1/* 2 * linux/cgroup-defs.h - basic definitions for cgroup 3 * 4 * This file provides basic type and interface. Include this file directly 5 * only if necessary to avoid cyclic dependencies. 6 */ 7#ifndef _LINUX_CGROUP_DEFS_H 8#define _LINUX_CGROUP_DEFS_H 9 10#include <linux/limits.h> 11#include <linux/list.h> 12#include <linux/idr.h> 13#include <linux/wait.h> 14#include <linux/mutex.h> 15#include <linux/rcupdate.h> 16#include <linux/percpu-refcount.h> 17#include <linux/percpu-rwsem.h> 18#include <linux/workqueue.h> 19 20#ifdef CONFIG_CGROUPS 21 22struct cgroup; 23struct cgroup_root; 24struct cgroup_subsys; 25struct cgroup_taskset; 26struct kernfs_node; 27struct kernfs_ops; 28struct kernfs_open_file; 29struct seq_file; 30 31#define MAX_CGROUP_TYPE_NAMELEN 32 32#define MAX_CGROUP_ROOT_NAMELEN 64 33#define MAX_CFTYPE_NAME 64 34 35/* define the enumeration of all cgroup subsystems */ 36#define SUBSYS(_x) _x ## _cgrp_id, 37enum cgroup_subsys_id { 38#include <linux/cgroup_subsys.h> 39 CGROUP_SUBSYS_COUNT, 40}; 41#undef SUBSYS 42 43/* bits in struct cgroup_subsys_state flags field */ 44enum { 45 CSS_NO_REF = (1 << 0), /* no reference counting for this css */ 46 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */ 47 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */ 48 CSS_VISIBLE = (1 << 3), /* css is visible to userland */ 49}; 50 51/* bits in struct cgroup flags field */ 52enum { 53 /* Control Group requires release notifications to userspace */ 54 CGRP_NOTIFY_ON_RELEASE, 55 /* 56 * Clone the parent's configuration when creating a new child 57 * cpuset cgroup. For historical reasons, this option can be 58 * specified at mount time and thus is implemented here. 59 */ 60 CGRP_CPUSET_CLONE_CHILDREN, 61}; 62 63/* cgroup_root->flags */ 64enum { 65 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */ 66 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */ 67}; 68 69/* cftype->flags */ 70enum { 71 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */ 72 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */ 73 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */ 74 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */ 75 76 /* internal flags, do not use outside cgroup core proper */ 77 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */ 78 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */ 79}; 80 81/* 82 * cgroup_file is the handle for a file instance created in a cgroup which 83 * is used, for example, to generate file changed notifications. This can 84 * be obtained by setting cftype->file_offset. 85 */ 86struct cgroup_file { 87 /* do not access any fields from outside cgroup core */ 88 struct kernfs_node *kn; 89}; 90 91/* 92 * Per-subsystem/per-cgroup state maintained by the system. This is the 93 * fundamental structural building block that controllers deal with. 94 * 95 * Fields marked with "PI:" are public and immutable and may be accessed 96 * directly without synchronization. 97 */ 98struct cgroup_subsys_state { 99 /* PI: the cgroup that this css is attached to */ 100 struct cgroup *cgroup; 101 102 /* PI: the cgroup subsystem that this css is attached to */ 103 struct cgroup_subsys *ss; 104 105 /* reference count - access via css_[try]get() and css_put() */ 106 struct percpu_ref refcnt; 107 108 /* PI: the parent css */ 109 struct cgroup_subsys_state *parent; 110 111 /* siblings list anchored at the parent's ->children */ 112 struct list_head sibling; 113 struct list_head children; 114 115 /* 116 * PI: Subsys-unique ID. 0 is unused and root is always 1. The 117 * matching css can be looked up using css_from_id(). 118 */ 119 int id; 120 121 unsigned int flags; 122 123 /* 124 * Monotonically increasing unique serial number which defines a 125 * uniform order among all csses. It's guaranteed that all 126 * ->children lists are in the ascending order of ->serial_nr and 127 * used to allow interrupting and resuming iterations. 128 */ 129 u64 serial_nr; 130 131 /* 132 * Incremented by online self and children. Used to guarantee that 133 * parents are not offlined before their children. 