blob: 10919fae2d3aa2e6ebc1e3cdc2d1636939c997ad [file] [log] [blame]
Vivek Goyal31e4c282009-12-03 12:59:42 -05001#ifndef _BLK_CGROUP_H
2#define _BLK_CGROUP_H
3/*
4 * Common Block IO controller cgroup interface
5 *
6 * Based on ideas and code from CFQ, CFS and BFQ:
7 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
8 *
9 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
10 * Paolo Valente <paolo.valente@unimore.it>
11 *
12 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
13 * Nauman Rafique <nauman@google.com>
14 */
15
16#include <linux/cgroup.h>
17
Vivek Goyal062a6442010-09-15 17:06:33 -040018enum blkio_policy_id {
19 BLKIO_POLICY_PROP = 0, /* Proportional Bandwidth division */
Vivek Goyal4c9eefa2010-09-15 17:06:34 -040020 BLKIO_POLICY_THROTL, /* Throttling */
Vivek Goyal062a6442010-09-15 17:06:33 -040021};
22
Vivek Goyal9355aed2010-10-01 21:16:41 +020023/* Max limits for throttle policy */
24#define THROTL_IOPS_MAX UINT_MAX
25
Ben Blum67523c42010-03-10 15:22:11 -080026#if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
27
28#ifndef CONFIG_BLK_CGROUP
29/* When blk-cgroup is a module, its subsys_id isn't a compile-time constant */
30extern struct cgroup_subsys blkio_subsys;
31#define blkio_subsys_id blkio_subsys.subsys_id
32#endif
Jens Axboe2f5ea472009-12-03 21:06:43 +010033
Divyesh Shah84c124d2010-04-09 08:31:19 +020034enum stat_type {
35 /* Total time spent (in ns) between request dispatch to the driver and
36 * request completion for IOs doen by this cgroup. This may not be
37 * accurate when NCQ is turned on. */
38 BLKIO_STAT_SERVICE_TIME = 0,
39 /* Total bytes transferred */
40 BLKIO_STAT_SERVICE_BYTES,
41 /* Total IOs serviced, post merge */
42 BLKIO_STAT_SERVICED,
43 /* Total time spent waiting in scheduler queue in ns */
44 BLKIO_STAT_WAIT_TIME,
Divyesh Shah812d4022010-04-08 21:14:23 -070045 /* Number of IOs merged */
46 BLKIO_STAT_MERGED,
Divyesh Shahcdc11842010-04-08 21:15:10 -070047 /* Number of IOs queued up */
48 BLKIO_STAT_QUEUED,
Divyesh Shah84c124d2010-04-09 08:31:19 +020049 /* All the single valued stats go below this */
50 BLKIO_STAT_TIME,
51 BLKIO_STAT_SECTORS,
Justin TerAvest167400d2011-03-12 16:54:00 +010052 /* Time not charged to this cgroup */
53 BLKIO_STAT_UNACCOUNTED_TIME,
Divyesh Shah84c124d2010-04-09 08:31:19 +020054#ifdef CONFIG_DEBUG_BLK_CGROUP
Divyesh Shahcdc11842010-04-08 21:15:10 -070055 BLKIO_STAT_AVG_QUEUE_SIZE,
Divyesh Shah812df482010-04-08 21:15:35 -070056 BLKIO_STAT_IDLE_TIME,
57 BLKIO_STAT_EMPTY_TIME,
58 BLKIO_STAT_GROUP_WAIT_TIME,
Divyesh Shah84c124d2010-04-09 08:31:19 +020059 BLKIO_STAT_DEQUEUE
60#endif
61};
62
63enum stat_sub_type {
64 BLKIO_STAT_READ = 0,
65 BLKIO_STAT_WRITE,
66 BLKIO_STAT_SYNC,
67 BLKIO_STAT_ASYNC,
68 BLKIO_STAT_TOTAL
Divyesh Shah303a3ac2010-04-01 15:01:24 -070069};
70
Divyesh Shah812df482010-04-08 21:15:35 -070071/* blkg state flags */
72enum blkg_state_flags {
73 BLKG_waiting = 0,
74 BLKG_idling,
75 BLKG_empty,
76};
77
Vivek Goyal062a6442010-09-15 17:06:33 -040078/* cgroup files owned by proportional weight policy */
79enum blkcg_file_name_prop {
80 BLKIO_PROP_weight = 1,
81 BLKIO_PROP_weight_device,
82 BLKIO_PROP_io_service_bytes,
83 BLKIO_PROP_io_serviced,
84 BLKIO_PROP_time,
85 BLKIO_PROP_sectors,
Justin TerAvest167400d2011-03-12 16:54:00 +010086 BLKIO_PROP_unaccounted_time,
Vivek Goyal062a6442010-09-15 17:06:33 -040087 BLKIO_PROP_io_service_time,
88 BLKIO_PROP_io_wait_time,
89 BLKIO_PROP_io_merged,
90 BLKIO_PROP_io_queued,
91 BLKIO_PROP_avg_queue_size,
92 BLKIO_PROP_group_wait_time,
93 BLKIO_PROP_idle_time,
94 BLKIO_PROP_empty_time,
95 BLKIO_PROP_dequeue,
96};
97
Vivek Goyal4c9eefa2010-09-15 17:06:34 -040098/* cgroup files owned by throttle policy */
