]>
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 | ||
f0054bb1 | 96 | static void wb_wakeup(struct bdi_writeback *wb) |
5acda9d1 | 97 | { |
f0054bb1 TH |
98 | spin_lock_bh(&wb->work_lock); |
99 | if (test_bit(WB_registered, &wb->state)) | |
100 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
101 | spin_unlock_bh(&wb->work_lock); | |
5acda9d1 JK |
102 | } |
103 | ||
f0054bb1 TH |
104 | static void wb_queue_work(struct bdi_writeback *wb, |
105 | struct wb_writeback_work *work) | |
6585027a | 106 | { |
f0054bb1 | 107 | trace_writeback_queue(wb->bdi, work); |
6585027a | 108 | |
f0054bb1 TH |
109 | spin_lock_bh(&wb->work_lock); |
110 | if (!test_bit(WB_registered, &wb->state)) { | |
5acda9d1 JK |
111 | if (work->done) |
112 | complete(work->done); | |
113 | goto out_unlock; | |
114 | } | |
f0054bb1 TH |
115 | list_add_tail(&work->list, &wb->work_list); |
116 | mod_delayed_work(bdi_wq, &wb->dwork, 0); | |
5acda9d1 | 117 | out_unlock: |
f0054bb1 | 118 | spin_unlock_bh(&wb->work_lock); |
1da177e4 LT |
119 | } |
120 | ||
f0054bb1 TH |
121 | static void __wb_start_writeback(struct bdi_writeback *wb, long nr_pages, |
122 | bool range_cyclic, enum wb_reason reason) | |
1da177e4 | 123 | { |
83ba7b07 | 124 | struct wb_writeback_work *work; |
03ba3782 | 125 | |
bcddc3f0 JA |
126 | /* |
127 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
128 | * wakeup the thread for old dirty data writeback | |
129 | */ | |
83ba7b07 CH |
130 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
131 | if (!work) { | |
f0054bb1 TH |
132 | trace_writeback_nowork(wb->bdi); |
133 | wb_wakeup(wb); | |
83ba7b07 | 134 | return; |
bcddc3f0 | 135 | } |
03ba3782 | 136 | |
83ba7b07 CH |
137 | work->sync_mode = WB_SYNC_NONE; |
138 | work->nr_pages = nr_pages; | |
139 | work->range_cyclic = range_cyclic; | |
0e175a18 | 140 | work->reason = reason; |
03ba3782 | 141 | |
f0054bb1 | 142 | wb_queue_work(wb, work); |
b6e51316 JA |
143 | } |
144 | ||
145 | /** | |
146 | * bdi_start_writeback - start writeback | |
147 | * @bdi: the backing device to write from | |
148 | * @nr_pages: the number of pages to write | |
786228ab | 149 | * @reason: reason why some writeback work was initiated |
b6e51316 JA |
150 | * |
151 | * Description: | |
152 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
25985edc | 153 | * started when this function returns, we make no guarantees on |
0e3c9a22 | 154 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
155 | * |
156 | */ | |
0e175a18 CW |
157 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, |
158 | enum wb_reason reason) | |
b6e51316 | 159 | { |
f0054bb1 | 160 | __wb_start_writeback(&bdi->wb, nr_pages, true, reason); |
c5444198 | 161 | } |
d3ddec76 | 162 | |
c5444198 CH |
163 | /** |
164 | * bdi_start_background_writeback - start background writeback | |
165 | * @bdi: the backing device to write from | |
166 | * | |
167 | * Description: | |
6585027a JK |
168 | * This makes sure WB_SYNC_NONE background writeback happens. When |
169 | * this function returns, it is only guaranteed that for given BDI | |
170 | * some IO is happening if we are over background dirty threshold. | |
171 | * Caller need not hold sb s_umount semaphore. | |
c5444198 CH |
172 | */ |
173 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
174 | { | |
6585027a JK |
175 | /* |
176 | * We just wake up the flusher thread. It will perform background | |
177 | * writeback as soon as there is no other work to do. | |
178 | */ | |
71927e84 | 179 | trace_writeback_wake_background(bdi); |
f0054bb1 | 180 | wb_wakeup(&bdi->wb); |
1da177e4 LT |
181 | } |
182 | ||
a66979ab DC |
183 | /* |
184 | * Remove the inode from the writeback list it is on. | |
185 | */ | |
186 | void inode_wb_list_del(struct inode *inode) | |
187 | { | |
f758eeab CH |
188 | struct backing_dev_info *bdi = inode_to_bdi(inode); |
189 | ||
190 | spin_lock(&bdi->wb.list_lock); | |
a66979ab | 191 | list_del_init(&inode->i_wb_list); |
f758eeab | 192 | spin_unlock(&bdi->wb.list_lock); |
a66979ab DC |
193 | } |
194 | ||
6610a0bc AM |
195 | /* |
196 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
197 | * furthest end of its superblock's dirty-inode list. | |
198 | * | |
199 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 200 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
201 | * the case then the inode must have been redirtied while it was being written |
202 | * out and we don't reset its dirtied_when. | |
203 | */ | |
f758eeab | 204 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) |
6610a0bc | 205 | { |
f758eeab | 206 | assert_spin_locked(&wb->list_lock); |
03ba3782 | 207 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 208 | struct inode *tail; |
6610a0bc | 209 | |
7ccf19a8 | 210 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 211 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
212 | inode->dirtied_when = jiffies; |
213 | } | |
7ccf19a8 | 214 | list_move(&inode->i_wb_list, &wb->b_dirty); |
6610a0bc AM |
215 | } |
216 | ||
c986d1e2 | 217 | /* |
66f3b8e2 | 218 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 219 | */ |
f758eeab | 220 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) |
c986d1e2 | 221 | { |
f758eeab | 222 | assert_spin_locked(&wb->list_lock); |
7ccf19a8 | 223 | list_move(&inode->i_wb_list, &wb->b_more_io); |
c986d1e2 AM |
224 | } |
225 | ||
1c0eeaf5 JE |
226 | static void inode_sync_complete(struct inode *inode) |
227 | { | |
365b94ae | 228 | inode->i_state &= ~I_SYNC; |
4eff96dd JK |
229 | /* If inode is clean an unused, put it into LRU now... */ |
230 | inode_add_lru(inode); | |
365b94ae | 231 | /* Waiters must see I_SYNC cleared before being woken up */ |
1c0eeaf5 JE |
232 | smp_mb(); |
233 | wake_up_bit(&inode->i_state, __I_SYNC); | |
234 | } | |
235 | ||
d2caa3c5 JL |
236 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
237 | { | |
238 | bool ret = time_after(inode->dirtied_when, t); | |
239 | #ifndef CONFIG_64BIT | |
240 | /* | |
241 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
242 | * It _appears_ to be in the future, but is actually in distant past. | |
243 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 244 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
245 | */ |
246 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
247 | #endif | |
248 | return ret; | |
249 | } | |
250 | ||
0ae45f63 TT |
251 | #define EXPIRE_DIRTY_ATIME 0x0001 |
252 | ||
2c136579 | 253 | /* |
0e2f2b23 | 254 | * Move expired (dirtied before work->older_than_this) dirty inodes from |
697e6fed | 255 | * @delaying_queue to @dispatch_queue. |
2c136579 | 256 | */ |
e84d0a4f | 257 | static int move_expired_inodes(struct list_head *delaying_queue, |
2c136579 | 258 | struct list_head *dispatch_queue, |
0ae45f63 | 259 | int flags, |
ad4e38dd | 260 | struct wb_writeback_work *work) |
2c136579 | 261 | { |
0ae45f63 TT |
262 | unsigned long *older_than_this = NULL; |
263 | unsigned long expire_time; | |
5c03449d SL |
264 | LIST_HEAD(tmp); |
265 | struct list_head *pos, *node; | |
cf137307 | 266 | struct super_block *sb = NULL; |
5c03449d | 267 | struct inode *inode; |
cf137307 | 268 | int do_sb_sort = 0; |
e84d0a4f | 269 | int moved = 0; |
5c03449d | 270 | |
0ae45f63 TT |
271 | if ((flags & EXPIRE_DIRTY_ATIME) == 0) |
272 | older_than_this = work->older_than_this; | |
a2f48706 TT |
273 | else if (!work->for_sync) { |
274 | expire_time = jiffies - (dirtytime_expire_interval * HZ); | |
