Commit | Line | Data |
---|---|---|
3fcfab16 AM |
1 | |
2 | #include <linux/wait.h> | |
3 | #include <linux/backing-dev.h> | |
03ba3782 JA |
4 | #include <linux/kthread.h> |
5 | #include <linux/freezer.h> | |
3fcfab16 | 6 | #include <linux/fs.h> |
26160158 | 7 | #include <linux/pagemap.h> |
03ba3782 | 8 | #include <linux/mm.h> |
3fcfab16 AM |
9 | #include <linux/sched.h> |
10 | #include <linux/module.h> | |
cf0ca9fe PZ |
11 | #include <linux/writeback.h> |
12 | #include <linux/device.h> | |
455b2864 | 13 | #include <trace/events/writeback.h> |
cf0ca9fe | 14 | |
c3c53206 JA |
15 | static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); |
16 | ||
26160158 | 17 | struct backing_dev_info default_backing_dev_info = { |
d993831f | 18 | .name = "default", |
26160158 JA |
19 | .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE, |
20 | .state = 0, | |
21 | .capabilities = BDI_CAP_MAP_COPY, | |
26160158 JA |
22 | }; |
23 | EXPORT_SYMBOL_GPL(default_backing_dev_info); | |
cf0ca9fe | 24 | |
5129a469 JE |
25 | struct backing_dev_info noop_backing_dev_info = { |
26 | .name = "noop", | |
976e48f8 | 27 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
5129a469 JE |
28 | }; |
29 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); | |
30 | ||
cf0ca9fe | 31 | static struct class *bdi_class; |
cfc4ba53 JA |
32 | |
33 | /* | |
181387da | 34 | * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side |
cfc4ba53 JA |
35 | * locking. |
36 | */ | |
03ba3782 | 37 | DEFINE_SPINLOCK(bdi_lock); |
66f3b8e2 | 38 | LIST_HEAD(bdi_list); |
03ba3782 | 39 | |
839a8e86 TH |
40 | /* bdi_wq serves all asynchronous writeback tasks */ |
41 | struct workqueue_struct *bdi_wq; | |
42 | ||
f758eeab CH |
43 | void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2) |
44 | { | |
45 | if (wb1 < wb2) { | |
46 | spin_lock(&wb1->list_lock); | |
47 | spin_lock_nested(&wb2->list_lock, 1); | |
48 | } else { | |
49 | spin_lock(&wb2->list_lock); | |
50 | spin_lock_nested(&wb1->list_lock, 1); | |
51 | } | |
52 | } | |
53 | ||
76f1418b MS |
54 | #ifdef CONFIG_DEBUG_FS |
55 | #include <linux/debugfs.h> | |
56 | #include <linux/seq_file.h> | |
57 | ||
58 | static struct dentry *bdi_debug_root; | |
59 | ||
60 | static void bdi_debug_init(void) | |
61 | { | |
62 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
63 | } | |
64 | ||
65 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
66 | { | |
67 | struct backing_dev_info *bdi = m->private; | |
c1955ce3 | 68 | struct bdi_writeback *wb = &bdi->wb; |
364aeb28 DR |
69 | unsigned long background_thresh; |
70 | unsigned long dirty_thresh; | |
71 | unsigned long bdi_thresh; | |
345227d7 | 72 | unsigned long nr_dirty, nr_io, nr_more_io; |
f09b00d3 JA |
73 | struct inode *inode; |
74 | ||
345227d7 | 75 | nr_dirty = nr_io = nr_more_io = 0; |
f758eeab | 76 | spin_lock(&wb->list_lock); |
7ccf19a8 | 77 | list_for_each_entry(inode, &wb->b_dirty, i_wb_list) |
c1955ce3 | 78 | nr_dirty++; |
7ccf19a8 | 79 | list_for_each_entry(inode, &wb->b_io, i_wb_list) |
c1955ce3 | 80 | nr_io++; |
7ccf19a8 | 81 | list_for_each_entry(inode, &wb->b_more_io, i_wb_list) |
c1955ce3 | 82 | nr_more_io++; |
f758eeab | 83 | spin_unlock(&wb->list_lock); |
76f1418b | 84 | |
16c4042f WF |
85 | global_dirty_limits(&background_thresh, &dirty_thresh); |
86 | bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh); | |
76f1418b MS |
87 | |
88 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
89 | seq_printf(m, | |
00821b00 WF |
90 | "BdiWriteback: %10lu kB\n" |
91 | "BdiReclaimable: %10lu kB\n" | |
