1 /* SPDX-License-Identifier: GPL-2.0 */ 2 #ifndef _LINUX_CGROUP_H 3 #define _LINUX_CGROUP_H 4 /* 5 * cgroup interface 6 * 7 * Copyright (C) 2003 BULL SA 8 * Copyright (C) 2004-2006 Silicon Graphics, Inc. 9 * 10 */ 11 12 #include <linux/sched.h> 13 #include <linux/cpumask.h> 14 #include <linux/nodemask.h> 15 #include <linux/rculist.h> 16 #include <linux/cgroupstats.h> 17 #include <linux/fs.h> 18 #include <linux/seq_file.h> 19 #include <linux/kernfs.h> 20 #include <linux/jump_label.h> 21 #include <linux/types.h> 22 #include <linux/ns_common.h> 23 #include <linux/nsproxy.h> 24 #include <linux/user_namespace.h> 25 #include <linux/refcount.h> 26 #include <linux/kernel_stat.h> 27 28 #include <linux/cgroup-defs.h> 29 30 #ifdef CONFIG_CGROUPS 31 32 /* 33 * All weight knobs on the default hierarhcy should use the following min, 34 * default and max values. The default value is the logarithmic center of 35 * MIN and MAX and allows 100x to be expressed in both directions. 36 */ 37 #define CGROUP_WEIGHT_MIN 1 38 #define CGROUP_WEIGHT_DFL 100 39 #define CGROUP_WEIGHT_MAX 10000 40 41 /* walk only threadgroup leaders */ 42 #define CSS_TASK_ITER_PROCS (1U << 0) 43 /* walk all threaded css_sets in the domain */ 44 #define CSS_TASK_ITER_THREADED (1U << 1) 45 46 /* a css_task_iter should be treated as an opaque object */ 47 struct css_task_iter { 48 struct cgroup_subsys *ss; 49 unsigned int flags; 50 51 struct list_head *cset_pos; 52 struct list_head *cset_head; 53 54 struct list_head *tcset_pos; 55 struct list_head *tcset_head; 56 57 struct list_head *task_pos; 58 struct list_head *tasks_head; 59 struct list_head *mg_tasks_head; 60 61 struct css_set *cur_cset; 62 struct css_set *cur_dcset; 63 struct task_struct *cur_task; 64 struct list_head iters_node; /* css_set->task_iters */ 65 }; 66 67 extern struct cgroup_root cgrp_dfl_root; 68 extern struct css_set init_css_set; 69 70 #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys; 71 #include <linux/cgroup_subsys.h> 72 #undef SUBSYS 73 74 #define SUBSYS(_x) \ 75 extern struct static_key_true _x ## _cgrp_subsys_enabled_key; \ 76 extern struct static_key_true _x ## _cgrp_subsys_on_dfl_key; 77 #include <linux/cgroup_subsys.h> 78 #undef SUBSYS 79 80 /** 81 * cgroup_subsys_enabled - fast test on whether a subsys is enabled 82 * @ss: subsystem in question 83 */ 84 #define cgroup_subsys_enabled(ss) \ 85 static_branch_likely(&ss ## _enabled_key) 86 87 /** 88 * cgroup_subsys_on_dfl - fast test on whether a subsys is on default hierarchy 89 * @ss: subsystem in question 90 */ 91 #define cgroup_subsys_on_dfl(ss) \ 92 static_branch_likely(&ss ## _on_dfl_key) 93 94 bool css_has_online_children(struct cgroup_subsys_state *css); 95 struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss); 96 struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgroup, 97 struct cgroup_subsys *ss); 98 struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup, 99 struct cgroup_subsys *ss); 100 struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, 101 struct cgroup_subsys *ss); 102 103 struct cgroup *cgroup_get_from_path(const char *path); 104 struct cgroup *cgroup_get_from_fd(int fd); 105 106 int cgroup_attach_task_all(struct task_struct *from, struct task_struct *); 107 int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from); 108 109 int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); 110 int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); 111 int cgroup_rm_cftypes(struct cftype *cfts); 112 void cgroup_file_notify(struct cgroup_file *cfile); 113 114 int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen); 115 int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry); 116 int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, 117 struct pid *pid, struct task_struct *tsk); 118 119 void cgroup_fork(struct task_struct *p); 120 extern int cgroup_can_fork(struct task_struct *p); 121 extern void cgroup_cancel_fork(struct task_struct *p); 122 extern void cgroup_post_fork(struct task_struct *p); 123 void cgroup_exit(struct task_struct *p); 124 void cgroup_release(struct task_struct *p); 125 void cgroup_free(struct task_struct *p); 126 127 int cgroup_init_early(void); 128 int cgroup_init(void); 129 130 /* 131 * Iteration helpers and macros. 