Commit | Line | Data |
---|---|---|
3fcfab16 AM |
1 | |
2 | #include <linux/wait.h> | |
3 | #include <linux/backing-dev.h> | |
03ba3782 JA |
4 | #include <linux/kthread.h> |
5 | #include <linux/freezer.h> | |
3fcfab16 | 6 | #include <linux/fs.h> |
26160158 | 7 | #include <linux/pagemap.h> |
03ba3782 | 8 | #include <linux/mm.h> |
3fcfab16 AM |
9 | #include <linux/sched.h> |
10 | #include <linux/module.h> | |
cf0ca9fe PZ |
11 | #include <linux/writeback.h> |
12 | #include <linux/device.h> | |
455b2864 | 13 | #include <trace/events/writeback.h> |
cf0ca9fe | 14 | |
c3c53206 JA |
15 | static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0); |
16 | ||
26160158 | 17 | struct backing_dev_info default_backing_dev_info = { |
d993831f | 18 | .name = "default", |
26160158 JA |
19 | .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE, |
20 | .state = 0, | |
21 | .capabilities = BDI_CAP_MAP_COPY, | |
26160158 JA |
22 | }; |
23 | EXPORT_SYMBOL_GPL(default_backing_dev_info); | |
cf0ca9fe | 24 | |
5129a469 JE |
25 | struct backing_dev_info noop_backing_dev_info = { |
26 | .name = "noop", | |
976e48f8 | 27 | .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK, |
5129a469 JE |
28 | }; |
29 | EXPORT_SYMBOL_GPL(noop_backing_dev_info); | |
30 | ||
cf0ca9fe | 31 | static struct class *bdi_class; |
cfc4ba53 JA |
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 | */ | |
03ba3782 | 38 | DEFINE_SPINLOCK(bdi_lock); |
66f3b8e2 | 39 | LIST_HEAD(bdi_list); |
03ba3782 JA |
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); | |
03ba3782 | 47 | |
f758eeab CH |
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 | ||
76f1418b MS |
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; | |
c1955ce3 | 73 | struct bdi_writeback *wb = &bdi->wb; |
364aeb28 DR |
74 | unsigned long background_thresh; |
75 | unsigned long dirty_thresh; | |
76 | unsigned long bdi_thresh; | |
345227d7 | 77 | unsigned long nr_dirty, nr_io, nr_more_io; |
f09b00d3 JA |
78 | struct inode *inode; |
79 | ||
345227d7 | 80 | nr_dirty = nr_io = nr_more_io = 0; |
f758eeab | 81 | spin_lock(&wb->list_lock); |
7ccf19a8 | 82 | list_for_each_entry(inode, &wb->b_dirty, i_wb_list) |
c1955ce3 | 83 | nr_dirty++; |
7ccf19a8 | 84 | list_for_each_entry(inode, &wb->b_io, i_wb_list) |
c1955ce3 | 85 | nr_io++; |
7ccf19a8 | 86 | list_for_each_entry(inode, &wb->b_more_io, i_wb_list) |
c1955ce3 | 87 | nr_more_io++; |
f758eeab | 88 | spin_unlock(&wb->list_lock); |
76f1418b | 89 | |
16c4042f WF |
90 | global_dirty_limits(&background_thresh, &dirty_thresh); |
91 | bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh); | |
76f1418b MS |
92 | |
93 | #define K(x) ((x) << (PAGE_SHIFT - 10)) | |
94 | seq_printf(m, | |
00821b00 WF |
95 | "BdiWriteback: %10lu kB\n" |
96 | "BdiReclaimable: %10lu kB\n" | |
97 | "BdiDirtyThresh: %10lu kB\n" | |
98 | "DirtyThresh: %10lu kB\n" | |
99 | "BackgroundThresh: %10lu kB\n" | |
100 | "BdiWritten: %10lu kB\n" | |
101 | "BdiWriteBandwidth: %10lu kBps\n" | |
102 | "b_dirty: %10lu\n" | |
103 | "b_io: %10lu\n" | |
104 | "b_more_io: %10lu\n" | |
105 | "bdi_list: %10u\n" | |
106 | "state: %10lx\n", | |
76f1418b MS |
107 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
108 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), | |
f7d2b1ec JK |
109 | K(bdi_thresh), |
110 | K(dirty_thresh), | |
111 | K(background_thresh), | |
112 | (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)), | |
00821b00 | 113 | (unsigned long) K(bdi->write_bandwidth), |
f7d2b1ec JK |
114 | nr_dirty, |
115 | nr_io, | |
116 | nr_more_io, | |
c1955ce3 | 117 | !list_empty(&bdi->bdi_list), bdi->state); |
76f1418b MS |
118 | #undef K |
119 | ||
120 | return 0; | |
121 | } | |
122 | ||
123 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) | |
124 | { | |
