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 JA |
17 | void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) |
18 | { | |
19 | } | |
20 | EXPORT_SYMBOL(default_unplug_io_fn); | |
21 | ||
22 | struct backing_dev_info default_backing_dev_info = { | |
d993831f | 23 | .name = "default", |
26160158 JA |
24 | .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE, |
25 | .state = 0, | |
26 | .capabilities = BDI_CAP_MAP_COPY, | |
27 | .unplug_io_fn = default_unplug_io_fn, | |
28 | }; | |
29 | EXPORT_SYMBOL_GPL(default_backing_dev_info); | |
cf0ca9fe | 30 | |
5129a469 JE |
31 | struct backing_dev_info noop_backing_dev_info = { |
32 | .name = "noop", | |
33 | }; | |
34 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); | |
35 | ||
cf0ca9fe | 36 | static struct class *bdi_class; |
cfc4ba53 JA |
37 | |
38 | /* | |
39 | * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as | |
40 | * reader side protection for bdi_pending_list. bdi_list has RCU reader side | |
41 | * locking. | |
42 | */ | |
03ba3782 | 43 | DEFINE_SPINLOCK(bdi_lock); |
66f3b8e2 | 44 | LIST_HEAD(bdi_list); |
03ba3782 JA |
45 | LIST_HEAD(bdi_pending_list); |
46 | ||
47 | static struct task_struct *sync_supers_tsk; | |
48 | static struct timer_list sync_supers_timer; | |
49 | ||
50 | static int bdi_sync_supers(void *); | |
51 | static void sync_supers_timer_fn(unsigned long); | |
03ba3782 | 52 | |
6f904ff0 | 53 | static void bdi_add_default_flusher_thread(struct backing_dev_info *bdi); |
cf0ca9fe | 54 | |
76f1418b MS |
55 | #ifdef CONFIG_DEBUG_FS |
56 | #include <linux/debugfs.h> | |
57 | #include <linux/seq_file.h> | |
58 | ||
59 | static struct dentry *bdi_debug_root; | |
60 | ||
61 | static void bdi_debug_init(void) | |
62 | { | |
63 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
64 | } | |
65 | ||
66 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
67 | { | |
68 | struct backing_dev_info *bdi = m->private; | |
c1955ce3 | 69 | struct bdi_writeback *wb = &bdi->wb; |
364aeb28 DR |
70 | unsigned long background_thresh; |
71 | unsigned long dirty_thresh; | |
72 | unsigned long bdi_thresh; | |
f09b00d3 JA |
73 | unsigned long nr_dirty, nr_io, nr_more_io, nr_wb; |
74 | struct inode *inode; | |
75 | ||
f09b00d3 JA |
76 | nr_wb = nr_dirty = nr_io = nr_more_io = 0; |
77 | spin_lock(&inode_lock); | |
c1955ce3 CH |
78 | list_for_each_entry(inode, &wb->b_dirty, i_list) |
79 | nr_dirty++; | |
80 | list_for_each_entry(inode, &wb->b_io, i_list) | |
81 | nr_io++; | |
82 | list_for_each_entry(inode, &wb->b_more_io, i_list) | |
83 | nr_more_io++; | |
f09b00d3 | 84 | spin_unlock(&inode_lock); |
76f1418b MS |
85 | |
86 | get_dirty_limits(&background_thresh, &dirty_thresh, &bdi_thresh, bdi); | |
87 | ||
88 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
89 | seq_printf(m, | |
90 | "BdiWriteback: %8lu kB\n" | |
91 | "BdiReclaimable: %8lu kB\n" | |
92 | "BdiDirtyThresh: %8lu kB\n" | |
93 | "DirtyThresh: %8lu kB\n" | |
f09b00d3 | 94 | "BackgroundThresh: %8lu kB\n" |
f09b00d3 JA |
95 | "b_dirty: %8lu\n" |
96 | "b_io: %8lu\n" | |
97 | "b_more_io: %8lu\n" | |
98 | "bdi_list: %8u\n" | |
c1955ce3 | 99 | "state: %8lx\n", |
76f1418b MS |
100 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
101 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), | |
f09b00d3 | 102 | K(bdi_thresh), K(dirty_thresh), |
c1955ce3 CH |
103 | K(background_thresh), nr_dirty, nr_io, nr_more_io, |
104 | !list_empty(&bdi->bdi_list), bdi->state); | |
76f1418b MS |
105 | #undef K |
106 | ||
107 | return 0; | |
108 | } | |
109 | ||
