Commit | Line | Data |
---|---|---|
b2441318 | 1 | /* SPDX-License-Identifier: GPL-2.0 */ |
b4a04ab7 TH |
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> | |
4b9502e6 | 17 | #include <linux/refcount.h> |
b4a04ab7 | 18 | #include <linux/percpu-refcount.h> |
7d7efec3 | 19 | #include <linux/percpu-rwsem.h> |
041cd640 | 20 | #include <linux/u64_stats_sync.h> |
b4a04ab7 | 21 | #include <linux/workqueue.h> |
30070984 | 22 | #include <linux/bpf-cgroup.h> |
2ce7135a | 23 | #include <linux/psi_types.h> |
b4a04ab7 TH |
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; | |
c80ef9e0 | 34 | struct seq_file; |
dc50537b | 35 | struct poll_table_struct; |
b4a04ab7 TH |
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 */ | |
88cb04b9 | 54 | CSS_VISIBLE = (1 << 3), /* css is visible to userland */ |
33c35aa4 | 55 | CSS_DYING = (1 << 4), /* css is dying */ |
b4a04ab7 TH |
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, | |
76f969e8 RG |
68 | |
69 | /* Control group has to be frozen. */ | |
70 | CGRP_FREEZE, | |
71 | ||
72 | /* Cgroup is frozen. */ | |
73 | CGRP_FROZEN, | |
b4a04ab7 TH |
74 | }; |
75 | ||
76 | /* cgroup_root->flags */ | |
77 | enum { | |
b4a04ab7 TH |
78 | CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */ |
79 | CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */ | |
5136f636 TH |
80 | |
81 | /* | |
82 | * Consider namespaces as delegation boundaries. If this flag is | |
83 | * set, controller specific interface files in a namespace root | |
84 | * aren't writeable from inside the namespace. | |
85 | */ | |
86 | CGRP_ROOT_NS_DELEGATE = (1 << 3), | |
e1cba4b8 WL |
87 | |
88 | /* | |
89 | * Enable cpuset controller in v1 cgroup to use v2 behavior. | |
90 | */ | |
91 | CGRP_ROOT_CPUSET_V2_MODE = (1 << 4), | |
9852ae3f CD |
92 | |
93 | /* | |
94 | * Enable legacy local memory.events. | |
95 | */ | |
96 | CGRP_ROOT_MEMORY_LOCAL_EVENTS = (1 << 5), | |
8a931f80 JW |
97 | |
98 | /* | |
99 | * Enable recursive subtree protection | |
100 | */ | |
101 | CGRP_ROOT_MEMORY_RECURSIVE_PROT = (1 << 6), | |
b4a04ab7 TH |
102 | }; |
103 | ||
104 | /* cftype->flags */ | |
105 | enum { | |
106 | CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */ | |
107 | CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */ | |
5136f636 TH |
108 | CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */ |
109 | ||
b4a04ab7 | 110 | CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */ |
7dbdb199 | 111 | CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */ |
5cf8114d | 112 | CFTYPE_DEBUG = (1 << 5), /* create when cgroup_debug */ |
b4a04ab7 TH |
113 | |
114 | /* internal flags, do not use outside cgroup core proper */ | |
115 | __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */ | |
116 | __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */ | |
117 | }; | |
118 | ||
6f60eade TH |
119 | /* |
120 | * cgroup_file is the handle for a file instance created in a cgroup which | |
121 | * is used, for example, to generate file changed notifications. This can | |
122 | * be obtained by setting cftype->file_offset. | |
123 | */ | |
124 | struct cgroup_file { | |
125 | /* do not access any fields from outside cgroup core */ | |
6f60eade | 126 | struct kernfs_node *kn; |
b12e3583 TH |
127 | unsigned long notified_at; |
128 | struct timer_list notify_timer; | |
6f60eade TH |
129 | }; |
130 | ||
b4a04ab7 TH |
131 | /* |
132 | * Per-subsystem/per-cgroup state maintained by the system. This is the | |
133 | * fundamental structural building block that controllers deal with. | |
134 | * | |
135 | * Fields marked with "PI:" are public and immutable and may be accessed | |
136 | * directly without synchronization. | |
137 | */ | |
138 | struct cgroup_subsys_state { | |
139 | /* PI: the cgroup that this css is attached to */ | |
140 | struct cgroup *cgroup; | |
141 | ||
142 | /* PI: the cgroup subsystem that this css is attached to */ | |
143 | struct cgroup_subsys *ss; | |
144 | ||
145 | /* reference count - access via css_[try]get() and css_put() */ | |
146 | struct percpu_ref refcnt; | |
147 | ||
b4a04ab7 TH |
148 | /* siblings list anchored at the parent's ->children */ |
149 | struct list_head sibling; | |
150 | struct list_head children; | |
151 | ||
8f53470b TH |
152 | /* flush target list anchored at cgrp->rstat_css_list */ |
153 | struct list_head rstat_css_node; | |
154 | ||
