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> |
fd1740b6 | 22 | #include <linux/bpf-cgroup-defs.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, | |
661ee628 CB |
74 | |
75 | /* Control group has to be killed. */ | |
76 | CGRP_KILL, | |
b4a04ab7 TH |
77 | }; |
78 | ||
79 | /* cgroup_root->flags */ | |
80 | enum { | |
b4a04ab7 TH |
81 | CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */ |
82 | CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */ | |
5136f636 TH |
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), | |
e1cba4b8 | 90 | |
6a010a49 TH |
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 | ||
e1cba4b8 WL |
104 | /* |
105 | * Enable cpuset controller in v1 cgroup to use v2 behavior. | |
106 | */ | |
6a010a49 | 107 | CGRP_ROOT_CPUSET_V2_MODE = (1 << 16), |
9852ae3f CD |
108 | |
109 | /* | |
110 | * Enable legacy local memory.events. | |
111 | */ | |
6a010a49 | 112 | CGRP_ROOT_MEMORY_LOCAL_EVENTS = (1 << 17), |
8a931f80 JW |
113 | |
114 | /* | |
115 | * Enable recursive subtree protection | |
116 | */ | |
6a010a49 | 117 | CGRP_ROOT_MEMORY_RECURSIVE_PROT = (1 << 18), |
8cba9576 NP |
118 | |
119 | /* | |
120 | * Enable hugetlb accounting for the memory controller. | |
121 | */ | |
122 | CGRP_ROOT_MEMORY_HUGETLB_ACCOUNTING = (1 << 19), | |
b4a04ab7 TH |
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 */ | |
5136f636 TH |
129 | CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */ |
130 | ||
b4a04ab7 | 131 | CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */ |
7dbdb199 | 132 | CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */ |
5cf8114d | 133 | CFTYPE_DEBUG = (1 << 5), /* create when cgroup_debug */ |
b4a04ab7 TH |
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 */ | |
0083d27b | 138 | __CFTYPE_ADDED = (1 << 18), |
b4a04ab7 TH |
139 | }; |
140 | ||
6f60eade TH |
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 */ | |
6f60eade | 148 | struct kernfs_node *kn; |
b12e3583 TH |
149 | unsigned long notified_at; |
150 | struct timer_list notify_timer; | |
6f60eade TH |
151 | }; |
152 | ||
b4a04ab7 TH |
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 | ||
b4a04ab7 TH |
170 | /* siblings list anchored at the parent's ->children */ |
171 | struct list_head sibling; | |
172 | struct list_head children; | |
173 | ||
8f53470b TH |
174 | /* flush target list anchored at cgrp->rstat_css_list */ |
175 | struct list_head rstat_css_node; | |
176 | ||
b4a04ab7 TH |
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 | ||
aa226ff4 TH |
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 | ||
b4a04ab7 | 199 | /* percpu_ref killing and RCU release */ |
b4a04ab7 | 200 | struct work_struct destroy_work; |
8f36aaec | 201 | struct rcu_work destroy_rwork; |
b8b1a2e5 TP |
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; | |
b4a04ab7 TH |
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 { | |
b4a04ab7 | 218 | /* |
5f617ebb TH |
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). | |
b4a04ab7 | 222 | */ |
5f617ebb TH |
223 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; |
224 | ||
225 | /* reference count */ | |
4b9502e6 | 226 | refcount_t refcount; |
5f617ebb | 227 | |
454000ad TH |
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 | ||
5f617ebb TH |
236 | /* the default cgroup associated with this css_set */ |
237 | struct cgroup *dfl_cgrp; | |
b4a04ab7 | 238 | |
73a7242a WL |
239 | /* internal task count, protected by css_set_lock */ |
240 | int nr_tasks; | |
241 | ||
