]>
Commit | Line | Data |
---|---|---|
1 | ||
2 | #include <linux/wait.h> | |
3 | #include <linux/backing-dev.h> | |
4 | #include <linux/kthread.h> | |
5 | #include <linux/freezer.h> | |
6 | #include <linux/fs.h> | |
7 | #include <linux/pagemap.h> | |
8 | #include <linux/mm.h> | |
9 | #include <linux/sched.h> | |
10 | #include <linux/module.h> | |
11 | #include <linux/writeback.h> | |
12 | #include <linux/device.h> | |
13 | #include <trace/events/writeback.h> | |
14 | ||
15 | static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); | |
16 | ||
17 | struct backing_dev_info default_backing_dev_info = { | |
18 | .name = "default", | |
19 | .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE, | |
20 | .state = 0, | |
21 | .capabilities = BDI_CAP_MAP_COPY, | |
22 | }; | |
23 | EXPORT_SYMBOL_GPL(default_backing_dev_info); | |
24 | ||
25 | struct backing_dev_info noop_backing_dev_info = { | |
26 | .name = "noop", | |
27 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, | |
28 | }; | |
29 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); | |
30 | ||
31 | static struct class *bdi_class; | |
32 | ||
33 | /* | |
34 | * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as | |
35 | * reader side protection for bdi_pending_list. bdi_list has RCU reader side | |
36 | * locking. | |
37 | */ | |
38 | DEFINE_SPINLOCK(bdi_lock); | |
39 | LIST_HEAD(bdi_list); | |
40 | LIST_HEAD(bdi_pending_list); | |
41 | ||
42 | static struct task_struct *sync_supers_tsk; | |
43 | static struct timer_list sync_supers_timer; | |
44 | ||
45 | static int bdi_sync_supers(void *); | |
46 | static void sync_supers_timer_fn(unsigned long); | |
47 | ||
48 | void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2) | |
49 | { | |
50 | if (wb1 < wb2) { | |
51 | spin_lock(&wb1->list_lock); | |
52 | spin_lock_nested(&wb2->list_lock, 1); | |
53 | } else { | |
54 | spin_lock(&wb2->list_lock); | |
55 | spin_lock_nested(&wb1->list_lock, 1); | |
56 | } | |
57 | } | |
58 | ||
59 | #ifdef CONFIG_DEBUG_FS | |
60 | #include <linux/debugfs.h> | |
61 | #include <linux/seq_file.h> | |
62 | ||
63 | static struct dentry *bdi_debug_root; | |
64 | ||
65 | static void bdi_debug_init(void) | |
66 | { | |
67 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
68 | } | |
69 | ||
70 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
71 | { | |
72 | struct backing_dev_info *bdi = m->private; | |
73 | struct bdi_writeback *wb = &bdi->wb; | |
74 | unsigned long background_thresh; | |
75 | unsigned long dirty_thresh; | |
76 | unsigned long bdi_thresh; | |
77 | unsigned long nr_dirty, nr_io, nr_more_io; | |
78 | struct inode *inode; | |
79 | ||
80 | nr_dirty = nr_io = nr_more_io = 0; | |
81 | spin_lock(&wb->list_lock); | |
82 | list_for_each_entry(inode, &wb->b_dirty, i_wb_list) | |
83 | nr_dirty++; | |
84 | list_for_each_entry(inode, &wb->b_io, i_wb_list) | |
85 | nr_io++; | |
86 | list_for_each_entry(inode, &wb->b_more_io, i_wb_list) | |
87 | nr_more_io++; | |
88 | spin_unlock(&wb->list_lock); | |
89 | ||
90 | global_dirty_limits(&background_thresh, &dirty_thresh); | |
91 | bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh); | |
92 | ||
93 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
94 | seq_printf(m, | |
95 | "BdiWriteback: %8lu kB\n" | |
96 | "BdiReclaimable: %8lu kB\n" | |
97 | "BdiDirtyThresh: %8lu kB\n" | |
98 | "DirtyThresh: %8lu kB\n" | |
99 | "BackgroundThresh: %8lu kB\n" | |
100 | "b_dirty: %8lu\n" | |
101 | "b_io: %8lu\n" | |
102 | "b_more_io: %8lu\n" | |
