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> | |
f5ff8422 | 17 | #include <linux/module.h> |
1da177e4 LT |
18 | #include <linux/spinlock.h> |
19 | #include <linux/sched.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/mm.h> | |
03ba3782 JA |
22 | #include <linux/kthread.h> |
23 | #include <linux/freezer.h> | |
1da177e4 LT |
24 | #include <linux/writeback.h> |
25 | #include <linux/blkdev.h> | |
26 | #include <linux/backing-dev.h> | |
27 | #include <linux/buffer_head.h> | |
07f3f05c | 28 | #include "internal.h" |
1da177e4 | 29 | |
66f3b8e2 | 30 | #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) |
f11b00f3 | 31 | |
d0bceac7 JA |
32 | /* |
33 | * We don't actually have pdflush, but this one is exported though /proc... | |
34 | */ | |
35 | int nr_pdflush_threads; | |
36 | ||
c4a77a6c JA |
37 | /* |
38 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
39 | */ | |
40 | struct wb_writeback_args { | |
41 | long nr_pages; | |
42 | struct super_block *sb; | |
43 | enum writeback_sync_modes sync_mode; | |
44 | int for_kupdate; | |
45 | int range_cyclic; | |
46 | }; | |
47 | ||
03ba3782 JA |
48 | /* |
49 | * Work items for the bdi_writeback threads | |
f11b00f3 | 50 | */ |
03ba3782 JA |
51 | struct bdi_work { |
52 | struct list_head list; | |
03ba3782 JA |
53 | struct rcu_head rcu_head; |
54 | ||
55 | unsigned long seen; | |
56 | atomic_t pending; | |
57 | ||
c4a77a6c | 58 | struct wb_writeback_args args; |
03ba3782 JA |
59 | |
60 | unsigned long state; | |
61 | }; | |
62 | ||
63 | enum { | |
64 | WS_USED_B = 0, | |
65 | WS_ONSTACK_B, | |
66 | }; | |
67 | ||
68 | #define WS_USED (1 << WS_USED_B) | |
69 | #define WS_ONSTACK (1 << WS_ONSTACK_B) | |
70 | ||
71 | static inline bool bdi_work_on_stack(struct bdi_work *work) | |
72 | { | |
73 | return test_bit(WS_ONSTACK_B, &work->state); | |
74 | } | |
75 | ||
76 | static inline void bdi_work_init(struct bdi_work *work, | |
77 | struct writeback_control *wbc) | |
78 | { | |
79 | INIT_RCU_HEAD(&work->rcu_head); | |
c4a77a6c JA |
80 | work->args.sb = wbc->sb; |
81 | work->args.nr_pages = wbc->nr_to_write; | |
82 | work->args.sync_mode = wbc->sync_mode; | |
83 | work->args.range_cyclic = wbc->range_cyclic; | |
84 | work->args.for_kupdate = 0; | |
03ba3782 JA |
85 | work->state = WS_USED; |
86 | } | |
87 | ||
f11b00f3 AB |
88 | /** |
89 | * writeback_in_progress - determine whether there is writeback in progress | |
90 | * @bdi: the device's backing_dev_info structure. | |
91 | * | |
03ba3782 JA |
92 | * Determine whether there is writeback waiting to be handled against a |
93 | * backing device. | |
f11b00f3 AB |
94 | */ |
95 | int writeback_in_progress(struct backing_dev_info *bdi) | |
96 | { | |
03ba3782 | 97 | return !list_empty(&bdi->work_list); |
f11b00f3 AB |
98 | } |
99 | ||
03ba3782 | 100 | static void bdi_work_clear(struct bdi_work *work) |
f11b00f3 | 101 | { |
03ba3782 JA |
102 | clear_bit(WS_USED_B, &work->state); |
103 | smp_mb__after_clear_bit(); | |
104 | wake_up_bit(&work->state, WS_USED_B); | |
f11b00f3 AB |
105 | } |
106 | ||
03ba3782 | 107 | static void bdi_work_free(struct rcu_head *head) |
4195f73d | 108 | { |
03ba3782 | 109 | struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); |
4195f73d | 110 | |
03ba3782 JA |
111 | if (!bdi_work_on_stack(work)) |
112 | kfree(work); | |
113 | else | |
114 | bdi_work_clear(work); | |
4195f73d NP |
115 | } |
116 | ||
03ba3782 | 117 | static void wb_work_complete(struct bdi_work *work) |
1da177e4 | 118 | { |
c4a77a6c | 119 | const enum writeback_sync_modes sync_mode = work->args.sync_mode; |
1da177e4 LT |
120 | |
121 | /* | |
03ba3782 JA |
122 | * For allocated work, we can clear the done/seen bit right here. |
123 | * For on-stack work, we need to postpone both the clear and free | |
124 | * to after the RCU grace period, since the stack could be invalidated | |
125 | * as soon as bdi_work_clear() has done the wakeup. | |
1da177e4 | 126 | */ |
03ba3782 JA |
127 | if (!bdi_work_on_stack(work)) |
128 | bdi_work_clear(work); | |
129 | if (sync_mode == WB_SYNC_NONE || bdi_work_on_stack(work)) | |
130 | call_rcu(&work->rcu_head, bdi_work_free); | |
131 | } | |
1da177e4 | 132 | |
03ba3782 JA |
133 | static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) |
134 | { | |
1da177e4 | 135 | /* |
03ba3782 JA |
136 | * The caller has retrieved the work arguments from this work, |
137 | * drop our reference. If this is the last ref, delete and free it | |
1da177e4 | 138 | */ |
03ba3782 JA |
139 | if (atomic_dec_and_test(&work->pending)) { |
140 | struct backing_dev_info *bdi = wb->bdi; | |
1da177e4 | 141 | |
03ba3782 JA |
142 | spin_lock(&bdi->wb_lock); |
143 | list_del_rcu(&work->list); | |
144 | spin_unlock(&bdi->wb_lock); | |
1da177e4 | 145 | |
03ba3782 JA |
146 | wb_work_complete(work); |
147 | } | |
148 | } | |
1da177e4 | 149 | |
03ba3782 JA |
150 | static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) |
151 | { | |
bcddc3f0 JA |
152 | work->seen = bdi->wb_mask; |
153 | BUG_ON(!work->seen); | |
154 | atomic_set(&work->pending, bdi->wb_cnt); | |
155 | BUG_ON(!bdi->wb_cnt); | |
1da177e4 | 156 | |
bcddc3f0 JA |
157 | /* |
158 | * Make sure stores are seen before it appears on the list | |
159 | */ | |
160 | smp_mb(); | |
1da177e4 | 161 | |
bcddc3f0 JA |
162 | spin_lock(&bdi->wb_lock); |
163 | list_add_tail_rcu(&work->list, &bdi->work_list); | |
164 | spin_unlock(&bdi->wb_lock); | |
03ba3782 JA |
165 | |
166 | /* | |
