]>
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 | * | |
e1f8e874 | 11 | * 10Apr2002 Andrew Morton |
1da177e4 LT |
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 | 18 | #include <linux/spinlock.h> |
5a0e3ad6 | 19 | #include <linux/slab.h> |
1da177e4 LT |
20 | #include <linux/sched.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/mm.h> | |
03ba3782 JA |
23 | #include <linux/kthread.h> |
24 | #include <linux/freezer.h> | |
1da177e4 LT |
25 | #include <linux/writeback.h> |
26 | #include <linux/blkdev.h> | |
27 | #include <linux/backing-dev.h> | |
28 | #include <linux/buffer_head.h> | |
455b2864 | 29 | #include <linux/tracepoint.h> |
07f3f05c | 30 | #include "internal.h" |
1da177e4 | 31 | |
c4a77a6c JA |
32 | /* |
33 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
34 | */ | |
83ba7b07 | 35 | struct wb_writeback_work { |
c4a77a6c JA |
36 | long nr_pages; |
37 | struct super_block *sb; | |
38 | enum writeback_sync_modes sync_mode; | |
52957fe1 HS |
39 | unsigned int for_kupdate:1; |
40 | unsigned int range_cyclic:1; | |
41 | unsigned int for_background:1; | |
c4a77a6c | 42 | |
8010c3b6 | 43 | struct list_head list; /* pending work list */ |
83ba7b07 | 44 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
45 | }; |
46 | ||
455b2864 DC |
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 | ||
f11b00f3 AB |
62 | /** |
63 | * writeback_in_progress - determine whether there is writeback in progress | |
64 | * @bdi: the device's backing_dev_info structure. | |
65 | * | |
03ba3782 JA |
66 | * Determine whether there is writeback waiting to be handled against a |
67 | * backing device. | |
f11b00f3 AB |
68 | */ |
69 | int writeback_in_progress(struct backing_dev_info *bdi) | |
70 | { | |
03ba3782 | 71 | return !list_empty(&bdi->work_list); |
f11b00f3 AB |
72 | } |
73 | ||
83ba7b07 CH |
74 | static void bdi_queue_work(struct backing_dev_info *bdi, |
75 | struct wb_writeback_work *work) | |
03ba3782 | 76 | { |
455b2864 DC |
77 | trace_writeback_queue(bdi, work); |
78 | ||
bcddc3f0 | 79 | spin_lock(&bdi->wb_lock); |
83ba7b07 | 80 | list_add_tail(&work->list, &bdi->work_list); |
bcddc3f0 | 81 | spin_unlock(&bdi->wb_lock); |
03ba3782 JA |
82 | |
83 | /* | |
84 | * If the default thread isn't there, make sure we add it. When | |
85 | * it gets created and wakes up, we'll run this work. | |
86 | */ | |
c1955ce3 | 87 | if (unlikely(!bdi->wb.task)) { |
455b2864 | 88 | trace_writeback_nothread(bdi, work); |
03ba3782 | 89 | wake_up_process(default_backing_dev_info.wb.task); |
c1955ce3 | 90 | } else { |
03ba3782 | 91 | struct bdi_writeback *wb = &bdi->wb; |
1da177e4 | 92 | |
1ef7d9aa | 93 | if (wb->task) |
03ba3782 | 94 | wake_up_process(wb->task); |
1da177e4 | 95 | } |
1da177e4 LT |
96 | } |
97 | ||
83ba7b07 CH |
98 | static void |
99 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
100 | bool range_cyclic, bool for_background) | |
1da177e4 | 101 | { |
83ba7b07 | 102 | struct wb_writeback_work *work; |
03ba3782 | 103 | |
bcddc3f0 JA |
104 | /* |
105 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
106 | * wakeup the thread for old dirty data writeback | |
107 | */ | |
83ba7b07 CH |
108 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
109 | if (!work) { | |
455b2864 DC |
110 | if (bdi->wb.task) { |
111 | trace_writeback_nowork(bdi); | |
83ba7b07 | 112 | wake_up_process(bdi->wb.task); |
455b2864 | 113 | } |
83ba7b07 | 114 | return; |
bcddc3f0 | 115 | } |
03ba3782 | 116 | |
83ba7b07 CH |
117 | work->sync_mode = WB_SYNC_NONE; |
118 | work->nr_pages = nr_pages; | |
119 | work->range_cyclic = range_cyclic; | |
120 | work->for_background = for_background; | |
03ba3782 | 121 | |
83ba7b07 | 122 | bdi_queue_work(bdi, work); |
b6e51316 JA |
123 | } |
124 | ||
125 | /** | |
126 | * bdi_start_writeback - start writeback | |
127 | * @bdi: the backing device to write from | |
128 | * @nr_pages: the number of pages to write | |
129 | * | |
130 | * Description: | |
131 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
132 | * started when this function returns, we make no guarentees on | |
0e3c9a22 | 133 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
134 | * |
135 | */ | |
c5444198 | 136 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) |
b6e51316 | 137 | { |
83ba7b07 | 138 | __bdi_start_writeback(bdi, nr_pages, true, false); |
c5444198 | 139 | } |
d3ddec76 | 140 | |
c5444198 CH |
141 | /** |
142 | * bdi_start_background_writeback - start background writeback | |
143 | * @bdi: the backing device to write from | |
144 | * | |
145 | * Description: | |
146 | * This does WB_SYNC_NONE background writeback. The IO is only | |
147 | * started when this function returns, we make no guarentees on | |
148 | * completion. Caller need not hold sb s_umount semaphore. | |
149 | */ | |
150 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
151 | { | |
83ba7b07 | 152 | __bdi_start_writeback(bdi, LONG_MAX, true, true); |
1da177e4 LT |
153 | } |
154 | ||
6610a0bc AM |
155 | /* |
156 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
157 | * furthest end of its superblock's dirty-inode list. | |
158 | * | |
159 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 160 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
161 | * the case then the inode must have been redirtied while it was being written |
162 | * out and we don't reset its dirtied_when. | |
163 | */ | |
164 | static void redirty_tail(struct inode *inode) | |
165 | { | |
03ba3782 | 166 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
6610a0bc | 167 | |
