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" | |
c8e28ce0 | 100 | "BdiDirtied: %10lu kB\n" |
00821b00 WF |
101 | "BdiWritten: %10lu kB\n" |
102 | "BdiWriteBandwidth: %10lu kBps\n" | |
103 | "b_dirty: %10lu\n" | |
104 | "b_io: %10lu\n" | |
105 | "b_more_io: %10lu\n" | |
106 | "bdi_list: %10u\n" | |
107 | "state: %10lx\n", | |
76f1418b MS |
108 | (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)), |
109 | (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)), | |
f7d2b1ec JK |
110 | K(bdi_thresh), |
111 | K(dirty_thresh), | |
112 | K(background_thresh), | |
c8e28ce0 | 113 | (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)), |
f7d2b1ec | 114 | (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)), |
00821b00 | 115 | (unsigned long) K(bdi->write_bandwidth), |
f7d2b1ec JK |
116 | nr_dirty, |
117 | nr_io, | |
118 | nr_more_io, | |
c1955ce3 | 119 | !list_empty(&bdi->bdi_list), bdi->state); |
76f1418b MS |
120 | #undef K |
121 | ||
122 | return 0; | |
123 | } | |
124 | ||
125 | static int bdi_debug_stats_open(struct inode *inode, struct file *file) | |
126 | { | |
127 | return single_open(file, bdi_debug_stats_show, inode->i_private); | |
128 | } | |
129 | ||
130 | static const struct file_operations bdi_debug_stats_fops = { | |
131 | .open = bdi_debug_stats_open, | |
132 | .read = seq_read, | |
133 | .llseek = seq_lseek, | |
134 | .release = single_release, | |
135 | }; | |
136 | ||
137 | static void bdi_debug_register(struct backing_dev_info *bdi, const char *name) | |
138 | { | |
139 | bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root); | |
140 | bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir, | |
141 | bdi, &bdi_debug_stats_fops); | |
142 | } | |
143 | ||
144 | static void bdi_debug_unregister(struct backing_dev_info *bdi) | |
145 | { | |
146 | debugfs_remove(bdi->debug_stats); | |
147 | debugfs_remove(bdi->debug_dir); | |
148 | } | |
149 | #else | |
150 | static inline void bdi_debug_init(void) | |
151 | { | |
152 | } | |
153 | static inline void bdi_debug_register(struct backing_dev_info *bdi, | |
154 | const char *name) | |
155 | { | |
156 | } | |
157 | static inline void bdi_debug_unregister(struct backing_dev_info *bdi) | |
158 | { | |
159 | } | |
160 | #endif | |
161 | ||
cf0ca9fe PZ |
162 | static ssize_t read_ahead_kb_store(struct device *dev, |
163 | struct device_attribute *attr, | |
164 | const char *buf, size_t count) | |
165 | { | |
166 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
167 | char *end; | |
168 | unsigned long read_ahead_kb; | |
169 | ssize_t ret = -EINVAL; | |
170 | ||
171 | read_ahead_kb = simple_strtoul(buf, &end, 10); | |
172 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
173 | bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10); | |
174 | ret = count; | |
175 | } | |
176 | return ret; | |
177 | } | |
178 | ||
179 | #define K(pages) ((pages) << (PAGE_SHIFT - 10)) | |
180 | ||
181 | #define BDI_SHOW(name, expr) \ | |
182 | static ssize_t name##_show(struct device *dev, \ | |
183 | struct device_attribute *attr, char *page) \ | |
184 | { \ | |
185 | struct backing_dev_info *bdi = dev_get_drvdata(dev); \ | |
186 | \ | |
187 | return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \ | |
188 | } | |
189 | ||
190 | BDI_SHOW(read_ahead_kb, K(bdi->ra_pages)) | |
191 | ||
189d3c4a PZ |
192 | static ssize_t min_ratio_store(struct device *dev, |
193 | struct device_attribute *attr, const char *buf, size_t count) | |
194 | { | |
195 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
196 | char *end; | |
