1/* SPDX-License-Identifier: GPL-2.0 */ 2/* 3 * linux/cgroup-defs.h - basic definitions for cgroup 4 * 5 * This file provides basic type and interface. Include this file directly 6 * only if necessary to avoid cyclic dependencies. 7 */ 8#ifndef _LINUX_CGROUP_DEFS_H 9#define _LINUX_CGROUP_DEFS_H 10 11#include <linux/limits.h> 12#include <linux/list.h> 13#include <linux/idr.h> 14#include <linux/wait.h> 15#include <linux/mutex.h> 16#include <linux/rcupdate.h> 17#include <linux/refcount.h> 18#include <linux/percpu-refcount.h> 19#include <linux/percpu-rwsem.h> 20#include <linux/u64_stats_sync.h> 21#include <linux/workqueue.h> 22#include <linux/bpf-cgroup.h> 23 24#ifdef CONFIG_CGROUPS 25 26struct cgroup; 27struct cgroup_root; 28struct cgroup_subsys; 29struct cgroup_taskset; 30struct kernfs_node; 31struct kernfs_ops; 32struct kernfs_open_file; 33struct seq_file; 34 35#define MAX_CGROUP_TYPE_NAMELEN 32 36#define MAX_CGROUP_ROOT_NAMELEN 64 37#define MAX_CFTYPE_NAME 64 38 39/* define the enumeration of all cgroup subsystems */ 40#define SUBSYS(_x) _x ## _cgrp_id, 41enum cgroup_subsys_id { 42#include <linux/cgroup_subsys.h> 43 CGROUP_SUBSYS_COUNT, 44}; 45#undef SUBSYS 46 47/* bits in struct cgroup_subsys_state flags field */ 48enum { 49 CSS_NO_REF = (1 << 0), /* no reference counting for this css */ 50 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */ 51 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */ 52 CSS_VISIBLE = (1 << 3), /* css is visible to userland */ 53 CSS_DYING = (1 << 4), /* css is dying */ 54}; 55 56/* bits in struct cgroup flags field */ 57enum { 58 /* Control Group requires release notifications to userspace */ 59 CGRP_NOTIFY_ON_RELEASE, 60 /* 61 * Clone the parent's configuration when creating a new child 62 * cpuset cgroup. For historical reasons, this option can be 63 * specified at mount time and thus is implemented here. 64 */ 65 CGRP_CPUSET_CLONE_CHILDREN, 66}; 67 68/* cgroup_root->flags */ 69enum { 70 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */ 71 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */ 72 73 /* 74 * Consider namespaces as delegation boundaries. If this flag is 75 * set, controller specific interface files in a namespace root 76 * aren't writeable from inside the namespace. 77 */ 78 CGRP_ROOT_NS_DELEGATE = (1 << 3), 79 80 /* 81 * Enable cpuset controller in v1 cgroup to use v2 behavior. 82 */ 83 CGRP_ROOT_CPUSET_V2_MODE = (1 << 4), 84}; 85 86/* cftype->flags */ 87enum { 88 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */ 89 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */ 90 CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */ 91 92 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */ 93 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */ 94 95 /* internal flags, do not use outside cgroup core proper */ 96 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */ 97 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */ 98}; 99 100/* 101 * cgroup_file is the handle for a file instance created in a cgroup which 102 * is used, for example, to generate file changed notifications. This can 103 * be obtained by setting cftype->file_offset. 104 */ 105struct cgroup_file { 106 /* do not access any fields from outside cgroup core */ 107 struct kernfs_node *kn; 108}; 109 110/* 111 * Per-subsystem/per-cgroup state maintained by the system. This is the 112 * fundamental structural building block that controllers deal with. 113 * 114 * Fields marked with "PI:" are public and immutable and may be accessed 115 * directly without synchronization. 116 */ 117struct cgroup_subsys_state { 118 /* PI: the cgroup that this css is attached to */ 119 struct cgroup *cgroup; 120 121 /* PI: the cgroup subsystem that this css is attached to */ 122 struct cgroup_subsys *ss; 123 124 /* reference count - access via css_[try]get() and css_put() */ 125 struct percpu_ref refcnt; 126 127 /* siblings list anchored at the parent's ->children */ 128 struct list_head sibling; 129 struct list_head children; 130 131 /* 132 * PI: Subsys-unique ID. 0 is unused and root is always 1. The 133 * matching css can be looked up using css_from_id(). 