b4a04ab7 TH |
155 | /* |
156 | * PI: Subsys-unique ID. 0 is unused and root is always 1. The | |
157 | * matching css can be looked up using css_from_id(). | |
158 | */ | |
159 | int id; | |
160 | ||
161 | unsigned int flags; | |
162 | ||
163 | /* | |
164 | * Monotonically increasing unique serial number which defines a | |
165 | * uniform order among all csses. It's guaranteed that all | |
166 | * ->children lists are in the ascending order of ->serial_nr and | |
167 | * used to allow interrupting and resuming iterations. | |
168 | */ | |
169 | u64 serial_nr; | |
170 | ||
aa226ff4 TH |
171 | /* |
172 | * Incremented by online self and children. Used to guarantee that | |
173 | * parents are not offlined before their children. | |
174 | */ | |
175 | atomic_t online_cnt; | |
176 | ||
b4a04ab7 | 177 | /* percpu_ref killing and RCU release */ |
b4a04ab7 | 178 | struct work_struct destroy_work; |
8f36aaec | 179 | struct rcu_work destroy_rwork; |
b8b1a2e5 TP |
180 | |
181 | /* | |
182 | * PI: the parent css. Placed here for cache proximity to following | |
183 | * fields of the containing structure. | |
184 | */ | |
185 | struct cgroup_subsys_state *parent; | |
b4a04ab7 TH |
186 | }; |
187 | ||
188 | /* | |
189 | * A css_set is a structure holding pointers to a set of | |
190 | * cgroup_subsys_state objects. This saves space in the task struct | |
191 | * object and speeds up fork()/exit(), since a single inc/dec and a | |
192 | * list_add()/del() can bump the reference count on the entire cgroup | |
193 | * set for a task. | |
194 | */ | |
195 | struct css_set { | |
b4a04ab7 | 196 | /* |
5f617ebb TH |
197 | * Set of subsystem states, one for each subsystem. This array is |
198 | * immutable after creation apart from the init_css_set during | |
199 | * subsystem registration (at boot time). | |
b4a04ab7 | 200 | */ |
5f617ebb TH |
201 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; |
202 | ||
203 | /* reference count */ | |
4b9502e6 | 204 | refcount_t refcount; |
5f617ebb | 205 | |
454000ad TH |
206 | /* |
207 | * For a domain cgroup, the following points to self. If threaded, | |
208 | * to the matching cset of the nearest domain ancestor. The | |
209 | * dom_cset provides access to the domain cgroup and its csses to | |
210 | * which domain level resource consumptions should be charged. | |
211 | */ | |
212 | struct css_set *dom_cset; | |
213 | ||
5f617ebb TH |
214 | /* the default cgroup associated with this css_set */ |
215 | struct cgroup *dfl_cgrp; | |
b4a04ab7 | 216 | |
73a7242a WL |
217 | /* internal task count, protected by css_set_lock */ |
218 | int nr_tasks; | |
219 | ||
b4a04ab7 TH |
220 | /* |
221 | * Lists running through all tasks using this cgroup group. | |
222 | * mg_tasks lists tasks which belong to this cset but are in the | |
223 | * process of being migrated out or in. Protected by | |
224 | * css_set_rwsem, but, during migration, once tasks are moved to | |
225 | * mg_tasks, it can be read safely while holding cgroup_mutex. | |
226 | */ | |
227 | struct list_head tasks; | |
228 | struct list_head mg_tasks; | |
c03cd773 | 229 | struct list_head dying_tasks; |
b4a04ab7 | 230 | |
5f617ebb TH |
231 | /* all css_task_iters currently walking this cset */ |
232 | struct list_head task_iters; | |
233 | ||
b4a04ab7 | 234 | /* |
5f617ebb TH |
235 | * On the default hierarhcy, ->subsys[ssid] may point to a css |
236 | * attached to an ancestor instead of the cgroup this css_set is | |
237 | * associated with. The following node is anchored at | |
238 | * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to | |
239 | * iterate through all css's attached to a given cgroup. | |
b4a04ab7 | 240 | */ |
5f617ebb | 241 | struct list_head e_cset_node[CGROUP_SUBSYS_COUNT]; |
b4a04ab7 | 242 | |
454000ad TH |
243 | /* all threaded csets whose ->dom_cset points to this cset */ |
244 | struct list_head threaded_csets; | |
245 | struct list_head threaded_csets_node; | |
246 | ||
5f617ebb TH |
247 | /* |
248 | * List running through all cgroup groups in the same hash | |
249 | * slot. Protected by css_set_lock | |
250 | */ | |
251 | struct hlist_node hlist; | |
b4a04ab7 TH |
252 | |
253 | /* | |
5f617ebb TH |
254 | * List of cgrp_cset_links pointing at cgroups referenced from this |
255 | * css_set. Protected by css_set_lock. | |
b4a04ab7 | 256 | */ |
5f617ebb | 257 | struct list_head cgrp_links; |