92 | "BdiDirtyThresh: %10lu kB\n" | |
93 | "DirtyThresh: %10lu kB\n" | |
94 | "BackgroundThresh: %10lu kB\n" | |
c8e28ce0 | 95 | "BdiDirtied: %10lu kB\n" |
00821b00 WF |
96 | "BdiWritten: %10lu kB\n" |
97 | "BdiWriteBandwidth: %10lu kBps\n" | |
98 | "b_dirty: %10lu\n" | |
99 | "b_io: %10lu\n" | |
100 | "b_more_io: %10lu\n" | |
101 | "bdi_list: %10u\n" | |
102 | "state: %10lx\n", | |
76f1418b MS |
103 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
104 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), | |
f7d2b1ec JK |
105 | K(bdi_thresh), |
106 | K(dirty_thresh), | |
107 | K(background_thresh), | |
c8e28ce0 | 108 | (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)), |
f7d2b1ec | 109 | (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)), |
00821b00 | 110 | (unsigned long) K(bdi->write_bandwidth), |
f7d2b1ec JK |
111 | nr_dirty, |
112 | nr_io, | |
113 | nr_more_io, | |
c1955ce3 | 114 | !list_empty(&bdi->bdi_list), bdi->state); |
76f1418b MS |
115 | #undef K |
116 | ||
117 | return 0; | |
118 | } | |
119 | ||
120 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) | |
121 | { | |
122 | return single_open(file, bdi_debug_stats_show, inode->i_private); | |
123 | } | |
124 | ||
125 | static const struct file_operations bdi_debug_stats_fops = { | |
126 | .open = bdi_debug_stats_open, | |
127 | .read = seq_read, | |
128 | .llseek = seq_lseek, | |
129 | .release = single_release, | |
130 | }; | |
131 | ||
132 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) | |
133 | { | |
134 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
135 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, | |
136 | bdi, &bdi_debug_stats_fops); | |
137 | } | |
138 | ||
139 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
140 | { | |
141 | debugfs_remove(bdi->debug_stats); | |
142 | debugfs_remove(bdi->debug_dir); | |
143 | } | |
144 | #else | |
145 | static inline void bdi_debug_init(void) | |
146 | { | |
147 | } | |
148 | static inline void bdi_debug_register(struct backing_dev_info *bdi, | |
149 | const char *name) | |
150 | { | |
151 | } | |
152 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
153 | { | |
154 | } | |
155 | #endif | |
156 | ||
cf0ca9fe PZ |
157 | static ssize_t read_ahead_kb_store(struct device *dev, |
158 | struct device_attribute *attr, | |
159 | const char *buf, size_t count) | |
160 | { | |
161 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
cf0ca9fe | 162 | unsigned long read_ahead_kb; |
7034ed13 | 163 | ssize_t ret; |
cf0ca9fe | 164 | |
7034ed13 NJ |
165 | ret = kstrtoul(buf, 10, &read_ahead_kb); |
166 | if (ret < 0) | |
167 | return ret; | |
168 | ||
169 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
170 | ||
171 | return count; | |
cf0ca9fe PZ |
172 | } |
173 | ||
174 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) | |
175 | ||
176 | #define BDI_SHOW(name, expr) \ | |
177 | static ssize_t name##_show(struct device *dev, \ | |
178 | struct device_attribute *attr, char *page) \ | |
179 | { \ | |
180 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
181 | \ | |
182 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ | |
d9e1241e GKH |
183 | } \ |
184 | static DEVICE_ATTR_RW(name); | |
cf0ca9fe PZ |
185 | |
186 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
187 | ||
189d3c4a PZ |
188 | static ssize_t min_ratio_store(struct device *dev, |
189 | struct device_attribute *attr, const char *buf, size_t count) | |
190 | { | |
191 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
