blk-cgroup.h 11 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365
  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. #include <linux/cgroup.h>
  16. #include <linux/u64_stats_sync.h>
  17. enum blkio_policy_id {
  18. BLKIO_POLICY_PROP = 0, /* Proportional Bandwidth division */
  19. BLKIO_POLICY_THROTL, /* Throttling */
  20. };
  21. /* Max limits for throttle policy */
  22. #define THROTL_IOPS_MAX UINT_MAX
  23. #if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
  24. #ifndef CONFIG_BLK_CGROUP
  25. /* When blk-cgroup is a module, its subsys_id isn't a compile-time constant */
  26. extern struct cgroup_subsys blkio_subsys;
  27. #define blkio_subsys_id blkio_subsys.subsys_id
  28. #endif
  29. enum stat_type {
  30. /* Total time spent (in ns) between request dispatch to the driver and
  31. * request completion for IOs doen by this cgroup. This may not be
  32. * accurate when NCQ is turned on. */
  33. BLKIO_STAT_SERVICE_TIME = 0,
  34. /* Total time spent waiting in scheduler queue in ns */
  35. BLKIO_STAT_WAIT_TIME,
  36. /* Number of IOs queued up */
  37. BLKIO_STAT_QUEUED,
  38. /* All the single valued stats go below this */
  39. BLKIO_STAT_TIME,
  40. #ifdef CONFIG_DEBUG_BLK_CGROUP
  41. /* Time not charged to this cgroup */
  42. BLKIO_STAT_UNACCOUNTED_TIME,
  43. BLKIO_STAT_AVG_QUEUE_SIZE,
  44. BLKIO_STAT_IDLE_TIME,
  45. BLKIO_STAT_EMPTY_TIME,
  46. BLKIO_STAT_GROUP_WAIT_TIME,
  47. BLKIO_STAT_DEQUEUE
  48. #endif
  49. };
  50. /* Per cpu stats */
  51. enum stat_type_cpu {
  52. BLKIO_STAT_CPU_SECTORS,
  53. /* Total bytes transferred */
  54. BLKIO_STAT_CPU_SERVICE_BYTES,
  55. /* Total IOs serviced, post merge */
  56. BLKIO_STAT_CPU_SERVICED,
  57. /* Number of IOs merged */
  58. BLKIO_STAT_CPU_MERGED,
  59. BLKIO_STAT_CPU_NR
  60. };
  61. enum stat_sub_type {
  62. BLKIO_STAT_READ = 0,
  63. BLKIO_STAT_WRITE,
  64. BLKIO_STAT_SYNC,
  65. BLKIO_STAT_ASYNC,
  66. BLKIO_STAT_TOTAL
  67. };
  68. /* blkg state flags */
  69. enum blkg_state_flags {
  70. BLKG_waiting = 0,
  71. BLKG_idling,
  72. BLKG_empty,
  73. };
  74. /* cgroup files owned by proportional weight policy */
  75. enum blkcg_file_name_prop {
  76. BLKIO_PROP_weight = 1,
  77. BLKIO_PROP_weight_device,
  78. BLKIO_PROP_io_service_bytes,
  79. BLKIO_PROP_io_serviced,
  80. BLKIO_PROP_time,
  81. BLKIO_PROP_sectors,
  82. BLKIO_PROP_unaccounted_time,
  83. BLKIO_PROP_io_service_time,
  84. BLKIO_PROP_io_wait_time,
  85. BLKIO_PROP_io_merged,
  86. BLKIO_PROP_io_queued,
  87. BLKIO_PROP_avg_queue_size,
  88. BLKIO_PROP_group_wait_time,
  89. BLKIO_PROP_idle_time,
  90. BLKIO_PROP_empty_time,
  91. BLKIO_PROP_dequeue,
  92. };
  93. /* cgroup files owned by throttle policy */
  94. enum blkcg_file_name_throtl {
  95. BLKIO_THROTL_read_bps_device,
  96. BLKIO_THROTL_write_bps_device,
  97. BLKIO_THROTL_read_iops_device,
  98. BLKIO_THROTL_write_iops_device,
  99. BLKIO_THROTL_io_service_bytes,
  100. BLKIO_THROTL_io_serviced,
  101. };
  102. struct blkio_cgroup {
  103. struct cgroup_subsys_state css;
  104. unsigned int weight;
  105. spinlock_t lock;
  106. struct hlist_head blkg_list;
  107. struct list_head policy_list; /* list of blkio_policy_node */
  108. };
  109. struct blkio_group_stats {
  110. /* total disk time and nr sectors dispatched by this group */
  111. uint64_t time;
  112. uint64_t stat_arr[BLKIO_STAT_QUEUED + 1][BLKIO_STAT_TOTAL];
  113. #ifdef CONFIG_DEBUG_BLK_CGROUP
  114. /* Time not charged to this cgroup */
  115. uint64_t unaccounted_time;
  116. /* Sum of number of IOs queued across all samples */
  117. uint64_t avg_queue_size_sum;
  118. /* Count of samples taken for average */
  119. uint64_t avg_queue_size_samples;
  120. /* How many times this group has been removed from service tree */
  121. unsigned long dequeue;
  122. /* Total time spent waiting for it to be assigned a timeslice. */
  123. uint64_t group_wait_time;
  124. uint64_t start_group_wait_time;
  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;
  135. #endif
  136. };
  137. /* Per cpu blkio group stats */
  138. struct blkio_group_stats_cpu {
  139. uint64_t sectors;
  140. uint64_t stat_arr_cpu[BLKIO_STAT_CPU_NR][BLKIO_STAT_TOTAL];
