diff options
author | Linus Torvalds <torvalds@ppc970.osdl.org> | 2005-04-16 18:20:36 -0400 |
---|---|---|
committer | Linus Torvalds <torvalds@ppc970.osdl.org> | 2005-04-16 18:20:36 -0400 |
commit | 1da177e4c3f41524e886b7f1b8a0c1fc7321cac2 (patch) | |
tree | 0bba044c4ce775e45a88a51686b5d9f90697ea9d /fs/fs-writeback.c |
Linux-2.6.12-rc2v2.6.12-rc2
Initial git repository build. I'm not bothering with the full history,
even though we have it. We can create a separate "historical" git
archive of that later if we want to, and in the meantime it's about
3.2GB when imported into git - space that would just make the early
git days unnecessarily complicated, when we don't have a lot of good
infrastructure for it.
Let it rip!
Diffstat (limited to 'fs/fs-writeback.c')
-rw-r--r-- | fs/fs-writeback.c | 695 |
1 files changed, 695 insertions, 0 deletions
diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c new file mode 100644 index 000000000000..d6efb36cab2a --- /dev/null +++ b/fs/fs-writeback.c | |||
@@ -0,0 +1,695 @@ | |||
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> | ||
17 | #include <linux/spinlock.h> | ||
18 | #include <linux/sched.h> | ||
19 | #include <linux/fs.h> | ||
20 | #include <linux/mm.h> | ||
21 | #include <linux/writeback.h> | ||
22 | #include <linux/blkdev.h> | ||
23 | #include <linux/backing-dev.h> | ||
24 | #include <linux/buffer_head.h> | ||
25 | |||
26 | extern struct super_block *blockdev_superblock; | ||
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 | |||
143 | /* | ||
144 | * Write a single inode's dirty pages and inode data out to disk. | ||
145 | * If `wait' is set, wait on the writeout. | ||
146 | * | ||
147 | * The whole writeout design is quite complex and fragile. We want to avoid | ||
148 | * starvation of particular inodes when others are being redirtied, prevent | ||
149 | * livelocks, etc. | ||
150 | * | ||
151 | * Called under inode_lock. | ||
152 | */ | ||
153 | static int | ||
154 | __sync_single_inode(struct inode *inode, struct writeback_control *wbc) | ||
155 | { | ||
156 | unsigned dirty; | ||
157 | struct address_space *mapping = inode->i_mapping; | ||
158 | struct super_block *sb = inode->i_sb; | ||
159 | int wait = wbc->sync_mode == WB_SYNC_ALL; | ||
160 | int ret; | ||
161 | |||
162 | BUG_ON(inode->i_state & I_LOCK); | ||
163 | |||
164 | /* Set I_LOCK, reset I_DIRTY */ | ||
165 | dirty = inode->i_state & I_DIRTY; | ||
166 | inode->i_state |= I_LOCK; | ||
167 | inode->i_state &= ~I_DIRTY; | ||
168 | |||
169 | spin_unlock(&inode_lock); | ||
170 | |||
171 | ret = do_writepages(mapping, wbc); | ||
172 | |||
173 | /* Don't write the inode if only I_DIRTY_PAGES was set */ | ||
174 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | ||
175 | int err = write_inode(inode, wait); | ||
176 | if (ret == 0) | ||
177 | ret = err; | ||
178 | } | ||
179 | |||
180 | if (wait) { | ||
181 | int err = filemap_fdatawait(mapping); | ||
182 | if (ret == 0) | ||
183 | ret = err; | ||
184 | } | ||
185 | |||
186 | spin_lock(&inode_lock); | ||
187 | inode->i_state &= ~I_LOCK; | ||
188 | if (!(inode->i_state & I_FREEING)) { | ||
189 | if (!(inode->i_state & I_DIRTY) && | ||
190 | mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | ||
