blob: 18b021e1c05f9eca406f1a34a81306daeb5239a1 [file] [log] [blame]
Vivek Goyal31e4c282009-12-03 12:59:42 -05001#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>
Vivek Goyal575969a2011-05-19 15:38:29 -040017#include <linux/u64_stats_sync.h>
Tejun Heo829fdb52012-04-01 14:38:43 -070018#include <linux/seq_file.h>
Vivek Goyal31e4c282009-12-03 12:59:42 -050019
Vivek Goyal9355aed2010-10-01 21:16:41 +020020/* Max limits for throttle policy */
21#define THROTL_IOPS_MAX UINT_MAX
22
Tejun Heo3381cb82012-04-01 14:38:44 -070023/* 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
Tejun Heof48ec1d2012-04-13 13:11:25 -070028#ifdef CONFIG_BLK_CGROUP
29
Tejun Heoedcb0722012-04-01 14:38:42 -070030enum 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,
Divyesh Shah303a3ac2010-04-01 15:01:24 -070038};
39
Tejun Heo3c798392012-04-16 13:57:25 -070040struct blkcg {
Tejun Heo36558c82012-04-16 13:57:24 -070041 struct cgroup_subsys_state css;
42 spinlock_t lock;
43 struct hlist_head blkg_list;
Tejun Heo9a9e8a22012-03-19 15:10:56 -070044
45 /* for policies to test whether associated blkcg has changed */
Tejun Heo36558c82012-04-16 13:57:24 -070046 uint64_t id;
Tejun Heo3381cb82012-04-01 14:38:44 -070047
Tejun Heo3c798392012-04-16 13:57:25 -070048 /* TODO: per-policy storage in blkcg */
Tejun Heo36558c82012-04-16 13:57:24 -070049 unsigned int cfq_weight; /* belongs to cfq */
Vivek Goyal31e4c282009-12-03 12:59:42 -050050};
51
Tejun Heoedcb0722012-04-01 14:38:42 -070052struct 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
Tejun Heof95a04a2012-04-16 13:57:26 -070062/*
63 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
64 * request_queue (q). This is used by blkcg policies which need to track
65 * information per blkcg - q pair.
66 *
67 * There can be multiple active blkcg policies and each has its private
68 * data on each blkg, the size of which is determined by
69 * blkcg_policy->pd_size. blkcg core allocates and frees such areas
70 * together with blkg and invokes pd_init/exit_fn() methods.
71 *
72 * Such private data must embed struct blkg_policy_data (pd) at the
73 * beginning and pd_size can't be smaller than pd.
74 */
Tejun Heo03814112012-03-05 13:15:14 -080075struct blkg_policy_data {
76 /* the blkg this per-policy data belongs to */
Tejun Heo3c798392012-04-16 13:57:25 -070077 struct blkcg_gq *blkg;
Tejun Heo03814112012-03-05 13:15:14 -080078
Tejun Heoa2b16932012-04-13 13:11:33 -070079 /* used during policy activation */
Tejun Heo36558c82012-04-16 13:57:24 -070080 struct list_head alloc_node;
Tejun Heo03814112012-03-05 13:15:14 -080081};
82
Tejun Heo3c798392012-04-16 13:57:25 -070083/* association between a blk cgroup and a request queue */
84struct blkcg_gq {
Tejun Heoc875f4d2012-03-05 13:15:22 -080085 /* Pointer to the associated request_queue */
Tejun Heo36558c82012-04-16 13:57:24 -070086 struct request_queue *q;
87 struct list_head q_node;
88 struct hlist_node blkcg_node;
Tejun Heo3c798392012-04-16 13:57:25 -070089 struct blkcg *blkcg;
Tejun Heo1adaf3d2012-03-05 13:15:15 -080090 /* reference count */
Tejun Heo36558c82012-04-16 13:57:24 -070091 int refcnt;
Vivek Goyal22084192009-12-03 12:59:49 -050092
Tejun Heo36558c82012-04-16 13:57:24 -070093 struct blkg_policy_data *pd[BLKCG_MAX_POLS];
Tejun Heo1adaf3d2012-03-05 13:15:15 -080094
Tejun Heo36558c82012-04-16 13:57:24 -070095 struct rcu_head rcu_head;
Vivek Goyal31e4c282009-12-03 12:59:42 -050096};
97
Tejun Heo3c798392012-04-16 13:57:25 -070098typedef void (blkcg_pol_init_pd_fn)(struct blkcg_gq *blkg);
99typedef void (blkcg_pol_exit_pd_fn)(struct blkcg_gq *blkg);
100typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkcg_gq *blkg);
Vivek Goyal3e252062009-12-04 10:36:42 -0500101
Tejun Heo3c798392012-04-16 13:57:25 -0700102struct blkcg_policy_ops {
103 blkcg_pol_init_pd_fn *pd_init_fn;
104 blkcg_pol_exit_pd_fn *pd_exit_fn;
105 blkcg_pol_reset_pd_stats_fn *pd_reset_stats_fn;
Vivek Goyal3e252062009-12-04 10:36:42 -0500106};
107
Tejun Heo3c798392012-04-16 13:57:25 -0700108struct blkcg_policy {
109 struct blkcg_policy_ops ops;
Tejun Heo36558c82012-04-16 13:57:24 -0700110 int plid;
111 /* policy specific private data size */
Tejun Heof95a04a2012-04-16 13:57:26 -0700112 size_t pd_size;
Tejun Heo36558c82012-04-16 13:57:24 -0700113 /* cgroup files for the policy */
114 struct cftype *cftypes;
Vivek Goyal3e252062009-12-04 10:36:42 -0500115};
116
Tejun Heo3c798392012-04-16 13:57:25 -0700117extern struct blkcg blkcg_root;
Tejun Heo36558c82012-04-16 13:57:24 -0700118
Tejun Heo3c798392012-04-16 13:57:25 -0700119struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup);
120struct blkcg *bio_blkcg(struct bio *bio);
121struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q);
122struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
123 struct request_queue *q);
Tejun Heo36558c82012-04-16 13:57:24 -0700124int blkcg_init_queue(struct request_queue *q);
125void blkcg_drain_queue(struct request_queue *q);
126void blkcg_exit_queue(struct request_queue *q);
Tejun Heo5efd6112012-03-05 13:15:12 -0800127
Vivek Goyal3e252062009-12-04 10:36:42 -0500128/* Blkio controller policy registration */
Tejun Heo3c798392012-04-16 13:57:25 -0700129int blkcg_policy_register(struct blkcg_policy *pol);
130void blkcg_policy_unregister(struct blkcg_policy *pol);
Tejun Heo36558c82012-04-16 13:57:24 -0700131int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700132 const struct blkcg_policy *pol);
Tejun Heo36558c82012-04-16 13:57:24 -0700133void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700134 const struct blkcg_policy *pol);
Vivek Goyal3e252062009-12-04 10:36:42 -0500135
Tejun Heo3c798392012-04-16 13:57:25 -0700136void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
Tejun Heof95a04a2012-04-16 13:57:26 -0700137 u64 (*prfill)(struct seq_file *,
138 struct blkg_policy_data *, int),
Tejun Heo3c798392012-04-16 13:57:25 -0700139 const struct blkcg_policy *pol, int data,
Tejun Heoec399342012-04-13 13:11:27 -0700140 bool show_total);
Tejun Heof95a04a2012-04-16 13:57:26 -0700141u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
142u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
Tejun Heo829fdb52012-04-01 14:38:43 -0700143 const struct blkg_rwstat *rwstat);
Tejun Heof95a04a2012-04-16 13:57:26 -0700144u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
145u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
146 int off);
Tejun Heo829fdb52012-04-01 14:38:43 -0700147
148struct blkg_conf_ctx {
Tejun Heo36558c82012-04-16 13:57:24 -0700149 struct gendisk *disk;
Tejun Heo3c798392012-04-16 13:57:25 -0700150 struct blkcg_gq *blkg;
Tejun Heo36558c82012-04-16 13:57:24 -0700151 u64 v;
Tejun Heo829fdb52012-04-01 14:38:43 -0700152};
153
Tejun Heo3c798392012-04-16 13:57:25 -0700154int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
155 const char *input, struct blkg_conf_ctx *ctx);
Tejun Heo829fdb52012-04-01 14:38:43 -0700156void blkg_conf_finish(struct blkg_conf_ctx *ctx);
157
158
Tejun Heo03814112012-03-05 13:15:14 -0800159/**
160 * blkg_to_pdata - get policy private data
161 * @blkg: blkg of interest
162 * @pol: policy of interest
163 *
164 * Return pointer to private data associated with the @blkg-@pol pair.
