Linux Audio

Check our new training course

Loading...
v6.2
  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
 23struct blkcg_gq;
 24struct blkg_policy_data;
 25
 26
 27/* percpu_counter batch for blkg_[rw]stats, per-cpu drift doesn't matter */
 28#define BLKG_STAT_CPU_BATCH	(INT_MAX / 2)
 29
 30#ifdef CONFIG_BLK_CGROUP
 31
 32enum blkg_iostat_type {
 33	BLKG_IOSTAT_READ,
 34	BLKG_IOSTAT_WRITE,
 35	BLKG_IOSTAT_DISCARD,
 36
 37	BLKG_IOSTAT_NR,
 38};
 39
 40struct blkg_iostat {
 41	u64				bytes[BLKG_IOSTAT_NR];
 42	u64				ios[BLKG_IOSTAT_NR];
 43};
 44
 45struct blkg_iostat_set {
 46	struct u64_stats_sync		sync;
 47	struct blkcg_gq		       *blkg;
 48	struct llist_node		lnode;
 49	int				lqueued;	/* queued in llist */
 50	struct blkg_iostat		cur;
 51	struct blkg_iostat		last;
 52};
 53
 54/* association between a blk cgroup and a request queue */
 55struct blkcg_gq {
 56	/* Pointer to the associated request_queue */
 57	struct request_queue		*q;
 58	struct list_head		q_node;
 59	struct hlist_node		blkcg_node;
 60	struct blkcg			*blkcg;
 61
 62	/* all non-root blkcg_gq's are guaranteed to have access to parent */
 63	struct blkcg_gq			*parent;
 64
 65	/* reference count */
 66	struct percpu_ref		refcnt;
 67
 68	/* is this blkg online? protected by both blkcg and q locks */
 69	bool				online;
 70
 71	struct blkg_iostat_set __percpu	*iostat_cpu;
 72	struct blkg_iostat_set		iostat;
 73
 74	struct blkg_policy_data		*pd[BLKCG_MAX_POLS];
 75
 76	spinlock_t			async_bio_lock;
 77	struct bio_list			async_bios;
 78	union {
 79		struct work_struct	async_bio_work;
 80		struct work_struct	free_work;
 81	};
 82
 83	atomic_t			use_delay;
 84	atomic64_t			delay_nsec;
 85	atomic64_t			delay_start;
 86	u64				last_delay;
 87	int				last_use;
 88
 89	struct rcu_head			rcu_head;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 90};
 91
 92struct blkcg {
 93	struct cgroup_subsys_state	css;
 94	spinlock_t			lock;
 95	refcount_t			online_pin;
 96
 97	struct radix_tree_root		blkg_tree;
 98	struct blkcg_gq	__rcu		*blkg_hint;
 99	struct hlist_head		blkg_list;
100
101	struct blkcg_policy_data	*cpd[BLKCG_MAX_POLS];
102
103	struct list_head		all_blkcgs_node;
104
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
105	/*
106	 * List of updated percpu blkg_iostat_set's since the last flush.
 
107	 */
108	struct llist_head __percpu	*lhead;
109
110#ifdef CONFIG_BLK_CGROUP_FC_APPID
111	char                            fc_app_id[FC_APPID_LEN];
112#endif
113#ifdef CONFIG_CGROUP_WRITEBACK
114	struct list_head		cgwb_list;
115#endif
116};
117
118static inline struct blkcg *css_to_blkcg(struct cgroup_subsys_state *css)
119{
120	return css ? container_of(css, struct blkcg, css) : NULL;
121}
122
123/*
124 * A blkcg_gq (blkg) is association between a block cgroup (blkcg) and a
125 * request_queue (q).  This is used by blkcg policies which need to track
126 * information per blkcg - q pair.
127 *
128 * There can be multiple active blkcg policies and each blkg:policy pair is
129 * represented by a blkg_policy_data which is allocated and freed by each
130 * policy's pd_alloc/free_fn() methods.  A policy can allocate private data
131 * area by allocating larger data structure which embeds blkg_policy_data
132 * at the beginning.
133 */
134struct blkg_policy_data {
135	/* the blkg and policy id this per-policy data belongs to */
136	struct blkcg_gq			*blkg;
137	int				plid;
138};
 
 
 
 
 
 
 
 
 
 
 
