]>
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); | |
250df6ed DC |
309 | while (inode->i_state & I_SYNC) { |
310 | spin_unlock(&inode->i_lock); | |
01c03194 CH |
311 | spin_unlock(&inode_lock); |
312 | __wait_on_bit(wqh, &wq, inode_wait, TASK_UNINTERRUPTIBLE); | |
313 | spin_lock(&inode_lock); | |
250df6ed | 314 | spin_lock(&inode->i_lock); |
58a9d3d8 | 315 | } |
01c03194 CH |
316 | } |
317 | ||
318 | /* | |
319 | * Write out an inode's dirty pages. Called under inode_lock. Either the | |
320 | * caller has ref on the inode (either via __iget or via syscall against an fd) | |
321 | * or the inode has I_WILL_FREE set (via generic_forget_inode) | |
322 | * | |
1da177e4 LT |
323 | * If `wait' is set, wait on the writeout. |
324 | * | |
325 | * The whole writeout design is quite complex and fragile. We want to avoid | |
326 | * starvation of particular inodes when others are being redirtied, prevent | |
327 | * livelocks, etc. | |
328 | * | |
329 | * Called under inode_lock. | |
330 | */ | |
331 | static int | |
01c03194 | 332 | writeback_single_inode(struct inode *inode, struct writeback_control *wbc) |
1da177e4 | 333 | { |
1da177e4 | 334 | struct address_space *mapping = inode->i_mapping; |
01c03194 | 335 | unsigned dirty; |
1da177e4 LT |
336 | int ret; |
337 | ||
250df6ed | 338 | spin_lock(&inode->i_lock); |
01c03194 CH |
339 | if (!atomic_read(&inode->i_count)) |
340 | WARN_ON(!(inode->i_state & (I_WILL_FREE|I_FREEING))); | |
341 | else | |
342 | WARN_ON(inode->i_state & I_WILL_FREE); | |
343 | ||
344 | if (inode->i_state & I_SYNC) { | |
345 | /* | |
346 | * If this inode is locked for writeback and we are not doing | |
66f3b8e2 | 347 | * writeback-for-data-integrity, move it to b_more_io so that |
01c03194 CH |
348 | * writeback can proceed with the other inodes on s_io. |
349 | * | |
350 | * We'll have another go at writing back this inode when we | |
66f3b8e2 | 351 | * completed a full scan of b_io. |
01c03194 | 352 | */ |
a9185b41 | 353 | if (wbc->sync_mode != WB_SYNC_ALL) { |
250df6ed | 354 | spin_unlock(&inode->i_lock); |
01c03194 CH |
355 | requeue_io(inode); |
356 | return 0; | |
357 | } | |
358 | ||
359 | /* | |
360 | * It's a data-integrity sync. We must wait. | |
361 | */ | |
362 | inode_wait_for_writeback(inode); | |
363 | } | |
364 | ||
1c0eeaf5 | 365 | BUG_ON(inode->i_state & I_SYNC); |
1da177e4 | 366 | |
5547e8aa | 367 | /* Set I_SYNC, reset I_DIRTY_PAGES */ |
1c0eeaf5 | 368 | inode->i_state |= I_SYNC; |
5547e8aa | 369 | inode->i_state &= ~I_DIRTY_PAGES; |
250df6ed | 370 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
371 | spin_unlock(&inode_lock); |
372 | ||
373 | ret = do_writepages(mapping, wbc); | |
374 | ||
26821ed4 CH |
375 | /* |
376 | * Make sure to wait on the data before writing out the metadata. | |
377 | * This is important for filesystems that modify metadata on data | |
378 | * I/O completion. | |
379 | */ | |
a9185b41 | 380 | if (wbc->sync_mode == WB_SYNC_ALL) { |
26821ed4 | 381 | int err = filemap_fdatawait(mapping); |
1da177e4 LT |
382 | if (ret == 0) |
383 | ret = err; | |
384 | } | |
385 | ||
5547e8aa DM |
386 | /* |
387 | * Some filesystems may redirty the inode during the writeback | |
388 | * due to delalloc, clear dirty metadata flags right before | |
389 | * write_inode() | |
390 | */ | |
391 | spin_lock(&inode_lock); | |
250df6ed | 392 | spin_lock(&inode->i_lock); |
5547e8aa DM |
393 | dirty = inode->i_state & I_DIRTY; |
394 | inode->i_state &= ~(I_DIRTY_SYNC | I_DIRTY_DATASYNC); | |
250df6ed | 395 | spin_unlock(&inode->i_lock); |
5547e8aa | 396 | spin_unlock(&inode_lock); |
26821ed4 CH |
397 | /* Don't write the inode if only I_DIRTY_PAGES was set */ |
398 | if (dirty & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
a9185b41 | 399 | int err = write_inode(inode, wbc); |
1da177e4 LT |
400 | if (ret == 0) |
401 | ret = err; | |
402 | } | |
403 | ||
404 | spin_lock(&inode_lock); | |
250df6ed | 405 | spin_lock(&inode->i_lock); |
1c0eeaf5 | 406 | inode->i_state &= ~I_SYNC; |
a4ffdde6 | 407 | if (!(inode->i_state & I_FREEING)) { |
23539afc | 408 | if (mapping_tagged(mapping, PAGECACHE_TAG_DIRTY)) { |
1da177e4 LT |
409 | /* |
410 | * We didn't write back all the pages. nfs_writepages() | |
a50aeb40 | 411 | * sometimes bales out without doing anything. |
1b43ef91 | 412 | */ |
a50aeb40 WF |
413 | inode->i_state |= I_DIRTY_PAGES; |
414 | if (wbc->nr_to_write <= 0) { | |
1da177e4 | 415 | /* |
a50aeb40 | 416 | * slice used up: queue for next turn |
1da177e4 | 417 | */ |
a50aeb40 | 418 | requeue_io(inode); |
1da177e4 LT |
419 | } else { |
420 | /* | |
a50aeb40 WF |
421 | * Writeback blocked by something other than |
422 | * congestion. Delay the inode for some time to | |
423 | * avoid spinning on the CPU (100% iowait) | |
424 | * retrying writeback of the dirty page/inode | |
425 | * that cannot be performed immediately. | |
1da177e4 | 426 | */ |
1b43ef91 | 427 | redirty_tail(inode); |
1da177e4 | 428 | } |
23539afc WF |
429 | } else if (inode->i_state & I_DIRTY) { |
430 | /* | |
431 | * Filesystems can dirty the inode during writeback | |