103 | "bdi_list: %8u\n" | |
104 | "state: %8lx\n", | |
105 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), | |
106 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), | |
107 | K(bdi_thresh), K(dirty_thresh), | |
108 | K(background_thresh), nr_dirty, nr_io, nr_more_io, | |
109 | !list_empty(&bdi->bdi_list), bdi->state); | |
110 | #undef K | |
111 | ||
112 | return 0; | |
113 | } | |
114 | ||
115 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) | |
116 | { | |
117 | return single_open(file, bdi_debug_stats_show, inode->i_private); | |
118 | } | |
119 | ||
120 | static const struct file_operations bdi_debug_stats_fops = { | |
121 | .open = bdi_debug_stats_open, | |
122 | .read = seq_read, | |
123 | .llseek = seq_lseek, | |
124 | .release = single_release, | |
125 | }; | |
126 | ||
127 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) | |
128 | { | |
129 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
130 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, | |
131 | bdi, &bdi_debug_stats_fops); | |
132 | } | |
133 | ||
134 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
135 | { | |
136 | debugfs_remove(bdi->debug_stats); | |
137 | debugfs_remove(bdi->debug_dir); | |
138 | } | |
139 | #else | |
140 | static inline void bdi_debug_init(void) | |
141 | { | |
142 | } | |
143 | static inline void bdi_debug_register(struct backing_dev_info *bdi, | |
144 | const char *name) | |
145 | { | |
146 | } | |
147 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
148 | { | |
149 | } | |
150 | #endif | |
151 | ||
152 | static ssize_t read_ahead_kb_store(struct device *dev, | |
153 | struct device_attribute *attr, | |
154 | const char *buf, size_t count) | |
155 | { | |
156 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
157 | char *end; | |
158 | unsigned long read_ahead_kb; | |
159 | ssize_t ret = -EINVAL; | |
160 | ||
161 | read_ahead_kb = simple_strtoul(buf, &end, 10); | |
162 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
163 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
164 | ret = count; | |
165 | } | |
166 | return ret; | |
167 | } | |
168 | ||
169 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) | |
170 | ||
171 | #define BDI_SHOW(name, expr) \ | |
172 | static ssize_t name##_show(struct device *dev, \ | |
173 | struct device_attribute *attr, char *page) \ | |
174 | { \ | |
175 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
176 | \ | |
177 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ | |
178 | } | |
179 | ||
180 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
181 | ||
182 | static ssize_t min_ratio_store(struct device *dev, | |
183 | struct device_attribute *attr, const char *buf, size_t count) | |
184 | { | |
185 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
186 | char *end; | |
187 | unsigned int ratio; | |
188 | ssize_t ret = -EINVAL; | |
189 | ||
190 | ratio = simple_strtoul(buf, &end, 10); | |
191 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
192 | ret = bdi_set_min_ratio(bdi, ratio); | |
193 | if (!ret) | |
194 | ret = count; | |
195 | } | |
196 | return ret; | |
197 | } | |
198 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
199 | ||
200 | static ssize_t max_ratio_store(struct device *dev, | |
201 | struct device_attribute *attr, const char *buf, size_t count) | |
202 | { | |
203 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
204 | char *end; | |
205 | unsigned int ratio; | |
206 | ssize_t ret = -EINVAL; | |
207 | ||
