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