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> | |
13 | ||
26160158 JA |
14 | void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) |
15 | { | |
16 | } | |
17 | EXPORT_SYMBOL(default_unplug_io_fn); | |
18 | ||
19 | struct backing_dev_info default_backing_dev_info = { | |
d993831f | 20 | .name = "default", |
26160158 JA |
21 | .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE, |
22 | .state = 0, | |
23 | .capabilities = BDI_CAP_MAP_COPY, | |
24 | .unplug_io_fn = default_unplug_io_fn, | |
25 | }; | |
26 | EXPORT_SYMBOL_GPL(default_backing_dev_info); | |
cf0ca9fe PZ |
27 | |
28 | static struct class *bdi_class; | |
03ba3782 | 29 | DEFINE_SPINLOCK(bdi_lock); |
66f3b8e2 | 30 | LIST_HEAD(bdi_list); |
03ba3782 JA |
31 | LIST_HEAD(bdi_pending_list); |
32 | ||
33 | static struct task_struct *sync_supers_tsk; | |
34 | static struct timer_list sync_supers_timer; | |
35 | ||
36 | static int bdi_sync_supers(void *); | |
37 | static void sync_supers_timer_fn(unsigned long); | |
38 | static void arm_supers_timer(void); | |
39 | ||
40 | static void bdi_add_default_flusher_task(struct backing_dev_info *bdi); | |
cf0ca9fe | 41 | |
76f1418b MS |
42 | #ifdef CONFIG_DEBUG_FS |
43 | #include <linux/debugfs.h> | |
44 | #include <linux/seq_file.h> | |
45 | ||
46 | static struct dentry *bdi_debug_root; | |
47 | ||
48 | static void bdi_debug_init(void) | |
49 | { | |
50 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
51 | } | |
52 | ||
53 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
54 | { | |
55 | struct backing_dev_info *bdi = m->private; | |
f09b00d3 | 56 | struct bdi_writeback *wb; |
364aeb28 DR |
57 | unsigned long background_thresh; |
58 | unsigned long dirty_thresh; | |
59 | unsigned long bdi_thresh; | |
f09b00d3 JA |
60 | unsigned long nr_dirty, nr_io, nr_more_io, nr_wb; |
61 | struct inode *inode; | |
62 | ||
63 | /* | |
64 | * inode lock is enough here, the bdi->wb_list is protected by | |
65 | * RCU on the reader side | |
66 | */ | |
67 | nr_wb = nr_dirty = nr_io = nr_more_io = 0; | |
68 | spin_lock(&inode_lock); | |
69 | list_for_each_entry(wb, &bdi->wb_list, list) { | |
70 | nr_wb++; | |
71 | list_for_each_entry(inode, &wb->b_dirty, i_list) | |
72 | nr_dirty++; | |
73 | list_for_each_entry(inode, &wb->b_io, i_list) | |
74 | nr_io++; | |
75 | list_for_each_entry(inode, &wb->b_more_io, i_list) | |
76 | nr_more_io++; | |
77 | } | |
78 | spin_unlock(&inode_lock); | |
76f1418b MS |
79 | |
80 | get_dirty_limits(&background_thresh, &dirty_thresh, &bdi_thresh, bdi); | |
81 | ||
82 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
83 | seq_printf(m, | |
84 | "BdiWriteback: %8lu kB\n" | |
85 | "BdiReclaimable: %8lu kB\n" | |
86 | "BdiDirtyThresh: %8lu kB\n" | |
87 | "DirtyThresh: %8lu kB\n" | |
f09b00d3 JA |
88 | "BackgroundThresh: %8lu kB\n" |
89 | "WriteBack threads:%8lu\n" | |
90 | "b_dirty: %8lu\n" | |
91 | "b_io: %8lu\n" | |
92 | "b_more_io: %8lu\n" | |
93 | "bdi_list: %8u\n" | |
94 | "state: %8lx\n" | |
95 | "wb_mask: %8lx\n" | |
96 | "wb_list: %8u\n" | |
97 | "wb_cnt: %8u\n", | |
76f1418b MS |
98 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
99 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), | |
f09b00d3 JA |
100 | K(bdi_thresh), K(dirty_thresh), |
101 | K(background_thresh), nr_wb, nr_dirty, nr_io, nr_more_io, | |
102 | !list_empty(&bdi->bdi_list), bdi->state, bdi->wb_mask, | |
103 | !list_empty(&bdi->wb_list), bdi->wb_cnt); | |
76f1418b MS |
104 | #undef K |
105 | ||
106 | return 0; | |
107 | } | |
108 | ||
109 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) | |
110 | { | |
111 | return single_open(file, bdi_debug_stats_show, inode->i_private); | |
112 | } | |
