blob: 62ca907e3aaf3671e9120167ccce4029338d0c92 [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};
Tejun Heoa212b102015-05-22 17:13:33 -040021EXPORT_SYMBOL_GPL(noop_backing_dev_info);
Jörn Engel5129a462010-04-25 08:54:42 +020022
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070023static struct class *bdi_class;
Jens Axboecfc4ba52009-09-14 13:12:40 +020024
25/*
Tejun Heo181387d2013-04-01 19:08:06 -070026 * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
Jens Axboecfc4ba52009-09-14 13:12:40 +020027 * locking.
28 */
Jens Axboe03ba3782009-09-09 09:08:54 +020029DEFINE_SPINLOCK(bdi_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +020030LIST_HEAD(bdi_list);
Jens Axboe03ba3782009-09-09 09:08:54 +020031
Tejun Heo839a8e82013-04-01 19:08:06 -070032/* bdi_wq serves all asynchronous writeback tasks */
33struct workqueue_struct *bdi_wq;
34
Miklos Szeredi76f14182008-04-30 00:54:36 -070035#ifdef CONFIG_DEBUG_FS
36#include <linux/debugfs.h>
37#include <linux/seq_file.h>
38
39static struct dentry *bdi_debug_root;
40
41static void bdi_debug_init(void)
42{
43 bdi_debug_root = debugfs_create_dir("bdi", NULL);
44}
45
46static int bdi_debug_stats_show(struct seq_file *m, void *v)
47{
48 struct backing_dev_info *bdi = m->private;
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020049 struct bdi_writeback *wb = &bdi->wb;
David Rientjes364aeb22009-01-06 14:39:29 -080050 unsigned long background_thresh;
51 unsigned long dirty_thresh;
Tejun Heo0d960a32015-05-22 18:23:19 -040052 unsigned long wb_thresh;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050053 unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time;
Jens Axboef09b00d2009-05-25 09:08:21 +020054 struct inode *inode;
55
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050056 nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0;
Christoph Hellwigf758eea2011-04-21 18:19:44 -060057 spin_lock(&wb->list_lock);
Dave Chinnerc7f54082015-03-04 14:07:22 -050058 list_for_each_entry(inode, &wb->b_dirty, i_io_list)
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020059 nr_dirty++;
Dave Chinnerc7f54082015-03-04 14:07:22 -050060 list_for_each_entry(inode, &wb->b_io, i_io_list)
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020061 nr_io++;
Dave Chinnerc7f54082015-03-04 14:07:22 -050062 list_for_each_entry(inode, &wb->b_more_io, i_io_list)
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020063 nr_more_io++;
Dave Chinnerc7f54082015-03-04 14:07:22 -050064 list_for_each_entry(inode, &wb->b_dirty_time, i_io_list)
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050065 if (inode->i_state & I_DIRTY_TIME)
66 nr_dirty_time++;
Christoph Hellwigf758eea2011-04-21 18:19:44 -060067 spin_unlock(&wb->list_lock);
Miklos Szeredi76f14182008-04-30 00:54:36 -070068
Wu Fengguang16c40422010-08-11 14:17:39 -070069 global_dirty_limits(&background_thresh, &dirty_thresh);
Tejun Heo0d960a32015-05-22 18:23:19 -040070 wb_thresh = wb_calc_thresh(wb, dirty_thresh);
Miklos Szeredi76f14182008-04-30 00:54:36 -070071
72#define K(x) ((x) << (PAGE_SHIFT - 10))
73 seq_printf(m,
Wu Fengguang00821b02010-08-29 11:28:45 -060074 "BdiWriteback: %10lu kB\n"
75 "BdiReclaimable: %10lu kB\n"
76 "BdiDirtyThresh: %10lu kB\n"
77 "DirtyThresh: %10lu kB\n"
78 "BackgroundThresh: %10lu kB\n"
Wu Fengguangc8e28ce2011-01-23 10:07:47 -060079 "BdiDirtied: %10lu kB\n"
Wu Fengguang00821b02010-08-29 11:28:45 -060080 "BdiWritten: %10lu kB\n"
81 "BdiWriteBandwidth: %10lu kBps\n"
82 "b_dirty: %10lu\n"
83 "b_io: %10lu\n"
84 "b_more_io: %10lu\n"
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050085 "b_dirty_time: %10lu\n"
Wu Fengguang00821b02010-08-29 11:28:45 -060086 "bdi_list: %10u\n"
87 "state: %10lx\n",
Tejun Heo93f78d82015-05-22 17:13:27 -040088 (unsigned long) K(wb_stat(wb, WB_WRITEBACK)),
89 (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)),
Tejun Heo0d960a32015-05-22 18:23:19 -040090 K(wb_thresh),
Jan Karaf7d2b1e2010-12-08 22:44:24 -060091 K(dirty_thresh),
92 K(background_thresh),
Tejun Heo93f78d82015-05-22 17:13:27 -040093 (unsigned long) K(wb_stat(wb, WB_DIRTIED)),
94 (unsigned long) K(wb_stat(wb, WB_WRITTEN)),
Tejun Heoa88a3412015-05-22 17:13:28 -040095 (unsigned long) K(wb->write_bandwidth),
Jan Karaf7d2b1e2010-12-08 22:44:24 -060096 nr_dirty,
97 nr_io,
98 nr_more_io,
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050099 nr_dirty_time,
Tejun Heo44522262015-05-22 17:13:26 -0400100 !list_empty(&bdi->bdi_list), bdi->wb.state);
Miklos Szeredi76f14182008-04-30 00:54:36 -0700101#undef K
102
103 return 0;
104}
105
106static int bdi_debug_stats_open(struct inode *inode, struct file *file)
107{
108 return single_open(file, bdi_debug_stats_show, inode->i_private);
109}
110
111static const struct file_operations bdi_debug_stats_fops = {
112 .open = bdi_debug_stats_open,
113 .read = seq_read,
114 .llseek = seq_lseek,
115 .release = single_release,
116};
117
118static void bdi_debug_register(struct backing_dev_info *bdi, const char *name)
119{
120 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
121 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
122 bdi, &bdi_debug_stats_fops);
123}
124
125static void bdi_debug_unregister(struct backing_dev_info *bdi)
126{
127 debugfs_remove(bdi->debug_stats);
128 debugfs_remove(bdi->debug_dir);
129}
130#else
131static inline void bdi_debug_init(void)
132{
133}
134static inline void bdi_debug_register(struct backing_dev_info *bdi,
135 const char *name)
136{
137}
138static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
139{
140}
