Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/fs-writeback.c | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds. | |
5 | * | |
6 | * Contains all the functions related to writing back and waiting | |
7 | * upon dirty inodes against superblocks, and writing back dirty | |
8 | * pages against inodes. ie: data writeback. Writeout of the | |
9 | * inode itself is not handled here. | |
10 | * | |
11 | * 10Apr2002 akpm@zip.com.au | |
12 | * Split out of fs/inode.c | |
13 | * Additions for address_space-based writeback | |
14 | */ | |
15 | ||
16 | #include <linux/kernel.h> | |
f5ff8422 | 17 | #include <linux/module.h> |
1da177e4 LT |
18 | #include <linux/spinlock.h> |
19 | #include <linux/sched.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/mm.h> | |
22 | #include <linux/writeback.h> | |
23 | #include <linux/blkdev.h> | |
24 | #include <linux/backing-dev.h> | |
25 | #include <linux/buffer_head.h> | |
07f3f05c | 26 | #include "internal.h" |
1da177e4 LT |
27 | |
28 | /** | |
29 | * __mark_inode_dirty - internal function | |
30 | * @inode: inode to mark | |
31 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
32 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
33 | * mark_inode_dirty_sync. | |
34 | * | |
35 | * Put the inode on the super block's dirty list. | |
36 | * | |
37 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
38 | * dirty list only if it is hashed or if it refers to a blockdev. | |
39 | * If it was not hashed, it will never be added to the dirty list | |
40 | * even if it is later hashed, as it will have been marked dirty already. | |
41 | * | |
42 | * In short, make sure you hash any inodes _before_ you start marking | |
43 | * them dirty. | |
44 | * | |
45 | * This function *must* be atomic for the I_DIRTY_PAGES case - | |
46 | * set_page_dirty() is called under spinlock in several places. | |
47 | * | |
48 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of | |
49 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
50 | * the kernel-internal blockdev inode represents the dirtying time of the | |
51 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
52 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
53 | * blockdev inode. | |
54 | */ | |
55 | void __mark_inode_dirty(struct inode *inode, int flags) | |
56 | { | |
57 | struct super_block *sb = inode->i_sb; | |
58 | ||
59 | /* | |
60 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
61 | * dirty the inode itself | |
62 | */ | |
63 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
64 | if (sb->s_op->dirty_inode) | |
65 | sb->s_op->dirty_inode(inode); | |
66 | } | |
67 | ||
68 | /* | |
69 | * make sure that changes are seen by all cpus before we test i_state | |
70 | * -- mikulas | |
71 | */ | |
72 | smp_mb(); | |
73 | ||
74 | /* avoid the locking if we can */ | |
75 | if ((inode->i_state & flags) == flags) | |
76 | return; | |
77 | ||
78 | if (unlikely(block_dump)) { | |
79 | struct dentry *dentry = NULL; | |
80 | const char *name = "?"; | |
81 | ||
82 | if (!list_empty(&inode->i_dentry)) { | |
83 | dentry = list_entry(inode->i_dentry.next, | |
84 | struct dentry, d_alias); | |
85 | if (dentry && dentry->d_name.name) | |
86 | name = (const char *) dentry->d_name.name; | |
87 | } | |
88 | ||
89 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) | |
90 | printk(KERN_DEBUG | |
91 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
92 | current->comm, current->pid, inode->i_ino, | |
93 | name, inode->i_sb->s_id); | |
94 | } | |
95 | ||
96 | spin_lock(&inode_lock); | |
97 | if ((inode->i_state & flags) != flags) { | |