110 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) | |
111 | { | |
112 | return single_open(file, bdi_debug_stats_show, inode->i_private); | |
113 | } | |
114 | ||
115 | static const struct file_operations bdi_debug_stats_fops = { | |
116 | .open = bdi_debug_stats_open, | |
117 | .read = seq_read, | |
118 | .llseek = seq_lseek, | |
119 | .release = single_release, | |
120 | }; | |
121 | ||
122 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) | |
123 | { | |
124 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
125 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, | |
126 | bdi, &bdi_debug_stats_fops); | |
127 | } | |
128 | ||
129 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
130 | { | |
131 | debugfs_remove(bdi->debug_stats); | |
132 | debugfs_remove(bdi->debug_dir); | |
133 | } | |
134 | #else | |
135 | static inline void bdi_debug_init(void) | |
136 | { | |
137 | } | |
138 | static inline void bdi_debug_register(struct backing_dev_info *bdi, | |
139 | const char *name) | |
140 | { | |
141 | } | |
142 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
143 | { | |
144 | } | |
145 | #endif | |
146 | ||
cf0ca9fe PZ |
147 | static ssize_t read_ahead_kb_store(struct device *dev, |
148 | struct device_attribute *attr, | |
149 | const char *buf, size_t count) | |
150 | { | |
151 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
152 | char *end; | |
153 | unsigned long read_ahead_kb; | |
154 | ssize_t ret = -EINVAL; | |
155 | ||
156 | read_ahead_kb = simple_strtoul(buf, &end, 10); | |
157 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
158 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
159 | ret = count; | |
160 | } | |
161 | return ret; | |
162 | } | |
163 | ||
164 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) | |
165 | ||
166 | #define BDI_SHOW(name, expr) \ | |
167 | static ssize_t name##_show(struct device *dev, \ | |
168 | struct device_attribute *attr, char *page) \ | |
169 | { \ | |
170 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
171 | \ | |
172 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ | |
173 | } | |
174 | ||
175 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
176 | ||
189d3c4a PZ |
177 | static ssize_t min_ratio_store(struct device *dev, |
178 | struct device_attribute *attr, const char *buf, size_t count) | |
179 | { | |
180 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
181 | char *end; | |
182 | unsigned int ratio; | |
183 | ssize_t ret = -EINVAL; | |
184 | ||
185 | ratio = simple_strtoul(buf, &end, 10); | |
186 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
187 | ret = bdi_set_min_ratio(bdi, ratio); | |
188 | if (!ret) | |
189 | ret = count; | |
190 | } | |
191 | return ret; | |
192 | } | |
193 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
194 | ||
a42dde04 PZ |
195 | static ssize_t max_ratio_store(struct device *dev, |
196 | struct device_attribute *attr, const char *buf, size_t count) | |
197 | { | |
198 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
199 | char *end; | |
200 | unsigned int ratio; | |
201 | ssize_t ret = -EINVAL; | |
202 | ||
203 | ratio = simple_strtoul(buf, &end, 10); | |
204 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
205 | ret = bdi_set_max_ratio(bdi, ratio); | |
206 | if (!ret) | |
207 | ret = count; | |
208 | } | |
209 | return ret; | |
210 | } | |
211 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
212 | ||
cf0ca9fe PZ |
213 | #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store) |
214 | ||
215 | static struct device_attribute bdi_dev_attrs[] = { | |