141#endif
142
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700143static ssize_t read_ahead_kb_store(struct device *dev,
144 struct device_attribute *attr,
145 const char *buf, size_t count)
146{
147 struct backing_dev_info *bdi = dev_get_drvdata(dev);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700148 unsigned long read_ahead_kb;
Namjae Jeon7034ed12012-08-25 16:57:27 +0800149 ssize_t ret;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700150
Namjae Jeon7034ed12012-08-25 16:57:27 +0800151 ret = kstrtoul(buf, 10, &read_ahead_kb);
152 if (ret < 0)
153 return ret;
154
155 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
156
157 return count;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700158}
159
160#define K(pages) ((pages) << (PAGE_SHIFT - 10))
161
162#define BDI_SHOW(name, expr) \
163static ssize_t name##_show(struct device *dev, \
164 struct device_attribute *attr, char *page) \
165{ \
166 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
167 \
168 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700169} \
170static DEVICE_ATTR_RW(name);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700171
172BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
173
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700174static ssize_t min_ratio_store(struct device *dev,
175 struct device_attribute *attr, const char *buf, size_t count)
176{
177 struct backing_dev_info *bdi = dev_get_drvdata(dev);
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700178 unsigned int ratio;
Namjae Jeon7034ed12012-08-25 16:57:27 +0800179 ssize_t ret;
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700180
Namjae Jeon7034ed12012-08-25 16:57:27 +0800181 ret = kstrtouint(buf, 10, &ratio);
182 if (ret < 0)
183 return ret;
184
185 ret = bdi_set_min_ratio(bdi, ratio);
186 if (!ret)
187 ret = count;
188
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700189 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);
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700197 unsigned int ratio;
Namjae Jeon7034ed12012-08-25 16:57:27 +0800198 ssize_t ret;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700199
Namjae Jeon7034ed12012-08-25 16:57:27 +0800200 ret = kstrtouint(buf, 10, &ratio);
201 if (ret < 0)
202 return ret;
203
204 ret = bdi_set_max_ratio(bdi, ratio);
205 if (!ret)
206 ret = count;
207
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700208 return ret;
209}
210BDI_SHOW(max_ratio, bdi->max_ratio)
211
Darrick J. Wong7d311cd2013-02-21 16:42:48 -0800212static ssize_t stable_pages_required_show(struct device *dev,
213 struct device_attribute *attr,
214 char *page)
215{
216 struct backing_dev_info *bdi = dev_get_drvdata(dev);
217
218 return snprintf(page, PAGE_SIZE-1, "%d\n",
219 bdi_cap_stable_pages_required(bdi) ? 1 : 0);
220}
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700221static DEVICE_ATTR_RO(stable_pages_required);
Darrick J. Wong7d311cd2013-02-21 16:42:48 -0800222
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700223static struct attribute *bdi_dev_attrs[] = {
224 &dev_attr_read_ahead_kb.attr,
225 &dev_attr_min_ratio.attr,
226 &dev_attr_max_ratio.attr,
227 &dev_attr_stable_pages_required.attr,
228 NULL,
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700229};
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700230ATTRIBUTE_GROUPS(bdi_dev);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700231
232static __init int bdi_class_init(void)
233{
234 bdi_class = class_create(THIS_MODULE, "bdi");
Anton Blanchard14421452010-04-02 09:46:55 +0200235 if (IS_ERR(bdi_class))
236 return PTR_ERR(bdi_class);
237
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700238 bdi_class->dev_groups = bdi_dev_groups;
Miklos Szeredi76f14182008-04-30 00:54:36 -0700239 bdi_debug_init();
Jan Kara33c17052017-02-02 15:56:51 +0100240
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700241 return 0;
242}
Miklos Szeredi76f14182008-04-30 00:54:36 -0700243postcore_initcall(bdi_class_init);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700244
Jens Axboe26160152009-03-17 09:35:06 +0100245static int __init default_bdi_init(void)
246{
247 int err;
248
Tejun Heo839a8e82013-04-01 19:08:06 -0700249 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
Tejun Heob5c872d2013-04-01 19:08:06 -0700250 WQ_UNBOUND | WQ_SYSFS, 0);
Tejun Heo839a8e82013-04-01 19:08:06 -0700251 if (!bdi_wq)
252 return -ENOMEM;
253
Jan Kara976e48f2010-09-21 11:48:55 +0200254 err = bdi_init(&noop_backing_dev_info);
Jens Axboe26160152009-03-17 09:35:06 +0100255
256 return err;
257}
258subsys_initcall(default_bdi_init);
259
Artem Bityutskiy64677162010-07-25 14:29:22 +0300260/*
Tejun Heof0054bb2015-05-22 17:13:30 -0400261 * This function is used when the first inode for this wb is marked dirty. It
Artem Bityutskiy64677162010-07-25 14:29:22 +0300262 * wakes-up the corresponding bdi thread which should then take care of the
263 * periodic background write-out of dirty inodes. Since the write-out would
264 * starts only 'dirty_writeback_interval' centisecs from now anyway, we just
265 * set up a timer which wakes the bdi thread up later.
266 *
267 * Note, we wouldn't bother setting up the timer, but this function is on the
268 * fast-path (used by '__mark_inode_dirty()'), so we save few context switches
269 * by delaying the wake-up.
Derek Basehore6ca738d2014-04-03 14:46:22 -0700270 *
271 * We have to be careful not to postpone flush work if it is scheduled for
272 * earlier. Thus we use queue_delayed_work().