134 */ 135 atomic_t online_cnt; 136 137 /* percpu_ref killing and RCU release */ 138 struct rcu_head rcu_head; 139 struct work_struct destroy_work; 140}; 141 142/* 143 * A css_set is a structure holding pointers to a set of 144 * cgroup_subsys_state objects. This saves space in the task struct 145 * object and speeds up fork()/exit(), since a single inc/dec and a 146 * list_add()/del() can bump the reference count on the entire cgroup 147 * set for a task. 148 */ 149struct css_set { 150 /* Reference count */ 151 atomic_t refcount; 152 153 /* 154 * List running through all cgroup groups in the same hash 155 * slot. Protected by css_set_lock 156 */ 157 struct hlist_node hlist; 158 159 /* 160 * Lists running through all tasks using this cgroup group. 161 * mg_tasks lists tasks which belong to this cset but are in the 162 * process of being migrated out or in. Protected by 163 * css_set_rwsem, but, during migration, once tasks are moved to 164 * mg_tasks, it can be read safely while holding cgroup_mutex. 165 */ 166 struct list_head tasks; 167 struct list_head mg_tasks; 168 169 /* 170 * List of cgrp_cset_links pointing at cgroups referenced from this 171 * css_set. Protected by css_set_lock. 172 */ 173 struct list_head cgrp_links; 174 175 /* the default cgroup associated with this css_set */ 176 struct cgroup *dfl_cgrp; 177 178 /* 179 * Set of subsystem states, one for each subsystem. This array is 180 * immutable after creation apart from the init_css_set during 181 * subsystem registration (at boot time). 182 */ 183 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; 184 185 /* 186 * List of csets participating in the on-going migration either as 187 * source or destination. Protected by cgroup_mutex. 188 */ 189 struct list_head mg_preload_node; 190 struct list_head mg_node; 191 192 /* 193 * If this cset is acting as the source of migration the following 194 * two fields are set. mg_src_cgrp and mg_dst_cgrp are 195 * respectively the source and destination cgroups of the on-going 196 * migration. mg_dst_cset is the destination cset the target tasks 197 * on this cset should be migrated to. Protected by cgroup_mutex. 198 */ 199 struct cgroup *mg_src_cgrp; 200 struct cgroup *mg_dst_cgrp; 201 struct css_set *mg_dst_cset; 202 203 /* 204 * On the default hierarhcy, ->subsys[ssid] may point to a css 205 * attached to an ancestor instead of the cgroup this css_set is 206 * associated with. The following node is anchored at 207 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to 208 * iterate through all css's attached to a given cgroup. 209 */ 210 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT]; 211 212 /* all css_task_iters currently walking this cset */ 213 struct list_head task_iters; 214 215 /* dead and being drained, ignore for migration */ 216 bool dead; 217 218 /* For RCU-protected deletion */ 219 struct rcu_head rcu_head; 220}; 221 222struct cgroup { 223 /* self css with NULL ->ss, points back to this cgroup */ 224 struct cgroup_subsys_state self; 225 226 unsigned long flags; /* "unsigned long" so bitops work */ 227 228 /* 229 * idr allocated in-hierarchy ID. 230 * 231 * ID 0 is not used, the ID of the root cgroup is always 1, and a 232 * new cgroup will be assigned with a smallest available ID. 233 * 234 * Allocating/Removing ID must be protected by cgroup_mutex. 235 */ 236 int id; 237 238 /* 239 * The depth this cgroup is at. The root is at depth zero and each 240 * step down the hierarchy increments the level. This along with 241 * ancestor_ids[] can determine whether a given cgroup is a 242 * descendant of another without traversing the hierarchy. 243 */ 244 int level; 245 246 /* 247 * Each non-empty css_set associated with this cgroup contributes 248 * one to populated_cnt. All children with non-zero popuplated_cnt 249 * of their own contribute one. The count is zero iff there's no 250 * task in this cgroup or its subtree. 251 */ 252 int populated_cnt; 253 254 struct kernfs_node *kn; /* cgroup kernfs entry */ 255 struct cgroup_file procs_file; /* handle for "cgroup.procs" */ 256 struct cgroup_file events_file; /* handle for "cgroup.events" */ 257 258 /* 259 * The bitmask of subsystems enabled on the child cgroups. 260 * ->subtree_control is the one configured through 261 * "cgroup.subtree_control" while ->child_ss_mask is the effective 262 * one which may have more subsystems enabled. Controller knobs 263 * are made available iff it's enabled in ->subtree_control. 