xref: /linux/block/blk-cgroup.h (revision c21614d5c0e76cf1d7a864f15d495fd3e5876acc)
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BLK_CGROUP_PRIVATE_H
3 #define _BLK_CGROUP_PRIVATE_H
4 /*
5  * block cgroup private header
6  *
7  * Based on ideas and code from CFQ, CFS and BFQ:
8  * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
9  *
10  * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
11  *		      Paolo Valente <paolo.valente@unimore.it>
12  *
13  * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
14  * 	              Nauman Rafique <nauman@google.com>
15  */
16 
17 #include <linux/blk-cgroup.h>
18 #include <linux/cgroup.h>
19 #include <linux/kthread.h>
20 #include <linux/blk-mq.h>
21 #include <linux/llist.h>
22 #include "blk.h"
23 
24 struct blkcg_gq;
25 struct blkg_policy_data;
26 
27 
28 /* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
29 #define BLKG_STAT_CPU_BATCH	(INT_MAX / 2)
30 
31 #ifdef CONFIG_BLK_CGROUP
32 
33 enum blkg_iostat_type {
34 	BLKG_IOSTAT_READ,
35 	BLKG_IOSTAT_WRITE,
36 	BLKG_IOSTAT_DISCARD,
37 
38 	BLKG_IOSTAT_NR,
39 };
40 
41 struct blkg_iostat {
42 	u64				bytes[BLKG_IOSTAT_NR];
43 	u64				ios[BLKG_IOSTAT_NR];
44 };
45 
46 struct blkg_iostat_set {
47 	struct u64_stats_sync		sync;
48 	struct blkcg_gq		       *blkg;
49 	struct llist_node		lnode;
50 	int				lqueued;	/* queued in llist */
51 	struct blkg_iostat		cur;
52 	struct blkg_iostat		last;
53 };
54 
55 /* association between a blk cgroup and a request queue */
56 struct blkcg_gq {
57 	/* Pointer to the associated request_queue */
58 	struct request_queue		*q;
59 	struct list_head		q_node;
60 	struct hlist_node		blkcg_node;
61 	struct blkcg			*blkcg;
62 
63 	/* all non-root blkcg_gq's are guaranteed to have access to parent */
64 	struct blkcg_gq			*parent;
65 
66 	/* reference count */
67 	struct percpu_ref		refcnt;
68 
69 	/* is this blkg online? protected by both blkcg and q locks */
70 	bool				online;
71 
72 	struct blkg_iostat_set __percpu	*iostat_cpu;
73 	struct blkg_iostat_set		iostat;
74 
75 	struct blkg_policy_data		*pd[BLKCG_MAX_POLS];
76 #ifdef CONFIG_BLK_CGROUP_PUNT_BIO
77 	spinlock_t			async_bio_lock;
78 	struct bio_list			async_bios;
79 #endif
80 	union {
81 		struct work_struct	async_bio_work;
82 		struct work_struct	free_work;
83 	};
84 
85 	atomic_t			use_delay;
86 	atomic64_t			delay_nsec;
87 	atomic64_t			delay_start;
88 	u64				last_delay;
89 	int				last_use;
90 
91 	struct rcu_head			rcu_head;
92 };
93 
94 struct blkcg {
95 	struct cgroup_subsys_state	css;
96 	spinlock_t			lock;
97 	refcount_t			online_pin;
98 
99 	struct radix_tree_root		blkg_tree;
100 	struct blkcg_gq	__rcu		*blkg_hint;
101 	struct hlist_head		blkg_list;
102 
103 	struct blkcg_policy_data	*cpd[BLKCG_MAX_POLS];
104 
105 	struct list_head		all_blkcgs_node;
106 
107 	/*
108 	 * List of updated percpu blkg_iostat_set's since the last flush.
109 	 */
110 	struct llist_head __percpu	*lhead;
111 
112 #ifdef CONFIG_BLK_CGROUP_FC_APPID
113 	char                            fc_app_id[FC_APPID_LEN];
114 #endif
115 #ifdef CONFIG_CGROUP_WRITEBACK
116 	struct list_head		cgwb_list;
117 #endif
118 };
119 
120 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
121 {
122 	return css ? container_of(css, struct blkcg, css) : NULL;
123 }
124 
125 /*
126  * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
127  * request_queue (q).  This is used by blkcg policies which need to track
128  * information per blkcg - q pair.