b4a04ab7 TH |
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 | |
13cc9ee8 | 246 | * css_set_lock, but, during migration, once tasks are moved to |
b4a04ab7 TH |
247 | * mg_tasks, it can be read safely while holding cgroup_mutex. |
248 | */ | |
249 | struct list_head tasks; | |
250 | struct list_head mg_tasks; | |
c03cd773 | 251 | struct list_head dying_tasks; |
b4a04ab7 | 252 | |
5f617ebb TH |
253 | /* all css_task_iters currently walking this cset */ |
254 | struct list_head task_iters; | |
255 | ||
b4a04ab7 | 256 | /* |
08b2b6fd | 257 | * On the default hierarchy, ->subsys[ssid] may point to a css |
5f617ebb TH |
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. | |
b4a04ab7 | 262 | */ |
5f617ebb | 263 | struct list_head e_cset_node[CGROUP_SUBSYS_COUNT]; |
b4a04ab7 | 264 | |
454000ad TH |
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 | ||
5f617ebb TH |
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; | |
b4a04ab7 TH |
274 | |
275 | /* | |
5f617ebb TH |
276 | * List of cgrp_cset_links pointing at cgroups referenced from this |
277 | * css_set. Protected by css_set_lock. | |
b4a04ab7 | 278 | */ |
5f617ebb | 279 | struct list_head cgrp_links; |
b4a04ab7 TH |
280 | |
281 | /* | |
282 | * List of csets participating in the on-going migration either as | |
283 | * source or destination. Protected by cgroup_mutex. | |
284 | */ | |
07fd5b6c TH |
285 | struct list_head mg_src_preload_node; |
286 | struct list_head mg_dst_preload_node; | |
b4a04ab7 TH |
287 | struct list_head mg_node; |
288 | ||
289 | /* | |
290 | * If this cset is acting as the source of migration the following | |
e4857982 TH |
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. | |
b4a04ab7 TH |
295 | */ |
296 | struct cgroup *mg_src_cgrp; | |
e4857982 | 297 | struct cgroup *mg_dst_cgrp; |
b4a04ab7 TH |
298 | struct css_set *mg_dst_cset; |
299 | ||
2b021cbf TH |
300 | /* dead and being drained, ignore for migration */ |
301 | bool dead; | |
302 | ||
b4a04ab7 TH |
303 | /* For RCU-protected deletion */ |
304 | struct rcu_head rcu_head; | |
305 | }; | |
306 | ||
d4ff749b TH |
307 | struct cgroup_base_stat { |
308 | struct task_cputime cputime; | |
1fcf54de JD |
309 | |
310 | #ifdef CONFIG_SCHED_CORE | |
311 | u64 forceidle_sum; | |
312 | #endif | |
d4ff749b TH |
313 | }; |
314 | ||
041cd640 | 315 | /* |
c58632b3 TH |
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. | |
041cd640 | 319 | * |
c58632b3 | 320 | * When a stat gets updated, the cgroup_rstat_cpu and its ancestors are |
041cd640 TH |
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. | |
d4ff749b TH |
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. | |
041cd640 | 334 | */ |
c58632b3 | 335 | struct cgroup_rstat_cpu { |
041cd640 | 336 | /* |
d4ff749b TH |
337 | * ->bsync protects ->bstat. These are the only fields which get |
338 | * updated in the hot path. | |
041cd640 | 339 | */ |
d4ff749b TH |
340 | struct u64_stats_sync bsync; |
341 | struct cgroup_base_stat bstat; | |
041cd640 TH |
342 | |
343 | /* | |
344 | * Snapshots at the last reading. These are used to calculate the | |
345 | * deltas to propagate to the global counters. | |
346 | */ | |
d4ff749b | 347 | struct cgroup_base_stat last_bstat; |
041cd640 | 348 | |
0437719c HJ |
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 | ||
041cd640 TH |
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 | * | |
c58632b3 | 372 | * Protected by per-cpu cgroup_rstat_cpu_lock. |
041cd640 TH |
373 | */ |
