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 | 19 | .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE, |
26160158 JA |
20 | }; |
21 | EXPORT_SYMBOL_GPL(default_backing_dev_info); | |
cf0ca9fe | 22 | |
5129a469 JE |
23 | struct backing_dev_info noop_backing_dev_info = { |
24 | .name = "noop", | |
976e48f8 | 25 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
5129a469 JE |
26 | }; |
27 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); | |
28 | ||
cf0ca9fe | 29 | static struct class *bdi_class; |
cfc4ba53 JA |
30 | |
31 | /* | |
181387da | 32 | * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side |
cfc4ba53 JA |
33 | * locking. |
34 | */ | |
03ba3782 | 35 | DEFINE_SPINLOCK(bdi_lock); |
66f3b8e2 | 36 | LIST_HEAD(bdi_list); |
03ba3782 | 37 | |
839a8e86 TH |
38 | /* bdi_wq serves all asynchronous writeback tasks */ |
39 | struct workqueue_struct *bdi_wq; | |
40 | ||
018a17bd | 41 | static void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2) |
f758eeab CH |
42 | { |
43 | if (wb1 < wb2) { | |
44 | spin_lock(&wb1->list_lock); | |
45 | spin_lock_nested(&wb2->list_lock, 1); | |
46 | } else { | |
47 | spin_lock(&wb2->list_lock); | |
48 | spin_lock_nested(&wb1->list_lock, 1); | |
49 | } | |
50 | } | |
51 | ||
76f1418b MS |
52 | #ifdef CONFIG_DEBUG_FS |
53 | #include <linux/debugfs.h> | |
54 | #include <linux/seq_file.h> | |
55 | ||
56 | static struct dentry *bdi_debug_root; | |
57 | ||
58 | static void bdi_debug_init(void) | |
59 | { | |
60 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
61 | } | |
62 | ||
63 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
64 | { | |
65 | struct backing_dev_info *bdi = m->private; | |
c1955ce3 | 66 | struct bdi_writeback *wb = &bdi->wb; |
364aeb28 DR |
67 | unsigned long background_thresh; |
68 | unsigned long dirty_thresh; | |
69 | unsigned long bdi_thresh; | |
345227d7 | 70 | unsigned long nr_dirty, nr_io, nr_more_io; |
f09b00d3 JA |
71 | struct inode *inode; |
72 | ||
345227d7 | 73 | nr_dirty = nr_io = nr_more_io = 0; |
f758eeab | 74 | spin_lock(&wb->list_lock); |
7ccf19a8 | 75 | list_for_each_entry(inode, &wb->b_dirty, i_wb_list) |
c1955ce3 | 76 | nr_dirty++; |
7ccf19a8 | 77 | list_for_each_entry(inode, &wb->b_io, i_wb_list) |
c1955ce3 | 78 | nr_io++; |
7ccf19a8 | 79 | list_for_each_entry(inode, &wb->b_more_io, i_wb_list) |
c1955ce3 | 80 | nr_more_io++; |
f758eeab | 81 | spin_unlock(&wb->list_lock); |
76f1418b | 82 | |
16c4042f WF |
83 | global_dirty_limits(&background_thresh, &dirty_thresh); |
84 | bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh); | |
76f1418b MS |
85 | |
86 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
87 | seq_printf(m, | |
00821b00 WF |
88 | "BdiWriteback: %10lu kB\n" |
89 | "BdiReclaimable: %10lu kB\n" | |
90 | "BdiDirtyThresh: %10lu kB\n" | |
91 | "DirtyThresh: %10lu kB\n" | |
92 | "BackgroundThresh: %10lu kB\n" | |
c8e28ce0 | 93 | "BdiDirtied: %10lu kB\n" |
00821b00 WF |
94 | "BdiWritten: %10lu kB\n" |
95 | "BdiWriteBandwidth: %10lu kBps\n" | |
96 | "b_dirty: %10lu\n" | |
97 | "b_io: %10lu\n" | |
98 | "b_more_io: %10lu\n" | |
99 | "bdi_list: %10u\n" | |
100 | "state: %10lx\n", | |
76f1418b MS |
101 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
102 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), | |
f7d2b1ec JK |
103 | K(bdi_thresh), |
104 | K(dirty_thresh), | |
105 | K(background_thresh), | |
