blob: a71d2904ffb97c1f6f2e9f9d5480d389b0ffccff [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>
Vivek Goyal575969a2011-05-19 15:38:29 -040017#include <linux/u64_stats_sync.h>
Vivek Goyal31e4c282009-12-03 12:59:42 -050018
Vivek Goyal062a6442010-09-15 17:06:33 -040019enum blkio_policy_id {
20 BLKIO_POLICY_PROP = 0, /* Proportional Bandwidth division */
Vivek Goyal4c9eefa2010-09-15 17:06:34 -040021 BLKIO_POLICY_THROTL, /* Throttling */
Vivek Goyal062a6442010-09-15 17:06:33 -040022};
23
Vivek Goyal9355aed2010-10-01 21:16:41 +020024/* Max limits for throttle policy */
25#define THROTL_IOPS_MAX UINT_MAX
26
Ben Blum67523c42010-03-10 15:22:11 -080027#if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
28
29#ifndef CONFIG_BLK_CGROUP
30/* When blk-cgroup is a module, its subsys_id isn't a compile-time constant */
31extern struct cgroup_subsys blkio_subsys;
32#define blkio_subsys_id blkio_subsys.subsys_id
33#endif
Jens Axboe2f5ea472009-12-03 21:06:43 +010034
Divyesh Shah84c124d2010-04-09 08:31:19 +020035enum stat_type {
36 /* Total time spent (in ns) between request dispatch to the driver and
37 * request completion for IOs doen by this cgroup. This may not be
38 * accurate when NCQ is turned on. */
39 BLKIO_STAT_SERVICE_TIME = 0,
Divyesh Shah84c124d2010-04-09 08:31:19 +020040 /* Total time spent waiting in scheduler queue in ns */
41 BLKIO_STAT_WAIT_TIME,
Divyesh Shahcdc11842010-04-08 21:15:10 -070042 /* Number of IOs queued up */
43 BLKIO_STAT_QUEUED,
Divyesh Shah84c124d2010-04-09 08:31:19 +020044 /* All the single valued stats go below this */
45 BLKIO_STAT_TIME,
Vivek Goyala23e6862011-05-19 15:38:20 -040046#ifdef CONFIG_DEBUG_BLK_CGROUP
Justin TerAvest167400d2011-03-12 16:54:00 +010047 /* Time not charged to this cgroup */
48 BLKIO_STAT_UNACCOUNTED_TIME,
Divyesh Shahcdc11842010-04-08 21:15:10 -070049 BLKIO_STAT_AVG_QUEUE_SIZE,
Divyesh Shah812df482010-04-08 21:15:35 -070050 BLKIO_STAT_IDLE_TIME,
51 BLKIO_STAT_EMPTY_TIME,
52 BLKIO_STAT_GROUP_WAIT_TIME,
Divyesh Shah84c124d2010-04-09 08:31:19 +020053 BLKIO_STAT_DEQUEUE
54#endif
55};
56
Vivek Goyal5624a4e2011-05-19 15:38:28 -040057/* Per cpu stats */
58enum stat_type_cpu {
59 BLKIO_STAT_CPU_SECTORS,
60 /* Total bytes transferred */
61 BLKIO_STAT_CPU_SERVICE_BYTES,
62 /* Total IOs serviced, post merge */
63 BLKIO_STAT_CPU_SERVICED,
Vivek Goyal317389a2011-05-23 10:02:19 +020064 /* Number of IOs merged */
65 BLKIO_STAT_CPU_MERGED,
Vivek Goyal5624a4e2011-05-19 15:38:28 -040066 BLKIO_STAT_CPU_NR
67};
68
Divyesh Shah84c124d2010-04-09 08:31:19 +020069enum stat_sub_type {
70 BLKIO_STAT_READ = 0,
71 BLKIO_STAT_WRITE,
72 BLKIO_STAT_SYNC,
73 BLKIO_STAT_ASYNC,
74 BLKIO_STAT_TOTAL
Divyesh Shah303a3ac2010-04-01 15:01:24 -070075};
76
Divyesh Shah812df482010-04-08 21:15:35 -070077/* blkg state flags */
78enum blkg_state_flags {
79 BLKG_waiting = 0,
80 BLKG_idling,
81 BLKG_empty,
82};
83
Vivek Goyal062a6442010-09-15 17:06:33 -040084/* cgroup files owned by proportional weight policy */
85enum blkcg_file_name_prop {
86 BLKIO_PROP_weight = 1,
87 BLKIO_PROP_weight_device,
88 BLKIO_PROP_io_service_bytes,
89 BLKIO_PROP_io_serviced,
90 BLKIO_PROP_time,
91 BLKIO_PROP_sectors,
Justin TerAvest167400d2011-03-12 16:54:00 +010092 BLKIO_PROP_unaccounted_time,
Vivek Goyal062a6442010-09-15 17:06:33 -040093 BLKIO_PROP_io_service_time,
94 BLKIO_PROP_io_wait_time,
95 BLKIO_PROP_io_merged,
96 BLKIO_PROP_io_queued,
97 BLKIO_PROP_avg_queue_size,
98 BLKIO_PROP_group_wait_time,
99 BLKIO_PROP_idle_time,
100 BLKIO_PROP_empty_time,
101 BLKIO_PROP_dequeue,
102};
103
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400104/* cgroup files owned by throttle policy */
105enum blkcg_file_name_throtl {
106 BLKIO_THROTL_read_bps_device,
107 BLKIO_THROTL_write_bps_device,
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400108 BLKIO_THROTL_read_iops_device,
109 BLKIO_THROTL_write_iops_device,
