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 */ |
035d10b2 TH |
22 | |
23 | BLKIO_NR_POLICIES, | |
062a644d VG |
24 | }; |
25 | ||
9355aede VG |
26 | /* Max limits for throttle policy */ |
27 | #define THROTL_IOPS_MAX UINT_MAX | |
28 | ||
32e380ae | 29 | #ifdef CONFIG_BLK_CGROUP |
2f5ea477 | 30 | |
84c124da DS |
31 | enum stat_type { |
32 | /* Total time spent (in ns) between request dispatch to the driver and | |
33 | * request completion for IOs doen by this cgroup. This may not be | |
34 | * accurate when NCQ is turned on. */ | |
35 | BLKIO_STAT_SERVICE_TIME = 0, | |
84c124da DS |
36 | /* Total time spent waiting in scheduler queue in ns */ |
37 | BLKIO_STAT_WAIT_TIME, | |
cdc1184c DS |
38 | /* Number of IOs queued up */ |
39 | BLKIO_STAT_QUEUED, | |
84c124da DS |
40 | /* All the single valued stats go below this */ |
41 | BLKIO_STAT_TIME, | |
a23e6869 | 42 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
167400d3 JT |
43 | /* Time not charged to this cgroup */ |
44 | BLKIO_STAT_UNACCOUNTED_TIME, | |
cdc1184c | 45 | BLKIO_STAT_AVG_QUEUE_SIZE, |
812df48d DS |
46 | BLKIO_STAT_IDLE_TIME, |
47 | BLKIO_STAT_EMPTY_TIME, | |
48 | BLKIO_STAT_GROUP_WAIT_TIME, | |
84c124da DS |
49 | BLKIO_STAT_DEQUEUE |
50 | #endif | |
51 | }; | |
52 | ||
5624a4e4 VG |
53 | /* Per cpu stats */ |
54 | enum stat_type_cpu { | |
55 | BLKIO_STAT_CPU_SECTORS, | |
56 | /* Total bytes transferred */ | |
57 | BLKIO_STAT_CPU_SERVICE_BYTES, | |
58 | /* Total IOs serviced, post merge */ | |
59 | BLKIO_STAT_CPU_SERVICED, | |
317389a7 VG |
60 | /* Number of IOs merged */ |
61 | BLKIO_STAT_CPU_MERGED, | |
5624a4e4 VG |
62 | BLKIO_STAT_CPU_NR |
63 | }; | |
64 | ||
84c124da DS |
65 | enum stat_sub_type { |
66 | BLKIO_STAT_READ = 0, | |
67 | BLKIO_STAT_WRITE, | |
68 | BLKIO_STAT_SYNC, | |
69 | BLKIO_STAT_ASYNC, | |
70 | BLKIO_STAT_TOTAL | |
303a3acb DS |
71 | }; |
72 | ||
812df48d DS |
73 | /* blkg state flags */ |
74 | enum blkg_state_flags { | |
75 | BLKG_waiting = 0, | |
76 | BLKG_idling, | |
77 | BLKG_empty, | |
78 | }; | |
79 | ||
062a644d VG |
80 | /* cgroup files owned by proportional weight policy */ |
81 | enum blkcg_file_name_prop { | |
82 | BLKIO_PROP_weight = 1, | |
83 | BLKIO_PROP_weight_device, | |
84 | BLKIO_PROP_io_service_bytes, | |
85 | BLKIO_PROP_io_serviced, | |
86 | BLKIO_PROP_time, | |
87 | BLKIO_PROP_sectors, | |
167400d3 | 88 | BLKIO_PROP_unaccounted_time, |
062a644d VG |
89 | BLKIO_PROP_io_service_time, |
90 | BLKIO_PROP_io_wait_time, | |
91 | BLKIO_PROP_io_merged, | |
92 | BLKIO_PROP_io_queued, | |
93 | BLKIO_PROP_avg_queue_size, | |
94 | BLKIO_PROP_group_wait_time, | |
95 | BLKIO_PROP_idle_time, | |
96 | BLKIO_PROP_empty_time, | |
97 | BLKIO_PROP_dequeue, | |
98 | }; | |
99 | ||
4c9eefa1 VG |
100 | /* cgroup files owned by throttle policy */ |
101 | enum blkcg_file_name_throtl { | |
102 | BLKIO_THROTL_read_bps_device, | |
103 | BLKIO_THROTL_write_bps_device, | |
7702e8f4 VG |
104 | BLKIO_THROTL_read_iops_device, |
105 | BLKIO_THROTL_write_iops_device, | |
4c9eefa1 VG |
106 | BLKIO_THROTL_io_service_bytes, |
107 | BLKIO_THROTL_io_serviced, | |
108 | }; | |
109 | ||
31e4c28d VG |
110 | struct blkio_cgroup { |
111 | struct cgroup_subsys_state css; | |
112 | unsigned int weight; | |
113 | spinlock_t lock; | |
114 | struct hlist_head blkg_list; | |
34d0f179 | 115 | struct list_head policy_list; /* list of blkio_policy_node */ |
31e4c28d VG |
116 | }; |
117 | ||
303a3acb DS |
118 | struct blkio_group_stats { |
119 | /* total disk time and nr sectors dispatched by this group */ | |
120 | uint64_t time; | |
cdc1184c | 121 | uint64_t stat_arr[BLKIO_STAT_QUEUED + 1][BLKIO_STAT_TOTAL]; |
303a3acb | 122 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
a23e6869 VG |
123 | /* Time not charged to this cgroup */ |
124 | uint64_t unaccounted_time; | |
125 | ||
cdc1184c DS |