165 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700166static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
167 struct blkcg_policy *pol)
Tejun Heo03814112012-03-05 13:15:14 -0800168{
Tejun Heof95a04a2012-04-16 13:57:26 -0700169 return blkg ? blkg->pd[pol->plid] : NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800170}
171
172/**
173 * pdata_to_blkg - get blkg associated with policy private data
Tejun Heof95a04a2012-04-16 13:57:26 -0700174 * @pd: policy private data of interest
Tejun Heo03814112012-03-05 13:15:14 -0800175 *
Tejun Heof95a04a2012-04-16 13:57:26 -0700176 * @pd is policy private data. Determine the blkg it's associated with.
Tejun Heo03814112012-03-05 13:15:14 -0800177 */
Tejun Heof95a04a2012-04-16 13:57:26 -0700178static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
Tejun Heo03814112012-03-05 13:15:14 -0800179{
Tejun Heof95a04a2012-04-16 13:57:26 -0700180 return pd ? pd->blkg : NULL;
Tejun Heo03814112012-03-05 13:15:14 -0800181}
182
Tejun Heo54e7ed12012-04-16 13:57:23 -0700183/**
184 * blkg_path - format cgroup path of blkg
185 * @blkg: blkg of interest
186 * @buf: target buffer
187 * @buflen: target buffer length
188 *
189 * Format the path of the cgroup of @blkg into @buf.
190 */
Tejun Heo3c798392012-04-16 13:57:25 -0700191static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
Vivek Goyalafc24d42010-04-26 19:27:56 +0200192{
Tejun Heo54e7ed12012-04-16 13:57:23 -0700193 int ret;
194
195 rcu_read_lock();
196 ret = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
197 rcu_read_unlock();
198 if (ret)
199 strncpy(buf, "<unavailable>", buflen);
200 return ret;
Vivek Goyalafc24d42010-04-26 19:27:56 +0200201}
202
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800203/**
204 * blkg_get - get a blkg reference
205 * @blkg: blkg to get
206 *
207 * The caller should be holding queue_lock and an existing reference.
208 */
Tejun Heo3c798392012-04-16 13:57:25 -0700209static inline void blkg_get(struct blkcg_gq *blkg)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800210{
211 lockdep_assert_held(blkg->q->queue_lock);
212 WARN_ON_ONCE(!blkg->refcnt);
213 blkg->refcnt++;
214}
215
Tejun Heo3c798392012-04-16 13:57:25 -0700216void __blkg_release(struct blkcg_gq *blkg);
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800217
218/**
219 * blkg_put - put a blkg reference
220 * @blkg: blkg to put
221 *
222 * The caller should be holding queue_lock.
223 */
Tejun Heo3c798392012-04-16 13:57:25 -0700224static inline void blkg_put(struct blkcg_gq *blkg)
Tejun Heo1adaf3d2012-03-05 13:15:15 -0800225{
226 lockdep_assert_held(blkg->q->queue_lock);
227 WARN_ON_ONCE(blkg->refcnt <= 0);
228 if (!--blkg->refcnt)
229 __blkg_release(blkg);
230}
231
Tejun Heoedcb0722012-04-01 14:38:42 -0700232/**
233 * blkg_stat_add - add a value to a blkg_stat
234 * @stat: target blkg_stat
235 * @val: value to add
236 *
237 * Add @val to @stat. The caller is responsible for synchronizing calls to
238 * this function.
239 */
240static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
241{
242 u64_stats_update_begin(&stat->syncp);
243 stat->cnt += val;
244 u64_stats_update_end(&stat->syncp);
245}
246
247/**
248 * blkg_stat_read - read the current value of a blkg_stat
249 * @stat: blkg_stat to read
250 *
251 * Read the current value of @stat. This function can be called without
252 * synchroniztion and takes care of u64 atomicity.
253 */
254static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
255{
256 unsigned int start;
257 uint64_t v;
258
259 do {
260 start = u64_stats_fetch_begin(&stat->syncp);
261 v = stat->cnt;
262 } while (u64_stats_fetch_retry(&stat->syncp, start));
263
264 return v;
265}
266
267/**
268 * blkg_stat_reset - reset a blkg_stat
269 * @stat: blkg_stat to reset
270 */
271static inline void blkg_stat_reset(struct blkg_stat *stat)
272{
273 stat->cnt = 0;
274}
275
276/**
277 * blkg_rwstat_add - add a value to a blkg_rwstat
278 * @rwstat: target blkg_rwstat
279 * @rw: mask of REQ_{WRITE|SYNC}
280 * @val: value to add
281 *
282 * Add @val to @rwstat. The counters are chosen according to @rw. The
283 * caller is responsible for synchronizing calls to this function.
