blob: eaa4a5bbe0634390fc802ebffdbad4291fa3b991 [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",
33};
34EXPORT_SYMBOL_GPL(noop_backing_dev_info);
35
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070036static struct class *bdi_class;
Jens Axboecfc4ba52009-09-14 13:12:40 +020037
38/*
39 * bdi_lock protects updates to bdi_list and bdi_pending_list, as well as
40 * reader side protection for bdi_pending_list. bdi_list has RCU reader side
41 * locking.
42 */
Jens Axboe03ba3782009-09-09 09:08:54 +020043DEFINE_SPINLOCK(bdi_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +020044LIST_HEAD(bdi_list);
Jens Axboe03ba3782009-09-09 09:08:54 +020045LIST_HEAD(bdi_pending_list);
46
47static struct task_struct *sync_supers_tsk;
48static struct timer_list sync_supers_timer;
49
50static int bdi_sync_supers(void *);
51static void sync_supers_timer_fn(unsigned long);
Jens Axboe03ba3782009-09-09 09:08:54 +020052
Miklos Szeredi76f14182008-04-30 00:54:36 -070053#ifdef CONFIG_DEBUG_FS
54#include <linux/debugfs.h>
55#include <linux/seq_file.h>
56
57static struct dentry *bdi_debug_root;
58
59static void bdi_debug_init(void)
60{
61 bdi_debug_root = debugfs_create_dir("bdi", NULL);
62}
63
64static int bdi_debug_stats_show(struct seq_file *m, void *v)
65{
66 struct backing_dev_info *bdi = m->private;
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020067 struct bdi_writeback *wb = &bdi->wb;
David Rientjes364aeb22009-01-06 14:39:29 -080068 unsigned long background_thresh;
69 unsigned long dirty_thresh;
70 unsigned long bdi_thresh;
Jens Axboef09b00d2009-05-25 09:08:21 +020071 unsigned long nr_dirty, nr_io, nr_more_io, nr_wb;
72 struct inode *inode;
73
Jens Axboef09b00d2009-05-25 09:08:21 +020074 nr_wb = nr_dirty = nr_io = nr_more_io = 0;
75 spin_lock(&inode_lock);
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020076 list_for_each_entry(inode, &wb->b_dirty, i_list)
77 nr_dirty++;
78 list_for_each_entry(inode, &wb->b_io, i_list)
79 nr_io++;
80 list_for_each_entry(inode, &wb->b_more_io, i_list)
81 nr_more_io++;
Jens Axboef09b00d2009-05-25 09:08:21 +020082 spin_unlock(&inode_lock);
Miklos Szeredi76f14182008-04-30 00:54:36 -070083
Wu Fengguang16c40422010-08-11 14:17:39 -070084 global_dirty_limits(&background_thresh, &dirty_thresh);
85 bdi_thresh = bdi_dirty_limit(bdi, dirty_thresh);
Miklos Szeredi76f14182008-04-30 00:54:36 -070086
87#define K(x) ((x) << (PAGE_SHIFT - 10))
88 seq_printf(m,
89 "BdiWriteback: %8lu kB\n"
90 "BdiReclaimable: %8lu kB\n"
91 "BdiDirtyThresh: %8lu kB\n"
92 "DirtyThresh: %8lu kB\n"
Jens Axboef09b00d2009-05-25 09:08:21 +020093 "BackgroundThresh: %8lu kB\n"
Jens Axboef09b00d2009-05-25 09:08:21 +020094 "b_dirty: %8lu\n"
95 "b_io: %8lu\n"
96 "b_more_io: %8lu\n"
97 "bdi_list: %8u\n"
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020098 "state: %8lx\n",
Miklos Szeredi76f14182008-04-30 00:54:36 -070099 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
100 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
Jens Axboef09b00d2009-05-25 09:08:21 +0200101 K(bdi_thresh), K(dirty_thresh),
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200102 K(background_thresh), nr_dirty, nr_io, nr_more_io,
103 !list_empty(&bdi->bdi_list), bdi->state);
Miklos Szeredi76f14182008-04-30 00:54:36 -0700104#undef K
105
106 return 0;
107}
108
109static int bdi_debug_stats_open(struct inode *inode, struct file *file)
110{
111 return single_open(file, bdi_debug_stats_show, inode->i_private);
112}
113
114static const struct file_operations bdi_debug_stats_fops = {
