Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 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 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 46 | /* internal flags */ |
| 47 | #define CSS_TASK_ITER_SKIPPED (1U << 16) |
| 48 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 49 | /* a css_task_iter should be treated as an opaque object */ |
| 50 | struct css_task_iter { |
| 51 | struct cgroup_subsys *ss; |
| 52 | unsigned int flags; |
| 53 | |
| 54 | struct list_head *cset_pos; |
| 55 | struct list_head *cset_head; |
| 56 | |
| 57 | struct list_head *tcset_pos; |
| 58 | struct list_head *tcset_head; |
| 59 | |
| 60 | struct list_head *task_pos; |
| 61 | struct list_head *tasks_head; |
| 62 | struct list_head *mg_tasks_head; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 63 | struct list_head *dying_tasks_head; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 64 | |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame^] | 65 | struct list_head *cur_tasks_head; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 66 | struct css_set *cur_cset; |
| 67 | struct css_set *cur_dcset; |
| 68 | struct task_struct *cur_task; |
| 69 | struct list_head iters_node; /* css_set->task_iters */ |
| 70 | }; |
| 71 | |
| 72 | extern struct cgroup_root cgrp_dfl_root; |
| 73 | extern struct css_set init_css_set; |
| 74 | |
| 75 | #define SUBSYS(_x) extern struct cgroup_subsys _x ## _cgrp_subsys; |
| 76 | #include <linux/cgroup_subsys.h> |
| 77 | #undef SUBSYS |
| 78 | |
| 79 | #define SUBSYS(_x) \ |
| 80 | extern struct static_key_true _x ## _cgrp_subsys_enabled_key; \ |
| 81 | extern struct static_key_true _x ## _cgrp_subsys_on_dfl_key; |
| 82 | #include <linux/cgroup_subsys.h> |
| 83 | #undef SUBSYS |
| 84 | |
| 85 | /** |
| 86 | * cgroup_subsys_enabled - fast test on whether a subsys is enabled |
| 87 | * @ss: subsystem in question |
| 88 | */ |
| 89 | #define cgroup_subsys_enabled(ss) \ |
| 90 | static_branch_likely(&ss ## _enabled_key) |
| 91 | |
| 92 | /** |
| 93 | * cgroup_subsys_on_dfl - fast test on whether a subsys is on default hierarchy |
| 94 | * @ss: subsystem in question |
| 95 | */ |
| 96 | #define cgroup_subsys_on_dfl(ss) \ |
| 97 | static_branch_likely(&ss ## _on_dfl_key) |
| 98 | |
| 99 | bool css_has_online_children(struct cgroup_subsys_state *css); |
| 100 | struct cgroup_subsys_state *css_from_id(int id, struct cgroup_subsys *ss); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 101 | struct cgroup_subsys_state *cgroup_e_css(struct cgroup *cgroup, |
| 102 | struct cgroup_subsys *ss); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 103 | struct cgroup_subsys_state *cgroup_get_e_css(struct cgroup *cgroup, |
| 104 | struct cgroup_subsys *ss); |
| 105 | struct cgroup_subsys_state *css_tryget_online_from_dir(struct dentry *dentry, |
| 106 | struct cgroup_subsys *ss); |
| 107 | |
| 108 | struct cgroup *cgroup_get_from_path(const char *path); |
| 109 | struct cgroup *cgroup_get_from_fd(int fd); |
| 110 | |
| 111 | int cgroup_attach_task_all(struct task_struct *from, struct task_struct *); |
| 112 | int cgroup_transfer_tasks(struct cgroup *to, struct cgroup *from); |
| 113 | |
| 114 | int cgroup_add_dfl_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); |
| 115 | int cgroup_add_legacy_cftypes(struct cgroup_subsys *ss, struct cftype *cfts); |
| 116 | int cgroup_rm_cftypes(struct cftype *cfts); |
| 117 | void cgroup_file_notify(struct cgroup_file *cfile); |
| 118 | |
| 119 | int task_cgroup_path(struct task_struct *task, char *buf, size_t buflen); |
| 120 | int cgroupstats_build(struct cgroupstats *stats, struct dentry *dentry); |
| 121 | int proc_cgroup_show(struct seq_file *m, struct pid_namespace *ns, |
| 122 | struct pid *pid, struct task_struct *tsk); |
| 123 | |
| 124 | void cgroup_fork(struct task_struct *p); |
| 125 | extern int cgroup_can_fork(struct task_struct *p); |
| 126 | extern void cgroup_cancel_fork(struct task_struct *p); |
| 127 | extern void cgroup_post_fork(struct task_struct *p); |
| 128 | void cgroup_exit(struct task_struct *p); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 129 | void cgroup_release(struct task_struct *p); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 130 | void cgroup_free(struct task_struct *p); |
| 131 | |
| 132 | int cgroup_init_early(void); |
| 133 | int cgroup_init(void); |
| 134 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 135 | int cgroup_parse_float(const char *input, unsigned dec_shift, s64 *v); |
| 136 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 137 | /* |
| 138 | * Iteration helpers and macros. |
| 139 | */ |
| 140 | |
| 141 | struct cgroup_subsys_state *css_next_child(struct cgroup_subsys_state *pos, |
| 142 | struct cgroup_subsys_state *parent); |
| 143 | struct cgroup_subsys_state *css_next_descendant_pre(struct cgroup_subsys_state *pos, |
