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 LT |
18 | #include <linux/spinlock.h> |
19 | #include <linux/sched.h> | |
20 | #include <linux/fs.h> | |
21 | #include <linux/mm.h> | |
03ba3782 JA |
22 | #include <linux/kthread.h> |
23 | #include <linux/freezer.h> | |
1da177e4 LT |
24 | #include <linux/writeback.h> |
25 | #include <linux/blkdev.h> | |
26 | #include <linux/backing-dev.h> | |
27 | #include <linux/buffer_head.h> | |
07f3f05c | 28 | #include "internal.h" |
1da177e4 | 29 | |
66f3b8e2 | 30 | #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) |
f11b00f3 | 31 | |
d0bceac7 JA |
32 | /* |
33 | * We don't actually have pdflush, but this one is exported though /proc... | |
34 | */ | |
35 | int nr_pdflush_threads; | |
36 | ||
c4a77a6c JA |
37 | /* |
38 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
39 | */ | |
40 | struct wb_writeback_args { | |
41 | long nr_pages; | |
42 | struct super_block *sb; | |
43 | enum writeback_sync_modes sync_mode; | |
d3ddec76 WF |
44 | int for_kupdate:1; |
45 | int range_cyclic:1; | |
46 | int for_background:1; | |
c4a77a6c JA |
47 | }; |
48 | ||
03ba3782 JA |
49 | /* |
50 | * Work items for the bdi_writeback threads | |
f11b00f3 | 51 | */ |
03ba3782 | 52 | struct bdi_work { |
8010c3b6 JA |
53 | struct list_head list; /* pending work list */ |
54 | struct rcu_head rcu_head; /* for RCU free/clear of work */ | |
03ba3782 | 55 | |
8010c3b6 JA |
56 | unsigned long seen; /* threads that have seen this work */ |
57 | atomic_t pending; /* number of threads still to do work */ | |
03ba3782 | 58 | |
8010c3b6 | 59 | struct wb_writeback_args args; /* writeback arguments */ |
03ba3782 | 60 | |
8010c3b6 | 61 | unsigned long state; /* flag bits, see WS_* */ |
03ba3782 JA |
62 | }; |
63 | ||
64 | enum { | |
65 | WS_USED_B = 0, | |
66 | WS_ONSTACK_B, | |
67 | }; | |
68 | ||
69 | #define WS_USED (1 << WS_USED_B) | |
70 | #define WS_ONSTACK (1 << WS_ONSTACK_B) | |
71 | ||
72 | static inline bool bdi_work_on_stack(struct bdi_work *work) | |
73 | { | |
74 | return test_bit(WS_ONSTACK_B, &work->state); | |
75 | } | |
76 | ||
77 | static inline void bdi_work_init(struct bdi_work *work, | |
b6e51316 | 78 | struct wb_writeback_args *args) |
03ba3782 JA |
79 | { |
80 | INIT_RCU_HEAD(&work->rcu_head); | |
b6e51316 | 81 | work->args = *args; |
03ba3782 JA |
82 | work->state = WS_USED; |
83 | } | |
84 | ||
f11b00f3 AB |
85 | /** |
86 | * writeback_in_progress - determine whether there is writeback in progress | |
87 | * @bdi: the device's backing_dev_info structure. | |
88 | * | |
03ba3782 JA |
89 | * Determine whether there is writeback waiting to be handled against a |
90 | * backing device. | |
f11b00f3 AB |
91 | */ |
92 | int writeback_in_progress(struct backing_dev_info *bdi) | |
93 | { | |
03ba3782 | 94 | return !list_empty(&bdi->work_list); |
f11b00f3 AB |
95 | } |
96 | ||
03ba3782 | 97 | static void bdi_work_clear(struct bdi_work *work) |
f11b00f3 | 98 | { |
03ba3782 JA |
99 | clear_bit(WS_USED_B, &work->state); |
100 | smp_mb__after_clear_bit(); | |
1ef7d9aa NP |
101 | /* |
102 | * work can have disappeared at this point. bit waitq functions | |
103 | * should be able to tolerate this, provided bdi_sched_wait does | |
104 | * not dereference it's pointer argument. | |
105 | */ | |
03ba3782 | 106 | wake_up_bit(&work->state, WS_USED_B); |
f11b00f3 AB |
107 | } |
108 | ||
03ba3782 | 109 | static void bdi_work_free(struct rcu_head *head) |
4195f73d | 110 | { |
03ba3782 | 111 | struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); |
4195f73d | 112 | |
03ba3782 JA |
113 | if (!bdi_work_on_stack(work)) |
114 | kfree(work); | |
115 | else | |
116 | bdi_work_clear(work); | |
4195f73d NP |
117 | } |
118 | ||
03ba3782 | 119 | static void wb_work_complete(struct bdi_work *work) |
1da177e4 | 120 | { |
c4a77a6c | 121 | const enum writeback_sync_modes sync_mode = work->args.sync_mode; |
77b9d059 | 122 | int onstack = bdi_work_on_stack(work); |
1da177e4 LT |
123 | |
124 | /* | |
03ba3782 JA |
125 | * For allocated work, we can clear the done/seen bit right here. |
126 | * For on-stack work, we need to postpone both the clear and free | |
127 | * to after the RCU grace period, since the stack could be invalidated | |
128 | * as soon as bdi_work_clear() has done the wakeup. | |
1da177e4 | 129 | */ |
77b9d059 | 130 | if (!onstack) |
03ba3782 | 131 | bdi_work_clear(work); |
77b9d059 | 132 | if (sync_mode == WB_SYNC_NONE || onstack) |
03ba3782 JA |
133 | call_rcu(&work->rcu_head, bdi_work_free); |
134 | } | |
1da177e4 | 135 | |
03ba3782 JA |
136 | static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) |
137 | { | |
1da177e4 | 138 | /* |
03ba3782 JA |
139 | * The caller has retrieved the work arguments from this work, |
140 | * drop our reference. If this is the last ref, delete and free it | |
1da177e4 | 141 | */ |
03ba3782 JA |
142 | if (atomic_dec_and_test(&work->pending)) { |
143 | struct backing_dev_info *bdi = wb->bdi; | |
1da177e4 | 144 | |
03ba3782 JA |
145 | spin_lock(&bdi->wb_lock); |
146 | list_del_rcu(&work->list); | |
147 | spin_unlock(&bdi->wb_lock); | |
1da177e4 | 148 | |
03ba3782 JA |
149 | wb_work_complete(work); |
150 | } | |
151 | } | |
1da177e4 | 152 | |
03ba3782 JA |
153 | static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) |
154 | { | |
bcddc3f0 JA |
155 | work->seen = bdi->wb_mask; |
156 | BUG_ON(!work->seen); | |
157 | atomic_set(&work->pending, bdi->wb_cnt); | |
158 | BUG_ON(!bdi->wb_cnt); | |
1da177e4 | 159 | |
bcddc3f0 | 160 | /* |
deed62ed NP |
161 | * list_add_tail_rcu() contains the necessary barriers to |
162 | * make sure the above stores are seen before the item is | |
163 | * noticed on the list | |
bcddc3f0 | 164 | */ |
bcddc3f0 JA |
165 | spin_lock(&bdi->wb_lock); |
166 | list_add_tail_rcu(&work->list, &bdi->work_list); | |
167 | spin_unlock(&bdi->wb_lock); | |
03ba3782 JA |
