Commit | Line | Data |
---|---|---|
b4a04ab7 TH |
1 | /* |
2 | * linux/cgroup-defs.h - basic definitions for cgroup | |
3 | * | |
4 | * This file provides basic type and interface. Include this file directly | |
5 | * only if necessary to avoid cyclic dependencies. | |
6 | */ | |
7 | #ifndef _LINUX_CGROUP_DEFS_H | |
8 | #define _LINUX_CGROUP_DEFS_H | |
9 | ||
10 | #include <linux/limits.h> | |
11 | #include <linux/list.h> | |
12 | #include <linux/idr.h> | |
13 | #include <linux/wait.h> | |
14 | #include <linux/mutex.h> | |
15 | #include <linux/rcupdate.h> | |
4b9502e6 | 16 | #include <linux/refcount.h> |
b4a04ab7 | 17 | #include <linux/percpu-refcount.h> |
7d7efec3 | 18 | #include <linux/percpu-rwsem.h> |
b4a04ab7 | 19 | #include <linux/workqueue.h> |
30070984 | 20 | #include <linux/bpf-cgroup.h> |
b4a04ab7 TH |
21 | |
22 | #ifdef CONFIG_CGROUPS | |
23 | ||
24 | struct cgroup; | |
25 | struct cgroup_root; | |
26 | struct cgroup_subsys; | |
27 | struct cgroup_taskset; | |
28 | struct kernfs_node; | |
29 | struct kernfs_ops; | |
30 | struct kernfs_open_file; | |
c80ef9e0 | 31 | struct seq_file; |
b4a04ab7 TH |
32 | |
33 | #define MAX_CGROUP_TYPE_NAMELEN 32 | |
34 | #define MAX_CGROUP_ROOT_NAMELEN 64 | |
35 | #define MAX_CFTYPE_NAME 64 | |
36 | ||
37 | /* define the enumeration of all cgroup subsystems */ | |
38 | #define SUBSYS(_x) _x ## _cgrp_id, | |
39 | enum cgroup_subsys_id { | |
40 | #include <linux/cgroup_subsys.h> | |
41 | CGROUP_SUBSYS_COUNT, | |
42 | }; | |
43 | #undef SUBSYS | |
44 | ||
45 | /* bits in struct cgroup_subsys_state flags field */ | |
46 | enum { | |
47 | CSS_NO_REF = (1 << 0), /* no reference counting for this css */ | |
48 | CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */ | |
49 | CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */ | |
88cb04b9 | 50 | CSS_VISIBLE = (1 << 3), /* css is visible to userland */ |
33c35aa4 | 51 | CSS_DYING = (1 << 4), /* css is dying */ |
b4a04ab7 TH |
52 | }; |
53 | ||
54 | /* bits in struct cgroup flags field */ | |
55 | enum { | |
56 | /* Control Group requires release notifications to userspace */ | |
57 | CGRP_NOTIFY_ON_RELEASE, | |
58 | /* | |
59 | * Clone the parent's configuration when creating a new child | |
60 | * cpuset cgroup. For historical reasons, this option can be | |
61 | * specified at mount time and thus is implemented here. | |
62 | */ | |
63 | CGRP_CPUSET_CLONE_CHILDREN, | |
64 | }; | |
65 | ||
66 | /* cgroup_root->flags */ | |
67 | enum { | |
b4a04ab7 TH |
68 | CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */ |
69 | CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */ | |
5136f636 TH |
70 | |
71 | /* | |
72 | * Consider namespaces as delegation boundaries. If this flag is | |
73 | * set, controller specific interface files in a namespace root | |
74 | * aren't writeable from inside the namespace. | |
75 | */ | |
76 | CGRP_ROOT_NS_DELEGATE = (1 << 3), | |
b4a04ab7 TH |
77 | }; |
78 | ||
79 | /* cftype->flags */ | |
80 | enum { | |
81 | CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */ | |
82 | CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */ | |
5136f636 TH |
83 | CFTYPE_NS_DELEGATABLE = (1 << 2), /* writeable beyond delegation boundaries */ |
84 | ||
b4a04ab7 | 85 | CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */ |
7dbdb199 | 86 | CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */ |
b4a04ab7 TH |
87 | |
88 | /* internal flags, do not use outside cgroup core proper */ | |
89 | __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */ | |
90 | __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */ | |
91 | }; | |
92 | ||
6f60eade TH |
93 | /* |
94 | * cgroup_file is the handle for a file instance created in a cgroup which | |
95 | * is used, for example, to generate file changed notifications. This can | |
96 | * be obtained by setting cftype->file_offset. | |
97 | */ | |
98 | struct cgroup_file { | |
