]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * fs/fs-writeback.c | |
3 | * | |
4 | * Copyright (C) 2002, Linus Torvalds. | |
5 | * | |
6 | * Contains all the functions related to writing back and waiting | |
7 | * upon dirty inodes against superblocks, and writing back dirty | |
8 | * pages against inodes. ie: data writeback. Writeout of the | |
9 | * inode itself is not handled here. | |
10 | * | |
e1f8e874 | 11 | * 10Apr2002 Andrew Morton |
1da177e4 LT |
12 | * Split out of fs/inode.c |
13 | * Additions for address_space-based writeback | |
14 | */ | |
15 | ||
16 | #include <linux/kernel.h> | |
f5ff8422 | 17 | #include <linux/module.h> |
1da177e4 | 18 | #include <linux/spinlock.h> |
5a0e3ad6 | 19 | #include <linux/slab.h> |
1da177e4 LT |
20 | #include <linux/sched.h> |
21 | #include <linux/fs.h> | |
22 | #include <linux/mm.h> | |
03ba3782 JA |
23 | #include <linux/kthread.h> |
24 | #include <linux/freezer.h> | |
1da177e4 LT |
25 | #include <linux/writeback.h> |
26 | #include <linux/blkdev.h> | |
27 | #include <linux/backing-dev.h> | |
28 | #include <linux/buffer_head.h> | |
455b2864 | 29 | #include <linux/tracepoint.h> |
07f3f05c | 30 | #include "internal.h" |
1da177e4 | 31 | |
c4a77a6c JA |
32 | /* |
33 | * Passed into wb_writeback(), essentially a subset of writeback_control | |
34 | */ | |
83ba7b07 | 35 | struct wb_writeback_work { |
c4a77a6c JA |
36 | long nr_pages; |
37 | struct super_block *sb; | |
38 | enum writeback_sync_modes sync_mode; | |
52957fe1 HS |
39 | unsigned int for_kupdate:1; |
40 | unsigned int range_cyclic:1; | |
41 | unsigned int for_background:1; | |
c4a77a6c | 42 | |
8010c3b6 | 43 | struct list_head list; /* pending work list */ |
83ba7b07 | 44 | struct completion *done; /* set if the caller waits */ |
03ba3782 JA |
45 | }; |
46 | ||
455b2864 DC |
47 | /* |
48 | * Include the creation of the trace points after defining the | |
49 | * wb_writeback_work structure so that the definition remains local to this | |
50 | * file. | |
51 | */ | |
52 | #define CREATE_TRACE_POINTS | |
53 | #include <trace/events/writeback.h> | |
54 | ||
455b2864 DC |
55 | /* |
56 | * We don't actually have pdflush, but this one is exported though /proc... | |
57 | */ | |
58 | int nr_pdflush_threads; | |
59 | ||
f11b00f3 AB |
60 | /** |
61 | * writeback_in_progress - determine whether there is writeback in progress | |
62 | * @bdi: the device's backing_dev_info structure. | |
63 | * | |
03ba3782 JA |
64 | * Determine whether there is writeback waiting to be handled against a |
65 | * backing device. | |
f11b00f3 AB |
66 | */ |
67 | int writeback_in_progress(struct backing_dev_info *bdi) | |
68 | { | |
81d73a32 | 69 | return test_bit(BDI_writeback_running, &bdi->state); |
f11b00f3 AB |
70 | } |
71 | ||
692ebd17 JK |
72 | static inline struct backing_dev_info *inode_to_bdi(struct inode *inode) |
73 | { | |
74 | struct super_block *sb = inode->i_sb; | |
692ebd17 | 75 | |
aaead25b CH |
76 | if (strcmp(sb->s_type->name, "bdev") == 0) |
77 | return inode->i_mapping->backing_dev_info; | |
78 | ||
79 | return sb->s_bdi; | |
692ebd17 JK |
80 | } |
81 | ||
7ccf19a8 NP |
82 | static inline struct inode *wb_inode(struct list_head *head) |
83 | { | |
84 | return list_entry(head, struct inode, i_wb_list); | |
85 | } | |
86 | ||
6585027a JK |
87 | /* Wakeup flusher thread or forker thread to fork it. Requires bdi->wb_lock. */ |
88 | static void bdi_wakeup_flusher(struct backing_dev_info *bdi) | |
03ba3782 | 89 | { |
fff5b85a AB |
90 | if (bdi->wb.task) { |
91 | wake_up_process(bdi->wb.task); | |
92 | } else { | |
93 | /* | |
94 | * The bdi thread isn't there, wake up the forker thread which | |
95 | * will create and run it. | |
96 | */ | |
03ba3782 | 97 | wake_up_process(default_backing_dev_info.wb.task); |
1da177e4 | 98 | } |
6585027a JK |
99 | } |
100 | ||
101 | static void bdi_queue_work(struct backing_dev_info *bdi, | |
102 | struct wb_writeback_work *work) | |
103 | { | |
104 | trace_writeback_queue(bdi, work); | |
105 | ||
106 | spin_lock_bh(&bdi->wb_lock); | |
107 | list_add_tail(&work->list, &bdi->work_list); | |
108 | if (!bdi->wb.task) | |
109 | trace_writeback_nothread(bdi, work); | |
110 | bdi_wakeup_flusher(bdi); | |
6467716a | 111 | spin_unlock_bh(&bdi->wb_lock); |
1da177e4 LT |
112 | } |
113 | ||
83ba7b07 CH |
114 | static void |
115 | __bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages, | |
6585027a | 116 | bool range_cyclic) |
1da177e4 | 117 | { |
83ba7b07 | 118 | struct wb_writeback_work *work; |
03ba3782 | 119 | |
bcddc3f0 JA |
120 | /* |
121 | * This is WB_SYNC_NONE writeback, so if allocation fails just | |
122 | * wakeup the thread for old dirty data writeback | |
123 | */ | |
83ba7b07 CH |
124 | work = kzalloc(sizeof(*work), GFP_ATOMIC); |
125 | if (!work) { | |
455b2864 DC |
126 | if (bdi->wb.task) { |
127 | trace_writeback_nowork(bdi); | |
83ba7b07 | 128 | wake_up_process(bdi->wb.task); |
455b2864 | 129 | } |
83ba7b07 | 130 | return; |
bcddc3f0 | 131 | } |
03ba3782 | 132 | |
83ba7b07 CH |
133 | work->sync_mode = WB_SYNC_NONE; |
134 | work->nr_pages = nr_pages; | |
135 | work->range_cyclic = range_cyclic; | |
03ba3782 | 136 | |
83ba7b07 | 137 | bdi_queue_work(bdi, work); |
b6e51316 JA |
138 | } |
139 | ||
140 | /** | |
141 | * bdi_start_writeback - start writeback | |
142 | * @bdi: the backing device to write from | |
143 | * @nr_pages: the number of pages to write | |
144 | * | |
145 | * Description: | |
146 | * This does WB_SYNC_NONE opportunistic writeback. The IO is only | |
147 | * started when this function returns, we make no guarentees on | |
0e3c9a22 | 148 | * completion. Caller need not hold sb s_umount semaphore. |
b6e51316 JA |
149 | * |
150 | */ | |
c5444198 | 151 | void bdi_start_writeback(struct backing_dev_info *bdi, long nr_pages) |
b6e51316 | 152 | { |
6585027a | 153 | __bdi_start_writeback(bdi, nr_pages, true); |
c5444198 | 154 | } |
d3ddec76 | 155 | |
c5444198 CH |
156 | /** |
157 | * bdi_start_background_writeback - start background writeback | |
158 | * @bdi: the backing device to write from | |
159 | * | |
160 | * Description: | |
6585027a JK |
