blob: ff85ecb7d6a103c94d4da80c340f24b919c259e3 [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
Jörn Engel5129a462010-04-25 08:54:42 +020017struct backing_dev_info noop_backing_dev_info = {
18 .name = "noop",
Jan Kara976e48f2010-09-21 11:48:55 +020019 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Jörn Engel5129a462010-04-25 08:54:42 +020020};
Jörn Engel5129a462010-04-25 08:54:42 +020021
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070022static struct class *bdi_class;
Jens Axboecfc4ba52009-09-14 13:12:40 +020023
24/*
Tejun Heo181387d2013-04-01 19:08:06 -070025 * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
Jens Axboecfc4ba52009-09-14 13:12:40 +020026 * locking.
27 */
Jens Axboe03ba3782009-09-09 09:08:54 +020028DEFINE_SPINLOCK(bdi_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +020029LIST_HEAD(bdi_list);
Jens Axboe03ba3782009-09-09 09:08:54 +020030
Tejun Heo839a8e82013-04-01 19:08:06 -070031/* bdi_wq serves all asynchronous writeback tasks */
32struct workqueue_struct *bdi_wq;
33
Miklos Szeredi76f14182008-04-30 00:54:36 -070034#ifdef CONFIG_DEBUG_FS
35#include <linux/debugfs.h>
36#include <linux/seq_file.h>
37
38static struct dentry *bdi_debug_root;
39
40static void bdi_debug_init(void)
41{
42 bdi_debug_root = debugfs_create_dir("bdi", NULL);
43}
44
45static int bdi_debug_stats_show(struct seq_file *m, void *v)
46{
47 struct backing_dev_info *bdi = m->private;
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020048 struct bdi_writeback *wb = &bdi->wb;
David Rientjes364aeb22009-01-06 14:39:29 -080049 unsigned long background_thresh;
50 unsigned long dirty_thresh;
51 unsigned long bdi_thresh;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050052 unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time;
Jens Axboef09b00d2009-05-25 09:08:21 +020053 struct inode *inode;
54
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050055 nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0;
Christoph Hellwigf758eea2011-04-21 18:19:44 -060056 spin_lock(&wb->list_lock);
Nick Piggin7ccf19a2010-10-21 11:49:30 +110057 list_for_each_entry(inode, &wb->b_dirty, i_wb_list)
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020058 nr_dirty++;
Nick Piggin7ccf19a2010-10-21 11:49:30 +110059 list_for_each_entry(inode, &wb->b_io, i_wb_list)
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020060 nr_io++;
Nick Piggin7ccf19a2010-10-21 11:49:30 +110061 list_for_each_entry(inode, &wb->b_more_io, i_wb_list)
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020062 nr_more_io++;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050063 list_for_each_entry(inode, &wb->b_dirty_time, i_wb_list)
64 if (inode->i_state & I_DIRTY_TIME)
65 nr_dirty_time++;
Christoph Hellwigf758eea2011-04-21 18:19:44 -060066 spin_unlock(&wb->list_lock);
Miklos Szeredi76f14182008-04-30 00:54:36 -070067
Wu Fengguang16c40422010-08-11 14:17:39 -070068 global_dirty_limits(&background_thresh, &dirty_thresh);
Tejun Heoa88a3412015-05-22 17:13:28 -040069 bdi_thresh = wb_dirty_limit(wb, dirty_thresh);
Miklos Szeredi76f14182008-04-30 00:54:36 -070070
71#define K(x) ((x) << (PAGE_SHIFT - 10))
72 seq_printf(m,
Wu Fengguang00821b02010-08-29 11:28:45 -060073 "BdiWriteback: %10lu kB\n"
74 "BdiReclaimable: %10lu kB\n"
75 "BdiDirtyThresh: %10lu kB\n"
76 "DirtyThresh: %10lu kB\n"
77 "BackgroundThresh: %10lu kB\n"
Wu Fengguangc8e28ce2011-01-23 10:07:47 -060078 "BdiDirtied: %10lu kB\n"
Wu Fengguang00821b02010-08-29 11:28:45 -060079 "BdiWritten: %10lu kB\n"
80 "BdiWriteBandwidth: %10lu kBps\n"
81 "b_dirty: %10lu\n"
82 "b_io: %10lu\n"
83 "b_more_io: %10lu\n"
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050084 "b_dirty_time: %10lu\n"
Wu Fengguang00821b02010-08-29 11:28:45 -060085 "bdi_list: %10u\n"