 
139
140/*
141 * Policies that need to keep per-blkcg data which is independent from any
142 * request_queue associated to it should implement cpd_alloc/free_fn()
143 * methods.  A policy can allocate private data area by allocating larger
144 * data structure which embeds blkcg_policy_data at the beginning.
145 * cpd_init() is invoked to let each policy handle per-blkcg data.
146 */
147struct blkcg_policy_data {
148	/* the blkcg and policy id this per-policy data belongs to */
149	struct blkcg			*blkcg;
150	int				plid;
151};
152
153typedef struct blkcg_policy_data *(blkcg_pol_alloc_cpd_fn)(gfp_t gfp);
154typedef void (blkcg_pol_init_cpd_fn)(struct blkcg_policy_data *cpd);
155typedef void (blkcg_pol_free_cpd_fn)(struct blkcg_policy_data *cpd);
156typedef void (blkcg_pol_bind_cpd_fn)(struct blkcg_policy_data *cpd);
157typedef struct blkg_policy_data *(blkcg_pol_alloc_pd_fn)(gfp_t gfp,
158				struct request_queue *q, struct blkcg *blkcg);
159typedef void (blkcg_pol_init_pd_fn)(struct blkg_policy_data *pd);
160typedef void (blkcg_pol_online_pd_fn)(struct blkg_policy_data *pd);
161typedef void (blkcg_pol_offline_pd_fn)(struct blkg_policy_data *pd);
162typedef void (blkcg_pol_free_pd_fn)(struct blkg_policy_data *pd);
163typedef void (blkcg_pol_reset_pd_stats_fn)(struct blkg_policy_data *pd);
164typedef void (blkcg_pol_stat_pd_fn)(struct blkg_policy_data *pd,
165				struct seq_file *s);
166
167struct blkcg_policy {
168	int				plid;
169	/* cgroup files for the policy */
170	struct cftype			*dfl_cftypes;
171	struct cftype			*legacy_cftypes;
172
173	/* operations */
174	blkcg_pol_alloc_cpd_fn		*cpd_alloc_fn;
175	blkcg_pol_init_cpd_fn		*cpd_init_fn;
176	blkcg_pol_free_cpd_fn		*cpd_free_fn;
177	blkcg_pol_bind_cpd_fn		*cpd_bind_fn;
178
179	blkcg_pol_alloc_pd_fn		*pd_alloc_fn;
180	blkcg_pol_init_pd_fn		*pd_init_fn;
181	blkcg_pol_online_pd_fn		*pd_online_fn;
182	blkcg_pol_offline_pd_fn		*pd_offline_fn;
183	blkcg_pol_free_pd_fn		*pd_free_fn;
184	blkcg_pol_reset_pd_stats_fn	*pd_reset_stats_fn;
185	blkcg_pol_stat_pd_fn		*pd_stat_fn;
 
 
 
