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 | ||
c3c53206 JA |
14 | static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); |
15 | ||
26160158 JA |
16 | void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page) |
17 | { | |
18 | } | |
19 | EXPORT_SYMBOL(default_unplug_io_fn); | |
20 | ||
21 | struct backing_dev_info default_backing_dev_info = { | |
d993831f | 22 | .name = "default", |
26160158 JA |
23 | .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE, |
24 | .state = 0, | |
25 | .capabilities = BDI_CAP_MAP_COPY, | |
26 | .unplug_io_fn = default_unplug_io_fn, | |
27 | }; | |
28 | EXPORT_SYMBOL_GPL(default_backing_dev_info); | |
cf0ca9fe | 29 | |
5129a469 JE |
30 | struct backing_dev_info noop_backing_dev_info = { |
31 | .name = "noop", | |
32 | }; | |
33 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); | |
34 | ||
cf0ca9fe | 35 | static struct class *bdi_class; |
cfc4ba53 JA |
36 | |
37 | /* | |
38 | * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as | |
39 | * reader side protection for bdi_pending_list. bdi_list has RCU reader side | |
40 | * locking. | |
41 | */ | |
03ba3782 | 42 | DEFINE_SPINLOCK(bdi_lock); |
66f3b8e2 | 43 | LIST_HEAD(bdi_list); |
03ba3782 JA |
44 | LIST_HEAD(bdi_pending_list); |
45 | ||
46 | static struct task_struct *sync_supers_tsk; | |
47 | static struct timer_list sync_supers_timer; | |
48 | ||
49 | static int bdi_sync_supers(void *); | |
50 | static void sync_supers_timer_fn(unsigned long); | |
51 | static void arm_supers_timer(void); | |
52 | ||
53 | static void bdi_add_default_flusher_task(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; | |
f09b00d3 | 69 | struct bdi_writeback *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 | ||
76 | /* | |
77 | * inode lock is enough here, the bdi->wb_list is protected by | |
78 | * RCU on the reader side | |
79 | */ | |
80 | nr_wb = nr_dirty = nr_io = nr_more_io = 0; | |
81 | spin_lock(&inode_lock); | |
82 | list_for_each_entry(wb, &bdi->wb_list, list) { | |
83 | nr_wb++; | |
84 | list_for_each_entry(inode, &wb->b_dirty, i_list) | |
85 | nr_dirty++; | |
86 | list_for_each_entry(inode, &wb->b_io, i_list) | |
87 | nr_io++; | |
88 | list_for_each_entry(inode, &wb->b_more_io, i_list) | |
89 | nr_more_io++; | |
90 | } | |
91 | spin_unlock(&inode_lock); | |
76f1418b MS |
92 | |
93 | get_dirty_limits(&background_thresh, &dirty_thresh, &bdi_thresh, bdi); | |
94 | ||
95 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
96 | seq_printf(m, | |
97 | "BdiWriteback: %8lu kB\n" | |
98 | "BdiReclaimable: %8lu kB\n" | |
99 | "BdiDirtyThresh: %8lu kB\n" | |
100 | "DirtyThresh: %8lu kB\n" | |
f09b00d3 | 101 | "BackgroundThresh: %8lu kB\n" |
961515f6 | 102 | "WritebackThreads: %8lu\n" |
f09b00d3 JA |
103 | "b_dirty: %8lu\n" |
104 | "b_io: %8lu\n" | |
105 | "b_more_io: %8lu\n" | |
106 | "bdi_list: %8u\n" | |
107 | "state: %8lx\n" | |
108 | "wb_mask: %8lx\n" | |
109 | "wb_list: %8u\n" | |
110 | "wb_cnt: %8u\n", | |
76f1418b MS |
111 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
112 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), | |
f09b00d3 JA |
113 | K(bdi_thresh), K(dirty_thresh), |
114 | K(background_thresh), nr_wb, nr_dirty, nr_io, nr_more_io, | |