86 "state: %10lx\n",
Tejun Heo93f78d82015-05-22 17:13:27 -040087 (unsigned long) K(wb_stat(wb, WB_WRITEBACK)),
88 (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)),
Jan Karaf7d2b1e2010-12-08 22:44:24 -060089 K(bdi_thresh),
90 K(dirty_thresh),
91 K(background_thresh),
Tejun Heo93f78d82015-05-22 17:13:27 -040092 (unsigned long) K(wb_stat(wb, WB_DIRTIED)),
93 (unsigned long) K(wb_stat(wb, WB_WRITTEN)),
Tejun Heoa88a3412015-05-22 17:13:28 -040094 (unsigned long) K(wb->write_bandwidth),
Jan Karaf7d2b1e2010-12-08 22:44:24 -060095 nr_dirty,
96 nr_io,
97 nr_more_io,
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050098 nr_dirty_time,
Tejun Heo44522262015-05-22 17:13:26 -040099 !list_empty(&bdi->bdi_list), bdi->wb.state);
Miklos Szeredi76f14182008-04-30 00:54:36 -0700100#undef K
101
102 return 0;
103}
104
105static int bdi_debug_stats_open(struct inode *inode, struct file *file)
106{
107 return single_open(file, bdi_debug_stats_show, inode->i_private);
108}
109
110static const struct file_operations bdi_debug_stats_fops = {
111 .open = bdi_debug_stats_open,
112 .read = seq_read,
113 .llseek = seq_lseek,
114 .release = single_release,
115};
116
117static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
118{
119 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
120 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
121 bdi, &bdi_debug_stats_fops);
122}
123
124static void bdi_debug_unregister(struct backing_dev_info *bdi)
125{
126 debugfs_remove(bdi->debug_stats);
127 debugfs_remove(bdi->debug_dir);
128}
129#else
130static inline void bdi_debug_init(void)
131{
132}
133static inline void bdi_debug_register(struct backing_dev_info *bdi,
134 const char *name)
135{
136}
137static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
138{
139}
140#endif
141
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700142static ssize_t read_ahead_kb_store(struct device *dev,
143 struct device_attribute *attr,
144 const char *buf, size_t count)
145{
146 struct backing_dev_info *bdi = dev_get_drvdata(dev);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700147 unsigned long read_ahead_kb;
Namjae Jeon7034ed12012-08-25 16:57:27 +0800148 ssize_t ret;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700149
Namjae Jeon7034ed12012-08-25 16:57:27 +0800150 ret = kstrtoul(buf, 10, &read_ahead_kb);
151 if (ret < 0)
152 return ret;
153
154 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
155
156 return count;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700157}
158
159#define K(pages) ((pages) << (PAGE_SHIFT - 10))
160
161#define BDI_SHOW(name, expr) \
162static ssize_t name##_show(struct device *dev, \
163 struct device_attribute *attr, char *page) \
164{ \
165 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
166 \
167 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700168} \
169static DEVICE_ATTR_RW(name);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700170
171BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
172
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700173static ssize_t min_ratio_store(struct device *dev,
174 struct device_attribute *attr, const char *buf, size_t count)
175{
176 struct backing_dev_info *bdi = dev_get_drvdata(dev);
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700177 unsigned int ratio;
Namjae Jeon7034ed12012-08-25 16:57:27 +0800178 ssize_t ret;
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700179
Namjae Jeon7034ed12012-08-25 16:57:27 +0800180 ret = kstrtouint(buf, 10, &ratio);
181 if (ret < 0)
182 return ret;
183
184 ret = bdi_set_min_ratio(bdi, ratio);
185 if (!ret)
186 ret = count;
187
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700188 return ret;
