blob: d014ee5fcbbd2fa8004ae642f83571ddd949a5e6 [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 +010017struct backing_dev_info default_backing_dev_info = {
Jens Axboed9938312009-06-12 14:45:52 +020018 .name = "default",
Jens Axboe26160152009-03-17 09:35:06 +010019 .ra_pages = VM_MAX_READAHEAD * 1024 / PAGE_CACHE_SIZE,
20 .state = 0,
21 .capabilities = BDI_CAP_MAP_COPY,
Jens Axboe26160152009-03-17 09:35:06 +010022};
23EXPORT_SYMBOL_GPL(default_backing_dev_info);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070024
Jörn Engel5129a462010-04-25 08:54:42 +020025struct backing_dev_info noop_backing_dev_info = {
26 .name = "noop",
Jan Kara976e48f2010-09-21 11:48:55 +020027 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Jörn Engel5129a462010-04-25 08:54:42 +020028};
29EXPORT_SYMBOL_GPL(noop_backing_dev_info);
30
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070031static struct class *bdi_class;
Jens Axboecfc4ba52009-09-14 13:12:40 +020032
33/*
Tejun Heo181387d2013-04-01 19:08:06 -070034 * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
Jens Axboecfc4ba52009-09-14 13:12:40 +020035 * locking.
36 */
Jens Axboe03ba3782009-09-09 09:08:54 +020037DEFINE_SPINLOCK(bdi_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +020038LIST_HEAD(bdi_list);
Jens Axboe03ba3782009-09-09 09:08:54 +020039
Tejun Heo839a8e82013-04-01 19:08:06 -070040/* bdi_wq serves all asynchronous writeback tasks */
41struct workqueue_struct *bdi_wq;
42
Christoph Hellwigf758eea2011-04-21 18:19:44 -060043void bdi_lock_two(struct bdi_writeback *wb1, struct bdi_writeback *wb2)
44{
45 if (wb1 < wb2) {
46 spin_lock(&wb1->list_lock);
47 spin_lock_nested(&wb2->list_lock, 1);
48 } else {
49 spin_lock(&wb2->list_lock);
50 spin_lock_nested(&wb1->list_lock, 1);
51 }
52}
53
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;
Gustavo F. Padovan345227d2011-05-20 21:23:37 +020072 unsigned long nr_dirty, nr_io, nr_more_io;
Jens Axboef09b00d2009-05-25 09:08:21 +020073 struct inode *inode;
74
Gustavo F. Padovan345227d2011-05-20 21:23:37 +020075 nr_dirty = nr_io = nr_more_io = 0;
Christoph Hellwigf758eea2011-04-21 18:19:44 -060076 spin_lock(&wb->list_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++;
Christoph Hellwigf758eea2011-04-21 18:19:44 -060083 spin_unlock(&wb->list_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,
Wu Fengguang00821b02010-08-29 11:28:45 -060090 "BdiWriteback: %10lu kB\n"
91 "BdiReclaimable: %10lu kB\n"
92 "BdiDirtyThresh: %10lu kB\n"
93 "DirtyThresh: %10lu kB\n"
94 "BackgroundThresh: %10lu kB\n"
Wu Fengguangc8e28ce2011-01-23 10:07:47 -060095 "BdiDirtied: %10lu kB\n"
Wu Fengguang00821b02010-08-29 11:28:45 -060096 "BdiWritten: %10lu kB\n"
97 "BdiWriteBandwidth: %10lu kBps\n"
98 "b_dirty: %10lu\n"
99 "b_io: %10lu\n"
100 "b_more_io: %10lu\n"
101 "bdi_list: %10u\n"
102 "state: %10lx\n",
Miklos Szeredi76f14182008-04-30 00:54:36 -0700103 (unsigned long) K(bdi_stat(bdi, BDI_WRITEBACK)),
104 (unsigned long) K(bdi_stat(bdi, BDI_RECLAIMABLE)),
Jan Karaf7d2b1e2010-12-08 22:44:24 -0600105 K(bdi_thresh),
106 K(dirty_thresh),
107 K(background_thresh),
Wu Fengguangc8e28ce2011-01-23 10:07:47 -0600108 (unsigned long) K(bdi_stat(bdi, BDI_DIRTIED)),
Jan Karaf7d2b1e2010-12-08 22:44:24 -0600109 (unsigned long) K(bdi_stat(bdi, BDI_WRITTEN)),
Wu Fengguang00821b02010-08-29 11:28:45 -0600110 (unsigned long) K(bdi->write_bandwidth),
Jan Karaf7d2b1e2010-12-08 22:44:24 -0600111 nr_dirty,
112 nr_io,
113 nr_more_io,
Christoph Hellwigc1955ce2010-06-19 23:08:06 +0200114 !list_empty(&bdi->bdi_list), bdi->state);
Miklos Szeredi76f14182008-04-30 00:54:36 -0700115#undef K
116
117 return 0;
118}
119
120static int bdi_debug_stats_open(struct inode *inode, struct file *file)
121{