191 | /* | ||
192 | * We didn't write back all the pages. nfs_writepages() | ||
193 | * sometimes bales out without doing anything. Redirty | ||
194 | * the inode. It is still on sb->s_io. | ||
195 | */ | ||
196 | if (wbc->for_kupdate) { | ||
197 | /* | ||
198 | * For the kupdate function we leave the inode | ||
199 | * at the head of sb_dirty so it will get more | ||
200 | * writeout as soon as the queue becomes | ||
201 | * uncongested. | ||
202 | */ | ||
203 | inode->i_state |= I_DIRTY_PAGES; | ||
204 | list_move_tail(&inode->i_list, &sb->s_dirty); | ||
205 | } else { | ||
206 | /* | ||
207 | * Otherwise fully redirty the inode so that | ||
208 | * other inodes on this superblock will get some | ||
209 | * writeout. Otherwise heavy writing to one | ||
210 | * file would indefinitely suspend writeout of | ||
211 | * all the other files. | ||
212 | */ | ||
213 | inode->i_state |= I_DIRTY_PAGES; | ||
214 | inode->dirtied_when = jiffies; | ||
215 | list_move(&inode->i_list, &sb->s_dirty); | ||
216 | } | ||
217 | } else if (inode->i_state & I_DIRTY) { | ||
218 | /* | ||
219 | * Someone redirtied the inode while were writing back | ||
220 | * the pages. | ||
221 | */ | ||
222 | list_move(&inode->i_list, &sb->s_dirty); | ||
223 | } else if (atomic_read(&inode->i_count)) { | ||
224 | /* | ||
225 | * The inode is clean, inuse | ||
226 | */ | ||
227 | list_move(&inode->i_list, &inode_in_use); | ||
228 | } else { | ||
229 | /* | ||
230 | * The inode is clean, unused | ||
231 | */ | ||
232 | list_move(&inode->i_list, &inode_unused); | ||
233 | inodes_stat.nr_unused++; | ||
234 | } | ||
235 | } | ||
236 | wake_up_inode(inode); | ||
237 | return ret; | ||
238 | } | ||
239 | |||
240 | /* | ||
241 | * Write out an inode's dirty pages. Called under inode_lock. | ||
242 | */ | ||
243 | static int | ||
244 | __writeback_single_inode(struct inode *inode, | ||
245 | struct writeback_control *wbc) | ||
246 | { | ||
247 | wait_queue_head_t *wqh; | ||
248 | |||
249 | if ((wbc->sync_mode != WB_SYNC_ALL) && (inode->i_state & I_LOCK)) { | ||
250 | list_move(&inode->i_list, &inode->i_sb->s_dirty); | ||
251 | return 0; | ||
252 | } | ||
253 | |||
254 | /* | ||
255 | * It's a data-integrity sync. We must wait. | ||
256 | */ | ||
257 | if (inode->i_state & I_LOCK) { | ||
258 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_LOCK); | ||
259 | |||
260 | wqh = bit_waitqueue(&inode->i_state, __I_LOCK); | ||
261 | do { | ||
262 | __iget(inode); | ||
263 | spin_unlock(&inode_lock); | ||
264 | __wait_on_bit(wqh, &wq, inode_wait, | ||
265 | TASK_UNINTERRUPTIBLE); | ||
266 | iput(inode); | ||
267 | spin_lock(&inode_lock); | ||
268 | } while (inode->i_state & I_LOCK); | ||
269 | } | ||
270 | return __sync_single_inode(inode, wbc); | ||
271 | } | ||
272 | |||
273 | /* | ||
274 | * Write out a superblock's list of dirty inodes. A wait will be performed | ||
275 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | ||
276 | * | ||
277 | * If older_than_this is non-NULL, then only write out inodes which | ||
278 | * had their first dirtying at a time earlier than *older_than_this. | ||
279 | * | ||
280 | * If we're a pdlfush thread, then implement pdflush collision avoidance | ||
281 | * against the entire list. | ||
