]>
Commit | Line | Data |
---|---|---|
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 | * | |
11 | * 10Apr2002 Andrew Morton | |
12 | * Split out of fs/inode.c | |
13 | * Additions for address_space-based writeback | |
14 | */ | |
15 | ||
16 | #include <linux/kernel.h> | |
17 | #include <linux/export.h> | |
18 | #include <linux/spinlock.h> | |
19 | #include <linux/slab.h> | |
20 | #include <linux/sched.h> | |
21 | #include <linux/fs.h> | |
22 | #include <linux/mm.h> | |
23 | #include <linux/pagemap.h> | |
24 | #include <linux/kthread.h> | |
25 | #include <linux/writeback.h> | |
26 | #include <linux/blkdev.h> | |
27 | #include <linux/backing-dev.h> | |
28 | #include <linux/tracepoint.h> | |
29 | #include "internal.h" | |
30 | ||
31 | /* | |
32 | * 4MB minimal write chunk size | |
33 | */ | |
34 | #define MIN_WRITEBACK_PAGES (4096UL >> (PAGE_CACHE_SHIFT - 10)) | |
35 | ||
36 | /* | |
37 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
38 | */ | |
39 | struct wb_writeback_work { | |
40 | long nr_pages; | |
41 | struct super_block *sb; | |
42 | unsigned long *older_than_this; | |
43 | enum writeback_sync_modes sync_mode; | |
44 | unsigned int tagged_writepages:1; | |
45 | unsigned int for_kupdate:1; | |
46 | unsigned int range_cyclic:1; | |
47 | unsigned int for_background:1; | |
48 | enum wb_reason reason; /* why was writeback initiated? */ | |
49 | ||
50 | struct list_head list; /* pending work list */ | |
51 | struct completion *done; /* set if the caller waits */ | |
52 | }; | |
53 | ||
54 | /** | |
55 | * writeback_in_progress - determine whether there is writeback in progress | |
56 | * @bdi: the device's backing_dev_info structure. | |
57 | * | |
58 | * Determine whether there is writeback waiting to be handled against a | |
59 | * backing device. | |
60 | */ | |
61 | int writeback_in_progress(struct backing_dev_info *bdi) | |
62 | { | |
63 | return test_bit(BDI_writeback_running, &bdi->state); | |
64 | } | |
65 | EXPORT_SYMBOL(writeback_in_progress); | |
66 | ||
67 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) | |
68 | { | |
69 | struct super_block *sb = inode->i_sb; | |
70 | ||
71 | if (strcmp(sb->s_type->name, "bdev") == 0) | |
72 | return inode->i_mapping->backing_dev_info; | |
73 | ||
74 | return sb->s_bdi; | |
75 | } | |
76 | ||
77 | static inline struct inode *wb_inode(struct list_head *head) | |
78 | { | |
79 | return list_entry(head, struct inode, i_wb_list); | |
80 | } | |
81 | ||
82 | /* | |
83 | * Include the creation of the trace points after defining the | |
84 | * wb_writeback_work structure and inline functions so that the definition | |
85 | * remains local to this file. | |
86 | */ | |
87 | #define CREATE_TRACE_POINTS | |
88 | #include <trace/events/writeback.h> | |
89 | ||
90 | static void bdi_queue_work(struct backing_dev_info *bdi, | |
91 | struct wb_writeback_work *work) | |
92 | { | |
93 | trace_writeback_queue(bdi, work); | |
94 | ||
95 | spin_lock_bh(&bdi->wb_lock); | |
96 | list_add_tail(&work->list, &bdi->work_list); | |
97 | spin_unlock_bh(&bdi->wb_lock); | |
98 | ||
99 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); | |
100 | } | |
101 | ||
102 | static void | |
103 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
104 | bool range_cyclic, enum wb_reason reason) | |
105 | { | |
106 | struct wb_writeback_work *work; | |
107 | ||
108 | /* | |
109 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
110 | * wakeup the thread for old dirty data writeback | |
111 | */ | |
112 | work = kzalloc(sizeof(*work), GFP_ATOMIC); | |
113 | if (!work) { | |
114 | trace_writeback_nowork(bdi); | |
115 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); | |
116 | return; | |
117 | } | |
118 | ||
119 | work->sync_mode = WB_SYNC_NONE; | |
120 | work->nr_pages = nr_pages; | |
121 | work->range_cyclic = range_cyclic; | |
122 | work->reason = reason; | |
123 | ||
124 | bdi_queue_work(bdi, work); | |
125 | } | |
126 | ||
127 | /** | |
128 | * bdi_start_writeback - start writeback | |
129 | * @bdi: the backing device to write from | |
130 | * @nr_pages: the number of pages to write | |
131 | * @reason: reason why some writeback work was initiated | |
132 | * | |
133 | * Description: | |
134 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
135 | * started when this function returns, we make no guarantees on | |
136 | * completion. Caller need not hold sb s_umount semaphore. | |
137 | * | |
138 | */ | |
139 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
140 | enum wb_reason reason) | |
141 | { | |
142 | __bdi_start_writeback(bdi, nr_pages, true, reason); | |
143 | } | |
144 | ||
145 | /** | |
146 | * bdi_start_background_writeback - start background writeback | |
147 | * @bdi: the backing device to write from | |
148 | * | |
149 | * Description: | |
150 | * This makes sure WB_SYNC_NONE background writeback happens. When | |
151 | * this function returns, it is only guaranteed that for given BDI | |
152 | * some IO is happening if we are over background dirty threshold. | |
153 | * Caller need not hold sb s_umount semaphore. | |
154 | */ | |
155 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
156 | { | |
157 | /* | |
158 | * We just wake up the flusher thread. It will perform background | |
159 | * writeback as soon as there is no other work to do. | |
160 | */ | |
161 | trace_writeback_wake_background(bdi); | |
162 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); | |
163 | } | |
164 | ||
165 | /* | |
166 | * Remove the inode from the writeback list it is on. | |
167 | */ | |
168 | void inode_wb_list_del(struct inode *inode) | |
169 | { | |
170 | struct backing_dev_info *bdi = inode_to_bdi(inode); | |
171 | ||
172 | spin_lock(&bdi->wb.list_lock); | |
173 | list_del_init(&inode->i_wb_list); | |
174 | spin_unlock(&bdi->wb.list_lock); | |
175 | } | |
176 | ||
177 | /* | |
178 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
179 | * furthest end of its superblock's dirty-inode list. | |
180 | * | |
181 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
182 | * already the most-recently-dirtied inode on the b_dirty list. If that is | |
183 | * the case then the inode must have been redirtied while it was being written | |
184 | * out and we don't reset its dirtied_when. | |
185 | */ | |
186 | static void redirty_tail(struct inode *inode, struct bdi_writeback *wb) | |
187 | { | |