189d3c4a | 192 | unsigned int ratio; |
7034ed13 NJ |
193 | ssize_t ret; |
194 | ||
195 | ret = kstrtouint(buf, 10, &ratio); | |
196 | if (ret < 0) | |
197 | return ret; | |
198 | ||
199 | ret = bdi_set_min_ratio(bdi, ratio); | |
200 | if (!ret) | |
201 | ret = count; | |
189d3c4a | 202 | |
189d3c4a PZ |
203 | return ret; |
204 | } | |
205 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
206 | ||
a42dde04 PZ |
207 | static ssize_t max_ratio_store(struct device *dev, |
208 | struct device_attribute *attr, const char *buf, size_t count) | |
209 | { | |
210 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
a42dde04 | 211 | unsigned int ratio; |
7034ed13 NJ |
212 | ssize_t ret; |
213 | ||
214 | ret = kstrtouint(buf, 10, &ratio); | |
215 | if (ret < 0) | |
216 | return ret; | |
217 | ||
218 | ret = bdi_set_max_ratio(bdi, ratio); | |
219 | if (!ret) | |
220 | ret = count; | |
a42dde04 | 221 | |
a42dde04 PZ |
222 | return ret; |
223 | } | |
224 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
225 | ||
7d311cda DW |
226 | static ssize_t stable_pages_required_show(struct device *dev, |
227 | struct device_attribute *attr, | |
228 | char *page) | |
229 | { | |
230 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
231 | ||
232 | return snprintf(page, PAGE_SIZE-1, "%d\n", | |
233 | bdi_cap_stable_pages_required(bdi) ? 1 : 0); | |
234 | } | |
d9e1241e GKH |
235 | static DEVICE_ATTR_RO(stable_pages_required); |
236 | ||
237 | static struct attribute *bdi_dev_attrs[] = { | |
238 | &dev_attr_read_ahead_kb.attr, | |
239 | &dev_attr_min_ratio.attr, | |
240 | &dev_attr_max_ratio.attr, | |
241 | &dev_attr_stable_pages_required.attr, | |
242 | NULL, | |
cf0ca9fe | 243 | }; |
d9e1241e | 244 | ATTRIBUTE_GROUPS(bdi_dev); |
cf0ca9fe PZ |
245 | |
246 | static __init int bdi_class_init(void) | |
247 | { | |
248 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
249 | if (IS_ERR(bdi_class)) |
250 | return PTR_ERR(bdi_class); | |
251 | ||
d9e1241e | 252 | bdi_class->dev_groups = bdi_dev_groups; |
76f1418b | 253 | bdi_debug_init(); |
cf0ca9fe PZ |
254 | return 0; |
255 | } | |
76f1418b | 256 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 257 | |
26160158 JA |
258 | static int __init default_bdi_init(void) |
259 | { | |
260 | int err; | |
261 | ||
839a8e86 | 262 | bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE | |
b5c872dd | 263 | WQ_UNBOUND | WQ_SYSFS, 0); |
839a8e86 TH |
264 | if (!bdi_wq) |
265 | return -ENOMEM; | |
266 | ||
26160158 JA |
267 | err = bdi_init(&default_backing_dev_info); |
268 | if (!err) | |
269 | bdi_register(&default_backing_dev_info, NULL, "default"); | |
976e48f8 | 270 | err = bdi_init(&noop_backing_dev_info); |
26160158 JA |
271 | |
272 | return err; | |
273 | } | |
274 | subsys_initcall(default_bdi_init); | |
275 | ||
03ba3782 JA |
276 | int bdi_has_dirty_io(struct backing_dev_info *bdi) |
277 | { | |
278 | return wb_has_dirty_io(&bdi->wb); | |
279 | } | |
280 | ||
6467716a AB |
281 | /* |
282 | * This function is used when the first inode for this bdi is marked dirty. It | |
283 | * wakes-up the corresponding bdi thread which should then take care of the | |
284 | * periodic background write-out of dirty inodes. Since the write-out would | |
285 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | |
286 | * set up a timer which wakes the bdi thread up later. | |
287 | * | |
288 | * Note, we wouldn't bother setting up the timer, but this function is on the | |
289 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | |
290 | * by delaying the wake-up. | |