98 | const int was_dirty = inode->i_state & I_DIRTY; | |
99 | ||
100 | inode->i_state |= flags; | |
101 | ||
102 | /* | |
103 | * If the inode is locked, just update its dirty state. | |
104 | * The unlocker will place the inode on the appropriate | |
105 | * superblock list, based upon its state. | |
106 | */ | |
107 | if (inode->i_state & I_LOCK) | |
108 | goto out; | |
109 | ||
110 | /* | |
111 | * Only add valid (hashed) inodes to the superblock's | |
112 | * dirty list. Add blockdev inodes as well. | |
113 | */ | |
114 | if (!S_ISBLK(inode->i_mode)) { | |
115 | if (hlist_unhashed(&inode->i_hash)) | |
116 | goto out; | |
117 | } | |
118 | if (inode->i_state & (I_FREEING|I_CLEAR)) | |
119 | goto out; | |
120 | ||
121 | /* | |
122 | * If the inode was already on s_dirty or s_io, don't | |
123 | * reposition it (that would break s_dirty time-ordering). | |
124 | */ | |
125 | if (!was_dirty) { | |
126 | inode->dirtied_when = jiffies; | |
127 | list_move(&inode->i_list, &sb->s_dirty); | |
128 | } | |
129 | } | |
130 | out: | |
131 | spin_unlock(&inode_lock); | |
132 | } | |
133 | ||
134 | EXPORT_SYMBOL(__mark_inode_dirty); | |
135 | ||
136 | static int write_inode(struct inode *inode, int sync) | |
137 | { | |
138 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) | |
139 | return inode->i_sb->s_op->write_inode(inode, sync); | |
140 | return 0; | |
141 | } | |
142 | ||
6610a0bc AM |
143 | /* |
144 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
145 | * furthest end of its superblock's dirty-inode list. | |
146 | * | |
147 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
148 | * already the most-recently-dirtied inode on the s_dirty list. If that is | |
149 | * the case then the inode must have been redirtied while it was being written | |
150 | * out and we don't reset its dirtied_when. | |
151 | */ | |
152 | static void redirty_tail(struct inode *inode) | |
153 | { | |
154 | struct super_block *sb = inode->i_sb; | |
155 | ||
156 | if (!list_empty(&sb->s_dirty)) { | |
157 | struct inode *tail_inode; | |
158 | ||
159 | tail_inode = list_entry(sb->s_dirty.next, struct inode, i_list); | |
160 | if (!time_after_eq(inode->dirtied_when, | |
161 | tail_inode->dirtied_when)) | |
162 | inode->dirtied_when = jiffies; | |
163 | } | |
164 | list_move(&inode->i_list, &sb->s_dirty); | |
165 | } | |
166 | ||
c986d1e2 AM |
167 | /* |
168 | * Redirty an inode, but mark it as the very next-to-be-written inode on its | |
169 | * superblock's dirty-inode list. | |
170 | * We need to preserve s_dirty's reverse-time-orderedness, so we cheat by | |
171 | * setting this inode's dirtied_when to the same value as that of the inode | |
172 | * which is presently head-of-list, if present head-of-list is newer than this | |
173 | * inode. (head-of-list is the least-recently-dirtied inode: the oldest one). | |
174 | */ | |
175 | static void redirty_head(struct inode *inode) | |
176 | { | |
177 | struct super_block *sb = inode->i_sb; | |
178 | ||
179 | if (!list_empty(&sb->s_dirty)) { | |
180 | struct inode *head_inode; | |
181 | ||
182 | head_inode = list_entry(sb->s_dirty.prev, struct inode, i_list); | |
183 | if (time_after(inode->dirtied_when, head_inode->dirtied_when)) | |
184 | inode->dirtied_when = head_inode->dirtied_when; | |
185 | } | |
186 | list_move_tail(&inode->i_list, &sb->s_dirty); | |
187 | } | |
188 | ||
1da177e4 LT |
189 | /* |
190 | * Write a single inode's dirty pages and inode data out to disk. | |
191 | * If `wait' is set, wait on the writeout. | |
192 | * | |
193 | * The whole writeout design is quite complex and fragile. We want to avoid | |