125 | return single_open(file, bdi_debug_stats_show, inode->i_private); | |
126 | } | |
127 | ||
128 | static const struct file_operations bdi_debug_stats_fops = { | |
129 | .open = bdi_debug_stats_open, | |
130 | .read = seq_read, | |
131 | .llseek = seq_lseek, | |
132 | .release = single_release, | |
133 | }; | |
134 | ||
135 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) | |
136 | { | |
137 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
138 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, | |
139 | bdi, &bdi_debug_stats_fops); | |
140 | } | |
141 | ||
142 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
143 | { | |
144 | debugfs_remove(bdi->debug_stats); | |
145 | debugfs_remove(bdi->debug_dir); | |
146 | } | |
147 | #else | |
148 | static inline void bdi_debug_init(void) | |
149 | { | |
150 | } | |
151 | static inline void bdi_debug_register(struct backing_dev_info *bdi, | |
152 | const char *name) | |
153 | { | |
154 | } | |
155 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
156 | { | |
157 | } | |
158 | #endif | |
159 | ||
cf0ca9fe PZ |
160 | static ssize_t read_ahead_kb_store(struct device *dev, |
161 | struct device_attribute *attr, | |
162 | const char *buf, size_t count) | |
163 | { | |
164 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
165 | char *end; | |
166 | unsigned long read_ahead_kb; | |
167 | ssize_t ret = -EINVAL; | |
168 | ||
169 | read_ahead_kb = simple_strtoul(buf, &end, 10); | |
170 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
171 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
172 | ret = count; | |
173 | } | |
174 | return ret; | |
175 | } | |
176 | ||
177 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) | |
178 | ||
179 | #define BDI_SHOW(name, expr) \ | |
180 | static ssize_t name##_show(struct device *dev, \ | |
181 | struct device_attribute *attr, char *page) \ | |
182 | { \ | |
183 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
184 | \ | |
185 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ | |
186 | } | |
187 | ||
188 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
189 | ||
189d3c4a PZ |
190 | static ssize_t min_ratio_store(struct device *dev, |
191 | struct device_attribute *attr, const char *buf, size_t count) | |
192 | { | |
193 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
194 | char *end; | |
195 | unsigned int ratio; | |
196 | ssize_t ret = -EINVAL; | |
197 | ||
198 | ratio = simple_strtoul(buf, &end, 10); | |
199 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
200 | ret = bdi_set_min_ratio(bdi, ratio); | |
201 | if (!ret) | |
202 | ret = count; | |
203 | } | |
204 | return ret; | |
205 | } | |
206 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
207 | ||
a42dde04 PZ |
208 | static ssize_t max_ratio_store(struct device *dev, |
209 | struct device_attribute *attr, const char *buf, size_t count) | |
210 | { | |
211 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
212 | char *end; | |
213 | unsigned int ratio; | |
214 | ssize_t ret = -EINVAL; | |
215 | ||
216 | ratio = simple_strtoul(buf, &end, 10); | |
217 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
218 | ret = bdi_set_max_ratio(bdi, ratio); | |
219 | if (!ret) | |
220 | ret = count; | |
221 | } | |
222 | return ret; | |
223 | } | |
224 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
225 | ||
cf0ca9fe PZ |
226 | #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store) |
227 | ||
228 | static struct device_attribute bdi_dev_attrs[] = { | |
229 | __ATTR_RW(read_ahead_kb), | |
189d3c4a | 230 | __ATTR_RW(min_ratio), |
a42dde04 | 231 | __ATTR_RW(max_ratio), |
cf0ca9fe PZ |
232 | __ATTR_NULL, |
233 | }; | |
234 | ||
235 | static __init int bdi_class_init(void) | |
236 | { | |