167 | * If the default thread isn't there, make sure we add it. When | |
168 | * it gets created and wakes up, we'll run this work. | |
169 | */ | |
170 | if (unlikely(list_empty_careful(&bdi->wb_list))) | |
171 | wake_up_process(default_backing_dev_info.wb.task); | |
172 | else { | |
173 | struct bdi_writeback *wb = &bdi->wb; | |
1da177e4 LT |
174 | |
175 | /* | |
bcddc3f0 JA |
176 | * End work now if this wb has no dirty IO pending. Otherwise |
177 | * wakeup the handling thread | |
1da177e4 | 178 | */ |
bcddc3f0 JA |
179 | if (!wb_has_dirty_io(wb)) |
180 | wb_clear_pending(wb, work); | |
181 | else if (wb->task) | |
03ba3782 | 182 | wake_up_process(wb->task); |
1da177e4 | 183 | } |
1da177e4 LT |
184 | } |
185 | ||
03ba3782 JA |
186 | /* |
187 | * Used for on-stack allocated work items. The caller needs to wait until | |
188 | * the wb threads have acked the work before it's safe to continue. | |
189 | */ | |
190 | static void bdi_wait_on_work_clear(struct bdi_work *work) | |
191 | { | |
192 | wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, | |
193 | TASK_UNINTERRUPTIBLE); | |
194 | } | |
1da177e4 | 195 | |
f11fcae8 JA |
196 | static void bdi_alloc_queue_work(struct backing_dev_info *bdi, |
197 | struct writeback_control *wbc) | |
1da177e4 | 198 | { |
03ba3782 JA |
199 | struct bdi_work *work; |
200 | ||
bcddc3f0 JA |
201 | /* |
202 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
203 | * wakeup the thread for old dirty data writeback | |
204 | */ | |
03ba3782 | 205 | work = kmalloc(sizeof(*work), GFP_ATOMIC); |
bcddc3f0 | 206 | if (work) { |
03ba3782 | 207 | bdi_work_init(work, wbc); |
bcddc3f0 JA |
208 | bdi_queue_work(bdi, work); |
209 | } else { | |
210 | struct bdi_writeback *wb = &bdi->wb; | |
03ba3782 | 211 | |
bcddc3f0 JA |
212 | if (wb->task) |
213 | wake_up_process(wb->task); | |
214 | } | |
03ba3782 JA |
215 | } |
216 | ||
217 | void bdi_start_writeback(struct writeback_control *wbc) | |
218 | { | |
f0fad8a5 CH |
219 | /* |
220 | * WB_SYNC_NONE is opportunistic writeback. If this allocation fails, | |
221 | * bdi_queue_work() will wake up the thread and flush old data. This | |
222 | * should ensure some amount of progress in freeing memory. | |
223 | */ | |
f11fcae8 JA |
224 | if (wbc->sync_mode != WB_SYNC_ALL) |
225 | bdi_alloc_queue_work(wbc->bdi, wbc); | |
226 | else { | |
f0fad8a5 | 227 | struct bdi_work work; |
03ba3782 | 228 | |
f0fad8a5 CH |
229 | bdi_work_init(&work, wbc); |
230 | work.state |= WS_ONSTACK; | |
03ba3782 | 231 | |
f0fad8a5 CH |
232 | bdi_queue_work(wbc->bdi, &work); |
233 | bdi_wait_on_work_clear(&work); | |
03ba3782 | 234 | } |
1da177e4 LT |
235 | } |
236 | ||
6610a0bc AM |
237 | /* |
238 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
239 | * furthest end of its superblock's dirty-inode list. | |
240 | * | |
241 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 242 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
243 | * the case then the inode must have been redirtied while it was being written |
244 | * out and we don't reset its dirtied_when. | |
245 | */ | |
246 | static void redirty_tail(struct inode *inode) | |
247 | { | |
03ba3782 | 248 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
6610a0bc | 249 | |
03ba3782 | 250 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 251 | struct inode *tail; |
6610a0bc | 252 | |
03ba3782 | 253 | tail = list_entry(wb->b_dirty.next, struct inode, i_list); |
66f3b8e2 | 254 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
255 | inode->dirtied_when = jiffies; |
256 | } | |
03ba3782 | 257 | list_move(&inode->i_list, &wb->b_dirty); |
6610a0bc AM |
258 | } |
259 | ||
c986d1e2 | 260 | /* |
66f3b8e2 | 261 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 262 | */ |
0e0f4fc2 | 263 | static void requeue_io(struct inode *inode) |
c986d1e2 | 264 | { |
03ba3782 JA |
265 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
266 | ||
267 | list_move(&inode->i_list, &wb->b_more_io); | |
c986d1e2 AM |
268 | } |
269 | ||
1c0eeaf5 JE |
270 | static void inode_sync_complete(struct inode *inode) |
271 | { | |
272 | /* | |
273 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
274 | */ | |
275 | smp_mb(); | |
276 | wake_up_bit(&inode->i_state, __I_SYNC); | |
277 | } | |
278 | ||
d2caa3c5 JL |
279 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
280 | { | |
281 | bool ret = time_after(inode->dirtied_when, t); | |
282 | #ifndef CONFIG_64BIT | |
283 | /* | |
284 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
285 | * It _appears_ to be in the future, but is actually in distant past. | |
286 | * This test is necessary to prevent such wrapped-around relative times | |
287 | * from permanently stopping the whole pdflush writeback. | |
288 | */ | |
289 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
290 | #endif | |
291 | return ret; | |
292 | } | |
293 | ||
2c136579 FW |
294 | /* |
295 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
296 | */ | |
297 | static void move_expired_inodes(struct list_head *delaying_queue, | |
298 | struct list_head *dispatch_queue, | |
299 | unsigned long *older_than_this) | |
300 | { | |
301 | while (!list_empty(delaying_queue)) { | |
302 | struct inode *inode = list_entry(delaying_queue->prev, | |
303 | struct inode, i_list); | |
304 | if (older_than_this && | |
d2caa3c5 | 305 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 FW |
306 | break; |
307 | list_move(&inode->i_list, dispatch_queue); | |