264 */ 265 u16 subtree_control; 266 u16 subtree_ss_mask; 267 u16 old_subtree_control; 268 u16 old_subtree_ss_mask; 269 270 /* Private pointers for each registered subsystem */ 271 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT]; 272 273 struct cgroup_root *root; 274 275 /* 276 * List of cgrp_cset_links pointing at css_sets with tasks in this 277 * cgroup. Protected by css_set_lock. 278 */ 279 struct list_head cset_links; 280 281 /* 282 * On the default hierarchy, a css_set for a cgroup with some 283 * susbsys disabled will point to css's which are associated with 284 * the closest ancestor which has the subsys enabled. The 285 * following lists all css_sets which point to this cgroup's css 286 * for the given subsystem. 287 */ 288 struct list_head e_csets[CGROUP_SUBSYS_COUNT]; 289 290 /* 291 * list of pidlists, up to two for each namespace (one for procs, one 292 * for tasks); created on demand. 293 */ 294 struct list_head pidlists; 295 struct mutex pidlist_mutex; 296 297 /* used to wait for offlining of csses */ 298 wait_queue_head_t offline_waitq; 299 300 /* used to schedule release agent */ 301 struct work_struct release_agent_work; 302 303 /* ids of the ancestors at each level including self */ 304 int ancestor_ids[]; 305}; 306 307/* 308 * A cgroup_root represents the root of a cgroup hierarchy, and may be 309 * associated with a kernfs_root to form an active hierarchy. This is 310 * internal to cgroup core. Don't access directly from controllers. 311 */ 312struct cgroup_root { 313 struct kernfs_root *kf_root; 314 315 /* The bitmask of subsystems attached to this hierarchy */ 316 unsigned int subsys_mask; 317 318 /* Unique id for this hierarchy. */ 319 int hierarchy_id; 320 321 /* The root cgroup. Root is destroyed on its release. */ 322 struct cgroup cgrp; 323 324 /* for cgrp->ancestor_ids[0] */ 325 int cgrp_ancestor_id_storage; 326 327 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */ 328 atomic_t nr_cgrps; 329 330 /* A list running through the active hierarchies */ 331 struct list_head root_list; 332 333 /* Hierarchy-specific flags */ 334 unsigned int flags; 335 336 /* IDs for cgroups in this hierarchy */ 337 struct idr cgroup_idr; 338 339 /* The path to use for release notifications. */ 340 char release_agent_path[PATH_MAX]; 341 342 /* The name for this hierarchy - may be empty */ 343 char name[MAX_CGROUP_ROOT_NAMELEN]; 344}; 345 346/* 347 * struct cftype: handler definitions for cgroup control files 348 * 349 * When reading/writing to a file: 350 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata 351 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata 352 */ 353struct cftype { 354 /* 355 * By convention, the name should begin with the name of the 356 * subsystem, followed by a period. Zero length string indicates 357 * end of cftype array. 358 */ 359 char name[MAX_CFTYPE_NAME]; 360 unsigned long private; 361 362 /* 363 * The maximum length of string, excluding trailing nul, that can 364 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed. 365 */ 366 size_t max_write_len; 367 368 /* CFTYPE_* flags */ 369 unsigned int flags; 370 371 /* 372 * If non-zero, should contain the offset from the start of css to 373 * a struct cgroup_file field. cgroup will record the handle of 374 * the created file into it. The recorded handle can be used as 375 * long as the containing css remains accessible. 376 */ 377 unsigned int file_offset; 378 379 /* 380 * Fields used for internal bookkeeping. Initialized automatically 381 * during registration. 382 */ 383 struct cgroup_subsys *ss; /* NULL for cgroup core files */ 384 struct list_head node; /* anchored at ss->cfts */ 385 struct kernfs_ops *kf_ops; 386 387 /* 388 * read_u64() is a shortcut for the common case of returning a 389 * single integer. Use it in place of read() 390 */ 391 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft); 392 /* 393 * read_s64() is a signed version of read_u64() 394 */ 395 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft); 396 397 /* generic seq_file read interface */ 398 int (*seq_show)(struct seq_file *sf, void *v); 399 400 /* optional ops, implement all or none */ 401 void *(*seq_start)(struct seq_file *sf, loff_t *ppos); 402 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos); 403 void (*seq_stop)(struct seq_file *sf, void *v); 404 405 /* 406 * write_u64() is a shortcut for the common case of accepting 407 * a single integer (as parsed by simple_strtoull) from 408 * userspace. Use in place of write(); return 0 or error. 