194 | * starvation of particular inodes when others are being redirtied, prevent | |
195 | * livelocks, etc. | |
196 | * | |
197 | * Called under inode_lock. | |
198 | */ | |
199 | static int | |
200 | __sync_single_inode(struct inode *inode, struct writeback_control *wbc) | |
201 | { | |
202 | unsigned dirty; | |
203 | struct address_space *mapping = inode->i_mapping; | |
1da177e4 LT |
204 | int wait = wbc->sync_mode == WB_SYNC_ALL; |
205 | int ret; | |
206 | ||
207 | BUG_ON(inode->i_state & I_LOCK); | |
208 | ||
209 | /* Set I_LOCK, reset I_DIRTY */ | |
210 | dirty = inode->i_state & I_DIRTY; | |
211 | inode->i_state |= I_LOCK; | |
212 | inode->i_state &= ~I_DIRTY; | |
213 | ||
214 | spin_unlock(&inode_lock); | |
215 | ||
216 | ret = do_writepages(mapping, wbc); | |
217 | ||
218 | /* Don't write the inode if only I_DIRTY_PAGES was set */ | |
219 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
220 | int err = write_inode(inode, wait); | |
221 | if (ret == 0) | |
222 | ret = err; | |
223 | } | |
224 | ||
225 | if (wait) { | |
226 | int err = filemap_fdatawait(mapping); | |
227 | if (ret == 0) | |
228 | ret = err; | |
229 | } | |
230 | ||
231 | spin_lock(&inode_lock); | |
232 | inode->i_state &= ~I_LOCK; | |
233 | if (!(inode->i_state & I_FREEING)) { | |
234 | if (!(inode->i_state & I_DIRTY) && | |
235 | mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
236 | /* | |
237 | * We didn't write back all the pages. nfs_writepages() | |
238 | * sometimes bales out without doing anything. Redirty | |
1b43ef91 AM |
239 | * the inode. It is moved from s_io onto s_dirty. |
240 | */ | |
241 | /* | |
242 | * akpm: if the caller was the kupdate function we put | |
243 | * this inode at the head of s_dirty so it gets first | |
244 | * consideration. Otherwise, move it to the tail, for | |
245 | * the reasons described there. I'm not really sure | |
246 | * how much sense this makes. Presumably I had a good | |
247 | * reasons for doing it this way, and I'd rather not | |
248 | * muck with it at present. | |
1da177e4 LT |
249 | */ |
250 | if (wbc->for_kupdate) { | |
251 | /* | |
252 | * For the kupdate function we leave the inode | |
253 | * at the head of sb_dirty so it will get more | |
254 | * writeout as soon as the queue becomes | |
255 | * uncongested. | |
256 | */ | |
257 | inode->i_state |= I_DIRTY_PAGES; | |
c986d1e2 | 258 | redirty_head(inode); |
1da177e4 LT |
259 | } else { |
260 | /* | |
261 | * Otherwise fully redirty the inode so that | |
262 | * other inodes on this superblock will get some | |
263 | * writeout. Otherwise heavy writing to one | |
264 | * file would indefinitely suspend writeout of | |
265 | * all the other files. | |
266 | */ | |
267 | inode->i_state |= I_DIRTY_PAGES; | |
1b43ef91 | 268 | redirty_tail(inode); |
1da177e4 LT |
269 | } |
270 | } else if (inode->i_state & I_DIRTY) { | |
271 | /* | |
272 | * Someone redirtied the inode while were writing back | |
273 | * the pages. | |
274 | */ | |
6610a0bc | 275 | redirty_tail(inode); |
1da177e4 LT |
276 | } else if (atomic_read(&inode->i_count)) { |
277 | /* | |
278 | * The inode is clean, inuse | |
279 | */ | |
280 | list_move(&inode->i_list, &inode_in_use); | |
281 | } else { | |
282 | /* | |
283 | * The inode is clean, unused | |
284 | */ | |
285 | list_move(&inode->i_list, &inode_unused); | |
1da177e4 LT |
286 | } |
287 | } | |
288 | wake_up_inode(inode); | |
289 | return ret; | |
290 | } | |
291 | ||
292 | /* | |
7f04c26d AA |
293 | * Write out an inode's dirty pages. Called under inode_lock. Either the |
294 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