03ba3782 | 168 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 169 | struct inode *tail; |
6610a0bc | 170 | |
03ba3782 | 171 | tail = list_entry(wb->b_dirty.next, struct inode, i_list); |
66f3b8e2 | 172 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
173 | inode->dirtied_when = jiffies; |
174 | } | |
03ba3782 | 175 | list_move(&inode->i_list, &wb->b_dirty); |
6610a0bc AM |
176 | } |
177 | ||
c986d1e2 | 178 | /* |
66f3b8e2 | 179 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 180 | */ |
0e0f4fc2 | 181 | static void requeue_io(struct inode *inode) |
c986d1e2 | 182 | { |
03ba3782 JA |
183 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
184 | ||
185 | list_move(&inode->i_list, &wb->b_more_io); | |
c986d1e2 AM |
186 | } |
187 | ||
1c0eeaf5 JE |
188 | static void inode_sync_complete(struct inode *inode) |
189 | { | |
190 | /* | |
191 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
192 | */ | |
193 | smp_mb(); | |
194 | wake_up_bit(&inode->i_state, __I_SYNC); | |
195 | } | |
196 | ||
d2caa3c5 JL |
197 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
198 | { | |
199 | bool ret = time_after(inode->dirtied_when, t); | |
200 | #ifndef CONFIG_64BIT | |
201 | /* | |
202 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
203 | * It _appears_ to be in the future, but is actually in distant past. | |
204 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 205 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
206 | */ |
207 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
208 | #endif | |
209 | return ret; | |
210 | } | |
211 | ||
2c136579 FW |
212 | /* |
213 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
214 | */ | |
215 | static void move_expired_inodes(struct list_head *delaying_queue, | |
216 | struct list_head *dispatch_queue, | |
217 | unsigned long *older_than_this) | |
218 | { | |
5c03449d SL |
219 | LIST_HEAD(tmp); |
220 | struct list_head *pos, *node; | |
cf137307 | 221 | struct super_block *sb = NULL; |
5c03449d | 222 | struct inode *inode; |
cf137307 | 223 | int do_sb_sort = 0; |
5c03449d | 224 | |
2c136579 | 225 | while (!list_empty(delaying_queue)) { |
5c03449d | 226 | inode = list_entry(delaying_queue->prev, struct inode, i_list); |
2c136579 | 227 | if (older_than_this && |
d2caa3c5 | 228 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 | 229 | break; |
cf137307 JA |
230 | if (sb && sb != inode->i_sb) |
231 | do_sb_sort = 1; | |
232 | sb = inode->i_sb; | |
5c03449d SL |
233 | list_move(&inode->i_list, &tmp); |
234 | } | |
235 | ||
cf137307 JA |
236 | /* just one sb in list, splice to dispatch_queue and we're done */ |
237 | if (!do_sb_sort) { | |
238 | list_splice(&tmp, dispatch_queue); | |
239 | return; | |
240 | } | |
241 | ||
5c03449d SL |
242 | /* Move inodes from one superblock together */ |
243 | while (!list_empty(&tmp)) { | |
244 | inode = list_entry(tmp.prev, struct inode, i_list); | |
245 | sb = inode->i_sb; | |
246 | list_for_each_prev_safe(pos, node, &tmp) { | |
247 | inode = list_entry(pos, struct inode, i_list); | |
248 | if (inode->i_sb == sb) | |
249 | list_move(&inode->i_list, dispatch_queue); | |
250 | } | |
2c136579 FW |
251 | } |
252 | } | |
253 | ||
254 | /* | |
255 | * Queue all expired dirty inodes for io, eldest first. | |
256 | */ | |
03ba3782 | 257 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 258 | { |
03ba3782 JA |
259 | list_splice_init(&wb->b_more_io, wb->b_io.prev); |
260 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); | |
66f3b8e2 JA |
261 | } |
262 | ||
a9185b41 | 263 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 264 | { |
03ba3782 | 265 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 266 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 267 | return 0; |
08d8e974 | 268 | } |
08d8e974 | 269 | |
1da177e4 | 270 | /* |
01c03194 CH |
271 | * Wait for writeback on an inode to complete. |
272 | */ | |
273 | static void inode_wait_for_writeback(struct inode *inode) | |
274 | { | |
275 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
276 | wait_queue_head_t *wqh; | |
277 | ||
278 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
58a9d3d8 | 279 | while (inode->i_state & I_SYNC) { |
01c03194 CH |
280 | spin_unlock(&inode_lock); |
281 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); | |
282 | spin_lock(&inode_lock); | |
58a9d3d8 | 283 | } |
01c03194 CH |
284 | } |
285 | ||
286 | /* | |
287 | * Write out an inode's dirty pages. Called under inode_lock. Either the | |
288 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
289 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
290 | * | |
1da177e4 LT |
291 | * If `wait' is set, wait on the writeout. |
292 | * | |
293 | * The whole writeout design is quite complex and fragile. We want to avoid | |
294 | * starvation of particular inodes when others are being redirtied, prevent | |
295 | * livelocks, etc. | |
296 | * | |
297 | * Called under inode_lock. | |
298 | */ | |
299 | static int | |
01c03194 | 300 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 301 | { |
1da177e4 | 302 | struct address_space *mapping = inode->i_mapping; |
01c03194 | 303 | unsigned dirty; |
1da177e4 LT |
304 | int ret; |
305 | ||
01c03194 CH |
306 | if (!atomic_read(&inode->i_count)) |
307 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
308 | else | |
309 | WARN_ON(inode->i_state & I_WILL_FREE); | |
310 | ||
311 | if (inode->i_state & I_SYNC) { | |
312 | /* | |
313 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 314 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
315 | * writeback can proceed with the other inodes on s_io. |