161 | * This makes sure WB_SYNC_NONE background writeback happens. When |
162 | * this function returns, it is only guaranteed that for given BDI | |
163 | * some IO is happening if we are over background dirty threshold. | |
164 | * Caller need not hold sb s_umount semaphore. | |
c5444198 CH |
165 | */ |
166 | void bdi_start_background_writeback(struct backing_dev_info *bdi) | |
167 | { | |
6585027a JK |
168 | /* |
169 | * We just wake up the flusher thread. It will perform background | |
170 | * writeback as soon as there is no other work to do. | |
171 | */ | |
71927e84 | 172 | trace_writeback_wake_background(bdi); |
6585027a JK |
173 | spin_lock_bh(&bdi->wb_lock); |
174 | bdi_wakeup_flusher(bdi); | |
175 | spin_unlock_bh(&bdi->wb_lock); | |
1da177e4 LT |
176 | } |
177 | ||
6610a0bc AM |
178 | /* |
179 | * Redirty an inode: set its when-it-was dirtied timestamp and move it to the | |
180 | * furthest end of its superblock's dirty-inode list. | |
181 | * | |
182 | * Before stamping the inode's ->dirtied_when, we check to see whether it is | |
66f3b8e2 | 183 | * already the most-recently-dirtied inode on the b_dirty list. If that is |
6610a0bc AM |
184 | * the case then the inode must have been redirtied while it was being written |
185 | * out and we don't reset its dirtied_when. | |
186 | */ | |
187 | static void redirty_tail(struct inode *inode) | |
188 | { | |
03ba3782 | 189 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
6610a0bc | 190 | |
03ba3782 | 191 | if (!list_empty(&wb->b_dirty)) { |
66f3b8e2 | 192 | struct inode *tail; |
6610a0bc | 193 | |
7ccf19a8 | 194 | tail = wb_inode(wb->b_dirty.next); |
66f3b8e2 | 195 | if (time_before(inode->dirtied_when, tail->dirtied_when)) |
6610a0bc AM |
196 | inode->dirtied_when = jiffies; |
197 | } | |
7ccf19a8 | 198 | list_move(&inode->i_wb_list, &wb->b_dirty); |
6610a0bc AM |
199 | } |
200 | ||
c986d1e2 | 201 | /* |
66f3b8e2 | 202 | * requeue inode for re-scanning after bdi->b_io list is exhausted. |
c986d1e2 | 203 | */ |
0e0f4fc2 | 204 | static void requeue_io(struct inode *inode) |
c986d1e2 | 205 | { |
03ba3782 JA |
206 | struct bdi_writeback *wb = &inode_to_bdi(inode)->wb; |
207 | ||
7ccf19a8 | 208 | list_move(&inode->i_wb_list, &wb->b_more_io); |
c986d1e2 AM |
209 | } |
210 | ||
1c0eeaf5 JE |
211 | static void inode_sync_complete(struct inode *inode) |
212 | { | |
213 | /* | |
214 | * Prevent speculative execution through spin_unlock(&inode_lock); | |
215 | */ | |
216 | smp_mb(); | |
217 | wake_up_bit(&inode->i_state, __I_SYNC); | |
218 | } | |
219 | ||
d2caa3c5 JL |
220 | static bool inode_dirtied_after(struct inode *inode, unsigned long t) |
221 | { | |
222 | bool ret = time_after(inode->dirtied_when, t); | |
223 | #ifndef CONFIG_64BIT | |
224 | /* | |
225 | * For inodes being constantly redirtied, dirtied_when can get stuck. | |
226 | * It _appears_ to be in the future, but is actually in distant past. | |
227 | * This test is necessary to prevent such wrapped-around relative times | |
5b0830cb | 228 | * from permanently stopping the whole bdi writeback. |
d2caa3c5 JL |
229 | */ |
230 | ret = ret && time_before_eq(inode->dirtied_when, jiffies); | |
231 | #endif | |
232 | return ret; | |
233 | } | |
234 | ||
2c136579 FW |
235 | /* |
236 | * Move expired dirty inodes from @delaying_queue to @dispatch_queue. | |
237 | */ | |
238 | static void move_expired_inodes(struct list_head *delaying_queue, | |
239 | struct list_head *dispatch_queue, | |
240 | unsigned long *older_than_this) | |
241 | { | |
5c03449d SL |
242 | LIST_HEAD(tmp); |
243 | struct list_head *pos, *node; | |
cf137307 | 244 | struct super_block *sb = NULL; |
5c03449d | 245 | struct inode *inode; |
cf137307 | 246 | int do_sb_sort = 0; |
5c03449d | 247 | |
2c136579 | 248 | while (!list_empty(delaying_queue)) { |
7ccf19a8 | 249 | inode = wb_inode(delaying_queue->prev); |
2c136579 | 250 | if (older_than_this && |
d2caa3c5 | 251 | inode_dirtied_after(inode, *older_than_this)) |
2c136579 | 252 | break; |
cf137307 JA |
253 | if (sb && sb != inode->i_sb) |
254 | do_sb_sort = 1; | |
255 | sb = inode->i_sb; | |
7ccf19a8 | 256 | list_move(&inode->i_wb_list, &tmp); |
5c03449d SL |
257 | } |
258 | ||
cf137307 JA |
259 | /* just one sb in list, splice to dispatch_queue and we're done */ |
260 | if (!do_sb_sort) { | |
261 | list_splice(&tmp, dispatch_queue); | |
262 | return; | |
263 | } | |
264 | ||
5c03449d SL |
265 | /* Move inodes from one superblock together */ |
266 | while (!list_empty(&tmp)) { | |
7ccf19a8 | 267 | sb = wb_inode(tmp.prev)->i_sb; |
5c03449d | 268 | list_for_each_prev_safe(pos, node, &tmp) { |
7ccf19a8 | 269 | inode = wb_inode(pos); |
5c03449d | 270 | if (inode->i_sb == sb) |
7ccf19a8 | 271 | list_move(&inode->i_wb_list, dispatch_queue); |
5c03449d | 272 | } |
2c136579 FW |
273 | } |
274 | } | |
275 | ||
276 | /* | |
277 | * Queue all expired dirty inodes for io, eldest first. | |
4ea879b9 WF |
278 | * Before |
279 | * newly dirtied b_dirty b_io b_more_io | |
280 | * =============> gf edc BA | |
281 | * After | |
282 | * newly dirtied b_dirty b_io b_more_io | |
283 | * =============> g fBAedc | |
284 | * | | |
285 | * +--> dequeue for IO | |
2c136579 | 286 | */ |
03ba3782 | 287 | static void queue_io(struct bdi_writeback *wb, unsigned long *older_than_this) |
66f3b8e2 | 288 | { |
4ea879b9 | 289 | list_splice_init(&wb->b_more_io, &wb->b_io); |
03ba3782 | 290 | move_expired_inodes(&wb->b_dirty, &wb->b_io, older_than_this); |
66f3b8e2 JA |
291 | } |
292 | ||
a9185b41 | 293 | static int write_inode(struct inode *inode, struct writeback_control *wbc) |
08d8e974 | 294 | { |
03ba3782 | 295 | if (inode->i_sb->s_op->write_inode && !is_bad_inode(inode)) |
a9185b41 | 296 | return inode->i_sb->s_op->write_inode(inode, wbc); |
03ba3782 | 297 | return 0; |
08d8e974 | 298 | } |
08d8e974 | 299 | |
1da177e4 | 300 | /* |
01c03194 CH |
301 | * Wait for writeback on an inode to complete. |
302 | */ | |
303 | static void inode_wait_for_writeback(struct inode *inode) | |
304 | { | |
305 | DEFINE_WAIT_BIT(wq, &inode->i_state, __I_SYNC); | |
306 | wait_queue_head_t *wqh; | |
307 | ||
308 | wqh = bit_waitqueue(&inode->i_state, __I_SYNC); | |
