blob: 72e6d0c55cfad9b51b173cca10c34011b035879e [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
Jörn Engel5129a462010-04-25 08:54:42 +020015struct backing_dev_info noop_backing_dev_info = {
16 .name = "noop",
Jan Kara976e48f2010-09-21 11:48:55 +020017 .capabilities = BDI_CAP_NO_ACCT_AND_WRITEBACK,
Jörn Engel5129a462010-04-25 08:54:42 +020018};
Tejun Heoa212b102015-05-22 17:13:33 -040019EXPORT_SYMBOL_GPL(noop_backing_dev_info);
Jörn Engel5129a462010-04-25 08:54:42 +020020
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -070021static struct class *bdi_class;
Jens Axboecfc4ba52009-09-14 13:12:40 +020022
23/*
Tejun Heo181387d2013-04-01 19:08:06 -070024 * bdi_lock protects updates to bdi_list. bdi_list has RCU reader side
Jens Axboecfc4ba52009-09-14 13:12:40 +020025 * locking.
26 */
Jens Axboe03ba3782009-09-09 09:08:54 +020027DEFINE_SPINLOCK(bdi_lock);
Jens Axboe66f3b8e2009-09-02 09:19:46 +020028LIST_HEAD(bdi_list);
Jens Axboe03ba3782009-09-09 09:08:54 +020029
Tejun Heo839a8e82013-04-01 19:08:06 -070030/* bdi_wq serves all asynchronous writeback tasks */
31struct workqueue_struct *bdi_wq;
32
Miklos Szeredi76f14182008-04-30 00:54:36 -070033#ifdef CONFIG_DEBUG_FS
34#include <linux/debugfs.h>
35#include <linux/seq_file.h>
36
37static struct dentry *bdi_debug_root;
38
39static void bdi_debug_init(void)
40{
41 bdi_debug_root = debugfs_create_dir("bdi", NULL);
42}
43
44static int bdi_debug_stats_show(struct seq_file *m, void *v)
45{
46 struct backing_dev_info *bdi = m->private;
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020047 struct bdi_writeback *wb = &bdi->wb;
David Rientjes364aeb22009-01-06 14:39:29 -080048 unsigned long background_thresh;
49 unsigned long dirty_thresh;
Tejun Heo0d960a32015-05-22 18:23:19 -040050 unsigned long wb_thresh;
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050051 unsigned long nr_dirty, nr_io, nr_more_io, nr_dirty_time;
Jens Axboef09b00d2009-05-25 09:08:21 +020052 struct inode *inode;
53
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050054 nr_dirty = nr_io = nr_more_io = nr_dirty_time = 0;
Christoph Hellwigf758eea2011-04-21 18:19:44 -060055 spin_lock(&wb->list_lock);
Dave Chinnerc7f54082015-03-04 14:07:22 -050056 list_for_each_entry(inode, &wb->b_dirty, i_io_list)
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020057 nr_dirty++;
Dave Chinnerc7f54082015-03-04 14:07:22 -050058 list_for_each_entry(inode, &wb->b_io, i_io_list)
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020059 nr_io++;
Dave Chinnerc7f54082015-03-04 14:07:22 -050060 list_for_each_entry(inode, &wb->b_more_io, i_io_list)
Christoph Hellwigc1955ce2010-06-19 23:08:06 +020061 nr_more_io++;
Dave Chinnerc7f54082015-03-04 14:07:22 -050062 list_for_each_entry(inode, &wb->b_dirty_time, i_io_list)
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050063 if (inode->i_state & I_DIRTY_TIME)
64 nr_dirty_time++;
Christoph Hellwigf758eea2011-04-21 18:19:44 -060065 spin_unlock(&wb->list_lock);
Miklos Szeredi76f14182008-04-30 00:54:36 -070066
Wu Fengguang16c40422010-08-11 14:17:39 -070067 global_dirty_limits(&background_thresh, &dirty_thresh);
Tejun Heo0d960a32015-05-22 18:23:19 -040068 wb_thresh = wb_calc_thresh(wb, dirty_thresh);
Miklos Szeredi76f14182008-04-30 00:54:36 -070069
70#define K(x) ((x) << (PAGE_SHIFT - 10))
71 seq_printf(m,
Wu Fengguang00821b02010-08-29 11:28:45 -060072 "BdiWriteback: %10lu kB\n"
73 "BdiReclaimable: %10lu kB\n"
74 "BdiDirtyThresh: %10lu kB\n"
75 "DirtyThresh: %10lu kB\n"
76 "BackgroundThresh: %10lu kB\n"
Wu Fengguangc8e28ce2011-01-23 10:07:47 -060077 "BdiDirtied: %10lu kB\n"
Wu Fengguang00821b02010-08-29 11:28:45 -060078 "BdiWritten: %10lu kB\n"
79 "BdiWriteBandwidth: %10lu kBps\n"
80 "b_dirty: %10lu\n"
81 "b_io: %10lu\n"
82 "b_more_io: %10lu\n"
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050083 "b_dirty_time: %10lu\n"
Wu Fengguang00821b02010-08-29 11:28:45 -060084 "bdi_list: %10u\n"
85 "state: %10lx\n",
Tejun Heo93f78d82015-05-22 17:13:27 -040086 (unsigned long) K(wb_stat(wb, WB_WRITEBACK)),
87 (unsigned long) K(wb_stat(wb, WB_RECLAIMABLE)),
Tejun Heo0d960a32015-05-22 18:23:19 -040088 K(wb_thresh),
Jan Karaf7d2b1e2010-12-08 22:44:24 -060089 K(dirty_thresh),
90 K(background_thresh),
Tejun Heo93f78d82015-05-22 17:13:27 -040091 (unsigned long) K(wb_stat(wb, WB_DIRTIED)),
92 (unsigned long) K(wb_stat(wb, WB_WRITTEN)),
Tejun Heoa88a3412015-05-22 17:13:28 -040093 (unsigned long) K(wb->write_bandwidth),
Jan Karaf7d2b1e2010-12-08 22:44:24 -060094 nr_dirty,
95 nr_io,
96 nr_more_io,
Theodore Ts'o0ae45f62015-02-02 00:37:00 -050097 nr_dirty_time,
Tejun Heo44522262015-05-22 17:13:26 -040098 !list_empty(&bdi->bdi_list), bdi->wb.state);
Miklos Szeredi76f14182008-04-30 00:54:36 -070099#undef K
100
101 return 0;
102}
Andy Shevchenko5ad35092018-04-05 16:23:16 -0700103DEFINE_SHOW_ATTRIBUTE(bdi_debug_stats);
Miklos Szeredi76f14182008-04-30 00:54:36 -0700104
weiping zhang97f07692017-10-31 18:37:54 +0800105static int bdi_debug_register(struct backing_dev_info *bdi, const char *name)
Miklos Szeredi76f14182008-04-30 00:54:36 -0700106{
weiping zhang97f07692017-10-31 18:37:54 +0800107 if (!bdi_debug_root)
108 return -ENOMEM;
109
Miklos Szeredi76f14182008-04-30 00:54:36 -0700110 bdi->debug_dir = debugfs_create_dir(name, bdi_debug_root);
weiping zhang97f07692017-10-31 18:37:54 +0800111 if (!bdi->debug_dir)
112 return -ENOMEM;
113
Miklos Szeredi76f14182008-04-30 00:54:36 -0700114 bdi->debug_stats = debugfs_create_file("stats", 0444, bdi->debug_dir,
115 bdi, &bdi_debug_stats_fops);
weiping zhang97f07692017-10-31 18:37:54 +0800116 if (!bdi->debug_stats) {
117 debugfs_remove(bdi->debug_dir);
