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-defs.h> 23 #include <linux/psi_types.h> 24 25 #ifdef CONFIG_CGROUPS 26 27 struct cgroup; 28 struct cgroup_root; 29 struct cgroup_subsys; 30 struct cgroup_taskset; 31 struct kernfs_node; 32 struct kernfs_ops; 33 struct kernfs_open_file; 34 struct seq_file; 35 struct poll_table_struct; 36 37 #define MAX_CGROUP_TYPE_NAMELEN 32 38 #define MAX_CGROUP_ROOT_NAMELEN 64 39 #define MAX_CFTYPE_NAME 64 40 41 /* define the enumeration of all cgroup subsystems */ 42 #define SUBSYS(_x) _x ## _cgrp_id, 43 enum cgroup_subsys_id { 44 #include <linux/cgroup_subsys.h> 45 CGROUP_SUBSYS_COUNT, 46 }; 47 #undef SUBSYS 48 49 /* bits in struct cgroup_subsys_state flags field */ 50 enum { 51 CSS_NO_REF = (1 << 0), /* no reference counting for this css */ 52 CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */ 53 CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */ 54 CSS_VISIBLE = (1 << 3), /* css is visible to userland */ 55 CSS_DYING = (1 << 4), /* css is dying */ 56 }; 57 58 /* bits in struct cgroup flags field */ 59 enum { 60 /* Control Group requires release notifications to userspace */ 61 CGRP_NOTIFY_ON_RELEASE, 62 /* 63 * Clone the parent's configuration when creating a new child 64 * cpuset cgroup. For historical reasons, this option can be 65 * specified at mount time and thus is implemented here. 66 */ 67 CGRP_CPUSET_CLONE_CHILDREN, 68 69 /* Control group has to be frozen. */ 70 CGRP_FREEZE, 71 72 /* Cgroup is frozen. */ 73 CGRP_FROZEN, 74 }; 75 76 /* cgroup_root->flags */ 77 enum { 78 CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */ 79 CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */ 80 81 /* 82 * Consider namespaces as delegation boundaries. If this flag is 83 * set, controller specific interface files in a namespace root 84 * aren't writeable from inside the namespace. 85 */ 86 CGRP_ROOT_NS_DELEGATE = (1 << 3), 87 88 /* 89 * Reduce latencies on dynamic cgroup modifications such as task 90 * migrations and controller on/offs by disabling percpu operation on 91 * cgroup_threadgroup_rwsem. This makes hot path operations such as 92 * forks and exits into the slow path and more expensive. 93 * 94 * The static usage pattern of creating a cgroup, enabling controllers, 95 * and then seeding it with CLONE_INTO_CGROUP doesn't require write 96 * locking cgroup_threadgroup_rwsem and thus doesn't benefit from 97 * favordynmod. 98 */ 99 CGRP_ROOT_FAVOR_DYNMODS = (1 << 4), 100 101 /* 102 * Enable cpuset controller in v1 cgroup to use v2 behavior. 103 */ 104 CGRP_ROOT_CPUSET_V2_MODE = (1 << 16), 105 106 /* 107 * Enable legacy local memory.events. 108 */ 109 CGRP_ROOT_MEMORY_LOCAL_EVENTS = (1 << 17), 110 111 /* 112 * Enable recursive subtree protection 113 */ 114 CGRP_ROOT_MEMORY_RECURSIVE_PROT = (1 << 18), 115 116 /* 117 * Enable hugetlb accounting for the memory controller. 118 */ 119 CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING = (1 << 19), 120 121 /* 122 * Enable legacy local pids.events. 123 */ 124 CGRP_ROOT_PIDS_LOCAL_EVENTS = (1 << 20), 125 }; 126 127 /* cftype->flags */ 128 enum { 129 CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */ 130 CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */ 131 CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */ 132 133 CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */ 134 CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */ 135 CFTYPE_DEBUG = (1 << 5), /* create when cgroup_debug */ 136 137 /* internal flags, do not use outside cgroup core proper */ 138 __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */ 139 __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */ 140 __CFTYPE_ADDED = (1 << 18), 141 }; 142 143 /* 144 * cgroup_file is the handle for a file instance created in a cgroup which 145 * is used, for example, to generate file changed notifications. This can 146 * be obtained by setting cftype->file_offset. 