]>
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/sched.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/mm.h> | |
22 | #include <linux/writeback.h> | |
23 | #include <linux/blkdev.h> | |
24 | #include <linux/backing-dev.h> | |
25 | #include <linux/buffer_head.h> | |
26 | #include "internal.h" | |
27 | ||
28 | ||
29 | /** | |
30 | * writeback_acquire - attempt to get exclusive writeback access to a device | |
31 | * @bdi: the device's backing_dev_info structure | |
32 | * | |
33 | * It is a waste of resources to have more than one pdflush thread blocked on | |
34 | * a single request queue. Exclusion at the request_queue level is obtained | |
35 | * via a flag in the request_queue's backing_dev_info.state. | |
36 | * | |
37 | * Non-request_queue-backed address_spaces will share default_backing_dev_info, | |
38 | * unless they implement their own. Which is somewhat inefficient, as this | |
39 | * may prevent concurrent writeback against multiple devices. | |
40 | */ | |
41 | static int writeback_acquire(struct backing_dev_info *bdi) | |
42 | { | |
43 | return !test_and_set_bit(BDI_pdflush, &bdi->state); | |
44 | } | |
45 | ||
46 | /** | |
47 | * writeback_in_progress - determine whether there is writeback in progress | |
48 | * @bdi: the device's backing_dev_info structure. | |
49 | * | |
50 | * Determine whether there is writeback in progress against a backing device. | |
51 | */ | |
52 | int writeback_in_progress(struct backing_dev_info *bdi) | |
53 | { | |
54 | return test_bit(BDI_pdflush, &bdi->state); | |
55 | } | |
56 | ||
57 | /** | |
58 | * writeback_release - relinquish exclusive writeback access against a device. | |
59 | * @bdi: the device's backing_dev_info structure | |
60 | */ | |
61 | static void writeback_release(struct backing_dev_info *bdi) | |
62 | { | |
63 | BUG_ON(!writeback_in_progress(bdi)); | |
64 | clear_bit(BDI_pdflush, &bdi->state); | |
65 | } | |
66 | ||
67 | /** | |
68 | * __mark_inode_dirty - internal function | |
69 | * @inode: inode to mark | |
70 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
71 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
72 | * mark_inode_dirty_sync. | |
73 | * | |
74 | * Put the inode on the super block's dirty list. | |
75 | * | |
76 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
77 | * dirty list only if it is hashed or if it refers to a blockdev. | |
78 | * If it was not hashed, it will never be added to the dirty list | |
79 | * even if it is later hashed, as it will have been marked dirty already. | |
80 | * | |
81 | * In short, make sure you hash any inodes _before_ you start marking | |
82 | * them dirty. | |
83 | * | |
84 | * This function *must* be atomic for the I_DIRTY_PAGES case - | |
85 | * set_page_dirty() is called under spinlock in several places. | |
86 | * | |
87 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of | |
88 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
89 | * the kernel-internal blockdev inode represents the dirtying time of the | |
90 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
91 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
92 | * blockdev inode. | |
93 | */ | |
94 | void __mark_inode_dirty(struct inode *inode, int flags) | |
95 | { | |
96 | struct super_block *sb = inode->i_sb; | |
97 | ||
98 | /* | |
99 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
100 | * dirty the inode itself | |
101 | */ | |
102 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
103 | if (sb->s_op->dirty_inode) | |
104 | sb->s_op->dirty_inode(inode); | |
105 | } | |
106 | ||
107 | /* | |