188 | assert_spin_locked(&wb->list_lock); | |
189 | if (!list_empty(&wb->b_dirty)) { | |
190 | struct inode *tail; | |
191 | ||
192 | tail = wb_inode(wb->b_dirty.next); | |
193 | if (time_before(inode->dirtied_when, tail->dirtied_when)) | |
194 | inode->dirtied_when = jiffies; | |
195 | } | |
196 | list_move(&inode->i_wb_list, &wb->b_dirty); | |
197 | } | |
198 | ||
199 | /* | |
200 | * requeue inode for re-scanning after bdi->b_io list is exhausted. | |
201 | */ | |
202 | static void requeue_io(struct inode *inode, struct bdi_writeback *wb) | |
203 | { | |
204 | assert_spin_locked(&wb->list_lock); | |
205 | list_move(&inode->i_wb_list, &wb->b_more_io); | |
206 | } | |
207 | ||
208 | static void inode_sync_complete(struct inode *inode) | |
209 | { | |
210 | inode->i_state &= ~I_SYNC; | |
211 | /* If inode is clean an unused, put it into LRU now... */ | |
212 | inode_add_lru(inode); | |
213 | /* Waiters must see I_SYNC cleared before being woken up */ | |
214 | smp_mb(); | |
215 | wake_up_bit(&inode->i_state, __I_SYNC); | |
216 | } | |
217 | ||
218 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) | |
219 | { | |
220 | bool ret = time_after(inode->dirtied_when, t); | |
221 | #ifndef CONFIG_64BIT | |
222 | /* | |
223 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
224 | * It _appears_ to be in the future, but is actually in distant past. | |
225 | * This test is necessary to prevent such wrapped-around relative times | |
226 | * from permanently stopping the whole bdi writeback. | |
227 | */ | |
228 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
229 | #endif | |
230 | return ret; | |
231 | } | |
232 | ||
233 | /* | |
234 | * Move expired (dirtied before work->older_than_this) dirty inodes from | |
235 | * @delaying_queue to @dispatch_queue. | |
236 | */ | |
237 | static int move_expired_inodes(struct list_head *delaying_queue, | |
238 | struct list_head *dispatch_queue, | |
239 | struct wb_writeback_work *work) | |
240 | { | |
241 | LIST_HEAD(tmp); | |
242 | struct list_head *pos, *node; | |
243 | struct super_block *sb = NULL; | |
244 | struct inode *inode; | |
245 | int do_sb_sort = 0; | |
246 | int moved = 0; | |
247 | ||
248 | while (!list_empty(delaying_queue)) { | |
249 | inode = wb_inode(delaying_queue->prev); | |
250 | if (work->older_than_this && | |
251 | inode_dirtied_after(inode, *work->older_than_this)) | |
252 | break; | |
253 | if (sb && sb != inode->i_sb) | |
254 | do_sb_sort = 1; | |
255 | sb = inode->i_sb; | |
256 | list_move(&inode->i_wb_list, &tmp); | |
257 | moved++; | |
258 | } | |
259 | ||
260 | /* just one sb in list, splice to dispatch_queue and we're done */ | |
261 | if (!do_sb_sort) { | |
262 | list_splice(&tmp, dispatch_queue); | |
263 | goto out; | |
264 | } | |
265 | ||
266 | /* Move inodes from one superblock together */ | |
267 | while (!list_empty(&tmp)) { | |
268 | sb = wb_inode(tmp.prev)->i_sb; | |
269 | list_for_each_prev_safe(pos, node, &tmp) { | |
270 | inode = wb_inode(pos); | |
271 | if (inode->i_sb == sb) | |
272 | list_move(&inode->i_wb_list, dispatch_queue); | |
273 | } | |
274 | } | |
275 | out: | |
276 | return moved; | |
277 | } | |
278 | ||
279 | /* | |
280 | * Queue all expired dirty inodes for io, eldest first. | |
281 | * Before | |
282 | * newly dirtied b_dirty b_io b_more_io | |
283 | * =============> gf edc BA | |
284 | * After | |
285 | * newly dirtied b_dirty b_io b_more_io | |
286 | * =============> g fBAedc | |
287 | * | | |
288 | * +--> dequeue for IO | |
289 | */ | |
290 | static void queue_io(struct bdi_writeback *wb, struct wb_writeback_work *work) | |
291 | { | |
292 | int moved; | |
293 | assert_spin_locked(&wb->list_lock); | |
294 | list_splice_init(&wb->b_more_io, &wb->b_io); | |
295 | moved = move_expired_inodes(&wb->b_dirty, &wb->b_io, work); | |
296 | trace_writeback_queue_io(wb, work, moved); | |
297 | } | |
298 | ||
299 | static int write_inode(struct inode *inode, struct writeback_control *wbc) | |
300 | { | |
301 | int ret; | |
302 | ||
303 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) { | |
304 | trace_writeback_write_inode_start(inode, wbc); | |
305 | ret = inode->i_sb->s_op->write_inode(inode, wbc); | |
306 | trace_writeback_write_inode(inode, wbc); | |
307 | return ret; | |
308 | } | |
309 | return 0; | |
310 | } | |
311 | ||
312 | /* | |
313 | * Wait for writeback on an inode to complete. Called with i_lock held. | |
314 | * Caller must make sure inode cannot go away when we drop i_lock. | |
315 | */ | |
316 | static void __inode_wait_for_writeback(struct inode *inode) | |
317 | __releases(inode->i_lock) | |
318 | __acquires(inode->i_lock) | |
319 | { | |
320 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
321 | wait_queue_head_t *wqh; | |
322 | ||
323 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
324 | while (inode->i_state & I_SYNC) { | |
325 | spin_unlock(&inode->i_lock); | |
326 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); | |
327 | spin_lock(&inode->i_lock); | |
328 | } | |
329 | } | |
330 | ||
331 | /* | |
332 | * Wait for writeback on an inode to complete. Caller must have inode pinned. | |
333 | */ | |
334 | void inode_wait_for_writeback(struct inode *inode) | |
335 | { | |
336 | spin_lock(&inode->i_lock); | |
337 | __inode_wait_for_writeback(inode); | |
338 | spin_unlock(&inode->i_lock); | |
339 | } | |
340 | ||
341 | /* | |
342 | * Sleep until I_SYNC is cleared. This function must be called with i_lock | |
343 | * held and drops it. It is aimed for callers not holding any inode reference | |
344 | * so once i_lock is dropped, inode can go away. | |
345 | */ | |
346 | static void inode_sleep_on_writeback(struct inode *inode) | |
347 | __releases(inode->i_lock) | |
348 | { | |
349 | DEFINE_WAIT(wait); | |
350 | wait_queue_head_t *wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
351 | int sleep; | |
352 | ||
353 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
354 | sleep = inode->i_state & I_SYNC; | |
355 | spin_unlock(&inode->i_lock); | |
356 | if (sleep) | |
357 | schedule(); | |
358 | finish_wait(wqh, &wait); | |
359 | } | |
360 | ||
361 | /* | |
362 | * Find proper writeback list for the inode depending on its current state and | |
363 | * possibly also change of its state while we were doing writeback. Here we | |
364 | * handle things such as livelock prevention or fairness of writeback among | |