115 | !list_empty(&bdi->bdi_list), bdi->state, bdi->wb_mask, | |
116 | !list_empty(&bdi->wb_list), bdi->wb_cnt); | |
76f1418b MS |
117 | #undef K |
118 | ||
119 | return 0; | |
120 | } | |
121 | ||
122 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) | |
123 | { | |
124 | return single_open(file, bdi_debug_stats_show, inode->i_private); | |
125 | } | |
126 | ||
127 | static const struct file_operations bdi_debug_stats_fops = { | |
128 | .open = bdi_debug_stats_open, | |
129 | .read = seq_read, | |
130 | .llseek = seq_lseek, | |
131 | .release = single_release, | |
132 | }; | |
133 | ||
134 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) | |
135 | { | |
136 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
137 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, | |
138 | bdi, &bdi_debug_stats_fops); | |
139 | } | |
140 | ||
141 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
142 | { | |
143 | debugfs_remove(bdi->debug_stats); | |
144 | debugfs_remove(bdi->debug_dir); | |
145 | } | |
146 | #else | |
147 | static inline void bdi_debug_init(void) | |
148 | { | |
149 | } | |
150 | static inline void bdi_debug_register(struct backing_dev_info *bdi, | |
151 | const char *name) | |
152 | { | |
153 | } | |
154 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
155 | { | |
156 | } | |
157 | #endif | |
158 | ||
cf0ca9fe PZ |
159 | static ssize_t read_ahead_kb_store(struct device *dev, |
160 | struct device_attribute *attr, | |
161 | const char *buf, size_t count) | |
162 | { | |
163 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
164 | char *end; | |
165 | unsigned long read_ahead_kb; | |
166 | ssize_t ret = -EINVAL; | |
167 | ||
168 | read_ahead_kb = simple_strtoul(buf, &end, 10); | |
169 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
170 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
171 | ret = count; | |
172 | } | |
173 | return ret; | |
174 | } | |
175 | ||
176 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) | |
177 | ||
178 | #define BDI_SHOW(name, expr) \ | |
179 | static ssize_t name##_show(struct device *dev, \ | |
180 | struct device_attribute *attr, char *page) \ | |
181 | { \ | |
182 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
183 | \ | |
184 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ | |
185 | } | |
186 | ||
187 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
188 | ||
189d3c4a PZ |
189 | static ssize_t min_ratio_store(struct device *dev, |
190 | struct device_attribute *attr, const char *buf, size_t count) | |
191 | { | |
192 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
193 | char *end; | |
194 | unsigned int ratio; | |
195 | ssize_t ret = -EINVAL; | |
196 | ||
197 | ratio = simple_strtoul(buf, &end, 10); | |
198 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
199 | ret = bdi_set_min_ratio(bdi, ratio); | |
200 | if (!ret) | |
201 | ret = count; | |
202 | } | |
203 | return ret; | |
204 | } | |
205 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
206 | ||
a42dde04 PZ |
207 | static ssize_t max_ratio_store(struct device *dev, |
208 | struct device_attribute *attr, const char *buf, size_t count) | |
209 | { | |
210 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
