]> bbs.cooldavid.org Git - net-next-2.6.git/blame - block/blk-cgroup.h
blkio: Core implementation of throttle policy
[net-next-2.6.git] / block / blk-cgroup.h
CommitLineData
31e4c28d
VG
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
062a644d
VG
18enum blkio_policy_id {
19 BLKIO_POLICY_PROP = 0, /* Proportional Bandwidth division */
4c9eefa1 20 BLKIO_POLICY_THROTL, /* Throttling */
062a644d
VG
21};
22
67523c48
BB
23#if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
24
25#ifndef CONFIG_BLK_CGROUP
26/* When blk-cgroup is a module, its subsys_id isn't a compile-time constant */
27extern struct cgroup_subsys blkio_subsys;
28#define blkio_subsys_id blkio_subsys.subsys_id
29#endif
2f5ea477 30
84c124da
DS
31enum stat_type {
32 /* Total time spent (in ns) between request dispatch to the driver and
33 * request completion for IOs doen by this cgroup. This may not be
34 * accurate when NCQ is turned on. */
35 BLKIO_STAT_SERVICE_TIME = 0,
36 /* Total bytes transferred */
37 BLKIO_STAT_SERVICE_BYTES,
38 /* Total IOs serviced, post merge */
39 BLKIO_STAT_SERVICED,
40 /* Total time spent waiting in scheduler queue in ns */
41 BLKIO_STAT_WAIT_TIME,
812d4026
DS
42 /* Number of IOs merged */
43 BLKIO_STAT_MERGED,
cdc1184c
DS
44 /* Number of IOs queued up */
45 BLKIO_STAT_QUEUED,
84c124da
DS
46 /* All the single valued stats go below this */
47 BLKIO_STAT_TIME,
48 BLKIO_STAT_SECTORS,
49#ifdef CONFIG_DEBUG_BLK_CGROUP
cdc1184c 50 BLKIO_STAT_AVG_QUEUE_SIZE,
812df48d
DS
51 BLKIO_STAT_IDLE_TIME,
52 BLKIO_STAT_EMPTY_TIME,
53 BLKIO_STAT_GROUP_WAIT_TIME,
84c124da
DS
54 BLKIO_STAT_DEQUEUE
55#endif
56};
57
58enum stat_sub_type {
59 BLKIO_STAT_READ = 0,
60 BLKIO_STAT_WRITE,
61 BLKIO_STAT_SYNC,
62 BLKIO_STAT_ASYNC,
63 BLKIO_STAT_TOTAL
303a3acb
DS
64};
65
812df48d
DS
66/* blkg state flags */
67enum blkg_state_flags {
68 BLKG_waiting = 0,
69 BLKG_idling,
70 BLKG_empty,
71};
72
062a644d
VG
73/* cgroup files owned by proportional weight policy */
74enum blkcg_file_name_prop {
75 BLKIO_PROP_weight = 1,
76 BLKIO_PROP_weight_device,
77 BLKIO_PROP_io_service_bytes,
78 BLKIO_PROP_io_serviced,
79 BLKIO_PROP_time,
80 BLKIO_PROP_sectors,
81 BLKIO_PROP_io_service_time,
82 BLKIO_PROP_io_wait_time,
83 BLKIO_PROP_io_merged,
84 BLKIO_PROP_io_queued,
85 BLKIO_PROP_avg_queue_size,
86 BLKIO_PROP_group_wait_time,
87 BLKIO_PROP_idle_time,
88 BLKIO_PROP_empty_time,
89 BLKIO_PROP_dequeue,
90};
91
4c9eefa1
VG
92/* cgroup files owned by throttle policy */
93enum blkcg_file_name_throtl {
94 BLKIO_THROTL_read_bps_device,
95 BLKIO_THROTL_write_bps_device,
96 BLKIO_THROTL_io_service_bytes,
97 BLKIO_THROTL_io_serviced,
98};
99
31e4c28d
VG
100struct blkio_cgroup {
101 struct cgroup_subsys_state css;
102 unsigned int weight;
103 spinlock_t lock;
104 struct hlist_head blkg_list;
34d0f179 105 struct list_head policy_list; /* list of blkio_policy_node */
31e4c28d
VG
106};
107
303a3acb
DS
108struct blkio_group_stats {
109 /* total disk time and nr sectors dispatched by this group */
110 uint64_t time;
111 uint64_t sectors;
cdc1184c 112 uint64_t stat_arr[BLKIO_STAT_QUEUED + 1][BLKIO_STAT_TOTAL];
303a3acb 113#ifdef CONFIG_DEBUG_BLK_CGROUP
cdc1184c
DS
114 /* Sum of number of IOs queued across all samples */
115 uint64_t avg_queue_size_sum;
116 /* Count of samples taken for average */
117 uint64_t avg_queue_size_samples;
303a3acb
DS
118 /* How many times this group has been removed from service tree */
119 unsigned long dequeue;
812df48d
DS
120
121 /* Total time spent waiting for it to be assigned a timeslice. */
122 uint64_t group_wait_time;
123 uint64_t start_group_wait_time;
124
125 /* Time spent idling for this blkio_group */
126 uint64_t idle_time;
127 uint64_t start_idle_time;
128 /*
129 * Total time when we have requests queued and do not contain the
130 * current active queue.