| 144 | struct cgroup_subsys_state *css); |
| 145 | struct cgroup_subsys_state *css_rightmost_descendant(struct cgroup_subsys_state *pos); |
| 146 | struct cgroup_subsys_state *css_next_descendant_post(struct cgroup_subsys_state *pos, |
| 147 | struct cgroup_subsys_state *css); |
| 148 | |
| 149 | struct task_struct *cgroup_taskset_first(struct cgroup_taskset *tset, |
| 150 | struct cgroup_subsys_state **dst_cssp); |
| 151 | struct task_struct *cgroup_taskset_next(struct cgroup_taskset *tset, |
| 152 | struct cgroup_subsys_state **dst_cssp); |
| 153 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 154 | void cgroup_enable_task_cg_lists(void); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 155 | void css_task_iter_start(struct cgroup_subsys_state *css, unsigned int flags, |
| 156 | struct css_task_iter *it); |
| 157 | struct task_struct *css_task_iter_next(struct css_task_iter *it); |
| 158 | void css_task_iter_end(struct css_task_iter *it); |
| 159 | |
| 160 | /** |
| 161 | * css_for_each_child - iterate through children of a css |
| 162 | * @pos: the css * to use as the loop cursor |
| 163 | * @parent: css whose children to walk |
| 164 | * |
| 165 | * Walk @parent's children. Must be called under rcu_read_lock(). |
| 166 | * |
| 167 | * If a subsystem synchronizes ->css_online() and the start of iteration, a |
| 168 | * css which finished ->css_online() is guaranteed to be visible in the |
| 169 | * future iterations and will stay visible until the last reference is put. |
| 170 | * A css which hasn't finished ->css_online() or already finished |
| 171 | * ->css_offline() may show up during traversal. It's each subsystem's |
| 172 | * responsibility to synchronize against on/offlining. |
| 173 | * |
| 174 | * It is allowed to temporarily drop RCU read lock during iteration. The |
| 175 | * caller is responsible for ensuring that @pos remains accessible until |
| 176 | * the start of the next iteration by, for example, bumping the css refcnt. |
| 177 | */ |
| 178 | #define css_for_each_child(pos, parent) \ |
| 179 | for ((pos) = css_next_child(NULL, (parent)); (pos); \ |
| 180 | (pos) = css_next_child((pos), (parent))) |
| 181 | |
| 182 | /** |
| 183 | * css_for_each_descendant_pre - pre-order walk of a css's descendants |
| 184 | * @pos: the css * to use as the loop cursor |
| 185 | * @root: css whose descendants to walk |
| 186 | * |
| 187 | * Walk @root's descendants. @root is included in the iteration and the |
| 188 | * first node to be visited. Must be called under rcu_read_lock(). |
| 189 | * |
| 190 | * If a subsystem synchronizes ->css_online() and the start of iteration, a |
| 191 | * css which finished ->css_online() is guaranteed to be visible in the |
| 192 | * future iterations and will stay visible until the last reference is put. |
| 193 | * A css which hasn't finished ->css_online() or already finished |
| 194 | * ->css_offline() may show up during traversal. It's each subsystem's |
| 195 | * responsibility to synchronize against on/offlining. |
| 196 | * |
| 197 | * For example, the following guarantees that a descendant can't escape |
| 198 | * state updates of its ancestors. |
| 199 | * |
| 200 | * my_online(@css) |
| 201 | * { |
| 202 | * Lock @css's parent and @css; |
| 203 | * Inherit state from the parent; |
| 204 | * Unlock both. |
| 205 | * } |
| 206 | * |
| 207 | * my_update_state(@css) |
| 208 | * { |
| 209 | * css_for_each_descendant_pre(@pos, @css) { |
| 210 | * Lock @pos; |
| 211 | * if (@pos == @css) |
| 212 | * Update @css's state; |
| 213 | * else |
| 214 | * Verify @pos is alive and inherit state from its parent; |
| 215 | * Unlock @pos; |
| 216 | * } |
| 217 | * } |
| 218 | * |
| 219 | * As long as the inheriting step, including checking the parent state, is |
| 220 | * enclosed inside @pos locking, double-locking the parent isn't necessary |
| 221 | * while inheriting. The state update to the parent is guaranteed to be |
| 222 | * visible by walking order and, as long as inheriting operations to the |
| 223 | * same @pos are atomic to each other, multiple updates racing each other |
| 224 | * still result in the correct state. It's guaranateed that at least one |
| 225 | * inheritance happens for any css after the latest update to its parent. |
| 226 | * |
| 227 | * If checking parent's state requires locking the parent, each inheriting |
| 228 | * iteration should lock and unlock both @pos->parent and @pos. |
| 229 | * |
| 230 | * Alternatively, a subsystem may choose to use a single global lock to |
| 231 | * synchronize ->css_online() and ->css_offline() against tree-walking |
| 232 | * operations. |
| 233 | * |
| 234 | * It is allowed to temporarily drop RCU read lock during iteration. The |