216 | __ATTR_RW(read_ahead_kb), | |
189d3c4a | 217 | __ATTR_RW(min_ratio), |
a42dde04 | 218 | __ATTR_RW(max_ratio), |
cf0ca9fe PZ |
219 | __ATTR_NULL, |
220 | }; | |
221 | ||
222 | static __init int bdi_class_init(void) | |
223 | { | |
224 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
225 | if (IS_ERR(bdi_class)) |
226 | return PTR_ERR(bdi_class); | |
227 | ||
cf0ca9fe | 228 | bdi_class->dev_attrs = bdi_dev_attrs; |
76f1418b | 229 | bdi_debug_init(); |
cf0ca9fe PZ |
230 | return 0; |
231 | } | |
76f1418b | 232 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 233 | |
26160158 JA |
234 | static int __init default_bdi_init(void) |
235 | { | |
236 | int err; | |
237 | ||
03ba3782 JA |
238 | sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers"); |
239 | BUG_ON(IS_ERR(sync_supers_tsk)); | |
240 | ||
241 | init_timer(&sync_supers_timer); | |
242 | setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0); | |
6423104b | 243 | bdi_arm_supers_timer(); |
03ba3782 | 244 | |
26160158 JA |
245 | err = bdi_init(&default_backing_dev_info); |
246 | if (!err) | |
247 | bdi_register(&default_backing_dev_info, NULL, "default"); | |
248 | ||
249 | return err; | |
250 | } | |
251 | subsys_initcall(default_bdi_init); | |
252 | ||
03ba3782 JA |
253 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
254 | { | |
255 | memset(wb, 0, sizeof(*wb)); | |
256 | ||
257 | wb->bdi = bdi; | |
258 | wb->last_old_flush = jiffies; | |
259 | INIT_LIST_HEAD(&wb->b_dirty); | |
260 | INIT_LIST_HEAD(&wb->b_io); | |
261 | INIT_LIST_HEAD(&wb->b_more_io); | |
262 | } | |
263 | ||
03ba3782 JA |
264 | int bdi_has_dirty_io(struct backing_dev_info *bdi) |
265 | { | |
266 | return wb_has_dirty_io(&bdi->wb); | |
267 | } | |
268 | ||
269 | static void bdi_flush_io(struct backing_dev_info *bdi) | |
270 | { | |
271 | struct writeback_control wbc = { | |
03ba3782 JA |
272 | .sync_mode = WB_SYNC_NONE, |
273 | .older_than_this = NULL, | |
274 | .range_cyclic = 1, | |
275 | .nr_to_write = 1024, | |
276 | }; | |
277 | ||
9c3a8ee8 | 278 | writeback_inodes_wb(&bdi->wb, &wbc); |
03ba3782 JA |
279 | } |
280 | ||
281 | /* | |
6f904ff0 | 282 | * kupdated() used to do this. We cannot do it from the bdi_forker_thread() |
03ba3782 JA |
283 | * or we risk deadlocking on ->s_umount. The longer term solution would be |
284 | * to implement sync_supers_bdi() or similar and simply do it from the | |
6f904ff0 | 285 | * bdi writeback thread individually. |
03ba3782 JA |
286 | */ |
287 | static int bdi_sync_supers(void *unused) | |
288 | { | |
289 | set_user_nice(current, 0); | |
290 | ||
291 | while (!kthread_should_stop()) { | |
292 | set_current_state(TASK_INTERRUPTIBLE); | |
293 | schedule(); | |
294 | ||
295 | /* | |
296 | * Do this periodically, like kupdated() did before. | |
297 | */ | |
298 | sync_supers(); | |
299 | } | |
300 | ||
301 | return 0; | |
302 | } | |
303 | ||
6423104b | 304 | void bdi_arm_supers_timer(void) |
03ba3782 JA |
305 | { |
306 | unsigned long next; | |
307 | ||
6423104b JA |
308 | if (!dirty_writeback_interval) |
309 | return; | |
310 | ||
03ba3782 JA |
311 | next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies; |
312 | mod_timer(&sync_supers_timer, round_jiffies_up(next)); | |
313 | } | |
314 | ||
315 | static void sync_supers_timer_fn(unsigned long unused) | |
316 | { | |
317 | wake_up_process(sync_supers_tsk); | |
6423104b | 318 | bdi_arm_supers_timer(); |
03ba3782 JA |
319 | } |
320 | ||
6f904ff0 | 321 | static int bdi_forker_thread(void *ptr) |
03ba3782 JA |
322 | { |