374 | struct cgroup *updated_children; /* terminated by self cgroup */ | |
375 | struct cgroup *updated_next; /* NULL iff not on the list */ | |
376 | }; | |
377 | ||
76f969e8 RG |
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 | ||
b4a04ab7 TH |
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 | ||
b11cfb58 TH |
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 | |
7f203bc8 | 406 | * ancestors[] can determine whether a given cgroup is a |
b11cfb58 TH |
407 | * descendant of another without traversing the hierarchy. |
408 | */ | |
409 | int level; | |
410 | ||
1a926e0b RG |
411 | /* Maximum allowed descent tree depth */ |
412 | int max_depth; | |
413 | ||
0679dee0 RG |
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. | |
1a926e0b | 418 | * max_descendants is a maximum allowed number of descent cgroups. |
4dcabece RG |
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. | |
0679dee0 RG |
424 | */ |
425 | int nr_descendants; | |
426 | int nr_dying_descendants; | |
1a926e0b | 427 | int max_descendants; |
0679dee0 | 428 | |
b4a04ab7 | 429 | /* |
0de0942d | 430 | * Each non-empty css_set associated with this cgroup contributes |
788b950c TH |
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 | |
454000ad TH |
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. | |
b4a04ab7 | 439 | */ |
788b950c | 440 | int nr_populated_csets; |
454000ad TH |
441 | int nr_populated_domain_children; |
442 | int nr_populated_threaded_children; | |
443 | ||
444 | int nr_threaded_children; /* # of live threaded child cgroups */ | |
b4a04ab7 TH |
445 | |
446 | struct kernfs_node *kn; /* cgroup kernfs entry */ | |
6f60eade TH |
447 | struct cgroup_file procs_file; /* handle for "cgroup.procs" */ |
448 | struct cgroup_file events_file; /* handle for "cgroup.events" */ | |
b4a04ab7 | 449 | |
34f26a15 CZ |
450 | /* handles for "{cpu,memory,io,irq}.pressure" */ |
451 | struct cgroup_file psi_files[NR_PSI_RESOURCES]; | |
452 | ||
b4a04ab7 TH |
453 | /* |
454 | * The bitmask of subsystems enabled on the child cgroups. | |
455 | * ->subtree_control is the one configured through | |
af3bf054 | 456 | * "cgroup.subtree_control" while ->subtree_ss_mask is the effective |
8699b776 TH |
457 | * one which may have more subsystems enabled. Controller knobs |
458 | * are made available iff it's enabled in ->subtree_control. | |
b4a04ab7 | 459 | */ |
6e5c8307 TH |
460 | u16 subtree_control; |
461 | u16 subtree_ss_mask; | |
15a27c36 TH |
462 | u16 old_subtree_control; |
463 | u16 old_subtree_ss_mask; | |
b4a04ab7 TH |
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 | ||
454000ad TH |
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; | |
479adb89 | 493 | struct cgroup *old_dom_cgrp; /* used while enabling threaded */ |
454000ad | 494 | |
c58632b3 TH |
495 | /* per-cpu recursive resource statistics */ |
496 | struct cgroup_rstat_cpu __percpu *rstat_cpu; | |
8f53470b | 497 | struct list_head rstat_css_list; |
c58632b3 | 498 | |
041cd640 | 499 | /* cgroup basic resource statistics */ |
1bb5ec2e | 500 | struct cgroup_base_stat last_bstat; |
d4ff749b TH |
501 | struct cgroup_base_stat bstat; |
502 | struct prev_cputime prev_cputime; /* for printing out cputime */ | |
041cd640 | 503 | |
b4a04ab7 TH |
504 | /* |
505 | * list of pidlists, up to two for each namespace (one for procs, one | |
506 | * for tasks); created on demand. | |
507 | */ | |
508 | struct list_head pidlists; | |
509 | struct mutex pidlist_mutex; | |
510 | ||
511 | /* used to wait for offlining of csses */ | |