Artem Bityutskiy64677162010-07-25 14:29:22 +0300273 */
Tejun Heof0054bb2015-05-22 17:13:30 -0400274void wb_wakeup_delayed(struct bdi_writeback *wb)
Artem Bityutskiy64677162010-07-25 14:29:22 +0300275{
276 unsigned long timeout;
277
278 timeout = msecs_to_jiffies(dirty_writeback_interval * 10);
Tejun Heof0054bb2015-05-22 17:13:30 -0400279 spin_lock_bh(&wb->work_lock);
280 if (test_bit(WB_registered, &wb->state))
281 queue_delayed_work(bdi_wq, &wb->dwork, timeout);
282 spin_unlock_bh(&wb->work_lock);
Jens Axboe03ba3782009-09-09 09:08:54 +0200283}
284
Jens Axboecfc4ba52009-09-14 13:12:40 +0200285/*
Tejun Heoa88a3412015-05-22 17:13:28 -0400286 * Initial write bandwidth: 100 MB/s
Jens Axboecfc4ba52009-09-14 13:12:40 +0200287 */
Tejun Heoa88a3412015-05-22 17:13:28 -0400288#define INIT_BW (100 << (20 - PAGE_SHIFT))
Jens Axboecfc4ba52009-09-14 13:12:40 +0200289
Tejun Heo8395cd92015-05-22 17:13:34 -0400290static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi,
Tejun Heoa13f35e2015-07-02 08:44:34 -0600291 int blkcg_id, gfp_t gfp)
Artem Bityutskiy64677162010-07-25 14:29:22 +0300292{
Tejun Heo93f78d82015-05-22 17:13:27 -0400293 int i, err;
294
Artem Bityutskiy64677162010-07-25 14:29:22 +0300295 memset(wb, 0, sizeof(*wb));
296
Jan Kara45502402017-03-23 01:36:55 +0100297 if (wb != &bdi->wb)
298 bdi_get(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300299 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 Heoa13f35e2015-07-02 08:44:34 -0600317 wb->congested = wb_congested_get_create(bdi, blkcg_id, gfp);
Jan Kara45502402017-03-23 01:36:55 +0100318 if (!wb->congested) {
319 err = -ENOMEM;
320 goto out_put_bdi;
321 }
Tejun Heoa13f35e2015-07-02 08:44:34 -0600322
Tejun Heo8395cd92015-05-22 17:13:34 -0400323 err = fprop_local_init_percpu(&wb->completions, gfp);
Tejun Heoa88a3412015-05-22 17:13:28 -0400324 if (err)
Tejun Heoa13f35e2015-07-02 08:44:34 -0600325 goto out_put_cong;
Tejun Heoa88a3412015-05-22 17:13:28 -0400326
Tejun Heo93f78d82015-05-22 17:13:27 -0400327 for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
Tejun Heo8395cd92015-05-22 17:13:34 -0400328 err = percpu_counter_init(&wb->stat[i], 0, gfp);
Tejun Heoa13f35e2015-07-02 08:44:34 -0600329 if (err)
330 goto out_destroy_stat;
Tejun Heo93f78d82015-05-22 17:13:27 -0400331 }
332
333 return 0;
Tejun Heoa13f35e2015-07-02 08:44:34 -0600334
335out_destroy_stat:
Rasmus Villemoes078c6c32016-02-11 16:13:06 -0800336 while (i--)
Tejun Heoa13f35e2015-07-02 08:44:34 -0600337 percpu_counter_destroy(&wb->stat[i]);
338 fprop_local_destroy_percpu(&wb->completions);
339out_put_cong:
340 wb_congested_put(wb->congested);
Jan Kara45502402017-03-23 01:36:55 +0100341out_put_bdi:
342 if (wb != &bdi->wb)
343 bdi_put(bdi);
Tejun Heoa13f35e2015-07-02 08:44:34 -0600344 return err;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700345}
Jens Axboe03ba3782009-09-09 09:08:54 +0200346
Jan Karacb9fad22017-03-23 01:36:56 +0100347static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb);
348
Tejun Heo46100072015-05-22 17:13:31 -0400349/*
350 * Remove bdi from the global list and shutdown any threads we have running
351 */
352static void wb_shutdown(struct bdi_writeback *wb)
353{
354 /* Make sure nobody queues further work */
355 spin_lock_bh(&wb->work_lock);
356 if (!test_and_clear_bit(WB_registered, &wb->state)) {
357 spin_unlock_bh(&wb->work_lock);
Jan Kara3d804e02017-03-23 01:36:57 +0100358 /*
359 * Wait for wb shutdown to finish if someone else is just
360 * running wb_shutdown(). Otherwise we could proceed to wb /
361 * bdi destruction before wb_shutdown() is finished.
362 */
363 wait_on_bit(&wb->state, WB_shutting_down, TASK_UNINTERRUPTIBLE);
Tejun Heo46100072015-05-22 17:13:31 -0400364 return;
365 }
Jan Kara3d804e02017-03-23 01:36:57 +0100366 set_bit(WB_shutting_down, &wb->state);
Tejun Heo46100072015-05-22 17:13:31 -0400367 spin_unlock_bh(&wb->work_lock);
368
Jan Karacb9fad22017-03-23 01:36:56 +0100369 cgwb_remove_from_bdi_list(wb);
Tejun Heo46100072015-05-22 17:13:31 -0400370 /*
371 * Drain work list and shutdown the delayed_work. !WB_registered
372 * tells wb_workfn() that @wb is dying and its work_list needs to
373 * be drained no matter what.
374 */
375 mod_delayed_work(bdi_wq, &wb->dwork, 0);
376 flush_delayed_work(&wb->dwork);
377 WARN_ON(!list_empty(&wb->work_list));
Jan Kara3d804e02017-03-23 01:36:57 +0100378 /*
379 * Make sure bit gets cleared after shutdown is finished. Matches with
380 * the barrier provided by test_and_clear_bit() above.
381 */
382 smp_wmb();
383 clear_bit(WB_shutting_down, &wb->state);
Tejun Heo46100072015-05-22 17:13:31 -0400384}
385
Tejun Heof0054bb2015-05-22 17:13:30 -0400386static void wb_exit(struct bdi_writeback *wb)
Tejun Heo93f78d82015-05-22 17:13:27 -0400387{
388 int i;
389
390 WARN_ON(delayed_work_pending(&wb->dwork));
391
392 for (i = 0; i < NR_WB_STAT_ITEMS; i++)
393 percpu_counter_destroy(&wb->stat[i]);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300394
Tejun Heoa88a3412015-05-22 17:13:28 -0400395 fprop_local_destroy_percpu(&wb->completions);
Tejun Heoa13f35e2015-07-02 08:44:34 -0600396 wb_congested_put(wb->congested);
Jan Kara45502402017-03-23 01:36:55 +0100397 if (wb != &wb->bdi->wb)
398 bdi_put(wb->bdi);
Tejun Heoa88a3412015-05-22 17:13:28 -0400399}
Wu Fengguange98be2d2010-08-29 11:22:30 -0600400
Tejun Heo52ebea72015-05-22 17:13:37 -0400401#ifdef CONFIG_CGROUP_WRITEBACK
402
403#include <linux/memcontrol.h>
404
405/*
406 * cgwb_lock protects bdi->cgwb_tree, bdi->cgwb_congested_tree,
407 * blkcg->cgwb_list, and memcg->cgwb_list. bdi->cgwb_tree is also RCU
Jan Karaa345cd52017-03-23 01:36:58 +0100408 * protected.
Tejun Heo52ebea72015-05-22 17:13:37 -0400409 */
410static DEFINE_SPINLOCK(cgwb_lock);
Tejun Heo52ebea72015-05-22 17:13:37 -0400411
412/**
413 * wb_congested_get_create - get or create a wb_congested
414 * @bdi: associated bdi
415 * @blkcg_id: ID of the associated blkcg
416 * @gfp: allocation mask
417 *
418 * Look up the wb_congested for @blkcg_id on @bdi. If missing, create one.
419 * The returned wb_congested has its reference count incremented. Returns
420 * NULL on failure.