186};
187
188extern struct blkcg blkcg_root;
189extern bool blkcg_debug_stats;
190
191int blkcg_init_disk(struct gendisk *disk);
192void blkcg_exit_disk(struct gendisk *disk);
193
194/* Blkio controller policy registration */
195int blkcg_policy_register(struct blkcg_policy *pol);
196void blkcg_policy_unregister(struct blkcg_policy *pol);
197int blkcg_activate_policy(struct request_queue *q,
198			  const struct blkcg_policy *pol);
199void blkcg_deactivate_policy(struct request_queue *q,
200			     const struct blkcg_policy *pol);
201
202const char *blkg_dev_name(struct blkcg_gq *blkg);
203void 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);
208u64 __blkg_prfill_u64(struct seq_file *sf, struct blkg_policy_data *pd, u64 v);
209
210struct blkg_conf_ctx {
211	struct block_device		*bdev;
212	struct blkcg_gq			*blkg;
213	char				*body;
214};
215
216struct block_device *blkcg_conf_open_bdev(char **inputp);
217int blkg_conf_prep(struct blkcg *blkcg, const struct blkcg_policy *pol,
218		   char *input, struct blkg_conf_ctx *ctx);
219void blkg_conf_finish(struct blkg_conf_ctx *ctx);
220
221/**
222 * bio_issue_as_root_blkg - see if this bio needs to be issued as root blkg
223 * @return: true if this bio needs to be submitted with the root blkg context.
224 *
225 * In order to avoid priority inversions we sometimes need to issue a bio as if
226 * it were attached to the root blkg, and then backcharge to the actual owning
227 * blkg.  The idea is we do bio_blkcg_css() to look up the actual context for
228 * the bio and attach the appropriate blkg to the bio.  Then we call this helper
229 * and if it is true run with the root blkg for that queue and then do any
230 * backcharging to the originating cgroup once the io is complete.
231 */
232static inline bool bio_issue_as_root_blkg(struct bio *bio)
233{
234	return (bio->bi_opf & (REQ_META | REQ_SWAP)) != 0;
235}
236
237/**
238 * blkg_lookup - lookup blkg for the specified blkcg - q pair
239 * @blkcg: blkcg of interest
240 * @q: request_queue of interest
241 *
242 * Lookup blkg for the @blkcg - @q pair.
243
244 * Must be called in a RCU critical section.
245 */
246static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg,
247					   struct request_queue *q)
248{
249	struct blkcg_gq *blkg;
250
251	WARN_ON_ONCE(!rcu_read_lock_held());
252
253	if (blkcg == &blkcg_root)
254		return q->root_blkg;
255
256	blkg = rcu_dereference(blkcg->blkg_hint);
257	if (blkg && blkg->q == q)
258		return blkg;
259
260	blkg = radix_tree_lookup(&blkcg->blkg_tree, q->id);
261	if (blkg && blkg->q != q)
262		blkg = NULL;
263	return blkg;
264}
265
266/**
267 * blkg_to_pdata - get policy private data
268 * @blkg: blkg of interest
269 * @pol: policy of interest
270 *
271 * Return pointer to private data associated with the @blkg-@pol pair.
272 */
273static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
274						  struct blkcg_policy *pol)
275{
276	return blkg ? blkg->pd[pol->plid] : NULL;
277}
278
279static inline struct blkcg_policy_data *blkcg_to_cpd(struct blkcg *blkcg,
280						     struct blkcg_policy *pol)
281{
282	return blkcg ? blkcg->cpd[pol->plid] : NULL;
283}
284
285/**
286 * pdata_to_blkg - get blkg associated with policy private data
287 * @pd: policy private data of interest
288 *
289 * @pd is policy private data.  Determine the blkg it's associated with.
290 */
291static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd)
292{
293	return pd ? pd->blkg : NULL;
294}
295
296static inline struct blkcg *cpd_to_blkcg(struct blkcg_policy_data *cpd)
297{
298	return cpd ? cpd->blkcg : NULL;
299}
300
301/**
302 * blkg_path - format cgroup path of blkg
303 * @blkg: blkg of interest
304 * @buf: target buffer
305 * @buflen: target buffer length
306 *
307 * Format the path of the cgroup of @blkg into @buf.
308 */
309static inline int blkg_path(struct blkcg_gq *blkg, char *buf, int buflen)
310{
311	return cgroup_path(blkg->blkcg->css.cgroup, buf, buflen);
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 */
320static 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 */
332static 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 */
341static 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
378bool __blkcg_punt_bio_submit(struct bio *bio);
379
380static inline bool blkcg_punt_bio_submit(struct bio *bio)
381{
382	if (bio->bi_opf & REQ_CGROUP_PUNT)
383		return __blkcg_punt_bio_submit(bio);
384	else
385		return false;
386}
387
388static inline void blkcg_bio_issue_init(struct bio *bio)
389{
390	bio_issue_init(&bio->bi_issue, bio_sectors(bio));
391}
392
393static inline void blkcg_use_delay(struct blkcg_gq *blkg)
394{
395	if (WARN_ON_ONCE(atomic_read(&blkg->use_delay) < 0))
396		return;
397	if (atomic_add_return(1, &blkg->use_delay) == 1)
398		atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
399}
400