b4a04ab7 TH |
258 | |
259 | /* | |
260 | * List of csets participating in the on-going migration either as | |
261 | * source or destination. Protected by cgroup_mutex. | |
262 | */ | |
263 | struct list_head mg_preload_node; | |
264 | struct list_head mg_node; | |
265 | ||
266 | /* | |
267 | * If this cset is acting as the source of migration the following | |
e4857982 TH |
268 | * two fields are set. mg_src_cgrp and mg_dst_cgrp are |
269 | * respectively the source and destination cgroups of the on-going | |
270 | * migration. mg_dst_cset is the destination cset the target tasks | |
271 | * on this cset should be migrated to. Protected by cgroup_mutex. | |
b4a04ab7 TH |
272 | */ |
273 | struct cgroup *mg_src_cgrp; | |
e4857982 | 274 | struct cgroup *mg_dst_cgrp; |
b4a04ab7 TH |
275 | struct css_set *mg_dst_cset; |
276 | ||
2b021cbf TH |
277 | /* dead and being drained, ignore for migration */ |
278 | bool dead; | |
279 | ||
b4a04ab7 TH |
280 | /* For RCU-protected deletion */ |
281 | struct rcu_head rcu_head; | |
282 | }; | |
283 | ||
d4ff749b TH |
284 | struct cgroup_base_stat { |
285 | struct task_cputime cputime; | |
286 | }; | |
287 | ||
041cd640 | 288 | /* |
c58632b3 TH |
289 | * rstat - cgroup scalable recursive statistics. Accounting is done |
290 | * per-cpu in cgroup_rstat_cpu which is then lazily propagated up the | |
291 | * hierarchy on reads. | |
041cd640 | 292 | * |
c58632b3 | 293 | * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are |
041cd640 TH |
294 | * linked into the updated tree. On the following read, propagation only |
295 | * considers and consumes the updated tree. This makes reading O(the | |
296 | * number of descendants which have been active since last read) instead of | |
297 | * O(the total number of descendants). | |
298 | * | |
299 | * This is important because there can be a lot of (draining) cgroups which | |
300 | * aren't active and stat may be read frequently. The combination can | |
301 | * become very expensive. By propagating selectively, increasing reading | |
302 | * frequency decreases the cost of each read. | |
d4ff749b TH |
303 | * |
304 | * This struct hosts both the fields which implement the above - | |
305 | * updated_children and updated_next - and the fields which track basic | |
306 | * resource statistics on top of it - bsync, bstat and last_bstat. | |
041cd640 | 307 | */ |
c58632b3 | 308 | struct cgroup_rstat_cpu { |
041cd640 | 309 | /* |
d4ff749b TH |
310 | * ->bsync protects ->bstat. These are the only fields which get |
311 | * updated in the hot path. | |
041cd640 | 312 | */ |
d4ff749b TH |
313 | struct u64_stats_sync bsync; |
314 | struct cgroup_base_stat bstat; | |
041cd640 TH |
315 | |
316 | /* | |
317 | * Snapshots at the last reading. These are used to calculate the | |
318 | * deltas to propagate to the global counters. | |
319 | */ | |
d4ff749b | 320 | struct cgroup_base_stat last_bstat; |
041cd640 TH |
321 | |
322 | /* | |
323 | * Child cgroups with stat updates on this cpu since the last read | |
324 | * are linked on the parent's ->updated_children through | |
325 | * ->updated_next. | |
326 | * | |
327 | * In addition to being more compact, singly-linked list pointing | |
328 | * to the cgroup makes it unnecessary for each per-cpu struct to | |
329 | * point back to the associated cgroup. | |
330 | * | |
c58632b3 | 331 | * Protected by per-cpu cgroup_rstat_cpu_lock. |
041cd640 TH |
332 | */ |
333 | struct cgroup *updated_children; /* terminated by self cgroup */ | |
334 | struct cgroup *updated_next; /* NULL iff not on the list */ | |
335 | }; | |
336 | ||
76f969e8 RG |
337 | struct cgroup_freezer_state { |
338 | /* Should the cgroup and its descendants be frozen. */ | |
339 | bool freeze; | |
340 | ||
341 | /* Should the cgroup actually be frozen? */ | |
342 | int e_freeze; | |
343 | ||
344 | /* Fields below are protected by css_set_lock */ | |
345 | ||
346 | /* Number of frozen descendant cgroups */ | |
347 | int nr_frozen_descendants; | |
348 | ||
349 | /* | |
350 | * Number of tasks, which are counted as frozen: | |
351 | * frozen, SIGSTOPped, and PTRACEd. | |
352 | */ | |
353 | int nr_frozen_tasks; | |
354 | }; | |
355 | ||
b4a04ab7 TH |
356 | struct cgroup { |
357 | /* self css with NULL ->ss, points back to this cgroup */ | |
358 | struct cgroup_subsys_state self; | |
359 | ||