421 */
422struct bdi_writeback_congested *
423wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
424{
425 struct bdi_writeback_congested *new_congested = NULL, *congested;
426 struct rb_node **node, *parent;
427 unsigned long flags;
Tejun Heo52ebea72015-05-22 17:13:37 -0400428retry:
429 spin_lock_irqsave(&cgwb_lock, flags);
430
431 node = &bdi->cgwb_congested_tree.rb_node;
432 parent = NULL;
433
434 while (*node != NULL) {
435 parent = *node;
436 congested = container_of(parent, struct bdi_writeback_congested,
437 rb_node);
438 if (congested->blkcg_id < blkcg_id)
439 node = &parent->rb_left;
440 else if (congested->blkcg_id > blkcg_id)
441 node = &parent->rb_right;
442 else
443 goto found;
444 }
445
446 if (new_congested) {
447 /* !found and storage for new one already allocated, insert */
448 congested = new_congested;
449 new_congested = NULL;
450 rb_link_node(&congested->rb_node, parent, node);
451 rb_insert_color(&congested->rb_node, &bdi->cgwb_congested_tree);
Tejun Heo52ebea72015-05-22 17:13:37 -0400452 goto found;
453 }
454
455 spin_unlock_irqrestore(&cgwb_lock, flags);
456
457 /* allocate storage for new one and retry */
458 new_congested = kzalloc(sizeof(*new_congested), gfp);
459 if (!new_congested)
460 return NULL;
461
462 atomic_set(&new_congested->refcnt, 0);
Jan Kara1448b922017-03-23 01:36:54 +0100463 new_congested->__bdi = bdi;
Tejun Heo52ebea72015-05-22 17:13:37 -0400464 new_congested->blkcg_id = blkcg_id;
465 goto retry;
466
467found:
468 atomic_inc(&congested->refcnt);
469 spin_unlock_irqrestore(&cgwb_lock, flags);
470 kfree(new_congested);
471 return congested;
472}
473
474/**
475 * wb_congested_put - put a wb_congested
476 * @congested: wb_congested to put
477 *
478 * Put @congested and destroy it if the refcnt reaches zero.
479 */
480void wb_congested_put(struct bdi_writeback_congested *congested)
481{
Tejun Heo52ebea72015-05-22 17:13:37 -0400482 unsigned long flags;
483
Tejun Heo52ebea72015-05-22 17:13:37 -0400484 local_irq_save(flags);
485 if (!atomic_dec_and_lock(&congested->refcnt, &cgwb_lock)) {
486 local_irq_restore(flags);
487 return;
488 }
489
Tejun Heoa20135f2015-07-01 20:53:37 -0400490 /* bdi might already have been destroyed leaving @congested unlinked */
Jan Kara1448b922017-03-23 01:36:54 +0100491 if (congested->__bdi) {
Tejun Heoa20135f2015-07-01 20:53:37 -0400492 rb_erase(&congested->rb_node,
Jan Kara1448b922017-03-23 01:36:54 +0100493 &congested->__bdi->cgwb_congested_tree);
494 congested->__bdi = NULL;
Tejun Heoa20135f2015-07-01 20:53:37 -0400495 }
496
Tejun Heo52ebea72015-05-22 17:13:37 -0400497 spin_unlock_irqrestore(&cgwb_lock, flags);
498 kfree(congested);
Tejun Heo52ebea72015-05-22 17:13:37 -0400499}
500
501static void cgwb_release_workfn(struct work_struct *work)
502{
503 struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
504 release_work);
Tejun Heo52ebea72015-05-22 17:13:37 -0400505
506 wb_shutdown(wb);
507
508 css_put(wb->memcg_css);
509 css_put(wb->blkcg_css);
Tejun Heo52ebea72015-05-22 17:13:37 -0400510
Tejun Heo841710a2015-05-22 18:23:33 -0400511 fprop_local_destroy_percpu(&wb->memcg_completions);
Tejun Heo52ebea72015-05-22 17:13:37 -0400512 percpu_ref_exit(&wb->refcnt);
513 wb_exit(wb);
514 kfree_rcu(wb, rcu);
Tejun Heo52ebea72015-05-22 17:13:37 -0400515}
516
517static void cgwb_release(struct percpu_ref *refcnt)
518{
519 struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
520 refcnt);
521 schedule_work(&wb->release_work);
522}
523
524static void cgwb_kill(struct bdi_writeback *wb)
525{
526 lockdep_assert_held(&cgwb_lock);
527
528 WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id));
529 list_del(&wb->memcg_node);
530 list_del(&wb->blkcg_node);
531 percpu_ref_kill(&wb->refcnt);
532}
533
Jan Karacb9fad22017-03-23 01:36:56 +0100534static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
535{
536 spin_lock_irq(&cgwb_lock);
537 list_del_rcu(&wb->bdi_node);
538 spin_unlock_irq(&cgwb_lock);
539}
540
Tejun Heo52ebea72015-05-22 17:13:37 -0400541static int cgwb_create(struct backing_dev_info *bdi,
542 struct cgroup_subsys_state *memcg_css, gfp_t gfp)
543{
544 struct mem_cgroup *memcg;
545 struct cgroup_subsys_state *blkcg_css;
546 struct blkcg *blkcg;
547 struct list_head *memcg_cgwb_list, *blkcg_cgwb_list;
548 struct bdi_writeback *wb;
549 unsigned long flags;
550 int ret = 0;
551
552 memcg = mem_cgroup_from_css(memcg_css);
Tejun Heoc165b3e2015-08-18 14:55:29 -0700553 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
Tejun Heo52ebea72015-05-22 17:13:37 -0400554 blkcg = css_to_blkcg(blkcg_css);
555 memcg_cgwb_list = mem_cgroup_cgwb_list(memcg);
556 blkcg_cgwb_list = &blkcg->cgwb_list;
557
558 /* look up again under lock and discard on blkcg mismatch */
559 spin_lock_irqsave(&cgwb_lock, flags);
560 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
561 if (wb && wb->blkcg_css != blkcg_css) {
562 cgwb_kill(wb);
563 wb = NULL;
564 }
565 spin_unlock_irqrestore(&cgwb_lock, flags);
566 if (wb)
567 goto out_put;
568
569 /* need to create a new one */
570 wb = kmalloc(sizeof(*wb), gfp);
571 if (!wb)
572 return -ENOMEM;
573
Tejun Heoa13f35e2015-07-02 08:44:34 -0600574 ret = wb_init(wb, bdi, blkcg_css->id, gfp);
Tejun Heo52ebea72015-05-22 17:13:37 -0400575 if (ret)
576 goto err_free;
577
578 ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
579 if (ret)
580 goto err_wb_exit;
581
Tejun Heo841710a2015-05-22 18:23:33 -0400582 ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
583 if (ret)
584 goto err_ref_exit;
585
Tejun Heo52ebea72015-05-22 17:13:37 -0400586 wb->memcg_css = memcg_css;
587 wb->blkcg_css = blkcg_css;
588 INIT_WORK(&wb->release_work, cgwb_release_workfn);
589 set_bit(WB_registered, &wb->state);
590
591 /*
592 * The root wb determines the registered state of the whole bdi and
593 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
594 * whether they're still online. Don't link @wb if any is dead.