6ca738d6 DB |
291 | * |
292 | * We have to be careful not to postpone flush work if it is scheduled for | |
293 | * earlier. Thus we use queue_delayed_work(). | |
6467716a AB |
294 | */ |
295 | void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi) | |
296 | { | |
297 | unsigned long timeout; | |
298 | ||
299 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | |
5acda9d1 JK |
300 | spin_lock_bh(&bdi->wb_lock); |
301 | if (test_bit(BDI_registered, &bdi->state)) | |
302 | queue_delayed_work(bdi_wq, &bdi->wb.dwork, timeout); | |
303 | spin_unlock_bh(&bdi->wb_lock); | |
03ba3782 JA |
304 | } |
305 | ||
cfc4ba53 JA |
306 | /* |
307 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
308 | */ | |
309 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
310 | { | |
311 | spin_lock_bh(&bdi_lock); | |
312 | list_del_rcu(&bdi->bdi_list); | |
313 | spin_unlock_bh(&bdi_lock); | |
314 | ||
ef323088 | 315 | synchronize_rcu_expedited(); |
cfc4ba53 JA |
316 | } |
317 | ||
cf0ca9fe PZ |
318 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
319 | const char *fmt, ...) | |
320 | { | |
cf0ca9fe | 321 | va_list args; |
cf0ca9fe PZ |
322 | struct device *dev; |
323 | ||
69fc208b | 324 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
c284de61 | 325 | return 0; |
f1d0b063 | 326 | |
cf0ca9fe | 327 | va_start(args, fmt); |
19051c50 | 328 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
cf0ca9fe | 329 | va_end(args); |
c284de61 AB |
330 | if (IS_ERR(dev)) |
331 | return PTR_ERR(dev); | |
66f3b8e2 | 332 | |
cf0ca9fe | 333 | bdi->dev = dev; |
cf0ca9fe | 334 | |
03ba3782 | 335 | bdi_debug_register(bdi, dev_name(dev)); |
500b067c | 336 | set_bit(BDI_registered, &bdi->state); |
c284de61 AB |
337 | |
338 | spin_lock_bh(&bdi_lock); | |
339 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
340 | spin_unlock_bh(&bdi_lock); | |
341 | ||
455b2864 | 342 | trace_writeback_bdi_register(bdi); |
c284de61 | 343 | return 0; |
cf0ca9fe PZ |
344 | } |
345 | EXPORT_SYMBOL(bdi_register); | |
346 | ||
347 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) | |
348 | { | |
349 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); | |
350 | } | |
351 | EXPORT_SYMBOL(bdi_register_dev); | |
352 | ||
03ba3782 JA |
353 | /* |
354 | * Remove bdi from the global list and shutdown any threads we have running | |
355 | */ | |
356 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) | |
66f3b8e2 | 357 | { |
03ba3782 JA |
358 | if (!bdi_cap_writeback_dirty(bdi)) |
359 | return; | |
360 | ||
361 | /* | |
fff5b85a | 362 | * Make sure nobody finds us on the bdi_list anymore |
03ba3782 | 363 | */ |
fff5b85a | 364 | bdi_remove_from_list(bdi); |
03ba3782 | 365 | |
5acda9d1 JK |
366 | /* Make sure nobody queues further work */ |
367 | spin_lock_bh(&bdi->wb_lock); | |
368 | clear_bit(BDI_registered, &bdi->state); | |
369 | spin_unlock_bh(&bdi->wb_lock); | |
370 | ||
03ba3782 | 371 | /* |
839a8e86 TH |
372 | * Drain work list and shutdown the delayed_work. At this point, |
373 | * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi | |
374 | * is dying and its work_list needs to be drained no matter what. | |
03ba3782 | 375 | */ |
839a8e86 TH |
376 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
377 | flush_delayed_work(&bdi->wb.dwork); | |
378 | WARN_ON(!list_empty(&bdi->work_list)); | |
03ba3782 JA |
379 | |
380 | /* | |
839a8e86 TH |
381 | * This shouldn't be necessary unless @bdi for some reason has |
382 | * unflushed dirty IO after work_list is drained. Do it anyway | |
383 | * just in case. | |