237 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
238 | if (IS_ERR(bdi_class)) |
239 | return PTR_ERR(bdi_class); | |
240 | ||
cf0ca9fe | 241 | bdi_class->dev_attrs = bdi_dev_attrs; |
76f1418b | 242 | bdi_debug_init(); |
cf0ca9fe PZ |
243 | return 0; |
244 | } | |
76f1418b | 245 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 246 | |
26160158 JA |
247 | static int __init default_bdi_init(void) |
248 | { | |
249 | int err; | |
250 | ||
03ba3782 JA |
251 | sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers"); |
252 | BUG_ON(IS_ERR(sync_supers_tsk)); | |
253 | ||
03ba3782 | 254 | setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0); |
6423104b | 255 | bdi_arm_supers_timer(); |
03ba3782 | 256 | |
26160158 JA |
257 | err = bdi_init(&default_backing_dev_info); |
258 | if (!err) | |
259 | bdi_register(&default_backing_dev_info, NULL, "default"); | |
976e48f8 | 260 | err = bdi_init(&noop_backing_dev_info); |
26160158 JA |
261 | |
262 | return err; | |
263 | } | |
264 | subsys_initcall(default_bdi_init); | |
265 | ||
03ba3782 JA |
266 | int bdi_has_dirty_io(struct backing_dev_info *bdi) |
267 | { | |
268 | return wb_has_dirty_io(&bdi->wb); | |
269 | } | |
270 | ||
03ba3782 | 271 | /* |
6f904ff0 | 272 | * kupdated() used to do this. We cannot do it from the bdi_forker_thread() |
03ba3782 JA |
273 | * or we risk deadlocking on ->s_umount. The longer term solution would be |
274 | * to implement sync_supers_bdi() or similar and simply do it from the | |
6f904ff0 | 275 | * bdi writeback thread individually. |
03ba3782 JA |
276 | */ |
277 | static int bdi_sync_supers(void *unused) | |
278 | { | |
279 | set_user_nice(current, 0); | |
280 | ||
281 | while (!kthread_should_stop()) { | |
282 | set_current_state(TASK_INTERRUPTIBLE); | |
283 | schedule(); | |
284 | ||
285 | /* | |
286 | * Do this periodically, like kupdated() did before. | |
287 | */ | |
288 | sync_supers(); | |
289 | } | |
290 | ||
291 | return 0; | |
292 | } | |
293 | ||
6423104b | 294 | void bdi_arm_supers_timer(void) |
03ba3782 JA |
295 | { |
296 | unsigned long next; | |
297 | ||
6423104b JA |
298 | if (!dirty_writeback_interval) |
299 | return; | |
300 | ||
03ba3782 JA |
301 | next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies; |
302 | mod_timer(&sync_supers_timer, round_jiffies_up(next)); | |
303 | } | |
304 | ||
305 | static void sync_supers_timer_fn(unsigned long unused) | |
306 | { | |
307 | wake_up_process(sync_supers_tsk); | |
6423104b | 308 | bdi_arm_supers_timer(); |
03ba3782 JA |
309 | } |
310 | ||
6467716a AB |
311 | static void wakeup_timer_fn(unsigned long data) |
312 | { | |
313 | struct backing_dev_info *bdi = (struct backing_dev_info *)data; | |
314 | ||
315 | spin_lock_bh(&bdi->wb_lock); | |
316 | if (bdi->wb.task) { | |
60332023 | 317 | trace_writeback_wake_thread(bdi); |
6467716a AB |
318 | wake_up_process(bdi->wb.task); |
319 | } else { | |
320 | /* | |
321 | * When bdi tasks are inactive for long time, they are killed. | |
322 | * In this case we have to wake-up the forker thread which | |
323 | * should create and run the bdi thread. | |
324 | */ | |
60332023 | 325 | trace_writeback_wake_forker_thread(bdi); |
6467716a AB |
326 | wake_up_process(default_backing_dev_info.wb.task); |
327 | } | |
328 | spin_unlock_bh(&bdi->wb_lock); | |
329 | } | |
330 | ||
331 | /* | |
332 | * This function is used when the first inode for this bdi is marked dirty. It | |
333 | * wakes-up the corresponding bdi thread which should then take care of the | |
334 | * periodic background write-out of dirty inodes. Since the write-out would | |
335 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | |
336 | * set up a timer which wakes the bdi thread up later. | |
337 | * | |
338 | * Note, we wouldn't bother setting up the timer, but this function is on the | |
