| 1 | #ifndef _LINUX_CGROUP_H |
| 2 | #define _LINUX_CGROUP_H |
| 3 | /* |
| 4 | * cgroup interface |
| 5 | * |
| 6 | * Copyright (C) 2003 BULL SA |
| 7 | * Copyright (C) 2004-2006 Silicon Graphics, Inc. |
| 8 | * |
| 9 | */ |
| 10 | |
| 11 | #include <linux/sched.h> |
| 12 | #include <linux/cpumask.h> |
| 13 | #include <linux/nodemask.h> |
| 14 | #include <linux/rcupdate.h> |
| 15 | #include <linux/rculist.h> |
| 16 | #include <linux/cgroupstats.h> |
| 17 | #include <linux/rwsem.h> |
| 18 | #include <linux/idr.h> |
| 19 | #include <linux/workqueue.h> |
| 20 | #include <linux/fs.h> |
| 21 | #include <linux/percpu-refcount.h> |
| 22 | #include <linux/seq_file.h> |
| 23 | #include <linux/kernfs.h> |
| 24 | #include <linux/wait.h> |
| 25 | |
| 26 | #ifdef CONFIG_CGROUPS |
| 27 | |
| 28 | struct cgroup_root; |
| 29 | struct cgroup_subsys; |
| 30 | struct inode; |
| 31 | struct cgroup; |
| 32 | |
| 33 | extern int cgroup_init_early(void); |
| 34 | extern int cgroup_init(void); |
| 35 | extern void cgroup_fork(struct task_struct *p); |
| 36 | extern void cgroup_post_fork(struct task_struct *p); |
| 37 | extern void cgroup_exit(struct task_struct *p); |
| 38 | extern int cgroupstats_build(struct cgroupstats *stats, |
| 39 | struct dentry *dentry); |
| 40 | |
| 41 | extern int proc_cgroup_show(struct seq_file *, void *); |
| 42 | |
| 43 | /* define the enumeration of all cgroup subsystems */ |
| 44 | #define SUBSYS(_x) _x ## _cgrp_id, |
| 45 | enum cgroup_subsys_id { |
| 46 | #include <linux/cgroup_subsys.h> |
| 47 | CGROUP_SUBSYS_COUNT, |
| 48 | }; |
| 49 | #undef SUBSYS |
| 50 | |
| 51 | /* Per-subsystem/per-cgroup state maintained by the system. */ |
| 52 | struct cgroup_subsys_state { |
| 53 | /* the cgroup that this css is attached to */ |
| 54 | struct cgroup *cgroup; |
| 55 | |
| 56 | /* the cgroup subsystem that this css is attached to */ |
| 57 | struct cgroup_subsys *ss; |
| 58 | |
| 59 | /* reference count - access via css_[try]get() and css_put() */ |
| 60 | struct percpu_ref refcnt; |
| 61 | |
| 62 | /* the parent css */ |
| 63 | struct cgroup_subsys_state *parent; |
| 64 | |
| 65 | unsigned int flags; |
| 66 | |
| 67 | /* percpu_ref killing and RCU release */ |
| 68 | struct rcu_head rcu_head; |
| 69 | struct work_struct destroy_work; |
| 70 | }; |
| 71 | |
| 72 | /* bits in struct cgroup_subsys_state flags field */ |
| 73 | enum { |
| 74 | CSS_ROOT = (1 << 0), /* this CSS is the root of the subsystem */ |
| 75 | CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */ |
| 76 | }; |
| 77 | |
| 78 | /** |
| 79 | * css_get - obtain a reference on the specified css |
| 80 | * @css: target css |
| 81 | * |
| 82 | * The caller must already have a reference. |
| 83 | */ |
| 84 | static inline void css_get(struct cgroup_subsys_state *css) |
| 85 | { |
| 86 | /* We don't need to reference count the root state */ |
| 87 | if (!(css->flags & CSS_ROOT)) |
| 88 | percpu_ref_get(&css->refcnt); |
| 89 | } |
| 90 | |
| 91 | /** |
| 92 | * css_tryget - try to obtain a reference on the specified css |
| 93 | * @css: target css |
| 94 | * |
| 95 | * Obtain a reference on @css if it's alive. The caller naturally needs to |
| 96 | * ensure that @css is accessible but doesn't have to be holding a |
| 97 | * reference on it - IOW, RCU protected access is good enough for this |
| 98 | * function. Returns %true if a reference count was successfully obtained; |
| 99 | * %false otherwise. |
| 100 | */ |
| 101 | static inline bool css_tryget(struct cgroup_subsys_state *css) |
| 102 | { |
| 103 | if (css->flags & CSS_ROOT) |
| 104 | return true; |
| 105 | return percpu_ref_tryget(&css->refcnt); |
| 106 | } |
| 107 | |
| 108 | /** |
| 109 | * css_put - put a css reference |
| 110 | * @css: target css |
| 111 | * |
| 112 | * Put a reference obtained via css_get() and css_tryget(). |
| 113 | */ |
| 114 | static inline void css_put(struct cgroup_subsys_state *css) |
| 115 | { |
| 116 | if (!(css->flags & CSS_ROOT)) |
| 117 | percpu_ref_put(&css->refcnt); |
| 118 | } |
| 119 | |
| 120 | /* bits in struct cgroup flags field */ |
| 121 | enum { |
| 122 | /* Control Group is dead */ |
| 123 | CGRP_DEAD, |
| 124 | /* |
| 125 | * Control Group has previously had a child cgroup or a task, |
| 126 | * but no longer (only if CGRP_NOTIFY_ON_RELEASE is set) |
| 127 | */ |
| 128 | CGRP_RELEASABLE, |
| 129 | /* Control Group requires release notifications to userspace */ |
| 130 | CGRP_NOTIFY_ON_RELEASE, |
| 131 | /* |