365 | * inodes. This function can be called only by flusher thread - noone else | |
366 | * processes all inodes in writeback lists and requeueing inodes behind flusher | |
367 | * thread's back can have unexpected consequences. | |
368 | */ | |
369 | static void requeue_inode(struct inode *inode, struct bdi_writeback *wb, | |
370 | struct writeback_control *wbc) | |
371 | { | |
372 | if (inode->i_state & I_FREEING) | |
373 | return; | |
374 | ||
375 | /* | |
376 | * Sync livelock prevention. Each inode is tagged and synced in one | |
377 | * shot. If still dirty, it will be redirty_tail()'ed below. Update | |
378 | * the dirty time to prevent enqueue and sync it again. | |
379 | */ | |
380 | if ((inode->i_state & I_DIRTY) && | |
381 | (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages)) | |
382 | inode->dirtied_when = jiffies; | |
383 | ||
384 | if (wbc->pages_skipped) { | |
385 | /* | |
386 | * writeback is not making progress due to locked | |
387 | * buffers. Skip this inode for now. | |
388 | */ | |
389 | redirty_tail(inode, wb); | |
390 | return; | |
391 | } | |
392 | ||
393 | if (mapping_tagged(inode->i_mapping, PAGECACHE_TAG_DIRTY)) { | |
394 | /* | |
395 | * We didn't write back all the pages. nfs_writepages() | |
396 | * sometimes bales out without doing anything. | |
397 | */ | |
398 | if (wbc->nr_to_write <= 0) { | |
399 | /* Slice used up. Queue for next turn. */ | |
400 | requeue_io(inode, wb); | |
401 | } else { | |
402 | /* | |
403 | * Writeback blocked by something other than | |
404 | * congestion. Delay the inode for some time to | |
405 | * avoid spinning on the CPU (100% iowait) | |
406 | * retrying writeback of the dirty page/inode | |
407 | * that cannot be performed immediately. | |
408 | */ | |
409 | redirty_tail(inode, wb); | |
410 | } | |
411 | } else if (inode->i_state & I_DIRTY) { | |
412 | /* | |
413 | * Filesystems can dirty the inode during writeback operations, | |
414 | * such as delayed allocation during submission or metadata | |
415 | * updates after data IO completion. | |
416 | */ | |
417 | redirty_tail(inode, wb); | |
418 | } else { | |
419 | /* The inode is clean. Remove from writeback lists. */ | |
420 | list_del_init(&inode->i_wb_list); | |
421 | } | |
422 | } | |
423 | ||
424 | /* | |
425 | * Write out an inode and its dirty pages. Do not update the writeback list | |
426 | * linkage. That is left to the caller. The caller is also responsible for | |
427 | * setting I_SYNC flag and calling inode_sync_complete() to clear it. | |
428 | */ | |
429 | static int | |
430 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) | |
431 | { | |
432 | struct address_space *mapping = inode->i_mapping; | |
433 | long nr_to_write = wbc->nr_to_write; | |
434 | unsigned dirty; | |
435 | int ret; | |
436 | ||
437 | WARN_ON(!(inode->i_state & I_SYNC)); | |
438 | ||
439 | trace_writeback_single_inode_start(inode, wbc, nr_to_write); | |
440 | ||
441 | ret = do_writepages(mapping, wbc); | |
442 | ||
443 | /* | |
444 | * Make sure to wait on the data before writing out the metadata. | |
445 | * This is important for filesystems that modify metadata on data | |
446 | * I/O completion. | |
447 | */ | |
448 | if (wbc->sync_mode == WB_SYNC_ALL) { | |
449 | int err = filemap_fdatawait(mapping); | |
450 | if (ret == 0) | |
451 | ret = err; | |
452 | } | |
453 | ||
454 | /* | |
455 | * Some filesystems may redirty the inode during the writeback | |
456 | * due to delalloc, clear dirty metadata flags right before | |
457 | * write_inode() | |
458 | */ | |
459 | spin_lock(&inode->i_lock); | |
460 | /* Clear I_DIRTY_PAGES if we've written out all dirty pages */ | |
461 | if (!mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) | |
462 | inode->i_state &= ~I_DIRTY_PAGES; | |
463 | dirty = inode->i_state & I_DIRTY; | |
464 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
465 | spin_unlock(&inode->i_lock); | |
466 | /* Don't write the inode if only I_DIRTY_PAGES was set */ | |
467 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
468 | int err = write_inode(inode, wbc); | |
469 | if (ret == 0) | |
470 | ret = err; | |
471 | } | |
472 | trace_writeback_single_inode(inode, wbc, nr_to_write); | |
473 | return ret; | |
474 | } | |
475 | ||
476 | /* | |
477 | * Write out an inode's dirty pages. Either the caller has an active reference | |
478 | * on the inode or the inode has I_WILL_FREE set. | |
479 | * | |
480 | * This function is designed to be called for writing back one inode which | |
481 | * we go e.g. from filesystem. Flusher thread uses __writeback_single_inode() | |
482 | * and does more profound writeback list handling in writeback_sb_inodes(). | |
483 | */ | |
484 | static int | |
485 | writeback_single_inode(struct inode *inode, struct bdi_writeback *wb, | |
486 | struct writeback_control *wbc) | |
487 | { | |
488 | int ret = 0; | |
489 | ||
490 | spin_lock(&inode->i_lock); | |
491 | if (!atomic_read(&inode->i_count)) | |
492 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
493 | else | |
494 | WARN_ON(inode->i_state & I_WILL_FREE); | |
495 | ||
496 | if (inode->i_state & I_SYNC) { | |
497 | if (wbc->sync_mode != WB_SYNC_ALL) | |
498 | goto out; | |
499 | /* | |
500 | * It's a data-integrity sync. We must wait. Since callers hold | |
501 | * inode reference or inode has I_WILL_FREE set, it cannot go | |
502 | * away under us. | |
503 | */ | |
504 | __inode_wait_for_writeback(inode); | |
505 | } | |
506 | WARN_ON(inode->i_state & I_SYNC); | |
507 | /* | |
508 | * Skip inode if it is clean. We don't want to mess with writeback | |
509 | * lists in this function since flusher thread may be doing for example | |
510 | * sync in parallel and if we move the inode, it could get skipped. So | |
511 | * here we make sure inode is on some writeback list and leave it there | |
512 | * unless we have completely cleaned the inode. | |
513 | */ | |
514 | if (!(inode->i_state & I_DIRTY)) | |
515 | goto out; | |
516 | inode->i_state |= I_SYNC; | |
517 | spin_unlock(&inode->i_lock); | |
518 | ||
519 | ret = __writeback_single_inode(inode, wbc); | |
520 | ||
521 | spin_lock(&wb->list_lock); | |
522 | spin_lock(&inode->i_lock); | |
523 | /* | |
524 | * If inode is clean, remove it from writeback lists. Otherwise don't | |
525 | * touch it. See comment above for explanation. | |
526 | */ | |
527 | if (!(inode->i_state & I_DIRTY)) | |