323 | struct bdi_writeback *me = ptr; | |
324 | ||
c1955ce3 CH |
325 | current->flags |= PF_FLUSHER | PF_SWAPWRITE; |
326 | set_freezable(); | |
327 | ||
328 | /* | |
329 | * Our parent may run at a different priority, just set us to normal | |
330 | */ | |
331 | set_user_nice(current, 0); | |
03ba3782 JA |
332 | |
333 | for (;;) { | |
94eac5e6 | 334 | struct task_struct *task; |
03ba3782 | 335 | struct backing_dev_info *bdi, *tmp; |
03ba3782 JA |
336 | |
337 | /* | |
338 | * Temporary measure, we want to make sure we don't see | |
339 | * dirty data on the default backing_dev_info | |
340 | */ | |
341 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) | |
342 | wb_do_writeback(me, 0); | |
343 | ||
cfc4ba53 | 344 | spin_lock_bh(&bdi_lock); |
c5f7ad23 | 345 | set_current_state(TASK_INTERRUPTIBLE); |
03ba3782 JA |
346 | |
347 | /* | |
348 | * Check if any existing bdi's have dirty data without | |
349 | * a thread registered. If so, set that up. | |
350 | */ | |
351 | list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) { | |
352 | if (bdi->wb.task) | |
353 | continue; | |
354 | if (list_empty(&bdi->work_list) && | |
355 | !bdi_has_dirty_io(bdi)) | |
356 | continue; | |
357 | ||
6f904ff0 | 358 | bdi_add_default_flusher_thread(bdi); |
03ba3782 JA |
359 | } |
360 | ||
c4ec7908 AB |
361 | /* Keep working if default bdi still has things to do */ |
362 | if (!list_empty(&me->bdi->work_list)) | |
363 | __set_current_state(TASK_RUNNING); | |
364 | ||
03ba3782 JA |
365 | if (list_empty(&bdi_pending_list)) { |
366 | unsigned long wait; | |
367 | ||
cfc4ba53 | 368 | spin_unlock_bh(&bdi_lock); |
03ba3782 | 369 | wait = msecs_to_jiffies(dirty_writeback_interval * 10); |
6423104b JA |
370 | if (wait) |
371 | schedule_timeout(wait); | |
372 | else | |
373 | schedule(); | |
03ba3782 JA |
374 | try_to_freeze(); |
375 | continue; | |
376 | } | |
377 | ||
378 | __set_current_state(TASK_RUNNING); | |
379 | ||
380 | /* | |
381 | * This is our real job - check for pending entries in | |
6f904ff0 | 382 | * bdi_pending_list, and create the threads that got added |
03ba3782 JA |
383 | */ |
384 | bdi = list_entry(bdi_pending_list.next, struct backing_dev_info, | |
385 | bdi_list); | |
386 | list_del_init(&bdi->bdi_list); | |
cfc4ba53 | 387 | spin_unlock_bh(&bdi_lock); |
03ba3782 | 388 | |
94eac5e6 AB |
389 | task = kthread_run(bdi_writeback_thread, &bdi->wb, "flush-%s", |
390 | dev_name(bdi->dev)); | |
391 | if (IS_ERR(task)) { | |
03ba3782 | 392 | /* |
94eac5e6 AB |
393 | * If thread creation fails, then readd the bdi back to |
394 | * the list and force writeout of the bdi from this | |
395 | * forker thread. That will free some memory and we can | |
396 | * try again. Add it to the tail so we get a chance to | |
397 | * flush other bdi's to free memory. | |
03ba3782 | 398 | */ |
cfc4ba53 | 399 | spin_lock_bh(&bdi_lock); |
03ba3782 | 400 | list_add_tail(&bdi->bdi_list, &bdi_pending_list); |
cfc4ba53 | 401 | spin_unlock_bh(&bdi_lock); |
03ba3782 JA |
402 | |
403 | bdi_flush_io(bdi); | |
94eac5e6 AB |
404 | } else |
405 | bdi->wb.task = task; | |
03ba3782 JA |
406 | } |
407 | ||
408 | return 0; | |
409 | } | |
410 | ||
cfc4ba53 JA |
411 | static void bdi_add_to_pending(struct rcu_head *head) |
412 | { | |
413 | struct backing_dev_info *bdi; | |
414 | ||
415 | bdi = container_of(head, struct backing_dev_info, rcu_head); | |
416 | INIT_LIST_HEAD(&bdi->bdi_list); | |
417 | ||
418 | spin_lock(&bdi_lock); | |
419 | list_add_tail(&bdi->bdi_list, &bdi_pending_list); | |
420 | spin_unlock(&bdi_lock); | |
421 | ||