208 | ratio = simple_strtoul(buf, &end, 10); | |
209 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
210 | ret = bdi_set_max_ratio(bdi, ratio); | |
211 | if (!ret) | |
212 | ret = count; | |
213 | } | |
214 | return ret; | |
215 | } | |
216 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
217 | ||
218 | #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store) | |
219 | ||
220 | static struct device_attribute bdi_dev_attrs[] = { | |
221 | __ATTR_RW(read_ahead_kb), | |
222 | __ATTR_RW(min_ratio), | |
223 | __ATTR_RW(max_ratio), | |
224 | __ATTR_NULL, | |
225 | }; | |
226 | ||
227 | static __init int bdi_class_init(void) | |
228 | { | |
229 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
230 | if (IS_ERR(bdi_class)) | |
231 | return PTR_ERR(bdi_class); | |
232 | ||
233 | bdi_class->dev_attrs = bdi_dev_attrs; | |
234 | bdi_debug_init(); | |
235 | return 0; | |
236 | } | |
237 | postcore_initcall(bdi_class_init); | |
238 | ||
239 | static int __init default_bdi_init(void) | |
240 | { | |
241 | int err; | |
242 | ||
243 | sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers"); | |
244 | BUG_ON(IS_ERR(sync_supers_tsk)); | |
245 | ||
246 | setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0); | |
247 | bdi_arm_supers_timer(); | |
248 | ||
249 | err = bdi_init(&default_backing_dev_info); | |
250 | if (!err) | |
251 | bdi_register(&default_backing_dev_info, NULL, "default"); | |
252 | err = bdi_init(&noop_backing_dev_info); | |
253 | ||
254 | return err; | |
255 | } | |
256 | subsys_initcall(default_bdi_init); | |
257 | ||
258 | int bdi_has_dirty_io(struct backing_dev_info *bdi) | |
259 | { | |
260 | return wb_has_dirty_io(&bdi->wb); | |
261 | } | |
262 | ||
263 | static void bdi_flush_io(struct backing_dev_info *bdi) | |
264 | { | |
265 | struct writeback_control wbc = { | |
266 | .sync_mode = WB_SYNC_NONE, | |
267 | .older_than_this = NULL, | |
268 | .range_cyclic = 1, | |
269 | .nr_to_write = 1024, | |
270 | }; | |
271 | ||
272 | writeback_inodes_wb(&bdi->wb, &wbc); | |
273 | } | |
274 | ||
275 | /* | |
276 | * kupdated() used to do this. We cannot do it from the bdi_forker_thread() | |
277 | * or we risk deadlocking on ->s_umount. The longer term solution would be | |
278 | * to implement sync_supers_bdi() or similar and simply do it from the | |
279 | * bdi writeback thread individually. | |
280 | */ | |
281 | static int bdi_sync_supers(void *unused) | |
282 | { | |
283 | set_user_nice(current, 0); | |
284 | ||
285 | while (!kthread_should_stop()) { | |
286 | set_current_state(TASK_INTERRUPTIBLE); | |
287 | schedule(); | |
288 | ||
289 | /* | |
290 | * Do this periodically, like kupdated() did before. | |
291 | */ | |
292 | sync_supers(); | |
293 | } | |
294 | ||
295 | return 0; | |
296 | } | |
297 | ||
298 | void bdi_arm_supers_timer(void) | |
299 | { | |
300 | unsigned long next; | |
301 | ||
302 | if (!dirty_writeback_interval) | |
303 | return; | |
304 | ||
305 | next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies; | |
306 | mod_timer(&sync_supers_timer, round_jiffies_up(next)); | |
307 | } | |
308 | ||
309 | static void sync_supers_timer_fn(unsigned long unused) | |
310 | { | |
311 | wake_up_process(sync_supers_tsk); | |
312 | bdi_arm_supers_timer(); | |
313 | } | |
314 | ||
315 | static void wakeup_timer_fn(unsigned long data) | |
316 | { | |
317 | struct backing_dev_info *bdi = (struct backing_dev_info *)data; | |
318 | ||
319 | spin_lock_bh(&bdi->wb_lock); | |
320 | if (bdi->wb.task) { | |
321 | trace_writeback_wake_thread(bdi); | |
