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; | |
53 | struct list_head wait_list; | |
54 | struct rcu_head rcu_head; | |
55 | ||
56 | unsigned long seen; | |
57 | atomic_t pending; | |
58 | ||
c4a77a6c | 59 | struct wb_writeback_args args; |
03ba3782 JA |
60 | |
61 | unsigned long state; | |
62 | }; | |
63 | ||
64 | enum { | |
65 | WS_USED_B = 0, | |
66 | WS_ONSTACK_B, | |
67 | }; | |
68 | ||
69 | #define WS_USED (1 << WS_USED_B) | |
70 | #define WS_ONSTACK (1 << WS_ONSTACK_B) | |
71 | ||
72 | static inline bool bdi_work_on_stack(struct bdi_work *work) | |
73 | { | |
74 | return test_bit(WS_ONSTACK_B, &work->state); | |
75 | } | |
76 | ||
77 | static inline void bdi_work_init(struct bdi_work *work, | |
78 | struct writeback_control *wbc) | |
79 | { | |
80 | INIT_RCU_HEAD(&work->rcu_head); | |
c4a77a6c JA |
81 | work->args.sb = wbc->sb; |
82 | work->args.nr_pages = wbc->nr_to_write; | |
83 | work->args.sync_mode = wbc->sync_mode; | |
84 | work->args.range_cyclic = wbc->range_cyclic; | |
85 | work->args.for_kupdate = 0; | |
03ba3782 JA |
86 | work->state = WS_USED; |
87 | } | |
88 | ||
f11b00f3 AB |
89 | /** |
90 | * writeback_in_progress - determine whether there is writeback in progress | |
91 | * @bdi: the device's backing_dev_info structure. | |
92 | * | |
03ba3782 JA |
93 | * Determine whether there is writeback waiting to be handled against a |
94 | * backing device. | |
f11b00f3 AB |
95 | */ |
96 | int writeback_in_progress(struct backing_dev_info *bdi) | |
97 | { | |
03ba3782 | 98 | return !list_empty(&bdi->work_list); |
f11b00f3 AB |
99 | } |
100 | ||
03ba3782 | 101 | static void bdi_work_clear(struct bdi_work *work) |
f11b00f3 | 102 | { |
03ba3782 JA |
103 | clear_bit(WS_USED_B, &work->state); |
104 | smp_mb__after_clear_bit(); | |
105 | wake_up_bit(&work->state, WS_USED_B); | |
f11b00f3 AB |
106 | } |
107 | ||
03ba3782 | 108 | static void bdi_work_free(struct rcu_head *head) |
4195f73d | 109 | { |
03ba3782 | 110 | struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); |
4195f73d | 111 | |
03ba3782 JA |
112 | if (!bdi_work_on_stack(work)) |
113 | kfree(work); | |
114 | else | |
115 | bdi_work_clear(work); | |
4195f73d NP |
116 | } |
117 | ||
03ba3782 | 118 | static void wb_work_complete(struct bdi_work *work) |
1da177e4 | 119 | { |
c4a77a6c | 120 | const enum writeback_sync_modes sync_mode = work->args.sync_mode; |
1da177e4 LT |
121 | |
122 | /* | |
03ba3782 JA |
123 | * For allocated work, we can clear the done/seen bit right here. |
124 | * For on-stack work, we need to postpone both the clear and free | |
125 | * to after the RCU grace period, since the stack could be invalidated | |
126 | * as soon as bdi_work_clear() has done the wakeup. | |
1da177e4 | 127 | */ |
03ba3782 JA |
128 | if (!bdi_work_on_stack(work)) |
129 | bdi_work_clear(work); | |
130 | if (sync_mode == WB_SYNC_NONE || bdi_work_on_stack(work)) | |
131 | call_rcu(&work->rcu_head, bdi_work_free); | |
132 | } | |
1da177e4 | 133 | |
03ba3782 JA |
134 | static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) |
135 | { | |
1da177e4 | 136 | /* |
03ba3782 JA |
137 | * The caller has retrieved the work arguments from this work, |
138 | * drop our reference. If this is the last ref, delete and free it | |
1da177e4 | 139 | */ |
03ba3782 JA |
140 | if (atomic_dec_and_test(&work->pending)) { |
141 | struct backing_dev_info *bdi = wb->bdi; | |
1da177e4 | 142 | |
03ba3782 JA |
143 | spin_lock(&bdi->wb_lock); |
144 | list_del_rcu(&work->list); | |
145 | spin_unlock(&bdi->wb_lock); | |
1da177e4 | 146 | |
03ba3782 JA |
147 | wb_work_complete(work); |
148 | } | |
149 | } | |
1da177e4 | 150 | |
03ba3782 JA |
151 | static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) |
152 | { | |
153 | if (work) { | |
154 | work->seen = bdi->wb_mask; | |
155 | BUG_ON(!work->seen); | |
156 | atomic_set(&work->pending, bdi->wb_cnt); | |
157 | BUG_ON(!bdi->wb_cnt); | |
1da177e4 LT |
158 | |
159 | /* | |
03ba3782 | 160 | * Make sure stores are seen before it appears on the list |
1da177e4 | 161 | */ |
03ba3782 | 162 | smp_mb(); |
1da177e4 | 163 | |
03ba3782 JA |
164 | spin_lock(&bdi->wb_lock); |
165 | list_add_tail_rcu(&work->list, &bdi->work_list); | |
166 | spin_unlock(&bdi->wb_lock); | |
167 | } | |
168 | ||
169 | /* | |
170 | * If the default thread isn't there, make sure we add it. When | |
171 | * it gets created and wakes up, we'll run this work. | |
172 | */ | |
173 | if (unlikely(list_empty_careful(&bdi->wb_list))) | |
174 | wake_up_process(default_backing_dev_info.wb.task); | |
175 | else { | |
176 | struct bdi_writeback *wb = &bdi->wb; | |
1da177e4 LT |
177 | |
178 | /* | |
03ba3782 JA |
179 | * If we failed allocating the bdi work item, wake up the wb |
180 | * thread always. As a safety precaution, it'll flush out | |
181 | * everything | |
1da177e4 | 182 | */ |
03ba3782 JA |
183 | if (!wb_has_dirty_io(wb)) { |
184 | if (work) | |
185 | wb_clear_pending(wb, work); | |
186 | } else if (wb->task) | |
187 | wake_up_process(wb->task); | |
1da177e4 | 188 | } |
1da177e4 LT |
189 | } |
190 | ||
03ba3782 JA |
191 | /* |
192 | * Used for on-stack allocated work items. The caller needs to wait until | |
193 | * the wb threads have acked the work before it's safe to continue. | |
194 | */ | |
195 | static void bdi_wait_on_work_clear(struct bdi_work *work) | |
196 | { | |
197 | wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, | |
198 | TASK_UNINTERRUPTIBLE); | |