122 return single_open(file, bdi_debug_stats_show, inode->i_private);
123}
124
125static const struct file_operations bdi_debug_stats_fops = {
126 .open = bdi_debug_stats_open,
127 .read = seq_read,
128 .llseek = seq_lseek,
129 .release = single_release,
130};
131
132static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
133{
134 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
135 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
136 bdi, &bdi_debug_stats_fops);
137}
138
139static void bdi_debug_unregister(struct backing_dev_info *bdi)
140{
141 debugfs_remove(bdi->debug_stats);
142 debugfs_remove(bdi->debug_dir);
143}
144#else
145static inline void bdi_debug_init(void)
146{
147}
148static inline void bdi_debug_register(struct backing_dev_info *bdi,
149 const char *name)
150{
151}
152static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
153{
154}
155#endif
156
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700157static ssize_t read_ahead_kb_store(struct device *dev,
158 struct device_attribute *attr,
159 const char *buf, size_t count)
160{
161 struct backing_dev_info *bdi = dev_get_drvdata(dev);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700162 unsigned long read_ahead_kb;
Namjae Jeon7034ed12012-08-25 16:57:27 +0800163 ssize_t ret;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700164
Namjae Jeon7034ed12012-08-25 16:57:27 +0800165 ret = kstrtoul(buf, 10, &read_ahead_kb);
166 if (ret < 0)
167 return ret;
168
169 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
170
171 return count;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700172}
173
174#define K(pages) ((pages) << (PAGE_SHIFT - 10))
175
176#define BDI_SHOW(name, expr) \
177static ssize_t name##_show(struct device *dev, \
178 struct device_attribute *attr, char *page) \
179{ \
180 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
181 \
182 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
183}
184
185BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
186
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700187static ssize_t min_ratio_store(struct device *dev,
188 struct device_attribute *attr, const char *buf, size_t count)
189{
190 struct backing_dev_info *bdi = dev_get_drvdata(dev);
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700191 unsigned int ratio;
Namjae Jeon7034ed12012-08-25 16:57:27 +0800192 ssize_t ret;
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700193
Namjae Jeon7034ed12012-08-25 16:57:27 +0800194 ret = kstrtouint(buf, 10, &ratio);
195 if (ret < 0)
196 return ret;
197
198 ret = bdi_set_min_ratio(bdi, ratio);
199 if (!ret)
200 ret = count;
201
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700202 return ret;
203}
204BDI_SHOW(min_ratio, bdi->min_ratio)
205
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700206static ssize_t max_ratio_store(struct device *dev,
207 struct device_attribute *attr, const char *buf, size_t count)
208{
209 struct backing_dev_info *bdi = dev_get_drvdata(dev);
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700210 unsigned int ratio;
Namjae Jeon7034ed12012-08-25 16:57:27 +0800211 ssize_t ret;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700212
Namjae Jeon7034ed12012-08-25 16:57:27 +0800213 ret = kstrtouint(buf, 10, &ratio);
214 if (ret < 0)
215 return ret;
216
217 ret = bdi_set_max_ratio(bdi, ratio);
218 if (!ret)
219 ret = count;
220
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700221 return ret;
222}
223BDI_SHOW(max_ratio, bdi->max_ratio)
224
Darrick J. Wong7d311cd2013-02-21 16:42:48 -0800225static ssize_t stable_pages_required_show(struct device *dev,
226 struct device_attribute *attr,
227 char *page)
228{
229 struct backing_dev_info *bdi = dev_get_drvdata(dev);
230