108 | * make sure that changes are seen by all cpus before we test i_state | |
109 | * -- mikulas | |
110 | */ | |
111 | smp_mb(); | |
112 | ||
113 | /* avoid the locking if we can */ | |
114 | if ((inode->i_state & flags) == flags) | |
115 | return; | |
116 | ||
117 | if (unlikely(block_dump)) { | |
118 | struct dentry *dentry = NULL; | |
119 | const char *name = "?"; | |
120 | ||
121 | if (!list_empty(&inode->i_dentry)) { | |
122 | dentry = list_entry(inode->i_dentry.next, | |
123 | struct dentry, d_alias); | |
124 | if (dentry && dentry->d_name.name) | |
125 | name = (const char *) dentry->d_name.name; | |
126 | } | |
127 | ||
128 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) | |
129 | printk(KERN_DEBUG | |
130 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
131 | current->comm, task_pid_nr(current), inode->i_ino, | |
132 | name, inode->i_sb->s_id); | |
133 | } | |
134 | ||
135 | spin_lock(&inode_lock); | |
136 | if ((inode->i_state & flags) != flags) { | |
137 | const int was_dirty = inode->i_state & I_DIRTY; | |
138 | ||
139 | inode->i_state |= flags; | |
140 | ||
141 | /* | |
142 | * If the inode is being synced, just update its dirty state. | |
143 | * The unlocker will place the inode on the appropriate | |
144 | * superblock list, based upon its state. | |
145 | */ | |
146 | if (inode->i_state & I_SYNC) | |
147 | goto out; | |
148 | ||
149 | /* | |
150 | * Only add valid (hashed) inodes to the superblock's | |
151 | * dirty list. Add blockdev inodes as well. | |
152 | */ | |
153 | if (!S_ISBLK(inode->i_mode)) { | |
154 | if (hlist_unhashed(&inode->i_hash)) | |
155 | goto out; | |
156 | } | |
157 | if (inode->i_state & (I_FREEING|I_CLEAR)) | |
158 | goto out; | |
159 | ||
160 | /* | |
161 | * If the inode was already on s_dirty/s_io/s_more_io, don't | |
162 | * reposition it (that would break s_dirty time-ordering). | |
163 | */ | |
164 | if (!was_dirty) { | |
165 | inode->dirtied_when = jiffies; | |
166 | list_move(&inode->i_list, &sb->s_dirty); | |
167 | } | |
168 | } | |
169 | out: | |
170 | spin_unlock(&inode_lock); | |
171 | } | |
172 | ||
173 | EXPORT_SYMBOL(__mark_inode_dirty); | |
174 | ||
175 | static int write_inode(struct inode *inode, int sync) | |
176 | { | |
177 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) | |
178 | return inode->i_sb->s_op->write_inode(inode, sync); | |
179 | return 0; | |
180 | } | |
181 | ||
182 | /* | |
183 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
184 | * furthest end of its superblock's dirty-inode list. | |
185 | * | |
186 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
187 | * already the most-recently-dirtied inode on the s_dirty list. If that is | |
188 | * the case then the inode must have been redirtied while it was being written | |
189 | * out and we don't reset its dirtied_when. | |
190 | */ | |
191 | static void redirty_tail(struct inode *inode) | |
192 | { | |
193 | struct super_block *sb = inode->i_sb; | |
194 | ||
195 | if (!list_empty(&sb->s_dirty)) { | |
196 | struct inode *tail_inode; | |
197 | ||
198 | tail_inode = list_entry(sb->s_dirty.next, struct inode, i_list); | |
199 | if (!time_after_eq(inode->dirtied_when, | |
200 | tail_inode->dirtied_when)) | |
201 | inode->dirtied_when = jiffies; | |
202 | } | |
203 | list_move(&inode->i_list, &sb->s_dirty); | |
204 | } | |
205 | ||
206 | /* | |
207 | * requeue inode for re-scanning after sb->s_io list is exhausted. | |
208 | */ | |
209 | static void requeue_io(struct inode *inode) | |
210 | { | |
211 | list_move(&inode->i_list, &inode->i_sb->s_more_io); | |
212 | } | |
213 | ||