295 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
1da177e4 LT |
296 | */ |
297 | static int | |
7f04c26d | 298 | __writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 LT |
299 | { |
300 | wait_queue_head_t *wqh; | |
301 | ||
7f04c26d | 302 | if (!atomic_read(&inode->i_count)) |
659603ef | 303 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); |
7f04c26d AA |
304 | else |
305 | WARN_ON(inode->i_state & I_WILL_FREE); | |
306 | ||
1da177e4 | 307 | if ((wbc->sync_mode != WB_SYNC_ALL) && (inode->i_state & I_LOCK)) { |
4b89eed9 LT |
308 | struct address_space *mapping = inode->i_mapping; |
309 | int ret; | |
310 | ||
c6945e77 | 311 | redirty_tail(inode); |
4b89eed9 LT |
312 | |
313 | /* | |
314 | * Even if we don't actually write the inode itself here, | |
315 | * we can at least start some of the data writeout.. | |
316 | */ | |
317 | spin_unlock(&inode_lock); | |
318 | ret = do_writepages(mapping, wbc); | |
319 | spin_lock(&inode_lock); | |
320 | return ret; | |
1da177e4 LT |
321 | } |
322 | ||
323 | /* | |
324 | * It's a data-integrity sync. We must wait. | |
325 | */ | |
326 | if (inode->i_state & I_LOCK) { | |
327 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_LOCK); | |
328 | ||
329 | wqh = bit_waitqueue(&inode->i_state, __I_LOCK); | |
330 | do { | |
1da177e4 LT |
331 | spin_unlock(&inode_lock); |
332 | __wait_on_bit(wqh, &wq, inode_wait, | |
333 | TASK_UNINTERRUPTIBLE); | |
1da177e4 LT |
334 | spin_lock(&inode_lock); |
335 | } while (inode->i_state & I_LOCK); | |
336 | } | |
337 | return __sync_single_inode(inode, wbc); | |
338 | } | |
339 | ||
340 | /* | |
341 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
342 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
343 | * | |
344 | * If older_than_this is non-NULL, then only write out inodes which | |
345 | * had their first dirtying at a time earlier than *older_than_this. | |
346 | * | |
347 | * If we're a pdlfush thread, then implement pdflush collision avoidance | |
348 | * against the entire list. | |
349 | * | |
350 | * WB_SYNC_HOLD is a hack for sys_sync(): reattach the inode to sb->s_dirty so | |
351 | * that it can be located for waiting on in __writeback_single_inode(). | |
352 | * | |
353 | * Called under inode_lock. | |
354 | * | |
355 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. | |
356 | * This function assumes that the blockdev superblock's inodes are backed by | |
357 | * a variety of queues, so all inodes are searched. For other superblocks, | |
358 | * assume that all inodes are backed by the same queue. | |
359 | * | |
360 | * FIXME: this linear search could get expensive with many fileystems. But | |
361 | * how to fix? We need to go from an address_space to all inodes which share | |
362 | * a queue with that address_space. (Easy: have a global "dirty superblocks" | |
363 | * list). | |
364 | * | |
365 | * The inodes to be written are parked on sb->s_io. They are moved back onto | |
366 | * sb->s_dirty as they are selected for writing. This way, none can be missed | |
367 | * on the writer throttling path, and we get decent balancing between many | |
368 | * throttled threads: we don't want them all piling up on __wait_on_inode. | |
369 | */ | |
370 | static void | |
371 | sync_sb_inodes(struct super_block *sb, struct writeback_control *wbc) | |
372 | { | |
373 | const unsigned long start = jiffies; /* livelock avoidance */ | |
374 | ||
375 | if (!wbc->for_kupdate || list_empty(&sb->s_io)) | |
376 | list_splice_init(&sb->s_dirty, &sb->s_io); | |
377 | ||
378 | while (!list_empty(&sb->s_io)) { | |
379 | struct inode *inode = list_entry(sb->s_io.prev, | |