131 */
132 uint64_t empty_time;
133 uint64_t start_empty_time;
134 uint16_t flags;
303a3acb
DS
135#endif
136};
137
31e4c28d
VG
138struct blkio_group {
139 /* An rcu protected unique identifier for the group */
140 void *key;
141 struct hlist_node blkcg_node;
b1c35769 142 unsigned short blkcg_id;
2868ef7b
VG
143 /* Store cgroup path */
144 char path[128];
22084190 145 /* The device MKDEV(major, minor), this group has been created for */
84c124da 146 dev_t dev;
062a644d
VG
147 /* policy which owns this blk group */
148 enum blkio_policy_id plid;
22084190 149
303a3acb
DS
150 /* Need to serialize the stats in the case of reset/update */
151 spinlock_t stats_lock;
152 struct blkio_group_stats stats;
31e4c28d
VG
153};
154
34d0f179
GJ
155struct blkio_policy_node {
156 struct list_head node;
157 dev_t dev;
062a644d
VG
158 /* This node belongs to max bw policy or porportional weight policy */
159 enum blkio_policy_id plid;
160 /* cgroup file to which this rule belongs to */
161 int fileid;
4c9eefa1
VG
162
163 union {
164 unsigned int weight;
165 /*
166 * Rate read/write in terms of byptes per second
167 * Whether this rate represents read or write is determined
168 * by file type "fileid".
169 */
170 u64 bps;
171 } val;
34d0f179
GJ
172};
173
174extern unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
175 dev_t dev);
4c9eefa1
VG
176extern uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg,
177 dev_t dev);
178extern uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg,
179 dev_t dev);
34d0f179 180
3e252066
VG
181typedef void (blkio_unlink_group_fn) (void *key, struct blkio_group *blkg);
182typedef void (blkio_update_group_weight_fn) (struct blkio_group *blkg,
183 unsigned int weight);
4c9eefa1
VG
184typedef void (blkio_update_group_read_bps_fn) (struct blkio_group *blkg,
185 u64 read_bps);
186typedef void (blkio_update_group_write_bps_fn) (struct blkio_group *blkg,
187 u64 write_bps);
3e252066
VG
188
189struct blkio_policy_ops {
190 blkio_unlink_group_fn *blkio_unlink_group_fn;
191 blkio_update_group_weight_fn *blkio_update_group_weight_fn;
4c9eefa1
VG
192 blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn;
193 blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn;
3e252066
VG
194};
195
196struct blkio_policy_type {
197 struct list_head list;
198 struct blkio_policy_ops ops;
062a644d 199 enum blkio_policy_id plid;
3e252066
VG
200};
201
202/* Blkio controller policy registration */
203extern void blkio_policy_register(struct blkio_policy_type *);
204extern void blkio_policy_unregister(struct blkio_policy_type *);
205
afc24d49
VG
206static inline char *blkg_path(struct blkio_group *blkg)
207{
208 return blkg->path;
209}
210
2f5ea477
JA
211#else
212
213struct blkio_group {
214};
215
3e252066
VG
216struct blkio_policy_type {
217};
218
219static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { }
220static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
221
afc24d49
VG
222static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
223
2f5ea477
JA
224#endif
225
31e4c28d
VG
226#define BLKIO_WEIGHT_MIN 100
227#define BLKIO_WEIGHT_MAX 1000
228#define BLKIO_WEIGHT_DEFAULT 500
229
2868ef7b 230#ifdef CONFIG_DEBUG_BLK_CGROUP
a11cdaa7 231void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg);
9195291e 232void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
22084190 233 unsigned long dequeue);
812df48d
DS
234void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg);
235void blkiocg_update_idle_time_stats(struct blkio_group *blkg);
e5ff082e 236void blkiocg_set_start_empty_time(struct blkio_group *blkg);
812df48d
DS
237
238#define BLKG_FLAG_FNS(name) \