512 | wait_queue_head_t offline_waitq; | |
513 | ||
514 | /* used to schedule release agent */ | |
515 | struct work_struct release_agent_work; | |
b11cfb58 | 516 | |
2ce7135a | 517 | /* used to track pressure stalls */ |
5f69a657 | 518 | struct psi_group *psi; |
2ce7135a | 519 | |
30070984 DM |
520 | /* used to store eBPF programs */ |
521 | struct cgroup_bpf bpf; | |
522 | ||
d09d8df3 JB |
523 | /* If there is block congestion on this cgroup. */ |
524 | atomic_t congestion_count; | |
525 | ||
76f969e8 RG |
526 | /* Used to store internal freezer state */ |
527 | struct cgroup_freezer_state freezer; | |
528 | ||
c4bcfb38 YS |
529 | #ifdef CONFIG_BPF_SYSCALL |
530 | struct bpf_local_storage __rcu *bpf_cgrp_storage; | |
531 | #endif | |
532 | ||
7f203bc8 TH |
533 | /* All ancestors including self */ |
534 | struct cgroup *ancestors[]; | |
b4a04ab7 TH |
535 | }; |
536 | ||
537 | /* | |
538 | * A cgroup_root represents the root of a cgroup hierarchy, and may be | |
539 | * associated with a kernfs_root to form an active hierarchy. This is | |
540 | * internal to cgroup core. Don't access directly from controllers. | |
541 | */ | |
542 | struct cgroup_root { | |
543 | struct kernfs_root *kf_root; | |
544 | ||
545 | /* The bitmask of subsystems attached to this hierarchy */ | |
546 | unsigned int subsys_mask; | |
547 | ||
548 | /* Unique id for this hierarchy. */ | |
549 | int hierarchy_id; | |
550 | ||
7f203bc8 TH |
551 | /* |
552 | * The root cgroup. The containing cgroup_root will be destroyed on its | |
553 | * release. cgrp->ancestors[0] will be used overflowing into the | |
554 | * following field. cgrp_ancestor_storage must immediately follow. | |
555 | */ | |
b4a04ab7 TH |
556 | struct cgroup cgrp; |
557 | ||
7f203bc8 TH |
558 | /* must follow cgrp for cgrp->ancestors[0], see above */ |
559 | struct cgroup *cgrp_ancestor_storage; | |
b11cfb58 | 560 | |
b4a04ab7 TH |
561 | /* Number of cgroups in the hierarchy, used only for /proc/cgroups */ |
562 | atomic_t nr_cgrps; | |
563 | ||
564 | /* A list running through the active hierarchies */ | |
565 | struct list_head root_list; | |
566 | ||
567 | /* Hierarchy-specific flags */ | |
568 | unsigned int flags; | |
569 | ||
b4a04ab7 TH |
570 | /* The path to use for release notifications. */ |
571 | char release_agent_path[PATH_MAX]; | |
572 | ||
573 | /* The name for this hierarchy - may be empty */ | |
574 | char name[MAX_CGROUP_ROOT_NAMELEN]; | |
575 | }; | |
576 | ||
577 | /* | |
578 | * struct cftype: handler definitions for cgroup control files | |
579 | * | |
580 | * When reading/writing to a file: | |
581 | * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata | |
582 | * - the 'cftype' of the file is file->f_path.dentry->d_fsdata | |
583 | */ | |
584 | struct cftype { | |
585 | /* | |
586 | * By convention, the name should begin with the name of the | |
587 | * subsystem, followed by a period. Zero length string indicates | |
588 | * end of cftype array. | |
589 | */ | |
590 | char name[MAX_CFTYPE_NAME]; | |
731a981e | 591 | unsigned long private; |
b4a04ab7 TH |
592 | |
593 | /* | |
594 | * The maximum length of string, excluding trailing nul, that can | |
595 | * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed. | |
596 | */ | |
597 | size_t max_write_len; | |
598 | ||
599 | /* CFTYPE_* flags */ | |
600 | unsigned int flags; | |
601 | ||
6f60eade TH |
602 | /* |
603 | * If non-zero, should contain the offset from the start of css to | |
604 | * a struct cgroup_file field. cgroup will record the handle of | |
