xref: /linux/block/blk-cgroup.h (revision 8bd435b30ecacb69bbb8b2d3e251f770b807c5b2)
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 
209355aedeSVivek Goyal /* Max limits for throttle policy */
219355aedeSVivek Goyal #define THROTL_IOPS_MAX		UINT_MAX
229355aedeSVivek Goyal 
233381cb8dSTejun Heo /* CFQ specific, out here for blkcg->cfq_weight */
243381cb8dSTejun Heo #define CFQ_WEIGHT_MIN		10
253381cb8dSTejun Heo #define CFQ_WEIGHT_MAX		1000
263381cb8dSTejun Heo #define CFQ_WEIGHT_DEFAULT	500
273381cb8dSTejun Heo 
28f48ec1d7STejun Heo #ifdef CONFIG_BLK_CGROUP
29f48ec1d7STejun Heo 
30edcb0722STejun Heo enum blkg_rwstat_type {
31edcb0722STejun Heo 	BLKG_RWSTAT_READ,
32edcb0722STejun Heo 	BLKG_RWSTAT_WRITE,
33edcb0722STejun Heo 	BLKG_RWSTAT_SYNC,
34edcb0722STejun Heo 	BLKG_RWSTAT_ASYNC,
35edcb0722STejun Heo 
36edcb0722STejun Heo 	BLKG_RWSTAT_NR,
37edcb0722STejun Heo 	BLKG_RWSTAT_TOTAL = BLKG_RWSTAT_NR,
38303a3acbSDivyesh Shah };
39303a3acbSDivyesh Shah 
4031e4c28dSVivek Goyal struct blkio_cgroup {
4131e4c28dSVivek Goyal 	struct cgroup_subsys_state css;
4231e4c28dSVivek Goyal 	spinlock_t lock;
4331e4c28dSVivek Goyal 	struct hlist_head blkg_list;
449a9e8a26STejun Heo 
459a9e8a26STejun Heo 	/* for policies to test whether associated blkcg has changed */
469a9e8a26STejun Heo 	uint64_t id;
473381cb8dSTejun Heo 
483381cb8dSTejun Heo 	/* TODO: per-policy storage in blkio_cgroup */
493381cb8dSTejun Heo 	unsigned int cfq_weight;	/* belongs to cfq */
5031e4c28dSVivek Goyal };
5131e4c28dSVivek Goyal 
52edcb0722STejun Heo struct blkg_stat {
53edf1b879STejun Heo 	struct u64_stats_sync		syncp;
54edcb0722STejun Heo 	uint64_t			cnt;
55edcb0722STejun Heo };
56edcb0722STejun Heo 
57edcb0722STejun Heo struct blkg_rwstat {
58edcb0722STejun Heo 	struct u64_stats_sync		syncp;
59edcb0722STejun Heo 	uint64_t			cnt[BLKG_RWSTAT_NR];
60edcb0722STejun Heo };
61edcb0722STejun Heo 
620381411eSTejun Heo /* per-blkg per-policy data */
630381411eSTejun Heo struct blkg_policy_data {
640381411eSTejun Heo 	/* the blkg this per-policy data belongs to */
650381411eSTejun Heo 	struct blkio_group *blkg;
660381411eSTejun Heo 
670381411eSTejun Heo 	/* pol->pdata_size bytes of private data used by policy impl */
680381411eSTejun Heo 	char pdata[] __aligned(__alignof__(unsigned long long));
690381411eSTejun Heo };
700381411eSTejun Heo 
7131e4c28dSVivek Goyal struct blkio_group {
72c875f4d0STejun Heo 	/* Pointer to the associated request_queue */
73c875f4d0STejun Heo 	struct request_queue *q;
74e8989faeSTejun Heo 	struct list_head q_node;
7531e4c28dSVivek Goyal 	struct hlist_node blkcg_node;
767ee9c562STejun Heo 	struct blkio_cgroup *blkcg;
772868ef7bSVivek Goyal 	/* Store cgroup path */
782868ef7bSVivek Goyal 	char path[128];
791adaf3ddSTejun Heo 	/* reference count */
801adaf3ddSTejun Heo 	int refcnt;
8122084190SVivek Goyal 
82*8bd435b3STejun Heo 	struct blkg_policy_data *pd[BLKCG_MAX_POLS];
831adaf3ddSTejun Heo 
841adaf3ddSTejun Heo 	struct rcu_head rcu_head;
8531e4c28dSVivek Goyal };
8631e4c28dSVivek Goyal 
870381411eSTejun Heo typedef void (blkio_init_group_fn)(struct blkio_group *blkg);
889ade5ea4STejun Heo typedef void (blkio_exit_group_fn)(struct blkio_group *blkg);
899ade5ea4STejun Heo typedef void (blkio_reset_group_stats_fn)(struct blkio_group *blkg);
903e252066SVivek Goyal 
913e252066SVivek Goyal struct blkio_policy_ops {
920381411eSTejun Heo 	blkio_init_group_fn *blkio_init_group_fn;