231 return snprintf(page, PAGE_SIZE-1, "%d\n",
232 bdi_cap_stable_pages_required(bdi) ? 1 : 0);
233}
234
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700235#define __ATTR_RW(attr) __ATTR(attr, 0644, attr##_show, attr##_store)
236
237static struct device_attribute bdi_dev_attrs[] = {
238 __ATTR_RW(read_ahead_kb),
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700239 __ATTR_RW(min_ratio),
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700240 __ATTR_RW(max_ratio),
Darrick J. Wong7d311cd2013-02-21 16:42:48 -0800241 __ATTR_RO(stable_pages_required),
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700242 __ATTR_NULL,
243};
244
245static __init int bdi_class_init(void)
246{
247 bdi_class = class_create(THIS_MODULE, "bdi");
Anton Blanchard14421452010-04-02 09:46:55 +0200248 if (IS_ERR(bdi_class))
249 return PTR_ERR(bdi_class);
250
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700251 bdi_class->dev_attrs = bdi_dev_attrs;
Miklos Szeredi76f14182008-04-30 00:54:36 -0700252 bdi_debug_init();
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700253 return 0;
254}
Miklos Szeredi76f14182008-04-30 00:54:36 -0700255postcore_initcall(bdi_class_init);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700256
Jens Axboe26160152009-03-17 09:35:06 +0100257static int __init default_bdi_init(void)
258{
259 int err;
260
Tejun Heo839a8e82013-04-01 19:08:06 -0700261 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
Tejun Heob5c872d2013-04-01 19:08:06 -0700262 WQ_UNBOUND | WQ_SYSFS, 0);
Tejun Heo839a8e82013-04-01 19:08:06 -0700263 if (!bdi_wq)
264 return -ENOMEM;
265
Jens Axboe26160152009-03-17 09:35:06 +0100266 err = bdi_init(&default_backing_dev_info);
267 if (!err)
268 bdi_register(&default_backing_dev_info, NULL, "default");
Jan Kara976e48f2010-09-21 11:48:55 +0200269 err = bdi_init(&noop_backing_dev_info);
Jens Axboe26160152009-03-17 09:35:06 +0100270
271 return err;
272}
273subsys_initcall(default_bdi_init);
274
Jens Axboe03ba3782009-09-09 09:08:54 +0200275int bdi_has_dirty_io(struct backing_dev_info *bdi)
276{
277 return wb_has_dirty_io(&bdi->wb);
278}
279
Artem Bityutskiy64677162010-07-25 14:29:22 +0300280/*
281 * This function is used when the first inode for this bdi is marked dirty. It
282 * wakes-up the corresponding bdi thread which should then take care of the
283 * periodic background write-out of dirty inodes. Since the write-out would
284 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
285 * set up a timer which wakes the bdi thread up later.
286 *
287 * Note, we wouldn't bother setting up the timer, but this function is on the
288 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
289 * by delaying the wake-up.
290 */
291void bdi_wakeup_thread_delayed(struct backing_dev_info *bdi)
292{
293 unsigned long timeout;
294
295 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
Tejun Heo839a8e82013-04-01 19:08:06 -0700296 mod_delayed_work(bdi_wq, &bdi->wb.dwork, timeout);
Jens Axboe03ba3782009-09-09 09:08:54 +0200297}
298
Jens Axboecfc4ba52009-09-14 13:12:40 +0200299/*
300 * Remove bdi from bdi_list, and ensure that it is no longer visible
301 */
302static void bdi_remove_from_list(struct backing_dev_info *bdi)
303{
304 spin_lock_bh(&bdi_lock);
305 list_del_rcu(&bdi->bdi_list);
306 spin_unlock_bh(&bdi_lock);
307
Mikulas Patockaef323082011-07-23 20:44:24 +0200308 synchronize_rcu_expedited();
Tejun Heo839a8e82013-04-01 19:08:06 -0700309
310 /* bdi_list is now unused, clear it to mark @bdi dying */
311 INIT_LIST_HEAD(&bdi->bdi_list);
Jens Axboecfc4ba52009-09-14 13:12:40 +0200312}
313
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700314int bdi_register(struct backing_dev_info *bdi, struct device *parent,
315 const char *fmt, ...)
