Commit | Line | Data |
---|---|---|
31e4c28d VG |
1 | #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> | |
575969a0 | 17 | #include <linux/u64_stats_sync.h> |
31e4c28d | 18 | |
062a644d VG |
19 | enum blkio_policy_id { |
20 | BLKIO_POLICY_PROP = 0, /* Proportional Bandwidth division */ | |
4c9eefa1 | 21 | BLKIO_POLICY_THROTL, /* Throttling */ |
062a644d VG |
22 | }; |
23 | ||
9355aede VG |
24 | /* Max limits for throttle policy */ |
25 | #define THROTL_IOPS_MAX UINT_MAX | |
26 | ||
32e380ae | 27 | #ifdef CONFIG_BLK_CGROUP |
2f5ea477 | 28 | |
84c124da DS |
29 | enum stat_type { |
30 | /* Total time spent (in ns) between request dispatch to the driver and | |
31 | * request completion for IOs doen by this cgroup. This may not be | |
32 | * accurate when NCQ is turned on. */ | |
33 | BLKIO_STAT_SERVICE_TIME = 0, | |
84c124da DS |
34 | /* Total time spent waiting in scheduler queue in ns */ |
35 | BLKIO_STAT_WAIT_TIME, | |
cdc1184c DS |
36 | /* Number of IOs queued up */ |
37 | BLKIO_STAT_QUEUED, | |
84c124da DS |
38 | /* All the single valued stats go below this */ |
39 | BLKIO_STAT_TIME, | |
a23e6869 | 40 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
167400d3 JT |
41 | /* Time not charged to this cgroup */ |
42 | BLKIO_STAT_UNACCOUNTED_TIME, | |
cdc1184c | 43 | BLKIO_STAT_AVG_QUEUE_SIZE, |
812df48d DS |
44 | BLKIO_STAT_IDLE_TIME, |
45 | BLKIO_STAT_EMPTY_TIME, | |
46 | BLKIO_STAT_GROUP_WAIT_TIME, | |
84c124da DS |
47 | BLKIO_STAT_DEQUEUE |
48 | #endif | |
49 | }; | |
50 | ||
5624a4e4 VG |
51 | /* Per cpu stats */ |
52 | enum stat_type_cpu { | |
53 | BLKIO_STAT_CPU_SECTORS, | |
54 | /* Total bytes transferred */ | |
55 | BLKIO_STAT_CPU_SERVICE_BYTES, | |
56 | /* Total IOs serviced, post merge */ | |
57 | BLKIO_STAT_CPU_SERVICED, | |
317389a7 VG |
58 | /* Number of IOs merged */ |
59 | BLKIO_STAT_CPU_MERGED, | |
5624a4e4 VG |
60 | BLKIO_STAT_CPU_NR |
61 | }; | |
62 | ||
84c124da DS |
63 | enum stat_sub_type { |
64 | BLKIO_STAT_READ = 0, | |
65 | BLKIO_STAT_WRITE, | |
66 | BLKIO_STAT_SYNC, | |
67 | BLKIO_STAT_ASYNC, | |
68 | BLKIO_STAT_TOTAL | |
303a3acb DS |
69 | }; |
70 | ||
812df48d DS |
71 | /* blkg state flags */ |
72 | enum blkg_state_flags { | |
73 | BLKG_waiting = 0, | |
74 | BLKG_idling, | |
75 | BLKG_empty, | |
76 | }; | |
77 | ||
062a644d VG |
78 | /* cgroup files owned by proportional weight policy */ |
79 | enum 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, | |
167400d3 | 86 | BLKIO_PROP_unaccounted_time, |
062a644d VG |
87 | 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 | ||
4c9eefa1 VG |
98 | /* cgroup files owned by throttle policy */ |
99 | enum blkcg_file_name_throtl { | |
100 | BLKIO_THROTL_read_bps_device, | |
101 | BLKIO_THROTL_write_bps_device, | |
7702e8f4 VG |
102 | BLKIO_THROTL_read_iops_device, |
103 | BLKIO_THROTL_write_iops_device, | |
4c9eefa1 VG |
104 | BLKIO_THROTL_io_service_bytes, |
105 | BLKIO_THROTL_io_serviced, | |
106 | }; | |
107 | ||
31e4c28d VG |
108 | struct blkio_cgroup { |
109 | struct cgroup_subsys_state css; | |
110 | unsigned int weight; | |
111 | spinlock_t lock; | |
112 | struct hlist_head blkg_list; | |
34d0f179 | 113 | struct list_head policy_list; /* list of blkio_policy_node */ |
31e4c28d VG |
114 | }; |
115 | ||
303a3acb DS |
116 | struct blkio_group_stats { |
117 | /* total disk time and nr sectors dispatched by this group */ | |
118 | uint64_t time; | |
cdc1184c | 119 | uint64_t stat_arr[BLKIO_STAT_QUEUED + 1][BLKIO_STAT_TOTAL]; |
303a3acb | 120 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