211 | char *end; | |
212 | unsigned int ratio; | |
213 | ssize_t ret = -EINVAL; | |
214 | ||
215 | ratio = simple_strtoul(buf, &end, 10); | |
216 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
217 | ret = bdi_set_max_ratio(bdi, ratio); | |
218 | if (!ret) | |
219 | ret = count; | |
220 | } | |
221 | return ret; | |
222 | } | |
223 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
224 | ||
cf0ca9fe PZ |
225 | #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store) |
226 | ||
227 | static struct device_attribute bdi_dev_attrs[] = { | |
228 | __ATTR_RW(read_ahead_kb), | |
189d3c4a | 229 | __ATTR_RW(min_ratio), |
a42dde04 | 230 | __ATTR_RW(max_ratio), |
cf0ca9fe PZ |
231 | __ATTR_NULL, |
232 | }; | |
233 | ||
234 | static __init int bdi_class_init(void) | |
235 | { | |
236 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
237 | if (IS_ERR(bdi_class)) |
238 | return PTR_ERR(bdi_class); | |
239 | ||
cf0ca9fe | 240 | bdi_class->dev_attrs = bdi_dev_attrs; |
76f1418b | 241 | bdi_debug_init(); |
cf0ca9fe PZ |
242 | return 0; |
243 | } | |
76f1418b | 244 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 245 | |
26160158 JA |
246 | static int __init default_bdi_init(void) |
247 | { | |
248 | int err; | |
249 | ||
03ba3782 JA |
250 | sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers"); |
251 | BUG_ON(IS_ERR(sync_supers_tsk)); | |
252 | ||
253 | init_timer(&sync_supers_timer); | |
254 | setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0); | |
255 | arm_supers_timer(); | |
256 | ||
26160158 JA |
257 | err = bdi_init(&default_backing_dev_info); |
258 | if (!err) | |
259 | bdi_register(&default_backing_dev_info, NULL, "default"); | |
260 | ||
261 | return err; | |
262 | } | |
263 | subsys_initcall(default_bdi_init); | |
264 | ||
03ba3782 JA |
265 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
266 | { | |
267 | memset(wb, 0, sizeof(*wb)); | |
268 | ||
269 | wb->bdi = bdi; | |
270 | wb->last_old_flush = jiffies; | |
271 | INIT_LIST_HEAD(&wb->b_dirty); | |
272 | INIT_LIST_HEAD(&wb->b_io); | |
273 | INIT_LIST_HEAD(&wb->b_more_io); | |
274 | } | |
275 | ||
276 | static void bdi_task_init(struct backing_dev_info *bdi, | |
277 | struct bdi_writeback *wb) | |
278 | { | |
279 | struct task_struct *tsk = current; | |
280 | ||
281 | spin_lock(&bdi->wb_lock); | |
282 | list_add_tail_rcu(&wb->list, &bdi->wb_list); | |
283 | spin_unlock(&bdi->wb_lock); | |
284 | ||
285 | tsk->flags |= PF_FLUSHER | PF_SWAPWRITE; | |
286 | set_freezable(); | |
287 | ||
288 | /* | |
289 | * Our parent may run at a different priority, just set us to normal | |
290 | */ | |
291 | set_user_nice(tsk, 0); | |
292 | } | |
293 | ||
294 | static int bdi_start_fn(void *ptr) | |
295 | { | |
296 | struct bdi_writeback *wb = ptr; | |
297 | struct backing_dev_info *bdi = wb->bdi; | |
298 | int ret; | |
299 | ||
300 | /* | |
301 | * Add us to the active bdi_list | |
302 | */ | |
cfc4ba53 JA |
303 | spin_lock_bh(&bdi_lock); |
304 | list_add_rcu(&bdi->bdi_list, &bdi_list); | |
305 | spin_unlock_bh(&bdi_lock); | |
03ba3782 JA |
306 | |
307 | bdi_task_init(bdi, wb); | |
308 | ||
309 | /* | |
310 | * Clear pending bit and wakeup anybody waiting to tear us down | |
311 | */ | |
