blob: 027100d30227fead0a4010d1feb0e2791a98fcaa [file] [log] [blame]
Andrew Morton3fcfab12006-10-19 23:28:16 -07001
2#include <linux/wait.h>
3#include <linux/backing-dev.h>
Jens Axboe03ba3782009-09-09 09:08:54 +02004#include <linux/kthread.h>
5#include <linux/freezer.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -07006#include <linux/fs.h>
Jens Axboe26160152009-03-17 09:35:06 +01007#include <linux/pagemap.h>
Jens Axboe03ba3782009-09-09 09:08:54 +02008#include <linux/mm.h>
Andrew Morton3fcfab12006-10-19 23:28:16 -07009#include <linux/sched.h>
10#include <linux/module.h>
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070011#include <linux/writeback.h>
12#include <linux/device.h>
Dave Chinner455b2862010-07-07 13:24:06 +100013#include <trace/events/writeback.h>
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070014
Jens Axboec3c53202010-04-22 11:37:01 +020015static atomic_long_t bdi_seq = ATOMIC_LONG_INIT(0);
16
Jens Axboe26160152009-03-17 09:35:06 +010017void default_unplug_io_fn(struct backing_dev_info *bdi, struct page *page)
18{
19}
20EXPORT_SYMBOL(default_unplug_io_fn);
21
22struct backing_dev_info default_backing_dev_info = {
Jens Axboed9938312009-06-12 14:45:52 +020023 .name = "default",
Jens Axboe26160152009-03-17 09:35:06 +010024 .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE,
25 .state = 0,
26 .capabilities = BDI_CAP_MAP_COPY,
27 .unplug_io_fn = default_unplug_io_fn,
28};
29EXPORT_SYMBOL_GPL(default_backing_dev_info);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070030
Jörn Engel5129a462010-04-25 08:54:42 +020031struct backing_dev_info noop_backing_dev_info = {
32 .name = "noop",
Jan Kara976e48f2010-09-21 11:48:55 +020033 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Jörn Engel5129a462010-04-25 08:54:42 +020034};
35EXPORT_SYMBOL_GPL(noop_backing_dev_info);
36
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070037static struct class *bdi_class;
Jens Axboecfc4ba52009-09-14 13:12:40 +020038
39/*
40 * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as
41 * reader side protection for bdi_pending_list. bdi_list has RCU reader side
42 * locking.
43 */
Jens Axboe03ba3782009-09-09 09:08:54 +020044DEFINE_SPINLOCK(bdi_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +020045LIST_HEAD(bdi_list);
Jens Axboe03ba3782009-09-09 09:08:54 +020046LIST_HEAD(bdi_pending_list);
47
48static struct task_struct *sync_supers_tsk;
49static struct timer_list sync_supers_timer;
50
51static int bdi_sync_supers(void *);
52static void sync_supers_timer_fn(unsigned long);
Jens Axboe03ba3782009-09-09 09:08:54 +020053
Miklos Szeredi76f14182008-04-30 00:54:36 -070054#ifdef CONFIG_DEBUG_FS
55#include <linux/debugfs.h>
56#include <linux/seq_file.h>
57
58static struct dentry *bdi_debug_root;
59
60static void bdi_debug_init(void)
61{
62 bdi_debug_root = debugfs_create_dir("bdi", NULL);
63}
64
65static int bdi_debug_stats_show(struct seq_file *m, void *v)
66{
67 struct backing_dev_info *bdi = m->private;
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020068 struct bdi_writeback *wb = &bdi->wb;
David Rientjes364aeb22009-01-06 14:39:29 -080069 unsigned long background_thresh;
70 unsigned long dirty_thresh;
71 unsigned long bdi_thresh;
Jens Axboef09b00d2009-05-25 09:08:21 +020072 unsigned long nr_dirty, nr_io, nr_more_io, nr_wb;
73 struct inode *inode;
74
Jens Axboef09b00d2009-05-25 09:08:21 +020075 nr_wb = nr_dirty = nr_io = nr_more_io = 0;
76 spin_lock(&inode_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +110077 list_for_each_entry(inode, &wb->b_dirty, i_wb_list)
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020078 nr_dirty++;
Nick Piggin7ccf19a2010-10-21 11:49:30 +110079 list_for_each_entry(inode, &wb->b_io, i_wb_list)
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020080 nr_io++;
Nick Piggin7ccf19a2010-10-21 11:49:30 +110081 list_for_each_entry(inode, &wb->b_more_io, i_wb_list)
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020082 nr_more_io++;
Jens Axboef09b00d2009-05-25 09:08:21 +020083 spin_unlock(&inode_lock);
Miklos Szeredi76f14182008-04-30 00:54:36 -070084
Wu Fengguang16c40422010-08-11 14:17:39 -070085 global_dirty_limits(&background_thresh, &dirty_thresh);
