]>
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; | |
115 | }; | |
116 | ||
303a3acb DS |
117 | struct blkio_group_stats { |
118 | /* total disk time and nr sectors dispatched by this group */ | |
119 | uint64_t time; | |
cdc1184c | 120 | uint64_t stat_arr[BLKIO_STAT_QUEUED + 1][BLKIO_STAT_TOTAL]; |
303a3acb | 121 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
a23e6869 VG |
122 | /* Time not charged to this cgroup */ |
123 | uint64_t unaccounted_time; | |
124 | ||
cdc1184c DS |
125 | /* Sum of number of IOs queued across all samples */ |
126 | uint64_t avg_queue_size_sum; | |
127 | /* Count of samples taken for average */ | |
128 | uint64_t avg_queue_size_samples; | |
303a3acb DS |
129 | /* How many times this group has been removed from service tree */ |
130 | unsigned long dequeue; | |
812df48d DS |
131 | |
132 | /* Total time spent waiting for it to be assigned a timeslice. */ | |
133 | uint64_t group_wait_time; | |
134 | uint64_t start_group_wait_time; | |
135 | ||
136 | /* Time spent idling for this blkio_group */ | |
137 | uint64_t idle_time; | |
138 | uint64_t start_idle_time; | |
139 | /* | |
140 | * Total time when we have requests queued and do not contain the | |
141 | * current active queue. | |
142 | */ | |
143 | uint64_t empty_time; | |
144 | uint64_t start_empty_time; | |
145 | uint16_t flags; | |
303a3acb DS |
146 | #endif |
147 | }; | |
148 | ||
5624a4e4 VG |
149 | /* Per cpu blkio group stats */ |
150 | struct blkio_group_stats_cpu { | |
151 | uint64_t sectors; | |
152 | uint64_t stat_arr_cpu[BLKIO_STAT_CPU_NR][BLKIO_STAT_TOTAL]; | |
575969a0 | 153 | struct u64_stats_sync syncp; |
5624a4e4 VG |
154 | }; |
155 | ||
e56da7e2 TH |
156 | struct blkio_group_conf { |
157 | unsigned int weight; | |
158 | unsigned int iops[2]; | |
159 | u64 bps[2]; | |
160 | }; | |
161 | ||
0381411e TH |
162 | /* per-blkg per-policy data */ |
163 | struct blkg_policy_data { | |
164 | /* the blkg this per-policy data belongs to */ | |
165 | struct blkio_group *blkg; | |
166 | ||
549d3aa8 TH |
167 | /* Configuration */ |
168 | struct blkio_group_conf conf; | |
169 | ||
170 | struct blkio_group_stats stats; | |
171 | /* Per cpu stats pointer */ | |
172 | struct blkio_group_stats_cpu __percpu *stats_cpu; | |
173 | ||
0381411e TH |
174 | /* pol->pdata_size bytes of private data used by policy impl */ |
175 | char pdata[] __aligned(__alignof__(unsigned long long)); | |
176 | }; | |
177 | ||
31e4c28d | 178 | struct blkio_group { |
ca32aefc TH |
179 | /* Pointer to the associated request_queue, RCU protected */ |
180 | struct request_queue __rcu *q; | |
31e4c28d | 181 | struct hlist_node blkcg_node; |
7ee9c562 | 182 | struct blkio_cgroup *blkcg; |
2868ef7b VG |
183 | /* Store cgroup path */ |
184 | char path[128]; | |
062a644d VG |
185 | /* policy which owns this blk group */ |
186 | enum blkio_policy_id plid; | |
1adaf3dd TH |
187 | /* reference count */ |
188 | int refcnt; | |
22084190 | 189 | |
303a3acb DS |
190 | /* Need to serialize the stats in the case of reset/update */ |
191 | spinlock_t stats_lock; | |
549d3aa8 | 192 | struct blkg_policy_data *pd[BLKIO_NR_POLICIES]; |
1adaf3dd TH |
193 | |
194 | struct rcu_head rcu_head; | |
31e4c28d VG |
195 | }; |
196 | ||
0381411e | 197 | typedef void (blkio_init_group_fn)(struct blkio_group *blkg); |
cd1604fa TH |
198 | typedef void (blkio_link_group_fn)(struct request_queue *q, |
199 | struct blkio_group *blkg); | |
ca32aefc TH |
200 | typedef void (blkio_unlink_group_fn)(struct request_queue *q, |
201 | struct blkio_group *blkg); | |