312 | clear_bit(BDI_pending, &bdi->state); | |
313 | smp_mb__after_clear_bit(); | |
314 | wake_up_bit(&bdi->state, BDI_pending); | |
315 | ||
316 | ret = bdi_writeback_task(wb); | |
317 | ||
318 | /* | |
319 | * Remove us from the list | |
320 | */ | |
321 | spin_lock(&bdi->wb_lock); | |
322 | list_del_rcu(&wb->list); | |
323 | spin_unlock(&bdi->wb_lock); | |
324 | ||
325 | /* | |
326 | * Flush any work that raced with us exiting. No new work | |
327 | * will be added, since this bdi isn't discoverable anymore. | |
328 | */ | |
329 | if (!list_empty(&bdi->work_list)) | |
330 | wb_do_writeback(wb, 1); | |
331 | ||
332 | wb->task = NULL; | |
333 | return ret; | |
334 | } | |
335 | ||
336 | int bdi_has_dirty_io(struct backing_dev_info *bdi) | |
337 | { | |
338 | return wb_has_dirty_io(&bdi->wb); | |
339 | } | |
340 | ||
341 | static void bdi_flush_io(struct backing_dev_info *bdi) | |
342 | { | |
343 | struct writeback_control wbc = { | |
344 | .bdi = bdi, | |
345 | .sync_mode = WB_SYNC_NONE, | |
346 | .older_than_this = NULL, | |
347 | .range_cyclic = 1, | |
348 | .nr_to_write = 1024, | |
349 | }; | |
350 | ||
351 | writeback_inodes_wbc(&wbc); | |
352 | } | |
353 | ||
354 | /* | |
355 | * kupdated() used to do this. We cannot do it from the bdi_forker_task() | |
356 | * or we risk deadlocking on ->s_umount. The longer term solution would be | |
357 | * to implement sync_supers_bdi() or similar and simply do it from the | |
358 | * bdi writeback tasks individually. | |
359 | */ | |
360 | static int bdi_sync_supers(void *unused) | |
361 | { | |
362 | set_user_nice(current, 0); | |
363 | ||
364 | while (!kthread_should_stop()) { | |
365 | set_current_state(TASK_INTERRUPTIBLE); | |
366 | schedule(); | |
367 | ||
368 | /* | |
369 | * Do this periodically, like kupdated() did before. | |
370 | */ | |
371 | sync_supers(); | |
372 | } | |
373 | ||
374 | return 0; | |
375 | } | |
376 | ||
377 | static void arm_supers_timer(void) | |
378 | { | |
379 | unsigned long next; | |
380 | ||
381 | next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies; | |
382 | mod_timer(&sync_supers_timer, round_jiffies_up(next)); | |
383 | } | |
384 | ||
385 | static void sync_supers_timer_fn(unsigned long unused) | |
386 | { | |
387 | wake_up_process(sync_supers_tsk); | |
388 | arm_supers_timer(); | |
389 | } | |
390 | ||
391 | static int bdi_forker_task(void *ptr) | |
392 | { | |
393 | struct bdi_writeback *me = ptr; | |
394 | ||
395 | bdi_task_init(me->bdi, me); | |
396 | ||
397 | for (;;) { | |
398 | struct backing_dev_info *bdi, *tmp; | |
399 | struct bdi_writeback *wb; | |
400 | ||
401 | /* | |
402 | * Temporary measure, we want to make sure we don't see | |
403 | * dirty data on the default backing_dev_info | |
404 | */ | |
405 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) | |
406 | wb_do_writeback(me, 0); | |
407 | ||
cfc4ba53 | 408 | spin_lock_bh(&bdi_lock); |
03ba3782 JA |
409 | |
410 | /* | |
411 | * Check if any existing bdi's have dirty data without | |
412 | * a thread registered. If so, set that up. | |
413 | */ | |
414 | list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) { | |
415 | if (bdi->wb.task) | |
416 | continue; | |
417 | if (list_empty(&bdi->work_list) && | |