| 132 | * Clone the parent's configuration when creating a new child |
| 133 | * cpuset cgroup. For historical reasons, this option can be |
| 134 | * specified at mount time and thus is implemented here. |
| 135 | */ |
| 136 | CGRP_CPUSET_CLONE_CHILDREN, |
| 137 | /* see the comment above CGRP_ROOT_SANE_BEHAVIOR for details */ |
| 138 | CGRP_SANE_BEHAVIOR, |
| 139 | }; |
| 140 | |
| 141 | struct cgroup { |
| 142 | unsigned long flags; /* "unsigned long" so bitops work */ |
| 143 | |
| 144 | /* |
| 145 | * idr allocated in-hierarchy ID. |
| 146 | * |
| 147 | * ID 0 is not used, the ID of the root cgroup is always 1, and a |
| 148 | * new cgroup will be assigned with a smallest available ID. |
| 149 | * |
| 150 | * Allocating/Removing ID must be protected by cgroup_mutex. |
| 151 | */ |
| 152 | int id; |
| 153 | |
| 154 | /* the number of attached css's */ |
| 155 | int nr_css; |
| 156 | |
| 157 | /* |
| 158 | * If this cgroup contains any tasks, it contributes one to |
| 159 | * populated_cnt. All children with non-zero popuplated_cnt of |
| 160 | * their own contribute one. The count is zero iff there's no task |
| 161 | * in this cgroup or its subtree. |
| 162 | */ |
| 163 | int populated_cnt; |
| 164 | |
| 165 | atomic_t refcnt; |
| 166 | |
| 167 | /* |
| 168 | * We link our 'sibling' struct into our parent's 'children'. |
| 169 | * Our children link their 'sibling' into our 'children'. |
| 170 | */ |
| 171 | struct list_head sibling; /* my parent's children */ |
| 172 | struct list_head children; /* my children */ |
| 173 | |
| 174 | struct cgroup *parent; /* my parent */ |
| 175 | struct kernfs_node *kn; /* cgroup kernfs entry */ |
| 176 | struct kernfs_node *control_kn; /* kn for "cgroup.subtree_control" */ |
| 177 | struct kernfs_node *populated_kn; /* kn for "cgroup.subtree_populated" */ |
| 178 | |
| 179 | /* |
| 180 | * Monotonically increasing unique serial number which defines a |
| 181 | * uniform order among all cgroups. It's guaranteed that all |
| 182 | * ->children lists are in the ascending order of ->serial_nr. |
| 183 | * It's used to allow interrupting and resuming iterations. |
| 184 | */ |
| 185 | u64 serial_nr; |
| 186 | |
| 187 | /* the bitmask of subsystems enabled on the child cgroups */ |
| 188 | unsigned int child_subsys_mask; |
| 189 | |
| 190 | /* Private pointers for each registered subsystem */ |
| 191 | struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT]; |
| 192 | |
| 193 | struct cgroup_root *root; |
| 194 | |
| 195 | /* |
| 196 | * List of cgrp_cset_links pointing at css_sets with tasks in this |
| 197 | * cgroup. Protected by css_set_lock. |
| 198 | */ |
| 199 | struct list_head cset_links; |
| 200 | |
| 201 | /* |
| 202 | * On the default hierarchy, a css_set for a cgroup with some |
| 203 | * susbsys disabled will point to css's which are associated with |
| 204 | * the closest ancestor which has the subsys enabled. The |
| 205 | * following lists all css_sets which point to this cgroup's css |
| 206 | * for the given subsystem. |
| 207 | */ |
| 208 | struct list_head e_csets[CGROUP_SUBSYS_COUNT]; |
| 209 | |
| 210 | /* |
| 211 | * Linked list running through all cgroups that can |
| 212 | * potentially be reaped by the release agent. Protected by |
| 213 | * release_list_lock |
| 214 | */ |
| 215 | struct list_head release_list; |
| 216 | |
| 217 | /* |
| 218 | * list of pidlists, up to two for each namespace (one for procs, one |
| 219 | * for tasks); created on demand. |
| 220 | */ |
| 221 | struct list_head pidlists; |
| 222 | struct mutex pidlist_mutex; |
| 223 | |
| 224 | /* dummy css with NULL ->ss, points back to this cgroup */ |
| 225 | struct cgroup_subsys_state dummy_css; |
| 226 | |
| 227 | /* For css percpu_ref killing and RCU-protected deletion */ |
| 228 | struct rcu_head rcu_head; |
| 229 | struct work_struct destroy_work; |
| 230 | |
| 231 | /* used to wait for offlining of csses */ |
| 232 | wait_queue_head_t offline_waitq; |
| 233 | }; |
| 234 | |
| 235 | #define MAX_CGROUP_ROOT_NAMELEN 64 |
| 236 | |
| 237 | /* cgroup_root->flags */ |
| 238 | enum { |
| 239 | /* |
| 240 | * Unfortunately, cgroup core and various controllers are riddled |
| 241 | * with idiosyncrasies and pointless options. The following flag, |
| 242 | * when set, will force sane behavior - some options are forced on, |
| 243 | * others are disallowed, and some controllers will change their |
| 244 | * hierarchical or other behaviors. |
| 245 | * |
| 246 | * The set of behaviors affected by this flag are still being |