322 | wake_up_process(bdi->wb.task); | |
323 | } else { | |
324 | /* | |
325 | * When bdi tasks are inactive for long time, they are killed. | |
326 | * In this case we have to wake-up the forker thread which | |
327 | * should create and run the bdi thread. | |
328 | */ | |
329 | trace_writeback_wake_forker_thread(bdi); | |
330 | wake_up_process(default_backing_dev_info.wb.task); | |
331 | } | |
332 | spin_unlock_bh(&bdi->wb_lock); | |
333 | } | |
334 | ||
335 | /* | |
336 | * This function is used when the first inode for this bdi is marked dirty. It | |
337 | * wakes-up the corresponding bdi thread which should then take care of the | |
338 | * periodic background write-out of dirty inodes. Since the write-out would | |
339 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | |
340 | * set up a timer which wakes the bdi thread up later. | |
341 | * | |
342 | * Note, we wouldn't bother setting up the timer, but this function is on the | |
343 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | |
344 | * by delaying the wake-up. | |
345 | */ | |
346 | void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi) | |
347 | { | |
348 | unsigned long timeout; | |
349 | ||
350 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | |
351 | mod_timer(&bdi->wb.wakeup_timer, jiffies + timeout); | |
352 | } | |
353 | ||
354 | /* | |
355 | * Calculate the longest interval (jiffies) bdi threads are allowed to be | |
356 | * inactive. | |
357 | */ | |
358 | static unsigned long bdi_longest_inactive(void) | |
359 | { | |
360 | unsigned long interval; | |
361 | ||
362 | interval = msecs_to_jiffies(dirty_writeback_interval * 10); | |
363 | return max(5UL * 60 * HZ, interval); | |
364 | } | |
365 | ||
366 | static int bdi_forker_thread(void *ptr) | |
367 | { | |
368 | struct bdi_writeback *me = ptr; | |
369 | ||
370 | current->flags |= PF_SWAPWRITE; | |
371 | set_freezable(); | |
372 | ||
373 | /* | |
374 | * Our parent may run at a different priority, just set us to normal | |
375 | */ | |
376 | set_user_nice(current, 0); | |
377 | ||
378 | for (;;) { | |
379 | struct task_struct *task = NULL; | |
380 | struct backing_dev_info *bdi; | |
381 | enum { | |
382 | NO_ACTION, /* Nothing to do */ | |
383 | FORK_THREAD, /* Fork bdi thread */ | |
384 | KILL_THREAD, /* Kill inactive bdi thread */ | |
385 | } action = NO_ACTION; | |
386 | ||
387 | /* | |
388 | * Temporary measure, we want to make sure we don't see | |
389 | * dirty data on the default backing_dev_info | |
390 | */ | |
391 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) { | |
392 | del_timer(&me->wakeup_timer); | |
393 | wb_do_writeback(me, 0); | |
394 | } | |
395 | ||
396 | spin_lock_bh(&bdi_lock); | |
397 | set_current_state(TASK_INTERRUPTIBLE); | |
398 | ||
399 | list_for_each_entry(bdi, &bdi_list, bdi_list) { | |
400 | bool have_dirty_io; | |
401 | ||
402 | if (!bdi_cap_writeback_dirty(bdi) || | |
403 | bdi_cap_flush_forker(bdi)) | |
404 | continue; | |
405 | ||
406 | WARN(!test_bit(BDI_registered, &bdi->state), | |
407 | "bdi %p/%s is not registered!\n", bdi, bdi->name); | |
408 | ||
409 | have_dirty_io = !list_empty(&bdi->work_list) || | |
410 | wb_has_dirty_io(&bdi->wb); | |
411 | ||
412 | /* | |
413 | * If the bdi has work to do, but the thread does not | |
414 | * exist - create it. | |
415 | */ | |
416 | if (!bdi->wb.task && have_dirty_io) { | |
417 | /* | |
418 | * Set the pending bit - if someone will try to | |
419 | * unregister this bdi - it'll wait on this bit. | |