199 | } | |
1da177e4 | 200 | |
03ba3782 | 201 | static struct bdi_work *bdi_alloc_work(struct writeback_control *wbc) |
1da177e4 | 202 | { |
03ba3782 JA |
203 | struct bdi_work *work; |
204 | ||
205 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
206 | if (work) | |
207 | bdi_work_init(work, wbc); | |
208 | ||
209 | return work; | |
210 | } | |
211 | ||
212 | void bdi_start_writeback(struct writeback_control *wbc) | |
213 | { | |
f0fad8a5 CH |
214 | /* |
215 | * WB_SYNC_NONE is opportunistic writeback. If this allocation fails, | |
216 | * bdi_queue_work() will wake up the thread and flush old data. This | |
217 | * should ensure some amount of progress in freeing memory. | |
218 | */ | |
219 | if (wbc->sync_mode != WB_SYNC_ALL) { | |
220 | struct bdi_work *w = bdi_alloc_work(wbc); | |
03ba3782 | 221 | |
f0fad8a5 CH |
222 | bdi_queue_work(wbc->bdi, w); |
223 | } else { | |
224 | struct bdi_work work; | |
03ba3782 | 225 | |
f0fad8a5 CH |
226 | bdi_work_init(&work, wbc); |
227 | work.state |= WS_ONSTACK; | |
03ba3782 | 228 | |
f0fad8a5 CH |
229 | bdi_queue_work(wbc->bdi, &work); |
230 | bdi_wait_on_work_clear(&work); | |
03ba3782 | 231 | } |
1da177e4 LT |
232 | } |
233 | ||
6610a0bc AM |
234 | /* |
235 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
236 | * furthest end of its superblock's dirty-inode list. | |
237 | * | |
238 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 239 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
240 | * the case then the inode must have been redirtied while it was being written |
241 | * out and we don't reset its dirtied_when. | |
242 | */ | |
243 | static void redirty_tail(struct inode *inode) | |
244 | { | |
03ba3782 | 245 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
6610a0bc | 246 | |
03ba3782 | 247 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 248 | struct inode *tail; |
6610a0bc | 249 | |
03ba3782 | 250 | tail = list_entry(wb->b_dirty.next, struct inode, i_list); |
66f3b8e2 | 251 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
252 | inode->dirtied_when = jiffies; |
253 | } | |
03ba3782 | 254 | list_move(&inode->i_list, &wb->b_dirty); |
6610a0bc AM |
255 | } |
256 | ||
c986d1e2 | 257 | /* |
66f3b8e2 | 258 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 259 | */ |
0e0f4fc2 | 260 | static void requeue_io(struct inode *inode) |
c986d1e2 | 261 | { |
03ba3782 JA |
262 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
263 | ||
264 | list_move(&inode->i_list, &wb->b_more_io); | |
c986d1e2 AM |
265 | } |
266 | ||
1c0eeaf5 JE |
267 | static void inode_sync_complete(struct inode *inode) |
268 | { | |
269 | /* | |
270 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
271 | */ | |
272 | smp_mb(); | |
273 | wake_up_bit(&inode->i_state, __I_SYNC); | |
274 | } | |
275 | ||
d2caa3c5 JL |
276 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
277 | { | |
278 | bool ret = time_after(inode->dirtied_when, t); | |
279 | #ifndef CONFIG_64BIT | |
280 | /* | |
281 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
282 | * It _appears_ to be in the future, but is actually in distant past. | |
283 | * This test is necessary to prevent such wrapped-around relative times | |
284 | * from permanently stopping the whole pdflush writeback. | |
285 | */ | |
286 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
287 | #endif | |
288 | return ret; | |
289 | } | |
290 | ||
2c136579 FW |
291 | /* |
292 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
293 | */ | |
294 | static void move_expired_inodes(struct list_head *delaying_queue, | |
295 | struct list_head *dispatch_queue, | |
296 | unsigned long *older_than_this) | |
297 | { | |
298 | while (!list_empty(delaying_queue)) { | |
299 | struct inode *inode = list_entry(delaying_queue->prev, | |
300 | struct inode, i_list); | |
301 | if (older_than_this && | |
d2caa3c5 | 302 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 FW |
303 | break; |
304 | list_move(&inode->i_list, dispatch_queue); | |
305 | } | |
306 | } | |
307 | ||
308 | /* | |
309 | * Queue all expired dirty inodes for io, eldest first. | |
310 | */ | |
03ba3782 | 311 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 312 | { |
03ba3782 JA |
313 | list_splice_init(&wb->b_more_io, wb->b_io.prev); |
314 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); | |
66f3b8e2 JA |
315 | } |
316 | ||
03ba3782 | 317 | static int write_inode(struct inode *inode, int sync) |
08d8e974 | 318 | { |
03ba3782 JA |
319 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
320 | return inode->i_sb->s_op->write_inode(inode, sync); | |
321 | return 0; | |
08d8e974 | 322 | } |
08d8e974 | 323 | |
1da177e4 | 324 | /* |
01c03194 CH |
325 | * Wait for writeback on an inode to complete. |
326 | */ | |
327 | static void inode_wait_for_writeback(struct inode *inode) | |
328 | { | |
329 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
330 | wait_queue_head_t *wqh; | |
331 | ||
332 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
333 | do { | |
334 | spin_unlock(&inode_lock); | |
335 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); | |
336 | spin_lock(&inode_lock); | |
337 | } while (inode->i_state & I_SYNC); | |
338 | } | |
339 | ||
340 | /* | |
341 | * Write out an inode's dirty pages. Called under inode_lock. Either the | |
342 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
343 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