409 */ 410 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft, 411 u64 val); 412 /* 413 * write_s64() is a signed version of write_u64() 414 */ 415 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft, 416 s64 val); 417 418 /* 419 * write() is the generic write callback which maps directly to 420 * kernfs write operation and overrides all other operations. 421 * Maximum write size is determined by ->max_write_len. Use 422 * of_css/cft() to access the associated css and cft. 423 */ 424 ssize_t (*write)(struct kernfs_open_file *of, 425 char *buf, size_t nbytes, loff_t off); 426 427#ifdef CONFIG_DEBUG_LOCK_ALLOC 428 struct lock_class_key lockdep_key; 429#endif 430}; 431 432/* 433 * Control Group subsystem type. 434 * See Documentation/cgroups/cgroups.txt for details 435 */ 436struct cgroup_subsys { 437 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css); 438 int (*css_online)(struct cgroup_subsys_state *css); 439 void (*css_offline)(struct cgroup_subsys_state *css); 440 void (*css_released)(struct cgroup_subsys_state *css); 441 void (*css_free)(struct cgroup_subsys_state *css); 442 void (*css_reset)(struct cgroup_subsys_state *css); 443 444 int (*can_attach)(struct cgroup_taskset *tset); 445 void (*cancel_attach)(struct cgroup_taskset *tset); 446 void (*attach)(struct cgroup_taskset *tset); 447 void (*post_attach)(void); 448 int (*can_fork)(struct task_struct *task); 449 void (*cancel_fork)(struct task_struct *task); 450 void (*fork)(struct task_struct *task); 451 void (*exit)(struct task_struct *task); 452 void (*free)(struct task_struct *task); 453 void (*bind)(struct cgroup_subsys_state *root_css); 454 455 bool early_init:1; 456 457 /* 458 * If %true, the controller, on the default hierarchy, doesn't show 459 * up in "cgroup.controllers" or "cgroup.subtree_control", is 460 * implicitly enabled on all cgroups on the default hierarchy, and 461 * bypasses the "no internal process" constraint. This is for 462 * utility type controllers which is transparent to userland. 463 * 464 * An implicit controller can be stolen from the default hierarchy 465 * anytime and thus must be okay with offline csses from previous 466 * hierarchies coexisting with csses for the current one. 467 */ 468 bool implicit_on_dfl:1; 469 470 /* 471 * If %false, this subsystem is properly hierarchical - 472 * configuration, resource accounting and restriction on a parent 473 * cgroup cover those of its children. If %true, hierarchy support 474 * is broken in some ways - some subsystems ignore hierarchy 475 * completely while others are only implemented half-way. 476 * 477 * It's now disallowed to create nested cgroups if the subsystem is 478 * broken and cgroup core will emit a warning message on such 479 * cases. Eventually, all subsystems will be made properly 480 * hierarchical and this will go away. 481 */ 482 bool broken_hierarchy:1; 483 bool warned_broken_hierarchy:1; 484 485 /* the following two fields are initialized automtically during boot */ 486 int id; 487 const char *name; 488 489 /* optional, initialized automatically during boot if not set */ 490 const char *legacy_name; 491 492 /* link to parent, protected by cgroup_lock() */ 493 struct cgroup_root *root; 494 495 /* idr for css->id */ 496 struct idr css_idr; 497 498 /* 499 * List of cftypes. Each entry is the first entry of an array 500 * terminated by zero length name. 501 */ 502 struct list_head cfts; 503 504 /* 505 * Base cftypes which are automatically registered. The two can 506 * point to the same array. 507 */ 508 struct cftype *dfl_cftypes; /* for the default hierarchy */ 509 struct cftype *legacy_cftypes; /* for the legacy hierarchies */ 510 511 /* 512 * A subsystem may depend on other subsystems. When such subsystem 513 * is enabled on a cgroup, the depended-upon subsystems are enabled 514 * together if available. Subsystems enabled due to dependency are 515 * not visible to userland until explicitly enabled. The following 516 * specifies the mask of subsystems that this one depends on. 517 */ 518 unsigned int depends_on; 519}; 520 521extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem; 522 523/** 524 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups 525 * @tsk: target task 526 * 527 * Called from threadgroup_change_begin() and allows cgroup operations to 528 * synchronize against threadgroup changes using a percpu_rw_semaphore. 