197 | unsigned int ratio; | |
198 | ssize_t ret = -EINVAL; | |
199 | ||
200 | ratio = simple_strtoul(buf, &end, 10); | |
201 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
202 | ret = bdi_set_min_ratio(bdi, ratio); | |
203 | if (!ret) | |
204 | ret = count; | |
205 | } | |
206 | return ret; | |
207 | } | |
208 | BDI_SHOW(min_ratio, bdi->min_ratio) | |
209 | ||
a42dde04 PZ |
210 | static ssize_t max_ratio_store(struct device *dev, |
211 | struct device_attribute *attr, const char *buf, size_t count) | |
212 | { | |
213 | struct backing_dev_info *bdi = dev_get_drvdata(dev); | |
214 | char *end; | |
215 | unsigned int ratio; | |
216 | ssize_t ret = -EINVAL; | |
217 | ||
218 | ratio = simple_strtoul(buf, &end, 10); | |
219 | if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) { | |
220 | ret = bdi_set_max_ratio(bdi, ratio); | |
221 | if (!ret) | |
222 | ret = count; | |
223 | } | |
224 | return ret; | |
225 | } | |
226 | BDI_SHOW(max_ratio, bdi->max_ratio) | |
227 | ||
cf0ca9fe PZ |
228 | #define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store) |
229 | ||
230 | static struct device_attribute bdi_dev_attrs[] = { | |
231 | __ATTR_RW(read_ahead_kb), | |
189d3c4a | 232 | __ATTR_RW(min_ratio), |
a42dde04 | 233 | __ATTR_RW(max_ratio), |
cf0ca9fe PZ |
234 | __ATTR_NULL, |
235 | }; | |
236 | ||
237 | static __init int bdi_class_init(void) | |
238 | { | |
239 | bdi_class = class_create(THIS_MODULE, "bdi"); | |
14421453 AB |
240 | if (IS_ERR(bdi_class)) |
241 | return PTR_ERR(bdi_class); | |
242 | ||
cf0ca9fe | 243 | bdi_class->dev_attrs = bdi_dev_attrs; |
76f1418b | 244 | bdi_debug_init(); |
cf0ca9fe PZ |
245 | return 0; |
246 | } | |
76f1418b | 247 | postcore_initcall(bdi_class_init); |
cf0ca9fe | 248 | |
26160158 JA |
249 | static int __init default_bdi_init(void) |
250 | { | |
251 | int err; | |
252 | ||
03ba3782 JA |
253 | sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers"); |
254 | BUG_ON(IS_ERR(sync_supers_tsk)); | |
255 | ||
03ba3782 | 256 | setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0); |
6423104b | 257 | bdi_arm_supers_timer(); |
03ba3782 | 258 | |
26160158 JA |
259 | err = bdi_init(&default_backing_dev_info); |
260 | if (!err) | |
261 | bdi_register(&default_backing_dev_info, NULL, "default"); | |
976e48f8 | 262 | err = bdi_init(&noop_backing_dev_info); |
26160158 JA |
263 | |
264 | return err; | |
265 | } | |
266 | subsys_initcall(default_bdi_init); | |
267 | ||
03ba3782 JA |
268 | int bdi_has_dirty_io(struct backing_dev_info *bdi) |
269 | { | |
270 | return wb_has_dirty_io(&bdi->wb); | |
271 | } | |
272 | ||
03ba3782 | 273 | /* |
6f904ff0 | 274 | * kupdated() used to do this. We cannot do it from the bdi_forker_thread() |
03ba3782 JA |
275 | * or we risk deadlocking on ->s_umount. The longer term solution would be |
276 | * to implement sync_supers_bdi() or similar and simply do it from the | |
6f904ff0 | 277 | * bdi writeback thread individually. |
03ba3782 JA |
278 | */ |
279 | static int bdi_sync_supers(void *unused) | |
280 | { | |
281 | set_user_nice(current, 0); | |
282 | ||
283 | while (!kthread_should_stop()) { | |
284 | set_current_state(TASK_INTERRUPTIBLE); | |
285 | schedule(); | |
286 | ||
287 | /* | |
288 | * Do this periodically, like kupdated() did before. | |
289 | */ | |
290 | sync_supers(); | |
291 | } | |
292 | ||
293 | return 0; | |
294 | } | |
295 | ||
6423104b | 296 | void bdi_arm_supers_timer(void) |
03ba3782 JA |
297 | { |
298 | unsigned long next; | |
299 | ||
6423104b JA |
300 | if (!dirty_writeback_interval) |
301 | return; | |
302 | ||
03ba3782 JA |
