xref: /linux/block/blk-cgroup.h (revision f48ec1d7885281a9c6cd7779d61f321d1b1fd741)
131e4c28dSVivek Goyal #ifndef _BLK_CGROUP_H
231e4c28dSVivek Goyal #define _BLK_CGROUP_H
331e4c28dSVivek Goyal /*
431e4c28dSVivek Goyal  * Common Block IO controller cgroup interface
531e4c28dSVivek Goyal  *
631e4c28dSVivek Goyal  * Based on ideas and code from CFQ, CFS and BFQ:
731e4c28dSVivek Goyal  * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
831e4c28dSVivek Goyal  *
931e4c28dSVivek Goyal  * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
1031e4c28dSVivek Goyal  *		      Paolo Valente <paolo.valente@unimore.it>
1131e4c28dSVivek Goyal  *
1231e4c28dSVivek Goyal  * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
1331e4c28dSVivek Goyal  * 	              Nauman Rafique <nauman@google.com>
1431e4c28dSVivek Goyal  */
1531e4c28dSVivek Goyal 
1631e4c28dSVivek Goyal #include <linux/cgroup.h>
17575969a0SVivek Goyal #include <linux/u64_stats_sync.h>
18829fdb50STejun Heo #include <linux/seq_file.h>
1931e4c28dSVivek Goyal 
20062a644dSVivek Goyal enum blkio_policy_id {
21062a644dSVivek Goyal 	BLKIO_POLICY_PROP = 0,		/* Proportional Bandwidth division */
224c9eefa1SVivek Goyal 	BLKIO_POLICY_THROTL,		/* Throttling */
23035d10b2STejun Heo 
24035d10b2STejun Heo 	BLKIO_NR_POLICIES,
25062a644dSVivek Goyal };
26062a644dSVivek Goyal 
279355aedeSVivek Goyal /* Max limits for throttle policy */
289355aedeSVivek Goyal #define THROTL_IOPS_MAX		UINT_MAX
299355aedeSVivek Goyal 
303381cb8dSTejun Heo /* CFQ specific, out here for blkcg->cfq_weight */
313381cb8dSTejun Heo #define CFQ_WEIGHT_MIN		10
323381cb8dSTejun Heo #define CFQ_WEIGHT_MAX		1000
333381cb8dSTejun Heo #define CFQ_WEIGHT_DEFAULT	500
343381cb8dSTejun Heo 
35*f48ec1d7STejun Heo #ifdef CONFIG_BLK_CGROUP
36*f48ec1d7STejun Heo 
37edcb0722STejun Heo enum blkg_rwstat_type {
38edcb0722STejun Heo 	BLKG_RWSTAT_READ,
39edcb0722STejun Heo 	BLKG_RWSTAT_WRITE,
40edcb0722STejun Heo 	BLKG_RWSTAT_SYNC,
41edcb0722STejun Heo 	BLKG_RWSTAT_ASYNC,
42edcb0722STejun Heo 
43edcb0722STejun Heo 	BLKG_RWSTAT_NR,
44edcb0722STejun Heo 	BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
45303a3acbSDivyesh Shah };
46303a3acbSDivyesh Shah 
4731e4c28dSVivek Goyal struct blkio_cgroup {
4831e4c28dSVivek Goyal 	struct cgroup_subsys_state css;
4931e4c28dSVivek Goyal 	spinlock_t lock;
5031e4c28dSVivek Goyal 	struct hlist_head blkg_list;
519a9e8a26STejun Heo 
529a9e8a26STejun Heo 	/* for policies to test whether associated blkcg has changed */
539a9e8a26STejun Heo 	uint64_t id;
543381cb8dSTejun Heo 
553381cb8dSTejun Heo 	/* TODO: per-policy storage in blkio_cgroup */
563381cb8dSTejun Heo 	unsigned int cfq_weight;	/* belongs to cfq */
5731e4c28dSVivek Goyal };
5831e4c28dSVivek Goyal 
59edcb0722STejun Heo struct blkg_stat {
60edf1b879STejun Heo 	struct u64_stats_sync		syncp;
61edcb0722STejun Heo 	uint64_t			cnt;
62edcb0722STejun Heo };
63edcb0722STejun Heo 
64edcb0722STejun Heo struct blkg_rwstat {
65edcb0722STejun Heo 	struct u64_stats_sync		syncp;