282 | * | ||
283 | * WB_SYNC_HOLD is a hack for sys_sync(): reattach the inode to sb->s_dirty so | ||
284 | * that it can be located for waiting on in __writeback_single_inode(). | ||
285 | * | ||
286 | * Called under inode_lock. | ||
287 | * | ||
288 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. | ||
289 | * This function assumes that the blockdev superblock's inodes are backed by | ||
290 | * a variety of queues, so all inodes are searched. For other superblocks, | ||
291 | * assume that all inodes are backed by the same queue. | ||
292 | * | ||
293 | * FIXME: this linear search could get expensive with many fileystems. But | ||
294 | * how to fix? We need to go from an address_space to all inodes which share | ||
295 | * a queue with that address_space. (Easy: have a global "dirty superblocks" | ||
296 | * list). | ||
297 | * | ||
298 | * The inodes to be written are parked on sb->s_io. They are moved back onto | ||
299 | * sb->s_dirty as they are selected for writing. This way, none can be missed | ||
300 | * on the writer throttling path, and we get decent balancing between many | ||
301 | * throttled threads: we don't want them all piling up on __wait_on_inode. | ||
302 | */ | ||
303 | static void | ||
304 | sync_sb_inodes(struct super_block *sb, struct writeback_control *wbc) | ||
305 | { | ||
306 | const unsigned long start = jiffies; /* livelock avoidance */ | ||
307 | |||
308 | if (!wbc->for_kupdate || list_empty(&sb->s_io)) | ||
309 | list_splice_init(&sb->s_dirty, &sb->s_io); | ||
310 | |||
311 | while (!list_empty(&sb->s_io)) { | ||
312 | struct inode *inode = list_entry(sb->s_io.prev, | ||
313 | struct inode, i_list); | ||
314 | struct address_space *mapping = inode->i_mapping; | ||
315 | struct backing_dev_info *bdi = mapping->backing_dev_info; | ||
316 | long pages_skipped; | ||
317 | |||
318 | if (!bdi_cap_writeback_dirty(bdi)) { | ||
319 | list_move(&inode->i_list, &sb->s_dirty); | ||
320 | if (sb == blockdev_superblock) { | ||
321 | /* | ||
322 | * Dirty memory-backed blockdev: the ramdisk | ||
323 | * driver does this. Skip just this inode | ||
324 | */ | ||
325 | continue; | ||
326 | } | ||
327 | /* | ||
328 | * Dirty memory-backed inode against a filesystem other | ||
329 | * than the kernel-internal bdev filesystem. Skip the | ||
330 | * entire superblock. | ||
331 | */ | ||
332 | break; | ||
333 | } | ||
334 | |||
335 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | ||
336 | wbc->encountered_congestion = 1; | ||
337 | if (sb != blockdev_superblock) | ||
338 | break; /* Skip a congested fs */ | ||
339 | list_move(&inode->i_list, &sb->s_dirty); | ||
340 | continue; /* Skip a congested blockdev */ | ||
341 | } | ||
342 | |||
343 | if (wbc->bdi && bdi != wbc->bdi) { | ||
344 | if (sb != blockdev_superblock) | ||
345 | break; /* fs has the wrong queue */ | ||
346 | list_move(&inode->i_list, &sb->s_dirty); | ||
347 | continue; /* blockdev has wrong queue */ | ||
348 | } | ||
349 | |||
350 | /* Was this inode dirtied after sync_sb_inodes was called? */ | ||
351 | if (time_after(inode->dirtied_when, start)) | ||
352 | break; | ||
353 | |||
354 | /* Was this inode dirtied too recently? */ | ||
