]> git.ipfire.org Git - thirdparty/linux.git/blame - block/blk-cgroup.h
blkcg: implement per-queue policy activation
[thirdparty/linux.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>
829fdb50 18#include <linux/seq_file.h>
31e4c28d 19
9355aede
VG
20/* Max limits for throttle policy */
21#define THROTL_IOPS_MAX UINT_MAX
22
3381cb8d
TH
23/* CFQ specific, out here for blkcg->cfq_weight */
24#define CFQ_WEIGHT_MIN 10
25#define CFQ_WEIGHT_MAX 1000
26#define CFQ_WEIGHT_DEFAULT 500
27
f48ec1d7
TH
28#ifdef CONFIG_BLK_CGROUP
29
edcb0722
TH
30enum blkg_rwstat_type {
31 BLKG_RWSTAT_READ,
32 BLKG_RWSTAT_WRITE,
33 BLKG_RWSTAT_SYNC,
34 BLKG_RWSTAT_ASYNC,
35
36 BLKG_RWSTAT_NR,
37 BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
303a3acb
DS
38};
39
31e4c28d
VG
40struct blkio_cgroup {
41 struct cgroup_subsys_state css;
31e4c28d
VG
42 spinlock_t lock;
43 struct hlist_head blkg_list;
9a9e8a26
TH
44
45 /* for policies to test whether associated blkcg has changed */
46 uint64_t id;
3381cb8d
TH
47
48 /* TODO: per-policy storage in blkio_cgroup */
49 unsigned int cfq_weight; /* belongs to cfq */
31e4c28d
VG
50};
51
edcb0722
TH
52struct blkg_stat {
53 struct u64_stats_sync syncp;
54 uint64_t cnt;
55};
56
57struct blkg_rwstat {
58 struct u64_stats_sync syncp;
59 uint64_t cnt[BLKG_RWSTAT_NR];
60};
61
0381411e
TH
62/* per-blkg per-policy data */
63struct blkg_policy_data {
64 /* the blkg this per-policy data belongs to */
65 struct blkio_group *blkg;
66
a2b1693b
TH
67 /* used during policy activation */
68 struct list_head alloc_node;
69
0381411e
TH
70 /* pol->pdata_size bytes of private data used by policy impl */
71 char pdata[] __aligned(__alignof__(unsigned long long));
72};
73
31e4c28d 74struct blkio_group {
c875f4d0
TH
75 /* Pointer to the associated request_queue */
76 struct request_queue *q;
e8989fae 77 struct list_head q_node;
31e4c28d 78 struct hlist_node blkcg_node;
7ee9c562 79 struct blkio_cgroup *blkcg;
2868ef7b
VG
80 /* Store cgroup path */
81 char path[128];
1adaf3dd
TH
82 /* reference count */
83 int refcnt;
22084190 84
8bd435b3 85 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
1adaf3dd
TH
86
87 struct rcu_head rcu_head;
31e4c28d
VG
88};
89
0381411e 90typedef void (blkio_init_group_fn)(struct blkio_group *blkg);
9ade5ea4
TH
91typedef void (blkio_exit_group_fn)(struct blkio_group *blkg);
92typedef void (blkio_reset_group_stats_fn)(struct blkio_group *blkg);
3e252066
VG
93
94struct blkio_policy_ops {
0381411e 95 blkio_init_group_fn *blkio_init_group_fn;
9ade5ea4
TH
96 blkio_exit_group_fn *blkio_exit_group_fn;
97 blkio_reset_group_stats_fn *blkio_reset_group_stats_fn;
3e252066
VG
98};
99
100struct blkio_policy_type {
3e252066 101 struct blkio_policy_ops ops;
8bd435b3 102 int plid;
0381411e 103 size_t pdata_size; /* policy specific private data size */
44ea53de 104 struct cftype *cftypes; /* cgroup files for the policy */
3e252066
VG
105};
106
5efd6113
TH
107extern int blkcg_init_queue(struct request_queue *q);
108extern void blkcg_drain_queue(struct request_queue *q);
109extern void blkcg_exit_queue(struct request_queue *q);
110
3e252066 111/* Blkio controller policy registration */
8bd435b3 112extern int blkio_policy_register(struct blkio_policy_type *);
3e252066 113extern void blkio_policy_unregister(struct blkio_policy_type *);
a2b1693b
TH
114extern int blkcg_activate_policy(struct request_queue *q,
115 const struct blkio_policy_type *pol);
116extern void blkcg_deactivate_policy(struct request_queue *q,