132 */ 133 134 struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, 135 struct cgroup_subsys_state *parent); 136 struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos, 137 struct cgroup_subsys_state *css); 138 struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos); 139 struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos, 140 struct cgroup_subsys_state *css); 141 142 struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, 143 struct cgroup_subsys_state **dst_cssp); 144 struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, 145 struct cgroup_subsys_state **dst_cssp); 146 147 void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags, 148 struct css_task_iter *it); 149 struct task_struct *css_task_iter_next(struct css_task_iter *it); 150 void css_task_iter_end(struct css_task_iter *it); 151 152 /** 153 * css_for_each_child - iterate through children of a css 154 * @pos: the css * to use as the loop cursor 155 * @parent: css whose children to walk 156 * 157 * Walk @parent's children. Must be called under rcu_read_lock(). 158 * 159 * If a subsystem synchronizes ->css_online() and the start of iteration, a 160 * css which finished ->css_online() is guaranteed to be visible in the 161 * future iterations and will stay visible until the last reference is put. 162 * A css which hasn't finished ->css_online() or already finished 163 * ->css_offline() may show up during traversal. It's each subsystem's 164 * responsibility to synchronize against on/offlining. 165 * 166 * It is allowed to temporarily drop RCU read lock during iteration. The 167 * caller is responsible for ensuring that @pos remains accessible until 168 * the start of the next iteration by, for example, bumping the css refcnt. 169 */ 170 #define css_for_each_child(pos, parent) \ 171 for ((pos) = css_next_child(NULL, (parent)); (pos); \ 172 (pos) = css_next_child((pos), (parent))) 173 174 /** 175 * css_for_each_descendant_pre - pre-order walk of a css's descendants 176 * @pos: the css * to use as the loop cursor 177 * @root: css whose descendants to walk 178 * 179 * Walk @root's descendants. @root is included in the iteration and the 180 * first node to be visited. Must be called under rcu_read_lock(). 181 * 182 * If a subsystem synchronizes ->css_online() and the start of iteration, a 183 * css which finished ->css_online() is guaranteed to be visible in the 184 * future iterations and will stay visible until the last reference is put. 185 * A css which hasn't finished ->css_online() or already finished 186 * ->css_offline() may show up during traversal. It's each subsystem's 187 * responsibility to synchronize against on/offlining. 188 * 189 * For example, the following guarantees that a descendant can't escape 190 * state updates of its ancestors. 191 * 192 * my_online(@css) 193 * { 194 * Lock @css's parent and @css; 195 * Inherit state from the parent; 196 * Unlock both. 197 * } 198 * 199 * my_update_state(@css) 200 * { 201 * css_for_each_descendant_pre(@pos, @css) { 202 * Lock @pos; 203 * if (@pos == @css) 204 * Update @css's state; 205 * else 206 * Verify @pos is alive and inherit state from its parent; 207 * Unlock @pos; 208 * } 209 * } 210 * 211 * As long as the inheriting step, including checking the parent state, is 212 * enclosed inside @pos locking, double-locking the parent isn't necessary 213 * while inheriting. The state update to the parent is guaranteed to be 214 * visible by walking order and, as long as inheriting operations to the 215 * same @pos are atomic to each other, multiple updates racing each other 216 * still result in the correct state. It's guaranateed that at least one 217 * inheritance happens for any css after the latest update to its parent. 