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400110 BLKIO_THROTL_io_service_bytes,
111 BLKIO_THROTL_io_serviced,
112};
113
Vivek Goyal31e4c282009-12-03 12:59:42 -0500114struct blkio_cgroup {
115 struct cgroup_subsys_state css;
116 unsigned int weight;
117 spinlock_t lock;
118 struct hlist_head blkg_list;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800119 struct list_head policy_list; /* list of blkio_policy_node */
Vivek Goyal31e4c282009-12-03 12:59:42 -0500120};
121
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700122struct blkio_group_stats {
123 /* total disk time and nr sectors dispatched by this group */
124 uint64_t time;
Divyesh Shahcdc11842010-04-08 21:15:10 -0700125 uint64_t stat_arr[BLKIO_STAT_QUEUED + 1][BLKIO_STAT_TOTAL];
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700126#ifdef CONFIG_DEBUG_BLK_CGROUP
Vivek Goyala23e6862011-05-19 15:38:20 -0400127 /* Time not charged to this cgroup */
128 uint64_t unaccounted_time;
129
Divyesh Shahcdc11842010-04-08 21:15:10 -0700130 /* Sum of number of IOs queued across all samples */
131 uint64_t avg_queue_size_sum;
132 /* Count of samples taken for average */
133 uint64_t avg_queue_size_samples;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700134 /* How many times this group has been removed from service tree */
135 unsigned long dequeue;
Divyesh Shah812df482010-04-08 21:15:35 -0700136
137 /* Total time spent waiting for it to be assigned a timeslice. */
138 uint64_t group_wait_time;
139 uint64_t start_group_wait_time;
140
141 /* Time spent idling for this blkio_group */
142 uint64_t idle_time;
143 uint64_t start_idle_time;
144 /*
145 * Total time when we have requests queued and do not contain the
146 * current active queue.
147 */
148 uint64_t empty_time;
149 uint64_t start_empty_time;
150 uint16_t flags;
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700151#endif
152};
153
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400154/* Per cpu blkio group stats */
155struct blkio_group_stats_cpu {
156 uint64_t sectors;
157 uint64_t stat_arr_cpu[BLKIO_STAT_CPU_NR][BLKIO_STAT_TOTAL];
Vivek Goyal575969a2011-05-19 15:38:29 -0400158 struct u64_stats_sync syncp;
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400159};
160
Vivek Goyal31e4c282009-12-03 12:59:42 -0500161struct blkio_group {
162 /* An rcu protected unique identifier for the group */
163 void *key;
164 struct hlist_node blkcg_node;
Vivek Goyalb1c35762009-12-03 12:59:47 -0500165 unsigned short blkcg_id;
Vivek Goyal2868ef72009-12-03 12:59:48 -0500166 /* Store cgroup path */
167 char path[128];
Vivek Goyal22084192009-12-03 12:59:49 -0500168 /* The device MKDEV(major, minor), this group has been created for */
Divyesh Shah84c124d2010-04-09 08:31:19 +0200169 dev_t dev;
Vivek Goyal062a6442010-09-15 17:06:33 -0400170 /* policy which owns this blk group */
171 enum blkio_policy_id plid;
Vivek Goyal22084192009-12-03 12:59:49 -0500172
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700173 /* Need to serialize the stats in the case of reset/update */
174 spinlock_t stats_lock;
175 struct blkio_group_stats stats;
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400176 /* Per cpu stats pointer */
177 struct blkio_group_stats_cpu __percpu *stats_cpu;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500178};
179
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800180struct blkio_policy_node {
181 struct list_head node;
182 dev_t dev;
Vivek Goyal062a6442010-09-15 17:06:33 -0400183 /* This node belongs to max bw policy or porportional weight policy */
184 enum blkio_policy_id plid;
185 /* cgroup file to which this rule belongs to */
186 int fileid;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400187
188 union {
189 unsigned int weight;
190 /*
191 * Rate read/write in terms of byptes per second
192 * Whether this rate represents read or write is determined
193 * by file type "fileid".