115 .open = bdi_debug_stats_open,
116 .read = seq_read,
117 .llseek = seq_lseek,
118 .release = single_release,
119};
120
121static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
122{
123 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
124 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
125 bdi, &bdi_debug_stats_fops);
126}
127
128static void bdi_debug_unregister(struct backing_dev_info *bdi)
129{
130 debugfs_remove(bdi->debug_stats);
131 debugfs_remove(bdi->debug_dir);
132}
133#else
134static inline void bdi_debug_init(void)
135{
136}
137static inline void bdi_debug_register(struct backing_dev_info *bdi,
138 const char *name)
139{
140}
141static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
142{
143}
144#endif
145
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700146static ssize_t read_ahead_kb_store(struct device *dev,
147 struct device_attribute *attr,
148 const char *buf, size_t count)
149{
150 struct backing_dev_info *bdi = dev_get_drvdata(dev);
151 char *end;
152 unsigned long read_ahead_kb;
153 ssize_t ret = -EINVAL;
154
155 read_ahead_kb = simple_strtoul(buf, &end, 10);
156 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
157 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
158 ret = count;
159 }
160 return ret;
161}
162
163#define K(pages) ((pages) << (PAGE_SHIFT - 10))
164
165#define BDI_SHOW(name, expr) \
166static ssize_t name##_show(struct device *dev, \
167 struct device_attribute *attr, char *page) \
168{ \
169 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
170 \
171 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
172}
173
174BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
175
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700176static ssize_t min_ratio_store(struct device *dev,
177 struct device_attribute *attr, const char *buf, size_t count)
178{
179 struct backing_dev_info *bdi = dev_get_drvdata(dev);
180 char *end;
181 unsigned int ratio;
182 ssize_t ret = -EINVAL;
183
184 ratio = simple_strtoul(buf, &end, 10);
185 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
186 ret = bdi_set_min_ratio(bdi, ratio);
187 if (!ret)
188 ret = count;
189 }
190 return ret;
191}
192BDI_SHOW(min_ratio, bdi->min_ratio)
193
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700194static ssize_t max_ratio_store(struct device *dev,
195 struct device_attribute *attr, const char *buf, size_t count)
196{
197 struct backing_dev_info *bdi = dev_get_drvdata(dev);
198 char *end;
199 unsigned int ratio;
200 ssize_t ret = -EINVAL;
201
202 ratio = simple_strtoul(buf, &end, 10);
203 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
204 ret = bdi_set_max_ratio(bdi, ratio);
205 if (!ret)
206 ret = count;
207 }
208 return ret;
209}
210BDI_SHOW(max_ratio, bdi->max_ratio)
211
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700212#define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
213
214static struct device_attribute bdi_dev_attrs[] = {
215 __ATTR_RW(read_ahead_kb),
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700216 __ATTR_RW(min_ratio),
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700217 __ATTR_RW(max_ratio),
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700218 __ATTR_NULL,
219};
220
221static __init int bdi_class_init(void)
222{
223 bdi_class = class_create(THIS_MODULE, "bdi");
Anton Blanchard14421452010-04-02 09:46:55 +0200224 if (IS_ERR(bdi_class))
225 return PTR_ERR(bdi_class);
226