86 bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
Miklos Szeredi76f14182008-04-30 00:54:36 -070087
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"
Jens Axboef09b00d2009-05-25 09:08:21 +020094 "BackgroundThresh: %8lu kB\n"
Jens Axboef09b00d2009-05-25 09:08:21 +020095 "b_dirty: %8lu\n"
96 "b_io: %8lu\n"
97 "b_more_io: %8lu\n"
98 "bdi_list: %8u\n"
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020099 "state: %8lx\n",
Miklos Szeredi76f14182008-04-30 00:54:36 -0700100 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
101 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
Jens Axboef09b00d2009-05-25 09:08:21 +0200102 K(bdi_thresh), K(dirty_thresh),
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200103 K(background_thresh), nr_dirty, nr_io, nr_more_io,
104 !list_empty(&bdi->bdi_list), bdi->state);
Miklos Szeredi76f14182008-04-30 00:54:36 -0700105#undef K
106
107 return 0;
108}
109
110static int bdi_debug_stats_open(struct inode *inode, struct file *file)
111{
112 return single_open(file, bdi_debug_stats_show, inode->i_private);
113}
114
115static const struct file_operations bdi_debug_stats_fops = {
116 .open = bdi_debug_stats_open,
117 .read = seq_read,
118 .llseek = seq_lseek,
119 .release = single_release,
120};
121
122static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
123{
124 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
125 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
126 bdi, &bdi_debug_stats_fops);
127}
128
129static void bdi_debug_unregister(struct backing_dev_info *bdi)
130{
131 debugfs_remove(bdi->debug_stats);
132 debugfs_remove(bdi->debug_dir);
133}
134#else
135static inline void bdi_debug_init(void)
136{
137}
138static inline void bdi_debug_register(struct backing_dev_info *bdi,
139 const char *name)
140{
141}
142static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
143{
144}
145#endif
146
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700147static ssize_t read_ahead_kb_store(struct device *dev,
148 struct device_attribute *attr,
149 const char *buf, size_t count)
150{
151 struct backing_dev_info *bdi = dev_get_drvdata(dev);
152 char *end;
153 unsigned long read_ahead_kb;
154 ssize_t ret = -EINVAL;
155
156 read_ahead_kb = simple_strtoul(buf, &end, 10);
157 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
158 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
159 ret = count;
160 }
161 return ret;
162}
163
164#define K(pages) ((pages) << (PAGE_SHIFT - 10))
165
166#define BDI_SHOW(name, expr) \
167static ssize_t name##_show(struct device *dev, \
168 struct device_attribute *attr, char *page) \
169{ \
170 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
171 \
172 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
173}
174
175BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
176
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700177static ssize_t min_ratio_store(struct device *dev,
178 struct device_attribute *attr, const char *buf, size_t count)
179{
180 struct backing_dev_info *bdi = dev_get_drvdata(dev);
181 char *end;
182 unsigned int ratio;
183 ssize_t ret = -EINVAL;
184
185 ratio = simple_strtoul(buf, &end, 10);
186 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
187 ret = bdi_set_min_ratio(bdi, ratio);
188 if (!ret)
189 ret = count;
190 }
191 return ret;
192}
193BDI_SHOW(min_ratio, bdi->min_ratio)
194
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700195static ssize_t max_ratio_store(struct device *dev,
196 struct device_attribute *attr, const char *buf, size_t count)
197{
198 struct backing_dev_info *bdi = dev_get_drvdata(dev);
199 char *end;
200 unsigned int ratio;
201 ssize_t ret = -EINVAL;
202
203 ratio = simple_strtoul(buf, &end, 10);
204 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
205 ret = bdi_set_max_ratio(bdi, ratio);
206 if (!ret)
207 ret = count;
208 }
209 return ret;
210}
211BDI_SHOW(max_ratio, bdi->max_ratio)
212
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700213#define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
214
215static struct device_attribute bdi_dev_attrs[] = {
