xref: /linux/block/blk-cgroup.h (revision 61d0b5a4b2777dcf5daef245e212b3c1fa8091ca)
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 #include <linux/blkdev.h>
21 
22 /* Max limits for throttle policy */
23 #define THROTL_IOPS_MAX		UINT_MAX
24 
25 /* CFQ specific, out here for blkcg->cfq_weight */
26 #define CFQ_WEIGHT_MIN		10
27 #define CFQ_WEIGHT_MAX		1000
28 #define CFQ_WEIGHT_DEFAULT	500
29 
30 #ifdef CONFIG_BLK_CGROUP
31 
32 enum blkg_rwstat_type {
33 	BLKG_RWSTAT_READ,
34 	BLKG_RWSTAT_WRITE,
35 	BLKG_RWSTAT_SYNC,
36 	BLKG_RWSTAT_ASYNC,
37 
38 	BLKG_RWSTAT_NR,
39 	BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
40 };
41 
42 struct blkcg_gq;
43 
44 struct blkcg {
45 	struct cgroup_subsys_state	css;
46 	spinlock_t			lock;
47 
48 	struct radix_tree_root		blkg_tree;
49 	struct blkcg_gq			*blkg_hint;
50 	struct hlist_head		blkg_list;
51 
52 	/* for policies to test whether associated blkcg has changed */
53 	uint64_t			id;
54 
55 	/* TODO: per-policy storage in blkcg */
56 	unsigned int			cfq_weight;	/* belongs to cfq */
57 	unsigned int			cfq_leaf_weight;
58 };
59 
60 struct blkg_stat {
61 	struct u64_stats_sync		syncp;
62 	uint64_t			cnt;
63 };
64 
65 struct blkg_rwstat {
66 	struct u64_stats_sync		syncp;
67 	uint64_t			cnt[BLKG_RWSTAT_NR];
68 };
69 
70 /*
71  * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
72  * request_queue (q).  This is used by blkcg policies which need to track
73  * information per blkcg - q pair.
74  *
75  * There can be multiple active blkcg policies and each has its private
76  * data on each blkg, the size of which is determined by
77  * blkcg_policy->pd_size.  blkcg core allocates and frees such areas
78  * together with blkg and invokes pd_init/exit_fn() methods.
79  *
80  * Such private data must embed struct blkg_policy_data (pd) at the
81  * beginning and pd_size can't be smaller than pd.
82  */
83 struct blkg_policy_data {
84 	/* the blkg and policy id this per-policy data belongs to */
85 	struct blkcg_gq			*blkg;
86 	int				plid;
87 
88 	/* used during policy activation */
89 	struct list_head		alloc_node;
90 };
91 
92 /* association between a blk cgroup and a request queue */
93 struct blkcg_gq {
94 	/* Pointer to the associated request_queue */
95 	struct request_queue		*q;
96 	struct list_head		q_node;
97 	struct hlist_node		blkcg_node;
98 	struct blkcg			*blkcg;
99 
100 	/* all non-root blkcg_gq's are guaranteed to have access to parent */
101 	struct blkcg_gq			*parent;
102 
103 	/* request allocation list for this blkcg-q pair */
104 	struct request_list		rl;
105 
106 	/* reference count */
107 	int				refcnt;
108 
109 	/* is this blkg online? protected by both blkcg and q locks */
110 	bool				online;
111 
112 	struct blkg_policy_data		*pd[BLKCG_MAX_POLS];
113 
114 	struct rcu_head			rcu_head;
115 };
116 
117 typedef void (blkcg_pol_init_pd_fn)(struct blkcg_gq *blkg);
118 typedef void (blkcg_pol_online_pd_fn)(struct blkcg_gq *blkg);
119 typedef void (blkcg_pol_offline_pd_fn)(struct blkcg_gq *blkg);
120 typedef void (blkcg_pol_exit_pd_fn)(struct blkcg_gq *blkg);
121 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkcg_gq *blkg);
122 
123 struct blkcg_policy {
124 	int				plid;
125 	/* policy specific private data size */
126 	size_t				pd_size;
127 	/* cgroup files for the policy */
128 	struct cftype			*cftypes;
129 
130 	/* operations */
131 	blkcg_pol_init_pd_fn		*pd_init_fn;
132 	blkcg_pol_online_pd_fn		*pd_online_fn;
133 	blkcg_pol_offline_pd_fn		*pd_offline_fn;
134 	blkcg_pol_exit_pd_fn		*pd_exit_fn;
135 	blkcg_pol_reset_pd_stats_fn	*pd_reset_stats_fn;
136 };
137 
138 extern struct blkcg blkcg_root;
139 
140 struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, struct request_queue *q);
141 struct blkcg_gq *blkg_lookup_create(struct blkcg *blkcg,
142 				    struct request_queue *q);
143 int blkcg_init_queue(struct request_queue *q);
144 void blkcg_drain_queue(struct request_queue *q);
145 void blkcg_exit_queue(struct request_queue *q);
146 
147 /* Blkio controller policy registration */