194 */
195 u64 bps;
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400196 unsigned int iops;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400197 } val;
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800198};
199
200extern unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
201 dev_t dev);
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400202extern uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg,
203 dev_t dev);
204extern uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg,
205 dev_t dev);
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400206extern unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg,
207 dev_t dev);
208extern unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg,
209 dev_t dev);
Gui Jianfeng34d0f172010-04-13 16:05:49 +0800210
Vivek Goyal3e252062009-12-04 10:36:42 -0500211typedef void (blkio_unlink_group_fn) (void *key, struct blkio_group *blkg);
Vivek Goyalfe071432010-10-01 14:49:49 +0200212
213typedef void (blkio_update_group_weight_fn) (void *key,
214 struct blkio_group *blkg, unsigned int weight);
215typedef void (blkio_update_group_read_bps_fn) (void * key,
216 struct blkio_group *blkg, u64 read_bps);
217typedef void (blkio_update_group_write_bps_fn) (void *key,
218 struct blkio_group *blkg, u64 write_bps);
219typedef void (blkio_update_group_read_iops_fn) (void *key,
220 struct blkio_group *blkg, unsigned int read_iops);
221typedef void (blkio_update_group_write_iops_fn) (void *key,
222 struct blkio_group *blkg, unsigned int write_iops);
Vivek Goyal3e252062009-12-04 10:36:42 -0500223
224struct blkio_policy_ops {
225 blkio_unlink_group_fn *blkio_unlink_group_fn;
226 blkio_update_group_weight_fn *blkio_update_group_weight_fn;
Vivek Goyal4c9eefa2010-09-15 17:06:34 -0400227 blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn;
228 blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn;
Vivek Goyal7702e8f2010-09-15 17:06:36 -0400229 blkio_update_group_read_iops_fn *blkio_update_group_read_iops_fn;
230 blkio_update_group_write_iops_fn *blkio_update_group_write_iops_fn;
Vivek Goyal3e252062009-12-04 10:36:42 -0500231};
232
233struct blkio_policy_type {
234 struct list_head list;
235 struct blkio_policy_ops ops;
Vivek Goyal062a6442010-09-15 17:06:33 -0400236 enum blkio_policy_id plid;
Vivek Goyal3e252062009-12-04 10:36:42 -0500237};
238
239/* Blkio controller policy registration */
240extern void blkio_policy_register(struct blkio_policy_type *);
241extern void blkio_policy_unregister(struct blkio_policy_type *);
242
Vivek Goyalafc24d42010-04-26 19:27:56 +0200243static inline char *blkg_path(struct blkio_group *blkg)
244{
245 return blkg->path;
246}
247
Jens Axboe2f5ea472009-12-03 21:06:43 +0100248#else
249
250struct blkio_group {
251};
252
Vivek Goyal3e252062009-12-04 10:36:42 -0500253struct blkio_policy_type {
254};
255
256static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { }
257static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
258
Vivek Goyalafc24d42010-04-26 19:27:56 +0200259static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
260
Jens Axboe2f5ea472009-12-03 21:06:43 +0100261#endif
262
Justin TerAvestdf457f82011-03-08 19:45:00 +0100263#define BLKIO_WEIGHT_MIN 10
Vivek Goyal31e4c282009-12-03 12:59:42 -0500264#define BLKIO_WEIGHT_MAX 1000
265#define BLKIO_WEIGHT_DEFAULT 500
266
Vivek Goyal2868ef72009-12-03 12:59:48 -0500267#ifdef CONFIG_DEBUG_BLK_CGROUP
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200268void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg);
Divyesh Shah91952912010-04-01 15:01:41 -0700269void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
Vivek Goyal22084192009-12-03 12:59:49 -0500270 unsigned long dequeue);
Divyesh Shah812df482010-04-08 21:15:35 -0700271void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg);
272void blkiocg_update_idle_time_stats(struct blkio_group *blkg);
Vivek Goyale5ff0822010-04-26 19:25:11 +0200273void blkiocg_set_start_empty_time(struct blkio_group *blkg);
Divyesh Shah812df482010-04-08 21:15:35 -0700274
275#define BLKG_FLAG_FNS(name) \
276static inline void blkio_mark_blkg_##name( \
277 struct blkio_group_stats *stats) \
278{ \
279 stats->flags |= (1 << BLKG_##name); \
280} \
281static inline void blkio_clear_blkg_##name( \
282 struct blkio_group_stats *stats) \
283{ \
284 stats->flags &= ~(1 << BLKG_##name); \
285} \
286static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \
287{ \
288 return (stats->flags & (1 << BLKG_##name)) != 0; \
289} \
290
291BLKG_FLAG_FNS(waiting)
292BLKG_FLAG_FNS(idling)
293BLKG_FLAG_FNS(empty)
294#undef BLKG_FLAG_FNS
Vivek Goyal2868ef72009-12-03 12:59:48 -0500295#else
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200296static inline void blkiocg_update_avg_queue_size_stats(
