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> | |
07f3f05c | 29 | #include "internal.h" |
1da177e4 | 30 | |
66f3b8e2 | 31 | #define inode_to_bdi(inode) ((inode)->i_mapping->backing_dev_info) |
f11b00f3 | 32 | |
d0bceac7 JA |
33 | /* |
34 | * We don't actually have pdflush, but this one is exported though /proc... | |
35 | */ | |
36 | int nr_pdflush_threads; | |
37 | ||
c4a77a6c JA |
38 | /* |
39 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
40 | */ | |
41 | struct wb_writeback_args { | |
42 | long nr_pages; | |
43 | struct super_block *sb; | |
44 | enum writeback_sync_modes sync_mode; | |
52957fe1 HS |
45 | unsigned int for_kupdate:1; |
46 | unsigned int range_cyclic:1; | |
47 | unsigned int for_background:1; | |
e8bebe2f | 48 | unsigned int sb_pinned:1; |
c4a77a6c JA |
49 | }; |
50 | ||
03ba3782 JA |
51 | /* |
52 | * Work items for the bdi_writeback threads | |
f11b00f3 | 53 | */ |
03ba3782 | 54 | struct bdi_work { |
8010c3b6 JA |
55 | struct list_head list; /* pending work list */ |
56 | struct rcu_head rcu_head; /* for RCU free/clear of work */ | |
03ba3782 | 57 | |
8010c3b6 JA |
58 | unsigned long seen; /* threads that have seen this work */ |
59 | atomic_t pending; /* number of threads still to do work */ | |
03ba3782 | 60 | |
8010c3b6 | 61 | struct wb_writeback_args args; /* writeback arguments */ |
03ba3782 | 62 | |
8010c3b6 | 63 | unsigned long state; /* flag bits, see WS_* */ |
03ba3782 JA |
64 | }; |
65 | ||
66 | enum { | |
67 | WS_USED_B = 0, | |
68 | WS_ONSTACK_B, | |
69 | }; | |
70 | ||
71 | #define WS_USED (1 << WS_USED_B) | |
72 | #define WS_ONSTACK (1 << WS_ONSTACK_B) | |
73 | ||
74 | static inline bool bdi_work_on_stack(struct bdi_work *work) | |
75 | { | |
76 | return test_bit(WS_ONSTACK_B, &work->state); | |
77 | } | |
78 | ||
79 | static inline void bdi_work_init(struct bdi_work *work, | |
b6e51316 | 80 | struct wb_writeback_args *args) |
03ba3782 JA |
81 | { |
82 | INIT_RCU_HEAD(&work->rcu_head); | |
b6e51316 | 83 | work->args = *args; |
03ba3782 JA |
84 | work->state = WS_USED; |
85 | } | |
86 | ||
f11b00f3 AB |
87 | /** |
88 | * writeback_in_progress - determine whether there is writeback in progress | |
89 | * @bdi: the device's backing_dev_info structure. | |
90 | * | |
03ba3782 JA |
91 | * Determine whether there is writeback waiting to be handled against a |
92 | * backing device. | |
f11b00f3 AB |
93 | */ |
94 | int writeback_in_progress(struct backing_dev_info *bdi) | |
95 | { | |
03ba3782 | 96 | return !list_empty(&bdi->work_list); |
f11b00f3 AB |
97 | } |
98 | ||
03ba3782 | 99 | static void bdi_work_clear(struct bdi_work *work) |
f11b00f3 | 100 | { |
03ba3782 JA |
101 | clear_bit(WS_USED_B, &work->state); |
102 | smp_mb__after_clear_bit(); | |
1ef7d9aa NP |
103 | /* |
104 | * work can have disappeared at this point. bit waitq functions | |
105 | * should be able to tolerate this, provided bdi_sched_wait does | |
106 | * not dereference it's pointer argument. | |
107 | */ | |
03ba3782 | 108 | wake_up_bit(&work->state, WS_USED_B); |
f11b00f3 AB |
109 | } |
110 | ||
03ba3782 | 111 | static void bdi_work_free(struct rcu_head *head) |
4195f73d | 112 | { |
03ba3782 | 113 | struct bdi_work *work = container_of(head, struct bdi_work, rcu_head); |
4195f73d | 114 | |
03ba3782 JA |
115 | if (!bdi_work_on_stack(work)) |
116 | kfree(work); | |
117 | else | |
118 | bdi_work_clear(work); | |
4195f73d NP |
119 | } |
120 | ||
03ba3782 | 121 | static void wb_work_complete(struct bdi_work *work) |
1da177e4 | 122 | { |
c4a77a6c | 123 | const enum writeback_sync_modes sync_mode = work->args.sync_mode; |
77b9d059 | 124 | int onstack = bdi_work_on_stack(work); |
1da177e4 LT |
125 | |
126 | /* | |
03ba3782 JA |
127 | * For allocated work, we can clear the done/seen bit right here. |
128 | * For on-stack work, we need to postpone both the clear and free | |
129 | * to after the RCU grace period, since the stack could be invalidated | |
130 | * as soon as bdi_work_clear() has done the wakeup. | |
1da177e4 | 131 | */ |
77b9d059 | 132 | if (!onstack) |
03ba3782 | 133 | bdi_work_clear(work); |
77b9d059 | 134 | if (sync_mode == WB_SYNC_NONE || onstack) |
03ba3782 JA |
135 | call_rcu(&work->rcu_head, bdi_work_free); |
136 | } | |
1da177e4 | 137 | |
03ba3782 JA |
138 | static void wb_clear_pending(struct bdi_writeback *wb, struct bdi_work *work) |
139 | { | |
1da177e4 | 140 | /* |
03ba3782 JA |
141 | * The caller has retrieved the work arguments from this work, |
142 | * drop our reference. If this is the last ref, delete and free it | |
1da177e4 | 143 | */ |
03ba3782 JA |
144 | if (atomic_dec_and_test(&work->pending)) { |
145 | struct backing_dev_info *bdi = wb->bdi; | |
1da177e4 | 146 | |
03ba3782 JA |
147 | spin_lock(&bdi->wb_lock); |
148 | list_del_rcu(&work->list); | |
149 | spin_unlock(&bdi->wb_lock); | |
1da177e4 | 150 | |
03ba3782 JA |
151 | wb_work_complete(work); |
152 | } | |
153 | } | |
1da177e4 | 154 | |
03ba3782 JA |
155 | static void bdi_queue_work(struct backing_dev_info *bdi, struct bdi_work *work) |
156 | { | |
bcddc3f0 JA |
157 | work->seen = bdi->wb_mask; |
158 | BUG_ON(!work->seen); | |
159 | atomic_set(&work->pending, bdi->wb_cnt); | |
160 | BUG_ON(!bdi->wb_cnt); | |
1da177e4 | 161 | |
bcddc3f0 | 162 | /* |
deed62ed NP |
163 | * list_add_tail_rcu() contains the necessary barriers to |
164 | * make sure the above stores are seen before the item is | |
165 | * noticed on the list | |
bcddc3f0 | 166 | */ |
bcddc3f0 JA |
167 | spin_lock(&bdi->wb_lock); |
168 | list_add_tail_rcu(&work->list, &bdi->work_list); | |
169 | spin_unlock(&bdi->wb_lock); | |
03ba3782 JA |
170 | |
171 | /* | |
172 | * If the default thread isn't there, make sure we add it. When | |
173 | * it gets created and wakes up, we'll run this work. | |
174 | */ | |
175 | if (unlikely(list_empty_careful(&bdi->wb_list))) | |
176 | wake_up_process(default_backing_dev_info.wb.task); | |
177 | else { | |
178 | struct bdi_writeback *wb = &bdi->wb; | |
1da177e4 | 179 | |
1ef7d9aa | 180 | if (wb->task) |
03ba3782 | 181 | wake_up_process(wb->task); |
1da177e4 | 182 | } |
1da177e4 LT |
183 | } |
184 | ||
03ba3782 JA |
185 | /* |