303 | next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies; |
304 | mod_timer(&sync_supers_timer, round_jiffies_up(next)); | |
305 | } | |
306 | ||
307 | static void sync_supers_timer_fn(unsigned long unused) | |
308 | { | |
309 | wake_up_process(sync_supers_tsk); | |
6423104b | 310 | bdi_arm_supers_timer(); |
03ba3782 JA |
311 | } |
312 | ||
6467716a AB |
313 | static void wakeup_timer_fn(unsigned long data) |
314 | { | |
315 | struct backing_dev_info *bdi = (struct backing_dev_info *)data; | |
316 | ||
317 | spin_lock_bh(&bdi->wb_lock); | |
318 | if (bdi->wb.task) { | |
60332023 | 319 | trace_writeback_wake_thread(bdi); |
6467716a AB |
320 | wake_up_process(bdi->wb.task); |
321 | } else { | |
322 | /* | |
323 | * When bdi tasks are inactive for long time, they are killed. | |
324 | * In this case we have to wake-up the forker thread which | |
325 | * should create and run the bdi thread. | |
326 | */ | |
60332023 | 327 | trace_writeback_wake_forker_thread(bdi); |
6467716a AB |
328 | wake_up_process(default_backing_dev_info.wb.task); |
329 | } | |
330 | spin_unlock_bh(&bdi->wb_lock); | |
331 | } | |
332 | ||
333 | /* | |
334 | * This function is used when the first inode for this bdi is marked dirty. It | |
335 | * wakes-up the corresponding bdi thread which should then take care of the | |
336 | * periodic background write-out of dirty inodes. Since the write-out would | |
337 | * starts only 'dirty_writeback_interval' centisecs from now anyway, we just | |
338 | * set up a timer which wakes the bdi thread up later. | |
339 | * | |
340 | * Note, we wouldn't bother setting up the timer, but this function is on the | |
341 | * fast-path (used by '__mark_inode_dirty()'), so we save few context switches | |
342 | * by delaying the wake-up. | |
343 | */ | |
344 | void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi) | |
345 | { | |
346 | unsigned long timeout; | |
347 | ||
348 | timeout = msecs_to_jiffies(dirty_writeback_interval * 10); | |
349 | mod_timer(&bdi->wb.wakeup_timer, jiffies + timeout); | |
350 | } | |
351 | ||
fff5b85a AB |
352 | /* |
353 | * Calculate the longest interval (jiffies) bdi threads are allowed to be | |
354 | * inactive. | |
355 | */ | |
356 | static unsigned long bdi_longest_inactive(void) | |
357 | { | |
358 | unsigned long interval; | |
359 | ||
360 | interval = msecs_to_jiffies(dirty_writeback_interval * 10); | |
361 | return max(5UL * 60 * HZ, interval); | |
362 | } | |
363 | ||
5a042aa4 JK |
364 | /* |
365 | * Clear pending bit and wakeup anybody waiting for flusher thread creation or | |
366 | * shutdown | |
367 | */ | |
368 | static void bdi_clear_pending(struct backing_dev_info *bdi) | |
369 | { | |
370 | clear_bit(BDI_pending, &bdi->state); | |
371 | smp_mb__after_clear_bit(); | |
372 | wake_up_bit(&bdi->state, BDI_pending); | |
373 | } | |
374 | ||
6f904ff0 | 375 | static int bdi_forker_thread(void *ptr) |
03ba3782 JA |
376 | { |
377 | struct bdi_writeback *me = ptr; | |
378 | ||
766f9164 | 379 | current->flags |= PF_SWAPWRITE; |
c1955ce3 CH |
380 | set_freezable(); |
381 | ||
382 | /* | |
383 | * Our parent may run at a different priority, just set us to normal | |
384 | */ | |
385 | set_user_nice(current, 0); | |
03ba3782 JA |
386 | |
387 | for (;;) { | |
fff5b85a | 388 | struct task_struct *task = NULL; |
78c40cb6 | 389 | struct backing_dev_info *bdi; |
adf39240 AB |
390 | enum { |
391 | NO_ACTION, /* Nothing to do */ | |
392 | FORK_THREAD, /* Fork bdi thread */ | |
fff5b85a | 393 | KILL_THREAD, /* Kill inactive bdi thread */ |
adf39240 | 394 | } action = NO_ACTION; |