| 247 | * determined and developed and the mount option for this flag is |
| 248 | * prefixed with __DEVEL__. The prefix will be dropped once we |
| 249 | * reach the point where all behaviors are compatible with the |
| 250 | * planned unified hierarchy, which will automatically turn on this |
| 251 | * flag. |
| 252 | * |
| 253 | * The followings are the behaviors currently affected this flag. |
| 254 | * |
| 255 | * - Mount options "noprefix", "xattr", "clone_children", |
| 256 | * "release_agent" and "name" are disallowed. |
| 257 | * |
| 258 | * - When mounting an existing superblock, mount options should |
| 259 | * match. |
| 260 | * |
| 261 | * - Remount is disallowed. |
| 262 | * |
| 263 | * - rename(2) is disallowed. |
| 264 | * |
| 265 | * - "tasks" is removed. Everything should be at process |
| 266 | * granularity. Use "cgroup.procs" instead. |
| 267 | * |
| 268 | * - "cgroup.procs" is not sorted. pids will be unique unless they |
| 269 | * got recycled inbetween reads. |
| 270 | * |
| 271 | * - "release_agent" and "notify_on_release" are removed. |
| 272 | * Replacement notification mechanism will be implemented. |
| 273 | * |
| 274 | * - "cgroup.clone_children" is removed. |
| 275 | * |
| 276 | * - "cgroup.subtree_populated" is available. Its value is 0 if |
| 277 | * the cgroup and its descendants contain no task; otherwise, 1. |
| 278 | * The file also generates kernfs notification which can be |
| 279 | * monitored through poll and [di]notify when the value of the |
| 280 | * file changes. |
| 281 | * |
| 282 | * - If mount is requested with sane_behavior but without any |
| 283 | * subsystem, the default unified hierarchy is mounted. |
| 284 | * |
| 285 | * - cpuset: tasks will be kept in empty cpusets when hotplug happens |
| 286 | * and take masks of ancestors with non-empty cpus/mems, instead of |
| 287 | * being moved to an ancestor. |
| 288 | * |
| 289 | * - cpuset: a task can be moved into an empty cpuset, and again it |
| 290 | * takes masks of ancestors. |
| 291 | * |
| 292 | * - memcg: use_hierarchy is on by default and the cgroup file for |
| 293 | * the flag is not created. |
| 294 | * |
| 295 | * - blkcg: blk-throttle becomes properly hierarchical. |
| 296 | */ |
| 297 | CGRP_ROOT_SANE_BEHAVIOR = (1 << 0), |
| 298 | |
| 299 | CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */ |
| 300 | CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */ |
| 301 | |
| 302 | /* mount options live below bit 16 */ |
| 303 | CGRP_ROOT_OPTION_MASK = (1 << 16) - 1, |
| 304 | }; |
| 305 | |
| 306 | /* |
| 307 | * A cgroup_root represents the root of a cgroup hierarchy, and may be |
| 308 | * associated with a kernfs_root to form an active hierarchy. This is |
| 309 | * internal to cgroup core. Don't access directly from controllers. |
| 310 | */ |
| 311 | struct cgroup_root { |
| 312 | struct kernfs_root *kf_root; |
| 313 | |
| 314 | /* The bitmask of subsystems attached to this hierarchy */ |
| 315 | unsigned int subsys_mask; |
| 316 | |
| 317 | /* Unique id for this hierarchy. */ |
| 318 | int hierarchy_id; |
| 319 | |
| 320 | /* The root cgroup. Root is destroyed on its release. */ |
| 321 | struct cgroup cgrp; |
| 322 | |
| 323 | /* Number of cgroups in the hierarchy, used only for /proc/cgroups */ |
| 324 | atomic_t nr_cgrps; |
| 325 | |
| 326 | /* A list running through the active hierarchies */ |
| 327 | struct list_head root_list; |
| 328 | |
| 329 | /* Hierarchy-specific flags */ |
| 330 | unsigned int flags; |
| 331 | |
| 332 | /* IDs for cgroups in this hierarchy */ |
| 333 | struct idr cgroup_idr; |
| 334 | |
| 335 | /* The path to use for release notifications. */ |
| 336 | char release_agent_path[PATH_MAX]; |
| 337 | |
| 338 | /* The name for this hierarchy - may be empty */ |
| 339 | char name[MAX_CGROUP_ROOT_NAMELEN]; |
| 340 | }; |
| 341 | |
| 342 | /* |
| 343 | * A css_set is a structure holding pointers to a set of |
| 344 | * cgroup_subsys_state objects. This saves space in the task struct |
| 345 | * object and speeds up fork()/exit(), since a single inc/dec and a |
| 346 | * list_add()/del() can bump the reference count on the entire cgroup |
| 347 | * set for a task. |
| 348 | */ |
| 349 | |
| 350 | struct css_set { |
| 351 | |
| 352 | /* Reference count */ |
| 353 | atomic_t refcount; |
| 354 | |
| 355 | /* |
| 356 | * List running through all cgroup groups in the same hash |
| 357 | * slot. Protected by css_set_lock |
| 358 | */ |