432 | * operations, such as delayed allocation during | |
433 | * submission or metadata updates after data IO | |
434 | * completion. | |
435 | */ | |
436 | redirty_tail(inode); | |
1da177e4 LT |
437 | } else { |
438 | /* | |
9e38d86f NP |
439 | * The inode is clean. At this point we either have |
440 | * a reference to the inode or it's on it's way out. | |
441 | * No need to add it back to the LRU. | |
1da177e4 | 442 | */ |
7ccf19a8 | 443 | list_del_init(&inode->i_wb_list); |
1da177e4 LT |
444 | } |
445 | } | |
1c0eeaf5 | 446 | inode_sync_complete(inode); |
250df6ed | 447 | spin_unlock(&inode->i_lock); |
1da177e4 LT |
448 | return ret; |
449 | } | |
450 | ||
03ba3782 | 451 | /* |
d19de7ed | 452 | * For background writeback the caller does not have the sb pinned |
03ba3782 JA |
453 | * before calling writeback. So make sure that we do pin it, so it doesn't |
454 | * go away while we are writing inodes from it. | |
03ba3782 | 455 | */ |
d19de7ed | 456 | static bool pin_sb_for_writeback(struct super_block *sb) |
03ba3782 | 457 | { |
03ba3782 | 458 | spin_lock(&sb_lock); |
29cb4859 CH |
459 | if (list_empty(&sb->s_instances)) { |
460 | spin_unlock(&sb_lock); | |
461 | return false; | |
462 | } | |
463 | ||
03ba3782 | 464 | sb->s_count++; |
29cb4859 CH |
465 | spin_unlock(&sb_lock); |
466 | ||
03ba3782 | 467 | if (down_read_trylock(&sb->s_umount)) { |
29cb4859 | 468 | if (sb->s_root) |
d19de7ed | 469 | return true; |
03ba3782 JA |
470 | up_read(&sb->s_umount); |
471 | } | |
29cb4859 CH |
472 | |
473 | put_super(sb); | |
d19de7ed | 474 | return false; |
03ba3782 JA |
475 | } |
476 | ||
f11c9c5c ES |
477 | /* |
478 | * Write a portion of b_io inodes which belong to @sb. | |
edadfb10 CH |
479 | * |
480 | * If @only_this_sb is true, then find and write all such | |
f11c9c5c ES |
481 | * inodes. Otherwise write only ones which go sequentially |
482 | * in reverse order. | |
edadfb10 | 483 | * |
f11c9c5c ES |
484 | * Return 1, if the caller writeback routine should be |
485 | * interrupted. Otherwise return 0. | |
486 | */ | |
edadfb10 CH |
487 | static int writeback_sb_inodes(struct super_block *sb, struct bdi_writeback *wb, |
488 | struct writeback_control *wbc, bool only_this_sb) | |
1da177e4 | 489 | { |
03ba3782 | 490 | while (!list_empty(&wb->b_io)) { |
1da177e4 | 491 | long pages_skipped; |
7ccf19a8 | 492 | struct inode *inode = wb_inode(wb->b_io.prev); |
edadfb10 CH |
493 | |
494 | if (inode->i_sb != sb) { | |
495 | if (only_this_sb) { | |
496 | /* | |
497 | * We only want to write back data for this | |
498 | * superblock, move all inodes not belonging | |
499 | * to it back onto the dirty list. | |
500 | */ | |
501 | redirty_tail(inode); | |
502 | continue; | |
503 | } | |
504 | ||
505 | /* | |
506 | * The inode belongs to a different superblock. | |
507 | * Bounce back to the caller to unpin this and | |
508 | * pin the next superblock. | |
509 | */ | |
f11c9c5c | 510 | return 0; |
edadfb10 CH |
511 | } |
512 | ||
9843b76a CH |
513 | /* |
514 | * Don't bother with new inodes or inodes beeing freed, first | |
515 | * kind does not need peridic writeout yet, and for the latter | |
516 | * kind writeout is handled by the freer. | |
517 | */ | |
250df6ed | 518 | spin_lock(&inode->i_lock); |
9843b76a | 519 | if (inode->i_state & (I_NEW | I_FREEING | I_WILL_FREE)) { |
250df6ed | 520 | spin_unlock(&inode->i_lock); |
7ef0d737 NP |
521 | requeue_io(inode); |
522 | continue; | |
523 | } | |
9843b76a | 524 | |
d2caa3c5 JL |
525 | /* |
526 | * Was this inode dirtied after sync_sb_inodes was called? | |
527 | * This keeps sync from extra jobs and livelock. | |
528 | */ | |
250df6ed DC |
529 | if (inode_dirtied_after(inode, wbc->wb_start)) { |
530 | spin_unlock(&inode->i_lock); | |
f11c9c5c | 531 | return 1; |
250df6ed | 532 | } |
1da177e4 | 533 | |
1da177e4 | 534 | __iget(inode); |
250df6ed DC |
535 | spin_unlock(&inode->i_lock); |
536 | ||
1da177e4 | 537 | pages_skipped = wbc->pages_skipped; |
01c03194 | 538 | writeback_single_inode(inode, wbc); |
1da177e4 LT |
539 | if (wbc->pages_skipped != pages_skipped) { |
540 | /* | |
541 | * writeback is not making progress due to locked | |
542 | * buffers. Skip this inode for now. | |
543 | */ | |
f57b9b7b | 544 | redirty_tail(inode); |
1da177e4 LT |
545 | } |
546 | spin_unlock(&inode_lock); | |
1da177e4 | 547 | iput(inode); |
4ffc8444 | 548 | cond_resched(); |
1da177e4 | 549 | spin_lock(&inode_lock); |
8bc3be27 FW |
550 | if (wbc->nr_to_write <= 0) { |
551 | wbc->more_io = 1; | |
f11c9c5c | 552 | return 1; |
8bc3be27 | 553 | } |
03ba3782 | 554 | if (!list_empty(&wb->b_more_io)) |
8bc3be27 | 555 | wbc->more_io = 1; |
1da177e4 | 556 | } |
f11c9c5c ES |
557 | /* b_io is empty */ |
558 | return 1; | |
559 | } | |
560 | ||
9c3a8ee8 CH |
561 | void writeback_inodes_wb(struct bdi_writeback *wb, |
562 | struct writeback_control *wbc) | |
f11c9c5c ES |
563 | { |
564 | int ret = 0; | |
565 | ||
7624ee72 JK |
566 | if (!wbc->wb_start) |
567 | wbc->wb_start = jiffies; /* livelock avoidance */ | |
f11c9c5c ES |
568 | spin_lock(&inode_lock); |