216 __ATTR_RW(read_ahead_kb),
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700217 __ATTR_RW(min_ratio),
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700218 __ATTR_RW(max_ratio),
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700219 __ATTR_NULL,
220};
221
222static __init int bdi_class_init(void)
223{
224 bdi_class = class_create(THIS_MODULE, "bdi");
Anton Blanchard14421452010-04-02 09:46:55 +0200225 if (IS_ERR(bdi_class))
226 return PTR_ERR(bdi_class);
227
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700228 bdi_class->dev_attrs = bdi_dev_attrs;
Miklos Szeredi76f14182008-04-30 00:54:36 -0700229 bdi_debug_init();
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700230 return 0;
231}
Miklos Szeredi76f14182008-04-30 00:54:36 -0700232postcore_initcall(bdi_class_init);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700233
Jens Axboe26160152009-03-17 09:35:06 +0100234static int __init default_bdi_init(void)
235{
236 int err;
237
Jens Axboe03ba3782009-09-09 09:08:54 +0200238 sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
239 BUG_ON(IS_ERR(sync_supers_tsk));
240
Jens Axboe03ba3782009-09-09 09:08:54 +0200241 setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0);
Jens Axboe64231042010-05-21 20:00:35 +0200242 bdi_arm_supers_timer();
Jens Axboe03ba3782009-09-09 09:08:54 +0200243
Jens Axboe26160152009-03-17 09:35:06 +0100244 err = bdi_init(&default_backing_dev_info);
245 if (!err)
246 bdi_register(&default_backing_dev_info, NULL, "default");
Jan Kara976e48f2010-09-21 11:48:55 +0200247 err = bdi_init(&noop_backing_dev_info);
Jens Axboe26160152009-03-17 09:35:06 +0100248
249 return err;
250}
251subsys_initcall(default_bdi_init);
252
Jens Axboe03ba3782009-09-09 09:08:54 +0200253int bdi_has_dirty_io(struct backing_dev_info *bdi)
254{
255 return wb_has_dirty_io(&bdi->wb);
256}
257
258static void bdi_flush_io(struct backing_dev_info *bdi)
259{
260 struct writeback_control wbc = {
Jens Axboe03ba3782009-09-09 09:08:54 +0200261 .sync_mode = WB_SYNC_NONE,
262 .older_than_this = NULL,
263 .range_cyclic = 1,
264 .nr_to_write = 1024,
265 };
266
Christoph Hellwig9c3a8ee2010-06-10 12:07:27 +0200267 writeback_inodes_wb(&bdi->wb, &wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200268}
269
270/*
Artem Bityutskiy6f904ff2010-07-25 14:29:11 +0300271 * kupdated() used to do this. We cannot do it from the bdi_forker_thread()
Jens Axboe03ba3782009-09-09 09:08:54 +0200272 * or we risk deadlocking on ->s_umount. The longer term solution would be
273 * to implement sync_supers_bdi() or similar and simply do it from the
Artem Bityutskiy6f904ff2010-07-25 14:29:11 +0300274 * bdi writeback thread individually.
Jens Axboe03ba3782009-09-09 09:08:54 +0200275 */
276static int bdi_sync_supers(void *unused)
277{
278 set_user_nice(current, 0);
279
280 while (!kthread_should_stop()) {
281 set_current_state(TASK_INTERRUPTIBLE);
282 schedule();
283
284 /*
285 * Do this periodically, like kupdated() did before.
286 */
287 sync_supers();
288 }
289
290 return 0;
291}
292
Jens Axboe64231042010-05-21 20:00:35 +0200293void bdi_arm_supers_timer(void)
Jens Axboe03ba3782009-09-09 09:08:54 +0200294{
295 unsigned long next;
296
Jens Axboe64231042010-05-21 20:00:35 +0200297 if (!dirty_writeback_interval)
298 return;
299
Jens Axboe03ba3782009-09-09 09:08:54 +0200300 next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
301 mod_timer(&sync_supers_timer, round_jiffies_up(next));
302}
303
304static void sync_supers_timer_fn(unsigned long unused)
305{
306 wake_up_process(sync_supers_tsk);
Jens Axboe64231042010-05-21 20:00:35 +0200307 bdi_arm_supers_timer();
Jens Axboe03ba3782009-09-09 09:08:54 +0200308}
309
Artem Bityutskiy64677162010-07-25 14:29:22 +0300310static void wakeup_timer_fn(unsigned long data)
311{
312 struct backing_dev_info *bdi = (struct backing_dev_info *)data;
313
314 spin_lock_bh(&bdi->wb_lock);
315 if (bdi->wb.task) {
Artem Bityutskiy60332022010-07-25 14:29:24 +0300316 trace_writeback_wake_thread(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300317 wake_up_process(bdi->wb.task);
318 } else {
319 /*
320 * When bdi tasks are inactive for long time, they are killed.