Divyesh Shahcdc11842010-04-08 21:15:10 -0700297 struct blkio_group *blkg) {}
Divyesh Shah91952912010-04-01 15:01:41 -0700298static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
299 unsigned long dequeue) {}
Divyesh Shah812df482010-04-08 21:15:35 -0700300static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
301{}
302static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg) {}
Vivek Goyale5ff0822010-04-26 19:25:11 +0200303static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg) {}
Vivek Goyal2868ef72009-12-03 12:59:48 -0500304#endif
305
Ben Blum67523c42010-03-10 15:22:11 -0800306#if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
Vivek Goyal31e4c282009-12-03 12:59:42 -0500307extern struct blkio_cgroup blkio_root_cgroup;
308extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
Vivek Goyal70087dc2011-05-16 15:24:08 +0200309extern struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500310extern void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
Vivek Goyal062a6442010-09-15 17:06:33 -0400311 struct blkio_group *blkg, void *key, dev_t dev,
312 enum blkio_policy_id plid);
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400313extern int blkio_alloc_blkg_stats(struct blkio_group *blkg);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500314extern int blkiocg_del_blkio_group(struct blkio_group *blkg);
315extern struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg,
316 void *key);
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700317void blkiocg_update_timeslice_used(struct blkio_group *blkg,
Justin TerAvest167400d2011-03-12 16:54:00 +0100318 unsigned long time,
319 unsigned long unaccounted_time);
Divyesh Shah84c124d2010-04-09 08:31:19 +0200320void blkiocg_update_dispatch_stats(struct blkio_group *blkg, uint64_t bytes,
321 bool direction, bool sync);
322void blkiocg_update_completion_stats(struct blkio_group *blkg,
323 uint64_t start_time, uint64_t io_start_time, bool direction, bool sync);
Divyesh Shah812d4022010-04-08 21:14:23 -0700324void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
325 bool sync);
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200326void blkiocg_update_io_add_stats(struct blkio_group *blkg,
Divyesh Shahcdc11842010-04-08 21:15:10 -0700327 struct blkio_group *curr_blkg, bool direction, bool sync);
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200328void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
Divyesh Shahcdc11842010-04-08 21:15:10 -0700329 bool direction, bool sync);
Vivek Goyal31e4c282009-12-03 12:59:42 -0500330#else
Jens Axboe2f5ea472009-12-03 21:06:43 +0100331struct cgroup;
Vivek Goyal31e4c282009-12-03 12:59:42 -0500332static inline struct blkio_cgroup *
333cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
Vivek Goyal70087dc2011-05-16 15:24:08 +0200334static inline struct blkio_cgroup *
335task_blkio_cgroup(struct task_struct *tsk) { return NULL; }
Vivek Goyal31e4c282009-12-03 12:59:42 -0500336
337static inline void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
Vivek Goyal062a6442010-09-15 17:06:33 -0400338 struct blkio_group *blkg, void *key, dev_t dev,
339 enum blkio_policy_id plid) {}
Vivek Goyal31e4c282009-12-03 12:59:42 -0500340
Vivek Goyal5624a4e2011-05-19 15:38:28 -0400341static inline int blkio_alloc_blkg_stats(struct blkio_group *blkg) { return 0; }
342
Vivek Goyal31e4c282009-12-03 12:59:42 -0500343static inline int
344blkiocg_del_blkio_group(struct blkio_group *blkg) { return 0; }
345
346static inline struct blkio_group *
347blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key) { return NULL; }
Divyesh Shah303a3ac2010-04-01 15:01:24 -0700348static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg,
Justin TerAvest167400d2011-03-12 16:54:00 +0100349 unsigned long time,
350 unsigned long unaccounted_time)
351{}
Divyesh Shah84c124d2010-04-09 08:31:19 +0200352static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
353 uint64_t bytes, bool direction, bool sync) {}
354static inline void blkiocg_update_completion_stats(struct blkio_group *blkg,
355 uint64_t start_time, uint64_t io_start_time, bool direction,
356 bool sync) {}
Divyesh Shah812d4022010-04-08 21:14:23 -0700357static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
358 bool direction, bool sync) {}
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200359static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg,
Divyesh Shahcdc11842010-04-08 21:15:10 -0700360 struct blkio_group *curr_blkg, bool direction, bool sync) {}
Divyesh Shaha11cdaa2010-04-13 19:59:17 +0200361static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
Divyesh Shahcdc11842010-04-08 21:15:10 -0700362 bool direction, bool sync) {}
Vivek Goyal31e4c282009-12-03 12:59:42 -0500363#endif
364#endif /* _BLK_CGROUP_H */