360 | unsigned long flags; /* "unsigned long" so bitops work */ | |
361 | ||
b11cfb58 TH |
362 | /* |
363 | * The depth this cgroup is at. The root is at depth zero and each | |
364 | * step down the hierarchy increments the level. This along with | |
365 | * ancestor_ids[] can determine whether a given cgroup is a | |
366 | * descendant of another without traversing the hierarchy. | |
367 | */ | |
368 | int level; | |
369 | ||
1a926e0b RG |
370 | /* Maximum allowed descent tree depth */ |
371 | int max_depth; | |
372 | ||
0679dee0 RG |
373 | /* |
374 | * Keep track of total numbers of visible and dying descent cgroups. | |
375 | * Dying cgroups are cgroups which were deleted by a user, | |
376 | * but are still existing because someone else is holding a reference. | |
1a926e0b | 377 | * max_descendants is a maximum allowed number of descent cgroups. |
4dcabece RG |
378 | * |
379 | * nr_descendants and nr_dying_descendants are protected | |
380 | * by cgroup_mutex and css_set_lock. It's fine to read them holding | |
381 | * any of cgroup_mutex and css_set_lock; for writing both locks | |
382 | * should be held. | |
0679dee0 RG |
383 | */ |
384 | int nr_descendants; | |
385 | int nr_dying_descendants; | |
1a926e0b | 386 | int max_descendants; |
0679dee0 | 387 | |
b4a04ab7 | 388 | /* |
0de0942d | 389 | * Each non-empty css_set associated with this cgroup contributes |
788b950c TH |
390 | * one to nr_populated_csets. The counter is zero iff this cgroup |
391 | * doesn't have any tasks. | |
392 | * | |
393 | * All children which have non-zero nr_populated_csets and/or | |
454000ad TH |
394 | * nr_populated_children of their own contribute one to either |
395 | * nr_populated_domain_children or nr_populated_threaded_children | |
396 | * depending on their type. Each counter is zero iff all cgroups | |
397 | * of the type in the subtree proper don't have any tasks. | |
b4a04ab7 | 398 | */ |
788b950c | 399 | int nr_populated_csets; |
454000ad TH |
400 | int nr_populated_domain_children; |
401 | int nr_populated_threaded_children; | |
402 | ||
403 | int nr_threaded_children; /* # of live threaded child cgroups */ | |
b4a04ab7 TH |
404 | |
405 | struct kernfs_node *kn; /* cgroup kernfs entry */ | |
6f60eade TH |
406 | struct cgroup_file procs_file; /* handle for "cgroup.procs" */ |
407 | struct cgroup_file events_file; /* handle for "cgroup.events" */ | |
b4a04ab7 TH |
408 | |
409 | /* | |
410 | * The bitmask of subsystems enabled on the child cgroups. | |
411 | * ->subtree_control is the one configured through | |
8699b776 TH |
412 | * "cgroup.subtree_control" while ->child_ss_mask is the effective |
413 | * one which may have more subsystems enabled. Controller knobs | |
414 | * are made available iff it's enabled in ->subtree_control. | |
b4a04ab7 | 415 | */ |
6e5c8307 TH |
416 | u16 subtree_control; |
417 | u16 subtree_ss_mask; | |
15a27c36 TH |
418 | u16 old_subtree_control; |
419 | u16 old_subtree_ss_mask; | |
b4a04ab7 TH |
420 | |
421 | /* Private pointers for each registered subsystem */ | |
422 | struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT]; | |
423 | ||
424 | struct cgroup_root *root; | |
425 | ||
426 | /* | |
427 | * List of cgrp_cset_links pointing at css_sets with tasks in this | |
428 | * cgroup. Protected by css_set_lock. | |
429 | */ | |
430 | struct list_head cset_links; | |
431 | ||
432 | /* | |
433 | * On the default hierarchy, a css_set for a cgroup with some | |
434 | * susbsys disabled will point to css's which are associated with | |
435 | * the closest ancestor which has the subsys enabled. The | |
436 | * following lists all css_sets which point to this cgroup's css | |
437 | * for the given subsystem. | |
438 | */ | |
439 | struct list_head e_csets[CGROUP_SUBSYS_COUNT]; | |
440 | ||
454000ad TH |
441 | /* |
442 | * If !threaded, self. If threaded, it points to the nearest | |
443 | * domain ancestor. Inside a threaded subtree, cgroups are exempt | |
444 | * from process granularity and no-internal-task constraint. | |
445 | * Domain level resource consumptions which aren't tied to a | |
446 | * specific task are charged to the dom_cgrp. | |
447 | */ | |
448 | struct cgroup *dom_cgrp; | |
479adb89 | 449 | struct cgroup *old_dom_cgrp; /* used while enabling threaded */ |
454000ad | 450 | |
c58632b3 TH |
451 | /* per-cpu recursive resource statistics */ |