355 | if (wbc->older_than_this && time_after(inode->dirtied_when, | ||
356 | *wbc->older_than_this)) | ||
357 | break; | ||
358 | |||
359 | /* Is another pdflush already flushing this queue? */ | ||
360 | if (current_is_pdflush() && !writeback_acquire(bdi)) | ||
361 | break; | ||
362 | |||
363 | BUG_ON(inode->i_state & I_FREEING); | ||
364 | __iget(inode); | ||
365 | pages_skipped = wbc->pages_skipped; | ||
366 | __writeback_single_inode(inode, wbc); | ||
367 | if (wbc->sync_mode == WB_SYNC_HOLD) { | ||
368 | inode->dirtied_when = jiffies; | ||
369 | list_move(&inode->i_list, &sb->s_dirty); | ||
370 | } | ||
371 | if (current_is_pdflush()) | ||
372 | writeback_release(bdi); | ||
373 | if (wbc->pages_skipped != pages_skipped) { | ||
374 | /* | ||
375 | * writeback is not making progress due to locked | ||
376 | * buffers. Skip this inode for now. | ||
377 | */ | ||
378 | list_move(&inode->i_list, &sb->s_dirty); | ||
379 | } | ||
380 | spin_unlock(&inode_lock); | ||
381 | cond_resched(); | ||
382 | iput(inode); | ||
383 | spin_lock(&inode_lock); | ||
384 | if (wbc->nr_to_write <= 0) | ||
385 | break; | ||
386 | } | ||
387 | return; /* Leave any unwritten inodes on s_io */ | ||
388 | } | ||
389 | |||
390 | /* | ||
391 | * Start writeback of dirty pagecache data against all unlocked inodes. | ||
392 | * | ||
393 | * Note: | ||
394 | * We don't need to grab a reference to superblock here. If it has non-empty | ||
395 | * ->s_dirty it's hadn't been killed yet and kill_super() won't proceed | ||
396 | * past sync_inodes_sb() until both the ->s_dirty and ->s_io lists are | ||
397 | * empty. Since __sync_single_inode() regains inode_lock before it finally moves | ||
398 | * inode from superblock lists we are OK. | ||
399 | * | ||
400 | * If `older_than_this' is non-zero then only flush inodes which have a | ||
401 | * flushtime older than *older_than_this. | ||
402 | * | ||
403 | * If `bdi' is non-zero then we will scan the first inode against each | ||
404 | * superblock until we find the matching ones. One group will be the dirty | ||
405 | * inodes against a filesystem. Then when we hit the dummy blockdev superblock, | ||
406 | * sync_sb_inodes will seekout the blockdev which matches `bdi'. Maybe not | ||
407 | * super-efficient but we're about to do a ton of I/O... | ||
408 | */ | ||
409 | void | ||
410 | writeback_inodes(struct writeback_control *wbc) | ||
411 | { | ||
412 | struct super_block *sb; | ||
413 | |||
414 | might_sleep(); | ||
415 | spin_lock(&sb_lock); | ||
416 | restart: | ||
417 | sb = sb_entry(super_blocks.prev); | ||
418 | for (; sb != sb_entry(&super_blocks); sb = sb_entry(sb->s_list.prev)) { | ||
419 | if (!list_empty(&sb->s_dirty) || !list_empty(&sb->s_io)) { | ||
420 | /* we're making our own get_super here */ | ||
421 | sb->s_count++; | ||
422 | spin_unlock(&sb_lock); | ||
423 | /* | ||
424 | * If we can't get the readlock, there's no sense in | ||
425 | * waiting around, most of the time the FS is going to | ||
426 | * be unmounted by the time it is released. | ||
427 | */ | ||
428 | if (down_read_trylock(&sb->s_umount)) { | ||
429 | if (sb->s_root) { | ||
430 | spin_lock(&inode_lock); | ||
431 | sync_sb_inodes(sb, wbc); | ||
432 | spin_unlock(&inode_lock); | ||
433 | } | ||