401static inline int blkcg_unuse_delay(struct blkcg_gq *blkg)
402{
403	int old = atomic_read(&blkg->use_delay);
404
405	if (WARN_ON_ONCE(old < 0))
406		return 0;
407	if (old == 0)
408		return 0;
409
410	/*
411	 * We do this song and dance because we can race with somebody else
412	 * adding or removing delay.  If we just did an atomic_dec we'd end up
413	 * negative and we'd already be in trouble.  We need to subtract 1 and
414	 * then check to see if we were the last delay so we can drop the
415	 * congestion count on the cgroup.
416	 */
417	while (old && !atomic_try_cmpxchg(&blkg->use_delay, &old, old - 1))
418		;
419
420	if (old == 0)
421		return 0;
422	if (old == 1)
423		atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
424	return 1;
425}
426
427/**
428 * blkcg_set_delay - Enable allocator delay mechanism with the specified delay amount
429 * @blkg: target blkg
430 * @delay: delay duration in nsecs
431 *
432 * When enabled with this function, the delay is not decayed and must be
433 * explicitly cleared with blkcg_clear_delay(). Must not be mixed with
434 * blkcg_[un]use_delay() and blkcg_add_delay() usages.
435 */
436static inline void blkcg_set_delay(struct blkcg_gq *blkg, u64 delay)
437{
438	int old = atomic_read(&blkg->use_delay);
439
440	/* We only want 1 person setting the congestion count for this blkg. */
441	if (!old && atomic_try_cmpxchg(&blkg->use_delay, &old, -1))
442		atomic_inc(&blkg->blkcg->css.cgroup->congestion_count);
443
444	atomic64_set(&blkg->delay_nsec, delay);
445}
446
447/**
448 * blkcg_clear_delay - Disable allocator delay mechanism
449 * @blkg: target blkg
450 *
451 * Disable use_delay mechanism. See blkcg_set_delay().
452 */
453static inline void blkcg_clear_delay(struct blkcg_gq *blkg)
454{
455	int old = atomic_read(&blkg->use_delay);
456
457	/* We only want 1 person clearing the congestion count for this blkg. */
458	if (old && atomic_try_cmpxchg(&blkg->use_delay, &old, 0))
459		atomic_dec(&blkg->blkcg->css.cgroup->congestion_count);
460}
461
462/**
463 * blk_cgroup_mergeable - Determine whether to allow or disallow merges
464 * @rq: request to merge into
465 * @bio: bio to merge
466 *
467 * @bio and @rq should belong to the same cgroup and their issue_as_root should
468 * match. The latter is necessary as we don't want to throttle e.g. a metadata
469 * update because it happens to be next to a regular IO.
470 */
471static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio)
472{
473	return rq->bio->bi_blkg == bio->bi_blkg &&
474		bio_issue_as_root_blkg(rq->bio) == bio_issue_as_root_blkg(bio);
475}
476
477void blk_cgroup_bio_start(struct bio *bio);
478void blkcg_add_delay(struct blkcg_gq *blkg, u64 now, u64 delta);
479#else	/* CONFIG_BLK_CGROUP */
480
481struct blkg_policy_data {
482};
483
484struct blkcg_policy_data {
485};
486
487struct blkcg_policy {
488};
489
490struct blkcg {
491};
492
493static inline struct blkcg_gq *blkg_lookup(struct blkcg *blkcg, void *key) { return NULL; }
494static inline int blkcg_init_disk(struct gendisk *disk) { return 0; }
495static inline void blkcg_exit_disk(struct gendisk *disk) { }
496static inline int blkcg_policy_register(struct blkcg_policy *pol) { return 0; }
497static inline void blkcg_policy_unregister(struct blkcg_policy *pol) { }
498static inline int blkcg_activate_policy(struct request_queue *q,
499					const struct blkcg_policy *pol) { return 0; }
500static inline void blkcg_deactivate_policy(struct request_queue *q,
501					   const struct blkcg_policy *pol) { }
502
503static inline struct blkg_policy_data *blkg_to_pd(struct blkcg_gq *blkg,
504						  struct blkcg_policy *pol) { return NULL; }
505static inline struct blkcg_gq *pd_to_blkg(struct blkg_policy_data *pd) { return NULL; }
506static inline char *blkg_path(struct blkcg_gq *blkg) { return NULL; }
507static inline void blkg_get(struct blkcg_gq *blkg) { }
508static inline void blkg_put(struct blkcg_gq *blkg) { }
509
510static inline bool blkcg_punt_bio_submit(struct bio *bio) { return false; }
511static inline void blkcg_bio_issue_init(struct bio *bio) { }
512static inline void blk_cgroup_bio_start(struct bio *bio) { }
513static inline bool blk_cgroup_mergeable(struct request *rq, struct bio *bio) { return true; }
514
515#define blk_queue_for_each_rl(rl, q)	\
516	for ((rl) = &(q)->root_rl; (rl); (rl) = NULL)
517
518#endif	/* CONFIG_BLK_CGROUP */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
519
520#endif /* _BLK_CGROUP_PRIVATE_H */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
v3.1
  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
 19enum blkio_policy_id {
 20	BLKIO_POLICY_PROP = 0,		/* Proportional Bandwidth division */
 21	BLKIO_POLICY_THROTL,		/* Throttling */
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 22};
 23
 24/* Max limits for throttle policy */
 25#define THROTL_IOPS_MAX		UINT_MAX
 