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700227 bdi_class->dev_attrs = bdi_dev_attrs;
Miklos Szeredi76f14182008-04-30 00:54:36 -0700228 bdi_debug_init();
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700229 return 0;
230}
Miklos Szeredi76f14182008-04-30 00:54:36 -0700231postcore_initcall(bdi_class_init);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700232
Jens Axboe26160152009-03-17 09:35:06 +0100233static int __init default_bdi_init(void)
234{
235 int err;
236
Jens Axboe03ba3782009-09-09 09:08:54 +0200237 sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
238 BUG_ON(IS_ERR(sync_supers_tsk));
239
Jens Axboe03ba3782009-09-09 09:08:54 +0200240 setup_timer(&sync_supers_timer, sync_supers_timer_fn, 0);
Jens Axboe64231042010-05-21 20:00:35 +0200241 bdi_arm_supers_timer();
Jens Axboe03ba3782009-09-09 09:08:54 +0200242
Jens Axboe26160152009-03-17 09:35:06 +0100243 err = bdi_init(&default_backing_dev_info);
244 if (!err)
245 bdi_register(&default_backing_dev_info, NULL, "default");
246
247 return err;
248}
249subsys_initcall(default_bdi_init);
250
Jens Axboe03ba3782009-09-09 09:08:54 +0200251int bdi_has_dirty_io(struct backing_dev_info *bdi)
252{
253 return wb_has_dirty_io(&bdi->wb);
254}
255
256static void bdi_flush_io(struct backing_dev_info *bdi)
257{
258 struct writeback_control wbc = {
Jens Axboe03ba3782009-09-09 09:08:54 +0200259 .sync_mode = WB_SYNC_NONE,
260 .older_than_this = NULL,
261 .range_cyclic = 1,
262 .nr_to_write = 1024,
263 };
264
Christoph Hellwig9c3a8ee2010-06-10 12:07:27 +0200265 writeback_inodes_wb(&bdi->wb, &wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200266}
267
268/*
Artem Bityutskiy6f904ff2010-07-25 14:29:11 +0300269 * kupdated() used to do this. We cannot do it from the bdi_forker_thread()
Jens Axboe03ba3782009-09-09 09:08:54 +0200270 * or we risk deadlocking on ->s_umount. The longer term solution would be
271 * to implement sync_supers_bdi() or similar and simply do it from the
Artem Bityutskiy6f904ff2010-07-25 14:29:11 +0300272 * bdi writeback thread individually.
Jens Axboe03ba3782009-09-09 09:08:54 +0200273 */
274static int bdi_sync_supers(void *unused)
275{
276 set_user_nice(current, 0);
277
278 while (!kthread_should_stop()) {
279 set_current_state(TASK_INTERRUPTIBLE);
280 schedule();
281
282 /*
283 * Do this periodically, like kupdated() did before.
284 */
285 sync_supers();
286 }
287
288 return 0;
289}
290
Jens Axboe64231042010-05-21 20:00:35 +0200291void bdi_arm_supers_timer(void)
Jens Axboe03ba3782009-09-09 09:08:54 +0200292{
293 unsigned long next;
294
Jens Axboe64231042010-05-21 20:00:35 +0200295 if (!dirty_writeback_interval)
296 return;
297
Jens Axboe03ba3782009-09-09 09:08:54 +0200298 next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
299 mod_timer(&sync_supers_timer, round_jiffies_up(next));
300}
301
302static void sync_supers_timer_fn(unsigned long unused)
303{
304 wake_up_process(sync_supers_tsk);
Jens Axboe64231042010-05-21 20:00:35 +0200305 bdi_arm_supers_timer();
Jens Axboe03ba3782009-09-09 09:08:54 +0200306}
307
Artem Bityutskiy64677162010-07-25 14:29:22 +0300308static void wakeup_timer_fn(unsigned long data)
309{
310 struct backing_dev_info *bdi = (struct backing_dev_info *)data;
311
312 spin_lock_bh(&bdi->wb_lock);
313 if (bdi->wb.task) {
Artem Bityutskiy60332022010-07-25 14:29:24 +0300314 trace_writeback_wake_thread(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300315 wake_up_process(bdi->wb.task);
316 } else {
317 /*
318 * When bdi tasks are inactive for long time, they are killed.