939ade5ea4STejun Heo 	blkio_exit_group_fn *blkio_exit_group_fn;
949ade5ea4STejun Heo 	blkio_reset_group_stats_fn *blkio_reset_group_stats_fn;
953e252066SVivek Goyal };
963e252066SVivek Goyal 
973e252066SVivek Goyal struct blkio_policy_type {
983e252066SVivek Goyal 	struct blkio_policy_ops ops;
99*8bd435b3STejun Heo 	int plid;
1000381411eSTejun Heo 	size_t pdata_size;		/* policy specific private data size */
10144ea53deSTejun Heo 	struct cftype *cftypes;		/* cgroup files for the policy */
1023e252066SVivek Goyal };
1033e252066SVivek Goyal 
1045efd6113STejun Heo extern int blkcg_init_queue(struct request_queue *q);
1055efd6113STejun Heo extern void blkcg_drain_queue(struct request_queue *q);
1065efd6113STejun Heo extern void blkcg_exit_queue(struct request_queue *q);
1075efd6113STejun Heo 
1083e252066SVivek Goyal /* Blkio controller policy registration */
109*8bd435b3STejun Heo extern int blkio_policy_register(struct blkio_policy_type *);
1103e252066SVivek Goyal extern void blkio_policy_unregister(struct blkio_policy_type *);
111e8989faeSTejun Heo extern void blkg_destroy_all(struct request_queue *q, bool destroy_root);
112e8989faeSTejun Heo extern void update_root_blkg_pd(struct request_queue *q,
113ec399347STejun Heo 				const struct blkio_policy_type *pol);
1143e252066SVivek Goyal 
115829fdb50STejun Heo void blkcg_print_blkgs(struct seq_file *sf, struct blkio_cgroup *blkcg,
116d366e7ecSTejun Heo 		       u64 (*prfill)(struct seq_file *, void *, int),
117ec399347STejun Heo 		       const struct blkio_policy_type *pol, int data,
118ec399347STejun Heo 		       bool show_total);
119d366e7ecSTejun Heo u64 __blkg_prfill_u64(struct seq_file *sf, void *pdata, u64 v);
120d366e7ecSTejun Heo u64 __blkg_prfill_rwstat(struct seq_file *sf, void *pdata,
121829fdb50STejun Heo 			 const struct blkg_rwstat *rwstat);
1225bc4afb1STejun Heo u64 blkg_prfill_stat(struct seq_file *sf, void *pdata, int off);
1235bc4afb1STejun Heo u64 blkg_prfill_rwstat(struct seq_file *sf, void *pdata, int off);
124829fdb50STejun Heo 
125829fdb50STejun Heo struct blkg_conf_ctx {
126829fdb50STejun Heo 	struct gendisk		*disk;
127829fdb50STejun Heo 	struct blkio_group	*blkg;
128829fdb50STejun Heo 	u64			v;
129829fdb50STejun Heo };
130829fdb50STejun Heo 
131829fdb50STejun Heo int blkg_conf_prep(struct blkio_cgroup *blkcg, const char *input,
132829fdb50STejun Heo 		   struct blkg_conf_ctx *ctx);
133829fdb50STejun Heo void blkg_conf_finish(struct blkg_conf_ctx *ctx);
134829fdb50STejun Heo 
135829fdb50STejun Heo 
1360381411eSTejun Heo /**
1370381411eSTejun Heo  * blkg_to_pdata - get policy private data
1380381411eSTejun Heo  * @blkg: blkg of interest
1390381411eSTejun Heo  * @pol: policy of interest
1400381411eSTejun Heo  *
1410381411eSTejun Heo  * Return pointer to private data associated with the @blkg-@pol pair.
1420381411eSTejun Heo  */
1430381411eSTejun Heo static inline void *blkg_to_pdata(struct blkio_group *blkg,
1440381411eSTejun Heo 			      struct blkio_policy_type *pol)
1450381411eSTejun Heo {
146549d3aa8STejun Heo 	return blkg ? blkg->pd[pol->plid]->pdata : NULL;
1470381411eSTejun Heo }
1480381411eSTejun Heo 
1490381411eSTejun Heo /**
1500381411eSTejun Heo  * pdata_to_blkg - get blkg associated with policy private data
1510381411eSTejun Heo  * @pdata: policy private data of interest
1520381411eSTejun Heo  *
153aaec55a0STejun Heo  * @pdata is policy private data.  Determine the blkg it's associated with.
