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; | |
cfc4ba53 JA |
29 | |
30 | /* | |
31 | * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as | |
32 | * reader side protection for bdi_pending_list. bdi_list has RCU reader side | |
33 | * locking. | |
34 | */ | |
03ba3782 | 35 | DEFINE_SPINLOCK(bdi_lock); |
66f3b8e2 | 36 | LIST_HEAD(bdi_list); |
03ba3782 JA |
37 | LIST_HEAD(bdi_pending_list); |
38 | ||
39 | static struct task_struct *sync_supers_tsk; | |
40 | static struct timer_list sync_supers_timer; | |
41 | ||
42 | static int bdi_sync_supers(void *); | |
43 | static void sync_supers_timer_fn(unsigned long); | |
44 | static void arm_supers_timer(void); | |
45 | ||
46 | static void bdi_add_default_flusher_task(struct backing_dev_info *bdi); | |
cf0ca9fe | 47 | |
76f1418b MS |
48 | #ifdef CONFIG_DEBUG_FS |
49 | #include <linux/debugfs.h> | |
50 | #include <linux/seq_file.h> | |
51 | ||
52 | static struct dentry *bdi_debug_root; | |
53 | ||
54 | static void bdi_debug_init(void) | |
55 | { | |
56 | bdi_debug_root = debugfs_create_dir("bdi", NULL); | |
57 | } | |
58 | ||
59 | static int bdi_debug_stats_show(struct seq_file *m, void *v) | |
60 | { | |
61 | struct backing_dev_info *bdi = m->private; | |
f09b00d3 | 62 | struct bdi_writeback *wb; |
364aeb28 DR |
63 | unsigned long background_thresh; |
64 | unsigned long dirty_thresh; | |
65 | unsigned long bdi_thresh; | |
f09b00d3 JA |
66 | unsigned long nr_dirty, nr_io, nr_more_io, nr_wb; |
67 | struct inode *inode; | |
68 | ||
69 | /* | |
70 | * inode lock is enough here, the bdi->wb_list is protected by | |
71 | * RCU on the reader side | |
72 | */ | |
73 | nr_wb = nr_dirty = nr_io = nr_more_io = 0; | |
74 | spin_lock(&inode_lock); | |
75 | list_for_each_entry(wb, &bdi->wb_list, list) { | |
76 | nr_wb++; | |
77 | list_for_each_entry(inode, &wb->b_dirty, i_list) | |
78 | nr_dirty++; | |
79 | list_for_each_entry(inode, &wb->b_io, i_list) | |
80 | nr_io++; | |
81 | list_for_each_entry(inode, &wb->b_more_io, i_list) | |
82 | nr_more_io++; | |
83 | } | |
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" |
961515f6 | 95 | "WritebackThreads: %8lu\n" |
f09b00d3 JA |
96 | "b_dirty: %8lu\n" |
97 | "b_io: %8lu\n" | |
98 | "b_more_io: %8lu\n" | |
99 | "bdi_list: %8u\n" | |
100 | "state: %8lx\n" | |
101 | "wb_mask: %8lx\n" | |
102 | "wb_list: %8u\n" | |
103 | "wb_cnt: %8u\n", | |
76f1418b MS |
104 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
105 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), | |
f09b00d3 JA |
106 | K(bdi_thresh), K(dirty_thresh), |
107 | K(background_thresh), nr_wb, nr_dirty, nr_io, nr_more_io, | |
108 | !list_empty(&bdi->bdi_list), bdi->state, bdi->wb_mask, | |
109 | !list_empty(&bdi->wb_list), bdi->wb_cnt); | |
76f1418b MS |
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 | ||
cf0ca9fe PZ |
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 | ||
189d3c4a PZ |
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 | ||
a42dde04 PZ |
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 | ||
cf0ca9fe PZ |
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), | |
189d3c4a | 222 | __ATTR_RW(min_ratio), |
a42dde04 | 223 | __ATTR_RW(max_ratio), |
cf0ca9fe PZ |
224 | __ATTR_NULL, |
225 | }; | |
226 | ||
227 | static __init int bdi_class_init(void) | |
228 | { | |
229 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
230 | bdi_class->dev_attrs = bdi_dev_attrs; | |
76f1418b | 231 | bdi_debug_init(); |
cf0ca9fe PZ |
232 | return 0; |
233 | } | |