0ae45f63 TT |
275 | older_than_this = &expire_time; |
276 | } | |
2c136579 | 277 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 278 | inode = wb_inode(delaying_queue->prev); |
0ae45f63 TT |
279 | if (older_than_this && |
280 | inode_dirtied_after(inode, *older_than_this)) | |
2c136579 | 281 | break; |
a8855990 JK |
282 | list_move(&inode->i_wb_list, &tmp); |
283 | moved++; | |
0ae45f63 TT |
284 | if (flags & EXPIRE_DIRTY_ATIME) |
285 | set_bit(__I_DIRTY_TIME_EXPIRED, &inode->i_state); | |
a8855990 JK |
286 | if (sb_is_blkdev_sb(inode->i_sb)) |
287 | continue; | |
cf137307 JA |
288 | if (sb && sb != inode->i_sb) |
289 | do_sb_sort = 1; | |
290 | sb = inode->i_sb; | |
5c03449d SL |
291 | } |
292 | ||
cf137307 JA |
293 | /* just one sb in list, splice to dispatch_queue and we're done */ |
294 | if (!do_sb_sort) { | |
295 | list_splice(&tmp, dispatch_queue); | |
e84d0a4f | 296 | goto out; |
cf137307 JA |
297 | } |
298 | ||
5c03449d SL |
299 | /* Move inodes from one superblock together */ |
300 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 301 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 302 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 303 | inode = wb_inode(pos); |
5c03449d | 304 | if (inode->i_sb == sb) |
7ccf19a8 | 305 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 306 | } |
2c136579 | 307 | } |
e84d0a4f WF |
308 | out: |
309 | return moved; | |
2c136579 FW |
310 | } |
311 | ||
312 | /* | |
313 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
314 | * Before |
315 | * newly dirtied b_dirty b_io b_more_io | |
316 | * =============> gf edc BA | |
317 | * After | |
318 | * newly dirtied b_dirty b_io b_more_io | |
319 | * =============> g fBAedc | |
320 | * | | |
321 | * +--> dequeue for IO | |
2c136579 | 322 | */ |
ad4e38dd | 323 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) |
66f3b8e2 | 324 | { |
e84d0a4f | 325 | int moved; |
0ae45f63 | 326 | |
f758eeab | 327 | assert_spin_locked(&wb->list_lock); |
4ea879b9 | 328 | list_splice_init(&wb->b_more_io, &wb->b_io); |
0ae45f63 TT |
329 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, 0, work); |
330 | moved += move_expired_inodes(&wb->b_dirty_time, &wb->b_io, | |
331 | EXPIRE_DIRTY_ATIME, work); | |
ad4e38dd | 332 | trace_writeback_queue_io(wb, work, moved); |
66f3b8e2 JA |
333 | } |
334 | ||
a9185b41 | 335 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 336 | { |
9fb0a7da TH |
337 | int ret; |
338 | ||
339 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
340 | trace_writeback_write_inode_start(inode, wbc); | |
341 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
342 | trace_writeback_write_inode(inode, wbc); | |
343 | return ret; | |
344 | } | |
03ba3782 | 345 | return 0; |
08d8e974 | 346 | } |
08d8e974 | 347 | |
1da177e4 | 348 | /* |
169ebd90 JK |
349 | * Wait for writeback on an inode to complete. Called with i_lock held. |
350 | * Caller must make sure inode cannot go away when we drop i_lock. | |
01c03194 | 351 | */ |
169ebd90 JK |
352 | static void __inode_wait_for_writeback(struct inode *inode) |
353 | __releases(inode->i_lock) | |
354 | __acquires(inode->i_lock) | |
01c03194 CH |
355 | { |
356 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
357 | wait_queue_head_t *wqh; | |
358 | ||
359 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
250df6ed DC |
360 | while (inode->i_state & I_SYNC) { |
361 | spin_unlock(&inode->i_lock); | |
74316201 N |
362 | __wait_on_bit(wqh, &wq, bit_wait, |
363 | TASK_UNINTERRUPTIBLE); | |
250df6ed | 364 | spin_lock(&inode->i_lock); |
58a9d3d8 | 365 | } |
01c03194 CH |
366 | } |
367 | ||
169ebd90 JK |
368 | /* |
369 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
370 | */ | |
371 | void inode_wait_for_writeback(struct inode *inode) | |
372 | { | |
373 | spin_lock(&inode->i_lock); | |
374 | __inode_wait_for_writeback(inode); | |
375 | spin_unlock(&inode->i_lock); | |
376 | } | |
377 | ||
378 | /* | |
379 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
380 | * held and drops it. It is aimed for callers not holding any inode reference | |
381 | * so once i_lock is dropped, inode can go away. | |
382 | */ | |
383 | static void inode_sleep_on_writeback(struct inode *inode) | |
384 | __releases(inode->i_lock) | |
385 | { | |
386 | DEFINE_WAIT(wait); | |
387 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
388 | int sleep; | |
389 | ||
390 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
391 | sleep = inode->i_state & I_SYNC; | |
392 | spin_unlock(&inode->i_lock); | |
393 | if (sleep) | |
394 | schedule(); | |
395 | finish_wait(wqh, &wait); | |
396 | } | |
397 | ||
ccb26b5a JK |
398 | /* |
399 | * Find proper writeback list for the inode depending on its current state and | |
400 | * possibly also change of its state while we were doing writeback. Here we | |
401 | * handle things such as livelock prevention or fairness of writeback among | |
402 | * inodes. This function can be called only by flusher thread - noone else | |
403 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
404 | * thread's back can have unexpected consequences. | |
405 | */ | |
406 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
407 | struct writeback_control *wbc) | |
408 | { | |
409 | if (inode->i_state & I_FREEING) | |
410 | return; | |
411 | ||
412 | /* | |
413 | * Sync livelock prevention. Each inode is tagged and synced in one | |
414 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
415 | * the dirty time to prevent enqueue and sync it again. | |
416 | */ | |
417 | if ((inode->i_state & I_DIRTY) && | |
418 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
419 | inode->dirtied_when = jiffies; | |
420 | ||
4f8ad655 JK |
421 | if (wbc->pages_skipped) { |
422 | /* | |
423 | * writeback is not making progress due to locked | |
424 | * buffers. Skip this inode for now. | |
425 | */ | |
426 | redirty_tail(inode, wb); | |
427 | return; | |
428 | } | |
429 | ||
ccb26b5a JK |
430 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { |
431 | /* | |
432 | * We didn't write back all the pages. nfs_writepages() | |
433 | * sometimes bales out without doing anything. | |
434 | */ | |
435 | if (wbc->nr_to_write <= 0) { | |
436 | /* Slice used up. Queue for next turn. */ | |
437 | requeue_io(inode, wb); | |
438 | } else { | |
439 | /* | |
440 | * Writeback blocked by something other than | |
441 | * congestion. Delay the inode for some time to | |
442 | * avoid spinning on the CPU (100% iowait) | |
443 | * retrying writeback of the dirty page/inode | |
444 | * that cannot be performed immediately. | |
445 | */ | |
446 | redirty_tail(inode, wb); | |
447 | } | |
448 | } else if (inode->i_state & I_DIRTY) { | |
449 | /* | |
450 | * Filesystems can dirty the inode during writeback operations, | |
451 | * such as delayed allocation during submission or metadata | |
452 | * updates after data IO completion. | |
453 | */ | |
454 | redirty_tail(inode, wb); | |
0ae45f63 | 455 | } else if (inode->i_state & I_DIRTY_TIME) { |
a2f48706 | 456 | inode->dirtied_when = jiffies; |
0ae45f63 | 457 | list_move(&inode->i_wb_list, &wb->b_dirty_time); |
ccb26b5a JK |
458 | } else { |
459 | /* The inode is clean. Remove from writeback lists. */ | |
460 | list_del_init(&inode->i_wb_list); | |
461 | } | |
462 | } | |
463 | ||
01c03194 | 464 | /* |
4f8ad655 JK |
465 | * Write out an inode and its dirty pages. Do not update the writeback list |
