]>
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 <[email protected]> | |
8 | * | |
9 | * Copyright (C) 2008 Fabio Checconi <[email protected]> | |
10 | * Paolo Valente <[email protected]> | |
11 | * | |
12 | * Copyright (C) 2009 Vivek Goyal <[email protected]> | |
13 | * Nauman Rafique <[email protected]> | |
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 | ||
67523c48 BB |
27 | #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 */ | |
31 | extern struct cgroup_subsys blkio_subsys; | |
32 | #define blkio_subsys_id blkio_subsys.subsys_id | |
33 | #endif | |
2f5ea477 | 34 | |
84c124da DS |
35 | enum 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, | |
84c124da DS |
40 | /* Total time spent waiting in scheduler queue in ns */ |
41 | BLKIO_STAT_WAIT_TIME, | |
cdc1184c DS |
42 | /* Number of IOs queued up */ |
43 | BLKIO_STAT_QUEUED, | |
84c124da DS |
44 | /* All the single valued stats go below this */ |
45 | BLKIO_STAT_TIME, | |
a23e6869 | 46 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
167400d3 JT |
47 | /* Time not charged to this cgroup */ |
48 | BLKIO_STAT_UNACCOUNTED_TIME, | |
cdc1184c | 49 | BLKIO_STAT_AVG_QUEUE_SIZE, |
812df48d DS |
50 | BLKIO_STAT_IDLE_TIME, |
51 | BLKIO_STAT_EMPTY_TIME, | |
52 | BLKIO_STAT_GROUP_WAIT_TIME, | |
84c124da DS |
53 | BLKIO_STAT_DEQUEUE |
54 | #endif | |
55 | }; | |
56 | ||
5624a4e4 VG |
57 | /* Per cpu stats */ |
58 | enum 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, | |
317389a7 VG |
64 | /* Number of IOs merged */ |
65 | BLKIO_STAT_CPU_MERGED, | |
5624a4e4 VG |
66 | BLKIO_STAT_CPU_NR |
67 | }; | |
68 | ||
84c124da DS |
69 | enum stat_sub_type { |
70 | BLKIO_STAT_READ = 0, | |
71 | BLKIO_STAT_WRITE, | |
72 | BLKIO_STAT_SYNC, | |
73 | BLKIO_STAT_ASYNC, | |
74 | BLKIO_STAT_TOTAL | |
303a3acb DS |
75 | }; |
76 | ||
812df48d DS |
77 | /* blkg state flags */ |
78 | enum blkg_state_flags { | |
79 | BLKG_waiting = 0, | |
80 | BLKG_idling, | |
81 | BLKG_empty, | |
82 | }; | |
83 | ||
062a644d VG |
84 | /* cgroup files owned by proportional weight policy */ |
85 | enum 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, | |
167400d3 | 92 | BLKIO_PROP_unaccounted_time, |
062a644d VG |
93 | 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 | ||
4c9eefa1 VG |
104 | /* cgroup files owned by throttle policy */ |
105 | enum blkcg_file_name_throtl { | |
106 | BLKIO_THROTL_read_bps_device, | |
107 | BLKIO_THROTL_write_bps_device, | |
7702e8f4 VG |
108 | BLKIO_THROTL_read_iops_device, |
109 | BLKIO_THROTL_write_iops_device, | |
4c9eefa1 VG |
110 | BLKIO_THROTL_io_service_bytes, |
111 | BLKIO_THROTL_io_serviced, | |
112 | }; | |
113 | ||
31e4c28d VG |
114 | struct blkio_cgroup { |
115 | struct cgroup_subsys_state css; | |
116 | unsigned int weight; | |
117 | spinlock_t lock; | |
118 | struct hlist_head blkg_list; | |
34d0f179 | 119 | struct list_head policy_list; /* list of blkio_policy_node */ |
31e4c28d VG |
120 | }; |
121 | ||
303a3acb DS |
122 | struct blkio_group_stats { |
123 | /* total disk time and nr sectors dispatched by this group */ | |
124 | uint64_t time; | |
cdc1184c | 125 | uint64_t stat_arr[BLKIO_STAT_QUEUED + 1][BLKIO_STAT_TOTAL]; |
303a3acb | 126 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
a23e6869 VG |
127 | /* Time not charged to this cgroup */ |
128 | uint64_t unaccounted_time; | |
129 | ||
cdc1184c DS |
130 | /* 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; | |
303a3acb DS |
134 | /* How many times this group has been removed from service tree */ |
135 | unsigned long dequeue; | |
812df48d DS |
136 | |
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; | |
303a3acb DS |
151 | #endif |
152 | }; | |
153 | ||
5624a4e4 VG |
154 | /* Per cpu blkio group stats */ |
155 | struct blkio_group_stats_cpu { | |
156 | uint64_t sectors; | |
157 | uint64_t stat_arr_cpu[BLKIO_STAT_CPU_NR][BLKIO_STAT_TOTAL]; | |
575969a0 | 158 | struct u64_stats_sync syncp; |
5624a4e4 VG |
159 | }; |
160 | ||
31e4c28d VG |
161 | struct blkio_group { |
162 | /* An rcu protected unique identifier for the group */ | |
163 | void *key; | |
164 | struct hlist_node blkcg_node; | |
b1c35769 | 165 | unsigned short blkcg_id; |
2868ef7b VG |
166 | /* Store cgroup path */ |
167 | char path[128]; | |
22084190 | 168 | /* The device MKDEV(major, minor), this group has been created for */ |
84c124da | 169 | dev_t dev; |
062a644d VG |
170 | /* policy which owns this blk group */ |