434 | up_read(&sb->s_umount); | ||
435 | } | ||
436 | spin_lock(&sb_lock); | ||
437 | if (__put_super_and_need_restart(sb)) | ||
438 | goto restart; | ||
439 | } | ||
440 | if (wbc->nr_to_write <= 0) | ||
441 | break; | ||
442 | } | ||
443 | spin_unlock(&sb_lock); | ||
444 | } | ||
445 | |||
446 | /* | ||
447 | * writeback and wait upon the filesystem's dirty inodes. The caller will | ||
448 | * do this in two passes - one to write, and one to wait. WB_SYNC_HOLD is | ||
449 | * used to park the written inodes on sb->s_dirty for the wait pass. | ||
450 | * | ||
451 | * A finite limit is set on the number of pages which will be written. | ||
452 | * To prevent infinite livelock of sys_sync(). | ||
453 | * | ||
454 | * We add in the number of potentially dirty inodes, because each inode write | ||
455 | * can dirty pagecache in the underlying blockdev. | ||
456 | */ | ||
457 | void sync_inodes_sb(struct super_block *sb, int wait) | ||
458 | { | ||
459 | struct writeback_control wbc = { | ||
460 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_HOLD, | ||
461 | }; | ||
462 | unsigned long nr_dirty = read_page_state(nr_dirty); | ||
463 | unsigned long nr_unstable = read_page_state(nr_unstable); | ||
464 | |||
465 | wbc.nr_to_write = nr_dirty + nr_unstable + | ||
466 | (inodes_stat.nr_inodes - inodes_stat.nr_unused) + | ||
467 | nr_dirty + nr_unstable; | ||
468 | wbc.nr_to_write += wbc.nr_to_write / 2; /* Bit more for luck */ | ||
469 | spin_lock(&inode_lock); | ||
470 | sync_sb_inodes(sb, &wbc); | ||
471 | spin_unlock(&inode_lock); | ||
472 | } | ||
473 | |||
474 | /* | ||
475 | * Rather lame livelock avoidance. | ||
476 | */ | ||
477 | static void set_sb_syncing(int val) | ||
478 | { | ||
479 | struct super_block *sb; | ||
480 | spin_lock(&sb_lock); | ||
481 | sb = sb_entry(super_blocks.prev); | ||
482 | for (; sb != sb_entry(&super_blocks); sb = sb_entry(sb->s_list.prev)) { | ||
483 | sb->s_syncing = val; | ||
484 | } | ||
485 | spin_unlock(&sb_lock); | ||
486 | } | ||
487 | |||
488 | /* | ||
489 | * Find a superblock with inodes that need to be synced | ||
490 | */ | ||
491 | static struct super_block *get_super_to_sync(void) | ||
492 | { | ||
493 | struct super_block *sb; | ||
494 | restart: | ||
495 | spin_lock(&sb_lock); | ||
496 | sb = sb_entry(super_blocks.prev); | ||
497 | for (; sb != sb_entry(&super_blocks); sb = sb_entry(sb->s_list.prev)) { | ||
498 | if (sb->s_syncing) | ||
499 | continue; | ||
500 | sb->s_syncing = 1; | ||
501 | sb->s_count++; | ||
502 | spin_unlock(&sb_lock); | ||
503 | down_read(&sb->s_umount); | ||
504 | if (!sb->s_root) { | ||
505 | drop_super(sb); | ||
506 | goto restart; | ||
507 | } | ||
508 | return sb; | ||
509 | } | ||
510 | spin_unlock(&sb_lock); | ||
511 | return NULL; | ||
512 | } | ||
513 | |||
514 | /** | ||
515 | * sync_inodes | ||
516 | * | ||
517 | * sync_inodes() goes through each super block's dirty inode list, writes the | ||
518 | * inodes out, waits on the writeout and puts the inodes back on the normal | ||
519 | * list. | ||
520 | * | ||
521 | * This is for sys_sync(). fsync_dev() uses the same algorithm. The subtle | ||
522 | * part of the sync functions is that the blockdev "superblock" is processed | ||