99enum blkcg_file_name_throtl {
100 BLKIO_THROTL_read_bps_device,
101 BLKIO_THROTL_write_bps_device,
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400102 BLKIO_THROTL_read_iops_device,
103 BLKIO_THROTL_write_iops_device,
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400104 BLKIO_THROTL_io_service_bytes,
105 BLKIO_THROTL_io_serviced,
106};
107
Vivek Goyal31e4c282009-12-03 12:59:42 -0500108struct blkio_cgroup {
109 struct cgroup_subsys_state css;
110 unsigned int weight;
111 spinlock_t lock;
112 struct hlist_head blkg_list;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800113 struct list_head policy_list; /* list of blkio_policy_node */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500114};
115
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700116struct blkio_group_stats {
117 /* total disk time and nr sectors dispatched by this group */
118 uint64_t time;
119 uint64_t sectors;
Justin TerAvest167400d2011-03-12 16:54:00 +0100120 /* Time not charged to this cgroup */
121 uint64_t unaccounted_time;
Divyesh Shahcdc11842010-04-08 21:15:10 -0700122 uint64_t stat_arr[BLKIO_STAT_QUEUED + 1][BLKIO_STAT_TOTAL];
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700123#ifdef CONFIG_DEBUG_BLK_CGROUP
Divyesh Shahcdc11842010-04-08 21:15:10 -0700124 /* Sum of number of IOs queued across all samples */
125 uint64_t avg_queue_size_sum;
126 /* Count of samples taken for average */
127 uint64_t avg_queue_size_samples;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700128 /* How many times this group has been removed from service tree */
129 unsigned long dequeue;
Divyesh Shah812df482010-04-08 21:15:35 -0700130
131 /* Total time spent waiting for it to be assigned a timeslice. */
132 uint64_t group_wait_time;
133 uint64_t start_group_wait_time;
134
135 /* Time spent idling for this blkio_group */
136 uint64_t idle_time;
137 uint64_t start_idle_time;
138 /*
139 * Total time when we have requests queued and do not contain the
140 * current active queue.
141 */
142 uint64_t empty_time;
143 uint64_t start_empty_time;
144 uint16_t flags;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700145#endif
146};
147
Vivek Goyal31e4c282009-12-03 12:59:42 -0500148struct blkio_group {
149 /* An rcu protected unique identifier for the group */
150 void *key;
151 struct hlist_node blkcg_node;
Vivek Goyalb1c35762009-12-03 12:59:47 -0500152 unsigned short blkcg_id;
Vivek Goyal2868ef72009-12-03 12:59:48 -0500153 /* Store cgroup path */
154 char path[128];
Vivek Goyal22084192009-12-03 12:59:49 -0500155 /* The device MKDEV(major, minor), this group has been created for */
Divyesh Shah84c124d2010-04-09 08:31:19 +0200156 dev_t dev;
Vivek Goyal062a6442010-09-15 17:06:33 -0400157 /* policy which owns this blk group */
158 enum blkio_policy_id plid;
Vivek Goyal22084192009-12-03 12:59:49 -0500159
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700160 /* Need to serialize the stats in the case of reset/update */
161 spinlock_t stats_lock;
162 struct blkio_group_stats stats;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500163};
164
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800165struct blkio_policy_node {
166 struct list_head node;
167 dev_t dev;
Vivek Goyal062a6442010-09-15 17:06:33 -0400168 /* This node belongs to max bw policy or porportional weight policy */
169 enum blkio_policy_id plid;
170 /* cgroup file to which this rule belongs to */
171 int fileid;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400172
173 union {
174 unsigned int weight;
175 /*
176 * Rate read/write in terms of byptes per second
177 * Whether this rate represents read or write is determined
178 * by file type "fileid".