126 | /* Sum of number of IOs queued across all samples */ |
127 | uint64_t avg_queue_size_sum; | |
128 | /* Count of samples taken for average */ | |
129 | uint64_t avg_queue_size_samples; | |
303a3acb DS |
130 | /* How many times this group has been removed from service tree */ |
131 | unsigned long dequeue; | |
812df48d DS |
132 | |
133 | /* Total time spent waiting for it to be assigned a timeslice. */ | |
134 | uint64_t group_wait_time; | |
135 | uint64_t start_group_wait_time; | |
136 | ||
137 | /* Time spent idling for this blkio_group */ | |
138 | uint64_t idle_time; | |
139 | uint64_t start_idle_time; | |
140 | /* | |
141 | * Total time when we have requests queued and do not contain the | |
142 | * current active queue. | |
143 | */ | |
144 | uint64_t empty_time; | |
145 | uint64_t start_empty_time; | |
146 | uint16_t flags; | |
303a3acb DS |
147 | #endif |
148 | }; | |
149 | ||
5624a4e4 VG |
150 | /* Per cpu blkio group stats */ |
151 | struct blkio_group_stats_cpu { | |
152 | uint64_t sectors; | |
153 | uint64_t stat_arr_cpu[BLKIO_STAT_CPU_NR][BLKIO_STAT_TOTAL]; | |
575969a0 | 154 | struct u64_stats_sync syncp; |
5624a4e4 VG |
155 | }; |
156 | ||
e56da7e2 TH |
157 | struct blkio_group_conf { |
158 | unsigned int weight; | |
159 | unsigned int iops[2]; | |
160 | u64 bps[2]; | |
161 | }; | |
162 | ||
31e4c28d | 163 | struct blkio_group { |
ca32aefc TH |
164 | /* Pointer to the associated request_queue, RCU protected */ |
165 | struct request_queue __rcu *q; | |
31e4c28d | 166 | struct hlist_node blkcg_node; |
b1c35769 | 167 | unsigned short blkcg_id; |
2868ef7b VG |
168 | /* Store cgroup path */ |
169 | char path[128]; | |
22084190 | 170 | /* The device MKDEV(major, minor), this group has been created for */ |
84c124da | 171 | dev_t dev; |
062a644d VG |
172 | /* policy which owns this blk group */ |
173 | enum blkio_policy_id plid; | |
22084190 | 174 | |
e56da7e2 TH |
175 | /* Configuration */ |
176 | struct blkio_group_conf conf; | |
177 | ||
303a3acb DS |
178 | /* Need to serialize the stats in the case of reset/update */ |
179 | spinlock_t stats_lock; | |
180 | struct blkio_group_stats stats; | |
5624a4e4 VG |
181 | /* Per cpu stats pointer */ |
182 | struct blkio_group_stats_cpu __percpu *stats_cpu; | |
31e4c28d VG |
183 | }; |
184 | ||
34d0f179 GJ |
185 | struct blkio_policy_node { |
186 | struct list_head node; | |
187 | dev_t dev; | |
062a644d VG |
188 | /* This node belongs to max bw policy or porportional weight policy */ |
189 | enum blkio_policy_id plid; | |
190 | /* cgroup file to which this rule belongs to */ | |
191 | int fileid; | |
4c9eefa1 VG |
192 | |
193 | union { | |
194 | unsigned int weight; | |
195 | /* | |
e890413a | 196 | * Rate read/write in terms of bytes per second |
4c9eefa1 VG |
197 | * Whether this rate represents read or write is determined |
198 | * by file type "fileid". | |
199 | */ | |
200 | u64 bps; | |
7702e8f4 | 201 | unsigned int iops; |
4c9eefa1 | 202 | } val; |
34d0f179 GJ |
203 | }; |
204 | ||
205 | extern unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg, | |
206 | dev_t dev); | |
4c9eefa1 VG |
207 | extern uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg, |
208 | dev_t dev); | |
209 | extern uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg, | |
210 | dev_t dev); | |
7702e8f4 VG |
211 | extern unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg, |
212 | dev_t dev); | |
213 | extern unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg, | |
214 | dev_t dev); | |
34d0f179 | 215 | |
cd1604fa TH |
216 | typedef struct blkio_group *(blkio_alloc_group_fn)(struct request_queue *q, |
217 | struct blkio_cgroup *blkcg); | |
218 | typedef void (blkio_link_group_fn)(struct request_queue *q, | |
219 | struct blkio_group *blkg); | |
ca32aefc TH |
220 | typedef void (blkio_unlink_group_fn)(struct request_queue *q, |