a23e6869 VG |
121 | /* Time not charged to this cgroup */ |
122 | uint64_t unaccounted_time; | |
123 | ||
cdc1184c DS |
124 | /* 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; | |
303a3acb DS |
128 | /* How many times this group has been removed from service tree */ |
129 | unsigned long dequeue; | |
812df48d DS |
130 | |
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; | |
303a3acb DS |
145 | #endif |
146 | }; | |
147 | ||
5624a4e4 VG |
148 | /* Per cpu blkio group stats */ |
149 | struct blkio_group_stats_cpu { | |
150 | uint64_t sectors; | |
151 | uint64_t stat_arr_cpu[BLKIO_STAT_CPU_NR][BLKIO_STAT_TOTAL]; | |
575969a0 | 152 | struct u64_stats_sync syncp; |
5624a4e4 VG |
153 | }; |
154 | ||
31e4c28d VG |
155 | struct blkio_group { |
156 | /* An rcu protected unique identifier for the group */ | |
157 | void *key; | |
158 | struct hlist_node blkcg_node; | |
b1c35769 | 159 | unsigned short blkcg_id; |
2868ef7b VG |
160 | /* Store cgroup path */ |
161 | char path[128]; | |
22084190 | 162 | /* The device MKDEV(major, minor), this group has been created for */ |
84c124da | 163 | dev_t dev; |
062a644d VG |
164 | /* policy which owns this blk group */ |
165 | enum blkio_policy_id plid; | |
22084190 | 166 | |
303a3acb DS |
167 | /* Need to serialize the stats in the case of reset/update */ |
168 | spinlock_t stats_lock; | |
169 | struct blkio_group_stats stats; | |
5624a4e4 VG |
170 | /* Per cpu stats pointer */ |
171 | struct blkio_group_stats_cpu __percpu *stats_cpu; | |
31e4c28d VG |
172 | }; |
173 | ||
34d0f179 GJ |
174 | struct blkio_policy_node { |
175 | struct list_head node; | |
176 | dev_t dev; | |
062a644d VG |
177 | /* This node belongs to max bw policy or porportional weight policy */ |
178 | enum blkio_policy_id plid; | |
179 | /* cgroup file to which this rule belongs to */ | |
180 | int fileid; | |
4c9eefa1 VG |
181 | |
182 | union { | |
183 | unsigned int weight; | |
184 | /* | |
e890413a | 185 | * Rate read/write in terms of bytes per second |
4c9eefa1 VG |
186 | * Whether this rate represents read or write is determined |
187 | * by file type "fileid". | |
188 | */ | |
189 | u64 bps; | |
7702e8f4 | 190 | unsigned int iops; |
4c9eefa1 | 191 | } val; |
34d0f179 GJ |
192 | }; |
193 | ||
194 | extern unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg, | |
195 | dev_t dev); | |
4c9eefa1 VG |
196 | extern uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, |
197 | dev_t dev); | |
198 | extern uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, | |
199 | dev_t dev); | |
7702e8f4 VG |
200 | extern unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, |
201 | dev_t dev); | |
202 | extern unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, | |
203 | dev_t dev); | |
34d0f179 | 204 | |
3e252066 | 205 | typedef void (blkio_unlink_group_fn) (void *key, struct blkio_group *blkg); |
72e06c25 | 206 | typedef bool (blkio_clear_queue_fn)(struct request_queue *q); |
fe071437 VG |
207 | typedef void (blkio_update_group_weight_fn) (void *key, |
208 | struct blkio_group *blkg, unsigned int weight); | |
209 | typedef void (blkio_update_group_read_bps_fn) (void * key, | |
210 | struct blkio_group *blkg, u64 read_bps); | |
211 | typedef void (blkio_update_group_write_bps_fn) (void *key, | |
212 | struct blkio_group *blkg, u64 write_bps); | |
213 | typedef void (blkio_update_group_read_iops_fn) (void *key, | |
214 | struct blkio_group *blkg, unsigned int read_iops); | |
215 | typedef void (blkio_update_group_write_iops_fn) (void *key, | |