319 * In this case we have to wake-up the forker thread which
320 * should create and run the bdi thread.
321 */
Artem Bityutskiy60332022010-07-25 14:29:24 +0300322 trace_writeback_wake_forker_thread(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300323 wake_up_process(default_backing_dev_info.wb.task);
324 }
325 spin_unlock_bh(&bdi->wb_lock);
326}
327
328/*
329 * This function is used when the first inode for this bdi is marked dirty. It
330 * wakes-up the corresponding bdi thread which should then take care of the
331 * periodic background write-out of dirty inodes. Since the write-out would
332 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
333 * set up a timer which wakes the bdi thread up later.
334 *
335 * Note, we wouldn't bother setting up the timer, but this function is on the
336 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
337 * by delaying the wake-up.
338 */
339void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
340{
341 unsigned long timeout;
342
343 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
344 mod_timer(&bdi->wb.wakeup_timer, jiffies + timeout);
345}
346
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300347/*
348 * Calculate the longest interval (jiffies) bdi threads are allowed to be
349 * inactive.
350 */
351static unsigned long bdi_longest_inactive(void)
352{
353 unsigned long interval;
354
355 interval = msecs_to_jiffies(dirty_writeback_interval * 10);
356 return max(5UL * 60 * HZ, interval);
357}
358
Artem Bityutskiy6f904ff2010-07-25 14:29:11 +0300359static int bdi_forker_thread(void *ptr)
Jens Axboe03ba3782009-09-09 09:08:54 +0200360{
361 struct bdi_writeback *me = ptr;
362
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200363 current->flags |= PF_FLUSHER | PF_SWAPWRITE;
364 set_freezable();
365
366 /*
367 * Our parent may run at a different priority, just set us to normal
368 */
369 set_user_nice(current, 0);
Jens Axboe03ba3782009-09-09 09:08:54 +0200370
371 for (;;) {
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300372 struct task_struct *task = NULL;
Artem Bityutskiy78c40cb2010-07-25 14:29:17 +0300373 struct backing_dev_info *bdi;
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300374 enum {
375 NO_ACTION, /* Nothing to do */
376 FORK_THREAD, /* Fork bdi thread */
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300377 KILL_THREAD, /* Kill inactive bdi thread */
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300378 } action = NO_ACTION;
Jens Axboe03ba3782009-09-09 09:08:54 +0200379
380 /*
381 * Temporary measure, we want to make sure we don't see
382 * dirty data on the default backing_dev_info
383 */
Artem Bityutskiy64677162010-07-25 14:29:22 +0300384 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list)) {
385 del_timer(&me->wakeup_timer);
Jens Axboe03ba3782009-09-09 09:08:54 +0200386 wb_do_writeback(me, 0);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300387 }
Jens Axboe03ba3782009-09-09 09:08:54 +0200388
Jens Axboecfc4ba52009-09-14 13:12:40 +0200389 spin_lock_bh(&bdi_lock);
Artem Bityutskiyc5f7ad22010-07-25 14:29:13 +0300390 set_current_state(TASK_INTERRUPTIBLE);
Jens Axboe03ba3782009-09-09 09:08:54 +0200391
Artem Bityutskiy78c40cb2010-07-25 14:29:17 +0300392 list_for_each_entry(bdi, &bdi_list, bdi_list) {
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300393 bool have_dirty_io;
394
395 if (!bdi_cap_writeback_dirty(bdi) ||
396 bdi_cap_flush_forker(bdi))
Jens Axboe03ba3782009-09-09 09:08:54 +0200397 continue;
398
Artem Bityutskiy080dcec2010-07-25 14:29:16 +0300399 WARN(!test_bit(BDI_registered, &bdi->state),
400 "bdi %p/%s is not registered!\n", bdi, bdi->name);
401
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300402 have_dirty_io = !list_empty(&bdi->work_list) ||
403 wb_has_dirty_io(&bdi->wb);
Artem Bityutskiy78c40cb2010-07-25 14:29:17 +0300404
405 /*
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300406 * If the bdi has work to do, but the thread does not
407 * exist - create it.