Tetsuo Handaf53823c2018-04-23 11:21:03 +0900118 bdi->debug_dir = NULL;
weiping zhang97f07692017-10-31 18:37:54 +0800119 return -ENOMEM;
120 }
121
122 return 0;
Miklos Szeredi76f14182008-04-30 00:54:36 -0700123}
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}
weiping zhang97f07692017-10-31 18:37:54 +0800134static inline int bdi_debug_register(struct backing_dev_info *bdi,
Miklos Szeredi76f14182008-04-30 00:54:36 -0700135 const char *name)
136{
weiping zhang97f07692017-10-31 18:37:54 +0800137 return 0;
Miklos Szeredi76f14182008-04-30 00:54:36 -0700138}
139static inline void bdi_debug_unregister(struct backing_dev_info *bdi)
140{
141}
142#endif
143
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700144static ssize_t read_ahead_kb_store(struct device *dev,
145 struct device_attribute *attr,
146 const char *buf, size_t count)
147{
148 struct backing_dev_info *bdi = dev_get_drvdata(dev);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700149 unsigned long read_ahead_kb;
Namjae Jeon7034ed12012-08-25 16:57:27 +0800150 ssize_t ret;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700151
Namjae Jeon7034ed12012-08-25 16:57:27 +0800152 ret = kstrtoul(buf, 10, &read_ahead_kb);
153 if (ret < 0)
154 return ret;
155
156 bdi->ra_pages = read_ahead_kb >> (PAGE_SHIFT - 10);
157
158 return count;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700159}
160
161#define K(pages) ((pages) << (PAGE_SHIFT - 10))
162
163#define BDI_SHOW(name, expr) \
164static ssize_t name##_show(struct device *dev, \
165 struct device_attribute *attr, char *page) \
166{ \
167 struct backing_dev_info *bdi = dev_get_drvdata(dev); \
168 \
169 return snprintf(page, PAGE_SIZE-1, "%lld\n", (long long)expr); \
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700170} \
171static DEVICE_ATTR_RW(name);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700172
173BDI_SHOW(read_ahead_kb, K(bdi->ra_pages))
174
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700175static ssize_t min_ratio_store(struct device *dev,
176 struct device_attribute *attr, const char *buf, size_t count)
177{
178 struct backing_dev_info *bdi = dev_get_drvdata(dev);
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700179 unsigned int ratio;
Namjae Jeon7034ed12012-08-25 16:57:27 +0800180 ssize_t ret;
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700181
Namjae Jeon7034ed12012-08-25 16:57:27 +0800182 ret = kstrtouint(buf, 10, &ratio);
183 if (ret < 0)
184 return ret;
185
186 ret = bdi_set_min_ratio(bdi, ratio);
187 if (!ret)
188 ret = count;
189
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700190 return ret;
191}
192BDI_SHOW(min_ratio, bdi->min_ratio)
193
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700194static ssize_t max_ratio_store(struct device *dev,
195 struct device_attribute *attr, const char *buf, size_t count)
196{
197 struct backing_dev_info *bdi = dev_get_drvdata(dev);
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700198 unsigned int ratio;
Namjae Jeon7034ed12012-08-25 16:57:27 +0800199 ssize_t ret;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700200
Namjae Jeon7034ed12012-08-25 16:57:27 +0800201 ret = kstrtouint(buf, 10, &ratio);
202 if (ret < 0)
203 return ret;
204
205 ret = bdi_set_max_ratio(bdi, ratio);
206 if (!ret)
207 ret = count;
208
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700209 return ret;
210}
211BDI_SHOW(max_ratio, bdi->max_ratio)
212
Darrick J. Wong7d311cd2013-02-21 16:42:48 -0800213static ssize_t stable_pages_required_show(struct device *dev,
214 struct device_attribute *attr,
215 char *page)
216{
217 struct backing_dev_info *bdi = dev_get_drvdata(dev);
218
219 return snprintf(page, PAGE_SIZE-1, "%d\n",
220 bdi_cap_stable_pages_required(bdi) ? 1 : 0);
221}
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700222static DEVICE_ATTR_RO(stable_pages_required);
Darrick J. Wong7d311cd2013-02-21 16:42:48 -0800223
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700224static struct attribute *bdi_dev_attrs[] = {
225 &dev_attr_read_ahead_kb.attr,
226 &dev_attr_min_ratio.attr,
227 &dev_attr_max_ratio.attr,
228 &dev_attr_stable_pages_required.attr,
229 NULL,
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700230};
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700231ATTRIBUTE_GROUPS(bdi_dev);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700232
233static __init int bdi_class_init(void)
234{
235 bdi_class = class_create(THIS_MODULE, "bdi");
Anton Blanchard14421452010-04-02 09:46:55 +0200236 if (IS_ERR(bdi_class))
237 return PTR_ERR(bdi_class);
238
Greg Kroah-Hartmand9e12412013-07-24 15:05:26 -0700239 bdi_class->dev_groups = bdi_dev_groups;
Miklos Szeredi76f14182008-04-30 00:54:36 -0700240 bdi_debug_init();
Jan Karad03f6cd2017-02-02 15:56:51 +0100241
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700242 return 0;
243}
Miklos Szeredi76f14182008-04-30 00:54:36 -0700244postcore_initcall(bdi_class_init);
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700245
Jan Kara2e82b842017-04-12 12:24:48 +0200246static int bdi_init(struct backing_dev_info *bdi);
247
Jens Axboe26160152009-03-17 09:35:06 +0100248static int __init default_bdi_init(void)
249{
250 int err;
251
Tejun Heo839a8e82013-04-01 19:08:06 -0700252 bdi_wq = alloc_workqueue("writeback", WQ_MEM_RECLAIM | WQ_FREEZABLE |
Tejun Heob5c872d2013-04-01 19:08:06 -0700253 WQ_UNBOUND | WQ_SYSFS, 0);
Tejun Heo839a8e82013-04-01 19:08:06 -0700254 if (!bdi_wq)
255 return -ENOMEM;
256
Jan Kara976e48f2010-09-21 11:48:55 +0200257 err = bdi_init(&noop_backing_dev_info);
Jens Axboe26160152009-03-17 09:35:06 +0100258
259 return err;
260}
261subsys_initcall(default_bdi_init);