58a9d3d8 | 309 | while (inode->i_state & I_SYNC) { |
01c03194 CH |
310 | spin_unlock(&inode_lock); |
311 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); | |
312 | spin_lock(&inode_lock); | |
58a9d3d8 | 313 | } |
01c03194 CH |
314 | } |
315 | ||
316 | /* | |
317 | * Write out an inode's dirty pages. Called under inode_lock. Either the | |
318 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
319 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
320 | * | |
1da177e4 LT |
321 | * If `wait' is set, wait on the writeout. |
322 | * | |
323 | * The whole writeout design is quite complex and fragile. We want to avoid | |
324 | * starvation of particular inodes when others are being redirtied, prevent | |
325 | * livelocks, etc. | |
326 | * | |
327 | * Called under inode_lock. | |
328 | */ | |
329 | static int | |
01c03194 | 330 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 331 | { |
1da177e4 | 332 | struct address_space *mapping = inode->i_mapping; |
01c03194 | 333 | unsigned dirty; |
1da177e4 LT |
334 | int ret; |
335 | ||
01c03194 CH |
336 | if (!atomic_read(&inode->i_count)) |
337 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
338 | else | |
339 | WARN_ON(inode->i_state & I_WILL_FREE); | |
340 | ||
341 | if (inode->i_state & I_SYNC) { | |
342 | /* | |
343 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 344 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
345 | * writeback can proceed with the other inodes on s_io. |
346 | * | |
347 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 348 | * completed a full scan of b_io. |
01c03194 | 349 | */ |
a9185b41 | 350 | if (wbc->sync_mode != WB_SYNC_ALL) { |
01c03194 CH |
351 | requeue_io(inode); |
352 | return 0; | |
353 | } | |
354 | ||
355 | /* | |
356 | * It's a data-integrity sync. We must wait. | |
357 | */ | |
358 | inode_wait_for_writeback(inode); | |
359 | } | |
360 | ||
1c0eeaf5 | 361 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 362 | |
5547e8aa | 363 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 364 | inode->i_state |= I_SYNC; |
5547e8aa | 365 | inode->i_state &= ~I_DIRTY_PAGES; |
1da177e4 LT |
366 | spin_unlock(&inode_lock); |
367 | ||
368 | ret = do_writepages(mapping, wbc); | |
369 | ||
26821ed4 CH |
370 | /* |
371 | * Make sure to wait on the data before writing out the metadata. | |
372 | * This is important for filesystems that modify metadata on data | |
373 | * I/O completion. | |
374 | */ | |
a9185b41 | 375 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 376 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
377 | if (ret == 0) |
378 | ret = err; | |
379 | } | |
380 | ||
5547e8aa DM |
381 | /* |
382 | * Some filesystems may redirty the inode during the writeback | |
383 | * due to delalloc, clear dirty metadata flags right before | |
384 | * write_inode() | |
385 | */ | |
386 | spin_lock(&inode_lock); | |
387 | dirty = inode->i_state & I_DIRTY; | |
388 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
389 | spin_unlock(&inode_lock); | |
26821ed4 CH |
390 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
391 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 392 | int err = write_inode(inode, wbc); |
1da177e4 LT |
393 | if (ret == 0) |
394 | ret = err; | |
395 | } | |
396 | ||
397 | spin_lock(&inode_lock); | |
1c0eeaf5 | 398 | inode->i_state &= ~I_SYNC; |
a4ffdde6 | 399 | if (!(inode->i_state & I_FREEING)) { |
23539afc | 400 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
1da177e4 LT |
401 | /* |
402 | * We didn't write back all the pages. nfs_writepages() | |
a50aeb40 | 403 | * sometimes bales out without doing anything. |
1b43ef91 | 404 | */ |
a50aeb40 WF |
405 | inode->i_state |= I_DIRTY_PAGES; |
406 | if (wbc->nr_to_write <= 0) { | |
1da177e4 | 407 | /* |
a50aeb40 | 408 | * slice used up: queue for next turn |
1da177e4 | 409 | */ |
a50aeb40 | 410 | requeue_io(inode); |
1da177e4 LT |
411 | } else { |
412 | /* | |
a50aeb40 WF |
413 | * Writeback blocked by something other than |
414 | * congestion. Delay the inode for some time to | |
415 | * avoid spinning on the CPU (100% iowait) | |
416 | * retrying writeback of the dirty page/inode | |
417 | * that cannot be performed immediately. | |
1da177e4 | 418 | */ |
1b43ef91 | 419 | redirty_tail(inode); |
1da177e4 | 420 | } |
23539afc WF |
421 | } else if (inode->i_state & I_DIRTY) { |
422 | /* | |
423 | * Filesystems can dirty the inode during writeback | |
424 | * operations, such as delayed allocation during | |
425 | * submission or metadata updates after data IO | |
426 | * completion. | |
427 | */ | |
428 | redirty_tail(inode); | |
1da177e4 LT |
429 | } else { |
430 | /* | |
9e38d86f NP |
431 | * The inode is clean. At this point we either have |
432 | * a reference to the inode or it's on it's way out. | |
433 | * No need to add it back to the LRU. | |
1da177e4 | 434 | */ |
7ccf19a8 | 435 | list_del_init(&inode->i_wb_list); |
1da177e4 LT |
436 | } |
437 | } | |
1c0eeaf5 | 438 | inode_sync_complete(inode); |
1da177e4 LT |
439 | return ret; |
440 | } | |
441 | ||
03ba3782 | 442 | /* |
d19de7ed | 443 | * For background writeback the caller does not have the sb pinned |
03ba3782 JA |
444 | * before calling writeback. So make sure that we do pin it, so it doesn't |
445 | * go away while we are writing inodes from it. | |
03ba3782 | 446 | */ |
d19de7ed | 447 | static bool pin_sb_for_writeback(struct super_block *sb) |
03ba3782 | 448 | { |
03ba3782 | 449 | spin_lock(&sb_lock); |
29cb4859 CH |
450 | if (list_empty(&sb->s_instances)) { |
451 | spin_unlock(&sb_lock); | |
452 | return false; | |
453 | } | |
454 | ||
03ba3782 | 455 | sb->s_count++; |
29cb4859 CH |
456 | spin_unlock(&sb_lock); |
457 | ||
03ba3782 | 458 | if (down_read_trylock(&sb->s_umount)) { |
29cb4859 | 459 | if (sb->s_root) |
d19de7ed | 460 | return true; |
03ba3782 JA |
461 | up_read(&sb->s_umount); |
462 | } | |
29cb4859 CH |
463 | |
464 | put_super(sb); | |
d19de7ed | 465 | return false; |
03ba3782 JA |
466 | } |
467 | ||
f11c9c5c ES |
468 | /* |
469 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 CH |
470 | * |
471 | * If @only_this_sb is true, then find and write all such | |