422 | /* | |
423 | * We are now on the pending list, wake up bdi_forker_task() | |
424 | * to finish the job and add us back to the active bdi_list | |
425 | */ | |
426 | wake_up_process(default_backing_dev_info.wb.task); | |
427 | } | |
428 | ||
03ba3782 | 429 | /* |
6f904ff0 | 430 | * Add the default flusher thread that gets created for any bdi |
03ba3782 JA |
431 | * that has dirty data pending writeout |
432 | */ | |
6f904ff0 | 433 | static void bdi_add_default_flusher_thread(struct backing_dev_info *bdi) |
03ba3782 JA |
434 | { |
435 | if (!bdi_cap_writeback_dirty(bdi)) | |
436 | return; | |
437 | ||
500b067c JA |
438 | if (WARN_ON(!test_bit(BDI_registered, &bdi->state))) { |
439 | printk(KERN_ERR "bdi %p/%s is not registered!\n", | |
440 | bdi, bdi->name); | |
441 | return; | |
442 | } | |
443 | ||
03ba3782 | 444 | /* |
6f904ff0 | 445 | * Check with the helper whether to proceed adding a thread. Will only |
03ba3782 | 446 | * abort if we two or more simultanous calls to |
6f904ff0 AB |
447 | * bdi_add_default_flusher_thread() occured, further additions will |
448 | * block waiting for previous additions to finish. | |
03ba3782 JA |
449 | */ |
450 | if (!test_and_set_bit(BDI_pending, &bdi->state)) { | |
cfc4ba53 | 451 | list_del_rcu(&bdi->bdi_list); |
03ba3782 JA |
452 | |
453 | /* | |
cfc4ba53 JA |
454 | * We must wait for the current RCU period to end before |
455 | * moving to the pending list. So schedule that operation | |
456 | * from an RCU callback. | |
03ba3782 | 457 | */ |
cfc4ba53 | 458 | call_rcu(&bdi->rcu_head, bdi_add_to_pending); |
03ba3782 JA |
459 | } |
460 | } | |
461 | ||
cfc4ba53 JA |
462 | /* |
463 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
464 | */ | |
465 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
466 | { | |
467 | spin_lock_bh(&bdi_lock); | |
468 | list_del_rcu(&bdi->bdi_list); | |
469 | spin_unlock_bh(&bdi_lock); | |
470 | ||
471 | synchronize_rcu(); | |
472 | } | |
473 | ||
cf0ca9fe PZ |
474 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
475 | const char *fmt, ...) | |
476 | { | |
cf0ca9fe PZ |
477 | va_list args; |
478 | int ret = 0; | |
479 | struct device *dev; | |
480 | ||
69fc208b | 481 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
f1d0b063 KS |
482 | goto exit; |
483 | ||
cf0ca9fe | 484 | va_start(args, fmt); |
19051c50 | 485 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
cf0ca9fe | 486 | va_end(args); |
cf0ca9fe PZ |
487 | if (IS_ERR(dev)) { |
488 | ret = PTR_ERR(dev); | |
489 | goto exit; | |
490 | } | |
491 | ||
cfc4ba53 JA |
492 | spin_lock_bh(&bdi_lock); |
493 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
494 | spin_unlock_bh(&bdi_lock); | |
66f3b8e2 | 495 | |
cf0ca9fe | 496 | bdi->dev = dev; |
cf0ca9fe | 497 | |
03ba3782 JA |
498 | /* |
499 | * Just start the forker thread for our default backing_dev_info, | |
500 | * and add other bdi's to the list. They will get a thread created | |
501 | * on-demand when they need it. | |
502 | */ | |
503 | if (bdi_cap_flush_forker(bdi)) { | |
504 | struct bdi_writeback *wb = &bdi->wb; | |
505 | ||
6f904ff0 | 506 | wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s", |
03ba3782 JA |
507 | dev_name(dev)); |
508 | if (IS_ERR(wb->task)) { | |
509 | wb->task = NULL; | |
510 | ret = -ENOMEM; | |
511 | ||
cfc4ba53 | 512 | bdi_remove_from_list(bdi); |
03ba3782 JA |
513 | goto exit; |
514 | } | |
515 | } | |
516 | ||
517 | bdi_debug_register(bdi, dev_name(dev)); | |
500b067c | 518 | set_bit(BDI_registered, &bdi->state); |
455b2864 | 519 | trace_writeback_bdi_register(bdi); |