420 | */ | |
421 | set_bit(BDI_pending, &bdi->state); | |
422 | action = FORK_THREAD; | |
423 | break; | |
424 | } | |
425 | ||
426 | spin_lock(&bdi->wb_lock); | |
427 | ||
428 | /* | |
429 | * If there is no work to do and the bdi thread was | |
430 | * inactive long enough - kill it. The wb_lock is taken | |
431 | * to make sure no-one adds more work to this bdi and | |
432 | * wakes the bdi thread up. | |
433 | */ | |
434 | if (bdi->wb.task && !have_dirty_io && | |
435 | time_after(jiffies, bdi->wb.last_active + | |
436 | bdi_longest_inactive())) { | |
437 | task = bdi->wb.task; | |
438 | bdi->wb.task = NULL; | |
439 | spin_unlock(&bdi->wb_lock); | |
440 | set_bit(BDI_pending, &bdi->state); | |
441 | action = KILL_THREAD; | |
442 | break; | |
443 | } | |
444 | spin_unlock(&bdi->wb_lock); | |
445 | } | |
446 | spin_unlock_bh(&bdi_lock); | |
447 | ||
448 | /* Keep working if default bdi still has things to do */ | |
449 | if (!list_empty(&me->bdi->work_list)) | |
450 | __set_current_state(TASK_RUNNING); | |
451 | ||
452 | switch (action) { | |
453 | case FORK_THREAD: | |
454 | __set_current_state(TASK_RUNNING); | |
455 | task = kthread_create(bdi_writeback_thread, &bdi->wb, | |
456 | "flush-%s", dev_name(bdi->dev)); | |
457 | if (IS_ERR(task)) { | |
458 | /* | |
459 | * If thread creation fails, force writeout of | |
460 | * the bdi from the thread. | |
461 | */ | |
462 | bdi_flush_io(bdi); | |
463 | } else { | |
464 | /* | |
465 | * The spinlock makes sure we do not lose | |
466 | * wake-ups when racing with 'bdi_queue_work()'. | |
467 | * And as soon as the bdi thread is visible, we | |
468 | * can start it. | |
469 | */ | |
470 | spin_lock_bh(&bdi->wb_lock); | |
471 | bdi->wb.task = task; | |
472 | spin_unlock_bh(&bdi->wb_lock); | |
473 | wake_up_process(task); | |
474 | } | |
475 | break; | |
476 | ||
477 | case KILL_THREAD: | |
478 | __set_current_state(TASK_RUNNING); | |
479 | kthread_stop(task); | |
480 | break; | |
481 | ||
482 | case NO_ACTION: | |
483 | if (!wb_has_dirty_io(me) || !dirty_writeback_interval) | |
484 | /* | |
485 | * There are no dirty data. The only thing we | |
486 | * should now care about is checking for | |
487 | * inactive bdi threads and killing them. Thus, | |
488 | * let's sleep for longer time, save energy and | |
489 | * be friendly for battery-driven devices. | |
490 | */ | |
491 | schedule_timeout(bdi_longest_inactive()); | |
492 | else | |
493 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); | |
494 | try_to_freeze(); | |
495 | /* Back to the main loop */ | |
496 | continue; | |
497 | } | |
498 | ||
499 | /* | |
500 | * Clear pending bit and wakeup anybody waiting to tear us down. | |
501 | */ | |
502 | clear_bit(BDI_pending, &bdi->state); | |
503 | smp_mb__after_clear_bit(); | |
504 | wake_up_bit(&bdi->state, BDI_pending); | |
505 | } | |
506 | ||
507 | return 0; | |
508 | } | |
509 | ||
510 | /* | |
511 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
512 | */ | |
513 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
514 | { | |
515 | spin_lock_bh(&bdi_lock); | |
516 | list_del_rcu(&bdi->bdi_list); | |
517 | spin_unlock_bh(&bdi_lock); | |
518 | ||
519 | synchronize_rcu(); | |
520 | } | |
521 | ||
522 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, | |
523 | const char *fmt, ...) | |
524 | { | |
525 | va_list args; | |
526 | struct device *dev; | |
527 | ||
528 | if (bdi->dev) /* The driver needs to use separate queues per device */ | |
529 | return 0; | |
530 | ||
531 | va_start(args, fmt); | |