452 | struct cgroup_rstat_cpu __percpu *rstat_cpu; | |
8f53470b | 453 | struct list_head rstat_css_list; |
c58632b3 | 454 | |
041cd640 | 455 | /* cgroup basic resource statistics */ |
1bb5ec2e | 456 | struct cgroup_base_stat last_bstat; |
d4ff749b TH |
457 | struct cgroup_base_stat bstat; |
458 | struct prev_cputime prev_cputime; /* for printing out cputime */ | |
041cd640 | 459 | |
b4a04ab7 TH |
460 | /* |
461 | * list of pidlists, up to two for each namespace (one for procs, one | |
462 | * for tasks); created on demand. | |
463 | */ | |
464 | struct list_head pidlists; | |
465 | struct mutex pidlist_mutex; | |
466 | ||
467 | /* used to wait for offlining of csses */ | |
468 | wait_queue_head_t offline_waitq; | |
469 | ||
470 | /* used to schedule release agent */ | |
471 | struct work_struct release_agent_work; | |
b11cfb58 | 472 | |
2ce7135a JW |
473 | /* used to track pressure stalls */ |
474 | struct psi_group psi; | |
475 | ||
30070984 DM |
476 | /* used to store eBPF programs */ |
477 | struct cgroup_bpf bpf; | |
478 | ||
d09d8df3 JB |
479 | /* If there is block congestion on this cgroup. */ |
480 | atomic_t congestion_count; | |
481 | ||
76f969e8 RG |
482 | /* Used to store internal freezer state */ |
483 | struct cgroup_freezer_state freezer; | |
484 | ||
b11cfb58 | 485 | /* ids of the ancestors at each level including self */ |
74321038 | 486 | u64 ancestor_ids[]; |
b4a04ab7 TH |
487 | }; |
488 | ||
489 | /* | |
490 | * A cgroup_root represents the root of a cgroup hierarchy, and may be | |
491 | * associated with a kernfs_root to form an active hierarchy. This is | |
492 | * internal to cgroup core. Don't access directly from controllers. | |
493 | */ | |
494 | struct cgroup_root { | |
495 | struct kernfs_root *kf_root; | |
496 | ||
497 | /* The bitmask of subsystems attached to this hierarchy */ | |
498 | unsigned int subsys_mask; | |
499 | ||
500 | /* Unique id for this hierarchy. */ | |
501 | int hierarchy_id; | |
502 | ||
503 | /* The root cgroup. Root is destroyed on its release. */ | |
504 | struct cgroup cgrp; | |
505 | ||
b11cfb58 | 506 | /* for cgrp->ancestor_ids[0] */ |
74321038 | 507 | u64 cgrp_ancestor_id_storage; |
b11cfb58 | 508 | |
b4a04ab7 TH |
509 | /* Number of cgroups in the hierarchy, used only for /proc/cgroups */ |
510 | atomic_t nr_cgrps; | |
511 | ||
512 | /* A list running through the active hierarchies */ | |
513 | struct list_head root_list; | |
514 | ||
515 | /* Hierarchy-specific flags */ | |
516 | unsigned int flags; | |
517 | ||
b4a04ab7 TH |
518 | /* The path to use for release notifications. */ |
519 | char release_agent_path[PATH_MAX]; | |
520 | ||
521 | /* The name for this hierarchy - may be empty */ | |
522 | char name[MAX_CGROUP_ROOT_NAMELEN]; | |
523 | }; | |
524 | ||
525 | /* | |
526 | * struct cftype: handler definitions for cgroup control files | |
527 | * | |
528 | * When reading/writing to a file: | |
529 | * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata | |
530 | * - the 'cftype' of the file is file->f_path.dentry->d_fsdata | |
531 | */ | |
532 | struct cftype { | |
533 | /* | |
534 | * By convention, the name should begin with the name of the | |
535 | * subsystem, followed by a period. Zero length string indicates | |
536 | * end of cftype array. | |
537 | */ | |
538 | char name[MAX_CFTYPE_NAME]; | |
731a981e | 539 | unsigned long private; |
b4a04ab7 TH |
540 | |
541 | /* | |
542 | * The maximum length of string, excluding trailing nul, that can | |
543 | * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed. | |
544 | */ | |
545 | size_t max_write_len; | |
546 | ||
547 | /* CFTYPE_* flags */ | |
548 | unsigned int flags; | |
549 | ||
6f60eade TH |
550 | /* |
551 | * If non-zero, should contain the offset from the start of css to | |
552 | * a struct cgroup_file field. cgroup will record the handle of | |
553 | * the created file into it. The recorded handle can be used as | |
554 | * long as the containing css remains accessible. | |
555 | */ | |
556 | unsigned int file_offset; | |
557 | ||
b4a04ab7 TH |
558 | /* |
559 | * Fields used for internal bookkeeping. Initialized automatically | |
560 | * during registration. | |
561 | */ | |
562 | struct cgroup_subsys *ss; /* NULL for cgroup core files */ | |
563 | struct list_head node; /* anchored at ss->cfts */ | |
564 | struct kernfs_ops *kf_ops; | |
565 | ||
e90cbebc TH |