| 359 | struct hlist_node hlist; |
| 360 | |
| 361 | /* |
| 362 | * Lists running through all tasks using this cgroup group. |
| 363 | * mg_tasks lists tasks which belong to this cset but are in the |
| 364 | * process of being migrated out or in. Protected by |
| 365 | * css_set_rwsem, but, during migration, once tasks are moved to |
| 366 | * mg_tasks, it can be read safely while holding cgroup_mutex. |
| 367 | */ |
| 368 | struct list_head tasks; |
| 369 | struct list_head mg_tasks; |
| 370 | |
| 371 | /* |
| 372 | * List of cgrp_cset_links pointing at cgroups referenced from this |
| 373 | * css_set. Protected by css_set_lock. |
| 374 | */ |
| 375 | struct list_head cgrp_links; |
| 376 | |
| 377 | /* the default cgroup associated with this css_set */ |
| 378 | struct cgroup *dfl_cgrp; |
| 379 | |
| 380 | /* |
| 381 | * Set of subsystem states, one for each subsystem. This array is |
| 382 | * immutable after creation apart from the init_css_set during |
| 383 | * subsystem registration (at boot time). |
| 384 | */ |
| 385 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; |
| 386 | |
| 387 | /* |
| 388 | * List of csets participating in the on-going migration either as |
| 389 | * source or destination. Protected by cgroup_mutex. |
| 390 | */ |
| 391 | struct list_head mg_preload_node; |
| 392 | struct list_head mg_node; |
| 393 | |
| 394 | /* |
| 395 | * If this cset is acting as the source of migration the following |
| 396 | * two fields are set. mg_src_cgrp is the source cgroup of the |
| 397 | * on-going migration and mg_dst_cset is the destination cset the |
| 398 | * target tasks on this cset should be migrated to. Protected by |
| 399 | * cgroup_mutex. |
| 400 | */ |
| 401 | struct cgroup *mg_src_cgrp; |
| 402 | struct css_set *mg_dst_cset; |
| 403 | |
| 404 | /* |
| 405 | * On the default hierarhcy, ->subsys[ssid] may point to a css |
| 406 | * attached to an ancestor instead of the cgroup this css_set is |
| 407 | * associated with. The following node is anchored at |
| 408 | * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to |
| 409 | * iterate through all css's attached to a given cgroup. |
| 410 | */ |
| 411 | struct list_head e_cset_node[CGROUP_SUBSYS_COUNT]; |
| 412 | |
| 413 | /* For RCU-protected deletion */ |
| 414 | struct rcu_head rcu_head; |
| 415 | }; |
| 416 | |
| 417 | /* |
| 418 | * struct cftype: handler definitions for cgroup control files |
| 419 | * |
| 420 | * When reading/writing to a file: |
| 421 | * - the cgroup to use is file->f_dentry->d_parent->d_fsdata |
| 422 | * - the 'cftype' of the file is file->f_dentry->d_fsdata |
| 423 | */ |
| 424 | |
| 425 | /* cftype->flags */ |
| 426 | enum { |
| 427 | CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */ |
| 428 | CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */ |
| 429 | CFTYPE_INSANE = (1 << 2), /* don't create if sane_behavior */ |
| 430 | CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */ |
| 431 | CFTYPE_ONLY_ON_DFL = (1 << 4), /* only on default hierarchy */ |
| 432 | }; |
| 433 | |
| 434 | #define MAX_CFTYPE_NAME 64 |
| 435 | |
| 436 | struct cftype { |
| 437 | /* |
| 438 | * By convention, the name should begin with the name of the |
| 439 | * subsystem, followed by a period. Zero length string indicates |
| 440 | * end of cftype array. |
| 441 | */ |
| 442 | char name[MAX_CFTYPE_NAME]; |
| 443 | int private; |
| 444 | /* |
| 445 | * If not 0, file mode is set to this value, otherwise it will |
| 446 | * be figured out automatically |
| 447 | */ |
| 448 | umode_t mode; |
| 449 | |
| 450 | /* |
| 451 | * The maximum length of string, excluding trailing nul, that can |
| 452 | * be passed to write_string. If < PAGE_SIZE-1, PAGE_SIZE-1 is |
| 453 | * assumed. |
| 454 | */ |
| 455 | size_t max_write_len; |
| 456 | |
| 457 | /* CFTYPE_* flags */ |
| 458 | unsigned int flags; |
| 459 | |
| 460 | /* |
| 461 | * Fields used for internal bookkeeping. Initialized automatically |
| 462 | * during registration. |
| 463 | */ |
| 464 | struct cgroup_subsys *ss; /* NULL for cgroup core files */ |
| 465 | struct list_head node; /* anchored at ss->cfts */ |
| 466 | struct kernfs_ops *kf_ops; |
| 467 | |
| 468 | /* |
| 469 | * read_u64() is a shortcut for the common case of returning a |
| 470 | * single integer. Use it in place of read() |
| 471 | */ |
| 472 | u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft); |
| 473 | /* |
| 474 | * read_s64() is a signed version of read_u64() |
| 475 | */ |