76f1418b | 234 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 235 | |
26160158 JA |
236 | static int __init default_bdi_init(void) |
237 | { | |
238 | int err; | |
239 | ||
03ba3782 JA |
240 | sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers"); |
241 | BUG_ON(IS_ERR(sync_supers_tsk)); | |
242 | ||
243 | init_timer(&sync_supers_timer); | |
244 | setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0); | |
245 | arm_supers_timer(); | |
246 | ||
26160158 JA |
247 | err = bdi_init(&default_backing_dev_info); |
248 | if (!err) | |
249 | bdi_register(&default_backing_dev_info, NULL, "default"); | |
250 | ||
251 | return err; | |
252 | } | |
253 | subsys_initcall(default_bdi_init); | |
254 | ||
03ba3782 JA |
255 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
256 | { | |
257 | memset(wb, 0, sizeof(*wb)); | |
258 | ||
259 | wb->bdi = bdi; | |
260 | wb->last_old_flush = jiffies; | |
261 | INIT_LIST_HEAD(&wb->b_dirty); | |
262 | INIT_LIST_HEAD(&wb->b_io); | |
263 | INIT_LIST_HEAD(&wb->b_more_io); | |
264 | } | |
265 | ||
266 | static void bdi_task_init(struct backing_dev_info *bdi, | |
267 | struct bdi_writeback *wb) | |
268 | { | |
269 | struct task_struct *tsk = current; | |
270 | ||
271 | spin_lock(&bdi->wb_lock); | |
272 | list_add_tail_rcu(&wb->list, &bdi->wb_list); | |
273 | spin_unlock(&bdi->wb_lock); | |
274 | ||
275 | tsk->flags |= PF_FLUSHER | PF_SWAPWRITE; | |
276 | set_freezable(); | |
277 | ||
278 | /* | |
279 | * Our parent may run at a different priority, just set us to normal | |
280 | */ | |
281 | set_user_nice(tsk, 0); | |
282 | } | |
283 | ||
284 | static int bdi_start_fn(void *ptr) | |
285 | { | |
286 | struct bdi_writeback *wb = ptr; | |
287 | struct backing_dev_info *bdi = wb->bdi; | |
288 | int ret; | |
289 | ||
290 | /* | |
291 | * Add us to the active bdi_list | |
292 | */ | |
cfc4ba53 JA |
293 | spin_lock_bh(&bdi_lock); |
294 | list_add_rcu(&bdi->bdi_list, &bdi_list); | |
295 | spin_unlock_bh(&bdi_lock); | |
03ba3782 JA |
296 | |
297 | bdi_task_init(bdi, wb); | |
298 | ||
299 | /* | |
300 | * Clear pending bit and wakeup anybody waiting to tear us down | |
301 | */ | |
302 | clear_bit(BDI_pending, &bdi->state); | |
303 | smp_mb__after_clear_bit(); | |
304 | wake_up_bit(&bdi->state, BDI_pending); | |
305 | ||
306 | ret = bdi_writeback_task(wb); | |
307 | ||
308 | /* | |
309 | * Remove us from the list | |
310 | */ | |
311 | spin_lock(&bdi->wb_lock); | |
312 | list_del_rcu(&wb->list); | |
313 | spin_unlock(&bdi->wb_lock); | |
314 | ||
315 | /* | |
316 | * Flush any work that raced with us exiting. No new work | |
317 | * will be added, since this bdi isn't discoverable anymore. | |
318 | */ | |
319 | if (!list_empty(&bdi->work_list)) | |
320 | wb_do_writeback(wb, 1); | |
321 | ||
322 | wb->task = NULL; | |
323 | return ret; | |
324 | } | |
325 | ||
326 | int bdi_has_dirty_io(struct backing_dev_info *bdi) | |
327 | { | |
328 | return wb_has_dirty_io(&bdi->wb); | |
329 | } | |
330 | ||
331 | static void bdi_flush_io(struct backing_dev_info *bdi) | |
332 | { | |
333 | struct writeback_control wbc = { | |
334 | .bdi = bdi, | |
335 | .sync_mode = WB_SYNC_NONE, | |
336 | .older_than_this = NULL, | |
337 | .range_cyclic = 1, | |
338 | .nr_to_write = 1024, | |
339 | }; | |
340 | ||
341 | writeback_inodes_wbc(&wbc); | |
342 | } | |
343 | ||
344 | /* | |
345 | * kupdated() used to do this. We cannot do it from the bdi_forker_task() | |
346 | * or we risk deadlocking on ->s_umount. The longer term solution would be | |