218 * 219 * If checking parent's state requires locking the parent, each inheriting 220 * iteration should lock and unlock both @pos->parent and @pos. 221 * 222 * Alternatively, a subsystem may choose to use a single global lock to 223 * synchronize ->css_online() and ->css_offline() against tree-walking 224 * operations. 225 * 226 * It is allowed to temporarily drop RCU read lock during iteration. The 227 * caller is responsible for ensuring that @pos remains accessible until 228 * the start of the next iteration by, for example, bumping the css refcnt. 229 */ 230 #define css_for_each_descendant_pre(pos, css) \ 231 for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \ 232 (pos) = css_next_descendant_pre((pos), (css))) 233 234 /** 235 * css_for_each_descendant_post - post-order walk of a css's descendants 236 * @pos: the css * to use as the loop cursor 237 * @css: css whose descendants to walk 238 * 239 * Similar to css_for_each_descendant_pre() but performs post-order 240 * traversal instead. @root is included in the iteration and the last 241 * node to be visited. 242 * 243 * If a subsystem synchronizes ->css_online() and the start of iteration, a 244 * css which finished ->css_online() is guaranteed to be visible in the 245 * future iterations and will stay visible until the last reference is put. 246 * A css which hasn't finished ->css_online() or already finished 247 * ->css_offline() may show up during traversal. It's each subsystem's 248 * responsibility to synchronize against on/offlining. 249 * 250 * Note that the walk visibility guarantee example described in pre-order 251 * walk doesn't apply the same to post-order walks. 252 */ 253 #define css_for_each_descendant_post(pos, css) \ 254 for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \ 255 (pos) = css_next_descendant_post((pos), (css))) 256 257 /** 258 * cgroup_taskset_for_each - iterate cgroup_taskset 259 * @task: the loop cursor 260 * @dst_css: the destination css 261 * @tset: taskset to iterate 262 * 263 * @tset may contain multiple tasks and they may belong to multiple 264 * processes. 265 * 266 * On the v2 hierarchy, there may be tasks from multiple processes and they 267 * may not share the source or destination csses. 268 * 269 * On traditional hierarchies, when there are multiple tasks in @tset, if a 270 * task of a process is in @tset, all tasks of the process are in @tset. 271 * Also, all are guaranteed to share the same source and destination csses. 272 * 273 * Iteration is not in any specific order. 274 */ 275 #define cgroup_taskset_for_each(task, dst_css, tset) \ 276 for ((task) = cgroup_taskset_first((tset), &(dst_css)); \ 277 (task); \ 278 (task) = cgroup_taskset_next((tset), &(dst_css))) 279 280 /** 281 * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset 282 * @leader: the loop cursor 283 * @dst_css: the destination css 284 * @tset: taskset to iterate 285 * 286 * Iterate threadgroup leaders of @tset. For single-task migrations, @tset 287 * may not contain any. 288 */ 289 #define cgroup_taskset_for_each_leader(leader, dst_css, tset) \ 290 for ((leader) = cgroup_taskset_first((tset), &(dst_css)); \ 291 (leader); \ 292 (leader) = cgroup_taskset_next((tset), &(dst_css))) \ 293 if ((leader) != (leader)->group_leader) \ 294 ; \ 295 else 296 297 /* 298 * Inline functions. 299 */ 300 301 /** 302 * css_get - obtain a reference on the specified css 303 * @css: target css 304 * 305 * The caller must already have a reference. 