308 | } | |
309 | } | |
310 | ||
311 | /* | |
312 | * Queue all expired dirty inodes for io, eldest first. | |
313 | */ | |
03ba3782 | 314 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 315 | { |
03ba3782 JA |
316 | list_splice_init(&wb->b_more_io, wb->b_io.prev); |
317 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); | |
66f3b8e2 JA |
318 | } |
319 | ||
03ba3782 | 320 | static int write_inode(struct inode *inode, int sync) |
08d8e974 | 321 | { |
03ba3782 JA |
322 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
323 | return inode->i_sb->s_op->write_inode(inode, sync); | |
324 | return 0; | |
08d8e974 | 325 | } |
08d8e974 | 326 | |
1da177e4 | 327 | /* |
01c03194 CH |
328 | * Wait for writeback on an inode to complete. |
329 | */ | |
330 | static void inode_wait_for_writeback(struct inode *inode) | |
331 | { | |
332 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
333 | wait_queue_head_t *wqh; | |
334 | ||
335 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
336 | do { | |
337 | spin_unlock(&inode_lock); | |
338 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); | |
339 | spin_lock(&inode_lock); | |
340 | } while (inode->i_state & I_SYNC); | |
341 | } | |
342 | ||
343 | /* | |
344 | * Write out an inode's dirty pages. Called under inode_lock. Either the | |
345 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
346 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
347 | * | |
1da177e4 LT |
348 | * If `wait' is set, wait on the writeout. |
349 | * | |
350 | * The whole writeout design is quite complex and fragile. We want to avoid | |
351 | * starvation of particular inodes when others are being redirtied, prevent | |
352 | * livelocks, etc. | |
353 | * | |
354 | * Called under inode_lock. | |
355 | */ | |
356 | static int | |
01c03194 | 357 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 358 | { |
1da177e4 | 359 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 360 | int wait = wbc->sync_mode == WB_SYNC_ALL; |
01c03194 | 361 | unsigned dirty; |
1da177e4 LT |
362 | int ret; |
363 | ||
01c03194 CH |
364 | if (!atomic_read(&inode->i_count)) |
365 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
366 | else | |
367 | WARN_ON(inode->i_state & I_WILL_FREE); | |
368 | ||
369 | if (inode->i_state & I_SYNC) { | |
370 | /* | |
371 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 372 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
373 | * writeback can proceed with the other inodes on s_io. |
374 | * | |
375 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 376 | * completed a full scan of b_io. |
01c03194 CH |
377 | */ |
378 | if (!wait) { | |
379 | requeue_io(inode); | |
380 | return 0; | |
381 | } | |
382 | ||
383 | /* | |
384 | * It's a data-integrity sync. We must wait. | |
385 | */ | |
386 | inode_wait_for_writeback(inode); | |
387 | } | |
388 | ||
1c0eeaf5 | 389 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 390 | |
1c0eeaf5 | 391 | /* Set I_SYNC, reset I_DIRTY */ |
1da177e4 | 392 | dirty = inode->i_state & I_DIRTY; |
1c0eeaf5 | 393 | inode->i_state |= I_SYNC; |
1da177e4 LT |
394 | inode->i_state &= ~I_DIRTY; |
395 | ||
396 | spin_unlock(&inode_lock); | |
397 | ||
398 | ret = do_writepages(mapping, wbc); | |
399 | ||
400 | /* Don't write the inode if only I_DIRTY_PAGES was set */ | |
401 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
402 | int err = write_inode(inode, wait); | |
403 | if (ret == 0) | |
404 | ret = err; | |
405 | } | |
406 | ||
407 | if (wait) { | |
408 | int err = filemap_fdatawait(mapping); | |
409 | if (ret == 0) | |
410 | ret = err; | |
411 | } | |
412 | ||
413 | spin_lock(&inode_lock); | |
1c0eeaf5 | 414 | inode->i_state &= ~I_SYNC; |
84a89245 | 415 | if (!(inode->i_state & (I_FREEING | I_CLEAR))) { |
1da177e4 LT |
416 | if (!(inode->i_state & I_DIRTY) && |
417 | mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
418 | /* | |
419 | * We didn't write back all the pages. nfs_writepages() | |
420 | * sometimes bales out without doing anything. Redirty | |
66f3b8e2 | 421 | * the inode; Move it from b_io onto b_more_io/b_dirty. |
1b43ef91 AM |
422 | */ |
423 | /* | |
424 | * akpm: if the caller was the kupdate function we put | |
66f3b8e2 | 425 | * this inode at the head of b_dirty so it gets first |
1b43ef91 AM |
426 | * consideration. Otherwise, move it to the tail, for |
427 | * the reasons described there. I'm not really sure | |
428 | * how much sense this makes. Presumably I had a good | |
429 | * reasons for doing it this way, and I'd rather not | |
430 | * muck with it at present. | |
1da177e4 LT |
431 | */ |
432 | if (wbc->for_kupdate) { | |
433 | /* | |
2c136579 | 434 | * For the kupdate function we move the inode |
66f3b8e2 | 435 | * to b_more_io so it will get more writeout as |
2c136579 | 436 | * soon as the queue becomes uncongested. |
1da177e4 LT |
437 | */ |
438 | inode->i_state |= I_DIRTY_PAGES; | |
8bc3be27 FW |
439 | if (wbc->nr_to_write <= 0) { |
440 | /* | |
441 | * slice used up: queue for next turn | |
442 | */ | |
443 | requeue_io(inode); | |
444 | } else { | |
445 | /* | |
446 | * somehow blocked: retry later | |
447 | */ | |
448 | redirty_tail(inode); | |
449 | } | |
1da177e4 LT |
450 | } else { |
451 | /* | |
452 | * Otherwise fully redirty the inode so that | |
453 | * other inodes on this superblock will get some | |
454 | * writeout. Otherwise heavy writing to one | |
455 | * file would indefinitely suspend writeout of | |