466 | * linkage. That is left to the caller. The caller is also responsible for | |
467 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. | |
1da177e4 LT |
468 | */ |
469 | static int | |
cd8ed2a4 | 470 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 471 | { |
1da177e4 | 472 | struct address_space *mapping = inode->i_mapping; |
251d6a47 | 473 | long nr_to_write = wbc->nr_to_write; |
01c03194 | 474 | unsigned dirty; |
1da177e4 LT |
475 | int ret; |
476 | ||
4f8ad655 | 477 | WARN_ON(!(inode->i_state & I_SYNC)); |
1da177e4 | 478 | |
9fb0a7da TH |
479 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); |
480 | ||
1da177e4 LT |
481 | ret = do_writepages(mapping, wbc); |
482 | ||
26821ed4 CH |
483 | /* |
484 | * Make sure to wait on the data before writing out the metadata. | |
485 | * This is important for filesystems that modify metadata on data | |
7747bd4b DC |
486 | * I/O completion. We don't do it for sync(2) writeback because it has a |
487 | * separate, external IO completion path and ->sync_fs for guaranteeing | |
488 | * inode metadata is written back correctly. | |
26821ed4 | 489 | */ |
7747bd4b | 490 | if (wbc->sync_mode == WB_SYNC_ALL && !wbc->for_sync) { |
26821ed4 | 491 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
492 | if (ret == 0) |
493 | ret = err; | |
494 | } | |
495 | ||
5547e8aa DM |
496 | /* |
497 | * Some filesystems may redirty the inode during the writeback | |
498 | * due to delalloc, clear dirty metadata flags right before | |
499 | * write_inode() | |
500 | */ | |
250df6ed | 501 | spin_lock(&inode->i_lock); |
9c6ac78e | 502 | |
5547e8aa | 503 | dirty = inode->i_state & I_DIRTY; |
a2f48706 TT |
504 | if (inode->i_state & I_DIRTY_TIME) { |
505 | if ((dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) || | |
506 | unlikely(inode->i_state & I_DIRTY_TIME_EXPIRED) || | |
507 | unlikely(time_after(jiffies, | |
508 | (inode->dirtied_time_when + | |
509 | dirtytime_expire_interval * HZ)))) { | |
510 | dirty |= I_DIRTY_TIME | I_DIRTY_TIME_EXPIRED; | |
511 | trace_writeback_lazytime(inode); | |
512 | } | |
513 | } else | |
514 | inode->i_state &= ~I_DIRTY_TIME_EXPIRED; | |
0ae45f63 | 515 | inode->i_state &= ~dirty; |
9c6ac78e TH |
516 | |
517 | /* | |
518 | * Paired with smp_mb() in __mark_inode_dirty(). This allows | |
519 | * __mark_inode_dirty() to test i_state without grabbing i_lock - | |
520 | * either they see the I_DIRTY bits cleared or we see the dirtied | |
521 | * inode. | |
522 | * | |
523 | * I_DIRTY_PAGES is always cleared together above even if @mapping | |
524 | * still has dirty pages. The flag is reinstated after smp_mb() if | |
525 | * necessary. This guarantees that either __mark_inode_dirty() | |
526 | * sees clear I_DIRTY_PAGES or we see PAGECACHE_TAG_DIRTY. | |
527 | */ | |
528 | smp_mb(); | |
529 | ||
530 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
531 | inode->i_state |= I_DIRTY_PAGES; | |
532 | ||
250df6ed | 533 | spin_unlock(&inode->i_lock); |
9c6ac78e | 534 | |
0ae45f63 TT |
535 | if (dirty & I_DIRTY_TIME) |
536 | mark_inode_dirty_sync(inode); | |
26821ed4 | 537 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
0ae45f63 | 538 | if (dirty & ~I_DIRTY_PAGES) { |
a9185b41 | 539 | int err = write_inode(inode, wbc); |
1da177e4 LT |
540 | if (ret == 0) |
541 | ret = err; | |
542 | } | |
4f8ad655 JK |
543 | trace_writeback_single_inode(inode, wbc, nr_to_write); |
544 | return ret; | |
545 | } | |
546 | ||
547 | /* | |
548 | * Write out an inode's dirty pages. Either the caller has an active reference | |
549 | * on the inode or the inode has I_WILL_FREE set. | |
550 | * | |
551 | * This function is designed to be called for writing back one inode which | |
552 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() | |
553 | * and does more profound writeback list handling in writeback_sb_inodes(). | |
554 | */ | |
555 | static int | |
556 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, | |
557 | struct writeback_control *wbc) | |
558 | { | |
559 | int ret = 0; | |
560 | ||
561 | spin_lock(&inode->i_lock); | |
562 | if (!atomic_read(&inode->i_count)) | |
563 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
564 | else | |
565 | WARN_ON(inode->i_state & I_WILL_FREE); | |
566 | ||
567 | if (inode->i_state & I_SYNC) { | |
568 | if (wbc->sync_mode != WB_SYNC_ALL) | |
569 | goto out; | |
570 | /* | |
169ebd90 JK |
571 | * It's a data-integrity sync. We must wait. Since callers hold |
572 | * inode reference or inode has I_WILL_FREE set, it cannot go | |
573 | * away under us. | |
4f8ad655 | 574 | */ |
169ebd90 | 575 | __inode_wait_for_writeback(inode); |
4f8ad655 JK |
576 | } |
577 | WARN_ON(inode->i_state & I_SYNC); | |
578 | /* | |
f9b0e058 JK |
579 | * Skip inode if it is clean and we have no outstanding writeback in |
580 | * WB_SYNC_ALL mode. We don't want to mess with writeback lists in this | |
581 | * function since flusher thread may be doing for example sync in | |
582 | * parallel and if we move the inode, it could get skipped. So here we | |
583 | * make sure inode is on some writeback list and leave it there unless | |
584 | * we have completely cleaned the inode. | |
4f8ad655 | 585 | */ |
0ae45f63 | 586 | if (!(inode->i_state & I_DIRTY_ALL) && |
f9b0e058 JK |
587 | (wbc->sync_mode != WB_SYNC_ALL || |
588 | !mapping_tagged(inode->i_mapping, PAGECACHE_TAG_WRITEBACK))) | |
4f8ad655 JK |
589 | goto out; |
590 | inode->i_state |= I_SYNC; | |
591 | spin_unlock(&inode->i_lock); | |
592 | ||
cd8ed2a4 | 593 | ret = __writeback_single_inode(inode, wbc); |
1da177e4 | 594 | |
f758eeab | 595 | spin_lock(&wb->list_lock); |
250df6ed | 596 | spin_lock(&inode->i_lock); |
4f8ad655 JK |
597 | /* |
598 | * If inode is clean, remove it from writeback lists. Otherwise don't | |
599 | * touch it. See comment above for explanation. | |
600 | */ | |
0ae45f63 | 601 | if (!(inode->i_state & I_DIRTY_ALL)) |
4f8ad655 JK |
602 | list_del_init(&inode->i_wb_list); |
603 | spin_unlock(&wb->list_lock); | |
1c0eeaf5 | 604 | inode_sync_complete(inode); |
4f8ad655 JK |
605 | out: |
606 | spin_unlock(&inode->i_lock); | |
1da177e4 LT |
607 | return ret; |
608 | } | |
609 | ||
a88a341a | 610 | static long writeback_chunk_size(struct bdi_writeback *wb, |
1a12d8bd | 611 | struct wb_writeback_work *work) |
d46db3d5 WF |
612 | { |
613 | long pages; | |
614 | ||
615 | /* | |
616 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
617 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
618 | * here avoids calling into writeback_inodes_wb() more than once. | |
619 | * | |
620 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
621 | * | |
622 | * wb_writeback() | |
623 | * writeback_sb_inodes() <== called only once | |
624 | * write_cache_pages() <== called once for each inode | |
625 | * (quickly) tag currently dirty pages | |
626 | * (maybe slowly) sync all tagged pages | |
627 | */ | |
628 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
629 | pages = LONG_MAX; | |
1a12d8bd | 630 | else { |
a88a341a | 631 | pages = min(wb->avg_write_bandwidth / 2, |
1a12d8bd WF |
632 | global_dirty_limit / DIRTY_SCOPE); |
633 | pages = min(pages, work->nr_pages); | |
634 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
635 | MIN_WRITEBACK_PAGES); | |
636 | } | |
d46db3d5 WF |
637 | |
638 | return pages; | |
639 | } | |
640 | ||
f11c9c5c ES |
641 | /* |
642 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 | 643 | * |
