xref: /linux/block/blk-cgroup.h (revision 26b0d14106954ae46d2f4f7eec3481828a210f7d)
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>
17 #include <linux/u64_stats_sync.h>
18 #include <linux/seq_file.h>
19 #include <linux/radix-tree.h>
20 
21 /* Max limits for throttle policy */
22 #define THROTL_IOPS_MAX		UINT_MAX
23 
24 /* CFQ specific, out here for blkcg->cfq_weight */
25 #define CFQ_WEIGHT_MIN		10
26 #define CFQ_WEIGHT_MAX		1000
27 #define CFQ_WEIGHT_DEFAULT	500
28 
29 #ifdef CONFIG_BLK_CGROUP
30 
31 enum blkg_rwstat_type {
32 	BLKG_RWSTAT_READ,
33 	BLKG_RWSTAT_WRITE,
34 	BLKG_RWSTAT_SYNC,
35 	BLKG_RWSTAT_ASYNC,
36 
37 	BLKG_RWSTAT_NR,
38 	BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
39 };
40 
41 struct blkcg_gq;
42 
43 struct blkcg {
44 	struct cgroup_subsys_state	css;
45 	spinlock_t			lock;
46 
47 	struct radix_tree_root		blkg_tree;
48 	struct blkcg_gq			*blkg_hint;
49 	struct hlist_head		blkg_list;
50 
51 	/* for policies to test whether associated blkcg has changed */
52 	uint64_t			id;
53 
54 	/* TODO: per-policy storage in blkcg */
55 	unsigned int			cfq_weight;	/* belongs to cfq */
56 };
57 
58 struct blkg_stat {
59 	struct u64_stats_sync		syncp;
60 	uint64_t			cnt;
61 };
62 
63 struct blkg_rwstat {
64 	struct u64_stats_sync		syncp;
65 	uint64_t			cnt[BLKG_RWSTAT_NR];
66 };
67 
68 /*
69  * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
70  * request_queue (q).  This is used by blkcg policies which need to track
71  * information per blkcg - q pair.
72  *
73  * There can be multiple active blkcg policies and each has its private
74  * data on each blkg, the size of which is determined by
75  * blkcg_policy->pd_size.  blkcg core allocates and frees such areas
76  * together with blkg and invokes pd_init/exit_fn() methods.
77  *
78  * Such private data must embed struct blkg_policy_data (pd) at the
79  * beginning and pd_size can't be smaller than pd.
80  */
81 struct blkg_policy_data {
82 	/* the blkg this per-policy data belongs to */
83 	struct blkcg_gq			*blkg;
84 
85 	/* used during policy activation */
86 	struct list_head		alloc_node;
87 };
88 
89 /* association between a blk cgroup and a request queue */
90 struct blkcg_gq {
91 	/* Pointer to the associated request_queue */
92 	struct request_queue		*q;
93 	struct list_head		q_node;
94 	struct hlist_node		blkcg_node;
95 	struct blkcg			*blkcg;
96 	/* reference count */
97 	int				refcnt;
98 
99 	struct blkg_policy_data		*pd[BLKCG_MAX_POLS];
100 
101 	struct rcu_head			rcu_head;
102 };
103 
104 typedef void (blkcg_pol_init_pd_fn)(struct blkcg_gq *blkg);
105 typedef void (blkcg_pol_exit_pd_fn)(struct blkcg_gq *blkg);
106 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkcg_gq *blkg);
107 
108 struct blkcg_policy {
109 	int				plid;
110 	/* policy specific private data size */
111 	size_t				pd_size;
112 	/* cgroup files for the policy */
113 	struct cftype			*cftypes;
114 
115 	/* operations */
116 	blkcg_pol_init_pd_fn		*pd_init_fn;
117 	blkcg_pol_exit_pd_fn		*pd_exit_fn;
118 	blkcg_pol_reset_pd_stats_fn	*pd_reset_stats_fn;
119 };
120 
121 extern struct blkcg blkcg_root;
122 
123 struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup);
124 struct blkcg *bio_blkcg(struct bio *bio);
125 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q);
126 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
127 				    struct request_queue *q);
128 int blkcg_init_queue(struct request_queue *q);
129 void blkcg_drain_queue(struct request_queue *q);
130 void blkcg_exit_queue(struct request_queue *q);
131 
132 /* Blkio controller policy registration */
133 int blkcg_policy_register(struct blkcg_policy *pol);