66edcb0722STejun Heo 	uint64_t			cnt[BLKG_RWSTAT_NR];
67edcb0722STejun Heo };
68edcb0722STejun Heo 
690381411eSTejun Heo /* per-blkg per-policy data */
700381411eSTejun Heo struct blkg_policy_data {
710381411eSTejun Heo 	/* the blkg this per-policy data belongs to */
720381411eSTejun Heo 	struct blkio_group *blkg;
730381411eSTejun Heo 
740381411eSTejun Heo 	/* pol->pdata_size bytes of private data used by policy impl */
750381411eSTejun Heo 	char pdata[] __aligned(__alignof__(unsigned long long));
760381411eSTejun Heo };
770381411eSTejun Heo 
7831e4c28dSVivek Goyal struct blkio_group {
79c875f4d0STejun Heo 	/* Pointer to the associated request_queue */
80c875f4d0STejun Heo 	struct request_queue *q;
81e8989faeSTejun Heo 	struct list_head q_node;
8231e4c28dSVivek Goyal 	struct hlist_node blkcg_node;
837ee9c562STejun Heo 	struct blkio_cgroup *blkcg;
842868ef7bSVivek Goyal 	/* Store cgroup path */
852868ef7bSVivek Goyal 	char path[128];
861adaf3ddSTejun Heo 	/* reference count */
871adaf3ddSTejun Heo 	int refcnt;
8822084190SVivek Goyal 
89549d3aa8STejun Heo 	struct blkg_policy_data *pd[BLKIO_NR_POLICIES];
901adaf3ddSTejun Heo 
911adaf3ddSTejun Heo 	struct rcu_head rcu_head;
9231e4c28dSVivek Goyal };
9331e4c28dSVivek Goyal 
940381411eSTejun Heo typedef void (blkio_init_group_fn)(struct blkio_group *blkg);
959ade5ea4STejun Heo typedef void (blkio_exit_group_fn)(struct blkio_group *blkg);
969ade5ea4STejun Heo typedef void (blkio_reset_group_stats_fn)(struct blkio_group *blkg);
973e252066SVivek Goyal 
983e252066SVivek Goyal struct blkio_policy_ops {
990381411eSTejun Heo 	blkio_init_group_fn *blkio_init_group_fn;
1009ade5ea4STejun Heo 	blkio_exit_group_fn *blkio_exit_group_fn;
1019ade5ea4STejun Heo 	blkio_reset_group_stats_fn *blkio_reset_group_stats_fn;
1023e252066SVivek Goyal };
1033e252066SVivek Goyal 
1043e252066SVivek Goyal struct blkio_policy_type {
1053e252066SVivek Goyal 	struct list_head list;
1063e252066SVivek Goyal 	struct blkio_policy_ops ops;
107062a644dSVivek Goyal 	enum blkio_policy_id plid;
1080381411eSTejun Heo 	size_t pdata_size;		/* policy specific private data size */
10944ea53deSTejun Heo 	struct cftype *cftypes;		/* cgroup files for the policy */
1103e252066SVivek Goyal };
1113e252066SVivek Goyal 
1125efd6113STejun Heo extern int blkcg_init_queue(struct request_queue *q);
1135efd6113STejun Heo extern void blkcg_drain_queue(struct request_queue *q);
1145efd6113STejun Heo extern void blkcg_exit_queue(struct request_queue *q);
1155efd6113STejun Heo 
1163e252066SVivek Goyal /* Blkio controller policy registration */
1173e252066SVivek Goyal extern void blkio_policy_register(struct blkio_policy_type *);
1183e252066SVivek Goyal extern void blkio_policy_unregister(struct blkio_policy_type *);
119e8989faeSTejun Heo extern void blkg_destroy_all(struct request_queue *q, bool destroy_root);
120e8989faeSTejun Heo extern void update_root_blkg_pd(struct request_queue *q,
121e8989faeSTejun Heo 				enum blkio_policy_id plid);