339 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | |
340 | * by delaying the wake-up. | |
341 | */ | |
342 | void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi) | |
343 | { | |
344 | unsigned long timeout; | |
345 | ||
346 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | |
347 | mod_timer(&bdi->wb.wakeup_timer, jiffies + timeout); | |
348 | } | |
349 | ||
fff5b85a AB |
350 | /* |
351 | * Calculate the longest interval (jiffies) bdi threads are allowed to be | |
352 | * inactive. | |
353 | */ | |
354 | static unsigned long bdi_longest_inactive(void) | |
355 | { | |
356 | unsigned long interval; | |
357 | ||
358 | interval = msecs_to_jiffies(dirty_writeback_interval * 10); | |
359 | return max(5UL * 60 * HZ, interval); | |
360 | } | |
361 | ||
6f904ff0 | 362 | static int bdi_forker_thread(void *ptr) |
03ba3782 JA |
363 | { |
364 | struct bdi_writeback *me = ptr; | |
365 | ||
766f9164 | 366 | current->flags |= PF_SWAPWRITE; |
c1955ce3 CH |
367 | set_freezable(); |
368 | ||
369 | /* | |
370 | * Our parent may run at a different priority, just set us to normal | |
371 | */ | |
372 | set_user_nice(current, 0); | |
03ba3782 JA |
373 | |
374 | for (;;) { | |
fff5b85a | 375 | struct task_struct *task = NULL; |
78c40cb6 | 376 | struct backing_dev_info *bdi; |
adf39240 AB |
377 | enum { |
378 | NO_ACTION, /* Nothing to do */ | |
379 | FORK_THREAD, /* Fork bdi thread */ | |
fff5b85a | 380 | KILL_THREAD, /* Kill inactive bdi thread */ |
adf39240 | 381 | } action = NO_ACTION; |
03ba3782 JA |
382 | |
383 | /* | |
384 | * Temporary measure, we want to make sure we don't see | |
385 | * dirty data on the default backing_dev_info | |
386 | */ | |
6467716a AB |
387 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) { |
388 | del_timer(&me->wakeup_timer); | |
03ba3782 | 389 | wb_do_writeback(me, 0); |
6467716a | 390 | } |
03ba3782 | 391 | |
cfc4ba53 | 392 | spin_lock_bh(&bdi_lock); |
c5f7ad23 | 393 | set_current_state(TASK_INTERRUPTIBLE); |
03ba3782 | 394 | |
78c40cb6 | 395 | list_for_each_entry(bdi, &bdi_list, bdi_list) { |
adf39240 AB |
396 | bool have_dirty_io; |
397 | ||
398 | if (!bdi_cap_writeback_dirty(bdi) || | |
399 | bdi_cap_flush_forker(bdi)) | |
03ba3782 JA |
400 | continue; |
401 | ||
080dcec4 AB |
402 | WARN(!test_bit(BDI_registered, &bdi->state), |
403 | "bdi %p/%s is not registered!\n", bdi, bdi->name); | |
404 | ||
adf39240 AB |
405 | have_dirty_io = !list_empty(&bdi->work_list) || |
406 | wb_has_dirty_io(&bdi->wb); | |
78c40cb6 AB |
407 | |
408 | /* | |
adf39240 AB |
409 | * If the bdi has work to do, but the thread does not |
410 | * exist - create it. | |
78c40cb6 | 411 | */ |
adf39240 AB |
412 | if (!bdi->wb.task && have_dirty_io) { |
413 | /* | |
414 | * Set the pending bit - if someone will try to | |
415 | * unregister this bdi - it'll wait on this bit. | |
416 | */ | |
417 | set_bit(BDI_pending, &bdi->state); | |
418 | action = FORK_THREAD; | |
419 | break; | |
420 | } | |
fff5b85a | 421 | |
6bf05d03 JA |
422 | spin_lock(&bdi->wb_lock); |
423 | ||
fff5b85a AB |
424 | /* |
425 | * If there is no work to do and the bdi thread was | |
426 | * inactive long enough - kill it. The wb_lock is taken | |
427 | * to make sure no-one adds more work to this bdi and | |
428 | * wakes the bdi thread up. | |
429 | */ | |
430 | if (bdi->wb.task && !have_dirty_io && | |
431 | time_after(jiffies, bdi->wb.last_active + | |
432 | bdi_longest_inactive())) { | |
433 | task = bdi->wb.task; | |
434 | bdi->wb.task = NULL; | |
435 | spin_unlock(&bdi->wb_lock); | |
436 | set_bit(BDI_pending, &bdi->state); | |
437 | action = KILL_THREAD; | |
438 | break; | |
439 | } | |
6bf05d03 | 440 | spin_unlock(&bdi->wb_lock); |
03ba3782 | 441 | } |