168 | |
169 | /* | |
170 | * If the default thread isn't there, make sure we add it. When | |
171 | * it gets created and wakes up, we'll run this work. | |
172 | */ | |
173 | if (unlikely(list_empty_careful(&bdi->wb_list))) | |
174 | wake_up_process(default_backing_dev_info.wb.task); | |
175 | else { | |
176 | struct bdi_writeback *wb = &bdi->wb; | |
1da177e4 | 177 | |
1ef7d9aa | 178 | if (wb->task) |
03ba3782 | 179 | wake_up_process(wb->task); |
1da177e4 | 180 | } |
1da177e4 LT |
181 | } |
182 | ||
03ba3782 JA |
183 | /* |
184 | * Used for on-stack allocated work items. The caller needs to wait until | |
185 | * the wb threads have acked the work before it's safe to continue. | |
186 | */ | |
187 | static void bdi_wait_on_work_clear(struct bdi_work *work) | |
188 | { | |
189 | wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, | |
190 | TASK_UNINTERRUPTIBLE); | |
191 | } | |
1da177e4 | 192 | |
f11fcae8 | 193 | static void bdi_alloc_queue_work(struct backing_dev_info *bdi, |
b6e51316 | 194 | struct wb_writeback_args *args) |
1da177e4 | 195 | { |
03ba3782 JA |
196 | struct bdi_work *work; |
197 | ||
bcddc3f0 JA |
198 | /* |
199 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
200 | * wakeup the thread for old dirty data writeback | |
201 | */ | |
03ba3782 | 202 | work = kmalloc(sizeof(*work), GFP_ATOMIC); |
bcddc3f0 | 203 | if (work) { |
b6e51316 | 204 | bdi_work_init(work, args); |
bcddc3f0 JA |
205 | bdi_queue_work(bdi, work); |
206 | } else { | |
207 | struct bdi_writeback *wb = &bdi->wb; | |
03ba3782 | 208 | |
bcddc3f0 JA |
209 | if (wb->task) |
210 | wake_up_process(wb->task); | |
211 | } | |
03ba3782 JA |
212 | } |
213 | ||
b6e51316 JA |
214 | /** |
215 | * bdi_sync_writeback - start and wait for writeback | |
216 | * @bdi: the backing device to write from | |
217 | * @sb: write inodes from this super_block | |
218 | * | |
219 | * Description: | |
220 | * This does WB_SYNC_ALL data integrity writeback and waits for the | |
221 | * IO to complete. Callers must hold the sb s_umount semaphore for | |
222 | * reading, to avoid having the super disappear before we are done. | |
223 | */ | |
224 | static void bdi_sync_writeback(struct backing_dev_info *bdi, | |
225 | struct super_block *sb) | |
03ba3782 | 226 | { |
b6e51316 JA |
227 | struct wb_writeback_args args = { |
228 | .sb = sb, | |
229 | .sync_mode = WB_SYNC_ALL, | |
230 | .nr_pages = LONG_MAX, | |
231 | .range_cyclic = 0, | |
232 | }; | |
233 | struct bdi_work work; | |
03ba3782 | 234 | |
b6e51316 JA |
235 | bdi_work_init(&work, &args); |
236 | work.state |= WS_ONSTACK; | |
03ba3782 | 237 | |
b6e51316 JA |
238 | bdi_queue_work(bdi, &work); |
239 | bdi_wait_on_work_clear(&work); | |
240 | } | |
241 | ||
242 | /** | |
243 | * bdi_start_writeback - start writeback | |
244 | * @bdi: the backing device to write from | |
4b6764fa | 245 | * @sb: write inodes from this super_block |
b6e51316 JA |
246 | * @nr_pages: the number of pages to write |
247 | * | |
248 | * Description: | |
249 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
250 | * started when this function returns, we make no guarentees on | |
251 | * completion. Caller need not hold sb s_umount semaphore. | |
252 | * | |
253 | */ | |
a72bfd4d JA |
254 | void bdi_start_writeback(struct backing_dev_info *bdi, struct super_block *sb, |
255 | long nr_pages) | |
b6e51316 JA |
256 | { |
257 | struct wb_writeback_args args = { | |
a72bfd4d | 258 | .sb = sb, |
b6e51316 JA |
259 | .sync_mode = WB_SYNC_NONE, |
260 | .nr_pages = nr_pages, | |
261 | .range_cyclic = 1, | |
262 | }; | |
263 | ||
d3ddec76 WF |
264 | /* |
265 | * We treat @nr_pages=0 as the special case to do background writeback, | |
266 | * ie. to sync pages until the background dirty threshold is reached. | |
267 | */ | |
268 | if (!nr_pages) { | |
269 | args.nr_pages = LONG_MAX; | |
270 | args.for_background = 1; | |
271 | } | |
272 | ||
b6e51316 | 273 | bdi_alloc_queue_work(bdi, &args); |
1da177e4 LT |
274 | } |
275 | ||
6610a0bc AM |
276 | /* |
277 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
278 | * furthest end of its superblock's dirty-inode list. | |
279 | * | |
280 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 281 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
282 | * the case then the inode must have been redirtied while it was being written |
283 | * out and we don't reset its dirtied_when. | |
284 | */ | |
285 | static void redirty_tail(struct inode *inode) | |
286 | { | |
03ba3782 | 287 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
6610a0bc | 288 | |
03ba3782 | 289 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 290 | struct inode *tail; |
6610a0bc | 291 | |
03ba3782 | 292 | tail = list_entry(wb->b_dirty.next, struct inode, i_list); |
66f3b8e2 | 293 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
294 | inode->dirtied_when = jiffies; |
295 | } | |
03ba3782 | 296 | list_move(&inode->i_list, &wb->b_dirty); |
6610a0bc AM |
297 | } |
298 | ||
c986d1e2 | 299 | /* |
66f3b8e2 | 300 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 301 | */ |
0e0f4fc2 | 302 | static void requeue_io(struct inode *inode) |
c986d1e2 | 303 | { |
03ba3782 JA |
304 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
305 | ||
306 | list_move(&inode->i_list, &wb->b_more_io); | |
c986d1e2 AM |
307 | } |
308 | ||
1c0eeaf5 JE |
309 | static void inode_sync_complete(struct inode *inode) |
310 | { | |
311 | /* | |
312 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
313 | */ | |
314 | smp_mb(); | |
315 | wake_up_bit(&inode->i_state, __I_SYNC); | |
316 | } | |
317 | ||
d2caa3c5 JL |
318 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
319 | { | |
320 | bool ret = time_after(inode->dirtied_when, t); | |
321 | #ifndef CONFIG_64BIT | |
322 | /* | |
323 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