Artem Bityutskiy78c40cb2010-07-25 14:29:17 +0300408 */
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300409 if (!bdi->wb.task && have_dirty_io) {
410 /*
411 * Set the pending bit - if someone will try to
412 * unregister this bdi - it'll wait on this bit.
413 */
414 set_bit(BDI_pending, &bdi->state);
415 action = FORK_THREAD;
416 break;
417 }
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300418
Jens Axboe6bf05d02010-08-04 13:34:31 +0200419 spin_lock(&bdi->wb_lock);
420
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300421 /*
422 * If there is no work to do and the bdi thread was
423 * inactive long enough - kill it. The wb_lock is taken
424 * to make sure no-one adds more work to this bdi and
425 * wakes the bdi thread up.
426 */
427 if (bdi->wb.task && !have_dirty_io &&
428 time_after(jiffies, bdi->wb.last_active +
429 bdi_longest_inactive())) {
430 task = bdi->wb.task;
431 bdi->wb.task = NULL;
432 spin_unlock(&bdi->wb_lock);
433 set_bit(BDI_pending, &bdi->state);
434 action = KILL_THREAD;
435 break;
436 }
Jens Axboe6bf05d02010-08-04 13:34:31 +0200437 spin_unlock(&bdi->wb_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200438 }
Artem Bityutskiy080dcec2010-07-25 14:29:16 +0300439 spin_unlock_bh(&bdi_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200440
Artem Bityutskiyc4ec7902010-07-25 14:29:14 +0300441 /* Keep working if default bdi still has things to do */
442 if (!list_empty(&me->bdi->work_list))
443 __set_current_state(TASK_RUNNING);
444
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300445 switch (action) {
446 case FORK_THREAD:
447 __set_current_state(TASK_RUNNING);
448 task = kthread_run(bdi_writeback_thread, &bdi->wb, "flush-%s",
449 dev_name(bdi->dev));
450 if (IS_ERR(task)) {
451 /*
452 * If thread creation fails, force writeout of
453 * the bdi from the thread.
454 */
455 bdi_flush_io(bdi);
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300456 } else {
457 /*
458 * The spinlock makes sure we do not lose
459 * wake-ups when racing with 'bdi_queue_work()'.
460 */
Artem Bityutskiy64677162010-07-25 14:29:22 +0300461 spin_lock_bh(&bdi->wb_lock);
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300462 bdi->wb.task = task;
Artem Bityutskiy64677162010-07-25 14:29:22 +0300463 spin_unlock_bh(&bdi->wb_lock);
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300464 }
465 break;
466
467 case KILL_THREAD:
468 __set_current_state(TASK_RUNNING);
469 kthread_stop(task);
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300470 break;
Jens Axboe03ba3782009-09-09 09:08:54 +0200471
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300472 case NO_ACTION:
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300473 if (!wb_has_dirty_io(me) || !dirty_writeback_interval)
474 /*
475 * There are no dirty data. The only thing we
476 * should now care about is checking for
477 * inactive bdi threads and killing them. Thus,
478 * let's sleep for longer time, save energy and
479 * be friendly for battery-driven devices.