d46db3d5 | 644 | * Return the number of pages and/or inodes written. |
f11c9c5c | 645 | */ |
d46db3d5 WF |
646 | static long writeback_sb_inodes(struct super_block *sb, |
647 | struct bdi_writeback *wb, | |
648 | struct wb_writeback_work *work) | |
1da177e4 | 649 | { |
d46db3d5 WF |
650 | struct writeback_control wbc = { |
651 | .sync_mode = work->sync_mode, | |
652 | .tagged_writepages = work->tagged_writepages, | |
653 | .for_kupdate = work->for_kupdate, | |
654 | .for_background = work->for_background, | |
7747bd4b | 655 | .for_sync = work->for_sync, |
d46db3d5 WF |
656 | .range_cyclic = work->range_cyclic, |
657 | .range_start = 0, | |
658 | .range_end = LLONG_MAX, | |
659 | }; | |
660 | unsigned long start_time = jiffies; | |
661 | long write_chunk; | |
662 | long wrote = 0; /* count both pages and inodes */ | |
663 | ||
03ba3782 | 664 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 665 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
666 | |
667 | if (inode->i_sb != sb) { | |
d46db3d5 | 668 | if (work->sb) { |
edadfb10 CH |
669 | /* |
670 | * We only want to write back data for this | |
671 | * superblock, move all inodes not belonging | |
672 | * to it back onto the dirty list. | |
673 | */ | |
f758eeab | 674 | redirty_tail(inode, wb); |
edadfb10 CH |
675 | continue; |
676 | } | |
677 | ||
678 | /* | |
679 | * The inode belongs to a different superblock. | |
680 | * Bounce back to the caller to unpin this and | |
681 | * pin the next superblock. | |
682 | */ | |
d46db3d5 | 683 | break; |
edadfb10 CH |
684 | } |
685 | ||
9843b76a | 686 | /* |
331cbdee WL |
687 | * Don't bother with new inodes or inodes being freed, first |
688 | * kind does not need periodic writeout yet, and for the latter | |
9843b76a CH |
689 | * kind writeout is handled by the freer. |
690 | */ | |
250df6ed | 691 | spin_lock(&inode->i_lock); |
9843b76a | 692 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 693 | spin_unlock(&inode->i_lock); |
fcc5c222 | 694 | redirty_tail(inode, wb); |
7ef0d737 NP |
695 | continue; |
696 | } | |
cc1676d9 JK |
697 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { |
698 | /* | |
699 | * If this inode is locked for writeback and we are not | |
700 | * doing writeback-for-data-integrity, move it to | |
701 | * b_more_io so that writeback can proceed with the | |
702 | * other inodes on s_io. | |
703 | * | |
704 | * We'll have another go at writing back this inode | |
705 | * when we completed a full scan of b_io. | |
706 | */ | |
707 | spin_unlock(&inode->i_lock); | |
708 | requeue_io(inode, wb); | |
709 | trace_writeback_sb_inodes_requeue(inode); | |
710 | continue; | |
711 | } | |
f0d07b7f JK |
712 | spin_unlock(&wb->list_lock); |
713 | ||
4f8ad655 JK |
714 | /* |
715 | * We already requeued the inode if it had I_SYNC set and we | |
716 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
717 | * WB_SYNC_ALL case. | |
718 | */ | |
169ebd90 JK |
719 | if (inode->i_state & I_SYNC) { |
720 | /* Wait for I_SYNC. This function drops i_lock... */ | |
721 | inode_sleep_on_writeback(inode); | |
722 | /* Inode may be gone, start again */ | |
ead188f9 | 723 | spin_lock(&wb->list_lock); |
169ebd90 JK |
724 | continue; |
725 | } | |
4f8ad655 JK |
726 | inode->i_state |= I_SYNC; |
727 | spin_unlock(&inode->i_lock); | |
169ebd90 | 728 | |
a88a341a | 729 | write_chunk = writeback_chunk_size(wb, work); |
d46db3d5 WF |
730 | wbc.nr_to_write = write_chunk; |
731 | wbc.pages_skipped = 0; | |
250df6ed | 732 | |
169ebd90 JK |
733 | /* |
734 | * We use I_SYNC to pin the inode in memory. While it is set | |
735 | * evict_inode() will wait so the inode cannot be freed. | |
736 | */ | |
cd8ed2a4 | 737 | __writeback_single_inode(inode, &wbc); |
250df6ed | 738 | |
d46db3d5 WF |
739 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
740 | wrote += write_chunk - wbc.nr_to_write; | |
4f8ad655 JK |
741 | spin_lock(&wb->list_lock); |
742 | spin_lock(&inode->i_lock); | |
0ae45f63 | 743 | if (!(inode->i_state & I_DIRTY_ALL)) |
d46db3d5 | 744 | wrote++; |
4f8ad655 JK |
745 | requeue_inode(inode, wb, &wbc); |
746 | inode_sync_complete(inode); | |
0f1b1fd8 | 747 | spin_unlock(&inode->i_lock); |
169ebd90 | 748 | cond_resched_lock(&wb->list_lock); |
d46db3d5 WF |
749 | /* |
750 | * bail out to wb_writeback() often enough to check | |
751 | * background threshold and other termination conditions. | |
752 | */ | |
753 | if (wrote) { | |
754 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
755 | break; | |
756 | if (work->nr_pages <= 0) | |
757 | break; | |
8bc3be27 | 758 | } |
1da177e4 | 759 | } |
d46db3d5 | 760 | return wrote; |
f11c9c5c ES |
761 | } |
762 | ||
d46db3d5 WF |
763 | static long __writeback_inodes_wb(struct bdi_writeback *wb, |
764 | struct wb_writeback_work *work) | |
f11c9c5c | 765 | { |
d46db3d5 WF |
766 | unsigned long start_time = jiffies; |
767 | long wrote = 0; | |
38f21977 | 768 | |
f11c9c5c | 769 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 770 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 771 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 772 | |
eb6ef3df | 773 | if (!trylock_super(sb)) { |
0e995816 | 774 | /* |
eb6ef3df | 775 | * trylock_super() may fail consistently due to |
0e995816 WF |
776 | * s_umount being grabbed by someone else. Don't use |
777 | * requeue_io() to avoid busy retrying the inode/sb. | |
778 | */ | |
779 | redirty_tail(inode, wb); | |
edadfb10 | 780 | continue; |
f11c9c5c | 781 | } |
d46db3d5 | 782 | wrote += writeback_sb_inodes(sb, wb, work); |
eb6ef3df | 783 | up_read(&sb->s_umount); |
f11c9c5c | 784 | |
d46db3d5 WF |
785 | /* refer to the same tests at the end of writeback_sb_inodes */ |
786 | if (wrote) { | |
787 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
788 | break; | |
789 | if (work->nr_pages <= 0) | |
790 | break; | |
791 | } | |
f11c9c5c | 792 | } |
66f3b8e2 | 793 | /* Leave any unwritten inodes on b_io */ |
d46db3d5 | 794 | return wrote; |
66f3b8e2 JA |
795 | } |
796 | ||
7d9f073b | 797 | static long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, |
0e175a18 | 798 | enum wb_reason reason) |
edadfb10 | 799 | { |
d46db3d5 WF |
800 | struct wb_writeback_work work = { |
801 | .nr_pages = nr_pages, | |
802 | .sync_mode = WB_SYNC_NONE, | |
803 | .range_cyclic = 1, | |
0e175a18 | 804 | .reason = reason, |
d46db3d5 | 805 | }; |
edadfb10 | 806 | |
f758eeab | 807 | spin_lock(&wb->list_lock); |
424b351f | 808 | if (list_empty(&wb->b_io)) |
ad4e38dd | 809 | queue_io(wb, &work); |
d46db3d5 | 810 | __writeback_inodes_wb(wb, &work); |
f758eeab | 811 | spin_unlock(&wb->list_lock); |
edadfb10 | 812 | |
d46db3d5 WF |
813 | return nr_pages - work.nr_pages; |
814 | } | |
03ba3782 | 815 | |
a88a341a | 816 | static bool over_bground_thresh(struct bdi_writeback *wb) |
03ba3782 JA |
817 | { |
818 | unsigned long background_thresh, dirty_thresh; | |
819 | ||
16c4042f | 820 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 | 821 | |
b00949aa WF |
822 | if (global_page_state(NR_FILE_DIRTY) + |
823 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) | |
824 | return true; | |
825 | ||
a88a341a | 826 | if (wb_stat(wb, WB_RECLAIMABLE) > wb_dirty_limit(wb, background_thresh)) |
b00949aa WF |
827 | return true; |
828 | ||
829 | return false; | |
03ba3782 JA |
830 | } |
831 | ||
e98be2d5 WF |
832 | /* |
833 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
834 | * only the flusher working on the first wb should do it. | |
835 | */ | |