03ba3782 JA |
395 | |
396 | /* | |
397 | * Temporary measure, we want to make sure we don't see | |
398 | * dirty data on the default backing_dev_info | |
399 | */ | |
6467716a AB |
400 | if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) { |
401 | del_timer(&me->wakeup_timer); | |
03ba3782 | 402 | wb_do_writeback(me, 0); |
6467716a | 403 | } |
03ba3782 | 404 | |
cfc4ba53 | 405 | spin_lock_bh(&bdi_lock); |
09f40f98 JK |
406 | /* |
407 | * In the following loop we are going to check whether we have | |
408 | * some work to do without any synchronization with tasks | |
20c8c628 AM |
409 | * waking us up to do work for them. Set the task state here |
410 | * so that we don't miss wakeups after verifying conditions. | |
09f40f98 | 411 | */ |
c5f7ad23 | 412 | set_current_state(TASK_INTERRUPTIBLE); |
03ba3782 | 413 | |
78c40cb6 | 414 | list_for_each_entry(bdi, &bdi_list, bdi_list) { |
adf39240 AB |
415 | bool have_dirty_io; |
416 | ||
417 | if (!bdi_cap_writeback_dirty(bdi) || | |
418 | bdi_cap_flush_forker(bdi)) | |
03ba3782 JA |
419 | continue; |
420 | ||
080dcec4 AB |
421 | WARN(!test_bit(BDI_registered, &bdi->state), |
422 | "bdi %p/%s is not registered!\n", bdi, bdi->name); | |
423 | ||
adf39240 AB |
424 | have_dirty_io = !list_empty(&bdi->work_list) || |
425 | wb_has_dirty_io(&bdi->wb); | |
78c40cb6 AB |
426 | |
427 | /* | |
adf39240 AB |
428 | * If the bdi has work to do, but the thread does not |
429 | * exist - create it. | |
78c40cb6 | 430 | */ |
adf39240 AB |
431 | if (!bdi->wb.task && have_dirty_io) { |
432 | /* | |
433 | * Set the pending bit - if someone will try to | |
434 | * unregister this bdi - it'll wait on this bit. | |
435 | */ | |
436 | set_bit(BDI_pending, &bdi->state); | |
437 | action = FORK_THREAD; | |
438 | break; | |
439 | } | |
fff5b85a | 440 | |
6bf05d03 JA |
441 | spin_lock(&bdi->wb_lock); |
442 | ||
fff5b85a AB |
443 | /* |
444 | * If there is no work to do and the bdi thread was | |
445 | * inactive long enough - kill it. The wb_lock is taken | |
446 | * to make sure no-one adds more work to this bdi and | |
447 | * wakes the bdi thread up. | |
448 | */ | |
449 | if (bdi->wb.task && !have_dirty_io && | |
450 | time_after(jiffies, bdi->wb.last_active + | |
451 | bdi_longest_inactive())) { | |
452 | task = bdi->wb.task; | |
453 | bdi->wb.task = NULL; | |
454 | spin_unlock(&bdi->wb_lock); | |
455 | set_bit(BDI_pending, &bdi->state); | |
456 | action = KILL_THREAD; | |
457 | break; | |
458 | } | |
6bf05d03 | 459 | spin_unlock(&bdi->wb_lock); |
03ba3782 | 460 | } |
080dcec4 | 461 | spin_unlock_bh(&bdi_lock); |
03ba3782 | 462 | |
c4ec7908 AB |
463 | /* Keep working if default bdi still has things to do */ |
464 | if (!list_empty(&me->bdi->work_list)) | |
465 | __set_current_state(TASK_RUNNING); | |
466 | ||
adf39240 AB |
467 | switch (action) { |
468 | case FORK_THREAD: | |
469 | __set_current_state(TASK_RUNNING); | |
6628bc74 AB |
470 | task = kthread_create(bdi_writeback_thread, &bdi->wb, |
471 | "flush-%s", dev_name(bdi->dev)); | |
adf39240 AB |
472 | if (IS_ERR(task)) { |
473 | /* | |
474 | * If thread creation fails, force writeout of | |
d46db3d5 WF |
475 | * the bdi from the thread. Hopefully 1024 is |
476 | * large enough for efficient IO. | |
adf39240 | 477 | */ |
0e175a18 CW |
478 | writeback_inodes_wb(&bdi->wb, 1024, |
479 | WB_REASON_FORKER_THREAD); | |
fff5b85a AB |
480 | } else { |
481 | /* | |
482 | * The spinlock makes sure we do not lose | |
483 | * wake-ups when racing with 'bdi_queue_work()'. | |
6628bc74 AB |
484 | * And as soon as the bdi thread is visible, we |
