4 * Common Block IO controller cgroup interface
6 * Based on ideas and code from CFQ, CFS and BFQ:
7 * Copyright (C) 2003 Jens Axboe <axboe@kernel.dk>
9 * Copyright (C) 2008 Fabio Checconi <fabio@gandalf.sssup.it>
10 * Paolo Valente <paolo.valente@unimore.it>
12 * Copyright (C) 2009 Vivek Goyal <vgoyal@redhat.com>
13 * Nauman Rafique <nauman@google.com>
16 #include <linux/cgroup.h>
18 enum blkio_policy_id {
19 BLKIO_POLICY_PROP = 0, /* Proportional Bandwidth division */
20 BLKIO_POLICY_THROTL, /* Throttling */
23 /* Max limits for throttle policy */
24 #define THROTL_IOPS_MAX UINT_MAX
26 #if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
28 #ifndef CONFIG_BLK_CGROUP
29 /* When blk-cgroup is a module, its subsys_id isn't a compile-time constant */
30 extern struct cgroup_subsys blkio_subsys;
31 #define blkio_subsys_id blkio_subsys.subsys_id
35 /* Total time spent (in ns) between request dispatch to the driver and
36 * request completion for IOs doen by this cgroup. This may not be
37 * accurate when NCQ is turned on. */
38 BLKIO_STAT_SERVICE_TIME = 0,
39 /* Total bytes transferred */
40 BLKIO_STAT_SERVICE_BYTES,
41 /* Total IOs serviced, post merge */
43 /* Total time spent waiting in scheduler queue in ns */
45 /* Number of IOs merged */
47 /* Number of IOs queued up */
49 /* All the single valued stats go below this */
52 #ifdef CONFIG_DEBUG_BLK_CGROUP
53 /* Time not charged to this cgroup */
54 BLKIO_STAT_UNACCOUNTED_TIME,
55 BLKIO_STAT_AVG_QUEUE_SIZE,
57 BLKIO_STAT_EMPTY_TIME,
58 BLKIO_STAT_GROUP_WAIT_TIME,
71 /* blkg state flags */
72 enum blkg_state_flags {
78 /* cgroup files owned by proportional weight policy */
79 enum blkcg_file_name_prop {
80 BLKIO_PROP_weight = 1,
81 BLKIO_PROP_weight_device,
82 BLKIO_PROP_io_service_bytes,
83 BLKIO_PROP_io_serviced,
86 BLKIO_PROP_unaccounted_time,
87 BLKIO_PROP_io_service_time,
88 BLKIO_PROP_io_wait_time,
91 BLKIO_PROP_avg_queue_size,
92 BLKIO_PROP_group_wait_time,
94 BLKIO_PROP_empty_time,
98 /* cgroup files owned by throttle policy */
99 enum blkcg_file_name_throtl {
100 BLKIO_THROTL_read_bps_device,
101 BLKIO_THROTL_write_bps_device,
102 BLKIO_THROTL_read_iops_device,
103 BLKIO_THROTL_write_iops_device,
104 BLKIO_THROTL_io_service_bytes,
105 BLKIO_THROTL_io_serviced,
108 struct blkio_cgroup {
109 struct cgroup_subsys_state css;
112 struct hlist_head blkg_list;
113 struct list_head policy_list; /* list of blkio_policy_node */
116 struct blkio_group_stats {
117 /* total disk time and nr sectors dispatched by this group */
120 uint64_t stat_arr[BLKIO_STAT_QUEUED + 1][BLKIO_STAT_TOTAL];
121 #ifdef CONFIG_DEBUG_BLK_CGROUP
122 /* Time not charged to this cgroup */
123 uint64_t unaccounted_time;
125 /* Sum of number of IOs queued across all samples */
126 uint64_t avg_queue_size_sum;
127 /* Count of samples taken for average */
128 uint64_t avg_queue_size_samples;
129 /* How many times this group has been removed from service tree */
130 unsigned long dequeue;
132 /* Total time spent waiting for it to be assigned a timeslice. */
133 uint64_t group_wait_time;
134 uint64_t start_group_wait_time;
136 /* Time spent idling for this blkio_group */
138 uint64_t start_idle_time;
140 * Total time when we have requests queued and do not contain the
141 * current active queue.
144 uint64_t start_empty_time;
150 /* An rcu protected unique identifier for the group */
152 struct hlist_node blkcg_node;
153 unsigned short blkcg_id;
154 /* Store cgroup path */
156 /* The device MKDEV(major, minor), this group has been created for */
158 /* policy which owns this blk group */
159 enum blkio_policy_id plid;
161 /* Need to serialize the stats in the case of reset/update */
162 spinlock_t stats_lock;
163 struct blkio_group_stats stats;
166 struct blkio_policy_node {
167 struct list_head node;
169 /* This node belongs to max bw policy or porportional weight policy */
170 enum blkio_policy_id plid;
171 /* cgroup file to which this rule belongs to */
177 * Rate read/write in terms of byptes per second
178 * Whether this rate represents read or write is determined
179 * by file type "fileid".