316 | * | |
317 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 318 | * completed a full scan of b_io. |
01c03194 | 319 | */ |
a9185b41 | 320 | if (wbc->sync_mode != WB_SYNC_ALL) { |
01c03194 CH |
321 | requeue_io(inode); |
322 | return 0; | |
323 | } | |
324 | ||
325 | /* | |
326 | * It's a data-integrity sync. We must wait. | |
327 | */ | |
328 | inode_wait_for_writeback(inode); | |
329 | } | |
330 | ||
1c0eeaf5 | 331 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 332 | |
5547e8aa | 333 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 334 | inode->i_state |= I_SYNC; |
5547e8aa | 335 | inode->i_state &= ~I_DIRTY_PAGES; |
1da177e4 LT |
336 | spin_unlock(&inode_lock); |
337 | ||
338 | ret = do_writepages(mapping, wbc); | |
339 | ||
26821ed4 CH |
340 | /* |
341 | * Make sure to wait on the data before writing out the metadata. | |
342 | * This is important for filesystems that modify metadata on data | |
343 | * I/O completion. | |
344 | */ | |
a9185b41 | 345 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 346 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
347 | if (ret == 0) |
348 | ret = err; | |
349 | } | |
350 | ||
5547e8aa DM |
351 | /* |
352 | * Some filesystems may redirty the inode during the writeback | |
353 | * due to delalloc, clear dirty metadata flags right before | |
354 | * write_inode() | |
355 | */ | |
356 | spin_lock(&inode_lock); | |
357 | dirty = inode->i_state & I_DIRTY; | |
358 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
359 | spin_unlock(&inode_lock); | |
26821ed4 CH |
360 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
361 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 362 | int err = write_inode(inode, wbc); |
1da177e4 LT |
363 | if (ret == 0) |
364 | ret = err; | |
365 | } | |
366 | ||
367 | spin_lock(&inode_lock); | |
1c0eeaf5 | 368 | inode->i_state &= ~I_SYNC; |
84a89245 | 369 | if (!(inode->i_state & (I_FREEING | I_CLEAR))) { |
b3af9468 | 370 | if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) { |
ae1b7f7d | 371 | /* |
b3af9468 WF |
372 | * More pages get dirtied by a fast dirtier. |
373 | */ | |
374 | goto select_queue; | |
375 | } else if (inode->i_state & I_DIRTY) { | |
376 | /* | |
377 | * At least XFS will redirty the inode during the | |
378 | * writeback (delalloc) and on io completion (isize). | |
ae1b7f7d WF |
379 | */ |
380 | redirty_tail(inode); | |
381 | } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
1da177e4 LT |
382 | /* |
383 | * We didn't write back all the pages. nfs_writepages() | |
384 | * sometimes bales out without doing anything. Redirty | |
66f3b8e2 | 385 | * the inode; Move it from b_io onto b_more_io/b_dirty. |
1b43ef91 AM |
386 | */ |
387 | /* | |
388 | * akpm: if the caller was the kupdate function we put | |
66f3b8e2 | 389 | * this inode at the head of b_dirty so it gets first |
1b43ef91 AM |
390 | * consideration. Otherwise, move it to the tail, for |
391 | * the reasons described there. I'm not really sure | |
392 | * how much sense this makes. Presumably I had a good | |
393 | * reasons for doing it this way, and I'd rather not | |
394 | * muck with it at present. | |
1da177e4 LT |
395 | */ |
396 | if (wbc->for_kupdate) { | |
397 | /* | |
2c136579 | 398 | * For the kupdate function we move the inode |
66f3b8e2 | 399 | * to b_more_io so it will get more writeout as |
2c136579 | 400 | * soon as the queue becomes uncongested. |
1da177e4 LT |
401 | */ |
402 | inode->i_state |= I_DIRTY_PAGES; | |
b3af9468 | 403 | select_queue: |
8bc3be27 FW |
404 | if (wbc->nr_to_write <= 0) { |
405 | /* | |
406 | * slice used up: queue for next turn | |
407 | */ | |
408 | requeue_io(inode); | |
409 | } else { | |
410 | /* | |
411 | * somehow blocked: retry later | |
412 | */ | |
413 | redirty_tail(inode); | |
414 | } | |
1da177e4 LT |
415 | } else { |
416 | /* | |
417 | * Otherwise fully redirty the inode so that | |
418 | * other inodes on this superblock will get some | |
419 | * writeout. Otherwise heavy writing to one | |
420 | * file would indefinitely suspend writeout of | |
421 | * all the other files. | |
422 | */ | |
423 | inode->i_state |= I_DIRTY_PAGES; | |
1b43ef91 | 424 | redirty_tail(inode); |
1da177e4 | 425 | } |
1da177e4 LT |
426 | } else if (atomic_read(&inode->i_count)) { |
427 | /* | |
428 | * The inode is clean, inuse | |
429 | */ | |
430 | list_move(&inode->i_list, &inode_in_use); | |
431 | } else { | |
432 | /* | |
433 | * The inode is clean, unused | |
434 | */ | |
435 | list_move(&inode->i_list, &inode_unused); | |
1da177e4 LT |
436 | } |
437 | } | |
1c0eeaf5 | 438 | inode_sync_complete(inode); |
1da177e4 LT |
439 | return ret; |
440 | } | |
441 | ||
03ba3782 | 442 | /* |
d19de7ed | 443 | * For background writeback the caller does not have the sb pinned |
03ba3782 JA |
444 | * before calling writeback. So make sure that we do pin it, so it doesn't |
445 | * go away while we are writing inodes from it. | |
03ba3782 | 446 | */ |
d19de7ed | 447 | static bool pin_sb_for_writeback(struct super_block *sb) |
03ba3782 | 448 | { |
03ba3782 | 449 | spin_lock(&sb_lock); |
29cb4859 CH |
450 | if (list_empty(&sb->s_instances)) { |
451 | spin_unlock(&sb_lock); | |
452 | return false; | |
453 | } | |
454 | ||
03ba3782 | 455 | sb->s_count++; |
29cb4859 CH |
456 | spin_unlock(&sb_lock); |
457 | ||
03ba3782 | 458 | if (down_read_trylock(&sb->s_umount)) { |
29cb4859 | 459 | if (sb->s_root) |
d19de7ed | 460 | return true; |
03ba3782 JA |
461 | up_read(&sb->s_umount); |
462 | } | |
29cb4859 CH |
463 | |
464 | put_super(sb); | |
