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 list_head q_node
[BLKIO_NR_POLICIES
];
182 struct hlist_node blkcg_node
;
183 struct blkio_cgroup
*blkcg
;
184 /* Store cgroup path */
186 /* policy which owns this blk group */
187 enum blkio_policy_id plid
;
188 /* reference count */
191 /* Need to serialize the stats in the case of reset/update */
192 spinlock_t stats_lock
;
193 struct blkg_policy_data
*pd
[BLKIO_NR_POLICIES
];
195 struct rcu_head rcu_head
;
198 typedef void (blkio_init_group_fn
)(struct blkio_group
*blkg
);
199 typedef void (blkio_link_group_fn
)(struct request_queue
*q
,
200 struct blkio_group
*blkg
);
201 typedef void (blkio_unlink_group_fn
)(struct request_queue
*q
,
202 struct blkio_group
*blkg
);
203 typedef bool (blkio_clear_queue_fn
)(struct request_queue
*q
);
204 typedef void (blkio_update_group_weight_fn
)(struct request_queue
*q
,
205 struct blkio_group
*blkg
, unsigned int weight
);
206 typedef void (blkio_update_group_read_bps_fn
)(struct request_queue
*q
,
207 struct blkio_group
*blkg
, u64 read_bps
);
208 typedef void (blkio_update_group_write_bps_fn
)(struct request_queue
*q
,
209 struct blkio_group
*blkg
, u64 write_bps
);
210 typedef void (blkio_update_group_read_iops_fn
)(struct request_queue
*q
,
211 struct blkio_group
*blkg
, unsigned int read_iops
);
212 typedef void (blkio_update_group_write_iops_fn
)(struct request_queue
*q
,
213 struct blkio_group
*blkg
, unsigned int write_iops
);
215 struct blkio_policy_ops
{
216 blkio_init_group_fn
*blkio_init_group_fn
;
217 blkio_link_group_fn
*blkio_link_group_fn
;
218 blkio_unlink_group_fn
*blkio_unlink_group_fn
;
219 blkio_clear_queue_fn
*blkio_clear_queue_fn
;
220 blkio_update_group_weight_fn
*blkio_update_group_weight_fn
;
221 blkio_update_group_read_bps_fn
*blkio_update_group_read_bps_fn
;
222 blkio_update_group_write_bps_fn
*blkio_update_group_write_bps_fn
;
223 blkio_update_group_read_iops_fn
*blkio_update_group_read_iops_fn
;
224 blkio_update_group_write_iops_fn
*blkio_update_group_write_iops_fn
;
227 struct blkio_policy_type
{
228 struct list_head list
;
229 struct blkio_policy_ops ops
;
230 enum blkio_policy_id plid
;
231 size_t pdata_size
; /* policy specific private data size */
234 extern int blkcg_init_queue(struct request_queue
*q
);
235 extern void blkcg_drain_queue(struct request_queue
*q
);
236 extern void blkcg_exit_queue(struct request_queue
*q
);
238 /* Blkio controller policy registration */
239 extern void blkio_policy_register(struct blkio_policy_type
*);
240 extern void blkio_policy_unregister(struct blkio_policy_type
*);
241 extern void blkg_destroy_all(struct request_queue
*q
);
244 * blkg_to_pdata - get policy private data
245 * @blkg: blkg of interest
246 * @pol: policy of interest
248 * Return pointer to private data associated with the @blkg-@pol pair.
250 static inline void *blkg_to_pdata(struct blkio_group
*blkg
,
251 struct blkio_policy_type
*pol
)
253 return blkg
? blkg
->pd
[pol
->plid
]->pdata
: NULL
;
257 * pdata_to_blkg - get blkg associated with policy private data
258 * @pdata: policy private data of interest
259 * @pol: policy @pdata is for
261 * @pdata is policy private data for @pol. Determine the blkg it's
264 static inline struct blkio_group
*pdata_to_blkg(void *pdata
,
265 struct blkio_policy_type
*pol
)
268 struct blkg_policy_data
*pd
=
269 container_of(pdata
, struct blkg_policy_data
, pdata
);
275 static inline char *blkg_path(struct blkio_group
*blkg
)
281 * blkg_get - get a blkg reference
284 * The caller should be holding queue_lock and an existing reference.