171 | enum blkio_policy_id plid; | |
22084190 | 172 | |
303a3acb DS |
173 | /* Need to serialize the stats in the case of reset/update */ |
174 | spinlock_t stats_lock; | |
175 | struct blkio_group_stats stats; | |
5624a4e4 VG |
176 | /* Per cpu stats pointer */ |
177 | struct blkio_group_stats_cpu __percpu *stats_cpu; | |
31e4c28d VG |
178 | }; |
179 | ||
34d0f179 GJ |
180 | struct blkio_policy_node { |
181 | struct list_head node; | |
182 | dev_t dev; | |
062a644d VG |
183 | /* 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; | |
4c9eefa1 VG |
187 | |
188 | union { | |
189 | unsigned int weight; | |
190 | /* | |
e890413a | 191 | * Rate read/write in terms of bytes per second |
4c9eefa1 VG |
192 | * Whether this rate represents read or write is determined |
193 | * by file type "fileid". | |
194 | */ | |
195 | u64 bps; | |
7702e8f4 | 196 | unsigned int iops; |
4c9eefa1 | 197 | } val; |
34d0f179 GJ |
198 | }; |
199 | ||
200 | extern unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg, | |
201 | dev_t dev); | |
4c9eefa1 VG |
202 | extern uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, |
203 | dev_t dev); | |
204 | extern uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, | |
205 | dev_t dev); | |
7702e8f4 VG |
206 | extern unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, |
207 | dev_t dev); | |
208 | extern unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, | |
209 | dev_t dev); | |
34d0f179 | 210 | |
3e252066 | 211 | typedef void (blkio_unlink_group_fn) (void *key, struct blkio_group *blkg); |
fe071437 VG |
212 | |
213 | typedef void (blkio_update_group_weight_fn) (void *key, | |
214 | struct blkio_group *blkg, unsigned int weight); | |
215 | typedef void (blkio_update_group_read_bps_fn) (void * key, | |
216 | struct blkio_group *blkg, u64 read_bps); | |
217 | typedef void (blkio_update_group_write_bps_fn) (void *key, | |
218 | struct blkio_group *blkg, u64 write_bps); | |
219 | typedef void (blkio_update_group_read_iops_fn) (void *key, | |
220 | struct blkio_group *blkg, unsigned int read_iops); | |
221 | typedef void (blkio_update_group_write_iops_fn) (void *key, | |
222 | struct blkio_group *blkg, unsigned int write_iops); | |
3e252066 VG |
223 | |
224 | struct blkio_policy_ops { | |
225 | blkio_unlink_group_fn *blkio_unlink_group_fn; | |
226 | blkio_update_group_weight_fn *blkio_update_group_weight_fn; | |
4c9eefa1 VG |
227 | 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; | |
7702e8f4 VG |
229 | 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; | |
3e252066 VG |
231 | }; |
232 | ||
233 | struct blkio_policy_type { | |
234 | struct list_head list; | |
235 | struct blkio_policy_ops ops; | |
062a644d | 236 | enum blkio_policy_id plid; |
3e252066 VG |
237 | }; |
238 | ||
239 | /* Blkio controller policy registration */ | |
240 | extern void blkio_policy_register(struct blkio_policy_type *); | |
241 | extern void blkio_policy_unregister(struct blkio_policy_type *); | |
242 | ||
afc24d49 VG |
243 | static inline char *blkg_path(struct blkio_group *blkg) |
244 | { | |
245 | return blkg->path; | |
246 | } | |
247 | ||
2f5ea477 JA |
248 | #else |
249 | ||
250 | struct blkio_group { | |
251 | }; | |
252 | ||
3e252066 VG |
253 | struct blkio_policy_type { |
254 | }; | |
255 | ||
256 | static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { } | |
257 | static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { } | |
258 | ||
afc24d49 VG |
259 | static inline char *blkg_path(struct blkio_group *blkg) { return NULL; } |
260 | ||
2f5ea477 JA |
261 | #endif |
262 | ||
df457f84 | 263 | #define BLKIO_WEIGHT_MIN 10 |
31e4c28d VG |
264 | #define BLKIO_WEIGHT_MAX 1000 |
265 | #define BLKIO_WEIGHT_DEFAULT 500 | |
266 | ||
2868ef7b | 267 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
a11cdaa7 | 268 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg); |
9195291e | 269 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
22084190 | 270 | unsigned long dequeue); |
812df48d DS |
271 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg); |
272 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg); | |
e5ff082e | 273 | void blkiocg_set_start_empty_time(struct blkio_group *blkg); |
812df48d DS |
274 | |
275 | #define BLKG_FLAG_FNS(name) \ | |
276 | static inline void blkio_mark_blkg_##name( \ | |
277 | struct blkio_group_stats *stats) \ | |
278 | { \ | |