306 */ 307 static inline void css_get(struct cgroup_subsys_state *css) 308 { 309 if (!(css->flags & CSS_NO_REF)) 310 percpu_ref_get(&css->refcnt); 311 } 312 313 /** 314 * css_get_many - obtain references on the specified css 315 * @css: target css 316 * @n: number of references to get 317 * 318 * The caller must already have a reference. 319 */ 320 static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n) 321 { 322 if (!(css->flags & CSS_NO_REF)) 323 percpu_ref_get_many(&css->refcnt, n); 324 } 325 326 /** 327 * css_tryget - try to obtain a reference on the specified css 328 * @css: target css 329 * 330 * Obtain a reference on @css unless it already has reached zero and is 331 * being released. This function doesn't care whether @css is on or 332 * offline. The caller naturally needs to ensure that @css is accessible 333 * but doesn't have to be holding a reference on it - IOW, RCU protected 334 * access is good enough for this function. Returns %true if a reference 335 * count was successfully obtained; %false otherwise. 336 */ 337 static inline bool css_tryget(struct cgroup_subsys_state *css) 338 { 339 if (!(css->flags & CSS_NO_REF)) 340 return percpu_ref_tryget(&css->refcnt); 341 return true; 342 } 343 344 /** 345 * css_tryget_online - try to obtain a reference on the specified css if online 346 * @css: target css 347 * 348 * Obtain a reference on @css if it's online. The caller naturally needs 349 * to ensure that @css is accessible but doesn't have to be holding a 350 * reference on it - IOW, RCU protected access is good enough for this 351 * function. Returns %true if a reference count was successfully obtained; 352 * %false otherwise. 353 */ 354 static inline bool css_tryget_online(struct cgroup_subsys_state *css) 355 { 356 if (!(css->flags & CSS_NO_REF)) 357 return percpu_ref_tryget_live(&css->refcnt); 358 return true; 359 } 360 361 /** 362 * css_is_dying - test whether the specified css is dying 363 * @css: target css 364 * 365 * Test whether @css is in the process of offlining or already offline. In 366 * most cases, ->css_online() and ->css_offline() callbacks should be 367 * enough; however, the actual offline operations are RCU delayed and this 368 * test returns %true also when @css is scheduled to be offlined. 369 * 370 * This is useful, for example, when the use case requires synchronous 371 * behavior with respect to cgroup removal. cgroup removal schedules css 372 * offlining but the css can seem alive while the operation is being 373 * delayed. If the delay affects user visible semantics, this test can be 374 * used to resolve the situation. 375 */ 376 static inline bool css_is_dying(struct cgroup_subsys_state *css) 377 { 378 return !(css->flags & CSS_NO_REF) && percpu_ref_is_dying(&css->refcnt); 379 } 380 381 /** 382 * css_put - put a css reference 383 * @css: target css 384 * 385 * Put a reference obtained via css_get() and css_tryget_online(). 386 */ 387 static inline void css_put(struct cgroup_subsys_state *css) 388 { 389 if (!(css->flags & CSS_NO_REF)) 390 percpu_ref_put(&css->refcnt); 391 } 392 393 /** 394 * css_put_many - put css references 395 * @css: target css 396 * @n: number of references to put 397 * 398 * Put references obtained via css_get() and css_tryget_online(). 399 */ 400 static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n) 401 { 402 if (!