344 | * | |
1da177e4 LT |
345 | * If `wait' is set, wait on the writeout. |
346 | * | |
347 | * The whole writeout design is quite complex and fragile. We want to avoid | |
348 | * starvation of particular inodes when others are being redirtied, prevent | |
349 | * livelocks, etc. | |
350 | * | |
351 | * Called under inode_lock. | |
352 | */ | |
353 | static int | |
01c03194 | 354 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 355 | { |
1da177e4 | 356 | struct address_space *mapping = inode->i_mapping; |
1da177e4 | 357 | int wait = wbc->sync_mode == WB_SYNC_ALL; |
01c03194 | 358 | unsigned dirty; |
1da177e4 LT |
359 | int ret; |
360 | ||
01c03194 CH |
361 | if (!atomic_read(&inode->i_count)) |
362 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
363 | else | |
364 | WARN_ON(inode->i_state & I_WILL_FREE); | |
365 | ||
366 | if (inode->i_state & I_SYNC) { | |
367 | /* | |
368 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 369 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
370 | * writeback can proceed with the other inodes on s_io. |
371 | * | |
372 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 373 | * completed a full scan of b_io. |
01c03194 CH |
374 | */ |
375 | if (!wait) { | |
376 | requeue_io(inode); | |
377 | return 0; | |
378 | } | |
379 | ||
380 | /* | |
381 | * It's a data-integrity sync. We must wait. | |
382 | */ | |
383 | inode_wait_for_writeback(inode); | |
384 | } | |
385 | ||
1c0eeaf5 | 386 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 387 | |
1c0eeaf5 | 388 | /* Set I_SYNC, reset I_DIRTY */ |
1da177e4 | 389 | dirty = inode->i_state & I_DIRTY; |
1c0eeaf5 | 390 | inode->i_state |= I_SYNC; |
1da177e4 LT |
391 | inode->i_state &= ~I_DIRTY; |
392 | ||
393 | spin_unlock(&inode_lock); | |
394 | ||
395 | ret = do_writepages(mapping, wbc); | |
396 | ||
397 | /* Don't write the inode if only I_DIRTY_PAGES was set */ | |
398 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
399 | int err = write_inode(inode, wait); | |
400 | if (ret == 0) | |
401 | ret = err; | |
402 | } | |
403 | ||
404 | if (wait) { | |
405 | int err = filemap_fdatawait(mapping); | |
406 | if (ret == 0) | |
407 | ret = err; | |
408 | } | |
409 | ||
410 | spin_lock(&inode_lock); | |
1c0eeaf5 | 411 | inode->i_state &= ~I_SYNC; |
84a89245 | 412 | if (!(inode->i_state & (I_FREEING | I_CLEAR))) { |
1da177e4 LT |
413 | if (!(inode->i_state & I_DIRTY) && |
414 | mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
415 | /* | |
416 | * We didn't write back all the pages. nfs_writepages() | |
417 | * sometimes bales out without doing anything. Redirty | |
66f3b8e2 | 418 | * the inode; Move it from b_io onto b_more_io/b_dirty. |
1b43ef91 AM |
419 | */ |
420 | /* | |
421 | * akpm: if the caller was the kupdate function we put | |
66f3b8e2 | 422 | * this inode at the head of b_dirty so it gets first |
1b43ef91 AM |
423 | * consideration. Otherwise, move it to the tail, for |
424 | * the reasons described there. I'm not really sure | |
425 | * how much sense this makes. Presumably I had a good | |
426 | * reasons for doing it this way, and I'd rather not | |
427 | * muck with it at present. | |
1da177e4 LT |
428 | */ |
429 | if (wbc->for_kupdate) { | |
430 | /* | |
2c136579 | 431 | * For the kupdate function we move the inode |
66f3b8e2 | 432 | * to b_more_io so it will get more writeout as |
2c136579 | 433 | * soon as the queue becomes uncongested. |
1da177e4 LT |
434 | */ |
435 | inode->i_state |= I_DIRTY_PAGES; | |
8bc3be27 FW |
436 | if (wbc->nr_to_write <= 0) { |
437 | /* | |
438 | * slice used up: queue for next turn | |
439 | */ | |
440 | requeue_io(inode); | |
441 | } else { | |
442 | /* | |
443 | * somehow blocked: retry later | |
444 | */ | |
445 | redirty_tail(inode); | |
446 | } | |
1da177e4 LT |
447 | } else { |
448 | /* | |
449 | * Otherwise fully redirty the inode so that | |
450 | * other inodes on this superblock will get some | |
451 | * writeout. Otherwise heavy writing to one | |
452 | * file would indefinitely suspend writeout of | |
453 | * all the other files. | |
454 | */ | |
455 | inode->i_state |= I_DIRTY_PAGES; | |
1b43ef91 | 456 | redirty_tail(inode); |
1da177e4 LT |
457 | } |
458 | } else if (inode->i_state & I_DIRTY) { | |
459 | /* | |
460 | * Someone redirtied the inode while were writing back | |
461 | * the pages. | |
462 | */ | |
6610a0bc | 463 | redirty_tail(inode); |
1da177e4 LT |
464 | } else if (atomic_read(&inode->i_count)) { |
465 | /* | |
466 | * The inode is clean, inuse | |
467 | */ | |
468 | list_move(&inode->i_list, &inode_in_use); | |
469 | } else { | |
470 | /* | |
471 | * The inode is clean, unused | |
472 | */ | |
473 | list_move(&inode->i_list, &inode_unused); | |
1da177e4 LT |
474 | } |
475 | } | |
1c0eeaf5 | 476 | inode_sync_complete(inode); |
1da177e4 LT |
477 | return ret; |
478 | } | |
479 | ||
03ba3782 JA |
480 | /* |
481 | * For WB_SYNC_NONE writeback, the caller does not have the sb pinned | |
482 | * before calling writeback. So make sure that we do pin it, so it doesn't | |
483 | * go away while we are writing inodes from it. | |
484 | * | |
485 | * Returns 0 if the super was successfully pinned (or pinning wasn't needed), | |
486 | * 1 if we failed. | |
487 | */ | |
488 | static int pin_sb_for_writeback(struct writeback_control *wbc, | |
489 | struct inode *inode) | |
490 | { | |
491 | struct super_block *sb = inode->i_sb; | |
492 | ||
493 | /* | |