418 | !bdi_has_dirty_io(bdi)) | |
419 | continue; | |
420 | ||
421 | bdi_add_default_flusher_task(bdi); | |
422 | } | |
423 | ||
424 | set_current_state(TASK_INTERRUPTIBLE); | |
425 | ||
426 | if (list_empty(&bdi_pending_list)) { | |
427 | unsigned long wait; | |
428 | ||
cfc4ba53 | 429 | spin_unlock_bh(&bdi_lock); |
03ba3782 JA |
430 | wait = msecs_to_jiffies(dirty_writeback_interval * 10); |
431 | schedule_timeout(wait); | |
432 | try_to_freeze(); | |
433 | continue; | |
434 | } | |
435 | ||
436 | __set_current_state(TASK_RUNNING); | |
437 | ||
438 | /* | |
439 | * This is our real job - check for pending entries in | |
440 | * bdi_pending_list, and create the tasks that got added | |
441 | */ | |
442 | bdi = list_entry(bdi_pending_list.next, struct backing_dev_info, | |
443 | bdi_list); | |
444 | list_del_init(&bdi->bdi_list); | |
cfc4ba53 | 445 | spin_unlock_bh(&bdi_lock); |
03ba3782 JA |
446 | |
447 | wb = &bdi->wb; | |
448 | wb->task = kthread_run(bdi_start_fn, wb, "flush-%s", | |
449 | dev_name(bdi->dev)); | |
450 | /* | |
451 | * If task creation fails, then readd the bdi to | |
452 | * the pending list and force writeout of the bdi | |
453 | * from this forker thread. That will free some memory | |
454 | * and we can try again. | |
455 | */ | |
456 | if (IS_ERR(wb->task)) { | |
457 | wb->task = NULL; | |
458 | ||
459 | /* | |
460 | * Add this 'bdi' to the back, so we get | |
461 | * a chance to flush other bdi's to free | |
462 | * memory. | |
463 | */ | |
cfc4ba53 | 464 | spin_lock_bh(&bdi_lock); |
03ba3782 | 465 | list_add_tail(&bdi->bdi_list, &bdi_pending_list); |
cfc4ba53 | 466 | spin_unlock_bh(&bdi_lock); |
03ba3782 JA |
467 | |
468 | bdi_flush_io(bdi); | |
469 | } | |
470 | } | |
471 | ||
472 | return 0; | |
473 | } | |
474 | ||
cfc4ba53 JA |
475 | static void bdi_add_to_pending(struct rcu_head *head) |
476 | { | |
477 | struct backing_dev_info *bdi; | |
478 | ||
479 | bdi = container_of(head, struct backing_dev_info, rcu_head); | |
480 | INIT_LIST_HEAD(&bdi->bdi_list); | |
481 | ||
482 | spin_lock(&bdi_lock); | |
483 | list_add_tail(&bdi->bdi_list, &bdi_pending_list); | |
484 | spin_unlock(&bdi_lock); | |
485 | ||
486 | /* | |
487 | * We are now on the pending list, wake up bdi_forker_task() | |
488 | * to finish the job and add us back to the active bdi_list | |
489 | */ | |
490 | wake_up_process(default_backing_dev_info.wb.task); | |
491 | } | |
492 | ||
03ba3782 JA |
493 | /* |
494 | * Add the default flusher task that gets created for any bdi | |
495 | * that has dirty data pending writeout | |
496 | */ | |
497 | void static bdi_add_default_flusher_task(struct backing_dev_info *bdi) | |
498 | { | |
499 | if (!bdi_cap_writeback_dirty(bdi)) | |
500 | return; | |
501 | ||
500b067c JA |
502 | if (WARN_ON(!test_bit(BDI_registered, &bdi->state))) { |
503 | printk(KERN_ERR "bdi %p/%s is not registered!\n", | |
504 | bdi, bdi->name); | |
505 | return; | |
506 | } | |
507 | ||
03ba3782 JA |
508 | /* |
509 | * Check with the helper whether to proceed adding a task. Will only | |
510 | * abort if we two or more simultanous calls to | |
511 | * bdi_add_default_flusher_task() occured, further additions will block | |
512 | * waiting for previous additions to finish. | |