d19de7ed | 465 | return false; |
03ba3782 JA |
466 | } |
467 | ||
f11c9c5c ES |
468 | /* |
469 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 CH |
470 | * |
471 | * If @only_this_sb is true, then find and write all such | |
f11c9c5c ES |
472 | * inodes. Otherwise write only ones which go sequentially |
473 | * in reverse order. | |
edadfb10 | 474 | * |
f11c9c5c ES |
475 | * Return 1, if the caller writeback routine should be |
476 | * interrupted. Otherwise return 0. | |
477 | */ | |
edadfb10 CH |
478 | static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, |
479 | struct writeback_control *wbc, bool only_this_sb) | |
1da177e4 | 480 | { |
03ba3782 | 481 | while (!list_empty(&wb->b_io)) { |
1da177e4 | 482 | long pages_skipped; |
f11c9c5c ES |
483 | struct inode *inode = list_entry(wb->b_io.prev, |
484 | struct inode, i_list); | |
edadfb10 CH |
485 | |
486 | if (inode->i_sb != sb) { | |
487 | if (only_this_sb) { | |
488 | /* | |
489 | * We only want to write back data for this | |
490 | * superblock, move all inodes not belonging | |
491 | * to it back onto the dirty list. | |
492 | */ | |
493 | redirty_tail(inode); | |
494 | continue; | |
495 | } | |
496 | ||
497 | /* | |
498 | * The inode belongs to a different superblock. | |
499 | * Bounce back to the caller to unpin this and | |
500 | * pin the next superblock. | |
501 | */ | |
f11c9c5c | 502 | return 0; |
edadfb10 CH |
503 | } |
504 | ||
84a89245 | 505 | if (inode->i_state & (I_NEW | I_WILL_FREE)) { |
7ef0d737 NP |
506 | requeue_io(inode); |
507 | continue; | |
508 | } | |
d2caa3c5 JL |
509 | /* |
510 | * Was this inode dirtied after sync_sb_inodes was called? | |
511 | * This keeps sync from extra jobs and livelock. | |
512 | */ | |
f11c9c5c ES |
513 | if (inode_dirtied_after(inode, wbc->wb_start)) |
514 | return 1; | |
1da177e4 | 515 | |
84a89245 | 516 | BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); |
1da177e4 LT |
517 | __iget(inode); |
518 | pages_skipped = wbc->pages_skipped; | |
01c03194 | 519 | writeback_single_inode(inode, wbc); |
1da177e4 LT |
520 | if (wbc->pages_skipped != pages_skipped) { |
521 | /* | |
522 | * writeback is not making progress due to locked | |
523 | * buffers. Skip this inode for now. | |
524 | */ | |
f57b9b7b | 525 | redirty_tail(inode); |
1da177e4 LT |
526 | } |
527 | spin_unlock(&inode_lock); | |
1da177e4 | 528 | iput(inode); |
4ffc8444 | 529 | cond_resched(); |
1da177e4 | 530 | spin_lock(&inode_lock); |
8bc3be27 FW |
531 | if (wbc->nr_to_write <= 0) { |
532 | wbc->more_io = 1; | |
f11c9c5c | 533 | return 1; |
8bc3be27 | 534 | } |
03ba3782 | 535 | if (!list_empty(&wb->b_more_io)) |
8bc3be27 | 536 | wbc->more_io = 1; |
1da177e4 | 537 | } |
f11c9c5c ES |
538 | /* b_io is empty */ |
539 | return 1; | |
540 | } | |
541 | ||
9c3a8ee8 CH |
542 | void writeback_inodes_wb(struct bdi_writeback *wb, |
543 | struct writeback_control *wbc) | |
f11c9c5c ES |
544 | { |
545 | int ret = 0; | |
546 | ||
547 | wbc->wb_start = jiffies; /* livelock avoidance */ | |
548 | spin_lock(&inode_lock); | |
549 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) | |
550 | queue_io(wb, wbc->older_than_this); | |
38f21977 | 551 | |
f11c9c5c ES |
552 | while (!list_empty(&wb->b_io)) { |
553 | struct inode *inode = list_entry(wb->b_io.prev, | |
554 | struct inode, i_list); | |
555 | struct super_block *sb = inode->i_sb; | |
9ecc2738 | 556 | |
edadfb10 CH |
557 | if (!pin_sb_for_writeback(sb)) { |
558 | requeue_io(inode); | |
559 | continue; | |
f11c9c5c | 560 | } |
edadfb10 CH |
561 | ret = writeback_sb_inodes(sb, wb, wbc, false); |
562 | drop_super(sb); | |
f11c9c5c | 563 | |
f11c9c5c ES |
564 | if (ret) |
565 | break; | |
566 | } | |
66f3b8e2 JA |
567 | spin_unlock(&inode_lock); |
568 | /* Leave any unwritten inodes on b_io */ | |
569 | } | |
570 | ||
edadfb10 CH |
571 | static void __writeback_inodes_sb(struct super_block *sb, |
572 | struct bdi_writeback *wb, struct writeback_control *wbc) | |
573 | { | |
574 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
575 | ||
576 | wbc->wb_start = jiffies; /* livelock avoidance */ | |
577 | spin_lock(&inode_lock); | |
578 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) | |
579 | queue_io(wb, wbc->older_than_this); | |
580 | writeback_sb_inodes(sb, wb, wbc, true); | |
581 | spin_unlock(&inode_lock); | |
582 | } | |
583 | ||
66f3b8e2 | 584 | /* |
03ba3782 JA |
585 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
586 | * operation. We do this so we don't hold I_SYNC against an inode for | |
587 | * enormous amounts of time, which would block a userspace task which has | |
588 | * been forced to throttle against that inode. Also, the code reevaluates | |
589 | * the dirty each time it has written this many pages. | |
590 | */ | |
591 | #define MAX_WRITEBACK_PAGES 1024 | |
592 | ||
593 | static inline bool over_bground_thresh(void) | |
594 | { | |
595 | unsigned long background_thresh, dirty_thresh; | |
596 | ||
597 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); | |
598 | ||
599 | return (global_page_state(NR_FILE_DIRTY) + | |
600 | global_page_state(NR_UNSTABLE_NFS) >= background_thresh); | |
601 | } | |
602 | ||
603 | /* | |
604 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 605 | * |
03ba3782 JA |
606 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
607 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
608 | * just walks the superblock inode list, writing back any inodes which are | |
609 | * older than a specific point in time. | |
66f3b8e2 | 610 | * |
03ba3782 JA |
611 | * Try to run once per dirty_writeback_interval. But if a writeback event |