  141. struct u64_stats_sync syncp;
  142. };
  143. struct blkio_group {
  144. /* An rcu protected unique identifier for the group */
  145. void *key;
  146. struct hlist_node blkcg_node;
  147. unsigned short blkcg_id;
  148. /* Store cgroup path */
  149. char path[128];
  150. /* The device MKDEV(major, minor), this group has been created for */
  151. dev_t dev;
  152. /* policy which owns this blk group */
  153. enum blkio_policy_id plid;
  154. /* Need to serialize the stats in the case of reset/update */
  155. spinlock_t stats_lock;
  156. struct blkio_group_stats stats;
  157. /* Per cpu stats pointer */
  158. struct blkio_group_stats_cpu __percpu *stats_cpu;
  159. };
  160. struct blkio_policy_node {
  161. struct list_head node;
  162. dev_t dev;
  163. /* This node belongs to max bw policy or porportional weight policy */
  164. enum blkio_policy_id plid;
  165. /* cgroup file to which this rule belongs to */
  166. int fileid;
  167. union {
  168. unsigned int weight;
  169. /*
  170. * Rate read/write in terms of byptes per second
  171. * Whether this rate represents read or write is determined
  172. * by file type "fileid".
  173. */
  174. u64 bps;
  175. unsigned int iops;
  176. } val;
  177. };
  178. extern unsigned int blkcg_get_weight(struct blkio_cgroup *blkcg,
  179. dev_t dev);
  180. extern uint64_t blkcg_get_read_bps(struct blkio_cgroup *blkcg,
  181. dev_t dev);
  182. extern uint64_t blkcg_get_write_bps(struct blkio_cgroup *blkcg,
  183. dev_t dev);
  184. extern unsigned int blkcg_get_read_iops(struct blkio_cgroup *blkcg,
  185. dev_t dev);
  186. extern unsigned int blkcg_get_write_iops(struct blkio_cgroup *blkcg,
  187. dev_t dev);
  188. typedef void (blkio_unlink_group_fn) (void *key, struct blkio_group *blkg);
  189. typedef void (blkio_update_group_weight_fn) (void *key,
  190. struct blkio_group *blkg, unsigned int weight);
  191. typedef void (blkio_update_group_read_bps_fn) (void * key,
  192. struct blkio_group *blkg, u64 read_bps);
  193. typedef void (blkio_update_group_write_bps_fn) (void *key,
  194. struct blkio_group *blkg, u64 write_bps);
  195. typedef void (blkio_update_group_read_iops_fn) (void *key,
  196. struct blkio_group *blkg, unsigned int read_iops);
  197. typedef void (blkio_update_group_write_iops_fn) (void *key,
  198. struct blkio_group *blkg, unsigned int write_iops);
  199. struct blkio_policy_ops {
  200. blkio_unlink_group_fn *blkio_unlink_group_fn;
  201. blkio_update_group_weight_fn *blkio_update_group_weight_fn;
  202. blkio_update_group_read_bps_fn *blkio_update_group_read_bps_fn;
  203. blkio_update_group_write_bps_fn *blkio_update_group_write_bps_fn;
  204. blkio_update_group_read_iops_fn *blkio_update_group_read_iops_fn;
  205. blkio_update_group_write_iops_fn *blkio_update_group_write_iops_fn;
  206. };
  207. struct blkio_policy_type {
  208. struct list_head list;
  209. struct blkio_policy_ops ops;
  210. enum blkio_policy_id plid;
  211. };
  212. /* Blkio controller policy registration */
  213. extern void blkio_policy_register(struct blkio_policy_type *);
  214. extern void blkio_policy_unregister(struct blkio_policy_type *);
  215. static inline char *blkg_path(struct blkio_group *blkg)
  216. {
  217. return blkg->path;
  218. }
  219. #else
  220. struct blkio_group {
  221. };
  222. struct blkio_policy_type {
  223. };
  224. static inline void blkio_policy_register(struct blkio_policy_type *blkiop) { }
  225. static inline void blkio_policy_unregister(struct blkio_policy_type *blkiop) { }
  226. static inline char *blkg_path(struct blkio_group *blkg) { return NULL; }
  227. #endif
  228. #define BLKIO_WEIGHT_MIN 10
  229. #define BLKIO_WEIGHT_MAX 1000
  230. #define BLKIO_WEIGHT_DEFAULT 500
  231. #ifdef CONFIG_DEBUG_BLK_CGROUP
  232. void blkiocg_update_avg_queue_size_stats(struct blkio_group *blkg);
  233. void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
  234. unsigned long dequeue);
  235. void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg);
  236. void blkiocg_update_idle_time_stats(struct blkio_group *blkg);
  237. void blkiocg_set_start_empty_time(struct blkio_group *blkg);
