xref: /linux/block/blk-cgroup.h (revision 805185b7c7a1069e407b6f7b3bc98e44d415f484)
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 	/* If there is block congestion on this cgroup. */
99 	atomic_t			congestion_count;
100 
101 	struct radix_tree_root		blkg_tree;
102 	struct blkcg_gq	__rcu		*blkg_hint;
103 	struct hlist_head		blkg_list;
104 
105 	struct blkcg_policy_data	*cpd[BLKCG_MAX_POLS];
106 
107 	struct list_head		all_blkcgs_node;
108 
109 	/*
110 	 * List of updated percpu blkg_iostat_set's since the last flush.
111 	 */
112 	struct llist_head __percpu	*lhead;
113 
114 #ifdef CONFIG_BLK_CGROUP_FC_APPID
115 	char                            fc_app_id[FC_APPID_LEN];
116 #endif
117 #ifdef CONFIG_CGROUP_WRITEBACK
118 	struct list_head		cgwb_list;
119 #endif
120 };
121 
122 static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
123 {
124 	return css ? container_of(css, struct blkcg, css) : NULL;
125 }
126 
127 /*
128  * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
129  * request_queue (q).  This is used by blkcg policies which need to track
130  * information per blkcg - q pair.
131  *
132  * There can be multiple active blkcg policies and each blkg:policy pair is
133  * represented by a blkg_policy_data which is allocated and freed by each
134  * policy's pd_alloc/free_fn() methods.  A policy can allocate private data
135  * area by allocating larger data structure which embeds blkg_policy_data
136  * at the beginning.
137  */
138 struct blkg_policy_data {
139 	/* the blkg and policy id this per-policy data belongs to */
140 	struct blkcg_gq			*blkg;
141 	int				plid;
142 	bool				online;
143 
144 	struct rcu_head			rcu_head;
145 };
146 
147 /*
148  * Policies that need to keep per-blkcg data which is independent from any
149  * request_queue associated to it should implement cpd_alloc/free_fn()
150  * methods.  A policy can allocate private data area by allocating larger
151  * data structure which embeds blkcg_policy_data at the beginning.
152  * cpd_init() is invoked to let each policy handle per-blkcg data.
153  */
154 struct blkcg_policy_data {
155 	/* the blkcg and policy id this per-policy data belongs to */
156 	struct blkcg			*blkcg;
157 	int				plid;
158 };
159 
160 typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
161 typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
162 typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
163 typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
164 typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(struct gendisk *disk,
165 		struct blkcg *blkcg, gfp_t gfp);
166 typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
167 typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
168 typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
169 typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
170 typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
171 typedef void (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd,
172 				struct seq_file *s);
173 
174 struct blkcg_policy {
175 	int				plid;
176 	/* cgroup files for the policy */
177 	struct cftype			*dfl_cftypes;
178 	struct cftype			*legacy_cftypes;
179 
180 	/* operations */
181 	blkcg_pol_alloc_cpd_fn		*cpd_alloc_fn;
182 	blkcg_pol_free_cpd_fn		*cpd_free_fn;
183 
184 	blkcg_pol_alloc_pd_fn		*pd_alloc_fn;
185 	blkcg_pol_init_pd_fn		*pd_init_fn;
186 	blkcg_pol_online_pd_fn		*pd_online_fn;
187 	blkcg_pol_offline_pd_fn		*pd_offline_fn;
188 	blkcg_pol_free_pd_fn		*pd_free_fn;
189 	blkcg_pol_reset_pd_stats_fn	*pd_reset_stats_fn;
190 	blkcg_pol_stat_pd_fn		*pd_stat_fn;
191 };
192 
193 extern struct blkcg blkcg_root;
194 extern bool blkcg_debug_stats;
195 
196 void blkg_init_queue(struct request_queue *q);
197 int blkcg_init_disk(struct gendisk *disk);
198 void blkcg_exit_disk(struct gendisk *disk);
199 
200 /* Blkio controller policy registration */
201 int blkcg_policy_register(struct blkcg_policy *pol);
202 void blkcg_policy_unregister(struct blkcg_policy *pol);
203 int blkcg_activate_policy(struct gendisk *disk, const struct blkcg_policy *pol);
204 void blkcg_deactivate_policy(struct gendisk *disk,
205 			     const struct blkcg_policy *pol);
206 
207 const char *blkg_dev_name(struct blkcg_gq *blkg);
208 void blkcg_print_blkgs(struct seq_file *sf, struct blkcg *blkcg,
209 		       u64 (*prfill)(struct seq_file *,
210 				     struct blkg_policy_data *, int),
211 		       const struct blkcg_policy *pol, int data,
212 		       bool show_total);
213 u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
214 
215 struct blkg_conf_ctx {
216 	char				*input;
217 	char				*body;
218 	struct block_device		*bdev;
219 	struct blkcg_gq			*blkg;
220 };
221 
222 void blkg_conf_init(struct blkg_conf_ctx *ctx, char *input);
223 int blkg_conf_open_bdev(struct blkg_conf_ctx *ctx)
224 	__cond_acquires(0, &ctx->bdev->bd_queue->rq_qos_mutex);
225 int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
226 		   struct blkg_conf_ctx *ctx)
227 	__cond_acquires(0, &ctx->bdev->bd_disk->queue->queue_lock);
228 void blkg_conf_unprep(struct blkg_conf_ctx *ctx)
229 	__releases(ctx->bdev->bd_disk->queue->queue_lock);
230 void blkg_conf_close_bdev(struct blkg_conf_ctx *ctx)
231 	__releases(&ctx->bdev->bd_queue->rq_qos_mutex);
232 
233 /**
234  * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
235  * @bio: the target &bio
236  *
237  * Return: true if this bio needs to be submitted with the root blkg context.