c8e28ce0 | 106 | (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)), |
f7d2b1ec | 107 | (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)), |
00821b00 | 108 | (unsigned long) K(bdi->write_bandwidth), |
f7d2b1ec JK |
109 | nr_dirty, |
110 | nr_io, | |
111 | nr_more_io, | |
c1955ce3 | 112 | !list_empty(&bdi->bdi_list), bdi->state); |
76f1418b MS |
113 | #undef K |
114 | ||
115 | return 0; | |
116 | } | |
117 | ||
118 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) | |
119 | { | |
120 | return single_open(file, bdi_debug_stats_show, inode->i_private); | |
121 | } | |
122 | ||
123 | static const struct file_operations bdi_debug_stats_fops = { | |
124 | .open = bdi_debug_stats_open, | |
125 | .read = seq_read, | |
126 | .llseek = seq_lseek, | |
127 | .release = single_release, | |
128 | }; | |
129 | ||
130 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) | |
131 | { | |
132 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
133 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, | |
134 | bdi, &bdi_debug_stats_fops); | |
135 | } | |
136 | ||
137 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
138 | { | |
139 | debugfs_remove(bdi->debug_stats); | |
140 | debugfs_remove(bdi->debug_dir); | |
141 | } | |
142 | #else | |
143 | static inline void bdi_debug_init(void) | |
144 | { | |
145 | } | |
146 | static inline void bdi_debug_register(struct backing_dev_info *bdi, | |
147 | const char *name) | |
148 | { | |
149 | } | |
150 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
151 | { | |
152 | } | |
153 | #endif | |
154 | ||
cf0ca9fe PZ |
155 | static ssize_t read_ahead_kb_store(struct device *dev, |
156 | struct device_attribute *attr, | |
157 | const char *buf, size_t count) | |
158 | { | |
159 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
cf0ca9fe | 160 | unsigned long read_ahead_kb; |
7034ed13 | 161 | ssize_t ret; |
cf0ca9fe | 162 | |
7034ed13 NJ |
163 | ret = kstrtoul(buf, 10, &read_ahead_kb); |
164 | if (ret < 0) | |
165 | return ret; | |
166 | ||
167 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
168 | ||
169 | return count; | |
cf0ca9fe PZ |
170 | } |
171 | ||
172 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) | |
173 | ||
174 | #define BDI_SHOW(name, expr) \ | |
175 | static ssize_t name##_show(struct device *dev, \ | |
176 | struct device_attribute *attr, char *page) \ | |
177 | { \ | |
178 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
179 | \ | |
180 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ | |
d9e1241e GKH |
181 | } \ |
182 | static DEVICE_ATTR_RW(name); | |
cf0ca9fe PZ |
183 | |
184 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
185 | ||
189d3c4a PZ |
186 | static ssize_t min_ratio_store(struct device *dev, |
187 | struct device_attribute *attr, const char *buf, size_t count) | |
188 | { | |
189 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
189d3c4a | 190 | unsigned int ratio; |
7034ed13 NJ |
191 | ssize_t ret; |
192 | ||
193 | ret = kstrtouint(buf, 10, &ratio); | |
194 | if (ret < 0) | |
195 | return ret; | |
196 | ||
197 | ret = bdi_set_min_ratio(bdi, ratio); | |
198 | if (!ret) | |
199 | ret = count; | |
189d3c4a | 200 | |
189d3c4a PZ |
201 | return ret; |
202 | } | |
203 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
204 | ||
a42dde04 PZ |
205 | static ssize_t max_ratio_store(struct device *dev, |
206 | struct device_attribute *attr, const char *buf, size_t count) | |
207 | { | |