186 | * Used for on-stack allocated work items. The caller needs to wait until | |
187 | * the wb threads have acked the work before it's safe to continue. | |
188 | */ | |
189 | static void bdi_wait_on_work_clear(struct bdi_work *work) | |
190 | { | |
191 | wait_on_bit(&work->state, WS_USED_B, bdi_sched_wait, | |
192 | TASK_UNINTERRUPTIBLE); | |
193 | } | |
1da177e4 | 194 | |
f11fcae8 | 195 | static void bdi_alloc_queue_work(struct backing_dev_info *bdi, |
f17625b3 | 196 | struct wb_writeback_args *args) |
1da177e4 | 197 | { |
03ba3782 JA |
198 | struct bdi_work *work; |
199 | ||
bcddc3f0 JA |
200 | /* |
201 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
202 | * wakeup the thread for old dirty data writeback | |
203 | */ | |
03ba3782 | 204 | work = kmalloc(sizeof(*work), GFP_ATOMIC); |
bcddc3f0 | 205 | if (work) { |
b6e51316 | 206 | bdi_work_init(work, args); |
bcddc3f0 JA |
207 | bdi_queue_work(bdi, work); |
208 | } else { | |
209 | struct bdi_writeback *wb = &bdi->wb; | |
03ba3782 | 210 | |
bcddc3f0 JA |
211 | if (wb->task) |
212 | wake_up_process(wb->task); | |
213 | } | |
03ba3782 JA |
214 | } |
215 | ||
b6e51316 JA |
216 | /** |
217 | * bdi_sync_writeback - start and wait for writeback | |
218 | * @bdi: the backing device to write from | |
219 | * @sb: write inodes from this super_block | |
220 | * | |
221 | * Description: | |
222 | * This does WB_SYNC_ALL data integrity writeback and waits for the | |
223 | * IO to complete. Callers must hold the sb s_umount semaphore for | |
224 | * reading, to avoid having the super disappear before we are done. | |
225 | */ | |
226 | static void bdi_sync_writeback(struct backing_dev_info *bdi, | |
227 | struct super_block *sb) | |
03ba3782 | 228 | { |
b6e51316 JA |
229 | struct wb_writeback_args args = { |
230 | .sb = sb, | |
231 | .sync_mode = WB_SYNC_ALL, | |
232 | .nr_pages = LONG_MAX, | |
233 | .range_cyclic = 0, | |
e913fc82 JA |
234 | /* |
235 | * Setting sb_pinned is not necessary for WB_SYNC_ALL, but | |
236 | * lets make it explicitly clear. | |
237 | */ | |
238 | .sb_pinned = 1, | |
b6e51316 JA |
239 | }; |
240 | struct bdi_work work; | |
03ba3782 | 241 | |
b6e51316 JA |
242 | bdi_work_init(&work, &args); |
243 | work.state |= WS_ONSTACK; | |
03ba3782 | 244 | |
b6e51316 JA |
245 | bdi_queue_work(bdi, &work); |
246 | bdi_wait_on_work_clear(&work); | |
247 | } | |
248 | ||
249 | /** | |
250 | * bdi_start_writeback - start writeback | |
251 | * @bdi: the backing device to write from | |
4b6764fa | 252 | * @sb: write inodes from this super_block |
b6e51316 | 253 | * @nr_pages: the number of pages to write |
e913fc82 | 254 | * @sb_locked: caller already holds sb umount sem. |
b6e51316 JA |
255 | * |
256 | * Description: | |
257 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
258 | * started when this function returns, we make no guarentees on | |
e913fc82 | 259 | * completion. Caller specifies whether sb umount sem is held already or not. |
b6e51316 JA |
260 | * |
261 | */ | |
a72bfd4d | 262 | void bdi_start_writeback(struct backing_dev_info *bdi, struct super_block *sb, |
e913fc82 | 263 | long nr_pages, int sb_locked) |
b6e51316 JA |
264 | { |
265 | struct wb_writeback_args args = { | |
a72bfd4d | 266 | .sb = sb, |
b6e51316 JA |
267 | .sync_mode = WB_SYNC_NONE, |
268 | .nr_pages = nr_pages, | |
269 | .range_cyclic = 1, | |
e913fc82 | 270 | .sb_pinned = sb_locked, |
b6e51316 JA |
271 | }; |
272 | ||
d3ddec76 WF |
273 | /* |
274 | * We treat @nr_pages=0 as the special case to do background writeback, | |
275 | * ie. to sync pages until the background dirty threshold is reached. | |
276 | */ | |
277 | if (!nr_pages) { | |
278 | args.nr_pages = LONG_MAX; | |
279 | args.for_background = 1; | |
280 | } | |
281 | ||
f17625b3 | 282 | bdi_alloc_queue_work(bdi, &args); |
1da177e4 LT |
283 | } |
284 | ||
6610a0bc AM |
285 | /* |
286 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
287 | * furthest end of its superblock's dirty-inode list. | |
288 | * | |
289 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 290 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
291 | * the case then the inode must have been redirtied while it was being written |
292 | * out and we don't reset its dirtied_when. | |
293 | */ | |
294 | static void redirty_tail(struct inode *inode) | |
295 | { | |
03ba3782 | 296 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
6610a0bc | 297 | |
03ba3782 | 298 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 299 | struct inode *tail; |
6610a0bc | 300 | |
03ba3782 | 301 | tail = list_entry(wb->b_dirty.next, struct inode, i_list); |
66f3b8e2 | 302 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
303 | inode->dirtied_when = jiffies; |
304 | } | |
03ba3782 | 305 | list_move(&inode->i_list, &wb->b_dirty); |
6610a0bc AM |
306 | } |
307 | ||
c986d1e2 | 308 | /* |
66f3b8e2 | 309 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 310 | */ |
0e0f4fc2 | 311 | static void requeue_io(struct inode *inode) |
c986d1e2 | 312 | { |
03ba3782 JA |
313 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
314 | ||
315 | list_move(&inode->i_list, &wb->b_more_io); | |
c986d1e2 AM |
316 | } |
317 | ||
1c0eeaf5 JE |
318 | static void inode_sync_complete(struct inode *inode) |
319 | { | |
320 | /* | |
321 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
322 | */ | |
323 | smp_mb(); | |
324 | wake_up_bit(&inode->i_state, __I_SYNC); | |
325 | } | |
326 | ||
d2caa3c5 JL |
327 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
328 | { | |
329 | bool ret = time_after(inode->dirtied_when, t); | |
330 | #ifndef CONFIG_64BIT | |
331 | /* | |
332 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
333 | * It _appears_ to be in the future, but is actually in distant past. | |
334 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 335 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
336 | */ |
337 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
338 | #endif | |
339 | return ret; | |
340 | } | |
341 | ||
2c136579 FW |
342 | /* |
343 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
344 | */ | |
345 | static void move_expired_inodes(struct list_head *delaying_queue, | |