134 */ 135 int id; 136 137 unsigned int flags; 138 139 /* 140 * Monotonically increasing unique serial number which defines a 141 * uniform order among all csses. It's guaranteed that all 142 * ->children lists are in the ascending order of ->serial_nr and 143 * used to allow interrupting and resuming iterations. 144 */ 145 u64 serial_nr; 146 147 /* 148 * Incremented by online self and children. Used to guarantee that 149 * parents are not offlined before their children. 150 */ 151 atomic_t online_cnt; 152 153 /* percpu_ref killing and RCU release */ 154 struct work_struct destroy_work; 155 struct rcu_work destroy_rwork; 156 157 /* 158 * PI: the parent css. Placed here for cache proximity to following 159 * fields of the containing structure. 160 */ 161 struct cgroup_subsys_state *parent; 162}; 163 164/* 165 * A css_set is a structure holding pointers to a set of 166 * cgroup_subsys_state objects. This saves space in the task struct 167 * object and speeds up fork()/exit(), since a single inc/dec and a 168 * list_add()/del() can bump the reference count on the entire cgroup 169 * set for a task. 170 */ 171struct css_set { 172 /* 173 * Set of subsystem states, one for each subsystem. This array is 174 * immutable after creation apart from the init_css_set during 175 * subsystem registration (at boot time). 176 */ 177 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; 178 179 /* reference count */ 180 refcount_t refcount; 181 182 /* 183 * For a domain cgroup, the following points to self. If threaded, 184 * to the matching cset of the nearest domain ancestor. The 185 * dom_cset provides access to the domain cgroup and its csses to 186 * which domain level resource consumptions should be charged. 187 */ 188 struct css_set *dom_cset; 189 190 /* the default cgroup associated with this css_set */ 191 struct cgroup *dfl_cgrp; 192 193 /* internal task count, protected by css_set_lock */ 194 int nr_tasks; 195 196 /* 197 * Lists running through all tasks using this cgroup group. 198 * mg_tasks lists tasks which belong to this cset but are in the 199 * process of being migrated out or in. Protected by 200 * css_set_rwsem, but, during migration, once tasks are moved to 201 * mg_tasks, it can be read safely while holding cgroup_mutex. 202 */ 203 struct list_head tasks; 204 struct list_head mg_tasks; 205 206 /* all css_task_iters currently walking this cset */ 207 struct list_head task_iters; 208 209 /* 210 * On the default hierarhcy, ->subsys[ssid] may point to a css 211 * attached to an ancestor instead of the cgroup this css_set is 212 * associated with. The following node is anchored at 213 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to 214 * iterate through all css's attached to a given cgroup. 215 */ 216 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT]; 217 218 /* all threaded csets whose ->dom_cset points to this cset */ 219 struct list_head threaded_csets; 220 struct list_head threaded_csets_node; 221 222 /* 223 * List running through all cgroup groups in the same hash 224 * slot. Protected by css_set_lock 225 */ 226 struct hlist_node hlist; 227 228 /* 229 * List of cgrp_cset_links pointing at cgroups referenced from this 230 * css_set. Protected by css_set_lock. 231 */ 232 struct list_head cgrp_links; 233 234 /* 235 * List of csets participating in the on-going migration either as 236 * source or destination. Protected by cgroup_mutex. 