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