346 | struct list_head *dispatch_queue, | |
347 | unsigned long *older_than_this) | |
348 | { | |
5c03449d SL |
349 | LIST_HEAD(tmp); |
350 | struct list_head *pos, *node; | |
cf137307 | 351 | struct super_block *sb = NULL; |
5c03449d | 352 | struct inode *inode; |
cf137307 | 353 | int do_sb_sort = 0; |
5c03449d | 354 | |
2c136579 | 355 | while (!list_empty(delaying_queue)) { |
5c03449d | 356 | inode = list_entry(delaying_queue->prev, struct inode, i_list); |
2c136579 | 357 | if (older_than_this && |
d2caa3c5 | 358 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 | 359 | break; |
cf137307 JA |
360 | if (sb && sb != inode->i_sb) |
361 | do_sb_sort = 1; | |
362 | sb = inode->i_sb; | |
5c03449d SL |
363 | list_move(&inode->i_list, &tmp); |
364 | } | |
365 | ||
cf137307 JA |
366 | /* just one sb in list, splice to dispatch_queue and we're done */ |
367 | if (!do_sb_sort) { | |
368 | list_splice(&tmp, dispatch_queue); | |
369 | return; | |
370 | } | |
371 | ||
5c03449d SL |
372 | /* Move inodes from one superblock together */ |
373 | while (!list_empty(&tmp)) { | |
374 | inode = list_entry(tmp.prev, struct inode, i_list); | |
375 | sb = inode->i_sb; | |
376 | list_for_each_prev_safe(pos, node, &tmp) { | |
377 | inode = list_entry(pos, struct inode, i_list); | |
378 | if (inode->i_sb == sb) | |
379 | list_move(&inode->i_list, dispatch_queue); | |
380 | } | |
2c136579 FW |
381 | } |
382 | } | |
383 | ||
384 | /* | |
385 | * Queue all expired dirty inodes for io, eldest first. | |
386 | */ | |
03ba3782 | 387 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 388 | { |
03ba3782 JA |
389 | list_splice_init(&wb->b_more_io, wb->b_io.prev); |
390 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); | |
66f3b8e2 JA |
391 | } |
392 | ||
a9185b41 | 393 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 394 | { |
03ba3782 | 395 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 396 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 397 | return 0; |
08d8e974 | 398 | } |
08d8e974 | 399 | |
1da177e4 | 400 | /* |
01c03194 CH |
401 | * Wait for writeback on an inode to complete. |
402 | */ | |
403 | static void inode_wait_for_writeback(struct inode *inode) | |
404 | { | |
405 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
406 | wait_queue_head_t *wqh; | |
407 | ||
408 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
409 | do { | |
410 | spin_unlock(&inode_lock); | |
411 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); | |
412 | spin_lock(&inode_lock); | |
413 | } while (inode->i_state & I_SYNC); | |
414 | } | |
415 | ||
416 | /* | |
417 | * Write out an inode's dirty pages. Called under inode_lock. Either the | |
418 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
419 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
420 | * | |
1da177e4 LT |
421 | * If `wait' is set, wait on the writeout. |
422 | * | |
423 | * The whole writeout design is quite complex and fragile. We want to avoid | |
424 | * starvation of particular inodes when others are being redirtied, prevent | |
425 | * livelocks, etc. | |
426 | * | |
427 | * Called under inode_lock. | |
428 | */ | |
429 | static int | |
01c03194 | 430 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 431 | { |
1da177e4 | 432 | struct address_space *mapping = inode->i_mapping; |
01c03194 | 433 | unsigned dirty; |
1da177e4 LT |
434 | int ret; |
435 | ||
01c03194 CH |
436 | if (!atomic_read(&inode->i_count)) |
437 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
438 | else | |
439 | WARN_ON(inode->i_state & I_WILL_FREE); | |
440 | ||
441 | if (inode->i_state & I_SYNC) { | |
442 | /* | |
443 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 444 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
445 | * writeback can proceed with the other inodes on s_io. |
446 | * | |
447 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 448 | * completed a full scan of b_io. |
01c03194 | 449 | */ |
a9185b41 | 450 | if (wbc->sync_mode != WB_SYNC_ALL) { |
01c03194 CH |
451 | requeue_io(inode); |
452 | return 0; | |
453 | } | |
454 | ||
455 | /* | |
456 | * It's a data-integrity sync. We must wait. | |
457 | */ | |
458 | inode_wait_for_writeback(inode); | |
459 | } | |
460 | ||
1c0eeaf5 | 461 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 462 | |
5547e8aa | 463 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 464 | inode->i_state |= I_SYNC; |
5547e8aa | 465 | inode->i_state &= ~I_DIRTY_PAGES; |
1da177e4 LT |
466 | spin_unlock(&inode_lock); |
467 | ||
468 | ret = do_writepages(mapping, wbc); | |
469 | ||
26821ed4 CH |
470 | /* |
471 | * Make sure to wait on the data before writing out the metadata. | |
472 | * This is important for filesystems that modify metadata on data | |
473 | * I/O completion. | |
474 | */ | |
a9185b41 | 475 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 476 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
477 | if (ret == 0) |
478 | ret = err; | |
479 | } | |
480 | ||
5547e8aa DM |
481 | /* |
482 | * Some filesystems may redirty the inode during the writeback | |
483 | * due to delalloc, clear dirty metadata flags right before | |
484 | * write_inode() | |
485 | */ | |
486 | spin_lock(&inode_lock); | |
487 | dirty = inode->i_state & I_DIRTY; | |
488 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
489 | spin_unlock(&inode_lock); | |
26821ed4 CH |
490 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
491 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 492 | int err = write_inode(inode, wbc); |
1da177e4 LT |
493 | if (ret == 0) |
494 | ret = err; | |
495 | } | |
496 | ||
497 | spin_lock(&inode_lock); | |
1c0eeaf5 | 498 | inode->i_state &= ~I_SYNC; |
84a89245 | 499 | if (!(inode->i_state & (I_FREEING | I_CLEAR))) { |
b3af9468 | 500 | if ((inode->i_state & I_DIRTY_PAGES) && wbc->for_kupdate) { |
ae1b7f7d | 501 | /* |
b3af9468 WF |
502 | * More pages get dirtied by a fast dirtier. |
503 | */ | |
504 | goto select_queue; | |
505 | } else if (inode->i_state & I_DIRTY) { | |
506 | /* | |
507 | * At least XFS will redirty the inode during the | |
508 | * writeback (delalloc) and on io completion (isize). | |
ae1b7f7d WF |