456 | * all the other files. | |
457 | */ | |
458 | inode->i_state |= I_DIRTY_PAGES; | |
1b43ef91 | 459 | redirty_tail(inode); |
1da177e4 LT |
460 | } |
461 | } else if (inode->i_state & I_DIRTY) { | |
462 | /* | |
463 | * Someone redirtied the inode while were writing back | |
464 | * the pages. | |
465 | */ | |
6610a0bc | 466 | redirty_tail(inode); |
1da177e4 LT |
467 | } else if (atomic_read(&inode->i_count)) { |
468 | /* | |
469 | * The inode is clean, inuse | |
470 | */ | |
471 | list_move(&inode->i_list, &inode_in_use); | |
472 | } else { | |
473 | /* | |
474 | * The inode is clean, unused | |
475 | */ | |
476 | list_move(&inode->i_list, &inode_unused); | |
1da177e4 LT |
477 | } |
478 | } | |
1c0eeaf5 | 479 | inode_sync_complete(inode); |
1da177e4 LT |
480 | return ret; |
481 | } | |
482 | ||
03ba3782 JA |
483 | /* |
484 | * For WB_SYNC_NONE writeback, the caller does not have the sb pinned | |
485 | * before calling writeback. So make sure that we do pin it, so it doesn't | |
486 | * go away while we are writing inodes from it. | |
487 | * | |
488 | * Returns 0 if the super was successfully pinned (or pinning wasn't needed), | |
489 | * 1 if we failed. | |
490 | */ | |
491 | static int pin_sb_for_writeback(struct writeback_control *wbc, | |
492 | struct inode *inode) | |
493 | { | |
494 | struct super_block *sb = inode->i_sb; | |
495 | ||
496 | /* | |
497 | * Caller must already hold the ref for this | |
498 | */ | |
499 | if (wbc->sync_mode == WB_SYNC_ALL) { | |
500 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
501 | return 0; | |
502 | } | |
503 | ||
504 | spin_lock(&sb_lock); | |
505 | sb->s_count++; | |
506 | if (down_read_trylock(&sb->s_umount)) { | |
507 | if (sb->s_root) { | |
508 | spin_unlock(&sb_lock); | |
509 | return 0; | |
510 | } | |
511 | /* | |
512 | * umounted, drop rwsem again and fall through to failure | |
513 | */ | |
514 | up_read(&sb->s_umount); | |
515 | } | |
516 | ||
517 | sb->s_count--; | |
518 | spin_unlock(&sb_lock); | |
519 | return 1; | |
520 | } | |
521 | ||
522 | static void unpin_sb_for_writeback(struct writeback_control *wbc, | |
523 | struct inode *inode) | |
524 | { | |
525 | struct super_block *sb = inode->i_sb; | |
526 | ||
527 | if (wbc->sync_mode == WB_SYNC_ALL) | |
528 | return; | |
529 | ||
530 | up_read(&sb->s_umount); | |
531 | put_super(sb); | |
532 | } | |
533 | ||
534 | static void writeback_inodes_wb(struct bdi_writeback *wb, | |
535 | struct writeback_control *wbc) | |
1da177e4 | 536 | { |
03ba3782 | 537 | struct super_block *sb = wbc->sb; |
66f3b8e2 | 538 | const int is_blkdev_sb = sb_is_blkdev_sb(sb); |
1da177e4 LT |
539 | const unsigned long start = jiffies; /* livelock avoidance */ |
540 | ||
ae8547b0 | 541 | spin_lock(&inode_lock); |
1da177e4 | 542 | |
03ba3782 JA |
543 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) |
544 | queue_io(wb, wbc->older_than_this); | |
66f3b8e2 | 545 | |
03ba3782 JA |
546 | while (!list_empty(&wb->b_io)) { |
547 | struct inode *inode = list_entry(wb->b_io.prev, | |
1da177e4 | 548 | struct inode, i_list); |
1da177e4 LT |
549 | long pages_skipped; |
550 | ||
66f3b8e2 JA |
551 | /* |
552 | * super block given and doesn't match, skip this inode | |
553 | */ | |
554 | if (sb && sb != inode->i_sb) { | |
555 | redirty_tail(inode); | |
556 | continue; | |
557 | } | |
558 | ||
03ba3782 | 559 | if (!bdi_cap_writeback_dirty(wb->bdi)) { |
9852a0e7 | 560 | redirty_tail(inode); |
66f3b8e2 | 561 | if (is_blkdev_sb) { |
1da177e4 LT |
562 | /* |
563 | * Dirty memory-backed blockdev: the ramdisk | |
564 | * driver does this. Skip just this inode | |
565 | */ | |
566 | continue; | |
567 | } | |
568 | /* | |
569 | * Dirty memory-backed inode against a filesystem other | |
570 | * than the kernel-internal bdev filesystem. Skip the | |
571 | * entire superblock. | |
572 | */ | |
573 | break; | |
574 | } | |
575 | ||
84a89245 | 576 | if (inode->i_state & (I_NEW | I_WILL_FREE)) { |
7ef0d737 NP |
577 | requeue_io(inode); |
578 | continue; | |
579 | } | |
580 | ||
03ba3782 | 581 | if (wbc->nonblocking && bdi_write_congested(wb->bdi)) { |
1da177e4 | 582 | wbc->encountered_congestion = 1; |
66f3b8e2 | 583 | if (!is_blkdev_sb) |
1da177e4 | 584 | break; /* Skip a congested fs */ |
0e0f4fc2 | 585 | requeue_io(inode); |
1da177e4 LT |
586 | continue; /* Skip a congested blockdev */ |
587 | } | |
588 | ||
d2caa3c5 JL |
589 | /* |
590 | * Was this inode dirtied after sync_sb_inodes was called? | |
591 | * This keeps sync from extra jobs and livelock. | |
592 | */ | |
593 | if (inode_dirtied_after(inode, start)) | |
1da177e4 LT |
594 | break; |
595 | ||
03ba3782 JA |
596 | if (pin_sb_for_writeback(wbc, inode)) { |
597 | requeue_io(inode); | |
598 | continue; | |
599 | } | |
1da177e4 | 600 | |
84a89245 | 601 | BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); |
1da177e4 LT |
602 | __iget(inode); |
603 | pages_skipped = wbc->pages_skipped; | |
01c03194 | 604 | writeback_single_inode(inode, wbc); |
03ba3782 | 605 | unpin_sb_for_writeback(wbc, inode); |
1da177e4 LT |
606 | if (wbc->pages_skipped != pages_skipped) { |
607 | /* | |
608 | * writeback is not making progress due to locked | |
609 | * buffers. Skip this inode for now. | |
610 | */ | |
f57b9b7b | 611 | redirty_tail(inode); |
1da177e4 LT |
612 | } |
613 | spin_unlock(&inode_lock); | |
1da177e4 | 614 | iput(inode); |
4ffc8444 | 615 | cond_resched(); |
1da177e4 | 616 | spin_lock(&inode_lock); |