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
Jens Axboecfc4ba52009-09-14 13:12:40 +0200290 */
Tejun Heoa88a3412015-05-22 17:13:28 -0400291#define INIT_BW (100 << (20 - PAGE_SHIFT))
Jens Axboecfc4ba52009-09-14 13:12:40 +0200292
Tejun Heo8395cd92015-05-22 17:13:34 -0400293static int wb_init(struct bdi_writeback *wb, struct backing_dev_info *bdi,
Tejun Heoa13f35e2015-07-02 08:44:34 -0600294 int blkcg_id, gfp_t gfp)
Artem Bityutskiy64677162010-07-25 14:29:22 +0300295{
Tejun Heo93f78d82015-05-22 17:13:27 -0400296 int i, err;
297
Artem Bityutskiy64677162010-07-25 14:29:22 +0300298 memset(wb, 0, sizeof(*wb));
299
Jan Kara810df542017-03-23 01:36:55 +0100300 if (wb != &bdi->wb)
301 bdi_get(bdi);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300302 wb->bdi = bdi;
303 wb->last_old_flush = jiffies;
304 INIT_LIST_HEAD(&wb->b_dirty);
305 INIT_LIST_HEAD(&wb->b_io);
306 INIT_LIST_HEAD(&wb->b_more_io);
Theodore Ts'o0ae45f62015-02-02 00:37:00 -0500307 INIT_LIST_HEAD(&wb->b_dirty_time);
Christoph Hellwigf758eea2011-04-21 18:19:44 -0600308 spin_lock_init(&wb->list_lock);
Tejun Heo93f78d82015-05-22 17:13:27 -0400309
Tejun Heoa88a3412015-05-22 17:13:28 -0400310 wb->bw_time_stamp = jiffies;
311 wb->balanced_dirty_ratelimit = INIT_BW;
312 wb->dirty_ratelimit = INIT_BW;
313 wb->write_bandwidth = INIT_BW;
314 wb->avg_write_bandwidth = INIT_BW;
315
Tejun Heof0054bb2015-05-22 17:13:30 -0400316 spin_lock_init(&wb->work_lock);
317 INIT_LIST_HEAD(&wb->work_list);
318 INIT_DELAYED_WORK(&wb->dwork, wb_workfn);
Jens Axboeb57d74a2016-09-01 10:20:33 -0600319 wb->dirty_sleep = jiffies;
Tejun Heof0054bb2015-05-22 17:13:30 -0400320
Tejun Heoa13f35e2015-07-02 08:44:34 -0600321 wb->congested = wb_congested_get_create(bdi, blkcg_id, gfp);
Jan Kara810df542017-03-23 01:36:55 +0100322 if (!wb->congested) {
323 err = -ENOMEM;
324 goto out_put_bdi;
325 }
Tejun Heoa13f35e2015-07-02 08:44:34 -0600326
Tejun Heo8395cd92015-05-22 17:13:34 -0400327 err = fprop_local_init_percpu(&wb->completions, gfp);
Tejun Heoa88a3412015-05-22 17:13:28 -0400328 if (err)
Tejun Heoa13f35e2015-07-02 08:44:34 -0600329 goto out_put_cong;
Tejun Heoa88a3412015-05-22 17:13:28 -0400330
Tejun Heo93f78d82015-05-22 17:13:27 -0400331 for (i = 0; i < NR_WB_STAT_ITEMS; i++) {
Tejun Heo8395cd92015-05-22 17:13:34 -0400332 err = percpu_counter_init(&wb->stat[i], 0, gfp);
Tejun Heoa13f35e2015-07-02 08:44:34 -0600333 if (err)
334 goto out_destroy_stat;
Tejun Heo93f78d82015-05-22 17:13:27 -0400335 }
336
337 return 0;
Tejun Heoa13f35e2015-07-02 08:44:34 -0600338
339out_destroy_stat:
Rasmus Villemoes078c6c32016-02-11 16:13:06 -0800340 while (i--)
Tejun Heoa13f35e2015-07-02 08:44:34 -0600341 percpu_counter_destroy(&wb->stat[i]);
342 fprop_local_destroy_percpu(&wb->completions);
343out_put_cong:
344 wb_congested_put(wb->congested);
Jan Kara810df542017-03-23 01:36:55 +0100345out_put_bdi:
346 if (wb != &bdi->wb)
347 bdi_put(bdi);
Tejun Heoa13f35e2015-07-02 08:44:34 -0600348 return err;
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700349}
Jens Axboe03ba3782009-09-09 09:08:54 +0200350
Jan Karae8cb72b2017-03-23 01:36:56 +0100351static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb);
352
Tejun Heo46100072015-05-22 17:13:31 -0400353/*
354 * Remove bdi from the global list and shutdown any threads we have running
355 */
356static void wb_shutdown(struct bdi_writeback *wb)
357{
358 /* Make sure nobody queues further work */
359 spin_lock_bh(&wb->work_lock);
360 if (!test_and_clear_bit(WB_registered, &wb->state)) {
361 spin_unlock_bh(&wb->work_lock);
362 return;
363 }
364 spin_unlock_bh(&wb->work_lock);
365
Jan Karae8cb72b2017-03-23 01:36:56 +0100366 cgwb_remove_from_bdi_list(wb);
Tejun Heo46100072015-05-22 17:13:31 -0400367 /*
368 * Drain work list and shutdown the delayed_work. !WB_registered
369 * tells wb_workfn() that @wb is dying and its work_list needs to
370 * be drained no matter what.
371 */
372 mod_delayed_work(bdi_wq, &wb->dwork, 0);
373 flush_delayed_work(&wb->dwork);
374 WARN_ON(!list_empty(&wb->work_list));
375}
376
Tejun Heof0054bb2015-05-22 17:13:30 -0400377static void wb_exit(struct bdi_writeback *wb)
Tejun Heo93f78d82015-05-22 17:13:27 -0400378{
379 int i;
380
381 WARN_ON(delayed_work_pending(&wb->dwork));
382
383 for (i = 0; i < NR_WB_STAT_ITEMS; i++)
384 percpu_counter_destroy(&wb->stat[i]);
Artem Bityutskiy64677162010-07-25 14:29:22 +0300385
Tejun Heoa88a3412015-05-22 17:13:28 -0400386 fprop_local_destroy_percpu(&wb->completions);
Tejun Heoa13f35e2015-07-02 08:44:34 -0600387 wb_congested_put(wb->congested);
Jan Kara810df542017-03-23 01:36:55 +0100388 if (wb != &wb->bdi->wb)
389 bdi_put(wb->bdi);
Tejun Heoa88a3412015-05-22 17:13:28 -0400390}
Wu Fengguange98be2d2010-08-29 11:22:30 -0600391
Tejun Heo52ebea72015-05-22 17:13:37 -0400392#ifdef CONFIG_CGROUP_WRITEBACK
393
394#include <linux/memcontrol.h>
395
396/*
397 * cgwb_lock protects bdi->cgwb_tree, bdi->cgwb_congested_tree,
398 * blkcg->cgwb_list, and memcg->cgwb_list. bdi->cgwb_tree is also RCU
Jan Kara45144512017-03-23 01:36:58 +0100399 * protected.
Tejun Heo52ebea72015-05-22 17:13:37 -0400400 */
401static DEFINE_SPINLOCK(cgwb_lock);
Tejun Heof1834642018-05-23 10:56:32 -0700402static struct workqueue_struct *cgwb_release_wq;
Tejun Heo52ebea72015-05-22 17:13:37 -0400403
404/**
405 * wb_congested_get_create - get or create a wb_congested
406 * @bdi: associated bdi
407 * @blkcg_id: ID of the associated blkcg
408 * @gfp: allocation mask
409 *
410 * Look up the wb_congested for @blkcg_id on @bdi. If missing, create one.
411 * The returned wb_congested has its reference count incremented. Returns
412 * NULL on failure.