| 235 | * caller is responsible for ensuring that @pos remains accessible until |
| 236 | * the start of the next iteration by, for example, bumping the css refcnt. |
| 237 | */ |
| 238 | #define css_for_each_descendant_pre(pos, css) \ |
| 239 | for ((pos) = css_next_descendant_pre(NULL, (css)); (pos); \ |
| 240 | (pos) = css_next_descendant_pre((pos), (css))) |
| 241 | |
| 242 | /** |
| 243 | * css_for_each_descendant_post - post-order walk of a css's descendants |
| 244 | * @pos: the css * to use as the loop cursor |
| 245 | * @css: css whose descendants to walk |
| 246 | * |
| 247 | * Similar to css_for_each_descendant_pre() but performs post-order |
| 248 | * traversal instead. @root is included in the iteration and the last |
| 249 | * node to be visited. |
| 250 | * |
| 251 | * If a subsystem synchronizes ->css_online() and the start of iteration, a |
| 252 | * css which finished ->css_online() is guaranteed to be visible in the |
| 253 | * future iterations and will stay visible until the last reference is put. |
| 254 | * A css which hasn't finished ->css_online() or already finished |
| 255 | * ->css_offline() may show up during traversal. It's each subsystem's |
| 256 | * responsibility to synchronize against on/offlining. |
| 257 | * |
| 258 | * Note that the walk visibility guarantee example described in pre-order |
| 259 | * walk doesn't apply the same to post-order walks. |
| 260 | */ |
| 261 | #define css_for_each_descendant_post(pos, css) \ |
| 262 | for ((pos) = css_next_descendant_post(NULL, (css)); (pos); \ |
| 263 | (pos) = css_next_descendant_post((pos), (css))) |
| 264 | |
| 265 | /** |
| 266 | * cgroup_taskset_for_each - iterate cgroup_taskset |
| 267 | * @task: the loop cursor |
| 268 | * @dst_css: the destination css |
| 269 | * @tset: taskset to iterate |
| 270 | * |
| 271 | * @tset may contain multiple tasks and they may belong to multiple |
| 272 | * processes. |
| 273 | * |
| 274 | * On the v2 hierarchy, there may be tasks from multiple processes and they |
| 275 | * may not share the source or destination csses. |
| 276 | * |
| 277 | * On traditional hierarchies, when there are multiple tasks in @tset, if a |
| 278 | * task of a process is in @tset, all tasks of the process are in @tset. |
| 279 | * Also, all are guaranteed to share the same source and destination csses. |
| 280 | * |
| 281 | * Iteration is not in any specific order. |
| 282 | */ |
| 283 | #define cgroup_taskset_for_each(task, dst_css, tset) \ |
| 284 | for ((task) = cgroup_taskset_first((tset), &(dst_css)); \ |
| 285 | (task); \ |
| 286 | (task) = cgroup_taskset_next((tset), &(dst_css))) |
| 287 | |
| 288 | /** |
| 289 | * cgroup_taskset_for_each_leader - iterate group leaders in a cgroup_taskset |
| 290 | * @leader: the loop cursor |
| 291 | * @dst_css: the destination css |
| 292 | * @tset: taskset to iterate |
| 293 | * |
| 294 | * Iterate threadgroup leaders of @tset. For single-task migrations, @tset |
| 295 | * may not contain any. |
| 296 | */ |
| 297 | #define cgroup_taskset_for_each_leader(leader, dst_css, tset) \ |
| 298 | for ((leader) = cgroup_taskset_first((tset), &(dst_css)); \ |
| 299 | (leader); \ |
| 300 | (leader) = cgroup_taskset_next((tset), &(dst_css))) \ |
| 301 | if ((leader) != (leader)->group_leader) \ |
| 302 | ; \ |
| 303 | else |
| 304 | |
| 305 | /* |
| 306 | * Inline functions. |
| 307 | */ |
| 308 | |
| 309 | /** |
| 310 | * css_get - obtain a reference on the specified css |
| 311 | * @css: target css |
| 312 | * |
| 313 | * The caller must already have a reference. |
| 314 | */ |
| 315 | static inline void css_get(struct cgroup_subsys_state *css) |
| 316 | { |
| 317 | if (!(css->flags & CSS_NO_REF)) |
| 318 | percpu_ref_get(&css->refcnt); |
| 319 | } |
| 320 | |
| 321 | /** |
| 322 | * css_get_many - obtain references on the specified css |
| 323 | * @css: target css |
| 324 | * @n: number of references to get |
| 325 | * |
| 326 | * The caller must already have a reference. |
| 327 | */ |
| 328 | static inline void css_get_many(struct cgroup_subsys_state *css, unsigned int n) |
| 329 | { |
| 330 | if (!(css->flags & CSS_NO_REF)) |
| 331 | percpu_ref_get_many(&css->refcnt, n); |
| 332 | } |
| 333 | |
| 334 | /** |
| 335 | * css_tryget - try to obtain a reference on the specified css |
| 336 | * @css: target css |
| 337 | * |
| 338 | * Obtain a reference on @css unless it already has reached zero and is |
| 339 | * being released. This function doesn't care whether @css is on or |
| 340 | * offline. The caller naturally needs to ensure that @css is accessible |
| 341 | * but doesn't have to be holding a reference on it - IOW, RCU protected |
| 342 | * access is good enough for this function. Returns %true if a reference |
| 343 | * count was successfully obtained; %false otherwise. |
| 344 | */ |