(css->flags & CSS_NO_REF)) 403 percpu_ref_put_many(&css->refcnt, n); 404 } 405 406 static inline void cgroup_get(struct cgroup *cgrp) 407 { 408 css_get(&cgrp->self); 409 } 410 411 static inline bool cgroup_tryget(struct cgroup *cgrp) 412 { 413 return css_tryget(&cgrp->self); 414 } 415 416 static inline void cgroup_put(struct cgroup *cgrp) 417 { 418 css_put(&cgrp->self); 419 } 420 421 /** 422 * task_css_set_check - obtain a task's css_set with extra access conditions 423 * @task: the task to obtain css_set for 424 * @__c: extra condition expression to be passed to rcu_dereference_check() 425 * 426 * A task's css_set is RCU protected, initialized and exited while holding 427 * task_lock(), and can only be modified while holding both cgroup_mutex 428 * and task_lock() while the task is alive. This macro verifies that the 429 * caller is inside proper critical section and returns @task's css_set. 430 * 431 * The caller can also specify additional allowed conditions via @__c, such 432 * as locks used during the cgroup_subsys::attach() methods. 433 */ 434 #ifdef CONFIG_PROVE_RCU 435 extern struct mutex cgroup_mutex; 436 extern spinlock_t css_set_lock; 437 #define task_css_set_check(task, __c) \ 438 rcu_dereference_check((task)->cgroups, \ 439 lockdep_is_held(&cgroup_mutex) || \ 440 lockdep_is_held(&css_set_lock) || \ 441 ((task)->flags & PF_EXITING) || (__c)) 442 #else 443 #define task_css_set_check(task, __c) \ 444 rcu_dereference((task)->cgroups) 445 #endif 446 447 /** 448 * task_css_check - obtain css for (task, subsys) w/ extra access conds 449 * @task: the target task 450 * @subsys_id: the target subsystem ID 451 * @__c: extra condition expression to be passed to rcu_dereference_check() 452 * 453 * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The 454 * synchronization rules are the same as task_css_set_check(). 455 */ 456 #define task_css_check(task, subsys_id, __c) \ 457 task_css_set_check((task), (__c))->subsys[(subsys_id)] 458 459 /** 460 * task_css_set - obtain a task's css_set 461 * @task: the task to obtain css_set for 462 * 463 * See task_css_set_check(). 464 */ 465 static inline struct css_set *task_css_set(struct task_struct *task) 466 { 467 return task_css_set_check(task, false); 468 } 469 470 /** 471 * task_css - obtain css for (task, subsys) 472 * @task: the target task 473 * @subsys_id: the target subsystem ID 474 * 475 * See task_css_check(). 476 */ 477 static inline struct cgroup_subsys_state *task_css(struct task_struct *task, 478 int subsys_id) 479 { 480 return task_css_check(task, subsys_id, false); 481 } 482 483 /** 484 * task_get_css - find and get the css for (task, subsys) 485 * @task: the target task 486 * @subsys_id: the target subsystem ID 487 * 488 * Find the css for the (@task, @subsys_id) combination, increment a 489 * reference on and return it. This function is guaranteed to return a 490 * valid css. 491 */ 492 static inline struct cgroup_subsys_state * 493 task_get_css(struct task_struct *task, int subsys_id) 494 { 495 struct cgroup_subsys_state *css; 496 497 rcu_read_lock(); 498 while (true) { 499 css = task_css(task, subsys_id); 500 if (likely(css_tryget_online(css))) 501 break; 502 cpu_relax(); 503 } 504 rcu_read_unlock(); 505 return css; 506 } 507 508 /** 509 * task_css_is_root - test whether a task belongs to the root css 510 * @task: the target task 511 * @subsys_id: the target subsystem ID 512 * 513 * Test whether @task belongs to the root css on the specified subsystem. 514 * May be invoked in any context. 515 */ 516 static inline bool task_css_is_root(struct task_struct *task, int subsys_id) 517 { 518 return task_css_check(task, subsys_id, true) == 519 init_css_set.subsys[subsys_id]; 520 } 521 522 static inline struct cgroup *task_cgroup(struct task_struct *task, 523 int subsys_id) 524 { 525 return task_css(task, subsys_id)->cgroup; 526 } 527 528 static inline struct cgroup *task_dfl_cgroup(struct task_struct *task) 529 { 530 return task_css_set(task)->dfl_cgrp; 531 } 532 533 static inline struct cgroup *cgroup_parent(struct cgroup *cgrp) 534 { 535 struct cgroup_subsys_state *parent_css = cgrp->self.parent; 536 537 if (parent_css) 538 return container_of(parent_css, struct cgroup, self); 539 return NULL; 540 } 541 542 /** 543 * cgroup_is_descendant - test ancestry 544 * @cgrp: the cgroup to be tested 545 * @ancestor: possible ancestor of @cgrp 546 * 547 * Test whether @cgrp is a descendant of @ancestor. It also returns %true 548 * if @cgrp == @ancestor. This function is safe to call as long as @cgrp 549 * and @ancestor are accessible. 