cgroup-internal.h 9.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299
  1. /* SPDX-License-Identifier: GPL-2.0 */
  2. #ifndef __CGROUP_INTERNAL_H
  3. #define __CGROUP_INTERNAL_H
  4. #include <linux/cgroup.h>
  5. #include <linux/kernfs.h>
  6. #include <linux/workqueue.h>
  7. #include <linux/list.h>
  8. #include <linux/refcount.h>
  9. #include <linux/fs_parser.h>
  10. #define TRACE_CGROUP_PATH_LEN 1024
  11. extern spinlock_t trace_cgroup_path_lock;
  12. extern char trace_cgroup_path[TRACE_CGROUP_PATH_LEN];
  13. extern void __init enable_debug_cgroup(void);
  14. /*
  15. * cgroup_path() takes a spin lock. It is good practice not to take
  16. * spin locks within trace point handlers, as they are mostly hidden
  17. * from normal view. As cgroup_path() can take the kernfs_rename_lock
  18. * spin lock, it is best to not call that function from the trace event
  19. * handler.
  20. *
  21. * Note: trace_cgroup_##type##_enabled() is a static branch that will only
  22. * be set when the trace event is enabled.
  23. */
  24. #define TRACE_CGROUP_PATH(type, cgrp, ...) \
  25. do { \
  26. if (trace_cgroup_##type##_enabled()) { \
  27. unsigned long flags; \
  28. spin_lock_irqsave(&trace_cgroup_path_lock, \
  29. flags); \
  30. cgroup_path(cgrp, trace_cgroup_path, \
  31. TRACE_CGROUP_PATH_LEN); \
  32. trace_cgroup_##type(cgrp, trace_cgroup_path, \
  33. ##__VA_ARGS__); \
  34. spin_unlock_irqrestore(&trace_cgroup_path_lock, \
  35. flags); \
  36. } \
  37. } while (0)
  38. /*
  39. * The cgroup filesystem superblock creation/mount context.
  40. */
  41. struct cgroup_fs_context {
  42. struct kernfs_fs_context kfc;
  43. struct cgroup_root *root;
  44. struct cgroup_namespace *ns;
  45. unsigned int flags; /* CGRP_ROOT_* flags */
  46. /* cgroup1 bits */
  47. bool cpuset_clone_children;
  48. bool none; /* User explicitly requested empty subsystem */
  49. bool all_ss; /* Seen 'all' option */
  50. u16 subsys_mask; /* Selected subsystems */
  51. char *name; /* Hierarchy name */
  52. char *release_agent; /* Path for release notifications */
  53. };
  54. static inline struct cgroup_fs_context *cgroup_fc2context(struct fs_context *fc)
  55. {
  56. struct kernfs_fs_context *kfc = fc->fs_private;
  57. return container_of(kfc, struct cgroup_fs_context, kfc);
  58. }
  59. struct cgroup_pidlist;
  60. struct cgroup_file_ctx {
  61. struct cgroup_namespace *ns;
  62. struct {
  63. void *trigger;
  64. } psi;
  65. struct {
  66. bool started;
  67. struct css_task_iter iter;
  68. } procs;
  69. struct {
  70. struct cgroup_pidlist *pidlist;
  71. } procs1;
  72. };
  73. /*
  74. * A cgroup can be associated with multiple css_sets as different tasks may
  75. * belong to different cgroups on different hierarchies. In the other
  76. * direction, a css_set is naturally associated with multiple cgroups.
  77. * This M:N relationship is represented by the following link structure
  78. * which exists for each association and allows traversing the associations
  79. * from both sides.
  80. */
  81. struct cgrp_cset_link {
  82. /* the cgroup and css_set this link associates */
  83. struct cgroup *cgrp;
  84. struct css_set *cset;
  85. /* list of cgrp_cset_links anchored at cgrp->cset_links */
  86. struct list_head cset_link;
  87. /* list of cgrp_cset_links anchored at css_set->cgrp_links */
  88. struct list_head cgrp_link;
  89. };
  90. /* used to track tasks and csets during migration */
  91. struct cgroup_taskset {
  92. /* the src and dst cset list running through cset->mg_node */
  93. struct list_head src_csets;
  94. struct list_head dst_csets;
  95. /* the number of tasks in the set */
  96. int nr_tasks;
  97. /* the subsys currently being processed */
  98. int ssid;
  99. /*
  100. * Fields for cgroup_taskset_*() iteration.