321 * In this case we have to wake-up the forker thread which
322 * should create and run the bdi thread.
323 */
Artem Bityutskiy60332022010-07-25 14:29:24 +0300324 trace_writeback_wake_forker_thread(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300325 wake_up_process(default_backing_dev_info.wb.task);
326 }
327 spin_unlock_bh(&bdi->wb_lock);
328}
329
330/*
331 * This function is used when the first inode for this bdi is marked dirty. It
332 * wakes-up the corresponding bdi thread which should then take care of the
333 * periodic background write-out of dirty inodes. Since the write-out would
334 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
335 * set up a timer which wakes the bdi thread up later.
336 *
337 * Note, we wouldn't bother setting up the timer, but this function is on the
338 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
339 * by delaying the wake-up.
340 */
341void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
342{
343 unsigned long timeout;
344
345 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
346 mod_timer(&bdi->wb.wakeup_timer, jiffies + timeout);
347}
348
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300349/*
350 * Calculate the longest interval (jiffies) bdi threads are allowed to be
351 * inactive.
352 */
353static unsigned long bdi_longest_inactive(void)
354{
355 unsigned long interval;
356
357 interval = msecs_to_jiffies(dirty_writeback_interval * 10);
358 return max(5UL * 60 * HZ, interval);
359}
360
Artem Bityutskiy6f904ff2010-07-25 14:29:11 +0300361static int bdi_forker_thread(void *ptr)
Jens Axboe03ba3782009-09-09 09:08:54 +0200362{
363 struct bdi_writeback *me = ptr;
364
Peter Zijlstra766f9162010-10-26 14:22:45 -0700365 current->flags |= PF_SWAPWRITE;
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200366 set_freezable();
367
368 /*
369 * Our parent may run at a different priority, just set us to normal
370 */
371 set_user_nice(current, 0);
Jens Axboe03ba3782009-09-09 09:08:54 +0200372
373 for (;;) {
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300374 struct task_struct *task = NULL;
Artem Bityutskiy78c40cb2010-07-25 14:29:17 +0300375 struct backing_dev_info *bdi;
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300376 enum {
377 NO_ACTION, /* Nothing to do */
378 FORK_THREAD, /* Fork bdi thread */
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300379 KILL_THREAD, /* Kill inactive bdi thread */
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300380 } action = NO_ACTION;
Jens Axboe03ba3782009-09-09 09:08:54 +0200381
382 /*
383 * Temporary measure, we want to make sure we don't see
384 * dirty data on the default backing_dev_info
385 */
Artem Bityutskiy64677162010-07-25 14:29:22 +0300386 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) {
387 del_timer(&me->wakeup_timer);
Jens Axboe03ba3782009-09-09 09:08:54 +0200388 wb_do_writeback(me, 0);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300389 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200390
Jens Axboecfc4ba52009-09-14 13:12:40 +0200391 spin_lock_bh(&bdi_lock);
Artem Bityutskiyc5f7ad22010-07-25 14:29:13 +0300392 set_current_state(TASK_INTERRUPTIBLE);
Jens Axboe03ba3782009-09-09 09:08:54 +0200393
Artem Bityutskiy78c40cb2010-07-25 14:29:17 +0300394 list_for_each_entry(bdi, &bdi_list, bdi_list) {
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300395 bool have_dirty_io;
396
397 if (!bdi_cap_writeback_dirty(bdi) ||
398 bdi_cap_flush_forker(bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200399 continue;
400
Artem Bityutskiy080dcec2010-07-25 14:29:16 +0300401 WARN(!test_bit(BDI_registered, &bdi->state),
402 "bdi %p/%s is not registered!\n", bdi, bdi->name);
403
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300404 have_dirty_io = !list_empty(&bdi->work_list) ||
405 wb_has_dirty_io(&bdi->wb);
Artem Bityutskiy78c40cb2010-07-25 14:29:17 +0300406
407 /*
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300408 * If the bdi has work to do, but the thread does not
409 * exist - create it.