595 * See wb_memcg_offline() and wb_blkcg_offline().
596 */
597 ret = -ENODEV;
598 spin_lock_irqsave(&cgwb_lock, flags);
599 if (test_bit(WB_registered, &bdi->wb.state) &&
600 blkcg_cgwb_list->next && memcg_cgwb_list->next) {
601 /* we might have raced another instance of this function */
602 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
603 if (!ret) {
Tejun Heob8175252015-10-02 14:47:05 -0400604 list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
Tejun Heo52ebea72015-05-22 17:13:37 -0400605 list_add(&wb->memcg_node, memcg_cgwb_list);
606 list_add(&wb->blkcg_node, blkcg_cgwb_list);
607 css_get(memcg_css);
608 css_get(blkcg_css);
609 }
610 }
611 spin_unlock_irqrestore(&cgwb_lock, flags);
612 if (ret) {
613 if (ret == -EEXIST)
614 ret = 0;
Tejun Heoa13f35e2015-07-02 08:44:34 -0600615 goto err_fprop_exit;
Tejun Heo52ebea72015-05-22 17:13:37 -0400616 }
617 goto out_put;
618
Tejun Heo841710a2015-05-22 18:23:33 -0400619err_fprop_exit:
620 fprop_local_destroy_percpu(&wb->memcg_completions);
Tejun Heo52ebea72015-05-22 17:13:37 -0400621err_ref_exit:
622 percpu_ref_exit(&wb->refcnt);
623err_wb_exit:
624 wb_exit(wb);
625err_free:
626 kfree(wb);
627out_put:
628 css_put(blkcg_css);
629 return ret;
630}
631
632/**
633 * wb_get_create - get wb for a given memcg, create if necessary
634 * @bdi: target bdi
635 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
636 * @gfp: allocation mask to use
637 *
638 * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to
639 * create one. The returned wb has its refcount incremented.
640 *
641 * This function uses css_get() on @memcg_css and thus expects its refcnt
642 * to be positive on invocation. IOW, rcu_read_lock() protection on
643 * @memcg_css isn't enough. try_get it before calling this function.
644 *
645 * A wb is keyed by its associated memcg. As blkcg implicitly enables
646 * memcg on the default hierarchy, memcg association is guaranteed to be
647 * more specific (equal or descendant to the associated blkcg) and thus can
648 * identify both the memcg and blkcg associations.
649 *
650 * Because the blkcg associated with a memcg may change as blkcg is enabled
651 * and disabled closer to root in the hierarchy, each wb keeps track of
652 * both the memcg and blkcg associated with it and verifies the blkcg on
653 * each lookup. On mismatch, the existing wb is discarded and a new one is
654 * created.
655 */
656struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
657 struct cgroup_subsys_state *memcg_css,
658 gfp_t gfp)
659{
660 struct bdi_writeback *wb;
661
Mel Gormand0164ad2015-11-06 16:28:21 -0800662 might_sleep_if(gfpflags_allow_blocking(gfp));
Tejun Heo52ebea72015-05-22 17:13:37 -0400663
664 if (!memcg_css->parent)
665 return &bdi->wb;
666
667 do {
668 rcu_read_lock();
669 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
670 if (wb) {
671 struct cgroup_subsys_state *blkcg_css;
672
673 /* see whether the blkcg association has changed */
674 blkcg_css = cgroup_get_e_css(memcg_css->cgroup,
Tejun Heoc165b3e2015-08-18 14:55:29 -0700675 &io_cgrp_subsys);
Tejun Heo52ebea72015-05-22 17:13:37 -0400676 if (unlikely(wb->blkcg_css != blkcg_css ||
677 !wb_tryget(wb)))
678 wb = NULL;
679 css_put(blkcg_css);
680 }
681 rcu_read_unlock();
682 } while (!wb && !cgwb_create(bdi, memcg_css, gfp));
683
684 return wb;
685}
686
Tejun Heoa13f35e2015-07-02 08:44:34 -0600687static int cgwb_bdi_init(struct backing_dev_info *bdi)
Tejun Heo52ebea72015-05-22 17:13:37 -0400688{
Tejun Heoa13f35e2015-07-02 08:44:34 -0600689 int ret;
690
Tejun Heo52ebea72015-05-22 17:13:37 -0400691 INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
692 bdi->cgwb_congested_tree = RB_ROOT;
Tejun Heoa13f35e2015-07-02 08:44:34 -0600693
694 ret = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL);
695 if (!ret) {
Johannes Weiner7d828602016-01-14 15:20:56 -0800696 bdi->wb.memcg_css = &root_mem_cgroup->css;
Tejun Heoa13f35e2015-07-02 08:44:34 -0600697 bdi->wb.blkcg_css = blkcg_root_css;
698 }
699 return ret;
Tejun Heo52ebea72015-05-22 17:13:37 -0400700}
701
Jan Karad8b63b62017-03-23 01:36:59 +0100702static void cgwb_bdi_unregister(struct backing_dev_info *bdi)
Tejun Heo52ebea72015-05-22 17:13:37 -0400703{
704 struct radix_tree_iter iter;
705 void **slot;
Jan Kara3d804e02017-03-23 01:36:57 +0100706 struct bdi_writeback *wb;
Tejun Heo52ebea72015-05-22 17:13:37 -0400707
708 WARN_ON(test_bit(WB_registered, &bdi->wb.state));
709
710 spin_lock_irq(&cgwb_lock);
711 radix_tree_for_each_slot(slot, &bdi->cgwb_tree, &iter, 0)
712 cgwb_kill(*slot);
Jan Kara3d804e02017-03-23 01:36:57 +0100713
714 while (!list_empty(&bdi->wb_list)) {
715 wb = list_first_entry(&bdi->wb_list, struct bdi_writeback,
716 bdi_node);
717 spin_unlock_irq(&cgwb_lock);
718 wb_shutdown(wb);
719 spin_lock_irq(&cgwb_lock);
720 }
Tejun Heo52ebea72015-05-22 17:13:37 -0400721 spin_unlock_irq(&cgwb_lock);
Tejun Heo52ebea72015-05-22 17:13:37 -0400722}
723
724/**
725 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
726 * @memcg: memcg being offlined
727 *
728 * Also prevents creation of any new wb's associated with @memcg.