147 */ 148 struct cgroup_file { 149 /* do not access any fields from outside cgroup core */ 150 struct kernfs_node *kn; 151 unsigned long notified_at; 152 struct timer_list notify_timer; 153 }; 154 155 /* 156 * Per-subsystem/per-cgroup state maintained by the system. This is the 157 * fundamental structural building block that controllers deal with. 158 * 159 * Fields marked with "PI:" are public and immutable and may be accessed 160 * directly without synchronization. 161 */ 162 struct cgroup_subsys_state { 163 /* PI: the cgroup that this css is attached to */ 164 struct cgroup *cgroup; 165 166 /* PI: the cgroup subsystem that this css is attached to */ 167 struct cgroup_subsys *ss; 168 169 /* reference count - access via css_[try]get() and css_put() */ 170 struct percpu_ref refcnt; 171 172 /* 173 * siblings list anchored at the parent's ->children 174 * 175 * linkage is protected by cgroup_mutex or RCU 176 */ 177 struct list_head sibling; 178 struct list_head children; 179 180 /* flush target list anchored at cgrp->rstat_css_list */ 181 struct list_head rstat_css_node; 182 183 /* 184 * PI: Subsys-unique ID. 0 is unused and root is always 1. The 185 * matching css can be looked up using css_from_id(). 186 */ 187 int id; 188 189 unsigned int flags; 190 191 /* 192 * Monotonically increasing unique serial number which defines a 193 * uniform order among all csses. It's guaranteed that all 194 * ->children lists are in the ascending order of ->serial_nr and 195 * used to allow interrupting and resuming iterations. 196 */ 197 u64 serial_nr; 198 199 /* 200 * Incremented by online self and children. Used to guarantee that 201 * parents are not offlined before their children. 202 */ 203 atomic_t online_cnt; 204 205 /* percpu_ref killing and RCU release */ 206 struct work_struct destroy_work; 207 struct rcu_work destroy_rwork; 208 209 /* 210 * PI: the parent css. Placed here for cache proximity to following 211 * fields of the containing structure. 212 */ 213 struct cgroup_subsys_state *parent; 214 215 /* 216 * Keep track of total numbers of visible descendant CSSes. 217 * The total number of dying CSSes is tracked in 218 * css->cgroup->nr_dying_subsys[ssid]. 219 * Protected by cgroup_mutex. 220 */ 221 int nr_descendants; 222 }; 223 224 /* 225 * A css_set is a structure holding pointers to a set of 226 * cgroup_subsys_state objects. This saves space in the task struct 227 * object and speeds up fork()/exit(), since a single inc/dec and a 228 * list_add()/del() can bump the reference count on the entire cgroup 229 * set for a task. 230 */ 231 struct css_set { 232 /* 233 * Set of subsystem states, one for each subsystem. This array is 234 * immutable after creation apart from the init_css_set during 235 * subsystem registration (at boot time). 236 */ 237 struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; 238 239 /* reference count */ 240 refcount_t refcount; 241 242 /* 243 * For a domain cgroup, the following points to self. If threaded, 244 * to the matching cset of the nearest domain ancestor. The 245 * dom_cset provides access to the domain cgroup and its csses to 246 * which domain level resource consumptions should be charged. 247 */ 248 struct css_set *dom_cset; 249 250 /* the default cgroup associated with this css_set */ 251 struct cgroup *dfl_cgrp; 252 253 /* internal task count, protected by css_set_lock */ 254 int nr_tasks; 255 256 /* 257 * Lists running through all tasks using this cgroup group. 258 * mg_tasks lists tasks which belong to this cset but are in the 259 * process of being migrated out or in. Protected by 260 * css_set_lock, but, during migration, once tasks are moved to 261 * mg_tasks, it can be read safely while holding cgroup_mutex. 262 */ 263 struct list_head tasks; 264 struct list_head mg_tasks; 265 struct list_head dying_tasks; 266 267 /* all css_task_iters currently walking this cset */ 268 struct list_head task_iters; 269 270 /* 271 * On the default hierarchy, ->subsys[ssid] may point to a css 272 * attached to an ancestor instead of the cgroup this css_set is 273 * associated with. The following node is anchored at 274 * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to 275 * iterate through all css's attached to a given cgroup. 