f11c9c5c ES |
472 | * inodes. Otherwise write only ones which go sequentially |
473 | * in reverse order. | |
edadfb10 | 474 | * |
f11c9c5c ES |
475 | * Return 1, if the caller writeback routine should be |
476 | * interrupted. Otherwise return 0. | |
477 | */ | |
edadfb10 CH |
478 | static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, |
479 | struct writeback_control *wbc, bool only_this_sb) | |
1da177e4 | 480 | { |
03ba3782 | 481 | while (!list_empty(&wb->b_io)) { |
1da177e4 | 482 | long pages_skipped; |
7ccf19a8 | 483 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
484 | |
485 | if (inode->i_sb != sb) { | |
486 | if (only_this_sb) { | |
487 | /* | |
488 | * We only want to write back data for this | |
489 | * superblock, move all inodes not belonging | |
490 | * to it back onto the dirty list. | |
491 | */ | |
492 | redirty_tail(inode); | |
493 | continue; | |
494 | } | |
495 | ||
496 | /* | |
497 | * The inode belongs to a different superblock. | |
498 | * Bounce back to the caller to unpin this and | |
499 | * pin the next superblock. | |
500 | */ | |
f11c9c5c | 501 | return 0; |
edadfb10 CH |
502 | } |
503 | ||
9843b76a CH |
504 | /* |
505 | * Don't bother with new inodes or inodes beeing freed, first | |
506 | * kind does not need peridic writeout yet, and for the latter | |
507 | * kind writeout is handled by the freer. | |
508 | */ | |
509 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { | |
7ef0d737 NP |
510 | requeue_io(inode); |
511 | continue; | |
512 | } | |
9843b76a | 513 | |
d2caa3c5 JL |
514 | /* |
515 | * Was this inode dirtied after sync_sb_inodes was called? | |
516 | * This keeps sync from extra jobs and livelock. | |
517 | */ | |
f11c9c5c ES |
518 | if (inode_dirtied_after(inode, wbc->wb_start)) |
519 | return 1; | |
1da177e4 | 520 | |
1da177e4 LT |
521 | __iget(inode); |
522 | pages_skipped = wbc->pages_skipped; | |
01c03194 | 523 | writeback_single_inode(inode, wbc); |
1da177e4 LT |
524 | if (wbc->pages_skipped != pages_skipped) { |
525 | /* | |
526 | * writeback is not making progress due to locked | |
527 | * buffers. Skip this inode for now. | |
528 | */ | |
f57b9b7b | 529 | redirty_tail(inode); |
1da177e4 LT |
530 | } |
531 | spin_unlock(&inode_lock); | |
1da177e4 | 532 | iput(inode); |
4ffc8444 | 533 | cond_resched(); |
1da177e4 | 534 | spin_lock(&inode_lock); |
8bc3be27 FW |
535 | if (wbc->nr_to_write <= 0) { |
536 | wbc->more_io = 1; | |
f11c9c5c | 537 | return 1; |
8bc3be27 | 538 | } |
03ba3782 | 539 | if (!list_empty(&wb->b_more_io)) |
8bc3be27 | 540 | wbc->more_io = 1; |
1da177e4 | 541 | } |
f11c9c5c ES |
542 | /* b_io is empty */ |
543 | return 1; | |
544 | } | |
545 | ||
9c3a8ee8 CH |
546 | void writeback_inodes_wb(struct bdi_writeback *wb, |
547 | struct writeback_control *wbc) | |
f11c9c5c ES |
548 | { |
549 | int ret = 0; | |
550 | ||
7624ee72 JK |
551 | if (!wbc->wb_start) |
552 | wbc->wb_start = jiffies; /* livelock avoidance */ | |
f11c9c5c ES |
553 | spin_lock(&inode_lock); |
554 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) | |
555 | queue_io(wb, wbc->older_than_this); | |
38f21977 | 556 | |
f11c9c5c | 557 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 558 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 559 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 560 | |
edadfb10 CH |
561 | if (!pin_sb_for_writeback(sb)) { |
562 | requeue_io(inode); | |
563 | continue; | |
f11c9c5c | 564 | } |
edadfb10 CH |
565 | ret = writeback_sb_inodes(sb, wb, wbc, false); |
566 | drop_super(sb); | |
f11c9c5c | 567 | |
f11c9c5c ES |
568 | if (ret) |
569 | break; | |
570 | } | |
66f3b8e2 JA |
571 | spin_unlock(&inode_lock); |
572 | /* Leave any unwritten inodes on b_io */ | |
573 | } | |
574 | ||
edadfb10 CH |
575 | static void __writeback_inodes_sb(struct super_block *sb, |
576 | struct bdi_writeback *wb, struct writeback_control *wbc) | |
577 | { | |
578 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
579 | ||
edadfb10 CH |
580 | spin_lock(&inode_lock); |
581 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) | |
582 | queue_io(wb, wbc->older_than_this); | |
583 | writeback_sb_inodes(sb, wb, wbc, true); | |
584 | spin_unlock(&inode_lock); | |
585 | } | |
586 | ||
66f3b8e2 | 587 | /* |
03ba3782 JA |
588 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
589 | * operation. We do this so we don't hold I_SYNC against an inode for | |
590 | * enormous amounts of time, which would block a userspace task which has | |
591 | * been forced to throttle against that inode. Also, the code reevaluates | |
592 | * the dirty each time it has written this many pages. | |
593 | */ | |
594 | #define MAX_WRITEBACK_PAGES 1024 | |
595 | ||
596 | static inline bool over_bground_thresh(void) | |
597 | { | |
598 | unsigned long background_thresh, dirty_thresh; | |
599 | ||
16c4042f | 600 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 JA |
601 | |
602 | return (global_page_state(NR_FILE_DIRTY) + | |
4cbec4c8 | 603 | global_page_state(NR_UNSTABLE_NFS) > background_thresh); |
03ba3782 JA |
604 | } |
605 | ||
606 | /* | |
607 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 608 | * |
03ba3782 JA |
609 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
610 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
611 | * just walks the superblock inode list, writing back any inodes which are | |
612 | * older than a specific point in time. | |
66f3b8e2 | 613 | * |
03ba3782 JA |
614 | * Try to run once per dirty_writeback_interval. But if a writeback event |
615 | * takes longer than a dirty_writeback_interval interval, then leave a | |
616 | * one-second gap. | |
66f3b8e2 | 617 | * |
03ba3782 JA |
618 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
619 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 620 | */ |
c4a77a6c | 621 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 622 | struct wb_writeback_work *work) |
66f3b8e2 | 623 | { |
03ba3782 | 624 | struct writeback_control wbc = { |
83ba7b07 | 625 | .sync_mode = work->sync_mode, |
03ba3782 | 626 | .older_than_this = NULL, |
83ba7b07 CH |
627 | .for_kupdate = work->for_kupdate, |