148 int blkcg_policy_register(struct blkcg_policy *pol);
149 void blkcg_policy_unregister(struct blkcg_policy *pol);
150 int blkcg_activate_policy(struct request_queue *q,
151 			  const struct blkcg_policy *pol);
152 void blkcg_deactivate_policy(struct request_queue *q,
153 			     const struct blkcg_policy *pol);
154 
155 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
156 		       u64 (*prfill)(struct seq_file *,
157 				     struct blkg_policy_data *, int),
158 		       const struct blkcg_policy *pol, int data,
159 		       bool show_total);
160 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
161 u64 __blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
162 			 const struct blkg_rwstat *rwstat);
163 u64 blkg_prfill_stat(struct seq_file *sf, struct blkg_policy_data *pd, int off);
164 u64 blkg_prfill_rwstat(struct seq_file *sf, struct blkg_policy_data *pd,
165 		       int off);
166 
167 u64 blkg_stat_recursive_sum(struct blkg_policy_data *pd, int off);
168 struct blkg_rwstat blkg_rwstat_recursive_sum(struct blkg_policy_data *pd,
169 					     int off);
170 
171 struct blkg_conf_ctx {
172 	struct gendisk			*disk;
173 	struct blkcg_gq			*blkg;
174 	u64				v;
175 };
176 
177 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
178 		   const char *input, struct blkg_conf_ctx *ctx);
179 void blkg_conf_finish(struct blkg_conf_ctx *ctx);
180 
181 
182 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup)
183 {
184 	return container_of(cgroup_subsys_state(cgroup, blkio_subsys_id),
185 			    struct blkcg, css);
186 }
187 
188 static inline struct blkcg *task_blkcg(struct task_struct *tsk)
189 {
190 	return container_of(task_subsys_state(tsk, blkio_subsys_id),
191 			    struct blkcg, css);
192 }
193 
194 static inline struct blkcg *bio_blkcg(struct bio *bio)
195 {
196 	if (bio && bio->bi_css)
197 		return container_of(bio->bi_css, struct blkcg, css);
198 	return task_blkcg(current);
199 }
200 
201 /**
202  * blkcg_parent - get the parent of a blkcg
203  * @blkcg: blkcg of interest
204  *
205  * Return the parent blkcg of @blkcg.  Can be called anytime.
206  */
207 static inline struct blkcg *blkcg_parent(struct blkcg *blkcg)
208 {
209 	struct cgroup *pcg = blkcg->css.cgroup->parent;
210 
211 	return pcg ? cgroup_to_blkcg(pcg) : NULL;
212 }
213 
214 /**
215  * blkg_to_pdata - get policy private data
216  * @blkg: blkg of interest
217  * @pol: policy of interest
218  *
219  * Return pointer to private data associated with the @blkg-@pol pair.
220  */
221 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
222 						  struct blkcg_policy *pol)
223 {
224 	return blkg ? blkg->pd[pol->plid] : NULL;
225 }
226 
227 /**
228  * pdata_to_blkg - get blkg associated with policy private data
229  * @pd: policy private data of interest
230  *
231  * @pd is policy private data.  Determine the blkg it's associated with.
232  */
233 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
234 {
235 	return pd ? pd->blkg : NULL;
236 }
237 
238 /**
239  * blkg_path - format cgroup path of blkg
240  * @blkg: blkg of interest
241  * @buf: target buffer
242  * @buflen: target buffer length
243  *
244  * Format the path of the cgroup of @blkg into @buf.
245  */
246 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
247 {
248 	int ret;
249 
250 	rcu_read_lock();
251 	ret = cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
252 	rcu_read_unlock();
253 	if (ret)
254 		strncpy(buf, "<unavailable>", buflen);
255 	return ret;
256 }
257 
258 /**
259  * blkg_get - get a blkg reference
260  * @blkg: blkg to get
261  *
262  * The caller should be holding queue_lock and an existing reference.
263  */
264 static inline void blkg_get(struct blkcg_gq *blkg)
265 {
266 	lockdep_assert_held(blkg->q->queue_lock);
267 	WARN_ON_ONCE(!blkg->refcnt);
268 	blkg->refcnt++;
269 }
270 
271 void __blkg_release(struct blkcg_gq *blkg);
272 
273 /**
274  * blkg_put - put a blkg reference
275  * @blkg: blkg to put
276  *
277  * The caller should be holding queue_lock.