| 345 | static inline bool css_tryget(struct cgroup_subsys_state *css) |
| 346 | { |
| 347 | if (!(css->flags & CSS_NO_REF)) |
| 348 | return percpu_ref_tryget(&css->refcnt); |
| 349 | return true; |
| 350 | } |
| 351 | |
| 352 | /** |
| 353 | * css_tryget_online - try to obtain a reference on the specified css if online |
| 354 | * @css: target css |
| 355 | * |
| 356 | * Obtain a reference on @css if it's online. The caller naturally needs |
| 357 | * to ensure that @css is accessible but doesn't have to be holding a |
| 358 | * reference on it - IOW, RCU protected access is good enough for this |
| 359 | * function. Returns %true if a reference count was successfully obtained; |
| 360 | * %false otherwise. |
| 361 | */ |
| 362 | static inline bool css_tryget_online(struct cgroup_subsys_state *css) |
| 363 | { |
| 364 | if (!(css->flags & CSS_NO_REF)) |
| 365 | return percpu_ref_tryget_live(&css->refcnt); |
| 366 | return true; |
| 367 | } |
| 368 | |
| 369 | /** |
| 370 | * css_is_dying - test whether the specified css is dying |
| 371 | * @css: target css |
| 372 | * |
| 373 | * Test whether @css is in the process of offlining or already offline. In |
| 374 | * most cases, ->css_online() and ->css_offline() callbacks should be |
| 375 | * enough; however, the actual offline operations are RCU delayed and this |
| 376 | * test returns %true also when @css is scheduled to be offlined. |
| 377 | * |
| 378 | * This is useful, for example, when the use case requires synchronous |
| 379 | * behavior with respect to cgroup removal. cgroup removal schedules css |
| 380 | * offlining but the css can seem alive while the operation is being |
| 381 | * delayed. If the delay affects user visible semantics, this test can be |
| 382 | * used to resolve the situation. |
| 383 | */ |
| 384 | static inline bool css_is_dying(struct cgroup_subsys_state *css) |
| 385 | { |
| 386 | return !(css->flags & CSS_NO_REF) && percpu_ref_is_dying(&css->refcnt); |
| 387 | } |
| 388 | |
| 389 | /** |
| 390 | * css_put - put a css reference |
| 391 | * @css: target css |
| 392 | * |
| 393 | * Put a reference obtained via css_get() and css_tryget_online(). |
| 394 | */ |
| 395 | static inline void css_put(struct cgroup_subsys_state *css) |
| 396 | { |
| 397 | if (!(css->flags & CSS_NO_REF)) |
| 398 | percpu_ref_put(&css->refcnt); |
| 399 | } |
| 400 | |
| 401 | /** |
| 402 | * css_put_many - put css references |
| 403 | * @css: target css |
| 404 | * @n: number of references to put |
| 405 | * |
| 406 | * Put references obtained via css_get() and css_tryget_online(). |
| 407 | */ |
| 408 | static inline void css_put_many(struct cgroup_subsys_state *css, unsigned int n) |
| 409 | { |
| 410 | if (!(css->flags & CSS_NO_REF)) |
| 411 | percpu_ref_put_many(&css->refcnt, n); |
| 412 | } |
| 413 | |
| 414 | static inline void cgroup_get(struct cgroup *cgrp) |
| 415 | { |
| 416 | css_get(&cgrp->self); |
| 417 | } |
| 418 | |
| 419 | static inline bool cgroup_tryget(struct cgroup *cgrp) |
| 420 | { |
| 421 | return css_tryget(&cgrp->self); |
| 422 | } |
| 423 | |
| 424 | static inline void cgroup_put(struct cgroup *cgrp) |
| 425 | { |
| 426 | css_put(&cgrp->self); |
| 427 | } |
| 428 | |
| 429 | /** |
| 430 | * task_css_set_check - obtain a task's css_set with extra access conditions |
| 431 | * @task: the task to obtain css_set for |
| 432 | * @__c: extra condition expression to be passed to rcu_dereference_check() |
| 433 | * |
| 434 | * A task's css_set is RCU protected, initialized and exited while holding |
| 435 | * task_lock(), and can only be modified while holding both cgroup_mutex |
| 436 | * and task_lock() while the task is alive. This macro verifies that the |
| 437 | * caller is inside proper critical section and returns @task's css_set. |
| 438 | * |
| 439 | * The caller can also specify additional allowed conditions via @__c, such |
| 440 | * as locks used during the cgroup_subsys::attach() methods. |
| 441 | */ |
| 442 | #ifdef CONFIG_PROVE_RCU |
| 443 | extern struct mutex cgroup_mutex; |
| 444 | extern spinlock_t css_set_lock; |
| 445 | #define task_css_set_check(task, __c) \ |
| 446 | rcu_dereference_check((task)->cgroups, \ |
| 447 | lockdep_is_held(&cgroup_mutex) || \ |
| 448 | lockdep_is_held(&css_set_lock) || \ |
| 449 | ((task)->flags & PF_EXITING) || (__c)) |
| 450 | #else |
| 451 | #define task_css_set_check(task, __c) \ |
| 452 | rcu_dereference((task)->cgroups) |
| 453 | #endif |
| 454 | |
| 455 | /** |
| 456 | * task_css_check - obtain css for (task, subsys) w/ extra access conds |
| 457 | * @task: the target task |
| 458 | * @subsys_id: the target subsystem ID |
| 459 | * @__c: extra condition expression to be passed to rcu_dereference_check() |
| 460 | * |
| 461 | * Return the cgroup_subsys_state for the (@task, @subsys_id) pair. The |