494 | * Caller must already hold the ref for this | |
495 | */ | |
496 | if (wbc->sync_mode == WB_SYNC_ALL) { | |
497 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
498 | return 0; | |
499 | } | |
500 | ||
501 | spin_lock(&sb_lock); | |
502 | sb->s_count++; | |
503 | if (down_read_trylock(&sb->s_umount)) { | |
504 | if (sb->s_root) { | |
505 | spin_unlock(&sb_lock); | |
506 | return 0; | |
507 | } | |
508 | /* | |
509 | * umounted, drop rwsem again and fall through to failure | |
510 | */ | |
511 | up_read(&sb->s_umount); | |
512 | } | |
513 | ||
514 | sb->s_count--; | |
515 | spin_unlock(&sb_lock); | |
516 | return 1; | |
517 | } | |
518 | ||
519 | static void unpin_sb_for_writeback(struct writeback_control *wbc, | |
520 | struct inode *inode) | |
521 | { | |
522 | struct super_block *sb = inode->i_sb; | |
523 | ||
524 | if (wbc->sync_mode == WB_SYNC_ALL) | |
525 | return; | |
526 | ||
527 | up_read(&sb->s_umount); | |
528 | put_super(sb); | |
529 | } | |
530 | ||
531 | static void writeback_inodes_wb(struct bdi_writeback *wb, | |
532 | struct writeback_control *wbc) | |
1da177e4 | 533 | { |
03ba3782 | 534 | struct super_block *sb = wbc->sb; |
66f3b8e2 | 535 | const int is_blkdev_sb = sb_is_blkdev_sb(sb); |
1da177e4 LT |
536 | const unsigned long start = jiffies; /* livelock avoidance */ |
537 | ||
ae8547b0 | 538 | spin_lock(&inode_lock); |
1da177e4 | 539 | |
03ba3782 JA |
540 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) |
541 | queue_io(wb, wbc->older_than_this); | |
66f3b8e2 | 542 | |
03ba3782 JA |
543 | while (!list_empty(&wb->b_io)) { |
544 | struct inode *inode = list_entry(wb->b_io.prev, | |
1da177e4 | 545 | struct inode, i_list); |
1da177e4 LT |
546 | long pages_skipped; |
547 | ||
66f3b8e2 JA |
548 | /* |
549 | * super block given and doesn't match, skip this inode | |
550 | */ | |
551 | if (sb && sb != inode->i_sb) { | |
552 | redirty_tail(inode); | |
553 | continue; | |
554 | } | |
555 | ||
03ba3782 | 556 | if (!bdi_cap_writeback_dirty(wb->bdi)) { |
9852a0e7 | 557 | redirty_tail(inode); |
66f3b8e2 | 558 | if (is_blkdev_sb) { |
1da177e4 LT |
559 | /* |
560 | * Dirty memory-backed blockdev: the ramdisk | |
561 | * driver does this. Skip just this inode | |
562 | */ | |
563 | continue; | |
564 | } | |
565 | /* | |
566 | * Dirty memory-backed inode against a filesystem other | |
567 | * than the kernel-internal bdev filesystem. Skip the | |
568 | * entire superblock. | |
569 | */ | |
570 | break; | |
571 | } | |
572 | ||
84a89245 | 573 | if (inode->i_state & (I_NEW | I_WILL_FREE)) { |
7ef0d737 NP |
574 | requeue_io(inode); |
575 | continue; | |
576 | } | |
577 | ||
03ba3782 | 578 | if (wbc->nonblocking && bdi_write_congested(wb->bdi)) { |
1da177e4 | 579 | wbc->encountered_congestion = 1; |
66f3b8e2 | 580 | if (!is_blkdev_sb) |
1da177e4 | 581 | break; /* Skip a congested fs */ |
0e0f4fc2 | 582 | requeue_io(inode); |
1da177e4 LT |
583 | continue; /* Skip a congested blockdev */ |
584 | } | |
585 | ||
d2caa3c5 JL |
586 | /* |
587 | * Was this inode dirtied after sync_sb_inodes was called? | |
588 | * This keeps sync from extra jobs and livelock. | |
589 | */ | |
590 | if (inode_dirtied_after(inode, start)) | |
1da177e4 LT |
591 | break; |
592 | ||
03ba3782 JA |
593 | if (pin_sb_for_writeback(wbc, inode)) { |
594 | requeue_io(inode); | |
595 | continue; | |
596 | } | |
1da177e4 | 597 | |
84a89245 | 598 | BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); |
1da177e4 LT |
599 | __iget(inode); |
600 | pages_skipped = wbc->pages_skipped; | |
01c03194 | 601 | writeback_single_inode(inode, wbc); |
03ba3782 | 602 | unpin_sb_for_writeback(wbc, inode); |
1da177e4 LT |
603 | if (wbc->pages_skipped != pages_skipped) { |
604 | /* | |
605 | * writeback is not making progress due to locked | |
606 | * buffers. Skip this inode for now. | |
607 | */ | |
f57b9b7b | 608 | redirty_tail(inode); |
1da177e4 LT |
609 | } |
610 | spin_unlock(&inode_lock); | |
1da177e4 | 611 | iput(inode); |
4ffc8444 | 612 | cond_resched(); |
1da177e4 | 613 | spin_lock(&inode_lock); |
8bc3be27 FW |
614 | if (wbc->nr_to_write <= 0) { |
615 | wbc->more_io = 1; | |
1da177e4 | 616 | break; |
8bc3be27 | 617 | } |
03ba3782 | 618 | if (!list_empty(&wb->b_more_io)) |
8bc3be27 | 619 | wbc->more_io = 1; |
1da177e4 | 620 | } |
38f21977 | 621 | |
66f3b8e2 JA |
622 | spin_unlock(&inode_lock); |
623 | /* Leave any unwritten inodes on b_io */ | |
624 | } | |
625 | ||
03ba3782 JA |
626 | void writeback_inodes_wbc(struct writeback_control *wbc) |
627 | { | |
628 | struct backing_dev_info *bdi = wbc->bdi; | |
629 | ||
630 | writeback_inodes_wb(&bdi->wb, wbc); | |
631 | } | |
632 | ||
66f3b8e2 | 633 | /* |
03ba3782 JA |
634 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
635 | * operation. We do this so we don't hold I_SYNC against an inode for | |
636 | * enormous amounts of time, which would block a userspace task which has | |
637 | * been forced to throttle against that inode. Also, the code reevaluates | |
638 | * the dirty each time it has written this many pages. | |
639 | */ | |
640 | #define MAX_WRITEBACK_PAGES 1024 | |
641 | ||
642 | static inline bool over_bground_thresh(void) | |
643 | { | |
644 | unsigned long background_thresh, dirty_thresh; | |
645 | ||
646 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); | |
647 | ||
648 | return (global_page_state(NR_FILE_DIRTY) + | |
649 | global_page_state(NR_UNSTABLE_NFS) >= background_thresh); | |