286 static inline void blkg_get(struct blkio_group
*blkg
)
288 lockdep_assert_held(blkg
->q
->queue_lock
);
289 WARN_ON_ONCE(!blkg
->refcnt
);
293 void __blkg_release(struct blkio_group
*blkg
);
296 * blkg_put - put a blkg reference
299 * The caller should be holding queue_lock.
301 static inline void blkg_put(struct blkio_group
*blkg
)
303 lockdep_assert_held(blkg
->q
->queue_lock
);
304 WARN_ON_ONCE(blkg
->refcnt
<= 0);
306 __blkg_release(blkg
);
314 struct blkio_policy_type
{
317 static inline int blkcg_init_queue(struct request_queue
*q
) { return 0; }
318 static inline void blkcg_drain_queue(struct request_queue
*q
) { }
319 static inline void blkcg_exit_queue(struct request_queue
*q
) { }
320 static inline void blkio_policy_register(struct blkio_policy_type
*blkiop
) { }
321 static inline void blkio_policy_unregister(struct blkio_policy_type
*blkiop
) { }
322 static inline void blkg_destroy_all(struct request_queue
*q
) { }
324 static inline void *blkg_to_pdata(struct blkio_group
*blkg
,
325 struct blkio_policy_type
*pol
) { return NULL
; }
326 static inline struct blkio_group
*pdata_to_blkg(void *pdata
,
327 struct blkio_policy_type
*pol
) { return NULL
; }
328 static inline char *blkg_path(struct blkio_group
*blkg
) { return NULL
; }
329 static inline void blkg_get(struct blkio_group
*blkg
) { }
330 static inline void blkg_put(struct blkio_group
*blkg
) { }
334 #define BLKIO_WEIGHT_MIN 10
335 #define BLKIO_WEIGHT_MAX 1000
336 #define BLKIO_WEIGHT_DEFAULT 500
338 #ifdef CONFIG_DEBUG_BLK_CGROUP
339 void blkiocg_update_avg_queue_size_stats(struct blkio_group
*blkg
,
340 struct blkio_policy_type
*pol
);
341 void blkiocg_update_dequeue_stats(struct blkio_group
*blkg
,
342 struct blkio_policy_type
*pol
,
343 unsigned long dequeue
);
344 void blkiocg_update_set_idle_time_stats(struct blkio_group
*blkg
,
345 struct blkio_policy_type
*pol
);
346 void blkiocg_update_idle_time_stats(struct blkio_group
*blkg
,
347 struct blkio_policy_type
*pol
);
348 void blkiocg_set_start_empty_time(struct blkio_group
*blkg
,
349 struct blkio_policy_type
*pol
);
351 #define BLKG_FLAG_FNS(name) \
352 static inline void blkio_mark_blkg_##name( \
353 struct blkio_group_stats *stats) \
355 stats->flags |= (1 << BLKG_##name); \
357 static inline void blkio_clear_blkg_##name( \
358 struct blkio_group_stats *stats) \
360 stats->flags &= ~(1 << BLKG_##name); \
362 static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \
364 return (stats->flags & (1 << BLKG_##name)) != 0; \
367 BLKG_FLAG_FNS(waiting)
368 BLKG_FLAG_FNS(idling
)
372 static inline void blkiocg_update_avg_queue_size_stats(struct blkio_group
*blkg
,
373 struct blkio_policy_type
*pol
) { }
374 static inline void blkiocg_update_dequeue_stats(struct blkio_group
*blkg
,
375 struct blkio_policy_type
*pol
, unsigned long dequeue
) { }
376 static inline void blkiocg_update_set_idle_time_stats(struct blkio_group
*blkg
,
377 struct blkio_policy_type
*pol
) { }
378 static inline void blkiocg_update_idle_time_stats(struct blkio_group