347 | * to implement sync_supers_bdi() or similar and simply do it from the | |
348 | * bdi writeback tasks individually. | |
349 | */ | |
350 | static int bdi_sync_supers(void *unused) | |
351 | { | |
352 | set_user_nice(current, 0); | |
353 | ||
354 | while (!kthread_should_stop()) { | |
355 | set_current_state(TASK_INTERRUPTIBLE); | |
356 | schedule(); | |
357 | ||
358 | /* | |
359 | * Do this periodically, like kupdated() did before. | |
360 | */ | |
361 | sync_supers(); | |
362 | } | |
363 | ||
364 | return 0; | |
365 | } | |
366 | ||
367 | static void arm_supers_timer(void) | |
368 | { | |
369 | unsigned long next; | |
370 | ||
371 | next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies; | |
372 | mod_timer(&sync_supers_timer, round_jiffies_up(next)); | |
373 | } | |
374 | ||
375 | static void sync_supers_timer_fn(unsigned long unused) | |
376 | { | |
377 | wake_up_process(sync_supers_tsk); | |
378 | arm_supers_timer(); | |
379 | } | |
380 | ||
381 | static int bdi_forker_task(void *ptr) | |
382 | { | |
383 | struct bdi_writeback *me = ptr; | |
384 | ||
385 | bdi_task_init(me->bdi, me); | |
386 | ||
387 | for (;;) { | |
388 | struct backing_dev_info *bdi, *tmp; | |
389 | struct bdi_writeback *wb; | |
390 | ||
391 | /* | |
392 | * Temporary measure, we want to make sure we don't see | |
393 | * dirty data on the default backing_dev_info | |
394 | */ | |
395 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) | |
396 | wb_do_writeback(me, 0); | |
397 | ||
cfc4ba53 | 398 | spin_lock_bh(&bdi_lock); |
03ba3782 JA |
399 | |
400 | /* | |
401 | * Check if any existing bdi's have dirty data without | |
402 | * a thread registered. If so, set that up. | |
403 | */ | |
404 | list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) { | |
405 | if (bdi->wb.task) | |
406 | continue; | |
407 | if (list_empty(&bdi->work_list) && | |
408 | !bdi_has_dirty_io(bdi)) | |
409 | continue; | |
410 | ||
411 | bdi_add_default_flusher_task(bdi); | |
412 | } | |
413 | ||
414 | set_current_state(TASK_INTERRUPTIBLE); | |
415 | ||
416 | if (list_empty(&bdi_pending_list)) { | |
417 | unsigned long wait; | |
418 | ||
cfc4ba53 | 419 | spin_unlock_bh(&bdi_lock); |
03ba3782 JA |
420 | wait = msecs_to_jiffies(dirty_writeback_interval * 10); |
421 | schedule_timeout(wait); | |
422 | try_to_freeze(); | |
423 | continue; | |
424 | } | |
425 | ||
426 | __set_current_state(TASK_RUNNING); | |
427 | ||
428 | /* | |
429 | * This is our real job - check for pending entries in | |
430 | * bdi_pending_list, and create the tasks that got added | |
431 | */ | |
432 | bdi = list_entry(bdi_pending_list.next, struct backing_dev_info, | |
433 | bdi_list); | |
434 | list_del_init(&bdi->bdi_list); | |
cfc4ba53 | 435 | spin_unlock_bh(&bdi_lock); |
03ba3782 JA |
436 | |
437 | wb = &bdi->wb; | |
438 | wb->task = kthread_run(bdi_start_fn, wb, "flush-%s", | |
439 | dev_name(bdi->dev)); | |
440 | /* | |
441 | * If task creation fails, then readd the bdi to | |
442 | * the pending list and force writeout of the bdi | |
443 | * from this forker thread. That will free some memory | |
444 | * and we can try again. | |
445 | */ | |
446 | if (IS_ERR(wb->task)) { | |
447 | wb->task = NULL; | |
448 | ||
449 | /* | |
450 | * Add this 'bdi' to the back, so we get | |
451 | * a chance to flush other bdi's to free | |
452 | * memory. | |
453 | */ | |
cfc4ba53 | 454 | spin_lock_bh(&bdi_lock); |
03ba3782 | 455 | list_add_tail(&bdi->bdi_list, &bdi_pending_list); |