413 */
414struct bdi_writeback_congested *
415wb_congested_get_create(struct backing_dev_info *bdi, int blkcg_id, gfp_t gfp)
416{
417 struct bdi_writeback_congested *new_congested = NULL, *congested;
418 struct rb_node **node, *parent;
419 unsigned long flags;
Tejun Heo52ebea72015-05-22 17:13:37 -0400420retry:
421 spin_lock_irqsave(&cgwb_lock, flags);
422
423 node = &bdi->cgwb_congested_tree.rb_node;
424 parent = NULL;
425
426 while (*node != NULL) {
427 parent = *node;
Geliang Tangbc712262017-02-22 15:45:52 -0800428 congested = rb_entry(parent, struct bdi_writeback_congested,
429 rb_node);
Tejun Heo52ebea72015-05-22 17:13:37 -0400430 if (congested->blkcg_id < blkcg_id)
431 node = &parent->rb_left;
432 else if (congested->blkcg_id > blkcg_id)
433 node = &parent->rb_right;
434 else
435 goto found;
436 }
437
438 if (new_congested) {
439 /* !found and storage for new one already allocated, insert */
440 congested = new_congested;
Tejun Heo52ebea72015-05-22 17:13:37 -0400441 rb_link_node(&congested->rb_node, parent, node);
442 rb_insert_color(&congested->rb_node, &bdi->cgwb_congested_tree);
Sebastian Andrzej Siewiore58dd0d2018-08-21 21:55:31 -0700443 spin_unlock_irqrestore(&cgwb_lock, flags);
444 return congested;
Tejun Heo52ebea72015-05-22 17:13:37 -0400445 }
446
447 spin_unlock_irqrestore(&cgwb_lock, flags);
448
449 /* allocate storage for new one and retry */
450 new_congested = kzalloc(sizeof(*new_congested), gfp);
451 if (!new_congested)
452 return NULL;
453
Sebastian Andrzej Siewiore58dd0d2018-08-21 21:55:31 -0700454 refcount_set(&new_congested->refcnt, 1);
Jan Karab7d680d2017-03-23 01:36:54 +0100455 new_congested->__bdi = bdi;
Tejun Heo52ebea72015-05-22 17:13:37 -0400456 new_congested->blkcg_id = blkcg_id;
457 goto retry;
458
459found:
Sebastian Andrzej Siewiore58dd0d2018-08-21 21:55:31 -0700460 refcount_inc(&congested->refcnt);
Tejun Heo52ebea72015-05-22 17:13:37 -0400461 spin_unlock_irqrestore(&cgwb_lock, flags);
462 kfree(new_congested);
463 return congested;
464}
465
466/**
467 * wb_congested_put - put a wb_congested
468 * @congested: wb_congested to put
469 *
470 * Put @congested and destroy it if the refcnt reaches zero.
471 */
472void wb_congested_put(struct bdi_writeback_congested *congested)
473{
Tejun Heo52ebea72015-05-22 17:13:37 -0400474 unsigned long flags;
475
Anna-Maria Gleixner060288a72018-08-21 21:55:35 -0700476 if (!refcount_dec_and_lock_irqsave(&congested->refcnt, &cgwb_lock, &flags))
Tejun Heo52ebea72015-05-22 17:13:37 -0400477 return;
Tejun Heo52ebea72015-05-22 17:13:37 -0400478
Tejun Heoa20135f2015-07-01 20:53:37 -0400479 /* bdi might already have been destroyed leaving @congested unlinked */
Jan Karab7d680d2017-03-23 01:36:54 +0100480 if (congested->__bdi) {
Tejun Heoa20135f2015-07-01 20:53:37 -0400481 rb_erase(&congested->rb_node,
Jan Karab7d680d2017-03-23 01:36:54 +0100482 &congested->__bdi->cgwb_congested_tree);
483 congested->__bdi = NULL;
Tejun Heoa20135f2015-07-01 20:53:37 -0400484 }
485
Tejun Heo52ebea72015-05-22 17:13:37 -0400486 spin_unlock_irqrestore(&cgwb_lock, flags);
487 kfree(congested);
Tejun Heo52ebea72015-05-22 17:13:37 -0400488}
489
490static void cgwb_release_workfn(struct work_struct *work)
491{
492 struct bdi_writeback *wb = container_of(work, struct bdi_writeback,
493 release_work);
Dennis Zhou (Facebook)59b57712018-08-31 16:22:43 -0400494 struct blkcg *blkcg = css_to_blkcg(wb->blkcg_css);
Tejun Heo52ebea72015-05-22 17:13:37 -0400495
Jan Kara3ee7e862018-06-18 15:46:58 +0200496 mutex_lock(&wb->bdi->cgwb_release_mutex);
Tejun Heo52ebea72015-05-22 17:13:37 -0400497 wb_shutdown(wb);
498
499 css_put(wb->memcg_css);
500 css_put(wb->blkcg_css);
Jan Kara3ee7e862018-06-18 15:46:58 +0200501 mutex_unlock(&wb->bdi->cgwb_release_mutex);
Tejun Heo52ebea72015-05-22 17:13:37 -0400502
Dennis Zhou (Facebook)59b57712018-08-31 16:22:43 -0400503 /* triggers blkg destruction if cgwb_refcnt becomes zero */
504 blkcg_cgwb_put(blkcg);
505
Tejun Heo841710a2015-05-22 18:23:33 -0400506 fprop_local_destroy_percpu(&wb->memcg_completions);
Tejun Heo52ebea72015-05-22 17:13:37 -0400507 percpu_ref_exit(&wb->refcnt);
508 wb_exit(wb);
509 kfree_rcu(wb, rcu);
Tejun Heo52ebea72015-05-22 17:13:37 -0400510}
511
512static void cgwb_release(struct percpu_ref *refcnt)
513{
514 struct bdi_writeback *wb = container_of(refcnt, struct bdi_writeback,
515 refcnt);
Tejun Heof1834642018-05-23 10:56:32 -0700516 queue_work(cgwb_release_wq, &wb->release_work);
Tejun Heo52ebea72015-05-22 17:13:37 -0400517}
518
519static void cgwb_kill(struct bdi_writeback *wb)
520{
521 lockdep_assert_held(&cgwb_lock);
522
523 WARN_ON(!radix_tree_delete(&wb->bdi->cgwb_tree, wb->memcg_css->id));
524 list_del(&wb->memcg_node);
525 list_del(&wb->blkcg_node);
526 percpu_ref_kill(&wb->refcnt);
527}
528
Jan Karae8cb72b2017-03-23 01:36:56 +0100529static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
530{
531 spin_lock_irq(&cgwb_lock);
532 list_del_rcu(&wb->bdi_node);
533 spin_unlock_irq(&cgwb_lock);
534}
535
Tejun Heo52ebea72015-05-22 17:13:37 -0400536static int cgwb_create(struct backing_dev_info *bdi,
537 struct cgroup_subsys_state *memcg_css, gfp_t gfp)
538{
539 struct mem_cgroup *memcg;
540 struct cgroup_subsys_state *blkcg_css;
541 struct blkcg *blkcg;
542 struct list_head *memcg_cgwb_list, *blkcg_cgwb_list;
543 struct bdi_writeback *wb;
544 unsigned long flags;
545 int ret = 0;
546
547 memcg = mem_cgroup_from_css(memcg_css);
Tejun Heoc165b3e2015-08-18 14:55:29 -0700548 blkcg_css = cgroup_get_e_css(memcg_css->cgroup, &io_cgrp_subsys);
Tejun Heo52ebea72015-05-22 17:13:37 -0400549 blkcg = css_to_blkcg(blkcg_css);
Wang Long9ccc3612018-06-07 17:07:19 -0700550 memcg_cgwb_list = &memcg->cgwb_list;
Tejun Heo52ebea72015-05-22 17:13:37 -0400551 blkcg_cgwb_list = &blkcg->cgwb_list;
552
553 /* look up again under lock and discard on blkcg mismatch */
554 spin_lock_irqsave(&cgwb_lock, flags);
555 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
556 if (wb && wb->blkcg_css != blkcg_css) {
557 cgwb_kill(wb);
558 wb = NULL;
559 }
560 spin_unlock_irqrestore(&cgwb_lock, flags);
561 if (wb)
562 goto out_put;
563
564 /* need to create a new one */