628 | .for_background = work->for_background, | |
629 | .range_cyclic = work->range_cyclic, | |
03ba3782 JA |
630 | }; |
631 | unsigned long oldest_jif; | |
632 | long wrote = 0; | |
a5989bdc | 633 | struct inode *inode; |
66f3b8e2 | 634 | |
03ba3782 JA |
635 | if (wbc.for_kupdate) { |
636 | wbc.older_than_this = &oldest_jif; | |
637 | oldest_jif = jiffies - | |
638 | msecs_to_jiffies(dirty_expire_interval * 10); | |
639 | } | |
c4a77a6c JA |
640 | if (!wbc.range_cyclic) { |
641 | wbc.range_start = 0; | |
642 | wbc.range_end = LLONG_MAX; | |
643 | } | |
38f21977 | 644 | |
7624ee72 | 645 | wbc.wb_start = jiffies; /* livelock avoidance */ |
03ba3782 JA |
646 | for (;;) { |
647 | /* | |
d3ddec76 | 648 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 649 | */ |
83ba7b07 | 650 | if (work->nr_pages <= 0) |
03ba3782 | 651 | break; |
66f3b8e2 | 652 | |
38f21977 | 653 | /* |
d3ddec76 WF |
654 | * For background writeout, stop when we are below the |
655 | * background dirty threshold | |
38f21977 | 656 | */ |
83ba7b07 | 657 | if (work->for_background && !over_bground_thresh()) |
03ba3782 | 658 | break; |
38f21977 | 659 | |
03ba3782 | 660 | wbc.more_io = 0; |
03ba3782 JA |
661 | wbc.nr_to_write = MAX_WRITEBACK_PAGES; |
662 | wbc.pages_skipped = 0; | |
028c2dd1 DC |
663 | |
664 | trace_wbc_writeback_start(&wbc, wb->bdi); | |
83ba7b07 CH |
665 | if (work->sb) |
666 | __writeback_inodes_sb(work->sb, wb, &wbc); | |
edadfb10 CH |
667 | else |
668 | writeback_inodes_wb(wb, &wbc); | |
028c2dd1 DC |
669 | trace_wbc_writeback_written(&wbc, wb->bdi); |
670 | ||
83ba7b07 | 671 | work->nr_pages -= MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
03ba3782 JA |
672 | wrote += MAX_WRITEBACK_PAGES - wbc.nr_to_write; |
673 | ||
674 | /* | |
71fd05a8 | 675 | * If we consumed everything, see if we have more |
03ba3782 | 676 | */ |
71fd05a8 JA |
677 | if (wbc.nr_to_write <= 0) |
678 | continue; | |
679 | /* | |
680 | * Didn't write everything and we don't have more IO, bail | |
681 | */ | |
682 | if (!wbc.more_io) | |
03ba3782 | 683 | break; |
71fd05a8 JA |
684 | /* |
685 | * Did we write something? Try for more | |
686 | */ | |
687 | if (wbc.nr_to_write < MAX_WRITEBACK_PAGES) | |
688 | continue; | |
689 | /* | |
690 | * Nothing written. Wait for some inode to | |
691 | * become available for writeback. Otherwise | |
692 | * we'll just busyloop. | |
693 | */ | |
694 | spin_lock(&inode_lock); | |
695 | if (!list_empty(&wb->b_more_io)) { | |
7ccf19a8 | 696 | inode = wb_inode(wb->b_more_io.prev); |
028c2dd1 | 697 | trace_wbc_writeback_wait(&wbc, wb->bdi); |
71fd05a8 | 698 | inode_wait_for_writeback(inode); |
03ba3782 | 699 | } |
71fd05a8 | 700 | spin_unlock(&inode_lock); |
03ba3782 JA |
701 | } |
702 | ||
703 | return wrote; | |
704 | } | |
705 | ||
706 | /* | |
83ba7b07 | 707 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 708 | */ |
83ba7b07 | 709 | static struct wb_writeback_work * |
08852b6d | 710 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 711 | { |
83ba7b07 | 712 | struct wb_writeback_work *work = NULL; |
03ba3782 | 713 | |
6467716a | 714 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 CH |
715 | if (!list_empty(&bdi->work_list)) { |
716 | work = list_entry(bdi->work_list.next, | |
717 | struct wb_writeback_work, list); | |
718 | list_del_init(&work->list); | |
03ba3782 | 719 | } |
6467716a | 720 | spin_unlock_bh(&bdi->wb_lock); |
83ba7b07 | 721 | return work; |
03ba3782 JA |
722 | } |
723 | ||
cdf01dd5 LT |
724 | /* |
725 | * Add in the number of potentially dirty inodes, because each inode | |
726 | * write can dirty pagecache in the underlying blockdev. | |
727 | */ | |
728 | static unsigned long get_nr_dirty_pages(void) | |
729 | { | |
730 | return global_page_state(NR_FILE_DIRTY) + | |
731 | global_page_state(NR_UNSTABLE_NFS) + | |
732 | get_nr_dirty_inodes(); | |
733 | } | |
734 | ||
6585027a JK |
735 | static long wb_check_background_flush(struct bdi_writeback *wb) |
736 | { | |
737 | if (over_bground_thresh()) { | |
738 | ||
739 | struct wb_writeback_work work = { | |
740 | .nr_pages = LONG_MAX, | |
741 | .sync_mode = WB_SYNC_NONE, | |
742 | .for_background = 1, | |
743 | .range_cyclic = 1, | |
744 | }; | |
745 | ||
746 | return wb_writeback(wb, &work); | |
747 | } | |
748 | ||
749 | return 0; | |
750 | } | |
751 | ||
03ba3782 JA |
752 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
753 | { | |
754 | unsigned long expired; | |
755 | long nr_pages; | |
756 | ||
69b62d01 JA |
757 | /* |
758 | * When set to zero, disable periodic writeback | |
759 | */ | |
760 | if (!dirty_writeback_interval) | |
761 | return 0; | |
762 | ||
03ba3782 JA |
763 | expired = wb->last_old_flush + |
764 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
765 | if (time_before(jiffies, expired)) | |
766 | return 0; | |
767 | ||
768 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 769 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 770 | |
c4a77a6c | 771 | if (nr_pages) { |
83ba7b07 | 772 | struct wb_writeback_work work = { |
c4a77a6c JA |
773 | .nr_pages = nr_pages, |
774 | .sync_mode = WB_SYNC_NONE, | |
775 | .for_kupdate = 1, | |
776 | .range_cyclic = 1, | |
777 | }; | |
778 | ||
83ba7b07 | 779 | return wb_writeback(wb, &work); |
c4a77a6c | 780 | } |
03ba3782 JA |
781 | |
782 | return 0; | |
783 | } | |
784 | ||
785 | /* | |
786 | * Retrieve work items and do the writeback they describe | |
787 | */ | |
788 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
789 | { | |
790 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 791 | struct wb_writeback_work *work; |
c4a77a6c | 792 | long wrote = 0; |
03ba3782 | 793 | |
81d73a32 | 794 | set_bit(BDI_writeback_running, &wb->bdi->state); |
08852b6d | 795 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 JA |
796 | /* |
797 | * Override sync mode, in case we must wait for completion | |
83ba7b07 | 798 | * because this thread is exiting now. |
03ba3782 JA |
799 | */ |
800 | if (force_wait) | |
83ba7b07 | 801 | work->sync_mode = WB_SYNC_ALL; |
03ba3782 | 802 | |
455b2864 DC |
803 | trace_writeback_exec(bdi, work); |
804 | ||