117 const struct blkio_policy_type *pol);
e8989fae 118extern void blkg_destroy_all(struct request_queue *q, bool destroy_root);
3e252066 119
829fdb50 120void blkcg_print_blkgs(struct seq_file *sf, struct blkio_cgroup *blkcg,
d366e7ec 121 u64 (*prfill)(struct seq_file *, void *, int),
ec399347
TH
122 const struct blkio_policy_type *pol, int data,
123 bool show_total);
d366e7ec
TH
124u64 __blkg_prfill_u64(struct seq_file *sf, void *pdata, u64 v);
125u64 __blkg_prfill_rwstat(struct seq_file *sf, void *pdata,
829fdb50 126 const struct blkg_rwstat *rwstat);
5bc4afb1
TH
127u64 blkg_prfill_stat(struct seq_file *sf, void *pdata, int off);
128u64 blkg_prfill_rwstat(struct seq_file *sf, void *pdata, int off);
829fdb50
TH
129
130struct blkg_conf_ctx {
131 struct gendisk *disk;
132 struct blkio_group *blkg;
133 u64 v;
134};
135
da8b0662
TH
136int blkg_conf_prep(struct blkio_cgroup *blkcg,
137 const struct blkio_policy_type *pol, const char *input,
829fdb50
TH
138 struct blkg_conf_ctx *ctx);
139void blkg_conf_finish(struct blkg_conf_ctx *ctx);
140
141
0381411e
TH
142/**
143 * blkg_to_pdata - get policy private data
144 * @blkg: blkg of interest
145 * @pol: policy of interest
146 *
147 * Return pointer to private data associated with the @blkg-@pol pair.
148 */
149static inline void *blkg_to_pdata(struct blkio_group *blkg,
150 struct blkio_policy_type *pol)
151{
549d3aa8 152 return blkg ? blkg->pd[pol->plid]->pdata : NULL;
0381411e
TH
153}
154
155/**
156 * pdata_to_blkg - get blkg associated with policy private data
157 * @pdata: policy private data of interest
0381411e 158 *
aaec55a0 159 * @pdata is policy private data. Determine the blkg it's associated with.
0381411e 160 */
aaec55a0 161static inline struct blkio_group *pdata_to_blkg(void *pdata)
0381411e
TH
162{
163 if (pdata) {
164 struct blkg_policy_data *pd =
165 container_of(pdata, struct blkg_policy_data, pdata);
166 return pd->blkg;
167 }
168 return NULL;
169}
170
afc24d49
VG
171static inline char *blkg_path(struct blkio_group *blkg)
172{
173 return blkg->path;
174}
175
1adaf3dd
TH
176/**
177 * blkg_get - get a blkg reference
178 * @blkg: blkg to get
179 *
180 * The caller should be holding queue_lock and an existing reference.
181 */
182static inline void blkg_get(struct blkio_group *blkg)
183{
184 lockdep_assert_held(blkg->q->queue_lock);
185 WARN_ON_ONCE(!blkg->refcnt);
186 blkg->refcnt++;
187}
188
189void __blkg_release(struct blkio_group *blkg);
190
191/**
192 * blkg_put - put a blkg reference
193 * @blkg: blkg to put
194 *
195 * The caller should be holding queue_lock.
196 */
197static inline void blkg_put(struct blkio_group *blkg)
198{
199 lockdep_assert_held(blkg->q->queue_lock);
200 WARN_ON_ONCE(blkg->refcnt <= 0);
201 if (!--blkg->refcnt)
202 __blkg_release(blkg);
203}
204
edcb0722
TH
205/**
206 * blkg_stat_add - add a value to a blkg_stat
207 * @stat: target blkg_stat
208 * @val: value to add
209 *
210 * Add @val to @stat. The caller is responsible for synchronizing calls to
211 * this function.
212 */
213static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
214{
215 u64_stats_update_begin(&stat->syncp);
216 stat->cnt += val;
217 u64_stats_update_end(&stat->syncp);
218}
219
220/**
221 * blkg_stat_read - read the current value of a blkg_stat
222 * @stat: blkg_stat to read
223 *
224 * Read the current value of @stat. This function can be called without
225 * synchroniztion and takes care of u64 atomicity.