1223e252066SVivek Goyal 
123829fdb50STejun Heo void blkcg_print_blkgs(struct seq_file *sf, struct blkio_cgroup *blkcg,
124d366e7ecSTejun Heo 		       u64 (*prfill)(struct seq_file *, void *, int),
125829fdb50STejun Heo 		       int pol, int data, bool show_total);
126d366e7ecSTejun Heo u64 __blkg_prfill_u64(struct seq_file *sf, void *pdata, u64 v);
127d366e7ecSTejun Heo u64 __blkg_prfill_rwstat(struct seq_file *sf, void *pdata,
128829fdb50STejun Heo 			 const struct blkg_rwstat *rwstat);
1295bc4afb1STejun Heo u64 blkg_prfill_stat(struct seq_file *sf, void *pdata, int off);
1305bc4afb1STejun Heo u64 blkg_prfill_rwstat(struct seq_file *sf, void *pdata, int off);
131829fdb50STejun Heo 
132829fdb50STejun Heo struct blkg_conf_ctx {
133829fdb50STejun Heo 	struct gendisk		*disk;
134829fdb50STejun Heo 	struct blkio_group	*blkg;
135829fdb50STejun Heo 	u64			v;
136829fdb50STejun Heo };
137829fdb50STejun Heo 
138829fdb50STejun Heo int blkg_conf_prep(struct blkio_cgroup *blkcg, const char *input,
139829fdb50STejun Heo 		   struct blkg_conf_ctx *ctx);
140829fdb50STejun Heo void blkg_conf_finish(struct blkg_conf_ctx *ctx);
141829fdb50STejun Heo 
142829fdb50STejun Heo 
1430381411eSTejun Heo /**
1440381411eSTejun Heo  * blkg_to_pdata - get policy private data
1450381411eSTejun Heo  * @blkg: blkg of interest
1460381411eSTejun Heo  * @pol: policy of interest
1470381411eSTejun Heo  *
1480381411eSTejun Heo  * Return pointer to private data associated with the @blkg-@pol pair.
1490381411eSTejun Heo  */
1500381411eSTejun Heo static inline void *blkg_to_pdata(struct blkio_group *blkg,
1510381411eSTejun Heo 			      struct blkio_policy_type *pol)
1520381411eSTejun Heo {
153549d3aa8STejun Heo 	return blkg ? blkg->pd[pol->plid]->pdata : NULL;
1540381411eSTejun Heo }
1550381411eSTejun Heo 
1560381411eSTejun Heo /**
1570381411eSTejun Heo  * pdata_to_blkg - get blkg associated with policy private data
1580381411eSTejun Heo  * @pdata: policy private data of interest
1590381411eSTejun Heo  *
160aaec55a0STejun Heo  * @pdata is policy private data.  Determine the blkg it's associated with.
1610381411eSTejun Heo  */
162aaec55a0STejun Heo static inline struct blkio_group *pdata_to_blkg(void *pdata)
1630381411eSTejun Heo {
1640381411eSTejun Heo 	if (pdata) {
1650381411eSTejun Heo 		struct blkg_policy_data *pd =
1660381411eSTejun Heo 			container_of(pdata, struct blkg_policy_data, pdata);
1670381411eSTejun Heo 		return pd->blkg;
1680381411eSTejun Heo 	}
1690381411eSTejun Heo 	return NULL;
1700381411eSTejun Heo }
1710381411eSTejun Heo 
172afc24d49SVivek Goyal static inline char *blkg_path(struct blkio_group *blkg)
173afc24d49SVivek Goyal {
174afc24d49SVivek Goyal 	return blkg->path;
175afc24d49SVivek Goyal }
176afc24d49SVivek Goyal 
1771adaf3ddSTejun Heo /**
1781adaf3ddSTejun Heo  * blkg_get - get a blkg reference
1791adaf3ddSTejun Heo  * @blkg: blkg to get
1801adaf3ddSTejun Heo  *
1811adaf3ddSTejun Heo  * The caller should be holding queue_lock and an existing reference.