237 */ 238 struct list_head mg_preload_node; 239 struct list_head mg_node; 240 241 /* 242 * If this cset is acting as the source of migration the following 243 * two fields are set. mg_src_cgrp and mg_dst_cgrp are 244 * respectively the source and destination cgroups of the on-going 245 * migration. mg_dst_cset is the destination cset the target tasks 246 * on this cset should be migrated to. Protected by cgroup_mutex. 247 */ 248 struct cgroup *mg_src_cgrp; 249 struct cgroup *mg_dst_cgrp; 250 struct css_set *mg_dst_cset; 251 252 /* dead and being drained, ignore for migration */ 253 bool dead; 254 255 /* For RCU-protected deletion */ 256 struct rcu_head rcu_head; 257}; 258 259/* 260 * cgroup basic resource usage statistics. Accounting is done per-cpu in 261 * cgroup_cpu_stat which is then lazily propagated up the hierarchy on 262 * reads. 263 * 264 * When a stat gets updated, the cgroup_cpu_stat and its ancestors are 265 * linked into the updated tree. On the following read, propagation only 266 * considers and consumes the updated tree. This makes reading O(the 267 * number of descendants which have been active since last read) instead of 268 * O(the total number of descendants). 269 * 270 * This is important because there can be a lot of (draining) cgroups which 271 * aren't active and stat may be read frequently. The combination can 272 * become very expensive. By propagating selectively, increasing reading 273 * frequency decreases the cost of each read. 274 */ 275struct cgroup_cpu_stat { 276 /* 277 * ->sync protects all the current counters. These are the only 278 * fields which get updated in the hot path. 279 */ 280 struct u64_stats_sync sync; 281 struct task_cputime cputime; 282 283 /* 284 * Snapshots at the last reading. These are used to calculate the 285 * deltas to propagate to the global counters. 286 */ 287 struct task_cputime last_cputime; 288 289 /* 290 * Child cgroups with stat updates on this cpu since the last read 291 * are linked on the parent's ->updated_children through 292 * ->updated_next. 293 * 294 * In addition to being more compact, singly-linked list pointing 295 * to the cgroup makes it unnecessary for each per-cpu struct to 296 * point back to the associated cgroup. 297 * 298 * Protected by per-cpu cgroup_cpu_stat_lock. 299 */ 300 struct cgroup *updated_children; /* terminated by self cgroup */ 301 struct cgroup *updated_next; /* NULL iff not on the list */ 302}; 303 304struct cgroup_stat { 305 /* per-cpu statistics are collected into the folowing global counters */ 306 struct task_cputime cputime; 307 struct prev_cputime prev_cputime; 308}; 309 310struct cgroup { 311 /* self css with NULL ->ss, points back to this cgroup */ 312 struct cgroup_subsys_state self; 313 314 unsigned long flags; /* "unsigned long" so bitops work */ 315 316 /* 317 * idr allocated in-hierarchy ID. 318 * 319 * ID 0 is not used, the ID of the root cgroup is always 1, and a 320 * new cgroup will be assigned with a smallest available ID. 321 * 322 * Allocating/Removing ID must be protected by cgroup_mutex. 323 */ 324 int id; 325 326 /* 327 * The depth this cgroup is at. The root is at depth zero and each 328 * step down the hierarchy increments the level. This along with 329 * ancestor_ids[] can determine whether a given cgroup is a 330 * descendant of another without traversing the hierarchy. 331 */ 332 int level; 333 334 /* Maximum allowed descent tree depth */ 335 int max_depth; 336 337 /* 338 * Keep track of total numbers of visible and dying descent cgroups. 339 * Dying cgroups are cgroups which were deleted by a user, 340 * but are still existing because someone else is holding a reference. 341 * max_descendants is a maximum allowed number of descent cgroups. 342 */ 343 int nr_descendants; 344 int nr_dying_descendants; 345 int max_descendants; 346 347 /* 348 * Each non-empty css_set associated with this cgroup contributes 349 * one to nr_populated_csets. The counter is zero iff this cgroup 350 * doesn't have any tasks. 