529 */ 530static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) 531{ 532 percpu_down_read(&cgroup_threadgroup_rwsem); 533} 534 535/** 536 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups 537 * @tsk: target task 538 * 539 * Called from threadgroup_change_end(). Counterpart of 540 * cgroup_threadcgroup_change_begin(). 541 */ 542static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) 543{ 544 percpu_up_read(&cgroup_threadgroup_rwsem); 545} 546 547#else /* CONFIG_CGROUPS */ 548 549#define CGROUP_SUBSYS_COUNT 0 550 551static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) {} 552static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {} 553 554#endif /* CONFIG_CGROUPS */ 555 556#ifdef CONFIG_SOCK_CGROUP_DATA 557 558/* 559 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains 560 * per-socket cgroup information except for memcg association. 561 * 562 * On legacy hierarchies, net_prio and net_cls controllers directly set 563 * attributes on each sock which can then be tested by the network layer. 564 * On the default hierarchy, each sock is associated with the cgroup it was 565 * created in and the networking layer can match the cgroup directly. 566 * 567 * To avoid carrying all three cgroup related fields separately in sock, 568 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer. 569 * On boot, sock_cgroup_data records the cgroup that the sock was created 570 * in so that cgroup2 matches can be made; however, once either net_prio or 571 * net_cls starts being used, the area is overriden to carry prioidx and/or 572 * classid. The two modes are distinguished by whether the lowest bit is 573 * set. Clear bit indicates cgroup pointer while set bit prioidx and 574 * classid. 575 * 576 * While userland may start using net_prio or net_cls at any time, once 577 * either is used, cgroup2 matching no longer works. There is no reason to 578 * mix the two and this is in line with how legacy and v2 compatibility is 579 * handled. On mode switch, cgroup references which are already being 580 * pointed to by socks may be leaked. While this can be remedied by adding 581 * synchronization around sock_cgroup_data, given that the number of leaked 582 * cgroups is bound and highly unlikely to be high, this seems to be the 583 * better trade-off. 584 */ 585struct sock_cgroup_data { 586 union { 587#ifdef __LITTLE_ENDIAN 588 struct { 589 u8 is_data; 590 u8 padding; 591 u16 prioidx; 592 u32 classid; 593 } __packed; 594#else 595 struct { 596 u32 classid; 597 u16 prioidx; 598 u8 padding; 599 u8 is_data; 600 } __packed; 601#endif 602 u64 val; 603 }; 604}; 605 606/* 607 * There's a theoretical window where the following accessors race with 608 * updaters and return part of the previous pointer as the prioidx or 609 * classid. Such races are short-lived and the result isn't critical. 610 */ 611static inline u16 sock_cgroup_prioidx(struct sock_cgroup_data *skcd) 612{ 613 /* fallback to 1 which is always the ID of the root cgroup */ 614 return (skcd->is_data & 1) ? skcd->prioidx : 1; 615} 616 617static inline u32 sock_cgroup_classid(struct sock_cgroup_data *skcd) 618{ 619 /* fallback to 0 which is the unconfigured default classid */ 620 return (skcd->is_data & 1) ? skcd->classid : 0; 621} 622 623/* 624 * If invoked concurrently, the updaters may clobber each other. The 625 * caller is responsible for synchronization. 626 */ 627static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd, 628 u16 prioidx) 629{ 630 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }}; 631 632 if (sock_cgroup_prioidx(&skcd_buf) == prioidx) 633 return; 634 635 if (!(skcd_buf.is_data & 1)) { 636 skcd_buf.val = 0; 637 skcd_buf.is_data = 1; 638 } 639 640 skcd_buf.prioidx = prioidx; 641 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */ 642} 643 644static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd, 645 u32 classid) 646{ 647 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }}; 648 649 if (sock_cgroup_classid(&skcd_buf) == classid) 650 return; 651 652 if (!(skcd_buf.is_data & 1)) { 653 skcd_buf.val = 0; 654 skcd_buf.is_data = 1; 655 } 656 657 skcd_buf.classid = classid; 658 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */ 659} 660 661#else /* CONFIG_SOCK_CGROUP_DATA */ 662 663struct sock_cgroup_data { 664}; 665 666#endif /* CONFIG_SOCK_CGROUP_DATA */ 667 668#endif /* _LINUX_CGROUP_DEFS_H */ 669