509 | */ |
510 | redirty_tail(inode); | |
511 | } else if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { | |
1da177e4 LT |
512 | /* |
513 | * We didn't write back all the pages. nfs_writepages() | |
514 | * sometimes bales out without doing anything. Redirty | |
66f3b8e2 | 515 | * the inode; Move it from b_io onto b_more_io/b_dirty. |
1b43ef91 AM |
516 | */ |
517 | /* | |
518 | * akpm: if the caller was the kupdate function we put | |
66f3b8e2 | 519 | * this inode at the head of b_dirty so it gets first |
1b43ef91 AM |
520 | * consideration. Otherwise, move it to the tail, for |
521 | * the reasons described there. I'm not really sure | |
522 | * how much sense this makes. Presumably I had a good | |
523 | * reasons for doing it this way, and I'd rather not | |
524 | * muck with it at present. | |
1da177e4 LT |
525 | */ |
526 | if (wbc->for_kupdate) { | |
527 | /* | |
2c136579 | 528 | * For the kupdate function we move the inode |
66f3b8e2 | 529 | * to b_more_io so it will get more writeout as |
2c136579 | 530 | * soon as the queue becomes uncongested. |
1da177e4 LT |
531 | */ |
532 | inode->i_state |= I_DIRTY_PAGES; | |
b3af9468 | 533 | select_queue: |
8bc3be27 FW |
534 | if (wbc->nr_to_write <= 0) { |
535 | /* | |
536 | * slice used up: queue for next turn | |
537 | */ | |
538 | requeue_io(inode); | |
539 | } else { | |
540 | /* | |
541 | * somehow blocked: retry later | |
542 | */ | |
543 | redirty_tail(inode); | |
544 | } | |
1da177e4 LT |
545 | } else { |
546 | /* | |
547 | * Otherwise fully redirty the inode so that | |
548 | * other inodes on this superblock will get some | |
549 | * writeout. Otherwise heavy writing to one | |
550 | * file would indefinitely suspend writeout of | |
551 | * all the other files. | |
552 | */ | |
553 | inode->i_state |= I_DIRTY_PAGES; | |
1b43ef91 | 554 | redirty_tail(inode); |
1da177e4 | 555 | } |
1da177e4 LT |
556 | } else if (atomic_read(&inode->i_count)) { |
557 | /* | |
558 | * The inode is clean, inuse | |
559 | */ | |
560 | list_move(&inode->i_list, &inode_in_use); | |
561 | } else { | |
562 | /* | |
563 | * The inode is clean, unused | |
564 | */ | |
565 | list_move(&inode->i_list, &inode_unused); | |
1da177e4 LT |
566 | } |
567 | } | |
1c0eeaf5 | 568 | inode_sync_complete(inode); |
1da177e4 LT |
569 | return ret; |
570 | } | |
571 | ||
f11c9c5c | 572 | static void unpin_sb_for_writeback(struct super_block *sb) |
9ecc2738 | 573 | { |
f11c9c5c ES |
574 | up_read(&sb->s_umount); |
575 | put_super(sb); | |
9ecc2738 JA |
576 | } |
577 | ||
f11c9c5c ES |
578 | enum sb_pin_state { |
579 | SB_PINNED, | |
580 | SB_NOT_PINNED, | |
581 | SB_PIN_FAILED | |
582 | }; | |
583 | ||
03ba3782 JA |
584 | /* |
585 | * For WB_SYNC_NONE writeback, the caller does not have the sb pinned | |
586 | * before calling writeback. So make sure that we do pin it, so it doesn't | |
587 | * go away while we are writing inodes from it. | |
03ba3782 | 588 | */ |
f11c9c5c ES |
589 | static enum sb_pin_state pin_sb_for_writeback(struct writeback_control *wbc, |
590 | struct super_block *sb) | |
03ba3782 | 591 | { |
03ba3782 JA |
592 | /* |
593 | * Caller must already hold the ref for this | |
594 | */ | |
e913fc82 | 595 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->sb_pinned) { |
03ba3782 | 596 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
f11c9c5c | 597 | return SB_NOT_PINNED; |
03ba3782 | 598 | } |
03ba3782 JA |
599 | spin_lock(&sb_lock); |
600 | sb->s_count++; | |
601 | if (down_read_trylock(&sb->s_umount)) { | |
602 | if (sb->s_root) { | |
603 | spin_unlock(&sb_lock); | |
f11c9c5c | 604 | return SB_PINNED; |
03ba3782 JA |
605 | } |
606 | /* | |
607 | * umounted, drop rwsem again and fall through to failure | |
608 | */ | |
609 | up_read(&sb->s_umount); | |
610 | } | |
03ba3782 JA |
611 | sb->s_count--; |
612 | spin_unlock(&sb_lock); | |
f11c9c5c | 613 | return SB_PIN_FAILED; |
03ba3782 JA |
614 | } |
615 | ||
f11c9c5c ES |
616 | /* |
617 | * Write a portion of b_io inodes which belong to @sb. | |
618 | * If @wbc->sb != NULL, then find and write all such | |
619 | * inodes. Otherwise write only ones which go sequentially | |
620 | * in reverse order. | |
621 | * Return 1, if the caller writeback routine should be | |
622 | * interrupted. Otherwise return 0. | |
623 | */ | |
624 | static int writeback_sb_inodes(struct super_block *sb, | |
625 | struct bdi_writeback *wb, | |
626 | struct writeback_control *wbc) | |
1da177e4 | 627 | { |
03ba3782 | 628 | while (!list_empty(&wb->b_io)) { |
1da177e4 | 629 | long pages_skipped; |
f11c9c5c ES |
630 | struct inode *inode = list_entry(wb->b_io.prev, |
631 | struct inode, i_list); | |
632 | if (wbc->sb && sb != inode->i_sb) { | |
633 | /* super block given and doesn't | |
634 | match, skip this inode */ | |
66f3b8e2 JA |
635 | redirty_tail(inode); |
636 | continue; | |
637 | } | |
f11c9c5c ES |
638 | if (sb != inode->i_sb) |
639 | /* finish with this superblock */ | |
640 | return 0; | |
84a89245 | 641 | if (inode->i_state & (I_NEW | I_WILL_FREE)) { |
7ef0d737 NP |
642 | requeue_io(inode); |
643 | continue; | |
644 | } | |
d2caa3c5 JL |
645 | /* |
646 | * Was this inode dirtied after sync_sb_inodes was called? | |
647 | * This keeps sync from extra jobs and livelock. | |
648 | */ | |
f11c9c5c ES |
649 | if (inode_dirtied_after(inode, wbc->wb_start)) |
650 | return 1; | |
1da177e4 | 651 | |
84a89245 | 652 | BUG_ON(inode->i_state & (I_FREEING | I_CLEAR)); |
1da177e4 LT |
653 | __iget(inode); |
654 | pages_skipped = wbc->pages_skipped; | |
01c03194 | 655 | writeback_single_inode(inode, wbc); |
1da177e4 LT |
656 | if (wbc->pages_skipped != pages_skipped) { |
657 | /* | |
658 | * writeback is not making progress due to locked | |
659 | * buffers. Skip this inode for now. | |
660 | */ | |
f57b9b7b | 661 | redirty_tail(inode); |
1da177e4 LT |
662 | } |
663 | spin_unlock(&inode_lock); | |
1da177e4 | 664 | iput(inode); |
4ffc8444 | 665 | cond_resched(); |
1da177e4 | 666 | spin_lock(&inode_lock); |
8bc3be27 FW |
667 | if (wbc->nr_to_write <= 0) { |
668 | wbc->more_io = 1; | |
f11c9c5c | 669 | return 1; |
8bc3be27 | 670 | } |
03ba3782 | 671 | if (!list_empty(&wb->b_more_io)) |