380 | struct inode, i_list); | |
381 | struct address_space *mapping = inode->i_mapping; | |
382 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
383 | long pages_skipped; | |
384 | ||
385 | if (!bdi_cap_writeback_dirty(bdi)) { | |
9852a0e7 | 386 | redirty_tail(inode); |
7b0de42d | 387 | if (sb_is_blkdev_sb(sb)) { |
1da177e4 LT |
388 | /* |
389 | * Dirty memory-backed blockdev: the ramdisk | |
390 | * driver does this. Skip just this inode | |
391 | */ | |
392 | continue; | |
393 | } | |
394 | /* | |
395 | * Dirty memory-backed inode against a filesystem other | |
396 | * than the kernel-internal bdev filesystem. Skip the | |
397 | * entire superblock. | |
398 | */ | |
399 | break; | |
400 | } | |
401 | ||
402 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
403 | wbc->encountered_congestion = 1; | |
7b0de42d | 404 | if (!sb_is_blkdev_sb(sb)) |
1da177e4 LT |
405 | break; /* Skip a congested fs */ |
406 | list_move(&inode->i_list, &sb->s_dirty); | |
407 | continue; /* Skip a congested blockdev */ | |
408 | } | |
409 | ||
410 | if (wbc->bdi && bdi != wbc->bdi) { | |
7b0de42d | 411 | if (!sb_is_blkdev_sb(sb)) |
1da177e4 LT |
412 | break; /* fs has the wrong queue */ |
413 | list_move(&inode->i_list, &sb->s_dirty); | |
414 | continue; /* blockdev has wrong queue */ | |
415 | } | |
416 | ||
417 | /* Was this inode dirtied after sync_sb_inodes was called? */ | |
418 | if (time_after(inode->dirtied_when, start)) | |
419 | break; | |
420 | ||
421 | /* Was this inode dirtied too recently? */ | |
422 | if (wbc->older_than_this && time_after(inode->dirtied_when, | |
423 | *wbc->older_than_this)) | |
424 | break; | |
425 | ||
426 | /* Is another pdflush already flushing this queue? */ | |
427 | if (current_is_pdflush() && !writeback_acquire(bdi)) | |
428 | break; | |
429 | ||
430 | BUG_ON(inode->i_state & I_FREEING); | |
431 | __iget(inode); | |
432 | pages_skipped = wbc->pages_skipped; | |
433 | __writeback_single_inode(inode, wbc); | |
434 | if (wbc->sync_mode == WB_SYNC_HOLD) { | |
435 | inode->dirtied_when = jiffies; | |
436 | list_move(&inode->i_list, &sb->s_dirty); | |
437 | } | |
438 | if (current_is_pdflush()) | |
439 | writeback_release(bdi); | |
440 | if (wbc->pages_skipped != pages_skipped) { | |
441 | /* | |
442 | * writeback is not making progress due to locked | |
443 | * buffers. Skip this inode for now. | |
444 | */ | |
f57b9b7b | 445 | redirty_tail(inode); |
1da177e4 LT |
446 | } |
447 | spin_unlock(&inode_lock); | |
1da177e4 | 448 | iput(inode); |
4ffc8444 | 449 | cond_resched(); |
1da177e4 LT |
450 | spin_lock(&inode_lock); |
451 | if (wbc->nr_to_write <= 0) | |
452 | break; | |
453 | } | |
454 | return; /* Leave any unwritten inodes on s_io */ | |
455 | } | |
456 | ||
457 | /* | |
458 | * Start writeback of dirty pagecache data against all unlocked inodes. | |
459 | * | |
460 | * Note: | |
461 | * We don't need to grab a reference to superblock here. If it has non-empty | |
462 | * ->s_dirty it's hadn't been killed yet and kill_super() won't proceed | |
463 | * past sync_inodes_sb() until both the ->s_dirty and ->s_io lists are | |
464 | * empty. Since __sync_single_inode() regains inode_lock before it finally moves | |
465 | * inode from superblock lists we are OK. | |
466 | * | |
467 | * If `older_than_this' is non-zero then only flush inodes which have a | |
468 | * flushtime older than *older_than_this. | |
469 | * | |
470 | * If `bdi' is non-zero then we will scan the first inode against each | |
471 | * superblock until we find the matching ones. One group will be the dirty | |