83ba7b07 | 805 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
806 | |
807 | /* | |
83ba7b07 CH |
808 | * Notify the caller of completion if this is a synchronous |
809 | * work item, otherwise just free it. | |
03ba3782 | 810 | */ |
83ba7b07 CH |
811 | if (work->done) |
812 | complete(work->done); | |
813 | else | |
814 | kfree(work); | |
03ba3782 JA |
815 | } |
816 | ||
817 | /* | |
818 | * Check for periodic writeback, kupdated() style | |
819 | */ | |
820 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 821 | wrote += wb_check_background_flush(wb); |
81d73a32 | 822 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
03ba3782 JA |
823 | |
824 | return wrote; | |
825 | } | |
826 | ||
827 | /* | |
828 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
829 | * wakes up periodically and does kupdated style flushing. | |
830 | */ | |
08243900 | 831 | int bdi_writeback_thread(void *data) |
03ba3782 | 832 | { |
08243900 CH |
833 | struct bdi_writeback *wb = data; |
834 | struct backing_dev_info *bdi = wb->bdi; | |
03ba3782 JA |
835 | long pages_written; |
836 | ||
766f9164 | 837 | current->flags |= PF_SWAPWRITE; |
08243900 | 838 | set_freezable(); |
ecd58403 | 839 | wb->last_active = jiffies; |
08243900 CH |
840 | |
841 | /* | |
842 | * Our parent may run at a different priority, just set us to normal | |
843 | */ | |
844 | set_user_nice(current, 0); | |
845 | ||
455b2864 DC |
846 | trace_writeback_thread_start(bdi); |
847 | ||
03ba3782 | 848 | while (!kthread_should_stop()) { |
6467716a AB |
849 | /* |
850 | * Remove own delayed wake-up timer, since we are already awake | |
851 | * and we'll take care of the preriodic write-back. | |
852 | */ | |
853 | del_timer(&wb->wakeup_timer); | |
854 | ||
03ba3782 JA |
855 | pages_written = wb_do_writeback(wb, 0); |
856 | ||
455b2864 DC |
857 | trace_writeback_pages_written(pages_written); |
858 | ||
03ba3782 | 859 | if (pages_written) |
ecd58403 | 860 | wb->last_active = jiffies; |
03ba3782 | 861 | |
297252c8 | 862 | set_current_state(TASK_INTERRUPTIBLE); |
b76b4014 | 863 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
f9eadbbd | 864 | __set_current_state(TASK_RUNNING); |
297252c8 | 865 | continue; |
03ba3782 JA |
866 | } |
867 | ||
253c34e9 | 868 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
fff5b85a | 869 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
253c34e9 AB |
870 | else { |
871 | /* | |
872 | * We have nothing to do, so can go sleep without any | |
873 | * timeout and save power. When a work is queued or | |
874 | * something is made dirty - we will be woken up. | |
875 | */ | |
297252c8 | 876 | schedule(); |
f9eadbbd | 877 | } |
69b62d01 | 878 | |
03ba3782 JA |
879 | try_to_freeze(); |
880 | } | |
881 | ||
fff5b85a | 882 | /* Flush any work that raced with us exiting */ |
08243900 CH |
883 | if (!list_empty(&bdi->work_list)) |
884 | wb_do_writeback(wb, 1); | |
455b2864 DC |
885 | |
886 | trace_writeback_thread_stop(bdi); | |
03ba3782 JA |
887 | return 0; |
888 | } | |
889 | ||
08243900 | 890 | |
03ba3782 | 891 | /* |
b8c2f347 CH |
892 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
893 | * the whole world. | |
03ba3782 | 894 | */ |
b8c2f347 | 895 | void wakeup_flusher_threads(long nr_pages) |
03ba3782 | 896 | { |
b8c2f347 | 897 | struct backing_dev_info *bdi; |
03ba3782 | 898 | |
83ba7b07 CH |
899 | if (!nr_pages) { |
900 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
b8c2f347 CH |
901 | global_page_state(NR_UNSTABLE_NFS); |
902 | } | |
03ba3782 | 903 | |
b8c2f347 | 904 | rcu_read_lock(); |
cfc4ba53 | 905 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
906 | if (!bdi_has_dirty_io(bdi)) |
907 | continue; | |
6585027a | 908 | __bdi_start_writeback(bdi, nr_pages, false); |
03ba3782 | 909 | } |
cfc4ba53 | 910 | rcu_read_unlock(); |
1da177e4 LT |
911 | } |
912 | ||
03ba3782 JA |
913 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
914 | { | |
915 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
916 | struct dentry *dentry; | |
917 | const char *name = "?"; | |
918 | ||
919 | dentry = d_find_alias(inode); | |
920 | if (dentry) { | |
921 | spin_lock(&dentry->d_lock); | |
922 | name = (const char *) dentry->d_name.name; | |
923 | } | |
924 | printk(KERN_DEBUG | |
925 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
926 | current->comm, task_pid_nr(current), inode->i_ino, | |
927 | name, inode->i_sb->s_id); | |
928 | if (dentry) { | |
929 | spin_unlock(&dentry->d_lock); | |
930 | dput(dentry); | |
931 | } | |
932 | } | |
933 | } | |
934 | ||
935 | /** | |
936 | * __mark_inode_dirty - internal function | |
937 | * @inode: inode to mark | |
938 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
939 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
940 | * mark_inode_dirty_sync. | |
1da177e4 | 941 | * |
03ba3782 JA |
942 | * Put the inode on the super block's dirty list. |
943 | * | |
944 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
945 | * dirty list only if it is hashed or if it refers to a blockdev. | |
946 | * If it was not hashed, it will never be added to the dirty list | |
947 | * even if it is later hashed, as it will have been marked dirty already. | |
948 | * | |
949 | * In short, make sure you hash any inodes _before_ you start marking | |
950 | * them dirty. | |
1da177e4 | 951 | * |
03ba3782 JA |
952 | * This function *must* be atomic for the I_DIRTY_PAGES case - |
953 | * set_page_dirty() is called under spinlock in several places. | |
1da177e4 | 954 | * |
03ba3782 JA |
955 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
956 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
957 | * the kernel-internal blockdev inode represents the dirtying time of the | |
958 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
959 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
960 | * blockdev inode. | |
1da177e4 | 961 | */ |
03ba3782 | 962 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 963 | { |
03ba3782 | 964 | struct super_block *sb = inode->i_sb; |
253c34e9 AB |
965 | struct backing_dev_info *bdi = NULL; |
966 | bool wakeup_bdi = false; | |
1da177e4 | 967 | |
03ba3782 JA |
968 | /* |