565 wb = kmalloc(sizeof(*wb), gfp);
Christophe JAILLET0b045bd2017-09-11 21:43:23 +0200566 if (!wb) {
567 ret = -ENOMEM;
568 goto out_put;
569 }
Tejun Heo52ebea72015-05-22 17:13:37 -0400570
Tejun Heoa13f35e2015-07-02 08:44:34 -0600571 ret = wb_init(wb, bdi, blkcg_css->id, gfp);
Tejun Heo52ebea72015-05-22 17:13:37 -0400572 if (ret)
573 goto err_free;
574
575 ret = percpu_ref_init(&wb->refcnt, cgwb_release, 0, gfp);
576 if (ret)
577 goto err_wb_exit;
578
Tejun Heo841710a2015-05-22 18:23:33 -0400579 ret = fprop_local_init_percpu(&wb->memcg_completions, gfp);
580 if (ret)
581 goto err_ref_exit;
582
Tejun Heo52ebea72015-05-22 17:13:37 -0400583 wb->memcg_css = memcg_css;
584 wb->blkcg_css = blkcg_css;
585 INIT_WORK(&wb->release_work, cgwb_release_workfn);
586 set_bit(WB_registered, &wb->state);
587
588 /*
589 * The root wb determines the registered state of the whole bdi and
590 * memcg_cgwb_list and blkcg_cgwb_list's next pointers indicate
591 * whether they're still online. Don't link @wb if any is dead.
592 * See wb_memcg_offline() and wb_blkcg_offline().
593 */
594 ret = -ENODEV;
595 spin_lock_irqsave(&cgwb_lock, flags);
596 if (test_bit(WB_registered, &bdi->wb.state) &&
597 blkcg_cgwb_list->next && memcg_cgwb_list->next) {
598 /* we might have raced another instance of this function */
599 ret = radix_tree_insert(&bdi->cgwb_tree, memcg_css->id, wb);
600 if (!ret) {
Tejun Heob8175252015-10-02 14:47:05 -0400601 list_add_tail_rcu(&wb->bdi_node, &bdi->wb_list);
Tejun Heo52ebea72015-05-22 17:13:37 -0400602 list_add(&wb->memcg_node, memcg_cgwb_list);
603 list_add(&wb->blkcg_node, blkcg_cgwb_list);
Dennis Zhou (Facebook)59b57712018-08-31 16:22:43 -0400604 blkcg_cgwb_get(blkcg);
Tejun Heo52ebea72015-05-22 17:13:37 -0400605 css_get(memcg_css);
606 css_get(blkcg_css);
607 }
608 }
609 spin_unlock_irqrestore(&cgwb_lock, flags);
610 if (ret) {
611 if (ret == -EEXIST)
612 ret = 0;
Tejun Heoa13f35e2015-07-02 08:44:34 -0600613 goto err_fprop_exit;
Tejun Heo52ebea72015-05-22 17:13:37 -0400614 }
615 goto out_put;
616
Tejun Heo841710a2015-05-22 18:23:33 -0400617err_fprop_exit:
618 fprop_local_destroy_percpu(&wb->memcg_completions);
Tejun Heo52ebea72015-05-22 17:13:37 -0400619err_ref_exit:
620 percpu_ref_exit(&wb->refcnt);
621err_wb_exit:
622 wb_exit(wb);
623err_free:
624 kfree(wb);
625out_put:
626 css_put(blkcg_css);
627 return ret;
628}
629
630/**
631 * wb_get_create - get wb for a given memcg, create if necessary
632 * @bdi: target bdi
633 * @memcg_css: cgroup_subsys_state of the target memcg (must have positive ref)
634 * @gfp: allocation mask to use
635 *
636 * Try to get the wb for @memcg_css on @bdi. If it doesn't exist, try to
637 * create one. The returned wb has its refcount incremented.
638 *
639 * This function uses css_get() on @memcg_css and thus expects its refcnt
640 * to be positive on invocation. IOW, rcu_read_lock() protection on
641 * @memcg_css isn't enough. try_get it before calling this function.
642 *
643 * A wb is keyed by its associated memcg. As blkcg implicitly enables
644 * memcg on the default hierarchy, memcg association is guaranteed to be
645 * more specific (equal or descendant to the associated blkcg) and thus can
646 * identify both the memcg and blkcg associations.
647 *
648 * Because the blkcg associated with a memcg may change as blkcg is enabled
649 * and disabled closer to root in the hierarchy, each wb keeps track of
650 * both the memcg and blkcg associated with it and verifies the blkcg on
651 * each lookup. On mismatch, the existing wb is discarded and a new one is
652 * created.
653 */
654struct bdi_writeback *wb_get_create(struct backing_dev_info *bdi,
655 struct cgroup_subsys_state *memcg_css,
656 gfp_t gfp)
657{
658 struct bdi_writeback *wb;
659
Mel Gormand0164ad2015-11-06 16:28:21 -0800660 might_sleep_if(gfpflags_allow_blocking(gfp));
Tejun Heo52ebea72015-05-22 17:13:37 -0400661
662 if (!memcg_css->parent)
663 return &bdi->wb;
664
665 do {
666 rcu_read_lock();
667 wb = radix_tree_lookup(&bdi->cgwb_tree, memcg_css->id);
668 if (wb) {
669 struct cgroup_subsys_state *blkcg_css;
670
671 /* see whether the blkcg association has changed */
672 blkcg_css = cgroup_get_e_css(memcg_css->cgroup,
Tejun Heoc165b3e2015-08-18 14:55:29 -0700673 &io_cgrp_subsys);
Tejun Heo52ebea72015-05-22 17:13:37 -0400674 if (unlikely(wb->blkcg_css != blkcg_css ||
675 !wb_tryget(wb)))
676 wb = NULL;
677 css_put(blkcg_css);
678 }
679 rcu_read_unlock();
680 } while (!wb && !cgwb_create(bdi, memcg_css, gfp));
681
682 return wb;
683}
684
Tejun Heoa13f35e2015-07-02 08:44:34 -0600685static int cgwb_bdi_init(struct backing_dev_info *bdi)
Tejun Heo52ebea72015-05-22 17:13:37 -0400686{
Tejun Heoa13f35e2015-07-02 08:44:34 -0600687 int ret;
688
Tejun Heo52ebea72015-05-22 17:13:37 -0400689 INIT_RADIX_TREE(&bdi->cgwb_tree, GFP_ATOMIC);
690 bdi->cgwb_congested_tree = RB_ROOT;
Jan Kara3ee7e862018-06-18 15:46:58 +0200691 mutex_init(&bdi->cgwb_release_mutex);
Tejun Heo7fc58542017-12-12 08:38:30 -0800692 init_rwsem(&bdi->wb_switch_rwsem);
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 Karab1c51af2017-03-23 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 Kara5318ce72017-03-23 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 Kara3ee7e862018-06-18 15:46:58 +0200713 spin_unlock_irq(&cgwb_lock);
Jan Kara5318ce72017-03-23 01:36:57 +0100714
Jan Kara3ee7e862018-06-18 15:46:58 +0200715 mutex_lock(&bdi->cgwb_release_mutex);
716 spin_lock_irq(&cgwb_lock);
Jan Kara5318ce72017-03-23 01:36:57 +0100717 while (!list_empty(&bdi->wb_list)) {
718 wb = list_first_entry(&bdi->wb_list, struct bdi_writeback,
719 bdi_node);
720 spin_unlock_irq(&cgwb_lock);
721 wb_shutdown(wb);
722 spin_lock_irq(&cgwb_lock);
723 }
Tejun Heo52ebea72015-05-22 17:13:37 -0400724 spin_unlock_irq(&cgwb_lock);
Jan Kara3ee7e862018-06-18 15:46:58 +0200725 mutex_unlock(&bdi->cgwb_release_mutex);
Tejun Heo52ebea72015-05-22 17:13:37 -0400726}
727
728/**
729 * wb_memcg_offline - kill all wb's associated with a memcg being offlined
730 * @memcg: memcg being offlined
731 *
732 * Also prevents creation of any new wb's associated with @memcg.