  101. *
  102. * Before migration is committed, the target migration tasks are on
  103. * ->mg_tasks of the csets on ->src_csets. After, on ->mg_tasks of
  104. * the csets on ->dst_csets. ->csets point to either ->src_csets
  105. * or ->dst_csets depending on whether migration is committed.
  106. *
  107. * ->cur_csets and ->cur_task point to the current task position
  108. * during iteration.
  109. */
  110. struct list_head *csets;
  111. struct css_set *cur_cset;
  112. struct task_struct *cur_task;
  113. };
  114. /* migration context also tracks preloading */
  115. struct cgroup_mgctx {
  116. /*
  117. * Preloaded source and destination csets. Used to guarantee
  118. * atomic success or failure on actual migration.
  119. */
  120. struct list_head preloaded_src_csets;
  121. struct list_head preloaded_dst_csets;
  122. /* tasks and csets to migrate */
  123. struct cgroup_taskset tset;
  124. /* subsystems affected by migration */
  125. u16 ss_mask;
  126. };
  127. #define CGROUP_TASKSET_INIT(tset) \
  128. { \
  129. .src_csets = LIST_HEAD_INIT(tset.src_csets), \
  130. .dst_csets = LIST_HEAD_INIT(tset.dst_csets), \
  131. .csets = &tset.src_csets, \
  132. }
  133. #define CGROUP_MGCTX_INIT(name) \
  134. { \
  135. LIST_HEAD_INIT(name.preloaded_src_csets), \
  136. LIST_HEAD_INIT(name.preloaded_dst_csets), \
  137. CGROUP_TASKSET_INIT(name.tset), \
  138. }
  139. #define DEFINE_CGROUP_MGCTX(name) \
  140. struct cgroup_mgctx name = CGROUP_MGCTX_INIT(name)
  141. extern spinlock_t css_set_lock;
  142. extern struct cgroup_subsys *cgroup_subsys[];
  143. extern struct list_head cgroup_roots;
  144. /* iterate across the hierarchies */
  145. #define for_each_root(root) \
  146. list_for_each_entry((root), &cgroup_roots, root_list)
  147. /**
  148. * for_each_subsys - iterate all enabled cgroup subsystems
  149. * @ss: the iteration cursor
  150. * @ssid: the index of @ss, CGROUP_SUBSYS_COUNT after reaching the end
  151. */
  152. #define for_each_subsys(ss, ssid) \
  153. for ((ssid) = 0; (ssid) < CGROUP_SUBSYS_COUNT && \
  154. (((ss) = cgroup_subsys[ssid]) || true); (ssid)++)
  155. static inline bool cgroup_is_dead(const struct cgroup *cgrp)
  156. {
  157. return !(cgrp->self.flags & CSS_ONLINE);
  158. }
  159. static inline bool notify_on_release(const struct cgroup *cgrp)
  160. {
  161. return test_bit(CGRP_NOTIFY_ON_RELEASE, &cgrp->flags);
  162. }
  163. void put_css_set_locked(struct css_set *cset);
  164. static inline void put_css_set(struct css_set *cset)
  165. {
  166. unsigned long flags;
  167. /*
  168. * Ensure that the refcount doesn't hit zero while any readers
  169. * can see it. Similar to atomic_dec_and_lock(), but for an
  170. * rwlock
  171. */
  172. if (refcount_dec_not_one(&cset->refcount))
  173. return;
  174. spin_lock_irqsave(&css_set_lock, flags);
  175. put_css_set_locked(cset);
  176. spin_unlock_irqrestore(&css_set_lock, flags);
  177. }
  178. /*
  179. * refcounted get/put for css_set objects
  180. */
  181. static inline void get_css_set(struct css_set *cset)
  182. {
  183. refcount_inc(&cset->refcount);
  184. }
  185. bool cgroup_ssid_enabled(int ssid);
  186. bool cgroup_on_dfl(const struct cgroup *cgrp);
  187. bool cgroup_is_thread_root(struct cgroup *cgrp);
  188. bool cgroup_is_threaded(struct cgroup *cgrp);