cf0ca9fe | 520 | exit: |
cf0ca9fe PZ |
521 | return ret; |
522 | } | |
523 | EXPORT_SYMBOL(bdi_register); | |
524 | ||
525 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) | |
526 | { | |
527 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); | |
528 | } | |
529 | EXPORT_SYMBOL(bdi_register_dev); | |
530 | ||
03ba3782 JA |
531 | /* |
532 | * Remove bdi from the global list and shutdown any threads we have running | |
533 | */ | |
534 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) | |
66f3b8e2 | 535 | { |
03ba3782 JA |
536 | if (!bdi_cap_writeback_dirty(bdi)) |
537 | return; | |
538 | ||
539 | /* | |
540 | * If setup is pending, wait for that to complete first | |
541 | */ | |
542 | wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait, | |
543 | TASK_UNINTERRUPTIBLE); | |
544 | ||
545 | /* | |
546 | * Make sure nobody finds us on the bdi_list anymore | |
547 | */ | |
cfc4ba53 | 548 | bdi_remove_from_list(bdi); |
03ba3782 JA |
549 | |
550 | /* | |
c1955ce3 | 551 | * Finally, kill the kernel thread. We don't need to be RCU |
c62b17a5 RD |
552 | * safe anymore, since the bdi is gone from visibility. Force |
553 | * unfreeze of the thread before calling kthread_stop(), otherwise | |
554 | * it would never exet if it is currently stuck in the refrigerator. | |
03ba3782 | 555 | */ |
c1955ce3 CH |
556 | if (bdi->wb.task) { |
557 | thaw_process(bdi->wb.task); | |
558 | kthread_stop(bdi->wb.task); | |
c62b17a5 | 559 | } |
66f3b8e2 JA |
560 | } |
561 | ||
592b09a4 JA |
562 | /* |
563 | * This bdi is going away now, make sure that no super_blocks point to it | |
564 | */ | |
565 | static void bdi_prune_sb(struct backing_dev_info *bdi) | |
566 | { | |
567 | struct super_block *sb; | |
568 | ||
569 | spin_lock(&sb_lock); | |
570 | list_for_each_entry(sb, &super_blocks, s_list) { | |
571 | if (sb->s_bdi == bdi) | |
572 | sb->s_bdi = NULL; | |
573 | } | |
574 | spin_unlock(&sb_lock); | |
575 | } | |
576 | ||
cf0ca9fe PZ |
577 | void bdi_unregister(struct backing_dev_info *bdi) |
578 | { | |
579 | if (bdi->dev) { | |
455b2864 | 580 | trace_writeback_bdi_unregister(bdi); |
8c4db335 JA |
581 | bdi_prune_sb(bdi); |
582 | ||
03ba3782 JA |
583 | if (!bdi_cap_flush_forker(bdi)) |
584 | bdi_wb_shutdown(bdi); | |
76f1418b | 585 | bdi_debug_unregister(bdi); |
cf0ca9fe PZ |
586 | device_unregister(bdi->dev); |
587 | bdi->dev = NULL; | |
588 | } | |
589 | } | |
590 | EXPORT_SYMBOL(bdi_unregister); | |
3fcfab16 | 591 | |
b2e8fb6e PZ |
592 | int bdi_init(struct backing_dev_info *bdi) |
593 | { | |
03ba3782 | 594 | int i, err; |
b2e8fb6e | 595 | |
cf0ca9fe PZ |
596 | bdi->dev = NULL; |
597 | ||
189d3c4a | 598 | bdi->min_ratio = 0; |
a42dde04 PZ |
599 | bdi->max_ratio = 100; |
600 | bdi->max_prop_frac = PROP_FRAC_BASE; | |
03ba3782 | 601 | spin_lock_init(&bdi->wb_lock); |
cfc4ba53 | 602 | INIT_RCU_HEAD(&bdi->rcu_head); |
66f3b8e2 | 603 | INIT_LIST_HEAD(&bdi->bdi_list); |
03ba3782 JA |
604 | INIT_LIST_HEAD(&bdi->work_list); |
605 | ||
606 | bdi_wb_init(&bdi->wb, bdi); | |
607 | ||
b2e8fb6e | 608 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
ea319518 | 609 | err = percpu_counter_init(&bdi->bdi_stat[i], 0); |
04fbfdc1 PZ |
610 | if (err) |
611 | goto err; | |
612 | } | |
613 | ||
614 | bdi->dirty_exceeded = 0; | |
615 | err = prop_local_init_percpu(&bdi->completions); | |
616 | ||
617 | if (err) { | |
618 | err: | |
4b01a0b1 | 619 | while (i--) |
04fbfdc1 | 620 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
b2e8fb6e PZ |
621 | } |
622 | ||
623 | return err; | |