03ba3782 | 384 | */ |
839a8e86 | 385 | cancel_delayed_work_sync(&bdi->wb.dwork); |
66f3b8e2 JA |
386 | } |
387 | ||
592b09a4 JA |
388 | /* |
389 | * This bdi is going away now, make sure that no super_blocks point to it | |
390 | */ | |
391 | static void bdi_prune_sb(struct backing_dev_info *bdi) | |
392 | { | |
393 | struct super_block *sb; | |
394 | ||
395 | spin_lock(&sb_lock); | |
396 | list_for_each_entry(sb, &super_blocks, s_list) { | |
397 | if (sb->s_bdi == bdi) | |
95f28604 | 398 | sb->s_bdi = &default_backing_dev_info; |
592b09a4 JA |
399 | } |
400 | spin_unlock(&sb_lock); | |
401 | } | |
402 | ||
cf0ca9fe PZ |
403 | void bdi_unregister(struct backing_dev_info *bdi) |
404 | { | |
2673b4cf RV |
405 | struct device *dev = bdi->dev; |
406 | ||
407 | if (dev) { | |
ccb6108f | 408 | bdi_set_min_ratio(bdi, 0); |
455b2864 | 409 | trace_writeback_bdi_unregister(bdi); |
8c4db335 JA |
410 | bdi_prune_sb(bdi); |
411 | ||
839a8e86 | 412 | bdi_wb_shutdown(bdi); |
76f1418b | 413 | bdi_debug_unregister(bdi); |
2673b4cf RV |
414 | |
415 | spin_lock_bh(&bdi->wb_lock); | |
cf0ca9fe | 416 | bdi->dev = NULL; |
2673b4cf RV |
417 | spin_unlock_bh(&bdi->wb_lock); |
418 | ||
419 | device_unregister(dev); | |
cf0ca9fe PZ |
420 | } |
421 | } | |
422 | EXPORT_SYMBOL(bdi_unregister); | |
3fcfab16 | 423 | |
6467716a AB |
424 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
425 | { | |
426 | memset(wb, 0, sizeof(*wb)); | |
427 | ||
428 | wb->bdi = bdi; | |
429 | wb->last_old_flush = jiffies; | |
430 | INIT_LIST_HEAD(&wb->b_dirty); | |
431 | INIT_LIST_HEAD(&wb->b_io); | |
432 | INIT_LIST_HEAD(&wb->b_more_io); | |
f758eeab | 433 | spin_lock_init(&wb->list_lock); |
839a8e86 | 434 | INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn); |
6467716a AB |
435 | } |
436 | ||
e98be2d5 WF |
437 | /* |
438 | * Initial write bandwidth: 100 MB/s | |
439 | */ | |
440 | #define INIT_BW (100 << (20 - PAGE_SHIFT)) | |
441 | ||
b2e8fb6e PZ |
442 | int bdi_init(struct backing_dev_info *bdi) |
443 | { | |
03ba3782 | 444 | int i, err; |
b2e8fb6e | 445 | |
cf0ca9fe PZ |
446 | bdi->dev = NULL; |
447 | ||
189d3c4a | 448 | bdi->min_ratio = 0; |
a42dde04 | 449 | bdi->max_ratio = 100; |
eb608e3a | 450 | bdi->max_prop_frac = FPROP_FRAC_BASE; |
03ba3782 | 451 | spin_lock_init(&bdi->wb_lock); |
66f3b8e2 | 452 | INIT_LIST_HEAD(&bdi->bdi_list); |
03ba3782 JA |
453 | INIT_LIST_HEAD(&bdi->work_list); |
454 | ||
455 | bdi_wb_init(&bdi->wb, bdi); | |
456 | ||
b2e8fb6e | 457 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
ea319518 | 458 | err = percpu_counter_init(&bdi->bdi_stat[i], 0); |
04fbfdc1 PZ |
459 | if (err) |
460 | goto err; | |
461 | } | |
462 | ||
463 | bdi->dirty_exceeded = 0; | |
e98be2d5 WF |
464 | |
465 | bdi->bw_time_stamp = jiffies; | |
466 | bdi->written_stamp = 0; | |
467 | ||
7381131c | 468 | bdi->balanced_dirty_ratelimit = INIT_BW; |
be3ffa27 | 469 | bdi->dirty_ratelimit = INIT_BW; |
e98be2d5 WF |
470 | bdi->write_bandwidth = INIT_BW; |
471 | bdi->avg_write_bandwidth = INIT_BW; | |
472 | ||
eb608e3a | 473 | err = fprop_local_init_percpu(&bdi->completions); |
04fbfdc1 PZ |
474 | |
475 | if (err) { | |
476 | err: | |
4b01a0b1 | 477 | while (i--) |
04fbfdc1 | 478 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
b2e8fb6e PZ |
479 | } |
480 | ||
481 | return err; | |
482 | } | |
483 | EXPORT_SYMBOL(bdi_init); | |
484 | ||
485 | void bdi_destroy(struct backing_dev_info *bdi) | |
486 | { | |
487 | int i; | |
488 | ||
ce5f8e77 JA |
489 | /* |
490 | * Splice our entries to the default_backing_dev_info, if this | |