472 | * inodes against a filesystem. Then when we hit the dummy blockdev superblock, | |
473 | * sync_sb_inodes will seekout the blockdev which matches `bdi'. Maybe not | |
474 | * super-efficient but we're about to do a ton of I/O... | |
475 | */ | |
476 | void | |
477 | writeback_inodes(struct writeback_control *wbc) | |
478 | { | |
479 | struct super_block *sb; | |
480 | ||
481 | might_sleep(); | |
482 | spin_lock(&sb_lock); | |
483 | restart: | |
484 | sb = sb_entry(super_blocks.prev); | |
485 | for (; sb != sb_entry(&super_blocks); sb = sb_entry(sb->s_list.prev)) { | |
486 | if (!list_empty(&sb->s_dirty) || !list_empty(&sb->s_io)) { | |
487 | /* we're making our own get_super here */ | |
488 | sb->s_count++; | |
489 | spin_unlock(&sb_lock); | |
490 | /* | |
491 | * If we can't get the readlock, there's no sense in | |
492 | * waiting around, most of the time the FS is going to | |
493 | * be unmounted by the time it is released. | |
494 | */ | |
495 | if (down_read_trylock(&sb->s_umount)) { | |
496 | if (sb->s_root) { | |
497 | spin_lock(&inode_lock); | |
498 | sync_sb_inodes(sb, wbc); | |
499 | spin_unlock(&inode_lock); | |
500 | } | |
501 | up_read(&sb->s_umount); | |
502 | } | |
503 | spin_lock(&sb_lock); | |
504 | if (__put_super_and_need_restart(sb)) | |
505 | goto restart; | |
506 | } | |
507 | if (wbc->nr_to_write <= 0) | |
508 | break; | |
509 | } | |
510 | spin_unlock(&sb_lock); | |
511 | } | |
512 | ||
513 | /* | |
514 | * writeback and wait upon the filesystem's dirty inodes. The caller will | |
515 | * do this in two passes - one to write, and one to wait. WB_SYNC_HOLD is | |
516 | * used to park the written inodes on sb->s_dirty for the wait pass. | |
517 | * | |
518 | * A finite limit is set on the number of pages which will be written. | |
519 | * To prevent infinite livelock of sys_sync(). | |
520 | * | |
521 | * We add in the number of potentially dirty inodes, because each inode write | |
522 | * can dirty pagecache in the underlying blockdev. | |
523 | */ | |
524 | void sync_inodes_sb(struct super_block *sb, int wait) | |
525 | { | |
526 | struct writeback_control wbc = { | |
527 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_HOLD, | |
111ebb6e OH |
528 | .range_start = 0, |
529 | .range_end = LLONG_MAX, | |
1da177e4 | 530 | }; |
b1e7a8fd | 531 | unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); |
fd39fc85 | 532 | unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); |
1da177e4 LT |
533 | |
534 | wbc.nr_to_write = nr_dirty + nr_unstable + | |
535 | (inodes_stat.nr_inodes - inodes_stat.nr_unused) + | |
536 | nr_dirty + nr_unstable; | |
537 | wbc.nr_to_write += wbc.nr_to_write / 2; /* Bit more for luck */ | |
538 | spin_lock(&inode_lock); | |
539 | sync_sb_inodes(sb, &wbc); | |
540 | spin_unlock(&inode_lock); | |
541 | } | |
542 | ||
543 | /* | |
544 | * Rather lame livelock avoidance. | |
545 | */ | |
546 | static void set_sb_syncing(int val) | |
547 | { | |
548 | struct super_block *sb; | |
549 | spin_lock(&sb_lock); | |
550 | sb = sb_entry(super_blocks.prev); | |
551 | for (; sb != sb_entry(&super_blocks); sb = sb_entry(sb->s_list.prev)) { | |
552 | sb->s_syncing = val; | |
553 | } | |
554 | spin_unlock(&sb_lock); | |
555 | } | |
556 | ||
1da177e4 | 557 | /** |
67be2dd1 MW |
558 | * sync_inodes - writes all inodes to disk |
559 | * @wait: wait for completion | |
1da177e4 LT |
560 | * |
561 | * sync_inodes() goes through each super block's dirty inode list, writes the | |
562 | * inodes out, waits on the writeout and puts the inodes back on the normal | |
563 | * list. | |
564 | * | |
565 | * This is for sys_sync(). fsync_dev() uses the same algorithm. The subtle | |