569 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) | |
570 | queue_io(wb, wbc->older_than_this); | |
38f21977 | 571 | |
f11c9c5c | 572 | while (!list_empty(&wb->b_io)) { |
7ccf19a8 | 573 | struct inode *inode = wb_inode(wb->b_io.prev); |
f11c9c5c | 574 | struct super_block *sb = inode->i_sb; |
9ecc2738 | 575 | |
edadfb10 CH |
576 | if (!pin_sb_for_writeback(sb)) { |
577 | requeue_io(inode); | |
578 | continue; | |
f11c9c5c | 579 | } |
edadfb10 CH |
580 | ret = writeback_sb_inodes(sb, wb, wbc, false); |
581 | drop_super(sb); | |
f11c9c5c | 582 | |
f11c9c5c ES |
583 | if (ret) |
584 | break; | |
585 | } | |
66f3b8e2 JA |
586 | spin_unlock(&inode_lock); |
587 | /* Leave any unwritten inodes on b_io */ | |
588 | } | |
589 | ||
edadfb10 CH |
590 | static void __writeback_inodes_sb(struct super_block *sb, |
591 | struct bdi_writeback *wb, struct writeback_control *wbc) | |
592 | { | |
593 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); | |
594 | ||
edadfb10 CH |
595 | spin_lock(&inode_lock); |
596 | if (!wbc->for_kupdate || list_empty(&wb->b_io)) | |
597 | queue_io(wb, wbc->older_than_this); | |
598 | writeback_sb_inodes(sb, wb, wbc, true); | |
599 | spin_unlock(&inode_lock); | |
600 | } | |
601 | ||
66f3b8e2 | 602 | /* |
03ba3782 JA |
603 | * The maximum number of pages to writeout in a single bdi flush/kupdate |
604 | * operation. We do this so we don't hold I_SYNC against an inode for | |
605 | * enormous amounts of time, which would block a userspace task which has | |
606 | * been forced to throttle against that inode. Also, the code reevaluates | |
607 | * the dirty each time it has written this many pages. | |
608 | */ | |
609 | #define MAX_WRITEBACK_PAGES 1024 | |
610 | ||
611 | static inline bool over_bground_thresh(void) | |
612 | { | |
613 | unsigned long background_thresh, dirty_thresh; | |
614 | ||
16c4042f | 615 | global_dirty_limits(&background_thresh, &dirty_thresh); |
03ba3782 JA |
616 | |
617 | return (global_page_state(NR_FILE_DIRTY) + | |
4cbec4c8 | 618 | global_page_state(NR_UNSTABLE_NFS) > background_thresh); |
03ba3782 JA |
619 | } |
620 | ||
621 | /* | |
622 | * Explicit flushing or periodic writeback of "old" data. | |
66f3b8e2 | 623 | * |
03ba3782 JA |
624 | * Define "old": the first time one of an inode's pages is dirtied, we mark the |
625 | * dirtying-time in the inode's address_space. So this periodic writeback code | |
626 | * just walks the superblock inode list, writing back any inodes which are | |
627 | * older than a specific point in time. | |
66f3b8e2 | 628 | * |
03ba3782 JA |
629 | * Try to run once per dirty_writeback_interval. But if a writeback event |
630 | * takes longer than a dirty_writeback_interval interval, then leave a | |
631 | * one-second gap. | |
66f3b8e2 | 632 | * |
03ba3782 JA |
633 | * older_than_this takes precedence over nr_to_write. So we'll only write back |
634 | * all dirty pages if they are all attached to "old" mappings. | |
66f3b8e2 | 635 | */ |
c4a77a6c | 636 | static long wb_writeback(struct bdi_writeback *wb, |
83ba7b07 | 637 | struct wb_writeback_work *work) |
66f3b8e2 | 638 | { |
03ba3782 | 639 | struct writeback_control wbc = { |
83ba7b07 | 640 | .sync_mode = work->sync_mode, |
03ba3782 | 641 | .older_than_this = NULL, |
83ba7b07 CH |
642 | .for_kupdate = work->for_kupdate, |
643 | .for_background = work->for_background, | |
644 | .range_cyclic = work->range_cyclic, | |
03ba3782 JA |
645 | }; |
646 | unsigned long oldest_jif; | |
647 | long wrote = 0; | |
b9543dac | 648 | long write_chunk; |
a5989bdc | 649 | struct inode *inode; |
66f3b8e2 | 650 | |
03ba3782 JA |
651 | if (wbc.for_kupdate) { |
652 | wbc.older_than_this = &oldest_jif; | |
653 | oldest_jif = jiffies - | |
654 | msecs_to_jiffies(dirty_expire_interval * 10); | |
655 | } | |
c4a77a6c JA |
656 | if (!wbc.range_cyclic) { |
657 | wbc.range_start = 0; | |
658 | wbc.range_end = LLONG_MAX; | |
659 | } | |
38f21977 | 660 | |
b9543dac JK |
661 | /* |
662 | * WB_SYNC_ALL mode does livelock avoidance by syncing dirty | |
663 | * inodes/pages in one big loop. Setting wbc.nr_to_write=LONG_MAX | |
664 | * here avoids calling into writeback_inodes_wb() more than once. | |
665 | * | |
666 | * The intended call sequence for WB_SYNC_ALL writeback is: | |
667 | * | |
668 | * wb_writeback() | |
669 | * __writeback_inodes_sb() <== called only once | |
670 | * write_cache_pages() <== called once for each inode | |
671 | * (quickly) tag currently dirty pages | |
672 | * (maybe slowly) sync all tagged pages | |
673 | */ | |
674 | if (wbc.sync_mode == WB_SYNC_NONE) | |
675 | write_chunk = MAX_WRITEBACK_PAGES; | |
676 | else | |
677 | write_chunk = LONG_MAX; | |
678 | ||
7624ee72 | 679 | wbc.wb_start = jiffies; /* livelock avoidance */ |
03ba3782 JA |
680 | for (;;) { |
681 | /* | |
d3ddec76 | 682 | * Stop writeback when nr_pages has been consumed |
03ba3782 | 683 | */ |
83ba7b07 | 684 | if (work->nr_pages <= 0) |
03ba3782 | 685 | break; |
66f3b8e2 | 686 | |
aa373cf5 JK |
687 | /* |
688 | * Background writeout and kupdate-style writeback may | |
689 | * run forever. Stop them if there is other work to do | |