72e06c25 | 202 | typedef bool (blkio_clear_queue_fn)(struct request_queue *q); |
ca32aefc | 203 | typedef void (blkio_update_group_weight_fn)(struct request_queue *q, |
fe071437 | 204 | struct blkio_group *blkg, unsigned int weight); |
ca32aefc | 205 | typedef void (blkio_update_group_read_bps_fn)(struct request_queue *q, |
fe071437 | 206 | struct blkio_group *blkg, u64 read_bps); |
ca32aefc | 207 | typedef void (blkio_update_group_write_bps_fn)(struct request_queue *q, |
fe071437 | 208 | struct blkio_group *blkg, u64 write_bps); |
ca32aefc | 209 | typedef void (blkio_update_group_read_iops_fn)(struct request_queue *q, |
fe071437 | 210 | struct blkio_group *blkg, unsigned int read_iops); |
ca32aefc | 211 | typedef void (blkio_update_group_write_iops_fn)(struct request_queue *q, |
fe071437 | 212 | struct blkio_group *blkg, unsigned int write_iops); |
3e252066 VG |
213 | |
214 | struct blkio_policy_ops { | |
0381411e | 215 | blkio_init_group_fn *blkio_init_group_fn; |
cd1604fa | 216 | blkio_link_group_fn *blkio_link_group_fn; |
3e252066 | 217 | blkio_unlink_group_fn *blkio_unlink_group_fn; |
72e06c25 | 218 | blkio_clear_queue_fn *blkio_clear_queue_fn; |
3e252066 | 219 | blkio_update_group_weight_fn *blkio_update_group_weight_fn; |
4c9eefa1 VG |
220 | blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn; |
221 | blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn; | |
7702e8f4 VG |
222 | blkio_update_group_read_iops_fn *blkio_update_group_read_iops_fn; |
223 | blkio_update_group_write_iops_fn *blkio_update_group_write_iops_fn; | |
3e252066 VG |
224 | }; |
225 | ||
226 | struct blkio_policy_type { | |
227 | struct list_head list; | |
228 | struct blkio_policy_ops ops; | |
062a644d | 229 | enum blkio_policy_id plid; |
0381411e | 230 | size_t pdata_size; /* policy specific private data size */ |
3e252066 VG |
231 | }; |
232 | ||
5efd6113 TH |
233 | extern int blkcg_init_queue(struct request_queue *q); |
234 | extern void blkcg_drain_queue(struct request_queue *q); | |
235 | extern void blkcg_exit_queue(struct request_queue *q); | |
236 | ||
3e252066 VG |
237 | /* Blkio controller policy registration */ |
238 | extern void blkio_policy_register(struct blkio_policy_type *); | |
239 | extern void blkio_policy_unregister(struct blkio_policy_type *); | |
72e06c25 | 240 | extern void blkg_destroy_all(struct request_queue *q); |
3e252066 | 241 | |
0381411e TH |
242 | /** |
243 | * blkg_to_pdata - get policy private data | |
244 | * @blkg: blkg of interest | |
245 | * @pol: policy of interest | |
246 | * | |
247 | * Return pointer to private data associated with the @blkg-@pol pair. | |
248 | */ | |
249 | static inline void *blkg_to_pdata(struct blkio_group *blkg, | |
250 | struct blkio_policy_type *pol) | |
251 | { | |
549d3aa8 | 252 | return blkg ? blkg->pd[pol->plid]->pdata : NULL; |
0381411e TH |
253 | } |
254 | ||
255 | /** | |
256 | * pdata_to_blkg - get blkg associated with policy private data | |
257 | * @pdata: policy private data of interest | |
258 | * @pol: policy @pdata is for | |
259 | * | |
260 | * @pdata is policy private data for @pol. Determine the blkg it's | |
261 | * associated with. | |
262 | */ | |
263 | static inline struct blkio_group *pdata_to_blkg(void *pdata, | |
264 | struct blkio_policy_type *pol) | |
265 | { | |
266 | if (pdata) { | |
267 | struct blkg_policy_data *pd = | |
268 | container_of(pdata, struct blkg_policy_data, pdata); | |
269 | return pd->blkg; | |
270 | } | |
271 | return NULL; | |
272 | } | |
273 | ||
afc24d49 VG |
274 | static inline char *blkg_path(struct blkio_group *blkg) |
275 | { | |