969 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
970 | * dirty the inode itself | |
971 | */ | |
972 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
973 | if (sb->s_op->dirty_inode) | |
974 | sb->s_op->dirty_inode(inode); | |
975 | } | |
976 | ||
977 | /* | |
978 | * make sure that changes are seen by all cpus before we test i_state | |
979 | * -- mikulas | |
980 | */ | |
981 | smp_mb(); | |
982 | ||
983 | /* avoid the locking if we can */ | |
984 | if ((inode->i_state & flags) == flags) | |
985 | return; | |
986 | ||
987 | if (unlikely(block_dump)) | |
988 | block_dump___mark_inode_dirty(inode); | |
989 | ||
990 | spin_lock(&inode_lock); | |
991 | if ((inode->i_state & flags) != flags) { | |
992 | const int was_dirty = inode->i_state & I_DIRTY; | |
993 | ||
994 | inode->i_state |= flags; | |
995 | ||
996 | /* | |
997 | * If the inode is being synced, just update its dirty state. | |
998 | * The unlocker will place the inode on the appropriate | |
999 | * superblock list, based upon its state. | |
1000 | */ | |
1001 | if (inode->i_state & I_SYNC) | |
1002 | goto out; | |
1003 | ||
1004 | /* | |
1005 | * Only add valid (hashed) inodes to the superblock's | |
1006 | * dirty list. Add blockdev inodes as well. | |
1007 | */ | |
1008 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1009 | if (inode_unhashed(inode)) |
03ba3782 JA |
1010 | goto out; |
1011 | } | |
a4ffdde6 | 1012 | if (inode->i_state & I_FREEING) |
03ba3782 JA |
1013 | goto out; |
1014 | ||
1015 | /* | |
1016 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1017 | * reposition it (that would break b_dirty time-ordering). | |
1018 | */ | |
1019 | if (!was_dirty) { | |
253c34e9 AB |
1020 | bdi = inode_to_bdi(inode); |
1021 | ||
1022 | if (bdi_cap_writeback_dirty(bdi)) { | |
1023 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1024 | "bdi-%s not registered\n", bdi->name); | |
1025 | ||
1026 | /* | |
1027 | * If this is the first dirty inode for this | |
1028 | * bdi, we have to wake-up the corresponding | |
1029 | * bdi thread to make sure background | |
1030 | * write-back happens later. | |
1031 | */ | |
1032 | if (!wb_has_dirty_io(&bdi->wb)) | |
1033 | wakeup_bdi = true; | |
500b067c | 1034 | } |
03ba3782 JA |
1035 | |
1036 | inode->dirtied_when = jiffies; | |
7ccf19a8 | 1037 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
1da177e4 | 1038 | } |
1da177e4 | 1039 | } |
03ba3782 JA |
1040 | out: |
1041 | spin_unlock(&inode_lock); | |
253c34e9 AB |
1042 | |
1043 | if (wakeup_bdi) | |
6467716a | 1044 | bdi_wakeup_thread_delayed(bdi); |
03ba3782 JA |
1045 | } |
1046 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1047 | ||
1048 | /* | |
1049 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1050 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1051 | * | |
1052 | * If older_than_this is non-NULL, then only write out inodes which | |
1053 | * had their first dirtying at a time earlier than *older_than_this. | |
1054 | * | |
03ba3782 JA |
1055 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1056 | * This function assumes that the blockdev superblock's inodes are backed by | |
1057 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1058 | * assume that all inodes are backed by the same queue. | |
1059 | * | |
1060 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1061 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1062 | * on the writer throttling path, and we get decent balancing between many | |
1063 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1064 | */ | |
b6e51316 | 1065 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1066 | { |
1067 | struct inode *inode, *old_inode = NULL; | |
1068 | ||
1069 | /* | |
1070 | * We need to be protected against the filesystem going from | |
1071 | * r/o to r/w or vice versa. | |
1072 | */ | |
b6e51316 | 1073 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 JA |
1074 | |
1075 | spin_lock(&inode_lock); | |
1076 | ||
1077 | /* | |
1078 | * Data integrity sync. Must wait for all pages under writeback, | |
1079 | * because there may have been pages dirtied before our sync | |
1080 | * call, but which had writeout started before we write it out. | |
1081 | * In which case, the inode may not be on the dirty list, but | |
1082 | * we still have to wait for that writeout. | |
1083 | */ | |
b6e51316 | 1084 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
03ba3782 JA |
1085 | struct address_space *mapping; |
1086 | ||
a4ffdde6 | 1087 | if (inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) |
03ba3782 JA |
1088 | continue; |
1089 | mapping = inode->i_mapping; | |
1090 | if (mapping->nrpages == 0) | |
1091 | continue; | |
1092 | __iget(inode); | |
1093 | spin_unlock(&inode_lock); | |
1094 | /* | |
1095 | * We hold a reference to 'inode' so it couldn't have | |
1096 | * been removed from s_inodes list while we dropped the | |
1097 | * inode_lock. We cannot iput the inode now as we can | |
1098 | * be holding the last reference and we cannot iput it | |
1099 | * under inode_lock. So we keep the reference and iput | |
1100 | * it later. | |
1101 | */ | |
1102 | iput(old_inode); | |
1103 | old_inode = inode; | |
1104 | ||
1105 | filemap_fdatawait(mapping); | |
1106 | ||
1107 | cond_resched(); | |
1108 | ||
1109 | spin_lock(&inode_lock); | |
1110 | } | |
1111 | spin_unlock(&inode_lock); | |
1112 | iput(old_inode); | |
1da177e4 LT |
1113 | } |
1114 | ||
d8a8559c | 1115 | /** |
3259f8be | 1116 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1117 | * @sb: the superblock |
3259f8be | 1118 | * @nr: the number of pages to write |
1da177e4 | 1119 | * |
d8a8559c JA |
1120 | * Start writeback on some inodes on this super_block. No guarantees are made |
1121 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1122 | * for IO completion of submitted IO. |
1da177e4 | 1123 | */ |
3259f8be | 1124 | void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr) |
1da177e4 | 1125 | { |
83ba7b07 CH |
1126 | DECLARE_COMPLETION_ONSTACK(done); |
1127 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1128 | .sb = sb, |
1129 | .sync_mode = WB_SYNC_NONE, | |
83ba7b07 | 1130 | .done = &done, |