99 | /* do not access any fields from outside cgroup core */ | |
6f60eade TH |
100 | struct kernfs_node *kn; |
101 | }; | |
102 | ||
b4a04ab7 TH |
103 | /* |
104 | * Per-subsystem/per-cgroup state maintained by the system. This is the | |
105 | * fundamental structural building block that controllers deal with. | |
106 | * | |
107 | * Fields marked with "PI:" are public and immutable and may be accessed | |
108 | * directly without synchronization. | |
109 | */ | |
110 | struct cgroup_subsys_state { | |
111 | /* PI: the cgroup that this css is attached to */ | |
112 | struct cgroup *cgroup; | |
113 | ||
114 | /* PI: the cgroup subsystem that this css is attached to */ | |
115 | struct cgroup_subsys *ss; | |
116 | ||
117 | /* reference count - access via css_[try]get() and css_put() */ | |
118 | struct percpu_ref refcnt; | |
119 | ||
b4a04ab7 TH |
120 | /* siblings list anchored at the parent's ->children */ |
121 | struct list_head sibling; | |
122 | struct list_head children; | |
123 | ||
124 | /* | |
125 | * PI: Subsys-unique ID. 0 is unused and root is always 1. The | |
126 | * matching css can be looked up using css_from_id(). | |
127 | */ | |
128 | int id; | |
129 | ||
130 | unsigned int flags; | |
131 | ||
132 | /* | |
133 | * Monotonically increasing unique serial number which defines a | |
134 | * uniform order among all csses. It's guaranteed that all | |
135 | * ->children lists are in the ascending order of ->serial_nr and | |
136 | * used to allow interrupting and resuming iterations. | |
137 | */ | |
138 | u64 serial_nr; | |
139 | ||
aa226ff4 TH |
140 | /* |
141 | * Incremented by online self and children. Used to guarantee that | |
142 | * parents are not offlined before their children. | |
143 | */ | |
144 | atomic_t online_cnt; | |
145 | ||
b4a04ab7 TH |
146 | /* percpu_ref killing and RCU release */ |
147 | struct rcu_head rcu_head; | |
148 | struct work_struct destroy_work; | |
b8b1a2e5 TP |
149 | |
150 | /* | |
151 | * PI: the parent css. Placed here for cache proximity to following | |
152 | * fields of the containing structure. | |
153 | */ | |
154 | struct cgroup_subsys_state *parent; | |
b4a04ab7 TH |
155 | }; |
156 | ||
157 | /* | |
158 | * A css_set is a structure holding pointers to a set of | |
159 | * cgroup_subsys_state objects. This saves space in the task struct | |
160 | * object and speeds up fork()/exit(), since a single inc/dec and a | |
161 | * list_add()/del() can bump the reference count on the entire cgroup | |
162 | * set for a task. | |
163 | */ | |
164 | struct css_set { | |
b4a04ab7 | 165 | /* |
5f617ebb TH |
166 | * Set of subsystem states, one for each subsystem. This array is |
167 | * immutable after creation apart from the init_css_set during | |
168 | * subsystem registration (at boot time). | |
b4a04ab7 | 169 | */ |
5f617ebb TH |
170 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; |
171 | ||
172 | /* reference count */ | |
4b9502e6 | 173 | refcount_t refcount; |
5f617ebb TH |
174 | |
175 | /* the default cgroup associated with this css_set */ | |
176 | struct cgroup *dfl_cgrp; | |
b4a04ab7 | 177 | |
73a7242a WL |
178 | /* internal task count, protected by css_set_lock */ |
179 | int nr_tasks; | |
180 | ||
b4a04ab7 TH |
181 | /* |
182 | * Lists running through all tasks using this cgroup group. | |
183 | * mg_tasks lists tasks which belong to this cset but are in the | |
184 | * process of being migrated out or in. Protected by | |
185 | * css_set_rwsem, but, during migration, once tasks are moved to | |
186 | * mg_tasks, it can be read safely while holding cgroup_mutex. | |
187 | */ | |
188 | struct list_head tasks; | |
189 | struct list_head mg_tasks; | |
190 | ||
5f617ebb TH |
191 | /* all css_task_iters currently walking this cset */ |
192 | struct list_head task_iters; | |
193 | ||
b4a04ab7 | 194 | /* |
5f617ebb TH |
195 | * On the default hierarhcy, ->subsys[ssid] may point to a css |
196 | * attached to an ancestor instead of the cgroup this css_set is | |