214 | static void inode_sync_complete(struct inode *inode) | |
215 | { | |
216 | /* | |
217 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
218 | */ | |
219 | smp_mb(); | |
220 | wake_up_bit(&inode->i_state, __I_SYNC); | |
221 | } | |
222 | ||
223 | /* | |
224 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
225 | */ | |
226 | static void move_expired_inodes(struct list_head *delaying_queue, | |
227 | struct list_head *dispatch_queue, | |
228 | unsigned long *older_than_this) | |
229 | { | |
230 | while (!list_empty(delaying_queue)) { | |
231 | struct inode *inode = list_entry(delaying_queue->prev, | |
232 | struct inode, i_list); | |
233 | if (older_than_this && | |
234 | time_after(inode->dirtied_when, *older_than_this)) | |
235 | break; | |
236 | list_move(&inode->i_list, dispatch_queue); | |
237 | } | |
238 | } | |
239 | ||
240 | /* | |
241 | * Queue all expired dirty inodes for io, eldest first. | |
242 | */ | |
243 | static void queue_io(struct super_block *sb, | |
244 | unsigned long *older_than_this) | |
245 | { | |
246 | list_splice_init(&sb->s_more_io, sb->s_io.prev); | |
247 | move_expired_inodes(&sb->s_dirty, &sb->s_io, older_than_this); | |
248 | } | |
249 | ||
250 | int sb_has_dirty_inodes(struct super_block *sb) | |
251 | { | |
252 | return !list_empty(&sb->s_dirty) || | |
253 | !list_empty(&sb->s_io) || | |
254 | !list_empty(&sb->s_more_io); | |
255 | } | |
256 | EXPORT_SYMBOL(sb_has_dirty_inodes); | |
257 | ||
258 | /* | |
259 | * Write a single inode's dirty pages and inode data out to disk. | |
260 | * If `wait' is set, wait on the writeout. | |
261 | * | |
262 | * The whole writeout design is quite complex and fragile. We want to avoid | |
263 | * starvation of particular inodes when others are being redirtied, prevent | |
264 | * livelocks, etc. | |
265 | * | |
266 | * Called under inode_lock. | |
267 | */ | |
268 | static int | |
269 | __sync_single_inode(struct inode *inode, struct writeback_control *wbc) | |
270 | { | |
271 | unsigned dirty; | |
272 | struct address_space *mapping = inode->i_mapping; | |
273 | int wait = wbc->sync_mode == WB_SYNC_ALL; | |
274 | int ret; | |
275 | ||
276 | BUG_ON(inode->i_state & I_SYNC); | |
277 | ||
278 | /* Set I_SYNC, reset I_DIRTY */ | |
279 | dirty = inode->i_state & I_DIRTY; | |
280 | inode->i_state |= I_SYNC; | |
281 | inode->i_state &= ~I_DIRTY; | |
282 | ||
283 | spin_unlock(&inode_lock); | |
284 | ||
285 | ret = do_writepages(mapping, wbc); | |
286 | ||
287 | /* Don't write the inode if only I_DIRTY_PAGES was set */ | |
288 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
289 | int err = write_inode(inode, wait); | |
290 | if (ret == 0) | |
291 | ret = err; | |
292 | } | |
293 | ||
294 | if (wait) { | |
295 | int err = filemap_fdatawait(mapping); | |
296 | if (ret == 0) | |
297 | ret = err; | |
298 | } | |
299 | ||
300 | spin_lock(&inode_lock); | |
301 | inode->i_state &= ~I_SYNC; | |
302 | if (!(inode->i_state & I_FREEING)) { | |
303 | if (!(inode->i_state & I_DIRTY) && | |
304 | mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
305 | /* | |
306 | * We didn't write back all the pages. nfs_writepages() | |
307 | * sometimes bales out without doing anything. Redirty | |
308 | * the inode; Move it from s_io onto s_more_io/s_dirty. | |
309 | */ | |
310 | /* | |
311 | * akpm: if the caller was the kupdate function we put | |
312 | * this inode at the head of s_dirty so it gets first | |
313 | * consideration. Otherwise, move it to the tail, for | |
314 | * the reasons described there. I'm not really sure | |
315 | * how much sense this makes. Presumably I had a good | |