650 | } | |
651 | ||
652 | /* | |
653 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 654 | * |
03ba3782 JA |
655 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
656 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
657 | * just walks the superblock inode list, writing back any inodes which are | |
658 | * older than a specific point in time. | |
66f3b8e2 | 659 | * |
03ba3782 JA |
660 | * Try to run once per dirty_writeback_interval. But if a writeback event |
661 | * takes longer than a dirty_writeback_interval interval, then leave a | |
662 | * one-second gap. | |
66f3b8e2 | 663 | * |
03ba3782 JA |
664 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
665 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 666 | */ |
c4a77a6c JA |
667 | static long wb_writeback(struct bdi_writeback *wb, |
668 | struct wb_writeback_args *args) | |
66f3b8e2 | 669 | { |
03ba3782 JA |
670 | struct writeback_control wbc = { |
671 | .bdi = wb->bdi, | |
c4a77a6c JA |
672 | .sb = args->sb, |
673 | .sync_mode = args->sync_mode, | |
03ba3782 | 674 | .older_than_this = NULL, |
c4a77a6c JA |
675 | .for_kupdate = args->for_kupdate, |
676 | .range_cyclic = args->range_cyclic, | |
03ba3782 JA |
677 | }; |
678 | unsigned long oldest_jif; | |
679 | long wrote = 0; | |
66f3b8e2 | 680 | |
03ba3782 JA |
681 | if (wbc.for_kupdate) { |
682 | wbc.older_than_this = &oldest_jif; | |
683 | oldest_jif = jiffies - | |
684 | msecs_to_jiffies(dirty_expire_interval * 10); | |
685 | } | |
c4a77a6c JA |
686 | if (!wbc.range_cyclic) { |
687 | wbc.range_start = 0; | |
688 | wbc.range_end = LLONG_MAX; | |
689 | } | |
38f21977 | 690 | |
03ba3782 JA |
691 | for (;;) { |
692 | /* | |
693 | * Don't flush anything for non-integrity writeback where | |
694 | * no nr_pages was given | |
695 | */ | |
c4a77a6c JA |
696 | if (!args->for_kupdate && args->nr_pages <= 0 && |
697 | args->sync_mode == WB_SYNC_NONE) | |
03ba3782 | 698 | break; |
66f3b8e2 | 699 | |
38f21977 | 700 | /* |
03ba3782 JA |
701 | * If no specific pages were given and this is just a |
702 | * periodic background writeout and we are below the | |
703 | * background dirty threshold, don't do anything | |
38f21977 | 704 | */ |
c4a77a6c JA |
705 | if (args->for_kupdate && args->nr_pages <= 0 && |
706 | !over_bground_thresh()) | |
03ba3782 | 707 | break; |
38f21977 | 708 | |
03ba3782 JA |
709 | wbc.more_io = 0; |
710 | wbc.encountered_congestion = 0; | |
711 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; | |
712 | wbc.pages_skipped = 0; | |
713 | writeback_inodes_wb(wb, &wbc); | |
c4a77a6c | 714 | args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
03ba3782 JA |
715 | wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
716 | ||
717 | /* | |
718 | * If we ran out of stuff to write, bail unless more_io got set | |
719 | */ | |
720 | if (wbc.nr_to_write > 0 || wbc.pages_skipped > 0) { | |
721 | if (wbc.more_io && !wbc.for_kupdate) | |
38f21977 | 722 | continue; |
03ba3782 JA |
723 | break; |
724 | } | |
725 | } | |
726 | ||
727 | return wrote; | |
728 | } | |
729 | ||
730 | /* | |
731 | * Return the next bdi_work struct that hasn't been processed by this | |
732 | * wb thread yet | |
733 | */ | |
734 | static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, | |
735 | struct bdi_writeback *wb) | |
736 | { | |
737 | struct bdi_work *work, *ret = NULL; | |
738 | ||
739 | rcu_read_lock(); | |
740 | ||
741 | list_for_each_entry_rcu(work, &bdi->work_list, list) { | |
742 | if (!test_and_clear_bit(wb->nr, &work->seen)) | |
743 | continue; | |
744 | ||
745 | ret = work; | |
746 | break; | |
747 | } | |
748 | ||
749 | rcu_read_unlock(); | |
750 | return ret; | |
751 | } | |
752 | ||
753 | static long wb_check_old_data_flush(struct bdi_writeback *wb) | |
754 | { | |
755 | unsigned long expired; | |
756 | long nr_pages; | |
757 | ||
758 | expired = wb->last_old_flush + | |
759 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
760 | if (time_before(jiffies, expired)) | |
761 | return 0; | |
762 | ||
763 | wb->last_old_flush = jiffies; | |
764 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
765 | global_page_state(NR_UNSTABLE_NFS) + | |
766 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); | |
767 | ||
c4a77a6c JA |
768 | if (nr_pages) { |
769 | struct wb_writeback_args args = { | |
770 | .nr_pages = nr_pages, | |
771 | .sync_mode = WB_SYNC_NONE, | |
772 | .for_kupdate = 1, | |
773 | .range_cyclic = 1, | |
774 | }; | |
775 | ||
776 | return wb_writeback(wb, &args); | |
777 | } | |
03ba3782 JA |
778 | |
779 | return 0; | |
780 | } | |
781 | ||
782 | /* | |
783 | * Retrieve work items and do the writeback they describe | |
784 | */ | |
785 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
786 | { | |
787 | struct backing_dev_info *bdi = wb->bdi; | |
788 | struct bdi_work *work; | |
c4a77a6c | 789 | long wrote = 0; |
03ba3782 JA |
790 | |
791 | while ((work = get_next_work_item(bdi, wb)) != NULL) { | |
c4a77a6c | 792 | struct wb_writeback_args args = work->args; |
03ba3782 JA |
793 | |
794 | /* | |
795 | * Override sync mode, in case we must wait for completion | |
796 | */ | |
797 | if (force_wait) | |
c4a77a6c | 798 | work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; |
03ba3782 JA |
799 | |
800 | /* | |
801 | * If this isn't a data integrity operation, just notify | |
802 | * that we have seen this work and we are now starting it. | |
803 | */ | |