1821adaf3ddSTejun Heo  */
1831adaf3ddSTejun Heo static inline void blkg_get(struct blkio_group *blkg)
1841adaf3ddSTejun Heo {
1851adaf3ddSTejun Heo 	lockdep_assert_held(blkg->q->queue_lock);
1861adaf3ddSTejun Heo 	WARN_ON_ONCE(!blkg->refcnt);
1871adaf3ddSTejun Heo 	blkg->refcnt++;
1881adaf3ddSTejun Heo }
1891adaf3ddSTejun Heo 
1901adaf3ddSTejun Heo void __blkg_release(struct blkio_group *blkg);
1911adaf3ddSTejun Heo 
1921adaf3ddSTejun Heo /**
1931adaf3ddSTejun Heo  * blkg_put - put a blkg reference
1941adaf3ddSTejun Heo  * @blkg: blkg to put
1951adaf3ddSTejun Heo  *
1961adaf3ddSTejun Heo  * The caller should be holding queue_lock.
1971adaf3ddSTejun Heo  */
1981adaf3ddSTejun Heo static inline void blkg_put(struct blkio_group *blkg)
1991adaf3ddSTejun Heo {
2001adaf3ddSTejun Heo 	lockdep_assert_held(blkg->q->queue_lock);
2011adaf3ddSTejun Heo 	WARN_ON_ONCE(blkg->refcnt <= 0);
2021adaf3ddSTejun Heo 	if (!--blkg->refcnt)
2031adaf3ddSTejun Heo 		__blkg_release(blkg);
2041adaf3ddSTejun Heo }
2051adaf3ddSTejun Heo 
206edcb0722STejun Heo /**
207edcb0722STejun Heo  * blkg_stat_add - add a value to a blkg_stat
208edcb0722STejun Heo  * @stat: target blkg_stat
209edcb0722STejun Heo  * @val: value to add
210edcb0722STejun Heo  *
211edcb0722STejun Heo  * Add @val to @stat.  The caller is responsible for synchronizing calls to
212edcb0722STejun Heo  * this function.
213edcb0722STejun Heo  */
214edcb0722STejun Heo static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
215edcb0722STejun Heo {
216edcb0722STejun Heo 	u64_stats_update_begin(&stat->syncp);
217edcb0722STejun Heo 	stat->cnt += val;
218edcb0722STejun Heo 	u64_stats_update_end(&stat->syncp);
219edcb0722STejun Heo }
220edcb0722STejun Heo 
221edcb0722STejun Heo /**
222edcb0722STejun Heo  * blkg_stat_read - read the current value of a blkg_stat
223edcb0722STejun Heo  * @stat: blkg_stat to read
224edcb0722STejun Heo  *
225edcb0722STejun Heo  * Read the current value of @stat.  This function can be called without
226edcb0722STejun Heo  * synchroniztion and takes care of u64 atomicity.