8bc3be27 | 672 | wbc->more_io = 1; |
1da177e4 | 673 | } |
f11c9c5c ES |
674 | /* b_io is empty */ |
675 | return 1; | |
676 | } | |
677 | ||
678 | static void writeback_inodes_wb(struct bdi_writeback *wb, | |
679 | struct writeback_control *wbc) | |
680 | { | |
681 | int ret = 0; | |
682 | ||
683 | wbc->wb_start = jiffies; /* livelock avoidance */ | |
684 | spin_lock(&inode_lock); | |
685 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) | |
686 | queue_io(wb, wbc->older_than_this); | |
38f21977 | 687 | |
f11c9c5c ES |
688 | while (!list_empty(&wb->b_io)) { |
689 | struct inode *inode = list_entry(wb->b_io.prev, | |
690 | struct inode, i_list); | |
691 | struct super_block *sb = inode->i_sb; | |
692 | enum sb_pin_state state; | |
9ecc2738 | 693 | |
f11c9c5c ES |
694 | if (wbc->sb && sb != wbc->sb) { |
695 | /* super block given and doesn't | |
696 | match, skip this inode */ | |
697 | redirty_tail(inode); | |
698 | continue; | |
699 | } | |
700 | state = pin_sb_for_writeback(wbc, sb); | |
701 | ||
702 | if (state == SB_PIN_FAILED) { | |
703 | requeue_io(inode); | |
704 | continue; | |
705 | } | |
706 | ret = writeback_sb_inodes(sb, wb, wbc); | |
707 | ||
708 | if (state == SB_PINNED) | |
709 | unpin_sb_for_writeback(sb); | |
710 | if (ret) | |
711 | break; | |
712 | } | |
66f3b8e2 JA |
713 | spin_unlock(&inode_lock); |
714 | /* Leave any unwritten inodes on b_io */ | |
715 | } | |
716 | ||
03ba3782 JA |
717 | void writeback_inodes_wbc(struct writeback_control *wbc) |
718 | { | |
719 | struct backing_dev_info *bdi = wbc->bdi; | |
720 | ||
721 | writeback_inodes_wb(&bdi->wb, wbc); | |
722 | } | |
723 | ||
66f3b8e2 | 724 | /* |
03ba3782 JA |
725 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
726 | * operation. We do this so we don't hold I_SYNC against an inode for | |
727 | * enormous amounts of time, which would block a userspace task which has | |
728 | * been forced to throttle against that inode. Also, the code reevaluates | |
729 | * the dirty each time it has written this many pages. | |
730 | */ | |
731 | #define MAX_WRITEBACK_PAGES 1024 | |
732 | ||
733 | static inline bool over_bground_thresh(void) | |
734 | { | |
735 | unsigned long background_thresh, dirty_thresh; | |
736 | ||
737 | get_dirty_limits(&background_thresh, &dirty_thresh, NULL, NULL); | |
738 | ||
739 | return (global_page_state(NR_FILE_DIRTY) + | |
740 | global_page_state(NR_UNSTABLE_NFS) >= background_thresh); | |
741 | } | |
742 | ||
743 | /* | |
744 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 745 | * |
03ba3782 JA |
746 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
747 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
748 | * just walks the superblock inode list, writing back any inodes which are | |
749 | * older than a specific point in time. | |
66f3b8e2 | 750 | * |
03ba3782 JA |
751 | * Try to run once per dirty_writeback_interval. But if a writeback event |
752 | * takes longer than a dirty_writeback_interval interval, then leave a | |
753 | * one-second gap. | |
66f3b8e2 | 754 | * |
03ba3782 JA |
755 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
756 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 757 | */ |
c4a77a6c JA |
758 | static long wb_writeback(struct bdi_writeback *wb, |
759 | struct wb_writeback_args *args) | |
66f3b8e2 | 760 | { |
03ba3782 JA |
761 | struct writeback_control wbc = { |
762 | .bdi = wb->bdi, | |
c4a77a6c JA |
763 | .sb = args->sb, |
764 | .sync_mode = args->sync_mode, | |
03ba3782 | 765 | .older_than_this = NULL, |
c4a77a6c | 766 | .for_kupdate = args->for_kupdate, |
b17621fe | 767 | .for_background = args->for_background, |
c4a77a6c | 768 | .range_cyclic = args->range_cyclic, |
e913fc82 | 769 | .sb_pinned = args->sb_pinned, |
03ba3782 JA |
770 | }; |
771 | unsigned long oldest_jif; | |
772 | long wrote = 0; | |
a5989bdc | 773 | struct inode *inode; |
66f3b8e2 | 774 | |
03ba3782 JA |
775 | if (wbc.for_kupdate) { |
776 | wbc.older_than_this = &oldest_jif; | |
777 | oldest_jif = jiffies - | |
778 | msecs_to_jiffies(dirty_expire_interval * 10); | |
779 | } | |
c4a77a6c JA |
780 | if (!wbc.range_cyclic) { |
781 | wbc.range_start = 0; | |
782 | wbc.range_end = LLONG_MAX; | |
783 | } | |
38f21977 | 784 | |
03ba3782 JA |
785 | for (;;) { |
786 | /* | |
d3ddec76 | 787 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 788 | */ |
d3ddec76 | 789 | if (args->nr_pages <= 0) |
03ba3782 | 790 | break; |
66f3b8e2 | 791 | |
38f21977 | 792 | /* |
d3ddec76 WF |
793 | * For background writeout, stop when we are below the |
794 | * background dirty threshold | |
38f21977 | 795 | */ |
d3ddec76 | 796 | if (args->for_background && !over_bground_thresh()) |
03ba3782 | 797 | break; |
38f21977 | 798 | |
03ba3782 | 799 | wbc.more_io = 0; |
03ba3782 JA |
800 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; |
801 | wbc.pages_skipped = 0; | |
802 | writeback_inodes_wb(wb, &wbc); | |
c4a77a6c | 803 | args->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
03ba3782 JA |
804 | wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
805 | ||
806 | /* | |
71fd05a8 | 807 | * If we consumed everything, see if we have more |
03ba3782 | 808 | */ |
71fd05a8 JA |
809 | if (wbc.nr_to_write <= 0) |
810 | continue; | |
811 | /* | |
812 | * Didn't write everything and we don't have more IO, bail | |
813 | */ | |
814 | if (!wbc.more_io) | |
03ba3782 | 815 | break; |
71fd05a8 JA |
816 | /* |
817 | * Did we write something? Try for more | |
818 | */ | |
819 | if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) | |
820 | continue; | |
821 | /* | |
822 | * Nothing written. Wait for some inode to | |
823 | * become available for writeback. Otherwise | |
824 | * we'll just busyloop. | |
825 | */ | |
826 | spin_lock(&inode_lock); | |
827 | if (!list_empty(&wb->b_more_io)) { | |
828 | inode = list_entry(wb->b_more_io.prev, | |
829 | struct inode, i_list); | |
830 | inode_wait_for_writeback(inode); | |
03ba3782 | 831 | } |
71fd05a8 | 832 | spin_unlock(&inode_lock); |
03ba3782 JA |
833 | } |
834 | ||
835 | return wrote; | |
836 | } | |
837 | ||
838 | /* | |
839 | * Return the next bdi_work struct that hasn't been processed by this | |