690 | * so that e.g. sync can proceed. They'll be restarted | |
691 | * after the other works are all done. | |
692 | */ | |
693 | if ((work->for_background || work->for_kupdate) && | |
694 | !list_empty(&wb->bdi->work_list)) | |
695 | break; | |
696 | ||
38f21977 | 697 | /* |
d3ddec76 WF |
698 | * For background writeout, stop when we are below the |
699 | * background dirty threshold | |
38f21977 | 700 | */ |
83ba7b07 | 701 | if (work->for_background && !over_bground_thresh()) |
03ba3782 | 702 | break; |
38f21977 | 703 | |
03ba3782 | 704 | wbc.more_io = 0; |
b9543dac | 705 | wbc.nr_to_write = write_chunk; |
03ba3782 | 706 | wbc.pages_skipped = 0; |
028c2dd1 DC |
707 | |
708 | trace_wbc_writeback_start(&wbc, wb->bdi); | |
83ba7b07 CH |
709 | if (work->sb) |
710 | __writeback_inodes_sb(work->sb, wb, &wbc); | |
edadfb10 CH |
711 | else |
712 | writeback_inodes_wb(wb, &wbc); | |
028c2dd1 DC |
713 | trace_wbc_writeback_written(&wbc, wb->bdi); |
714 | ||
b9543dac JK |
715 | work->nr_pages -= write_chunk - wbc.nr_to_write; |
716 | wrote += write_chunk - wbc.nr_to_write; | |
03ba3782 JA |
717 | |
718 | /* | |
71fd05a8 | 719 | * If we consumed everything, see if we have more |
03ba3782 | 720 | */ |
71fd05a8 JA |
721 | if (wbc.nr_to_write <= 0) |
722 | continue; | |
723 | /* | |
724 | * Didn't write everything and we don't have more IO, bail | |
725 | */ | |
726 | if (!wbc.more_io) | |
03ba3782 | 727 | break; |
71fd05a8 JA |
728 | /* |
729 | * Did we write something? Try for more | |
730 | */ | |
b9543dac | 731 | if (wbc.nr_to_write < write_chunk) |
71fd05a8 JA |
732 | continue; |
733 | /* | |
734 | * Nothing written. Wait for some inode to | |
735 | * become available for writeback. Otherwise | |
736 | * we'll just busyloop. | |
737 | */ | |
738 | spin_lock(&inode_lock); | |
739 | if (!list_empty(&wb->b_more_io)) { | |
7ccf19a8 | 740 | inode = wb_inode(wb->b_more_io.prev); |
028c2dd1 | 741 | trace_wbc_writeback_wait(&wbc, wb->bdi); |
250df6ed | 742 | spin_lock(&inode->i_lock); |
71fd05a8 | 743 | inode_wait_for_writeback(inode); |
250df6ed | 744 | spin_unlock(&inode->i_lock); |
03ba3782 | 745 | } |
71fd05a8 | 746 | spin_unlock(&inode_lock); |
03ba3782 JA |
747 | } |
748 | ||
749 | return wrote; | |
750 | } | |
751 | ||
752 | /* | |
83ba7b07 | 753 | * Return the next wb_writeback_work struct that hasn't been processed yet. |
03ba3782 | 754 | */ |
83ba7b07 | 755 | static struct wb_writeback_work * |
08852b6d | 756 | get_next_work_item(struct backing_dev_info *bdi) |
03ba3782 | 757 | { |
83ba7b07 | 758 | struct wb_writeback_work *work = NULL; |
03ba3782 | 759 | |
6467716a | 760 | spin_lock_bh(&bdi->wb_lock); |
83ba7b07 CH |
761 | if (!list_empty(&bdi->work_list)) { |
762 | work = list_entry(bdi->work_list.next, | |
763 | struct wb_writeback_work, list); | |
764 | list_del_init(&work->list); | |
03ba3782 | 765 | } |
6467716a | 766 | spin_unlock_bh(&bdi->wb_lock); |
83ba7b07 | 767 | return work; |
03ba3782 JA |
768 | } |
769 | ||
cdf01dd5 LT |
770 | /* |
771 | * Add in the number of potentially dirty inodes, because each inode | |
772 | * write can dirty pagecache in the underlying blockdev. | |
773 | */ | |
774 | static unsigned long get_nr_dirty_pages(void) | |
775 | { | |
776 | return global_page_state(NR_FILE_DIRTY) + | |
777 | global_page_state(NR_UNSTABLE_NFS) + | |
778 | get_nr_dirty_inodes(); | |
779 | } | |
780 | ||
6585027a JK |
781 | static long wb_check_background_flush(struct bdi_writeback *wb) |
782 | { | |
783 | if (over_bground_thresh()) { | |
784 | ||
785 | struct wb_writeback_work work = { | |
786 | .nr_pages = LONG_MAX, | |
787 | .sync_mode = WB_SYNC_NONE, | |
788 | .for_background = 1, | |
789 | .range_cyclic = 1, | |
790 | }; | |
791 | ||
792 | return wb_writeback(wb, &work); | |
793 | } | |
794 | ||
795 | return 0; | |
796 | } | |
797 | ||
03ba3782 JA |
798 | static long wb_check_old_data_flush(struct bdi_writeback *wb) |
799 | { | |
800 | unsigned long expired; | |
801 | long nr_pages; | |
802 | ||
69b62d01 JA |
803 | /* |
804 | * When set to zero, disable periodic writeback | |
805 | */ | |
806 | if (!dirty_writeback_interval) | |
807 | return 0; | |
808 | ||
03ba3782 JA |
809 | expired = wb->last_old_flush + |
810 | msecs_to_jiffies(dirty_writeback_interval * 10); | |
811 | if (time_before(jiffies, expired)) | |
812 | return 0; | |
813 | ||
814 | wb->last_old_flush = jiffies; | |
cdf01dd5 | 815 | nr_pages = get_nr_dirty_pages(); |
03ba3782 | 816 | |
c4a77a6c | 817 | if (nr_pages) { |
83ba7b07 | 818 | struct wb_writeback_work work = { |
c4a77a6c JA |
819 | .nr_pages = nr_pages, |
820 | .sync_mode = WB_SYNC_NONE, | |
821 | .for_kupdate = 1, | |
822 | .range_cyclic = 1, | |
823 | }; | |
824 | ||
83ba7b07 | 825 | return wb_writeback(wb, &work); |
c4a77a6c | 826 | } |
03ba3782 JA |
827 | |
828 | return 0; | |
829 | } | |
830 | ||
831 | /* | |
832 | * Retrieve work items and do the writeback they describe | |
833 | */ | |
834 | long wb_do_writeback(struct bdi_writeback *wb, int force_wait) | |
835 | { | |
836 | struct backing_dev_info *bdi = wb->bdi; | |