513 | */ | |
514 | if (!test_and_set_bit(BDI_pending, &bdi->state)) { | |
cfc4ba53 | 515 | list_del_rcu(&bdi->bdi_list); |
03ba3782 JA |
516 | |
517 | /* | |
cfc4ba53 JA |
518 | * We must wait for the current RCU period to end before |
519 | * moving to the pending list. So schedule that operation | |
520 | * from an RCU callback. | |
03ba3782 | 521 | */ |
cfc4ba53 | 522 | call_rcu(&bdi->rcu_head, bdi_add_to_pending); |
03ba3782 JA |
523 | } |
524 | } | |
525 | ||
cfc4ba53 JA |
526 | /* |
527 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
528 | */ | |
529 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
530 | { | |
531 | spin_lock_bh(&bdi_lock); | |
532 | list_del_rcu(&bdi->bdi_list); | |
533 | spin_unlock_bh(&bdi_lock); | |
534 | ||
535 | synchronize_rcu(); | |
536 | } | |
537 | ||
cf0ca9fe PZ |
538 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
539 | const char *fmt, ...) | |
540 | { | |
cf0ca9fe PZ |
541 | va_list args; |
542 | int ret = 0; | |
543 | struct device *dev; | |
544 | ||
69fc208b | 545 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
f1d0b063 KS |
546 | goto exit; |
547 | ||
cf0ca9fe | 548 | va_start(args, fmt); |
19051c50 | 549 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
cf0ca9fe | 550 | va_end(args); |
cf0ca9fe PZ |
551 | if (IS_ERR(dev)) { |
552 | ret = PTR_ERR(dev); | |
553 | goto exit; | |
554 | } | |
555 | ||
cfc4ba53 JA |
556 | spin_lock_bh(&bdi_lock); |
557 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
558 | spin_unlock_bh(&bdi_lock); | |
66f3b8e2 | 559 | |
cf0ca9fe | 560 | bdi->dev = dev; |
cf0ca9fe | 561 | |
03ba3782 JA |
562 | /* |
563 | * Just start the forker thread for our default backing_dev_info, | |
564 | * and add other bdi's to the list. They will get a thread created | |
565 | * on-demand when they need it. | |
566 | */ | |
567 | if (bdi_cap_flush_forker(bdi)) { | |
568 | struct bdi_writeback *wb = &bdi->wb; | |
569 | ||
570 | wb->task = kthread_run(bdi_forker_task, wb, "bdi-%s", | |
571 | dev_name(dev)); | |
572 | if (IS_ERR(wb->task)) { | |
573 | wb->task = NULL; | |
574 | ret = -ENOMEM; | |
575 | ||
cfc4ba53 | 576 | bdi_remove_from_list(bdi); |
03ba3782 JA |
577 | goto exit; |
578 | } | |
579 | } | |
580 | ||
581 | bdi_debug_register(bdi, dev_name(dev)); | |
500b067c | 582 | set_bit(BDI_registered, &bdi->state); |
cf0ca9fe | 583 | exit: |
cf0ca9fe PZ |
584 | return ret; |
585 | } | |
586 | EXPORT_SYMBOL(bdi_register); | |
587 | ||
588 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) | |
589 | { | |
590 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); | |
591 | } | |
592 | EXPORT_SYMBOL(bdi_register_dev); | |
593 | ||
03ba3782 JA |
594 | /* |
595 | * Remove bdi from the global list and shutdown any threads we have running | |
596 | */ | |
597 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) | |
66f3b8e2 | 598 | { |
03ba3782 JA |
599 | struct bdi_writeback *wb; |
600 | ||
601 | if (!bdi_cap_writeback_dirty(bdi)) | |
602 | return; | |
603 | ||
604 | /* | |
605 | * If setup is pending, wait for that to complete first | |
606 | */ | |
607 | wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait, | |
608 | TASK_UNINTERRUPTIBLE); | |
609 | ||