080dcec4 | 442 | spin_unlock_bh(&bdi_lock); |
03ba3782 | 443 | |
c4ec7908 AB |
444 | /* Keep working if default bdi still has things to do */ |
445 | if (!list_empty(&me->bdi->work_list)) | |
446 | __set_current_state(TASK_RUNNING); | |
447 | ||
adf39240 AB |
448 | switch (action) { |
449 | case FORK_THREAD: | |
450 | __set_current_state(TASK_RUNNING); | |
6628bc74 AB |
451 | task = kthread_create(bdi_writeback_thread, &bdi->wb, |
452 | "flush-%s", dev_name(bdi->dev)); | |
adf39240 AB |
453 | if (IS_ERR(task)) { |
454 | /* | |
455 | * If thread creation fails, force writeout of | |
d46db3d5 WF |
456 | * the bdi from the thread. Hopefully 1024 is |
457 | * large enough for efficient IO. | |
adf39240 | 458 | */ |
d46db3d5 | 459 | writeback_inodes_wb(&bdi->wb, 1024); |
fff5b85a AB |
460 | } else { |
461 | /* | |
462 | * The spinlock makes sure we do not lose | |
463 | * wake-ups when racing with 'bdi_queue_work()'. | |
6628bc74 AB |
464 | * And as soon as the bdi thread is visible, we |
465 | * can start it. | |
fff5b85a | 466 | */ |
6467716a | 467 | spin_lock_bh(&bdi->wb_lock); |
adf39240 | 468 | bdi->wb.task = task; |
6467716a | 469 | spin_unlock_bh(&bdi->wb_lock); |
6628bc74 | 470 | wake_up_process(task); |
fff5b85a AB |
471 | } |
472 | break; | |
473 | ||
474 | case KILL_THREAD: | |
475 | __set_current_state(TASK_RUNNING); | |
476 | kthread_stop(task); | |
adf39240 | 477 | break; |
03ba3782 | 478 | |
adf39240 | 479 | case NO_ACTION: |
253c34e9 AB |
480 | if (!wb_has_dirty_io(me) || !dirty_writeback_interval) |
481 | /* | |
482 | * There are no dirty data. The only thing we | |
483 | * should now care about is checking for | |
484 | * inactive bdi threads and killing them. Thus, | |
485 | * let's sleep for longer time, save energy and | |
486 | * be friendly for battery-driven devices. | |
487 | */ | |
488 | schedule_timeout(bdi_longest_inactive()); | |
6423104b | 489 | else |
253c34e9 | 490 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
03ba3782 | 491 | try_to_freeze(); |
adf39240 | 492 | /* Back to the main loop */ |
03ba3782 JA |
493 | continue; |
494 | } | |
fff5b85a AB |
495 | |
496 | /* | |
497 | * Clear pending bit and wakeup anybody waiting to tear us down. | |
498 | */ | |
499 | clear_bit(BDI_pending, &bdi->state); | |
500 | smp_mb__after_clear_bit(); | |
501 | wake_up_bit(&bdi->state, BDI_pending); | |
03ba3782 JA |
502 | } |
503 | ||
504 | return 0; | |
505 | } | |
506 | ||
cfc4ba53 JA |
507 | /* |
508 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
509 | */ | |
510 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
511 | { | |
512 | spin_lock_bh(&bdi_lock); | |
513 | list_del_rcu(&bdi->bdi_list); | |
514 | spin_unlock_bh(&bdi_lock); | |
515 | ||
ef323088 | 516 | synchronize_rcu_expedited(); |
cfc4ba53 JA |
517 | } |
518 | ||
cf0ca9fe PZ |
519 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
520 | const char *fmt, ...) | |
521 | { | |
cf0ca9fe | 522 | va_list args; |
cf0ca9fe PZ |
523 | struct device *dev; |
524 | ||
69fc208b | 525 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
c284de61 | 526 | return 0; |
f1d0b063 | 527 | |
cf0ca9fe | 528 | va_start(args, fmt); |
19051c50 | 529 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
cf0ca9fe | 530 | va_end(args); |
c284de61 AB |
531 | if (IS_ERR(dev)) |
532 | return PTR_ERR(dev); | |
66f3b8e2 | 533 | |
cf0ca9fe | 534 | bdi->dev = dev; |
cf0ca9fe | 535 | |
03ba3782 JA |
536 | /* |
537 | * Just start the forker thread for our default backing_dev_info, | |
538 | * and add other bdi's to the list. They will get a thread created | |
539 | * on-demand when they need it. | |
540 | */ | |
541 | if (bdi_cap_flush_forker(bdi)) { | |
542 | struct bdi_writeback *wb = &bdi->wb; | |
543 | ||
6f904ff0 | 544 | wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s", |