*blkg
,
379 struct blkio_policy_type
*pol
) { }
380 static inline void blkiocg_set_start_empty_time(struct blkio_group
*blkg
,
381 struct blkio_policy_type
*pol
) { }
384 #ifdef CONFIG_BLK_CGROUP
385 extern struct blkio_cgroup blkio_root_cgroup
;
386 extern struct blkio_cgroup
*cgroup_to_blkio_cgroup(struct cgroup
*cgroup
);
387 extern struct blkio_cgroup
*task_blkio_cgroup(struct task_struct
*tsk
);
388 extern int blkiocg_del_blkio_group(struct blkio_group
*blkg
);
389 extern struct blkio_group
*blkg_lookup(struct blkio_cgroup
*blkcg
,
390 struct request_queue
*q
,
391 enum blkio_policy_id plid
);
392 struct blkio_group
*blkg_lookup_create(struct blkio_cgroup
*blkcg
,
393 struct request_queue
*q
,
394 enum blkio_policy_id plid
,
396 void blkiocg_update_timeslice_used(struct blkio_group
*blkg
,
397 struct blkio_policy_type
*pol
,
399 unsigned long unaccounted_time
);
400 void blkiocg_update_dispatch_stats(struct blkio_group
*blkg
,
401 struct blkio_policy_type
*pol
,
402 uint64_t bytes
, bool direction
, bool sync
);
403 void blkiocg_update_completion_stats(struct blkio_group
*blkg
,
404 struct blkio_policy_type
*pol
,
406 uint64_t io_start_time
, bool direction
,
408 void blkiocg_update_io_merged_stats(struct blkio_group
*blkg
,
409 struct blkio_policy_type
*pol
,
410 bool direction
, bool sync
);
411 void blkiocg_update_io_add_stats(struct blkio_group
*blkg
,
412 struct blkio_policy_type
*pol
,
413 struct blkio_group
*curr_blkg
, bool direction
,
415 void blkiocg_update_io_remove_stats(struct blkio_group
*blkg
,
416 struct blkio_policy_type
*pol
,
417 bool direction
, bool sync
);
420 static inline struct blkio_cgroup
*
421 cgroup_to_blkio_cgroup(struct cgroup
*cgroup
) { return NULL
; }
422 static inline struct blkio_cgroup
*
423 task_blkio_cgroup(struct task_struct
*tsk
) { return NULL
; }
426 blkiocg_del_blkio_group(struct blkio_group
*blkg
) { return 0; }
428 static inline struct blkio_group
*blkg_lookup(struct blkio_cgroup
*blkcg
,
429 void *key
) { return NULL
; }
430 static inline void blkiocg_update_timeslice_used(struct blkio_group
*blkg
,
431 struct blkio_policy_type
*pol
, unsigned long time
,
432 unsigned long unaccounted_time
) { }
433 static inline void blkiocg_update_dispatch_stats(struct blkio_group
*blkg
,
434 struct blkio_policy_type
*pol
, uint64_t bytes
,
435 bool direction
, bool sync
) { }
436 static inline void blkiocg_update_completion_stats(struct blkio_group
*blkg
,
437 struct blkio_policy_type
*pol
, uint64_t start_time
,
438 uint64_t io_start_time
, bool direction
, bool sync
) { }
439 static inline void blkiocg_update_io_merged_stats(struct blkio_group
*blkg
,
440 struct blkio_policy_type
*pol
, bool direction
,
442 static inline void blkiocg_update_io_add_stats(struct blkio_group
*blkg
,
443 struct blkio_policy_type
*pol
,
444 struct blkio_group
*curr_blkg
, bool direction
,
446 static inline void blkiocg_update_io_remove_stats(struct blkio_group
*blkg
,
447 struct blkio_policy_type
*pol
, bool direction
,
450 #endif /* _BLK_CGROUP_H */