528 | list_del_init(&inode->i_wb_list); | |
529 | spin_unlock(&wb->list_lock); | |
530 | inode_sync_complete(inode); | |
531 | out: | |
532 | spin_unlock(&inode->i_lock); | |
533 | return ret; | |
534 | } | |
535 | ||
536 | static long writeback_chunk_size(struct backing_dev_info *bdi, | |
537 | struct wb_writeback_work *work) | |
538 | { | |
539 | long pages; | |
540 | ||
541 | /* | |
542 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
543 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
544 | * here avoids calling into writeback_inodes_wb() more than once. | |
545 | * | |
546 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
547 | * | |
548 | * wb_writeback() | |
549 | * writeback_sb_inodes() <== called only once | |
550 | * write_cache_pages() <== called once for each inode | |
551 | * (quickly) tag currently dirty pages | |
552 | * (maybe slowly) sync all tagged pages | |
553 | */ | |
554 | if (work->sync_mode == WB_SYNC_ALL || work->tagged_writepages) | |
555 | pages = LONG_MAX; | |
556 | else { | |
557 | pages = min(bdi->avg_write_bandwidth / 2, | |
558 | global_dirty_limit / DIRTY_SCOPE); | |
559 | pages = min(pages, work->nr_pages); | |
560 | pages = round_down(pages + MIN_WRITEBACK_PAGES, | |
561 | MIN_WRITEBACK_PAGES); | |
562 | } | |
563 | ||
564 | return pages; | |
565 | } | |
566 | ||
567 | /* | |
568 | * Write a portion of b_io inodes which belong to @sb. | |
569 | * | |
570 | * Return the number of pages and/or inodes written. | |
571 | */ | |
572 | static long writeback_sb_inodes(struct super_block *sb, | |
573 | struct bdi_writeback *wb, | |
574 | struct wb_writeback_work *work) | |
575 | { | |
576 | struct writeback_control wbc = { | |
577 | .sync_mode = work->sync_mode, | |
578 | .tagged_writepages = work->tagged_writepages, | |
579 | .for_kupdate = work->for_kupdate, | |
580 | .for_background = work->for_background, | |
581 | .range_cyclic = work->range_cyclic, | |
582 | .range_start = 0, | |
583 | .range_end = LLONG_MAX, | |
584 | }; | |
585 | unsigned long start_time = jiffies; | |
586 | long write_chunk; | |
587 | long wrote = 0; /* count both pages and inodes */ | |
588 | ||
589 | while (!list_empty(&wb->b_io)) { | |
590 | struct inode *inode = wb_inode(wb->b_io.prev); | |
591 | ||
592 | if (inode->i_sb != sb) { | |
593 | if (work->sb) { | |
594 | /* | |
595 | * We only want to write back data for this | |
596 | * superblock, move all inodes not belonging | |
597 | * to it back onto the dirty list. | |
598 | */ | |
599 | redirty_tail(inode, wb); | |
600 | continue; | |
601 | } | |
602 | ||
603 | /* | |
604 | * The inode belongs to a different superblock. | |
605 | * Bounce back to the caller to unpin this and | |
606 | * pin the next superblock. | |
607 | */ | |
608 | break; | |
609 | } | |
610 | ||
611 | /* | |
612 | * Don't bother with new inodes or inodes being freed, first | |
613 | * kind does not need periodic writeout yet, and for the latter | |
614 | * kind writeout is handled by the freer. | |
615 | */ | |
616 | spin_lock(&inode->i_lock); | |
617 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
618 | spin_unlock(&inode->i_lock); | |
619 | redirty_tail(inode, wb); | |
620 | continue; | |
621 | } | |
622 | if ((inode->i_state & I_SYNC) && wbc.sync_mode != WB_SYNC_ALL) { | |
623 | /* | |
624 | * If this inode is locked for writeback and we are not | |
625 | * doing writeback-for-data-integrity, move it to | |
626 | * b_more_io so that writeback can proceed with the | |
627 | * other inodes on s_io. | |
628 | * | |
629 | * We'll have another go at writing back this inode | |
630 | * when we completed a full scan of b_io. | |
631 | */ | |
632 | spin_unlock(&inode->i_lock); | |
633 | requeue_io(inode, wb); | |
634 | trace_writeback_sb_inodes_requeue(inode); | |
635 | continue; | |
636 | } | |
637 | spin_unlock(&wb->list_lock); | |
638 | ||
639 | /* | |
640 | * We already requeued the inode if it had I_SYNC set and we | |
641 | * are doing WB_SYNC_NONE writeback. So this catches only the | |
642 | * WB_SYNC_ALL case. | |
643 | */ | |
644 | if (inode->i_state & I_SYNC) { | |
645 | /* Wait for I_SYNC. This function drops i_lock... */ | |
646 | inode_sleep_on_writeback(inode); | |
647 | /* Inode may be gone, start again */ | |
648 | spin_lock(&wb->list_lock); | |
649 | continue; | |
650 | } | |
651 | inode->i_state |= I_SYNC; | |
652 | spin_unlock(&inode->i_lock); | |
653 | ||
654 | write_chunk = writeback_chunk_size(wb->bdi, work); | |
655 | wbc.nr_to_write = write_chunk; | |
656 | wbc.pages_skipped = 0; | |
657 | ||
658 | /* | |
659 | * We use I_SYNC to pin the inode in memory. While it is set | |
660 | * evict_inode() will wait so the inode cannot be freed. | |
661 | */ | |
662 | __writeback_single_inode(inode, &wbc); | |
663 | ||
664 | work->nr_pages -= write_chunk - wbc.nr_to_write; | |
665 | wrote += write_chunk - wbc.nr_to_write; | |
666 | spin_lock(&wb->list_lock); | |
667 | spin_lock(&inode->i_lock); | |
668 | if (!(inode->i_state & I_DIRTY)) | |
669 | wrote++; | |
670 | requeue_inode(inode, wb, &wbc); | |
671 | inode_sync_complete(inode); | |
672 | spin_unlock(&inode->i_lock); | |
673 | cond_resched_lock(&wb->list_lock); | |
674 | /* | |
675 | * bail out to wb_writeback() often enough to check | |
676 | * background threshold and other termination conditions. | |
677 | */ | |
678 | if (wrote) { | |
679 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
680 | break; | |
681 | if (work->nr_pages <= 0) | |
682 | break; | |
683 | } | |
684 | } | |
685 | return wrote; | |
686 | } | |
687 | ||
688 | static long __writeback_inodes_wb(struct bdi_writeback *wb, | |
689 | struct wb_writeback_work *work) | |
690 | { | |
691 | unsigned long start_time = jiffies; | |
692 | long wrote = 0; | |
693 | ||
694 | while (!list_empty(&wb->b_io)) { | |
695 | struct inode *inode = wb_inode(wb->b_io.prev); | |
696 | struct super_block *sb = inode->i_sb; | |
697 | ||
698 | if (!grab_super_passive(sb)) { | |
699 | /* | |
700 | * grab_super_passive() may fail consistently due to | |
701 | * s_umount being grabbed by someone else. Don't use | |
702 | * requeue_io() to avoid busy retrying the inode/sb. | |
703 | */ | |
704 | redirty_tail(inode, wb); | |
705 | continue; | |
706 | } | |
707 | wrote += writeback_sb_inodes(sb, wb, work); | |
708 | drop_super(sb); | |
709 | ||
710 | /* refer to the same tests at the end of writeback_sb_inodes */ | |