208 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
a42dde04 | 209 | unsigned int ratio; |
7034ed13 NJ |
210 | ssize_t ret; |
211 | ||
212 | ret = kstrtouint(buf, 10, &ratio); | |
213 | if (ret < 0) | |
214 | return ret; | |
215 | ||
216 | ret = bdi_set_max_ratio(bdi, ratio); | |
217 | if (!ret) | |
218 | ret = count; | |
a42dde04 | 219 | |
a42dde04 PZ |
220 | return ret; |
221 | } | |
222 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
223 | ||
7d311cda DW |
224 | static ssize_t stable_pages_required_show(struct device *dev, |
225 | struct device_attribute *attr, | |
226 | char *page) | |
227 | { | |
228 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
229 | ||
230 | return snprintf(page, PAGE_SIZE-1, "%d\n", | |
231 | bdi_cap_stable_pages_required(bdi) ? 1 : 0); | |
232 | } | |
d9e1241e GKH |
233 | static DEVICE_ATTR_RO(stable_pages_required); |
234 | ||
235 | static struct attribute *bdi_dev_attrs[] = { | |
236 | &dev_attr_read_ahead_kb.attr, | |
237 | &dev_attr_min_ratio.attr, | |
238 | &dev_attr_max_ratio.attr, | |
239 | &dev_attr_stable_pages_required.attr, | |
240 | NULL, | |
cf0ca9fe | 241 | }; |
d9e1241e | 242 | ATTRIBUTE_GROUPS(bdi_dev); |
cf0ca9fe PZ |
243 | |
244 | static __init int bdi_class_init(void) | |
245 | { | |
246 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
247 | if (IS_ERR(bdi_class)) |
248 | return PTR_ERR(bdi_class); | |
249 | ||
d9e1241e | 250 | bdi_class->dev_groups = bdi_dev_groups; |
76f1418b | 251 | bdi_debug_init(); |
cf0ca9fe PZ |
252 | return 0; |
253 | } | |
76f1418b | 254 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 255 | |
26160158 JA |
256 | static int __init default_bdi_init(void) |
257 | { | |
258 | int err; | |
259 | ||
839a8e86 | 260 | bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE | |
b5c872dd | 261 | WQ_UNBOUND | WQ_SYSFS, 0); |
839a8e86 TH |
262 | if (!bdi_wq) |
263 | return -ENOMEM; | |
264 | ||
26160158 JA |
265 | err = bdi_init(&default_backing_dev_info); |
266 | if (!err) | |
267 | bdi_register(&default_backing_dev_info, NULL, "default"); | |
976e48f8 | 268 | err = bdi_init(&noop_backing_dev_info); |
26160158 JA |
269 | |
270 | return err; | |
271 | } | |
272 | subsys_initcall(default_bdi_init); | |
273 | ||
03ba3782 JA |
274 | int bdi_has_dirty_io(struct backing_dev_info *bdi) |
275 | { | |
276 | return wb_has_dirty_io(&bdi->wb); | |
277 | } | |
278 | ||
6467716a AB |
279 | /* |
280 | * This function is used when the first inode for this bdi is marked dirty. It | |
281 | * wakes-up the corresponding bdi thread which should then take care of the | |
282 | * periodic background write-out of dirty inodes. Since the write-out would | |
283 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | |
284 | * set up a timer which wakes the bdi thread up later. | |
285 | * | |
286 | * Note, we wouldn't bother setting up the timer, but this function is on the | |
287 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | |
288 | * by delaying the wake-up. | |
6ca738d6 DB |
289 | * |
290 | * We have to be careful not to postpone flush work if it is scheduled for | |
291 | * earlier. Thus we use queue_delayed_work(). | |
6467716a AB |
292 | */ |
293 | void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi) | |
294 | { | |
295 | unsigned long timeout; | |
296 | ||
297 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | |
5acda9d1 JK |
298 | spin_lock_bh(&bdi->wb_lock); |
