blob: dbc66815a0fe00a591695a211adb8d280ad79979 [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
84 get_dirty_limits(&background_thresh, &dirty_thresh, &bdi_thresh, bdi);
85
86#define K(x) ((x) << (PAGE_SHIFT - 10))
87 seq_printf(m,
88 "BdiWriteback: %8lu kB\n"
89 "BdiReclaimable: %8lu kB\n"
90 "BdiDirtyThresh: %8lu kB\n"
91 "DirtyThresh: %8lu kB\n"
Jens Axboef09b00d2009-05-25 09:08:21 +020092 "BackgroundThresh: %8lu kB\n"
Jens Axboef09b00d2009-05-25 09:08:21 +020093 "b_dirty: %8lu\n"
94 "b_io: %8lu\n"
95 "b_more_io: %8lu\n"
96 "bdi_list: %8u\n"
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020097 "state: %8lx\n",
Miklos Szeredi76f14182008-04-30 00:54:36 -070098 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
99 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
Jens Axboef09b00d2009-05-25 09:08:21 +0200100 K(bdi_thresh), K(dirty_thresh),
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200101 K(background_thresh), nr_dirty, nr_io, nr_more_io,
102 !list_empty(&bdi->bdi_list), bdi->state);
Miklos Szeredi76f14182008-04-30 00:54:36 -0700103#undef K
104
105 return 0;
106}
107
108static int bdi_debug_stats_open(struct inode *inode, struct file *file)
109{
110 return single_open(file, bdi_debug_stats_show, inode->i_private);
111}
112
113static const struct file_operations bdi_debug_stats_fops = {
114 .open = bdi_debug_stats_open,
115 .read = seq_read,
116 .llseek = seq_lseek,
117 .release = single_release,
118};
119
120static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
121{
122 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
123 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
124 bdi, &bdi_debug_stats_fops);
125}
126
127static void bdi_debug_unregister(struct backing_dev_info *bdi)
128{
129 debugfs_remove(bdi->debug_stats);
130 debugfs_remove(bdi->debug_dir);
131}
132#else
133static inline void bdi_debug_init(void)
134{
135}
136static inline void bdi_debug_register(struct backing_dev_info *bdi,
137 const char *name)
138{
139}
140static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
141{
142}
143#endif
144
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700145static ssize_t read_ahead_kb_store(struct device *dev,
146 struct device_attribute *attr,
147 const char *buf, size_t count)
148{
149 struct backing_dev_info *bdi = dev_get_drvdata(dev);
150 char *end;
151 unsigned long read_ahead_kb;
152 ssize_t ret = -EINVAL;
153
154 read_ahead_kb = simple_strtoul(buf, &end, 10);
155 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
156 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
157 ret = count;
158 }
159 return ret;
160}
161
162#define K(pages) ((pages) << (PAGE_SHIFT - 10))
163
164#define BDI_SHOW(name, expr) \
165static ssize_t name##_show(struct device *dev, \
166 struct device_attribute *attr, char *page) \
167{ \
168 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
169 \
170 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
171}
172
173BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
174
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700175static ssize_t min_ratio_store(struct device *dev,
176 struct device_attribute *attr, const char *buf, size_t count)
177{
178 struct backing_dev_info *bdi = dev_get_drvdata(dev);
179 char *end;
180 unsigned int ratio;
181 ssize_t ret = -EINVAL;
182
183 ratio = simple_strtoul(buf, &end, 10);
184 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
185 ret = bdi_set_min_ratio(bdi, ratio);
186 if (!ret)
187 ret = count;
188 }
189 return ret;
190}