610 | /* | |
611 | * Make sure nobody finds us on the bdi_list anymore | |
612 | */ | |
cfc4ba53 | 613 | bdi_remove_from_list(bdi); |
03ba3782 JA |
614 | |
615 | /* | |
616 | * Finally, kill the kernel threads. We don't need to be RCU | |
c62b17a5 RD |
617 | * safe anymore, since the bdi is gone from visibility. Force |
618 | * unfreeze of the thread before calling kthread_stop(), otherwise | |
619 | * it would never exet if it is currently stuck in the refrigerator. | |
03ba3782 | 620 | */ |
c62b17a5 | 621 | list_for_each_entry(wb, &bdi->wb_list, list) { |
bf7ec5bb | 622 | thaw_process(wb->task); |
03ba3782 | 623 | kthread_stop(wb->task); |
c62b17a5 | 624 | } |
66f3b8e2 JA |
625 | } |
626 | ||
592b09a4 JA |
627 | /* |
628 | * This bdi is going away now, make sure that no super_blocks point to it | |
629 | */ | |
630 | static void bdi_prune_sb(struct backing_dev_info *bdi) | |
631 | { | |
632 | struct super_block *sb; | |
633 | ||
634 | spin_lock(&sb_lock); | |
635 | list_for_each_entry(sb, &super_blocks, s_list) { | |
636 | if (sb->s_bdi == bdi) | |
637 | sb->s_bdi = NULL; | |
638 | } | |
639 | spin_unlock(&sb_lock); | |
640 | } | |
641 | ||
cf0ca9fe PZ |
642 | void bdi_unregister(struct backing_dev_info *bdi) |
643 | { | |
644 | if (bdi->dev) { | |
8c4db335 JA |
645 | bdi_prune_sb(bdi); |
646 | ||
03ba3782 JA |
647 | if (!bdi_cap_flush_forker(bdi)) |
648 | bdi_wb_shutdown(bdi); | |
76f1418b | 649 | bdi_debug_unregister(bdi); |
cf0ca9fe PZ |
650 | device_unregister(bdi->dev); |
651 | bdi->dev = NULL; | |
652 | } | |
653 | } | |
654 | EXPORT_SYMBOL(bdi_unregister); | |
3fcfab16 | 655 | |
b2e8fb6e PZ |
656 | int bdi_init(struct backing_dev_info *bdi) |
657 | { | |
03ba3782 | 658 | int i, err; |
b2e8fb6e | 659 | |
cf0ca9fe PZ |
660 | bdi->dev = NULL; |
661 | ||
189d3c4a | 662 | bdi->min_ratio = 0; |
a42dde04 PZ |
663 | bdi->max_ratio = 100; |
664 | bdi->max_prop_frac = PROP_FRAC_BASE; | |
03ba3782 | 665 | spin_lock_init(&bdi->wb_lock); |
cfc4ba53 | 666 | INIT_RCU_HEAD(&bdi->rcu_head); |
66f3b8e2 | 667 | INIT_LIST_HEAD(&bdi->bdi_list); |
03ba3782 JA |
668 | INIT_LIST_HEAD(&bdi->wb_list); |
669 | INIT_LIST_HEAD(&bdi->work_list); | |
670 | ||
671 | bdi_wb_init(&bdi->wb, bdi); | |
672 | ||
673 | /* | |
674 | * Just one thread support for now, hard code mask and count | |
675 | */ | |
676 | bdi->wb_mask = 1; | |
677 | bdi->wb_cnt = 1; | |
189d3c4a | 678 | |
b2e8fb6e | 679 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
ea319518 | 680 | err = percpu_counter_init(&bdi->bdi_stat[i], 0); |
04fbfdc1 PZ |
681 | if (err) |
682 | goto err; | |
683 | } | |
684 | ||
685 | bdi->dirty_exceeded = 0; | |
686 | err = prop_local_init_percpu(&bdi->completions); | |
687 | ||
688 | if (err) { | |
689 | err: | |
4b01a0b1 | 690 | while (i--) |
04fbfdc1 | 691 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
b2e8fb6e PZ |
692 | } |
693 | ||
694 | return err; | |
695 | } | |
696 | EXPORT_SYMBOL(bdi_init); | |
697 | ||
698 | void bdi_destroy(struct backing_dev_info *bdi) | |
699 | { | |
700 | int i; | |
701 | ||
ce5f8e77 JA |
702 | /* |
703 | * Splice our entries to the default_backing_dev_info, if this | |
704 | * bdi disappears | |
705 | */ | |
706 | if (bdi_has_dirty_io(bdi)) { | |
