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>
17 #include <linux/u64_stats_sync.h>
19 enum blkio_policy_id {
20 BLKIO_POLICY_PROP = 0, /* Proportional Bandwidth division */
21 BLKIO_POLICY_THROTL, /* Throttling */
26 /* Max limits for throttle policy */
27 #define THROTL_IOPS_MAX UINT_MAX
29 #ifdef CONFIG_BLK_CGROUP
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 time spent waiting in scheduler queue in ns */
38 /* Number of IOs queued up */
40 /* All the single valued stats go below this */
42 #ifdef CONFIG_DEBUG_BLK_CGROUP
43 /* Time not charged to this cgroup */
44 BLKIO_STAT_UNACCOUNTED_TIME,
45 BLKIO_STAT_AVG_QUEUE_SIZE,
47 BLKIO_STAT_EMPTY_TIME,
48 BLKIO_STAT_GROUP_WAIT_TIME,
55 BLKIO_STAT_CPU_SECTORS,
56 /* Total bytes transferred */
57 BLKIO_STAT_CPU_SERVICE_BYTES,
58 /* Total IOs serviced, post merge */
59 BLKIO_STAT_CPU_SERVICED,
60 /* Number of IOs merged */
61 BLKIO_STAT_CPU_MERGED,
73 /* blkg state flags */
74 enum blkg_state_flags {
80 /* cgroup files owned by proportional weight policy */
81 enum blkcg_file_name_prop {
82 BLKIO_PROP_weight = 1,
83 BLKIO_PROP_weight_device,
84 BLKIO_PROP_io_service_bytes,
85 BLKIO_PROP_io_serviced,
88 BLKIO_PROP_unaccounted_time,
89 BLKIO_PROP_io_service_time,
90 BLKIO_PROP_io_wait_time,
93 BLKIO_PROP_avg_queue_size,
94 BLKIO_PROP_group_wait_time,
96 BLKIO_PROP_empty_time,
100 /* cgroup files owned by throttle policy */
101 enum blkcg_file_name_throtl {
102 BLKIO_THROTL_read_bps_device,
103 BLKIO_THROTL_write_bps_device,
104 BLKIO_THROTL_read_iops_device,
105 BLKIO_THROTL_write_iops_device,
106 BLKIO_THROTL_io_service_bytes,
107 BLKIO_THROTL_io_serviced,
110 struct blkio_cgroup {
111 struct cgroup_subsys_state css;
114 struct hlist_head blkg_list;
117 struct blkio_group_stats {
118 /* 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;
149 /* Per cpu blkio group stats */
150 struct blkio_group_stats_cpu {
152 uint64_t stat_arr_cpu[BLKIO_STAT_CPU_NR][BLKIO_STAT_TOTAL];
153 struct u64_stats_sync syncp;
156 struct blkio_group_conf {
158 unsigned int iops[2];
162 /* per-blkg per-policy data */
163 struct blkg_policy_data {
164 /* the blkg this per-policy data belongs to */
165 struct blkio_group *blkg;
168 struct blkio_group_conf conf;
170 struct blkio_group_stats stats;
171 /* Per cpu stats pointer */
172 struct blkio_group_stats_cpu __percpu *stats_cpu;
174 /* pol->pdata_size bytes of private data used by policy impl */
175 char pdata[] __aligned(__alignof__(unsigned long long));
179 /* Pointer to the associated request_queue, RCU protected */
180 struct request_queue __rcu *q;
181 struct hlist_node blkcg_node;
182 struct blkio_cgroup *blkcg;
183 /* Store cgroup path */
185 /* policy which owns this blk group */
186 enum blkio_policy_id plid;
187 /* reference count */
190 /* Need to serialize the stats in the case of reset/update */
191 spinlock_t stats_lock;
192 struct blkg_policy_data *pd[BLKIO_NR_POLICIES];
194 struct rcu_head rcu_head;
197 typedef void (blkio_init_group_fn)(struct blkio_group *blkg);
198 typedef void (blkio_link_group_fn)(struct request_queue *q,
199 struct blkio_group *blkg);