Artem Bityutskiy78c40cb2010-07-25 14:29:17 +0300410 */
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300411 if (!bdi->wb.task && have_dirty_io) {
412 /*
413 * Set the pending bit - if someone will try to
414 * unregister this bdi - it'll wait on this bit.
415 */
416 set_bit(BDI_pending, &bdi->state);
417 action = FORK_THREAD;
418 break;
419 }
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300420
Jens Axboe6bf05d02010-08-04 13:34:31 +0200421 spin_lock(&bdi->wb_lock);
422
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300423 /*
424 * If there is no work to do and the bdi thread was
425 * inactive long enough - kill it. The wb_lock is taken
426 * to make sure no-one adds more work to this bdi and
427 * wakes the bdi thread up.
428 */
429 if (bdi->wb.task && !have_dirty_io &&
430 time_after(jiffies, bdi->wb.last_active +
431 bdi_longest_inactive())) {
432 task = bdi->wb.task;
433 bdi->wb.task = NULL;
434 spin_unlock(&bdi->wb_lock);
435 set_bit(BDI_pending, &bdi->state);
436 action = KILL_THREAD;
437 break;
438 }
Jens Axboe6bf05d02010-08-04 13:34:31 +0200439 spin_unlock(&bdi->wb_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200440 }
Artem Bityutskiy080dcec2010-07-25 14:29:16 +0300441 spin_unlock_bh(&bdi_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200442
Artem Bityutskiyc4ec7902010-07-25 14:29:14 +0300443 /* Keep working if default bdi still has things to do */
444 if (!list_empty(&me->bdi->work_list))
445 __set_current_state(TASK_RUNNING);
446
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300447 switch (action) {
448 case FORK_THREAD:
449 __set_current_state(TASK_RUNNING);
Artem Bityutskiy6628bc72010-08-27 09:15:09 +0200450 task = kthread_create(bdi_writeback_thread, &bdi->wb,
451 "flush-%s", dev_name(bdi->dev));
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300452 if (IS_ERR(task)) {
453 /*
454 * If thread creation fails, force writeout of
455 * the bdi from the thread.
456 */
457 bdi_flush_io(bdi);
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300458 } else {
459 /*
460 * The spinlock makes sure we do not lose
461 * wake-ups when racing with 'bdi_queue_work()'.
Artem Bityutskiy6628bc72010-08-27 09:15:09 +0200462 * And as soon as the bdi thread is visible, we
463 * can start it.
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300464 */
Artem Bityutskiy64677162010-07-25 14:29:22 +0300465 spin_lock_bh(&bdi->wb_lock);
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300466 bdi->wb.task = task;
Artem Bityutskiy64677162010-07-25 14:29:22 +0300467 spin_unlock_bh(&bdi->wb_lock);
Artem Bityutskiy6628bc72010-08-27 09:15:09 +0200468 wake_up_process(task);
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300469 }
470 break;
471
472 case KILL_THREAD:
473 __set_current_state(TASK_RUNNING);
474 kthread_stop(task);
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300475 break;
Jens Axboe03ba3782009-09-09 09:08:54 +0200476
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300477 case NO_ACTION:
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300478 if (!wb_has_dirty_io(me) || !dirty_writeback_interval)
479 /*
480 * There are no dirty data. The only thing we
481 * should now care about is checking for
482 * inactive bdi threads and killing them. Thus,
483 * let's sleep for longer time, save energy and
484 * be friendly for battery-driven devices.
485 */
486 schedule_timeout(bdi_longest_inactive());
Jens Axboe64231042010-05-21 20:00:35 +0200487 else
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300488 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Jens Axboe03ba3782009-09-09 09:08:54 +0200489 try_to_freeze();
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300490 /* Back to the main loop */
Jens Axboe03ba3782009-09-09 09:08:54 +0200491 continue;
492 }
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300493
494 /*
495 * Clear pending bit and wakeup anybody waiting to tear us down.
496 */
497 clear_bit(BDI_pending, &bdi->state);
498 smp_mb__after_clear_bit();
499 wake_up_bit(&bdi->state, BDI_pending);
Jens Axboe03ba3782009-09-09 09:08:54 +0200500 }
501
502 return 0;
503}
504
Jens Axboecfc4ba52009-09-14 13:12:40 +0200505/*
506 * Remove bdi from bdi_list, and ensure that it is no longer visible
507 */
508static void bdi_remove_from_list(struct backing_dev_info *bdi)
509{
510 spin_lock_bh(&bdi_lock);
511 list_del_rcu(&bdi->bdi_list);
512 spin_unlock_bh(&bdi_lock);
513
514 synchronize_rcu();
515}
516
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700517int bdi_register(struct backing_dev_info *bdi, struct device *parent,
518 const char *fmt, ...)