316{
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700317 va_list args;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700318 struct device *dev;
319
Andrew Morton69fc2082008-12-09 13:14:06 -0800320 if (bdi->dev) /* The driver needs to use separate queues per device */
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300321 return 0;
Kay Sieversf1d0b062008-12-02 10:31:50 -0800322
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700323 va_start(args, fmt);
Greg Kroah-Hartman19051c52008-05-15 13:44:08 -0700324 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700325 va_end(args);
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300326 if (IS_ERR(dev))
327 return PTR_ERR(dev);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200328
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700329 bdi->dev = dev;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700330
Jens Axboe03ba3782009-09-09 09:08:54 +0200331 bdi_debug_register(bdi, dev_name(dev));
Jens Axboe500b0672009-09-09 09:10:25 +0200332 set_bit(BDI_registered, &bdi->state);
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300333
334 spin_lock_bh(&bdi_lock);
335 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
336 spin_unlock_bh(&bdi_lock);
337
Dave Chinner455b2862010-07-07 13:24:06 +1000338 trace_writeback_bdi_register(bdi);
Artem Bityutskiyc284de62010-07-25 14:29:25 +0300339 return 0;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700340}
341EXPORT_SYMBOL(bdi_register);
342
343int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
344{
345 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
346}
347EXPORT_SYMBOL(bdi_register_dev);
348
Jens Axboe03ba3782009-09-09 09:08:54 +0200349/*
350 * Remove bdi from the global list and shutdown any threads we have running
351 */
352static void bdi_wb_shutdown(struct backing_dev_info *bdi)
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200353{
Jens Axboe03ba3782009-09-09 09:08:54 +0200354 if (!bdi_cap_writeback_dirty(bdi))
355 return;
356
357 /*
Artem Bityutskiyfff5b852010-07-25 14:29:20 +0300358 * Make sure nobody finds us on the bdi_list anymore
359 */
360 bdi_remove_from_list(bdi);
361
362 /*
Tejun Heo839a8e82013-04-01 19:08:06 -0700363 * Drain work list and shutdown the delayed_work. At this point,
364 * @bdi->bdi_list is empty telling bdi_Writeback_workfn() that @bdi
365 * is dying and its work_list needs to be drained no matter what.
Jens Axboe03ba3782009-09-09 09:08:54 +0200366 */
Tejun Heo839a8e82013-04-01 19:08:06 -0700367 mod_delayed_work(bdi_wq, &bdi->wb.dwork, 0);
368 flush_delayed_work(&bdi->wb.dwork);
369 WARN_ON(!list_empty(&bdi->work_list));
Jens Axboe03ba3782009-09-09 09:08:54 +0200370
371 /*
Tejun Heo839a8e82013-04-01 19:08:06 -0700372 * This shouldn't be necessary unless @bdi for some reason has
373 * unflushed dirty IO after work_list is drained. Do it anyway
374 * just in case.