134 void blkcg_policy_unregister(struct blkcg_policy *pol);
135 int blkcg_activate_policy(struct request_queue *q,
136 			  const struct blkcg_policy *pol);
137 void blkcg_deactivate_policy(struct request_queue *q,
138 			     const struct blkcg_policy *pol);
139 
140 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
141 		       u64 (*prfill)(struct seq_file *,
142 				     struct blkg_policy_data *, int),
143 		       const struct blkcg_policy *pol, int data,
144 		       bool show_total);
145 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
146 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
147 			 const struct blkg_rwstat *rwstat);
148 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
149 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
150 		       int off);
151 
152 struct blkg_conf_ctx {
153 	struct gendisk			*disk;
154 	struct blkcg_gq			*blkg;
155 	u64				v;
156 };
157 
158 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
159 		   const char *input, struct blkg_conf_ctx *ctx);
160 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
161 
162 
163 /**
164  * blkg_to_pdata - get policy private data
165  * @blkg: blkg of interest
166  * @pol: policy of interest
167  *
168  * Return pointer to private data associated with the @blkg-@pol pair.
169  */
170 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
171 						  struct blkcg_policy *pol)
172 {
173 	return blkg ? blkg->pd[pol->plid] : NULL;
174 }
175 
176 /**
177  * pdata_to_blkg - get blkg associated with policy private data
178  * @pd: policy private data of interest
179  *
180  * @pd is policy private data.  Determine the blkg it's associated with.
181  */
182 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
183 {
184 	return pd ? pd->blkg : NULL;
185 }
186 
187 /**
188  * blkg_path - format cgroup path of blkg
189  * @blkg: blkg of interest
190  * @buf: target buffer
191  * @buflen: target buffer length
192  *
193  * Format the path of the cgroup of @blkg into @buf.
194  */
195 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
196 {
197 	int ret;
198 
199 	rcu_read_lock();
200 	ret = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
201 	rcu_read_unlock();
202 	if (ret)
203 		strncpy(buf, "<unavailable>", buflen);
204 	return ret;
205 }
206 
207 /**
208  * blkg_get - get a blkg reference
209  * @blkg: blkg to get
210  *
211  * The caller should be holding queue_lock and an existing reference.
212  */
213 static inline void blkg_get(struct blkcg_gq *blkg)
214 {
215 	lockdep_assert_held(blkg->q->queue_lock);
216 	WARN_ON_ONCE(!blkg->refcnt);
217 	blkg->refcnt++;
218 }
219 
220 void __blkg_release(struct blkcg_gq *blkg);
221 
222 /**
223  * blkg_put - put a blkg reference
224  * @blkg: blkg to put
225  *
226  * The caller should be holding queue_lock.
227  */
228 static inline void blkg_put(struct blkcg_gq *blkg)
229 {
230 	lockdep_assert_held(blkg->q->queue_lock);
231 	WARN_ON_ONCE(blkg->refcnt <= 0);
232 	if (!--blkg->refcnt)
233 		__blkg_release(blkg);
234 }
235 
236 /**
237  * blkg_stat_add - add a value to a blkg_stat
238  * @stat: target blkg_stat
239  * @val: value to add
240  *
241  * Add @val to @stat.  The caller is responsible for synchronizing calls to
242  * this function.
243  */
244 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
245 {
246 	u64_stats_update_begin(&stat->syncp);
247 	stat->cnt += val;
248 	u64_stats_update_end(&stat->syncp);
249 }
250 
251 /**
252  * blkg_stat_read - read the current value of a blkg_stat
253  * @stat: blkg_stat to read
254  *
255  * Read the current value of @stat.  This function can be called without
256  * synchroniztion and takes care of u64 atomicity.