523 | * last. This is because the write_inode() function of a typical fs will | ||
524 | * perform no I/O, but will mark buffers in the blockdev mapping as dirty. | ||
525 | * What we want to do is to perform all that dirtying first, and then write | ||
526 | * back all those inode blocks via the blockdev mapping in one sweep. So the | ||
527 | * additional (somewhat redundant) sync_blockdev() calls here are to make | ||
528 | * sure that really happens. Because if we call sync_inodes_sb(wait=1) with | ||
529 | * outstanding dirty inodes, the writeback goes block-at-a-time within the | ||
530 | * filesystem's write_inode(). This is extremely slow. | ||
531 | */ | ||
532 | void sync_inodes(int wait) | ||
533 | { | ||
534 | struct super_block *sb; | ||
535 | |||
536 | set_sb_syncing(0); | ||
537 | while ((sb = get_super_to_sync()) != NULL) { | ||
538 | sync_inodes_sb(sb, 0); | ||
539 | sync_blockdev(sb->s_bdev); | ||
540 | drop_super(sb); | ||
541 | } | ||
542 | if (wait) { | ||
543 | set_sb_syncing(0); | ||
544 | while ((sb = get_super_to_sync()) != NULL) { | ||
545 | sync_inodes_sb(sb, 1); | ||
546 | sync_blockdev(sb->s_bdev); | ||
547 | drop_super(sb); | ||
548 | } | ||
549 | } | ||
550 | } | ||
551 | |||
552 | /** | ||
553 | * write_inode_now - write an inode to disk | ||
554 | * @inode: inode to write to disk | ||
555 | * @sync: whether the write should be synchronous or not | ||
556 | * | ||
557 | * This function commits an inode to disk immediately if it is | ||
558 | * dirty. This is primarily needed by knfsd. | ||
559 | */ | ||
560 | |||
561 | int write_inode_now(struct inode *inode, int sync) | ||
562 | { | ||
563 | int ret; | ||
564 | struct writeback_control wbc = { | ||
565 | .nr_to_write = LONG_MAX, | ||
566 | .sync_mode = WB_SYNC_ALL, | ||
567 | }; | ||
568 | |||
569 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | ||
570 | return 0; | ||
571 | |||
572 | might_sleep(); | ||
573 | spin_lock(&inode_lock); | ||
574 | ret = __writeback_single_inode(inode, &wbc); | ||
575 | spin_unlock(&inode_lock); | ||
576 | if (sync) | ||
577 | wait_on_inode(inode); | ||
578 | return ret; | ||
579 | } | ||
580 | EXPORT_SYMBOL(write_inode_now); | ||
581 | |||
582 | /** | ||
583 | * sync_inode - write an inode and its pages to disk. | ||
584 | * @inode: the inode to sync | ||
585 | * @wbc: controls the writeback mode | ||
586 | * | ||
587 | * sync_inode() will write an inode and its pages to disk. It will also | ||
588 | * correctly update the inode on its superblock's dirty inode lists and will | ||
589 | * update inode->i_state. | ||
590 | * | ||
591 | * The caller must have a ref on the inode. | ||
592 | */ | ||
593 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | ||
594 | { | ||
595 | int ret; | ||
596 | |||
597 | spin_lock(&inode_lock); | ||
598 | ret = __writeback_single_inode(inode, wbc); | ||
599 | spin_unlock(&inode_lock); | ||
600 | return ret; | ||
601 | } | ||
602 | EXPORT_SYMBOL(sync_inode); | ||
603 | |||
604 | /** | ||
605 | * generic_osync_inode - flush all dirty data for a given inode to disk | ||
606 | * @inode: inode to write | ||
607 | * @what: what to write and wait upon | ||
608 | * | ||
609 | * This can be called by file_write functions for files which have the | ||
