 1578 | |
d8a8559c JA |
1579 | /** |
1580 | * sync_inodes_sb - sync sb inode pages | |
0dc83bd3 | 1581 | * @sb: the superblock |
d8a8559c JA |
1582 | * |
1583 | * This function writes and waits on any dirty inode belonging to this | |
0dc83bd3 | 1584 | * super_block. |
d8a8559c | 1585 | */ |
0dc83bd3 | 1586 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1587 | { |
83ba7b07 CH |
1588 | DECLARE_COMPLETION_ONSTACK(done); |
1589 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1590 | .sb = sb, |
1591 | .sync_mode = WB_SYNC_ALL, | |
1592 | .nr_pages = LONG_MAX, | |
1593 | .range_cyclic = 0, | |
83ba7b07 | 1594 | .done = &done, |
0e175a18 | 1595 | .reason = WB_REASON_SYNC, |
7747bd4b | 1596 | .for_sync = 1, |
3c4d7165 | 1597 | }; |
e7972912 | 1598 | struct backing_dev_info *bdi = sb->s_bdi; |
3c4d7165 | 1599 | |
6eedc701 | 1600 | /* Nothing to do? */ |
e7972912 | 1601 | if (!bdi_has_dirty_io(bdi) || bdi == &noop_backing_dev_info) |
6eedc701 | 1602 | return; |
cf37e972 CH |
1603 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1604 | ||
e7972912 | 1605 | wb_queue_work(&bdi->wb, &work); |
83ba7b07 CH |
1606 | wait_for_completion(&done); |
1607 | ||
b6e51316 | 1608 | wait_sb_inodes(sb); |
1da177e4 | 1609 | } |
d8a8559c | 1610 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1611 | |
1da177e4 | 1612 | /** |
7f04c26d AA |
1613 | * write_inode_now - write an inode to disk |
1614 | * @inode: inode to write to disk | |
1615 | * @sync: whether the write should be synchronous or not | |
1616 | * | |
1617 | * This function commits an inode to disk immediately if it is dirty. This is | |
1618 | * primarily needed by knfsd. | |
1da177e4 | 1619 | * |
7f04c26d | 1620 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1621 | */ |
1da177e4 LT |
1622 | int write_inode_now(struct inode *inode, int sync) |
1623 | { | |
f758eeab | 1624 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1625 | struct writeback_control wbc = { |
1626 | .nr_to_write = LONG_MAX, | |
18914b18 | 1627 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1628 | .range_start = 0, |
1629 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1630 | }; |
1631 | ||
1632 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1633 | wbc.nr_to_write = 0; |
1da177e4 LT |
1634 | |
1635 | might_sleep(); | |
4f8ad655 | 1636 | return writeback_single_inode(inode, wb, &wbc); |
1da177e4 LT |
1637 | } |
1638 | EXPORT_SYMBOL(write_inode_now); | |
1639 | ||
1640 | /** | |
1641 | * sync_inode - write an inode and its pages to disk. | |
1642 | * @inode: the inode to sync | |
1643 | * @wbc: controls the writeback mode | |
1644 | * | |
1645 | * sync_inode() will write an inode and its pages to disk. It will also | |
1646 | * correctly update the inode on its superblock's dirty inode lists and will | |
1647 | * update inode->i_state. | |
1648 | * | |
1649 | * The caller must have a ref on the inode. | |
1650 | */ | |
1651 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1652 | { | |
4f8ad655 | 1653 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); |
1da177e4 LT |
1654 | } |
1655 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1656 | |
1657 | /** | |
c691b9d9 | 1658 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1659 | * @inode: the inode to sync |
1660 | * @wait: wait for I/O to complete. | |
1661 | * | |
c691b9d9 | 1662 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1663 | * |
1664 | * Note: only writes the actual inode, no associated data or other metadata. | |
1665 | */ | |
1666 | int sync_inode_metadata(struct inode *inode, int wait) | |
1667 | { | |
1668 | struct writeback_control wbc = { | |
1669 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1670 | .nr_to_write = 0, /* metadata-only */ | |
1671 | }; | |
1672 | ||
1673 | return sync_inode(inode, &wbc); | |
1674 | } | |
1675 | EXPORT_SYMBOL(sync_inode_metadata); |