550 */ 551 static inline bool cgroup_is_descendant(struct cgroup *cgrp, 552 struct cgroup *ancestor) 553 { 554 if (cgrp->root != ancestor->root || cgrp->level < ancestor->level) 555 return false; 556 return cgrp->ancestor_ids[ancestor->level] == ancestor->id; 557 } 558 559 /** 560 * cgroup_ancestor - find ancestor of cgroup 561 * @cgrp: cgroup to find ancestor of 562 * @ancestor_level: level of ancestor to find starting from root 563 * 564 * Find ancestor of cgroup at specified level starting from root if it exists 565 * and return pointer to it. Return NULL if @cgrp doesn't have ancestor at 566 * @ancestor_level. 567 * 568 * This function is safe to call as long as @cgrp is accessible. 569 */ 570 static inline struct cgroup *cgroup_ancestor(struct cgroup *cgrp, 571 int ancestor_level) 572 { 573 if (cgrp->level < ancestor_level) 574 return NULL; 575 while (cgrp && cgrp->level > ancestor_level) 576 cgrp = cgroup_parent(cgrp); 577 return cgrp; 578 } 579 580 /** 581 * task_under_cgroup_hierarchy - test task's membership of cgroup ancestry 582 * @task: the task to be tested 583 * @ancestor: possible ancestor of @task's cgroup 584 * 585 * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor. 586 * It follows all the same rules as cgroup_is_descendant, and only applies 587 * to the default hierarchy. 588 */ 589 static inline bool task_under_cgroup_hierarchy(struct task_struct *task, 590 struct cgroup *ancestor) 591 { 592 struct css_set *cset = task_css_set(task); 593 594 return cgroup_is_descendant(cset->dfl_cgrp, ancestor); 595 } 596 597 /* no synchronization, the result can only be used as a hint */ 598 static inline bool cgroup_is_populated(struct cgroup *cgrp) 599 { 600 return cgrp->nr_populated_csets + cgrp->nr_populated_domain_children + 601 cgrp->nr_populated_threaded_children; 602 } 603 604 /* returns ino associated with a cgroup */ 605 static inline ino_t cgroup_ino(struct cgroup *cgrp) 606 { 607 return cgrp->kn->id.ino; 608 } 609 610 /* cft/css accessors for cftype->write() operation */ 611 static inline struct cftype *of_cft(struct kernfs_open_file *of) 612 { 613 return of->kn->priv; 614 } 615 616 struct cgroup_subsys_state *of_css(struct kernfs_open_file *of); 617 618 /* cft/css accessors for cftype->seq_*() operations */ 619 static inline struct cftype *seq_cft(struct seq_file *seq) 620 { 621 return of_cft(seq->private); 622 } 623 624 static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq) 625 { 626 return of_css(seq->private); 627 } 628 629 /* 630 * Name / path handling functions. All are thin wrappers around the kernfs 631 * counterparts and can be called under any context. 632 */ 633 634 static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen) 635 { 636 return kernfs_name(cgrp->kn, buf, buflen); 637 } 638 639 static inline int cgroup_path(struct cgroup *cgrp, char *buf, size_t buflen) 640 { 641 return kernfs_path(cgrp->kn, buf, buflen); 642 } 643 644 static inline void pr_cont_cgroup_name(struct cgroup *cgrp) 645 { 646 pr_cont_kernfs_name(cgrp->kn); 647 } 648 649 static inline void pr_cont_cgroup_path(struct cgroup *cgrp) 650 { 651 pr_cont_kernfs_path(cgrp->kn); 652 } 653 654 static inline struct psi_group *cgroup_psi(struct cgroup *cgrp) 655 { 656 return &cgrp->psi; 657 } 658 659 static inline void cgroup_init_kthreadd(void) 660 { 661 /* 662 * kthreadd is inherited by all kthreads, keep it in the root so 663 * that the new kthreads are guaranteed to stay in the root until 664 * initialization is finished. 