186 extern unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
188 extern uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg,
190 extern uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg,
192 extern unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg,
194 extern unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg,
197 typedef void (blkio_unlink_group_fn) (void *key, struct blkio_group *blkg);
199 typedef void (blkio_update_group_weight_fn) (void *key,
200 struct blkio_group *blkg, unsigned int weight);
201 typedef void (blkio_update_group_read_bps_fn) (void * key,
202 struct blkio_group *blkg, u64 read_bps);
203 typedef void (blkio_update_group_write_bps_fn) (void *key,
204 struct blkio_group *blkg, u64 write_bps);
205 typedef void (blkio_update_group_read_iops_fn) (void *key,
206 struct blkio_group *blkg, unsigned int read_iops);
207 typedef void (blkio_update_group_write_iops_fn) (void *key,
208 struct blkio_group *blkg, unsigned int write_iops);
210 struct blkio_policy_ops {
211 blkio_unlink_group_fn *blkio_unlink_group_fn;
212 blkio_update_group_weight_fn *blkio_update_group_weight_fn;
213 blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn;
214 blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn;
215 blkio_update_group_read_iops_fn *blkio_update_group_read_iops_fn;
216 blkio_update_group_write_iops_fn *blkio_update_group_write_iops_fn;
219 struct blkio_policy_type {
220 struct list_head list;
221 struct blkio_policy_ops ops;
222 enum blkio_policy_id plid;
225 /* Blkio controller policy registration */
226 extern void blkio_policy_register(struct blkio_policy_type *);
227 extern void blkio_policy_unregister(struct blkio_policy_type *);
229 static inline char *blkg_path(struct blkio_group *blkg)
239 struct blkio_policy_type {
242 static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { }
243 static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
245 static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
249 #define BLKIO_WEIGHT_MIN 10
250 #define BLKIO_WEIGHT_MAX 1000
251 #define BLKIO_WEIGHT_DEFAULT 500
253 #ifdef CONFIG_DEBUG_BLK_CGROUP
254 void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg);
255 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
256 unsigned long dequeue);
257 void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg);
258 void blkiocg_update_idle_time_stats(struct blkio_group *blkg);
259 void blkiocg_set_start_empty_time(struct blkio_group *blkg);
261 #define BLKG_FLAG_FNS(name) \
262 static inline void blkio_mark_blkg_##name( \
263 struct blkio_group_stats *stats) \
265 stats->flags |= (1 << BLKG_##name); \
267 static inline void blkio_clear_blkg_##name( \
268 struct blkio_group_stats *stats) \
270 stats->flags &= ~(1 << BLKG_##name); \
272 static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \
274 return (stats->flags & (1 << BLKG_##name)) != 0; \
277 BLKG_FLAG_FNS(waiting)
278 BLKG_FLAG_FNS(idling)
282 static inline void blkiocg_update_avg_queue_size_stats(
283 struct blkio_group *blkg) {}
284 static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
285 unsigned long dequeue) {}
286 static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
288 static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg) {}
289 static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg) {}
292 #if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
293 extern struct blkio_cgroup blkio_root_cgroup;
294 extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
295 extern struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk);
296 extern void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
297 struct blkio_group *blkg, void *key, dev_t dev,
298 enum blkio_policy_id plid);
299 extern int blkiocg_del_blkio_group(struct blkio_group *blkg);
300 extern struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg,
302 void blkiocg_update_timeslice_used(struct blkio_group *blkg,
304 unsigned long unaccounted_time);
305 void blkiocg_update_dispatch_stats(struct blkio_group *blkg, uint64_t bytes,
306 bool direction, bool sync);
307 void blkiocg_update_completion_stats(struct blkio_group *blkg,
308 uint64_t start_time, uint64_t io_start_time, bool direction, bool sync);
309 void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
311 void blkiocg_update_io_add_stats(struct blkio_group *blkg,
312 struct blkio_group *curr_blkg, bool direction, bool sync);
313 void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
314 bool direction, bool sync);
317 static inline struct blkio_cgroup *
318 cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
319 static inline struct blkio_cgroup *
320 task_blkio_cgroup(struct task_struct *tsk) { return NULL; }
322 static inline void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
323 struct blkio_group *blkg, void *key, dev_t dev,
324 enum blkio_policy_id plid) {}
327 blkiocg_del_blkio_group(struct blkio_group *blkg) { return 0; }
329 static inline struct blkio_group *
330 blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key) { return NULL; }
331 static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg,
333 unsigned long unaccounted_time)
335 static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
336 uint64_t bytes, bool direction, bool sync) {}
337 static inline void blkiocg_update_completion_stats(struct blkio_group *blkg,
338 uint64_t start_time, uint64_t io_start_time, bool direction,
340 static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
341 bool direction, bool sync) {}
342 static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg,
343 struct blkio_group *curr_blkg, bool direction, bool sync) {}
344 static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
345 bool direction, bool sync) {}
347 #endif /* _BLK_CGROUP_H */