03ba3782 | 545 | dev_name(dev)); |
c284de61 AB |
546 | if (IS_ERR(wb->task)) |
547 | return PTR_ERR(wb->task); | |
03ba3782 JA |
548 | } |
549 | ||
550 | bdi_debug_register(bdi, dev_name(dev)); | |
500b067c | 551 | set_bit(BDI_registered, &bdi->state); |
c284de61 AB |
552 | |
553 | spin_lock_bh(&bdi_lock); | |
554 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
555 | spin_unlock_bh(&bdi_lock); | |
556 | ||
455b2864 | 557 | trace_writeback_bdi_register(bdi); |
c284de61 | 558 | return 0; |
cf0ca9fe PZ |
559 | } |
560 | EXPORT_SYMBOL(bdi_register); | |
561 | ||
562 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) | |
563 | { | |
564 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); | |
565 | } | |
566 | EXPORT_SYMBOL(bdi_register_dev); | |
567 | ||
03ba3782 JA |
568 | /* |
569 | * Remove bdi from the global list and shutdown any threads we have running | |
570 | */ | |
571 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) | |
66f3b8e2 | 572 | { |
03ba3782 JA |
573 | if (!bdi_cap_writeback_dirty(bdi)) |
574 | return; | |
575 | ||
576 | /* | |
fff5b85a | 577 | * Make sure nobody finds us on the bdi_list anymore |
03ba3782 | 578 | */ |
fff5b85a | 579 | bdi_remove_from_list(bdi); |
03ba3782 JA |
580 | |
581 | /* | |
fff5b85a | 582 | * If setup is pending, wait for that to complete first |
03ba3782 | 583 | */ |
fff5b85a AB |
584 | wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait, |
585 | TASK_UNINTERRUPTIBLE); | |
03ba3782 JA |
586 | |
587 | /* | |
c1955ce3 | 588 | * Finally, kill the kernel thread. We don't need to be RCU |
c62b17a5 RD |
589 | * safe anymore, since the bdi is gone from visibility. Force |
590 | * unfreeze of the thread before calling kthread_stop(), otherwise | |
591 | * it would never exet if it is currently stuck in the refrigerator. | |
03ba3782 | 592 | */ |
c1955ce3 CH |
593 | if (bdi->wb.task) { |
594 | thaw_process(bdi->wb.task); | |
595 | kthread_stop(bdi->wb.task); | |
c62b17a5 | 596 | } |
66f3b8e2 JA |
597 | } |
598 | ||
592b09a4 JA |
599 | /* |
600 | * This bdi is going away now, make sure that no super_blocks point to it | |
601 | */ | |
602 | static void bdi_prune_sb(struct backing_dev_info *bdi) | |
603 | { | |
604 | struct super_block *sb; | |
605 | ||
606 | spin_lock(&sb_lock); | |
607 | list_for_each_entry(sb, &super_blocks, s_list) { | |
608 | if (sb->s_bdi == bdi) | |
95f28604 | 609 | sb->s_bdi = &default_backing_dev_info; |
592b09a4 JA |
610 | } |
611 | spin_unlock(&sb_lock); | |
612 | } | |
613 | ||
cf0ca9fe PZ |
614 | void bdi_unregister(struct backing_dev_info *bdi) |
615 | { | |
616 | if (bdi->dev) { | |
ccb6108f | 617 | bdi_set_min_ratio(bdi, 0); |
455b2864 | 618 | trace_writeback_bdi_unregister(bdi); |
8c4db335 | 619 | bdi_prune_sb(bdi); |
6467716a | 620 | del_timer_sync(&bdi->wb.wakeup_timer); |
8c4db335 | 621 | |
03ba3782 JA |
622 | if (!bdi_cap_flush_forker(bdi)) |
623 | bdi_wb_shutdown(bdi); | |
76f1418b | 624 | bdi_debug_unregister(bdi); |
cf0ca9fe PZ |
625 | device_unregister(bdi->dev); |
626 | bdi->dev = NULL; | |
627 | } | |
628 | } | |
629 | EXPORT_SYMBOL(bdi_unregister); | |
3fcfab16 | 630 | |
6467716a AB |
631 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
632 | { | |
633 | memset(wb, 0, sizeof(*wb)); | |
634 | ||
635 | wb->bdi = bdi; | |
636 | wb->last_old_flush = jiffies; | |
637 | INIT_LIST_HEAD(&wb->b_dirty); | |
638 | INIT_LIST_HEAD(&wb->b_io); | |
639 | INIT_LIST_HEAD(&wb->b_more_io); | |
f758eeab | 640 | spin_lock_init(&wb->list_lock); |
6467716a AB |
641 | setup_timer(&wb->wakeup_timer, wakeup_timer_fn, (unsigned long)bdi); |
642 | } | |
643 | ||
e98be2d5 WF |
644 | /* |
645 | * Initial write bandwidth: 100 MB/s | |
646 | */ | |
647 | #define INIT_BW (100 << (20 - PAGE_SHIFT)) | |