316 | * reasons for doing it this way, and I'd rather not | |
317 | * muck with it at present. | |
318 | */ | |
319 | if (wbc->for_kupdate) { | |
320 | /* | |
321 | * For the kupdate function we move the inode | |
322 | * to s_more_io so it will get more writeout as | |
323 | * soon as the queue becomes uncongested. | |
324 | */ | |
325 | inode->i_state |= I_DIRTY_PAGES; | |
326 | if (wbc->nr_to_write <= 0) { | |
327 | /* | |
328 | * slice used up: queue for next turn | |
329 | */ | |
330 | requeue_io(inode); | |
331 | } else { | |
332 | /* | |
333 | * somehow blocked: retry later | |
334 | */ | |
335 | redirty_tail(inode); | |
336 | } | |
337 | } else { | |
338 | /* | |
339 | * Otherwise fully redirty the inode so that | |
340 | * other inodes on this superblock will get some | |
341 | * writeout. Otherwise heavy writing to one | |
342 | * file would indefinitely suspend writeout of | |
343 | * all the other files. | |
344 | */ | |
345 | inode->i_state |= I_DIRTY_PAGES; | |
346 | redirty_tail(inode); | |
347 | } | |
348 | } else if (inode->i_state & I_DIRTY) { | |
349 | /* | |
350 | * Someone redirtied the inode while were writing back | |
351 | * the pages. | |
352 | */ | |
353 | redirty_tail(inode); | |
354 | } else if (atomic_read(&inode->i_count)) { | |
355 | /* | |
356 | * The inode is clean, inuse | |
357 | */ | |
358 | list_move(&inode->i_list, &inode_in_use); | |
359 | } else { | |
360 | /* | |
361 | * The inode is clean, unused | |
362 | */ | |
363 | list_move(&inode->i_list, &inode_unused); | |
364 | } | |
365 | } | |
366 | inode_sync_complete(inode); | |
367 | return ret; | |
368 | } | |
369 | ||
370 | /* | |
371 | * Write out an inode's dirty pages. Called under inode_lock. Either the | |
372 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
373 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
374 | */ | |
375 | static int | |
376 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) | |
377 | { | |
378 | wait_queue_head_t *wqh; | |
379 | ||
380 | if (!atomic_read(&inode->i_count)) | |
381 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
382 | else | |
383 | WARN_ON(inode->i_state & I_WILL_FREE); | |
384 | ||
385 | if ((wbc->sync_mode != WB_SYNC_ALL) && (inode->i_state & I_SYNC)) { | |
386 | /* | |
387 | * We're skipping this inode because it's locked, and we're not | |
388 | * doing writeback-for-data-integrity. Move it to s_more_io so | |
389 | * that writeback can proceed with the other inodes on s_io. | |
390 | * We'll have another go at writing back this inode when we | |
391 | * completed a full scan of s_io. | |
392 | */ | |
393 | requeue_io(inode); | |
394 | return 0; | |
395 | } | |
396 | ||
397 | /* | |
398 | * It's a data-integrity sync. We must wait. | |
399 | */ | |
400 | if (inode->i_state & I_SYNC) { | |
401 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
402 | ||
403 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
404 | do { | |
405 | spin_unlock(&inode_lock); | |
406 | __wait_on_bit(wqh, &wq, inode_wait, | |
407 | TASK_UNINTERRUPTIBLE); | |
408 | spin_lock(&inode_lock); | |
409 | } while (inode->i_state & I_SYNC); | |
410 | } | |
411 | return __sync_single_inode(inode, wbc); | |
412 | } | |
413 | ||
414 | /* | |
415 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
416 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
417 | * | |
418 | * If older_than_this is non-NULL, then only write out inodes which | |
419 | * had their first dirtying at a time earlier than *older_than_this. | |
420 | * | |
421 | * If we're a pdlfush thread, then implement pdflush collision avoidance | |