605 | * the created file into it. The recorded handle can be used as | |
606 | * long as the containing css remains accessible. | |
607 | */ | |
608 | unsigned int file_offset; | |
609 | ||
b4a04ab7 TH |
610 | /* |
611 | * Fields used for internal bookkeeping. Initialized automatically | |
612 | * during registration. | |
613 | */ | |
614 | struct cgroup_subsys *ss; /* NULL for cgroup core files */ | |
615 | struct list_head node; /* anchored at ss->cfts */ | |
616 | struct kernfs_ops *kf_ops; | |
617 | ||
e90cbebc TH |
618 | int (*open)(struct kernfs_open_file *of); |
619 | void (*release)(struct kernfs_open_file *of); | |
620 | ||
b4a04ab7 TH |
621 | /* |
622 | * read_u64() is a shortcut for the common case of returning a | |
623 | * single integer. Use it in place of read() | |
624 | */ | |
625 | u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft); | |
626 | /* | |
627 | * read_s64() is a signed version of read_u64() | |
628 | */ | |
629 | s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft); | |
630 | ||
631 | /* generic seq_file read interface */ | |
632 | int (*seq_show)(struct seq_file *sf, void *v); | |
633 | ||
634 | /* optional ops, implement all or none */ | |
635 | void *(*seq_start)(struct seq_file *sf, loff_t *ppos); | |
636 | void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos); | |
637 | void (*seq_stop)(struct seq_file *sf, void *v); | |
638 | ||
639 | /* | |
640 | * write_u64() is a shortcut for the common case of accepting | |
641 | * a single integer (as parsed by simple_strtoull) from | |
642 | * userspace. Use in place of write(); return 0 or error. | |
643 | */ | |
644 | int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft, | |
645 | u64 val); | |
646 | /* | |
647 | * write_s64() is a signed version of write_u64() | |
648 | */ | |
649 | int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft, | |
650 | s64 val); | |
651 | ||
652 | /* | |
653 | * write() is the generic write callback which maps directly to | |
654 | * kernfs write operation and overrides all other operations. | |
655 | * Maximum write size is determined by ->max_write_len. Use | |
656 | * of_css/cft() to access the associated css and cft. | |
657 | */ | |
658 | ssize_t (*write)(struct kernfs_open_file *of, | |
659 | char *buf, size_t nbytes, loff_t off); | |
660 | ||
dc50537b JW |
661 | __poll_t (*poll)(struct kernfs_open_file *of, |
662 | struct poll_table_struct *pt); | |
663 | ||
b4a04ab7 TH |
664 | #ifdef CONFIG_DEBUG_LOCK_ALLOC |
665 | struct lock_class_key lockdep_key; | |
666 | #endif | |
667 | }; | |
668 | ||
669 | /* | |
670 | * Control Group subsystem type. | |
da82c92f | 671 | * See Documentation/admin-guide/cgroup-v1/cgroups.rst for details |
b4a04ab7 TH |
672 | */ |
673 | struct cgroup_subsys { | |
674 | struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css); | |
675 | int (*css_online)(struct cgroup_subsys_state *css); | |
676 | void (*css_offline)(struct cgroup_subsys_state *css); | |
677 | void (*css_released)(struct cgroup_subsys_state *css); | |
678 | void (*css_free)(struct cgroup_subsys_state *css); | |
679 | void (*css_reset)(struct cgroup_subsys_state *css); | |
8f53470b | 680 | void (*css_rstat_flush)(struct cgroup_subsys_state *css, int cpu); |
d41bf8c9 TH |
681 | int (*css_extra_stat_show)(struct seq_file *seq, |
682 | struct cgroup_subsys_state *css); | |
677ea015 JD |
683 | int (*css_local_stat_show)(struct seq_file *seq, |
684 | struct cgroup_subsys_state *css); | |