Jens Axboe03ba3782009-09-09 09:08:54 +0200375 */
Tejun Heo839a8e82013-04-01 19:08:06 -0700376 cancel_delayed_work_sync(&bdi->wb.dwork);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200377}
378
Jens Axboe592b09a2009-10-29 11:46:12 +0100379/*
380 * This bdi is going away now, make sure that no super_blocks point to it
381 */
382static void bdi_prune_sb(struct backing_dev_info *bdi)
383{
384 struct super_block *sb;
385
386 spin_lock(&sb_lock);
387 list_for_each_entry(sb, &super_blocks, s_list) {
388 if (sb->s_bdi == bdi)
Jens Axboe95f28602011-03-17 11:13:12 +0100389 sb->s_bdi = &default_backing_dev_info;
Jens Axboe592b09a2009-10-29 11:46:12 +0100390 }
391 spin_unlock(&sb_lock);
392}
393
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700394void bdi_unregister(struct backing_dev_info *bdi)
395{
Rabin Vincent2673b4c2012-01-29 12:17:33 -0600396 struct device *dev = bdi->dev;
397
398 if (dev) {
Peter Zijlstraccb61082011-07-25 17:11:57 -0700399 bdi_set_min_ratio(bdi, 0);
Dave Chinner455b2862010-07-07 13:24:06 +1000400 trace_writeback_bdi_unregister(bdi);
Jens Axboe8c4db332009-11-03 20:18:44 +0100401 bdi_prune_sb(bdi);
402
Tejun Heo839a8e82013-04-01 19:08:06 -0700403 bdi_wb_shutdown(bdi);
Miklos Szeredi76f14182008-04-30 00:54:36 -0700404 bdi_debug_unregister(bdi);
Rabin Vincent2673b4c2012-01-29 12:17:33 -0600405
406 spin_lock_bh(&bdi->wb_lock);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700407 bdi->dev = NULL;
Rabin Vincent2673b4c2012-01-29 12:17:33 -0600408 spin_unlock_bh(&bdi->wb_lock);
409
410 device_unregister(dev);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700411 }
412}
413EXPORT_SYMBOL(bdi_unregister);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700414
Artem Bityutskiy64677162010-07-25 14:29:22 +0300415static void bdi_wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
416{
417 memset(wb, 0, sizeof(*wb));
418
419 wb->bdi = bdi;
420 wb->last_old_flush = jiffies;
421 INIT_LIST_HEAD(&wb->b_dirty);
422 INIT_LIST_HEAD(&wb->b_io);
423 INIT_LIST_HEAD(&wb->b_more_io);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600424 spin_lock_init(&wb->list_lock);
Tejun Heo839a8e82013-04-01 19:08:06 -0700425 INIT_DELAYED_WORK(&wb->dwork, bdi_writeback_workfn);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300426}
427
Wu Fengguange98be2d2010-08-29 11:22:30 -0600428/*
429 * Initial write bandwidth: 100 MB/s
430 */
431#define INIT_BW (100 << (20 - PAGE_SHIFT))
432
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700433int bdi_init(struct backing_dev_info *bdi)
434{
Jens Axboe03ba3782009-09-09 09:08:54 +0200435 int i, err;
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700436
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700437 bdi->dev = NULL;
438
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700439 bdi->min_ratio = 0;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700440 bdi->max_ratio = 100;
Jan Karaeb608e32012-05-24 18:59:11 +0200441 bdi->max_prop_frac = FPROP_FRAC_BASE;
Jens Axboe03ba3782009-09-09 09:08:54 +0200442 spin_lock_init(&bdi->wb_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200443 INIT_LIST_HEAD(&bdi->bdi_list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200444 INIT_LIST_HEAD(&bdi->work_list);
445
446 bdi_wb_init(&bdi->wb, bdi);
447
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700448 for (i = 0; i < NR_BDI_STAT_ITEMS; i++) {
Peter Zijlstraea319512008-12-26 15:08:55 +0100449 err = percpu_counter_init(&bdi->bdi_stat[i], 0);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700450 if (err)
451 goto err;
452 }
453
454 bdi->dirty_exceeded = 0;
Wu Fengguange98be2d2010-08-29 11:22:30 -0600455
456 bdi->bw_time_stamp = jiffies;
457 bdi->written_stamp = 0;
458
Wu Fengguang73811312011-08-26 15:53:24 -0600459 bdi->balanced_dirty_ratelimit = INIT_BW;
Wu Fengguangbe3ffa22011-06-12 10:51:31 -0600460 bdi->dirty_ratelimit = INIT_BW;
Wu Fengguange98be2d2010-08-29 11:22:30 -0600461 bdi->write_bandwidth = INIT_BW;
462 bdi->avg_write_bandwidth = INIT_BW;
463
Jan Karaeb608e32012-05-24 18:59:11 +0200464 err = fprop_local_init_percpu(&bdi->completions);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700465
466 if (err) {
467err:
Denis Cheng4b01a0b2007-12-04 23:45:07 -0800468 while (i--)
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700469 percpu_counter_destroy(&bdi->bdi_stat[i]);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700470 }
471
472 return err;
473}
474EXPORT_SYMBOL(bdi_init);
475
476void bdi_destroy(struct backing_dev_info *bdi)