| 462 | * synchronization rules are the same as task_css_set_check(). |
| 463 | */ |
| 464 | #define task_css_check(task, subsys_id, __c) \ |
| 465 | task_css_set_check((task), (__c))->subsys[(subsys_id)] |
| 466 | |
| 467 | /** |
| 468 | * task_css_set - obtain a task's css_set |
| 469 | * @task: the task to obtain css_set for |
| 470 | * |
| 471 | * See task_css_set_check(). |
| 472 | */ |
| 473 | static inline struct css_set *task_css_set(struct task_struct *task) |
| 474 | { |
| 475 | return task_css_set_check(task, false); |
| 476 | } |
| 477 | |
| 478 | /** |
| 479 | * task_css - obtain css for (task, subsys) |
| 480 | * @task: the target task |
| 481 | * @subsys_id: the target subsystem ID |
| 482 | * |
| 483 | * See task_css_check(). |
| 484 | */ |
| 485 | static inline struct cgroup_subsys_state *task_css(struct task_struct *task, |
| 486 | int subsys_id) |
| 487 | { |
| 488 | return task_css_check(task, subsys_id, false); |
| 489 | } |
| 490 | |
| 491 | /** |
| 492 | * task_get_css - find and get the css for (task, subsys) |
| 493 | * @task: the target task |
| 494 | * @subsys_id: the target subsystem ID |
| 495 | * |
| 496 | * Find the css for the (@task, @subsys_id) combination, increment a |
| 497 | * reference on and return it. This function is guaranteed to return a |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 498 | * valid css. The returned css may already have been offlined. |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 499 | */ |
| 500 | static inline struct cgroup_subsys_state * |
| 501 | task_get_css(struct task_struct *task, int subsys_id) |
| 502 | { |
| 503 | struct cgroup_subsys_state *css; |
| 504 | |
| 505 | rcu_read_lock(); |
| 506 | while (true) { |
| 507 | css = task_css(task, subsys_id); |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 508 | /* |
| 509 | * Can't use css_tryget_online() here. A task which has |
| 510 | * PF_EXITING set may stay associated with an offline css. |
| 511 | * If such task calls this function, css_tryget_online() |
| 512 | * will keep failing. |
| 513 | */ |
| 514 | if (likely(css_tryget(css))) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 515 | break; |
| 516 | cpu_relax(); |
| 517 | } |
| 518 | rcu_read_unlock(); |
| 519 | return css; |
| 520 | } |
| 521 | |
| 522 | /** |
| 523 | * task_css_is_root - test whether a task belongs to the root css |
| 524 | * @task: the target task |
| 525 | * @subsys_id: the target subsystem ID |
| 526 | * |
| 527 | * Test whether @task belongs to the root css on the specified subsystem. |
| 528 | * May be invoked in any context. |
| 529 | */ |
| 530 | static inline bool task_css_is_root(struct task_struct *task, int subsys_id) |
| 531 | { |
| 532 | return task_css_check(task, subsys_id, true) == |
| 533 | init_css_set.subsys[subsys_id]; |
| 534 | } |
| 535 | |
| 536 | static inline struct cgroup *task_cgroup(struct task_struct *task, |
| 537 | int subsys_id) |
| 538 | { |
| 539 | return task_css(task, subsys_id)->cgroup; |
| 540 | } |
| 541 | |
| 542 | static inline struct cgroup *task_dfl_cgroup(struct task_struct *task) |
| 543 | { |
| 544 | return task_css_set(task)->dfl_cgrp; |
| 545 | } |
| 546 | |
| 547 | static inline struct cgroup *cgroup_parent(struct cgroup *cgrp) |
| 548 | { |
| 549 | struct cgroup_subsys_state *parent_css = cgrp->self.parent; |
| 550 | |
| 551 | if (parent_css) |
| 552 | return container_of(parent_css, struct cgroup, self); |
| 553 | return NULL; |
| 554 | } |
| 555 | |
| 556 | /** |
| 557 | * cgroup_is_descendant - test ancestry |
| 558 | * @cgrp: the cgroup to be tested |
| 559 | * @ancestor: possible ancestor of @cgrp |
| 560 | * |
| 561 | * Test whether @cgrp is a descendant of @ancestor. It also returns %true |
| 562 | * if @cgrp == @ancestor. This function is safe to call as long as @cgrp |
| 563 | * and @ancestor are accessible. |
| 564 | */ |
| 565 | static inline bool cgroup_is_descendant(struct cgroup *cgrp, |
| 566 | struct cgroup *ancestor) |
| 567 | { |
| 568 | if (cgrp->root != ancestor->root || cgrp->level < ancestor->level) |
| 569 | return false; |
| 570 | return cgrp->ancestor_ids[ancestor->level] == ancestor->id; |
| 571 | } |
| 572 | |
| 573 | /** |
| 574 | * cgroup_ancestor - find ancestor of cgroup |
| 575 | * @cgrp: cgroup to find ancestor of |
| 576 | * @ancestor_level: level of ancestor to find starting from root |
| 577 | * |
| 578 | * Find ancestor of cgroup at specified level starting from root if it exists |
| 579 | * and return pointer to it. Return NULL if @cgrp doesn't have ancestor at |
| 580 | * @ancestor_level. |
| 581 | * |
| 582 | * This function is safe to call as long as @cgrp is accessible. |
| 583 | */ |
| 584 | static inline struct cgroup *cgroup_ancestor(struct cgroup *cgrp, |
| 585 | int ancestor_level) |