665 */ 666 current->no_cgroup_migration = 1; 667 } 668 669 static inline void cgroup_kthread_ready(void) 670 { 671 /* 672 * This kthread finished initialization. The creator should have 673 * set PF_NO_SETAFFINITY if this kthread should stay in the root. 674 */ 675 current->no_cgroup_migration = 0; 676 } 677 678 static inline union kernfs_node_id *cgroup_get_kernfs_id(struct cgroup *cgrp) 679 { 680 return &cgrp->kn->id; 681 } 682 683 void cgroup_path_from_kernfs_id(const union kernfs_node_id *id, 684 char *buf, size_t buflen); 685 #else /* !CONFIG_CGROUPS */ 686 687 struct cgroup_subsys_state; 688 struct cgroup; 689 690 static inline void css_put(struct cgroup_subsys_state *css) {} 691 static inline int cgroup_attach_task_all(struct task_struct *from, 692 struct task_struct *t) { return 0; } 693 static inline int cgroupstats_build(struct cgroupstats *stats, 694 struct dentry *dentry) { return -EINVAL; } 695 696 static inline void cgroup_fork(struct task_struct *p) {} 697 static inline int cgroup_can_fork(struct task_struct *p) { return 0; } 698 static inline void cgroup_cancel_fork(struct task_struct *p) {} 699 static inline void cgroup_post_fork(struct task_struct *p) {} 700 static inline void cgroup_exit(struct task_struct *p) {} 701 static inline void cgroup_release(struct task_struct *p) {} 702 static inline void cgroup_free(struct task_struct *p) {} 703 704 static inline int cgroup_init_early(void) { return 0; } 705 static inline int cgroup_init(void) { return 0; } 706 static inline void cgroup_init_kthreadd(void) {} 707 static inline void cgroup_kthread_ready(void) {} 708 static inline union kernfs_node_id *cgroup_get_kernfs_id(struct cgroup *cgrp) 709 { 710 return NULL; 711 } 712 713 static inline struct cgroup *cgroup_parent(struct cgroup *cgrp) 714 { 715 return NULL; 716 } 717 718 static inline struct psi_group *cgroup_psi(struct cgroup *cgrp) 719 { 720 return NULL; 721 } 722 723 static inline bool task_under_cgroup_hierarchy(struct task_struct *task, 724 struct cgroup *ancestor) 725 { 726 return true; 727 } 728 729 static inline void cgroup_path_from_kernfs_id(const union kernfs_node_id *id, 730 char *buf, size_t buflen) {} 731 #endif /* !CONFIG_CGROUPS */ 732 733 #ifdef CONFIG_CGROUPS 734 /* 735 * cgroup scalable recursive statistics. 736 */ 737 void cgroup_rstat_updated(struct cgroup *cgrp, int cpu); 738 void cgroup_rstat_flush(struct cgroup *cgrp); 739 void cgroup_rstat_flush_irqsafe(struct cgroup *cgrp); 740 void cgroup_rstat_flush_hold(struct cgroup *cgrp); 741 void cgroup_rstat_flush_release(void); 742 743 /* 744 * Basic resource stats. 745 */ 746 #ifdef CONFIG_CGROUP_CPUACCT 747 void cpuacct_charge(struct task_struct *tsk, u64 cputime); 748 void cpuacct_account_field(struct task_struct *tsk, int index, u64 val); 749 #else 750 static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} 751 static inline void cpuacct_account_field(struct task_struct *tsk, int index, 752 u64 val) {} 753 #endif 754 755 void __cgroup_account_cputime(struct cgroup *cgrp, u64 delta_exec); 756 void __cgroup_account_cputime_field(struct cgroup *cgrp, 757 enum cpu_usage_stat index, u64 delta_exec); 758 759 static inline void cgroup_account_cputime(struct task_struct *task, 760 u64 delta_exec) 761 { 762 struct cgroup *cgrp; 763 764 cpuacct_charge(task, delta_exec); 765 766 rcu_read_lock(); 767 cgrp = task_dfl_cgroup(task); 768 if (cgroup_parent(cgrp)) 769 __cgroup_account_cputime(cgrp, delta_exec); 770 rcu_read_unlock(); 771 } 772 773 static inline void cgroup_account_cputime_field(struct task_struct *task, 774 enum cpu_usage_stat index, 775 u64 delta_exec) 776 { 777 struct cgroup *cgrp; 778 779 cpuacct_account_field(task, index, delta_exec); 780 781 rcu_read_lock(); 782 cgrp = task_dfl_cgroup(task); 783 if (cgroup_parent(cgrp)) 784 __cgroup_account_cputime_field(cgrp, index, delta_exec); 785 rcu_read_unlock(); 786 } 787 788 #else /* CONFIG_CGROUPS */ 789 