| 476 | s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft); |
| 477 | |
| 478 | /* generic seq_file read interface */ |
| 479 | int (*seq_show)(struct seq_file *sf, void *v); |
| 480 | |
| 481 | /* optional ops, implement all or none */ |
| 482 | void *(*seq_start)(struct seq_file *sf, loff_t *ppos); |
| 483 | void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos); |
| 484 | void (*seq_stop)(struct seq_file *sf, void *v); |
| 485 | |
| 486 | /* |
| 487 | * write_u64() is a shortcut for the common case of accepting |
| 488 | * a single integer (as parsed by simple_strtoull) from |
| 489 | * userspace. Use in place of write(); return 0 or error. |
| 490 | */ |
| 491 | int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft, |
| 492 | u64 val); |
| 493 | /* |
| 494 | * write_s64() is a signed version of write_u64() |
| 495 | */ |
| 496 | int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft, |
| 497 | s64 val); |
| 498 | |
| 499 | /* |
| 500 | * write_string() is passed a nul-terminated kernelspace |
| 501 | * buffer of maximum length determined by max_write_len. |
| 502 | * Returns 0 or -ve error code. |
| 503 | */ |
| 504 | int (*write_string)(struct cgroup_subsys_state *css, struct cftype *cft, |
| 505 | char *buffer); |
| 506 | /* |
| 507 | * trigger() callback can be used to get some kick from the |
| 508 | * userspace, when the actual string written is not important |
| 509 | * at all. The private field can be used to determine the |
| 510 | * kick type for multiplexing. |
| 511 | */ |
| 512 | int (*trigger)(struct cgroup_subsys_state *css, unsigned int event); |
| 513 | |
| 514 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
| 515 | struct lock_class_key lockdep_key; |
| 516 | #endif |
| 517 | }; |
| 518 | |
| 519 | extern struct cgroup_root cgrp_dfl_root; |
| 520 | |
| 521 | static inline bool cgroup_on_dfl(const struct cgroup *cgrp) |
| 522 | { |
| 523 | return cgrp->root == &cgrp_dfl_root; |
| 524 | } |
| 525 | |
| 526 | /* |
| 527 | * See the comment above CGRP_ROOT_SANE_BEHAVIOR for details. This |
| 528 | * function can be called as long as @cgrp is accessible. |
| 529 | */ |
| 530 | static inline bool cgroup_sane_behavior(const struct cgroup *cgrp) |
| 531 | { |
| 532 | return cgrp->root->flags & CGRP_ROOT_SANE_BEHAVIOR; |
| 533 | } |
| 534 | |
| 535 | /* no synchronization, the result can only be used as a hint */ |
| 536 | static inline bool cgroup_has_tasks(struct cgroup *cgrp) |
| 537 | { |
| 538 | return !list_empty(&cgrp->cset_links); |
| 539 | } |
| 540 | |
| 541 | /* returns ino associated with a cgroup, 0 indicates unmounted root */ |
| 542 | static inline ino_t cgroup_ino(struct cgroup *cgrp) |
| 543 | { |
| 544 | if (cgrp->kn) |
| 545 | return cgrp->kn->ino; |
| 546 | else |
| 547 | return 0; |
| 548 | } |
| 549 | |
| 550 | static inline struct cftype *seq_cft(struct seq_file *seq) |
| 551 | { |
| 552 | struct kernfs_open_file *of = seq->private; |
| 553 | |
| 554 | return of->kn->priv; |
| 555 | } |
| 556 | |
| 557 | struct cgroup_subsys_state *seq_css(struct seq_file *seq); |
| 558 | |
| 559 | /* |
| 560 | * Name / path handling functions. All are thin wrappers around the kernfs |
| 561 | * counterparts and can be called under any context. |
| 562 | */ |
| 563 | |
| 564 | static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen) |
| 565 | { |
| 566 | return kernfs_name(cgrp->kn, buf, buflen); |
| 567 | } |
| 568 | |
| 569 | static inline char * __must_check cgroup_path(struct cgroup *cgrp, char *buf, |
| 570 | size_t buflen) |
| 571 | { |
| 572 | return kernfs_path(cgrp->kn, buf, buflen); |
| 573 | } |
| 574 | |
| 575 | static inline void pr_cont_cgroup_name(struct cgroup *cgrp) |
| 576 | { |
| 577 | pr_cont_kernfs_name(cgrp->kn); |
| 578 | } |
| 579 | |
| 580 | static inline void pr_cont_cgroup_path(struct cgroup *cgrp) |
| 581 | { |
| 582 | pr_cont_kernfs_path(cgrp->kn); |
| 583 | } |
| 584 | |
| 585 | char *task_cgroup_path(struct task_struct *task, char *buf, size_t buflen); |
| 586 | |
| 587 | int cgroup_add_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); |
| 588 | int cgroup_rm_cftypes(struct cftype *cfts); |
| 589 | |
| 590 | bool cgroup_is_descendant(struct cgroup *cgrp, struct cgroup *ancestor); |
| 591 | |
| 592 | /* |
| 593 | * Control Group taskset, used to pass around set of tasks to cgroup_subsys |
| 594 | * methods. |
| 595 | */ |
| 596 | struct cgroup_taskset; |
| 597 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset); |