566 | * part of the sync functions is that the blockdev "superblock" is processed | |
567 | * last. This is because the write_inode() function of a typical fs will | |
568 | * perform no I/O, but will mark buffers in the blockdev mapping as dirty. | |
569 | * What we want to do is to perform all that dirtying first, and then write | |
570 | * back all those inode blocks via the blockdev mapping in one sweep. So the | |
571 | * additional (somewhat redundant) sync_blockdev() calls here are to make | |
572 | * sure that really happens. Because if we call sync_inodes_sb(wait=1) with | |
573 | * outstanding dirty inodes, the writeback goes block-at-a-time within the | |
574 | * filesystem's write_inode(). This is extremely slow. | |
575 | */ | |
618f0636 | 576 | static void __sync_inodes(int wait) |
1da177e4 LT |
577 | { |
578 | struct super_block *sb; | |
579 | ||
618f0636 KK |
580 | spin_lock(&sb_lock); |
581 | restart: | |
582 | list_for_each_entry(sb, &super_blocks, s_list) { | |
583 | if (sb->s_syncing) | |
584 | continue; | |
585 | sb->s_syncing = 1; | |
586 | sb->s_count++; | |
587 | spin_unlock(&sb_lock); | |
588 | down_read(&sb->s_umount); | |
589 | if (sb->s_root) { | |
590 | sync_inodes_sb(sb, wait); | |
591 | sync_blockdev(sb->s_bdev); | |
592 | } | |
593 | up_read(&sb->s_umount); | |
594 | spin_lock(&sb_lock); | |
595 | if (__put_super_and_need_restart(sb)) | |
596 | goto restart; | |
1da177e4 | 597 | } |
618f0636 KK |
598 | spin_unlock(&sb_lock); |
599 | } | |
600 | ||
601 | void sync_inodes(int wait) | |
602 | { | |
603 | set_sb_syncing(0); | |
604 | __sync_inodes(0); | |
605 | ||
1da177e4 LT |
606 | if (wait) { |
607 | set_sb_syncing(0); | |
618f0636 | 608 | __sync_inodes(1); |
1da177e4 LT |
609 | } |
610 | } | |
611 | ||
612 | /** | |
7f04c26d AA |
613 | * write_inode_now - write an inode to disk |
614 | * @inode: inode to write to disk | |
615 | * @sync: whether the write should be synchronous or not | |
616 | * | |
617 | * This function commits an inode to disk immediately if it is dirty. This is | |
618 | * primarily needed by knfsd. | |
1da177e4 | 619 | * |
7f04c26d | 620 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 621 | */ |
1da177e4 LT |
622 | int write_inode_now(struct inode *inode, int sync) |
623 | { | |
624 | int ret; | |
625 | struct writeback_control wbc = { | |
626 | .nr_to_write = LONG_MAX, | |
627 | .sync_mode = WB_SYNC_ALL, | |
111ebb6e OH |
628 | .range_start = 0, |
629 | .range_end = LLONG_MAX, | |
1da177e4 LT |
630 | }; |
631 | ||
632 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 633 | wbc.nr_to_write = 0; |
1da177e4 LT |
634 | |
635 | might_sleep(); | |
636 | spin_lock(&inode_lock); | |
637 | ret = __writeback_single_inode(inode, &wbc); | |
638 | spin_unlock(&inode_lock); | |
639 | if (sync) | |
640 | wait_on_inode(inode); | |
641 | return ret; | |
642 | } | |
643 | EXPORT_SYMBOL(write_inode_now); | |
644 | ||
645 | /** | |
646 | * sync_inode - write an inode and its pages to disk. | |
647 | * @inode: the inode to sync | |
648 | * @wbc: controls the writeback mode | |
649 | * | |
650 | * sync_inode() will write an inode and its pages to disk. It will also | |
651 | * correctly update the inode on its superblock's dirty inode lists and will | |
652 | * update inode->i_state. | |
653 | * | |
654 | * The caller must have a ref on the inode. | |
655 | */ | |
656 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
657 | { | |
658 | int ret; | |
659 | ||
660 | spin_lock(&inode_lock); | |
661 | ret = __writeback_single_inode(inode, wbc); | |
662 | spin_unlock(&inode_lock); | |