299 | if (test_bit(BDI_registered, &bdi->state)) | |
300 | queue_delayed_work(bdi_wq, &bdi->wb.dwork, timeout); | |
301 | spin_unlock_bh(&bdi->wb_lock); | |
03ba3782 JA |
302 | } |
303 | ||
cfc4ba53 JA |
304 | /* |
305 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
306 | */ | |
307 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
308 | { | |
309 | spin_lock_bh(&bdi_lock); | |
310 | list_del_rcu(&bdi->bdi_list); | |
311 | spin_unlock_bh(&bdi_lock); | |
312 | ||
ef323088 | 313 | synchronize_rcu_expedited(); |
cfc4ba53 JA |
314 | } |
315 | ||
cf0ca9fe PZ |
316 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
317 | const char *fmt, ...) | |
318 | { | |
cf0ca9fe | 319 | va_list args; |
cf0ca9fe PZ |
320 | struct device *dev; |
321 | ||
69fc208b | 322 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
c284de61 | 323 | return 0; |
f1d0b063 | 324 | |
cf0ca9fe | 325 | va_start(args, fmt); |
19051c50 | 326 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
cf0ca9fe | 327 | va_end(args); |
c284de61 AB |
328 | if (IS_ERR(dev)) |
329 | return PTR_ERR(dev); | |
66f3b8e2 | 330 | |
cf0ca9fe | 331 | bdi->dev = dev; |
cf0ca9fe | 332 | |
03ba3782 | 333 | bdi_debug_register(bdi, dev_name(dev)); |
500b067c | 334 | set_bit(BDI_registered, &bdi->state); |
c284de61 AB |
335 | |
336 | spin_lock_bh(&bdi_lock); | |
337 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
338 | spin_unlock_bh(&bdi_lock); | |
339 | ||
455b2864 | 340 | trace_writeback_bdi_register(bdi); |
c284de61 | 341 | return 0; |
cf0ca9fe PZ |
342 | } |
343 | EXPORT_SYMBOL(bdi_register); | |
344 | ||
345 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) | |
346 | { | |
347 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); | |
348 | } | |
349 | EXPORT_SYMBOL(bdi_register_dev); | |
350 | ||
03ba3782 JA |
351 | /* |
352 | * Remove bdi from the global list and shutdown any threads we have running | |
353 | */ | |
354 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) | |
66f3b8e2 | 355 | { |
03ba3782 JA |
356 | if (!bdi_cap_writeback_dirty(bdi)) |
357 | return; | |
358 | ||
359 | /* | |
fff5b85a | 360 | * Make sure nobody finds us on the bdi_list anymore |
03ba3782 | 361 | */ |
fff5b85a | 362 | bdi_remove_from_list(bdi); |
03ba3782 | 363 | |
5acda9d1 JK |
364 | /* Make sure nobody queues further work */ |
365 | spin_lock_bh(&bdi->wb_lock); | |
366 | clear_bit(BDI_registered, &bdi->state); | |
367 | spin_unlock_bh(&bdi->wb_lock); | |
368 | ||
03ba3782 | 369 | /* |
839a8e86 TH |
370 | * Drain work list and shutdown the delayed_work. At this point, |
371 | * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi | |
372 | * is dying and its work_list needs to be drained no matter what. | |
03ba3782 | 373 | */ |
839a8e86 TH |
374 | mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0); |
375 | flush_delayed_work(&bdi->wb.dwork); | |
376 | WARN_ON(!list_empty(&bdi->work_list)); | |
c0ea1c22 | 377 | WARN_ON(delayed_work_pending(&bdi->wb.dwork)); |
66f3b8e2 JA |
378 | } |
379 | ||
592b09a4 JA |
380 | /* |
381 | * This bdi is going away now, make sure that no super_blocks point to it | |
382 | */ | |
383 | static void bdi_prune_sb(struct backing_dev_info *bdi) | |
384 | { | |
385 | struct super_block *sb; | |
386 | ||
387 | spin_lock(&sb_lock); | |
388 | list_for_each_entry(sb, &super_blocks, s_list) { | |
389 | if (sb->s_bdi == bdi) | |
95f28604 | 390 | sb->s_bdi = &default_backing_dev_info; |