cfc4ba53 | 456 | spin_unlock_bh(&bdi_lock); |
03ba3782 JA |
457 | |
458 | bdi_flush_io(bdi); | |
459 | } | |
460 | } | |
461 | ||
462 | return 0; | |
463 | } | |
464 | ||
cfc4ba53 JA |
465 | static void bdi_add_to_pending(struct rcu_head *head) |
466 | { | |
467 | struct backing_dev_info *bdi; | |
468 | ||
469 | bdi = container_of(head, struct backing_dev_info, rcu_head); | |
470 | INIT_LIST_HEAD(&bdi->bdi_list); | |
471 | ||
472 | spin_lock(&bdi_lock); | |
473 | list_add_tail(&bdi->bdi_list, &bdi_pending_list); | |
474 | spin_unlock(&bdi_lock); | |
475 | ||
476 | /* | |
477 | * We are now on the pending list, wake up bdi_forker_task() | |
478 | * to finish the job and add us back to the active bdi_list | |
479 | */ | |
480 | wake_up_process(default_backing_dev_info.wb.task); | |
481 | } | |
482 | ||
03ba3782 JA |
483 | /* |
484 | * Add the default flusher task that gets created for any bdi | |
485 | * that has dirty data pending writeout | |
486 | */ | |
487 | void static bdi_add_default_flusher_task(struct backing_dev_info *bdi) | |
488 | { | |
489 | if (!bdi_cap_writeback_dirty(bdi)) | |
490 | return; | |
491 | ||
500b067c JA |
492 | if (WARN_ON(!test_bit(BDI_registered, &bdi->state))) { |
493 | printk(KERN_ERR "bdi %p/%s is not registered!\n", | |
494 | bdi, bdi->name); | |
495 | return; | |
496 | } | |
497 | ||
03ba3782 JA |
498 | /* |
499 | * Check with the helper whether to proceed adding a task. Will only | |
500 | * abort if we two or more simultanous calls to | |
501 | * bdi_add_default_flusher_task() occured, further additions will block | |
502 | * waiting for previous additions to finish. | |
503 | */ | |
504 | if (!test_and_set_bit(BDI_pending, &bdi->state)) { | |
cfc4ba53 | 505 | list_del_rcu(&bdi->bdi_list); |
03ba3782 JA |
506 | |
507 | /* | |
cfc4ba53 JA |
508 | * We must wait for the current RCU period to end before |
509 | * moving to the pending list. So schedule that operation | |
510 | * from an RCU callback. | |
03ba3782 | 511 | */ |
cfc4ba53 | 512 | call_rcu(&bdi->rcu_head, bdi_add_to_pending); |
03ba3782 JA |
513 | } |
514 | } | |
515 | ||
cfc4ba53 JA |
516 | /* |
517 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
518 | */ | |
519 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
520 | { | |
521 | spin_lock_bh(&bdi_lock); | |
522 | list_del_rcu(&bdi->bdi_list); | |
523 | spin_unlock_bh(&bdi_lock); | |
524 | ||
525 | synchronize_rcu(); | |
526 | } | |
527 | ||
cf0ca9fe PZ |
528 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
529 | const char *fmt, ...) | |
530 | { | |
cf0ca9fe PZ |
531 | va_list args; |
532 | int ret = 0; | |
533 | struct device *dev; | |
534 | ||
69fc208b | 535 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
f1d0b063 KS |
536 | goto exit; |
537 | ||
cf0ca9fe | 538 | va_start(args, fmt); |
19051c50 | 539 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
cf0ca9fe | 540 | va_end(args); |
cf0ca9fe PZ |
541 | if (IS_ERR(dev)) { |
542 | ret = PTR_ERR(dev); | |
543 | goto exit; | |
544 | } | |
545 | ||
cfc4ba53 JA |
546 | spin_lock_bh(&bdi_lock); |
547 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
548 | spin_unlock_bh(&bdi_lock); | |
66f3b8e2 | 549 | |
cf0ca9fe | 550 | bdi->dev = dev; |
cf0ca9fe | 551 | |
03ba3782 JA |
552 | /* |
553 | * Just start the forker thread for our default backing_dev_info, | |
554 | * and add other bdi's to the list. They will get a thread created | |
555 | * on-demand when they need it. | |
556 | */ | |
557 | if (bdi_cap_flush_forker(bdi)) { | |