351 * 352 * All children which have non-zero nr_populated_csets and/or 353 * nr_populated_children of their own contribute one to either 354 * nr_populated_domain_children or nr_populated_threaded_children 355 * depending on their type. Each counter is zero iff all cgroups 356 * of the type in the subtree proper don't have any tasks. 357 */ 358 int nr_populated_csets; 359 int nr_populated_domain_children; 360 int nr_populated_threaded_children; 361 362 int nr_threaded_children; /* # of live threaded child cgroups */ 363 364 struct kernfs_node *kn; /* cgroup kernfs entry */ 365 struct cgroup_file procs_file; /* handle for "cgroup.procs" */ 366 struct cgroup_file events_file; /* handle for "cgroup.events" */ 367 368 /* 369 * The bitmask of subsystems enabled on the child cgroups. 370 * ->subtree_control is the one configured through 371 * "cgroup.subtree_control" while ->child_ss_mask is the effective 372 * one which may have more subsystems enabled. Controller knobs 373 * are made available iff it's enabled in ->subtree_control. 374 */ 375 u16 subtree_control; 376 u16 subtree_ss_mask; 377 u16 old_subtree_control; 378 u16 old_subtree_ss_mask; 379 380 /* Private pointers for each registered subsystem */ 381 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT]; 382 383 struct cgroup_root *root; 384 385 /* 386 * List of cgrp_cset_links pointing at css_sets with tasks in this 387 * cgroup. Protected by css_set_lock. 388 */ 389 struct list_head cset_links; 390 391 /* 392 * On the default hierarchy, a css_set for a cgroup with some 393 * susbsys disabled will point to css's which are associated with 394 * the closest ancestor which has the subsys enabled. The 395 * following lists all css_sets which point to this cgroup's css 396 * for the given subsystem. 397 */ 398 struct list_head e_csets[CGROUP_SUBSYS_COUNT]; 399 400 /* 401 * If !threaded, self. If threaded, it points to the nearest 402 * domain ancestor. Inside a threaded subtree, cgroups are exempt 403 * from process granularity and no-internal-task constraint. 404 * Domain level resource consumptions which aren't tied to a 405 * specific task are charged to the dom_cgrp. 406 */ 407 struct cgroup *dom_cgrp; 408 409 /* cgroup basic resource statistics */ 410 struct cgroup_cpu_stat __percpu *cpu_stat; 411 struct cgroup_stat pending_stat; /* pending from children */ 412 struct cgroup_stat stat; 413 414 /* 415 * list of pidlists, up to two for each namespace (one for procs, one 416 * for tasks); created on demand. 417 */ 418 struct list_head pidlists; 419 struct mutex pidlist_mutex; 420 421 /* used to wait for offlining of csses */ 422 wait_queue_head_t offline_waitq; 423 424 /* used to schedule release agent */ 425 struct work_struct release_agent_work; 426 427 /* used to store eBPF programs */ 428 struct cgroup_bpf bpf; 429 430 /* ids of the ancestors at each level including self */ 431 int ancestor_ids[]; 432}; 433 434/* 435 * A cgroup_root represents the root of a cgroup hierarchy, and may be 436 * associated with a kernfs_root to form an active hierarchy. This is 437 * internal to cgroup core. Don't access directly from controllers. 438 */ 439struct cgroup_root { 440 struct kernfs_root *kf_root; 441 442 /* The bitmask of subsystems attached to this hierarchy */ 443 unsigned int subsys_mask; 444 445 /* Unique id for this hierarchy. */ 446 int hierarchy_id; 447 448 /* The root cgroup. Root is destroyed on its release. */ 449 struct cgroup cgrp; 450 451 /* for cgrp->ancestor_ids[0] */ 452 int cgrp_ancestor_id_storage; 453 454 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */ 455 atomic_t nr_cgrps; 456 457 /* A list running through the active hierarchies */ 458 struct list_head root_list; 459 460 /* Hierarchy-specific flags */ 461 unsigned int flags; 462 463 /* IDs for cgroups in this hierarchy */ 464 struct idr cgroup_idr; 465 466 /* The path to use for release notifications. */ 467 char release_agent_path[PATH_MAX]; 468 469 /* The name for this hierarchy - may be empty */ 470 char name[MAX_CGROUP_ROOT_NAMELEN]; 471}; 472 473/* 474 * struct cftype: handler definitions for cgroup control files 475 * 476 * When reading/writing to a file: 477 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata 478 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata 479 */ 480struct cftype { 481 /* 482 * By convention, the name should begin with the name of the 483 * subsystem, followed by a period. Zero length string indicates 484 * end of cftype array. 