| 586 | { |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 587 | if (cgrp->level < ancestor_level) |
| 588 | return NULL; |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 589 | while (cgrp && cgrp->level > ancestor_level) |
| 590 | cgrp = cgroup_parent(cgrp); |
| 591 | return cgrp; |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 592 | } |
| 593 | |
| 594 | /** |
| 595 | * task_under_cgroup_hierarchy - test task's membership of cgroup ancestry |
| 596 | * @task: the task to be tested |
| 597 | * @ancestor: possible ancestor of @task's cgroup |
| 598 | * |
| 599 | * Tests whether @task's default cgroup hierarchy is a descendant of @ancestor. |
| 600 | * It follows all the same rules as cgroup_is_descendant, and only applies |
| 601 | * to the default hierarchy. |
| 602 | */ |
| 603 | static inline bool task_under_cgroup_hierarchy(struct task_struct *task, |
| 604 | struct cgroup *ancestor) |
| 605 | { |
| 606 | struct css_set *cset = task_css_set(task); |
| 607 | |
| 608 | return cgroup_is_descendant(cset->dfl_cgrp, ancestor); |
| 609 | } |
| 610 | |
| 611 | /* no synchronization, the result can only be used as a hint */ |
| 612 | static inline bool cgroup_is_populated(struct cgroup *cgrp) |
| 613 | { |
| 614 | return cgrp->nr_populated_csets + cgrp->nr_populated_domain_children + |
| 615 | cgrp->nr_populated_threaded_children; |
| 616 | } |
| 617 | |
| 618 | /* returns ino associated with a cgroup */ |
| 619 | static inline ino_t cgroup_ino(struct cgroup *cgrp) |
| 620 | { |
| 621 | return cgrp->kn->id.ino; |
| 622 | } |
| 623 | |
| 624 | /* cft/css accessors for cftype->write() operation */ |
| 625 | static inline struct cftype *of_cft(struct kernfs_open_file *of) |
| 626 | { |
| 627 | return of->kn->priv; |
| 628 | } |
| 629 | |
| 630 | struct cgroup_subsys_state *of_css(struct kernfs_open_file *of); |
| 631 | |
| 632 | /* cft/css accessors for cftype->seq_*() operations */ |
| 633 | static inline struct cftype *seq_cft(struct seq_file *seq) |
| 634 | { |
| 635 | return of_cft(seq->private); |
| 636 | } |
| 637 | |
| 638 | static inline struct cgroup_subsys_state *seq_css(struct seq_file *seq) |
| 639 | { |
| 640 | return of_css(seq->private); |
| 641 | } |
| 642 | |
| 643 | /* |
| 644 | * Name / path handling functions. All are thin wrappers around the kernfs |
| 645 | * counterparts and can be called under any context. |
| 646 | */ |
| 647 | |
| 648 | static inline int cgroup_name(struct cgroup *cgrp, char *buf, size_t buflen) |
| 649 | { |
| 650 | return kernfs_name(cgrp->kn, buf, buflen); |
| 651 | } |
| 652 | |
| 653 | static inline int cgroup_path(struct cgroup *cgrp, char *buf, size_t buflen) |
| 654 | { |
| 655 | return kernfs_path(cgrp->kn, buf, buflen); |
| 656 | } |
| 657 | |
| 658 | static inline void pr_cont_cgroup_name(struct cgroup *cgrp) |
| 659 | { |
| 660 | pr_cont_kernfs_name(cgrp->kn); |
| 661 | } |
| 662 | |
| 663 | static inline void pr_cont_cgroup_path(struct cgroup *cgrp) |
| 664 | { |
| 665 | pr_cont_kernfs_path(cgrp->kn); |
| 666 | } |
| 667 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 668 | static inline struct psi_group *cgroup_psi(struct cgroup *cgrp) |
| 669 | { |
| 670 | return &cgrp->psi; |
| 671 | } |
| 672 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 673 | static inline void cgroup_init_kthreadd(void) |
| 674 | { |
| 675 | /* |
| 676 | * kthreadd is inherited by all kthreads, keep it in the root so |
| 677 | * that the new kthreads are guaranteed to stay in the root until |
| 678 | * initialization is finished. |
| 679 | */ |
| 680 | current->no_cgroup_migration = 1; |
| 681 | } |
| 682 | |
| 683 | static inline void cgroup_kthread_ready(void) |
| 684 | { |
| 685 | /* |
| 686 | * This kthread finished initialization. The creator should have |
| 687 | * set PF_NO_SETAFFINITY if this kthread should stay in the root. |
| 688 | */ |
| 689 | current->no_cgroup_migration = 0; |
| 690 | } |
| 691 | |
| 692 | static inline union kernfs_node_id *cgroup_get_kernfs_id(struct cgroup *cgrp) |
| 693 | { |
| 694 | return &cgrp->kn->id; |
| 695 | } |
| 696 | |
| 697 | void cgroup_path_from_kernfs_id(const union kernfs_node_id *id, |
| 698 | char *buf, size_t buflen); |
| 699 | #else /* !CONFIG_CGROUPS */ |
| 700 | |
| 701 | struct cgroup_subsys_state; |
| 702 | struct cgroup; |
| 703 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 704 | static inline void css_get(struct cgroup_subsys_state *css) {} |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 705 | static inline void css_put(struct cgroup_subsys_state *css) {} |
| 706 | static inline int cgroup_attach_task_all(struct task_struct *from, |
| 707 | struct task_struct *t) { return 0; } |
| 708 | static inline int cgroupstats_build(struct cgroupstats *stats, |
| 709 | struct dentry *dentry) { return -EINVAL; } |
| 710 | |