532 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); | |
533 | va_end(args); | |
534 | if (IS_ERR(dev)) | |
535 | return PTR_ERR(dev); | |
536 | ||
537 | bdi->dev = dev; | |
538 | ||
539 | /* | |
540 | * Just start the forker thread for our default backing_dev_info, | |
541 | * and add other bdi's to the list. They will get a thread created | |
542 | * on-demand when they need it. | |
543 | */ | |
544 | if (bdi_cap_flush_forker(bdi)) { | |
545 | struct bdi_writeback *wb = &bdi->wb; | |
546 | ||
547 | wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s", | |
548 | dev_name(dev)); | |
549 | if (IS_ERR(wb->task)) | |
550 | return PTR_ERR(wb->task); | |
551 | } | |
552 | ||
553 | bdi_debug_register(bdi, dev_name(dev)); | |
554 | set_bit(BDI_registered, &bdi->state); | |
555 | ||
556 | spin_lock_bh(&bdi_lock); | |
557 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
558 | spin_unlock_bh(&bdi_lock); | |
559 | ||
560 | trace_writeback_bdi_register(bdi); | |
561 | return 0; | |
562 | } | |
563 | EXPORT_SYMBOL(bdi_register); | |
564 | ||
565 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) | |
566 | { | |
567 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); | |
568 | } | |
569 | EXPORT_SYMBOL(bdi_register_dev); | |
570 | ||
571 | /* | |
572 | * Remove bdi from the global list and shutdown any threads we have running | |
573 | */ | |
574 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) | |
575 | { | |
576 | if (!bdi_cap_writeback_dirty(bdi)) | |
577 | return; | |
578 | ||
579 | /* | |
580 | * Make sure nobody finds us on the bdi_list anymore | |
581 | */ | |
582 | bdi_remove_from_list(bdi); | |
583 | ||
584 | /* | |
585 | * If setup is pending, wait for that to complete first | |
586 | */ | |
587 | wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait, | |
588 | TASK_UNINTERRUPTIBLE); | |
589 | ||
590 | /* | |
591 | * Finally, kill the kernel thread. We don't need to be RCU | |
592 | * safe anymore, since the bdi is gone from visibility. Force | |
593 | * unfreeze of the thread before calling kthread_stop(), otherwise | |
594 | * it would never exet if it is currently stuck in the refrigerator. | |
595 | */ | |
596 | if (bdi->wb.task) { | |
597 | thaw_process(bdi->wb.task); | |
598 | kthread_stop(bdi->wb.task); | |
599 | } | |
600 | } | |
601 | ||
602 | /* | |
603 | * This bdi is going away now, make sure that no super_blocks point to it | |
604 | */ | |
605 | static void bdi_prune_sb(struct backing_dev_info *bdi) | |
606 | { | |
607 | struct super_block *sb; | |
608 | ||
609 | spin_lock(&sb_lock); | |
610 | list_for_each_entry(sb, &super_blocks, s_list) { | |
611 | if (sb->s_bdi == bdi) | |
612 | sb->s_bdi = &default_backing_dev_info; | |
613 | } | |
614 | spin_unlock(&sb_lock); | |
615 | } | |
616 | ||
617 | void bdi_unregister(struct backing_dev_info *bdi) | |
618 | { | |
619 | if (bdi->dev) { | |
620 | trace_writeback_bdi_unregister(bdi); | |
621 | bdi_prune_sb(bdi); | |
622 | del_timer_sync(&bdi->wb.wakeup_timer); | |
623 | ||
624 | if (!bdi_cap_flush_forker(bdi)) | |
625 | bdi_wb_shutdown(bdi); | |
626 | bdi_debug_unregister(bdi); | |
627 | device_unregister(bdi->dev); | |
628 | bdi->dev = NULL; | |
629 | } | |
630 | } | |
631 | EXPORT_SYMBOL(bdi_unregister); | |
632 | ||
633 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) | |
634 | { | |
635 | memset(wb, 0, sizeof(*wb)); | |
636 | ||
637 | wb->bdi = bdi; | |
638 | wb->last_old_flush = jiffies; | |
639 | INIT_LIST_HEAD(&wb->b_dirty); | |