227edcb0722STejun Heo  */
228edcb0722STejun Heo static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
229edcb0722STejun Heo {
230edcb0722STejun Heo 	unsigned int start;
231edcb0722STejun Heo 	uint64_t v;
232edcb0722STejun Heo 
233edcb0722STejun Heo 	do {
234edcb0722STejun Heo 		start = u64_stats_fetch_begin(&stat->syncp);
235edcb0722STejun Heo 		v = stat->cnt;
236edcb0722STejun Heo 	} while (u64_stats_fetch_retry(&stat->syncp, start));
237edcb0722STejun Heo 
238edcb0722STejun Heo 	return v;
239edcb0722STejun Heo }
240edcb0722STejun Heo 
241edcb0722STejun Heo /**
242edcb0722STejun Heo  * blkg_stat_reset - reset a blkg_stat
243edcb0722STejun Heo  * @stat: blkg_stat to reset
244edcb0722STejun Heo  */
245edcb0722STejun Heo static inline void blkg_stat_reset(struct blkg_stat *stat)
246edcb0722STejun Heo {
247edcb0722STejun Heo 	stat->cnt = 0;
248edcb0722STejun Heo }
249edcb0722STejun Heo 
250edcb0722STejun Heo /**
251edcb0722STejun Heo  * blkg_rwstat_add - add a value to a blkg_rwstat
252edcb0722STejun Heo  * @rwstat: target blkg_rwstat
253edcb0722STejun Heo  * @rw: mask of REQ_{WRITE|SYNC}
254edcb0722STejun Heo  * @val: value to add
255edcb0722STejun Heo  *
256edcb0722STejun Heo  * Add @val to @rwstat.  The counters are chosen according to @rw.  The
257edcb0722STejun Heo  * caller is responsible for synchronizing calls to this function.
258edcb0722STejun Heo  */
259edcb0722STejun Heo static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
260edcb0722STejun Heo 				   int rw, uint64_t val)
261edcb0722STejun Heo {
262edcb0722STejun Heo 	u64_stats_update_begin(&rwstat->syncp);
263edcb0722STejun Heo 
264edcb0722STejun Heo 	if (rw & REQ_WRITE)
265edcb0722STejun Heo 		rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
266edcb0722STejun Heo 	else
267edcb0722STejun Heo 		rwstat->cnt[BLKG_RWSTAT_READ] += val;
268edcb0722STejun Heo 	if (rw & REQ_SYNC)
269edcb0722STejun Heo 		rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
270edcb0722STejun Heo 	else
271edcb0722STejun Heo 		rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
272edcb0722STejun Heo 
273edcb0722STejun Heo 	u64_stats_update_end(&rwstat->syncp);
274edcb0722STejun Heo }
275edcb0722STejun Heo 
276edcb0722STejun Heo /**
277edcb0722STejun Heo  * blkg_rwstat_read - read the current values of a blkg_rwstat
278edcb0722STejun Heo  * @rwstat: blkg_rwstat to read
279edcb0722STejun Heo  *
280edcb0722STejun Heo  * Read the current snapshot of @rwstat and return it as the return value.
281edcb0722STejun Heo  * This function can be called without synchronization and takes care of
282edcb0722STejun Heo  * u64 atomicity.
283edcb0722STejun Heo  */
284edcb0722STejun Heo static struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
285edcb0722STejun Heo {
286edcb0722STejun Heo 	unsigned int start;
287edcb0722STejun Heo 	struct blkg_rwstat tmp;
288edcb0722STejun Heo 
289edcb0722STejun Heo 	do {
290edcb0722STejun Heo 		start = u64_stats_fetch_begin(&rwstat->syncp);
291edcb0722STejun Heo 		tmp = *rwstat;
292edcb0722STejun Heo 	} while (u64_stats_fetch_retry(&rwstat->syncp, start));
293edcb0722STejun Heo 
294edcb0722STejun Heo 	return tmp;
295edcb0722STejun Heo }
296edcb0722STejun Heo 
297edcb0722STejun Heo /**
298edcb0722STejun Heo  * blkg_rwstat_sum - read the total count of a blkg_rwstat
299edcb0722STejun Heo  * @rwstat: blkg_rwstat to read
300edcb0722STejun Heo  *
301edcb0722STejun Heo  * Return the total count of @rwstat regardless of the IO direction.  This
302edcb0722STejun Heo  * function can be called without synchronization and takes care of u64
303edcb0722STejun Heo  * atomicity.