711 | if (wrote) { | |
712 | if (time_is_before_jiffies(start_time + HZ / 10UL)) | |
713 | break; | |
714 | if (work->nr_pages <= 0) | |
715 | break; | |
716 | } | |
717 | } | |
718 | /* Leave any unwritten inodes on b_io */ | |
719 | return wrote; | |
720 | } | |
721 | ||
722 | long writeback_inodes_wb(struct bdi_writeback *wb, long nr_pages, | |
723 | enum wb_reason reason) | |
724 | { | |
725 | struct wb_writeback_work work = { | |
726 | .nr_pages = nr_pages, | |
727 | .sync_mode = WB_SYNC_NONE, | |
728 | .range_cyclic = 1, | |
729 | .reason = reason, | |
730 | }; | |
731 | ||
732 | spin_lock(&wb->list_lock); | |
733 | if (list_empty(&wb->b_io)) | |
734 | queue_io(wb, &work); | |
735 | __writeback_inodes_wb(wb, &work); | |
736 | spin_unlock(&wb->list_lock); | |
737 | ||
738 | return nr_pages - work.nr_pages; | |
739 | } | |
740 | ||
741 | static bool over_bground_thresh(struct backing_dev_info *bdi) | |
742 | { | |
743 | unsigned long background_thresh, dirty_thresh; | |
744 | ||
745 | global_dirty_limits(&background_thresh, &dirty_thresh); | |
746 | ||
747 | if (global_page_state(NR_FILE_DIRTY) + | |
748 | global_page_state(NR_UNSTABLE_NFS) > background_thresh) | |
749 | return true; | |
750 | ||
751 | if (bdi_stat(bdi, BDI_RECLAIMABLE) > | |
752 | bdi_dirty_limit(bdi, background_thresh)) | |
753 | return true; | |
754 | ||
755 | return false; | |
756 | } | |
757 | ||
758 | /* | |
759 | * Called under wb->list_lock. If there are multiple wb per bdi, | |
760 | * only the flusher working on the first wb should do it. | |
761 | */ | |
762 | static void wb_update_bandwidth(struct bdi_writeback *wb, | |
763 | unsigned long start_time) | |
764 | { | |
765 | __bdi_update_bandwidth(wb->bdi, 0, 0, 0, 0, 0, start_time); | |
766 | } | |
767 | ||
768 | /* | |
769 | * Explicit flushing or periodic writeback of "old" data. | |
770 | * | |
771 | * Define "old": the first time one of an inode's pages is dirtied, we mark the | |
772 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
773 | * just walks the superblock inode list, writing back any inodes which are | |
774 | * older than a specific point in time. | |
775 | * | |
776 | * Try to run once per dirty_writeback_interval. But if a writeback event | |
777 | * takes longer than a dirty_writeback_interval interval, then leave a | |
778 | * one-second gap. | |
779 | * | |
780 | * older_than_this takes precedence over nr_to_write. So we'll only write back | |
781 | * all dirty pages if they are all attached to "old" mappings. | |
782 | */ | |
783 | static long wb_writeback(struct bdi_writeback *wb, | |
784 | struct wb_writeback_work *work) | |
785 | { | |
786 | unsigned long wb_start = jiffies; | |
787 | long nr_pages = work->nr_pages; | |
788 | unsigned long oldest_jif; | |
789 | struct inode *inode; | |
790 | long progress; | |
791 | ||
792 | oldest_jif = jiffies; | |
793 | work->older_than_this = &oldest_jif; | |
794 | ||
795 | spin_lock(&wb->list_lock); | |
796 | for (;;) { | |
797 | /* | |
798 | * Stop writeback when nr_pages has been consumed | |
799 | */ | |
800 | if (work->nr_pages <= 0) | |
801 | break; | |
802 | ||
803 | /* | |
804 | * Background writeout and kupdate-style writeback may | |
805 | * run forever. Stop them if there is other work to do | |
806 | * so that e.g. sync can proceed. They'll be restarted | |
807 | * after the other works are all done. | |
808 | */ | |
809 | if ((work->for_background || work->for_kupdate) && | |
810 | !list_empty(&wb->bdi->work_list)) | |
811 | break; | |
812 | ||
813 | /* | |
814 | * For background writeout, stop when we are below the | |
815 | * background dirty threshold | |
816 | */ | |
817 | if (work->for_background && !over_bground_thresh(wb->bdi)) | |
818 | break; | |
819 | ||
820 | /* | |
821 | * Kupdate and background works are special and we want to | |
822 | * include all inodes that need writing. Livelock avoidance is | |
823 | * handled by these works yielding to any other work so we are | |
824 | * safe. | |
825 | */ | |
826 | if (work->for_kupdate) { | |
827 | oldest_jif = jiffies - | |
828 | msecs_to_jiffies(dirty_expire_interval * 10); | |
829 | } else if (work->for_background) | |
830 | oldest_jif = jiffies; | |
831 | ||
832 | trace_writeback_start(wb->bdi, work); | |
833 | if (list_empty(&wb->b_io)) | |
834 | queue_io(wb, work); | |
835 | if (work->sb) | |
836 | progress = writeback_sb_inodes(work->sb, wb, work); | |
837 | else | |
838 | progress = __writeback_inodes_wb(wb, work); | |
839 | trace_writeback_written(wb->bdi, work); | |
840 | ||
841 | wb_update_bandwidth(wb, wb_start); | |
842 | ||
843 | /* | |
844 | * Did we write something? Try for more | |
845 | * | |
846 | * Dirty inodes are moved to b_io for writeback in batches. | |
847 | * The completion of the current batch does not necessarily | |
848 | * mean the overall work is done. So we keep looping as long | |
849 | * as made some progress on cleaning pages or inodes. | |
850 | */ | |
851 | if (progress) | |
852 | continue; | |
853 | /* | |
854 | * No more inodes for IO, bail | |
855 | */ | |
856 | if (list_empty(&wb->b_more_io)) | |
857 | break; | |
858 | /* | |
859 | * Nothing written. Wait for some inode to | |
860 | * become available for writeback. Otherwise | |
861 | * we'll just busyloop. | |
862 | */ | |
863 | if (!list_empty(&wb->b_more_io)) { | |
864 | trace_writeback_wait(wb->bdi, work); | |
865 | inode = wb_inode(wb->b_more_io.prev); | |
866 | spin_lock(&inode->i_lock); | |
867 | spin_unlock(&wb->list_lock); | |
868 | /* This function drops i_lock... */ | |
869 | inode_sleep_on_writeback(inode); | |
870 | spin_lock(&wb->list_lock); | |
871 | } | |
872 | } | |
873 | spin_unlock(&wb->list_lock); | |
874 | ||
875 | return nr_pages - work->nr_pages; | |
876 | } | |
877 | ||
878 | /* | |
879 | * Return the next wb_writeback_work struct that hasn't been processed yet. | |
880 | */ | |
881 | static struct wb_writeback_work * | |
882 | get_next_work_item(struct backing_dev_info *bdi) | |
883 | { | |
884 | struct wb_writeback_work *work = NULL; | |
885 | ||
886 | spin_lock_bh(&bdi->wb_lock); | |
887 | if (!list_empty(&bdi->work_list)) { | |
888 | work = list_entry(bdi->work_list.next, | |
889 | struct wb_writeback_work, list); | |
890 | list_del_init(&work->list); | |
891 | } | |
892 | spin_unlock_bh(&bdi->wb_lock); | |
893 | return work; | |
894 | } | |