221 | struct blkio_group *blkg); | |
72e06c25 | 222 | typedef bool (blkio_clear_queue_fn)(struct request_queue *q); |
ca32aefc | 223 | typedef void (blkio_update_group_weight_fn)(struct request_queue *q, |
fe071437 | 224 | struct blkio_group *blkg, unsigned int weight); |
ca32aefc | 225 | typedef void (blkio_update_group_read_bps_fn)(struct request_queue *q, |
fe071437 | 226 | struct blkio_group *blkg, u64 read_bps); |
ca32aefc | 227 | typedef void (blkio_update_group_write_bps_fn)(struct request_queue *q, |
fe071437 | 228 | struct blkio_group *blkg, u64 write_bps); |
ca32aefc | 229 | typedef void (blkio_update_group_read_iops_fn)(struct request_queue *q, |
fe071437 | 230 | struct blkio_group *blkg, unsigned int read_iops); |
ca32aefc | 231 | typedef void (blkio_update_group_write_iops_fn)(struct request_queue *q, |
fe071437 | 232 | struct blkio_group *blkg, unsigned int write_iops); |
3e252066 VG |
233 | |
234 | struct blkio_policy_ops { | |
cd1604fa TH |
235 | blkio_alloc_group_fn *blkio_alloc_group_fn; |
236 | blkio_link_group_fn *blkio_link_group_fn; | |
3e252066 | 237 | blkio_unlink_group_fn *blkio_unlink_group_fn; |
72e06c25 | 238 | blkio_clear_queue_fn *blkio_clear_queue_fn; |
3e252066 | 239 | blkio_update_group_weight_fn *blkio_update_group_weight_fn; |
4c9eefa1 VG |
240 | blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn; |
241 | blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn; | |
7702e8f4 VG |
242 | blkio_update_group_read_iops_fn *blkio_update_group_read_iops_fn; |
243 | blkio_update_group_write_iops_fn *blkio_update_group_write_iops_fn; | |
3e252066 VG |
244 | }; |
245 | ||
246 | struct blkio_policy_type { | |
247 | struct list_head list; | |
248 | struct blkio_policy_ops ops; | |
062a644d | 249 | enum blkio_policy_id plid; |
3e252066 VG |
250 | }; |
251 | ||
252 | /* Blkio controller policy registration */ | |
253 | extern void blkio_policy_register(struct blkio_policy_type *); | |
254 | extern void blkio_policy_unregister(struct blkio_policy_type *); | |
72e06c25 | 255 | extern void blkg_destroy_all(struct request_queue *q); |
3e252066 | 256 | |
afc24d49 VG |
257 | static inline char *blkg_path(struct blkio_group *blkg) |
258 | { | |
259 | return blkg->path; | |
260 | } | |
261 | ||
2f5ea477 JA |
262 | #else |
263 | ||
264 | struct blkio_group { | |
265 | }; | |
266 | ||
3e252066 VG |
267 | struct blkio_policy_type { |
268 | }; | |
269 | ||
270 | static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { } | |
271 | static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { } | |
72e06c25 | 272 | static inline void blkg_destroy_all(struct request_queue *q) { } |
3e252066 | 273 | |
afc24d49 VG |
274 | static inline char *blkg_path(struct blkio_group *blkg) { return NULL; } |
275 | ||
2f5ea477 JA |
276 | #endif |
277 | ||
df457f84 | 278 | #define BLKIO_WEIGHT_MIN 10 |
31e4c28d VG |
279 | #define BLKIO_WEIGHT_MAX 1000 |
280 | #define BLKIO_WEIGHT_DEFAULT 500 | |
281 | ||
2868ef7b | 282 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
a11cdaa7 | 283 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg); |
9195291e | 284 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
22084190 | 285 | unsigned long dequeue); |
812df48d DS |
286 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg); |
287 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg); | |
e5ff082e | 288 | void blkiocg_set_start_empty_time(struct blkio_group *blkg); |
812df48d DS |
289 | |
290 | #define BLKG_FLAG_FNS(name) \ | |
291 | static inline void blkio_mark_blkg_##name( \ | |
292 | struct blkio_group_stats *stats) \ | |
293 | { \ | |
294 | stats->flags |= (1 << BLKG_##name); \ | |
295 | } \ | |
296 | static inline void blkio_clear_blkg_##name( \ | |
297 | struct blkio_group_stats *stats) \ | |
298 | { \ | |
299 | stats->flags &= ~(1 << BLKG_##name); \ | |
300 | } \ | |