729 */
730void wb_memcg_offline(struct mem_cgroup *memcg)
731{
732 LIST_HEAD(to_destroy);
733 struct list_head *memcg_cgwb_list = mem_cgroup_cgwb_list(memcg);
734 struct bdi_writeback *wb, *next;
735
736 spin_lock_irq(&cgwb_lock);
737 list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
738 cgwb_kill(wb);
739 memcg_cgwb_list->next = NULL; /* prevent new wb's */
740 spin_unlock_irq(&cgwb_lock);
741}
742
743/**
744 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
745 * @blkcg: blkcg being offlined
746 *
747 * Also prevents creation of any new wb's associated with @blkcg.
748 */
749void wb_blkcg_offline(struct blkcg *blkcg)
750{
751 LIST_HEAD(to_destroy);
752 struct bdi_writeback *wb, *next;
753
754 spin_lock_irq(&cgwb_lock);
755 list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node)
756 cgwb_kill(wb);
757 blkcg->cgwb_list.next = NULL; /* prevent new wb's */
758 spin_unlock_irq(&cgwb_lock);
759}
760
Jan Karaa8f57f72017-03-08 17:48:32 +0100761static void cgwb_bdi_exit(struct backing_dev_info *bdi)
762{
763 struct rb_node *rbn;
764
765 spin_lock_irq(&cgwb_lock);
766 while ((rbn = rb_first(&bdi->cgwb_congested_tree))) {
767 struct bdi_writeback_congested *congested =
768 rb_entry(rbn, struct bdi_writeback_congested, rb_node);
769
770 rb_erase(rbn, &bdi->cgwb_congested_tree);
Jan Kara1448b922017-03-23 01:36:54 +0100771 congested->__bdi = NULL; /* mark @congested unlinked */
Jan Karaa8f57f72017-03-08 17:48:32 +0100772 }
773 spin_unlock_irq(&cgwb_lock);
774}
775
Jan Karacb9fad22017-03-23 01:36:56 +0100776static void cgwb_bdi_register(struct backing_dev_info *bdi)
777{
778 spin_lock_irq(&cgwb_lock);
779 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
780 spin_unlock_irq(&cgwb_lock);
781}
782
Tejun Heo52ebea72015-05-22 17:13:37 -0400783#else /* CONFIG_CGROUP_WRITEBACK */
784
Tejun Heoa13f35e2015-07-02 08:44:34 -0600785static int cgwb_bdi_init(struct backing_dev_info *bdi)
786{
787 int err;
788
789 bdi->wb_congested = kzalloc(sizeof(*bdi->wb_congested), GFP_KERNEL);
790 if (!bdi->wb_congested)
791 return -ENOMEM;
792
Tejun Heo1cb3de82017-02-08 15:19:07 -0500793 atomic_set(&bdi->wb_congested->refcnt, 1);
794
Tejun Heoa13f35e2015-07-02 08:44:34 -0600795 err = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL);
796 if (err) {
Tejun Heo1cb3de82017-02-08 15:19:07 -0500797 wb_congested_put(bdi->wb_congested);
Tejun Heoa13f35e2015-07-02 08:44:34 -0600798 return err;
799 }
800 return 0;
801}
802
Jan Karad8b63b62017-03-23 01:36:59 +0100803static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { }
Jan Karaa8f57f72017-03-08 17:48:32 +0100804
805static void cgwb_bdi_exit(struct backing_dev_info *bdi)
Tejun Heo1cb3de82017-02-08 15:19:07 -0500806{
807 wb_congested_put(bdi->wb_congested);
808}
Tejun Heo52ebea72015-05-22 17:13:37 -0400809
Jan Karacb9fad22017-03-23 01:36:56 +0100810static void cgwb_bdi_register(struct backing_dev_info *bdi)
811{
812 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
813}
814
815static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
816{
817 list_del_rcu(&wb->bdi_node);
818}
819
Tejun Heo52ebea72015-05-22 17:13:37 -0400820#endif /* CONFIG_CGROUP_WRITEBACK */
821
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700822int bdi_init(struct backing_dev_info *bdi)
823{
Tejun Heob8175252015-10-02 14:47:05 -0400824 int ret;
825
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700826 bdi->dev = NULL;
827
Jan Kara33c17052017-02-02 15:56:51 +0100828 kref_init(&bdi->refcnt);
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700829 bdi->min_ratio = 0;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700830 bdi->max_ratio = 100;
Jan Karaeb608e32012-05-24 18:59:11 +0200831 bdi->max_prop_frac = FPROP_FRAC_BASE;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200832 INIT_LIST_HEAD(&bdi->bdi_list);
Tejun Heob8175252015-10-02 14:47:05 -0400833 INIT_LIST_HEAD(&bdi->wb_list);
Tejun Heocc395d72015-05-22 17:13:58 -0400834 init_waitqueue_head(&bdi->wb_waitq);
Jens Axboe03ba3782009-09-09 09:08:54 +0200835
Tejun Heob8175252015-10-02 14:47:05 -0400836 ret = cgwb_bdi_init(bdi);
837
Tejun Heob8175252015-10-02 14:47:05 -0400838 return ret;
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700839}
840EXPORT_SYMBOL(bdi_init);
841
Jan Kara33c17052017-02-02 15:56:51 +0100842struct backing_dev_info *bdi_alloc_node(gfp_t gfp_mask, int node_id)
843{
844 struct backing_dev_info *bdi;
845
846 bdi = kmalloc_node(sizeof(struct backing_dev_info),
847 gfp_mask | __GFP_ZERO, node_id);
848 if (!bdi)
849 return NULL;
850
851 if (bdi_init(bdi)) {
852 kfree(bdi);
853 return NULL;
854 }
855 return bdi;
856}
857
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700858int bdi_register(struct backing_dev_info *bdi, struct device *parent,
859 const char *fmt, ...)