324 | * It _appears_ to be in the future, but is actually in distant past. | |
325 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 326 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
327 | */ |
328 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
329 | #endif | |
330 | return ret; | |
331 | } | |
332 | ||
2c136579 FW |
333 | /* |
334 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
335 | */ | |
336 | static void move_expired_inodes(struct list_head *delaying_queue, | |
337 | struct list_head *dispatch_queue, | |
338 | unsigned long *older_than_this) | |
339 | { | |
5c03449d SL |
340 | LIST_HEAD(tmp); |
341 | struct list_head *pos, *node; | |
cf137307 | 342 | struct super_block *sb = NULL; |
5c03449d | 343 | struct inode *inode; |
cf137307 | 344 | int do_sb_sort = 0; |
5c03449d | 345 | |
2c136579 | 346 | while (!list_empty(delaying_queue)) { |
5c03449d | 347 | inode = list_entry(delaying_queue->prev, struct inode, i_list); |
2c136579 | 348 | if (older_than_this && |
d2caa3c5 | 349 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 | 350 | break; |
cf137307 JA |
351 | if (sb && sb != inode->i_sb) |
352 | do_sb_sort = 1; | |
353 | sb = inode->i_sb; | |
5c03449d SL |
354 | list_move(&inode->i_list, &tmp); |
355 | } | |
356 | ||
cf137307 JA |
357 | /* just one sb in list, splice to dispatch_queue and we're done */ |
358 | if (!do_sb_sort) { | |
359 | list_splice(&tmp, dispatch_queue); | |
360 | return; | |
361 | } | |
362 | ||
5c03449d SL |
363 | /* Move inodes from one superblock together */ |
364 | while (!list_empty(&tmp)) { | |
365 | inode = list_entry(tmp.prev, struct inode, i_list); | |
366 | sb = inode->i_sb; | |
367 | list_for_each_prev_safe(pos, node, &tmp) { | |
368 | inode = list_entry(pos, struct inode, i_list); | |
369 | if (inode->i_sb == sb) | |
370 | list_move(&inode->i_list, dispatch_queue); | |
371 | } | |
2c136579 FW |
372 | } |
373 | } | |
374 | ||
375 | /* | |
376 | * Queue all expired dirty inodes for io, eldest first. | |
377 | */ | |
03ba3782 | 378 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 379 | { |
03ba3782 JA |
380 | list_splice_init(&wb->b_more_io, wb->b_io.prev); |
381 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); | |
66f3b8e2 JA |
382 | } |
383 | ||
a9185b41 | 384 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 385 | { |
03ba3782 | 386 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 387 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 388 | return 0; |
08d8e974 | 389 | } |
08d8e974 | 390 | |
1da177e4 | 391 | /* |
01c03194 CH |
392 | * Wait for writeback on an inode to complete. |
393 | */ | |
394 | static void inode_wait_for_writeback(struct inode *inode) | |
395 | { | |
396 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
397 | wait_queue_head_t *wqh; | |
398 | ||
399 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
400 | do { | |
401 | spin_unlock(&inode_lock); | |
402 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); | |
403 | spin_lock(&inode_lock); | |
404 | } while (inode->i_state & I_SYNC); | |
405 | } | |
406 | ||
407 | /* | |
408 | * Write out an inode's dirty pages. Called under inode_lock. Either the | |
409 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
410 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
411 | * | |
1da177e4 LT |
412 | * If `wait' is set, wait on the writeout. |
413 | * | |
414 | * The whole writeout design is quite complex and fragile. We want to avoid | |
415 | * starvation of particular inodes when others are being redirtied, prevent | |
416 | * livelocks, etc. | |
417 | * | |
418 | * Called under inode_lock. | |
419 | */ | |
420 | static int | |
01c03194 | 421 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 422 | { |
1da177e4 | 423 | struct address_space *mapping = inode->i_mapping; |
01c03194 | 424 | unsigned dirty; |
1da177e4 LT |
425 | int ret; |
426 | ||
01c03194 CH |
427 | if (!atomic_read(&inode->i_count)) |
428 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
429 | else | |
430 | WARN_ON(inode->i_state & I_WILL_FREE); | |
431 | ||
432 | if (inode->i_state & I_SYNC) { | |
433 | /* | |
434 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 435 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
436 | * writeback can proceed with the other inodes on s_io. |
437 | * | |
438 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 439 | * completed a full scan of b_io. |
01c03194 | 440 | */ |
a9185b41 | 441 | if (wbc->sync_mode != WB_SYNC_ALL) { |
01c03194 CH |
442 | requeue_io(inode); |
443 | return 0; | |
444 | } | |
445 | ||
446 | /* | |
447 | * It's a data-integrity sync. We must wait. | |
448 | */ | |
449 | inode_wait_for_writeback(inode); | |
450 | } | |
451 | ||
1c0eeaf5 | 452 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 453 | |
1c0eeaf5 | 454 | /* Set I_SYNC, reset I_DIRTY */ |
1da177e4 | 455 | dirty = inode->i_state & I_DIRTY; |
1c0eeaf5 | 456 | inode->i_state |= I_SYNC; |
1da177e4 LT |
457 | inode->i_state &= ~I_DIRTY; |
458 | ||
459 | spin_unlock(&inode_lock); | |
460 | ||
461 | ret = do_writepages(mapping, wbc); | |
462 | ||
26821ed4 CH |
463 | /* |
464 | * Make sure to wait on the data before writing out the metadata. | |
465 | * This is important for filesystems that modify metadata on data | |
466 | * I/O completion. | |
467 | */ | |
a9185b41 | 468 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 469 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
470 | if (ret == 0) |
471 | ret = err; | |
472 | } | |
473 | ||
26821ed4 CH |
474 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
475 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 476 | int err = write_inode(inode, wbc); |
1da177e4 LT |
477 | if (ret == 0) |
478 | ret = err; | |
479 | } | |
480 | ||
481 | spin_lock(&inode_lock); | |
1c0eeaf5 | 482 | inode->i_state &= ~I_SYNC; |