566 | int (*open)(struct kernfs_open_file *of); |
567 | void (*release)(struct kernfs_open_file *of); | |
568 | ||
b4a04ab7 TH |
569 | /* |
570 | * read_u64() is a shortcut for the common case of returning a | |
571 | * single integer. Use it in place of read() | |
572 | */ | |
573 | u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft); | |
574 | /* | |
575 | * read_s64() is a signed version of read_u64() | |
576 | */ | |
577 | s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft); | |
578 | ||
579 | /* generic seq_file read interface */ | |
580 | int (*seq_show)(struct seq_file *sf, void *v); | |
581 | ||
582 | /* optional ops, implement all or none */ | |
583 | void *(*seq_start)(struct seq_file *sf, loff_t *ppos); | |
584 | void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos); | |
585 | void (*seq_stop)(struct seq_file *sf, void *v); | |
586 | ||
587 | /* | |
588 | * write_u64() is a shortcut for the common case of accepting | |
589 | * a single integer (as parsed by simple_strtoull) from | |
590 | * userspace. Use in place of write(); return 0 or error. | |
591 | */ | |
592 | int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft, | |
593 | u64 val); | |
594 | /* | |
595 | * write_s64() is a signed version of write_u64() | |
596 | */ | |
597 | int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft, | |
598 | s64 val); | |
599 | ||
600 | /* | |
601 | * write() is the generic write callback which maps directly to | |
602 | * kernfs write operation and overrides all other operations. | |
603 | * Maximum write size is determined by ->max_write_len. Use | |
604 | * of_css/cft() to access the associated css and cft. | |
605 | */ | |
606 | ssize_t (*write)(struct kernfs_open_file *of, | |
607 | char *buf, size_t nbytes, loff_t off); | |
608 | ||
dc50537b JW |
609 | __poll_t (*poll)(struct kernfs_open_file *of, |
610 | struct poll_table_struct *pt); | |
611 | ||
b4a04ab7 TH |
612 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
613 | struct lock_class_key lockdep_key; | |
614 | #endif | |
615 | }; | |
616 | ||
617 | /* | |
618 | * Control Group subsystem type. | |
da82c92f | 619 | * See Documentation/admin-guide/cgroup-v1/cgroups.rst for details |
b4a04ab7 TH |
620 | */ |
621 | struct cgroup_subsys { | |
622 | struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css); | |
623 | int (*css_online)(struct cgroup_subsys_state *css); | |
624 | void (*css_offline)(struct cgroup_subsys_state *css); | |
625 | void (*css_released)(struct cgroup_subsys_state *css); | |
626 | void (*css_free)(struct cgroup_subsys_state *css); | |
627 | void (*css_reset)(struct cgroup_subsys_state *css); | |
8f53470b | 628 | void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu); |
d41bf8c9 TH |
629 | int (*css_extra_stat_show)(struct seq_file *seq, |
630 | struct cgroup_subsys_state *css); | |
b4a04ab7 | 631 | |
1f7dd3e5 TH |
632 | int (*can_attach)(struct cgroup_taskset *tset); |
633 | void (*cancel_attach)(struct cgroup_taskset *tset); | |
634 | void (*attach)(struct cgroup_taskset *tset); | |
5cf1cacb | 635 | void (*post_attach)(void); |
ef2c41cf CB |
636 | int (*can_fork)(struct task_struct *task, |
637 | struct css_set *cset); | |
638 | void (*cancel_fork)(struct task_struct *task, struct css_set *cset); | |
b53202e6 | 639 | void (*fork)(struct task_struct *task); |
2e91fa7f | 640 | void (*exit)(struct task_struct *task); |
51bee5ab | 641 | void (*release)(struct task_struct *task); |
b4a04ab7 TH |
642 | void (*bind)(struct cgroup_subsys_state *root_css); |
643 | ||
b38e42e9 | 644 | bool early_init:1; |
b4a04ab7 | 645 | |
f6d635ad TH |
646 | /* |
647 | * If %true, the controller, on the default hierarchy, doesn't show | |
648 | * up in "cgroup.controllers" or "cgroup.subtree_control", is | |
649 | * implicitly enabled on all cgroups on the default hierarchy, and | |
650 | * bypasses the "no internal process" constraint. This is for | |
651 | * utility type controllers which is transparent to userland. | |
652 | * | |
653 | * An implicit controller can be stolen from the default hierarchy | |
654 | * anytime and thus must be okay with offline csses from previous | |
655 | * hierarchies coexisting with csses for the current one. | |
656 | */ | |
657 | bool implicit_on_dfl:1; | |
658 | ||
8cfd8147 TH |
659 | /* |
660 | * If %true, the controller, supports threaded mode on the default | |