1540381411eSTejun Heo  */
155aaec55a0STejun Heo static inline struct blkio_group *pdata_to_blkg(void *pdata)
1560381411eSTejun Heo {
1570381411eSTejun Heo 	if (pdata) {
1580381411eSTejun Heo 		struct blkg_policy_data *pd =
1590381411eSTejun Heo 			container_of(pdata, struct blkg_policy_data, pdata);
1600381411eSTejun Heo 		return pd->blkg;
1610381411eSTejun Heo 	}
1620381411eSTejun Heo 	return NULL;
1630381411eSTejun Heo }
1640381411eSTejun Heo 
165afc24d49SVivek Goyal static inline char *blkg_path(struct blkio_group *blkg)
166afc24d49SVivek Goyal {
167afc24d49SVivek Goyal 	return blkg->path;
168afc24d49SVivek Goyal }
169afc24d49SVivek Goyal 
1701adaf3ddSTejun Heo /**
1711adaf3ddSTejun Heo  * blkg_get - get a blkg reference
1721adaf3ddSTejun Heo  * @blkg: blkg to get
1731adaf3ddSTejun Heo  *
1741adaf3ddSTejun Heo  * The caller should be holding queue_lock and an existing reference.
1751adaf3ddSTejun Heo  */
1761adaf3ddSTejun Heo static inline void blkg_get(struct blkio_group *blkg)
1771adaf3ddSTejun Heo {
1781adaf3ddSTejun Heo 	lockdep_assert_held(blkg->q->queue_lock);
1791adaf3ddSTejun Heo 	WARN_ON_ONCE(!blkg->refcnt);
1801adaf3ddSTejun Heo 	blkg->refcnt++;
1811adaf3ddSTejun Heo }
1821adaf3ddSTejun Heo 
1831adaf3ddSTejun Heo void __blkg_release(struct blkio_group *blkg);
1841adaf3ddSTejun Heo 
1851adaf3ddSTejun Heo /**
1861adaf3ddSTejun Heo  * blkg_put - put a blkg reference
1871adaf3ddSTejun Heo  * @blkg: blkg to put
1881adaf3ddSTejun Heo  *
1891adaf3ddSTejun Heo  * The caller should be holding queue_lock.
1901adaf3ddSTejun Heo  */
1911adaf3ddSTejun Heo static inline void blkg_put(struct blkio_group *blkg)
1921adaf3ddSTejun Heo {
1931adaf3ddSTejun Heo 	lockdep_assert_held(blkg->q->queue_lock);
1941adaf3ddSTejun Heo 	WARN_ON_ONCE(blkg->refcnt <= 0);
1951adaf3ddSTejun Heo 	if (!--blkg->refcnt)
1961adaf3ddSTejun Heo 		__blkg_release(blkg);
1971adaf3ddSTejun Heo }
1981adaf3ddSTejun Heo 
199edcb0722STejun Heo /**
200edcb0722STejun Heo  * blkg_stat_add - add a value to a blkg_stat
201edcb0722STejun Heo  * @stat: target blkg_stat
202edcb0722STejun Heo  * @val: value to add
203edcb0722STejun Heo  *
204edcb0722STejun Heo  * Add @val to @stat.  The caller is responsible for synchronizing calls to
205edcb0722STejun Heo  * this function.
206edcb0722STejun Heo  */
207edcb0722STejun Heo static inline void blkg_stat_add(struct blkg_stat *stat, uint64_t val)
208edcb0722STejun Heo {
209edcb0722STejun Heo 	u64_stats_update_begin(&stat->syncp);
210edcb0722STejun Heo 	stat->cnt += val;
211edcb0722STejun Heo 	u64_stats_update_end(&stat->syncp);
212edcb0722STejun Heo }
213edcb0722STejun Heo 
214edcb0722STejun Heo /**
215edcb0722STejun Heo  * blkg_stat_read - read the current value of a blkg_stat
216edcb0722STejun Heo  * @stat: blkg_stat to read
217edcb0722STejun Heo  *
218edcb0722STejun Heo  * Read the current value of @stat.  This function can be called without
219edcb0722STejun Heo  * synchroniztion and takes care of u64 atomicity.