422 | * against the entire list. | |
423 | * | |
424 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. | |
425 | * This function assumes that the blockdev superblock's inodes are backed by | |
426 | * a variety of queues, so all inodes are searched. For other superblocks, | |
427 | * assume that all inodes are backed by the same queue. | |
428 | * | |
429 | * FIXME: this linear search could get expensive with many fileystems. But | |
430 | * how to fix? We need to go from an address_space to all inodes which share | |
431 | * a queue with that address_space. (Easy: have a global "dirty superblocks" | |
432 | * list). | |
433 | * | |
434 | * The inodes to be written are parked on sb->s_io. They are moved back onto | |
435 | * sb->s_dirty as they are selected for writing. This way, none can be missed | |
436 | * on the writer throttling path, and we get decent balancing between many | |
437 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
438 | */ | |
439 | void generic_sync_sb_inodes(struct super_block *sb, | |
440 | struct writeback_control *wbc) | |
441 | { | |
442 | const unsigned long start = jiffies; /* livelock avoidance */ | |
443 | int sync = wbc->sync_mode == WB_SYNC_ALL; | |
444 | ||
445 | spin_lock(&inode_lock); | |
446 | if (!wbc->for_kupdate || list_empty(&sb->s_io)) | |
447 | queue_io(sb, wbc->older_than_this); | |
448 | ||
449 | while (!list_empty(&sb->s_io)) { | |
450 | struct inode *inode = list_entry(sb->s_io.prev, | |
451 | struct inode, i_list); | |
452 | struct address_space *mapping = inode->i_mapping; | |
453 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
454 | long pages_skipped; | |
455 | ||
456 | if (!bdi_cap_writeback_dirty(bdi)) { | |
457 | redirty_tail(inode); | |
458 | if (sb_is_blkdev_sb(sb)) { | |
459 | /* | |
460 | * Dirty memory-backed blockdev: the ramdisk | |
461 | * driver does this. Skip just this inode | |
462 | */ | |
463 | continue; | |
464 | } | |
465 | /* | |
466 | * Dirty memory-backed inode against a filesystem other | |
467 | * than the kernel-internal bdev filesystem. Skip the | |
468 | * entire superblock. | |
469 | */ | |
470 | break; | |
471 | } | |
472 | ||
473 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
474 | wbc->encountered_congestion = 1; | |
475 | if (!sb_is_blkdev_sb(sb)) | |
476 | break; /* Skip a congested fs */ | |
477 | requeue_io(inode); | |
478 | continue; /* Skip a congested blockdev */ | |
479 | } | |
480 | ||
481 | if (wbc->bdi && bdi != wbc->bdi) { | |
482 | if (!sb_is_blkdev_sb(sb)) | |
483 | break; /* fs has the wrong queue */ | |
484 | requeue_io(inode); | |
485 | continue; /* blockdev has wrong queue */ | |
486 | } | |
487 | ||
488 | /* Was this inode dirtied after sync_sb_inodes was called? */ | |
489 | if (time_after(inode->dirtied_when, start)) | |
490 | break; | |
491 | ||
492 | /* Is another pdflush already flushing this queue? */ | |
493 | if (current_is_pdflush() && !writeback_acquire(bdi)) | |
494 | break; | |
495 | ||
496 | BUG_ON(inode->i_state & I_FREEING); | |
497 | __iget(inode); | |
498 | pages_skipped = wbc->pages_skipped; | |
499 | __writeback_single_inode(inode, wbc); | |
500 | if (current_is_pdflush()) | |
501 | writeback_release(bdi); | |
502 | if (wbc->pages_skipped != pages_skipped) { | |
503 | /* | |
504 | * writeback is not making progress due to locked | |
505 | * buffers. Skip this inode for now. | |
506 | */ | |
507 | redirty_tail(inode); | |
508 | } | |
509 | spin_unlock(&inode_lock); | |
510 | iput(inode); | |
511 | cond_resched(); | |
512 | spin_lock(&inode_lock); | |
513 | if (wbc->nr_to_write <= 0) { | |