238  *
239  * In order to avoid priority inversions we sometimes need to issue a bio as if
240  * it were attached to the root blkg, and then backcharge to the actual owning
241  * blkg.  The idea is we do bio_blkcg_css() to look up the actual context for
242  * the bio and attach the appropriate blkg to the bio.  Then we call this helper
243  * and if it is true run with the root blkg for that queue and then do any
244  * backcharging to the originating cgroup once the io is complete.
245  */
246 static inline bool bio_issue_as_root_blkg(struct bio *bio)
247 {
248 	return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
249 }
250 
251 /**
252  * blkg_lookup - lookup blkg for the specified blkcg - q pair
253  * @blkcg: blkcg of interest
254  * @q: request_queue of interest
255  *
256  * Lookup blkg for the @blkcg - @q pair.
257  *
258  * Must be called in a RCU critical section.
259  */
260 static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
261 					   struct request_queue *q)
262 {
263 	struct blkcg_gq *blkg;
264 
265 	if (blkcg == &blkcg_root)
266 		return q->root_blkg;
267 
268 	blkg = rcu_dereference_check(blkcg->blkg_hint,
269 			lockdep_is_held(&q->queue_lock));
270 	if (blkg && blkg->q == q)
271 		return blkg;
272 
273 	blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
274 	if (blkg && blkg->q != q)
275 		blkg = NULL;
276 	return blkg;
277 }
278 
279 /**
280  * blkg_to_pd - get policy private data
281  * @blkg: blkg of interest
282  * @pol: policy of interest
283  *
284  * Return pointer to private data associated with the @blkg-@pol pair.
285  */
286 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
287 						  const struct blkcg_policy *pol)
288 {
289 	return blkg ? READ_ONCE(blkg->pd[pol->plid]) : NULL;
290 }
291 
292 static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
293 						     struct blkcg_policy *pol)
294 {
295 	return blkcg ? blkcg->cpd[pol->plid] : NULL;
296 }
297 
298 /**
299  * pd_to_blkg - get blkg associated with policy private data
300  * @pd: policy private data of interest
301  *
302  * @pd is policy private data.  Determine the blkg it's associated with.
303  */
304 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
305 {
306 	return pd ? pd->blkg : NULL;
307 }
308 
309 static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
310 {
311 	return cpd ? cpd->blkcg : NULL;
312 }
313 
314 /**
315  * blkg_get - get a blkg reference
316  * @blkg: blkg to get
317  *
318  * The caller should be holding an existing reference.
319  */
320 static inline void blkg_get(struct blkcg_gq *blkg)
321 {
322 	percpu_ref_get(&blkg->refcnt);
323 }
324 
325 /**
326  * blkg_tryget - try and get a blkg reference
327  * @blkg: blkg to get
328  *
329  * This is for use when doing an RCU lookup of the blkg.  We may be in the midst
330  * of freeing this blkg, so we can only use it if the refcnt is not zero.
331  */
332 static inline bool blkg_tryget(struct blkcg_gq *blkg)
333 {
334 	return blkg && percpu_ref_tryget(&blkg->refcnt);
335 }
336 
337 /**
338  * blkg_put - put a blkg reference
339  * @blkg: blkg to put
340  */
341 static inline void blkg_put(struct blkcg_gq *blkg)
342 {
343 	percpu_ref_put(&blkg->refcnt);
344 }
345 
346 /**
347  * blkg_for_each_descendant_pre - pre-order walk of a blkg's descendants
348  * @d_blkg: loop cursor pointing to the current descendant
349  * @pos_css: used for iteration
350  * @p_blkg: target blkg to walk descendants of
351  *
352  * Walk @c_blkg through the descendants of @p_blkg.  Must be used with RCU
353  * read locked.  If called under either blkcg or queue lock, the iteration
354  * is guaranteed to include all and only online blkgs.  The caller may
355  * update @pos_css by calling css_rightmost_descendant() to skip subtree.
356  * @p_blkg is included in the iteration and the first node to be visited.