8bc3be27 FW |
617 | if (wbc->nr_to_write <= 0) { |
618 | wbc->more_io = 1; | |
1da177e4 | 619 | break; |
8bc3be27 | 620 | } |
03ba3782 | 621 | if (!list_empty(&wb->b_more_io)) |
8bc3be27 | 622 | wbc->more_io = 1; |
1da177e4 | 623 | } |
38f21977 | 624 | |
66f3b8e2 JA |
625 | spin_unlock(&inode_lock); |
626 | /* Leave any unwritten inodes on b_io */ | |
627 | } | |
628 | ||
03ba3782 JA |
629 | void writeback_inodes_wbc(struct writeback_control *wbc) |
630 | { | |
631 | struct backing_dev_info *bdi = wbc->bdi; | |
632 | ||
633 | writeback_inodes_wb(&bdi->wb, wbc); | |
634 | } | |
635 | ||
66f3b8e2 | 636 | /* |
03ba3782 JA |
637 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
638 | * operation. We do this so we don't hold I_SYNC against an inode for | |
639 | * enormous amounts of time, which would block a userspace task which has | |
640 | * been forced to throttle against that inode. Also, the code reevaluates | |
641 | * the dirty each time it has written this many pages. | |
642 | */ | |
643 | #define MAX_WRITEBACK_PAGES 1024 | |
644 | ||
645 | static inline bool over_bground_thresh(void) | |
646 | { | |
647 | unsigned long background_thresh, dirty_thresh; | |
648 | ||
649 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); | |
650 | ||
651 | return (global_page_state(NR_FILE_DIRTY) + | |
652 | global_page_state(NR_UNSTABLE_NFS) >= background_thresh); | |
653 | } | |
654 | ||
655 | /* | |
656 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 657 | * |
03ba3782 JA |
658 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
659 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
660 | * just walks the superblock inode list, writing back any inodes which are | |
661 | * older than a specific point in time. | |
66f3b8e2 | 662 | * |
03ba3782 JA |
663 | * Try to run once per dirty_writeback_interval. But if a writeback event |
664 | * takes longer than a dirty_writeback_interval interval, then leave a | |
665 | * one-second gap. | |
66f3b8e2 | 666 | * |
03ba3782 JA |
667 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
668 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 669 | */ |
c4a77a6c JA |
670 | static long wb_writeback(struct bdi_writeback *wb, |
671 | struct wb_writeback_args *args) | |
66f3b8e2 | 672 | { |
03ba3782 JA |
673 | struct writeback_control wbc = { |
674 | .bdi = wb->bdi, | |
c4a77a6c JA |
675 | .sb = args->sb, |
676 | .sync_mode = args->sync_mode, | |
03ba3782 | 677 | .older_than_this = NULL, |
c4a77a6c JA |
678 | .for_kupdate = args->for_kupdate, |
679 | .range_cyclic = args->range_cyclic, | |
03ba3782 JA |
680 | }; |
681 | unsigned long oldest_jif; | |
682 | long wrote = 0; | |
66f3b8e2 | 683 | |
03ba3782 JA |
684 | if (wbc.for_kupdate) { |
685 | wbc.older_than_this = &oldest_jif; | |
686 | oldest_jif = jiffies - | |
687 | msecs_to_jiffies(dirty_expire_interval * 10); | |
688 | } | |
c4a77a6c JA |
689 | if (!wbc.range_cyclic) { |
690 | wbc.range_start = 0; | |
691 | wbc.range_end = LLONG_MAX; | |
692 | } | |
38f21977 | 693 | |
03ba3782 JA |
694 | for (;;) { |
695 | /* | |
696 | * Don't flush anything for non-integrity writeback where | |
697 | * no nr_pages was given | |
698 | */ | |
c4a77a6c JA |
699 | if (!args->for_kupdate && args->nr_pages <= 0 && |
700 | args->sync_mode == WB_SYNC_NONE) | |
03ba3782 | 701 | break; |
66f3b8e2 | 702 | |
38f21977 | 703 | /* |
03ba3782 JA |
704 | * If no specific pages were given and this is just a |
705 | * periodic background writeout and we are below the | |
706 | * background dirty threshold, don't do anything | |
38f21977 | 707 | */ |
c4a77a6c JA |
708 | if (args->for_kupdate && args->nr_pages <= 0 && |
709 | !over_bground_thresh()) | |
03ba3782 | 710 | break; |
38f21977 | 711 | |
03ba3782 JA |
712 | wbc.more_io = 0; |
713 | wbc.encountered_congestion = 0; | |
714 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; | |
715 | wbc.pages_skipped = 0; | |
716 | writeback_inodes_wb(wb, &wbc); | |
c4a77a6c | 717 | args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
03ba3782 JA |
718 | wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
719 | ||
720 | /* | |
721 | * If we ran out of stuff to write, bail unless more_io got set | |
722 | */ | |
723 | if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) { | |
724 | if (wbc.more_io && !wbc.for_kupdate) | |
38f21977 | 725 | continue; |
03ba3782 JA |
726 | break; |
727 | } | |
728 | } | |
729 | ||
730 | return wrote; | |
731 | } | |
732 | ||
733 | /* | |
734 | * Return the next bdi_work struct that hasn't been processed by this | |
735 | * wb thread yet | |
736 | */ | |
737 | static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, | |
738 | struct bdi_writeback *wb) | |
739 | { | |
740 | struct bdi_work *work, *ret = NULL; | |
741 | ||
742 | rcu_read_lock(); | |
743 | ||
744 | list_for_each_entry_rcu(work, &bdi->work_list, list) { | |
745 | if (!test_and_clear_bit(wb->nr, &work->seen)) | |
746 | continue; | |
747 | ||
748 | ret = work; | |
749 | break; | |
750 | } | |
751 | ||
752 | rcu_read_unlock(); | |
753 | return ret; | |
754 | } | |
755 | ||
756 | static long wb_check_old_data_flush(struct bdi_writeback *wb) | |
757 | { | |
758 | unsigned long expired; | |
759 | long nr_pages; | |
760 | ||
761 | expired = wb->last_old_flush + | |
762 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
763 | if (time_before(jiffies, expired)) | |
764 | return 0; | |
765 | ||
766 | wb->last_old_flush = jiffies; | |
