]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - block/blk-cgroup.h
blkcg: use double locking instead of RCU for blkg synchronization
[mirror_ubuntu-artful-kernel.git] / block / blk-cgroup.h
CommitLineData
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
19enum 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
31enum 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 */
54enum 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
65enum 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 */
74enum 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 */
81enum 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 */
101enum 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
110struct 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
117struct 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 */
150struct 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
156struct 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 */
163struct 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 178struct blkio_group {
ca32aefc
TH
179 /* Pointer to the associated request_queue, RCU protected */
180 struct request_queue __rcu *q;
e8989fae 181 struct list_head q_node;
31e4c28d 182 struct hlist_node blkcg_node;
7ee9c562 183 struct blkio_cgroup *blkcg;
2868ef7b
VG
184 /* Store cgroup path */
185 char path[128];
1adaf3dd
TH
186 /* reference count */
187 int refcnt;
22084190 188
303a3acb
DS
189 /* Need to serialize the stats in the case of reset/update */
190 spinlock_t stats_lock;
549d3aa8 191 struct blkg_policy_data *pd[BLKIO_NR_POLICIES];
1adaf3dd
TH
192
193 struct rcu_head rcu_head;
31e4c28d
VG
194};
195
0381411e 196typedef void (blkio_init_group_fn)(struct blkio_group *blkg);
ca32aefc 197typedef void (blkio_update_group_weight_fn)(struct request_queue *q,
fe071437 198 struct blkio_group *blkg, unsigned int weight);
ca32aefc 199typedef void (blkio_update_group_read_bps_fn)(struct request_queue *q,
fe071437 200 struct blkio_group *blkg, u64 read_bps);
ca32aefc 201typedef void (blkio_update_group_write_bps_fn)(struct request_queue *q,
fe071437 202 struct blkio_group *blkg, u64 write_bps);
ca32aefc 203typedef void (blkio_update_group_read_iops_fn)(struct request_queue *q,
fe071437 204 struct blkio_group *blkg, unsigned int read_iops);
ca32aefc 205typedef void (blkio_update_group_write_iops_fn)(struct request_queue *q,
fe071437 206 struct blkio_group *blkg, unsigned int write_iops);
3e252066
VG
207
208struct blkio_policy_ops {
0381411e 209 blkio_init_group_fn *blkio_init_group_fn;
3e252066 210 blkio_update_group_weight_fn *blkio_update_group_weight_fn;
4c9eefa1
VG
211 blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn;
212 blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn;
7702e8f4
VG
213 blkio_update_group_read_iops_fn *blkio_update_group_read_iops_fn;
214 blkio_update_group_write_iops_fn *blkio_update_group_write_iops_fn;
3e252066
VG
215};
216
217struct blkio_policy_type {
218 struct list_head list;
219 struct blkio_policy_ops ops;
062a644d 220 enum blkio_policy_id plid;
0381411e 221 size_t pdata_size; /* policy specific private data size */
3e252066
VG
222};
223
5efd6113
TH
224extern int blkcg_init_queue(struct request_queue *q);
225extern void blkcg_drain_queue(struct request_queue *q);
226extern void blkcg_exit_queue(struct request_queue *q);
227
3e252066
VG
228/* Blkio controller policy registration */
229extern void blkio_policy_register(struct blkio_policy_type *);
230extern void blkio_policy_unregister(struct blkio_policy_type *);
e8989fae
TH
231extern void blkg_destroy_all(struct request_queue *q, bool destroy_root);
232extern void update_root_blkg_pd(struct request_queue *q,
233 enum blkio_policy_id plid);
3e252066 234
0381411e
TH
235/**
236 * blkg_to_pdata - get policy private data
237 * @blkg: blkg of interest
238 * @pol: policy of interest
239 *
240 * Return pointer to private data associated with the @blkg-@pol pair.