84a89245 | 483 | if (!(inode->i_state & (I_FREEING | I_CLEAR))) { |
b3af9468 | 484 | if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) { |
ae1b7f7d | 485 | /* |
b3af9468 WF |
486 | * More pages get dirtied by a fast dirtier. |
487 | */ | |
488 | goto select_queue; | |
489 | } else if (inode->i_state & I_DIRTY) { | |
490 | /* | |
491 | * At least XFS will redirty the inode during the | |
492 | * writeback (delalloc) and on io completion (isize). | |
ae1b7f7d WF |
493 | */ |
494 | redirty_tail(inode); | |
495 | } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
1da177e4 LT |
496 | /* |
497 | * We didn't write back all the pages. nfs_writepages() | |
498 | * sometimes bales out without doing anything. Redirty | |
66f3b8e2 | 499 | * the inode; Move it from b_io onto b_more_io/b_dirty. |
1b43ef91 AM |
500 | */ |
501 | /* | |
502 | * akpm: if the caller was the kupdate function we put | |
66f3b8e2 | 503 | * this inode at the head of b_dirty so it gets first |
1b43ef91 AM |
504 | * consideration. Otherwise, move it to the tail, for |
505 | * the reasons described there. I'm not really sure | |
506 | * how much sense this makes. Presumably I had a good | |
507 | * reasons for doing it this way, and I'd rather not | |
508 | * muck with it at present. | |
1da177e4 LT |
509 | */ |
510 | if (wbc->for_kupdate) { | |
511 | /* | |
2c136579 | 512 | * For the kupdate function we move the inode |
66f3b8e2 | 513 | * to b_more_io so it will get more writeout as |
2c136579 | 514 | * soon as the queue becomes uncongested. |
1da177e4 LT |
515 | */ |
516 | inode->i_state |= I_DIRTY_PAGES; | |
b3af9468 | 517 | select_queue: |
8bc3be27 FW |
518 | if (wbc->nr_to_write <= 0) { |
519 | /* | |
520 | * slice used up: queue for next turn | |
521 | */ | |
522 | requeue_io(inode); | |
523 | } else { | |
524 | /* | |
525 | * somehow blocked: retry later | |
526 | */ | |
527 | redirty_tail(inode); | |
528 | } | |
1da177e4 LT |
529 | } else { |
530 | /* | |
531 | * Otherwise fully redirty the inode so that | |
532 | * other inodes on this superblock will get some | |
533 | * writeout. Otherwise heavy writing to one | |
534 | * file would indefinitely suspend writeout of | |
535 | * all the other files. | |
536 | */ | |
537 | inode->i_state |= I_DIRTY_PAGES; | |
1b43ef91 | 538 | redirty_tail(inode); |
1da177e4 | 539 | } |
1da177e4 LT |
540 | } else if (atomic_read(&inode->i_count)) { |
541 | /* | |
542 | * The inode is clean, inuse | |
543 | */ | |
544 | list_move(&inode->i_list, &inode_in_use); | |
545 | } else { | |
546 | /* | |
547 | * The inode is clean, unused | |
548 | */ | |
549 | list_move(&inode->i_list, &inode_unused); | |
1da177e4 LT |
550 | } |
551 | } | |
1c0eeaf5 | 552 | inode_sync_complete(inode); |
1da177e4 LT |
553 | return ret; |
554 | } | |
555 | ||
9ecc2738 JA |
556 | static void unpin_sb_for_writeback(struct super_block **psb) |
557 | { | |
558 | struct super_block *sb = *psb; | |
559 | ||
560 | if (sb) { | |
561 | up_read(&sb->s_umount); | |
562 | put_super(sb); | |
563 | *psb = NULL; | |
564 | } | |
565 | } | |
566 | ||
03ba3782 JA |
567 | /* |
568 | * For WB_SYNC_NONE writeback, the caller does not have the sb pinned | |
569 | * before calling writeback. So make sure that we do pin it, so it doesn't | |
570 | * go away while we are writing inodes from it. | |
571 | * | |
572 | * Returns 0 if the super was successfully pinned (or pinning wasn't needed), | |
573 | * 1 if we failed. | |
574 | */ | |
575 | static int pin_sb_for_writeback(struct writeback_control *wbc, | |
9ecc2738 | 576 | struct inode *inode, struct super_block **psb) |
03ba3782 JA |
577 | { |
578 | struct super_block *sb = inode->i_sb; | |
579 | ||
9ecc2738 JA |
580 | /* |
581 | * If this sb is already pinned, nothing more to do. If not and | |
582 | * *psb is non-NULL, unpin the old one first | |
583 | */ | |
584 | if (sb == *psb) | |
585 | return 0; | |
586 | else if (*psb) | |
587 | unpin_sb_for_writeback(psb); | |
588 | ||
03ba3782 JA |
589 | /* |
590 | * Caller must already hold the ref for this | |
591 | */ | |
592 | if (wbc->sync_mode == WB_SYNC_ALL) { | |
593 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
594 | return 0; | |
595 | } | |
596 | ||
597 | spin_lock(&sb_lock); | |
598 | sb->s_count++; | |
599 | if (down_read_trylock(&sb->s_umount)) { | |
600 | if (sb->s_root) { | |
601 | spin_unlock(&sb_lock); | |
9ecc2738 | 602 | goto pinned; |
03ba3782 JA |
603 | } |
604 | /* | |
605 | * umounted, drop rwsem again and fall through to failure | |
606 | */ | |
607 | up_read(&sb->s_umount); | |
608 | } | |
609 | ||
610 | sb->s_count--; | |
611 | spin_unlock(&sb_lock); | |
612 | return 1; | |
9ecc2738 JA |
613 | pinned: |
614 | *psb = sb; | |
615 | return 0; | |
03ba3782 JA |
616 | } |
617 | ||
618 | static void writeback_inodes_wb(struct bdi_writeback *wb, | |
619 | struct writeback_control *wbc) | |
1da177e4 | 620 | { |
9ecc2738 | 621 | struct super_block *sb = wbc->sb, *pin_sb = NULL; |
1da177e4 LT |
622 | const unsigned long start = jiffies; /* livelock avoidance */ |
623 | ||
ae8547b0 | 624 | spin_lock(&inode_lock); |
1da177e4 | 625 | |
03ba3782 JA |
626 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) |
627 | queue_io(wb, wbc->older_than_this); | |
66f3b8e2 | 628 | |
03ba3782 JA |
629 | while (!list_empty(&wb->b_io)) { |
630 | struct inode *inode = list_entry(wb->b_io.prev, | |
1da177e4 | 631 | struct inode, i_list); |
1da177e4 LT |
632 | long pages_skipped; |
633 | ||
66f3b8e2 JA |
634 | /* |
635 | * super block given and doesn't match, skip this inode | |
636 | */ | |
637 | if (sb && sb != inode->i_sb) { | |
638 | redirty_tail(inode); | |
639 | continue; | |
640 | } | |
641 | ||
84a89245 | 642 | if (inode->i_state & (I_NEW | I_WILL_FREE)) { |
7ef0d737 NP |
643 | requeue_io(inode); |
644 | continue; | |
645 | } | |
646 | ||
d2caa3c5 JL |
647 | /* |
648 | * Was this inode dirtied after sync_sb_inodes was called? | |