8010c3b6 JA |
840 | * wb thread yet. ->seen is initially set for each thread that exists |
841 | * for this device, when a thread first notices a piece of work it | |
842 | * clears its bit. Depending on writeback type, the thread will notify | |
843 | * completion on either receiving the work (WB_SYNC_NONE) or after | |
844 | * it is done (WB_SYNC_ALL). | |
03ba3782 JA |
845 | */ |
846 | static struct bdi_work *get_next_work_item(struct backing_dev_info *bdi, | |
847 | struct bdi_writeback *wb) | |
848 | { | |
849 | struct bdi_work *work, *ret = NULL; | |
850 | ||
851 | rcu_read_lock(); | |
852 | ||
853 | list_for_each_entry_rcu(work, &bdi->work_list, list) { | |
77fad5e6 | 854 | if (!test_bit(wb->nr, &work->seen)) |
03ba3782 | 855 | continue; |
77fad5e6 | 856 | clear_bit(wb->nr, &work->seen); |
03ba3782 JA |
857 | |
858 | ret = work; | |
859 | break; | |
860 | } | |
861 | ||
862 | rcu_read_unlock(); | |
863 | return ret; | |
864 | } | |
865 | ||
866 | static long wb_check_old_data_flush(struct bdi_writeback *wb) | |
867 | { | |
868 | unsigned long expired; | |
869 | long nr_pages; | |
870 | ||
69b62d01 JA |
871 | /* |
872 | * When set to zero, disable periodic writeback | |
873 | */ | |
874 | if (!dirty_writeback_interval) | |
875 | return 0; | |
876 | ||
03ba3782 JA |
877 | expired = wb->last_old_flush + |
878 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
879 | if (time_before(jiffies, expired)) | |
880 | return 0; | |
881 | ||
882 | wb->last_old_flush = jiffies; | |
883 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
884 | global_page_state(NR_UNSTABLE_NFS) + | |
885 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); | |
886 | ||
c4a77a6c JA |
887 | if (nr_pages) { |
888 | struct wb_writeback_args args = { | |
889 | .nr_pages = nr_pages, | |
890 | .sync_mode = WB_SYNC_NONE, | |
891 | .for_kupdate = 1, | |
892 | .range_cyclic = 1, | |
893 | }; | |
894 | ||
895 | return wb_writeback(wb, &args); | |
896 | } | |
03ba3782 JA |
897 | |
898 | return 0; | |
899 | } | |
900 | ||
901 | /* | |
902 | * Retrieve work items and do the writeback they describe | |
903 | */ | |
904 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
905 | { | |
906 | struct backing_dev_info *bdi = wb->bdi; | |
907 | struct bdi_work *work; | |
c4a77a6c | 908 | long wrote = 0; |
03ba3782 JA |
909 | |
910 | while ((work = get_next_work_item(bdi, wb)) != NULL) { | |
c4a77a6c | 911 | struct wb_writeback_args args = work->args; |
03ba3782 JA |
912 | |
913 | /* | |
914 | * Override sync mode, in case we must wait for completion | |
915 | */ | |
916 | if (force_wait) | |
c4a77a6c | 917 | work->args.sync_mode = args.sync_mode = WB_SYNC_ALL; |
03ba3782 JA |
918 | |
919 | /* | |
920 | * If this isn't a data integrity operation, just notify | |
921 | * that we have seen this work and we are now starting it. | |
922 | */ | |
f17625b3 | 923 | if (args.sync_mode == WB_SYNC_NONE) |
03ba3782 JA |
924 | wb_clear_pending(wb, work); |
925 | ||
c4a77a6c | 926 | wrote += wb_writeback(wb, &args); |
03ba3782 JA |
927 | |
928 | /* | |
929 | * This is a data integrity writeback, so only do the | |
930 | * notification when we have completed the work. | |
931 | */ | |
f17625b3 | 932 | if (args.sync_mode == WB_SYNC_ALL) |
03ba3782 JA |
933 | wb_clear_pending(wb, work); |
934 | } | |
935 | ||
936 | /* | |
937 | * Check for periodic writeback, kupdated() style | |
938 | */ | |
939 | wrote += wb_check_old_data_flush(wb); | |
940 | ||
941 | return wrote; | |
942 | } | |
943 | ||
944 | /* | |
945 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
946 | * wakes up periodically and does kupdated style flushing. | |
947 | */ | |
948 | int bdi_writeback_task(struct bdi_writeback *wb) | |
949 | { | |
950 | unsigned long last_active = jiffies; | |
951 | unsigned long wait_jiffies = -1UL; | |
952 | long pages_written; | |
953 | ||
954 | while (!kthread_should_stop()) { | |
955 | pages_written = wb_do_writeback(wb, 0); | |
956 | ||
957 | if (pages_written) | |
958 | last_active = jiffies; | |
959 | else if (wait_jiffies != -1UL) { | |
960 | unsigned long max_idle; | |
961 | ||
38f21977 | 962 | /* |
03ba3782 JA |
963 | * Longest period of inactivity that we tolerate. If we |
964 | * see dirty data again later, the task will get | |
965 | * recreated automatically. | |
38f21977 | 966 | */ |
03ba3782 JA |
967 | max_idle = max(5UL * 60 * HZ, wait_jiffies); |
968 | if (time_after(jiffies, max_idle + last_active)) | |
969 | break; | |
970 | } | |
971 | ||
69b62d01 JA |
972 | if (dirty_writeback_interval) { |
973 | wait_jiffies = msecs_to_jiffies(dirty_writeback_interval * 10); | |
974 | schedule_timeout_interruptible(wait_jiffies); | |
f9eadbbd JA |
975 | } else { |
976 | set_current_state(TASK_INTERRUPTIBLE); | |
977 | if (list_empty_careful(&wb->bdi->work_list) && | |
978 | !kthread_should_stop()) | |
979 | schedule(); | |
980 | __set_current_state(TASK_RUNNING); | |
981 | } | |
69b62d01 | 982 | |
03ba3782 JA |
983 | try_to_freeze(); |
984 | } | |
985 | ||
986 | return 0; | |
987 | } | |
988 | ||
989 | /* | |
b6e51316 JA |
990 | * Schedule writeback for all backing devices. This does WB_SYNC_NONE |
991 | * writeback, for integrity writeback see bdi_sync_writeback(). | |
03ba3782 | 992 | */ |
b6e51316 | 993 | static void bdi_writeback_all(struct super_block *sb, long nr_pages) |
03ba3782 | 994 | { |
b6e51316 JA |
995 | struct wb_writeback_args args = { |
996 | .sb = sb, | |
997 | .nr_pages = nr_pages, | |
998 | .sync_mode = WB_SYNC_NONE, | |
999 | }; | |
03ba3782 | 1000 | struct backing_dev_info *bdi; |
03ba3782 | 1001 | |
cfc4ba53 | 1002 | rcu_read_lock(); |
03ba3782 | 1003 | |
cfc4ba53 | 1004 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
1005 | if (!bdi_has_dirty_io(bdi)) |
1006 | continue; | |
38f21977 | 1007 | |
f17625b3 | 1008 | bdi_alloc_queue_work(bdi, &args); |
03ba3782 JA |
1009 | } |
1010 | ||
cfc4ba53 | 1011 | rcu_read_unlock(); |
1da177e4 LT |
1012 | } |
1013 | ||
1014 | /* | |
03ba3782 JA |
1015 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
1016 | * the whole world. | |
1017 | */ | |
1018 | void wakeup_flusher_threads(long nr_pages) | |
1019 | { | |