3259f8be | 1131 | .nr_pages = nr, |
3c4d7165 | 1132 | }; |
d8a8559c | 1133 | |
cf37e972 | 1134 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
83ba7b07 CH |
1135 | bdi_queue_work(sb->s_bdi, &work); |
1136 | wait_for_completion(&done); | |
e913fc82 | 1137 | } |
3259f8be CM |
1138 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1139 | ||
1140 | /** | |
1141 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1142 | * @sb: the superblock | |
1143 | * | |
1144 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1145 | * on how many (if any) will be written, and this function does not wait | |
1146 | * for IO completion of submitted IO. | |
1147 | */ | |
1148 | void writeback_inodes_sb(struct super_block *sb) | |
1149 | { | |
925d169f | 1150 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages()); |
3259f8be | 1151 | } |
0e3c9a22 | 1152 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1153 | |
17bd55d0 ES |
1154 | /** |
1155 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1156 | * @sb: the superblock | |
1157 | * | |
1158 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1159 | * Returns 1 if writeback was started, 0 if not. | |
1160 | */ | |
1161 | int writeback_inodes_sb_if_idle(struct super_block *sb) | |
1162 | { | |
1163 | if (!writeback_in_progress(sb->s_bdi)) { | |
cf37e972 | 1164 | down_read(&sb->s_umount); |
17bd55d0 | 1165 | writeback_inodes_sb(sb); |
cf37e972 | 1166 | up_read(&sb->s_umount); |
17bd55d0 ES |
1167 | return 1; |
1168 | } else | |
1169 | return 0; | |
1170 | } | |
1171 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1172 | ||
3259f8be CM |
1173 | /** |
1174 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1175 | * @sb: the superblock | |
1176 | * @nr: the number of pages to write | |
1177 | * | |
1178 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1179 | * Returns 1 if writeback was started, 0 if not. | |
1180 | */ | |
1181 | int writeback_inodes_sb_nr_if_idle(struct super_block *sb, | |
1182 | unsigned long nr) | |
1183 | { | |
1184 | if (!writeback_in_progress(sb->s_bdi)) { | |
1185 | down_read(&sb->s_umount); | |
1186 | writeback_inodes_sb_nr(sb, nr); | |
1187 | up_read(&sb->s_umount); | |
1188 | return 1; | |
1189 | } else | |
1190 | return 0; | |
1191 | } | |
1192 | EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); | |
1193 | ||
d8a8559c JA |
1194 | /** |
1195 | * sync_inodes_sb - sync sb inode pages | |
1196 | * @sb: the superblock | |
1197 | * | |
1198 | * This function writes and waits on any dirty inode belonging to this | |
1199 | * super_block. The number of pages synced is returned. | |
1200 | */ | |
b6e51316 | 1201 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1202 | { |
83ba7b07 CH |
1203 | DECLARE_COMPLETION_ONSTACK(done); |
1204 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1205 | .sb = sb, |
1206 | .sync_mode = WB_SYNC_ALL, | |
1207 | .nr_pages = LONG_MAX, | |
1208 | .range_cyclic = 0, | |
83ba7b07 | 1209 | .done = &done, |
3c4d7165 CH |
1210 | }; |
1211 | ||
cf37e972 CH |
1212 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1213 | ||
83ba7b07 CH |
1214 | bdi_queue_work(sb->s_bdi, &work); |
1215 | wait_for_completion(&done); | |
1216 | ||
b6e51316 | 1217 | wait_sb_inodes(sb); |
1da177e4 | 1218 | } |
d8a8559c | 1219 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1220 | |
1da177e4 | 1221 | /** |
7f04c26d AA |
1222 | * write_inode_now - write an inode to disk |
1223 | * @inode: inode to write to disk | |
1224 | * @sync: whether the write should be synchronous or not | |
1225 | * | |
1226 | * This function commits an inode to disk immediately if it is dirty. This is | |
1227 | * primarily needed by knfsd. | |
1da177e4 | 1228 | * |
7f04c26d | 1229 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1230 | */ |
1da177e4 LT |
1231 | int write_inode_now(struct inode *inode, int sync) |
1232 | { | |
1233 | int ret; | |
1234 | struct writeback_control wbc = { | |
1235 | .nr_to_write = LONG_MAX, | |
18914b18 | 1236 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1237 | .range_start = 0, |
1238 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1239 | }; |
1240 | ||
1241 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1242 | wbc.nr_to_write = 0; |
1da177e4 LT |
1243 | |
1244 | might_sleep(); | |
1245 | spin_lock(&inode_lock); | |
01c03194 | 1246 | ret = writeback_single_inode(inode, &wbc); |
1da177e4 LT |
1247 | spin_unlock(&inode_lock); |
1248 | if (sync) | |
1c0eeaf5 | 1249 | inode_sync_wait(inode); |
1da177e4 LT |
1250 | return ret; |
1251 | } | |
1252 | EXPORT_SYMBOL(write_inode_now); | |
1253 | ||
1254 | /** | |
1255 | * sync_inode - write an inode and its pages to disk. | |
1256 | * @inode: the inode to sync | |
1257 | * @wbc: controls the writeback mode | |
1258 | * | |
1259 | * sync_inode() will write an inode and its pages to disk. It will also | |
1260 | * correctly update the inode on its superblock's dirty inode lists and will | |
1261 | * update inode->i_state. | |
1262 | * | |
1263 | * The caller must have a ref on the inode. | |
1264 | */ | |
1265 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1266 | { | |
1267 | int ret; | |
1268 | ||
1269 | spin_lock(&inode_lock); | |
01c03194 | 1270 | ret = writeback_single_inode(inode, wbc); |
1da177e4 LT |
1271 | spin_unlock(&inode_lock); |
1272 | return ret; | |
1273 | } | |
1274 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1275 | |
1276 | /** | |
1277 | * sync_inode - write an inode to disk | |
1278 | * @inode: the inode to sync | |
1279 | * @wait: wait for I/O to complete. | |
1280 | * | |
1281 | * Write an inode to disk and adjust it's dirty state after completion. | |
1282 | * | |
1283 | * Note: only writes the actual inode, no associated data or other metadata. | |
1284 | */ | |
1285 | int sync_inode_metadata(struct inode *inode, int wait) | |
1286 | { | |
1287 | struct writeback_control wbc = { | |
1288 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1289 | .nr_to_write = 0, /* metadata-only */ | |
1290 | }; | |
1291 | ||
1292 | return sync_inode(inode, &wbc); | |
1293 | } | |
1294 | EXPORT_SYMBOL(sync_inode_metadata); |