191BDI_SHOW(min_ratio, bdi->min_ratio)
192
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700193static ssize_t max_ratio_store(struct device *dev,
194 struct device_attribute *attr, const char *buf, size_t count)
195{
196 struct backing_dev_info *bdi = dev_get_drvdata(dev);
197 char *end;
198 unsigned int ratio;
199 ssize_t ret = -EINVAL;
200
201 ratio = simple_strtoul(buf, &end, 10);
202 if (*buf && (end[0] == '\0' || (end[0] == '\n' && end[1] == '\0'))) {
203 ret = bdi_set_max_ratio(bdi, ratio);
204 if (!ret)
205 ret = count;
206 }
207 return ret;
208}
209BDI_SHOW(max_ratio, bdi->max_ratio)
210
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700211#define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
212
213static struct device_attribute bdi_dev_attrs[] = {
214 __ATTR_RW(read_ahead_kb),
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700215 __ATTR_RW(min_ratio),
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700216 __ATTR_RW(max_ratio),
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700217 __ATTR_NULL,
218};
219
220static __init int bdi_class_init(void)
221{
222 bdi_class = class_create(THIS_MODULE, "bdi");
Anton Blanchard14421452010-04-02 09:46:55 +0200223 if (IS_ERR(bdi_class))
224 return PTR_ERR(bdi_class);
225
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700226 bdi_class->dev_attrs = bdi_dev_attrs;
Miklos Szeredi76f14182008-04-30 00:54:36 -0700227 bdi_debug_init();
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700228 return 0;
229}
Miklos Szeredi76f14182008-04-30 00:54:36 -0700230postcore_initcall(bdi_class_init);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700231
Jens Axboe26160152009-03-17 09:35:06 +0100232static int __init default_bdi_init(void)
233{
234 int err;
235
Jens Axboe03ba3782009-09-09 09:08:54 +0200236 sync_supers_tsk = kthread_run(bdi_sync_supers, NULL, "sync_supers");
237 BUG_ON(IS_ERR(sync_supers_tsk));
238
239 init_timer(&sync_supers_timer);
240 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 +0200251static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
252{
253 memset(wb, 0, sizeof(*wb));
254
255 wb->bdi = bdi;
256 wb->last_old_flush = jiffies;
257 INIT_LIST_HEAD(&wb->b_dirty);
258 INIT_LIST_HEAD(&wb->b_io);
259 INIT_LIST_HEAD(&wb->b_more_io);
260}
261
Jens Axboe03ba3782009-09-09 09:08:54 +0200262int bdi_has_dirty_io(struct backing_dev_info *bdi)
263{
264 return wb_has_dirty_io(&bdi->wb);
265}
266
267static void bdi_flush_io(struct backing_dev_info *bdi)
268{
269 struct writeback_control wbc = {
Jens Axboe03ba3782009-09-09 09:08:54 +0200270 .sync_mode = WB_SYNC_NONE,
271 .older_than_this = NULL,
272 .range_cyclic = 1,
273 .nr_to_write = 1024,
274 };
275
Christoph Hellwig9c3a8ee2010-06-10 12:07:27 +0200276 writeback_inodes_wb(&bdi->wb, &wbc);
Jens Axboe03ba3782009-09-09 09:08:54 +0200277}
278
279/*
Artem Bityutskiy6f904ff2010-07-25 14:29:11 +0300280 * kupdated() used to do this. We cannot do it from the bdi_forker_thread()
Jens Axboe03ba3782009-09-09 09:08:54 +0200281 * or we risk deadlocking on ->s_umount. The longer term solution would be
282 * to implement sync_supers_bdi() or similar and simply do it from the
Artem Bityutskiy6f904ff2010-07-25 14:29:11 +0300283 * bdi writeback thread individually.
Jens Axboe03ba3782009-09-09 09:08:54 +0200284 */
285static int bdi_sync_supers(void *unused)
286{
287 set_user_nice(current, 0);
288
289 while (!kthread_should_stop()) {
290 set_current_state(TASK_INTERRUPTIBLE);
291 schedule();
292
293 /*
294 * Do this periodically, like kupdated() did before.