767 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
768 | global_page_state(NR_UNSTABLE_NFS) + | |
769 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); | |
770 | ||
c4a77a6c JA |
771 | if (nr_pages) { |
772 | struct wb_writeback_args args = { | |
773 | .nr_pages = nr_pages, | |
774 | .sync_mode = WB_SYNC_NONE, | |
775 | .for_kupdate = 1, | |
776 | .range_cyclic = 1, | |
777 | }; | |
778 | ||
779 | return wb_writeback(wb, &args); | |
780 | } | |
03ba3782 JA |
781 | |
782 | return 0; | |
783 | } | |
784 | ||
785 | /* | |
786 | * Retrieve work items and do the writeback they describe | |
787 | */ | |
788 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
789 | { | |
790 | struct backing_dev_info *bdi = wb->bdi; | |
791 | struct bdi_work *work; | |
c4a77a6c | 792 | long wrote = 0; |
03ba3782 JA |
793 | |
794 | while ((work = get_next_work_item(bdi, wb)) != NULL) { | |
c4a77a6c | 795 | struct wb_writeback_args args = work->args; |
03ba3782 JA |
796 | |
797 | /* | |
798 | * Override sync mode, in case we must wait for completion | |
799 | */ | |
800 | if (force_wait) | |
c4a77a6c | 801 | work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; |
03ba3782 JA |
802 | |
803 | /* | |
804 | * If this isn't a data integrity operation, just notify | |
805 | * that we have seen this work and we are now starting it. | |
806 | */ | |
c4a77a6c | 807 | if (args.sync_mode == WB_SYNC_NONE) |
03ba3782 JA |
808 | wb_clear_pending(wb, work); |
809 | ||
c4a77a6c | 810 | wrote += wb_writeback(wb, &args); |
03ba3782 JA |
811 | |
812 | /* | |
813 | * This is a data integrity writeback, so only do the | |
814 | * notification when we have completed the work. | |
815 | */ | |
c4a77a6c | 816 | if (args.sync_mode == WB_SYNC_ALL) |
03ba3782 JA |
817 | wb_clear_pending(wb, work); |
818 | } | |
819 | ||
820 | /* | |
821 | * Check for periodic writeback, kupdated() style | |
822 | */ | |
823 | wrote += wb_check_old_data_flush(wb); | |
824 | ||
825 | return wrote; | |
826 | } | |
827 | ||
828 | /* | |
829 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
830 | * wakes up periodically and does kupdated style flushing. | |
831 | */ | |
832 | int bdi_writeback_task(struct bdi_writeback *wb) | |
833 | { | |
834 | unsigned long last_active = jiffies; | |
835 | unsigned long wait_jiffies = -1UL; | |
836 | long pages_written; | |
837 | ||
838 | while (!kthread_should_stop()) { | |
839 | pages_written = wb_do_writeback(wb, 0); | |
840 | ||
841 | if (pages_written) | |
842 | last_active = jiffies; | |
843 | else if (wait_jiffies != -1UL) { | |
844 | unsigned long max_idle; | |
845 | ||
38f21977 | 846 | /* |
03ba3782 JA |
847 | * Longest period of inactivity that we tolerate. If we |
848 | * see dirty data again later, the task will get | |
849 | * recreated automatically. | |
38f21977 | 850 | */ |
03ba3782 JA |
851 | max_idle = max(5UL * 60 * HZ, wait_jiffies); |
852 | if (time_after(jiffies, max_idle + last_active)) | |
853 | break; | |
854 | } | |
855 | ||
856 | wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); | |
857 | set_current_state(TASK_INTERRUPTIBLE); | |
858 | schedule_timeout(wait_jiffies); | |
859 | try_to_freeze(); | |
860 | } | |
861 | ||
862 | return 0; | |
863 | } | |
864 | ||
865 | /* | |
f11fcae8 JA |
866 | * Schedule writeback for all backing devices. Can only be used for |
867 | * WB_SYNC_NONE writeback, WB_SYNC_ALL should use bdi_start_writeback() | |
868 | * and pass in the superblock. | |
03ba3782 JA |
869 | */ |
870 | static void bdi_writeback_all(struct writeback_control *wbc) | |
871 | { | |
03ba3782 | 872 | struct backing_dev_info *bdi; |
03ba3782 | 873 | |
f11fcae8 JA |
874 | WARN_ON(wbc->sync_mode == WB_SYNC_ALL); |
875 | ||
cfc4ba53 | 876 | rcu_read_lock(); |
03ba3782 | 877 | |
cfc4ba53 | 878 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
879 | if (!bdi_has_dirty_io(bdi)) |
880 | continue; | |
38f21977 | 881 | |
f11fcae8 | 882 | bdi_alloc_queue_work(bdi, wbc); |
03ba3782 JA |
883 | } |
884 | ||
cfc4ba53 | 885 | rcu_read_unlock(); |
1da177e4 LT |
886 | } |
887 | ||
888 | /* | |
03ba3782 JA |
889 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
890 | * the whole world. | |
891 | */ | |
892 | void wakeup_flusher_threads(long nr_pages) | |
893 | { | |
894 | struct writeback_control wbc = { | |
895 | .sync_mode = WB_SYNC_NONE, | |
896 | .older_than_this = NULL, | |
897 | .range_cyclic = 1, | |
898 | }; | |
899 | ||
900 | if (nr_pages == 0) | |
901 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
902 | global_page_state(NR_UNSTABLE_NFS); | |
903 | wbc.nr_to_write = nr_pages; | |
904 | bdi_writeback_all(&wbc); | |
905 | } | |
906 | ||
907 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) | |
908 | { | |
909 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
910 | struct dentry *dentry; | |
911 | const char *name = "?"; | |
912 | ||
913 | dentry = d_find_alias(inode); | |
914 | if (dentry) { | |
915 | spin_lock(&dentry->d_lock); | |
916 | name = (const char *) dentry->d_name.name; | |
917 | } | |
918 | printk(KERN_DEBUG | |
919 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
920 | current->comm, task_pid_nr(current), inode->i_ino, | |
921 | name, inode->i_sb->s_id); | |
922 | if (dentry) { | |
923 | spin_unlock(&dentry->d_lock); | |
924 | dput(dentry); | |
925 | } | |
926 | } | |
927 | } | |
928 | ||
929 | /** | |
930 | * __mark_inode_dirty - internal function | |
931 | * @inode: inode to mark | |