480 */
481 schedule_timeout(bdi_longest_inactive());
Jens Axboe64231042010-05-21 20:00:35 +0200482 else
Artem Bityutskiy253c34e2010-07-25 14:29:21 +0300483 schedule_timeout(msecs_to_jiffies(dirty_writeback_interval * 10));
Jens Axboe03ba3782009-09-09 09:08:54 +0200484 try_to_freeze();
Artem Bityutskiyadf39242010-07-25 14:29:19 +0300485 /* Back to the main loop */
Jens Axboe03ba3782009-09-09 09:08:54 +0200486 continue;
487 }
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300488
489 /*
490 * Clear pending bit and wakeup anybody waiting to tear us down.
491 */
492 clear_bit(BDI_pending, &bdi->state);
493 smp_mb__after_clear_bit();
494 wake_up_bit(&bdi->state, BDI_pending);
Jens Axboe03ba3782009-09-09 09:08:54 +0200495 }
496
497 return 0;
498}
499
Jens Axboecfc4ba52009-09-14 13:12:40 +0200500/*
501 * Remove bdi from bdi_list, and ensure that it is no longer visible
502 */
503static void bdi_remove_from_list(struct backing_dev_info *bdi)
504{
505 spin_lock_bh(&bdi_lock);
506 list_del_rcu(&bdi->bdi_list);
507 spin_unlock_bh(&bdi_lock);
508
509 synchronize_rcu();
510}
511
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700512int bdi_register(struct backing_dev_info *bdi, struct device *parent,
513 const char *fmt, ...)
514{
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700515 va_list args;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700516 struct device *dev;
517
Andrew Morton69fc2082008-12-09 13:14:06 -0800518 if (bdi->dev) /* The driver needs to use separate queues per device */
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300519 return 0;
Kay Sieversf1d0b062008-12-02 10:31:50 -0800520
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700521 va_start(args, fmt);
Greg Kroah-Hartman19051c52008-05-15 13:44:08 -0700522 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700523 va_end(args);
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300524 if (IS_ERR(dev))
525 return PTR_ERR(dev);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200526
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700527 bdi->dev = dev;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700528
Jens Axboe03ba3782009-09-09 09:08:54 +0200529 /*
530 * Just start the forker thread for our default backing_dev_info,
531 * and add other bdi's to the list. They will get a thread created
532 * on-demand when they need it.
533 */
534 if (bdi_cap_flush_forker(bdi)) {
535 struct bdi_writeback *wb = &bdi->wb;
536
Artem Bityutskiy6f904ff2010-07-25 14:29:11 +0300537 wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s",
Jens Axboe03ba3782009-09-09 09:08:54 +0200538 dev_name(dev));
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300539 if (IS_ERR(wb->task))
540 return PTR_ERR(wb->task);
Jens Axboe03ba3782009-09-09 09:08:54 +0200541 }
542
543 bdi_debug_register(bdi, dev_name(dev));
Jens Axboe500b0672009-09-09 09:10:25 +0200544 set_bit(BDI_registered, &bdi->state);
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300545
546 spin_lock_bh(&bdi_lock);
547 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
548 spin_unlock_bh(&bdi_lock);
549
Dave Chinner455b2862010-07-07 13:24:06 +1000550 trace_writeback_bdi_register(bdi);
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300551 return 0;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700552}
553EXPORT_SYMBOL(bdi_register);
554
555int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
556{
557 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
558}
559EXPORT_SYMBOL(bdi_register_dev);
560
Jens Axboe03ba3782009-09-09 09:08:54 +0200561/*
562 * Remove bdi from the global list and shutdown any threads we have running
563 */
564static void bdi_wb_shutdown(struct backing_dev_info *bdi)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200565{
Jens Axboe03ba3782009-09-09 09:08:54 +0200566 if (!bdi_cap_writeback_dirty(bdi))
567 return;
568
569 /*
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300570 * Make sure nobody finds us on the bdi_list anymore
571 */
572 bdi_remove_from_list(bdi);
573
574 /*
Jens Axboe03ba3782009-09-09 09:08:54 +0200575 * If setup is pending, wait for that to complete first
576 */
577 wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
578 TASK_UNINTERRUPTIBLE);
579
580 /*
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200581 * Finally, kill the kernel thread. We don't need to be RCU
Romit Dasguptac62b17a2009-11-12 13:08:11 +0100582 * safe anymore, since the bdi is gone from visibility. Force
583 * unfreeze of the thread before calling kthread_stop(), otherwise
584 * it would never exet if it is currently stuck in the refrigerator.