c4a77a6c | 804 | if (args.sync_mode == WB_SYNC_NONE) |
03ba3782 JA |
805 | wb_clear_pending(wb, work); |
806 | ||
c4a77a6c | 807 | wrote += wb_writeback(wb, &args); |
03ba3782 JA |
808 | |
809 | /* | |
810 | * This is a data integrity writeback, so only do the | |
811 | * notification when we have completed the work. | |
812 | */ | |
c4a77a6c | 813 | if (args.sync_mode == WB_SYNC_ALL) |
03ba3782 JA |
814 | wb_clear_pending(wb, work); |
815 | } | |
816 | ||
817 | /* | |
818 | * Check for periodic writeback, kupdated() style | |
819 | */ | |
820 | wrote += wb_check_old_data_flush(wb); | |
821 | ||
822 | return wrote; | |
823 | } | |
824 | ||
825 | /* | |
826 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
827 | * wakes up periodically and does kupdated style flushing. | |
828 | */ | |
829 | int bdi_writeback_task(struct bdi_writeback *wb) | |
830 | { | |
831 | unsigned long last_active = jiffies; | |
832 | unsigned long wait_jiffies = -1UL; | |
833 | long pages_written; | |
834 | ||
835 | while (!kthread_should_stop()) { | |
836 | pages_written = wb_do_writeback(wb, 0); | |
837 | ||
838 | if (pages_written) | |
839 | last_active = jiffies; | |
840 | else if (wait_jiffies != -1UL) { | |
841 | unsigned long max_idle; | |
842 | ||
38f21977 | 843 | /* |
03ba3782 JA |
844 | * Longest period of inactivity that we tolerate. If we |
845 | * see dirty data again later, the task will get | |
846 | * recreated automatically. | |
38f21977 | 847 | */ |
03ba3782 JA |
848 | max_idle = max(5UL * 60 * HZ, wait_jiffies); |
849 | if (time_after(jiffies, max_idle + last_active)) | |
850 | break; | |
851 | } | |
852 | ||
853 | wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); | |
854 | set_current_state(TASK_INTERRUPTIBLE); | |
855 | schedule_timeout(wait_jiffies); | |
856 | try_to_freeze(); | |
857 | } | |
858 | ||
859 | return 0; | |
860 | } | |
861 | ||
862 | /* | |
863 | * Schedule writeback for all backing devices. Expensive! If this is a data | |
864 | * integrity operation, writeback will be complete when this returns. If | |
865 | * we are simply called for WB_SYNC_NONE, then writeback will merely be | |
866 | * scheduled to run. | |
867 | */ | |
868 | static void bdi_writeback_all(struct writeback_control *wbc) | |
869 | { | |
870 | const bool must_wait = wbc->sync_mode == WB_SYNC_ALL; | |
871 | struct backing_dev_info *bdi; | |
872 | struct bdi_work *work; | |
873 | LIST_HEAD(list); | |
874 | ||
875 | restart: | |
876 | spin_lock(&bdi_lock); | |
877 | ||
878 | list_for_each_entry(bdi, &bdi_list, bdi_list) { | |
879 | struct bdi_work *work; | |
880 | ||
881 | if (!bdi_has_dirty_io(bdi)) | |
882 | continue; | |
38f21977 | 883 | |
03ba3782 JA |
884 | /* |
885 | * If work allocation fails, do the writes inline. We drop | |
886 | * the lock and restart the list writeout. This should be OK, | |
887 | * since this happens rarely and because the writeout should | |
888 | * eventually make more free memory available. | |
889 | */ | |
890 | work = bdi_alloc_work(wbc); | |
891 | if (!work) { | |
892 | struct writeback_control __wbc; | |
38f21977 | 893 | |
03ba3782 JA |
894 | /* |
895 | * Not a data integrity writeout, just continue | |
896 | */ | |
897 | if (!must_wait) | |
898 | continue; | |
38f21977 | 899 | |
03ba3782 JA |
900 | spin_unlock(&bdi_lock); |
901 | __wbc = *wbc; | |
902 | __wbc.bdi = bdi; | |
903 | writeback_inodes_wbc(&__wbc); | |
904 | goto restart; | |
38f21977 | 905 | } |
03ba3782 JA |
906 | if (must_wait) |
907 | list_add_tail(&work->wait_list, &list); | |
908 | ||
909 | bdi_queue_work(bdi, work); | |
910 | } | |
911 | ||
912 | spin_unlock(&bdi_lock); | |
913 | ||
914 | /* | |
915 | * If this is for WB_SYNC_ALL, wait for pending work to complete | |
916 | * before returning. | |
917 | */ | |
918 | while (!list_empty(&list)) { | |
919 | work = list_entry(list.next, struct bdi_work, wait_list); | |
920 | list_del(&work->wait_list); | |
921 | bdi_wait_on_work_clear(work); | |
922 | call_rcu(&work->rcu_head, bdi_work_free); | |
66f3b8e2 | 923 | } |
1da177e4 LT |
924 | } |
925 | ||
926 | /* | |
03ba3782 JA |
927 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
928 | * the whole world. | |
929 | */ | |
930 | void wakeup_flusher_threads(long nr_pages) | |
931 | { | |
932 | struct writeback_control wbc = { | |
933 | .sync_mode = WB_SYNC_NONE, | |
934 | .older_than_this = NULL, | |
935 | .range_cyclic = 1, | |
936 | }; | |
937 | ||
938 | if (nr_pages == 0) | |
939 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
940 | global_page_state(NR_UNSTABLE_NFS); | |
941 | wbc.nr_to_write = nr_pages; | |
942 | bdi_writeback_all(&wbc); | |
943 | } | |
944 | ||
945 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) | |
946 | { | |
947 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
948 | struct dentry *dentry; | |
949 | const char *name = "?"; | |
950 | ||
951 | dentry = d_find_alias(inode); | |
952 | if (dentry) { | |
953 | spin_lock(&dentry->d_lock); | |
954 | name = (const char *) dentry->d_name.name; | |
955 | } | |
956 | printk(KERN_DEBUG | |
957 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
958 | current->comm, task_pid_nr(current), inode->i_ino, | |
959 | name, inode->i_sb->s_id); | |
960 | if (dentry) { | |
961 | spin_unlock(&dentry->d_lock); | |
962 | dput(dentry); | |
963 | } | |
964 | } | |
965 | } | |
966 | ||
967 | /** | |
968 | * __mark_inode_dirty - internal function | |
969 | * @inode: inode to mark | |