485 */ 486 char name[MAX_CFTYPE_NAME]; 487 unsigned long private; 488 489 /* 490 * The maximum length of string, excluding trailing nul, that can 491 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed. 492 */ 493 size_t max_write_len; 494 495 /* CFTYPE_* flags */ 496 unsigned int flags; 497 498 /* 499 * If non-zero, should contain the offset from the start of css to 500 * a struct cgroup_file field. cgroup will record the handle of 501 * the created file into it. The recorded handle can be used as 502 * long as the containing css remains accessible. 503 */ 504 unsigned int file_offset; 505 506 /* 507 * Fields used for internal bookkeeping. Initialized automatically 508 * during registration. 509 */ 510 struct cgroup_subsys *ss; /* NULL for cgroup core files */ 511 struct list_head node; /* anchored at ss->cfts */ 512 struct kernfs_ops *kf_ops; 513 514 int (*open)(struct kernfs_open_file *of); 515 void (*release)(struct kernfs_open_file *of); 516 517 /* 518 * read_u64() is a shortcut for the common case of returning a 519 * single integer. Use it in place of read() 520 */ 521 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft); 522 /* 523 * read_s64() is a signed version of read_u64() 524 */ 525 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft); 526 527 /* generic seq_file read interface */ 528 int (*seq_show)(struct seq_file *sf, void *v); 529 530 /* optional ops, implement all or none */ 531 void *(*seq_start)(struct seq_file *sf, loff_t *ppos); 532 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos); 533 void (*seq_stop)(struct seq_file *sf, void *v); 534 535 /* 536 * write_u64() is a shortcut for the common case of accepting 537 * a single integer (as parsed by simple_strtoull) from 538 * userspace. Use in place of write(); return 0 or error. 539 */ 540 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft, 541 u64 val); 542 /* 543 * write_s64() is a signed version of write_u64() 544 */ 545 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft, 546 s64 val); 547 548 /* 549 * write() is the generic write callback which maps directly to 550 * kernfs write operation and overrides all other operations. 551 * Maximum write size is determined by ->max_write_len. Use 552 * of_css/cft() to access the associated css and cft. 553 */ 554 ssize_t (*write)(struct kernfs_open_file *of, 555 char *buf, size_t nbytes, loff_t off); 556 557#ifdef CONFIG_DEBUG_LOCK_ALLOC 558 struct lock_class_key lockdep_key; 559#endif 560}; 561 562/* 563 * Control Group subsystem type. 564 * See Documentation/cgroup-v1/cgroups.txt for details 565 */ 566struct cgroup_subsys { 567 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css); 568 int (*css_online)(struct cgroup_subsys_state *css); 569 void (*css_offline)(struct cgroup_subsys_state *css); 570 void (*css_released)(struct cgroup_subsys_state *css); 571 void (*css_free)(struct cgroup_subsys_state *css); 572 void (*css_reset)(struct cgroup_subsys_state *css); 573 int (*css_extra_stat_show)(struct seq_file *seq, 574 struct cgroup_subsys_state *css); 575 576 int (*can_attach)(struct cgroup_taskset *tset); 577 void (*cancel_attach)(struct cgroup_taskset *tset); 578 void (*attach)(struct cgroup_taskset *tset); 579 void (*post_attach)(void); 580 int (*can_fork)(struct task_struct *task); 581 void (*cancel_fork)(struct