648 | ||
b2e8fb6e PZ |
649 | int bdi_init(struct backing_dev_info *bdi) |
650 | { | |
03ba3782 | 651 | int i, err; |
b2e8fb6e | 652 | |
cf0ca9fe PZ |
653 | bdi->dev = NULL; |
654 | ||
189d3c4a | 655 | bdi->min_ratio = 0; |
a42dde04 PZ |
656 | bdi->max_ratio = 100; |
657 | bdi->max_prop_frac = PROP_FRAC_BASE; | |
03ba3782 | 658 | spin_lock_init(&bdi->wb_lock); |
66f3b8e2 | 659 | INIT_LIST_HEAD(&bdi->bdi_list); |
03ba3782 JA |
660 | INIT_LIST_HEAD(&bdi->work_list); |
661 | ||
662 | bdi_wb_init(&bdi->wb, bdi); | |
663 | ||
b2e8fb6e | 664 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
ea319518 | 665 | err = percpu_counter_init(&bdi->bdi_stat[i], 0); |
04fbfdc1 PZ |
666 | if (err) |
667 | goto err; | |
668 | } | |
669 | ||
670 | bdi->dirty_exceeded = 0; | |
e98be2d5 WF |
671 | |
672 | bdi->bw_time_stamp = jiffies; | |
673 | bdi->written_stamp = 0; | |
674 | ||
675 | bdi->write_bandwidth = INIT_BW; | |
676 | bdi->avg_write_bandwidth = INIT_BW; | |
677 | ||
04fbfdc1 PZ |
678 | err = prop_local_init_percpu(&bdi->completions); |
679 | ||
680 | if (err) { | |
681 | err: | |
4b01a0b1 | 682 | while (i--) |
04fbfdc1 | 683 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
b2e8fb6e PZ |
684 | } |
685 | ||
686 | return err; | |
687 | } | |
688 | EXPORT_SYMBOL(bdi_init); | |
689 | ||
690 | void bdi_destroy(struct backing_dev_info *bdi) | |
691 | { | |
692 | int i; | |
693 | ||
ce5f8e77 JA |
694 | /* |
695 | * Splice our entries to the default_backing_dev_info, if this | |
696 | * bdi disappears | |
697 | */ | |
698 | if (bdi_has_dirty_io(bdi)) { | |
699 | struct bdi_writeback *dst = &default_backing_dev_info.wb; | |
700 | ||
f758eeab | 701 | bdi_lock_two(&bdi->wb, dst); |
ce5f8e77 JA |
702 | list_splice(&bdi->wb.b_dirty, &dst->b_dirty); |
703 | list_splice(&bdi->wb.b_io, &dst->b_io); | |
704 | list_splice(&bdi->wb.b_more_io, &dst->b_more_io); | |
f758eeab CH |
705 | spin_unlock(&bdi->wb.list_lock); |
706 | spin_unlock(&dst->list_lock); | |
ce5f8e77 | 707 | } |
66f3b8e2 | 708 | |
cf0ca9fe PZ |
709 | bdi_unregister(bdi); |
710 | ||
b2e8fb6e PZ |
711 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
712 | percpu_counter_destroy(&bdi->bdi_stat[i]); | |
04fbfdc1 PZ |
713 | |
714 | prop_local_destroy_percpu(&bdi->completions); | |
b2e8fb6e PZ |
715 | } |
716 | EXPORT_SYMBOL(bdi_destroy); | |
717 | ||
c3c53206 JA |
718 | /* |
719 | * For use from filesystems to quickly init and register a bdi associated | |
720 | * with dirty writeback | |
721 | */ | |
722 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name, | |
723 | unsigned int cap) | |
724 | { | |
725 | char tmp[32]; | |
726 | int err; | |
727 | ||
728 | bdi->name = name; | |
729 | bdi->capabilities = cap; | |
730 | err = bdi_init(bdi); | |
731 | if (err) | |
732 | return err; | |
733 | ||
734 | sprintf(tmp, "%.28s%s", name, "-%d"); | |
735 | err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq)); | |
736 | if (err) { | |
737 | bdi_destroy(bdi); | |
738 | return err; | |
739 | } | |
740 | ||
741 | return 0; | |
742 | } | |
743 | EXPORT_SYMBOL(bdi_setup_and_register); | |
744 | ||
3fcfab16 AM |
745 | static wait_queue_head_t congestion_wqh[2] = { |
746 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
747 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
748 | }; | |
0e093d99 | 749 | static atomic_t nr_bdi_congested[2]; |
3fcfab16 | 750 | |
1faa16d2 | 751 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
752 | { |
753 | enum bdi_state bit; | |
1faa16d2 | 754 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 | 755 | |
1faa16d2 | 756 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
0e093d99 MG |
757 | if (test_and_clear_bit(bit, &bdi->state)) |
758 | atomic_dec(&nr_bdi_congested[sync]); | |
3fcfab16 AM |
759 | smp_mb__after_clear_bit(); |