514 | wbc->more_io = 1; | |
515 | break; | |
516 | } | |
517 | if (!list_empty(&sb->s_more_io)) | |
518 | wbc->more_io = 1; | |
519 | } | |
520 | ||
521 | if (sync) { | |
522 | struct inode *inode, *old_inode = NULL; | |
523 | ||
524 | /* | |
525 | * Data integrity sync. Must wait for all pages under writeback, | |
526 | * because there may have been pages dirtied before our sync | |
527 | * call, but which had writeout started before we write it out. | |
528 | * In which case, the inode may not be on the dirty list, but | |
529 | * we still have to wait for that writeout. | |
530 | */ | |
531 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { | |
532 | struct address_space *mapping; | |
533 | ||
534 | if (inode->i_state & (I_FREEING|I_WILL_FREE)) | |
535 | continue; | |
536 | mapping = inode->i_mapping; | |
537 | if (mapping->nrpages == 0) | |
538 | continue; | |
539 | __iget(inode); | |
540 | spin_unlock(&inode_lock); | |
541 | /* | |
542 | * We hold a reference to 'inode' so it couldn't have | |
543 | * been removed from s_inodes list while we dropped the | |
544 | * inode_lock. We cannot iput the inode now as we can | |
545 | * be holding the last reference and we cannot iput it | |
546 | * under inode_lock. So we keep the reference and iput | |
547 | * it later. | |
548 | */ | |
549 | iput(old_inode); | |
550 | old_inode = inode; | |
551 | ||
552 | filemap_fdatawait(mapping); | |
553 | ||
554 | cond_resched(); | |
555 | ||
556 | spin_lock(&inode_lock); | |
557 | } | |
558 | spin_unlock(&inode_lock); | |
559 | iput(old_inode); | |
560 | } else | |
561 | spin_unlock(&inode_lock); | |
562 | ||
563 | return; /* Leave any unwritten inodes on s_io */ | |
564 | } | |
565 | EXPORT_SYMBOL_GPL(generic_sync_sb_inodes); | |
566 | ||
567 | static void sync_sb_inodes(struct super_block *sb, | |
568 | struct writeback_control *wbc) | |
569 | { | |
570 | generic_sync_sb_inodes(sb, wbc); | |
571 | } | |
572 | ||
573 | /* | |
574 | * Start writeback of dirty pagecache data against all unlocked inodes. | |
575 | * | |
576 | * Note: | |
577 | * We don't need to grab a reference to superblock here. If it has non-empty | |
578 | * ->s_dirty it's hadn't been killed yet and kill_super() won't proceed | |
579 | * past sync_inodes_sb() until the ->s_dirty/s_io/s_more_io lists are all | |
580 | * empty. Since __sync_single_inode() regains inode_lock before it finally moves | |
581 | * inode from superblock lists we are OK. | |
582 | * | |
583 | * If `older_than_this' is non-zero then only flush inodes which have a | |
584 | * flushtime older than *older_than_this. | |
585 | * | |
586 | * If `bdi' is non-zero then we will scan the first inode against each | |
587 | * superblock until we find the matching ones. One group will be the dirty | |
588 | * inodes against a filesystem. Then when we hit the dummy blockdev superblock, | |
589 | * sync_sb_inodes will seekout the blockdev which matches `bdi'. Maybe not | |
590 | * super-efficient but we're about to do a ton of I/O... | |
591 | */ | |
592 | void | |
593 | writeback_inodes(struct writeback_control *wbc) | |
594 | { | |
595 | struct super_block *sb; | |
596 | ||
597 | might_sleep(); | |
598 | spin_lock(&sb_lock); | |
599 | restart: | |
600 | list_for_each_entry_reverse(sb, &super_blocks, s_list) { | |
601 | if (sb_has_dirty_inodes(sb)) { | |
602 | /* we're making our own get_super here */ | |
603 | sb->s_count++; | |
604 | spin_unlock(&sb_lock); | |
605 | /* | |
606 | * If we can't get the readlock, there's no sense in | |
607 | * waiting around, most of the time the FS is going to | |