83ba7b07 | 837 | struct wb_writeback_work *work; |
c4a77a6c | 838 | long wrote = 0; |
03ba3782 | 839 | |
81d73a32 | 840 | set_bit(BDI_writeback_running, &wb->bdi->state); |
08852b6d | 841 | while ((work = get_next_work_item(bdi)) != NULL) { |
03ba3782 JA |
842 | /* |
843 | * Override sync mode, in case we must wait for completion | |
83ba7b07 | 844 | * because this thread is exiting now. |
03ba3782 JA |
845 | */ |
846 | if (force_wait) | |
83ba7b07 | 847 | work->sync_mode = WB_SYNC_ALL; |
03ba3782 | 848 | |
455b2864 DC |
849 | trace_writeback_exec(bdi, work); |
850 | ||
83ba7b07 | 851 | wrote += wb_writeback(wb, work); |
03ba3782 JA |
852 | |
853 | /* | |
83ba7b07 CH |
854 | * Notify the caller of completion if this is a synchronous |
855 | * work item, otherwise just free it. | |
03ba3782 | 856 | */ |
83ba7b07 CH |
857 | if (work->done) |
858 | complete(work->done); | |
859 | else | |
860 | kfree(work); | |
03ba3782 JA |
861 | } |
862 | ||
863 | /* | |
864 | * Check for periodic writeback, kupdated() style | |
865 | */ | |
866 | wrote += wb_check_old_data_flush(wb); | |
6585027a | 867 | wrote += wb_check_background_flush(wb); |
81d73a32 | 868 | clear_bit(BDI_writeback_running, &wb->bdi->state); |
03ba3782 JA |
869 | |
870 | return wrote; | |
871 | } | |
872 | ||
873 | /* | |
874 | * Handle writeback of dirty data for the device backed by this bdi. Also | |
875 | * wakes up periodically and does kupdated style flushing. | |
876 | */ | |
08243900 | 877 | int bdi_writeback_thread(void *data) |
03ba3782 | 878 | { |
08243900 CH |
879 | struct bdi_writeback *wb = data; |
880 | struct backing_dev_info *bdi = wb->bdi; | |
03ba3782 JA |
881 | long pages_written; |
882 | ||
766f9164 | 883 | current->flags |= PF_SWAPWRITE; |
08243900 | 884 | set_freezable(); |
ecd58403 | 885 | wb->last_active = jiffies; |
08243900 CH |
886 | |
887 | /* | |
888 | * Our parent may run at a different priority, just set us to normal | |
889 | */ | |
890 | set_user_nice(current, 0); | |
891 | ||
455b2864 DC |
892 | trace_writeback_thread_start(bdi); |
893 | ||
03ba3782 | 894 | while (!kthread_should_stop()) { |
6467716a AB |
895 | /* |
896 | * Remove own delayed wake-up timer, since we are already awake | |
897 | * and we'll take care of the preriodic write-back. | |
898 | */ | |
899 | del_timer(&wb->wakeup_timer); | |
900 | ||
03ba3782 JA |
901 | pages_written = wb_do_writeback(wb, 0); |
902 | ||
455b2864 DC |
903 | trace_writeback_pages_written(pages_written); |
904 | ||
03ba3782 | 905 | if (pages_written) |
ecd58403 | 906 | wb->last_active = jiffies; |
03ba3782 | 907 | |
297252c8 | 908 | set_current_state(TASK_INTERRUPTIBLE); |
b76b4014 | 909 | if (!list_empty(&bdi->work_list) || kthread_should_stop()) { |
f9eadbbd | 910 | __set_current_state(TASK_RUNNING); |
297252c8 | 911 | continue; |
03ba3782 JA |
912 | } |
913 | ||
253c34e9 | 914 | if (wb_has_dirty_io(wb) && dirty_writeback_interval) |
fff5b85a | 915 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
253c34e9 AB |
916 | else { |
917 | /* | |
918 | * We have nothing to do, so can go sleep without any | |
919 | * timeout and save power. When a work is queued or | |
920 | * something is made dirty - we will be woken up. | |
921 | */ | |
297252c8 | 922 | schedule(); |
f9eadbbd | 923 | } |
69b62d01 | 924 | |
03ba3782 JA |
925 | try_to_freeze(); |
926 | } | |
927 | ||
fff5b85a | 928 | /* Flush any work that raced with us exiting */ |
08243900 CH |
929 | if (!list_empty(&bdi->work_list)) |
930 | wb_do_writeback(wb, 1); | |
455b2864 DC |
931 | |
932 | trace_writeback_thread_stop(bdi); | |
03ba3782 JA |
933 | return 0; |
934 | } | |
935 | ||
08243900 | 936 | |
03ba3782 | 937 | /* |
b8c2f347 CH |
938 | * Start writeback of `nr_pages' pages. If `nr_pages' is zero, write back |
939 | * the whole world. | |
03ba3782 | 940 | */ |
b8c2f347 | 941 | void wakeup_flusher_threads(long nr_pages) |
03ba3782 | 942 | { |
b8c2f347 | 943 | struct backing_dev_info *bdi; |
03ba3782 | 944 | |
83ba7b07 CH |
945 | if (!nr_pages) { |
946 | nr_pages = global_page_state(NR_FILE_DIRTY) + | |
b8c2f347 CH |
947 | global_page_state(NR_UNSTABLE_NFS); |
948 | } | |
03ba3782 | 949 | |
b8c2f347 | 950 | rcu_read_lock(); |
cfc4ba53 | 951 | list_for_each_entry_rcu(bdi, &bdi_list, bdi_list) { |
03ba3782 JA |
952 | if (!bdi_has_dirty_io(bdi)) |
953 | continue; | |
6585027a | 954 | __bdi_start_writeback(bdi, nr_pages, false); |
03ba3782 | 955 | } |
cfc4ba53 | 956 | rcu_read_unlock(); |
1da177e4 LT |
957 | } |
958 | ||
03ba3782 JA |
959 | static noinline void block_dump___mark_inode_dirty(struct inode *inode) |
960 | { | |
961 | if (inode->i_ino || strcmp(inode->i_sb->s_id, "bdev")) { | |
962 | struct dentry *dentry; | |
963 | const char *name = "?"; | |
964 | ||
965 | dentry = d_find_alias(inode); | |
966 | if (dentry) { | |
967 | spin_lock(&dentry->d_lock); | |
968 | name = (const char *) dentry->d_name.name; | |
969 | } | |
970 | printk(KERN_DEBUG | |
971 | "%s(%d): dirtied inode %lu (%s) on %s\n", | |
972 | current->comm, task_pid_nr(current), inode->i_ino, | |
973 | name, inode->i_sb->s_id); | |
974 | if (dentry) { | |