733 */
734void wb_memcg_offline(struct mem_cgroup *memcg)
735{
Wang Long9ccc3612018-06-07 17:07:19 -0700736 struct list_head *memcg_cgwb_list = &memcg->cgwb_list;
Tejun Heo52ebea72015-05-22 17:13:37 -0400737 struct bdi_writeback *wb, *next;
738
739 spin_lock_irq(&cgwb_lock);
740 list_for_each_entry_safe(wb, next, memcg_cgwb_list, memcg_node)
741 cgwb_kill(wb);
742 memcg_cgwb_list->next = NULL; /* prevent new wb's */
743 spin_unlock_irq(&cgwb_lock);
744}
745
746/**
747 * wb_blkcg_offline - kill all wb's associated with a blkcg being offlined
748 * @blkcg: blkcg being offlined
749 *
750 * Also prevents creation of any new wb's associated with @blkcg.
751 */
752void wb_blkcg_offline(struct blkcg *blkcg)
753{
Tejun Heo52ebea72015-05-22 17:13:37 -0400754 struct bdi_writeback *wb, *next;
755
756 spin_lock_irq(&cgwb_lock);
757 list_for_each_entry_safe(wb, next, &blkcg->cgwb_list, blkcg_node)
758 cgwb_kill(wb);
759 blkcg->cgwb_list.next = NULL; /* prevent new wb's */
760 spin_unlock_irq(&cgwb_lock);
761}
762
Jan Karadf23de52017-03-08 17:48:32 +0100763static void cgwb_bdi_exit(struct backing_dev_info *bdi)
764{
765 struct rb_node *rbn;
766
767 spin_lock_irq(&cgwb_lock);
768 while ((rbn = rb_first(&bdi->cgwb_congested_tree))) {
769 struct bdi_writeback_congested *congested =
770 rb_entry(rbn, struct bdi_writeback_congested, rb_node);
771
772 rb_erase(rbn, &bdi->cgwb_congested_tree);
Jan Karab7d680d2017-03-23 01:36:54 +0100773 congested->__bdi = NULL; /* mark @congested unlinked */
Jan Karadf23de52017-03-08 17:48:32 +0100774 }
775 spin_unlock_irq(&cgwb_lock);
776}
777
Jan Karae8cb72b2017-03-23 01:36:56 +0100778static void cgwb_bdi_register(struct backing_dev_info *bdi)
779{
780 spin_lock_irq(&cgwb_lock);
781 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
782 spin_unlock_irq(&cgwb_lock);
783}
784
Tejun Heof1834642018-05-23 10:56:32 -0700785static int __init cgwb_init(void)
786{
787 /*
788 * There can be many concurrent release work items overwhelming
789 * system_wq. Put them in a separate wq and limit concurrency.
790 * There's no point in executing many of these in parallel.
791 */
792 cgwb_release_wq = alloc_workqueue("cgwb_release", 0, 1);
793 if (!cgwb_release_wq)
794 return -ENOMEM;
795
796 return 0;
797}
798subsys_initcall(cgwb_init);
799
Tejun Heo52ebea72015-05-22 17:13:37 -0400800#else /* CONFIG_CGROUP_WRITEBACK */
801
Tejun Heoa13f35e2015-07-02 08:44:34 -0600802static int cgwb_bdi_init(struct backing_dev_info *bdi)
803{
804 int err;
805
806 bdi->wb_congested = kzalloc(sizeof(*bdi->wb_congested), GFP_KERNEL);
807 if (!bdi->wb_congested)
808 return -ENOMEM;
809
Sebastian Andrzej Siewiore58dd0d2018-08-21 21:55:31 -0700810 refcount_set(&bdi->wb_congested->refcnt, 1);
Tejun Heo5f478e42017-02-08 15:19:07 -0500811
Tejun Heoa13f35e2015-07-02 08:44:34 -0600812 err = wb_init(&bdi->wb, bdi, 1, GFP_KERNEL);
813 if (err) {
Tejun Heo5f478e42017-02-08 15:19:07 -0500814 wb_congested_put(bdi->wb_congested);
Tejun Heoa13f35e2015-07-02 08:44:34 -0600815 return err;
816 }
817 return 0;
818}
819
Jan Karab1c51af2017-03-23 01:36:59 +0100820static void cgwb_bdi_unregister(struct backing_dev_info *bdi) { }
Jan Karadf23de52017-03-08 17:48:32 +0100821
822static void cgwb_bdi_exit(struct backing_dev_info *bdi)
Tejun Heo5f478e42017-02-08 15:19:07 -0500823{
824 wb_congested_put(bdi->wb_congested);
825}
Tejun Heo52ebea72015-05-22 17:13:37 -0400826
Jan Karae8cb72b2017-03-23 01:36:56 +0100827static void cgwb_bdi_register(struct backing_dev_info *bdi)
828{
829 list_add_tail_rcu(&bdi->wb.bdi_node, &bdi->wb_list);
830}
831
832static void cgwb_remove_from_bdi_list(struct bdi_writeback *wb)
833{
834 list_del_rcu(&wb->bdi_node);
835}
836
Tejun Heo52ebea72015-05-22 17:13:37 -0400837#endif /* CONFIG_CGROUP_WRITEBACK */
838
Jan Kara2e82b842017-04-12 12:24:48 +0200839static int bdi_init(struct backing_dev_info *bdi)
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700840{
Tejun Heob8175252015-10-02 14:47:05 -0400841 int ret;
842
Peter Zijlstracf0ca9f2008-04-30 00:54:32 -0700843 bdi->dev = NULL;
844
Jan Karad03f6cd2017-02-02 15:56:51 +0100845 kref_init(&bdi->refcnt);
Peter Zijlstra189d3c42008-04-30 00:54:35 -0700846 bdi->min_ratio = 0;
Peter Zijlstraa42dde02008-04-30 00:54:36 -0700847 bdi->max_ratio = 100;
Jan Karaeb608e32012-05-24 18:59:11 +0200848 bdi->max_prop_frac = FPROP_FRAC_BASE;
Jens Axboe66f3b8e2009-09-02 09:19:46 +0200849 INIT_LIST_HEAD(&bdi->bdi_list);
Tejun Heob8175252015-10-02 14:47:05 -0400850 INIT_LIST_HEAD(&bdi->wb_list);
Tejun Heocc395d72015-05-22 17:13:58 -0400851 init_waitqueue_head(&bdi->wb_waitq);
Jens Axboe03ba3782009-09-09 09:08:54 +0200852
Tejun Heob8175252015-10-02 14:47:05 -0400853 ret = cgwb_bdi_init(bdi);
854
Tejun Heob8175252015-10-02 14:47:05 -0400855 return ret;
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700856}
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700857
Jan Karad03f6cd2017-02-02 15:56:51 +0100858struct backing_dev_info *bdi_alloc_node(gfp_t gfp_mask, int node_id)
859{
860 struct backing_dev_info *bdi;
861
862 bdi = kmalloc_node(sizeof(struct backing_dev_info),