836 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
837 | unsigned long start_time) | |
838 | { | |
a88a341a | 839 | __wb_update_bandwidth(wb, 0, 0, 0, 0, 0, start_time); |
e98be2d5 WF |
840 | } |
841 | ||
03ba3782 JA |
842 | /* |
843 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 844 | * |
03ba3782 JA |
845 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
846 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
847 | * just walks the superblock inode list, writing back any inodes which are | |
848 | * older than a specific point in time. | |
66f3b8e2 | 849 | * |
03ba3782 JA |
850 | * Try to run once per dirty_writeback_interval. But if a writeback event |
851 | * takes longer than a dirty_writeback_interval interval, then leave a | |
852 | * one-second gap. | |
66f3b8e2 | 853 | * |
03ba3782 JA |
854 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
855 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 856 | */ |
c4a77a6c | 857 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 858 | struct wb_writeback_work *work) |
66f3b8e2 | 859 | { |
e98be2d5 | 860 | unsigned long wb_start = jiffies; |
d46db3d5 | 861 | long nr_pages = work->nr_pages; |
0dc83bd3 | 862 | unsigned long oldest_jif; |
a5989bdc | 863 | struct inode *inode; |
d46db3d5 | 864 | long progress; |
66f3b8e2 | 865 | |
0dc83bd3 JK |
866 | oldest_jif = jiffies; |
867 | work->older_than_this = &oldest_jif; | |
38f21977 | 868 | |
e8dfc305 | 869 | spin_lock(&wb->list_lock); |
03ba3782 JA |
870 | for (;;) { |
871 | /* | |
d3ddec76 | 872 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 873 | */ |
83ba7b07 | 874 | if (work->nr_pages <= 0) |
03ba3782 | 875 | break; |
66f3b8e2 | 876 | |
aa373cf5 JK |
877 | /* |
878 | * Background writeout and kupdate-style writeback may | |
879 | * run forever. Stop them if there is other work to do | |
880 | * so that e.g. sync can proceed. They'll be restarted | |
881 | * after the other works are all done. | |
882 | */ | |
883 | if ((work->for_background || work->for_kupdate) && | |
f0054bb1 | 884 | !list_empty(&wb->work_list)) |
aa373cf5 JK |
885 | break; |
886 | ||
38f21977 | 887 | /* |
d3ddec76 WF |
888 | * For background writeout, stop when we are below the |
889 | * background dirty threshold | |
38f21977 | 890 | */ |
a88a341a | 891 | if (work->for_background && !over_bground_thresh(wb)) |
03ba3782 | 892 | break; |
38f21977 | 893 | |
1bc36b64 JK |
894 | /* |
895 | * Kupdate and background works are special and we want to | |
896 | * include all inodes that need writing. Livelock avoidance is | |
897 | * handled by these works yielding to any other work so we are | |
898 | * safe. | |
899 | */ | |
ba9aa839 | 900 | if (work->for_kupdate) { |
0dc83bd3 | 901 | oldest_jif = jiffies - |
ba9aa839 | 902 | msecs_to_jiffies(dirty_expire_interval * 10); |
1bc36b64 | 903 | } else if (work->for_background) |
0dc83bd3 | 904 | oldest_jif = jiffies; |
028c2dd1 | 905 | |
d46db3d5 | 906 | trace_writeback_start(wb->bdi, work); |
e8dfc305 | 907 | if (list_empty(&wb->b_io)) |
ad4e38dd | 908 | queue_io(wb, work); |
83ba7b07 | 909 | if (work->sb) |
d46db3d5 | 910 | progress = writeback_sb_inodes(work->sb, wb, work); |
edadfb10 | 911 | else |
d46db3d5 WF |
912 | progress = __writeback_inodes_wb(wb, work); |
913 | trace_writeback_written(wb->bdi, work); | |
028c2dd1 | 914 | |
e98be2d5 | 915 | wb_update_bandwidth(wb, wb_start); |
03ba3782 JA |
916 | |
917 | /* | |
e6fb6da2 WF |
918 | * Did we write something? Try for more |
919 | * | |
920 | * Dirty inodes are moved to b_io for writeback in batches. | |
921 | * The completion of the current batch does not necessarily | |
922 | * mean the overall work is done. So we keep looping as long | |
923 | * as made some progress on cleaning pages or inodes. | |
03ba3782 | 924 | */ |
d46db3d5 | 925 | if (progress) |
71fd05a8 JA |
926 | continue; |
927 | /* | |
e6fb6da2 | 928 | * No more inodes for IO, bail |
71fd05a8 | 929 | */ |
b7a2441f | 930 | if (list_empty(&wb->b_more_io)) |
03ba3782 | 931 | break; |
71fd05a8 JA |
932 | /* |
933 | * Nothing written. Wait for some inode to | |
934 | * become available for writeback. Otherwise | |
935 | * we'll just busyloop. | |
936 | */ | |
71fd05a8 | 937 | if (!list_empty(&wb->b_more_io)) { |
d46db3d5 | 938 | trace_writeback_wait(wb->bdi, work); |
7ccf19a8 | 939 | inode = wb_inode(wb->b_more_io.prev); |
250df6ed | 940 | spin_lock(&inode->i_lock); |
f0d07b7f | 941 | spin_unlock(&wb->list_lock); |
169ebd90 JK |
942 | /* This function drops i_lock... */ |
943 | inode_sleep_on_writeback(inode); | |
f0d07b7f | 944 | spin_lock(&wb->list_lock); |
03ba3782 JA |
945 | } |
946 | } | |
e8dfc305 | 947 | spin_unlock(&wb->list_lock); |
03ba3782 | 948 | |
d46db3d5 | 949 | return nr_pages - work->nr_pages; |
03ba3782 JA |
950 | } |
951 | ||
952 | /* | |
83ba7b07 | 953 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 954 | */ |
f0054bb1 | 955 | static struct wb_writeback_work *get_next_work_item(struct bdi_writeback *wb) |
03ba3782 | 956 | { |
83ba7b07 | 957 | struct wb_writeback_work *work = NULL; |
03ba3782 | 958 | |
f0054bb1 TH |
959 | spin_lock_bh(&wb->work_lock); |
960 | if (!list_empty(&wb->work_list)) { | |
961 | work = list_entry(wb->work_list.next, | |
83ba7b07 CH |
962 | struct wb_writeback_work, list); |
963 | list_del_init(&work->list); | |
03ba3782 | 964 | } |
f0054bb1 | 965 | spin_unlock_bh(&wb->work_lock); |
83ba7b07 | 966 | return work; |
03ba3782 JA |
967 | } |
968 | ||
cdf01dd5 LT |
969 | /* |
970 | * Add in the number of potentially dirty inodes, because each inode | |
971 | * write can dirty pagecache in the underlying blockdev. | |
972 | */ | |
973 | static unsigned long get_nr_dirty_pages(void) | |
974 | { | |
975 | return global_page_state(NR_FILE_DIRTY) + | |
976 | global_page_state(NR_UNSTABLE_NFS) + | |
977 | get_nr_dirty_inodes(); | |
978 | } | |
979 | ||
6585027a JK |
980 | static long wb_check_background_flush(struct bdi_writeback *wb) |
981 | { | |
a88a341a | 982 | if (over_bground_thresh(wb)) { |
6585027a JK |
983 | |
984 | struct wb_writeback_work work = { | |
985 | .nr_pages = LONG_MAX, | |
986 | .sync_mode = WB_SYNC_NONE, | |
987 | .for_background = 1, | |
988 | .range_cyclic = 1, | |
0e175a18 | 989 | .reason = WB_REASON_BACKGROUND, |
6585027a JK |
990 | }; |
991 | ||
992 | return wb_writeback(wb, &work); | |
993 | } | |
994 | ||
995 | return 0; | |
996 | } | |
997 | ||
03ba3782 JA |
998 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
999 | { | |
1000 | unsigned long expired; | |
1001 | long nr_pages; | |
1002 | ||
69b62d01 JA |
1003 | /* |
1004 | * When set to zero, disable periodic writeback | |
1005 | */ | |
1006 | if (!dirty_writeback_interval) | |
1007 | return 0; | |
1008 | ||
03ba3782 JA |
1009 | expired = wb->last_old_flush + |
1010 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
1011 | if (time_before(jiffies, expired)) | |
1012 | return 0; | |
1013 | ||
1014 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 1015 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 1016 | |
c4a77a6c | 1017 | if (nr_pages) { |
83ba7b07 | 1018 | struct wb_writeback_work work = { |
c4a77a6c JA |
1019 | .nr_pages = nr_pages, |
1020 | .sync_mode = WB_SYNC_NONE, | |
1021 | .for_kupdate = 1, | |
1022 | .range_cyclic = 1, | |
0e175a18 | 1023 | .reason = WB_REASON_PERIODIC, |
c4a77a6c JA |
1024 | }; |
1025 | ||
83ba7b07 | 1026 | return wb_writeback(wb, &work); |
c4a77a6c | 1027 | } |
03ba3782 JA |