491 | * bdi disappears | |
492 | */ | |
493 | if (bdi_has_dirty_io(bdi)) { | |
494 | struct bdi_writeback *dst = &default_backing_dev_info.wb; | |
495 | ||
f758eeab | 496 | bdi_lock_two(&bdi->wb, dst); |
ce5f8e77 JA |
497 | list_splice(&bdi->wb.b_dirty, &dst->b_dirty); |
498 | list_splice(&bdi->wb.b_io, &dst->b_io); | |
499 | list_splice(&bdi->wb.b_more_io, &dst->b_more_io); | |
f758eeab CH |
500 | spin_unlock(&bdi->wb.list_lock); |
501 | spin_unlock(&dst->list_lock); | |
ce5f8e77 | 502 | } |
66f3b8e2 | 503 | |
cf0ca9fe PZ |
504 | bdi_unregister(bdi); |
505 | ||
7a401a97 | 506 | /* |
839a8e86 TH |
507 | * If bdi_unregister() had already been called earlier, the dwork |
508 | * could still be pending because bdi_prune_sb() can race with the | |
509 | * bdi_wakeup_thread_delayed() calls from __mark_inode_dirty(). | |
7a401a97 | 510 | */ |
839a8e86 | 511 | cancel_delayed_work_sync(&bdi->wb.dwork); |
7a401a97 | 512 | |
b2e8fb6e PZ |
513 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
514 | percpu_counter_destroy(&bdi->bdi_stat[i]); | |
04fbfdc1 | 515 | |
eb608e3a | 516 | fprop_local_destroy_percpu(&bdi->completions); |
b2e8fb6e PZ |
517 | } |
518 | EXPORT_SYMBOL(bdi_destroy); | |
519 | ||
c3c53206 JA |
520 | /* |
521 | * For use from filesystems to quickly init and register a bdi associated | |
522 | * with dirty writeback | |
523 | */ | |
524 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name, | |
525 | unsigned int cap) | |
526 | { | |
c3c53206 JA |
527 | int err; |
528 | ||
529 | bdi->name = name; | |
530 | bdi->capabilities = cap; | |
531 | err = bdi_init(bdi); | |
532 | if (err) | |
533 | return err; | |
534 | ||
02aa2a37 KC |
535 | err = bdi_register(bdi, NULL, "%.28s-%ld", name, |
536 | atomic_long_inc_return(&bdi_seq)); | |
c3c53206 JA |
537 | if (err) { |
538 | bdi_destroy(bdi); | |
539 | return err; | |
540 | } | |
541 | ||
542 | return 0; | |
543 | } | |
544 | EXPORT_SYMBOL(bdi_setup_and_register); | |
545 | ||
3fcfab16 AM |
546 | static wait_queue_head_t congestion_wqh[2] = { |
547 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
548 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
549 | }; | |
0e093d99 | 550 | static atomic_t nr_bdi_congested[2]; |
3fcfab16 | 551 | |
1faa16d2 | 552 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
553 | { |
554 | enum bdi_state bit; | |
1faa16d2 | 555 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 | 556 | |
1faa16d2 | 557 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
0e093d99 MG |
558 | if (test_and_clear_bit(bit, &bdi->state)) |
559 | atomic_dec(&nr_bdi_congested[sync]); | |
4e857c58 | 560 | smp_mb__after_atomic(); |
3fcfab16 AM |
561 | if (waitqueue_active(wqh)) |
562 | wake_up(wqh); | |
563 | } | |
564 | EXPORT_SYMBOL(clear_bdi_congested); | |
565 | ||
1faa16d2 | 566 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
567 | { |
568 | enum bdi_state bit; | |
569 | ||
1faa16d2 | 570 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
0e093d99 MG |
571 | if (!test_and_set_bit(bit, &bdi->state)) |
572 | atomic_inc(&nr_bdi_congested[sync]); | |
3fcfab16 AM |
573 | } |
574 | EXPORT_SYMBOL(set_bdi_congested); | |
575 | ||
576 | /** | |
577 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 578 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
579 | * @timeout: timeout in jiffies |
580 | * | |
581 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
582 | * write congestion. If no backing_devs are congested then just wait for the | |