113 | ||
114 | static const struct file_operations bdi_debug_stats_fops = { | |
115 | .open = bdi_debug_stats_open, | |
116 | .read = seq_read, | |
117 | .llseek = seq_lseek, | |
118 | .release = single_release, | |
119 | }; | |
120 | ||
121 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) | |
122 | { | |
123 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
124 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, | |
125 | bdi, &bdi_debug_stats_fops); | |
126 | } | |
127 | ||
128 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
129 | { | |
130 | debugfs_remove(bdi->debug_stats); | |
131 | debugfs_remove(bdi->debug_dir); | |
132 | } | |
133 | #else | |
134 | static inline void bdi_debug_init(void) | |
135 | { | |
136 | } | |
137 | static inline void bdi_debug_register(struct backing_dev_info *bdi, | |
138 | const char *name) | |
139 | { | |
140 | } | |
141 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
142 | { | |
143 | } | |
144 | #endif | |
145 | ||
cf0ca9fe PZ |
146 | static ssize_t read_ahead_kb_store(struct device *dev, |
147 | struct device_attribute *attr, | |
148 | const char *buf, size_t count) | |
149 | { | |
150 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
151 | char *end; | |
152 | unsigned long read_ahead_kb; | |
153 | ssize_t ret = -EINVAL; | |
154 | ||
155 | read_ahead_kb = simple_strtoul(buf, &end, 10); | |
156 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
157 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
158 | ret = count; | |
159 | } | |
160 | return ret; | |
161 | } | |
162 | ||
163 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) | |
164 | ||
165 | #define BDI_SHOW(name, expr) \ | |
166 | static ssize_t name##_show(struct device *dev, \ | |
167 | struct device_attribute *attr, char *page) \ | |
168 | { \ | |
169 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
170 | \ | |
171 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ | |
172 | } | |
173 | ||
174 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
175 | ||
189d3c4a PZ |
176 | static ssize_t min_ratio_store(struct device *dev, |
177 | struct device_attribute *attr, const char *buf, size_t count) | |
178 | { | |
179 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
180 | char *end; | |
181 | unsigned int ratio; | |
182 | ssize_t ret = -EINVAL; | |
183 | ||
184 | ratio = simple_strtoul(buf, &end, 10); | |
185 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
186 | ret = bdi_set_min_ratio(bdi, ratio); | |
187 | if (!ret) | |
188 | ret = count; | |
189 | } | |
190 | return ret; | |
191 | } | |
192 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
193 | ||
a42dde04 PZ |
194 | static ssize_t max_ratio_store(struct device *dev, |
195 | struct device_attribute *attr, const char *buf, size_t count) | |
196 | { | |
197 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
198 | char *end; | |
199 | unsigned int ratio; | |
200 | ssize_t ret = -EINVAL; | |
201 | ||
202 | ratio = simple_strtoul(buf, &end, 10); | |
203 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
204 | ret = bdi_set_max_ratio(bdi, ratio); | |
205 | if (!ret) | |
206 | ret = count; | |
207 | } | |
208 | return ret; | |
209 | } | |
210 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
211 | ||
cf0ca9fe PZ |
212 | #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store) |
213 | ||
214 | static struct device_attribute bdi_dev_attrs[] = { | |
215 | __ATTR_RW(read_ahead_kb), | |
189d3c4a | 216 | __ATTR_RW(min_ratio), |
a42dde04 | 217 | __ATTR_RW(max_ratio), |
cf0ca9fe PZ |
218 | __ATTR_NULL, |
219 | }; | |
220 | ||
221 | static __init int bdi_class_init(void) | |
222 | { | |