task_struct *task); 582 void (*fork)(struct task_struct *task); 583 void (*exit)(struct task_struct *task); 584 void (*free)(struct task_struct *task); 585 void (*bind)(struct cgroup_subsys_state *root_css); 586 587 bool early_init:1; 588 589 /* 590 * If %true, the controller, on the default hierarchy, doesn't show 591 * up in "cgroup.controllers" or "cgroup.subtree_control", is 592 * implicitly enabled on all cgroups on the default hierarchy, and 593 * bypasses the "no internal process" constraint. This is for 594 * utility type controllers which is transparent to userland. 595 * 596 * An implicit controller can be stolen from the default hierarchy 597 * anytime and thus must be okay with offline csses from previous 598 * hierarchies coexisting with csses for the current one. 599 */ 600 bool implicit_on_dfl:1; 601 602 /* 603 * If %true, the controller, supports threaded mode on the default 604 * hierarchy. In a threaded subtree, both process granularity and 605 * no-internal-process constraint are ignored and a threaded 606 * controllers should be able to handle that. 607 * 608 * Note that as an implicit controller is automatically enabled on 609 * all cgroups on the default hierarchy, it should also be 610 * threaded. implicit && !threaded is not supported. 611 */ 612 bool threaded:1; 613 614 /* 615 * If %false, this subsystem is properly hierarchical - 616 * configuration, resource accounting and restriction on a parent 617 * cgroup cover those of its children. If %true, hierarchy support 618 * is broken in some ways - some subsystems ignore hierarchy 619 * completely while others are only implemented half-way. 620 * 621 * It's now disallowed to create nested cgroups if the subsystem is 622 * broken and cgroup core will emit a warning message on such 623 * cases. Eventually, all subsystems will be made properly 624 * hierarchical and this will go away. 625 */ 626 bool broken_hierarchy:1; 627 bool warned_broken_hierarchy:1; 628 629 /* the following two fields are initialized automtically during boot */ 630 int id; 631 const char *name; 632 633 /* optional, initialized automatically during boot if not set */ 634 const char *legacy_name; 635 636 /* link to parent, protected by cgroup_lock() */ 637 struct cgroup_root *root; 638 639 /* idr for css->id */ 640 struct idr css_idr; 641 642 /* 643 * List of cftypes. Each entry is the first entry of an array 644 * terminated by zero length name. 645 */ 646 struct list_head cfts; 647 648 /* 649 * Base cftypes which are automatically registered. The two can 650 * point to the same array. 651 */ 652 struct cftype *dfl_cftypes; /* for the default hierarchy */ 653 struct cftype *legacy_cftypes; /* for the legacy hierarchies */ 654 655 /* 656 * A subsystem may depend on other subsystems. When such subsystem 657 * is enabled on a cgroup, the depended-upon subsystems are enabled 658 * together if available. Subsystems enabled due to dependency are 659 * not visible to userland until explicitly enabled. The following 660 * specifies the mask of subsystems that this one depends on. 661 */ 662 unsigned int depends_on; 663}; 664 665extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem; 666 667/** 668 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups 669 * @tsk: target task 670 * 671 * Allows cgroup operations to synchronize against threadgroup changes 672 * using a percpu_rw_semaphore. 673 */ 674static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) 675{ 676 percpu_down_read(&cgroup_threadgroup_rwsem); 677} 678 679/** 680 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups 681 * @tsk: target task 682 * 683 * Counterpart of cgroup_threadcgroup_change_begin(). 