860{
861 va_list args;
862 struct device *dev;
863
864 if (bdi->dev) /* The driver needs to use separate queues per device */
865 return 0;
866
867 va_start(args, fmt);
868 dev = device_create_vargs(bdi_class, parent, MKDEV(0, 0), bdi, fmt, args);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700869 va_end(args);
870 if (IS_ERR(dev))
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700871 return PTR_ERR(dev);
872
Jan Karacb9fad22017-03-23 01:36:56 +0100873 cgwb_bdi_register(bdi);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700874 bdi->dev = dev;
875
876 bdi_debug_register(bdi, dev_name(dev));
Tejun Heo46100072015-05-22 17:13:31 -0400877 set_bit(WB_registered, &bdi->wb.state);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700878
879 spin_lock_bh(&bdi_lock);
880 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
881 spin_unlock_bh(&bdi_lock);
882
883 trace_writeback_bdi_register(bdi);
884 return 0;
885}
886EXPORT_SYMBOL(bdi_register);
887
888int bdi_register_dev(struct backing_dev_info *bdi, dev_t dev)
889{
890 return bdi_register(bdi, NULL, "%u:%u", MAJOR(dev), MINOR(dev));
891}
892EXPORT_SYMBOL(bdi_register_dev);
893
Dan Williamsdf08c322016-07-31 11:15:13 -0700894int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner)
895{
896 int rc;
897
898 rc = bdi_register(bdi, NULL, "%u:%u", MAJOR(owner->devt),
899 MINOR(owner->devt));
900 if (rc)
901 return rc;
Jan Karade03e632017-03-08 17:48:31 +0100902 /* Leaking owner reference... */
903 WARN_ON(bdi->owner);
Dan Williamsdf08c322016-07-31 11:15:13 -0700904 bdi->owner = owner;
905 get_device(owner);
906 return 0;
907}
908EXPORT_SYMBOL(bdi_register_owner);
909
Andrew Morton3fcfab12006-10-19 23:28:16 -0700910/*
Tejun Heo46100072015-05-22 17:13:31 -0400911 * Remove bdi from bdi_list, and ensure that it is no longer visible
Andrew Morton3fcfab12006-10-19 23:28:16 -0700912 */
Tejun Heo46100072015-05-22 17:13:31 -0400913static void bdi_remove_from_list(struct backing_dev_info *bdi)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700914{
Tejun Heo46100072015-05-22 17:13:31 -0400915 spin_lock_bh(&bdi_lock);
916 list_del_rcu(&bdi->bdi_list);
917 spin_unlock_bh(&bdi_lock);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700918
Tejun Heo46100072015-05-22 17:13:31 -0400919 synchronize_rcu_expedited();
Andrew Morton3fcfab12006-10-19 23:28:16 -0700920}
921
Tejun Heob02176f2015-09-08 12:20:22 -0400922void bdi_unregister(struct backing_dev_info *bdi)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700923{
Tejun Heof0054bb2015-05-22 17:13:30 -0400924 /* make sure nobody finds us on the bdi_list anymore */
925 bdi_remove_from_list(bdi);
926 wb_shutdown(&bdi->wb);
Jan Karad8b63b62017-03-23 01:36:59 +0100927 cgwb_bdi_unregister(bdi);
Rabin Vincent7a401a92011-11-11 13:29:04 +0100928
Christoph Hellwigc4db59d2015-01-20 14:05:00 -0700929 if (bdi->dev) {
930 bdi_debug_unregister(bdi);
931 device_unregister(bdi->dev);
932 bdi->dev = NULL;
933 }
Dan Williamsdf08c322016-07-31 11:15:13 -0700934
935 if (bdi->owner) {
936 put_device(bdi->owner);
937 bdi->owner = NULL;
938 }
Tejun Heob02176f2015-09-08 12:20:22 -0400939}
Christoph Hellwigc4db59d2015-01-20 14:05:00 -0700940
Jan Kara33c17052017-02-02 15:56:51 +0100941static void bdi_exit(struct backing_dev_info *bdi)
Tejun Heob02176f2015-09-08 12:20:22 -0400942{
943 WARN_ON_ONCE(bdi->dev);
Tejun Heof0054bb2015-05-22 17:13:30 -0400944 wb_exit(&bdi->wb);
Jan Karaa8f57f72017-03-08 17:48:32 +0100945 cgwb_bdi_exit(bdi);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700946}
Tejun Heob02176f2015-09-08 12:20:22 -0400947
Jan Kara33c17052017-02-02 15:56:51 +0100948static void release_bdi(struct kref *ref)
949{
950 struct backing_dev_info *bdi =
951 container_of(ref, struct backing_dev_info, refcnt);
952
953 bdi_exit(bdi);
954 kfree(bdi);
955}
956
957void bdi_put(struct backing_dev_info *bdi)
958{
959 kref_put(&bdi->refcnt, release_bdi);
960}
961
Tejun Heob02176f2015-09-08 12:20:22 -0400962void bdi_destroy(struct backing_dev_info *bdi)
963{
964 bdi_unregister(bdi);
965 bdi_exit(bdi);
966}
Andrew Morton3fcfab12006-10-19 23:28:16 -0700967EXPORT_SYMBOL(bdi_destroy);
968
Jens Axboec3c53202010-04-22 11:37:01 +0200969/*
970 * For use from filesystems to quickly init and register a bdi associated
971 * with dirty writeback
972 */
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100973int bdi_setup_and_register(struct backing_dev_info *bdi, char *name)
Jens Axboec3c53202010-04-22 11:37:01 +0200974{
Jens Axboec3c53202010-04-22 11:37:01 +0200975 int err;
976
977 bdi->name = name;
Christoph Hellwigb4caecd2015-01-14 10:42:32 +0100978 bdi->capabilities = 0;
Jens Axboec3c53202010-04-22 11:37:01 +0200979 err = bdi_init(bdi);
980 if (err)
981 return err;
982
Kees Cook02aa2a32013-07-03 15:04:56 -0700983 err = bdi_register(bdi, NULL, "%.28s-%ld", name,
984 atomic_long_inc_return(&bdi_seq));
Jens Axboec3c53202010-04-22 11:37:01 +0200985 if (err) {
986 bdi_destroy(bdi);
987 return err;
988 }
989
990 return 0;
991}
992EXPORT_SYMBOL(bdi_setup_and_register);
993
Andrew Morton3fcfab12006-10-19 23:28:16 -0700994static wait_queue_head_t congestion_wqh[2] = {
995 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
996 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
997 };
Tejun Heoec8a6f22015-05-22 17:13:41 -0400998static atomic_t nr_wb_congested[2];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700999
Tejun Heoec8a6f22015-05-22 17:13:41 -04001000void clear_wb_congested(struct bdi_writeback_congested *congested, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -07001001{
Jens Axboe1faa16d2009-04-06 14:48:01 +02001002 wait_queue_head_t *wqh = &congestion_wqh[sync];
Kaixu Xiac877ef82016-03-31 13:19:41 +00001003 enum wb_congested_state bit;
Andrew Morton3fcfab12006-10-19 23:28:16 -07001004
Tejun Heo44522262015-05-22 17:13:26 -04001005 bit = sync ? WB_sync_congested : WB_async_congested;
Tejun Heoec8a6f22015-05-22 17:13:41 -04001006 if (test_and_clear_bit(bit, &congested->state))
1007 atomic_dec(&nr_wb_congested[sync]);
Peter Zijlstra4e857c52014-03-17 18:06:10 +01001008 smp_mb__after_atomic();
Andrew Morton3fcfab12006-10-19 23:28:16 -07001009 if (waitqueue_active(wqh))
1010 wake_up(wqh);
1011}
Tejun Heoec8a6f22015-05-22 17:13:41 -04001012EXPORT_SYMBOL(clear_wb_congested);
Andrew Morton3fcfab12006-10-19 23:28:16 -07001013
Tejun Heoec8a6f22015-05-22 17:13:41 -04001014void set_wb_congested(struct bdi_writeback_congested *congested, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -07001015{
Kaixu Xiac877ef82016-03-31 13:19:41 +00001016 enum wb_congested_state bit;
Andrew Morton3fcfab12006-10-19 23:28:16 -07001017
Tejun Heo44522262015-05-22 17:13:26 -04001018 bit = sync ? WB_sync_congested : WB_async_congested;
Tejun Heoec8a6f22015-05-22 17:13:41 -04001019 if (!test_and_set_bit(bit, &congested->state))
1020 atomic_inc(&nr_wb_congested[sync]);
Andrew Morton3fcfab12006-10-19 23:28:16 -07001021}
Tejun Heoec8a6f22015-05-22 17:13:41 -04001022EXPORT_SYMBOL(set_wb_congested);
Andrew Morton3fcfab12006-10-19 23:28:16 -07001023
1024/**
1025 * congestion_wait - wait for a backing_dev to become uncongested
Jens Axboe8aa7e842009-07-09 14:52:32 +02001026 * @sync: SYNC or ASYNC IO
Andrew Morton3fcfab12006-10-19 23:28:16 -07001027 * @timeout: timeout in jiffies
1028 *
1029 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
1030 * write congestion. If no backing_devs are congested then just wait for the
1031 * next write to be completed.