608 | * be unmounted by the time it is released. | |
609 | */ | |
610 | if (down_read_trylock(&sb->s_umount)) { | |
611 | if (sb->s_root) | |
612 | sync_sb_inodes(sb, wbc); | |
613 | up_read(&sb->s_umount); | |
614 | } | |
615 | spin_lock(&sb_lock); | |
616 | if (__put_super_and_need_restart(sb)) | |
617 | goto restart; | |
618 | } | |
619 | if (wbc->nr_to_write <= 0) | |
620 | break; | |
621 | } | |
622 | spin_unlock(&sb_lock); | |
623 | } | |
624 | ||
625 | /* | |
626 | * writeback and wait upon the filesystem's dirty inodes. The caller will | |
627 | * do this in two passes - one to write, and one to wait. | |
628 | * | |
629 | * A finite limit is set on the number of pages which will be written. | |
630 | * To prevent infinite livelock of sys_sync(). | |
631 | * | |
632 | * We add in the number of potentially dirty inodes, because each inode write | |
633 | * can dirty pagecache in the underlying blockdev. | |
634 | */ | |
635 | void sync_inodes_sb(struct super_block *sb, int wait) | |
636 | { | |
637 | struct writeback_control wbc = { | |
638 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
639 | .range_start = 0, | |
640 | .range_end = LLONG_MAX, | |
641 | }; | |
642 | ||
643 | if (!wait) { | |
644 | unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); | |
645 | unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); | |
646 | ||
647 | wbc.nr_to_write = nr_dirty + nr_unstable + | |
648 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); | |
649 | } else | |
650 | wbc.nr_to_write = LONG_MAX; /* doesn't actually matter */ | |
651 | ||
652 | sync_sb_inodes(sb, &wbc); | |
653 | } | |
654 | ||
655 | /** | |
656 | * sync_inodes - writes all inodes to disk | |
657 | * @wait: wait for completion | |
658 | * | |
659 | * sync_inodes() goes through each super block's dirty inode list, writes the | |
660 | * inodes out, waits on the writeout and puts the inodes back on the normal | |
661 | * list. | |
662 | * | |
663 | * This is for sys_sync(). fsync_dev() uses the same algorithm. The subtle | |
664 | * part of the sync functions is that the blockdev "superblock" is processed | |
665 | * last. This is because the write_inode() function of a typical fs will | |
666 | * perform no I/O, but will mark buffers in the blockdev mapping as dirty. | |
667 | * What we want to do is to perform all that dirtying first, and then write | |
668 | * back all those inode blocks via the blockdev mapping in one sweep. So the | |
669 | * additional (somewhat redundant) sync_blockdev() calls here are to make | |
670 | * sure that really happens. Because if we call sync_inodes_sb(wait=1) with | |
671 | * outstanding dirty inodes, the writeback goes block-at-a-time within the | |
672 | * filesystem's write_inode(). This is extremely slow. | |
673 | */ | |
674 | static void __sync_inodes(int wait) | |
675 | { | |
676 | struct super_block *sb; | |
677 | ||
678 | spin_lock(&sb_lock); | |
679 | restart: | |
680 | list_for_each_entry(sb, &super_blocks, s_list) { | |
681 | sb->s_count++; | |
682 | spin_unlock(&sb_lock); | |
683 | down_read(&sb->s_umount); | |
684 | if (sb->s_root) { | |
685 | sync_inodes_sb(sb, wait); | |
686 | sync_blockdev(sb->s_bdev); | |
687 | } | |
688 | up_read(&sb->s_umount); | |
689 | spin_lock(&sb_lock); | |
690 | if (__put_super_and_need_restart(sb)) | |
691 | goto restart; | |
692 | } | |
693 | spin_unlock(&sb_lock); | |
694 | } | |
695 | ||
696 | void sync_inodes(int wait) | |
697 | { | |
698 | __sync_inodes(0); | |
699 | ||
700 | if (wait) | |
701 | __sync_inodes(1); | |
702 | } | |
703 | ||
704 | /** | |
705 | * write_inode_now - write an inode to disk | |