226 */
227static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
228{
229 unsigned int start;
230 uint64_t v;
231
232 do {
233 start = u64_stats_fetch_begin(&stat->syncp);
234 v = stat->cnt;
235 } while (u64_stats_fetch_retry(&stat->syncp, start));
236
237 return v;
238}
239
240/**
241 * blkg_stat_reset - reset a blkg_stat
242 * @stat: blkg_stat to reset
243 */
244static inline void blkg_stat_reset(struct blkg_stat *stat)
245{
246 stat->cnt = 0;
247}
248
249/**
250 * blkg_rwstat_add - add a value to a blkg_rwstat
251 * @rwstat: target blkg_rwstat
252 * @rw: mask of REQ_{WRITE|SYNC}
253 * @val: value to add
254 *
255 * Add @val to @rwstat. The counters are chosen according to @rw. The
256 * caller is responsible for synchronizing calls to this function.
257 */
258static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
259 int rw, uint64_t val)
260{
261 u64_stats_update_begin(&rwstat->syncp);
262
263 if (rw & REQ_WRITE)
264 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
265 else
266 rwstat->cnt[BLKG_RWSTAT_READ] += val;
267 if (rw & REQ_SYNC)
268 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
269 else
270 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
271
272 u64_stats_update_end(&rwstat->syncp);
273}
274
275/**
276 * blkg_rwstat_read - read the current values of a blkg_rwstat
277 * @rwstat: blkg_rwstat to read
278 *
279 * Read the current snapshot of @rwstat and return it as the return value.
280 * This function can be called without synchronization and takes care of
281 * u64 atomicity.
282 */
283static struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
284{
285 unsigned int start;
286 struct blkg_rwstat tmp;
287
288 do {
289 start = u64_stats_fetch_begin(&rwstat->syncp);
290 tmp = *rwstat;
291 } while (u64_stats_fetch_retry(&rwstat->syncp, start));
292
293 return tmp;
294}
295
296/**
297 * blkg_rwstat_sum - read the total count of a blkg_rwstat
298 * @rwstat: blkg_rwstat to read
299 *
300 * Return the total count of @rwstat regardless of the IO direction. This
301 * function can be called without synchronization and takes care of u64
302 * atomicity.
303 */
304static inline uint64_t blkg_rwstat_sum(struct blkg_rwstat *rwstat)
305{
306 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
307
308 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
309}
310
311/**
312 * blkg_rwstat_reset - reset a blkg_rwstat
313 * @rwstat: blkg_rwstat to reset
314 */
315static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
316{
317 memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
318}
319
2f5ea477
JA
320#else
321
322struct blkio_group {
323};
324
3e252066
VG
325struct blkio_policy_type {
326};
327
5efd6113
TH
328static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
329static inline void blkcg_drain_queue(struct request_queue *q) { }
330static inline void blkcg_exit_queue(struct request_queue *q) { }
8bd435b3 331static inline int blkio_policy_register(struct blkio_policy_type *blkiop) { return 0; }
3e252066 332static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
a2b1693b
TH
333static inline int blkcg_activate_policy(struct request_queue *q,
334 const struct blkio_policy_type *pol) { return 0; }
335static inline void blkcg_deactivate_policy(struct request_queue *q,
336 const struct blkio_policy_type *pol) { }
03aa264a 337static inline void blkg_destroy_all(struct request_queue *q,
03aa264a 338 bool destory_root) { }
3e252066 339
0381411e
TH
340static inline void *blkg_to_pdata(struct blkio_group *blkg,
341 struct blkio_policy_type *pol) { return NULL; }
342static inline struct blkio_group *pdata_to_blkg(void *pdata,
343 struct blkio_policy_type *pol) { return NULL; }
afc24d49 344static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
1adaf3dd
TH
345static inline void blkg_get(struct blkio_group *blkg) { }
346static inline void blkg_put(struct blkio_group *blkg) { }
afc24d49 347
2f5ea477
JA
348#endif
349
32e380ae 350#ifdef CONFIG_BLK_CGROUP
31e4c28d
VG
351extern struct blkio_cgroup blkio_root_cgroup;
352extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
4f85cb96 353extern struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio);
cd1604fa 354extern struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
e8989fae 355 struct request_queue *q);
cd1604fa
TH
356struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg,
357 struct request_queue *q,
cd1604fa 358 bool for_root);
31e4c28d 359#else
2f5ea477 360struct cgroup;
31e4c28d
VG
361static inline struct blkio_cgroup *
362cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
70087dc3 363static inline struct blkio_cgroup *
4f85cb96 364bio_blkio_cgroup(struct bio *bio) { return NULL; }
31e4c28d 365
cd1604fa
TH
366static inline struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
367 void *key) { return NULL; }
31e4c28d
VG
368#endif
369#endif /* _BLK_CGROUP_H */