301 | static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \ | |
302 | { \ | |
303 | return (stats->flags & (1 << BLKG_##name)) != 0; \ | |
304 | } \ | |
305 | ||
306 | BLKG_FLAG_FNS(waiting) | |
307 | BLKG_FLAG_FNS(idling) | |
308 | BLKG_FLAG_FNS(empty) | |
309 | #undef BLKG_FLAG_FNS | |
2868ef7b | 310 | #else |
a11cdaa7 | 311 | static inline void blkiocg_update_avg_queue_size_stats( |
cdc1184c | 312 | struct blkio_group *blkg) {} |
9195291e DS |
313 | static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
314 | unsigned long dequeue) {} | |
812df48d DS |
315 | static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg) |
316 | {} | |
317 | static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg) {} | |
e5ff082e | 318 | static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg) {} |
2868ef7b VG |
319 | #endif |
320 | ||
32e380ae | 321 | #ifdef CONFIG_BLK_CGROUP |
31e4c28d VG |
322 | extern struct blkio_cgroup blkio_root_cgroup; |
323 | extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup); | |
70087dc3 | 324 | extern struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk); |
31e4c28d | 325 | extern int blkiocg_del_blkio_group(struct blkio_group *blkg); |
cd1604fa TH |
326 | extern struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg, |
327 | struct request_queue *q, | |
328 | enum blkio_policy_id plid); | |
329 | struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg, | |
330 | struct request_queue *q, | |
331 | enum blkio_policy_id plid, | |
332 | bool for_root); | |
303a3acb | 333 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, |
167400d3 JT |
334 | unsigned long time, |
335 | unsigned long unaccounted_time); | |
84c124da DS |
336 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, uint64_t bytes, |
337 | bool direction, bool sync); | |
338 | void blkiocg_update_completion_stats(struct blkio_group *blkg, | |
339 | uint64_t start_time, uint64_t io_start_time, bool direction, bool sync); | |
812d4026 DS |
340 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction, |
341 | bool sync); | |
a11cdaa7 | 342 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c | 343 | struct blkio_group *curr_blkg, bool direction, bool sync); |
a11cdaa7 | 344 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c | 345 | bool direction, bool sync); |
31e4c28d | 346 | #else |
2f5ea477 | 347 | struct cgroup; |
31e4c28d VG |
348 | static inline struct blkio_cgroup * |
349 | cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; } | |
70087dc3 VG |
350 | static inline struct blkio_cgroup * |
351 | task_blkio_cgroup(struct task_struct *tsk) { return NULL; } | |
31e4c28d | 352 | |
31e4c28d VG |
353 | static inline int |
354 | blkiocg_del_blkio_group(struct blkio_group *blkg) { return 0; } | |
355 | ||
cd1604fa TH |
356 | static inline struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg, |
357 | void *key) { return NULL; } | |
303a3acb | 358 | static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg, |
167400d3 JT |
359 | unsigned long time, |
360 | unsigned long unaccounted_time) | |
361 | {} | |
84c124da DS |
362 | static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
363 | uint64_t bytes, bool direction, bool sync) {} | |
364 | static inline void blkiocg_update_completion_stats(struct blkio_group *blkg, | |
365 | uint64_t start_time, uint64_t io_start_time, bool direction, | |
366 | bool sync) {} | |
812d4026 DS |
367 | static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg, |
368 | bool direction, bool sync) {} | |
a11cdaa7 | 369 | static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c | 370 | struct blkio_group *curr_blkg, bool direction, bool sync) {} |
a11cdaa7 | 371 | static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c | 372 | bool direction, bool sync) {} |
31e4c28d VG |
373 | #endif |
374 | #endif /* _BLK_CGROUP_H */ |