276 */ 277 struct list_head e_cset_node[CGROUP_SUBSYS_COUNT]; 278 279 /* all threaded csets whose ->dom_cset points to this cset */ 280 struct list_head threaded_csets; 281 struct list_head threaded_csets_node; 282 283 /* 284 * List running through all cgroup groups in the same hash 285 * slot. Protected by css_set_lock 286 */ 287 struct hlist_node hlist; 288 289 /* 290 * List of cgrp_cset_links pointing at cgroups referenced from this 291 * css_set. Protected by css_set_lock. 292 */ 293 struct list_head cgrp_links; 294 295 /* 296 * List of csets participating in the on-going migration either as 297 * source or destination. Protected by cgroup_mutex. 298 */ 299 struct list_head mg_src_preload_node; 300 struct list_head mg_dst_preload_node; 301 struct list_head mg_node; 302 303 /* 304 * If this cset is acting as the source of migration the following 305 * two fields are set. mg_src_cgrp and mg_dst_cgrp are 306 * respectively the source and destination cgroups of the on-going 307 * migration. mg_dst_cset is the destination cset the target tasks 308 * on this cset should be migrated to. Protected by cgroup_mutex. 309 */ 310 struct cgroup *mg_src_cgrp; 311 struct cgroup *mg_dst_cgrp; 312 struct css_set *mg_dst_cset; 313 314 /* dead and being drained, ignore for migration */ 315 bool dead; 316 317 /* For RCU-protected deletion */ 318 struct rcu_head rcu_head; 319 }; 320 321 struct cgroup_base_stat { 322 struct task_cputime cputime; 323 324 #ifdef CONFIG_SCHED_CORE 325 u64 forceidle_sum; 326 #endif 327 u64 ntime; 328 }; 329 330 /* 331 * rstat - cgroup scalable recursive statistics. Accounting is done 332 * per-cpu in cgroup_rstat_cpu which is then lazily propagated up the 333 * hierarchy on reads. 334 * 335 * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are 336 * linked into the updated tree. On the following read, propagation only 337 * considers and consumes the updated tree. This makes reading O(the 338 * number of descendants which have been active since last read) instead of 339 * O(the total number of descendants). 340 * 341 * This is important because there can be a lot of (draining) cgroups which 342 * aren't active and stat may be read frequently. The combination can 343 * become very expensive. By propagating selectively, increasing reading 344 * frequency decreases the cost of each read. 345 * 346 * This struct hosts both the fields which implement the above - 347 * updated_children and updated_next - and the fields which track basic 348 * resource statistics on top of it - bsync, bstat and last_bstat. 349 */ 350 struct cgroup_rstat_cpu { 351 /* 352 * ->bsync protects ->bstat. These are the only fields which get 353 * updated in the hot path. 354 */ 355 struct u64_stats_sync bsync; 356 struct cgroup_base_stat bstat; 357 358 /* 359 * Snapshots at the last reading. These are used to calculate the 360 * deltas to propagate to the global counters. 361 */ 362 struct cgroup_base_stat last_bstat; 363 364 /* 365 * This field is used to record the cumulative per-cpu time of 366 * the cgroup and its descendants. Currently it can be read via 367 * eBPF/drgn etc, and we are still trying to determine how to 368 * expose it in the cgroupfs interface. 369 */ 370 struct cgroup_base_stat subtree_bstat; 371 372 /* 373 * Snapshots at the last reading. These are used to calculate the 374 * deltas to propagate to the per-cpu subtree_bstat. 375 */ 376 struct cgroup_base_stat last_subtree_bstat; 377 378 /* 379 * Child cgroups with stat updates on this cpu since the last read 380 * are linked on the parent's ->updated_children through 381 * ->updated_next. 382 * 383 * In addition to being more compact, singly-linked list pointing 384 * to the cgroup makes it unnecessary for each per-cpu struct to 385 * point back to the associated cgroup. 386 * 387 * Protected by per-cpu cgroup_rstat_cpu_lock. 388 */ 389 struct cgroup *updated_children; /* terminated by self cgroup */ 390 struct cgroup *updated_next; /* NULL iff not on the list */ 391 }; 392 393 struct cgroup_freezer_state { 394 /* Should the cgroup and its descendants be frozen. */ 395 bool freeze; 396 397 /* Should the cgroup actually be frozen? */ 398 bool e_freeze; 399 400 /* Fields below are protected by css_set_lock */ 401 402 /* Number of frozen descendant cgroups */ 403 int nr_frozen_descendants; 404 405 /* 406 * Number of tasks, which are counted as frozen: 407 * frozen, SIGSTOPped, and PTRACEd. 