 
 
 
 
 26
 27#if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
 
 28
 29#ifndef CONFIG_BLK_CGROUP
 30/* When blk-cgroup is a module, its subsys_id isn't a compile-time constant */
 31extern struct cgroup_subsys blkio_subsys;
 32#define blkio_subsys_id blkio_subsys.subsys_id
 33#endif
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 34
 35enum stat_type {
 36	/* Total time spent (in ns) between request dispatch to the driver and
 37	 * request completion for IOs doen by this cgroup. This may not be
 38	 * accurate when NCQ is turned on. */
 39	BLKIO_STAT_SERVICE_TIME = 0,
 40	/* Total time spent waiting in scheduler queue in ns */
 41	BLKIO_STAT_WAIT_TIME,
 42	/* Number of IOs queued up */
 43	BLKIO_STAT_QUEUED,
 44	/* All the single valued stats go below this */
 45	BLKIO_STAT_TIME,
 46#ifdef CONFIG_DEBUG_BLK_CGROUP
 47	/* Time not charged to this cgroup */
 48	BLKIO_STAT_UNACCOUNTED_TIME,
 49	BLKIO_STAT_AVG_QUEUE_SIZE,
 50	BLKIO_STAT_IDLE_TIME,
 51	BLKIO_STAT_EMPTY_TIME,
 52	BLKIO_STAT_GROUP_WAIT_TIME,
 53	BLKIO_STAT_DEQUEUE
 54#endif
 55};
 56
 57/* Per cpu stats */
 58enum stat_type_cpu {
 59	BLKIO_STAT_CPU_SECTORS,
 60	/* Total bytes transferred */
 61	BLKIO_STAT_CPU_SERVICE_BYTES,
 62	/* Total IOs serviced, post merge */
 63	BLKIO_STAT_CPU_SERVICED,
 64	/* Number of IOs merged */
 65	BLKIO_STAT_CPU_MERGED,
 66	BLKIO_STAT_CPU_NR
 67};
 68
 69enum stat_sub_type {
 70	BLKIO_STAT_READ = 0,
 71	BLKIO_STAT_WRITE,
 72	BLKIO_STAT_SYNC,
 73	BLKIO_STAT_ASYNC,
 74	BLKIO_STAT_TOTAL
 75};
 76
 77/* blkg state flags */
 78enum blkg_state_flags {
 79	BLKG_waiting = 0,
 80	BLKG_idling,
 81	BLKG_empty,
 82};
 83
 84/* cgroup files owned by proportional weight policy */
 85enum blkcg_file_name_prop {
 86	BLKIO_PROP_weight = 1,
 87	BLKIO_PROP_weight_device,
 88	BLKIO_PROP_io_service_bytes,
 89	BLKIO_PROP_io_serviced,
 90	BLKIO_PROP_time,
 91	BLKIO_PROP_sectors,
 92	BLKIO_PROP_unaccounted_time,
 93	BLKIO_PROP_io_service_time,
 94	BLKIO_PROP_io_wait_time,
 95	BLKIO_PROP_io_merged,
 96	BLKIO_PROP_io_queued,
 97	BLKIO_PROP_avg_queue_size,
 98	BLKIO_PROP_group_wait_time,
 99	BLKIO_PROP_idle_time,
100	BLKIO_PROP_empty_time,
101	BLKIO_PROP_dequeue,
102};
103
104/* cgroup files owned by throttle policy */
105enum blkcg_file_name_throtl {
106	BLKIO_THROTL_read_bps_device,
107	BLKIO_THROTL_write_bps_device,
108	BLKIO_THROTL_read_iops_device,
109	BLKIO_THROTL_write_iops_device,
110	BLKIO_THROTL_io_service_bytes,
111	BLKIO_THROTL_io_serviced,
112};
113
114struct blkio_cgroup {
115	struct cgroup_subsys_state css;
116	unsigned int weight;
117	spinlock_t lock;
118	struct hlist_head blkg_list;
119	struct list_head policy_list; /* list of blkio_policy_node */
120};
121
122struct blkio_group_stats {
123	/* total disk time and nr sectors dispatched by this group */
124	uint64_t time;
125	uint64_t stat_arr[BLKIO_STAT_QUEUED + 1][BLKIO_STAT_TOTAL];
126#ifdef CONFIG_DEBUG_BLK_CGROUP
127	/* Time not charged to this cgroup */
128	uint64_t unaccounted_time;
129
130	/* Sum of number of IOs queued across all samples */
131	uint64_t avg_queue_size_sum;
132	/* Count of samples taken for average */
133	uint64_t avg_queue_size_samples;
134	/* How many times this group has been removed from service tree */
135	unsigned long dequeue;
136
137	/* Total time spent waiting for it to be assigned a timeslice. */
138	uint64_t group_wait_time;
139	uint64_t start_group_wait_time;
140
141	/* Time spent idling for this blkio_group */
142	uint64_t idle_time;
143	uint64_t start_idle_time;
144	/*
145	 * Total time when we have requests queued and do not contain the
146	 * current active queue.
147	 */
148	uint64_t empty_time;
149	uint64_t start_empty_time;
150	uint16_t flags;
 