485 | * can start it. | |
fff5b85a | 486 | */ |
6467716a | 487 | spin_lock_bh(&bdi->wb_lock); |
adf39240 | 488 | bdi->wb.task = task; |
6467716a | 489 | spin_unlock_bh(&bdi->wb_lock); |
6628bc74 | 490 | wake_up_process(task); |
fff5b85a | 491 | } |
5a042aa4 | 492 | bdi_clear_pending(bdi); |
fff5b85a AB |
493 | break; |
494 | ||
495 | case KILL_THREAD: | |
496 | __set_current_state(TASK_RUNNING); | |
497 | kthread_stop(task); | |
5a042aa4 | 498 | bdi_clear_pending(bdi); |
adf39240 | 499 | break; |
03ba3782 | 500 | |
adf39240 | 501 | case NO_ACTION: |
253c34e9 AB |
502 | if (!wb_has_dirty_io(me) || !dirty_writeback_interval) |
503 | /* | |
504 | * There are no dirty data. The only thing we | |
505 | * should now care about is checking for | |
506 | * inactive bdi threads and killing them. Thus, | |
507 | * let's sleep for longer time, save energy and | |
508 | * be friendly for battery-driven devices. | |
509 | */ | |
510 | schedule_timeout(bdi_longest_inactive()); | |
6423104b | 511 | else |
253c34e9 | 512 | schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10)); |
03ba3782 | 513 | try_to_freeze(); |
5a042aa4 | 514 | break; |
03ba3782 | 515 | } |
03ba3782 JA |
516 | } |
517 | ||
518 | return 0; | |
519 | } | |
520 | ||
cfc4ba53 JA |
521 | /* |
522 | * Remove bdi from bdi_list, and ensure that it is no longer visible | |
523 | */ | |
524 | static void bdi_remove_from_list(struct backing_dev_info *bdi) | |
525 | { | |
526 | spin_lock_bh(&bdi_lock); | |
527 | list_del_rcu(&bdi->bdi_list); | |
528 | spin_unlock_bh(&bdi_lock); | |
529 | ||
ef323088 | 530 | synchronize_rcu_expedited(); |
cfc4ba53 JA |
531 | } |
532 | ||
cf0ca9fe PZ |
533 | int bdi_register(struct backing_dev_info *bdi, struct device *parent, |
534 | const char *fmt, ...) | |
535 | { | |
cf0ca9fe | 536 | va_list args; |
cf0ca9fe PZ |
537 | struct device *dev; |
538 | ||
69fc208b | 539 | if (bdi->dev) /* The driver needs to use separate queues per device */ |
c284de61 | 540 | return 0; |
f1d0b063 | 541 | |
cf0ca9fe | 542 | va_start(args, fmt); |
19051c50 | 543 | dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args); |
cf0ca9fe | 544 | va_end(args); |
c284de61 AB |
545 | if (IS_ERR(dev)) |
546 | return PTR_ERR(dev); | |
66f3b8e2 | 547 | |
cf0ca9fe | 548 | bdi->dev = dev; |
cf0ca9fe | 549 | |
03ba3782 JA |
550 | /* |
551 | * Just start the forker thread for our default backing_dev_info, | |
552 | * and add other bdi's to the list. They will get a thread created | |
553 | * on-demand when they need it. | |
554 | */ | |
555 | if (bdi_cap_flush_forker(bdi)) { | |
556 | struct bdi_writeback *wb = &bdi->wb; | |
557 | ||
6f904ff0 | 558 | wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s", |
03ba3782 | 559 | dev_name(dev)); |
c284de61 AB |
560 | if (IS_ERR(wb->task)) |
561 | return PTR_ERR(wb->task); | |
03ba3782 JA |
562 | } |
563 | ||
564 | bdi_debug_register(bdi, dev_name(dev)); | |
500b067c | 565 | set_bit(BDI_registered, &bdi->state); |
c284de61 AB |
566 | |
567 | spin_lock_bh(&bdi_lock); | |
568 | list_add_tail_rcu(&bdi->bdi_list, &bdi_list); | |
569 | spin_unlock_bh(&bdi_lock); | |
570 | ||
455b2864 | 571 | trace_writeback_bdi_register(bdi); |
c284de61 | 572 | return 0; |
cf0ca9fe PZ |
573 | } |
574 | EXPORT_SYMBOL(bdi_register); | |
575 | ||
576 | int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev) | |
577 | { | |
578 | return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev)); | |
579 | } | |
580 | EXPORT_SYMBOL(bdi_register_dev); | |
581 | ||
03ba3782 JA |
582 | /* |