278  */
279 static inline void blkg_put(struct blkcg_gq *blkg)
280 {
281 	lockdep_assert_held(blkg->q->queue_lock);
282 	WARN_ON_ONCE(blkg->refcnt <= 0);
283 	if (!--blkg->refcnt)
284 		__blkg_release(blkg);
285 }
286 
287 /**
288  * blk_get_rl - get request_list to use
289  * @q: request_queue of interest
290  * @bio: bio which will be attached to the allocated request (may be %NULL)
291  *
292  * The caller wants to allocate a request from @q to use for @bio.  Find
293  * the request_list to use and obtain a reference on it.  Should be called
294  * under queue_lock.  This function is guaranteed to return non-%NULL
295  * request_list.
296  */
297 static inline struct request_list *blk_get_rl(struct request_queue *q,
298 					      struct bio *bio)
299 {
300 	struct blkcg *blkcg;
301 	struct blkcg_gq *blkg;
302 
303 	rcu_read_lock();
304 
305 	blkcg = bio_blkcg(bio);
306 
307 	/* bypass blkg lookup and use @q->root_rl directly for root */
308 	if (blkcg == &blkcg_root)
309 		goto root_rl;
310 
311 	/*
312 	 * Try to use blkg->rl.  blkg lookup may fail under memory pressure
313 	 * or if either the blkcg or queue is going away.  Fall back to
314 	 * root_rl in such cases.
315 	 */
316 	blkg = blkg_lookup_create(blkcg, q);
317 	if (unlikely(IS_ERR(blkg)))
318 		goto root_rl;
319 
320 	blkg_get(blkg);
321 	rcu_read_unlock();
322 	return &blkg->rl;
323 root_rl:
324 	rcu_read_unlock();
325 	return &q->root_rl;
326 }
327 
328 /**
329  * blk_put_rl - put request_list
330  * @rl: request_list to put
331  *
332  * Put the reference acquired by blk_get_rl().  Should be called under
333  * queue_lock.
334  */
335 static inline void blk_put_rl(struct request_list *rl)
336 {
337 	/* root_rl may not have blkg set */
338 	if (rl->blkg && rl->blkg->blkcg != &blkcg_root)
339 		blkg_put(rl->blkg);
340 }
341 
342 /**
343  * blk_rq_set_rl - associate a request with a request_list
344  * @rq: request of interest
345  * @rl: target request_list
346  *
347  * Associate @rq with @rl so that accounting and freeing can know the
348  * request_list @rq came from.
349  */
350 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl)
351 {
352 	rq->rl = rl;
353 }
354 
355 /**
356  * blk_rq_rl - return the request_list a request came from
357  * @rq: request of interest
358  *
359  * Return the request_list @rq is allocated from.
360  */
361 static inline struct request_list *blk_rq_rl(struct request *rq)
362 {
363 	return rq->rl;
364 }
365 
366 struct request_list *__blk_queue_next_rl(struct request_list *rl,
367 					 struct request_queue *q);
368 /**
369  * blk_queue_for_each_rl - iterate through all request_lists of a request_queue
370  *
371  * Should be used under queue_lock.
372  */
373 #define blk_queue_for_each_rl(rl, q)	\
374 	for ((rl) = &(q)->root_rl; (rl); (rl) = __blk_queue_next_rl((rl), (q)))
375 
376 /**
377  * blkg_stat_add - add a value to a blkg_stat
378  * @stat: target blkg_stat
379  * @val: value to add
380  *
381  * Add @val to @stat.  The caller is responsible for synchronizing calls to
382  * this function.
383  */
384 static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
385 {
386 	u64_stats_update_begin(&stat->syncp);
387 	stat->cnt += val;
388 	u64_stats_update_end(&stat->syncp);
389 }
390 
391 /**
392  * blkg_stat_read - read the current value of a blkg_stat
393  * @stat: blkg_stat to read
394  *
395  * Read the current value of @stat.  This function can be called without
396  * synchroniztion and takes care of u64 atomicity.
397  */
398 static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
399 {
400 	unsigned int start;
401 	uint64_t v;
402 
403 	do {
404 		start = u64_stats_fetch_begin(&stat->syncp);
405 		v = stat->cnt;
406 	} while (u64_stats_fetch_retry(&stat->syncp, start));
407 
408 	return v;
409 }
410 
411 /**
412  * blkg_stat_reset - reset a blkg_stat
413  * @stat: blkg_stat to reset
414  */
415 static inline void blkg_stat_reset(struct blkg_stat *stat)
416 {
417 	stat->cnt = 0;
418 }
419 
420 /**
421  * blkg_stat_merge - merge a blkg_stat into another
422  * @to: the destination blkg_stat
423  * @from: the source
424  *
425  * Add @from's count to @to.