295 */
296 sync_supers();
297 }
298
299 return 0;
300}
301
Jens Axboe64231042010-05-21 20:00:35 +0200302void bdi_arm_supers_timer(void)
Jens Axboe03ba3782009-09-09 09:08:54 +0200303{
304 unsigned long next;
305
Jens Axboe64231042010-05-21 20:00:35 +0200306 if (!dirty_writeback_interval)
307 return;
308
Jens Axboe03ba3782009-09-09 09:08:54 +0200309 next = msecs_to_jiffies(dirty_writeback_interval * 10) + jiffies;
310 mod_timer(&sync_supers_timer, round_jiffies_up(next));
311}
312
313static void sync_supers_timer_fn(unsigned long unused)
314{
315 wake_up_process(sync_supers_tsk);
Jens Axboe64231042010-05-21 20:00:35 +0200316 bdi_arm_supers_timer();
Jens Axboe03ba3782009-09-09 09:08:54 +0200317}
318
Artem Bityutskiy6f904ff2010-07-25 14:29:11 +0300319static int bdi_forker_thread(void *ptr)
Jens Axboe03ba3782009-09-09 09:08:54 +0200320{
321 struct bdi_writeback *me = ptr;
322
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200323 current->flags |= PF_FLUSHER | PF_SWAPWRITE;
324 set_freezable();
325
326 /*
327 * Our parent may run at a different priority, just set us to normal
328 */
329 set_user_nice(current, 0);
Jens Axboe03ba3782009-09-09 09:08:54 +0200330
331 for (;;) {
Artem Bityutskiy080dcec2010-07-25 14:29:16 +0300332 bool fork = false;
Artem Bityutskiy94eac5e2010-07-25 14:29:12 +0300333 struct task_struct *task;
Jens Axboe03ba3782009-09-09 09:08:54 +0200334 struct backing_dev_info *bdi, *tmp;
Jens Axboe03ba3782009-09-09 09:08:54 +0200335
336 /*
337 * Temporary measure, we want to make sure we don't see
338 * dirty data on the default backing_dev_info
339 */
340 if (wb_has_dirty_io(me) || !list_empty(&me->bdi->work_list))
341 wb_do_writeback(me, 0);
342
Jens Axboecfc4ba52009-09-14 13:12:40 +0200343 spin_lock_bh(&bdi_lock);
Artem Bityutskiyc5f7ad22010-07-25 14:29:13 +0300344 set_current_state(TASK_INTERRUPTIBLE);
Jens Axboe03ba3782009-09-09 09:08:54 +0200345
346 /*
347 * Check if any existing bdi's have dirty data without
348 * a thread registered. If so, set that up.
349 */
350 list_for_each_entry_safe(bdi, tmp, &bdi_list, bdi_list) {
Artem Bityutskiy080dcec2010-07-25 14:29:16 +0300351 if (!bdi_cap_writeback_dirty(bdi))
352 continue;
Jens Axboe03ba3782009-09-09 09:08:54 +0200353 if (bdi->wb.task)
354 continue;
355 if (list_empty(&bdi->work_list) &&
356 !bdi_has_dirty_io(bdi))
357 continue;
358
Artem Bityutskiy080dcec2010-07-25 14:29:16 +0300359 WARN(!test_bit(BDI_registered, &bdi->state),
360 "bdi %p/%s is not registered!\n", bdi, bdi->name);
361
362 list_del_rcu(&bdi->bdi_list);
363 fork = true;
364 break;
Jens Axboe03ba3782009-09-09 09:08:54 +0200365 }
Artem Bityutskiy080dcec2010-07-25 14:29:16 +0300366 spin_unlock_bh(&bdi_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200367
Artem Bityutskiyc4ec7902010-07-25 14:29:14 +0300368 /* Keep working if default bdi still has things to do */
369 if (!list_empty(&me->bdi->work_list))
370 __set_current_state(TASK_RUNNING);
371
Artem Bityutskiy080dcec2010-07-25 14:29:16 +0300372 if (!fork) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200373 unsigned long wait;
374
Jens Axboe03ba3782009-09-09 09:08:54 +0200375 wait = msecs_to_jiffies(dirty_writeback_interval * 10);
Jens Axboe64231042010-05-21 20:00:35 +0200376 if (wait)
377 schedule_timeout(wait);
378 else
379 schedule();
Jens Axboe03ba3782009-09-09 09:08:54 +0200380 try_to_freeze();
381 continue;
382 }
383
384 __set_current_state(TASK_RUNNING);
385
386 /*
Artem Bityutskiy080dcec2010-07-25 14:29:16 +0300387 * Set the pending bit - if someone will try to unregister this
388 * bdi - it'll wait on this bit.