200 typedef void (blkio_unlink_group_fn)(struct request_queue *q,
201 struct blkio_group *blkg);
202 typedef bool (blkio_clear_queue_fn)(struct request_queue *q);
203 typedef void (blkio_update_group_weight_fn)(struct request_queue *q,
204 struct blkio_group *blkg, unsigned int weight);
205 typedef void (blkio_update_group_read_bps_fn)(struct request_queue *q,
206 struct blkio_group *blkg, u64 read_bps);
207 typedef void (blkio_update_group_write_bps_fn)(struct request_queue *q,
208 struct blkio_group *blkg, u64 write_bps);
209 typedef void (blkio_update_group_read_iops_fn)(struct request_queue *q,
210 struct blkio_group *blkg, unsigned int read_iops);
211 typedef void (blkio_update_group_write_iops_fn)(struct request_queue *q,
212 struct blkio_group *blkg, unsigned int write_iops);
214 struct blkio_policy_ops {
215 blkio_init_group_fn *blkio_init_group_fn;
216 blkio_link_group_fn *blkio_link_group_fn;
217 blkio_unlink_group_fn *blkio_unlink_group_fn;
218 blkio_clear_queue_fn *blkio_clear_queue_fn;
219 blkio_update_group_weight_fn *blkio_update_group_weight_fn;
220 blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn;
221 blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn;
222 blkio_update_group_read_iops_fn *blkio_update_group_read_iops_fn;
223 blkio_update_group_write_iops_fn *blkio_update_group_write_iops_fn;
226 struct blkio_policy_type {
227 struct list_head list;
228 struct blkio_policy_ops ops;
229 enum blkio_policy_id plid;
230 size_t pdata_size; /* policy specific private data size */
233 extern int blkcg_init_queue(struct request_queue *q);
234 extern void blkcg_drain_queue(struct request_queue *q);
235 extern void blkcg_exit_queue(struct request_queue *q);
237 /* Blkio controller policy registration */
238 extern void blkio_policy_register(struct blkio_policy_type *);
239 extern void blkio_policy_unregister(struct blkio_policy_type *);
240 extern void blkg_destroy_all(struct request_queue *q);
243 * blkg_to_pdata - get policy private data
244 * @blkg: blkg of interest
245 * @pol: policy of interest
247 * Return pointer to private data associated with the @blkg-@pol pair.
249 static inline void *blkg_to_pdata(struct blkio_group *blkg,
250 struct blkio_policy_type *pol)
252 return blkg ? blkg->pd[pol->plid]->pdata : NULL;
256 * pdata_to_blkg - get blkg associated with policy private data
257 * @pdata: policy private data of interest
258 * @pol: policy @pdata is for
260 * @pdata is policy private data for @pol. Determine the blkg it's
263 static inline struct blkio_group *pdata_to_blkg(void *pdata,
264 struct blkio_policy_type *pol)
267 struct blkg_policy_data *pd =
268 container_of(pdata, struct blkg_policy_data, pdata);
274 static inline char *blkg_path(struct blkio_group *blkg)
280 * blkg_get - get a blkg reference
283 * The caller should be holding queue_lock and an existing reference.
285 static inline void blkg_get(struct blkio_group *blkg)
287 lockdep_assert_held(blkg->q->queue_lock);
288 WARN_ON_ONCE(!blkg->refcnt);
292 void __blkg_release(struct blkio_group *blkg);
295 * blkg_put - put a blkg reference
298 * The caller should be holding queue_lock.