216 | struct blkio_group *blkg, unsigned int write_iops); | |
3e252066 VG |
217 | |
218 | struct blkio_policy_ops { | |
219 | blkio_unlink_group_fn *blkio_unlink_group_fn; | |
72e06c25 | 220 | blkio_clear_queue_fn *blkio_clear_queue_fn; |
3e252066 | 221 | blkio_update_group_weight_fn *blkio_update_group_weight_fn; |
4c9eefa1 VG |
222 | blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn; |
223 | blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn; | |
7702e8f4 VG |
224 | blkio_update_group_read_iops_fn *blkio_update_group_read_iops_fn; |
225 | blkio_update_group_write_iops_fn *blkio_update_group_write_iops_fn; | |
3e252066 VG |
226 | }; |
227 | ||
228 | struct blkio_policy_type { | |
229 | struct list_head list; | |
230 | struct blkio_policy_ops ops; | |
062a644d | 231 | enum blkio_policy_id plid; |
3e252066 VG |
232 | }; |
233 | ||
234 | /* Blkio controller policy registration */ | |
235 | extern void blkio_policy_register(struct blkio_policy_type *); | |
236 | extern void blkio_policy_unregister(struct blkio_policy_type *); | |
72e06c25 | 237 | extern void blkg_destroy_all(struct request_queue *q); |
3e252066 | 238 | |
afc24d49 VG |
239 | static inline char *blkg_path(struct blkio_group *blkg) |
240 | { | |
241 | return blkg->path; | |
242 | } | |
243 | ||
2f5ea477 JA |
244 | #else |
245 | ||
246 | struct blkio_group { | |
247 | }; | |
248 | ||
3e252066 VG |
249 | struct blkio_policy_type { |
250 | }; | |
251 | ||
252 | static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { } | |
253 | static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { } | |
72e06c25 | 254 | static inline void blkg_destroy_all(struct request_queue *q) { } |
3e252066 | 255 | |
afc24d49 VG |
256 | static inline char *blkg_path(struct blkio_group *blkg) { return NULL; } |
257 | ||
2f5ea477 JA |
258 | #endif |
259 | ||
df457f84 | 260 | #define BLKIO_WEIGHT_MIN 10 |
31e4c28d VG |
261 | #define BLKIO_WEIGHT_MAX 1000 |
262 | #define BLKIO_WEIGHT_DEFAULT 500 | |
263 | ||
2868ef7b | 264 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
a11cdaa7 | 265 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg); |
9195291e | 266 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
22084190 | 267 | unsigned long dequeue); |
812df48d DS |
268 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg); |
269 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg); | |
e5ff082e | 270 | void blkiocg_set_start_empty_time(struct blkio_group *blkg); |
812df48d DS |
271 | |
272 | #define BLKG_FLAG_FNS(name) \ | |
273 | static inline void blkio_mark_blkg_##name( \ | |
274 | struct blkio_group_stats *stats) \ | |
275 | { \ | |
276 | stats->flags |= (1 << BLKG_##name); \ | |
277 | } \ | |
278 | static inline void blkio_clear_blkg_##name( \ | |
279 | struct blkio_group_stats *stats) \ | |
280 | { \ | |
281 | stats->flags &= ~(1 << BLKG_##name); \ | |
282 | } \ | |
283 | static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \ | |
284 | { \ | |
285 | return (stats->flags & (1 << BLKG_##name)) != 0; \ | |
286 | } \ | |
287 | ||
288 | BLKG_FLAG_FNS(waiting) | |
289 | BLKG_FLAG_FNS(idling) | |
290 | BLKG_FLAG_FNS(empty) | |
291 | #undef BLKG_FLAG_FNS | |
2868ef7b | 292 | #else |
a11cdaa7 | 293 | static inline void blkiocg_update_avg_queue_size_stats( |
cdc1184c | 294 | struct blkio_group *blkg) {} |
9195291e DS |
295 | static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
296 | unsigned long dequeue) {} | |
812df48d DS |
297 | static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg) |
298 | {} | |