239static inline void blkio_mark_blkg_##name( \
240 struct blkio_group_stats *stats) \
241{ \
242 stats->flags |= (1 << BLKG_##name); \
243} \
244static inline void blkio_clear_blkg_##name( \
245 struct blkio_group_stats *stats) \
246{ \
247 stats->flags &= ~(1 << BLKG_##name); \
248} \
249static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \
250{ \
251 return (stats->flags & (1 << BLKG_##name)) != 0; \
252} \
253
254BLKG_FLAG_FNS(waiting)
255BLKG_FLAG_FNS(idling)
256BLKG_FLAG_FNS(empty)
257#undef BLKG_FLAG_FNS
2868ef7b 258#else
a11cdaa7 259static inline void blkiocg_update_avg_queue_size_stats(
cdc1184c 260 struct blkio_group *blkg) {}
9195291e
DS
261static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
262 unsigned long dequeue) {}
812df48d
DS
263static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
264{}
265static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg) {}
e5ff082e 266static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg) {}
2868ef7b
VG
267#endif
268
67523c48 269#if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
31e4c28d
VG
270extern struct blkio_cgroup blkio_root_cgroup;
271extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
272extern void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
062a644d
VG
273 struct blkio_group *blkg, void *key, dev_t dev,
274 enum blkio_policy_id plid);
31e4c28d
VG
275extern int blkiocg_del_blkio_group(struct blkio_group *blkg);
276extern struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg,
277 void *key);
303a3acb
DS
278void blkiocg_update_timeslice_used(struct blkio_group *blkg,
279 unsigned long time);
84c124da
DS
280void blkiocg_update_dispatch_stats(struct blkio_group *blkg, uint64_t bytes,
281 bool direction, bool sync);
282void blkiocg_update_completion_stats(struct blkio_group *blkg,
283 uint64_t start_time, uint64_t io_start_time, bool direction, bool sync);
812d4026
DS
284void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
285 bool sync);
a11cdaa7 286void blkiocg_update_io_add_stats(struct blkio_group *blkg,
cdc1184c 287 struct blkio_group *curr_blkg, bool direction, bool sync);
a11cdaa7 288void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
cdc1184c 289 bool direction, bool sync);
31e4c28d 290#else
2f5ea477 291struct cgroup;
31e4c28d
VG
292static inline struct blkio_cgroup *
293cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
294
295static inline void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
062a644d
VG
296 struct blkio_group *blkg, void *key, dev_t dev,
297 enum blkio_policy_id plid) {}
31e4c28d
VG
298
299static inline int
300blkiocg_del_blkio_group(struct blkio_group *blkg) { return 0; }
301
302static inline struct blkio_group *
303blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key) { return NULL; }
303a3acb 304static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg,
9a0785b0 305 unsigned long time) {}
84c124da
DS
306static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
307 uint64_t bytes, bool direction, bool sync) {}
308static inline void blkiocg_update_completion_stats(struct blkio_group *blkg,
309 uint64_t start_time, uint64_t io_start_time, bool direction,
310 bool sync) {}
812d4026
DS
311static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
312 bool direction, bool sync) {}
a11cdaa7 313static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg,
cdc1184c 314 struct blkio_group *curr_blkg, bool direction, bool sync) {}
a11cdaa7 315static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
cdc1184c 316 bool direction, bool sync) {}
31e4c28d
VG
317#endif
318#endif /* _BLK_CGROUP_H */