197 | * associated with. The following node is anchored at | |
198 | * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to | |
199 | * iterate through all css's attached to a given cgroup. | |
b4a04ab7 | 200 | */ |
5f617ebb | 201 | struct list_head e_cset_node[CGROUP_SUBSYS_COUNT]; |
b4a04ab7 | 202 | |
5f617ebb TH |
203 | /* |
204 | * List running through all cgroup groups in the same hash | |
205 | * slot. Protected by css_set_lock | |
206 | */ | |
207 | struct hlist_node hlist; | |
b4a04ab7 TH |
208 | |
209 | /* | |
5f617ebb TH |
210 | * List of cgrp_cset_links pointing at cgroups referenced from this |
211 | * css_set. Protected by css_set_lock. | |
b4a04ab7 | 212 | */ |
5f617ebb | 213 | struct list_head cgrp_links; |
b4a04ab7 TH |
214 | |
215 | /* | |
216 | * List of csets participating in the on-going migration either as | |
217 | * source or destination. Protected by cgroup_mutex. | |
218 | */ | |
219 | struct list_head mg_preload_node; | |
220 | struct list_head mg_node; | |
221 | ||
222 | /* | |
223 | * If this cset is acting as the source of migration the following | |
e4857982 TH |
224 | * two fields are set. mg_src_cgrp and mg_dst_cgrp are |
225 | * respectively the source and destination cgroups of the on-going | |
226 | * migration. mg_dst_cset is the destination cset the target tasks | |
227 | * on this cset should be migrated to. Protected by cgroup_mutex. | |
b4a04ab7 TH |
228 | */ |
229 | struct cgroup *mg_src_cgrp; | |
e4857982 | 230 | struct cgroup *mg_dst_cgrp; |
b4a04ab7 TH |
231 | struct css_set *mg_dst_cset; |
232 | ||
2b021cbf TH |
233 | /* dead and being drained, ignore for migration */ |
234 | bool dead; | |
235 | ||
b4a04ab7 TH |
236 | /* For RCU-protected deletion */ |
237 | struct rcu_head rcu_head; | |
238 | }; | |
239 | ||
240 | struct cgroup { | |
241 | /* self css with NULL ->ss, points back to this cgroup */ | |
242 | struct cgroup_subsys_state self; | |
243 | ||
244 | unsigned long flags; /* "unsigned long" so bitops work */ | |
245 | ||
246 | /* | |
247 | * idr allocated in-hierarchy ID. | |
248 | * | |
249 | * ID 0 is not used, the ID of the root cgroup is always 1, and a | |
250 | * new cgroup will be assigned with a smallest available ID. | |
251 | * | |
252 | * Allocating/Removing ID must be protected by cgroup_mutex. | |
253 | */ | |
254 | int id; | |
255 | ||
b11cfb58 TH |
256 | /* |
257 | * The depth this cgroup is at. The root is at depth zero and each | |
258 | * step down the hierarchy increments the level. This along with | |
259 | * ancestor_ids[] can determine whether a given cgroup is a | |
260 | * descendant of another without traversing the hierarchy. | |
261 | */ | |
262 | int level; | |
263 | ||
b4a04ab7 | 264 | /* |
0de0942d | 265 | * Each non-empty css_set associated with this cgroup contributes |
788b950c TH |
266 | * one to nr_populated_csets. The counter is zero iff this cgroup |
267 | * doesn't have any tasks. | |
268 | * | |
269 | * All children which have non-zero nr_populated_csets and/or | |
270 | * nr_populated_children of their own contribute one to | |
271 | * nr_populated_children. The counter is zero iff this cgroup's | |
272 | * subtree proper doesn't have any tasks. | |
b4a04ab7 | 273 | */ |
788b950c TH |
274 | int nr_populated_csets; |
275 | int nr_populated_children; | |
b4a04ab7 TH |
276 | |
277 | struct kernfs_node *kn; /* cgroup kernfs entry */ | |
6f60eade TH |
278 | struct cgroup_file procs_file; /* handle for "cgroup.procs" */ |
279 | struct cgroup_file events_file; /* handle for "cgroup.events" */ | |
b4a04ab7 TH |
280 | |
281 | /* | |
282 | * The bitmask of subsystems enabled on the child cgroups. | |
283 | * ->subtree_control is the one configured through | |
8699b776 TH |
284 | * "cgroup.subtree_control" while ->child_ss_mask is the effective |
285 | * one which may have more subsystems enabled. Controller knobs | |