299 | static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg) {} | |
e5ff082e | 300 | static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg) {} |
2868ef7b VG |
301 | #endif |
302 | ||
32e380ae | 303 | #ifdef CONFIG_BLK_CGROUP |
31e4c28d VG |
304 | extern struct blkio_cgroup blkio_root_cgroup; |
305 | extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup); | |
70087dc3 | 306 | extern struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk); |
31e4c28d | 307 | extern void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg, |
062a644d VG |
308 | struct blkio_group *blkg, void *key, dev_t dev, |
309 | enum blkio_policy_id plid); | |
5624a4e4 | 310 | extern int blkio_alloc_blkg_stats(struct blkio_group *blkg); |
31e4c28d VG |
311 | extern int blkiocg_del_blkio_group(struct blkio_group *blkg); |
312 | extern struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, | |
313 | void *key); | |
303a3acb | 314 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, |
167400d3 JT |
315 | unsigned long time, |
316 | unsigned long unaccounted_time); | |
84c124da DS |
317 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, uint64_t bytes, |
318 | bool direction, bool sync); | |
319 | void blkiocg_update_completion_stats(struct blkio_group *blkg, | |
320 | uint64_t start_time, uint64_t io_start_time, bool direction, bool sync); | |
812d4026 DS |
321 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction, |
322 | bool sync); | |
a11cdaa7 | 323 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c | 324 | struct blkio_group *curr_blkg, bool direction, bool sync); |
a11cdaa7 | 325 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c | 326 | bool direction, bool sync); |
31e4c28d | 327 | #else |
2f5ea477 | 328 | struct cgroup; |
31e4c28d VG |
329 | static inline struct blkio_cgroup * |
330 | cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; } | |
70087dc3 VG |
331 | static inline struct blkio_cgroup * |
332 | task_blkio_cgroup(struct task_struct *tsk) { return NULL; } | |
31e4c28d VG |
333 | |
334 | static inline void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg, | |
062a644d VG |
335 | struct blkio_group *blkg, void *key, dev_t dev, |
336 | enum blkio_policy_id plid) {} | |
31e4c28d | 337 | |
5624a4e4 VG |
338 | static inline int blkio_alloc_blkg_stats(struct blkio_group *blkg) { return 0; } |
339 | ||
31e4c28d VG |
340 | static inline int |
341 | blkiocg_del_blkio_group(struct blkio_group *blkg) { return 0; } | |
342 | ||
343 | static inline struct blkio_group * | |
344 | blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key) { return NULL; } | |
303a3acb | 345 | static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg, |
167400d3 JT |
346 | unsigned long time, |
347 | unsigned long unaccounted_time) | |
348 | {} | |
84c124da DS |
349 | static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
350 | uint64_t bytes, bool direction, bool sync) {} | |
351 | static inline void blkiocg_update_completion_stats(struct blkio_group *blkg, | |
352 | uint64_t start_time, uint64_t io_start_time, bool direction, | |
353 | bool sync) {} | |
812d4026 DS |
354 | static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg, |
355 | bool direction, bool sync) {} | |
a11cdaa7 | 356 | static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c | 357 | struct blkio_group *curr_blkg, bool direction, bool sync) {} |
a11cdaa7 | 358 | static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c | 359 | bool direction, bool sync) {} |
31e4c28d VG |
360 | #endif |
361 | #endif /* _BLK_CGROUP_H */ |