Jens Axboe03ba3782009-09-09 09:08:54 +0200389 */
Artem Bityutskiy080dcec2010-07-25 14:29:16 +0300390 set_bit(BDI_pending, &bdi->state);
391
392 /* Make sure no one uses the picked bdi */
393 synchronize_rcu();
Jens Axboe03ba3782009-09-09 09:08:54 +0200394
Artem Bityutskiy94eac5e2010-07-25 14:29:12 +0300395 task = kthread_run(bdi_writeback_thread, &bdi->wb, "flush-%s",
396 dev_name(bdi->dev));
397 if (IS_ERR(task)) {
Jens Axboe03ba3782009-09-09 09:08:54 +0200398 /*
Artem Bityutskiy94eac5e2010-07-25 14:29:12 +0300399 * If thread creation fails, then readd the bdi back to
400 * the list and force writeout of the bdi from this
401 * forker thread. That will free some memory and we can
402 * try again. Add it to the tail so we get a chance to
403 * flush other bdi's to free memory.
Jens Axboe03ba3782009-09-09 09:08:54 +0200404 */
Jens Axboecfc4ba52009-09-14 13:12:40 +0200405 spin_lock_bh(&bdi_lock);
Artem Bityutskiy080dcec2010-07-25 14:29:16 +0300406 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
Jens Axboecfc4ba52009-09-14 13:12:40 +0200407 spin_unlock_bh(&bdi_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200408
409 bdi_flush_io(bdi);
Artem Bityutskiy94eac5e2010-07-25 14:29:12 +0300410 } else
411 bdi->wb.task = task;
Jens Axboe03ba3782009-09-09 09:08:54 +0200412 }
413
414 return 0;
415}
416
Jens Axboecfc4ba52009-09-14 13:12:40 +0200417/*
418 * Remove bdi from bdi_list, and ensure that it is no longer visible
419 */
420static void bdi_remove_from_list(struct backing_dev_info *bdi)
421{
422 spin_lock_bh(&bdi_lock);
423 list_del_rcu(&bdi->bdi_list);
424 spin_unlock_bh(&bdi_lock);
425
426 synchronize_rcu();
427}
428
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700429int bdi_register(struct backing_dev_info *bdi, struct device *parent,
430 const char *fmt, ...)
431{
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700432 va_list args;
433 int ret = 0;
434 struct device *dev;
435
Andrew Morton69fc2082008-12-09 13:14:06 -0800436 if (bdi->dev) /* The driver needs to use separate queues per device */
Kay Sieversf1d0b062008-12-02 10:31:50 -0800437 goto exit;
438
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700439 va_start(args, fmt);
Greg Kroah-Hartman19051c52008-05-15 13:44:08 -0700440 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700441 va_end(args);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700442 if (IS_ERR(dev)) {
443 ret = PTR_ERR(dev);
444 goto exit;
445 }
446
Jens Axboecfc4ba52009-09-14 13:12:40 +0200447 spin_lock_bh(&bdi_lock);
448 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
449 spin_unlock_bh(&bdi_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200450
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700451 bdi->dev = dev;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700452
Jens Axboe03ba3782009-09-09 09:08:54 +0200453 /*
454 * Just start the forker thread for our default backing_dev_info,
455 * and add other bdi's to the list. They will get a thread created
456 * on-demand when they need it.