649 | * This keeps sync from extra jobs and livelock. | |
650 | */ | |
651 | if (inode_dirtied_after(inode, start)) | |
1da177e4 LT |
652 | break; |
653 | ||
9ecc2738 | 654 | if (pin_sb_for_writeback(wbc, inode, &pin_sb)) { |
03ba3782 JA |
655 | requeue_io(inode); |
656 | continue; | |
657 | } | |
1da177e4 | 658 | |
84a89245 | 659 | BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); |
1da177e4 LT |
660 | __iget(inode); |
661 | pages_skipped = wbc->pages_skipped; | |
01c03194 | 662 | writeback_single_inode(inode, wbc); |
1da177e4 LT |
663 | if (wbc->pages_skipped != pages_skipped) { |
664 | /* | |
665 | * writeback is not making progress due to locked | |
666 | * buffers. Skip this inode for now. | |
667 | */ | |
f57b9b7b | 668 | redirty_tail(inode); |
1da177e4 LT |
669 | } |
670 | spin_unlock(&inode_lock); | |
1da177e4 | 671 | iput(inode); |
4ffc8444 | 672 | cond_resched(); |
1da177e4 | 673 | spin_lock(&inode_lock); |
8bc3be27 FW |
674 | if (wbc->nr_to_write <= 0) { |
675 | wbc->more_io = 1; | |
1da177e4 | 676 | break; |
8bc3be27 | 677 | } |
03ba3782 | 678 | if (!list_empty(&wb->b_more_io)) |
8bc3be27 | 679 | wbc->more_io = 1; |
1da177e4 | 680 | } |
38f21977 | 681 | |
9ecc2738 JA |
682 | unpin_sb_for_writeback(&pin_sb); |
683 | ||
66f3b8e2 JA |
684 | spin_unlock(&inode_lock); |
685 | /* Leave any unwritten inodes on b_io */ | |
686 | } | |
687 | ||
03ba3782 JA |
688 | void writeback_inodes_wbc(struct writeback_control *wbc) |
689 | { | |
690 | struct backing_dev_info *bdi = wbc->bdi; | |
691 | ||
692 | writeback_inodes_wb(&bdi->wb, wbc); | |
693 | } | |
694 | ||
66f3b8e2 | 695 | /* |
03ba3782 JA |
696 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
697 | * operation. We do this so we don't hold I_SYNC against an inode for | |
698 | * enormous amounts of time, which would block a userspace task which has | |
699 | * been forced to throttle against that inode. Also, the code reevaluates | |
700 | * the dirty each time it has written this many pages. | |
701 | */ | |
702 | #define MAX_WRITEBACK_PAGES 1024 | |
703 | ||
704 | static inline bool over_bground_thresh(void) | |
705 | { | |
706 | unsigned long background_thresh, dirty_thresh; | |
707 | ||
708 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); | |
709 | ||
710 | return (global_page_state(NR_FILE_DIRTY) + | |
711 | global_page_state(NR_UNSTABLE_NFS) >= background_thresh); | |
712 | } | |
713 | ||
714 | /* | |
715 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 716 | * |
03ba3782 JA |
717 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
718 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
719 | * just walks the superblock inode list, writing back any inodes which are | |
720 | * older than a specific point in time. | |
66f3b8e2 | 721 | * |
03ba3782 JA |
722 | * Try to run once per dirty_writeback_interval. But if a writeback event |
723 | * takes longer than a dirty_writeback_interval interval, then leave a | |
724 | * one-second gap. | |
66f3b8e2 | 725 | * |
03ba3782 JA |
726 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
727 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 728 | */ |
c4a77a6c JA |
729 | static long wb_writeback(struct bdi_writeback *wb, |
730 | struct wb_writeback_args *args) | |
66f3b8e2 | 731 | { |
03ba3782 JA |
732 | struct writeback_control wbc = { |
733 | .bdi = wb->bdi, | |
c4a77a6c JA |
734 | .sb = args->sb, |
735 | .sync_mode = args->sync_mode, | |
03ba3782 | 736 | .older_than_this = NULL, |
c4a77a6c | 737 | .for_kupdate = args->for_kupdate, |
b17621fe | 738 | .for_background = args->for_background, |
c4a77a6c | 739 | .range_cyclic = args->range_cyclic, |
03ba3782 JA |
740 | }; |
741 | unsigned long oldest_jif; | |
742 | long wrote = 0; | |
a5989bdc | 743 | struct inode *inode; |
66f3b8e2 | 744 | |
03ba3782 JA |
745 | if (wbc.for_kupdate) { |
746 | wbc.older_than_this = &oldest_jif; | |
747 | oldest_jif = jiffies - | |
748 | msecs_to_jiffies(dirty_expire_interval * 10); | |
749 | } | |
c4a77a6c JA |
750 | if (!wbc.range_cyclic) { |
751 | wbc.range_start = 0; | |
752 | wbc.range_end = LLONG_MAX; | |
753 | } | |
38f21977 | 754 | |
03ba3782 JA |
755 | for (;;) { |
756 | /* | |
d3ddec76 | 757 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 758 | */ |
d3ddec76 | 759 | if (args->nr_pages <= 0) |
03ba3782 | 760 | break; |
66f3b8e2 | 761 | |
38f21977 | 762 | /* |
d3ddec76 WF |
763 | * For background writeout, stop when we are below the |
764 | * background dirty threshold | |
38f21977 | 765 | */ |
d3ddec76 | 766 | if (args->for_background && !over_bground_thresh()) |
03ba3782 | 767 | break; |
38f21977 | 768 | |
03ba3782 | 769 | wbc.more_io = 0; |
03ba3782 JA |
770 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; |
771 | wbc.pages_skipped = 0; | |
772 | writeback_inodes_wb(wb, &wbc); | |
c4a77a6c | 773 | args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
03ba3782 JA |
774 | wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
775 | ||
776 | /* | |
71fd05a8 | 777 | * If we consumed everything, see if we have more |
03ba3782 | 778 | */ |
71fd05a8 JA |
779 | if (wbc.nr_to_write <= 0) |
780 | continue; | |
781 | /* | |
782 | * Didn't write everything and we don't have more IO, bail | |
783 | */ | |
784 | if (!wbc.more_io) | |
03ba3782 | 785 | break; |
71fd05a8 JA |
786 | /* |
787 | * Did we write something? Try for more | |
788 | */ | |
789 | if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) | |
790 | continue; | |
791 | /* | |
792 | * Nothing written. Wait for some inode to | |
793 | * become available for writeback. Otherwise | |
794 | * we'll just busyloop. | |
795 | */ | |
796 | spin_lock(&inode_lock); | |
797 | if (!list_empty(&wb->b_more_io)) { | |
798 | inode = list_entry(wb->b_more_io.prev, | |
799 | struct inode, i_list); | |
800 | inode_wait_for_writeback(inode); | |