129  *
130  * There can be multiple active blkcg policies and each blkg:policy pair is
131  * represented by a blkg_policy_data which is allocated and freed by each
132  * policy's pd_alloc/free_fn() methods.  A policy can allocate private data
133  * area by allocating larger data structure which embeds blkg_policy_data
134  * at the beginning.
135  */
136 struct blkg_policy_data {
137 	/* the blkg and policy id this per-policy data belongs to */
138 	struct blkcg_gq			*blkg;
139 	int				plid;
140 	bool				online;
141 };
142 
143 /*
144  * Policies that need to keep per-blkcg data which is independent from any
145  * request_queue associated to it should implement cpd_alloc/free_fn()
146  * methods.  A policy can allocate private data area by allocating larger
147  * data structure which embeds blkcg_policy_data at the beginning.
148  * cpd_init() is invoked to let each policy handle per-blkcg data.
149  */
150 struct blkcg_policy_data {
151 	/* the blkcg and policy id this per-policy data belongs to */
152 	struct blkcg			*blkcg;
153 	int				plid;
154 };
155 
156 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
157 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
158 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
159 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
160 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(struct gendisk *disk,
161 		struct blkcg *blkcg, gfp_t gfp);
162 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
163 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
164 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
165 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
166 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
167 typedef void (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd,
168 				struct seq_file *s);
169 
170 struct blkcg_policy {
171 	int				plid;
172 	/* cgroup files for the policy */
173 	struct cftype			*dfl_cftypes;
174 	struct cftype			*legacy_cftypes;
175 
176 	/* operations */
177 	blkcg_pol_alloc_cpd_fn		*cpd_alloc_fn;
178 	blkcg_pol_free_cpd_fn		*cpd_free_fn;
179 
180 	blkcg_pol_alloc_pd_fn		*pd_alloc_fn;
181 	blkcg_pol_init_pd_fn		*pd_init_fn;
182 	blkcg_pol_online_pd_fn		*pd_online_fn;
183 	blkcg_pol_offline_pd_fn		*pd_offline_fn;
184 	blkcg_pol_free_pd_fn		*pd_free_fn;
185 	blkcg_pol_reset_pd_stats_fn	*pd_reset_stats_fn;
186 	blkcg_pol_stat_pd_fn		*pd_stat_fn;
187 };
188 
189 extern struct blkcg blkcg_root;
190 extern bool blkcg_debug_stats;
191 
192 int blkcg_init_disk(struct gendisk *disk);
193 void blkcg_exit_disk(struct gendisk *disk);
194 
195 /* Blkio controller policy registration */
196 int blkcg_policy_register(struct blkcg_policy *pol);
197 void blkcg_policy_unregister(struct blkcg_policy *pol);
198 int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol);
199 void blkcg_deactivate_policy(struct gendisk *disk,
200 			     const struct blkcg_policy *pol);
201 
202 const char *blkg_dev_name(struct blkcg_gq *blkg);
203 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
204 		       u64 (*prfill)(struct seq_file *,
205 				     struct blkg_policy_data *, int),
206 		       const struct blkcg_policy *pol, int data,
207 		       bool show_total);
208 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
209 
210 struct blkg_conf_ctx {
211 	char				*input;
212 	char				*body;
213 	struct block_device		*bdev;
214 	struct blkcg_gq			*blkg;
215 };
216 
217 void blkg_conf_init(struct blkg_conf_ctx *ctx, char *input);
218 int blkg_conf_open_bdev(struct blkg_conf_ctx *ctx);
219 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
220 		   struct blkg_conf_ctx *ctx);
221 void blkg_conf_exit(struct blkg_conf_ctx *ctx);
222 
223 /**
224  * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
225  * @return: true if this bio needs to be submitted with the root blkg context.
226  *
227  * In order to avoid priority inversions we sometimes need to issue a bio as if
228  * it were attached to the root blkg, and then backcharge to the actual owning
229  * blkg.  The idea is we do bio_blkcg_css() to look up the actual context for
230  * the bio and attach the appropriate blkg to the bio.  Then we call this helper
231  * and if it is true run with the root blkg for that queue and then do any
232  * backcharging to the originating cgroup once the io is complete.
233  */
234 static inline bool bio_issue_as_root_blkg(struct bio *bio)
235 {
236 	return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
237 }
238 
239 /**
240  * blkg_lookup - lookup blkg for the specified blkcg - q pair
241  * @blkcg: blkcg of interest
242  * @q: request_queue of interest
243  *
244  * Lookup blkg for the @blkcg - @q pair.