179 */
180 u64 bps;
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400181 unsigned int iops;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400182 } val;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800183};
184
185extern unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
186 dev_t dev);
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400187extern uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg,
188 dev_t dev);
189extern uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg,
190 dev_t dev);
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400191extern unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg,
192 dev_t dev);
193extern unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg,
194 dev_t dev);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800195
Vivek Goyal3e252062009-12-04 10:36:42 -0500196typedef void (blkio_unlink_group_fn) (void *key, struct blkio_group *blkg);
Vivek Goyalfe071432010-10-01 14:49:49 +0200197
198typedef void (blkio_update_group_weight_fn) (void *key,
199 struct blkio_group *blkg, unsigned int weight);
200typedef void (blkio_update_group_read_bps_fn) (void * key,
201 struct blkio_group *blkg, u64 read_bps);
202typedef void (blkio_update_group_write_bps_fn) (void *key,
203 struct blkio_group *blkg, u64 write_bps);
204typedef void (blkio_update_group_read_iops_fn) (void *key,
205 struct blkio_group *blkg, unsigned int read_iops);
206typedef void (blkio_update_group_write_iops_fn) (void *key,
207 struct blkio_group *blkg, unsigned int write_iops);
Vivek Goyal3e252062009-12-04 10:36:42 -0500208
209struct blkio_policy_ops {
210 blkio_unlink_group_fn *blkio_unlink_group_fn;
211 blkio_update_group_weight_fn *blkio_update_group_weight_fn;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400212 blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn;
213 blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn;
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400214 blkio_update_group_read_iops_fn *blkio_update_group_read_iops_fn;
215 blkio_update_group_write_iops_fn *blkio_update_group_write_iops_fn;
Vivek Goyal3e252062009-12-04 10:36:42 -0500216};
217
218struct blkio_policy_type {
219 struct list_head list;
220 struct blkio_policy_ops ops;
Vivek Goyal062a6442010-09-15 17:06:33 -0400221 enum blkio_policy_id plid;
Vivek Goyal3e252062009-12-04 10:36:42 -0500222};
223
224/* Blkio controller policy registration */
225extern void blkio_policy_register(struct blkio_policy_type *);
226extern void blkio_policy_unregister(struct blkio_policy_type *);
227
Vivek Goyalafc24d42010-04-26 19:27:56 +0200228static inline char *blkg_path(struct blkio_group *blkg)
229{
230 return blkg->path;
231}
232
Jens Axboe2f5ea472009-12-03 21:06:43 +0100233#else
234
235struct blkio_group {
236};
237
Vivek Goyal3e252062009-12-04 10:36:42 -0500238struct blkio_policy_type {
239};
240
241static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { }
242static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
243
Vivek Goyalafc24d42010-04-26 19:27:56 +0200244static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
245
Jens Axboe2f5ea472009-12-03 21:06:43 +0100246#endif
247
Justin TerAvestdf457f82011-03-08 19:45:00 +0100248#define BLKIO_WEIGHT_MIN 10
Vivek Goyal31e4c282009-12-03 12:59:42 -0500249#define BLKIO_WEIGHT_MAX 1000
250#define BLKIO_WEIGHT_DEFAULT 500
251
Vivek Goyal2868ef72009-12-03 12:59:48 -0500252#ifdef CONFIG_DEBUG_BLK_CGROUP
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200253void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg);
Divyesh Shah91952912010-04-01 15:01:41 -0700254void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
Vivek Goyal22084192009-12-03 12:59:49 -0500255 unsigned long dequeue);
Divyesh Shah812df482010-04-08 21:15:35 -0700256void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg);
257void blkiocg_update_idle_time_stats(struct blkio_group *blkg);
Vivek Goyale5ff0822010-04-26 19:25:11 +0200258void blkiocg_set_start_empty_time(struct blkio_group *blkg);
Divyesh Shah812df482010-04-08 21:15:35 -0700259
260#define BLKG_FLAG_FNS(name) \
261static inline void blkio_mark_blkg_##name( \
262 struct blkio_group_stats *stats) \
263{ \
264 stats->flags |= (1 << BLKG_##name); \
265} \
266static inline void blkio_clear_blkg_##name( \
267 struct blkio_group_stats *stats) \
268{ \
269 stats->flags &= ~(1 << BLKG_##name); \
270} \
271static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \
272{ \