790 static inline void cgroup_account_cputime(struct task_struct *task, 791 u64 delta_exec) {} 792 static inline void cgroup_account_cputime_field(struct task_struct *task, 793 enum cpu_usage_stat index, 794 u64 delta_exec) {} 795 796 #endif /* CONFIG_CGROUPS */ 797 798 /* 799 * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data 800 * definition in cgroup-defs.h. 801 */ 802 #ifdef CONFIG_SOCK_CGROUP_DATA 803 804 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) 805 extern spinlock_t cgroup_sk_update_lock; 806 #endif 807 808 void cgroup_sk_alloc_disable(void); 809 void cgroup_sk_alloc(struct sock_cgroup_data *skcd); 810 void cgroup_sk_free(struct sock_cgroup_data *skcd); 811 812 static inline struct cgroup *sock_cgroup_ptr(struct sock_cgroup_data *skcd) 813 { 814 #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) 815 unsigned long v; 816 817 /* 818 * @skcd->val is 64bit but the following is safe on 32bit too as we 819 * just need the lower ulong to be written and read atomically. 820 */ 821 v = READ_ONCE(skcd->val); 822 823 if (v & 1) 824 return &cgrp_dfl_root.cgrp; 825 826 return (struct cgroup *)(unsigned long)v ?: &cgrp_dfl_root.cgrp; 827 #else 828 return (struct cgroup *)(unsigned long)skcd->val; 829 #endif 830 } 831 832 #else /* CONFIG_CGROUP_DATA */ 833 834 static inline void cgroup_sk_alloc(struct sock_cgroup_data *skcd) {} 835 static inline void cgroup_sk_free(struct sock_cgroup_data *skcd) {} 836 837 #endif /* CONFIG_CGROUP_DATA */ 838 839 struct cgroup_namespace { 840 refcount_t count; 841 struct ns_common ns; 842 struct user_namespace *user_ns; 843 struct ucounts *ucounts; 844 struct css_set *root_cset; 845 }; 846 847 extern struct cgroup_namespace init_cgroup_ns; 848 849 #ifdef CONFIG_CGROUPS 850 851 void free_cgroup_ns(struct cgroup_namespace *ns); 852 853 struct cgroup_namespace *copy_cgroup_ns(unsigned long flags, 854 struct user_namespace *user_ns, 855 struct cgroup_namespace *old_ns); 856 857 int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, 858 struct cgroup_namespace *ns); 859 860 #else /* !CONFIG_CGROUPS */ 861 862 static inline void free_cgroup_ns(struct cgroup_namespace *ns) { } 863 static inline struct cgroup_namespace * 864 copy_cgroup_ns(unsigned long flags, struct user_namespace *user_ns, 865 struct cgroup_namespace *old_ns) 866 { 867 return old_ns; 868 } 869 870 #endif /* !CONFIG_CGROUPS */ 871 872 static inline void get_cgroup_ns(struct cgroup_namespace *ns) 873 { 874 if (ns) 875 refcount_inc(&ns->count); 876 } 877 878 static inline void put_cgroup_ns(struct cgroup_namespace *ns) 879 { 880 if (ns && refcount_dec_and_test(&ns->count)) 881 free_cgroup_ns(ns); 882 } 883 884 #ifdef CONFIG_CGROUPS 885 886 void cgroup_enter_frozen(void); 887 void cgroup_leave_frozen(bool always_leave); 888 void cgroup_update_frozen(struct cgroup *cgrp); 889 void cgroup_freeze(struct cgroup *cgrp, bool freeze); 890 void cgroup_freezer_migrate_task(struct task_struct *task, struct cgroup *src, 891 struct cgroup *dst); 892 893 static inline bool cgroup_task_freeze(struct task_struct *task) 894 { 895 bool ret; 896 897 if (task->flags & PF_KTHREAD) 898 return false; 899 900 rcu_read_lock(); 901 ret = test_bit(CGRP_FREEZE, &task_dfl_cgroup(task)->flags); 902 rcu_read_unlock(); 903 904 return ret; 905 } 906 907 static inline bool cgroup_task_frozen(struct task_struct *task) 908 { 909 return task->frozen; 910 } 911 912 #else /* !CONFIG_CGROUPS */ 913 914 static inline void cgroup_enter_frozen(void) { } 915 static inline void cgroup_leave_frozen(bool always_leave) { } 916 static inline bool cgroup_task_freeze(struct task_struct *task) 917 { 918 return false; 919 } 920 static inline bool cgroup_task_frozen(struct task_struct *task) 921 { 922 return false; 923 } 924 925 #endif /* !CONFIG_CGROUPS */ 926 927 #endif /* _LINUX_CGROUP_H */ 928