| 711 | static inline void cgroup_fork(struct task_struct *p) {} |
| 712 | static inline int cgroup_can_fork(struct task_struct *p) { return 0; } |
| 713 | static inline void cgroup_cancel_fork(struct task_struct *p) {} |
| 714 | static inline void cgroup_post_fork(struct task_struct *p) {} |
| 715 | static inline void cgroup_exit(struct task_struct *p) {} |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 716 | static inline void cgroup_release(struct task_struct *p) {} |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 717 | static inline void cgroup_free(struct task_struct *p) {} |
| 718 | |
| 719 | static inline int cgroup_init_early(void) { return 0; } |
| 720 | static inline int cgroup_init(void) { return 0; } |
| 721 | static inline void cgroup_init_kthreadd(void) {} |
| 722 | static inline void cgroup_kthread_ready(void) {} |
| 723 | static inline union kernfs_node_id *cgroup_get_kernfs_id(struct cgroup *cgrp) |
| 724 | { |
| 725 | return NULL; |
| 726 | } |
| 727 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 728 | static inline struct cgroup *cgroup_parent(struct cgroup *cgrp) |
| 729 | { |
| 730 | return NULL; |
| 731 | } |
| 732 | |
| 733 | static inline struct psi_group *cgroup_psi(struct cgroup *cgrp) |
| 734 | { |
| 735 | return NULL; |
| 736 | } |
| 737 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 738 | static inline bool task_under_cgroup_hierarchy(struct task_struct *task, |
| 739 | struct cgroup *ancestor) |
| 740 | { |
| 741 | return true; |
| 742 | } |
| 743 | |
| 744 | static inline void cgroup_path_from_kernfs_id(const union kernfs_node_id *id, |
| 745 | char *buf, size_t buflen) {} |
| 746 | #endif /* !CONFIG_CGROUPS */ |
| 747 | |
| 748 | #ifdef CONFIG_CGROUPS |
| 749 | /* |
| 750 | * cgroup scalable recursive statistics. |
| 751 | */ |
| 752 | void cgroup_rstat_updated(struct cgroup *cgrp, int cpu); |
| 753 | void cgroup_rstat_flush(struct cgroup *cgrp); |
| 754 | void cgroup_rstat_flush_irqsafe(struct cgroup *cgrp); |
| 755 | void cgroup_rstat_flush_hold(struct cgroup *cgrp); |
| 756 | void cgroup_rstat_flush_release(void); |
| 757 | |
| 758 | /* |
| 759 | * Basic resource stats. |
| 760 | */ |
| 761 | #ifdef CONFIG_CGROUP_CPUACCT |
| 762 | void cpuacct_charge(struct task_struct *tsk, u64 cputime); |
| 763 | void cpuacct_account_field(struct task_struct *tsk, int index, u64 val); |
| 764 | #else |
| 765 | static inline void cpuacct_charge(struct task_struct *tsk, u64 cputime) {} |
| 766 | static inline void cpuacct_account_field(struct task_struct *tsk, int index, |
| 767 | u64 val) {} |
| 768 | #endif |
| 769 | |
| 770 | void __cgroup_account_cputime(struct cgroup *cgrp, u64 delta_exec); |
| 771 | void __cgroup_account_cputime_field(struct cgroup *cgrp, |
| 772 | enum cpu_usage_stat index, u64 delta_exec); |
| 773 | |
| 774 | static inline void cgroup_account_cputime(struct task_struct *task, |
| 775 | u64 delta_exec) |
| 776 | { |
| 777 | struct cgroup *cgrp; |
| 778 | |
| 779 | cpuacct_charge(task, delta_exec); |
| 780 | |
| 781 | rcu_read_lock(); |
| 782 | cgrp = task_dfl_cgroup(task); |
| 783 | if (cgroup_parent(cgrp)) |
| 784 | __cgroup_account_cputime(cgrp, delta_exec); |
| 785 | rcu_read_unlock(); |
| 786 | } |
| 787 | |
| 788 | static inline void cgroup_account_cputime_field(struct task_struct *task, |
| 789 | enum cpu_usage_stat index, |
| 790 | u64 delta_exec) |
| 791 | { |
| 792 | struct cgroup *cgrp; |
| 793 | |
| 794 | cpuacct_account_field(task, index, delta_exec); |
| 795 | |
| 796 | rcu_read_lock(); |
| 797 | cgrp = task_dfl_cgroup(task); |
| 798 | if (cgroup_parent(cgrp)) |
| 799 | __cgroup_account_cputime_field(cgrp, index, delta_exec); |
| 800 | rcu_read_unlock(); |
| 801 | } |
| 802 | |
| 803 | #else /* CONFIG_CGROUPS */ |
| 804 | |
| 805 | static inline void cgroup_account_cputime(struct task_struct *task, |
| 806 | u64 delta_exec) {} |
| 807 | static inline void cgroup_account_cputime_field(struct task_struct *task, |
| 808 | enum cpu_usage_stat index, |
| 809 | u64 delta_exec) {} |
| 810 | |
| 811 | #endif /* CONFIG_CGROUPS */ |
| 812 | |
| 813 | /* |
| 814 | * sock->sk_cgrp_data handling. For more info, see sock_cgroup_data |
| 815 | * definition in cgroup-defs.h. |
| 816 | */ |
| 817 | #ifdef CONFIG_SOCK_CGROUP_DATA |
| 818 | |
| 819 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) |
| 820 | extern spinlock_t cgroup_sk_update_lock; |
| 821 | #endif |
| 822 | |
| 823 | void cgroup_sk_alloc_disable(void); |
| 824 | void cgroup_sk_alloc(struct sock_cgroup_data *skcd); |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame^] | 825 | void cgroup_sk_clone(struct sock_cgroup_data *skcd); |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 826 | void cgroup_sk_free(struct sock_cgroup_data *skcd); |
| 827 | |
| 828 | static inline struct cgroup *sock_cgroup_ptr(struct sock_cgroup_data *skcd) |