706 | * @inode: inode to write to disk | |
707 | * @sync: whether the write should be synchronous or not | |
708 | * | |
709 | * This function commits an inode to disk immediately if it is dirty. This is | |
710 | * primarily needed by knfsd. | |
711 | * | |
712 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. | |
713 | */ | |
714 | int write_inode_now(struct inode *inode, int sync) | |
715 | { | |
716 | int ret; | |
717 | struct writeback_control wbc = { | |
718 | .nr_to_write = LONG_MAX, | |
719 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, | |
720 | .range_start = 0, | |
721 | .range_end = LLONG_MAX, | |
722 | }; | |
723 | ||
724 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
725 | wbc.nr_to_write = 0; | |
726 | ||
727 | might_sleep(); | |
728 | spin_lock(&inode_lock); | |
729 | ret = __writeback_single_inode(inode, &wbc); | |
730 | spin_unlock(&inode_lock); | |
731 | if (sync) | |
732 | inode_sync_wait(inode); | |
733 | return ret; | |
734 | } | |
735 | EXPORT_SYMBOL(write_inode_now); | |
736 | ||
737 | /** | |
738 | * sync_inode - write an inode and its pages to disk. | |
739 | * @inode: the inode to sync | |
740 | * @wbc: controls the writeback mode | |
741 | * | |
742 | * sync_inode() will write an inode and its pages to disk. It will also | |
743 | * correctly update the inode on its superblock's dirty inode lists and will | |
744 | * update inode->i_state. | |
745 | * | |
746 | * The caller must have a ref on the inode. | |
747 | */ | |
748 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
749 | { | |
750 | int ret; | |
751 | ||
752 | spin_lock(&inode_lock); | |
753 | ret = __writeback_single_inode(inode, wbc); | |
754 | spin_unlock(&inode_lock); | |
755 | return ret; | |
756 | } | |
757 | EXPORT_SYMBOL(sync_inode); | |
758 | ||
759 | /** | |
760 | * generic_osync_inode - flush all dirty data for a given inode to disk | |
761 | * @inode: inode to write | |
762 | * @mapping: the address_space that should be flushed | |
763 | * @what: what to write and wait upon | |
764 | * | |
765 | * This can be called by file_write functions for files which have the | |
766 | * O_SYNC flag set, to flush dirty writes to disk. | |
767 | * | |
768 | * @what is a bitmask, specifying which part of the inode's data should be | |
769 | * written and waited upon. | |
770 | * | |
771 | * OSYNC_DATA: i_mapping's dirty data | |
772 | * OSYNC_METADATA: the buffers at i_mapping->private_list | |
773 | * OSYNC_INODE: the inode itself | |
774 | */ | |
775 | ||
776 | int generic_osync_inode(struct inode *inode, struct address_space *mapping, int what) | |
777 | { | |
778 | int err = 0; | |
779 | int need_write_inode_now = 0; | |
780 | int err2; | |
781 | ||
782 | if (what & OSYNC_DATA) | |
783 | err = filemap_fdatawrite(mapping); | |
784 | if (what & (OSYNC_METADATA|OSYNC_DATA)) { | |
785 | err2 = sync_mapping_buffers(mapping); | |
786 | if (!err) | |
787 | err = err2; | |
788 | } | |
789 | if (what & OSYNC_DATA) { | |
790 | err2 = filemap_fdatawait(mapping); | |
791 | if (!err) | |
792 | err = err2; | |
793 | } | |
794 | ||
795 | spin_lock(&inode_lock); | |
796 | if ((inode->i_state & I_DIRTY) && | |
797 | ((what & OSYNC_INODE) || (inode->i_state & I_DIRTY_DATASYNC))) | |
798 | need_write_inode_now = 1; | |
799 | spin_unlock(&inode_lock); | |
800 | ||
801 | if (need_write_inode_now) { | |
802 | err2 = write_inode_now(inode, 1); | |
803 | if (!err) | |
804 | err = err2; | |
805 | } | |
806 | else | |
807 | inode_sync_wait(inode); | |
808 | ||
809 | return err; | |
810 | } | |
811 | EXPORT_SYMBOL(generic_osync_inode); |