286 | * are made available iff it's enabled in ->subtree_control. | |
b4a04ab7 | 287 | */ |
6e5c8307 TH |
288 | u16 subtree_control; |
289 | u16 subtree_ss_mask; | |
15a27c36 TH |
290 | u16 old_subtree_control; |
291 | u16 old_subtree_ss_mask; | |
b4a04ab7 TH |
292 | |
293 | /* Private pointers for each registered subsystem */ | |
294 | struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT]; | |
295 | ||
296 | struct cgroup_root *root; | |
297 | ||
298 | /* | |
299 | * List of cgrp_cset_links pointing at css_sets with tasks in this | |
300 | * cgroup. Protected by css_set_lock. | |
301 | */ | |
302 | struct list_head cset_links; | |
303 | ||
304 | /* | |
305 | * On the default hierarchy, a css_set for a cgroup with some | |
306 | * susbsys disabled will point to css's which are associated with | |
307 | * the closest ancestor which has the subsys enabled. The | |
308 | * following lists all css_sets which point to this cgroup's css | |
309 | * for the given subsystem. | |
310 | */ | |
311 | struct list_head e_csets[CGROUP_SUBSYS_COUNT]; | |
312 | ||
313 | /* | |
314 | * list of pidlists, up to two for each namespace (one for procs, one | |
315 | * for tasks); created on demand. | |
316 | */ | |
317 | struct list_head pidlists; | |
318 | struct mutex pidlist_mutex; | |
319 | ||
320 | /* used to wait for offlining of csses */ | |
321 | wait_queue_head_t offline_waitq; | |
322 | ||
323 | /* used to schedule release agent */ | |
324 | struct work_struct release_agent_work; | |
b11cfb58 | 325 | |
30070984 DM |
326 | /* used to store eBPF programs */ |
327 | struct cgroup_bpf bpf; | |
328 | ||
b11cfb58 TH |
329 | /* ids of the ancestors at each level including self */ |
330 | int ancestor_ids[]; | |
b4a04ab7 TH |
331 | }; |
332 | ||
333 | /* | |
334 | * A cgroup_root represents the root of a cgroup hierarchy, and may be | |
335 | * associated with a kernfs_root to form an active hierarchy. This is | |
336 | * internal to cgroup core. Don't access directly from controllers. | |
337 | */ | |
338 | struct cgroup_root { | |
339 | struct kernfs_root *kf_root; | |
340 | ||
341 | /* The bitmask of subsystems attached to this hierarchy */ | |
342 | unsigned int subsys_mask; | |
343 | ||
344 | /* Unique id for this hierarchy. */ | |
345 | int hierarchy_id; | |
346 | ||
347 | /* The root cgroup. Root is destroyed on its release. */ | |
348 | struct cgroup cgrp; | |
349 | ||
b11cfb58 TH |
350 | /* for cgrp->ancestor_ids[0] */ |
351 | int cgrp_ancestor_id_storage; | |
352 | ||
b4a04ab7 TH |
353 | /* Number of cgroups in the hierarchy, used only for /proc/cgroups */ |
354 | atomic_t nr_cgrps; | |
355 | ||
356 | /* A list running through the active hierarchies */ | |
357 | struct list_head root_list; | |
358 | ||
359 | /* Hierarchy-specific flags */ | |
360 | unsigned int flags; | |
361 | ||
362 | /* IDs for cgroups in this hierarchy */ | |
363 | struct idr cgroup_idr; | |
364 | ||
365 | /* The path to use for release notifications. */ | |
366 | char release_agent_path[PATH_MAX]; | |
367 | ||
368 | /* The name for this hierarchy - may be empty */ | |
369 | char name[MAX_CGROUP_ROOT_NAMELEN]; | |
370 | }; | |
371 | ||
372 | /* | |
373 | * struct cftype: handler definitions for cgroup control files | |
374 | * | |
375 | * When reading/writing to a file: | |
376 | * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata | |
377 | * - the 'cftype' of the file is file->f_path.dentry->d_fsdata | |
378 | */ | |
379 | struct cftype { | |
380 | /* | |
381 | * By convention, the name should begin with the name of the | |
382 | * subsystem, followed by a period. Zero length string indicates | |
383 | * end of cftype array. | |
384 | */ | |
385 | char name[MAX_CFTYPE_NAME]; | |
731a981e | 386 | unsigned long private; |
b4a04ab7 TH |
387 | |
388 | /* | |
389 | * The maximum length of string, excluding trailing nul, that can | |
390 | * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed. | |
391 | */ | |