583 | * Remove bdi from the global list and shutdown any threads we have running | |
584 | */ | |
585 | static void bdi_wb_shutdown(struct backing_dev_info *bdi) | |
66f3b8e2 | 586 | { |
03ba3782 JA |
587 | if (!bdi_cap_writeback_dirty(bdi)) |
588 | return; | |
589 | ||
590 | /* | |
fff5b85a | 591 | * Make sure nobody finds us on the bdi_list anymore |
03ba3782 | 592 | */ |
fff5b85a | 593 | bdi_remove_from_list(bdi); |
03ba3782 JA |
594 | |
595 | /* | |
fff5b85a | 596 | * If setup is pending, wait for that to complete first |
03ba3782 | 597 | */ |
fff5b85a AB |
598 | wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait, |
599 | TASK_UNINTERRUPTIBLE); | |
03ba3782 JA |
600 | |
601 | /* | |
c1955ce3 | 602 | * Finally, kill the kernel thread. We don't need to be RCU |
c62b17a5 RD |
603 | * safe anymore, since the bdi is gone from visibility. Force |
604 | * unfreeze of the thread before calling kthread_stop(), otherwise | |
605 | * it would never exet if it is currently stuck in the refrigerator. | |
03ba3782 | 606 | */ |
c1955ce3 CH |
607 | if (bdi->wb.task) { |
608 | thaw_process(bdi->wb.task); | |
609 | kthread_stop(bdi->wb.task); | |
c62b17a5 | 610 | } |
66f3b8e2 JA |
611 | } |
612 | ||
592b09a4 JA |
613 | /* |
614 | * This bdi is going away now, make sure that no super_blocks point to it | |
615 | */ | |
616 | static void bdi_prune_sb(struct backing_dev_info *bdi) | |
617 | { | |
618 | struct super_block *sb; | |
619 | ||
620 | spin_lock(&sb_lock); | |
621 | list_for_each_entry(sb, &super_blocks, s_list) { | |
622 | if (sb->s_bdi == bdi) | |
95f28604 | 623 | sb->s_bdi = &default_backing_dev_info; |
592b09a4 JA |
624 | } |
625 | spin_unlock(&sb_lock); | |
626 | } | |
627 | ||
cf0ca9fe PZ |
628 | void bdi_unregister(struct backing_dev_info *bdi) |
629 | { | |
630 | if (bdi->dev) { | |
ccb6108f | 631 | bdi_set_min_ratio(bdi, 0); |
455b2864 | 632 | trace_writeback_bdi_unregister(bdi); |
8c4db335 | 633 | bdi_prune_sb(bdi); |
6467716a | 634 | del_timer_sync(&bdi->wb.wakeup_timer); |
8c4db335 | 635 | |
03ba3782 JA |
636 | if (!bdi_cap_flush_forker(bdi)) |
637 | bdi_wb_shutdown(bdi); | |
76f1418b | 638 | bdi_debug_unregister(bdi); |
cf0ca9fe PZ |
639 | device_unregister(bdi->dev); |
640 | bdi->dev = NULL; | |
641 | } | |
642 | } | |
643 | EXPORT_SYMBOL(bdi_unregister); | |
3fcfab16 | 644 | |
6467716a AB |
645 | static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi) |
646 | { | |
647 | memset(wb, 0, sizeof(*wb)); | |
648 | ||
649 | wb->bdi = bdi; | |
650 | wb->last_old_flush = jiffies; | |
651 | INIT_LIST_HEAD(&wb->b_dirty); | |
652 | INIT_LIST_HEAD(&wb->b_io); | |
653 | INIT_LIST_HEAD(&wb->b_more_io); | |
f758eeab | 654 | spin_lock_init(&wb->list_lock); |
6467716a AB |
655 | setup_timer(&wb->wakeup_timer, wakeup_timer_fn, (unsigned long)bdi); |
656 | } | |
657 | ||
e98be2d5 WF |
658 | /* |
659 | * Initial write bandwidth: 100 MB/s | |
660 | */ | |
661 | #define INIT_BW (100 << (20 - PAGE_SHIFT)) | |
662 | ||
b2e8fb6e PZ |
663 | int bdi_init(struct backing_dev_info *bdi) |
664 | { | |
03ba3782 | 665 | int i, err; |
b2e8fb6e | 666 | |
cf0ca9fe PZ |
667 | bdi->dev = NULL; |
668 | ||
189d3c4a | 669 | bdi->min_ratio = 0; |
a42dde04 PZ |
670 | bdi->max_ratio = 100; |
671 | bdi->max_prop_frac = PROP_FRAC_BASE; | |
03ba3782 | 672 | spin_lock_init(&bdi->wb_lock); |
66f3b8e2 | 673 | INIT_LIST_HEAD(&bdi->bdi_list); |
03ba3782 JA |
674 | INIT_LIST_HEAD(&bdi->work_list); |
675 | ||
676 | bdi_wb_init(&bdi->wb, bdi); | |
677 | ||
b2e8fb6e | 678 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) { |
ea319518 | 679 | err = percpu_counter_init(&bdi->bdi_stat[i], 0); |