245 
246  * Must be called in a RCU critical section.
247  */
248 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
249 					   struct request_queue *q)
250 {
251 	struct blkcg_gq *blkg;
252 
253 	if (blkcg == &blkcg_root)
254 		return q->root_blkg;
255 
256 	blkg = rcu_dereference_check(blkcg->blkg_hint,
257 			lockdep_is_held(&q->queue_lock));
258 	if (blkg && blkg->q == q)
259 		return blkg;
260 
261 	blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
262 	if (blkg && blkg->q != q)
263 		blkg = NULL;
264 	return blkg;
265 }
266 
267 /**
268  * blkg_to_pdata - get policy private data
269  * @blkg: blkg of interest
270  * @pol: policy of interest
271  *
272  * Return pointer to private data associated with the @blkg-@pol pair.
273  */
274 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
275 						  struct blkcg_policy *pol)
276 {
277 	return blkg ? blkg->pd[pol->plid] : NULL;
278 }
279 
280 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
281 						     struct blkcg_policy *pol)
282 {
283 	return blkcg ? blkcg->cpd[pol->plid] : NULL;
284 }
285 
286 /**
287  * pdata_to_blkg - get blkg associated with policy private data
288  * @pd: policy private data of interest
289  *
290  * @pd is policy private data.  Determine the blkg it's associated with.
291  */
292 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
293 {
294 	return pd ? pd->blkg : NULL;
295 }
296 
297 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
298 {
299 	return cpd ? cpd->blkcg : NULL;
300 }
301 
302 /**
303  * blkg_path - format cgroup path of blkg
304  * @blkg: blkg of interest
305  * @buf: target buffer
306  * @buflen: target buffer length
307  *
308  * Format the path of the cgroup of @blkg into @buf.
309  */
310 static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
311 {
312 	return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
313 }
314 
315 /**
316  * blkg_get - get a blkg reference
317  * @blkg: blkg to get
318  *
319  * The caller should be holding an existing reference.
320  */
321 static inline void blkg_get(struct blkcg_gq *blkg)
322 {
323 	percpu_ref_get(&blkg->refcnt);
324 }
325 
326 /**
327  * blkg_tryget - try and get a blkg reference
328  * @blkg: blkg to get
329  *
330  * This is for use when doing an RCU lookup of the blkg.  We may be in the midst
331  * of freeing this blkg, so we can only use it if the refcnt is not zero.
332  */
333 static inline bool blkg_tryget(struct blkcg_gq *blkg)
334 {
335 	return blkg && percpu_ref_tryget(&blkg->refcnt);
336 }
337 
338 /**
339  * blkg_put - put a blkg reference
340  * @blkg: blkg to put
341  */
342 static inline void blkg_put(struct blkcg_gq *blkg)
343 {
344 	percpu_ref_put(&blkg->refcnt);
345 }
346 
347 /**
348  * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
349  * @d_blkg: loop cursor pointing to the current descendant
350  * @pos_css: used for iteration
351  * @p_blkg: target blkg to walk descendants of
352  *
353  * Walk @c_blkg through the descendants of @p_blkg.  Must be used with RCU
354  * read locked.  If called under either blkcg or queue lock, the iteration
355  * is guaranteed to include all and only online blkgs.  The caller may
356  * update @pos_css by calling css_rightmost_descendant() to skip subtree.
357  * @p_blkg is included in the iteration and the first node to be visited.
358  */
359 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg)		\
360 	css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css)	\
361 		if (((d_blkg) = blkg_lookup(css_to_blkcg(pos_css),	\
362 					    (p_blkg)->q)))
363 
364 /**
365  * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
366  * @d_blkg: loop cursor pointing to the current descendant
367  * @pos_css: used for iteration
368  * @p_blkg: target blkg to walk descendants of
369  *
370  * Similar to blkg_for_each_descendant_pre() but performs post-order
371  * traversal instead.  Synchronization rules are the same.  @p_blkg is
372  * included in the iteration and the last node to be visited.