932 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
933 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
934 | * mark_inode_dirty_sync. | |
1da177e4 | 935 | * |
03ba3782 JA |
936 | * Put the inode on the super block's dirty list. |
937 | * | |
938 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
939 | * dirty list only if it is hashed or if it refers to a blockdev. | |
940 | * If it was not hashed, it will never be added to the dirty list | |
941 | * even if it is later hashed, as it will have been marked dirty already. | |
942 | * | |
943 | * In short, make sure you hash any inodes _before_ you start marking | |
944 | * them dirty. | |
1da177e4 | 945 | * |
03ba3782 JA |
946 | * This function *must* be atomic for the I_DIRTY_PAGES case - |
947 | * set_page_dirty() is called under spinlock in several places. | |
1da177e4 | 948 | * |
03ba3782 JA |
949 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
950 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
951 | * the kernel-internal blockdev inode represents the dirtying time of the | |
952 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
953 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
954 | * blockdev inode. | |
1da177e4 | 955 | */ |
03ba3782 | 956 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 957 | { |
03ba3782 | 958 | struct super_block *sb = inode->i_sb; |
1da177e4 | 959 | |
03ba3782 JA |
960 | /* |
961 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
962 | * dirty the inode itself | |
963 | */ | |
964 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
965 | if (sb->s_op->dirty_inode) | |
966 | sb->s_op->dirty_inode(inode); | |
967 | } | |
968 | ||
969 | /* | |
970 | * make sure that changes are seen by all cpus before we test i_state | |
971 | * -- mikulas | |
972 | */ | |
973 | smp_mb(); | |
974 | ||
975 | /* avoid the locking if we can */ | |
976 | if ((inode->i_state & flags) == flags) | |
977 | return; | |
978 | ||
979 | if (unlikely(block_dump)) | |
980 | block_dump___mark_inode_dirty(inode); | |
981 | ||
982 | spin_lock(&inode_lock); | |
983 | if ((inode->i_state & flags) != flags) { | |
984 | const int was_dirty = inode->i_state & I_DIRTY; | |
985 | ||
986 | inode->i_state |= flags; | |
987 | ||
988 | /* | |
989 | * If the inode is being synced, just update its dirty state. | |
990 | * The unlocker will place the inode on the appropriate | |
991 | * superblock list, based upon its state. | |
992 | */ | |
993 | if (inode->i_state & I_SYNC) | |
994 | goto out; | |
995 | ||
996 | /* | |
997 | * Only add valid (hashed) inodes to the superblock's | |
998 | * dirty list. Add blockdev inodes as well. | |
999 | */ | |
1000 | if (!S_ISBLK(inode->i_mode)) { | |
1001 | if (hlist_unhashed(&inode->i_hash)) | |
1002 | goto out; | |
1003 | } | |
1004 | if (inode->i_state & (I_FREEING|I_CLEAR)) | |
1005 | goto out; | |
1006 | ||
1007 | /* | |
1008 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1009 | * reposition it (that would break b_dirty time-ordering). | |
1010 | */ | |
1011 | if (!was_dirty) { | |
1012 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; | |
500b067c JA |
1013 | struct backing_dev_info *bdi = wb->bdi; |
1014 | ||
1015 | if (bdi_cap_writeback_dirty(bdi) && | |
1016 | !test_bit(BDI_registered, &bdi->state)) { | |
1017 | WARN_ON(1); | |
1018 | printk(KERN_ERR "bdi-%s not registered\n", | |
1019 | bdi->name); | |
1020 | } | |
03ba3782 JA |
1021 | |
1022 | inode->dirtied_when = jiffies; | |
1023 | list_move(&inode->i_list, &wb->b_dirty); | |
1da177e4 | 1024 | } |
1da177e4 | 1025 | } |
03ba3782 JA |
1026 | out: |
1027 | spin_unlock(&inode_lock); | |
1028 | } | |
1029 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1030 | ||
1031 | /* | |
1032 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1033 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1034 | * | |
1035 | * If older_than_this is non-NULL, then only write out inodes which | |
1036 | * had their first dirtying at a time earlier than *older_than_this. | |
1037 | * | |
1038 | * If we're a pdlfush thread, then implement pdflush collision avoidance | |
1039 | * against the entire list. | |
1040 | * | |
1041 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. | |
1042 | * This function assumes that the blockdev superblock's inodes are backed by | |
1043 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1044 | * assume that all inodes are backed by the same queue. | |
1045 | * | |
1046 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1047 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1048 | * on the writer throttling path, and we get decent balancing between many | |
1049 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1050 | */ | |
1051 | static void wait_sb_inodes(struct writeback_control *wbc) | |
1052 | { | |
1053 | struct inode *inode, *old_inode = NULL; | |
1054 | ||
1055 | /* | |
1056 | * We need to be protected against the filesystem going from | |
1057 | * r/o to r/w or vice versa. | |
1058 | */ | |
1059 | WARN_ON(!rwsem_is_locked(&wbc->sb->s_umount)); | |
1060 | ||
1061 | spin_lock(&inode_lock); | |
1062 | ||
1063 | /* | |
1064 | * Data integrity sync. Must wait for all pages under writeback, | |
1065 | * because there may have been pages dirtied before our sync | |
1066 | * call, but which had writeout started before we write it out. | |
1067 | * In which case, the inode may not be on the dirty list, but | |