03ba3782 JA |
1020 | if (nr_pages == 0) |
1021 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
1022 | global_page_state(NR_UNSTABLE_NFS); | |
b6e51316 | 1023 | bdi_writeback_all(NULL, nr_pages); |
03ba3782 JA |
1024 | } |
1025 | ||
1026 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) | |
1027 | { | |
1028 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
1029 | struct dentry *dentry; | |
1030 | const char *name = "?"; | |
1031 | ||
1032 | dentry = d_find_alias(inode); | |
1033 | if (dentry) { | |
1034 | spin_lock(&dentry->d_lock); | |
1035 | name = (const char *) dentry->d_name.name; | |
1036 | } | |
1037 | printk(KERN_DEBUG | |
1038 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
1039 | current->comm, task_pid_nr(current), inode->i_ino, | |
1040 | name, inode->i_sb->s_id); | |
1041 | if (dentry) { | |
1042 | spin_unlock(&dentry->d_lock); | |
1043 | dput(dentry); | |
1044 | } | |
1045 | } | |
1046 | } | |
1047 | ||
1048 | /** | |
1049 | * __mark_inode_dirty - internal function | |
1050 | * @inode: inode to mark | |
1051 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
1052 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
1053 | * mark_inode_dirty_sync. | |
1da177e4 | 1054 | * |
03ba3782 JA |
1055 | * Put the inode on the super block's dirty list. |
1056 | * | |
1057 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
1058 | * dirty list only if it is hashed or if it refers to a blockdev. | |
1059 | * If it was not hashed, it will never be added to the dirty list | |
1060 | * even if it is later hashed, as it will have been marked dirty already. | |
1061 | * | |
1062 | * In short, make sure you hash any inodes _before_ you start marking | |
1063 | * them dirty. | |
1da177e4 | 1064 | * |
03ba3782 JA |
1065 | * This function *must* be atomic for the I_DIRTY_PAGES case - |
1066 | * set_page_dirty() is called under spinlock in several places. | |
1da177e4 | 1067 | * |
03ba3782 JA |
1068 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1069 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1070 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1071 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1072 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1073 | * blockdev inode. | |
1da177e4 | 1074 | */ |
03ba3782 | 1075 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1076 | { |
03ba3782 | 1077 | struct super_block *sb = inode->i_sb; |
1da177e4 | 1078 | |
03ba3782 JA |
1079 | /* |
1080 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1081 | * dirty the inode itself | |
1082 | */ | |
1083 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1084 | if (sb->s_op->dirty_inode) | |
1085 | sb->s_op->dirty_inode(inode); | |
1086 | } | |
1087 | ||
1088 | /* | |
1089 | * make sure that changes are seen by all cpus before we test i_state | |
1090 | * -- mikulas | |
1091 | */ | |
1092 | smp_mb(); | |
1093 | ||
1094 | /* avoid the locking if we can */ | |
1095 | if ((inode->i_state & flags) == flags) | |
1096 | return; | |
1097 | ||
1098 | if (unlikely(block_dump)) | |
1099 | block_dump___mark_inode_dirty(inode); | |
1100 | ||
1101 | spin_lock(&inode_lock); | |
1102 | if ((inode->i_state & flags) != flags) { | |
1103 | const int was_dirty = inode->i_state & I_DIRTY; | |
1104 | ||
1105 | inode->i_state |= flags; | |
1106 | ||
1107 | /* | |
1108 | * If the inode is being synced, just update its dirty state. | |
1109 | * The unlocker will place the inode on the appropriate | |
1110 | * superblock list, based upon its state. | |
1111 | */ | |
1112 | if (inode->i_state & I_SYNC) | |
1113 | goto out; | |
1114 | ||
1115 | /* | |
1116 | * Only add valid (hashed) inodes to the superblock's | |
1117 | * dirty list. Add blockdev inodes as well. | |
1118 | */ | |
1119 | if (!S_ISBLK(inode->i_mode)) { | |
1120 | if (hlist_unhashed(&inode->i_hash)) | |
1121 | goto out; | |
1122 | } | |
1123 | if (inode->i_state & (I_FREEING|I_CLEAR)) | |
1124 | goto out; | |
1125 | ||
1126 | /* | |
1127 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1128 | * reposition it (that would break b_dirty time-ordering). | |
1129 | */ | |
1130 | if (!was_dirty) { | |
1131 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; | |
500b067c JA |
1132 | struct backing_dev_info *bdi = wb->bdi; |
1133 | ||
1134 | if (bdi_cap_writeback_dirty(bdi) && | |
1135 | !test_bit(BDI_registered, &bdi->state)) { | |
1136 | WARN_ON(1); | |
1137 | printk(KERN_ERR "bdi-%s not registered\n", | |
1138 | bdi->name); | |
1139 | } | |
03ba3782 JA |
1140 | |
1141 | inode->dirtied_when = jiffies; | |
1142 | list_move(&inode->i_list, &wb->b_dirty); | |
1da177e4 | 1143 | } |
1da177e4 | 1144 | } |
03ba3782 JA |
1145 | out: |
1146 | spin_unlock(&inode_lock); | |
1147 | } | |
1148 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1149 | ||
1150 | /* | |
1151 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1152 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1153 | * | |
1154 | * If older_than_this is non-NULL, then only write out inodes which | |
1155 | * had their first dirtying at a time earlier than *older_than_this. | |
1156 | * | |
03ba3782 JA |
1157 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1158 | * This function assumes that the blockdev superblock's inodes are backed by | |
1159 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1160 | * assume that all inodes are backed by the same queue. | |
1161 | * | |
1162 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1163 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1164 | * on the writer throttling path, and we get decent balancing between many | |
1165 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1166 | */ | |
b6e51316 | 1167 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1168 | { |
1169 | struct inode *inode, *old_inode = NULL; | |
1170 | ||
1171 | /* | |
1172 | * We need to be protected against the filesystem going from | |
1173 | * r/o to r/w or vice versa. | |
1174 | */ | |
b6e51316 | 1175 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 JA |
1176 | |
1177 | spin_lock(&inode_lock); | |
1178 | ||
1179 | /* | |
1180 | * Data integrity sync. Must wait for all pages under writeback, | |