895 | ||
896 | /* | |
897 | * Add in the number of potentially dirty inodes, because each inode | |
898 | * write can dirty pagecache in the underlying blockdev. | |
899 | */ | |
900 | static unsigned long get_nr_dirty_pages(void) | |
901 | { | |
902 | return global_page_state(NR_FILE_DIRTY) + | |
903 | global_page_state(NR_UNSTABLE_NFS) + | |
904 | get_nr_dirty_inodes(); | |
905 | } | |
906 | ||
907 | static long wb_check_background_flush(struct bdi_writeback *wb) | |
908 | { | |
909 | if (over_bground_thresh(wb->bdi)) { | |
910 | ||
911 | struct wb_writeback_work work = { | |
912 | .nr_pages = LONG_MAX, | |
913 | .sync_mode = WB_SYNC_NONE, | |
914 | .for_background = 1, | |
915 | .range_cyclic = 1, | |
916 | .reason = WB_REASON_BACKGROUND, | |
917 | }; | |
918 | ||
919 | return wb_writeback(wb, &work); | |
920 | } | |
921 | ||
922 | return 0; | |
923 | } | |
924 | ||
925 | static long wb_check_old_data_flush(struct bdi_writeback *wb) | |
926 | { | |
927 | unsigned long expired; | |
928 | long nr_pages; | |
929 | ||
930 | /* | |
931 | * When set to zero, disable periodic writeback | |
932 | */ | |
933 | if (!dirty_writeback_interval) | |
934 | return 0; | |
935 | ||
936 | expired = wb->last_old_flush + | |
937 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
938 | if (time_before(jiffies, expired)) | |
939 | return 0; | |
940 | ||
941 | wb->last_old_flush = jiffies; | |
942 | nr_pages = get_nr_dirty_pages(); | |
943 | ||
944 | if (nr_pages) { | |
945 | struct wb_writeback_work work = { | |
946 | .nr_pages = nr_pages, | |
947 | .sync_mode = WB_SYNC_NONE, | |
948 | .for_kupdate = 1, | |
949 | .range_cyclic = 1, | |
950 | .reason = WB_REASON_PERIODIC, | |
951 | }; | |
952 | ||
953 | return wb_writeback(wb, &work); | |
954 | } | |
955 | ||
956 | return 0; | |
957 | } | |
958 | ||
959 | /* | |
960 | * Retrieve work items and do the writeback they describe | |
961 | */ | |
962 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
963 | { | |
964 | struct backing_dev_info *bdi = wb->bdi; | |
965 | struct wb_writeback_work *work; | |
966 | long wrote = 0; | |
967 | ||
968 | set_bit(BDI_writeback_running, &wb->bdi->state); | |
969 | while ((work = get_next_work_item(bdi)) != NULL) { | |
970 | /* | |
971 | * Override sync mode, in case we must wait for completion | |
972 | * because this thread is exiting now. | |
973 | */ | |
974 | if (force_wait) | |
975 | work->sync_mode = WB_SYNC_ALL; | |
976 | ||
977 | trace_writeback_exec(bdi, work); | |
978 | ||
979 | wrote += wb_writeback(wb, work); | |
980 | ||
981 | /* | |
982 | * Notify the caller of completion if this is a synchronous | |
983 | * work item, otherwise just free it. | |
984 | */ | |
985 | if (work->done) | |
986 | complete(work->done); | |
987 | else | |
988 | kfree(work); | |
989 | } | |
990 | ||
991 | /* | |
992 | * Check for periodic writeback, kupdated() style | |
993 | */ | |
994 | wrote += wb_check_old_data_flush(wb); | |
995 | wrote += wb_check_background_flush(wb); | |
996 | clear_bit(BDI_writeback_running, &wb->bdi->state); | |
997 | ||
998 | return wrote; | |
999 | } | |
1000 | ||
1001 | /* | |
1002 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
1003 | * reschedules periodically and does kupdated style flushing. | |
1004 | */ | |
1005 | void bdi_writeback_workfn(struct work_struct *work) | |
1006 | { | |
1007 | struct bdi_writeback *wb = container_of(to_delayed_work(work), | |
1008 | struct bdi_writeback, dwork); | |
1009 | struct backing_dev_info *bdi = wb->bdi; | |
1010 | long pages_written; | |
1011 | ||
1012 | set_worker_desc("flush-%s", dev_name(bdi->dev)); | |
1013 | current->flags |= PF_SWAPWRITE; | |
1014 | ||
1015 | if (likely(!current_is_workqueue_rescuer() || | |
1016 | list_empty(&bdi->bdi_list))) { | |
1017 | /* | |
1018 | * The normal path. Keep writing back @bdi until its | |
1019 | * work_list is empty. Note that this path is also taken | |
1020 | * if @bdi is shutting down even when we're running off the | |
1021 | * rescuer as work_list needs to be drained. | |
1022 | */ | |
1023 | do { | |
1024 | pages_written = wb_do_writeback(wb, 0); | |
1025 | trace_writeback_pages_written(pages_written); | |
1026 | } while (!list_empty(&bdi->work_list)); | |
1027 | } else { | |
1028 | /* | |
1029 | * bdi_wq can't get enough workers and we're running off | |
1030 | * the emergency worker. Don't hog it. Hopefully, 1024 is | |
1031 | * enough for efficient IO. | |
1032 | */ | |
1033 | pages_written = writeback_inodes_wb(&bdi->wb, 1024, | |
1034 | WB_REASON_FORKER_THREAD); | |
1035 | trace_writeback_pages_written(pages_written); | |
1036 | } | |
1037 | ||
1038 | if (!list_empty(&bdi->work_list) || | |
1039 | (wb_has_dirty_io(wb) && dirty_writeback_interval)) | |
1040 | queue_delayed_work(bdi_wq, &wb->dwork, | |
1041 | msecs_to_jiffies(dirty_writeback_interval * 10)); | |
1042 | ||
1043 | current->flags &= ~PF_SWAPWRITE; | |
1044 | } | |
1045 | ||
1046 | /* | |
1047 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back | |
1048 | * the whole world. | |
1049 | */ | |
1050 | void wakeup_flusher_threads(long nr_pages, enum wb_reason reason) | |
1051 | { | |
1052 | struct backing_dev_info *bdi; | |
1053 | ||
1054 | if (!nr_pages) { | |
1055 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
1056 | global_page_state(NR_UNSTABLE_NFS); | |
1057 | } | |
1058 | ||
1059 | rcu_read_lock(); | |
1060 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { | |
1061 | if (!bdi_has_dirty_io(bdi)) | |
1062 | continue; | |
1063 | __bdi_start_writeback(bdi, nr_pages, false, reason); | |
1064 | } | |
1065 | rcu_read_unlock(); | |
1066 | } | |
1067 | ||
1068 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) | |
1069 | { | |
1070 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1071 | struct dentry *dentry; | |
1072 | const char *name = "?"; | |
1073 | ||
1074 | dentry = d_find_alias(inode); | |
1075 | if (dentry) { | |
1076 | spin_lock(&dentry->d_lock); | |
1077 | name = (const char *) dentry->d_name.name; | |
1078 | } | |
1079 | printk(KERN_DEBUG | |
1080 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1081 | current->comm, task_pid_nr(current), inode->i_ino, | |
1082 | name, inode->i_sb->s_id); | |
1083 | if (dentry) { | |
1084 | spin_unlock(&dentry->d_lock); | |
1085 | dput(dentry); | |
1086 | } | |
1087 | } | |
1088 | } | |
1089 | ||
1090 | /** | |