241 */
242static inline void *blkg_to_pdata(struct blkio_group *blkg,
243 struct blkio_policy_type *pol)
244{
549d3aa8 245 return blkg ? blkg->pd[pol->plid]->pdata : NULL;
0381411e
TH
246}
247
248/**
249 * pdata_to_blkg - get blkg associated with policy private data
250 * @pdata: policy private data of interest
251 * @pol: policy @pdata is for
252 *
253 * @pdata is policy private data for @pol. Determine the blkg it's
254 * associated with.
255 */
256static inline struct blkio_group *pdata_to_blkg(void *pdata,
257 struct blkio_policy_type *pol)
258{
259 if (pdata) {
260 struct blkg_policy_data *pd =
261 container_of(pdata, struct blkg_policy_data, pdata);
262 return pd->blkg;
263 }
264 return NULL;
265}
266
afc24d49
VG
267static inline char *blkg_path(struct blkio_group *blkg)
268{
269 return blkg->path;
270}
271
1adaf3dd
TH
272/**
273 * blkg_get - get a blkg reference
274 * @blkg: blkg to get
275 *
276 * The caller should be holding queue_lock and an existing reference.
277 */
278static inline void blkg_get(struct blkio_group *blkg)
279{
280 lockdep_assert_held(blkg->q->queue_lock);
281 WARN_ON_ONCE(!blkg->refcnt);
282 blkg->refcnt++;
283}
284
285void __blkg_release(struct blkio_group *blkg);
286
287/**
288 * blkg_put - put a blkg reference
289 * @blkg: blkg to put
290 *
291 * The caller should be holding queue_lock.
292 */
293static inline void blkg_put(struct blkio_group *blkg)
294{
295 lockdep_assert_held(blkg->q->queue_lock);
296 WARN_ON_ONCE(blkg->refcnt <= 0);
297 if (!--blkg->refcnt)
298 __blkg_release(blkg);
299}
300
2f5ea477
JA
301#else
302
303struct blkio_group {
304};
305
3e252066
VG
306struct blkio_policy_type {
307};
308
5efd6113
TH
309static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
310static inline void blkcg_drain_queue(struct request_queue *q) { }
311static inline void blkcg_exit_queue(struct request_queue *q) { }
3e252066
VG
312static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { }
313static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
03aa264a 314static inline void blkg_destroy_all(struct request_queue *q,
03aa264a 315 bool destory_root) { }
e8989fae
TH
316static inline void update_root_blkg_pd(struct request_queue *q,
317 enum blkio_policy_id plid) { }
3e252066 318
0381411e
TH
319static inline void *blkg_to_pdata(struct blkio_group *blkg,
320 struct blkio_policy_type *pol) { return NULL; }
321static inline struct blkio_group *pdata_to_blkg(void *pdata,
322 struct blkio_policy_type *pol) { return NULL; }
afc24d49 323static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
1adaf3dd
TH
324static inline void blkg_get(struct blkio_group *blkg) { }
325static inline void blkg_put(struct blkio_group *blkg) { }
afc24d49 326
2f5ea477
JA
327#endif
328
df457f84 329#define BLKIO_WEIGHT_MIN 10
31e4c28d
VG
330#define BLKIO_WEIGHT_MAX 1000
331#define BLKIO_WEIGHT_DEFAULT 500
332
2868ef7b 333#ifdef CONFIG_DEBUG_BLK_CGROUP
c1768268
TH
334void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg,
335 struct blkio_policy_type *pol);
9195291e 336void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
c1768268
TH
337 struct blkio_policy_type *pol,
338 unsigned long dequeue);
339void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg,
340 struct blkio_policy_type *pol);
341void blkiocg_update_idle_time_stats(struct blkio_group *blkg,
342 struct blkio_policy_type *pol);
343void blkiocg_set_start_empty_time(struct blkio_group *blkg,
344 struct blkio_policy_type *pol);
812df48d
DS
345
346#define BLKG_FLAG_FNS(name) \
347static inline void blkio_mark_blkg_##name( \
348 struct blkio_group_stats *stats) \
349{ \
350 stats->flags |= (1 << BLKG_##name); \
351} \
352static inline void blkio_clear_blkg_##name( \
353 struct blkio_group_stats *stats) \
354{ \
355 stats->flags &= ~(1 << BLKG_##name); \
356} \
357static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \
358{ \
359 return (stats->flags & (1 << BLKG_##name)) != 0; \
360} \
361
362BLKG_FLAG_FNS(waiting)
363BLKG_FLAG_FNS(idling)
364BLKG_FLAG_FNS(empty)
365#undef BLKG_FLAG_FNS
2868ef7b 366#else
c1768268
TH
367static inline void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg,
368 struct blkio_policy_type *pol) { }
9195291e 369static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
c1768268
TH
370 struct blkio_policy_type *pol, unsigned long dequeue) { }
371static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg,
372 struct blkio_policy_type *pol) { }
373static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg,
374 struct blkio_policy_type *pol) { }
375static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg,
376 struct blkio_policy_type *pol) { }
2868ef7b
VG
377#endif
378
32e380ae 379#ifdef CONFIG_BLK_CGROUP
31e4c28d
VG
380extern struct blkio_cgroup blkio_root_cgroup;
381extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
70087dc3 382extern struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk);
cd1604fa 383extern struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
e8989fae 384 struct request_queue *q);
cd1604fa
TH
385struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg,
386 struct request_queue *q,
387 enum blkio_policy_id plid,
388 bool for_root);
303a3acb 389void blkiocg_update_timeslice_used(struct blkio_group *blkg,
c1768268
TH
390 struct blkio_policy_type *pol,
391 unsigned long time,
392 unsigned long unaccounted_time);
393void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
394 struct blkio_policy_type *pol,
395 uint64_t bytes, bool direction, bool sync);
84c124da 396void blkiocg_update_completion_stats(struct blkio_group *blkg,
c1768268
TH
397 struct blkio_policy_type *pol,
398 uint64_t start_time,
399 uint64_t io_start_time, bool direction,
400 bool sync);
401void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
402 struct blkio_policy_type *pol,
403 bool direction, bool sync);
a11cdaa7 404void blkiocg_update_io_add_stats(struct blkio_group *blkg,
c1768268
TH
405 struct blkio_policy_type *pol,
406 struct blkio_group *curr_blkg, bool direction,
407 bool sync);
a11cdaa7 408void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
c1768268
TH
409 struct blkio_policy_type *pol,
410 bool direction, bool sync);
31e4c28d 411#else
2f5ea477 412struct cgroup;
31e4c28d
VG
413static inline struct blkio_cgroup *
414cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
70087dc3
VG
415static inline struct blkio_cgroup *
416task_blkio_cgroup(struct task_struct *tsk) { return NULL; }
31e4c28d 417
cd1604fa
TH
418static inline struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
419 void *key) { return NULL; }
303a3acb 420static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg,
c1768268
TH
421 struct blkio_policy_type *pol, unsigned long time,
422 unsigned long unaccounted_time) { }
84c124da 423static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
c1768268
TH
424 struct blkio_policy_type *pol, uint64_t bytes,
425 bool direction, bool sync) { }
84c124da 426static inline void blkiocg_update_completion_stats(struct blkio_group *blkg,
c1768268
TH
427 struct blkio_policy_type *pol, uint64_t start_time,
428 uint64_t io_start_time, bool direction, bool sync) { }
812d4026 429static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
c1768268
TH
430 struct blkio_policy_type *pol, bool direction,
431 bool sync) { }
a11cdaa7 432static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg,
c1768268
TH
433 struct blkio_policy_type *pol,
434 struct blkio_group *curr_blkg, bool direction,
435 bool sync) { }
a11cdaa7 436static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
c1768268
TH
437 struct blkio_policy_type *pol, bool direction,
438 bool sync) { }
31e4c28d
VG
439#endif
440#endif /* _BLK_CGROUP_H */