592b09a4 JA |
391 | } |
392 | spin_unlock(&sb_lock); | |
393 | } | |
394 | ||
cf0ca9fe PZ |
395 | void bdi_unregister(struct backing_dev_info *bdi) |
396 | { | |
b6875734 | 397 | if (bdi->dev) { |
ccb6108f | 398 | bdi_set_min_ratio(bdi, 0); |
455b2864 | 399 | trace_writeback_bdi_unregister(bdi); |
8c4db335 JA |
400 | bdi_prune_sb(bdi); |
401 | ||
839a8e86 | 402 | bdi_wb_shutdown(bdi); |
76f1418b | 403 | bdi_debug_unregister(bdi); |
b6875734 | 404 | device_unregister(bdi->dev); |
cf0ca9fe PZ |
405 | bdi->dev = NULL; |
406 | } | |
407 | } | |
408 | EXPORT_SYMBOL(bdi_unregister); | |
3fcfab16 | 409 | |
6467716a AB |
410 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
411 | { | |
412 | memset(wb, 0, sizeof(*wb)); | |
413 | ||
414 | wb->bdi = bdi; | |
415 | wb->last_old_flush = jiffies; | |
416 | INIT_LIST_HEAD(&wb->b_dirty); | |
417 | INIT_LIST_HEAD(&wb->b_io); | |
418 | INIT_LIST_HEAD(&wb->b_more_io); | |
f758eeab | 419 | spin_lock_init(&wb->list_lock); |
839a8e86 | 420 | INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn); |
6467716a AB |
421 | } |
422 | ||
e98be2d5 WF |
423 | /* |
424 | * Initial write bandwidth: 100 MB/s | |
425 | */ | |
426 | #define INIT_BW (100 << (20 - PAGE_SHIFT)) | |
427 | ||
b2e8fb6e PZ |
428 | int bdi_init(struct backing_dev_info *bdi) |
429 | { | |
03ba3782 | 430 | int i, err; |
b2e8fb6e | 431 | |
cf0ca9fe PZ |
432 | bdi->dev = NULL; |
433 | ||
189d3c4a | 434 | bdi->min_ratio = 0; |
a42dde04 | 435 | bdi->max_ratio = 100; |
eb608e3a | 436 | bdi->max_prop_frac = FPROP_FRAC_BASE; |
03ba3782 | 437 | spin_lock_init(&bdi->wb_lock); |
66f3b8e2 | 438 | INIT_LIST_HEAD(&bdi->bdi_list); |
03ba3782 JA |
439 | INIT_LIST_HEAD(&bdi->work_list); |
440 | ||
441 | bdi_wb_init(&bdi->wb, bdi); | |
442 | ||
b2e8fb6e | 443 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
908c7f19 | 444 | err = percpu_counter_init(&bdi->bdi_stat[i], 0, GFP_KERNEL); |
04fbfdc1 PZ |
445 | if (err) |
446 | goto err; | |
447 | } | |
448 | ||
449 | bdi->dirty_exceeded = 0; | |
e98be2d5 WF |
450 | |
451 | bdi->bw_time_stamp = jiffies; | |
452 | bdi->written_stamp = 0; | |
453 | ||
7381131c | 454 | bdi->balanced_dirty_ratelimit = INIT_BW; |
be3ffa27 | 455 | bdi->dirty_ratelimit = INIT_BW; |
e98be2d5 WF |
456 | bdi->write_bandwidth = INIT_BW; |
457 | bdi->avg_write_bandwidth = INIT_BW; | |
458 | ||
20ae0079 | 459 | err = fprop_local_init_percpu(&bdi->completions, GFP_KERNEL); |
04fbfdc1 PZ |
460 | |
461 | if (err) { | |
462 | err: | |
4b01a0b1 | 463 | while (i--) |
04fbfdc1 | 464 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
b2e8fb6e PZ |
465 | } |
466 | ||
467 | return err; | |
468 | } | |
469 | EXPORT_SYMBOL(bdi_init); | |
470 | ||
471 | void bdi_destroy(struct backing_dev_info *bdi) | |
472 | { | |
473 | int i; | |
474 | ||
ce5f8e77 | 475 | /* |
1a1e4530 TH |
476 | * Splice our entries to the default_backing_dev_info. This |
477 | * condition shouldn't happen. @wb must be empty at this point and | |
478 | * dirty inodes on it might cause other issues. This workaround is | |
479 | * added by ce5f8e779519 ("writeback: splice dirty inode entries to | |
480 | * default bdi on bdi_destroy()") without root-causing the issue. | |
481 | * | |
482 | * http://lkml.kernel.org/g/1253038617-30204-11-git-send-email-jens.axboe@oracle.com | |
483 | * http://thread.gmane.org/gmane.linux.file-systems/35341/focus=35350 | |
484 | * | |
485 | * We should probably add WARN_ON() to find out whether it still | |