04fbfdc1 PZ |
680 | if (err) |
681 | goto err; | |
682 | } | |
683 | ||
684 | bdi->dirty_exceeded = 0; | |
e98be2d5 WF |
685 | |
686 | bdi->bw_time_stamp = jiffies; | |
687 | bdi->written_stamp = 0; | |
688 | ||
7381131c | 689 | bdi->balanced_dirty_ratelimit = INIT_BW; |
be3ffa27 | 690 | bdi->dirty_ratelimit = INIT_BW; |
e98be2d5 WF |
691 | bdi->write_bandwidth = INIT_BW; |
692 | bdi->avg_write_bandwidth = INIT_BW; | |
693 | ||
04fbfdc1 PZ |
694 | err = prop_local_init_percpu(&bdi->completions); |
695 | ||
696 | if (err) { | |
697 | err: | |
4b01a0b1 | 698 | while (i--) |
04fbfdc1 | 699 | percpu_counter_destroy(&bdi->bdi_stat[i]); |
b2e8fb6e PZ |
700 | } |
701 | ||
702 | return err; | |
703 | } | |
704 | EXPORT_SYMBOL(bdi_init); | |
705 | ||
706 | void bdi_destroy(struct backing_dev_info *bdi) | |
707 | { | |
708 | int i; | |
709 | ||
ce5f8e77 JA |
710 | /* |
711 | * Splice our entries to the default_backing_dev_info, if this | |
712 | * bdi disappears | |
713 | */ | |
714 | if (bdi_has_dirty_io(bdi)) { | |
715 | struct bdi_writeback *dst = &default_backing_dev_info.wb; | |
716 | ||
f758eeab | 717 | bdi_lock_two(&bdi->wb, dst); |
ce5f8e77 JA |
718 | list_splice(&bdi->wb.b_dirty, &dst->b_dirty); |
719 | list_splice(&bdi->wb.b_io, &dst->b_io); | |
720 | list_splice(&bdi->wb.b_more_io, &dst->b_more_io); | |
f758eeab CH |
721 | spin_unlock(&bdi->wb.list_lock); |
722 | spin_unlock(&dst->list_lock); | |
ce5f8e77 | 723 | } |
66f3b8e2 | 724 | |
cf0ca9fe PZ |
725 | bdi_unregister(bdi); |
726 | ||
b2e8fb6e PZ |
727 | for (i = 0; i < NR_BDI_STAT_ITEMS; i++) |
728 | percpu_counter_destroy(&bdi->bdi_stat[i]); | |
04fbfdc1 PZ |
729 | |
730 | prop_local_destroy_percpu(&bdi->completions); | |
b2e8fb6e PZ |
731 | } |
732 | EXPORT_SYMBOL(bdi_destroy); | |
733 | ||
c3c53206 JA |
734 | /* |
735 | * For use from filesystems to quickly init and register a bdi associated | |
736 | * with dirty writeback | |
737 | */ | |
738 | int bdi_setup_and_register(struct backing_dev_info *bdi, char *name, | |
739 | unsigned int cap) | |
740 | { | |
741 | char tmp[32]; | |
742 | int err; | |
743 | ||
744 | bdi->name = name; | |
745 | bdi->capabilities = cap; | |
746 | err = bdi_init(bdi); | |
747 | if (err) | |
748 | return err; | |
749 | ||
750 | sprintf(tmp, "%.28s%s", name, "-%d"); | |
751 | err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq)); | |
752 | if (err) { | |
753 | bdi_destroy(bdi); | |
754 | return err; | |
755 | } | |
756 | ||
757 | return 0; | |
758 | } | |
759 | EXPORT_SYMBOL(bdi_setup_and_register); | |
760 | ||
3fcfab16 AM |
761 | static wait_queue_head_t congestion_wqh[2] = { |
762 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]), | |
763 | __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1]) | |
764 | }; | |
0e093d99 | 765 | static atomic_t nr_bdi_congested[2]; |
3fcfab16 | 766 | |
1faa16d2 | 767 | void clear_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
768 | { |
769 | enum bdi_state bit; | |
1faa16d2 | 770 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 | 771 | |
1faa16d2 | 772 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
0e093d99 MG |
773 | if (test_and_clear_bit(bit, &bdi->state)) |
774 | atomic_dec(&nr_bdi_congested[sync]); | |
3fcfab16 AM |
775 | smp_mb__after_clear_bit(); |
776 | if (waitqueue_active(wqh)) | |
777 | wake_up(wqh); | |
778 | } | |
779 | EXPORT_SYMBOL(clear_bdi_congested); | |
780 | ||
1faa16d2 | 781 | void set_bdi_congested(struct backing_dev_info *bdi, int sync) |
3fcfab16 AM |
782 | { |
783 | enum bdi_state bit; | |
784 | ||