558 | struct bdi_writeback *wb = &bdi->wb; | |
559 | ||
560 | wb->task = kthread_run(bdi_forker_task, wb, "bdi-%s", | |
561 | dev_name(dev)); | |
562 | if (IS_ERR(wb->task)) { | |
563 | wb->task = NULL; | |
564 | ret = -ENOMEM; | |
565 | ||
cfc4ba53 | 566 | bdi_remove_from_list(bdi); |
03ba3782 JA |
567 | goto exit; |
568 | } | |
569 | } | |
570 | ||
571 | bdi_debug_register(bdi, dev_name(dev)); | |
500b067c | 572 | set_bit(BDI_registered, &bdi->state); |
cf0ca9fe | 573 | exit: |
cf0ca9fe PZ |
574 | return ret; |
575 | } | |
576 | EXPORT_SYMBOL(bdi_register); | |
577 | ||
578 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) | |
579 | { | |
580 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); | |
581 | } | |
582 | EXPORT_SYMBOL(bdi_register_dev); | |
583 | ||
03ba3782 JA |
584 | /* |
585 | * Remove bdi from the global list and shutdown any threads we have running | |
586 | */ | |
587 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) | |
66f3b8e2 | 588 | { |
03ba3782 JA |
589 | struct bdi_writeback *wb; |
590 | ||
591 | if (!bdi_cap_writeback_dirty(bdi)) | |
592 | return; | |
593 | ||
594 | /* | |
595 | * If setup is pending, wait for that to complete first | |
596 | */ | |
597 | wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait, | |
598 | TASK_UNINTERRUPTIBLE); | |
599 | ||
600 | /* | |
601 | * Make sure nobody finds us on the bdi_list anymore | |
602 | */ | |
cfc4ba53 | 603 | bdi_remove_from_list(bdi); |
03ba3782 JA |
604 | |
605 | /* | |
606 | * Finally, kill the kernel threads. We don't need to be RCU | |
c62b17a5 RD |
607 | * safe anymore, since the bdi is gone from visibility. Force |
608 | * unfreeze of the thread before calling kthread_stop(), otherwise | |
609 | * it would never exet if it is currently stuck in the refrigerator. | |
03ba3782 | 610 | */ |
c62b17a5 | 611 | list_for_each_entry(wb, &bdi->wb_list, list) { |
bf7ec5bb | 612 | thaw_process(wb->task); |
03ba3782 | 613 | kthread_stop(wb->task); |
c62b17a5 | 614 | } |
66f3b8e2 JA |
615 | } |
616 | ||
592b09a4 JA |
617 | /* |
618 | * This bdi is going away now, make sure that no super_blocks point to it | |
619 | */ | |
620 | static void bdi_prune_sb(struct backing_dev_info *bdi) | |
621 | { | |
622 | struct super_block *sb; | |
623 | ||
624 | spin_lock(&sb_lock); | |
625 | list_for_each_entry(sb, &super_blocks, s_list) { | |
626 | if (sb->s_bdi == bdi) | |
627 | sb->s_bdi = NULL; | |
628 | } | |
629 | spin_unlock(&sb_lock); | |
630 | } | |
631 | ||
cf0ca9fe PZ |
632 | void bdi_unregister(struct backing_dev_info *bdi) |
633 | { | |
634 | if (bdi->dev) { | |
8c4db335 JA |
635 | bdi_prune_sb(bdi); |
636 | ||
03ba3782 JA |
637 | if (!bdi_cap_flush_forker(bdi)) |
638 | bdi_wb_shutdown(bdi); | |
76f1418b | 639 | bdi_debug_unregister(bdi); |
cf0ca9fe PZ |
640 | device_unregister(bdi->dev); |
641 | bdi->dev = NULL; | |
642 | } | |
643 | } | |
644 | EXPORT_SYMBOL(bdi_unregister); | |
3fcfab16 | 645 | |
b2e8fb6e PZ |
646 | int bdi_init(struct backing_dev_info *bdi) |
647 | { | |
03ba3782 | 648 | int i, err; |
b2e8fb6e | 649 | |
cf0ca9fe PZ |
650 | bdi->dev = NULL; |
651 | ||
189d3c4a | 652 | bdi->min_ratio = 0; |
a42dde04 PZ |
653 | bdi->max_ratio = 100; |
654 | bdi->max_prop_frac = PROP_FRAC_BASE; | |
03ba3782 | 655 | spin_lock_init(&bdi->wb_lock); |
cfc4ba53 | 656 | INIT_RCU_HEAD(&bdi->rcu_head); |
66f3b8e2 | 657 | INIT_LIST_HEAD(&bdi->bdi_list); |
03ba3782 JA |
658 | INIT_LIST_HEAD(&bdi->wb_list); |
659 | INIT_LIST_HEAD(&bdi->work_list); | |
660 | ||