273 return (stats->flags & (1 << BLKG_##name)) != 0; \
274} \
275
276BLKG_FLAG_FNS(waiting)
277BLKG_FLAG_FNS(idling)
278BLKG_FLAG_FNS(empty)
279#undef BLKG_FLAG_FNS
Vivek Goyal2868ef72009-12-03 12:59:48 -0500280#else
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200281static inline void blkiocg_update_avg_queue_size_stats(
Divyesh Shahcdc11842010-04-08 21:15:10 -0700282 struct blkio_group *blkg) {}
Divyesh Shah91952912010-04-01 15:01:41 -0700283static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
284 unsigned long dequeue) {}
Divyesh Shah812df482010-04-08 21:15:35 -0700285static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
286{}
287static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg) {}
Vivek Goyale5ff0822010-04-26 19:25:11 +0200288static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg) {}
Vivek Goyal2868ef72009-12-03 12:59:48 -0500289#endif
290
Ben Blum67523c42010-03-10 15:22:11 -0800291#if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500292extern struct blkio_cgroup blkio_root_cgroup;
293extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
294extern void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
Vivek Goyal062a6442010-09-15 17:06:33 -0400295 struct blkio_group *blkg, void *key, dev_t dev,
296 enum blkio_policy_id plid);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500297extern int blkiocg_del_blkio_group(struct blkio_group *blkg);
298extern struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg,
299 void *key);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700300void blkiocg_update_timeslice_used(struct blkio_group *blkg,
Justin TerAvest167400d2011-03-12 16:54:00 +0100301 unsigned long time,
302 unsigned long unaccounted_time);
Divyesh Shah84c124d2010-04-09 08:31:19 +0200303void blkiocg_update_dispatch_stats(struct blkio_group *blkg, uint64_t bytes,
304 bool direction, bool sync);
305void blkiocg_update_completion_stats(struct blkio_group *blkg,
306 uint64_t start_time, uint64_t io_start_time, bool direction, bool sync);
Divyesh Shah812d4022010-04-08 21:14:23 -0700307void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
308 bool sync);
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200309void blkiocg_update_io_add_stats(struct blkio_group *blkg,
Divyesh Shahcdc11842010-04-08 21:15:10 -0700310 struct blkio_group *curr_blkg, bool direction, bool sync);
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200311void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
Divyesh Shahcdc11842010-04-08 21:15:10 -0700312 bool direction, bool sync);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500313#else
Jens Axboe2f5ea472009-12-03 21:06:43 +0100314struct cgroup;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500315static inline struct blkio_cgroup *
316cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
317
318static inline void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
Vivek Goyal062a6442010-09-15 17:06:33 -0400319 struct blkio_group *blkg, void *key, dev_t dev,
320 enum blkio_policy_id plid) {}
Vivek Goyal31e4c282009-12-03 12:59:42 -0500321
322static inline int
323blkiocg_del_blkio_group(struct blkio_group *blkg) { return 0; }
324
325static inline struct blkio_group *
326blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key) { return NULL; }
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700327static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg,
Justin TerAvest167400d2011-03-12 16:54:00 +0100328 unsigned long time,
329 unsigned long unaccounted_time)
330{}
Divyesh Shah84c124d2010-04-09 08:31:19 +0200331static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
332 uint64_t bytes, bool direction, bool sync) {}
333static inline void blkiocg_update_completion_stats(struct blkio_group *blkg,
334 uint64_t start_time, uint64_t io_start_time, bool direction,
335 bool sync) {}
Divyesh Shah812d4022010-04-08 21:14:23 -0700336static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
337 bool direction, bool sync) {}
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200338static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg,
Divyesh Shahcdc11842010-04-08 21:15:10 -0700339 struct blkio_group *curr_blkg, bool direction, bool sync) {}
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200340static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
Divyesh Shahcdc11842010-04-08 21:15:10 -0700341 bool direction, bool sync) {}
Vivek Goyal31e4c282009-12-03 12:59:42 -0500342#endif
343#endif /* _BLK_CGROUP_H */