Jens Axboe03ba3782009-09-09 09:08:54 +0200585 */
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200586 if (bdi->wb.task) {
587 thaw_process(bdi->wb.task);
588 kthread_stop(bdi->wb.task);
Romit Dasguptac62b17a2009-11-12 13:08:11 +0100589 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200590}
591
Jens Axboe592b09a2009-10-29 11:46:12 +0100592/*
593 * This bdi is going away now, make sure that no super_blocks point to it
594 */
595static void bdi_prune_sb(struct backing_dev_info *bdi)
596{
597 struct super_block *sb;
598
599 spin_lock(&sb_lock);
600 list_for_each_entry(sb, &super_blocks, s_list) {
601 if (sb->s_bdi == bdi)
602 sb->s_bdi = NULL;
603 }
604 spin_unlock(&sb_lock);
605}
606
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700607void bdi_unregister(struct backing_dev_info *bdi)
608{
609 if (bdi->dev) {
Dave Chinner455b2862010-07-07 13:24:06 +1000610 trace_writeback_bdi_unregister(bdi);
Jens Axboe8c4db332009-11-03 20:18:44 +0100611 bdi_prune_sb(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300612 del_timer_sync(&bdi->wb.wakeup_timer);
Jens Axboe8c4db332009-11-03 20:18:44 +0100613
Jens Axboe03ba3782009-09-09 09:08:54 +0200614 if (!bdi_cap_flush_forker(bdi))
615 bdi_wb_shutdown(bdi);
Miklos Szeredi76f14182008-04-30 00:54:36 -0700616 bdi_debug_unregister(bdi);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700617 device_unregister(bdi->dev);
618 bdi->dev = NULL;
619 }
620}
621EXPORT_SYMBOL(bdi_unregister);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700622
Artem Bityutskiy64677162010-07-25 14:29:22 +0300623static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
624{
625 memset(wb, 0, sizeof(*wb));
626
627 wb->bdi = bdi;
628 wb->last_old_flush = jiffies;
629 INIT_LIST_HEAD(&wb->b_dirty);
630 INIT_LIST_HEAD(&wb->b_io);
631 INIT_LIST_HEAD(&wb->b_more_io);
632 setup_timer(&wb->wakeup_timer, wakeup_timer_fn, (unsigned long)bdi);
633}
634
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700635int bdi_init(struct backing_dev_info *bdi)
636{
Jens Axboe03ba3782009-09-09 09:08:54 +0200637 int i, err;
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700638
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700639 bdi->dev = NULL;
640
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700641 bdi->min_ratio = 0;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700642 bdi->max_ratio = 100;
643 bdi->max_prop_frac = PROP_FRAC_BASE;
Jens Axboe03ba3782009-09-09 09:08:54 +0200644 spin_lock_init(&bdi->wb_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200645 INIT_LIST_HEAD(&bdi->bdi_list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200646 INIT_LIST_HEAD(&bdi->work_list);
647
648 bdi_wb_init(&bdi->wb, bdi);
649
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700650 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
Peter Zijlstraea319512008-12-26 15:08:55 +0100651 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700652 if (err)
653 goto err;
654 }
655
656 bdi->dirty_exceeded = 0;
657 err = prop_local_init_percpu(&bdi->completions);
658
659 if (err) {
660err:
Denis Cheng4b01a0b2007-12-04 23:45:07 -0800661 while (i--)
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700662 percpu_counter_destroy(&bdi->bdi_stat[i]);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700663 }
664
665 return err;
666}
667EXPORT_SYMBOL(bdi_init);
668
669void bdi_destroy(struct backing_dev_info *bdi)
670{
671 int i;
672
Jens Axboece5f8e72009-09-14 12:57:56 +0200673 /*
674 * Splice our entries to the default_backing_dev_info, if this
675 * bdi disappears
676 */
677 if (bdi_has_dirty_io(bdi)) {