457 */
458 if (bdi_cap_flush_forker(bdi)) {
459 struct bdi_writeback *wb = &bdi->wb;
460
Artem Bityutskiy6f904ff2010-07-25 14:29:11 +0300461 wb->task = kthread_run(bdi_forker_thread, wb, "bdi-%s",
Jens Axboe03ba3782009-09-09 09:08:54 +0200462 dev_name(dev));
463 if (IS_ERR(wb->task)) {
464 wb->task = NULL;
465 ret = -ENOMEM;
466
Jens Axboecfc4ba52009-09-14 13:12:40 +0200467 bdi_remove_from_list(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +0200468 goto exit;
469 }
470 }
471
472 bdi_debug_register(bdi, dev_name(dev));
Jens Axboe500b0672009-09-09 09:10:25 +0200473 set_bit(BDI_registered, &bdi->state);
Dave Chinner455b2862010-07-07 13:24:06 +1000474 trace_writeback_bdi_register(bdi);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700475exit:
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700476 return ret;
477}
478EXPORT_SYMBOL(bdi_register);
479
480int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
481{
482 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
483}
484EXPORT_SYMBOL(bdi_register_dev);
485
Jens Axboe03ba3782009-09-09 09:08:54 +0200486/*
487 * Remove bdi from the global list and shutdown any threads we have running
488 */
489static void bdi_wb_shutdown(struct backing_dev_info *bdi)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200490{
Jens Axboe03ba3782009-09-09 09:08:54 +0200491 if (!bdi_cap_writeback_dirty(bdi))
492 return;
493
494 /*
495 * If setup is pending, wait for that to complete first
496 */
497 wait_on_bit(&bdi->state, BDI_pending, bdi_sched_wait,
498 TASK_UNINTERRUPTIBLE);
499
500 /*
501 * Make sure nobody finds us on the bdi_list anymore
502 */
Jens Axboecfc4ba52009-09-14 13:12:40 +0200503 bdi_remove_from_list(bdi);
Jens Axboe03ba3782009-09-09 09:08:54 +0200504
505 /*
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200506 * Finally, kill the kernel thread. We don't need to be RCU
Romit Dasguptac62b17a2009-11-12 13:08:11 +0100507 * safe anymore, since the bdi is gone from visibility. Force
508 * unfreeze of the thread before calling kthread_stop(), otherwise
509 * it would never exet if it is currently stuck in the refrigerator.
Jens Axboe03ba3782009-09-09 09:08:54 +0200510 */
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200511 if (bdi->wb.task) {
512 thaw_process(bdi->wb.task);
513 kthread_stop(bdi->wb.task);
Romit Dasguptac62b17a2009-11-12 13:08:11 +0100514 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200515}
516
Jens Axboe592b09a2009-10-29 11:46:12 +0100517/*
518 * This bdi is going away now, make sure that no super_blocks point to it
519 */
520static void bdi_prune_sb(struct backing_dev_info *bdi)
521{
522 struct super_block *sb;
523
524 spin_lock(&sb_lock);
525 list_for_each_entry(sb, &super_blocks, s_list) {
526 if (sb->s_bdi == bdi)
527 sb->s_bdi = NULL;
528 }
529 spin_unlock(&sb_lock);
530}
531
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700532void bdi_unregister(struct backing_dev_info *bdi)
533{
534 if (bdi->dev) {
Dave Chinner455b2862010-07-07 13:24:06 +1000535 trace_writeback_bdi_unregister(bdi);
Jens Axboe8c4db332009-11-03 20:18:44 +0100536 bdi_prune_sb(bdi);
537
Jens Axboe03ba3782009-09-09 09:08:54 +0200538 if (!bdi_cap_flush_forker(bdi))
539 bdi_wb_shutdown(bdi);
Miklos Szeredi76f14182008-04-30 00:54:36 -0700540 bdi_debug_unregister(bdi);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700541 device_unregister(bdi->dev);
542 bdi->dev = NULL;
543 }
544}
545EXPORT_SYMBOL(bdi_unregister);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700546
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700547int bdi_init(struct backing_dev_info *bdi)
548{
Jens Axboe03ba3782009-09-09 09:08:54 +0200549 int i, err;
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700550
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700551 bdi->dev = NULL;
552
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700553 bdi->min_ratio = 0;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700554 bdi->max_ratio = 100;
555 bdi->max_prop_frac = PROP_FRAC_BASE;
Jens Axboe03ba3782009-09-09 09:08:54 +0200556 spin_lock_init(&bdi->wb_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200557 INIT_LIST_HEAD(&bdi->bdi_list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200558 INIT_LIST_HEAD(&bdi->work_list);
559
560 bdi_wb_init(&bdi->wb, bdi);
561
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700562 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