  189. struct cgroup_root *cgroup_root_from_kf(struct kernfs_root *kf_root);
  190. struct cgroup *task_cgroup_from_root(struct task_struct *task,
  191. struct cgroup_root *root);
  192. struct cgroup *cgroup_kn_lock_live(struct kernfs_node *kn, bool drain_offline);
  193. void cgroup_kn_unlock(struct kernfs_node *kn);
  194. int cgroup_path_ns_locked(struct cgroup *cgrp, char *buf, size_t buflen,
  195. struct cgroup_namespace *ns);
  196. void cgroup_favor_dynmods(struct cgroup_root *root, bool favor);
  197. void cgroup_free_root(struct cgroup_root *root);
  198. void init_cgroup_root(struct cgroup_fs_context *ctx);
  199. int cgroup_setup_root(struct cgroup_root *root, u16 ss_mask);
  200. int rebind_subsystems(struct cgroup_root *dst_root, u16 ss_mask);
  201. int cgroup_do_get_tree(struct fs_context *fc);
  202. int cgroup_migrate_vet_dst(struct cgroup *dst_cgrp);
  203. void cgroup_migrate_finish(struct cgroup_mgctx *mgctx);
  204. void cgroup_migrate_add_src(struct css_set *src_cset, struct cgroup *dst_cgrp,
  205. struct cgroup_mgctx *mgctx);
  206. int cgroup_migrate_prepare_dst(struct cgroup_mgctx *mgctx);
  207. int cgroup_migrate(struct task_struct *leader, bool threadgroup,
  208. struct cgroup_mgctx *mgctx);
  209. int cgroup_attach_task(struct cgroup *dst_cgrp, struct task_struct *leader,
  210. bool threadgroup);
  211. void cgroup_attach_lock(bool lock_threadgroup);
  212. void cgroup_attach_unlock(bool lock_threadgroup);
  213. struct task_struct *cgroup_procs_write_start(char *buf, bool threadgroup,
  214. bool *locked,
  215. struct cgroup *dst_cgrp);
  216. __acquires(&cgroup_threadgroup_rwsem);
  217. void cgroup_procs_write_finish(struct task_struct *task, bool locked)
  218. __releases(&cgroup_threadgroup_rwsem);
  219. void cgroup_lock_and_drain_offline(struct cgroup *cgrp);
  220. int cgroup_mkdir(struct kernfs_node *parent_kn, const char *name, umode_t mode);
  221. int cgroup_rmdir(struct kernfs_node *kn);
  222. int cgroup_show_path(struct seq_file *sf, struct kernfs_node *kf_node,
  223. struct kernfs_root *kf_root);
  224. int __cgroup_task_count(const struct cgroup *cgrp);
  225. int cgroup_task_count(const struct cgroup *cgrp);
  226. /*
  227. * rstat.c
  228. */
  229. int cgroup_rstat_init(struct cgroup *cgrp);
  230. void cgroup_rstat_exit(struct cgroup *cgrp);
  231. void cgroup_rstat_boot(void);
  232. void cgroup_base_stat_cputime_show(struct seq_file *seq);
  233. /*
  234. * namespace.c
  235. */
  236. extern const struct proc_ns_operations cgroupns_operations;
  237. /*
  238. * cgroup-v1.c
  239. */
  240. extern struct cftype cgroup1_base_files[];
  241. extern struct kernfs_syscall_ops cgroup1_kf_syscall_ops;
  242. extern const struct fs_parameter_spec cgroup1_fs_parameters[];
  243. int proc_cgroupstats_show(struct seq_file *m, void *v);
  244. bool cgroup1_ssid_disabled(int ssid);
  245. void cgroup1_pidlist_destroy_all(struct cgroup *cgrp);
  246. void cgroup1_release_agent(struct work_struct *work);
  247. void cgroup1_check_for_release(struct cgroup *cgrp);
  248. int cgroup1_parse_param(struct fs_context *fc, struct fs_parameter *param);
  249. int cgroup1_get_tree(struct fs_context *fc);
  250. int cgroup1_reconfigure(struct fs_context *ctx);
  251. #endif /* __CGROUP_INTERNAL_H */