1028 | |
1029 | return 0; | |
1030 | } | |
1031 | ||
1032 | /* | |
1033 | * Retrieve work items and do the writeback they describe | |
1034 | */ | |
25d130ba | 1035 | static long wb_do_writeback(struct bdi_writeback *wb) |
03ba3782 | 1036 | { |
83ba7b07 | 1037 | struct wb_writeback_work *work; |
c4a77a6c | 1038 | long wrote = 0; |
03ba3782 | 1039 | |
4452226e | 1040 | set_bit(WB_writeback_running, &wb->state); |
f0054bb1 | 1041 | while ((work = get_next_work_item(wb)) != NULL) { |
03ba3782 | 1042 | |
f0054bb1 | 1043 | trace_writeback_exec(wb->bdi, work); |
455b2864 | 1044 | |
83ba7b07 | 1045 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
1046 | |
1047 | /* | |
83ba7b07 CH |
1048 | * Notify the caller of completion if this is a synchronous |
1049 | * work item, otherwise just free it. | |
03ba3782 | 1050 | */ |
83ba7b07 CH |
1051 | if (work->done) |
1052 | complete(work->done); | |
1053 | else | |
1054 | kfree(work); | |
03ba3782 JA |
1055 | } |
1056 | ||
1057 | /* | |
1058 | * Check for periodic writeback, kupdated() style | |
1059 | */ | |
1060 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 1061 | wrote += wb_check_background_flush(wb); |
4452226e | 1062 | clear_bit(WB_writeback_running, &wb->state); |
03ba3782 JA |
1063 | |
1064 | return wrote; | |
1065 | } | |
1066 | ||
1067 | /* | |
1068 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
839a8e86 | 1069 | * reschedules periodically and does kupdated style flushing. |
03ba3782 | 1070 | */ |
f0054bb1 | 1071 | void wb_workfn(struct work_struct *work) |
03ba3782 | 1072 | { |
839a8e86 TH |
1073 | struct bdi_writeback *wb = container_of(to_delayed_work(work), |
1074 | struct bdi_writeback, dwork); | |
03ba3782 JA |
1075 | long pages_written; |
1076 | ||
f0054bb1 | 1077 | set_worker_desc("flush-%s", dev_name(wb->bdi->dev)); |
766f9164 | 1078 | current->flags |= PF_SWAPWRITE; |
455b2864 | 1079 | |
839a8e86 | 1080 | if (likely(!current_is_workqueue_rescuer() || |
4452226e | 1081 | !test_bit(WB_registered, &wb->state))) { |
6467716a | 1082 | /* |
f0054bb1 | 1083 | * The normal path. Keep writing back @wb until its |
839a8e86 | 1084 | * work_list is empty. Note that this path is also taken |
f0054bb1 | 1085 | * if @wb is shutting down even when we're running off the |
839a8e86 | 1086 | * rescuer as work_list needs to be drained. |
6467716a | 1087 | */ |
839a8e86 | 1088 | do { |
25d130ba | 1089 | pages_written = wb_do_writeback(wb); |
839a8e86 | 1090 | trace_writeback_pages_written(pages_written); |
f0054bb1 | 1091 | } while (!list_empty(&wb->work_list)); |
839a8e86 TH |
1092 | } else { |
1093 | /* | |
1094 | * bdi_wq can't get enough workers and we're running off | |
1095 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
1096 | * enough for efficient IO. | |
1097 | */ | |
f0054bb1 | 1098 | pages_written = writeback_inodes_wb(wb, 1024, |
839a8e86 | 1099 | WB_REASON_FORKER_THREAD); |
455b2864 | 1100 | trace_writeback_pages_written(pages_written); |
03ba3782 JA |
1101 | } |
1102 | ||
f0054bb1 | 1103 | if (!list_empty(&wb->work_list)) |
6ca738d6 DB |
1104 | mod_delayed_work(bdi_wq, &wb->dwork, 0); |
1105 | else if (wb_has_dirty_io(wb) && dirty_writeback_interval) | |
f0054bb1 | 1106 | wb_wakeup_delayed(wb); |
455b2864 | 1107 | |
839a8e86 | 1108 | current->flags &= ~PF_SWAPWRITE; |
03ba3782 JA |
1109 | } |
1110 | ||
1111 | /* | |
b8c2f347 CH |
1112 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
1113 | * the whole world. | |
03ba3782 | 1114 | */ |
0e175a18 | 1115 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) |
03ba3782 | 1116 | { |
b8c2f347 | 1117 | struct backing_dev_info *bdi; |
03ba3782 | 1118 | |
47df3dde JK |
1119 | if (!nr_pages) |
1120 | nr_pages = get_nr_dirty_pages(); | |
03ba3782 | 1121 | |
b8c2f347 | 1122 | rcu_read_lock(); |
cfc4ba53 | 1123 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
1124 | if (!bdi_has_dirty_io(bdi)) |
1125 | continue; | |
f0054bb1 | 1126 | __wb_start_writeback(&bdi->wb, nr_pages, false, reason); |
03ba3782 | 1127 | } |
cfc4ba53 | 1128 | rcu_read_unlock(); |
1da177e4 LT |
1129 | } |
1130 | ||
a2f48706 TT |
1131 | /* |
1132 | * Wake up bdi's periodically to make sure dirtytime inodes gets | |
1133 | * written back periodically. We deliberately do *not* check the | |
1134 | * b_dirtytime list in wb_has_dirty_io(), since this would cause the | |
1135 | * kernel to be constantly waking up once there are any dirtytime | |
1136 | * inodes on the system. So instead we define a separate delayed work | |
1137 | * function which gets called much more rarely. (By default, only | |
1138 | * once every 12 hours.) | |
1139 | * | |
1140 | * If there is any other write activity going on in the file system, | |
1141 | * this function won't be necessary. But if the only thing that has | |
1142 | * happened on the file system is a dirtytime inode caused by an atime | |
1143 | * update, we need this infrastructure below to make sure that inode | |
1144 | * eventually gets pushed out to disk. | |
1145 | */ | |
1146 | static void wakeup_dirtytime_writeback(struct work_struct *w); | |
1147 | static DECLARE_DELAYED_WORK(dirtytime_work, wakeup_dirtytime_writeback); | |
1148 | ||
1149 | static void wakeup_dirtytime_writeback(struct work_struct *w) | |
1150 | { | |
1151 | struct backing_dev_info *bdi; | |
1152 | ||
1153 | rcu_read_lock(); | |
1154 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { | |
1155 | if (list_empty(&bdi->wb.b_dirty_time)) | |
1156 | continue; | |
f0054bb1 | 1157 | wb_wakeup(&bdi->wb); |
a2f48706 TT |
1158 | } |
1159 | rcu_read_unlock(); | |
1160 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
1161 | } | |
1162 | ||
1163 | static int __init start_dirtytime_writeback(void) | |
1164 | { | |
1165 | schedule_delayed_work(&dirtytime_work, dirtytime_expire_interval * HZ); | |
1166 | return 0; | |
1167 | } | |
1168 | __initcall(start_dirtytime_writeback); | |
1169 | ||
1efff914 TT |
1170 | int dirtytime_interval_handler(struct ctl_table *table, int write, |
1171 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
1172 | { | |
1173 | int ret; | |
1174 | ||
1175 | ret = proc_dointvec_minmax(table, write, buffer, lenp, ppos); | |
1176 | if (ret == 0 && write) | |
1177 | mod_delayed_work(system_wq, &dirtytime_work, 0); | |
1178 | return ret; | |
1179 | } | |
1180 | ||
03ba3782 JA |
1181 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
1182 | { | |
1183 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1184 | struct dentry *dentry; | |
1185 | const char *name = "?"; | |
1186 | ||
1187 | dentry = d_find_alias(inode); | |
1188 | if (dentry) { | |
1189 | spin_lock(&dentry->d_lock); | |
1190 | name = (const char *) dentry->d_name.name; | |
1191 | } | |
1192 | printk(KERN_DEBUG | |
1193 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1194 | current->comm, task_pid_nr(current), inode->i_ino, | |
1195 | name, inode->i_sb->s_id); | |
1196 | if (dentry) { | |
1197 | spin_unlock(&dentry->d_lock); | |
1198 | dput(dentry); | |
1199 | } | |
1200 | } | |
1201 | } | |
1202 | ||
1203 | /** | |
1204 | * __mark_inode_dirty - internal function | |
1205 | * @inode: inode to mark | |
1206 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1207 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1208 | * mark_inode_dirty_sync. | |
1da177e4 | 1209 | * |
03ba3782 JA |
1210 | * Put the inode on the super block's dirty list. |
1211 | * | |