970 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
971 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
972 | * mark_inode_dirty_sync. | |
1da177e4 | 973 | * |
03ba3782 JA |
974 | * Put the inode on the super block's dirty list. |
975 | * | |
976 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
977 | * dirty list only if it is hashed or if it refers to a blockdev. | |
978 | * If it was not hashed, it will never be added to the dirty list | |
979 | * even if it is later hashed, as it will have been marked dirty already. | |
980 | * | |
981 | * In short, make sure you hash any inodes _before_ you start marking | |
982 | * them dirty. | |
1da177e4 | 983 | * |
03ba3782 JA |
984 | * This function *must* be atomic for the I_DIRTY_PAGES case - |
985 | * set_page_dirty() is called under spinlock in several places. | |
1da177e4 | 986 | * |
03ba3782 JA |
987 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
988 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
989 | * the kernel-internal blockdev inode represents the dirtying time of the | |
990 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
991 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
992 | * blockdev inode. | |
1da177e4 | 993 | */ |
03ba3782 | 994 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 995 | { |
03ba3782 | 996 | struct super_block *sb = inode->i_sb; |
1da177e4 | 997 | |
03ba3782 JA |
998 | /* |
999 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1000 | * dirty the inode itself | |
1001 | */ | |
1002 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1003 | if (sb->s_op->dirty_inode) | |
1004 | sb->s_op->dirty_inode(inode); | |
1005 | } | |
1006 | ||
1007 | /* | |
1008 | * make sure that changes are seen by all cpus before we test i_state | |
1009 | * -- mikulas | |
1010 | */ | |
1011 | smp_mb(); | |
1012 | ||
1013 | /* avoid the locking if we can */ | |
1014 | if ((inode->i_state & flags) == flags) | |
1015 | return; | |
1016 | ||
1017 | if (unlikely(block_dump)) | |
1018 | block_dump___mark_inode_dirty(inode); | |
1019 | ||
1020 | spin_lock(&inode_lock); | |
1021 | if ((inode->i_state & flags) != flags) { | |
1022 | const int was_dirty = inode->i_state & I_DIRTY; | |
1023 | ||
1024 | inode->i_state |= flags; | |
1025 | ||
1026 | /* | |
1027 | * If the inode is being synced, just update its dirty state. | |
1028 | * The unlocker will place the inode on the appropriate | |
1029 | * superblock list, based upon its state. | |
1030 | */ | |
1031 | if (inode->i_state & I_SYNC) | |
1032 | goto out; | |
1033 | ||
1034 | /* | |
1035 | * Only add valid (hashed) inodes to the superblock's | |
1036 | * dirty list. Add blockdev inodes as well. | |
1037 | */ | |
1038 | if (!S_ISBLK(inode->i_mode)) { | |
1039 | if (hlist_unhashed(&inode->i_hash)) | |
1040 | goto out; | |
1041 | } | |
1042 | if (inode->i_state & (I_FREEING|I_CLEAR)) | |
1043 | goto out; | |
1044 | ||
1045 | /* | |
1046 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1047 | * reposition it (that would break b_dirty time-ordering). | |
1048 | */ | |
1049 | if (!was_dirty) { | |
1050 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; | |
500b067c JA |
1051 | struct backing_dev_info *bdi = wb->bdi; |
1052 | ||
1053 | if (bdi_cap_writeback_dirty(bdi) && | |
1054 | !test_bit(BDI_registered, &bdi->state)) { | |
1055 | WARN_ON(1); | |
1056 | printk(KERN_ERR "bdi-%s not registered\n", | |
1057 | bdi->name); | |
1058 | } | |
03ba3782 JA |
1059 | |
1060 | inode->dirtied_when = jiffies; | |
1061 | list_move(&inode->i_list, &wb->b_dirty); | |
1da177e4 | 1062 | } |
1da177e4 | 1063 | } |
03ba3782 JA |
1064 | out: |
1065 | spin_unlock(&inode_lock); | |
1066 | } | |
1067 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1068 | ||
1069 | /* | |
1070 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1071 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1072 | * | |
1073 | * If older_than_this is non-NULL, then only write out inodes which | |
1074 | * had their first dirtying at a time earlier than *older_than_this. | |
1075 | * | |
1076 | * If we're a pdlfush thread, then implement pdflush collision avoidance | |
1077 | * against the entire list. | |
1078 | * | |
1079 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. | |
1080 | * This function assumes that the blockdev superblock's inodes are backed by | |
1081 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1082 | * assume that all inodes are backed by the same queue. | |
1083 | * | |
1084 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1085 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1086 | * on the writer throttling path, and we get decent balancing between many | |
1087 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1088 | */ | |
1089 | static void wait_sb_inodes(struct writeback_control *wbc) | |
1090 | { | |
1091 | struct inode *inode, *old_inode = NULL; | |
1092 | ||
1093 | /* | |
1094 | * We need to be protected against the filesystem going from | |
1095 | * r/o to r/w or vice versa. | |
1096 | */ | |
1097 | WARN_ON(!rwsem_is_locked(&wbc->sb->s_umount)); | |
1098 | ||
1099 | spin_lock(&inode_lock); | |
1100 | ||
1101 | /* | |
1102 | * Data integrity sync. Must wait for all pages under writeback, | |
1103 | * because there may have been pages dirtied before our sync | |
1104 | * call, but which had writeout started before we write it out. | |
1105 | * In which case, the inode may not be on the dirty list, but | |