220edcb0722STejun Heo  */
221edcb0722STejun Heo static inline uint64_t blkg_stat_read(struct blkg_stat *stat)
222edcb0722STejun Heo {
223edcb0722STejun Heo 	unsigned int start;
224edcb0722STejun Heo 	uint64_t v;
225edcb0722STejun Heo 
226edcb0722STejun Heo 	do {
227edcb0722STejun Heo 		start = u64_stats_fetch_begin(&stat->syncp);
228edcb0722STejun Heo 		v = stat->cnt;
229edcb0722STejun Heo 	} while (u64_stats_fetch_retry(&stat->syncp, start));
230edcb0722STejun Heo 
231edcb0722STejun Heo 	return v;
232edcb0722STejun Heo }
233edcb0722STejun Heo 
234edcb0722STejun Heo /**
235edcb0722STejun Heo  * blkg_stat_reset - reset a blkg_stat
236edcb0722STejun Heo  * @stat: blkg_stat to reset
237edcb0722STejun Heo  */
238edcb0722STejun Heo static inline void blkg_stat_reset(struct blkg_stat *stat)
239edcb0722STejun Heo {
240edcb0722STejun Heo 	stat->cnt = 0;
241edcb0722STejun Heo }
242edcb0722STejun Heo 
243edcb0722STejun Heo /**
244edcb0722STejun Heo  * blkg_rwstat_add - add a value to a blkg_rwstat
245edcb0722STejun Heo  * @rwstat: target blkg_rwstat
246edcb0722STejun Heo  * @rw: mask of REQ_{WRITE|SYNC}
247edcb0722STejun Heo  * @val: value to add
248edcb0722STejun Heo  *
249edcb0722STejun Heo  * Add @val to @rwstat.  The counters are chosen according to @rw.  The
250edcb0722STejun Heo  * caller is responsible for synchronizing calls to this function.
251edcb0722STejun Heo  */
252edcb0722STejun Heo static inline void blkg_rwstat_add(struct blkg_rwstat *rwstat,
253edcb0722STejun Heo 				   int rw, uint64_t val)
254edcb0722STejun Heo {
255edcb0722STejun Heo 	u64_stats_update_begin(&rwstat->syncp);
256edcb0722STejun Heo 
257edcb0722STejun Heo 	if (rw & REQ_WRITE)
258edcb0722STejun Heo 		rwstat->cnt[BLKG_RWSTAT_WRITE] += val;
259edcb0722STejun Heo 	else
260edcb0722STejun Heo 		rwstat->cnt[BLKG_RWSTAT_READ] += val;
261edcb0722STejun Heo 	if (rw & REQ_SYNC)
262edcb0722STejun Heo 		rwstat->cnt[BLKG_RWSTAT_SYNC] += val;
263edcb0722STejun Heo 	else
264edcb0722STejun Heo 		rwstat->cnt[BLKG_RWSTAT_ASYNC] += val;
265edcb0722STejun Heo 
266edcb0722STejun Heo 	u64_stats_update_end(&rwstat->syncp);
267edcb0722STejun Heo }
268edcb0722STejun Heo 
269edcb0722STejun Heo /**
270edcb0722STejun Heo  * blkg_rwstat_read - read the current values of a blkg_rwstat
271edcb0722STejun Heo  * @rwstat: blkg_rwstat to read
272edcb0722STejun Heo  *
273edcb0722STejun Heo  * Read the current snapshot of @rwstat and return it as the return value.
274edcb0722STejun Heo  * This function can be called without synchronization and takes care of
275edcb0722STejun Heo  * u64 atomicity.
276edcb0722STejun Heo  */
277edcb0722STejun Heo static struct blkg_rwstat blkg_rwstat_read(struct blkg_rwstat *rwstat)
278edcb0722STejun Heo {
279edcb0722STejun Heo 	unsigned int start;
280edcb0722STejun Heo 	struct blkg_rwstat tmp;
281edcb0722STejun Heo 
282edcb0722STejun Heo 	do {
283edcb0722STejun Heo 		start = u64_stats_fetch_begin(&rwstat->syncp);
284edcb0722STejun Heo 		tmp = *rwstat;
285edcb0722STejun Heo 	} while (u64_stats_fetch_retry(&rwstat->syncp, start));
286edcb0722STejun Heo 
287edcb0722STejun Heo 	return tmp;
288edcb0722STejun Heo }
289edcb0722STejun Heo 
290edcb0722STejun Heo /**
291edcb0722STejun Heo  * blkg_rwstat_sum - read the total count of a blkg_rwstat
292edcb0722STejun Heo  * @rwstat: blkg_rwstat to read
293edcb0722STejun Heo  *
294edcb0722STejun Heo  * Return the total count of @rwstat regardless of the IO direction.  This
295edcb0722STejun Heo  * function can be called without synchronization and takes care of u64
296edcb0722STejun Heo  * atomicity.