| 598 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset); |
| 599 | |
| 600 | /** |
| 601 | * cgroup_taskset_for_each - iterate cgroup_taskset |
| 602 | * @task: the loop cursor |
| 603 | * @tset: taskset to iterate |
| 604 | */ |
| 605 | #define cgroup_taskset_for_each(task, tset) \ |
| 606 | for ((task) = cgroup_taskset_first((tset)); (task); \ |
| 607 | (task) = cgroup_taskset_next((tset))) |
| 608 | |
| 609 | /* |
| 610 | * Control Group subsystem type. |
| 611 | * See Documentation/cgroups/cgroups.txt for details |
| 612 | */ |
| 613 | |
| 614 | struct cgroup_subsys { |
| 615 | struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css); |
| 616 | int (*css_online)(struct cgroup_subsys_state *css); |
| 617 | void (*css_offline)(struct cgroup_subsys_state *css); |
| 618 | void (*css_free)(struct cgroup_subsys_state *css); |
| 619 | |
| 620 | int (*can_attach)(struct cgroup_subsys_state *css, |
| 621 | struct cgroup_taskset *tset); |
| 622 | void (*cancel_attach)(struct cgroup_subsys_state *css, |
| 623 | struct cgroup_taskset *tset); |
| 624 | void (*attach)(struct cgroup_subsys_state *css, |
| 625 | struct cgroup_taskset *tset); |
| 626 | void (*fork)(struct task_struct *task); |
| 627 | void (*exit)(struct cgroup_subsys_state *css, |
| 628 | struct cgroup_subsys_state *old_css, |
| 629 | struct task_struct *task); |
| 630 | void (*bind)(struct cgroup_subsys_state *root_css); |
| 631 | |
| 632 | int disabled; |
| 633 | int early_init; |
| 634 | |
| 635 | /* |
| 636 | * If %false, this subsystem is properly hierarchical - |
| 637 | * configuration, resource accounting and restriction on a parent |
| 638 | * cgroup cover those of its children. If %true, hierarchy support |
| 639 | * is broken in some ways - some subsystems ignore hierarchy |
| 640 | * completely while others are only implemented half-way. |
| 641 | * |
| 642 | * It's now disallowed to create nested cgroups if the subsystem is |
| 643 | * broken and cgroup core will emit a warning message on such |
| 644 | * cases. Eventually, all subsystems will be made properly |
| 645 | * hierarchical and this will go away. |
| 646 | */ |
| 647 | bool broken_hierarchy; |
| 648 | bool warned_broken_hierarchy; |
| 649 | |
| 650 | /* the following two fields are initialized automtically during boot */ |
| 651 | int id; |
| 652 | #define MAX_CGROUP_TYPE_NAMELEN 32 |
| 653 | const char *name; |
| 654 | |
| 655 | /* link to parent, protected by cgroup_lock() */ |
| 656 | struct cgroup_root *root; |
| 657 | |
| 658 | /* |
| 659 | * List of cftypes. Each entry is the first entry of an array |
| 660 | * terminated by zero length name. |
| 661 | */ |
| 662 | struct list_head cfts; |
| 663 | |
| 664 | /* base cftypes, automatically registered with subsys itself */ |
| 665 | struct cftype *base_cftypes; |
| 666 | }; |
| 667 | |
| 668 | #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys; |
| 669 | #include <linux/cgroup_subsys.h> |
| 670 | #undef SUBSYS |
| 671 | |
| 672 | /** |
| 673 | * css_parent - find the parent css |
| 674 | * @css: the target cgroup_subsys_state |
| 675 | * |
| 676 | * Return the parent css of @css. This function is guaranteed to return |
| 677 | * non-NULL parent as long as @css isn't the root. |
| 678 | */ |
| 679 | static inline |
| 680 | struct cgroup_subsys_state *css_parent(struct cgroup_subsys_state *css) |
| 681 | { |
| 682 | return css->parent; |
| 683 | } |
| 684 | |
| 685 | /** |
| 686 | * task_css_set_check - obtain a task's css_set with extra access conditions |
| 687 | * @task: the task to obtain css_set for |
| 688 | * @__c: extra condition expression to be passed to rcu_dereference_check() |
| 689 | * |
| 690 | * A task's css_set is RCU protected, initialized and exited while holding |
| 691 | * task_lock(), and can only be modified while holding both cgroup_mutex |
| 692 | * and task_lock() while the task is alive. This macro verifies that the |
| 693 | * caller is inside proper critical section and returns @task's css_set. |
| 694 | * |
| 695 | * The caller can also specify additional allowed conditions via @__c, such |
| 696 | * as locks used during the cgroup_subsys::attach() methods. |
| 697 | */ |
| 698 | #ifdef CONFIG_PROVE_RCU |
| 699 | extern struct mutex cgroup_mutex; |
| 700 | extern struct rw_semaphore css_set_rwsem; |
| 701 | #define task_css_set_check(task, __c) \ |
| 702 | rcu_dereference_check((task)->cgroups, \ |
| 703 | lockdep_is_held(&cgroup_mutex) || \ |
| 704 | lockdep_is_held(&css_set_rwsem) || \ |