| 829 | { |
| 830 | #if defined(CONFIG_CGROUP_NET_PRIO) || defined(CONFIG_CGROUP_NET_CLASSID) |
| 831 | unsigned long v; |
| 832 | |
| 833 | /* |
| 834 | * @skcd->val is 64bit but the following is safe on 32bit too as we |
| 835 | * just need the lower ulong to be written and read atomically. |
| 836 | */ |
| 837 | v = READ_ONCE(skcd->val); |
| 838 | |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame^] | 839 | if (v & 3) |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 840 | return &cgrp_dfl_root.cgrp; |
| 841 | |
| 842 | return (struct cgroup *)(unsigned long)v ?: &cgrp_dfl_root.cgrp; |
| 843 | #else |
| 844 | return (struct cgroup *)(unsigned long)skcd->val; |
| 845 | #endif |
| 846 | } |
| 847 | |
| 848 | #else /* CONFIG_CGROUP_DATA */ |
| 849 | |
| 850 | static inline void cgroup_sk_alloc(struct sock_cgroup_data *skcd) {} |
Olivier Deprez | 0e64123 | 2021-09-23 10:07:05 +0200 | [diff] [blame^] | 851 | static inline void cgroup_sk_clone(struct sock_cgroup_data *skcd) {} |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 852 | static inline void cgroup_sk_free(struct sock_cgroup_data *skcd) {} |
| 853 | |
| 854 | #endif /* CONFIG_CGROUP_DATA */ |
| 855 | |
| 856 | struct cgroup_namespace { |
| 857 | refcount_t count; |
| 858 | struct ns_common ns; |
| 859 | struct user_namespace *user_ns; |
| 860 | struct ucounts *ucounts; |
| 861 | struct css_set *root_cset; |
| 862 | }; |
| 863 | |
| 864 | extern struct cgroup_namespace init_cgroup_ns; |
| 865 | |
| 866 | #ifdef CONFIG_CGROUPS |
| 867 | |
| 868 | void free_cgroup_ns(struct cgroup_namespace *ns); |
| 869 | |
| 870 | struct cgroup_namespace *copy_cgroup_ns(unsigned long flags, |
| 871 | struct user_namespace *user_ns, |
| 872 | struct cgroup_namespace *old_ns); |
| 873 | |
| 874 | int cgroup_path_ns(struct cgroup *cgrp, char *buf, size_t buflen, |
| 875 | struct cgroup_namespace *ns); |
| 876 | |
| 877 | #else /* !CONFIG_CGROUPS */ |
| 878 | |
| 879 | static inline void free_cgroup_ns(struct cgroup_namespace *ns) { } |
| 880 | static inline struct cgroup_namespace * |
| 881 | copy_cgroup_ns(unsigned long flags, struct user_namespace *user_ns, |
| 882 | struct cgroup_namespace *old_ns) |
| 883 | { |
| 884 | return old_ns; |
| 885 | } |
| 886 | |
| 887 | #endif /* !CONFIG_CGROUPS */ |
| 888 | |
| 889 | static inline void get_cgroup_ns(struct cgroup_namespace *ns) |
| 890 | { |
| 891 | if (ns) |
| 892 | refcount_inc(&ns->count); |
| 893 | } |
| 894 | |
| 895 | static inline void put_cgroup_ns(struct cgroup_namespace *ns) |
| 896 | { |
| 897 | if (ns && refcount_dec_and_test(&ns->count)) |
| 898 | free_cgroup_ns(ns); |
| 899 | } |
| 900 | |
David Brazdil | 0f672f6 | 2019-12-10 10:32:29 +0000 | [diff] [blame] | 901 | #ifdef CONFIG_CGROUPS |
| 902 | |
| 903 | void cgroup_enter_frozen(void); |
| 904 | void cgroup_leave_frozen(bool always_leave); |
| 905 | void cgroup_update_frozen(struct cgroup *cgrp); |
| 906 | void cgroup_freeze(struct cgroup *cgrp, bool freeze); |
| 907 | void cgroup_freezer_migrate_task(struct task_struct *task, struct cgroup *src, |
| 908 | struct cgroup *dst); |
| 909 | |
| 910 | static inline bool cgroup_task_freeze(struct task_struct *task) |
| 911 | { |
| 912 | bool ret; |
| 913 | |
| 914 | if (task->flags & PF_KTHREAD) |
| 915 | return false; |
| 916 | |
| 917 | rcu_read_lock(); |
| 918 | ret = test_bit(CGRP_FREEZE, &task_dfl_cgroup(task)->flags); |
| 919 | rcu_read_unlock(); |
| 920 | |
| 921 | return ret; |
| 922 | } |
| 923 | |
| 924 | static inline bool cgroup_task_frozen(struct task_struct *task) |
| 925 | { |
| 926 | return task->frozen; |
| 927 | } |
| 928 | |
| 929 | #else /* !CONFIG_CGROUPS */ |
| 930 | |
| 931 | static inline void cgroup_enter_frozen(void) { } |
| 932 | static inline void cgroup_leave_frozen(bool always_leave) { } |
| 933 | static inline bool cgroup_task_freeze(struct task_struct *task) |
| 934 | { |
| 935 | return false; |
| 936 | } |
| 937 | static inline bool cgroup_task_frozen(struct task_struct *task) |
| 938 | { |
| 939 | return false; |
| 940 | } |
| 941 | |
| 942 | #endif /* !CONFIG_CGROUPS */ |
| 943 | |
| 944 | #ifdef CONFIG_CGROUP_BPF |
| 945 | static inline void cgroup_bpf_get(struct cgroup *cgrp) |
| 946 | { |
| 947 | percpu_ref_get(&cgrp->bpf.refcnt); |
| 948 | } |
| 949 | |
| 950 | static inline void cgroup_bpf_put(struct cgroup *cgrp) |
| 951 | { |
| 952 | percpu_ref_put(&cgrp->bpf.refcnt); |
| 953 | } |
| 954 | |
| 955 | #else /* CONFIG_CGROUP_BPF */ |
| 956 | |
| 957 | static inline void cgroup_bpf_get(struct cgroup *cgrp) {} |
| 958 | static inline void cgroup_bpf_put(struct cgroup *cgrp) {} |
| 959 | |
| 960 | #endif /* CONFIG_CGROUP_BPF */ |
| 961 | |
Andrew Scull | b4b6d4a | 2019-01-02 15:54:55 +0000 | [diff] [blame] | 962 | #endif /* _LINUX_CGROUP_H */ |