477{
478 int i;
479
Jens Axboece5f8e72009-09-14 12:57:56 +0200480 /*
481 * Splice our entries to the default_backing_dev_info, if this
482 * bdi disappears
483 */
484 if (bdi_has_dirty_io(bdi)) {
485 struct bdi_writeback *dst = &default_backing_dev_info.wb;
486
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600487 bdi_lock_two(&bdi->wb, dst);
Jens Axboece5f8e72009-09-14 12:57:56 +0200488 list_splice(&bdi->wb.b_dirty, &dst->b_dirty);
489 list_splice(&bdi->wb.b_io, &dst->b_io);
490 list_splice(&bdi->wb.b_more_io, &dst->b_more_io);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600491 spin_unlock(&bdi->wb.list_lock);
492 spin_unlock(&dst->list_lock);
Jens Axboece5f8e72009-09-14 12:57:56 +0200493 }
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200494
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700495 bdi_unregister(bdi);
496
Rabin Vincent7a401a92011-11-11 13:29:04 +0100497 /*
Tejun Heo839a8e82013-04-01 19:08:06 -0700498 * If bdi_unregister() had already been called earlier, the dwork
499 * could still be pending because bdi_prune_sb() can race with the
500 * bdi_wakeup_thread_delayed() calls from __mark_inode_dirty().
Rabin Vincent7a401a92011-11-11 13:29:04 +0100501 */
Tejun Heo839a8e82013-04-01 19:08:06 -0700502 cancel_delayed_work_sync(&bdi->wb.dwork);
Rabin Vincent7a401a92011-11-11 13:29:04 +0100503
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700504 for (i = 0; i < NR_BDI_STAT_ITEMS; i++)
505 percpu_counter_destroy(&bdi->bdi_stat[i]);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700506
Jan Karaeb608e32012-05-24 18:59:11 +0200507 fprop_local_destroy_percpu(&bdi->completions);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700508}
509EXPORT_SYMBOL(bdi_destroy);
510
Jens Axboec3c53202010-04-22 11:37:01 +0200511/*
512 * For use from filesystems to quickly init and register a bdi associated
513 * with dirty writeback
514 */
515int bdi_setup_and_register(struct backing_dev_info *bdi, char *name,
516 unsigned int cap)
517{
Jens Axboec3c53202010-04-22 11:37:01 +0200518 int err;
519
520 bdi->name = name;
521 bdi->capabilities = cap;
522 err = bdi_init(bdi);
523 if (err)
524 return err;
525
Kees Cook02aa2a32013-07-03 15:04:56 -0700526 err = bdi_register(bdi, NULL, "%.28s-%ld", name,
527 atomic_long_inc_return(&bdi_seq));
Jens Axboec3c53202010-04-22 11:37:01 +0200528 if (err) {
529 bdi_destroy(bdi);
530 return err;
531 }
532
533 return 0;
534}
535EXPORT_SYMBOL(bdi_setup_and_register);
536
Andrew Morton3fcfab12006-10-19 23:28:16 -0700537static wait_queue_head_t congestion_wqh[2] = {
538 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
539 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
540 };
Mel Gorman0e093d992010-10-26 14:21:45 -0700541static atomic_t nr_bdi_congested[2];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700542
Jens Axboe1faa16d2009-04-06 14:48:01 +0200543void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700544{
545 enum bdi_state bit;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200546 wait_queue_head_t *wqh = &congestion_wqh[sync];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700547
Jens Axboe1faa16d2009-04-06 14:48:01 +0200548 bit = sync ? BDI_sync_congested : BDI_async_congested;
Mel Gorman0e093d992010-10-26 14:21:45 -0700549 if (test_and_clear_bit(bit, &bdi->state))
550 atomic_dec(&nr_bdi_congested[sync]);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700551 smp_mb__after_clear_bit();
552 if (waitqueue_active(wqh))
553 wake_up(wqh);
554}
555EXPORT_SYMBOL(clear_bdi_congested);
556
Jens Axboe1faa16d2009-04-06 14:48:01 +0200557void set_bdi_congested(struct backing_dev_info *bdi, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700558{
559 enum bdi_state bit;
560
Jens Axboe1faa16d2009-04-06 14:48:01 +0200561 bit = sync ? BDI_sync_congested : BDI_async_congested;
Mel Gorman0e093d992010-10-26 14:21:45 -0700562 if (!test_and_set_bit(bit, &bdi->state))
563 atomic_inc(&nr_bdi_congested[sync]);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700564}
565EXPORT_SYMBOL(set_bdi_congested);
566
567/**
568 * congestion_wait - wait for a backing_dev to become uncongested
Jens Axboe8aa7e842009-07-09 14:52:32 +0200569 * @sync: SYNC or ASYNC IO
Andrew Morton3fcfab12006-10-19 23:28:16 -0700570 * @timeout: timeout in jiffies
571 *
572 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
573 * write congestion. If no backing_devs are congested then just wait for the
574 * next write to be completed.