519{
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700520 va_list args;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700521 struct device *dev;
522
Andrew Morton69fc2082008-12-09 13:14:06 -0800523 if (bdi->dev) /* The driver needs to use separate queues per device */
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300524 return 0;
Kay Sieversf1d0b062008-12-02 10:31:50 -0800525
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700526 va_start(args, fmt);
Greg Kroah-Hartman19051c52008-05-15 13:44:08 -0700527 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700528 va_end(args);
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300529 if (IS_ERR(dev))
530 return PTR_ERR(dev);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200531
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700532 bdi->dev = dev;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700533
Jens Axboe03ba3782009-09-09 09:08:54 +0200534 /*
535 * Just start the forker thread for our default backing_dev_info,
536 * and add other bdi's to the list. They will get a thread created
537 * on-demand when they need it.
538 */
539 if (bdi_cap_flush_forker(bdi)) {
540 struct bdi_writeback *wb = &bdi->wb;
541
Artem Bityutskiy6f904ff2010-07-25 14:29:11 +0300542 wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s",
Jens Axboe03ba3782009-09-09 09:08:54 +0200543 dev_name(dev));
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300544 if (IS_ERR(wb->task))
545 return PTR_ERR(wb->task);
Jens Axboe03ba3782009-09-09 09:08:54 +0200546 }
547
548 bdi_debug_register(bdi, dev_name(dev));
Jens Axboe500b0672009-09-09 09:10:25 +0200549 set_bit(BDI_registered, &bdi->state);
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300550
551 spin_lock_bh(&bdi_lock);
552 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
553 spin_unlock_bh(&bdi_lock);
554
Dave Chinner455b2862010-07-07 13:24:06 +1000555 trace_writeback_bdi_register(bdi);
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300556 return 0;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700557}
558EXPORT_SYMBOL(bdi_register);
559
560int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
561{
562 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
563}
564EXPORT_SYMBOL(bdi_register_dev);
565
Jens Axboe03ba3782009-09-09 09:08:54 +0200566/*
567 * Remove bdi from the global list and shutdown any threads we have running
568 */
569static void bdi_wb_shutdown(struct backing_dev_info *bdi)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200570{
Jens Axboe03ba3782009-09-09 09:08:54 +0200571 if (!bdi_cap_writeback_dirty(bdi))
572 return;
573
574 /*
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300575 * Make sure nobody finds us on the bdi_list anymore
576 */
577 bdi_remove_from_list(bdi);
578
579 /*
Jens Axboe03ba3782009-09-09 09:08:54 +0200580 * If setup is pending, wait for that to complete first
581 */
582 wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
583 TASK_UNINTERRUPTIBLE);
584
585 /*
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200586 * Finally, kill the kernel thread. We don't need to be RCU
Romit Dasguptac62b17a2009-11-12 13:08:11 +0100587 * safe anymore, since the bdi is gone from visibility. Force
588 * unfreeze of the thread before calling kthread_stop(), otherwise
589 * it would never exet if it is currently stuck in the refrigerator.