640 | INIT_LIST_HEAD(&wb->b_io); | |
641 | INIT_LIST_HEAD(&wb->b_more_io); | |
642 | spin_lock_init(&wb->list_lock); | |
643 | setup_timer(&wb->wakeup_timer, wakeup_timer_fn, (unsigned long)bdi); | |
644 | } | |
645 | ||
646 | int bdi_init(struct backing_dev_info *bdi) | |
647 | { | |
648 | int i, err; | |
649 | ||
650 | bdi->dev = NULL; | |
651 | ||
652 | bdi->min_ratio = 0; | |
653 | bdi->max_ratio = 100; | |
654 | bdi->max_prop_frac = PROP_FRAC_BASE; | |
655 | spin_lock_init(&bdi->wb_lock); | |
656 | INIT_LIST_HEAD(&bdi->bdi_list); | |
657 | INIT_LIST_HEAD(&bdi->work_list); | |
658 | ||
659 | bdi_wb_init(&bdi->wb, bdi); | |
660 | ||
661 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { | |
662 | err = percpu_counter_init(&bdi->bdi_stat[i], 0); | |
663 | if (err) | |
664 | goto err; | |
665 | } | |
666 | ||
667 | bdi->dirty_exceeded = 0; | |
668 | err = prop_local_init_percpu(&bdi->completions); | |
669 | ||
670 | if (err) { | |
671 | err: | |
672 | while (i--) | |
673 | percpu_counter_destroy(&bdi->bdi_stat[i]); | |
674 | } | |
675 | ||
676 | return err; | |
677 | } | |
678 | EXPORT_SYMBOL(bdi_init); | |
679 | ||
680 | void bdi_destroy(struct backing_dev_info *bdi) | |
681 | { | |
682 | int i; | |
683 | ||
684 | /* | |
685 | * Splice our entries to the default_backing_dev_info, if this | |
686 | * bdi disappears | |
687 | */ | |
688 | if (bdi_has_dirty_io(bdi)) { | |
689 | struct bdi_writeback *dst = &default_backing_dev_info.wb; | |
690 | ||
691 | bdi_lock_two(&bdi->wb, dst); | |
692 | list_splice(&bdi->wb.b_dirty, &dst->b_dirty); | |
693 | list_splice(&bdi->wb.b_io, &dst->b_io); | |
694 | list_splice(&bdi->wb.b_more_io, &dst->b_more_io); | |
695 | spin_unlock(&bdi->wb.list_lock); | |
696 | spin_unlock(&dst->list_lock); | |
697 | } | |
698 | ||
699 | bdi_unregister(bdi); | |
700 | ||
701 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) | |
702 | percpu_counter_destroy(&bdi->bdi_stat[i]); | |
703 | ||
704 | prop_local_destroy_percpu(&bdi->completions); | |
705 | } | |
706 | EXPORT_SYMBOL(bdi_destroy); | |
707 | ||
708 | /* | |
709 | * For use from filesystems to quickly init and register a bdi associated | |
710 | * with dirty writeback | |
711 | */ | |
712 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name, | |
713 | unsigned int cap) | |
714 | { | |
715 | char tmp[32]; | |
716 | int err; | |
717 | ||
718 | bdi->name = name; | |
719 | bdi->capabilities = cap; | |
720 | err = bdi_init(bdi); | |
721 | if (err) | |
722 | return err; | |
723 | ||
724 | sprintf(tmp, "%.28s%s", name, "-%d"); | |
725 | err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq)); | |
726 | if (err) { | |
727 | bdi_destroy(bdi); | |
728 | return err; | |
729 | } | |
730 | ||
731 | return 0; | |
732 | } | |
733 | EXPORT_SYMBOL(bdi_setup_and_register); | |
734 | ||
735 | static wait_queue_head_t congestion_wqh[2] = { | |
736 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
737 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
738 | }; | |
739 | static atomic_t nr_bdi_congested[2]; | |
740 | ||
741 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) | |
742 | { | |
743 | enum bdi_state bit; | |
744 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
745 | ||
746 | bit = sync ? BDI_sync_congested : BDI_async_congested; | |
747 | if (test_and_clear_bit(bit, &bdi->state)) | |
748 | atomic_dec(&nr_bdi_congested[sync]); | |
749 | smp_mb__after_clear_bit(); | |
750 | if (waitqueue_active(wqh)) | |
751 | wake_up(wqh); | |
752 | } | |