408 */ 409 int nr_frozen_tasks; 410 }; 411 412 struct cgroup { 413 /* self css with NULL ->ss, points back to this cgroup */ 414 struct cgroup_subsys_state self; 415 416 unsigned long flags; /* "unsigned long" so bitops work */ 417 418 /* 419 * The depth this cgroup is at. The root is at depth zero and each 420 * step down the hierarchy increments the level. This along with 421 * ancestors[] can determine whether a given cgroup is a 422 * descendant of another without traversing the hierarchy. 423 */ 424 int level; 425 426 /* Maximum allowed descent tree depth */ 427 int max_depth; 428 429 /* 430 * Keep track of total numbers of visible and dying descent cgroups. 431 * Dying cgroups are cgroups which were deleted by a user, 432 * but are still existing because someone else is holding a reference. 433 * max_descendants is a maximum allowed number of descent cgroups. 434 * 435 * nr_descendants and nr_dying_descendants are protected 436 * by cgroup_mutex and css_set_lock. It's fine to read them holding 437 * any of cgroup_mutex and css_set_lock; for writing both locks 438 * should be held. 439 */ 440 int nr_descendants; 441 int nr_dying_descendants; 442 int max_descendants; 443 444 /* 445 * Each non-empty css_set associated with this cgroup contributes 446 * one to nr_populated_csets. The counter is zero iff this cgroup 447 * doesn't have any tasks. 448 * 449 * All children which have non-zero nr_populated_csets and/or 450 * nr_populated_children of their own contribute one to either 451 * nr_populated_domain_children or nr_populated_threaded_children 452 * depending on their type. Each counter is zero iff all cgroups 453 * of the type in the subtree proper don't have any tasks. 454 */ 455 int nr_populated_csets; 456 int nr_populated_domain_children; 457 int nr_populated_threaded_children; 458 459 int nr_threaded_children; /* # of live threaded child cgroups */ 460 461 /* sequence number for cgroup.kill, serialized by css_set_lock. */ 462 unsigned int kill_seq; 463 464 struct kernfs_node *kn; /* cgroup kernfs entry */ 465 struct cgroup_file procs_file; /* handle for "cgroup.procs" */ 466 struct cgroup_file events_file; /* handle for "cgroup.events" */ 467 468 /* handles for "{cpu,memory,io,irq}.pressure" */ 469 struct cgroup_file psi_files[NR_PSI_RESOURCES]; 470 471 /* 472 * The bitmask of subsystems enabled on the child cgroups. 473 * ->subtree_control is the one configured through 474 * "cgroup.subtree_control" while ->subtree_ss_mask is the effective 475 * one which may have more subsystems enabled. Controller knobs 476 * are made available iff it's enabled in ->subtree_control. 477 */ 478 u16 subtree_control; 479 u16 subtree_ss_mask; 480 u16 old_subtree_control; 481 u16 old_subtree_ss_mask; 482 483 /* Private pointers for each registered subsystem */ 484 struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT]; 485 486 /* 487 * Keep track of total number of dying CSSes at and below this cgroup. 488 * Protected by cgroup_mutex. 489 */ 490 int nr_dying_subsys[CGROUP_SUBSYS_COUNT]; 491 492 struct cgroup_root *root; 493 494 /* 495 * List of cgrp_cset_links pointing at css_sets with tasks in this 496 * cgroup. Protected by css_set_lock. 497 */ 498 struct list_head cset_links; 499 500 /* 501 * On the default hierarchy, a css_set for a cgroup with some 502 * susbsys disabled will point to css's which are associated with 503 * the closest ancestor which has the subsys enabled. The 504 * following lists all css_sets which point to this cgroup's css 505 * for the given subsystem. 506 */ 507 struct list_head e_csets[CGROUP_SUBSYS_COUNT]; 508 509 /* 510 * If !threaded, self. If threaded, it points to the nearest 511 * domain ancestor. Inside a threaded subtree, cgroups are exempt 512 * from process granularity and no-internal-task constraint. 