1068 | * we still have to wait for that writeout. | |
1069 | */ | |
1070 | list_for_each_entry(inode, &wbc->sb->s_inodes, i_sb_list) { | |
1071 | struct address_space *mapping; | |
1072 | ||
1073 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) | |
1074 | continue; | |
1075 | mapping = inode->i_mapping; | |
1076 | if (mapping->nrpages == 0) | |
1077 | continue; | |
1078 | __iget(inode); | |
1079 | spin_unlock(&inode_lock); | |
1080 | /* | |
1081 | * We hold a reference to 'inode' so it couldn't have | |
1082 | * been removed from s_inodes list while we dropped the | |
1083 | * inode_lock. We cannot iput the inode now as we can | |
1084 | * be holding the last reference and we cannot iput it | |
1085 | * under inode_lock. So we keep the reference and iput | |
1086 | * it later. | |
1087 | */ | |
1088 | iput(old_inode); | |
1089 | old_inode = inode; | |
1090 | ||
1091 | filemap_fdatawait(mapping); | |
1092 | ||
1093 | cond_resched(); | |
1094 | ||
1095 | spin_lock(&inode_lock); | |
1096 | } | |
1097 | spin_unlock(&inode_lock); | |
1098 | iput(old_inode); | |
1da177e4 LT |
1099 | } |
1100 | ||
d8a8559c JA |
1101 | /** |
1102 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1103 | * @sb: the superblock | |
1da177e4 | 1104 | * |
d8a8559c JA |
1105 | * Start writeback on some inodes on this super_block. No guarantees are made |
1106 | * on how many (if any) will be written, and this function does not wait | |
1107 | * for IO completion of submitted IO. The number of pages submitted is | |
1108 | * returned. | |
1da177e4 | 1109 | */ |
d8a8559c | 1110 | long writeback_inodes_sb(struct super_block *sb) |
1da177e4 LT |
1111 | { |
1112 | struct writeback_control wbc = { | |
03ba3782 | 1113 | .sb = sb, |
d8a8559c | 1114 | .sync_mode = WB_SYNC_NONE, |
111ebb6e OH |
1115 | .range_start = 0, |
1116 | .range_end = LLONG_MAX, | |
1da177e4 | 1117 | }; |
d8a8559c JA |
1118 | unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); |
1119 | unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); | |
1120 | long nr_to_write; | |
1da177e4 | 1121 | |
d8a8559c | 1122 | nr_to_write = nr_dirty + nr_unstable + |
38f21977 | 1123 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); |
38f21977 | 1124 | |
d8a8559c | 1125 | wbc.nr_to_write = nr_to_write; |
03ba3782 | 1126 | bdi_writeback_all(&wbc); |
d8a8559c JA |
1127 | return nr_to_write - wbc.nr_to_write; |
1128 | } | |
1129 | EXPORT_SYMBOL(writeback_inodes_sb); | |
1130 | ||
1131 | /** | |
1132 | * sync_inodes_sb - sync sb inode pages | |
1133 | * @sb: the superblock | |
1134 | * | |
1135 | * This function writes and waits on any dirty inode belonging to this | |
1136 | * super_block. The number of pages synced is returned. | |
1137 | */ | |
1138 | long sync_inodes_sb(struct super_block *sb) | |
1139 | { | |
1140 | struct writeback_control wbc = { | |
03ba3782 | 1141 | .sb = sb, |
f11fcae8 | 1142 | .bdi = sb->s_bdi, |
d8a8559c JA |
1143 | .sync_mode = WB_SYNC_ALL, |
1144 | .range_start = 0, | |
1145 | .range_end = LLONG_MAX, | |
1146 | }; | |
1147 | long nr_to_write = LONG_MAX; /* doesn't actually matter */ | |
1148 | ||
1149 | wbc.nr_to_write = nr_to_write; | |
f11fcae8 | 1150 | bdi_start_writeback(&wbc); |
03ba3782 | 1151 | wait_sb_inodes(&wbc); |
d8a8559c | 1152 | return nr_to_write - wbc.nr_to_write; |
1da177e4 | 1153 | } |
d8a8559c | 1154 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1155 | |
1da177e4 | 1156 | /** |
7f04c26d AA |
1157 | * write_inode_now - write an inode to disk |
1158 | * @inode: inode to write to disk | |
1159 | * @sync: whether the write should be synchronous or not | |
1160 | * | |
1161 | * This function commits an inode to disk immediately if it is dirty. This is | |
1162 | * primarily needed by knfsd. | |
1da177e4 | 1163 | * |
7f04c26d | 1164 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1165 | */ |
1da177e4 LT |
1166 | int write_inode_now(struct inode *inode, int sync) |
1167 | { | |
1168 | int ret; | |
1169 | struct writeback_control wbc = { | |
1170 | .nr_to_write = LONG_MAX, | |
18914b18 | 1171 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1172 | .range_start = 0, |
1173 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1174 | }; |
1175 | ||
1176 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1177 | wbc.nr_to_write = 0; |
1da177e4 LT |
1178 | |
1179 | might_sleep(); | |
1180 | spin_lock(&inode_lock); | |
01c03194 | 1181 | ret = writeback_single_inode(inode, &wbc); |
1da177e4 LT |
1182 | spin_unlock(&inode_lock); |
1183 | if (sync) | |
1c0eeaf5 | 1184 | inode_sync_wait(inode); |
1da177e4 LT |
1185 | return ret; |
1186 | } | |
1187 | EXPORT_SYMBOL(write_inode_now); | |
1188 | ||
1189 | /** | |
1190 | * sync_inode - write an inode and its pages to disk. | |
1191 | * @inode: the inode to sync | |
1192 | * @wbc: controls the writeback mode | |
1193 | * | |
1194 | * sync_inode() will write an inode and its pages to disk. It will also | |
1195 | * correctly update the inode on its superblock's dirty inode lists and will | |
1196 | * update inode->i_state. | |
1197 | * | |
1198 | * The caller must have a ref on the inode. | |
1199 | */ | |
1200 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1201 | { | |
1202 | int ret; | |
1203 | ||
1204 | spin_lock(&inode_lock); | |
01c03194 | 1205 | ret = writeback_single_inode(inode, wbc); |
1da177e4 LT |
1206 | spin_unlock(&inode_lock); |
1207 | return ret; | |
1208 | } | |
1209 | EXPORT_SYMBOL(sync_inode); |