189}
190BDI_SHOW(min_ratio, bdi->min_ratio)
191
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700192static ssize_t max_ratio_store(struct device *dev,
193 struct device_attribute *attr, const char *buf, size_t count)
194{
195 struct backing_dev_info *bdi = dev_get_drvdata(dev);
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700196 unsigned int ratio;
Namjae Jeon7034ed12012-08-25 16:57:27 +0800197 ssize_t ret;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700198
Namjae Jeon7034ed12012-08-25 16:57:27 +0800199 ret = kstrtouint(buf, 10, &ratio);
200 if (ret < 0)
201 return ret;
202
203 ret = bdi_set_max_ratio(bdi, ratio);
204 if (!ret)
205 ret = count;
206
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700207 return ret;
208}
209BDI_SHOW(max_ratio, bdi->max_ratio)
210
Darrick J. Wong7d311cd2013-02-21 16:42:48 -0800211static ssize_t stable_pages_required_show(struct device *dev,
212 struct device_attribute *attr,
213 char *page)
214{
215 struct backing_dev_info *bdi = dev_get_drvdata(dev);
216
217 return snprintf(page, PAGE_SIZE-1, "%d\n",
218 bdi_cap_stable_pages_required(bdi) ? 1 : 0);
219}
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700220static DEVICE_ATTR_RO(stable_pages_required);
Darrick J. Wong7d311cd2013-02-21 16:42:48 -0800221
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700222static struct attribute *bdi_dev_attrs[] = {
223 &dev_attr_read_ahead_kb.attr,
224 &dev_attr_min_ratio.attr,
225 &dev_attr_max_ratio.attr,
226 &dev_attr_stable_pages_required.attr,
227 NULL,
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700228};
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700229ATTRIBUTE_GROUPS(bdi_dev);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700230
231static __init int bdi_class_init(void)
232{
233 bdi_class = class_create(THIS_MODULE, "bdi");
Anton Blanchard14421452010-04-02 09:46:55 +0200234 if (IS_ERR(bdi_class))
235 return PTR_ERR(bdi_class);
236
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700237 bdi_class->dev_groups = bdi_dev_groups;
Miklos Szeredi76f14182008-04-30 00:54:36 -0700238 bdi_debug_init();
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700239 return 0;
240}
Miklos Szeredi76f14182008-04-30 00:54:36 -0700241postcore_initcall(bdi_class_init);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700242
Jens Axboe26160152009-03-17 09:35:06 +0100243static int __init default_bdi_init(void)
244{
245 int err;
246
Tejun Heo839a8e82013-04-01 19:08:06 -0700247 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
Tejun Heob5c872d2013-04-01 19:08:06 -0700248 WQ_UNBOUND | WQ_SYSFS, 0);
Tejun Heo839a8e82013-04-01 19:08:06 -0700249 if (!bdi_wq)
250 return -ENOMEM;
251
Jan Kara976e48f2010-09-21 11:48:55 +0200252 err = bdi_init(&noop_backing_dev_info);
Jens Axboe26160152009-03-17 09:35:06 +0100253
254 return err;
255}
256subsys_initcall(default_bdi_init);
257
Jens Axboe03ba3782009-09-09 09:08:54 +0200258int bdi_has_dirty_io(struct backing_dev_info *bdi)
259{
260 return wb_has_dirty_io(&bdi->wb);
261}
262
Artem Bityutskiy64677162010-07-25 14:29:22 +0300263/*
Tejun Heof0054bb2015-05-22 17:13:30 -0400264 * This function is used when the first inode for this wb is marked dirty. It
Artem Bityutskiy64677162010-07-25 14:29:22 +0300265 * wakes-up the corresponding bdi thread which should then take care of the
266 * periodic background write-out of dirty inodes. Since the write-out would
267 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
268 * set up a timer which wakes the bdi thread up later.
269 *
270 * Note, we wouldn't bother setting up the timer, but this function is on the
271 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
272 * by delaying the wake-up.