583 | * next write to be completed. | |
584 | */ | |
8aa7e847 | 585 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
586 | { |
587 | long ret; | |
52bb9198 | 588 | unsigned long start = jiffies; |
3fcfab16 | 589 | DEFINE_WAIT(wait); |
8aa7e847 | 590 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
591 | |
592 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
593 | ret = io_schedule_timeout(timeout); | |
594 | finish_wait(wqh, &wait); | |
52bb9198 MG |
595 | |
596 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), | |
597 | jiffies_to_usecs(jiffies - start)); | |
598 | ||
3fcfab16 AM |
599 | return ret; |
600 | } | |
601 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 602 | |
0e093d99 MG |
603 | /** |
604 | * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes | |
605 | * @zone: A zone to check if it is heavily congested | |
606 | * @sync: SYNC or ASYNC IO | |
607 | * @timeout: timeout in jiffies | |
608 | * | |
609 | * In the event of a congested backing_dev (any backing_dev) and the given | |
610 | * @zone has experienced recent congestion, this waits for up to @timeout | |
611 | * jiffies for either a BDI to exit congestion of the given @sync queue | |
612 | * or a write to complete. | |
613 | * | |
25985edc | 614 | * In the absence of zone congestion, cond_resched() is called to yield |
0e093d99 MG |
615 | * the processor if necessary but otherwise does not sleep. |
616 | * | |
617 | * The return value is 0 if the sleep is for the full timeout. Otherwise, | |
618 | * it is the number of jiffies that were still remaining when the function | |
619 | * returned. return_value == timeout implies the function did not sleep. | |
620 | */ | |
621 | long wait_iff_congested(struct zone *zone, int sync, long timeout) | |
622 | { | |
623 | long ret; | |
624 | unsigned long start = jiffies; | |
625 | DEFINE_WAIT(wait); | |
626 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
627 | ||
628 | /* | |
629 | * If there is no congestion, or heavy congestion is not being | |
630 | * encountered in the current zone, yield if necessary instead | |
631 | * of sleeping on the congestion queue | |
632 | */ | |
633 | if (atomic_read(&nr_bdi_congested[sync]) == 0 || | |
57054651 | 634 | !test_bit(ZONE_CONGESTED, &zone->flags)) { |
0e093d99 MG |
635 | cond_resched(); |
636 | ||
637 | /* In case we scheduled, work out time remaining */ | |
638 | ret = timeout - (jiffies - start); | |
639 | if (ret < 0) | |
640 | ret = 0; | |
641 | ||
642 | goto out; | |
643 | } | |
644 | ||
645 | /* Sleep until uncongested or a write happens */ | |
646 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
647 | ret = io_schedule_timeout(timeout); | |
648 | finish_wait(wqh, &wait); | |
649 | ||
650 | out: | |
651 | trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout), | |
652 | jiffies_to_usecs(jiffies - start)); | |
653 | ||
654 | return ret; | |
655 | } | |
656 | EXPORT_SYMBOL(wait_iff_congested); | |
3965c9ae WL |
657 | |
658 | int pdflush_proc_obsolete(struct ctl_table *table, int write, | |
659 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
660 | { | |
661 | char kbuf[] = "0\n"; | |
662 | ||
4c3bffc2 | 663 | if (*ppos || *lenp < sizeof(kbuf)) { |
3965c9ae WL |
664 | *lenp = 0; |
665 | return 0; | |
666 | } | |
667 | ||
668 | if (copy_to_user(buffer, kbuf, sizeof(kbuf))) | |
669 | return -EFAULT; | |
670 | printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n", | |
671 | table->procname); | |
672 | ||
673 | *lenp = 2; | |
674 | *ppos += *lenp; | |
675 | return 2; | |
676 | } |