753 | EXPORT_SYMBOL(clear_bdi_congested); | |
754 | ||
755 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) | |
756 | { | |
757 | enum bdi_state bit; | |
758 | ||
759 | bit = sync ? BDI_sync_congested : BDI_async_congested; | |
760 | if (!test_and_set_bit(bit, &bdi->state)) | |
761 | atomic_inc(&nr_bdi_congested[sync]); | |
762 | } | |
763 | EXPORT_SYMBOL(set_bdi_congested); | |
764 | ||
765 | /** | |
766 | * congestion_wait - wait for a backing_dev to become uncongested | |
767 | * @sync: SYNC or ASYNC IO | |
768 | * @timeout: timeout in jiffies | |
769 | * | |
770 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
771 | * write congestion. If no backing_devs are congested then just wait for the | |
772 | * next write to be completed. | |
773 | */ | |
774 | long congestion_wait(int sync, long timeout) | |
775 | { | |
776 | long ret; | |
777 | unsigned long start = jiffies; | |
778 | DEFINE_WAIT(wait); | |
779 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
780 | ||
781 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
782 | ret = io_schedule_timeout(timeout); | |
783 | finish_wait(wqh, &wait); | |
784 | ||
785 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), | |
786 | jiffies_to_usecs(jiffies - start)); | |
787 | ||
788 | return ret; | |
789 | } | |
790 | EXPORT_SYMBOL(congestion_wait); | |
791 | ||
792 | /** | |
793 | * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes | |
794 | * @zone: A zone to check if it is heavily congested | |
795 | * @sync: SYNC or ASYNC IO | |
796 | * @timeout: timeout in jiffies | |
797 | * | |
798 | * In the event of a congested backing_dev (any backing_dev) and the given | |
799 | * @zone has experienced recent congestion, this waits for up to @timeout | |
800 | * jiffies for either a BDI to exit congestion of the given @sync queue | |
801 | * or a write to complete. | |
802 | * | |
803 | * In the absence of zone congestion, cond_resched() is called to yield | |
804 | * the processor if necessary but otherwise does not sleep. | |
805 | * | |
806 | * The return value is 0 if the sleep is for the full timeout. Otherwise, | |
807 | * it is the number of jiffies that were still remaining when the function | |
808 | * returned. return_value == timeout implies the function did not sleep. | |
809 | */ | |
810 | long wait_iff_congested(struct zone *zone, int sync, long timeout) | |
811 | { | |
812 | long ret; | |
813 | unsigned long start = jiffies; | |
814 | DEFINE_WAIT(wait); | |
815 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
816 | ||
817 | /* | |
818 | * If there is no congestion, or heavy congestion is not being | |
819 | * encountered in the current zone, yield if necessary instead | |
820 | * of sleeping on the congestion queue | |
821 | */ | |
822 | if (atomic_read(&nr_bdi_congested[sync]) == 0 || | |
823 | !zone_is_reclaim_congested(zone)) { | |
824 | cond_resched(); | |
825 | ||
826 | /* In case we scheduled, work out time remaining */ | |
827 | ret = timeout - (jiffies - start); | |
828 | if (ret < 0) | |
829 | ret = 0; | |
830 | ||
831 | goto out; | |
832 | } | |
833 | ||
834 | /* Sleep until uncongested or a write happens */ | |
835 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
836 | ret = io_schedule_timeout(timeout); | |
837 | finish_wait(wqh, &wait); | |
838 | ||
839 | out: | |
840 | trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout), | |
841 | jiffies_to_usecs(jiffies - start)); | |
842 | ||
843 | return ret; | |
844 | } | |
845 | EXPORT_SYMBOL(wait_iff_congested); |