1faa16d2 | 785 | bit = sync ? BDI_sync_congested : BDI_async_congested; |
0e093d99 MG |
786 | if (!test_and_set_bit(bit, &bdi->state)) |
787 | atomic_inc(&nr_bdi_congested[sync]); | |
3fcfab16 AM |
788 | } |
789 | EXPORT_SYMBOL(set_bdi_congested); | |
790 | ||
791 | /** | |
792 | * congestion_wait - wait for a backing_dev to become uncongested | |
8aa7e847 | 793 | * @sync: SYNC or ASYNC IO |
3fcfab16 AM |
794 | * @timeout: timeout in jiffies |
795 | * | |
796 | * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit | |
797 | * write congestion. If no backing_devs are congested then just wait for the | |
798 | * next write to be completed. | |
799 | */ | |
8aa7e847 | 800 | long congestion_wait(int sync, long timeout) |
3fcfab16 AM |
801 | { |
802 | long ret; | |
52bb9198 | 803 | unsigned long start = jiffies; |
3fcfab16 | 804 | DEFINE_WAIT(wait); |
8aa7e847 | 805 | wait_queue_head_t *wqh = &congestion_wqh[sync]; |
3fcfab16 AM |
806 | |
807 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
808 | ret = io_schedule_timeout(timeout); | |
809 | finish_wait(wqh, &wait); | |
52bb9198 MG |
810 | |
811 | trace_writeback_congestion_wait(jiffies_to_usecs(timeout), | |
812 | jiffies_to_usecs(jiffies - start)); | |
813 | ||
3fcfab16 AM |
814 | return ret; |
815 | } | |
816 | EXPORT_SYMBOL(congestion_wait); | |
04fbfdc1 | 817 | |
0e093d99 MG |
818 | /** |
819 | * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes | |
820 | * @zone: A zone to check if it is heavily congested | |
821 | * @sync: SYNC or ASYNC IO | |
822 | * @timeout: timeout in jiffies | |
823 | * | |
824 | * In the event of a congested backing_dev (any backing_dev) and the given | |
825 | * @zone has experienced recent congestion, this waits for up to @timeout | |
826 | * jiffies for either a BDI to exit congestion of the given @sync queue | |
827 | * or a write to complete. | |
828 | * | |
25985edc | 829 | * In the absence of zone congestion, cond_resched() is called to yield |
0e093d99 MG |
830 | * the processor if necessary but otherwise does not sleep. |
831 | * | |
832 | * The return value is 0 if the sleep is for the full timeout. Otherwise, | |
833 | * it is the number of jiffies that were still remaining when the function | |
834 | * returned. return_value == timeout implies the function did not sleep. | |
835 | */ | |
836 | long wait_iff_congested(struct zone *zone, int sync, long timeout) | |
837 | { | |
838 | long ret; | |
839 | unsigned long start = jiffies; | |
840 | DEFINE_WAIT(wait); | |
841 | wait_queue_head_t *wqh = &congestion_wqh[sync]; | |
842 | ||
843 | /* | |
844 | * If there is no congestion, or heavy congestion is not being | |
845 | * encountered in the current zone, yield if necessary instead | |
846 | * of sleeping on the congestion queue | |
847 | */ | |
848 | if (atomic_read(&nr_bdi_congested[sync]) == 0 || | |
849 | !zone_is_reclaim_congested(zone)) { | |
850 | cond_resched(); | |
851 | ||
852 | /* In case we scheduled, work out time remaining */ | |
853 | ret = timeout - (jiffies - start); | |
854 | if (ret < 0) | |
855 | ret = 0; | |
856 | ||
857 | goto out; | |
858 | } | |
859 | ||
860 | /* Sleep until uncongested or a write happens */ | |
861 | prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE); | |
862 | ret = io_schedule_timeout(timeout); | |
863 | finish_wait(wqh, &wait); | |
864 | ||
865 | out: | |
866 | trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout), | |
867 | jiffies_to_usecs(jiffies - start)); | |
868 | ||
869 | return ret; | |
870 | } | |
871 | EXPORT_SYMBOL(wait_iff_congested); |