03ba3782 | 801 | } |
71fd05a8 | 802 | spin_unlock(&inode_lock); |
03ba3782 JA |
803 | } |
804 | ||
805 | return wrote; | |
806 | } | |
807 | ||
808 | /* | |
809 | * Return the next bdi_work struct that hasn't been processed by this | |
8010c3b6 JA |
810 | * wb thread yet. ->seen is initially set for each thread that exists |
811 | * for this device, when a thread first notices a piece of work it | |
812 | * clears its bit. Depending on writeback type, the thread will notify | |
813 | * completion on either receiving the work (WB_SYNC_NONE) or after | |
814 | * it is done (WB_SYNC_ALL). | |
03ba3782 JA |
815 | */ |
816 | static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, | |
817 | struct bdi_writeback *wb) | |
818 | { | |
819 | struct bdi_work *work, *ret = NULL; | |
820 | ||
821 | rcu_read_lock(); | |
822 | ||
823 | list_for_each_entry_rcu(work, &bdi->work_list, list) { | |
77fad5e6 | 824 | if (!test_bit(wb->nr, &work->seen)) |
03ba3782 | 825 | continue; |
77fad5e6 | 826 | clear_bit(wb->nr, &work->seen); |
03ba3782 JA |
827 | |
828 | ret = work; | |
829 | break; | |
830 | } | |
831 | ||
832 | rcu_read_unlock(); | |
833 | return ret; | |
834 | } | |
835 | ||
836 | static long wb_check_old_data_flush(struct bdi_writeback *wb) | |
837 | { | |
838 | unsigned long expired; | |
839 | long nr_pages; | |
840 | ||
841 | expired = wb->last_old_flush + | |
842 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
843 | if (time_before(jiffies, expired)) | |
844 | return 0; | |
845 | ||
846 | wb->last_old_flush = jiffies; | |
847 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
848 | global_page_state(NR_UNSTABLE_NFS) + | |
849 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); | |
850 | ||
c4a77a6c JA |
851 | if (nr_pages) { |
852 | struct wb_writeback_args args = { | |
853 | .nr_pages = nr_pages, | |
854 | .sync_mode = WB_SYNC_NONE, | |
855 | .for_kupdate = 1, | |
856 | .range_cyclic = 1, | |
857 | }; | |
858 | ||
859 | return wb_writeback(wb, &args); | |
860 | } | |
03ba3782 JA |
861 | |
862 | return 0; | |
863 | } | |
864 | ||
865 | /* | |
866 | * Retrieve work items and do the writeback they describe | |
867 | */ | |
868 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
869 | { | |
870 | struct backing_dev_info *bdi = wb->bdi; | |
871 | struct bdi_work *work; | |
c4a77a6c | 872 | long wrote = 0; |
03ba3782 JA |
873 | |
874 | while ((work = get_next_work_item(bdi, wb)) != NULL) { | |
c4a77a6c | 875 | struct wb_writeback_args args = work->args; |
03ba3782 JA |
876 | |
877 | /* | |
878 | * Override sync mode, in case we must wait for completion | |
879 | */ | |
880 | if (force_wait) | |
c4a77a6c | 881 | work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; |
03ba3782 JA |
882 | |
883 | /* | |
884 | * If this isn't a data integrity operation, just notify | |
885 | * that we have seen this work and we are now starting it. | |
886 | */ | |
c4a77a6c | 887 | if (args.sync_mode == WB_SYNC_NONE) |
03ba3782 JA |
888 | wb_clear_pending(wb, work); |
889 | ||
c4a77a6c | 890 | wrote += wb_writeback(wb, &args); |
03ba3782 JA |
891 | |
892 | /* | |
893 | * This is a data integrity writeback, so only do the | |
894 | * notification when we have completed the work. | |
895 | */ | |
c4a77a6c | 896 | if (args.sync_mode == WB_SYNC_ALL) |
03ba3782 JA |
897 | wb_clear_pending(wb, work); |
898 | } | |
899 | ||
900 | /* | |
901 | * Check for periodic writeback, kupdated() style | |
902 | */ | |
903 | wrote += wb_check_old_data_flush(wb); | |
904 | ||
905 | return wrote; | |
906 | } | |
907 | ||
908 | /* | |
909 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
910 | * wakes up periodically and does kupdated style flushing. | |
911 | */ | |
912 | int bdi_writeback_task(struct bdi_writeback *wb) | |
913 | { | |
914 | unsigned long last_active = jiffies; | |
915 | unsigned long wait_jiffies = -1UL; | |
916 | long pages_written; | |
917 | ||
918 | while (!kthread_should_stop()) { | |
919 | pages_written = wb_do_writeback(wb, 0); | |
920 | ||
921 | if (pages_written) | |
922 | last_active = jiffies; | |
923 | else if (wait_jiffies != -1UL) { | |
924 | unsigned long max_idle; | |
925 | ||
38f21977 | 926 | /* |
03ba3782 JA |
927 | * Longest period of inactivity that we tolerate. If we |
928 | * see dirty data again later, the task will get | |
929 | * recreated automatically. | |
38f21977 | 930 | */ |
03ba3782 JA |
931 | max_idle = max(5UL * 60 * HZ, wait_jiffies); |
932 | if (time_after(jiffies, max_idle + last_active)) | |
933 | break; | |
934 | } | |
935 | ||
936 | wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); | |
49db0414 | 937 | schedule_timeout_interruptible(wait_jiffies); |
03ba3782 JA |
938 | try_to_freeze(); |
939 | } | |
940 | ||
941 | return 0; | |
942 | } | |
943 | ||
944 | /* | |
b6e51316 JA |
945 | * Schedule writeback for all backing devices. This does WB_SYNC_NONE |
946 | * writeback, for integrity writeback see bdi_sync_writeback(). | |
03ba3782 | 947 | */ |
b6e51316 | 948 | static void bdi_writeback_all(struct super_block *sb, long nr_pages) |
03ba3782 | 949 | { |
b6e51316 JA |
950 | struct wb_writeback_args args = { |
951 | .sb = sb, | |
952 | .nr_pages = nr_pages, | |
953 | .sync_mode = WB_SYNC_NONE, | |
954 | }; | |
03ba3782 | 955 | struct backing_dev_info *bdi; |
03ba3782 | 956 | |
cfc4ba53 | 957 | rcu_read_lock(); |
03ba3782 | 958 | |
cfc4ba53 | 959 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
960 | if (!bdi_has_dirty_io(bdi)) |
961 | continue; | |
38f21977 | 962 | |
b6e51316 | 963 | bdi_alloc_queue_work(bdi, &args); |
03ba3782 JA |
964 | } |
965 | ||
cfc4ba53 | 966 | rcu_read_unlock(); |
1da177e4 LT |
967 | } |
968 | ||
969 | /* | |
03ba3782 JA |
970 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
971 | * the whole world. | |
972 | */ | |
973 | void wakeup_flusher_threads(long nr_pages) | |