486 | * happens and track it down if so. | |
ce5f8e77 JA |
487 | */ |
488 | if (bdi_has_dirty_io(bdi)) { | |
489 | struct bdi_writeback *dst = &default_backing_dev_info.wb; | |
490 | ||
f758eeab | 491 | bdi_lock_two(&bdi->wb, dst); |
ce5f8e77 JA |
492 | list_splice(&bdi->wb.b_dirty, &dst->b_dirty); |
493 | list_splice(&bdi->wb.b_io, &dst->b_io); | |
494 | list_splice(&bdi->wb.b_more_io, &dst->b_more_io); | |
f758eeab CH |
495 | spin_unlock(&bdi->wb.list_lock); |
496 | spin_unlock(&dst->list_lock); | |
ce5f8e77 | 497 | } |
66f3b8e2 | 498 | |
cf0ca9fe PZ |
499 | bdi_unregister(bdi); |
500 | ||
c0ea1c22 | 501 | WARN_ON(delayed_work_pending(&bdi->wb.dwork)); |
7a401a97 | 502 | |
b2e8fb6e PZ |
503 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
504 | percpu_counter_destroy(&bdi->bdi_stat[i]); | |
04fbfdc1 | 505 | |
eb608e3a | 506 | fprop_local_destroy_percpu(&bdi->completions); |
b2e8fb6e PZ |
507 | } |
508 | EXPORT_SYMBOL(bdi_destroy); | |
509 | ||
c3c53206 JA |
510 | /* |
511 | * For use from filesystems to quickly init and register a bdi associated | |
512 | * with dirty writeback | |
513 | */ | |
b4caecd4 | 514 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name) |
c3c53206 | 515 | { |
c3c53206 JA |
516 | int err; |
517 | ||
518 | bdi->name = name; | |
b4caecd4 | 519 | bdi->capabilities = 0; |
c3c53206 JA |
520 | err = bdi_init(bdi); |
521 | if (err) | |
522 | return err; | |
523 | ||
02aa2a37 KC |
524 | err = bdi_register(bdi, NULL, "%.28s-%ld", name, |
525 | atomic_long_inc_return(&bdi_seq)); | |
c3c53206 JA |
526 | if (err) { |
527 | bdi_destroy(bdi); | |
528 | return err; | |
529 | } | |
530 | ||
531 | return 0; | |
532 | } | |
533 | EXPORT_SYMBOL(bdi_setup_and_register); | |
534 | ||
3fcfab16 AM |
535 | static wait_queue_head_t congestion_wqh[2] = { |
536 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
537 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
538 | }; | |
0e093d99 | 539 | static atomic_t nr_bdi_congested[2]; |
3fcfab16 | 540 | |
1faa16d2 | 541 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
542 | { |
543 | enum bdi_state bit; | |
1faa16d2 | 544 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 | 545 | |
1faa16d2 | 546 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
0e093d99 MG |
547 | if (test_and_clear_bit(bit, &bdi->state)) |
548 | atomic_dec(&nr_bdi_congested[sync]); | |
4e857c58 | 549 | smp_mb__after_atomic(); |
3fcfab16 AM |
550 | if (waitqueue_active(wqh)) |
551 | wake_up(wqh); | |
552 | } | |
553 | EXPORT_SYMBOL(clear_bdi_congested); | |
554 | ||
1faa16d2 | 555 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
556 | { |
557 | enum bdi_state bit; | |
558 | ||
1faa16d2 | 559 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
0e093d99 MG |
560 | if (!test_and_set_bit(bit, &bdi->state)) |
561 | atomic_inc(&nr_bdi_congested[sync]); | |
3fcfab16 AM |
562 | } |
563 | EXPORT_SYMBOL(set_bdi_congested); | |
564 | ||
565 | /** | |
566 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 567 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
568 | * @timeout: timeout in jiffies |
569 | * | |
570 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
571 | * write congestion. If no backing_devs are congested then just wait for the | |
572 | * next write to be completed. | |
573 | */ | |
8aa7e847 | 574 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
575 | { |