392 | size_t max_write_len; | |
393 | ||
394 | /* CFTYPE_* flags */ | |
395 | unsigned int flags; | |
396 | ||
6f60eade TH |
397 | /* |
398 | * If non-zero, should contain the offset from the start of css to | |
399 | * a struct cgroup_file field. cgroup will record the handle of | |
400 | * the created file into it. The recorded handle can be used as | |
401 | * long as the containing css remains accessible. | |
402 | */ | |
403 | unsigned int file_offset; | |
404 | ||
b4a04ab7 TH |
405 | /* |
406 | * Fields used for internal bookkeeping. Initialized automatically | |
407 | * during registration. | |
408 | */ | |
409 | struct cgroup_subsys *ss; /* NULL for cgroup core files */ | |
410 | struct list_head node; /* anchored at ss->cfts */ | |
411 | struct kernfs_ops *kf_ops; | |
412 | ||
e90cbebc TH |
413 | int (*open)(struct kernfs_open_file *of); |
414 | void (*release)(struct kernfs_open_file *of); | |
415 | ||
b4a04ab7 TH |
416 | /* |
417 | * read_u64() is a shortcut for the common case of returning a | |
418 | * single integer. Use it in place of read() | |
419 | */ | |
420 | u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft); | |
421 | /* | |
422 | * read_s64() is a signed version of read_u64() | |
423 | */ | |
424 | s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft); | |
425 | ||
426 | /* generic seq_file read interface */ | |
427 | int (*seq_show)(struct seq_file *sf, void *v); | |
428 | ||
429 | /* optional ops, implement all or none */ | |
430 | void *(*seq_start)(struct seq_file *sf, loff_t *ppos); | |
431 | void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos); | |
432 | void (*seq_stop)(struct seq_file *sf, void *v); | |
433 | ||
434 | /* | |
435 | * write_u64() is a shortcut for the common case of accepting | |
436 | * a single integer (as parsed by simple_strtoull) from | |
437 | * userspace. Use in place of write(); return 0 or error. | |
438 | */ | |
439 | int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft, | |
440 | u64 val); | |
441 | /* | |
442 | * write_s64() is a signed version of write_u64() | |
443 | */ | |
444 | int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft, | |
445 | s64 val); | |
446 | ||
447 | /* | |
448 | * write() is the generic write callback which maps directly to | |
449 | * kernfs write operation and overrides all other operations. | |
450 | * Maximum write size is determined by ->max_write_len. Use | |
451 | * of_css/cft() to access the associated css and cft. | |
452 | */ | |
453 | ssize_t (*write)(struct kernfs_open_file *of, | |
454 | char *buf, size_t nbytes, loff_t off); | |
455 | ||
456 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
457 | struct lock_class_key lockdep_key; | |
458 | #endif | |
459 | }; | |
460 | ||
461 | /* | |
462 | * Control Group subsystem type. | |
463 | * See Documentation/cgroups/cgroups.txt for details | |
464 | */ | |
465 | struct cgroup_subsys { | |
466 | struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css); | |
467 | int (*css_online)(struct cgroup_subsys_state *css); | |
468 | void (*css_offline)(struct cgroup_subsys_state *css); | |
469 | void (*css_released)(struct cgroup_subsys_state *css); | |
470 | void (*css_free)(struct cgroup_subsys_state *css); | |
471 | void (*css_reset)(struct cgroup_subsys_state *css); | |
b4a04ab7 | 472 | |
1f7dd3e5 TH |
473 | int (*can_attach)(struct cgroup_taskset *tset); |
474 | void (*cancel_attach)(struct cgroup_taskset *tset); | |
475 | void (*attach)(struct cgroup_taskset *tset); | |
5cf1cacb | 476 | void (*post_attach)(void); |
b53202e6 ON |
477 | int (*can_fork)(struct task_struct *task); |
478 | void (*cancel_fork)(struct task_struct *task); | |
479 | void (*fork)(struct task_struct *task); | |
2e91fa7f | 480 | void (*exit)(struct task_struct *task); |
afcf6c8b | 481 | void (*free)(struct task_struct *task); |
b4a04ab7 TH |
482 | void (*bind)(struct cgroup_subsys_state *root_css); |
483 | ||