Derek Basehore6ca738d2014-04-03 14:46:22 -0700273 *
274 * We have to be careful not to postpone flush work if it is scheduled for
275 * earlier. Thus we use queue_delayed_work().
Artem Bityutskiy64677162010-07-25 14:29:22 +0300276 */
Tejun Heof0054bb2015-05-22 17:13:30 -0400277void wb_wakeup_delayed(struct bdi_writeback *wb)
Artem Bityutskiy64677162010-07-25 14:29:22 +0300278{
279 unsigned long timeout;
280
281 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
Tejun Heof0054bb2015-05-22 17:13:30 -0400282 spin_lock_bh(&wb->work_lock);
283 if (test_bit(WB_registered, &wb->state))
284 queue_delayed_work(bdi_wq, &wb->dwork, timeout);
285 spin_unlock_bh(&wb->work_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200286}
287
Jens Axboecfc4ba52009-09-14 13:12:40 +0200288/*
Tejun Heoa88a3412015-05-22 17:13:28 -0400289 * Initial write bandwidth: 100 MB/s
290 */
291#define INIT_BW (100 << (20 - PAGE_SHIFT))
292
Tejun Heof0054bb2015-05-22 17:13:30 -0400293static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi)
Artem Bityutskiy64677162010-07-25 14:29:22 +0300294{
Tejun Heo93f78d82015-05-22 17:13:27 -0400295 int i, err;
296
Artem Bityutskiy64677162010-07-25 14:29:22 +0300297 memset(wb, 0, sizeof(*wb));
298
299 wb->bdi = bdi;
300 wb->last_old_flush = jiffies;
301 INIT_LIST_HEAD(&wb->b_dirty);
302 INIT_LIST_HEAD(&wb->b_io);
303 INIT_LIST_HEAD(&wb->b_more_io);
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500304 INIT_LIST_HEAD(&wb->b_dirty_time);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600305 spin_lock_init(&wb->list_lock);
Tejun Heo93f78d82015-05-22 17:13:27 -0400306
Tejun Heoa88a3412015-05-22 17:13:28 -0400307 wb->bw_time_stamp = jiffies;
308 wb->balanced_dirty_ratelimit = INIT_BW;
309 wb->dirty_ratelimit = INIT_BW;
310 wb->write_bandwidth = INIT_BW;
311 wb->avg_write_bandwidth = INIT_BW;
312
Tejun Heof0054bb2015-05-22 17:13:30 -0400313 spin_lock_init(&wb->work_lock);
314 INIT_LIST_HEAD(&wb->work_list);
315 INIT_DELAYED_WORK(&wb->dwork, wb_workfn);
316
Tejun Heoa88a3412015-05-22 17:13:28 -0400317 err = fprop_local_init_percpu(&wb->completions, GFP_KERNEL);
318 if (err)
319 return err;
320
Tejun Heo93f78d82015-05-22 17:13:27 -0400321 for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
322 err = percpu_counter_init(&wb->stat[i], 0, GFP_KERNEL);
323 if (err) {
324 while (--i)
325 percpu_counter_destroy(&wb->stat[i]);
Tejun Heoa88a3412015-05-22 17:13:28 -0400326 fprop_local_destroy_percpu(&wb->completions);
Tejun Heo93f78d82015-05-22 17:13:27 -0400327 return err;
328 }
329 }
330
331 return 0;
332}
333
Tejun Heo46100072015-05-22 17:13:31 -0400334/*
335 * Remove bdi from the global list and shutdown any threads we have running
336 */
337static void wb_shutdown(struct bdi_writeback *wb)
338{
339 /* Make sure nobody queues further work */
340 spin_lock_bh(&wb->work_lock);
341 if (!test_and_clear_bit(WB_registered, &wb->state)) {
342 spin_unlock_bh(&wb->work_lock);
343 return;
344 }
345 spin_unlock_bh(&wb->work_lock);
346
347 /*
348 * Drain work list and shutdown the delayed_work. !WB_registered
349 * tells wb_workfn() that @wb is dying and its work_list needs to
350 * be drained no matter what.