 
 
 
151#endif
152};
153
154/* Per cpu blkio group stats */
155struct blkio_group_stats_cpu {
156	uint64_t sectors;
157	uint64_t stat_arr_cpu[BLKIO_STAT_CPU_NR][BLKIO_STAT_TOTAL];
158	struct u64_stats_sync syncp;
159};
160
161struct blkio_group {
162	/* An rcu protected unique identifier for the group */
163	void *key;
164	struct hlist_node blkcg_node;
165	unsigned short blkcg_id;
166	/* Store cgroup path */
167	char path[128];
168	/* The device MKDEV(major, minor), this group has been created for */
169	dev_t dev;
170	/* policy which owns this blk group */
171	enum blkio_policy_id plid;
172
173	/* Need to serialize the stats in the case of reset/update */
174	spinlock_t stats_lock;
175	struct blkio_group_stats stats;
176	/* Per cpu stats pointer */
177	struct blkio_group_stats_cpu __percpu *stats_cpu;
178};
179
180struct blkio_policy_node {
181	struct list_head node;
182	dev_t dev;
183	/* This node belongs to max bw policy or porportional weight policy */
184	enum blkio_policy_id plid;
185	/* cgroup file to which this rule belongs to */
186	int fileid;
187
188	union {
189		unsigned int weight;
190		/*
191		 * Rate read/write in terms of byptes per second
192		 * Whether this rate represents read or write is determined
193		 * by file type "fileid".
194		 */
195		u64 bps;
196		unsigned int iops;
197	} val;
198};
199
200extern unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
201				     dev_t dev);
202extern uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg,
203				     dev_t dev);
204extern uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg,
205				     dev_t dev);
206extern unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg,
207				     dev_t dev);
208extern unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg,
209				     dev_t dev);
210
211typedef void (blkio_unlink_group_fn) (void *key, struct blkio_group *blkg);
212
213typedef void (blkio_update_group_weight_fn) (void *key,
214			struct blkio_group *blkg, unsigned int weight);
215typedef void (blkio_update_group_read_bps_fn) (void * key,
216			struct blkio_group *blkg, u64 read_bps);
217typedef void (blkio_update_group_write_bps_fn) (void *key,
218			struct blkio_group *blkg, u64 write_bps);
219typedef void (blkio_update_group_read_iops_fn) (void *key,
220			struct blkio_group *blkg, unsigned int read_iops);
221typedef void (blkio_update_group_write_iops_fn) (void *key,
222			struct blkio_group *blkg, unsigned int write_iops);
223
224struct blkio_policy_ops {
225	blkio_unlink_group_fn *blkio_unlink_group_fn;
226	blkio_update_group_weight_fn *blkio_update_group_weight_fn;
227	blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn;
228	blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn;
229	blkio_update_group_read_iops_fn *blkio_update_group_read_iops_fn;
230	blkio_update_group_write_iops_fn *blkio_update_group_write_iops_fn;
231};
232
233struct blkio_policy_type {
234	struct list_head list;
235	struct blkio_policy_ops ops;
236	enum blkio_policy_id plid;
237};
238
 
 
 
 
 