223 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
224 | bdi_class->dev_attrs = bdi_dev_attrs; | |
76f1418b | 225 | bdi_debug_init(); |
cf0ca9fe PZ |
226 | return 0; |
227 | } | |
76f1418b | 228 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 229 | |
26160158 JA |
230 | static int __init default_bdi_init(void) |
231 | { | |
232 | int err; | |
233 | ||
03ba3782 JA |
234 | sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers"); |
235 | BUG_ON(IS_ERR(sync_supers_tsk)); | |
236 | ||
237 | init_timer(&sync_supers_timer); | |
238 | setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0); | |
239 | arm_supers_timer(); | |
240 | ||
26160158 JA |
241 | err = bdi_init(&default_backing_dev_info); |
242 | if (!err) | |
243 | bdi_register(&default_backing_dev_info, NULL, "default"); | |
244 | ||
245 | return err; | |
246 | } | |
247 | subsys_initcall(default_bdi_init); | |
248 | ||
03ba3782 JA |
249 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
250 | { | |
251 | memset(wb, 0, sizeof(*wb)); | |
252 | ||
253 | wb->bdi = bdi; | |
254 | wb->last_old_flush = jiffies; | |
255 | INIT_LIST_HEAD(&wb->b_dirty); | |
256 | INIT_LIST_HEAD(&wb->b_io); | |
257 | INIT_LIST_HEAD(&wb->b_more_io); | |
258 | } | |
259 | ||
260 | static void bdi_task_init(struct backing_dev_info *bdi, | |
261 | struct bdi_writeback *wb) | |
262 | { | |
263 | struct task_struct *tsk = current; | |
264 | ||
265 | spin_lock(&bdi->wb_lock); | |
266 | list_add_tail_rcu(&wb->list, &bdi->wb_list); | |
267 | spin_unlock(&bdi->wb_lock); | |
268 | ||
269 | tsk->flags |= PF_FLUSHER | PF_SWAPWRITE; | |
270 | set_freezable(); | |
271 | ||
272 | /* | |
273 | * Our parent may run at a different priority, just set us to normal | |
274 | */ | |
275 | set_user_nice(tsk, 0); | |
276 | } | |
277 | ||
278 | static int bdi_start_fn(void *ptr) | |
279 | { | |
280 | struct bdi_writeback *wb = ptr; | |
281 | struct backing_dev_info *bdi = wb->bdi; | |
282 | int ret; | |
283 | ||
284 | /* | |
285 | * Add us to the active bdi_list | |
286 | */ | |
287 | spin_lock(&bdi_lock); | |
288 | list_add(&bdi->bdi_list, &bdi_list); | |
289 | spin_unlock(&bdi_lock); | |
290 | ||
291 | bdi_task_init(bdi, wb); | |
292 | ||
293 | /* | |
294 | * Clear pending bit and wakeup anybody waiting to tear us down | |
295 | */ | |
296 | clear_bit(BDI_pending, &bdi->state); | |
297 | smp_mb__after_clear_bit(); | |
298 | wake_up_bit(&bdi->state, BDI_pending); | |
299 | ||
300 | ret = bdi_writeback_task(wb); | |
301 | ||
302 | /* | |
303 | * Remove us from the list | |
304 | */ | |
305 | spin_lock(&bdi->wb_lock); | |
306 | list_del_rcu(&wb->list); | |
307 | spin_unlock(&bdi->wb_lock); | |
308 | ||
309 | /* | |
310 | * Flush any work that raced with us exiting. No new work | |
311 | * will be added, since this bdi isn't discoverable anymore. | |
312 | */ | |
313 | if (!list_empty(&bdi->work_list)) | |
314 | wb_do_writeback(wb, 1); | |
315 | ||
316 | wb->task = NULL; | |
317 | return ret; | |
318 | } | |
319 | ||
320 | int bdi_has_dirty_io(struct backing_dev_info *bdi) | |
321 | { | |
322 | return wb_has_dirty_io(&bdi->wb); | |
323 | } | |
324 | ||
325 | static void bdi_flush_io(struct backing_dev_info *bdi) | |
326 | { | |
327 | struct writeback_control wbc = { | |
328 | .bdi = bdi, | |
329 | .sync_mode = WB_SYNC_NONE, | |
330 | .older_than_this = NULL, | |
331 | .range_cyclic = 1, | |
332 | .nr_to_write = 1024, | |
333 | }; | |
334 | ||
335 | writeback_inodes_wbc(&wbc); | |
336 | } | |
337 | ||
338 | /* | |
339 | * kupdated() used to do this. We cannot do it from the bdi_forker_task() | |
340 | * or we risk deadlocking on ->s_umount. The longer term solution would be | |