513 * Domain level resource consumptions which aren't tied to a 514 * specific task are charged to the dom_cgrp. 515 */ 516 struct cgroup *dom_cgrp; 517 struct cgroup *old_dom_cgrp; /* used while enabling threaded */ 518 519 /* per-cpu recursive resource statistics */ 520 struct cgroup_rstat_cpu __percpu *rstat_cpu; 521 struct list_head rstat_css_list; 522 523 /* 524 * Add padding to separate the read mostly rstat_cpu and 525 * rstat_css_list into a different cacheline from the following 526 * rstat_flush_next and *bstat fields which can have frequent updates. 527 */ 528 CACHELINE_PADDING(_pad_); 529 530 /* 531 * A singly-linked list of cgroup structures to be rstat flushed. 532 * This is a scratch field to be used exclusively by 533 * cgroup_rstat_flush_locked() and protected by cgroup_rstat_lock. 534 */ 535 struct cgroup *rstat_flush_next; 536 537 /* cgroup basic resource statistics */ 538 struct cgroup_base_stat last_bstat; 539 struct cgroup_base_stat bstat; 540 struct prev_cputime prev_cputime; /* for printing out cputime */ 541 542 /* 543 * list of pidlists, up to two for each namespace (one for procs, one 544 * for tasks); created on demand. 545 */ 546 struct list_head pidlists; 547 struct mutex pidlist_mutex; 548 549 /* used to wait for offlining of csses */ 550 wait_queue_head_t offline_waitq; 551 552 /* used to schedule release agent */ 553 struct work_struct release_agent_work; 554 555 /* used to track pressure stalls */ 556 struct psi_group *psi; 557 558 /* used to store eBPF programs */ 559 struct cgroup_bpf bpf; 560 561 /* Used to store internal freezer state */ 562 struct cgroup_freezer_state freezer; 563 564 #ifdef CONFIG_BPF_SYSCALL 565 struct bpf_local_storage __rcu *bpf_cgrp_storage; 566 #endif 567 568 /* All ancestors including self */ 569 struct cgroup *ancestors[]; 570 }; 571 572 /* 573 * A cgroup_root represents the root of a cgroup hierarchy, and may be 574 * associated with a kernfs_root to form an active hierarchy. This is 575 * internal to cgroup core. Don't access directly from controllers. 576 */ 577 struct cgroup_root { 578 struct kernfs_root *kf_root; 579 580 /* The bitmask of subsystems attached to this hierarchy */ 581 unsigned int subsys_mask; 582 583 /* Unique id for this hierarchy. */ 584 int hierarchy_id; 585 586 /* A list running through the active hierarchies */ 587 struct list_head root_list; 588 struct rcu_head rcu; /* Must be near the top */ 589 590 /* 591 * The root cgroup. The containing cgroup_root will be destroyed on its 592 * release. cgrp->ancestors[0] will be used overflowing into the 593 * following field. cgrp_ancestor_storage must immediately follow. 594 */ 595 struct cgroup cgrp; 596 597 /* must follow cgrp for cgrp->ancestors[0], see above */ 598 struct cgroup *cgrp_ancestor_storage; 599 600 /* Number of cgroups in the hierarchy, used only for /proc/cgroups */ 601 atomic_t nr_cgrps; 602 603 /* Hierarchy-specific flags */ 604 unsigned int flags; 605 606 /* The path to use for release notifications. */ 607 char release_agent_path[PATH_MAX]; 608 609 /* The name for this hierarchy - may be empty */ 610 char name[MAX_CGROUP_ROOT_NAMELEN]; 611 }; 612 613 /* 614 * struct cftype: handler definitions for cgroup control files 615 * 616 * When reading/writing to a file: 617 * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata 618 * - the 'cftype' of the file is file->f_path.dentry->d_fsdata 619 */ 620 struct cftype { 621 /* 622 * Name of the subsystem is prepended in cgroup_file_name(). 623 * Zero length string indicates end of cftype array. 624 */ 625 char name[MAX_CFTYPE_NAME]; 626 unsigned long private; 627 628 /* 629 * The maximum length of string, excluding trailing nul, that can 630 * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed. 631 */ 632 size_t max_write_len; 633 634 /* CFTYPE_* flags */ 635 unsigned int flags; 636 637 /* 638 * If non-zero, should contain the offset from the start of css to 639 * a struct cgroup_file field. cgroup will record the handle of 640 * the created file into it. The recorded handle can be used as 641 * long as the containing css remains accessible. 