| 705 | ((task)->flags & PF_EXITING) || (__c)) |
| 706 | #else |
| 707 | #define task_css_set_check(task, __c) \ |
| 708 | rcu_dereference((task)->cgroups) |
| 709 | #endif |
| 710 | |
| 711 | /** |
| 712 | * task_css_check - obtain css for (task, subsys) w/ extra access conds |
| 713 | * @task: the target task |
| 714 | * @subsys_id: the target subsystem ID |
| 715 | * @__c: extra condition expression to be passed to rcu_dereference_check() |
| 716 | * |
| 717 | * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The |
| 718 | * synchronization rules are the same as task_css_set_check(). |
| 719 | */ |
| 720 | #define task_css_check(task, subsys_id, __c) \ |
| 721 | task_css_set_check((task), (__c))->subsys[(subsys_id)] |
| 722 | |
| 723 | /** |
| 724 | * task_css_set - obtain a task's css_set |
| 725 | * @task: the task to obtain css_set for |
| 726 | * |
| 727 | * See task_css_set_check(). |
| 728 | */ |
| 729 | static inline struct css_set *task_css_set(struct task_struct *task) |
| 730 | { |
| 731 | return task_css_set_check(task, false); |
| 732 | } |
| 733 | |
| 734 | /** |
| 735 | * task_css - obtain css for (task, subsys) |
| 736 | * @task: the target task |
| 737 | * @subsys_id: the target subsystem ID |
| 738 | * |
| 739 | * See task_css_check(). |
| 740 | */ |
| 741 | static inline struct cgroup_subsys_state *task_css(struct task_struct *task, |
| 742 | int subsys_id) |
| 743 | { |
| 744 | return task_css_check(task, subsys_id, false); |
| 745 | } |
| 746 | |
| 747 | static inline struct cgroup *task_cgroup(struct task_struct *task, |
| 748 | int subsys_id) |
| 749 | { |
| 750 | return task_css(task, subsys_id)->cgroup; |
| 751 | } |
| 752 | |
| 753 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
| 754 | struct cgroup_subsys_state *parent); |
| 755 | |
| 756 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss); |
| 757 | |
| 758 | /** |
| 759 | * css_for_each_child - iterate through children of a css |
| 760 | * @pos: the css * to use as the loop cursor |
| 761 | * @parent: css whose children to walk |
| 762 | * |
| 763 | * Walk @parent's children. Must be called under rcu_read_lock(). A child |
| 764 | * css which hasn't finished ->css_online() or already has finished |
| 765 | * ->css_offline() may show up during traversal and it's each subsystem's |
| 766 | * responsibility to verify that each @pos is alive. |
| 767 | * |
| 768 | * If a subsystem synchronizes against the parent in its ->css_online() and |
| 769 | * before starting iterating, a css which finished ->css_online() is |
| 770 | * guaranteed to be visible in the future iterations. |
| 771 | * |
| 772 | * It is allowed to temporarily drop RCU read lock during iteration. The |
| 773 | * caller is responsible for ensuring that @pos remains accessible until |
| 774 | * the start of the next iteration by, for example, bumping the css refcnt. |
| 775 | */ |
| 776 | #define css_for_each_child(pos, parent) \ |
| 777 | for ((pos) = css_next_child(NULL, (parent)); (pos); \ |
| 778 | (pos) = css_next_child((pos), (parent))) |
| 779 | |
| 780 | struct cgroup_subsys_state * |
| 781 | css_next_descendant_pre(struct cgroup_subsys_state *pos, |
| 782 | struct cgroup_subsys_state *css); |
| 783 | |
| 784 | struct cgroup_subsys_state * |
| 785 | css_rightmost_descendant(struct cgroup_subsys_state *pos); |
| 786 | |
| 787 | /** |
| 788 | * css_for_each_descendant_pre - pre-order walk of a css's descendants |
| 789 | * @pos: the css * to use as the loop cursor |
| 790 | * @root: css whose descendants to walk |
| 791 | * |
| 792 | * Walk @root's descendants. @root is included in the iteration and the |
| 793 | * first node to be visited. Must be called under rcu_read_lock(). A |
| 794 | * descendant css which hasn't finished ->css_online() or already has |
| 795 | * finished ->css_offline() may show up during traversal and it's each |
| 796 | * subsystem's responsibility to verify that each @pos is alive. |
| 797 | * |
| 798 | * If a subsystem synchronizes against the parent in its ->css_online() and |
| 799 | * before starting iterating, and synchronizes against @pos on each |
| 800 | * iteration, any descendant css which finished ->css_online() is |
| 801 | * guaranteed to be visible in the future iterations. |
| 802 | * |
| 803 | * In other words, the following guarantees that a descendant can't escape |
| 804 | * state updates of its ancestors. |
| 805 | * |
| 806 | * my_online(@css) |
| 807 | * { |
| 808 | * Lock @css's parent and @css; |