663 | return ret; | |
664 | } | |
665 | EXPORT_SYMBOL(sync_inode); | |
666 | ||
667 | /** | |
668 | * generic_osync_inode - flush all dirty data for a given inode to disk | |
669 | * @inode: inode to write | |
67be2dd1 | 670 | * @mapping: the address_space that should be flushed |
1da177e4 LT |
671 | * @what: what to write and wait upon |
672 | * | |
673 | * This can be called by file_write functions for files which have the | |
674 | * O_SYNC flag set, to flush dirty writes to disk. | |
675 | * | |
676 | * @what is a bitmask, specifying which part of the inode's data should be | |
b8887e6e | 677 | * written and waited upon. |
1da177e4 LT |
678 | * |
679 | * OSYNC_DATA: i_mapping's dirty data | |
680 | * OSYNC_METADATA: the buffers at i_mapping->private_list | |
681 | * OSYNC_INODE: the inode itself | |
682 | */ | |
683 | ||
684 | int generic_osync_inode(struct inode *inode, struct address_space *mapping, int what) | |
685 | { | |
686 | int err = 0; | |
687 | int need_write_inode_now = 0; | |
688 | int err2; | |
689 | ||
1da177e4 LT |
690 | if (what & OSYNC_DATA) |
691 | err = filemap_fdatawrite(mapping); | |
692 | if (what & (OSYNC_METADATA|OSYNC_DATA)) { | |
693 | err2 = sync_mapping_buffers(mapping); | |
694 | if (!err) | |
695 | err = err2; | |
696 | } | |
697 | if (what & OSYNC_DATA) { | |
698 | err2 = filemap_fdatawait(mapping); | |
699 | if (!err) | |
700 | err = err2; | |
701 | } | |
1da177e4 LT |
702 | |
703 | spin_lock(&inode_lock); | |
704 | if ((inode->i_state & I_DIRTY) && | |
705 | ((what & OSYNC_INODE) || (inode->i_state & I_DIRTY_DATASYNC))) | |
706 | need_write_inode_now = 1; | |
707 | spin_unlock(&inode_lock); | |
708 | ||
709 | if (need_write_inode_now) { | |
710 | err2 = write_inode_now(inode, 1); | |
711 | if (!err) | |
712 | err = err2; | |
713 | } | |
714 | else | |
715 | wait_on_inode(inode); | |
716 | ||
717 | return err; | |
718 | } | |
719 | ||
720 | EXPORT_SYMBOL(generic_osync_inode); | |
721 | ||
722 | /** | |
723 | * writeback_acquire: attempt to get exclusive writeback access to a device | |
724 | * @bdi: the device's backing_dev_info structure | |
725 | * | |
726 | * It is a waste of resources to have more than one pdflush thread blocked on | |
727 | * a single request queue. Exclusion at the request_queue level is obtained | |
728 | * via a flag in the request_queue's backing_dev_info.state. | |
729 | * | |
730 | * Non-request_queue-backed address_spaces will share default_backing_dev_info, | |
731 | * unless they implement their own. Which is somewhat inefficient, as this | |
732 | * may prevent concurrent writeback against multiple devices. | |
733 | */ | |
734 | int writeback_acquire(struct backing_dev_info *bdi) | |
735 | { | |
736 | return !test_and_set_bit(BDI_pdflush, &bdi->state); | |
737 | } | |
738 | ||
739 | /** | |
740 | * writeback_in_progress: determine whether there is writeback in progress | |
1da177e4 | 741 | * @bdi: the device's backing_dev_info structure. |
b8887e6e RD |
742 | * |
743 | * Determine whether there is writeback in progress against a backing device. | |
1da177e4 LT |
744 | */ |
745 | int writeback_in_progress(struct backing_dev_info *bdi) | |
746 | { | |
747 | return test_bit(BDI_pdflush, &bdi->state); | |
748 | } | |
749 | ||
750 | /** | |
751 | * writeback_release: relinquish exclusive writeback access against a device. | |
752 | * @bdi: the device's backing_dev_info structure | |
753 | */ | |
754 | void writeback_release(struct backing_dev_info *bdi) | |
755 | { | |
756 | BUG_ON(!writeback_in_progress(bdi)); | |
757 | clear_bit(BDI_pdflush, &bdi->state); | |
758 | } |