b38e42e9 | 484 | bool early_init:1; |
b4a04ab7 | 485 | |
f6d635ad TH |
486 | /* |
487 | * If %true, the controller, on the default hierarchy, doesn't show | |
488 | * up in "cgroup.controllers" or "cgroup.subtree_control", is | |
489 | * implicitly enabled on all cgroups on the default hierarchy, and | |
490 | * bypasses the "no internal process" constraint. This is for | |
491 | * utility type controllers which is transparent to userland. | |
492 | * | |
493 | * An implicit controller can be stolen from the default hierarchy | |
494 | * anytime and thus must be okay with offline csses from previous | |
495 | * hierarchies coexisting with csses for the current one. | |
496 | */ | |
497 | bool implicit_on_dfl:1; | |
498 | ||
b4a04ab7 TH |
499 | /* |
500 | * If %false, this subsystem is properly hierarchical - | |
501 | * configuration, resource accounting and restriction on a parent | |
502 | * cgroup cover those of its children. If %true, hierarchy support | |
503 | * is broken in some ways - some subsystems ignore hierarchy | |
504 | * completely while others are only implemented half-way. | |
505 | * | |
506 | * It's now disallowed to create nested cgroups if the subsystem is | |
507 | * broken and cgroup core will emit a warning message on such | |
508 | * cases. Eventually, all subsystems will be made properly | |
509 | * hierarchical and this will go away. | |
510 | */ | |
b38e42e9 TH |
511 | bool broken_hierarchy:1; |
512 | bool warned_broken_hierarchy:1; | |
b4a04ab7 TH |
513 | |
514 | /* the following two fields are initialized automtically during boot */ | |
515 | int id; | |
516 | const char *name; | |
517 | ||
3e1d2eed TH |
518 | /* optional, initialized automatically during boot if not set */ |
519 | const char *legacy_name; | |
520 | ||
b4a04ab7 TH |
521 | /* link to parent, protected by cgroup_lock() */ |
522 | struct cgroup_root *root; | |
523 | ||
524 | /* idr for css->id */ | |
525 | struct idr css_idr; | |
526 | ||
527 | /* | |
528 | * List of cftypes. Each entry is the first entry of an array | |
529 | * terminated by zero length name. | |
530 | */ | |
531 | struct list_head cfts; | |
532 | ||
533 | /* | |
534 | * Base cftypes which are automatically registered. The two can | |
535 | * point to the same array. | |
536 | */ | |
537 | struct cftype *dfl_cftypes; /* for the default hierarchy */ | |
538 | struct cftype *legacy_cftypes; /* for the legacy hierarchies */ | |
539 | ||
540 | /* | |
541 | * A subsystem may depend on other subsystems. When such subsystem | |
542 | * is enabled on a cgroup, the depended-upon subsystems are enabled | |
543 | * together if available. Subsystems enabled due to dependency are | |
544 | * not visible to userland until explicitly enabled. The following | |
545 | * specifies the mask of subsystems that this one depends on. | |
546 | */ | |
547 | unsigned int depends_on; | |
548 | }; | |
549 | ||
1ed13287 TH |
550 | extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
551 | ||
552 | /** | |
553 | * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups | |
554 | * @tsk: target task | |
555 | * | |
780de9dd IM |
556 | * Allows cgroup operations to synchronize against threadgroup changes |
557 | * using a percpu_rw_semaphore. | |
1ed13287 TH |
558 | */ |
559 | static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) | |
560 | { | |
561 | percpu_down_read(&cgroup_threadgroup_rwsem); | |
562 | } | |
563 | ||
564 | /** | |
565 | * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups | |
566 | * @tsk: target task | |
567 | * | |
780de9dd | 568 | * Counterpart of cgroup_threadcgroup_change_begin(). |
1ed13287 TH |
569 | */ |
570 | static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) | |
571 | { | |
572 | percpu_up_read(&cgroup_threadgroup_rwsem); | |
573 | } | |
7d7efec3 TH |
574 | |
575 | #else /* CONFIG_CGROUPS */ | |
576 | ||
cb4a3167 AS |
577 | #define CGROUP_SUBSYS_COUNT 0 |
578 | ||
780de9dd IM |