1032 */
Jens Axboe8aa7e842009-07-09 14:52:32 +02001033long congestion_wait(int sync, long timeout)
Andrew Morton3fcfab12006-10-19 23:28:16 -07001034{
1035 long ret;
Mel Gorman52bb9192010-10-26 14:21:41 -07001036 unsigned long start = jiffies;
Andrew Morton3fcfab12006-10-19 23:28:16 -07001037 DEFINE_WAIT(wait);
Jens Axboe8aa7e842009-07-09 14:52:32 +02001038 wait_queue_head_t *wqh = &congestion_wqh[sync];
Andrew Morton3fcfab12006-10-19 23:28:16 -07001039
1040 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1041 ret = io_schedule_timeout(timeout);
1042 finish_wait(wqh, &wait);
Mel Gorman52bb9192010-10-26 14:21:41 -07001043
1044 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
1045 jiffies_to_usecs(jiffies - start));
1046
Andrew Morton3fcfab12006-10-19 23:28:16 -07001047 return ret;
1048}
1049EXPORT_SYMBOL(congestion_wait);
1050
Mel Gorman0e093d992010-10-26 14:21:45 -07001051/**
Mel Gorman599d0c92016-07-28 15:45:31 -07001052 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a pgdat to complete writes
1053 * @pgdat: A pgdat to check if it is heavily congested
Mel Gorman0e093d992010-10-26 14:21:45 -07001054 * @sync: SYNC or ASYNC IO
1055 * @timeout: timeout in jiffies
1056 *
1057 * In the event of a congested backing_dev (any backing_dev) and the given
Mel Gorman599d0c92016-07-28 15:45:31 -07001058 * @pgdat has experienced recent congestion, this waits for up to @timeout
Mel Gorman0e093d992010-10-26 14:21:45 -07001059 * jiffies for either a BDI to exit congestion of the given @sync queue
1060 * or a write to complete.
1061 *
Mel Gorman599d0c92016-07-28 15:45:31 -07001062 * In the absence of pgdat congestion, cond_resched() is called to yield
Michal Hockoede37712016-05-20 16:57:03 -07001063 * the processor if necessary but otherwise does not sleep.
Mel Gorman0e093d992010-10-26 14:21:45 -07001064 *
1065 * The return value is 0 if the sleep is for the full timeout. Otherwise,
1066 * it is the number of jiffies that were still remaining when the function
1067 * returned. return_value == timeout implies the function did not sleep.
1068 */
Mel Gorman599d0c92016-07-28 15:45:31 -07001069long wait_iff_congested(struct pglist_data *pgdat, int sync, long timeout)
Mel Gorman0e093d992010-10-26 14:21:45 -07001070{
1071 long ret;
1072 unsigned long start = jiffies;
1073 DEFINE_WAIT(wait);
1074 wait_queue_head_t *wqh = &congestion_wqh[sync];
1075
1076 /*
1077 * If there is no congestion, or heavy congestion is not being
Mel Gorman599d0c92016-07-28 15:45:31 -07001078 * encountered in the current pgdat, yield if necessary instead
Mel Gorman0e093d992010-10-26 14:21:45 -07001079 * of sleeping on the congestion queue
1080 */
Tejun Heoec8a6f22015-05-22 17:13:41 -04001081 if (atomic_read(&nr_wb_congested[sync]) == 0 ||
Mel Gorman599d0c92016-07-28 15:45:31 -07001082 !test_bit(PGDAT_CONGESTED, &pgdat->flags)) {
Michal Hockoede37712016-05-20 16:57:03 -07001083 cond_resched();
Mel Gorman599d0c92016-07-28 15:45:31 -07001084
Mel Gorman0e093d992010-10-26 14:21:45 -07001085 /* In case we scheduled, work out time remaining */
1086 ret = timeout - (jiffies - start);
1087 if (ret < 0)
1088 ret = 0;
1089
1090 goto out;
1091 }
1092
1093 /* Sleep until uncongested or a write happens */
1094 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1095 ret = io_schedule_timeout(timeout);
1096 finish_wait(wqh, &wait);
1097
1098out:
1099 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
1100 jiffies_to_usecs(jiffies - start));
1101
1102 return ret;
1103}
1104EXPORT_SYMBOL(wait_iff_congested);
Wanpeng Li3965c9a2012-07-31 16:41:52 -07001105
1106int pdflush_proc_obsolete(struct ctl_table *table, int write,
1107 void __user *buffer, size_t *lenp, loff_t *ppos)
1108{
1109 char kbuf[] = "0\n";
1110
Chen Gang4c3bffc2013-09-11 14:22:44 -07001111 if (*ppos || *lenp < sizeof(kbuf)) {
Wanpeng Li3965c9a2012-07-31 16:41:52 -07001112 *lenp = 0;
1113 return 0;
1114 }
1115
1116 if (copy_to_user(buffer, kbuf, sizeof(kbuf)))
1117 return -EFAULT;
Joe Perches11705322016-03-17 14:19:50 -07001118 pr_warn_once("%s exported in /proc is scheduled for removal\n",
1119 table->procname);
Wanpeng Li3965c9a2012-07-31 16:41:52 -07001120
1121 *lenp = 2;
1122 *ppos += *lenp;
1123 return 2;
1124}