642 */ 643 unsigned int file_offset; 644 645 /* 646 * Fields used for internal bookkeeping. Initialized automatically 647 * during registration. 648 */ 649 struct cgroup_subsys *ss; /* NULL for cgroup core files */ 650 struct list_head node; /* anchored at ss->cfts */ 651 struct kernfs_ops *kf_ops; 652 653 int (*open)(struct kernfs_open_file *of); 654 void (*release)(struct kernfs_open_file *of); 655 656 /* 657 * read_u64() is a shortcut for the common case of returning a 658 * single integer. Use it in place of read() 659 */ 660 u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft); 661 /* 662 * read_s64() is a signed version of read_u64() 663 */ 664 s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft); 665 666 /* generic seq_file read interface */ 667 int (*seq_show)(struct seq_file *sf, void *v); 668 669 /* optional ops, implement all or none */ 670 void *(*seq_start)(struct seq_file *sf, loff_t *ppos); 671 void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos); 672 void (*seq_stop)(struct seq_file *sf, void *v); 673 674 /* 675 * write_u64() is a shortcut for the common case of accepting 676 * a single integer (as parsed by simple_strtoull) from 677 * userspace. Use in place of write(); return 0 or error. 678 */ 679 int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft, 680 u64 val); 681 /* 682 * write_s64() is a signed version of write_u64() 683 */ 684 int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft, 685 s64 val); 686 687 /* 688 * write() is the generic write callback which maps directly to 689 * kernfs write operation and overrides all other operations. 690 * Maximum write size is determined by ->max_write_len. Use 691 * of_css/cft() to access the associated css and cft. 692 */ 693 ssize_t (*write)(struct kernfs_open_file *of, 694 char *buf, size_t nbytes, loff_t off); 695 696 __poll_t (*poll)(struct kernfs_open_file *of, 697 struct poll_table_struct *pt); 698 699 struct lock_class_key lockdep_key; 700 }; 701 702 /* 703 * Control Group subsystem type. 704 * See Documentation/admin-guide/cgroup-v1/cgroups.rst for details 705 */ 706 struct cgroup_subsys { 707 struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css); 708 int (*css_online)(struct cgroup_subsys_state *css); 709 void (*css_offline)(struct cgroup_subsys_state *css); 710 void (*css_released)(struct cgroup_subsys_state *css); 711 void (*css_free)(struct cgroup_subsys_state *css); 712 void (*css_reset)(struct cgroup_subsys_state *css); 713 void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu); 714 int (*css_extra_stat_show)(struct seq_file *seq, 715 struct cgroup_subsys_state *css); 716 int (*css_local_stat_show)(struct seq_file *seq, 717 struct cgroup_subsys_state *css); 718 719 int (*can_attach)(struct cgroup_taskset *tset); 720 void (*cancel_attach)(struct cgroup_taskset *tset); 721 void (*attach)(struct cgroup_taskset *tset); 722 void (*post_attach)(void); 723 int (*can_fork)(struct task_struct *task, 724 struct css_set *cset); 725 void (*cancel_fork)(struct task_struct *task, struct css_set *cset); 726 void (*fork)(struct task_struct *task); 727 void (*exit)(struct task_struct *task); 728 void (*release)(struct task_struct *task); 729 void (*bind)(struct cgroup_subsys_state *root_css); 730 731 bool early_init:1; 732 733 /* 734 * If %true, the controller, on the default hierarchy, doesn't show 735 * up in "cgroup.controllers" or "cgroup.subtree_control", is 736 * implicitly enabled on all cgroups on the default hierarchy, and 737 * bypasses the "no internal process" constraint. This is for 738 * utility type controllers which is transparent to userland. 739 * 740 * An implicit controller can be stolen from the default hierarchy 741 * anytime and thus must be okay with offline csses from previous 742 * hierarchies coexisting with csses for the current one. 743 */ 744 bool implicit_on_dfl:1; 745 746 /* 747 * If %true, the controller, supports threaded mode on the default 748 * hierarchy. In a threaded subtree, both process granularity and 749 * no-internal-process constraint are ignored and a threaded 750 * controllers should be able to handle that. 751 * 752 * Note that as an implicit controller is automatically enabled on 753 * all cgroups on the default hierarchy, it should also be 754 * threaded. implicit && !threaded is not supported. 