612 | * takes longer than a dirty_writeback_interval interval, then leave a | |
613 | * one-second gap. | |
66f3b8e2 | 614 | * |
03ba3782 JA |
615 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
616 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 617 | */ |
c4a77a6c | 618 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 619 | struct wb_writeback_work *work) |
66f3b8e2 | 620 | { |
03ba3782 | 621 | struct writeback_control wbc = { |
83ba7b07 | 622 | .sync_mode = work->sync_mode, |
03ba3782 | 623 | .older_than_this = NULL, |
83ba7b07 CH |
624 | .for_kupdate = work->for_kupdate, |
625 | .for_background = work->for_background, | |
626 | .range_cyclic = work->range_cyclic, | |
03ba3782 JA |
627 | }; |
628 | unsigned long oldest_jif; | |
629 | long wrote = 0; | |
a5989bdc | 630 | struct inode *inode; |
66f3b8e2 | 631 | |
03ba3782 JA |
632 | if (wbc.for_kupdate) { |
633 | wbc.older_than_this = &oldest_jif; | |
634 | oldest_jif = jiffies - | |
635 | msecs_to_jiffies(dirty_expire_interval * 10); | |
636 | } | |
c4a77a6c JA |
637 | if (!wbc.range_cyclic) { |
638 | wbc.range_start = 0; | |
639 | wbc.range_end = LLONG_MAX; | |
640 | } | |
38f21977 | 641 | |
03ba3782 JA |
642 | for (;;) { |
643 | /* | |
d3ddec76 | 644 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 645 | */ |
83ba7b07 | 646 | if (work->nr_pages <= 0) |
03ba3782 | 647 | break; |
66f3b8e2 | 648 | |
38f21977 | 649 | /* |
d3ddec76 WF |
650 | * For background writeout, stop when we are below the |
651 | * background dirty threshold | |
38f21977 | 652 | */ |
83ba7b07 | 653 | if (work->for_background && !over_bground_thresh()) |
03ba3782 | 654 | break; |
38f21977 | 655 | |
03ba3782 | 656 | wbc.more_io = 0; |
03ba3782 JA |
657 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; |
658 | wbc.pages_skipped = 0; | |
028c2dd1 DC |
659 | |
660 | trace_wbc_writeback_start(&wbc, wb->bdi); | |
83ba7b07 CH |
661 | if (work->sb) |
662 | __writeback_inodes_sb(work->sb, wb, &wbc); | |
edadfb10 CH |
663 | else |
664 | writeback_inodes_wb(wb, &wbc); | |
028c2dd1 DC |
665 | trace_wbc_writeback_written(&wbc, wb->bdi); |
666 | ||
83ba7b07 | 667 | work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
03ba3782 JA |
668 | wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
669 | ||
670 | /* | |
71fd05a8 | 671 | * If we consumed everything, see if we have more |
03ba3782 | 672 | */ |
71fd05a8 JA |
673 | if (wbc.nr_to_write <= 0) |
674 | continue; | |
675 | /* | |
676 | * Didn't write everything and we don't have more IO, bail | |
677 | */ | |
678 | if (!wbc.more_io) | |
03ba3782 | 679 | break; |
71fd05a8 JA |
680 | /* |
681 | * Did we write something? Try for more | |
682 | */ | |
683 | if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) | |
684 | continue; | |
685 | /* | |
686 | * Nothing written. Wait for some inode to | |
687 | * become available for writeback. Otherwise | |
688 | * we'll just busyloop. | |
689 | */ | |
690 | spin_lock(&inode_lock); | |
691 | if (!list_empty(&wb->b_more_io)) { | |
692 | inode = list_entry(wb->b_more_io.prev, | |
693 | struct inode, i_list); | |
028c2dd1 | 694 | trace_wbc_writeback_wait(&wbc, wb->bdi); |
71fd05a8 | 695 | inode_wait_for_writeback(inode); |
03ba3782 | 696 | } |
71fd05a8 | 697 | spin_unlock(&inode_lock); |
03ba3782 JA |
698 | } |
699 | ||
700 | return wrote; | |
701 | } | |
702 | ||
703 | /* | |
83ba7b07 | 704 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 705 | */ |
83ba7b07 | 706 | static struct wb_writeback_work * |
08852b6d | 707 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 708 | { |
83ba7b07 | 709 | struct wb_writeback_work *work = NULL; |
03ba3782 | 710 | |
83ba7b07 CH |
711 | spin_lock(&bdi->wb_lock); |
712 | if (!list_empty(&bdi->work_list)) { | |
713 | work = list_entry(bdi->work_list.next, | |
714 | struct wb_writeback_work, list); | |
715 | list_del_init(&work->list); | |
03ba3782 | 716 | } |
83ba7b07 CH |
717 | spin_unlock(&bdi->wb_lock); |
718 | return work; | |
03ba3782 JA |
719 | } |
720 | ||
721 | static long wb_check_old_data_flush(struct bdi_writeback *wb) | |
722 | { | |
723 | unsigned long expired; | |
724 | long nr_pages; | |
725 | ||
69b62d01 JA |
726 | /* |
727 | * When set to zero, disable periodic writeback | |
728 | */ | |
729 | if (!dirty_writeback_interval) | |
730 | return 0; | |
731 | ||
03ba3782 JA |
732 | expired = wb->last_old_flush + |
733 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
734 | if (time_before(jiffies, expired)) | |
735 | return 0; | |
736 | ||
737 | wb->last_old_flush = jiffies; | |
738 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
739 | global_page_state(NR_UNSTABLE_NFS) + | |
740 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); | |
741 | ||
c4a77a6c | 742 | if (nr_pages) { |
83ba7b07 | 743 | struct wb_writeback_work work = { |
c4a77a6c JA |
744 | .nr_pages = nr_pages, |
745 | .sync_mode = WB_SYNC_NONE, | |
746 | .for_kupdate = 1, | |
747 | .range_cyclic = 1, | |
748 | }; | |
749 | ||
83ba7b07 | 750 | return wb_writeback(wb, &work); |
c4a77a6c | 751 | } |
03ba3782 JA |
752 | |
753 | return 0; | |
754 | } | |
755 | ||
756 | /* | |
757 | * Retrieve work items and do the writeback they describe | |
758 | */ | |
759 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
760 | { | |
761 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 762 | struct wb_writeback_work *work; |
c4a77a6c | 763 | long wrote = 0; |
03ba3782 | 764 | |
08852b6d | 765 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 JA |
766 | /* |