341 | * to implement sync_supers_bdi() or similar and simply do it from the | |
342 | * bdi writeback tasks individually. | |
343 | */ | |
344 | static int bdi_sync_supers(void *unused) | |
345 | { | |
346 | set_user_nice(current, 0); | |
347 | ||
348 | while (!kthread_should_stop()) { | |
349 | set_current_state(TASK_INTERRUPTIBLE); | |
350 | schedule(); | |
351 | ||
352 | /* | |
353 | * Do this periodically, like kupdated() did before. | |
354 | */ | |
355 | sync_supers(); | |
356 | } | |
357 | ||
358 | return 0; | |
359 | } | |
360 | ||
361 | static void arm_supers_timer(void) | |
362 | { | |
363 | unsigned long next; | |
364 | ||
365 | next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies; | |
366 | mod_timer(&sync_supers_timer, round_jiffies_up(next)); | |
367 | } | |
368 | ||
369 | static void sync_supers_timer_fn(unsigned long unused) | |
370 | { | |
371 | wake_up_process(sync_supers_tsk); | |
372 | arm_supers_timer(); | |
373 | } | |
374 | ||
375 | static int bdi_forker_task(void *ptr) | |
376 | { | |
377 | struct bdi_writeback *me = ptr; | |
378 | ||
379 | bdi_task_init(me->bdi, me); | |
380 | ||
381 | for (;;) { | |
382 | struct backing_dev_info *bdi, *tmp; | |
383 | struct bdi_writeback *wb; | |
384 | ||
385 | /* | |
386 | * Temporary measure, we want to make sure we don't see | |
387 | * dirty data on the default backing_dev_info | |
388 | */ | |
389 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) | |
390 | wb_do_writeback(me, 0); | |
391 | ||
392 | spin_lock(&bdi_lock); | |
393 | ||
394 | /* | |
395 | * Check if any existing bdi's have dirty data without | |
396 | * a thread registered. If so, set that up. | |
397 | */ | |
398 | list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) { | |
399 | if (bdi->wb.task) | |
400 | continue; | |
401 | if (list_empty(&bdi->work_list) && | |
402 | !bdi_has_dirty_io(bdi)) | |
403 | continue; | |
404 | ||
405 | bdi_add_default_flusher_task(bdi); | |
406 | } | |
407 | ||
408 | set_current_state(TASK_INTERRUPTIBLE); | |
409 | ||
410 | if (list_empty(&bdi_pending_list)) { | |
411 | unsigned long wait; | |
412 | ||
413 | spin_unlock(&bdi_lock); | |
414 | wait = msecs_to_jiffies(dirty_writeback_interval * 10); | |
415 | schedule_timeout(wait); | |
416 | try_to_freeze(); | |
417 | continue; | |
418 | } | |
419 | ||
420 | __set_current_state(TASK_RUNNING); | |
421 | ||
422 | /* | |
423 | * This is our real job - check for pending entries in | |
424 | * bdi_pending_list, and create the tasks that got added | |
425 | */ | |
426 | bdi = list_entry(bdi_pending_list.next, struct backing_dev_info, | |
427 | bdi_list); | |
428 | list_del_init(&bdi->bdi_list); | |
429 | spin_unlock(&bdi_lock); | |
430 | ||
431 | wb = &bdi->wb; | |
432 | wb->task = kthread_run(bdi_start_fn, wb, "flush-%s", | |
433 | dev_name(bdi->dev)); | |
434 | /* | |
435 | * If task creation fails, then readd the bdi to | |
436 | * the pending list and force writeout of the bdi | |
437 | * from this forker thread. That will free some memory | |
438 | * and we can try again. | |
439 | */ | |
440 | if (IS_ERR(wb->task)) { | |
441 | wb->task = NULL; | |
442 | ||
443 | /* | |
444 | * Add this 'bdi' to the back, so we get | |
445 | * a chance to flush other bdi's to free | |
446 | * memory. | |
447 | */ | |
448 | spin_lock(&bdi_lock); | |
449 | list_add_tail(&bdi->bdi_list, &bdi_pending_list); | |
450 | spin_unlock(&bdi_lock); | |
451 | ||
452 | bdi_flush_io(bdi); | |
453 | } | |
454 | } | |
455 | ||
456 | return 0; | |
457 | } | |
458 | ||
459 | /* | |
460 | * Add the default flusher task that gets created for any bdi | |
461 | * that has dirty data pending writeout | |
462 | */ | |
463 | void static bdi_add_default_flusher_task(struct backing_dev_info *bdi) | |