357  */
358 #define blkg_for_each_descendant_pre(d_blkg, pos_css, p_blkg)		\
359 	css_for_each_descendant_pre((pos_css), &(p_blkg)->blkcg->css)	\
360 		if (((d_blkg) = blkg_lookup(css_to_blkcg(pos_css),	\
361 					    (p_blkg)->q)))
362 
363 /**
364  * blkg_for_each_descendant_post - post-order walk of a blkg's descendants
365  * @d_blkg: loop cursor pointing to the current descendant
366  * @pos_css: used for iteration
367  * @p_blkg: target blkg to walk descendants of
368  *
369  * Similar to blkg_for_each_descendant_pre() but performs post-order
370  * traversal instead.  Synchronization rules are the same.  @p_blkg is
371  * included in the iteration and the last node to be visited.
372  */
373 #define blkg_for_each_descendant_post(d_blkg, pos_css, p_blkg)		\
374 	css_for_each_descendant_post((pos_css), &(p_blkg)->blkcg->css)	\
375 		if (((d_blkg) = blkg_lookup(css_to_blkcg(pos_css),	\
376 					    (p_blkg)->q)))
377 
378 static inline void blkcg_use_delay(struct blkcg_gq *blkg)
379 {
380 	if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0))
381 		return;
382 	if (atomic_add_return(1, &blkg->use_delay) == 1)
383 		atomic_inc(&blkg->blkcg->congestion_count);
384 }
385 
386 static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
387 {
388 	int old = atomic_read(&blkg->use_delay);
389 
390 	if (WARN_ON_ONCE(old < 0))
391 		return 0;
392 	if (old == 0)
393 		return 0;
394 
395 	/*
396 	 * We do this song and dance because we can race with somebody else
397 	 * adding or removing delay.  If we just did an atomic_dec we'd end up
398 	 * negative and we'd already be in trouble.  We need to subtract 1 and
399 	 * then check to see if we were the last delay so we can drop the
400 	 * congestion count on the cgroup.
401 	 */
402 	while (old && !atomic_try_cmpxchg(&blkg->use_delay, &old, old - 1))
403 		;
404 
405 	if (old == 0)
406 		return 0;
407 	if (old == 1)
408 		atomic_dec(&blkg->blkcg->congestion_count);
409 	return 1;
410 }
411 
412 /**
413  * blkcg_set_delay - Enable allocator delay mechanism with the specified delay amount
414  * @blkg: target blkg
415  * @delay: delay duration in nsecs
416  *
417  * When enabled with this function, the delay is not decayed and must be
418  * explicitly cleared with blkcg_clear_delay(). Must not be mixed with
419  * blkcg_[un]use_delay() and blkcg_add_delay() usages.
420  */
421 static inline void blkcg_set_delay(struct blkcg_gq *blkg, u64 delay)
422 {
423 	int old = atomic_read(&blkg->use_delay);
424 
425 	/* We only want 1 person setting the congestion count for this blkg. */
426 	if (!old && atomic_try_cmpxchg(&blkg->use_delay, &old, -1))
427 		atomic_inc(&blkg->blkcg->congestion_count);
428 
429 	atomic64_set(&blkg->delay_nsec, delay);
430 }
431 
432 /**
433  * blkcg_clear_delay - Disable allocator delay mechanism
434  * @blkg: target blkg
435  *
436  * Disable use_delay mechanism. See blkcg_set_delay().
437  */
438 static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
439 {
440 	int old = atomic_read(&blkg->use_delay);
441 
442 	/* We only want 1 person clearing the congestion count for this blkg. */
443 	if (old && atomic_try_cmpxchg(&blkg->use_delay, &old, 0))
444 		atomic_dec(&blkg->blkcg->congestion_count);
445 }
446 
447 /**
448  * blk_cgroup_mergeable - Determine whether to allow or disallow merges
449  * @rq: request to merge into
450  * @bio: bio to merge
451  *
452  * @bio and @rq should belong to the same cgroup and their issue_as_root should
453  * match. The latter is necessary as we don't want to throttle e.g. a metadata
454  * update because it happens to be next to a regular IO.
455  */
456 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio)
457 {
458 	return rq->bio->bi_blkg == bio->bi_blkg &&
459 		bio_issue_as_root_blkg(rq->bio) == bio_issue_as_root_blkg(bio);
460 }
461 
462 static inline bool blkcg_policy_enabled(struct request_queue *q,
463 				const struct blkcg_policy *pol)
464 {
465 	return pol && test_bit(pol->plid, q->blkcg_pols);
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 void blkg_init_queue(struct request_queue *q) { }
486 static inline int blkcg_init_disk(struct gendisk *disk) { return 0; }
487 static inline void blkcg_exit_disk(struct gendisk *disk) { }
488 static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
489 static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
490 static inline int blkcg_activate_policy(struct gendisk *disk,
491 					const struct blkcg_policy *pol) { return 0; }
492 static inline void blkcg_deactivate_policy(struct gendisk *disk,
493 					   const struct blkcg_policy *pol) { }
494 
495 static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
496 						  const struct blkcg_policy *pol) { return NULL; }
497 static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { 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 blk_cgroup_bio_start(struct bio *bio) { }
501 static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) { return true; }
502 
503 #define blk_queue_for_each_rl(rl, q)	\
504 	for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
505 
506 #endif	/* CONFIG_BLK_CGROUP */
507 
508 #endif /* _BLK_CGROUP_PRIVATE_H */
509