974 | { | |
03ba3782 JA |
975 | if (nr_pages == 0) |
976 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
977 | global_page_state(NR_UNSTABLE_NFS); | |
b6e51316 | 978 | bdi_writeback_all(NULL, nr_pages); |
03ba3782 JA |
979 | } |
980 | ||
981 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) | |
982 | { | |
983 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
984 | struct dentry *dentry; | |
985 | const char *name = "?"; | |
986 | ||
987 | dentry = d_find_alias(inode); | |
988 | if (dentry) { | |
989 | spin_lock(&dentry->d_lock); | |
990 | name = (const char *) dentry->d_name.name; | |
991 | } | |
992 | printk(KERN_DEBUG | |
993 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
994 | current->comm, task_pid_nr(current), inode->i_ino, | |
995 | name, inode->i_sb->s_id); | |
996 | if (dentry) { | |
997 | spin_unlock(&dentry->d_lock); | |
998 | dput(dentry); | |
999 | } | |
1000 | } | |
1001 | } | |
1002 | ||
1003 | /** | |
1004 | * __mark_inode_dirty - internal function | |
1005 | * @inode: inode to mark | |
1006 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1007 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1008 | * mark_inode_dirty_sync. | |
1da177e4 | 1009 | * |
03ba3782 JA |
1010 | * Put the inode on the super block's dirty list. |
1011 | * | |
1012 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1013 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1014 | * If it was not hashed, it will never be added to the dirty list | |
1015 | * even if it is later hashed, as it will have been marked dirty already. | |
1016 | * | |
1017 | * In short, make sure you hash any inodes _before_ you start marking | |
1018 | * them dirty. | |
1da177e4 | 1019 | * |
03ba3782 JA |
1020 | * This function *must* be atomic for the I_DIRTY_PAGES case - |
1021 | * set_page_dirty() is called under spinlock in several places. | |
1da177e4 | 1022 | * |
03ba3782 JA |
1023 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1024 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1025 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1026 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1027 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1028 | * blockdev inode. | |
1da177e4 | 1029 | */ |
03ba3782 | 1030 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1031 | { |
03ba3782 | 1032 | struct super_block *sb = inode->i_sb; |
1da177e4 | 1033 | |
03ba3782 JA |
1034 | /* |
1035 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1036 | * dirty the inode itself | |
1037 | */ | |
1038 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1039 | if (sb->s_op->dirty_inode) | |
1040 | sb->s_op->dirty_inode(inode); | |
1041 | } | |
1042 | ||
1043 | /* | |
1044 | * make sure that changes are seen by all cpus before we test i_state | |
1045 | * -- mikulas | |
1046 | */ | |
1047 | smp_mb(); | |
1048 | ||
1049 | /* avoid the locking if we can */ | |
1050 | if ((inode->i_state & flags) == flags) | |
1051 | return; | |
1052 | ||
1053 | if (unlikely(block_dump)) | |
1054 | block_dump___mark_inode_dirty(inode); | |
1055 | ||
1056 | spin_lock(&inode_lock); | |
1057 | if ((inode->i_state & flags) != flags) { | |
1058 | const int was_dirty = inode->i_state & I_DIRTY; | |
1059 | ||
1060 | inode->i_state |= flags; | |
1061 | ||
1062 | /* | |
1063 | * If the inode is being synced, just update its dirty state. | |
1064 | * The unlocker will place the inode on the appropriate | |
1065 | * superblock list, based upon its state. | |
1066 | */ | |
1067 | if (inode->i_state & I_SYNC) | |
1068 | goto out; | |
1069 | ||
1070 | /* | |
1071 | * Only add valid (hashed) inodes to the superblock's | |
1072 | * dirty list. Add blockdev inodes as well. | |
1073 | */ | |
1074 | if (!S_ISBLK(inode->i_mode)) { | |
1075 | if (hlist_unhashed(&inode->i_hash)) | |
1076 | goto out; | |
1077 | } | |
1078 | if (inode->i_state & (I_FREEING|I_CLEAR)) | |
1079 | goto out; | |
1080 | ||
1081 | /* | |
1082 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1083 | * reposition it (that would break b_dirty time-ordering). | |
1084 | */ | |
1085 | if (!was_dirty) { | |
1086 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; | |
500b067c JA |
1087 | struct backing_dev_info *bdi = wb->bdi; |
1088 | ||
1089 | if (bdi_cap_writeback_dirty(bdi) && | |
1090 | !test_bit(BDI_registered, &bdi->state)) { | |
1091 | WARN_ON(1); | |
1092 | printk(KERN_ERR "bdi-%s not registered\n", | |
1093 | bdi->name); | |
1094 | } | |
03ba3782 JA |
1095 | |
1096 | inode->dirtied_when = jiffies; | |
1097 | list_move(&inode->i_list, &wb->b_dirty); | |
1da177e4 | 1098 | } |
1da177e4 | 1099 | } |
03ba3782 JA |
1100 | out: |
1101 | spin_unlock(&inode_lock); | |
1102 | } | |
1103 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1104 | ||
1105 | /* | |
1106 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1107 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1108 | * | |
1109 | * If older_than_this is non-NULL, then only write out inodes which | |
1110 | * had their first dirtying at a time earlier than *older_than_this. | |
1111 | * | |
03ba3782 JA |
1112 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1113 | * This function assumes that the blockdev superblock's inodes are backed by | |
1114 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1115 | * assume that all inodes are backed by the same queue. | |
1116 | * | |
1117 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1118 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1119 | * on the writer throttling path, and we get decent balancing between many | |
1120 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1121 | */ | |
b6e51316 | 1122 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1123 | { |
1124 | struct inode *inode, *old_inode = NULL; | |
1125 | ||
1126 | /* | |
1127 | * We need to be protected against the filesystem going from | |