1106 | * we still have to wait for that writeout. | |
1107 | */ | |
1108 | list_for_each_entry(inode, &wbc->sb->s_inodes, i_sb_list) { | |
1109 | struct address_space *mapping; | |
1110 | ||
1111 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) | |
1112 | continue; | |
1113 | mapping = inode->i_mapping; | |
1114 | if (mapping->nrpages == 0) | |
1115 | continue; | |
1116 | __iget(inode); | |
1117 | spin_unlock(&inode_lock); | |
1118 | /* | |
1119 | * We hold a reference to 'inode' so it couldn't have | |
1120 | * been removed from s_inodes list while we dropped the | |
1121 | * inode_lock. We cannot iput the inode now as we can | |
1122 | * be holding the last reference and we cannot iput it | |
1123 | * under inode_lock. So we keep the reference and iput | |
1124 | * it later. | |
1125 | */ | |
1126 | iput(old_inode); | |
1127 | old_inode = inode; | |
1128 | ||
1129 | filemap_fdatawait(mapping); | |
1130 | ||
1131 | cond_resched(); | |
1132 | ||
1133 | spin_lock(&inode_lock); | |
1134 | } | |
1135 | spin_unlock(&inode_lock); | |
1136 | iput(old_inode); | |
1da177e4 LT |
1137 | } |
1138 | ||
d8a8559c JA |
1139 | /** |
1140 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1141 | * @sb: the superblock | |
1da177e4 | 1142 | * |
d8a8559c JA |
1143 | * Start writeback on some inodes on this super_block. No guarantees are made |
1144 | * on how many (if any) will be written, and this function does not wait | |
1145 | * for IO completion of submitted IO. The number of pages submitted is | |
1146 | * returned. | |
1da177e4 | 1147 | */ |
d8a8559c | 1148 | long writeback_inodes_sb(struct super_block *sb) |
1da177e4 LT |
1149 | { |
1150 | struct writeback_control wbc = { | |
03ba3782 | 1151 | .sb = sb, |
d8a8559c | 1152 | .sync_mode = WB_SYNC_NONE, |
111ebb6e OH |
1153 | .range_start = 0, |
1154 | .range_end = LLONG_MAX, | |
1da177e4 | 1155 | }; |
d8a8559c JA |
1156 | unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); |
1157 | unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); | |
1158 | long nr_to_write; | |
1da177e4 | 1159 | |
d8a8559c | 1160 | nr_to_write = nr_dirty + nr_unstable + |
38f21977 | 1161 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); |
38f21977 | 1162 | |
d8a8559c | 1163 | wbc.nr_to_write = nr_to_write; |
03ba3782 | 1164 | bdi_writeback_all(&wbc); |
d8a8559c JA |
1165 | return nr_to_write - wbc.nr_to_write; |
1166 | } | |
1167 | EXPORT_SYMBOL(writeback_inodes_sb); | |
1168 | ||
1169 | /** | |
1170 | * sync_inodes_sb - sync sb inode pages | |
1171 | * @sb: the superblock | |
1172 | * | |
1173 | * This function writes and waits on any dirty inode belonging to this | |
1174 | * super_block. The number of pages synced is returned. | |
1175 | */ | |
1176 | long sync_inodes_sb(struct super_block *sb) | |
1177 | { | |
1178 | struct writeback_control wbc = { | |
03ba3782 | 1179 | .sb = sb, |
d8a8559c JA |
1180 | .sync_mode = WB_SYNC_ALL, |
1181 | .range_start = 0, | |
1182 | .range_end = LLONG_MAX, | |
1183 | }; | |
1184 | long nr_to_write = LONG_MAX; /* doesn't actually matter */ | |
1185 | ||
1186 | wbc.nr_to_write = nr_to_write; | |
03ba3782 JA |
1187 | bdi_writeback_all(&wbc); |
1188 | wait_sb_inodes(&wbc); | |
d8a8559c | 1189 | return nr_to_write - wbc.nr_to_write; |
1da177e4 | 1190 | } |
d8a8559c | 1191 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1192 | |
1da177e4 | 1193 | /** |
7f04c26d AA |
1194 | * write_inode_now - write an inode to disk |
1195 | * @inode: inode to write to disk | |
1196 | * @sync: whether the write should be synchronous or not | |
1197 | * | |
1198 | * This function commits an inode to disk immediately if it is dirty. This is | |
1199 | * primarily needed by knfsd. | |
1da177e4 | 1200 | * |
7f04c26d | 1201 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1202 | */ |
1da177e4 LT |
1203 | int write_inode_now(struct inode *inode, int sync) |
1204 | { | |
1205 | int ret; | |
1206 | struct writeback_control wbc = { | |
1207 | .nr_to_write = LONG_MAX, | |
18914b18 | 1208 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1209 | .range_start = 0, |
1210 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1211 | }; |
1212 | ||
1213 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1214 | wbc.nr_to_write = 0; |
1da177e4 LT |
1215 | |
1216 | might_sleep(); | |
1217 | spin_lock(&inode_lock); | |
01c03194 | 1218 | ret = writeback_single_inode(inode, &wbc); |
1da177e4 LT |
1219 | spin_unlock(&inode_lock); |
1220 | if (sync) | |
1c0eeaf5 | 1221 | inode_sync_wait(inode); |
1da177e4 LT |
1222 | return ret; |
1223 | } | |
1224 | EXPORT_SYMBOL(write_inode_now); | |
1225 | ||
1226 | /** | |
1227 | * sync_inode - write an inode and its pages to disk. | |
1228 | * @inode: the inode to sync | |
1229 | * @wbc: controls the writeback mode | |
1230 | * | |
1231 | * sync_inode() will write an inode and its pages to disk. It will also | |
1232 | * correctly update the inode on its superblock's dirty inode lists and will | |
1233 | * update inode->i_state. | |
1234 | * | |
1235 | * The caller must have a ref on the inode. | |
1236 | */ | |
1237 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1238 | { | |
1239 | int ret; | |
1240 | ||
1241 | spin_lock(&inode_lock); | |
01c03194 | 1242 | ret = writeback_single_inode(inode, wbc); |
1da177e4 LT |
1243 | spin_unlock(&inode_lock); |
1244 | return ret; | |
1245 | } | |
1246 | EXPORT_SYMBOL(sync_inode); |