755 */ 756 bool threaded:1; 757 758 /* the following two fields are initialized automatically during boot */ 759 int id; 760 const char *name; 761 762 /* optional, initialized automatically during boot if not set */ 763 const char *legacy_name; 764 765 /* link to parent, protected by cgroup_lock() */ 766 struct cgroup_root *root; 767 768 /* idr for css->id */ 769 struct idr css_idr; 770 771 /* 772 * List of cftypes. Each entry is the first entry of an array 773 * terminated by zero length name. 774 */ 775 struct list_head cfts; 776 777 /* 778 * Base cftypes which are automatically registered. The two can 779 * point to the same array. 780 */ 781 struct cftype *dfl_cftypes; /* for the default hierarchy */ 782 struct cftype *legacy_cftypes; /* for the legacy hierarchies */ 783 784 /* 785 * A subsystem may depend on other subsystems. When such subsystem 786 * is enabled on a cgroup, the depended-upon subsystems are enabled 787 * together if available. Subsystems enabled due to dependency are 788 * not visible to userland until explicitly enabled. The following 789 * specifies the mask of subsystems that this one depends on. 790 */ 791 unsigned int depends_on; 792 }; 793 794 extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem; 795 796 struct cgroup_of_peak { 797 unsigned long value; 798 struct list_head list; 799 }; 800 801 /** 802 * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups 803 * @tsk: target task 804 * 805 * Allows cgroup operations to synchronize against threadgroup changes 806 * using a percpu_rw_semaphore. 807 */ 808 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) 809 { 810 percpu_down_read(&cgroup_threadgroup_rwsem); 811 } 812 813 /** 814 * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups 815 * @tsk: target task 816 * 817 * Counterpart of cgroup_threadcgroup_change_begin(). 818 */ 819 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) 820 { 821 percpu_up_read(&cgroup_threadgroup_rwsem); 822 } 823 824 #else /* CONFIG_CGROUPS */ 825 826 #define CGROUP_SUBSYS_COUNT 0 827 828 static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) 829 { 830 might_sleep(); 831 } 832 833 static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {} 834 835 #endif /* CONFIG_CGROUPS */ 836 837 #ifdef CONFIG_SOCK_CGROUP_DATA 838 839 /* 840 * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains 841 * per-socket cgroup information except for memcg association. 842 * 843 * On legacy hierarchies, net_prio and net_cls controllers directly 844 * set attributes on each sock which can then be tested by the network 845 * layer. On the default hierarchy, each sock is associated with the 846 * cgroup it was created in and the networking layer can match the 847 * cgroup directly. 848 */ 849 struct sock_cgroup_data { 850 struct cgroup *cgroup; /* v2 */ 851 #ifdef CONFIG_CGROUP_NET_CLASSID 852 u32 classid; /* v1 */ 853 #endif 854 #ifdef CONFIG_CGROUP_NET_PRIO 855 u16 prioidx; /* v1 */ 856 #endif 857 }; 858 859 static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd) 860 { 861 #ifdef CONFIG_CGROUP_NET_PRIO 862 return READ_ONCE(skcd->prioidx); 863 #else 864 return 1; 865 #endif 866 } 867 868 static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd) 869 { 870 #ifdef CONFIG_CGROUP_NET_CLASSID 871 return READ_ONCE(skcd->classid); 872 #else 873 return 0; 874 #endif 875 } 876 877 static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd, 878 u16 prioidx) 879 { 880 #ifdef CONFIG_CGROUP_NET_PRIO 881 WRITE_ONCE(skcd->prioidx, prioidx); 882 #endif 883 } 884 885 static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd, 886 u32 classid) 887 { 888 #ifdef CONFIG_CGROUP_NET_CLASSID 889 WRITE_ONCE(skcd->classid, classid); 890 #endif 891 } 892 893 #else /* CONFIG_SOCK_CGROUP_DATA */ 894 895 struct sock_cgroup_data { 896 }; 897 898 #endif /* CONFIG_SOCK_CGROUP_DATA */ 899 900 #endif /* _LINUX_CGROUP_DEFS_H */ 901