575 */
Jens Axboe8aa7e842009-07-09 14:52:32 +0200576long congestion_wait(int sync, long timeout)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700577{
578 long ret;
Mel Gorman52bb9192010-10-26 14:21:41 -0700579 unsigned long start = jiffies;
Andrew Morton3fcfab12006-10-19 23:28:16 -0700580 DEFINE_WAIT(wait);
Jens Axboe8aa7e842009-07-09 14:52:32 +0200581 wait_queue_head_t *wqh = &congestion_wqh[sync];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700582
583 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
584 ret = io_schedule_timeout(timeout);
585 finish_wait(wqh, &wait);
Mel Gorman52bb9192010-10-26 14:21:41 -0700586
587 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
588 jiffies_to_usecs(jiffies - start));
589
Andrew Morton3fcfab12006-10-19 23:28:16 -0700590 return ret;
591}
592EXPORT_SYMBOL(congestion_wait);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700593
Mel Gorman0e093d992010-10-26 14:21:45 -0700594/**
595 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
596 * @zone: A zone to check if it is heavily congested
597 * @sync: SYNC or ASYNC IO
598 * @timeout: timeout in jiffies
599 *
600 * In the event of a congested backing_dev (any backing_dev) and the given
601 * @zone has experienced recent congestion, this waits for up to @timeout
602 * jiffies for either a BDI to exit congestion of the given @sync queue
603 * or a write to complete.
604 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300605 * In the absence of zone congestion, cond_resched() is called to yield
Mel Gorman0e093d992010-10-26 14:21:45 -0700606 * the processor if necessary but otherwise does not sleep.
607 *
608 * The return value is 0 if the sleep is for the full timeout. Otherwise,
609 * it is the number of jiffies that were still remaining when the function
610 * returned. return_value == timeout implies the function did not sleep.
611 */
612long wait_iff_congested(struct zone *zone, int sync, long timeout)
613{
614 long ret;
615 unsigned long start = jiffies;
616 DEFINE_WAIT(wait);
617 wait_queue_head_t *wqh = &congestion_wqh[sync];
618
619 /*
620 * If there is no congestion, or heavy congestion is not being
621 * encountered in the current zone, yield if necessary instead
622 * of sleeping on the congestion queue
623 */
624 if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
625 !zone_is_reclaim_congested(zone)) {
626 cond_resched();
627
628 /* In case we scheduled, work out time remaining */
629 ret = timeout - (jiffies - start);
630 if (ret < 0)
631 ret = 0;
632
633 goto out;
634 }
635
636 /* Sleep until uncongested or a write happens */
637 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
638 ret = io_schedule_timeout(timeout);
639 finish_wait(wqh, &wait);
640
641out:
642 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
643 jiffies_to_usecs(jiffies - start));
644
645 return ret;
646}
647EXPORT_SYMBOL(wait_iff_congested);
Wanpeng Li3965c9a2012-07-31 16:41:52 -0700648
649int pdflush_proc_obsolete(struct ctl_table *table, int write,
650 void __user *buffer, size_t *lenp, loff_t *ppos)
651{
652 char kbuf[] = "0\n";
653
654 if (*ppos) {
655 *lenp = 0;
656 return 0;
657 }
658
659 if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
660 return -EFAULT;
661 printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",
662 table->procname);
663
664 *lenp = 2;
665 *ppos += *lenp;
666 return 2;
667}