1212 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1213 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1214 | * If it was not hashed, it will never be added to the dirty list | |
1215 | * even if it is later hashed, as it will have been marked dirty already. | |
1216 | * | |
1217 | * In short, make sure you hash any inodes _before_ you start marking | |
1218 | * them dirty. | |
1da177e4 | 1219 | * |
03ba3782 JA |
1220 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1221 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1222 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1223 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1224 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1225 | * blockdev inode. | |
1da177e4 | 1226 | */ |
0ae45f63 | 1227 | #define I_DIRTY_INODE (I_DIRTY_SYNC | I_DIRTY_DATASYNC) |
03ba3782 | 1228 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1229 | { |
03ba3782 | 1230 | struct super_block *sb = inode->i_sb; |
253c34e9 | 1231 | struct backing_dev_info *bdi = NULL; |
0ae45f63 TT |
1232 | int dirtytime; |
1233 | ||
1234 | trace_writeback_mark_inode_dirty(inode, flags); | |
1da177e4 | 1235 | |
03ba3782 JA |
1236 | /* |
1237 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1238 | * dirty the inode itself | |
1239 | */ | |
0ae45f63 | 1240 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC | I_DIRTY_TIME)) { |
9fb0a7da TH |
1241 | trace_writeback_dirty_inode_start(inode, flags); |
1242 | ||
03ba3782 | 1243 | if (sb->s_op->dirty_inode) |
aa385729 | 1244 | sb->s_op->dirty_inode(inode, flags); |
9fb0a7da TH |
1245 | |
1246 | trace_writeback_dirty_inode(inode, flags); | |
03ba3782 | 1247 | } |
0ae45f63 TT |
1248 | if (flags & I_DIRTY_INODE) |
1249 | flags &= ~I_DIRTY_TIME; | |
1250 | dirtytime = flags & I_DIRTY_TIME; | |
03ba3782 JA |
1251 | |
1252 | /* | |
9c6ac78e TH |
1253 | * Paired with smp_mb() in __writeback_single_inode() for the |
1254 | * following lockless i_state test. See there for details. | |
03ba3782 JA |
1255 | */ |
1256 | smp_mb(); | |
1257 | ||
0ae45f63 TT |
1258 | if (((inode->i_state & flags) == flags) || |
1259 | (dirtytime && (inode->i_state & I_DIRTY_INODE))) | |
03ba3782 JA |
1260 | return; |
1261 | ||
1262 | if (unlikely(block_dump)) | |
1263 | block_dump___mark_inode_dirty(inode); | |
1264 | ||
250df6ed | 1265 | spin_lock(&inode->i_lock); |
0ae45f63 TT |
1266 | if (dirtytime && (inode->i_state & I_DIRTY_INODE)) |
1267 | goto out_unlock_inode; | |
03ba3782 JA |
1268 | if ((inode->i_state & flags) != flags) { |
1269 | const int was_dirty = inode->i_state & I_DIRTY; | |
1270 | ||
0ae45f63 TT |
1271 | if (flags & I_DIRTY_INODE) |
1272 | inode->i_state &= ~I_DIRTY_TIME; | |
03ba3782 JA |
1273 | inode->i_state |= flags; |
1274 | ||
1275 | /* | |
1276 | * If the inode is being synced, just update its dirty state. | |
1277 | * The unlocker will place the inode on the appropriate | |
1278 | * superblock list, based upon its state. | |
1279 | */ | |
1280 | if (inode->i_state & I_SYNC) | |
250df6ed | 1281 | goto out_unlock_inode; |
03ba3782 JA |
1282 | |
1283 | /* | |
1284 | * Only add valid (hashed) inodes to the superblock's | |
1285 | * dirty list. Add blockdev inodes as well. | |
1286 | */ | |
1287 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1288 | if (inode_unhashed(inode)) |
250df6ed | 1289 | goto out_unlock_inode; |
03ba3782 | 1290 | } |
a4ffdde6 | 1291 | if (inode->i_state & I_FREEING) |
250df6ed | 1292 | goto out_unlock_inode; |
03ba3782 JA |
1293 | |
1294 | /* | |
1295 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1296 | * reposition it (that would break b_dirty time-ordering). | |
1297 | */ | |
1298 | if (!was_dirty) { | |
a66979ab | 1299 | bool wakeup_bdi = false; |
253c34e9 AB |
1300 | bdi = inode_to_bdi(inode); |
1301 | ||
146d7009 JB |
1302 | spin_unlock(&inode->i_lock); |
1303 | spin_lock(&bdi->wb.list_lock); | |
253c34e9 | 1304 | if (bdi_cap_writeback_dirty(bdi)) { |
4452226e | 1305 | WARN(!test_bit(WB_registered, &bdi->wb.state), |
253c34e9 AB |
1306 | "bdi-%s not registered\n", bdi->name); |
1307 | ||
1308 | /* | |
1309 | * If this is the first dirty inode for this | |
1310 | * bdi, we have to wake-up the corresponding | |
1311 | * bdi thread to make sure background | |
1312 | * write-back happens later. | |
1313 | */ | |
1314 | if (!wb_has_dirty_io(&bdi->wb)) | |
1315 | wakeup_bdi = true; | |
500b067c | 1316 | } |
03ba3782 JA |
1317 | |
1318 | inode->dirtied_when = jiffies; | |
a2f48706 TT |
1319 | if (dirtytime) |
1320 | inode->dirtied_time_when = jiffies; | |
1321 | if (inode->i_state & (I_DIRTY_INODE | I_DIRTY_PAGES)) | |
1322 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); | |
1323 | else | |
1324 | list_move(&inode->i_wb_list, | |
1325 | &bdi->wb.b_dirty_time); | |
f758eeab | 1326 | spin_unlock(&bdi->wb.list_lock); |
0ae45f63 | 1327 | trace_writeback_dirty_inode_enqueue(inode); |
a66979ab DC |
1328 | |
1329 | if (wakeup_bdi) | |
f0054bb1 | 1330 | wb_wakeup_delayed(&bdi->wb); |
a66979ab | 1331 | return; |
1da177e4 | 1332 | } |
1da177e4 | 1333 | } |
250df6ed DC |
1334 | out_unlock_inode: |
1335 | spin_unlock(&inode->i_lock); | |
253c34e9 | 1336 | |
03ba3782 JA |
1337 | } |
1338 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1339 | ||
b6e51316 | 1340 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1341 | { |
1342 | struct inode *inode, *old_inode = NULL; | |
1343 | ||
1344 | /* | |
1345 | * We need to be protected against the filesystem going from | |
1346 | * r/o to r/w or vice versa. | |
1347 | */ | |
b6e51316 | 1348 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1349 | |
55fa6091 | 1350 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1351 | |
1352 | /* | |
1353 | * Data integrity sync. Must wait for all pages under writeback, | |
1354 | * because there may have been pages dirtied before our sync | |
1355 | * call, but which had writeout started before we write it out. | |
1356 | * In which case, the inode may not be on the dirty list, but | |
1357 | * we still have to wait for that writeout. | |
1358 | */ | |
b6e51316 | 1359 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1360 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1361 | |
250df6ed DC |
1362 | spin_lock(&inode->i_lock); |
1363 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1364 | (mapping->nrpages == 0)) { | |
1365 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1366 | continue; |
250df6ed | 1367 | } |
03ba3782 | 1368 | __iget(inode); |
250df6ed | 1369 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1370 | spin_unlock(&inode_sb_list_lock); |
1371 | ||
03ba3782 | 1372 | /* |
55fa6091 DC |
1373 | * We hold a reference to 'inode' so it couldn't have been |
1374 | * removed from s_inodes list while we dropped the | |
1375 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1376 | * be holding the last reference and we cannot iput it under | |
1377 | * inode_sb_list_lock. So we keep the reference and iput it | |
1378 | * later. | |
03ba3782 JA |
1379 | */ |
1380 | iput(old_inode); | |
1381 | old_inode = inode; | |
1382 | ||
1383 | filemap_fdatawait(mapping); | |
1384 | ||
1385 | cond_resched(); | |
1386 | ||
55fa6091 | 1387 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1388 | } |
55fa6091 | 1389 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1390 | iput(old_inode); |
1da177e4 LT |
1391 | } |
1392 | ||
d8a8559c | 1393 | /** |
3259f8be | 1394 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1395 | * @sb: the superblock |