b4a04ab7 | 685 | |
1f7dd3e5 TH |
686 | int (*can_attach)(struct cgroup_taskset *tset); |
687 | void (*cancel_attach)(struct cgroup_taskset *tset); | |
688 | void (*attach)(struct cgroup_taskset *tset); | |
5cf1cacb | 689 | void (*post_attach)(void); |
ef2c41cf CB |
690 | int (*can_fork)(struct task_struct *task, |
691 | struct css_set *cset); | |
692 | void (*cancel_fork)(struct task_struct *task, struct css_set *cset); | |
b53202e6 | 693 | void (*fork)(struct task_struct *task); |
2e91fa7f | 694 | void (*exit)(struct task_struct *task); |
51bee5ab | 695 | void (*release)(struct task_struct *task); |
b4a04ab7 TH |
696 | void (*bind)(struct cgroup_subsys_state *root_css); |
697 | ||
b38e42e9 | 698 | bool early_init:1; |
b4a04ab7 | 699 | |
f6d635ad TH |
700 | /* |
701 | * If %true, the controller, on the default hierarchy, doesn't show | |
702 | * up in "cgroup.controllers" or "cgroup.subtree_control", is | |
703 | * implicitly enabled on all cgroups on the default hierarchy, and | |
704 | * bypasses the "no internal process" constraint. This is for | |
705 | * utility type controllers which is transparent to userland. | |
706 | * | |
707 | * An implicit controller can be stolen from the default hierarchy | |
708 | * anytime and thus must be okay with offline csses from previous | |
709 | * hierarchies coexisting with csses for the current one. | |
710 | */ | |
711 | bool implicit_on_dfl:1; | |
712 | ||
8cfd8147 TH |
713 | /* |
714 | * If %true, the controller, supports threaded mode on the default | |
715 | * hierarchy. In a threaded subtree, both process granularity and | |
716 | * no-internal-process constraint are ignored and a threaded | |
717 | * controllers should be able to handle that. | |
718 | * | |
719 | * Note that as an implicit controller is automatically enabled on | |
720 | * all cgroups on the default hierarchy, it should also be | |
721 | * threaded. implicit && !threaded is not supported. | |
722 | */ | |
723 | bool threaded:1; | |
724 | ||
08b2b6fd | 725 | /* the following two fields are initialized automatically during boot */ |
b4a04ab7 TH |
726 | int id; |
727 | const char *name; | |
728 | ||
3e1d2eed TH |
729 | /* optional, initialized automatically during boot if not set */ |
730 | const char *legacy_name; | |
731 | ||
b4a04ab7 TH |
732 | /* link to parent, protected by cgroup_lock() */ |
733 | struct cgroup_root *root; | |
734 | ||
735 | /* idr for css->id */ | |
736 | struct idr css_idr; | |
737 | ||
738 | /* | |
739 | * List of cftypes. Each entry is the first entry of an array | |
740 | * terminated by zero length name. | |
741 | */ | |
742 | struct list_head cfts; | |
743 | ||
744 | /* | |
745 | * Base cftypes which are automatically registered. The two can | |
746 | * point to the same array. | |
747 | */ | |
748 | struct cftype *dfl_cftypes; /* for the default hierarchy */ | |
749 | struct cftype *legacy_cftypes; /* for the legacy hierarchies */ | |
750 | ||
751 | /* | |
752 | * A subsystem may depend on other subsystems. When such subsystem | |
753 | * is enabled on a cgroup, the depended-upon subsystems are enabled | |
754 | * together if available. Subsystems enabled due to dependency are | |
755 | * not visible to userland until explicitly enabled. The following | |
756 | * specifies the mask of subsystems that this one depends on. | |
757 | */ | |
758 | unsigned int depends_on; | |
759 | }; | |
760 | ||
1ed13287 TH |
761 | extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
762 | ||
763 | /** | |