863 gfp_mask | __GFP_ZERO, node_id);
864 if (!bdi)
865 return NULL;
866
867 if (bdi_init(bdi)) {
868 kfree(bdi);
869 return NULL;
870 }
871 return bdi;
872}
Jan Kara62bf42a2017-04-12 12:24:27 +0200873EXPORT_SYMBOL(bdi_alloc_node);
Jan Karad03f6cd2017-02-02 15:56:51 +0100874
Jan Kara7c4cc302017-04-12 12:24:49 +0200875int bdi_register_va(struct backing_dev_info *bdi, const char *fmt, va_list args)
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700876{
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700877 struct device *dev;
878
879 if (bdi->dev) /* The driver needs to use separate queues per device */
880 return 0;
881
Jan Kara7c4cc302017-04-12 12:24:49 +0200882 dev = device_create_vargs(bdi_class, NULL, MKDEV(0, 0), bdi, fmt, args);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700883 if (IS_ERR(dev))
884 return PTR_ERR(dev);
885
Jan Karae8cb72b2017-03-23 01:36:56 +0100886 cgwb_bdi_register(bdi);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700887 bdi->dev = dev;
888
Jens Axboe6d0e4822017-12-21 10:01:30 -0700889 bdi_debug_register(bdi, dev_name(dev));
Tejun Heo46100072015-05-22 17:13:31 -0400890 set_bit(WB_registered, &bdi->wb.state);
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700891
892 spin_lock_bh(&bdi_lock);
893 list_add_tail_rcu(&bdi->bdi_list, &bdi_list);
894 spin_unlock_bh(&bdi_lock);
895
896 trace_writeback_bdi_register(bdi);
897 return 0;
898}
Jan Karabaf7a612017-04-12 12:24:25 +0200899EXPORT_SYMBOL(bdi_register_va);
900
Jan Kara7c4cc302017-04-12 12:24:49 +0200901int bdi_register(struct backing_dev_info *bdi, const char *fmt, ...)
Jan Karabaf7a612017-04-12 12:24:25 +0200902{
903 va_list args;
904 int ret;
905
906 va_start(args, fmt);
Jan Kara7c4cc302017-04-12 12:24:49 +0200907 ret = bdi_register_va(bdi, fmt, args);
Jan Karabaf7a612017-04-12 12:24:25 +0200908 va_end(args);
909 return ret;
910}
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700911EXPORT_SYMBOL(bdi_register);
912
Dan Williamsdf08c322016-07-31 11:15:13 -0700913int bdi_register_owner(struct backing_dev_info *bdi, struct device *owner)
914{
915 int rc;
916
Jan Kara7c4cc302017-04-12 12:24:49 +0200917 rc = bdi_register(bdi, "%u:%u", MAJOR(owner->devt), MINOR(owner->devt));
Dan Williamsdf08c322016-07-31 11:15:13 -0700918 if (rc)
919 return rc;
Jan Karab6f8fec2017-03-08 17:48:31 +0100920 /* Leaking owner reference... */
921 WARN_ON(bdi->owner);
Dan Williamsdf08c322016-07-31 11:15:13 -0700922 bdi->owner = owner;
923 get_device(owner);
924 return 0;
925}
926EXPORT_SYMBOL(bdi_register_owner);
927
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700928/*
Tejun Heo46100072015-05-22 17:13:31 -0400929 * Remove bdi from bdi_list, and ensure that it is no longer visible
Peter Zijlstrab2e8fb62007-10-16 23:25:47 -0700930 */
Tejun Heo46100072015-05-22 17:13:31 -0400931static void bdi_remove_from_list(struct backing_dev_info *bdi)
Peter Zijlstra04fbfdc2007-10-16 23:25:50 -0700932{
Tejun Heo46100072015-05-22 17:13:31 -0400933 spin_lock_bh(&bdi_lock);
934 list_del_rcu(&bdi->bdi_list);
935 spin_unlock_bh(&bdi_lock);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700936
Tejun Heo46100072015-05-22 17:13:31 -0400937 synchronize_rcu_expedited();
Andrew Morton3fcfab12006-10-19 23:28:16 -0700938}
939
Tejun Heob02176f2015-09-08 12:20:22 -0400940void bdi_unregister(struct backing_dev_info *bdi)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700941{
Tejun Heof0054bb2015-05-22 17:13:30 -0400942 /* make sure nobody finds us on the bdi_list anymore */
943 bdi_remove_from_list(bdi);
944 wb_shutdown(&bdi->wb);
Jan Karab1c51af2017-03-23 01:36:59 +0100945 cgwb_bdi_unregister(bdi);
Rabin Vincent7a401a92011-11-11 13:29:04 +0100946
Christoph Hellwigc4db59d2015-01-20 14:05:00 -0700947 if (bdi->dev) {
948 bdi_debug_unregister(bdi);
949 device_unregister(bdi->dev);
950 bdi->dev = NULL;
951 }
Dan Williamsdf08c322016-07-31 11:15:13 -0700952
953 if (bdi->owner) {
954 put_device(bdi->owner);
955 bdi->owner = NULL;
956 }
Tejun Heob02176f2015-09-08 12:20:22 -0400957}
Christoph Hellwigc4db59d2015-01-20 14:05:00 -0700958
Jan Karad03f6cd2017-02-02 15:56:51 +0100959static void release_bdi(struct kref *ref)
960{
961 struct backing_dev_info *bdi =
962 container_of(ref, struct backing_dev_info, refcnt);
963
Jan Kara5af110b2017-04-12 12:24:26 +0200964 if (test_bit(WB_registered, &bdi->wb.state))
965 bdi_unregister(bdi);
Jan Kara2e82b842017-04-12 12:24:48 +0200966 WARN_ON_ONCE(bdi->dev);
967 wb_exit(&bdi->wb);
968 cgwb_bdi_exit(bdi);
Jan Karad03f6cd2017-02-02 15:56:51 +0100969 kfree(bdi);
970}
971
972void bdi_put(struct backing_dev_info *bdi)
973{
974 kref_put(&bdi->refcnt, release_bdi);
975}
Jan Kara62bf42a2017-04-12 12:24:27 +0200976EXPORT_SYMBOL(bdi_put);
Jan Karad03f6cd2017-02-02 15:56:51 +0100977
Andrew Morton3fcfab12006-10-19 23:28:16 -0700978static wait_queue_head_t congestion_wqh[2] = {
979 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[0]),
980 __WAIT_QUEUE_HEAD_INITIALIZER(congestion_wqh[1])
981 };
Tejun Heoec8a6f22015-05-22 17:13:41 -0400982static atomic_t nr_wb_congested[2];
Andrew Morton3fcfab12006-10-19 23:28:16 -0700983