Peter Zijlstraea319512008-12-26 15:08:55 +0100563 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700564 if (err)
565 goto err;
566 }
567
568 bdi->dirty_exceeded = 0;
569 err = prop_local_init_percpu(&bdi->completions);
570
571 if (err) {
572err:
Denis Cheng4b01a0b2007-12-04 23:45:07 -0800573 while (i--)
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700574 percpu_counter_destroy(&bdi->bdi_stat[i]);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700575 }
576
577 return err;
578}
579EXPORT_SYMBOL(bdi_init);
580
581void bdi_destroy(struct backing_dev_info *bdi)
582{
583 int i;
584
Jens Axboece5f8e72009-09-14 12:57:56 +0200585 /*
586 * Splice our entries to the default_backing_dev_info, if this
587 * bdi disappears
588 */
589 if (bdi_has_dirty_io(bdi)) {
590 struct bdi_writeback *dst = &default_backing_dev_info.wb;
591
592 spin_lock(&inode_lock);
593 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
594 list_splice(&bdi->wb.b_io, &dst->b_io);
595 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
596 spin_unlock(&inode_lock);
597 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200598
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700599 bdi_unregister(bdi);
600
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700601 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
602 percpu_counter_destroy(&bdi->bdi_stat[i]);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700603
604 prop_local_destroy_percpu(&bdi->completions);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700605}
606EXPORT_SYMBOL(bdi_destroy);
607
Jens Axboec3c53202010-04-22 11:37:01 +0200608/*
609 * For use from filesystems to quickly init and register a bdi associated
610 * with dirty writeback
611 */
612int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
613 unsigned int cap)
614{
615 char tmp[32];
616 int err;
617
618 bdi->name = name;
619 bdi->capabilities = cap;
620 err = bdi_init(bdi);
621 if (err)
622 return err;
623
624 sprintf(tmp, "%.28s%s", name, "-%d");
625 err = bdi_register(bdi, NULL, tmp, atomic_long_inc_return(&bdi_seq));
626 if (err) {
627 bdi_destroy(bdi);
628 return err;
629 }
630
631 return 0;
632}
633EXPORT_SYMBOL(bdi_setup_and_register);
634
Andrew Morton3fcfab12006-10-19 23:28:16 -0700635static wait_queue_head_t congestion_wqh[2] = {
636 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
637 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
638 };
639
Jens Axboe1faa16d2009-04-06 14:48:01 +0200640void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700641{
642 enum bdi_state bit;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200643 wait_queue_head_t *wqh = &congestion_wqh[sync];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700644
Jens Axboe1faa16d2009-04-06 14:48:01 +0200645 bit = sync ? BDI_sync_congested : BDI_async_congested;
Andrew Morton3fcfab12006-10-19 23:28:16 -0700646 clear_bit(bit, &bdi->state);
647 smp_mb__after_clear_bit();
648 if (waitqueue_active(wqh))
649 wake_up(wqh);
650}
651EXPORT_SYMBOL(clear_bdi_congested);
652
Jens Axboe1faa16d2009-04-06 14:48:01 +0200653void set_bdi_congested(struct backing_dev_info *bdi, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700654{
655 enum bdi_state bit;
656
Jens Axboe1faa16d2009-04-06 14:48:01 +0200657 bit = sync ? BDI_sync_congested : BDI_async_congested;
Andrew Morton3fcfab12006-10-19 23:28:16 -0700658 set_bit(bit, &bdi->state);
659}
660EXPORT_SYMBOL(set_bdi_congested);
661
662/**
663 * congestion_wait - wait for a backing_dev to become uncongested
Jens Axboe8aa7e842009-07-09 14:52:32 +0200664 * @sync: SYNC or ASYNC IO
Andrew Morton3fcfab12006-10-19 23:28:16 -0700665 * @timeout: timeout in jiffies
666 *
667 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
668 * write congestion. If no backing_devs are congested then just wait for the
669 * next write to be completed.
670 */
Jens Axboe8aa7e842009-07-09 14:52:32 +0200671long congestion_wait(int sync, long timeout)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700672{
673 long ret;
674 DEFINE_WAIT(wait);
Jens Axboe8aa7e842009-07-09 14:52:32 +0200675 wait_queue_head_t *wqh = &congestion_wqh[sync];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700676
677 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
678 ret = io_schedule_timeout(timeout);
679 finish_wait(wqh, &wait);
680 return ret;
681}
682EXPORT_SYMBOL(congestion_wait);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700683