1091 | * __mark_inode_dirty - internal function | |
1092 | * @inode: inode to mark | |
1093 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1094 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1095 | * mark_inode_dirty_sync. | |
1096 | * | |
1097 | * Put the inode on the super block's dirty list. | |
1098 | * | |
1099 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1100 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1101 | * If it was not hashed, it will never be added to the dirty list | |
1102 | * even if it is later hashed, as it will have been marked dirty already. | |
1103 | * | |
1104 | * In short, make sure you hash any inodes _before_ you start marking | |
1105 | * them dirty. | |
1106 | * | |
1107 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of | |
1108 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1109 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1110 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1111 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1112 | * blockdev inode. | |
1113 | */ | |
1114 | void __mark_inode_dirty(struct inode *inode, int flags) | |
1115 | { | |
1116 | struct super_block *sb = inode->i_sb; | |
1117 | struct backing_dev_info *bdi = NULL; | |
1118 | ||
1119 | /* | |
1120 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1121 | * dirty the inode itself | |
1122 | */ | |
1123 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1124 | trace_writeback_dirty_inode_start(inode, flags); | |
1125 | ||
1126 | if (sb->s_op->dirty_inode) | |
1127 | sb->s_op->dirty_inode(inode, flags); | |
1128 | ||
1129 | trace_writeback_dirty_inode(inode, flags); | |
1130 | } | |
1131 | ||
1132 | /* | |
1133 | * make sure that changes are seen by all cpus before we test i_state | |
1134 | * -- mikulas | |
1135 | */ | |
1136 | smp_mb(); | |
1137 | ||
1138 | /* avoid the locking if we can */ | |
1139 | if ((inode->i_state & flags) == flags) | |
1140 | return; | |
1141 | ||
1142 | if (unlikely(block_dump)) | |
1143 | block_dump___mark_inode_dirty(inode); | |
1144 | ||
1145 | spin_lock(&inode->i_lock); | |
1146 | if ((inode->i_state & flags) != flags) { | |
1147 | const int was_dirty = inode->i_state & I_DIRTY; | |
1148 | ||
1149 | inode->i_state |= flags; | |
1150 | ||
1151 | /* | |
1152 | * If the inode is being synced, just update its dirty state. | |
1153 | * The unlocker will place the inode on the appropriate | |
1154 | * superblock list, based upon its state. | |
1155 | */ | |
1156 | if (inode->i_state & I_SYNC) | |
1157 | goto out_unlock_inode; | |
1158 | ||
1159 | /* | |
1160 | * Only add valid (hashed) inodes to the superblock's | |
1161 | * dirty list. Add blockdev inodes as well. | |
1162 | */ | |
1163 | if (!S_ISBLK(inode->i_mode)) { | |
1164 | if (inode_unhashed(inode)) | |
1165 | goto out_unlock_inode; | |
1166 | } | |
1167 | if (inode->i_state & I_FREEING) | |
1168 | goto out_unlock_inode; | |
1169 | ||
1170 | /* | |
1171 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1172 | * reposition it (that would break b_dirty time-ordering). | |
1173 | */ | |
1174 | if (!was_dirty) { | |
1175 | bool wakeup_bdi = false; | |
1176 | bdi = inode_to_bdi(inode); | |
1177 | ||
1178 | if (bdi_cap_writeback_dirty(bdi)) { | |
1179 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1180 | "bdi-%s not registered\n", bdi->name); | |
1181 | ||
1182 | /* | |
1183 | * If this is the first dirty inode for this | |
1184 | * bdi, we have to wake-up the corresponding | |
1185 | * bdi thread to make sure background | |
1186 | * write-back happens later. | |
1187 | */ | |
1188 | if (!wb_has_dirty_io(&bdi->wb)) | |
1189 | wakeup_bdi = true; | |
1190 | } | |
1191 | ||
1192 | spin_unlock(&inode->i_lock); | |
1193 | spin_lock(&bdi->wb.list_lock); | |
1194 | inode->dirtied_when = jiffies; | |
1195 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); | |
1196 | spin_unlock(&bdi->wb.list_lock); | |
1197 | ||
1198 | if (wakeup_bdi) | |
1199 | bdi_wakeup_thread_delayed(bdi); | |
1200 | return; | |
1201 | } | |
1202 | } | |
1203 | out_unlock_inode: | |
1204 | spin_unlock(&inode->i_lock); | |
1205 | ||
1206 | } | |
1207 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1208 | ||
1209 | static void wait_sb_inodes(struct super_block *sb) | |
1210 | { | |
1211 | struct inode *inode, *old_inode = NULL; | |
1212 | ||
1213 | /* | |
1214 | * We need to be protected against the filesystem going from | |
1215 | * r/o to r/w or vice versa. | |
1216 | */ | |
1217 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
1218 | ||
1219 | spin_lock(&inode_sb_list_lock); | |
1220 | ||
1221 | /* | |
1222 | * Data integrity sync. Must wait for all pages under writeback, | |
1223 | * because there may have been pages dirtied before our sync | |
1224 | * call, but which had writeout started before we write it out. | |
1225 | * In which case, the inode may not be on the dirty list, but | |
1226 | * we still have to wait for that writeout. | |
1227 | */ | |
1228 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { | |
1229 | struct address_space *mapping = inode->i_mapping; | |
1230 | ||
1231 | spin_lock(&inode->i_lock); | |
1232 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1233 | (mapping->nrpages == 0)) { | |
1234 | spin_unlock(&inode->i_lock); | |
1235 | continue; | |
1236 | } | |
1237 | __iget(inode); | |
1238 | spin_unlock(&inode->i_lock); | |
1239 | spin_unlock(&inode_sb_list_lock); | |
1240 | ||
1241 | /* | |
1242 | * We hold a reference to 'inode' so it couldn't have been | |
1243 | * removed from s_inodes list while we dropped the | |
1244 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1245 | * be holding the last reference and we cannot iput it under | |
1246 | * inode_sb_list_lock. So we keep the reference and iput it | |
1247 | * later. | |
1248 | */ | |
1249 | iput(old_inode); | |
1250 | old_inode = inode; | |
1251 | ||
1252 | filemap_fdatawait(mapping); | |
1253 | ||
1254 | cond_resched(); | |
1255 | ||
1256 | spin_lock(&inode_sb_list_lock); | |
1257 | } | |
1258 | spin_unlock(&inode_sb_list_lock); | |
1259 | iput(old_inode); | |
1260 | } | |
1261 | ||
1262 | /** | |
1263 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block | |
1264 | * @sb: the superblock | |
1265 | * @nr: the number of pages to write | |
1266 | * @reason: reason why some writeback work initiated | |
1267 | * | |