661 | * hierarchy. In a threaded subtree, both process granularity and | |
662 | * no-internal-process constraint are ignored and a threaded | |
663 | * controllers should be able to handle that. | |
664 | * | |
665 | * Note that as an implicit controller is automatically enabled on | |
666 | * all cgroups on the default hierarchy, it should also be | |
667 | * threaded. implicit && !threaded is not supported. | |
668 | */ | |
669 | bool threaded:1; | |
670 | ||
b4a04ab7 TH |
671 | /* |
672 | * If %false, this subsystem is properly hierarchical - | |
673 | * configuration, resource accounting and restriction on a parent | |
674 | * cgroup cover those of its children. If %true, hierarchy support | |
675 | * is broken in some ways - some subsystems ignore hierarchy | |
676 | * completely while others are only implemented half-way. | |
677 | * | |
678 | * It's now disallowed to create nested cgroups if the subsystem is | |
679 | * broken and cgroup core will emit a warning message on such | |
680 | * cases. Eventually, all subsystems will be made properly | |
681 | * hierarchical and this will go away. | |
682 | */ | |
b38e42e9 TH |
683 | bool broken_hierarchy:1; |
684 | bool warned_broken_hierarchy:1; | |
b4a04ab7 TH |
685 | |
686 | /* the following two fields are initialized automtically during boot */ | |
687 | int id; | |
688 | const char *name; | |
689 | ||
3e1d2eed TH |
690 | /* optional, initialized automatically during boot if not set */ |
691 | const char *legacy_name; | |
692 | ||
b4a04ab7 TH |
693 | /* link to parent, protected by cgroup_lock() */ |
694 | struct cgroup_root *root; | |
695 | ||
696 | /* idr for css->id */ | |
697 | struct idr css_idr; | |
698 | ||
699 | /* | |
700 | * List of cftypes. Each entry is the first entry of an array | |
701 | * terminated by zero length name. | |
702 | */ | |
703 | struct list_head cfts; | |
704 | ||
705 | /* | |
706 | * Base cftypes which are automatically registered. The two can | |
707 | * point to the same array. | |
708 | */ | |
709 | struct cftype *dfl_cftypes; /* for the default hierarchy */ | |
710 | struct cftype *legacy_cftypes; /* for the legacy hierarchies */ | |
711 | ||
712 | /* | |
713 | * A subsystem may depend on other subsystems. When such subsystem | |
714 | * is enabled on a cgroup, the depended-upon subsystems are enabled | |
715 | * together if available. Subsystems enabled due to dependency are | |
716 | * not visible to userland until explicitly enabled. The following | |
717 | * specifies the mask of subsystems that this one depends on. | |
718 | */ | |
719 | unsigned int depends_on; | |
720 | }; | |
721 | ||
1ed13287 TH |
722 | extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
723 | ||
724 | /** | |
725 | * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups | |
726 | * @tsk: target task | |
727 | * | |
780de9dd IM |
728 | * Allows cgroup operations to synchronize against threadgroup changes |
729 | * using a percpu_rw_semaphore. | |
1ed13287 TH |
730 | */ |
731 | static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) | |
732 | { | |
733 | percpu_down_read(&cgroup_threadgroup_rwsem); | |
734 | } | |
735 | ||
736 | /** | |
737 | * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups | |
738 | * @tsk: target task | |
739 | * | |
780de9dd | 740 | * Counterpart of cgroup_threadcgroup_change_begin(). |
1ed13287 TH |
741 | */ |
742 | static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) | |
743 | { | |
744 | percpu_up_read(&cgroup_threadgroup_rwsem); | |
745 | } | |
7d7efec3 TH |
746 | |
747 | #else /* CONFIG_CGROUPS */ | |
748 | ||
cb4a3167 AS |
749 | #define CGROUP_SUBSYS_COUNT 0 |
750 | ||
780de9dd IM |
751 | static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) |
752 | { | |
753 | might_sleep(); | |
754 | } | |
755 | ||
7d7efec3 TH |
756 | static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {} |
757 | ||
b4a04ab7 | 758 | #endif /* CONFIG_CGROUPS */ |
7d7efec3 | 759 | |
2a56a1fe TH |
760 | #ifdef CONFIG_SOCK_CGROUP_DATA |
761 | ||
bd1060a1 TH |
762 | /* |
763 | * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains | |
764 | * per-socket cgroup information except for memcg association. | |
765 | * | |
766 | * On legacy hierarchies, net_prio and net_cls controllers directly set | |
767 | * attributes on each sock which can then be tested by the network layer. | |