975 | spin_unlock(&dentry->d_lock); | |
976 | dput(dentry); | |
977 | } | |
978 | } | |
979 | } | |
980 | ||
981 | /** | |
982 | * __mark_inode_dirty - internal function | |
983 | * @inode: inode to mark | |
984 | * @flags: what kind of dirty (i.e. I_DIRTY_SYNC) | |
985 | * Mark an inode as dirty. Callers should use mark_inode_dirty or | |
986 | * mark_inode_dirty_sync. | |
1da177e4 | 987 | * |
03ba3782 JA |
988 | * Put the inode on the super block's dirty list. |
989 | * | |
990 | * CAREFUL! We mark it dirty unconditionally, but move it onto the | |
991 | * dirty list only if it is hashed or if it refers to a blockdev. | |
992 | * If it was not hashed, it will never be added to the dirty list | |
993 | * even if it is later hashed, as it will have been marked dirty already. | |
994 | * | |
995 | * In short, make sure you hash any inodes _before_ you start marking | |
996 | * them dirty. | |
1da177e4 | 997 | * |
03ba3782 JA |
998 | * This function *must* be atomic for the I_DIRTY_PAGES case - |
999 | * set_page_dirty() is called under spinlock in several places. | |
1da177e4 | 1000 | * |
03ba3782 JA |
1001 | * Note that for blockdevs, inode->dirtied_when represents the dirtying time of |
1002 | * the block-special inode (/dev/hda1) itself. And the ->dirtied_when field of | |
1003 | * the kernel-internal blockdev inode represents the dirtying time of the | |
1004 | * blockdev's pages. This is why for I_DIRTY_PAGES we always use | |
1005 | * page->mapping->host, so the page-dirtying time is recorded in the internal | |
1006 | * blockdev inode. | |
1da177e4 | 1007 | */ |
03ba3782 | 1008 | void __mark_inode_dirty(struct inode *inode, int flags) |
1da177e4 | 1009 | { |
03ba3782 | 1010 | struct super_block *sb = inode->i_sb; |
253c34e9 AB |
1011 | struct backing_dev_info *bdi = NULL; |
1012 | bool wakeup_bdi = false; | |
1da177e4 | 1013 | |
03ba3782 JA |
1014 | /* |
1015 | * Don't do this for I_DIRTY_PAGES - that doesn't actually | |
1016 | * dirty the inode itself | |
1017 | */ | |
1018 | if (flags & (I_DIRTY_SYNC | I_DIRTY_DATASYNC)) { | |
1019 | if (sb->s_op->dirty_inode) | |
1020 | sb->s_op->dirty_inode(inode); | |
1021 | } | |
1022 | ||
1023 | /* | |
1024 | * make sure that changes are seen by all cpus before we test i_state | |
1025 | * -- mikulas | |
1026 | */ | |
1027 | smp_mb(); | |
1028 | ||
1029 | /* avoid the locking if we can */ | |
1030 | if ((inode->i_state & flags) == flags) | |
1031 | return; | |
1032 | ||
1033 | if (unlikely(block_dump)) | |
1034 | block_dump___mark_inode_dirty(inode); | |
1035 | ||
1036 | spin_lock(&inode_lock); | |
250df6ed | 1037 | spin_lock(&inode->i_lock); |
03ba3782 JA |
1038 | if ((inode->i_state & flags) != flags) { |
1039 | const int was_dirty = inode->i_state & I_DIRTY; | |
1040 | ||
1041 | inode->i_state |= flags; | |
1042 | ||
1043 | /* | |
1044 | * If the inode is being synced, just update its dirty state. | |
1045 | * The unlocker will place the inode on the appropriate | |
1046 | * superblock list, based upon its state. | |
1047 | */ | |
1048 | if (inode->i_state & I_SYNC) | |
250df6ed | 1049 | goto out_unlock_inode; |
03ba3782 JA |
1050 | |
1051 | /* | |
1052 | * Only add valid (hashed) inodes to the superblock's | |
1053 | * dirty list. Add blockdev inodes as well. | |
1054 | */ | |
1055 | if (!S_ISBLK(inode->i_mode)) { | |
1d3382cb | 1056 | if (inode_unhashed(inode)) |
250df6ed | 1057 | goto out_unlock_inode; |
03ba3782 | 1058 | } |
a4ffdde6 | 1059 | if (inode->i_state & I_FREEING) |
250df6ed | 1060 | goto out_unlock_inode; |
03ba3782 | 1061 | |
250df6ed | 1062 | spin_unlock(&inode->i_lock); |
03ba3782 JA |
1063 | /* |
1064 | * If the inode was already on b_dirty/b_io/b_more_io, don't | |
1065 | * reposition it (that would break b_dirty time-ordering). | |
1066 | */ | |
1067 | if (!was_dirty) { | |
253c34e9 AB |
1068 | bdi = inode_to_bdi(inode); |
1069 | ||
1070 | if (bdi_cap_writeback_dirty(bdi)) { | |
1071 | WARN(!test_bit(BDI_registered, &bdi->state), | |
1072 | "bdi-%s not registered\n", bdi->name); | |
1073 | ||
1074 | /* | |
1075 | * If this is the first dirty inode for this | |
1076 | * bdi, we have to wake-up the corresponding | |
1077 | * bdi thread to make sure background | |
1078 | * write-back happens later. | |
1079 | */ | |
1080 | if (!wb_has_dirty_io(&bdi->wb)) | |
1081 | wakeup_bdi = true; | |
500b067c | 1082 | } |
03ba3782 JA |
1083 | |
1084 | inode->dirtied_when = jiffies; | |
7ccf19a8 | 1085 | list_move(&inode->i_wb_list, &bdi->wb.b_dirty); |
1da177e4 | 1086 | } |
250df6ed | 1087 | goto out; |
1da177e4 | 1088 | } |
250df6ed DC |
1089 | out_unlock_inode: |
1090 | spin_unlock(&inode->i_lock); | |
03ba3782 JA |
1091 | out: |
1092 | spin_unlock(&inode_lock); | |
253c34e9 AB |
1093 | |
1094 | if (wakeup_bdi) | |
6467716a | 1095 | bdi_wakeup_thread_delayed(bdi); |
03ba3782 JA |
1096 | } |
1097 | EXPORT_SYMBOL(__mark_inode_dirty); | |
1098 | ||
1099 | /* | |
1100 | * Write out a superblock's list of dirty inodes. A wait will be performed | |
1101 | * upon no inodes, all inodes or the final one, depending upon sync_mode. | |
1102 | * | |