Tejun Heoec8a6f22015-05-22 17:13:41 -0400984void clear_wb_congested(struct bdi_writeback_congested *congested, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700985{
Jens Axboe1faa16d2009-04-06 14:48:01 +0200986 wait_queue_head_t *wqh = &congestion_wqh[sync];
Kaixu Xiac877ef82016-03-31 13:19:41 +0000987 enum wb_congested_state bit;
Andrew Morton3fcfab12006-10-19 23:28:16 -0700988
Tejun Heo44522262015-05-22 17:13:26 -0400989 bit = sync ? WB_sync_congested : WB_async_congested;
Tejun Heoec8a6f22015-05-22 17:13:41 -0400990 if (test_and_clear_bit(bit, &congested->state))
991 atomic_dec(&nr_wb_congested[sync]);
Peter Zijlstra4e857c52014-03-17 18:06:10 +0100992 smp_mb__after_atomic();
Andrew Morton3fcfab12006-10-19 23:28:16 -0700993 if (waitqueue_active(wqh))
994 wake_up(wqh);
995}
Tejun Heoec8a6f22015-05-22 17:13:41 -0400996EXPORT_SYMBOL(clear_wb_congested);
Andrew Morton3fcfab12006-10-19 23:28:16 -0700997
Tejun Heoec8a6f22015-05-22 17:13:41 -0400998void set_wb_congested(struct bdi_writeback_congested *congested, int sync)
Andrew Morton3fcfab12006-10-19 23:28:16 -0700999{
Kaixu Xiac877ef82016-03-31 13:19:41 +00001000 enum wb_congested_state bit;
Andrew Morton3fcfab12006-10-19 23:28:16 -07001001
Tejun Heo44522262015-05-22 17:13:26 -04001002 bit = sync ? WB_sync_congested : WB_async_congested;
Tejun Heoec8a6f22015-05-22 17:13:41 -04001003 if (!test_and_set_bit(bit, &congested->state))
1004 atomic_inc(&nr_wb_congested[sync]);
Andrew Morton3fcfab12006-10-19 23:28:16 -07001005}
Tejun Heoec8a6f22015-05-22 17:13:41 -04001006EXPORT_SYMBOL(set_wb_congested);
Andrew Morton3fcfab12006-10-19 23:28:16 -07001007
1008/**
1009 * congestion_wait - wait for a backing_dev to become uncongested
Jens Axboe8aa7e842009-07-09 14:52:32 +02001010 * @sync: SYNC or ASYNC IO
Andrew Morton3fcfab12006-10-19 23:28:16 -07001011 * @timeout: timeout in jiffies
1012 *
1013 * Waits for up to @timeout jiffies for a backing_dev (any backing_dev) to exit
1014 * write congestion. If no backing_devs are congested then just wait for the
1015 * next write to be completed.
1016 */
Jens Axboe8aa7e842009-07-09 14:52:32 +02001017long congestion_wait(int sync, long timeout)
Andrew Morton3fcfab12006-10-19 23:28:16 -07001018{
1019 long ret;
Mel Gorman52bb9192010-10-26 14:21:41 -07001020 unsigned long start = jiffies;
Andrew Morton3fcfab12006-10-19 23:28:16 -07001021 DEFINE_WAIT(wait);
Jens Axboe8aa7e842009-07-09 14:52:32 +02001022 wait_queue_head_t *wqh = &congestion_wqh[sync];
Andrew Morton3fcfab12006-10-19 23:28:16 -07001023
1024 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1025 ret = io_schedule_timeout(timeout);
1026 finish_wait(wqh, &wait);
Mel Gorman52bb9192010-10-26 14:21:41 -07001027
1028 trace_writeback_congestion_wait(jiffies_to_usecs(timeout),
1029 jiffies_to_usecs(jiffies - start));
1030
Andrew Morton3fcfab12006-10-19 23:28:16 -07001031 return ret;
1032}
1033EXPORT_SYMBOL(congestion_wait);
1034
Mel Gorman0e093d992010-10-26 14:21:45 -07001035/**
Mel Gorman599d0c92016-07-28 15:45:31 -07001036 * wait_iff_congested - Conditionally wait for a backing_dev to become uncongested or a pgdat to complete writes
Mel Gorman0e093d992010-10-26 14:21:45 -07001037 * @sync: SYNC or ASYNC IO
1038 * @timeout: timeout in jiffies
1039 *
Andrey Ryabinine3c1ac52018-04-10 16:28:03 -07001040 * In the event of a congested backing_dev (any backing_dev) this waits
1041 * for up to @timeout jiffies for either a BDI to exit congestion of the
1042 * given @sync queue or a write to complete.
Mel Gorman0e093d992010-10-26 14:21:45 -07001043 *
1044 * The return value is 0 if the sleep is for the full timeout. Otherwise,
1045 * it is the number of jiffies that were still remaining when the function
1046 * returned. return_value == timeout implies the function did not sleep.
1047 */
Andrey Ryabinine3c1ac52018-04-10 16:28:03 -07001048long wait_iff_congested(int sync, long timeout)
Mel Gorman0e093d992010-10-26 14:21:45 -07001049{
1050 long ret;
1051 unsigned long start = jiffies;
1052 DEFINE_WAIT(wait);
1053 wait_queue_head_t *wqh = &congestion_wqh[sync];
1054
1055 /*
Andrey Ryabinine3c1ac52018-04-10 16:28:03 -07001056 * If there is no congestion, yield if necessary instead
Mel Gorman0e093d992010-10-26 14:21:45 -07001057 * of sleeping on the congestion queue
1058 */
Andrey Ryabinine3c1ac52018-04-10 16:28:03 -07001059 if (atomic_read(&nr_wb_congested[sync]) == 0) {
Michal Hockoede37712016-05-20 16:57:03 -07001060 cond_resched();
Mel Gorman599d0c92016-07-28 15:45:31 -07001061
Mel Gorman0e093d992010-10-26 14:21:45 -07001062 /* In case we scheduled, work out time remaining */
1063 ret = timeout - (jiffies - start);
1064 if (ret < 0)
1065 ret = 0;
1066
1067 goto out;
1068 }
1069
1070 /* Sleep until uncongested or a write happens */
1071 prepare_to_wait(wqh, &wait, TASK_UNINTERRUPTIBLE);
1072 ret = io_schedule_timeout(timeout);
1073 finish_wait(wqh, &wait);
1074
1075out:
1076 trace_writeback_wait_iff_congested(jiffies_to_usecs(timeout),
1077 jiffies_to_usecs(jiffies - start));
1078
1079 return ret;
1080}
1081EXPORT_SYMBOL(wait_iff_congested);