768 | * On the default hierarchy, each sock is associated with the cgroup it was | |
769 | * created in and the networking layer can match the cgroup directly. | |
770 | * | |
771 | * To avoid carrying all three cgroup related fields separately in sock, | |
772 | * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer. | |
773 | * On boot, sock_cgroup_data records the cgroup that the sock was created | |
774 | * in so that cgroup2 matches can be made; however, once either net_prio or | |
775 | * net_cls starts being used, the area is overriden to carry prioidx and/or | |
776 | * classid. The two modes are distinguished by whether the lowest bit is | |
777 | * set. Clear bit indicates cgroup pointer while set bit prioidx and | |
778 | * classid. | |
779 | * | |
780 | * While userland may start using net_prio or net_cls at any time, once | |
781 | * either is used, cgroup2 matching no longer works. There is no reason to | |
782 | * mix the two and this is in line with how legacy and v2 compatibility is | |
783 | * handled. On mode switch, cgroup references which are already being | |
784 | * pointed to by socks may be leaked. While this can be remedied by adding | |
785 | * synchronization around sock_cgroup_data, given that the number of leaked | |
786 | * cgroups is bound and highly unlikely to be high, this seems to be the | |
787 | * better trade-off. | |
788 | */ | |
2a56a1fe | 789 | struct sock_cgroup_data { |
bd1060a1 TH |
790 | union { |
791 | #ifdef __LITTLE_ENDIAN | |
792 | struct { | |
ad0f75e5 CW |
793 | u8 is_data : 1; |
794 | u8 no_refcnt : 1; | |
14b032b8 | 795 | u8 unused : 6; |
bd1060a1 TH |
796 | u8 padding; |
797 | u16 prioidx; | |
798 | u32 classid; | |
799 | } __packed; | |
800 | #else | |
801 | struct { | |
802 | u32 classid; | |
803 | u16 prioidx; | |
804 | u8 padding; | |
14b032b8 | 805 | u8 unused : 6; |
ad0f75e5 CW |
806 | u8 no_refcnt : 1; |
807 | u8 is_data : 1; | |
bd1060a1 TH |
808 | } __packed; |
809 | #endif | |
810 | u64 val; | |
811 | }; | |
2a56a1fe TH |
812 | }; |
813 | ||
bd1060a1 TH |
814 | /* |
815 | * There's a theoretical window where the following accessors race with | |
816 | * updaters and return part of the previous pointer as the prioidx or | |
817 | * classid. Such races are short-lived and the result isn't critical. | |
818 | */ | |
4dcb31d4 | 819 | static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd) |
2a56a1fe | 820 | { |
bd1060a1 TH |
821 | /* fallback to 1 which is always the ID of the root cgroup */ |
822 | return (skcd->is_data & 1) ? skcd->prioidx : 1; | |
2a56a1fe TH |
823 | } |
824 | ||
4dcb31d4 | 825 | static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd) |
2a56a1fe | 826 | { |
bd1060a1 TH |
827 | /* fallback to 0 which is the unconfigured default classid */ |
828 | return (skcd->is_data & 1) ? skcd->classid : 0; | |
2a56a1fe TH |
829 | } |
830 | ||
bd1060a1 TH |
831 | /* |
832 | * If invoked concurrently, the updaters may clobber each other. The | |
833 | * caller is responsible for synchronization. | |
834 | */ | |
2a56a1fe TH |
835 | static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd, |
836 | u16 prioidx) | |
837 | { | |
ad2c8c73 | 838 | struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }}; |
bd1060a1 TH |
839 | |
840 | if (sock_cgroup_prioidx(&skcd_buf) == prioidx) | |
841 | return; | |
842 | ||
843 | if (!(skcd_buf.is_data & 1)) { | |
844 | skcd_buf.val = 0; | |
845 | skcd_buf.is_data = 1; | |
846 | } | |
847 | ||
848 | skcd_buf.prioidx = prioidx; | |
849 | WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */ | |
2a56a1fe TH |
850 | } |
851 | ||
852 | static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd, | |
853 | u32 classid) | |
854 | { | |
ad2c8c73 | 855 | struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }}; |
bd1060a1 TH |
856 | |
857 | if (sock_cgroup_classid(&skcd_buf) == classid) | |
858 | return; | |
859 | ||
860 | if (!(skcd_buf.is_data & 1)) { | |
861 | skcd_buf.val = 0; | |
862 | skcd_buf.is_data = 1; | |
863 | } | |
864 | ||
865 | skcd_buf.classid = classid; | |
866 | WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */ | |
2a56a1fe TH |
867 | } |
868 | ||
869 | #else /* CONFIG_SOCK_CGROUP_DATA */ | |
870 | ||
871 | struct sock_cgroup_data { | |
872 | }; | |
873 | ||
874 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
875 | ||
b4a04ab7 | 876 | #endif /* _LINUX_CGROUP_DEFS_H */ |