678 struct bdi_writeback *dst = &default_backing_dev_info.wb;
679
680 spin_lock(&inode_lock);
681 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
682 list_splice(&bdi->wb.b_io, &dst->b_io);
683 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
684 spin_unlock(&inode_lock);
685 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200686
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700687 bdi_unregister(bdi);
688
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700689 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
690 percpu_counter_destroy(&bdi->bdi_stat[i]);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700691
692 prop_local_destroy_percpu(&bdi->completions);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700693}
694EXPORT_SYMBOL(bdi_destroy);
695
Jens Axboec3c53202010-04-22 11:37:01 +0200696/*
697 * For use from filesystems to quickly init and register a bdi associated
698 * with dirty writeback
699 */
700int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
701 unsigned int cap)
702{
703 char tmp[32];
704 int err;
705
706 bdi->name = name;
707 bdi->capabilities = cap;
708 err = bdi_init(bdi);
709 if (err)
710 return err;
711
712 sprintf(tmp, "%.28s%s", name, "-%d");
713 err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq));
714 if (err) {
715 bdi_destroy(bdi);
716 return err;
717 }
718
719 return 0;
720}
721EXPORT_SYMBOL(bdi_setup_and_register);
722
Andrew Morton3fcfab12006-10-19 23:28:16 -0700723static wait_queue_head_t congestion_wqh[2] = {
724 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
725 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
726 };
727
Jens Axboe1faa16d2009-04-06 14:48:01 +0200728void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700729{
730 enum bdi_state bit;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200731 wait_queue_head_t *wqh = &congestion_wqh[sync];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700732
Jens Axboe1faa16d2009-04-06 14:48:01 +0200733 bit = sync ? BDI_sync_congested : BDI_async_congested;
Andrew Morton3fcfab12006-10-19 23:28:16 -0700734 clear_bit(bit, &bdi->state);
735 smp_mb__after_clear_bit();
736 if (waitqueue_active(wqh))
737 wake_up(wqh);
738}
739EXPORT_SYMBOL(clear_bdi_congested);
740
Jens Axboe1faa16d2009-04-06 14:48:01 +0200741void set_bdi_congested(struct backing_dev_info *bdi, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700742{
743 enum bdi_state bit;
744
Jens Axboe1faa16d2009-04-06 14:48:01 +0200745 bit = sync ? BDI_sync_congested : BDI_async_congested;
Andrew Morton3fcfab12006-10-19 23:28:16 -0700746 set_bit(bit, &bdi->state);
747}
748EXPORT_SYMBOL(set_bdi_congested);
749
750/**
751 * congestion_wait - wait for a backing_dev to become uncongested
Jens Axboe8aa7e842009-07-09 14:52:32 +0200752 * @sync: SYNC or ASYNC IO
Andrew Morton3fcfab12006-10-19 23:28:16 -0700753 * @timeout: timeout in jiffies
754 *
755 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
756 * write congestion. If no backing_devs are congested then just wait for the
757 * next write to be completed.
758 */
Jens Axboe8aa7e842009-07-09 14:52:32 +0200759long congestion_wait(int sync, long timeout)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700760{
761 long ret;
762 DEFINE_WAIT(wait);
Jens Axboe8aa7e842009-07-09 14:52:32 +0200763 wait_queue_head_t *wqh = &congestion_wqh[sync];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700764
765 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
766 ret = io_schedule_timeout(timeout);
767 finish_wait(wqh, &wait);
768 return ret;
769}
770EXPORT_SYMBOL(congestion_wait);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700771