Jens Axboe03ba3782009-09-09 09:08:54 +0200590 */
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200591 if (bdi->wb.task) {
592 thaw_process(bdi->wb.task);
593 kthread_stop(bdi->wb.task);
Romit Dasguptac62b17a2009-11-12 13:08:11 +0100594 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200595}
596
Jens Axboe592b09a2009-10-29 11:46:12 +0100597/*
598 * This bdi is going away now, make sure that no super_blocks point to it
599 */
600static void bdi_prune_sb(struct backing_dev_info *bdi)
601{
602 struct super_block *sb;
603
604 spin_lock(&sb_lock);
605 list_for_each_entry(sb, &super_blocks, s_list) {
606 if (sb->s_bdi == bdi)
607 sb->s_bdi = NULL;
608 }
609 spin_unlock(&sb_lock);
610}
611
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700612void bdi_unregister(struct backing_dev_info *bdi)
613{
614 if (bdi->dev) {
Dave Chinner455b2862010-07-07 13:24:06 +1000615 trace_writeback_bdi_unregister(bdi);
Jens Axboe8c4db332009-11-03 20:18:44 +0100616 bdi_prune_sb(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300617 del_timer_sync(&bdi->wb.wakeup_timer);
Jens Axboe8c4db332009-11-03 20:18:44 +0100618
Jens Axboe03ba3782009-09-09 09:08:54 +0200619 if (!bdi_cap_flush_forker(bdi))
620 bdi_wb_shutdown(bdi);
Miklos Szeredi76f14182008-04-30 00:54:36 -0700621 bdi_debug_unregister(bdi);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700622 device_unregister(bdi->dev);
623 bdi->dev = NULL;
624 }
625}
626EXPORT_SYMBOL(bdi_unregister);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700627
Artem Bityutskiy64677162010-07-25 14:29:22 +0300628static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
629{
630 memset(wb, 0, sizeof(*wb));
631
632 wb->bdi = bdi;
633 wb->last_old_flush = jiffies;
634 INIT_LIST_HEAD(&wb->b_dirty);
635 INIT_LIST_HEAD(&wb->b_io);
636 INIT_LIST_HEAD(&wb->b_more_io);
637 setup_timer(&wb->wakeup_timer, wakeup_timer_fn, (unsigned long)bdi);
638}
639
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700640int bdi_init(struct backing_dev_info *bdi)
641{
Jens Axboe03ba3782009-09-09 09:08:54 +0200642 int i, err;
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700643
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700644 bdi->dev = NULL;
645
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700646 bdi->min_ratio = 0;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700647 bdi->max_ratio = 100;
648 bdi->max_prop_frac = PROP_FRAC_BASE;
Jens Axboe03ba3782009-09-09 09:08:54 +0200649 spin_lock_init(&bdi->wb_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200650 INIT_LIST_HEAD(&bdi->bdi_list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200651 INIT_LIST_HEAD(&bdi->work_list);
652
653 bdi_wb_init(&bdi->wb, bdi);
654
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700655 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
Peter Zijlstraea319512008-12-26 15:08:55 +0100656 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700657 if (err)
658 goto err;
659 }
660
661 bdi->dirty_exceeded = 0;
662 err = prop_local_init_percpu(&bdi->completions);
663
664 if (err) {
665err:
Denis Cheng4b01a0b2007-12-04 23:45:07 -0800666 while (i--)
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700667 percpu_counter_destroy(&bdi->bdi_stat[i]);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700668 }
669
670 return err;
671}
672EXPORT_SYMBOL(bdi_init);
673
674void bdi_destroy(struct backing_dev_info *bdi)
675{
676 int i;
677
Jens Axboece5f8e72009-09-14 12:57:56 +0200678 /*
679 * Splice our entries to the default_backing_dev_info, if this
680 * bdi disappears
681 */
682 if (bdi_has_dirty_io(bdi)) {
683 struct bdi_writeback *dst = &default_backing_dev_info.wb;
684
685 spin_lock(&inode_lock);
686 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
687 list_splice(&bdi->wb.b_io, &dst->b_io);
688 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
689 spin_unlock(&inode_lock);
690 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200691
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700692 bdi_unregister(bdi);
693
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700694 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
695 percpu_counter_destroy(&bdi->bdi_stat[i]);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700696
697 prop_local_destroy_percpu(&bdi->completions);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700698}
699EXPORT_SYMBOL(bdi_destroy);
700
Jens Axboec3c53202010-04-22 11:37:01 +0200701/*
702 * For use from filesystems to quickly init and register a bdi associated
703 * with dirty writeback
704 */
705int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
706 unsigned int cap)
707{
708 char tmp[32];
709 int err;
710
711 bdi->name = name;
712 bdi->capabilities = cap;
713 err = bdi_init(bdi);
714 if (err)
715 return err;
716
717 sprintf(tmp, "%.28s%s", name, "-%d");
718 err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq));
719 if (err) {