464 | { | |
465 | if (!bdi_cap_writeback_dirty(bdi)) | |
466 | return; | |
467 | ||
500b067c JA |
468 | if (WARN_ON(!test_bit(BDI_registered, &bdi->state))) { |
469 | printk(KERN_ERR "bdi %p/%s is not registered!\n", | |
470 | bdi, bdi->name); | |
471 | return; | |
472 | } | |
473 | ||
03ba3782 JA |
474 | /* |
475 | * Check with the helper whether to proceed adding a task. Will only | |
476 | * abort if we two or more simultanous calls to | |
477 | * bdi_add_default_flusher_task() occured, further additions will block | |
478 | * waiting for previous additions to finish. | |
479 | */ | |
480 | if (!test_and_set_bit(BDI_pending, &bdi->state)) { | |
481 | list_move_tail(&bdi->bdi_list, &bdi_pending_list); | |
482 | ||
483 | /* | |
484 | * We are now on the pending list, wake up bdi_forker_task() | |
485 | * to finish the job and add us back to the active bdi_list | |
486 | */ | |
487 | wake_up_process(default_backing_dev_info.wb.task); | |
488 | } | |
489 | } | |
490 | ||
cf0ca9fe PZ |
491 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
492 | const char *fmt, ...) | |
493 | { | |
cf0ca9fe PZ |
494 | va_list args; |
495 | int ret = 0; | |
496 | struct device *dev; | |
497 | ||
69fc208b | 498 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
f1d0b063 KS |
499 | goto exit; |
500 | ||
cf0ca9fe | 501 | va_start(args, fmt); |
19051c50 | 502 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
cf0ca9fe | 503 | va_end(args); |
cf0ca9fe PZ |
504 | if (IS_ERR(dev)) { |
505 | ret = PTR_ERR(dev); | |
506 | goto exit; | |
507 | } | |
508 | ||
03ba3782 | 509 | spin_lock(&bdi_lock); |
66f3b8e2 | 510 | list_add_tail(&bdi->bdi_list, &bdi_list); |
03ba3782 | 511 | spin_unlock(&bdi_lock); |
66f3b8e2 | 512 | |
cf0ca9fe | 513 | bdi->dev = dev; |
cf0ca9fe | 514 | |
03ba3782 JA |
515 | /* |
516 | * Just start the forker thread for our default backing_dev_info, | |
517 | * and add other bdi's to the list. They will get a thread created | |
518 | * on-demand when they need it. | |
519 | */ | |
520 | if (bdi_cap_flush_forker(bdi)) { | |
521 | struct bdi_writeback *wb = &bdi->wb; | |
522 | ||
523 | wb->task = kthread_run(bdi_forker_task, wb, "bdi-%s", | |
524 | dev_name(dev)); | |
525 | if (IS_ERR(wb->task)) { | |
526 | wb->task = NULL; | |
527 | ret = -ENOMEM; | |
528 | ||
529 | spin_lock(&bdi_lock); | |
530 | list_del(&bdi->bdi_list); | |
531 | spin_unlock(&bdi_lock); | |
532 | goto exit; | |
533 | } | |
534 | } | |
535 | ||
536 | bdi_debug_register(bdi, dev_name(dev)); | |
500b067c | 537 | set_bit(BDI_registered, &bdi->state); |
cf0ca9fe | 538 | exit: |
cf0ca9fe PZ |
539 | return ret; |
540 | } | |
541 | EXPORT_SYMBOL(bdi_register); | |
542 | ||
543 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) | |
544 | { | |
545 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); | |
546 | } | |
547 | EXPORT_SYMBOL(bdi_register_dev); | |
548 | ||
03ba3782 JA |
549 | /* |
550 | * Remove bdi from the global list and shutdown any threads we have running | |
551 | */ | |
552 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) | |
66f3b8e2 | 553 | { |
03ba3782 JA |
554 | struct bdi_writeback *wb; |
555 | ||
556 | if (!bdi_cap_writeback_dirty(bdi)) | |
557 | return; | |
558 | ||
559 | /* | |
560 | * If setup is pending, wait for that to complete first | |
561 | */ | |
562 | wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait, | |
563 | TASK_UNINTERRUPTIBLE); | |
564 | ||
565 | /* | |
566 | * Make sure nobody finds us on the bdi_list anymore | |
567 | */ | |
568 | spin_lock(&bdi_lock); | |
66f3b8e2 | 569 | list_del(&bdi->bdi_list); |
03ba3782 JA |
570 | spin_unlock(&bdi_lock); |
571 | ||