351 */
352 mod_delayed_work(bdi_wq, &wb->dwork, 0);
353 flush_delayed_work(&wb->dwork);
354 WARN_ON(!list_empty(&wb->work_list));
355}
356
Tejun Heof0054bb2015-05-22 17:13:30 -0400357static void wb_exit(struct bdi_writeback *wb)
Tejun Heo93f78d82015-05-22 17:13:27 -0400358{
359 int i;
360
361 WARN_ON(delayed_work_pending(&wb->dwork));
362
363 for (i = 0; i < NR_WB_STAT_ITEMS; i++)
364 percpu_counter_destroy(&wb->stat[i]);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300365
Tejun Heoa88a3412015-05-22 17:13:28 -0400366 fprop_local_destroy_percpu(&wb->completions);
367}
Wu Fengguange98be2d2010-08-29 11:22:30 -0600368
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700369int bdi_init(struct backing_dev_info *bdi)
370{
Tejun Heo93f78d82015-05-22 17:13:27 -0400371 int err;
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700372
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700373 bdi->dev = NULL;
374
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700375 bdi->min_ratio = 0;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700376 bdi->max_ratio = 100;
Jan Karaeb608e32012-05-24 18:59:11 +0200377 bdi->max_prop_frac = FPROP_FRAC_BASE;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200378 INIT_LIST_HEAD(&bdi->bdi_list);
Jens Axboe03ba3782009-09-09 09:08:54 +0200379
Tejun Heof0054bb2015-05-22 17:13:30 -0400380 err = wb_init(&bdi->wb, bdi);
Tejun Heo93f78d82015-05-22 17:13:27 -0400381 if (err)
382 return err;
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700383
Tejun Heo93f78d82015-05-22 17:13:27 -0400384 return 0;
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700385}
386EXPORT_SYMBOL(bdi_init);
387
Tejun Heo46100072015-05-22 17:13:31 -0400388int bdi_register(struct backing_dev_info *bdi, struct device *parent,
389 const char *fmt, ...)
390{
391 va_list args;
392 struct device *dev;
393
394 if (bdi->dev) /* The driver needs to use separate queues per device */
395 return 0;
396
397 va_start(args, fmt);
398 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
399 va_end(args);
400 if (IS_ERR(dev))
401 return PTR_ERR(dev);
402
403 bdi->dev = dev;
404
405 bdi_debug_register(bdi, dev_name(dev));
406 set_bit(WB_registered, &bdi->wb.state);
407
408 spin_lock_bh(&bdi_lock);
409 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
410 spin_unlock_bh(&bdi_lock);
411
412 trace_writeback_bdi_register(bdi);
413 return 0;
414}
415EXPORT_SYMBOL(bdi_register);
416
417int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
418{
419 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
420}
421EXPORT_SYMBOL(bdi_register_dev);
422
423/*
424 * Remove bdi from bdi_list, and ensure that it is no longer visible
425 */
426static void bdi_remove_from_list(struct backing_dev_info *bdi)
427{
428 spin_lock_bh(&bdi_lock);
429 list_del_rcu(&bdi->bdi_list);
430 spin_unlock_bh(&bdi_lock);
431
432 synchronize_rcu_expedited();
433}
434
435/*
436 * Called when the device behind @bdi has been removed or ejected.
437 *
438 * We can't really do much here except for reducing the dirty ratio at
439 * the moment. In the future we should be able to set a flag so that
440 * the filesystem can handle errors at mark_inode_dirty time instead
441 * of only at writeback time.