276 | return blkg->path; | |
277 | } | |
278 | ||
1adaf3dd TH |
279 | /** |
280 | * blkg_get - get a blkg reference | |
281 | * @blkg: blkg to get | |
282 | * | |
283 | * The caller should be holding queue_lock and an existing reference. | |
284 | */ | |
285 | static inline void blkg_get(struct blkio_group *blkg) | |
286 | { | |
287 | lockdep_assert_held(blkg->q->queue_lock); | |
288 | WARN_ON_ONCE(!blkg->refcnt); | |
289 | blkg->refcnt++; | |
290 | } | |
291 | ||
292 | void __blkg_release(struct blkio_group *blkg); | |
293 | ||
294 | /** | |
295 | * blkg_put - put a blkg reference | |
296 | * @blkg: blkg to put | |
297 | * | |
298 | * The caller should be holding queue_lock. | |
299 | */ | |
300 | static inline void blkg_put(struct blkio_group *blkg) | |
301 | { | |
302 | lockdep_assert_held(blkg->q->queue_lock); | |
303 | WARN_ON_ONCE(blkg->refcnt <= 0); | |
304 | if (!--blkg->refcnt) | |
305 | __blkg_release(blkg); | |
306 | } | |
307 | ||
2f5ea477 JA |
308 | #else |
309 | ||
310 | struct blkio_group { | |
311 | }; | |
312 | ||
3e252066 VG |
313 | struct blkio_policy_type { |
314 | }; | |
315 | ||
5efd6113 TH |
316 | static inline int blkcg_init_queue(struct request_queue *q) { return 0; } |
317 | static inline void blkcg_drain_queue(struct request_queue *q) { } | |
318 | static inline void blkcg_exit_queue(struct request_queue *q) { } | |
3e252066 VG |
319 | static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { } |
320 | static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { } | |
72e06c25 | 321 | static inline void blkg_destroy_all(struct request_queue *q) { } |
3e252066 | 322 | |
0381411e TH |
323 | static inline void *blkg_to_pdata(struct blkio_group *blkg, |
324 | struct blkio_policy_type *pol) { return NULL; } | |
325 | static inline struct blkio_group *pdata_to_blkg(void *pdata, | |
326 | struct blkio_policy_type *pol) { return NULL; } | |
afc24d49 | 327 | static inline char *blkg_path(struct blkio_group *blkg) { return NULL; } |
1adaf3dd TH |
328 | static inline void blkg_get(struct blkio_group *blkg) { } |
329 | static inline void blkg_put(struct blkio_group *blkg) { } | |
afc24d49 | 330 | |
2f5ea477 JA |
331 | #endif |
332 | ||
df457f84 | 333 | #define BLKIO_WEIGHT_MIN 10 |
31e4c28d VG |
334 | #define BLKIO_WEIGHT_MAX 1000 |
335 | #define BLKIO_WEIGHT_DEFAULT 500 | |
336 | ||
2868ef7b | 337 | #ifdef CONFIG_DEBUG_BLK_CGROUP |
a11cdaa7 | 338 | void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg); |
9195291e | 339 | void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
22084190 | 340 | unsigned long dequeue); |
812df48d DS |
341 | void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg); |
342 | void blkiocg_update_idle_time_stats(struct blkio_group *blkg); | |
e5ff082e | 343 | void blkiocg_set_start_empty_time(struct blkio_group *blkg); |
812df48d DS |
344 | |
345 | #define BLKG_FLAG_FNS(name) \ | |
346 | static inline void blkio_mark_blkg_##name( \ | |
347 | struct blkio_group_stats *stats) \ | |
348 | { \ | |
349 | stats->flags |= (1 << BLKG_##name); \ | |
350 | } \ | |
351 | static inline void blkio_clear_blkg_##name( \ | |
352 | struct blkio_group_stats *stats) \ | |
353 | { \ | |
354 | stats->flags &= ~(1 << BLKG_##name); \ | |
355 | } \ | |
356 | static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \ | |
357 | { \ | |
358 | return (stats->flags & (1 << BLKG_##name)) != 0; \ | |
359 | } \ | |
360 | ||
361 | BLKG_FLAG_FNS(waiting) | |
362 | BLKG_FLAG_FNS(idling) | |
363 | BLKG_FLAG_FNS(empty) | |
364 | #undef BLKG_FLAG_FNS | |
2868ef7b | 365 | #else |