1268 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1269 | * on how many (if any) will be written, and this function does not wait | |
1270 | * for IO completion of submitted IO. | |
1271 | */ | |
1272 | void writeback_inodes_sb_nr(struct super_block *sb, | |
1273 | unsigned long nr, | |
1274 | enum wb_reason reason) | |
1275 | { | |
1276 | DECLARE_COMPLETION_ONSTACK(done); | |
1277 | struct wb_writeback_work work = { | |
1278 | .sb = sb, | |
1279 | .sync_mode = WB_SYNC_NONE, | |
1280 | .tagged_writepages = 1, | |
1281 | .done = &done, | |
1282 | .nr_pages = nr, | |
1283 | .reason = reason, | |
1284 | }; | |
1285 | ||
1286 | if (sb->s_bdi == &noop_backing_dev_info) | |
1287 | return; | |
1288 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
1289 | bdi_queue_work(sb->s_bdi, &work); | |
1290 | wait_for_completion(&done); | |
1291 | } | |
1292 | EXPORT_SYMBOL(writeback_inodes_sb_nr); | |
1293 | ||
1294 | /** | |
1295 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1296 | * @sb: the superblock | |
1297 | * @reason: reason why some writeback work was initiated | |
1298 | * | |
1299 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1300 | * on how many (if any) will be written, and this function does not wait | |
1301 | * for IO completion of submitted IO. | |
1302 | */ | |
1303 | void writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) | |
1304 | { | |
1305 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); | |
1306 | } | |
1307 | EXPORT_SYMBOL(writeback_inodes_sb); | |
1308 | ||
1309 | /** | |
1310 | * try_to_writeback_inodes_sb_nr - try to start writeback if none underway | |
1311 | * @sb: the superblock | |
1312 | * @nr: the number of pages to write | |
1313 | * @reason: the reason of writeback | |
1314 | * | |
1315 | * Invoke writeback_inodes_sb_nr if no writeback is currently underway. | |
1316 | * Returns 1 if writeback was started, 0 if not. | |
1317 | */ | |
1318 | int try_to_writeback_inodes_sb_nr(struct super_block *sb, | |
1319 | unsigned long nr, | |
1320 | enum wb_reason reason) | |
1321 | { | |
1322 | if (writeback_in_progress(sb->s_bdi)) | |
1323 | return 1; | |
1324 | ||
1325 | if (!down_read_trylock(&sb->s_umount)) | |
1326 | return 0; | |
1327 | ||
1328 | writeback_inodes_sb_nr(sb, nr, reason); | |
1329 | up_read(&sb->s_umount); | |
1330 | return 1; | |
1331 | } | |
1332 | EXPORT_SYMBOL(try_to_writeback_inodes_sb_nr); | |
1333 | ||
1334 | /** | |
1335 | * try_to_writeback_inodes_sb - try to start writeback if none underway | |
1336 | * @sb: the superblock | |
1337 | * @reason: reason why some writeback work was initiated | |
1338 | * | |
1339 | * Implement by try_to_writeback_inodes_sb_nr() | |
1340 | * Returns 1 if writeback was started, 0 if not. | |
1341 | */ | |
1342 | int try_to_writeback_inodes_sb(struct super_block *sb, enum wb_reason reason) | |
1343 | { | |
1344 | return try_to_writeback_inodes_sb_nr(sb, get_nr_dirty_pages(), reason); | |
1345 | } | |
1346 | EXPORT_SYMBOL(try_to_writeback_inodes_sb); | |
1347 | ||
1348 | /** | |
1349 | * sync_inodes_sb - sync sb inode pages | |
1350 | * @sb: the superblock | |
1351 | * | |
1352 | * This function writes and waits on any dirty inode belonging to this | |
1353 | * super_block. | |
1354 | */ | |
1355 | void sync_inodes_sb(struct super_block *sb) | |
1356 | { | |
1357 | DECLARE_COMPLETION_ONSTACK(done); | |
1358 | struct wb_writeback_work work = { | |
1359 | .sb = sb, | |
1360 | .sync_mode = WB_SYNC_ALL, | |
1361 | .nr_pages = LONG_MAX, | |
1362 | .range_cyclic = 0, | |
1363 | .done = &done, | |
1364 | .reason = WB_REASON_SYNC, | |
1365 | }; | |
1366 | ||
1367 | /* Nothing to do? */ | |
1368 | if (sb->s_bdi == &noop_backing_dev_info) | |
1369 | return; | |
1370 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
1371 | ||
1372 | bdi_queue_work(sb->s_bdi, &work); | |
1373 | wait_for_completion(&done); | |
1374 | ||
1375 | wait_sb_inodes(sb); | |
1376 | } | |
1377 | EXPORT_SYMBOL(sync_inodes_sb); | |
1378 | ||
1379 | /** | |
1380 | * write_inode_now - write an inode to disk | |
1381 | * @inode: inode to write to disk | |
1382 | * @sync: whether the write should be synchronous or not | |
1383 | * | |
1384 | * This function commits an inode to disk immediately if it is dirty. This is | |
1385 | * primarily needed by knfsd. | |
1386 | * | |
1387 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. | |
1388 | */ | |
1389 | int write_inode_now(struct inode *inode, int sync) | |
1390 | { | |
1391 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; | |
1392 | struct writeback_control wbc = { | |
1393 | .nr_to_write = LONG_MAX, | |
1394 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1395 | .range_start = 0, | |
1396 | .range_end = LLONG_MAX, | |
1397 | }; | |
1398 | ||
1399 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
1400 | wbc.nr_to_write = 0; | |
1401 | ||
1402 | might_sleep(); | |
1403 | return writeback_single_inode(inode, wb, &wbc); | |
1404 | } | |
1405 | EXPORT_SYMBOL(write_inode_now); | |
1406 | ||
1407 | /** | |
1408 | * sync_inode - write an inode and its pages to disk. | |
1409 | * @inode: the inode to sync | |
1410 | * @wbc: controls the writeback mode | |
1411 | * | |
1412 | * sync_inode() will write an inode and its pages to disk. It will also | |
1413 | * correctly update the inode on its superblock's dirty inode lists and will | |
1414 | * update inode->i_state. | |
1415 | * | |
1416 | * The caller must have a ref on the inode. | |
1417 | */ | |
1418 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1419 | { | |
1420 | return writeback_single_inode(inode, &inode_to_bdi(inode)->wb, wbc); | |
1421 | } | |
1422 | EXPORT_SYMBOL(sync_inode); | |
1423 | ||
1424 | /** | |
1425 | * sync_inode_metadata - write an inode to disk | |
1426 | * @inode: the inode to sync | |
1427 | * @wait: wait for I/O to complete. | |
1428 | * | |
1429 | * Write an inode to disk and adjust its dirty state after completion. | |
1430 | * | |
1431 | * Note: only writes the actual inode, no associated data or other metadata. | |
1432 | */ | |
1433 | int sync_inode_metadata(struct inode *inode, int wait) | |
1434 | { | |
1435 | struct writeback_control wbc = { | |
1436 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1437 | .nr_to_write = 0, /* metadata-only */ | |
1438 | }; | |
1439 | ||
1440 | return sync_inode(inode, &wbc); | |
1441 | } | |
1442 | EXPORT_SYMBOL(sync_inode_metadata); |