610 | * O_SYNC flag set, to flush dirty writes to disk. | ||
611 | * | ||
612 | * @what is a bitmask, specifying which part of the inode's data should be | ||
613 | * written and waited upon: | ||
614 | * | ||
615 | * OSYNC_DATA: i_mapping's dirty data | ||
616 | * OSYNC_METADATA: the buffers at i_mapping->private_list | ||
617 | * OSYNC_INODE: the inode itself | ||
618 | */ | ||
619 | |||
620 | int generic_osync_inode(struct inode *inode, struct address_space *mapping, int what) | ||
621 | { | ||
622 | int err = 0; | ||
623 | int need_write_inode_now = 0; | ||
624 | int err2; | ||
625 | |||
626 | current->flags |= PF_SYNCWRITE; | ||
627 | if (what & OSYNC_DATA) | ||
628 | err = filemap_fdatawrite(mapping); | ||
629 | if (what & (OSYNC_METADATA|OSYNC_DATA)) { | ||
630 | err2 = sync_mapping_buffers(mapping); | ||
631 | if (!err) | ||
632 | err = err2; | ||
633 | } | ||
634 | if (what & OSYNC_DATA) { | ||
635 | err2 = filemap_fdatawait(mapping); | ||
636 | if (!err) | ||
637 | err = err2; | ||
638 | } | ||
639 | current->flags &= ~PF_SYNCWRITE; | ||
640 | |||
641 | spin_lock(&inode_lock); | ||
642 | if ((inode->i_state & I_DIRTY) && | ||
643 | ((what & OSYNC_INODE) || (inode->i_state & I_DIRTY_DATASYNC))) | ||
644 | need_write_inode_now = 1; | ||
645 | spin_unlock(&inode_lock); | ||
646 | |||
647 | if (need_write_inode_now) { | ||
648 | err2 = write_inode_now(inode, 1); | ||
649 | if (!err) | ||
650 | err = err2; | ||
651 | } | ||
652 | else | ||
653 | wait_on_inode(inode); | ||
654 | |||
655 | return err; | ||
656 | } | ||
657 | |||
658 | EXPORT_SYMBOL(generic_osync_inode); | ||
659 | |||
660 | /** | ||
661 | * writeback_acquire: attempt to get exclusive writeback access to a device | ||
662 | * @bdi: the device's backing_dev_info structure | ||
663 | * | ||
664 | * It is a waste of resources to have more than one pdflush thread blocked on | ||
665 | * a single request queue. Exclusion at the request_queue level is obtained | ||
666 | * via a flag in the request_queue's backing_dev_info.state. | ||
667 | * | ||
668 | * Non-request_queue-backed address_spaces will share default_backing_dev_info, | ||
669 | * unless they implement their own. Which is somewhat inefficient, as this | ||
670 | * may prevent concurrent writeback against multiple devices. | ||
671 | */ | ||
672 | int writeback_acquire(struct backing_dev_info *bdi) | ||
673 | { | ||
674 | return !test_and_set_bit(BDI_pdflush, &bdi->state); | ||
675 | } | ||
676 | |||
677 | /** | ||
678 | * writeback_in_progress: determine whether there is writeback in progress | ||
679 | * against a backing device. | ||
680 | * @bdi: the device's backing_dev_info structure. | ||
681 | */ | ||
682 | int writeback_in_progress(struct backing_dev_info *bdi) | ||
683 | { | ||
684 | return test_bit(BDI_pdflush, &bdi->state); | ||
685 | } | ||
686 | |||
687 | /** | ||
688 | * writeback_release: relinquish exclusive writeback access against a device. | ||
689 | * @bdi: the device's backing_dev_info structure | ||
690 | */ | ||
691 | void writeback_release(struct backing_dev_info *bdi) | ||
692 | { | ||
693 | BUG_ON(!writeback_in_progress(bdi)); | ||
694 | clear_bit(BDI_pdflush, &bdi->state); | ||
695 | } | ||