767 | * Override sync mode, in case we must wait for completion | |
83ba7b07 | 768 | * because this thread is exiting now. |
03ba3782 JA |
769 | */ |
770 | if (force_wait) | |
83ba7b07 | 771 | work->sync_mode = WB_SYNC_ALL; |
03ba3782 | 772 | |
455b2864 DC |
773 | trace_writeback_exec(bdi, work); |
774 | ||
83ba7b07 | 775 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
776 | |
777 | /* | |
83ba7b07 CH |
778 | * Notify the caller of completion if this is a synchronous |
779 | * work item, otherwise just free it. | |
03ba3782 | 780 | */ |
83ba7b07 CH |
781 | if (work->done) |
782 | complete(work->done); | |
783 | else | |
784 | kfree(work); | |
03ba3782 JA |
785 | } |
786 | ||
787 | /* | |
788 | * Check for periodic writeback, kupdated() style | |
789 | */ | |
790 | wrote += wb_check_old_data_flush(wb); | |
791 | ||
792 | return wrote; | |
793 | } | |
794 | ||
795 | /* | |
796 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
797 | * wakes up periodically and does kupdated style flushing. | |
798 | */ | |
08243900 | 799 | int bdi_writeback_thread(void *data) |
03ba3782 | 800 | { |
08243900 CH |
801 | struct bdi_writeback *wb = data; |
802 | struct backing_dev_info *bdi = wb->bdi; | |
03ba3782 JA |
803 | unsigned long last_active = jiffies; |
804 | unsigned long wait_jiffies = -1UL; | |
805 | long pages_written; | |
806 | ||
08243900 CH |
807 | current->flags |= PF_FLUSHER | PF_SWAPWRITE; |
808 | set_freezable(); | |
809 | ||
810 | /* | |
811 | * Our parent may run at a different priority, just set us to normal | |
812 | */ | |
813 | set_user_nice(current, 0); | |
814 | ||
815 | /* | |
816 | * Clear pending bit and wakeup anybody waiting to tear us down | |
817 | */ | |
818 | clear_bit(BDI_pending, &bdi->state); | |
819 | smp_mb__after_clear_bit(); | |
820 | wake_up_bit(&bdi->state, BDI_pending); | |
821 | ||
455b2864 DC |
822 | trace_writeback_thread_start(bdi); |
823 | ||
03ba3782 JA |
824 | while (!kthread_should_stop()) { |
825 | pages_written = wb_do_writeback(wb, 0); | |
826 | ||
455b2864 DC |
827 | trace_writeback_pages_written(pages_written); |
828 | ||
03ba3782 JA |
829 | if (pages_written) |
830 | last_active = jiffies; | |
831 | else if (wait_jiffies != -1UL) { | |
832 | unsigned long max_idle; | |
833 | ||
38f21977 | 834 | /* |
03ba3782 | 835 | * Longest period of inactivity that we tolerate. If we |
6f904ff0 | 836 | * see dirty data again later, the thread will get |
03ba3782 | 837 | * recreated automatically. |
38f21977 | 838 | */ |
03ba3782 JA |
839 | max_idle = max(5UL * 60 * HZ, wait_jiffies); |
840 | if (time_after(jiffies, max_idle + last_active)) | |
841 | break; | |
842 | } | |
843 | ||
297252c8 AB |
844 | set_current_state(TASK_INTERRUPTIBLE); |
845 | if (!list_empty(&bdi->work_list)) { | |
f9eadbbd | 846 | __set_current_state(TASK_RUNNING); |
297252c8 | 847 | continue; |
f9eadbbd | 848 | } |
69b62d01 | 849 | |
297252c8 AB |
850 | if (dirty_writeback_interval) { |
851 | wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); | |
852 | schedule_timeout(wait_jiffies); | |
853 | } else | |
854 | schedule(); | |
855 | ||
03ba3782 JA |
856 | try_to_freeze(); |
857 | } | |
858 | ||
08243900 CH |
859 | wb->task = NULL; |
860 | ||
861 | /* | |
862 | * Flush any work that raced with us exiting. No new work | |
863 | * will be added, since this bdi isn't discoverable anymore. | |
864 | */ | |
865 | if (!list_empty(&bdi->work_list)) | |
866 | wb_do_writeback(wb, 1); | |
455b2864 DC |
867 | |
868 | trace_writeback_thread_stop(bdi); | |
03ba3782 JA |
869 | return 0; |
870 | } | |
871 | ||
08243900 | 872 | |
03ba3782 | 873 | /* |
b8c2f347 CH |
874 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
875 | * the whole world. | |
03ba3782 | 876 | */ |
b8c2f347 | 877 | void wakeup_flusher_threads(long nr_pages) |
03ba3782 | 878 | { |
b8c2f347 | 879 | struct backing_dev_info *bdi; |
03ba3782 | 880 | |
83ba7b07 CH |
881 | if (!nr_pages) { |
882 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
b8c2f347 CH |
883 | global_page_state(NR_UNSTABLE_NFS); |
884 | } | |
03ba3782 | 885 | |
b8c2f347 | 886 | rcu_read_lock(); |
cfc4ba53 | 887 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
888 | if (!bdi_has_dirty_io(bdi)) |
889 | continue; | |
83ba7b07 | 890 | __bdi_start_writeback(bdi, nr_pages, false, false); |
03ba3782 | 891 | } |
cfc4ba53 | 892 | rcu_read_unlock(); |
1da177e4 LT |
893 | } |
894 | ||
03ba3782 JA |
895 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
896 | { | |
897 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
898 | struct dentry *dentry; | |
899 | const char *name = "?"; | |
900 | ||
901 | dentry = d_find_alias(inode); | |
902 | if (dentry) { | |
903 | spin_lock(&dentry->d_lock); | |
904 | name = (const char *) dentry->d_name.name; | |
905 | } | |
906 | printk(KERN_DEBUG | |
907 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
908 | current->comm, task_pid_nr(current), inode->i_ino, | |
909 | name, inode->i_sb->s_id); | |
910 | if (dentry) { | |
911 | spin_unlock(&dentry->d_lock); | |
912 | dput(dentry); | |
913 | } | |
914 | } | |
915 | } | |
916 | ||
917 | /** | |
918 | * __mark_inode_dirty - internal function | |
919 | * @inode: inode to mark | |
920 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
921 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
922 | * mark_inode_dirty_sync. | |
1da177e4 | 923 | * |
03ba3782 JA |
924 | * Put the inode on the super block's dirty list. |
925 | * | |
926 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
927 | * dirty list only if it is hashed or if it refers to a blockdev. | |