442 */
443void bdi_unregister(struct backing_dev_info *bdi)
444{
445 if (WARN_ON_ONCE(!bdi->dev))
446 return;
447
448 bdi_set_min_ratio(bdi, 0);
449}
450EXPORT_SYMBOL(bdi_unregister);
451
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700452void bdi_destroy(struct backing_dev_info *bdi)
453{
Tejun Heof0054bb2015-05-22 17:13:30 -0400454 /* make sure nobody finds us on the bdi_list anymore */
455 bdi_remove_from_list(bdi);
456 wb_shutdown(&bdi->wb);
Rabin Vincent7a401a92011-11-11 13:29:04 +0100457
Christoph Hellwigc4db59d2015-01-20 14:05:00 -0700458 if (bdi->dev) {
459 bdi_debug_unregister(bdi);
460 device_unregister(bdi->dev);
461 bdi->dev = NULL;
462 }
463
Tejun Heof0054bb2015-05-22 17:13:30 -0400464 wb_exit(&bdi->wb);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700465}
466EXPORT_SYMBOL(bdi_destroy);
467
Jens Axboec3c53202010-04-22 11:37:01 +0200468/*
469 * For use from filesystems to quickly init and register a bdi associated
470 * with dirty writeback
471 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100472int bdi_setup_and_register(struct backing_dev_info *bdi, char *name)
Jens Axboec3c53202010-04-22 11:37:01 +0200473{
Jens Axboec3c53202010-04-22 11:37:01 +0200474 int err;
475
476 bdi->name = name;
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100477 bdi->capabilities = 0;
Jens Axboec3c53202010-04-22 11:37:01 +0200478 err = bdi_init(bdi);
479 if (err)
480 return err;
481
Kees Cook02aa2a32013-07-03 15:04:56 -0700482 err = bdi_register(bdi, NULL, "%.28s-%ld", name,
483 atomic_long_inc_return(&bdi_seq));
Jens Axboec3c53202010-04-22 11:37:01 +0200484 if (err) {
485 bdi_destroy(bdi);
486 return err;
487 }
488
489 return 0;
490}
491EXPORT_SYMBOL(bdi_setup_and_register);
492
Andrew Morton3fcfab12006-10-19 23:28:16 -0700493static wait_queue_head_t congestion_wqh[2] = {
494 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
495 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
496 };
Mel Gorman0e093d992010-10-26 14:21:45 -0700497static atomic_t nr_bdi_congested[2];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700498
Jens Axboe1faa16d2009-04-06 14:48:01 +0200499void clear_bdi_congested(struct backing_dev_info *bdi, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700500{
Tejun Heo44522262015-05-22 17:13:26 -0400501 enum wb_state bit;
Jens Axboe1faa16d2009-04-06 14:48:01 +0200502 wait_queue_head_t *wqh = &congestion_wqh[sync];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700503
Tejun Heo44522262015-05-22 17:13:26 -0400504 bit = sync ? WB_sync_congested : WB_async_congested;
505 if (test_and_clear_bit(bit, &bdi->wb.state))
Mel Gorman0e093d992010-10-26 14:21:45 -0700506 atomic_dec(&nr_bdi_congested[sync]);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100507 smp_mb__after_atomic();
Andrew Morton3fcfab12006-10-19 23:28:16 -0700508 if (waitqueue_active(wqh))
509 wake_up(wqh);
510}
511EXPORT_SYMBOL(clear_bdi_congested);
512
Jens Axboe1faa16d2009-04-06 14:48:01 +0200513void set_bdi_congested(struct backing_dev_info *bdi, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700514{
Tejun Heo44522262015-05-22 17:13:26 -0400515 enum wb_state bit;
Andrew Morton3fcfab12006-10-19 23:28:16 -0700516
Tejun Heo44522262015-05-22 17:13:26 -0400517 bit = sync ? WB_sync_congested : WB_async_congested;
518 if (!test_and_set_bit(bit, &bdi->wb.state))
Mel Gorman0e093d992010-10-26 14:21:45 -0700519 atomic_inc(&nr_bdi_congested[sync]);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700520}
521EXPORT_SYMBOL(set_bdi_congested);
522
523/**
524 * congestion_wait - wait for a backing_dev to become uncongested
Jens Axboe8aa7e842009-07-09 14:52:32 +0200525 * @sync: SYNC or ASYNC IO
Andrew Morton3fcfab12006-10-19 23:28:16 -0700526 * @timeout: timeout in jiffies
527 *
528 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
529 * write congestion. If no backing_devs are congested then just wait for the
530 * next write to be completed.