684 */ 685static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) 686{ 687 percpu_up_read(&cgroup_threadgroup_rwsem); 688} 689 690#else /* CONFIG_CGROUPS */ 691 692#define CGROUP_SUBSYS_COUNT 0 693 694static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) 695{ 696 might_sleep(); 697} 698 699static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {} 700 701#endif /* CONFIG_CGROUPS */ 702 703#ifdef CONFIG_SOCK_CGROUP_DATA 704 705/* 706 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains 707 * per-socket cgroup information except for memcg association. 708 * 709 * On legacy hierarchies, net_prio and net_cls controllers directly set 710 * attributes on each sock which can then be tested by the network layer. 711 * On the default hierarchy, each sock is associated with the cgroup it was 712 * created in and the networking layer can match the cgroup directly. 713 * 714 * To avoid carrying all three cgroup related fields separately in sock, 715 * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer. 716 * On boot, sock_cgroup_data records the cgroup that the sock was created 717 * in so that cgroup2 matches can be made; however, once either net_prio or 718 * net_cls starts being used, the area is overriden to carry prioidx and/or 719 * classid. The two modes are distinguished by whether the lowest bit is 720 * set. Clear bit indicates cgroup pointer while set bit prioidx and 721 * classid. 722 * 723 * While userland may start using net_prio or net_cls at any time, once 724 * either is used, cgroup2 matching no longer works. There is no reason to 725 * mix the two and this is in line with how legacy and v2 compatibility is 726 * handled. On mode switch, cgroup references which are already being 727 * pointed to by socks may be leaked. While this can be remedied by adding 728 * synchronization around sock_cgroup_data, given that the number of leaked 729 * cgroups is bound and highly unlikely to be high, this seems to be the 730 * better trade-off. 731 */ 732struct sock_cgroup_data { 733 union { 734#ifdef __LITTLE_ENDIAN 735 struct { 736 u8 is_data; 737 u8 padding; 738 u16 prioidx; 739 u32 classid; 740 } __packed; 741#else 742 struct { 743 u32 classid; 744 u16 prioidx; 745 u8 padding; 746 u8 is_data; 747 } __packed; 748#endif 749 u64 val; 750 }; 751}; 752 753/* 754 * There's a theoretical window where the following accessors race with 755 * updaters and return part of the previous pointer as the prioidx or 756 * classid. Such races are short-lived and the result isn't critical. 757 */ 758static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd) 759{ 760 /* fallback to 1 which is always the ID of the root cgroup */ 761 return (skcd->is_data & 1) ? skcd->prioidx : 1; 762} 763 764static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd) 765{ 766 /* fallback to 0 which is the unconfigured default classid */ 767 return (skcd->is_data & 1) ? skcd->classid : 0; 768} 769 770/* 771 * If invoked concurrently, the updaters may clobber each other. The 772 * caller is responsible for synchronization. 773 */ 774static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd, 775 u16 prioidx) 776{ 777 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }}; 778 779 if (sock_cgroup_prioidx(&skcd_buf) == prioidx) 780 return; 781 782 if (!(skcd_buf.is_data & 1)) { 783 skcd_buf.val = 0; 784 skcd_buf.is_data = 1; 785 } 786 787 skcd_buf.prioidx = prioidx; 788 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */ 789} 790 791static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd, 792 u32 classid) 793{ 794 struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }}; 795 796 if (sock_cgroup_classid(&skcd_buf) == classid) 797 return; 798 799 if (!(skcd_buf.is_data & 1)) { 800 skcd_buf.val = 0; 801 skcd_buf.is_data = 1; 802 } 803 804 skcd_buf.classid = classid; 805 WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */ 806} 807 808#else /* CONFIG_SOCK_CGROUP_DATA */ 809 810struct sock_cgroup_data { 811}; 812 813#endif /* CONFIG_SOCK_CGROUP_DATA */ 814 815#endif /* _LINUX_CGROUP_DEFS_H */ 816