1103 | * If older_than_this is non-NULL, then only write out inodes which | |
1104 | * had their first dirtying at a time earlier than *older_than_this. | |
1105 | * | |
03ba3782 JA |
1106 | * If `bdi' is non-zero then we're being asked to writeback a specific queue. |
1107 | * This function assumes that the blockdev superblock's inodes are backed by | |
1108 | * a variety of queues, so all inodes are searched. For other superblocks, | |
1109 | * assume that all inodes are backed by the same queue. | |
1110 | * | |
1111 | * The inodes to be written are parked on bdi->b_io. They are moved back onto | |
1112 | * bdi->b_dirty as they are selected for writing. This way, none can be missed | |
1113 | * on the writer throttling path, and we get decent balancing between many | |
1114 | * throttled threads: we don't want them all piling up on inode_sync_wait. | |
1115 | */ | |
b6e51316 | 1116 | static void wait_sb_inodes(struct super_block *sb) |
03ba3782 JA |
1117 | { |
1118 | struct inode *inode, *old_inode = NULL; | |
1119 | ||
1120 | /* | |
1121 | * We need to be protected against the filesystem going from | |
1122 | * r/o to r/w or vice versa. | |
1123 | */ | |
b6e51316 | 1124 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
03ba3782 | 1125 | |
55fa6091 | 1126 | spin_lock(&inode_sb_list_lock); |
03ba3782 JA |
1127 | |
1128 | /* | |
1129 | * Data integrity sync. Must wait for all pages under writeback, | |
1130 | * because there may have been pages dirtied before our sync | |
1131 | * call, but which had writeout started before we write it out. | |
1132 | * In which case, the inode may not be on the dirty list, but | |
1133 | * we still have to wait for that writeout. | |
1134 | */ | |
b6e51316 | 1135 | list_for_each_entry(inode, &sb->s_inodes, i_sb_list) { |
250df6ed | 1136 | struct address_space *mapping = inode->i_mapping; |
03ba3782 | 1137 | |
250df6ed DC |
1138 | spin_lock(&inode->i_lock); |
1139 | if ((inode->i_state & (I_FREEING|I_WILL_FREE|I_NEW)) || | |
1140 | (mapping->nrpages == 0)) { | |
1141 | spin_unlock(&inode->i_lock); | |
03ba3782 | 1142 | continue; |
250df6ed | 1143 | } |
03ba3782 | 1144 | __iget(inode); |
250df6ed | 1145 | spin_unlock(&inode->i_lock); |
55fa6091 DC |
1146 | spin_unlock(&inode_sb_list_lock); |
1147 | ||
03ba3782 | 1148 | /* |
55fa6091 DC |
1149 | * We hold a reference to 'inode' so it couldn't have been |
1150 | * removed from s_inodes list while we dropped the | |
1151 | * inode_sb_list_lock. We cannot iput the inode now as we can | |
1152 | * be holding the last reference and we cannot iput it under | |
1153 | * inode_sb_list_lock. So we keep the reference and iput it | |
1154 | * later. | |
03ba3782 JA |
1155 | */ |
1156 | iput(old_inode); | |
1157 | old_inode = inode; | |
1158 | ||
1159 | filemap_fdatawait(mapping); | |
1160 | ||
1161 | cond_resched(); | |
1162 | ||
55fa6091 | 1163 | spin_lock(&inode_sb_list_lock); |
03ba3782 | 1164 | } |
55fa6091 | 1165 | spin_unlock(&inode_sb_list_lock); |
03ba3782 | 1166 | iput(old_inode); |
1da177e4 LT |
1167 | } |
1168 | ||
d8a8559c | 1169 | /** |
3259f8be | 1170 | * writeback_inodes_sb_nr - writeback dirty inodes from given super_block |
d8a8559c | 1171 | * @sb: the superblock |
3259f8be | 1172 | * @nr: the number of pages to write |
1da177e4 | 1173 | * |
d8a8559c JA |
1174 | * Start writeback on some inodes on this super_block. No guarantees are made |
1175 | * on how many (if any) will be written, and this function does not wait | |
3259f8be | 1176 | * for IO completion of submitted IO. |
1da177e4 | 1177 | */ |
3259f8be | 1178 | void writeback_inodes_sb_nr(struct super_block *sb, unsigned long nr) |
1da177e4 | 1179 | { |
83ba7b07 CH |
1180 | DECLARE_COMPLETION_ONSTACK(done); |
1181 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1182 | .sb = sb, |
1183 | .sync_mode = WB_SYNC_NONE, | |
83ba7b07 | 1184 | .done = &done, |
3259f8be | 1185 | .nr_pages = nr, |
3c4d7165 | 1186 | }; |
d8a8559c | 1187 | |
cf37e972 | 1188 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
83ba7b07 CH |
1189 | bdi_queue_work(sb->s_bdi, &work); |
1190 | wait_for_completion(&done); | |
e913fc82 | 1191 | } |
3259f8be CM |
1192 | EXPORT_SYMBOL(writeback_inodes_sb_nr); |
1193 | ||
1194 | /** | |
1195 | * writeback_inodes_sb - writeback dirty inodes from given super_block | |
1196 | * @sb: the superblock | |
1197 | * | |
1198 | * Start writeback on some inodes on this super_block. No guarantees are made | |
1199 | * on how many (if any) will be written, and this function does not wait | |
1200 | * for IO completion of submitted IO. | |
1201 | */ | |
1202 | void writeback_inodes_sb(struct super_block *sb) | |
1203 | { | |
925d169f | 1204 | return writeback_inodes_sb_nr(sb, get_nr_dirty_pages()); |
3259f8be | 1205 | } |
0e3c9a22 | 1206 | EXPORT_SYMBOL(writeback_inodes_sb); |
e913fc82 | 1207 | |
17bd55d0 ES |
1208 | /** |
1209 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1210 | * @sb: the superblock | |
1211 | * | |
1212 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1213 | * Returns 1 if writeback was started, 0 if not. | |