3259f8be | 1396 | * @nr: the number of pages to write |
786228ab | 1397 | * @reason: reason why some writeback work initiated |
1da177e4 | 1398 | * |
d8a8559c JA |
1399 | * Start writeback on some inodes on this super_block. No guarantees are made |
1400 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1401 | * for IO completion of submitted IO. |
1da177e4 | 1402 | */ |
0e175a18 CW |
1403 | void writeback_inodes_sb_nr(struct super_block *sb, |
1404 | unsigned long nr, | |
1405 | enum wb_reason reason) | |
1da177e4 | 1406 | { |
83ba7b07 CH |
1407 | DECLARE_COMPLETION_ONSTACK(done); |
1408 | struct wb_writeback_work work = { | |
6e6938b6 WF |
1409 | .sb = sb, |
1410 | .sync_mode = WB_SYNC_NONE, | |
1411 | .tagged_writepages = 1, | |
1412 | .done = &done, | |
1413 | .nr_pages = nr, | |
0e175a18 | 1414 | .reason = reason, |
3c4d7165 | 1415 | }; |
d8a8559c | 1416 | |
6eedc701 JK |
1417 | if (sb->s_bdi == &noop_backing_dev_info) |
1418 | return; | |
cf37e972 | 1419 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
f0054bb1 | 1420 | wb_queue_work(&sb->s_bdi->wb, &work); |
83ba7b07 | 1421 | wait_for_completion(&done); |
e913fc82 | 1422 | } |
3259f8be CM |
1423 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1424 | ||
1425 | /** | |
1426 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1427 | * @sb: the superblock | |
786228ab | 1428 | * @reason: reason why some writeback work was initiated |
3259f8be CM |
1429 | * |
1430 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1431 | * on how many (if any) will be written, and this function does not wait | |
1432 | * for IO completion of submitted IO. | |
1433 | */ | |
0e175a18 | 1434 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1435 | { |
0e175a18 | 1436 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1437 | } |
0e3c9a22 | 1438 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1439 | |
17bd55d0 | 1440 | /** |
10ee27a0 | 1441 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway |
17bd55d0 | 1442 | * @sb: the superblock |
10ee27a0 MX |
1443 | * @nr: the number of pages to write |
1444 | * @reason: the reason of writeback | |
17bd55d0 | 1445 | * |
10ee27a0 | 1446 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. |
17bd55d0 ES |
1447 | * Returns 1 if writeback was started, 0 if not. |
1448 | */ | |
10ee27a0 MX |
1449 | int try_to_writeback_inodes_sb_nr(struct super_block *sb, |
1450 | unsigned long nr, | |
1451 | enum wb_reason reason) | |
17bd55d0 | 1452 | { |
10ee27a0 | 1453 | if (writeback_in_progress(sb->s_bdi)) |
17bd55d0 | 1454 | return 1; |
10ee27a0 MX |
1455 | |
1456 | if (!down_read_trylock(&sb->s_umount)) | |
17bd55d0 | 1457 | return 0; |
10ee27a0 MX |
1458 | |
1459 | writeback_inodes_sb_nr(sb, nr, reason); | |
1460 | up_read(&sb->s_umount); | |
1461 | return 1; | |
17bd55d0 | 1462 | } |
10ee27a0 | 1463 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); |
17bd55d0 | 1464 | |
3259f8be | 1465 | /** |
10ee27a0 | 1466 | * try_to_writeback_inodes_sb - try to start writeback if none underway |
3259f8be | 1467 | * @sb: the superblock |
786228ab | 1468 | * @reason: reason why some writeback work was initiated |
3259f8be | 1469 | * |
10ee27a0 | 1470 | * Implement by try_to_writeback_inodes_sb_nr() |
3259f8be CM |
1471 | * Returns 1 if writeback was started, 0 if not. |
1472 | */ | |
10ee27a0 | 1473 | int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) |
3259f8be | 1474 | { |
10ee27a0 | 1475 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); |
3259f8be | 1476 | } |
10ee27a0 | 1477 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); |
3259f8be | 1478 | |
d8a8559c JA |
1479 | /** |
1480 | * sync_inodes_sb - sync sb inode pages | |
0dc83bd3 | 1481 | * @sb: the superblock |
d8a8559c JA |
1482 | * |
1483 | * This function writes and waits on any dirty inode belonging to this | |
0dc83bd3 | 1484 | * super_block. |
d8a8559c | 1485 | */ |
0dc83bd3 | 1486 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1487 | { |
83ba7b07 CH |
1488 | DECLARE_COMPLETION_ONSTACK(done); |
1489 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1490 | .sb = sb, |
1491 | .sync_mode = WB_SYNC_ALL, | |
1492 | .nr_pages = LONG_MAX, | |
1493 | .range_cyclic = 0, | |
83ba7b07 | 1494 | .done = &done, |
0e175a18 | 1495 | .reason = WB_REASON_SYNC, |
7747bd4b | 1496 | .for_sync = 1, |
3c4d7165 CH |
1497 | }; |
1498 | ||
6eedc701 JK |
1499 | /* Nothing to do? */ |
1500 | if (sb->s_bdi == &noop_backing_dev_info) | |
1501 | return; | |
cf37e972 CH |
1502 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1503 | ||
f0054bb1 | 1504 | wb_queue_work(&sb->s_bdi->wb, &work); |
83ba7b07 CH |
1505 | wait_for_completion(&done); |
1506 | ||
b6e51316 | 1507 | wait_sb_inodes(sb); |
1da177e4 | 1508 | } |
d8a8559c | 1509 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1510 | |
1da177e4 | 1511 | /** |
7f04c26d AA |
1512 | * write_inode_now - write an inode to disk |
1513 | * @inode: inode to write to disk | |
1514 | * @sync: whether the write should be synchronous or not | |
1515 | * | |
1516 | * This function commits an inode to disk immediately if it is dirty. This is | |
1517 | * primarily needed by knfsd. | |
1da177e4 | 1518 | * |
7f04c26d | 1519 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1520 | */ |
1da177e4 LT |
1521 | int write_inode_now(struct inode *inode, int sync) |
1522 | { | |
f758eeab | 1523 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
1da177e4 LT |
1524 | struct writeback_control wbc = { |
1525 | .nr_to_write = LONG_MAX, | |
18914b18 | 1526 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1527 | .range_start = 0, |
1528 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1529 | }; |
1530 | ||
1531 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1532 | wbc.nr_to_write = 0; |
1da177e4 LT |
1533 | |
1534 | might_sleep(); | |
4f8ad655 | 1535 | return writeback_single_inode(inode, wb, &wbc); |
1da177e4 LT |
1536 | } |
1537 | EXPORT_SYMBOL(write_inode_now); | |
1538 | ||
1539 | /** | |
1540 | * sync_inode - write an inode and its pages to disk. | |
1541 | * @inode: the inode to sync | |
1542 | * @wbc: controls the writeback mode | |
1543 | * | |
1544 | * sync_inode() will write an inode and its pages to disk. It will also | |
1545 | * correctly update the inode on its superblock's dirty inode lists and will | |
1546 | * update inode->i_state. | |
1547 | * | |
1548 | * The caller must have a ref on the inode. | |
1549 | */ | |
1550 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1551 | { | |
4f8ad655 | 1552 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); |
1da177e4 LT |
1553 | } |
1554 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1555 | |
1556 | /** | |
c691b9d9 | 1557 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1558 | * @inode: the inode to sync |
1559 | * @wait: wait for I/O to complete. | |
1560 | * | |
c691b9d9 | 1561 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1562 | * |
1563 | * Note: only writes the actual inode, no associated data or other metadata. | |
1564 | */ | |
1565 | int sync_inode_metadata(struct inode *inode, int wait) | |
1566 | { | |
1567 | struct writeback_control wbc = { | |
1568 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1569 | .nr_to_write = 0, /* metadata-only */ | |
1570 | }; | |
1571 | ||
1572 | return sync_inode(inode, &wbc); | |
1573 | } | |
1574 | EXPORT_SYMBOL(sync_inode_metadata); |