928 | * If it was not hashed, it will never be added to the dirty list | |
929 | * even if it is later hashed, as it will have been marked dirty already. | |
930 | * | |
931 | * In short, make sure you hash any inodes _before_ you start marking | |
932 | * them dirty. | |
1da177e4 | 933 | * |
03ba3782 JA |
934 | * This function *must* be atomic for the I_DIRTY_PAGES case - |
935 | * set_page_dirty() is called under spinlock in several places. | |
1da177e4 | 936 | * |
03ba3782 JA |
937 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
938 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
939 | * the kernel-internal blockdev inode represents the dirtying time of the | |
940 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
941 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
942 | * blockdev inode. | |
1da177e4 | 943 | */ |
03ba3782 | 944 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 945 | { |
03ba3782 | 946 | struct super_block *sb = inode->i_sb; |
1da177e4 | 947 | |
03ba3782 JA |
948 | /* |
949 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
950 | * dirty the inode itself | |
951 | */ | |
952 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
953 | if (sb->s_op->dirty_inode) | |
954 | sb->s_op->dirty_inode(inode); | |
955 | } | |
956 | ||
957 | /* | |
958 | * make sure that changes are seen by all cpus before we test i_state | |
959 | * -- mikulas | |
960 | */ | |
961 | smp_mb(); | |
962 | ||
963 | /* avoid the locking if we can */ | |
964 | if ((inode->i_state & flags) == flags) | |
965 | return; | |
966 | ||
967 | if (unlikely(block_dump)) | |
968 | block_dump___mark_inode_dirty(inode); | |
969 | ||
970 | spin_lock(&inode_lock); | |
971 | if ((inode->i_state & flags) != flags) { | |
972 | const int was_dirty = inode->i_state & I_DIRTY; | |
973 | ||
974 | inode->i_state |= flags; | |
975 | ||
976 | /* | |
977 | * If the inode is being synced, just update its dirty state. | |
978 | * The unlocker will place the inode on the appropriate | |
979 | * superblock list, based upon its state. | |
980 | */ | |
981 | if (inode->i_state & I_SYNC) | |
982 | goto out; | |
983 | ||
984 | /* | |
985 | * Only add valid (hashed) inodes to the superblock's | |
986 | * dirty list. Add blockdev inodes as well. | |
987 | */ | |
988 | if (!S_ISBLK(inode->i_mode)) { | |
989 | if (hlist_unhashed(&inode->i_hash)) | |
990 | goto out; | |
991 | } | |
992 | if (inode->i_state & (I_FREEING|I_CLEAR)) | |
993 | goto out; | |
994 | ||
995 | /* | |
996 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
997 | * reposition it (that would break b_dirty time-ordering). | |
998 | */ | |
999 | if (!was_dirty) { | |
1000 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; | |
500b067c JA |
1001 | struct backing_dev_info *bdi = wb->bdi; |
1002 | ||
1003 | if (bdi_cap_writeback_dirty(bdi) && | |
1004 | !test_bit(BDI_registered, &bdi->state)) { | |
1005 | WARN_ON(1); | |
1006 | printk(KERN_ERR "bdi-%s not registered\n", | |
1007 | bdi->name); | |
1008 | } | |
03ba3782 JA |
1009 | |
1010 | inode->dirtied_when = jiffies; | |
1011 | list_move(&inode->i_list, &wb->b_dirty); | |
1da177e4 | 1012 | } |
1da177e4 | 1013 | } |
03ba3782 JA |
1014 | out: |
1015 | spin_unlock(&inode_lock); | |
1016 | } | |
1017 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1018 | ||
1019 | /* | |
1020 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1021 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1022 | * | |
1023 | * If older_than_this is non-NULL, then only write out inodes which | |
1024 | * had their first dirtying at a time earlier than *older_than_this. | |
1025 | * | |
03ba3782 JA |
1026 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1027 | * This function assumes that the blockdev superblock's inodes are backed by | |
1028 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1029 | * assume that all inodes are backed by the same queue. | |
1030 | * | |
1031 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1032 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1033 | * on the writer throttling path, and we get decent balancing between many | |
1034 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1035 | */ | |
b6e51316 | 1036 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1037 | { |
1038 | struct inode *inode, *old_inode = NULL; | |
1039 | ||
1040 | /* | |
1041 | * We need to be protected against the filesystem going from | |
1042 | * r/o to r/w or vice versa. | |
1043 | */ | |
b6e51316 | 1044 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 JA |
1045 | |
1046 | spin_lock(&inode_lock); | |
1047 | ||
1048 | /* | |
1049 | * Data integrity sync. Must wait for all pages under writeback, | |
1050 | * because there may have been pages dirtied before our sync | |
1051 | * call, but which had writeout started before we write it out. | |
1052 | * In which case, the inode may not be on the dirty list, but | |
1053 | * we still have to wait for that writeout. | |
1054 | */ | |
b6e51316 | 1055 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
03ba3782 JA |
1056 | struct address_space *mapping; |
1057 | ||
1058 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) | |
1059 | continue; | |
1060 | mapping = inode->i_mapping; | |
1061 | if (mapping->nrpages == 0) | |
1062 | continue; | |
1063 | __iget(inode); | |
1064 | spin_unlock(&inode_lock); | |
1065 | /* | |
1066 | * We hold a reference to 'inode' so it couldn't have | |
1067 | * been removed from s_inodes list while we dropped the | |
1068 | * inode_lock. We cannot iput the inode now as we can | |