720 bdi_destroy(bdi);
721 return err;
722 }
723
724 return 0;
725}
726EXPORT_SYMBOL(bdi_setup_and_register);
727
Andrew Morton3fcfab12006-10-19 23:28:16 -0700728static wait_queue_head_t congestion_wqh[2] = {
729 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
730 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
731 };
Mel Gorman0e093d992010-10-26 14:21:45 -0700732static atomic_t nr_bdi_congested[2];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700733
Jens Axboe1faa16d2009-04-06 14:48:01 +0200734void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700735{
736 enum bdi_state bit;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200737 wait_queue_head_t *wqh = &congestion_wqh[sync];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700738
Jens Axboe1faa16d2009-04-06 14:48:01 +0200739 bit = sync ? BDI_sync_congested : BDI_async_congested;
Mel Gorman0e093d992010-10-26 14:21:45 -0700740 if (test_and_clear_bit(bit, &bdi->state))
741 atomic_dec(&nr_bdi_congested[sync]);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700742 smp_mb__after_clear_bit();
743 if (waitqueue_active(wqh))
744 wake_up(wqh);
745}
746EXPORT_SYMBOL(clear_bdi_congested);
747
Jens Axboe1faa16d2009-04-06 14:48:01 +0200748void set_bdi_congested(struct backing_dev_info *bdi, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700749{
750 enum bdi_state bit;
751
Jens Axboe1faa16d2009-04-06 14:48:01 +0200752 bit = sync ? BDI_sync_congested : BDI_async_congested;
Mel Gorman0e093d992010-10-26 14:21:45 -0700753 if (!test_and_set_bit(bit, &bdi->state))
754 atomic_inc(&nr_bdi_congested[sync]);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700755}
756EXPORT_SYMBOL(set_bdi_congested);
757
758/**
759 * congestion_wait - wait for a backing_dev to become uncongested
Jens Axboe8aa7e842009-07-09 14:52:32 +0200760 * @sync: SYNC or ASYNC IO
Andrew Morton3fcfab12006-10-19 23:28:16 -0700761 * @timeout: timeout in jiffies
762 *
763 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
764 * write congestion. If no backing_devs are congested then just wait for the
765 * next write to be completed.
766 */
Jens Axboe8aa7e842009-07-09 14:52:32 +0200767long congestion_wait(int sync, long timeout)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700768{
769 long ret;
Mel Gorman52bb9192010-10-26 14:21:41 -0700770 unsigned long start = jiffies;
Andrew Morton3fcfab12006-10-19 23:28:16 -0700771 DEFINE_WAIT(wait);
Jens Axboe8aa7e842009-07-09 14:52:32 +0200772 wait_queue_head_t *wqh = &congestion_wqh[sync];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700773
774 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
775 ret = io_schedule_timeout(timeout);
776 finish_wait(wqh, &wait);
Mel Gorman52bb9192010-10-26 14:21:41 -0700777
778 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
779 jiffies_to_usecs(jiffies - start));
780
Andrew Morton3fcfab12006-10-19 23:28:16 -0700781 return ret;
782}
783EXPORT_SYMBOL(congestion_wait);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700784
Mel Gorman0e093d992010-10-26 14:21:45 -0700785/**
786 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
787 * @zone: A zone to check if it is heavily congested
788 * @sync: SYNC or ASYNC IO
789 * @timeout: timeout in jiffies
790 *
791 * In the event of a congested backing_dev (any backing_dev) and the given
792 * @zone has experienced recent congestion, this waits for up to @timeout
793 * jiffies for either a BDI to exit congestion of the given @sync queue
794 * or a write to complete.
795 *
796 * In the absense of zone congestion, cond_resched() is called to yield
797 * the processor if necessary but otherwise does not sleep.
798 *
799 * The return value is 0 if the sleep is for the full timeout. Otherwise,
800 * it is the number of jiffies that were still remaining when the function
801 * returned. return_value == timeout implies the function did not sleep.
802 */
803long wait_iff_congested(struct zone *zone, int sync, long timeout)
804{
805 long ret;
806 unsigned long start = jiffies;
807 DEFINE_WAIT(wait);
808 wait_queue_head_t *wqh = &congestion_wqh[sync];
809
810 /*
811 * If there is no congestion, or heavy congestion is not being
812 * encountered in the current zone, yield if necessary instead
813 * of sleeping on the congestion queue
814 */
815 if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
816 !zone_is_reclaim_congested(zone)) {
817 cond_resched();
818
819 /* In case we scheduled, work out time remaining */
820 ret = timeout - (jiffies - start);
821 if (ret < 0)
822 ret = 0;
823
824 goto out;
825 }
826
827 /* Sleep until uncongested or a write happens */
828 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
829 ret = io_schedule_timeout(timeout);
830 finish_wait(wqh, &wait);
831
832out:
833 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
834 jiffies_to_usecs(jiffies - start));
835
836 return ret;
837}
838EXPORT_SYMBOL(wait_iff_congested);