300 static inline void blkg_put(struct blkio_group *blkg)
302 lockdep_assert_held(blkg->q->queue_lock);
303 WARN_ON_ONCE(blkg->refcnt <= 0);
305 __blkg_release(blkg);
313 struct blkio_policy_type {
316 static inline int blkcg_init_queue(struct request_queue *q) { return 0; }
317 static inline void blkcg_drain_queue(struct request_queue *q) { }
318 static inline void blkcg_exit_queue(struct request_queue *q) { }
319 static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { }
320 static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
321 static inline void blkg_destroy_all(struct request_queue *q) { }
323 static inline void *blkg_to_pdata(struct blkio_group *blkg,
324 struct blkio_policy_type *pol) { return NULL; }
325 static inline struct blkio_group *pdata_to_blkg(void *pdata,
326 struct blkio_policy_type *pol) { return NULL; }
327 static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
328 static inline void blkg_get(struct blkio_group *blkg) { }
329 static inline void blkg_put(struct blkio_group *blkg) { }
333 #define BLKIO_WEIGHT_MIN 10
334 #define BLKIO_WEIGHT_MAX 1000
335 #define BLKIO_WEIGHT_DEFAULT 500
337 #ifdef CONFIG_DEBUG_BLK_CGROUP
338 void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg,
339 struct blkio_policy_type *pol);
340 void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
341 struct blkio_policy_type *pol,
342 unsigned long dequeue);
343 void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg,
344 struct blkio_policy_type *pol);
345 void blkiocg_update_idle_time_stats(struct blkio_group *blkg,
346 struct blkio_policy_type *pol);
347 void blkiocg_set_start_empty_time(struct blkio_group *blkg,
348 struct blkio_policy_type *pol);
350 #define BLKG_FLAG_FNS(name) \
351 static inline void blkio_mark_blkg_##name( \
352 struct blkio_group_stats *stats) \
354 stats->flags |= (1 << BLKG_##name); \
356 static inline void blkio_clear_blkg_##name( \
357 struct blkio_group_stats *stats) \
359 stats->flags &= ~(1 << BLKG_##name); \
361 static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \
363 return (stats->flags & (1 << BLKG_##name)) != 0; \
366 BLKG_FLAG_FNS(waiting)
367 BLKG_FLAG_FNS(idling)
371 static inline void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg,
372 struct blkio_policy_type *pol) { }
373 static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
374 struct blkio_policy_type *pol, unsigned long dequeue) { }
375 static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg,
376 struct blkio_policy_type *pol) { }
377 static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg,
378 struct blkio_policy_type *pol) { }
379 static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg,
380 struct blkio_policy_type *pol) { }
383 #ifdef CONFIG_BLK_CGROUP
384 extern struct blkio_cgroup blkio_root_cgroup;
385 extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
386 extern struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk);
387 extern int blkiocg_del_blkio_group(struct blkio_group *blkg);
388 extern struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
389 struct request_queue *q,
390 enum blkio_policy_id plid);
391 struct blkio_group *blkg_lookup_create(struct blkio_cgroup *blkcg,
392 struct request_queue *q,
393 enum blkio_policy_id plid,
395 void blkiocg_update_timeslice_used(struct blkio_group *blkg,
396 struct blkio_policy_type *pol,
398 unsigned long unaccounted_time);
399 void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
400 struct blkio_policy_type *pol,
401 uint64_t bytes, bool direction, bool sync);
402 void blkiocg_update_completion_stats(struct blkio_group *blkg,
403 struct blkio_policy_type *pol,
405 uint64_t io_start_time, bool direction,
407 void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
408 struct blkio_policy_type *pol,
409 bool direction, bool sync);
410 void blkiocg_update_io_add_stats(struct blkio_group *blkg,
411 struct blkio_policy_type *pol,
412 struct blkio_group *curr_blkg, bool direction,
414 void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
415 struct blkio_policy_type *pol,
416 bool direction, bool sync);
419 static inline struct blkio_cgroup *
420 cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
421 static inline struct blkio_cgroup *
422 task_blkio_cgroup(struct task_struct *tsk) { return NULL; }
425 blkiocg_del_blkio_group(struct blkio_group *blkg) { return 0; }
427 static inline struct blkio_group *blkg_lookup(struct blkio_cgroup *blkcg,
428 void *key) { return NULL; }
429 static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg,
430 struct blkio_policy_type *pol, unsigned long time,
431 unsigned long unaccounted_time) { }
432 static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
433 struct blkio_policy_type *pol, uint64_t bytes,
434 bool direction, bool sync) { }
435 static inline void blkiocg_update_completion_stats(struct blkio_group *blkg,
436 struct blkio_policy_type *pol, uint64_t start_time,
437 uint64_t io_start_time, bool direction, bool sync) { }
438 static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
439 struct blkio_policy_type *pol, bool direction,
441 static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg,
442 struct blkio_policy_type *pol,
443 struct blkio_group *curr_blkg, bool direction,
445 static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
446 struct blkio_policy_type *pol, bool direction,
449 #endif /* _BLK_CGROUP_H */