1128 | * r/o to r/w or vice versa. | |
1129 | */ | |
b6e51316 | 1130 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 JA |
1131 | |
1132 | spin_lock(&inode_lock); | |
1133 | ||
1134 | /* | |
1135 | * Data integrity sync. Must wait for all pages under writeback, | |
1136 | * because there may have been pages dirtied before our sync | |
1137 | * call, but which had writeout started before we write it out. | |
1138 | * In which case, the inode may not be on the dirty list, but | |
1139 | * we still have to wait for that writeout. | |
1140 | */ | |
b6e51316 | 1141 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
03ba3782 JA |
1142 | struct address_space *mapping; |
1143 | ||
1144 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) | |
1145 | continue; | |
1146 | mapping = inode->i_mapping; | |
1147 | if (mapping->nrpages == 0) | |
1148 | continue; | |
1149 | __iget(inode); | |
1150 | spin_unlock(&inode_lock); | |
1151 | /* | |
1152 | * We hold a reference to 'inode' so it couldn't have | |
1153 | * been removed from s_inodes list while we dropped the | |
1154 | * inode_lock. We cannot iput the inode now as we can | |
1155 | * be holding the last reference and we cannot iput it | |
1156 | * under inode_lock. So we keep the reference and iput | |
1157 | * it later. | |
1158 | */ | |
1159 | iput(old_inode); | |
1160 | old_inode = inode; | |
1161 | ||
1162 | filemap_fdatawait(mapping); | |
1163 | ||
1164 | cond_resched(); | |
1165 | ||
1166 | spin_lock(&inode_lock); | |
1167 | } | |
1168 | spin_unlock(&inode_lock); | |
1169 | iput(old_inode); | |
1da177e4 LT |
1170 | } |
1171 | ||
d8a8559c JA |
1172 | /** |
1173 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1174 | * @sb: the superblock | |
1da177e4 | 1175 | * |
d8a8559c JA |
1176 | * Start writeback on some inodes on this super_block. No guarantees are made |
1177 | * on how many (if any) will be written, and this function does not wait | |
1178 | * for IO completion of submitted IO. The number of pages submitted is | |
1179 | * returned. | |
1da177e4 | 1180 | */ |
b6e51316 | 1181 | void writeback_inodes_sb(struct super_block *sb) |
1da177e4 | 1182 | { |
d8a8559c JA |
1183 | unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); |
1184 | unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); | |
1185 | long nr_to_write; | |
1da177e4 | 1186 | |
d8a8559c | 1187 | nr_to_write = nr_dirty + nr_unstable + |
38f21977 | 1188 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); |
38f21977 | 1189 | |
a72bfd4d | 1190 | bdi_start_writeback(sb->s_bdi, sb, nr_to_write); |
d8a8559c JA |
1191 | } |
1192 | EXPORT_SYMBOL(writeback_inodes_sb); | |
1193 | ||
17bd55d0 ES |
1194 | /** |
1195 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1196 | * @sb: the superblock | |
1197 | * | |
1198 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1199 | * Returns 1 if writeback was started, 0 if not. | |
1200 | */ | |
1201 | int writeback_inodes_sb_if_idle(struct super_block *sb) | |
1202 | { | |
1203 | if (!writeback_in_progress(sb->s_bdi)) { | |
1204 | writeback_inodes_sb(sb); | |
1205 | return 1; | |
1206 | } else | |
1207 | return 0; | |
1208 | } | |
1209 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1210 | ||
d8a8559c JA |
1211 | /** |
1212 | * sync_inodes_sb - sync sb inode pages | |
1213 | * @sb: the superblock | |
1214 | * | |
1215 | * This function writes and waits on any dirty inode belonging to this | |
1216 | * super_block. The number of pages synced is returned. | |
1217 | */ | |
b6e51316 | 1218 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1219 | { |
b6e51316 JA |
1220 | bdi_sync_writeback(sb->s_bdi, sb); |
1221 | wait_sb_inodes(sb); | |
1da177e4 | 1222 | } |
d8a8559c | 1223 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1224 | |
1da177e4 | 1225 | /** |
7f04c26d AA |
1226 | * write_inode_now - write an inode to disk |
1227 | * @inode: inode to write to disk | |
1228 | * @sync: whether the write should be synchronous or not | |
1229 | * | |
1230 | * This function commits an inode to disk immediately if it is dirty. This is | |
1231 | * primarily needed by knfsd. | |
1da177e4 | 1232 | * |
7f04c26d | 1233 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1234 | */ |
1da177e4 LT |
1235 | int write_inode_now(struct inode *inode, int sync) |
1236 | { | |
1237 | int ret; | |
1238 | struct writeback_control wbc = { | |
1239 | .nr_to_write = LONG_MAX, | |
18914b18 | 1240 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1241 | .range_start = 0, |
1242 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1243 | }; |
1244 | ||
1245 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1246 | wbc.nr_to_write = 0; |
1da177e4 LT |
1247 | |
1248 | might_sleep(); | |
1249 | spin_lock(&inode_lock); | |
01c03194 | 1250 | ret = writeback_single_inode(inode, &wbc); |
1da177e4 LT |
1251 | spin_unlock(&inode_lock); |
1252 | if (sync) | |
1c0eeaf5 | 1253 | inode_sync_wait(inode); |
1da177e4 LT |
1254 | return ret; |
1255 | } | |
1256 | EXPORT_SYMBOL(write_inode_now); | |
1257 | ||
1258 | /** | |
1259 | * sync_inode - write an inode and its pages to disk. | |
1260 | * @inode: the inode to sync | |
1261 | * @wbc: controls the writeback mode | |
1262 | * | |
1263 | * sync_inode() will write an inode and its pages to disk. It will also | |
1264 | * correctly update the inode on its superblock's dirty inode lists and will | |
1265 | * update inode->i_state. | |
1266 | * | |
1267 | * The caller must have a ref on the inode. | |
1268 | */ | |
1269 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1270 | { | |
1271 | int ret; | |
1272 | ||
1273 | spin_lock(&inode_lock); | |
01c03194 | 1274 | ret = writeback_single_inode(inode, wbc); |
1da177e4 LT |
1275 | spin_unlock(&inode_lock); |
1276 | return ret; | |
1277 | } | |
1278 | EXPORT_SYMBOL(sync_inode); |