764 | * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups | |
765 | * @tsk: target task | |
766 | * | |
780de9dd IM |
767 | * Allows cgroup operations to synchronize against threadgroup changes |
768 | * using a percpu_rw_semaphore. | |
1ed13287 TH |
769 | */ |
770 | static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) | |
771 | { | |
772 | percpu_down_read(&cgroup_threadgroup_rwsem); | |
773 | } | |
774 | ||
775 | /** | |
776 | * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups | |
777 | * @tsk: target task | |
778 | * | |
780de9dd | 779 | * Counterpart of cgroup_threadcgroup_change_begin(). |
1ed13287 TH |
780 | */ |
781 | static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) | |
782 | { | |
783 | percpu_up_read(&cgroup_threadgroup_rwsem); | |
784 | } | |
7d7efec3 TH |
785 | |
786 | #else /* CONFIG_CGROUPS */ | |
787 | ||
cb4a3167 AS |
788 | #define CGROUP_SUBSYS_COUNT 0 |
789 | ||
780de9dd IM |
790 | static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) |
791 | { | |
792 | might_sleep(); | |
793 | } | |
794 | ||
7d7efec3 TH |
795 | static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {} |
796 | ||
b4a04ab7 | 797 | #endif /* CONFIG_CGROUPS */ |
7d7efec3 | 798 | |
2a56a1fe TH |
799 | #ifdef CONFIG_SOCK_CGROUP_DATA |
800 | ||
bd1060a1 TH |
801 | /* |
802 | * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains | |
803 | * per-socket cgroup information except for memcg association. | |
804 | * | |
8520e224 DB |
805 | * On legacy hierarchies, net_prio and net_cls controllers directly |
806 | * set attributes on each sock which can then be tested by the network | |
807 | * layer. On the default hierarchy, each sock is associated with the | |
808 | * cgroup it was created in and the networking layer can match the | |
809 | * cgroup directly. | |
bd1060a1 | 810 | */ |
2a56a1fe | 811 | struct sock_cgroup_data { |
8520e224 DB |
812 | struct cgroup *cgroup; /* v2 */ |
813 | #ifdef CONFIG_CGROUP_NET_CLASSID | |
814 | u32 classid; /* v1 */ | |
815 | #endif | |
816 | #ifdef CONFIG_CGROUP_NET_PRIO | |
817 | u16 prioidx; /* v1 */ | |
bd1060a1 | 818 | #endif |
2a56a1fe TH |
819 | }; |
820 | ||
4dcb31d4 | 821 | static inline u16 sock_cgroup_prioidx(const struct sock_cgroup_data *skcd) |
2a56a1fe | 822 | { |
8520e224 DB |
823 | #ifdef CONFIG_CGROUP_NET_PRIO |
824 | return READ_ONCE(skcd->prioidx); | |
825 | #else | |
826 | return 1; | |
827 | #endif | |
2a56a1fe TH |
828 | } |
829 | ||
4dcb31d4 | 830 | static inline u32 sock_cgroup_classid(const struct sock_cgroup_data *skcd) |
2a56a1fe | 831 | { |
8520e224 DB |
832 | #ifdef CONFIG_CGROUP_NET_CLASSID |
833 | return READ_ONCE(skcd->classid); | |
834 | #else | |
835 | return 0; | |
836 | #endif | |
2a56a1fe TH |
837 | } |
838 | ||
839 | static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd, | |
840 | u16 prioidx) | |
841 | { | |
8520e224 DB |
842 | #ifdef CONFIG_CGROUP_NET_PRIO |
843 | WRITE_ONCE(skcd->prioidx, prioidx); | |
844 | #endif | |
2a56a1fe TH |
845 | } |
846 | ||
847 | static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd, | |
848 | u32 classid) | |
849 | { | |
8520e224 DB |
850 | #ifdef CONFIG_CGROUP_NET_CLASSID |
851 | WRITE_ONCE(skcd->classid, classid); | |
852 | #endif | |
2a56a1fe TH |
853 | } |
854 | ||
855 | #else /* CONFIG_SOCK_CGROUP_DATA */ | |
856 | ||
857 | struct sock_cgroup_data { | |
858 | }; | |
859 | ||
860 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
861 | ||
b4a04ab7 | 862 | #endif /* _LINUX_CGROUP_DEFS_H */ |