624 | } | |
625 | EXPORT_SYMBOL(bdi_init); | |
626 | ||
627 | void bdi_destroy(struct backing_dev_info *bdi) | |
628 | { | |
629 | int i; | |
630 | ||
ce5f8e77 JA |
631 | /* |
632 | * Splice our entries to the default_backing_dev_info, if this | |
633 | * bdi disappears | |
634 | */ | |
635 | if (bdi_has_dirty_io(bdi)) { | |
636 | struct bdi_writeback *dst = &default_backing_dev_info.wb; | |
637 | ||
638 | spin_lock(&inode_lock); | |
639 | list_splice(&bdi->wb.b_dirty, &dst->b_dirty); | |
640 | list_splice(&bdi->wb.b_io, &dst->b_io); | |
641 | list_splice(&bdi->wb.b_more_io, &dst->b_more_io); | |
642 | spin_unlock(&inode_lock); | |
643 | } | |
66f3b8e2 | 644 | |
cf0ca9fe PZ |
645 | bdi_unregister(bdi); |
646 | ||
b2e8fb6e PZ |
647 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
648 | percpu_counter_destroy(&bdi->bdi_stat[i]); | |
04fbfdc1 PZ |
649 | |
650 | prop_local_destroy_percpu(&bdi->completions); | |
b2e8fb6e PZ |
651 | } |
652 | EXPORT_SYMBOL(bdi_destroy); | |
653 | ||
c3c53206 JA |
654 | /* |
655 | * For use from filesystems to quickly init and register a bdi associated | |
656 | * with dirty writeback | |
657 | */ | |
658 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name, | |
659 | unsigned int cap) | |
660 | { | |
661 | char tmp[32]; | |
662 | int err; | |
663 | ||
664 | bdi->name = name; | |
665 | bdi->capabilities = cap; | |
666 | err = bdi_init(bdi); | |
667 | if (err) | |
668 | return err; | |
669 | ||
670 | sprintf(tmp, "%.28s%s", name, "-%d"); | |
671 | err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq)); | |
672 | if (err) { | |
673 | bdi_destroy(bdi); | |
674 | return err; | |
675 | } | |
676 | ||
677 | return 0; | |
678 | } | |
679 | EXPORT_SYMBOL(bdi_setup_and_register); | |
680 | ||
3fcfab16 AM |
681 | static wait_queue_head_t congestion_wqh[2] = { |
682 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
683 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
684 | }; | |
685 | ||
1faa16d2 | 686 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
687 | { |
688 | enum bdi_state bit; | |
1faa16d2 | 689 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 | 690 | |
1faa16d2 | 691 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
3fcfab16 AM |
692 | clear_bit(bit, &bdi->state); |
693 | smp_mb__after_clear_bit(); | |
694 | if (waitqueue_active(wqh)) | |
695 | wake_up(wqh); | |
696 | } | |
697 | EXPORT_SYMBOL(clear_bdi_congested); | |
698 | ||
1faa16d2 | 699 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
700 | { |
701 | enum bdi_state bit; | |
702 | ||
1faa16d2 | 703 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
3fcfab16 AM |
704 | set_bit(bit, &bdi->state); |
705 | } | |
706 | EXPORT_SYMBOL(set_bdi_congested); | |
707 | ||
708 | /** | |
709 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 710 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
711 | * @timeout: timeout in jiffies |
712 | * | |
713 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
714 | * write congestion. If no backing_devs are congested then just wait for the | |
715 | * next write to be completed. | |
716 | */ | |
8aa7e847 | 717 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
718 | { |
719 | long ret; | |
720 | DEFINE_WAIT(wait); | |
8aa7e847 | 721 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
722 | |
723 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
724 | ret = io_schedule_timeout(timeout); | |
725 | finish_wait(wqh, &wait); | |
726 | return ret; | |
727 | } | |
728 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 729 |