707 | struct bdi_writeback *dst = &default_backing_dev_info.wb; | |
708 | ||
709 | spin_lock(&inode_lock); | |
710 | list_splice(&bdi->wb.b_dirty, &dst->b_dirty); | |
711 | list_splice(&bdi->wb.b_io, &dst->b_io); | |
712 | list_splice(&bdi->wb.b_more_io, &dst->b_more_io); | |
713 | spin_unlock(&inode_lock); | |
714 | } | |
66f3b8e2 | 715 | |
cf0ca9fe PZ |
716 | bdi_unregister(bdi); |
717 | ||
b2e8fb6e PZ |
718 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
719 | percpu_counter_destroy(&bdi->bdi_stat[i]); | |
04fbfdc1 PZ |
720 | |
721 | prop_local_destroy_percpu(&bdi->completions); | |
b2e8fb6e PZ |
722 | } |
723 | EXPORT_SYMBOL(bdi_destroy); | |
724 | ||
c3c53206 JA |
725 | /* |
726 | * For use from filesystems to quickly init and register a bdi associated | |
727 | * with dirty writeback | |
728 | */ | |
729 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name, | |
730 | unsigned int cap) | |
731 | { | |
732 | char tmp[32]; | |
733 | int err; | |
734 | ||
735 | bdi->name = name; | |
736 | bdi->capabilities = cap; | |
737 | err = bdi_init(bdi); | |
738 | if (err) | |
739 | return err; | |
740 | ||
741 | sprintf(tmp, "%.28s%s", name, "-%d"); | |
742 | err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq)); | |
743 | if (err) { | |
744 | bdi_destroy(bdi); | |
745 | return err; | |
746 | } | |
747 | ||
748 | return 0; | |
749 | } | |
750 | EXPORT_SYMBOL(bdi_setup_and_register); | |
751 | ||
3fcfab16 AM |
752 | static wait_queue_head_t congestion_wqh[2] = { |
753 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
754 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
755 | }; | |
756 | ||
1faa16d2 | 757 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
758 | { |
759 | enum bdi_state bit; | |
1faa16d2 | 760 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 | 761 | |
1faa16d2 | 762 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
3fcfab16 AM |
763 | clear_bit(bit, &bdi->state); |
764 | smp_mb__after_clear_bit(); | |
765 | if (waitqueue_active(wqh)) | |
766 | wake_up(wqh); | |
767 | } | |
768 | EXPORT_SYMBOL(clear_bdi_congested); | |
769 | ||
1faa16d2 | 770 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
771 | { |
772 | enum bdi_state bit; | |
773 | ||
1faa16d2 | 774 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
3fcfab16 AM |
775 | set_bit(bit, &bdi->state); |
776 | } | |
777 | EXPORT_SYMBOL(set_bdi_congested); | |
778 | ||
779 | /** | |
780 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 781 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
782 | * @timeout: timeout in jiffies |
783 | * | |
784 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
785 | * write congestion. If no backing_devs are congested then just wait for the | |
786 | * next write to be completed. | |
787 | */ | |
8aa7e847 | 788 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
789 | { |
790 | long ret; | |
791 | DEFINE_WAIT(wait); | |
8aa7e847 | 792 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
793 | |
794 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
795 | ret = io_schedule_timeout(timeout); | |
796 | finish_wait(wqh, &wait); | |
797 | return ret; | |
798 | } | |
799 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 800 |