661 | bdi_wb_init(&bdi->wb, bdi); | |
662 | ||
663 | /* | |
664 | * Just one thread support for now, hard code mask and count | |
665 | */ | |
666 | bdi->wb_mask = 1; | |
667 | bdi->wb_cnt = 1; | |
189d3c4a | 668 | |
b2e8fb6e | 669 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
ea319518 | 670 | err = percpu_counter_init(&bdi->bdi_stat[i], 0); |
04fbfdc1 PZ |
671 | if (err) |
672 | goto err; | |
673 | } | |
674 | ||
675 | bdi->dirty_exceeded = 0; | |
676 | err = prop_local_init_percpu(&bdi->completions); | |
677 | ||
678 | if (err) { | |
679 | err: | |
4b01a0b1 | 680 | while (i--) |
04fbfdc1 | 681 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
b2e8fb6e PZ |
682 | } |
683 | ||
684 | return err; | |
685 | } | |
686 | EXPORT_SYMBOL(bdi_init); | |
687 | ||
688 | void bdi_destroy(struct backing_dev_info *bdi) | |
689 | { | |
690 | int i; | |
691 | ||
ce5f8e77 JA |
692 | /* |
693 | * Splice our entries to the default_backing_dev_info, if this | |
694 | * bdi disappears | |
695 | */ | |
696 | if (bdi_has_dirty_io(bdi)) { | |
697 | struct bdi_writeback *dst = &default_backing_dev_info.wb; | |
698 | ||
699 | spin_lock(&inode_lock); | |
700 | list_splice(&bdi->wb.b_dirty, &dst->b_dirty); | |
701 | list_splice(&bdi->wb.b_io, &dst->b_io); | |
702 | list_splice(&bdi->wb.b_more_io, &dst->b_more_io); | |
703 | spin_unlock(&inode_lock); | |
704 | } | |
66f3b8e2 | 705 | |
cf0ca9fe PZ |
706 | bdi_unregister(bdi); |
707 | ||
b2e8fb6e PZ |
708 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
709 | percpu_counter_destroy(&bdi->bdi_stat[i]); | |
04fbfdc1 PZ |
710 | |
711 | prop_local_destroy_percpu(&bdi->completions); | |
b2e8fb6e PZ |
712 | } |
713 | EXPORT_SYMBOL(bdi_destroy); | |
714 | ||
3fcfab16 AM |
715 | static wait_queue_head_t congestion_wqh[2] = { |
716 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
717 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
718 | }; | |
719 | ||
1faa16d2 | 720 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
721 | { |
722 | enum bdi_state bit; | |
1faa16d2 | 723 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 | 724 | |
1faa16d2 | 725 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
3fcfab16 AM |
726 | clear_bit(bit, &bdi->state); |
727 | smp_mb__after_clear_bit(); | |
728 | if (waitqueue_active(wqh)) | |
729 | wake_up(wqh); | |
730 | } | |
731 | EXPORT_SYMBOL(clear_bdi_congested); | |
732 | ||
1faa16d2 | 733 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
734 | { |
735 | enum bdi_state bit; | |
736 | ||
1faa16d2 | 737 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
3fcfab16 AM |
738 | set_bit(bit, &bdi->state); |
739 | } | |
740 | EXPORT_SYMBOL(set_bdi_congested); | |
741 | ||
742 | /** | |
743 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 744 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
745 | * @timeout: timeout in jiffies |
746 | * | |
747 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
748 | * write congestion. If no backing_devs are congested then just wait for the | |
749 | * next write to be completed. | |
750 | */ | |
8aa7e847 | 751 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
752 | { |
753 | long ret; | |
754 | DEFINE_WAIT(wait); | |
8aa7e847 | 755 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
756 | |
757 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
758 | ret = io_schedule_timeout(timeout); | |
759 | finish_wait(wqh, &wait); | |
760 | return ret; | |
761 | } | |
762 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 763 |