284 */
285static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
286 int rw, uint64_t val)
287{
288 u64_stats_update_begin(&rwstat->syncp);
289
290 if (rw & REQ_WRITE)
291 rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
292 else
293 rwstat->cnt[BLKG_RWSTAT_READ] += val;
294 if (rw & REQ_SYNC)
295 rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
296 else
297 rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
298
299 u64_stats_update_end(&rwstat->syncp);
300}
301
302/**
303 * blkg_rwstat_read - read the current values of a blkg_rwstat
304 * @rwstat: blkg_rwstat to read
305 *
306 * Read the current snapshot of @rwstat and return it as the return value.
307 * This function can be called without synchronization and takes care of
308 * u64 atomicity.
309 */
Tejun Heoc94bed892012-04-16 13:57:22 -0700310static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
Tejun Heoedcb0722012-04-01 14:38:42 -0700311{
312 unsigned int start;
313 struct blkg_rwstat tmp;
314
315 do {
316 start = u64_stats_fetch_begin(&rwstat->syncp);
317 tmp = *rwstat;
318 } while (u64_stats_fetch_retry(&rwstat->syncp, start));
319
320 return tmp;
321}
322
323/**
324 * blkg_rwstat_sum - read the total count of a blkg_rwstat
325 * @rwstat: blkg_rwstat to read
326 *
327 * Return the total count of @rwstat regardless of the IO direction. This
328 * function can be called without synchronization and takes care of u64
329 * atomicity.
330 */
331static inline uint64_t blkg_rwstat_sum(struct blkg_rwstat *rwstat)
332{
333 struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
334
335 return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
336}
337
338/**
339 * blkg_rwstat_reset - reset a blkg_rwstat
340 * @rwstat: blkg_rwstat to reset
341 */
342static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
343{
344 memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
345}
346
Tejun Heo36558c82012-04-16 13:57:24 -0700347#else /* CONFIG_BLK_CGROUP */
348
349struct cgroup;
Jens Axboe2f5ea472009-12-03 21:06:43 +0100350
Tejun Heof95a04a2012-04-16 13:57:26 -0700351struct blkg_policy_data {
352};
353
Tejun Heo3c798392012-04-16 13:57:25 -0700354struct blkcg_gq {
Jens Axboe2f5ea472009-12-03 21:06:43 +0100355};
356
Tejun Heo3c798392012-04-16 13:57:25 -0700357struct blkcg_policy {
Vivek Goyal3e252062009-12-04 10:36:42 -0500358};
359
Tejun Heo3c798392012-04-16 13:57:25 -0700360static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup) { return NULL; }
361static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
362static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
Tejun Heo5efd6112012-03-05 13:15:12 -0800363static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
364static inline void blkcg_drain_queue(struct request_queue *q) { }
365static inline void blkcg_exit_queue(struct request_queue *q) { }
Tejun Heo3c798392012-04-16 13:57:25 -0700366static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
367static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
Tejun Heoa2b16932012-04-13 13:11:33 -0700368static inline int blkcg_activate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700369 const struct blkcg_policy *pol) { return 0; }
Tejun Heoa2b16932012-04-13 13:11:33 -0700370static inline void blkcg_deactivate_policy(struct request_queue *q,
Tejun Heo3c798392012-04-16 13:57:25 -0700371 const struct blkcg_policy *pol) { }
Vivek Goyal3e252062009-12-04 10:36:42 -0500372
Tejun Heof95a04a2012-04-16 13:57:26 -0700373static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
374 struct blkcg_policy *pol) { return NULL; }
375static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
Tejun Heo3c798392012-04-16 13:57:25 -0700376static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
377static inline void blkg_get(struct blkcg_gq *blkg) { }
378static inline void blkg_put(struct blkcg_gq *blkg) { }
Vivek Goyalafc24d42010-04-26 19:27:56 +0200379
Tejun Heo36558c82012-04-16 13:57:24 -0700380#endif /* CONFIG_BLK_CGROUP */
381#endif /* _BLK_CGROUP_H */