760 | if (waitqueue_active(wqh)) | |
761 | wake_up(wqh); | |
762 | } | |
763 | EXPORT_SYMBOL(clear_bdi_congested); | |
764 | ||
1faa16d2 | 765 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
766 | { |
767 | enum bdi_state bit; | |
768 | ||
1faa16d2 | 769 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
0e093d99 MG |
770 | if (!test_and_set_bit(bit, &bdi->state)) |
771 | atomic_inc(&nr_bdi_congested[sync]); | |
3fcfab16 AM |
772 | } |
773 | EXPORT_SYMBOL(set_bdi_congested); | |
774 | ||
775 | /** | |
776 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 777 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
778 | * @timeout: timeout in jiffies |
779 | * | |
780 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
781 | * write congestion. If no backing_devs are congested then just wait for the | |
782 | * next write to be completed. | |
783 | */ | |
8aa7e847 | 784 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
785 | { |
786 | long ret; | |
52bb9198 | 787 | unsigned long start = jiffies; |
3fcfab16 | 788 | DEFINE_WAIT(wait); |
8aa7e847 | 789 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
790 | |
791 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
792 | ret = io_schedule_timeout(timeout); | |
793 | finish_wait(wqh, &wait); | |
52bb9198 MG |
794 | |
795 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), | |
796 | jiffies_to_usecs(jiffies - start)); | |
797 | ||
3fcfab16 AM |
798 | return ret; |
799 | } | |
800 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 801 | |
0e093d99 MG |
802 | /** |
803 | * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes | |
804 | * @zone: A zone to check if it is heavily congested | |
805 | * @sync: SYNC or ASYNC IO | |
806 | * @timeout: timeout in jiffies | |
807 | * | |
808 | * In the event of a congested backing_dev (any backing_dev) and the given | |
809 | * @zone has experienced recent congestion, this waits for up to @timeout | |
810 | * jiffies for either a BDI to exit congestion of the given @sync queue | |
811 | * or a write to complete. | |
812 | * | |
25985edc | 813 | * In the absence of zone congestion, cond_resched() is called to yield |
0e093d99 MG |
814 | * the processor if necessary but otherwise does not sleep. |
815 | * | |
816 | * The return value is 0 if the sleep is for the full timeout. Otherwise, | |
817 | * it is the number of jiffies that were still remaining when the function | |
818 | * returned. return_value == timeout implies the function did not sleep. | |
819 | */ | |
820 | long wait_iff_congested(struct zone *zone, int sync, long timeout) | |
821 | { | |
822 | long ret; | |
823 | unsigned long start = jiffies; | |
824 | DEFINE_WAIT(wait); | |
825 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
826 | ||
827 | /* | |
828 | * If there is no congestion, or heavy congestion is not being | |
829 | * encountered in the current zone, yield if necessary instead | |
830 | * of sleeping on the congestion queue | |
831 | */ | |
832 | if (atomic_read(&nr_bdi_congested[sync]) == 0 || | |
833 | !zone_is_reclaim_congested(zone)) { | |
834 | cond_resched(); | |
835 | ||
836 | /* In case we scheduled, work out time remaining */ | |
837 | ret = timeout - (jiffies - start); | |
838 | if (ret < 0) | |
839 | ret = 0; | |
840 | ||
841 | goto out; | |
842 | } | |
843 | ||
844 | /* Sleep until uncongested or a write happens */ | |
845 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
846 | ret = io_schedule_timeout(timeout); | |
847 | finish_wait(wqh, &wait); | |
848 | ||
849 | out: | |
850 | trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout), | |
851 | jiffies_to_usecs(jiffies - start)); | |
852 | ||
853 | return ret; | |
854 | } | |
855 | EXPORT_SYMBOL(wait_iff_congested); |