a11cdaa7 | 366 | static inline void blkiocg_update_avg_queue_size_stats( |
cdc1184c | 367 | struct blkio_group *blkg) {} |
9195291e DS |
368 | static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg, |
369 | unsigned long dequeue) {} | |
812df48d DS |
370 | static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg) |
371 | {} | |
372 | static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg) {} | |
e5ff082e | 373 | static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg) {} |
2868ef7b VG |
374 | #endif |
375 | ||
32e380ae | 376 | #ifdef CONFIG_BLK_CGROUP |
31e4c28d VG |
377 | extern struct blkio_cgroup blkio_root_cgroup; |
378 | extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup); | |
70087dc3 | 379 | extern struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk); |
31e4c28d | 380 | extern int blkiocg_del_blkio_group(struct blkio_group *blkg); |
cd1604fa TH |
381 | extern struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg, |
382 | struct request_queue *q, | |
383 | enum blkio_policy_id plid); | |
384 | struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg, | |
385 | struct request_queue *q, | |
386 | enum blkio_policy_id plid, | |
387 | bool for_root); | |
303a3acb | 388 | void blkiocg_update_timeslice_used(struct blkio_group *blkg, |
167400d3 JT |
389 | unsigned long time, |
390 | unsigned long unaccounted_time); | |
84c124da DS |
391 | void blkiocg_update_dispatch_stats(struct blkio_group *blkg, uint64_t bytes, |
392 | bool direction, bool sync); | |
393 | void blkiocg_update_completion_stats(struct blkio_group *blkg, | |
394 | uint64_t start_time, uint64_t io_start_time, bool direction, bool sync); | |
812d4026 DS |
395 | void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction, |
396 | bool sync); | |
a11cdaa7 | 397 | void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c | 398 | struct blkio_group *curr_blkg, bool direction, bool sync); |
a11cdaa7 | 399 | void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c | 400 | bool direction, bool sync); |
31e4c28d | 401 | #else |
2f5ea477 | 402 | struct cgroup; |
31e4c28d VG |
403 | static inline struct blkio_cgroup * |
404 | cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; } | |
70087dc3 VG |
405 | static inline struct blkio_cgroup * |
406 | task_blkio_cgroup(struct task_struct *tsk) { return NULL; } | |
31e4c28d | 407 | |
31e4c28d VG |
408 | static inline int |
409 | blkiocg_del_blkio_group(struct blkio_group *blkg) { return 0; } | |
410 | ||
cd1604fa TH |
411 | static inline struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg, |
412 | void *key) { return NULL; } | |
303a3acb | 413 | static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg, |
167400d3 JT |
414 | unsigned long time, |
415 | unsigned long unaccounted_time) | |
416 | {} | |
84c124da DS |
417 | static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg, |
418 | uint64_t bytes, bool direction, bool sync) {} | |
419 | static inline void blkiocg_update_completion_stats(struct blkio_group *blkg, | |
420 | uint64_t start_time, uint64_t io_start_time, bool direction, | |
421 | bool sync) {} | |
812d4026 DS |
422 | static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg, |
423 | bool direction, bool sync) {} | |
a11cdaa7 | 424 | static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg, |
cdc1184c | 425 | struct blkio_group *curr_blkg, bool direction, bool sync) {} |
a11cdaa7 | 426 | static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg, |
cdc1184c | 427 | bool direction, bool sync) {} |
31e4c28d VG |
428 | #endif |
429 | #endif /* _BLK_CGROUP_H */ |