426  */
427 static inline void blkg_stat_merge(struct blkg_stat *to, struct blkg_stat *from)
428 {
429 	blkg_stat_add(to, blkg_stat_read(from));
430 }
431 
432 /**
433  * blkg_rwstat_add - add a value to a blkg_rwstat
434  * @rwstat: target blkg_rwstat
435  * @rw: mask of REQ_{WRITE|SYNC}
436  * @val: value to add
437  *
438  * Add @val to @rwstat.  The counters are chosen according to @rw.  The
439  * caller is responsible for synchronizing calls to this function.
440  */
441 static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
442 				   int rw, uint64_t val)
443 {
444 	u64_stats_update_begin(&rwstat->syncp);
445 
446 	if (rw & REQ_WRITE)
447 		rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
448 	else
449 		rwstat->cnt[BLKG_RWSTAT_READ] += val;
450 	if (rw & REQ_SYNC)
451 		rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
452 	else
453 		rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
454 
455 	u64_stats_update_end(&rwstat->syncp);
456 }
457 
458 /**
459  * blkg_rwstat_read - read the current values of a blkg_rwstat
460  * @rwstat: blkg_rwstat to read
461  *
462  * Read the current snapshot of @rwstat and return it as the return value.
463  * This function can be called without synchronization and takes care of
464  * u64 atomicity.
465  */
466 static inline struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
467 {
468 	unsigned int start;
469 	struct blkg_rwstat tmp;
470 
471 	do {
472 		start = u64_stats_fetch_begin(&rwstat->syncp);
473 		tmp = *rwstat;
474 	} while (u64_stats_fetch_retry(&rwstat->syncp, start));
475 
476 	return tmp;
477 }
478 
479 /**
480  * blkg_rwstat_total - read the total count of a blkg_rwstat
481  * @rwstat: blkg_rwstat to read
482  *
483  * Return the total count of @rwstat regardless of the IO direction.  This
484  * function can be called without synchronization and takes care of u64
485  * atomicity.
486  */
487 static inline uint64_t blkg_rwstat_total(struct blkg_rwstat *rwstat)
488 {
489 	struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
490 
491 	return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
492 }
493 
494 /**
495  * blkg_rwstat_reset - reset a blkg_rwstat
496  * @rwstat: blkg_rwstat to reset
497  */
498 static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
499 {
500 	memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
501 }
502 
503 /**
504  * blkg_rwstat_merge - merge a blkg_rwstat into another
505  * @to: the destination blkg_rwstat
506  * @from: the source
507  *
508  * Add @from's counts to @to.
509  */
510 static inline void blkg_rwstat_merge(struct blkg_rwstat *to,
511 				     struct blkg_rwstat *from)
512 {
513 	struct blkg_rwstat v = blkg_rwstat_read(from);
514 	int i;
515 
516 	u64_stats_update_begin(&to->syncp);
517 	for (i = 0; i < BLKG_RWSTAT_NR; i++)
518 		to->cnt[i] += v.cnt[i];
519 	u64_stats_update_end(&to->syncp);
520 }
521 
522 #else	/* CONFIG_BLK_CGROUP */
523 
524 struct cgroup;
525 struct blkcg;
526 
527 struct blkg_policy_data {
528 };
529 
530 struct blkcg_gq {
531 };
532 
533 struct blkcg_policy {
534 };
535 
536 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
537 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
538 static inline void blkcg_drain_queue(struct request_queue *q) { }
539 static inline void blkcg_exit_queue(struct request_queue *q) { }
540 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
541 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
542 static inline int blkcg_activate_policy(struct request_queue *q,
543 					const struct blkcg_policy *pol) { return 0; }
544 static inline void blkcg_deactivate_policy(struct request_queue *q,
545 					   const struct blkcg_policy *pol) { }
546 
547 static inline struct blkcg *cgroup_to_blkcg(struct cgroup *cgroup) { return NULL; }
548 static inline struct blkcg *bio_blkcg(struct bio *bio) { return NULL; }
549 
550 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
551 						  struct blkcg_policy *pol) { return NULL; }
552 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
553 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
554 static inline void blkg_get(struct blkcg_gq *blkg) { }
555 static inline void blkg_put(struct blkcg_gq *blkg) { }
556 
557 static inline struct request_list *blk_get_rl(struct request_queue *q,
558 					      struct bio *bio) { return &q->root_rl; }
559 static inline void blk_put_rl(struct request_list *rl) { }
560 static inline void blk_rq_set_rl(struct request *rq, struct request_list *rl) { }
561 static inline struct request_list *blk_rq_rl(struct request *rq) { return &rq->q->root_rl; }
562 
563 #define blk_queue_for_each_rl(rl, q)	\
564 	for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
565 
566 #endif	/* CONFIG_BLK_CGROUP */
567 #endif	/* _BLK_CGROUP_H */
568