576 | long ret; | |
52bb9198 | 577 | unsigned long start = jiffies; |
3fcfab16 | 578 | DEFINE_WAIT(wait); |
8aa7e847 | 579 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
580 | |
581 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
582 | ret = io_schedule_timeout(timeout); | |
583 | finish_wait(wqh, &wait); | |
52bb9198 MG |
584 | |
585 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), | |
586 | jiffies_to_usecs(jiffies - start)); | |
587 | ||
3fcfab16 AM |
588 | return ret; |
589 | } | |
590 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 591 | |
0e093d99 MG |
592 | /** |
593 | * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes | |
594 | * @zone: A zone to check if it is heavily congested | |
595 | * @sync: SYNC or ASYNC IO | |
596 | * @timeout: timeout in jiffies | |
597 | * | |
598 | * In the event of a congested backing_dev (any backing_dev) and the given | |
599 | * @zone has experienced recent congestion, this waits for up to @timeout | |
600 | * jiffies for either a BDI to exit congestion of the given @sync queue | |
601 | * or a write to complete. | |
602 | * | |
25985edc | 603 | * In the absence of zone congestion, cond_resched() is called to yield |
0e093d99 MG |
604 | * the processor if necessary but otherwise does not sleep. |
605 | * | |
606 | * The return value is 0 if the sleep is for the full timeout. Otherwise, | |
607 | * it is the number of jiffies that were still remaining when the function | |
608 | * returned. return_value == timeout implies the function did not sleep. | |
609 | */ | |
610 | long wait_iff_congested(struct zone *zone, int sync, long timeout) | |
611 | { | |
612 | long ret; | |
613 | unsigned long start = jiffies; | |
614 | DEFINE_WAIT(wait); | |
615 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
616 | ||
617 | /* | |
618 | * If there is no congestion, or heavy congestion is not being | |
619 | * encountered in the current zone, yield if necessary instead | |
620 | * of sleeping on the congestion queue | |
621 | */ | |
622 | if (atomic_read(&nr_bdi_congested[sync]) == 0 || | |
57054651 | 623 | !test_bit(ZONE_CONGESTED, &zone->flags)) { |
0e093d99 MG |
624 | cond_resched(); |
625 | ||
626 | /* In case we scheduled, work out time remaining */ | |
627 | ret = timeout - (jiffies - start); | |
628 | if (ret < 0) | |
629 | ret = 0; | |
630 | ||
631 | goto out; | |
632 | } | |
633 | ||
634 | /* Sleep until uncongested or a write happens */ | |
635 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
636 | ret = io_schedule_timeout(timeout); | |
637 | finish_wait(wqh, &wait); | |
638 | ||
639 | out: | |
640 | trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout), | |
641 | jiffies_to_usecs(jiffies - start)); | |
642 | ||
643 | return ret; | |
644 | } | |
645 | EXPORT_SYMBOL(wait_iff_congested); | |
3965c9ae WL |
646 | |
647 | int pdflush_proc_obsolete(struct ctl_table *table, int write, | |
648 | void __user *buffer, size_t *lenp, loff_t *ppos) | |
649 | { | |
650 | char kbuf[] = "0\n"; | |
651 | ||
4c3bffc2 | 652 | if (*ppos || *lenp < sizeof(kbuf)) { |
3965c9ae WL |
653 | *lenp = 0; |
654 | return 0; | |
655 | } | |
656 | ||
657 | if (copy_to_user(buffer, kbuf, sizeof(kbuf))) | |
658 | return -EFAULT; | |
659 | printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n", | |
660 | table->procname); | |
661 | ||
662 | *lenp = 2; | |
663 | *ppos += *lenp; | |
664 | return 2; | |
665 | } |