  238. #define BLKG_FLAG_FNS(name) \
  239. static inline void blkio_mark_blkg_##name( \
  240. struct blkio_group_stats *stats) \
  241. { \
  242. stats->flags |= (1 << BLKG_##name); \
  243. } \
  244. static inline void blkio_clear_blkg_##name( \
  245. struct blkio_group_stats *stats) \
  246. { \
  247. stats->flags &= ~(1 << BLKG_##name); \
  248. } \
  249. static inline int blkio_blkg_##name(struct blkio_group_stats *stats) \
  250. { \
  251. return (stats->flags & (1 << BLKG_##name)) != 0; \
  252. } \
  253. BLKG_FLAG_FNS(waiting)
  254. BLKG_FLAG_FNS(idling)
  255. BLKG_FLAG_FNS(empty)
  256. #undef BLKG_FLAG_FNS
  257. #else
  258. static inline void blkiocg_update_avg_queue_size_stats(
  259. struct blkio_group *blkg) {}
  260. static inline void blkiocg_update_dequeue_stats(struct blkio_group *blkg,
  261. unsigned long dequeue) {}
  262. static inline void blkiocg_update_set_idle_time_stats(struct blkio_group *blkg)
  263. {}
  264. static inline void blkiocg_update_idle_time_stats(struct blkio_group *blkg) {}
  265. static inline void blkiocg_set_start_empty_time(struct blkio_group *blkg) {}
  266. #endif
  267. #if defined(CONFIG_BLK_CGROUP) || defined(CONFIG_BLK_CGROUP_MODULE)
  268. extern struct blkio_cgroup blkio_root_cgroup;
  269. extern struct blkio_cgroup *cgroup_to_blkio_cgroup(struct cgroup *cgroup);
  270. extern struct blkio_cgroup *task_blkio_cgroup(struct task_struct *tsk);
  271. extern void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
  272. struct blkio_group *blkg, void *key, dev_t dev,
  273. enum blkio_policy_id plid);
  274. extern int blkio_alloc_blkg_stats(struct blkio_group *blkg);
  275. extern int blkiocg_del_blkio_group(struct blkio_group *blkg);
  276. extern struct blkio_group *blkiocg_lookup_group(struct blkio_cgroup *blkcg,
  277. void *key);
  278. void blkiocg_update_timeslice_used(struct blkio_group *blkg,
  279. unsigned long time,
  280. unsigned long unaccounted_time);
  281. void blkiocg_update_dispatch_stats(struct blkio_group *blkg, uint64_t bytes,
  282. bool direction, bool sync);
  283. void blkiocg_update_completion_stats(struct blkio_group *blkg,
  284. uint64_t start_time, uint64_t io_start_time, bool direction, bool sync);
  285. void blkiocg_update_io_merged_stats(struct blkio_group *blkg, bool direction,
  286. bool sync);
  287. void blkiocg_update_io_add_stats(struct blkio_group *blkg,
  288. struct blkio_group *curr_blkg, bool direction, bool sync);
  289. void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
  290. bool direction, bool sync);
  291. #else
  292. struct cgroup;
  293. static inline struct blkio_cgroup *
  294. cgroup_to_blkio_cgroup(struct cgroup *cgroup) { return NULL; }
  295. static inline struct blkio_cgroup *
  296. task_blkio_cgroup(struct task_struct *tsk) { return NULL; }
  297. static inline void blkiocg_add_blkio_group(struct blkio_cgroup *blkcg,
  298. struct blkio_group *blkg, void *key, dev_t dev,
  299. enum blkio_policy_id plid) {}
  300. static inline int blkio_alloc_blkg_stats(struct blkio_group *blkg) { return 0; }
  301. static inline int
  302. blkiocg_del_blkio_group(struct blkio_group *blkg) { return 0; }
  303. static inline struct blkio_group *
  304. blkiocg_lookup_group(struct blkio_cgroup *blkcg, void *key) { return NULL; }
  305. static inline void blkiocg_update_timeslice_used(struct blkio_group *blkg,
  306. unsigned long time,
  307. unsigned long unaccounted_time)
  308. {}
  309. static inline void blkiocg_update_dispatch_stats(struct blkio_group *blkg,
  310. uint64_t bytes, bool direction, bool sync) {}
  311. static inline void blkiocg_update_completion_stats(struct blkio_group *blkg,
  312. uint64_t start_time, uint64_t io_start_time, bool direction,
  313. bool sync) {}
  314. static inline void blkiocg_update_io_merged_stats(struct blkio_group *blkg,
  315. bool direction, bool sync) {}
  316. static inline void blkiocg_update_io_add_stats(struct blkio_group *blkg,
  317. struct blkio_group *curr_blkg, bool direction, bool sync) {}
  318. static inline void blkiocg_update_io_remove_stats(struct blkio_group *blkg,
  319. bool direction, bool sync) {}
  320. #endif
  321. #endif /* _BLK_CGROUP_H */