297edcb0722STejun Heo  */
298edcb0722STejun Heo static inline uint64_t blkg_rwstat_sum(struct blkg_rwstat *rwstat)
299edcb0722STejun Heo {
300edcb0722STejun Heo 	struct blkg_rwstat tmp = blkg_rwstat_read(rwstat);
301edcb0722STejun Heo 
302edcb0722STejun Heo 	return tmp.cnt[BLKG_RWSTAT_READ] + tmp.cnt[BLKG_RWSTAT_WRITE];
303edcb0722STejun Heo }
304edcb0722STejun Heo 
305edcb0722STejun Heo /**
306edcb0722STejun Heo  * blkg_rwstat_reset - reset a blkg_rwstat
307edcb0722STejun Heo  * @rwstat: blkg_rwstat to reset
308edcb0722STejun Heo  */
309edcb0722STejun Heo static inline void blkg_rwstat_reset(struct blkg_rwstat *rwstat)
310edcb0722STejun Heo {
311edcb0722STejun Heo 	memset(rwstat->cnt, 0, sizeof(rwstat->cnt));
312edcb0722STejun Heo }
313edcb0722STejun Heo 
3142f5ea477SJens Axboe #else
3152f5ea477SJens Axboe 
3162f5ea477SJens Axboe struct blkio_group {
3172f5ea477SJens Axboe };
3182f5ea477SJens Axboe 
3193e252066SVivek Goyal struct blkio_policy_type {
3203e252066SVivek Goyal };
3213e252066SVivek Goyal 
3225efd6113STejun Heo static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
3235efd6113STejun Heo static inline void blkcg_drain_queue(struct request_queue *q) { }
3245efd6113STejun Heo static inline void blkcg_exit_queue(struct request_queue *q) { }
325*8bd435b3STejun Heo static inline int blkio_policy_register(struct blkio_policy_type *blkiop) { return 0; }
3263e252066SVivek Goyal static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
32703aa264aSTejun Heo static inline void blkg_destroy_all(struct request_queue *q,
32803aa264aSTejun Heo 				    bool destory_root) { }
329e8989faeSTejun Heo static inline void update_root_blkg_pd(struct request_queue *q,
330ec399347STejun Heo 				       const struct blkio_policy_type *pol) { }
3313e252066SVivek Goyal 
3320381411eSTejun Heo static inline void *blkg_to_pdata(struct blkio_group *blkg,
3330381411eSTejun Heo 				struct blkio_policy_type *pol) { return NULL; }
3340381411eSTejun Heo static inline struct blkio_group *pdata_to_blkg(void *pdata,
3350381411eSTejun Heo 				struct blkio_policy_type *pol) { return NULL; }
336afc24d49SVivek Goyal static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
3371adaf3ddSTejun Heo static inline void blkg_get(struct blkio_group *blkg) { }
3381adaf3ddSTejun Heo static inline void blkg_put(struct blkio_group *blkg) { }
339afc24d49SVivek Goyal 
3402f5ea477SJens Axboe #endif
3412f5ea477SJens Axboe 
34232e380aeSTejun Heo #ifdef CONFIG_BLK_CGROUP
34331e4c28dSVivek Goyal extern struct blkio_cgroup blkio_root_cgroup;
34431e4c28dSVivek Goyal extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
3454f85cb96STejun Heo extern struct blkio_cgroup *bio_blkio_cgroup(struct bio *bio);
346cd1604faSTejun Heo extern struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
347e8989faeSTejun Heo 				       struct request_queue *q);
348cd1604faSTejun Heo struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg,
349cd1604faSTejun Heo 				       struct request_queue *q,
350cd1604faSTejun Heo 				       bool for_root);
35131e4c28dSVivek Goyal #else
3522f5ea477SJens Axboe struct cgroup;
35331e4c28dSVivek Goyal static inline struct blkio_cgroup *
35431e4c28dSVivek Goyal cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
35570087dc3SVivek Goyal static inline struct blkio_cgroup *
3564f85cb96STejun Heo bio_blkio_cgroup(struct bio *bio) { return NULL; }
35731e4c28dSVivek Goyal 
358cd1604faSTejun Heo static inline struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
359cd1604faSTejun Heo 					      void *key) { return NULL; }
36031e4c28dSVivek Goyal #endif
36131e4c28dSVivek Goyal #endif /* _BLK_CGROUP_H */
362