 
239/* Blkio controller policy registration */
240extern void blkio_policy_register(struct blkio_policy_type *);
241extern void blkio_policy_unregister(struct blkio_policy_type *);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
242
243static inline char *blkg_path(struct blkio_group *blkg)
 
 
 
 
 
 
 
 
244{
245	return blkg->path;
246}
247
248#else
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
249
250struct blkio_group {
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
251};
252
253struct blkio_policy_type {
254};
255
256static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { }
257static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
 
 
 
 
 
 
 
 
 
 
 
 
 
258
259static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
 
 
 
 
 
260
261#endif
 
 
 
 
 
 
262
263#define BLKIO_WEIGHT_MIN	10
264#define BLKIO_WEIGHT_MAX	1000
265#define BLKIO_WEIGHT_DEFAULT	500
266
267#ifdef CONFIG_DEBUG_BLK_CGROUP
268void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg);
269void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
270				unsigned long dequeue);
271void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg);
272void blkiocg_update_idle_time_stats(struct blkio_group *blkg);
273void blkiocg_set_start_empty_time(struct blkio_group *blkg);
274
275#define BLKG_FLAG_FNS(name)						\
276static inline void blkio_mark_blkg_##name(				\
277		struct blkio_group_stats *stats)			\
278{									\
279	stats->flags |= (1 << BLKG_##name);				\
280}									\
281static inline void blkio_clear_blkg_##name(				\
282		struct blkio_group_stats *stats)			\
283{									\
284	stats->flags &= ~(1 << BLKG_##name);				\
285}									\
286static inline int blkio_blkg_##name(struct blkio_group_stats *stats)	\
287{									\
288	return (stats->flags & (1 << BLKG_##name)) != 0;		\
289}									\
290
291BLKG_FLAG_FNS(waiting)
292BLKG_FLAG_FNS(idling)
293BLKG_FLAG_FNS(empty)
294#undef BLKG_FLAG_FNS
295#else
296static inline void blkiocg_update_avg_queue_size_stats(
297						struct blkio_group *blkg) {}
298static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
299						unsigned long dequeue) {}
300static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
301{}
302static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg) {}
303static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg) {}
304#endif
305
306#if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
307extern struct blkio_cgroup blkio_root_cgroup;
308extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
309extern struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk);
310extern void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
311	struct blkio_group *blkg, void *key, dev_t dev,
312	enum blkio_policy_id plid);
313extern int blkio_alloc_blkg_stats(struct blkio_group *blkg);
314extern int blkiocg_del_blkio_group(struct blkio_group *blkg);
315extern struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg,
316						void *key);
317void blkiocg_update_timeslice_used(struct blkio_group *blkg,
318					unsigned long time,
319					unsigned long unaccounted_time);
320void blkiocg_update_dispatch_stats(struct blkio_group *blkg, uint64_t bytes,
321						bool direction, bool sync);
322void blkiocg_update_completion_stats(struct blkio_group *blkg,
323	uint64_t start_time, uint64_t io_start_time, bool direction, bool sync);
324void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
325					bool sync);
326void blkiocg_update_io_add_stats(struct blkio_group *blkg,
327		struct blkio_group *curr_blkg, bool direction, bool sync);
328void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
329					bool direction, bool sync);
330#else
331struct cgroup;
332static inline struct blkio_cgroup *
333cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
334static inline struct blkio_cgroup *
335task_blkio_cgroup(struct task_struct *tsk) { return NULL; }
336
337static inline void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
338		struct blkio_group *blkg, void *key, dev_t dev,
339		enum blkio_policy_id plid) {}
340
341static inline int blkio_alloc_blkg_stats(struct blkio_group *blkg) { return 0; }
342
343static inline int
344blkiocg_del_blkio_group(struct blkio_group *blkg) { return 0; }
345
346static inline struct blkio_group *
347blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key) { return NULL; }
348static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg,
349						unsigned long time,
350						unsigned long unaccounted_time)
351{}
352static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
353				uint64_t bytes, bool direction, bool sync) {}
354static inline void blkiocg_update_completion_stats(struct blkio_group *blkg,
355		uint64_t start_time, uint64_t io_start_time, bool direction,
356		bool sync) {}
357static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
358						bool direction, bool sync) {}
359static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg,
360		struct blkio_group *curr_blkg, bool direction, bool sync) {}
361static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
362						bool direction, bool sync) {}
363#endif
364#endif /* _BLK_CGROUP_H */