1181 | * because there may have been pages dirtied before our sync | |
1182 | * call, but which had writeout started before we write it out. | |
1183 | * In which case, the inode may not be on the dirty list, but | |
1184 | * we still have to wait for that writeout. | |
1185 | */ | |
b6e51316 | 1186 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
03ba3782 JA |
1187 | struct address_space *mapping; |
1188 | ||
1189 | if (inode->i_state & (I_FREEING|I_CLEAR|I_WILL_FREE|I_NEW)) | |
1190 | continue; | |
1191 | mapping = inode->i_mapping; | |
1192 | if (mapping->nrpages == 0) | |
1193 | continue; | |
1194 | __iget(inode); | |
1195 | spin_unlock(&inode_lock); | |
1196 | /* | |
1197 | * We hold a reference to 'inode' so it couldn't have | |
1198 | * been removed from s_inodes list while we dropped the | |
1199 | * inode_lock. We cannot iput the inode now as we can | |
1200 | * be holding the last reference and we cannot iput it | |
1201 | * under inode_lock. So we keep the reference and iput | |
1202 | * it later. | |
1203 | */ | |
1204 | iput(old_inode); | |
1205 | old_inode = inode; | |
1206 | ||
1207 | filemap_fdatawait(mapping); | |
1208 | ||
1209 | cond_resched(); | |
1210 | ||
1211 | spin_lock(&inode_lock); | |
1212 | } | |
1213 | spin_unlock(&inode_lock); | |
1214 | iput(old_inode); | |
1da177e4 LT |
1215 | } |
1216 | ||
e913fc82 JA |
1217 | static void __writeback_inodes_sb(struct super_block *sb, int sb_locked) |
1218 | { | |
1219 | unsigned long nr_dirty = global_page_state(NR_FILE_DIRTY); | |
1220 | unsigned long nr_unstable = global_page_state(NR_UNSTABLE_NFS); | |
1221 | long nr_to_write; | |
1222 | ||
1223 | nr_to_write = nr_dirty + nr_unstable + | |
1224 | (inodes_stat.nr_inodes - inodes_stat.nr_unused); | |
1225 | ||
1226 | bdi_start_writeback(sb->s_bdi, sb, nr_to_write, sb_locked); | |
1227 | } | |
1228 | ||
d8a8559c JA |
1229 | /** |
1230 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1231 | * @sb: the superblock | |
1da177e4 | 1232 | * |
d8a8559c JA |
1233 | * Start writeback on some inodes on this super_block. No guarantees are made |
1234 | * on how many (if any) will be written, and this function does not wait | |
1235 | * for IO completion of submitted IO. The number of pages submitted is | |
1236 | * returned. | |
1da177e4 | 1237 | */ |
b6e51316 | 1238 | void writeback_inodes_sb(struct super_block *sb) |
1da177e4 | 1239 | { |
e913fc82 | 1240 | __writeback_inodes_sb(sb, 0); |
d8a8559c JA |
1241 | } |
1242 | EXPORT_SYMBOL(writeback_inodes_sb); | |
1243 | ||
e913fc82 JA |
1244 | /** |
1245 | * writeback_inodes_sb_locked - writeback dirty inodes from given super_block | |
1246 | * @sb: the superblock | |
1247 | * | |
1248 | * Like writeback_inodes_sb(), except the caller already holds the | |
1249 | * sb umount sem. | |
1250 | */ | |
1251 | void writeback_inodes_sb_locked(struct super_block *sb) | |
1252 | { | |
1253 | __writeback_inodes_sb(sb, 1); | |
1254 | } | |
1255 | ||
17bd55d0 ES |
1256 | /** |
1257 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1258 | * @sb: the superblock | |
1259 | * | |
1260 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1261 | * Returns 1 if writeback was started, 0 if not. | |
1262 | */ | |
1263 | int writeback_inodes_sb_if_idle(struct super_block *sb) | |
1264 | { | |
1265 | if (!writeback_in_progress(sb->s_bdi)) { | |
1266 | writeback_inodes_sb(sb); | |
1267 | return 1; | |
1268 | } else | |
1269 | return 0; | |
1270 | } | |
1271 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1272 | ||
d8a8559c JA |
1273 | /** |
1274 | * sync_inodes_sb - sync sb inode pages | |
1275 | * @sb: the superblock | |
1276 | * | |
1277 | * This function writes and waits on any dirty inode belonging to this | |
1278 | * super_block. The number of pages synced is returned. | |
1279 | */ | |
b6e51316 | 1280 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1281 | { |
b6e51316 JA |
1282 | bdi_sync_writeback(sb->s_bdi, sb); |
1283 | wait_sb_inodes(sb); | |
1da177e4 | 1284 | } |
d8a8559c | 1285 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1286 | |
1da177e4 | 1287 | /** |
7f04c26d AA |
1288 | * write_inode_now - write an inode to disk |
1289 | * @inode: inode to write to disk | |
1290 | * @sync: whether the write should be synchronous or not | |
1291 | * | |
1292 | * This function commits an inode to disk immediately if it is dirty. This is | |
1293 | * primarily needed by knfsd. | |
1da177e4 | 1294 | * |
7f04c26d | 1295 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1296 | */ |
1da177e4 LT |
1297 | int write_inode_now(struct inode *inode, int sync) |
1298 | { | |
1299 | int ret; | |
1300 | struct writeback_control wbc = { | |
1301 | .nr_to_write = LONG_MAX, | |
18914b18 | 1302 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1303 | .range_start = 0, |
1304 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1305 | }; |
1306 | ||
1307 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1308 | wbc.nr_to_write = 0; |
1da177e4 LT |
1309 | |
1310 | might_sleep(); | |
1311 | spin_lock(&inode_lock); | |
01c03194 | 1312 | ret = writeback_single_inode(inode, &wbc); |
1da177e4 LT |
1313 | spin_unlock(&inode_lock); |
1314 | if (sync) | |
1c0eeaf5 | 1315 | inode_sync_wait(inode); |
1da177e4 LT |
1316 | return ret; |
1317 | } | |
1318 | EXPORT_SYMBOL(write_inode_now); | |
1319 | ||
1320 | /** | |
1321 | * sync_inode - write an inode and its pages to disk. | |
1322 | * @inode: the inode to sync | |
1323 | * @wbc: controls the writeback mode | |
1324 | * | |
1325 | * sync_inode() will write an inode and its pages to disk. It will also | |
1326 | * correctly update the inode on its superblock's dirty inode lists and will | |
1327 | * update inode->i_state. | |
1328 | * | |
1329 | * The caller must have a ref on the inode. | |
1330 | */ | |
1331 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1332 | { | |
1333 | int ret; | |
1334 | ||
1335 | spin_lock(&inode_lock); | |
01c03194 | 1336 | ret = writeback_single_inode(inode, wbc); |
1da177e4 LT |
1337 | spin_unlock(&inode_lock); |
1338 | return ret; | |
1339 | } | |
1340 | EXPORT_SYMBOL(sync_inode); |