257  */
258 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
259 {
260 	unsigned int start;
261 	uint64_t v;
262 
263 	do {
264 		start = u64_stats_fetch_begin(&stat->syncp);
265 		v = stat->cnt;
266 	} while (u64_stats_fetch_retry(&stat->syncp, start));
267 
268 	return v;
269 }
270 
271 /**
272  * blkg_stat_reset - reset a blkg_stat
273  * @stat: blkg_stat to reset
274  */
275 static inline void blkg_stat_reset(struct blkg_stat *stat)
276 {
277 	stat->cnt = 0;
278 }
279 
280 /**
281  * blkg_rwstat_add - add a value to a blkg_rwstat
282  * @rwstat: target blkg_rwstat
283  * @rw: mask of REQ_{WRITE|SYNC}
284  * @val: value to add
285  *
286  * Add @val to @rwstat.  The counters are chosen according to @rw.  The
287  * caller is responsible for synchronizing calls to this function.
288  */
289 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
290 				   int rw, uint64_t val)
291 {
292 	u64_stats_update_begin(&rwstat->syncp);
293 
294 	if (rw & REQ_WRITE)
295 		rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
296 	else
297 		rwstat->cnt[BLKG_RWSTAT_READ] += val;
298 	if (rw & REQ_SYNC)
299 		rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
300 	else
301 		rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
302 
303 	u64_stats_update_end(&rwstat->syncp);
304 }
305 
306 /**
307  * blkg_rwstat_read - read the current values of a blkg_rwstat
308  * @rwstat: blkg_rwstat to read
309  *
310  * Read the current snapshot of @rwstat and return it as the return value.
311  * This function can be called without synchronization and takes care of
312  * u64 atomicity.
313  */
314 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
315 {
316 	unsigned int start;
317 	struct blkg_rwstat tmp;
318 
319 	do {
320 		start = u64_stats_fetch_begin(&rwstat->syncp);
321 		tmp = *rwstat;
322 	} while (u64_stats_fetch_retry(&rwstat->syncp, start));
323 
324 	return tmp;
325 }
326 
327 /**
328  * blkg_rwstat_sum - read the total count of a blkg_rwstat
329  * @rwstat: blkg_rwstat to read
330  *
331  * Return the total count of @rwstat regardless of the IO direction.  This
332  * function can be called without synchronization and takes care of u64
333  * atomicity.
334  */
335 static inline uint64_t blkg_rwstat_sum(struct blkg_rwstat *rwstat)
336 {
337 	struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
338 
339 	return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
340 }
341 
342 /**
343  * blkg_rwstat_reset - reset a blkg_rwstat
344  * @rwstat: blkg_rwstat to reset
345  */
346 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
347 {
348 	memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
349 }
350 
351 #else	/* CONFIG_BLK_CGROUP */
352 
353 struct cgroup;
354 
355 struct blkg_policy_data {
356 };
357 
358 struct blkcg_gq {
359 };
360 
361 struct blkcg_policy {
362 };
363 
364 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup) { return NULL; }
365 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
366 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
367 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
368 static inline void blkcg_drain_queue(struct request_queue *q) { }
369 static inline void blkcg_exit_queue(struct request_queue *q) { }
370 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
371 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
372 static inline int blkcg_activate_policy(struct request_queue *q,
373 					const struct blkcg_policy *pol) { return 0; }
374 static inline void blkcg_deactivate_policy(struct request_queue *q,
375 					   const struct blkcg_policy *pol) { }
376 
377 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
378 						  struct blkcg_policy *pol) { return NULL; }
379 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
380 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
381 static inline void blkg_get(struct blkcg_gq *blkg) { }
382 static inline void blkg_put(struct blkcg_gq *blkg) { }
383 
384 #endif	/* CONFIG_BLK_CGROUP */
385 #endif	/* _BLK_CGROUP_H */
386