279 | stats->flags |= (1 << BLKG_##name); \ | |
280 | } \ | |
281 | static inline void blkio_clear_blkg_##name( \ | |
282 | struct blkio_group_stats *stats) \ | |
283 | { \ | |
284 | stats->flags &= ~(1 << BLKG_##name); \ | |
285 | } \ | |
286 | static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \ | |
287 | { \ | |
288 | return (stats->flags & (1 << BLKG_##name)) != 0; \ | |
289 | } \ | |
290 | ||
291 | BLKG_FLAG_FNS(waiting) | |
292 | BLKG_FLAG_FNS(idling) | |
293 | BLKG_FLAG_FNS(empty) | |
294 | #undef BLKG_FLAG_FNS | |
2868ef7b | 295 | #else |
a11cdaa7 | 296 | static inline void blkiocg_update_avg_queue_size_stats( |
cdc1184c | 297 | struct blkio_group *blkg) {} |
9195291e DS |
298 | static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
299 | unsigned long dequeue) {} | |
812df48d DS |
300 | static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg) |
301 | {} | |
302 | static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg) {} | |
e5ff082e | 303 | static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg) {} |
2868ef7b VG |
304 | #endif |
305 | ||
67523c48 | 306 | #if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE) |
31e4c28d VG |
307 | extern struct blkio_cgroup blkio_root_cgroup; |
308 | extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup); | |
70087dc3 | 309 | extern struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk); |
31e4c28d | 310 | extern void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg, |
062a644d VG |
311 | struct blkio_group *blkg, void *key, dev_t dev, |
312 | enum blkio_policy_id plid); | |
5624a4e4 | 313 | extern int blkio_alloc_blkg_stats(struct blkio_group *blkg); |
31e4c28d VG |
314 | extern int blkiocg_del_blkio_group(struct blkio_group *blkg); |
315 | extern struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg, | |
316 | void *key); | |
303a3acb | 317 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, |
167400d3 JT |
318 | unsigned long time, |
319 | unsigned long unaccounted_time); | |
84c124da DS |
320 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, uint64_t bytes, |
321 | bool direction, bool sync); | |
322 | void blkiocg_update_completion_stats(struct blkio_group *blkg, | |
323 | uint64_t start_time, uint64_t io_start_time, bool direction, bool sync); | |
812d4026 DS |
324 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction, |
325 | bool sync); | |
a11cdaa7 | 326 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c | 327 | struct blkio_group *curr_blkg, bool direction, bool sync); |
a11cdaa7 | 328 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c | 329 | bool direction, bool sync); |
31e4c28d | 330 | #else |
2f5ea477 | 331 | struct cgroup; |
31e4c28d VG |
332 | static inline struct blkio_cgroup * |
333 | cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; } | |
70087dc3 VG |
334 | static inline struct blkio_cgroup * |
335 | task_blkio_cgroup(struct task_struct *tsk) { return NULL; } | |
31e4c28d VG |
336 | |
337 | static inline void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg, | |
062a644d VG |
338 | struct blkio_group *blkg, void *key, dev_t dev, |
339 | enum blkio_policy_id plid) {} | |
31e4c28d | 340 | |
5624a4e4 VG |
341 | static inline int blkio_alloc_blkg_stats(struct blkio_group *blkg) { return 0; } |
342 | ||
31e4c28d VG |
343 | static inline int |
344 | blkiocg_del_blkio_group(struct blkio_group *blkg) { return 0; } | |
345 | ||
346 | static inline struct blkio_group * | |
347 | blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key) { return NULL; } | |
303a3acb | 348 | static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg, |
167400d3 JT |
349 | unsigned long time, |
350 | unsigned long unaccounted_time) | |
351 | {} | |
84c124da DS |
352 | static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
353 | uint64_t bytes, bool direction, bool sync) {} | |
354 | static 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) {} | |
812d4026 DS |
357 | static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg, |
358 | bool direction, bool sync) {} | |
a11cdaa7 | 359 | static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c | 360 | struct blkio_group *curr_blkg, bool direction, bool sync) {} |
a11cdaa7 | 361 | static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c | 362 | bool direction, bool sync) {} |
31e4c28d VG |
363 | #endif |
364 | #endif /* _BLK_CGROUP_H */ |