531 */
Jens Axboe8aa7e842009-07-09 14:52:32 +0200532long congestion_wait(int sync, long timeout)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700533{
534 long ret;
Mel Gorman52bb9192010-10-26 14:21:41 -0700535 unsigned long start = jiffies;
Andrew Morton3fcfab12006-10-19 23:28:16 -0700536 DEFINE_WAIT(wait);
Jens Axboe8aa7e842009-07-09 14:52:32 +0200537 wait_queue_head_t *wqh = &congestion_wqh[sync];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700538
539 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
540 ret = io_schedule_timeout(timeout);
541 finish_wait(wqh, &wait);
Mel Gorman52bb9192010-10-26 14:21:41 -0700542
543 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
544 jiffies_to_usecs(jiffies - start));
545
Andrew Morton3fcfab12006-10-19 23:28:16 -0700546 return ret;
547}
548EXPORT_SYMBOL(congestion_wait);
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700549
Mel Gorman0e093d992010-10-26 14:21:45 -0700550/**
551 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a zone to complete writes
552 * @zone: A zone to check if it is heavily congested
553 * @sync: SYNC or ASYNC IO
554 * @timeout: timeout in jiffies
555 *
556 * In the event of a congested backing_dev (any backing_dev) and the given
557 * @zone has experienced recent congestion, this waits for up to @timeout
558 * jiffies for either a BDI to exit congestion of the given @sync queue
559 * or a write to complete.
560 *
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300561 * In the absence of zone congestion, cond_resched() is called to yield
Mel Gorman0e093d992010-10-26 14:21:45 -0700562 * the processor if necessary but otherwise does not sleep.
563 *
564 * The return value is 0 if the sleep is for the full timeout. Otherwise,
565 * it is the number of jiffies that were still remaining when the function
566 * returned. return_value == timeout implies the function did not sleep.
567 */
568long wait_iff_congested(struct zone *zone, int sync, long timeout)
569{
570 long ret;
571 unsigned long start = jiffies;
572 DEFINE_WAIT(wait);
573 wait_queue_head_t *wqh = &congestion_wqh[sync];
574
575 /*
576 * If there is no congestion, or heavy congestion is not being
577 * encountered in the current zone, yield if necessary instead
578 * of sleeping on the congestion queue
579 */
580 if (atomic_read(&nr_bdi_congested[sync]) == 0 ||
Johannes Weiner57054652014-10-09 15:28:17 -0700581 !test_bit(ZONE_CONGESTED, &zone->flags)) {
Mel Gorman0e093d992010-10-26 14:21:45 -0700582 cond_resched();
583
584 /* In case we scheduled, work out time remaining */
585 ret = timeout - (jiffies - start);
586 if (ret < 0)
587 ret = 0;
588
589 goto out;
590 }
591
592 /* Sleep until uncongested or a write happens */
593 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
594 ret = io_schedule_timeout(timeout);
595 finish_wait(wqh, &wait);
596
597out:
598 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
599 jiffies_to_usecs(jiffies - start));
600
601 return ret;
602}
603EXPORT_SYMBOL(wait_iff_congested);
Wanpeng Li3965c9a2012-07-31 16:41:52 -0700604
605int pdflush_proc_obsolete(struct ctl_table *table, int write,
606 void __user *buffer, size_t *lenp, loff_t *ppos)
607{
608 char kbuf[] = "0\n";
609
Chen Gang4c3bffc2013-09-11 14:22:44 -0700610 if (*ppos || *lenp < sizeof(kbuf)) {
Wanpeng Li3965c9a2012-07-31 16:41:52 -0700611 *lenp = 0;
612 return 0;
613 }
614
615 if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
616 return -EFAULT;
617 printk_once(KERN_WARNING "%s exported in /proc is scheduled for removal\n",
618 table->procname);
619
620 *lenp = 2;
621 *ppos += *lenp;
622 return 2;
623}