| 809 | * Inherit state from the parent; |
| 810 | * Unlock both. |
| 811 | * } |
| 812 | * |
| 813 | * my_update_state(@css) |
| 814 | * { |
| 815 | * css_for_each_descendant_pre(@pos, @css) { |
| 816 | * Lock @pos; |
| 817 | * if (@pos == @css) |
| 818 | * Update @css's state; |
| 819 | * else |
| 820 | * Verify @pos is alive and inherit state from its parent; |
| 821 | * Unlock @pos; |
| 822 | * } |
| 823 | * } |
| 824 | * |
| 825 | * As long as the inheriting step, including checking the parent state, is |
| 826 | * enclosed inside @pos locking, double-locking the parent isn't necessary |
| 827 | * while inheriting. The state update to the parent is guaranteed to be |
| 828 | * visible by walking order and, as long as inheriting operations to the |
| 829 | * same @pos are atomic to each other, multiple updates racing each other |
| 830 | * still result in the correct state. It's guaranateed that at least one |
| 831 | * inheritance happens for any css after the latest update to its parent. |
| 832 | * |
| 833 | * If checking parent's state requires locking the parent, each inheriting |
| 834 | * iteration should lock and unlock both @pos->parent and @pos. |
| 835 | * |
| 836 | * Alternatively, a subsystem may choose to use a single global lock to |
| 837 | * synchronize ->css_online() and ->css_offline() against tree-walking |
| 838 | * operations. |
| 839 | * |
| 840 | * It is allowed to temporarily drop RCU read lock during iteration. The |
| 841 | * caller is responsible for ensuring that @pos remains accessible until |
| 842 | * the start of the next iteration by, for example, bumping the css refcnt. |
| 843 | */ |
| 844 | #define css_for_each_descendant_pre(pos, css) \ |
| 845 | for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \ |
| 846 | (pos) = css_next_descendant_pre((pos), (css))) |
| 847 | |
| 848 | struct cgroup_subsys_state * |
| 849 | css_next_descendant_post(struct cgroup_subsys_state *pos, |
| 850 | struct cgroup_subsys_state *css); |
| 851 | |
| 852 | /** |
| 853 | * css_for_each_descendant_post - post-order walk of a css's descendants |
| 854 | * @pos: the css * to use as the loop cursor |
| 855 | * @css: css whose descendants to walk |
| 856 | * |
| 857 | * Similar to css_for_each_descendant_pre() but performs post-order |
| 858 | * traversal instead. @root is included in the iteration and the last |
| 859 | * node to be visited. Note that the walk visibility guarantee described |
| 860 | * in pre-order walk doesn't apply the same to post-order walks. |
| 861 | */ |
| 862 | #define css_for_each_descendant_post(pos, css) \ |
| 863 | for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \ |
| 864 | (pos) = css_next_descendant_post((pos), (css))) |
| 865 | |
| 866 | /* A css_task_iter should be treated as an opaque object */ |
| 867 | struct css_task_iter { |
| 868 | struct cgroup_subsys *ss; |
| 869 | |
| 870 | struct list_head *cset_pos; |
| 871 | struct list_head *cset_head; |
| 872 | |
| 873 | struct list_head *task_pos; |
| 874 | struct list_head *tasks_head; |
| 875 | struct list_head *mg_tasks_head; |
| 876 | }; |
| 877 | |
| 878 | void css_task_iter_start(struct cgroup_subsys_state *css, |
| 879 | struct css_task_iter *it); |
| 880 | struct task_struct *css_task_iter_next(struct css_task_iter *it); |
| 881 | void css_task_iter_end(struct css_task_iter *it); |
| 882 | |
| 883 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *); |
| 884 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from); |
| 885 | |
| 886 | struct cgroup_subsys_state *css_tryget_from_dir(struct dentry *dentry, |
| 887 | struct cgroup_subsys *ss); |
| 888 | |
| 889 | #else /* !CONFIG_CGROUPS */ |
| 890 | |
| 891 | static inline int cgroup_init_early(void) { return 0; } |
| 892 | static inline int cgroup_init(void) { return 0; } |
| 893 | static inline void cgroup_fork(struct task_struct *p) {} |
| 894 | static inline void cgroup_post_fork(struct task_struct *p) {} |
| 895 | static inline void cgroup_exit(struct task_struct *p) {} |
| 896 | |
| 897 | static inline int cgroupstats_build(struct cgroupstats *stats, |
| 898 | struct dentry *dentry) |
| 899 | { |
| 900 | return -EINVAL; |
| 901 | } |
| 902 | |
| 903 | /* No cgroups - nothing to do */ |
| 904 | static inline int cgroup_attach_task_all(struct task_struct *from, |
| 905 | struct task_struct *t) |
| 906 | { |
| 907 | return 0; |
| 908 | } |
| 909 | |
| 910 | #endif /* !CONFIG_CGROUPS */ |
| 911 | |
| 912 | #endif /* _LINUX_CGROUP_H */ |