1069 | * be holding the last reference and we cannot iput it | |
1070 | * under inode_lock. So we keep the reference and iput | |
1071 | * it later. | |
1072 | */ | |
1073 | iput(old_inode); | |
1074 | old_inode = inode; | |
1075 | ||
1076 | filemap_fdatawait(mapping); | |
1077 | ||
1078 | cond_resched(); | |
1079 | ||
1080 | spin_lock(&inode_lock); | |
1081 | } | |
1082 | spin_unlock(&inode_lock); | |
1083 | iput(old_inode); | |
1da177e4 LT |
1084 | } |
1085 | ||
d8a8559c JA |
1086 | /** |
1087 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1088 | * @sb: the superblock | |
1da177e4 | 1089 | * |
d8a8559c JA |
1090 | * Start writeback on some inodes on this super_block. No guarantees are made |
1091 | * on how many (if any) will be written, and this function does not wait | |
1092 | * for IO completion of submitted IO. The number of pages submitted is | |
1093 | * returned. | |
1da177e4 | 1094 | */ |
b6e51316 | 1095 | void writeback_inodes_sb(struct super_block *sb) |
1da177e4 | 1096 | { |
0e3c9a22 JA |
1097 | unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); |
1098 | unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); | |
83ba7b07 CH |
1099 | DECLARE_COMPLETION_ONSTACK(done); |
1100 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1101 | .sb = sb, |
1102 | .sync_mode = WB_SYNC_NONE, | |
83ba7b07 | 1103 | .done = &done, |
3c4d7165 | 1104 | }; |
d8a8559c | 1105 | |
cf37e972 CH |
1106 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1107 | ||
83ba7b07 | 1108 | work.nr_pages = nr_dirty + nr_unstable + |
0e3c9a22 JA |
1109 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); |
1110 | ||
83ba7b07 CH |
1111 | bdi_queue_work(sb->s_bdi, &work); |
1112 | wait_for_completion(&done); | |
e913fc82 | 1113 | } |
0e3c9a22 | 1114 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1115 | |
17bd55d0 ES |
1116 | /** |
1117 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1118 | * @sb: the superblock | |
1119 | * | |
1120 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1121 | * Returns 1 if writeback was started, 0 if not. | |
1122 | */ | |
1123 | int writeback_inodes_sb_if_idle(struct super_block *sb) | |
1124 | { | |
1125 | if (!writeback_in_progress(sb->s_bdi)) { | |
cf37e972 | 1126 | down_read(&sb->s_umount); |
17bd55d0 | 1127 | writeback_inodes_sb(sb); |
cf37e972 | 1128 | up_read(&sb->s_umount); |
17bd55d0 ES |
1129 | return 1; |
1130 | } else | |
1131 | return 0; | |
1132 | } | |
1133 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1134 | ||
d8a8559c JA |
1135 | /** |
1136 | * sync_inodes_sb - sync sb inode pages | |
1137 | * @sb: the superblock | |
1138 | * | |
1139 | * This function writes and waits on any dirty inode belonging to this | |
1140 | * super_block. The number of pages synced is returned. | |
1141 | */ | |
b6e51316 | 1142 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1143 | { |
83ba7b07 CH |
1144 | DECLARE_COMPLETION_ONSTACK(done); |
1145 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1146 | .sb = sb, |
1147 | .sync_mode = WB_SYNC_ALL, | |
1148 | .nr_pages = LONG_MAX, | |
1149 | .range_cyclic = 0, | |
83ba7b07 | 1150 | .done = &done, |
3c4d7165 CH |
1151 | }; |
1152 | ||
cf37e972 CH |
1153 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1154 | ||
83ba7b07 CH |
1155 | bdi_queue_work(sb->s_bdi, &work); |
1156 | wait_for_completion(&done); | |
1157 | ||
b6e51316 | 1158 | wait_sb_inodes(sb); |
1da177e4 | 1159 | } |
d8a8559c | 1160 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1161 | |
1da177e4 | 1162 | /** |
7f04c26d AA |
1163 | * write_inode_now - write an inode to disk |
1164 | * @inode: inode to write to disk | |
1165 | * @sync: whether the write should be synchronous or not | |
1166 | * | |
1167 | * This function commits an inode to disk immediately if it is dirty. This is | |
1168 | * primarily needed by knfsd. | |
1da177e4 | 1169 | * |
7f04c26d | 1170 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1171 | */ |
1da177e4 LT |
1172 | int write_inode_now(struct inode *inode, int sync) |
1173 | { | |
1174 | int ret; | |
1175 | struct writeback_control wbc = { | |
1176 | .nr_to_write = LONG_MAX, | |
18914b18 | 1177 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1178 | .range_start = 0, |
1179 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1180 | }; |
1181 | ||
1182 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1183 | wbc.nr_to_write = 0; |
1da177e4 LT |
1184 | |
1185 | might_sleep(); | |
1186 | spin_lock(&inode_lock); | |
01c03194 | 1187 | ret = writeback_single_inode(inode, &wbc); |
1da177e4 LT |
1188 | spin_unlock(&inode_lock); |
1189 | if (sync) | |
1c0eeaf5 | 1190 | inode_sync_wait(inode); |
1da177e4 LT |
1191 | return ret; |
1192 | } | |
1193 | EXPORT_SYMBOL(write_inode_now); | |
1194 | ||
1195 | /** | |
1196 | * sync_inode - write an inode and its pages to disk. | |
1197 | * @inode: the inode to sync | |
1198 | * @wbc: controls the writeback mode | |
1199 | * | |
1200 | * sync_inode() will write an inode and its pages to disk. It will also | |
1201 | * correctly update the inode on its superblock's dirty inode lists and will | |
1202 | * update inode->i_state. | |
1203 | * | |
1204 | * The caller must have a ref on the inode. | |
1205 | */ | |
1206 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1207 | { | |
1208 | int ret; | |
1209 | ||
1210 | spin_lock(&inode_lock); | |
01c03194 | 1211 | ret = writeback_single_inode(inode, wbc); |
1da177e4 LT |
1212 | spin_unlock(&inode_lock); |
1213 | return ret; | |
1214 | } | |
1215 | EXPORT_SYMBOL(sync_inode); |