304edcb0722STejun Heo  */
305edcb0722STejun Heo static inline uint64_t blkg_rwstat_sum(struct blkg_rwstat *rwstat)
306edcb0722STejun Heo {
307edcb0722STejun Heo 	struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
308edcb0722STejun Heo 
309edcb0722STejun Heo 	return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
310edcb0722STejun Heo }
311edcb0722STejun Heo 
312edcb0722STejun Heo /**
313edcb0722STejun Heo  * blkg_rwstat_reset - reset a blkg_rwstat
314edcb0722STejun Heo  * @rwstat: blkg_rwstat to reset
315edcb0722STejun Heo  */
316edcb0722STejun Heo static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
317edcb0722STejun Heo {
318edcb0722STejun Heo 	memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
319edcb0722STejun Heo }
320edcb0722STejun Heo 
3212f5ea477SJens Axboe #else
3222f5ea477SJens Axboe 
3232f5ea477SJens Axboe struct blkio_group {
3242f5ea477SJens Axboe };
3252f5ea477SJens Axboe 
3263e252066SVivek Goyal struct blkio_policy_type {
3273e252066SVivek Goyal };
3283e252066SVivek Goyal 
3295efd6113STejun Heo static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
3305efd6113STejun Heo static inline void blkcg_drain_queue(struct request_queue *q) { }
3315efd6113STejun Heo static inline void blkcg_exit_queue(struct request_queue *q) { }
3323e252066SVivek Goyal static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { }
3333e252066SVivek Goyal static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
33403aa264aSTejun Heo static inline void blkg_destroy_all(struct request_queue *q,
33503aa264aSTejun Heo 				    bool destory_root) { }
336e8989faeSTejun Heo static inline void update_root_blkg_pd(struct request_queue *q,
337e8989faeSTejun Heo 				       enum blkio_policy_id plid) { }
3383e252066SVivek Goyal 
3390381411eSTejun Heo static inline void *blkg_to_pdata(struct blkio_group *blkg,
3400381411eSTejun Heo 				struct blkio_policy_type *pol) { return NULL; }
3410381411eSTejun Heo static inline struct blkio_group *pdata_to_blkg(void *pdata,
3420381411eSTejun Heo 				struct blkio_policy_type *pol) { return NULL; }
343afc24d49SVivek Goyal static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
3441adaf3ddSTejun Heo static inline void blkg_get(struct blkio_group *blkg) { }
3451adaf3ddSTejun Heo static inline void blkg_put(struct blkio_group *blkg) { }
346afc24d49SVivek Goyal 
3472f5ea477SJens Axboe #endif
3482f5ea477SJens Axboe 
34932e380aeSTejun Heo #ifdef CONFIG_BLK_CGROUP
35031e4c28dSVivek Goyal extern struct blkio_cgroup blkio_root_cgroup;
35131e4c28dSVivek Goyal extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
3524f85cb96STejun Heo extern struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio);
353cd1604faSTejun Heo extern struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
354e8989faeSTejun Heo 				       struct request_queue *q);
355cd1604faSTejun Heo struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg,
356cd1604faSTejun Heo 				       struct request_queue *q,
357cd1604faSTejun Heo 				       bool for_root);
35831e4c28dSVivek Goyal #else
3592f5ea477SJens Axboe struct cgroup;
36031e4c28dSVivek Goyal static inline struct blkio_cgroup *
36131e4c28dSVivek Goyal cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
36270087dc3SVivek Goyal static inline struct blkio_cgroup *
3634f85cb96STejun Heo bio_blkio_cgroup(struct bio *bio) { return NULL; }
36431e4c28dSVivek Goyal 
365cd1604faSTejun Heo static inline struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
366cd1604faSTejun Heo 					      void *key) { return NULL; }
36731e4c28dSVivek Goyal #endif
36831e4c28dSVivek Goyal #endif /* _BLK_CGROUP_H */
369