572 | /* | |
573 | * Finally, kill the kernel threads. We don't need to be RCU | |
574 | * safe anymore, since the bdi is gone from visibility. | |
575 | */ | |
576 | list_for_each_entry(wb, &bdi->wb_list, list) | |
577 | kthread_stop(wb->task); | |
66f3b8e2 JA |
578 | } |
579 | ||
cf0ca9fe PZ |
580 | void bdi_unregister(struct backing_dev_info *bdi) |
581 | { | |
582 | if (bdi->dev) { | |
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); |
66f3b8e2 | 602 | INIT_LIST_HEAD(&bdi->bdi_list); |
03ba3782 JA |
603 | INIT_LIST_HEAD(&bdi->wb_list); |
604 | INIT_LIST_HEAD(&bdi->work_list); | |
605 | ||
606 | bdi_wb_init(&bdi->wb, bdi); | |
607 | ||
608 | /* | |
609 | * Just one thread support for now, hard code mask and count | |
610 | */ | |
611 | bdi->wb_mask = 1; | |
612 | bdi->wb_cnt = 1; | |
189d3c4a | 613 | |
b2e8fb6e | 614 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
ea319518 | 615 | err = percpu_counter_init(&bdi->bdi_stat[i], 0); |
04fbfdc1 PZ |
616 | if (err) |
617 | goto err; | |
618 | } | |
619 | ||
620 | bdi->dirty_exceeded = 0; | |
621 | err = prop_local_init_percpu(&bdi->completions); | |
622 | ||
623 | if (err) { | |
624 | err: | |
4b01a0b1 | 625 | while (i--) |
04fbfdc1 | 626 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
b2e8fb6e PZ |
627 | } |
628 | ||
629 | return err; | |
630 | } | |
631 | EXPORT_SYMBOL(bdi_init); | |
632 | ||
633 | void bdi_destroy(struct backing_dev_info *bdi) | |
634 | { | |
635 | int i; | |
636 | ||
03ba3782 | 637 | WARN_ON(bdi_has_dirty_io(bdi)); |
66f3b8e2 | 638 | |
cf0ca9fe PZ |
639 | bdi_unregister(bdi); |
640 | ||
b2e8fb6e PZ |
641 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
642 | percpu_counter_destroy(&bdi->bdi_stat[i]); | |
04fbfdc1 PZ |
643 | |
644 | prop_local_destroy_percpu(&bdi->completions); | |
b2e8fb6e PZ |
645 | } |
646 | EXPORT_SYMBOL(bdi_destroy); | |
647 | ||
3fcfab16 AM |
648 | static wait_queue_head_t congestion_wqh[2] = { |
649 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
650 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
651 | }; | |
652 | ||
1faa16d2 | 653 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
654 | { |
655 | enum bdi_state bit; | |
1faa16d2 | 656 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 | 657 | |
1faa16d2 | 658 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
3fcfab16 AM |
659 | clear_bit(bit, &bdi->state); |
660 | smp_mb__after_clear_bit(); | |
661 | if (waitqueue_active(wqh)) | |
662 | wake_up(wqh); | |
663 | } | |
664 | EXPORT_SYMBOL(clear_bdi_congested); | |
665 | ||
1faa16d2 | 666 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
667 | { |
668 | enum bdi_state bit; | |
669 | ||
1faa16d2 | 670 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
3fcfab16 AM |
671 | set_bit(bit, &bdi->state); |
672 | } | |
673 | EXPORT_SYMBOL(set_bdi_congested); | |
674 | ||
675 | /** | |
676 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 677 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
678 | * @timeout: timeout in jiffies |
679 | * | |
680 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
681 | * write congestion. If no backing_devs are congested then just wait for the | |
682 | * next write to be completed. | |
683 | */ | |
8aa7e847 | 684 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
685 | { |
686 | long ret; | |
687 | DEFINE_WAIT(wait); | |
8aa7e847 | 688 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
689 | |
690 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
691 | ret = io_schedule_timeout(timeout); | |
692 | finish_wait(wqh, &wait); | |
693 | return ret; | |
694 | } | |
695 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 696 |