1214 | */ | |
1215 | int writeback_inodes_sb_if_idle(struct super_block *sb) | |
1216 | { | |
1217 | if (!writeback_in_progress(sb->s_bdi)) { | |
cf37e972 | 1218 | down_read(&sb->s_umount); |
17bd55d0 | 1219 | writeback_inodes_sb(sb); |
cf37e972 | 1220 | up_read(&sb->s_umount); |
17bd55d0 ES |
1221 | return 1; |
1222 | } else | |
1223 | return 0; | |
1224 | } | |
1225 | EXPORT_SYMBOL(writeback_inodes_sb_if_idle); | |
1226 | ||
3259f8be CM |
1227 | /** |
1228 | * writeback_inodes_sb_if_idle - start writeback if none underway | |
1229 | * @sb: the superblock | |
1230 | * @nr: the number of pages to write | |
1231 | * | |
1232 | * Invoke writeback_inodes_sb if no writeback is currently underway. | |
1233 | * Returns 1 if writeback was started, 0 if not. | |
1234 | */ | |
1235 | int writeback_inodes_sb_nr_if_idle(struct super_block *sb, | |
1236 | unsigned long nr) | |
1237 | { | |
1238 | if (!writeback_in_progress(sb->s_bdi)) { | |
1239 | down_read(&sb->s_umount); | |
1240 | writeback_inodes_sb_nr(sb, nr); | |
1241 | up_read(&sb->s_umount); | |
1242 | return 1; | |
1243 | } else | |
1244 | return 0; | |
1245 | } | |
1246 | EXPORT_SYMBOL(writeback_inodes_sb_nr_if_idle); | |
1247 | ||
d8a8559c JA |
1248 | /** |
1249 | * sync_inodes_sb - sync sb inode pages | |
1250 | * @sb: the superblock | |
1251 | * | |
1252 | * This function writes and waits on any dirty inode belonging to this | |
cb9ef8d5 | 1253 | * super_block. |
d8a8559c | 1254 | */ |
b6e51316 | 1255 | void sync_inodes_sb(struct super_block *sb) |
d8a8559c | 1256 | { |
83ba7b07 CH |
1257 | DECLARE_COMPLETION_ONSTACK(done); |
1258 | struct wb_writeback_work work = { | |
3c4d7165 CH |
1259 | .sb = sb, |
1260 | .sync_mode = WB_SYNC_ALL, | |
1261 | .nr_pages = LONG_MAX, | |
1262 | .range_cyclic = 0, | |
83ba7b07 | 1263 | .done = &done, |
3c4d7165 CH |
1264 | }; |
1265 | ||
cf37e972 CH |
1266 | WARN_ON(!rwsem_is_locked(&sb->s_umount)); |
1267 | ||
83ba7b07 CH |
1268 | bdi_queue_work(sb->s_bdi, &work); |
1269 | wait_for_completion(&done); | |
1270 | ||
b6e51316 | 1271 | wait_sb_inodes(sb); |
1da177e4 | 1272 | } |
d8a8559c | 1273 | EXPORT_SYMBOL(sync_inodes_sb); |
1da177e4 | 1274 | |
1da177e4 | 1275 | /** |
7f04c26d AA |
1276 | * write_inode_now - write an inode to disk |
1277 | * @inode: inode to write to disk | |
1278 | * @sync: whether the write should be synchronous or not | |
1279 | * | |
1280 | * This function commits an inode to disk immediately if it is dirty. This is | |
1281 | * primarily needed by knfsd. | |
1da177e4 | 1282 | * |
7f04c26d | 1283 | * The caller must either have a ref on the inode or must have set I_WILL_FREE. |
1da177e4 | 1284 | */ |
1da177e4 LT |
1285 | int write_inode_now(struct inode *inode, int sync) |
1286 | { | |
1287 | int ret; | |
1288 | struct writeback_control wbc = { | |
1289 | .nr_to_write = LONG_MAX, | |
18914b18 | 1290 | .sync_mode = sync ? WB_SYNC_ALL : WB_SYNC_NONE, |
111ebb6e OH |
1291 | .range_start = 0, |
1292 | .range_end = LLONG_MAX, | |
1da177e4 LT |
1293 | }; |
1294 | ||
1295 | if (!mapping_cap_writeback_dirty(inode->i_mapping)) | |
49364ce2 | 1296 | wbc.nr_to_write = 0; |
1da177e4 LT |
1297 | |
1298 | might_sleep(); | |
1299 | spin_lock(&inode_lock); | |
01c03194 | 1300 | ret = writeback_single_inode(inode, &wbc); |
1da177e4 LT |
1301 | spin_unlock(&inode_lock); |
1302 | if (sync) | |
1c0eeaf5 | 1303 | inode_sync_wait(inode); |
1da177e4 LT |
1304 | return ret; |
1305 | } | |
1306 | EXPORT_SYMBOL(write_inode_now); | |
1307 | ||
1308 | /** | |
1309 | * sync_inode - write an inode and its pages to disk. | |
1310 | * @inode: the inode to sync | |
1311 | * @wbc: controls the writeback mode | |
1312 | * | |
1313 | * sync_inode() will write an inode and its pages to disk. It will also | |
1314 | * correctly update the inode on its superblock's dirty inode lists and will | |
1315 | * update inode->i_state. | |
1316 | * | |
1317 | * The caller must have a ref on the inode. | |
1318 | */ | |
1319 | int sync_inode(struct inode *inode, struct writeback_control *wbc) | |
1320 | { | |
1321 | int ret; | |
1322 | ||
1323 | spin_lock(&inode_lock); | |
01c03194 | 1324 | ret = writeback_single_inode(inode, wbc); |
1da177e4 LT |
1325 | spin_unlock(&inode_lock); |
1326 | return ret; | |
1327 | } | |
1328 | EXPORT_SYMBOL(sync_inode); | |
c3765016 CH |
1329 | |
1330 | /** | |
c691b9d9 | 1331 | * sync_inode_metadata - write an inode to disk |
c3765016 CH |
1332 | * @inode: the inode to sync |
1333 | * @wait: wait for I/O to complete. | |
1334 | * | |
c691b9d9 | 1335 | * Write an inode to disk and adjust its dirty state after completion. |
c3765016 CH |
1336 | * |
1337 | * Note: only writes the actual inode, no associated data or other metadata. | |
1338 | */ | |
1339 | int sync_inode_metadata(struct inode *inode, int wait) | |
1340 | { | |
1341 | struct writeback_control wbc = { | |
1342 | .sync_mode = wait ? WB_SYNC_ALL : WB_SYNC_NONE, | |
1343 | .nr_to_write = 0, /* metadata-only */ | |
1344 | }; | |
1345 | ||
1346 | return sync_inode(inode, &wbc); | |
1347 | } | |
1348 | EXPORT_SYMBOL(sync_inode_metadata); |