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