373  */
374 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg)		\
375 	css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css)	\
376 		if (((d_blkg) = blkg_lookup(css_to_blkcg(pos_css),	\
377 					    (p_blkg)->q)))
378 
379 static inline void blkcg_bio_issue_init(struct bio *bio)
380 {
381 	bio_issue_init(&bio->bi_issue, bio_sectors(bio));
382 }
383 
384 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
385 {
386 	if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0))
387 		return;
388 	if (atomic_add_return(1, &blkg->use_delay) == 1)
389 		atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
390 }
391 
392 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
393 {
394 	int old = atomic_read(&blkg->use_delay);
395 
396 	if (WARN_ON_ONCE(old < 0))
397 		return 0;
398 	if (old == 0)
399 		return 0;
400 
401 	/*
402 	 * We do this song and dance because we can race with somebody else
403 	 * adding or removing delay.  If we just did an atomic_dec we'd end up
404 	 * negative and we'd already be in trouble.  We need to subtract 1 and
405 	 * then check to see if we were the last delay so we can drop the
406 	 * congestion count on the cgroup.
407 	 */
408 	while (old && !atomic_try_cmpxchg(&blkg->use_delay, &old, old - 1))
409 		;
410 
411 	if (old == 0)
412 		return 0;
413 	if (old == 1)
414 		atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
415 	return 1;
416 }
417 
418 /**
419  * blkcg_set_delay - Enable allocator delay mechanism with the specified delay amount
420  * @blkg: target blkg
421  * @delay: delay duration in nsecs
422  *
423  * When enabled with this function, the delay is not decayed and must be
424  * explicitly cleared with blkcg_clear_delay(). Must not be mixed with
425  * blkcg_[un]use_delay() and blkcg_add_delay() usages.
426  */
427 static inline void blkcg_set_delay(struct blkcg_gq *blkg, u64 delay)
428 {
429 	int old = atomic_read(&blkg->use_delay);
430 
431 	/* We only want 1 person setting the congestion count for this blkg. */
432 	if (!old && atomic_try_cmpxchg(&blkg->use_delay, &old, -1))
433 		atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
434 
435 	atomic64_set(&blkg->delay_nsec, delay);
436 }
437 
438 /**
439  * blkcg_clear_delay - Disable allocator delay mechanism
440  * @blkg: target blkg
441  *
442  * Disable use_delay mechanism. See blkcg_set_delay().
443  */
444 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
445 {
446 	int old = atomic_read(&blkg->use_delay);
447 
448 	/* We only want 1 person clearing the congestion count for this blkg. */
449 	if (old && atomic_try_cmpxchg(&blkg->use_delay, &old, 0))
450 		atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
451 }
452 
453 /**
454  * blk_cgroup_mergeable - Determine whether to allow or disallow merges
455  * @rq: request to merge into
456  * @bio: bio to merge
457  *
458  * @bio and @rq should belong to the same cgroup and their issue_as_root should
459  * match. The latter is necessary as we don't want to throttle e.g. a metadata
460  * update because it happens to be next to a regular IO.
461  */
462 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio)
463 {
464 	return rq->bio->bi_blkg == bio->bi_blkg &&
465 		bio_issue_as_root_blkg(rq->bio) == bio_issue_as_root_blkg(bio);
466 }
467 
468 void blk_cgroup_bio_start(struct bio *bio);
469 void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
470 #else	/* CONFIG_BLK_CGROUP */
471 
472 struct blkg_policy_data {
473 };
474 
475 struct blkcg_policy_data {
476 };
477 
478 struct blkcg_policy {
479 };
480 
481 struct blkcg {
482 };
483 
484 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
485 static inline int blkcg_init_disk(struct gendisk *disk) { return 0; }
486 static inline void blkcg_exit_disk(struct gendisk *disk) { }
487 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
488 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
489 static inline int blkcg_activate_policy(struct gendisk *disk,
490 					const struct blkcg_policy *pol) { return 0; }
491 static inline void blkcg_deactivate_policy(struct gendisk *disk,
492 					   const struct blkcg_policy *pol) { }
493 
494 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
495 						  struct blkcg_policy *pol) { return NULL; }
496 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
497 static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
498 static inline void blkg_get(struct blkcg_gq *blkg) { }
499 static inline void blkg_put(struct blkcg_gq *blkg) { }
500 static inline void blkcg_bio_issue_init(struct bio *bio) { }
501 static inline void blk_cgroup_bio_start(struct bio *bio) { }
502 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) { return true; }
503 
504 #define blk_queue_for_each_rl(rl, q)	\
505 	for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
506 
507 #endif	/* CONFIG_BLK_CGROUP */
508 
509 #endif /* _BLK_CGROUP_PRIVATE_H */
510