579 | static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) |
580 | { | |
581 | might_sleep(); | |
582 | } | |
583 | ||
7d7efec3 TH |
584 | static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {} |
585 | ||
b4a04ab7 | 586 | #endif /* CONFIG_CGROUPS */ |
7d7efec3 | 587 | |
2a56a1fe TH |
588 | #ifdef CONFIG_SOCK_CGROUP_DATA |
589 | ||
bd1060a1 TH |
590 | /* |
591 | * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains | |
592 | * per-socket cgroup information except for memcg association. | |
593 | * | |
594 | * On legacy hierarchies, net_prio and net_cls controllers directly set | |
595 | * attributes on each sock which can then be tested by the network layer. | |
596 | * On the default hierarchy, each sock is associated with the cgroup it was | |
597 | * created in and the networking layer can match the cgroup directly. | |
598 | * | |
599 | * To avoid carrying all three cgroup related fields separately in sock, | |
600 | * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer. | |
601 | * On boot, sock_cgroup_data records the cgroup that the sock was created | |
602 | * in so that cgroup2 matches can be made; however, once either net_prio or | |
603 | * net_cls starts being used, the area is overriden to carry prioidx and/or | |
604 | * classid. The two modes are distinguished by whether the lowest bit is | |
605 | * set. Clear bit indicates cgroup pointer while set bit prioidx and | |
606 | * classid. | |
607 | * | |
608 | * While userland may start using net_prio or net_cls at any time, once | |
609 | * either is used, cgroup2 matching no longer works. There is no reason to | |
610 | * mix the two and this is in line with how legacy and v2 compatibility is | |
611 | * handled. On mode switch, cgroup references which are already being | |
612 | * pointed to by socks may be leaked. While this can be remedied by adding | |
613 | * synchronization around sock_cgroup_data, given that the number of leaked | |
614 | * cgroups is bound and highly unlikely to be high, this seems to be the | |
615 | * better trade-off. | |
616 | */ | |
2a56a1fe | 617 | struct sock_cgroup_data { |
bd1060a1 TH |
618 | union { |
619 | #ifdef __LITTLE_ENDIAN | |
620 | struct { | |
621 | u8 is_data; | |
622 | u8 padding; | |
623 | u16 prioidx; | |
624 | u32 classid; | |
625 | } __packed; | |
626 | #else | |
627 | struct { | |
628 | u32 classid; | |
629 | u16 prioidx; | |
630 | u8 padding; | |
631 | u8 is_data; | |
632 | } __packed; | |
633 | #endif | |
634 | u64 val; | |
635 | }; | |
2a56a1fe TH |
636 | }; |
637 | ||
bd1060a1 TH |
638 | /* |
639 | * There's a theoretical window where the following accessors race with | |
640 | * updaters and return part of the previous pointer as the prioidx or | |
641 | * classid. Such races are short-lived and the result isn't critical. | |
642 | */ | |
2a56a1fe TH |
643 | static inline u16 sock_cgroup_prioidx(struct sock_cgroup_data *skcd) |
644 | { | |
bd1060a1 TH |
645 | /* fallback to 1 which is always the ID of the root cgroup */ |
646 | return (skcd->is_data & 1) ? skcd->prioidx : 1; | |
2a56a1fe TH |
647 | } |
648 | ||
649 | static inline u32 sock_cgroup_classid(struct sock_cgroup_data *skcd) | |
650 | { | |
bd1060a1 TH |
651 | /* fallback to 0 which is the unconfigured default classid */ |
652 | return (skcd->is_data & 1) ? skcd->classid : 0; | |
2a56a1fe TH |
653 | } |
654 | ||
bd1060a1 TH |
655 | /* |
656 | * If invoked concurrently, the updaters may clobber each other. The | |
657 | * caller is responsible for synchronization. | |
658 | */ | |
2a56a1fe TH |
659 | static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd, |
660 | u16 prioidx) | |
661 | { | |
ad2c8c73 | 662 | struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }}; |
bd1060a1 TH |
663 | |
664 | if (sock_cgroup_prioidx(&skcd_buf) == prioidx) | |
665 | return; | |
666 | ||
667 | if (!(skcd_buf.is_data & 1)) { | |
668 | skcd_buf.val = 0; | |
669 | skcd_buf.is_data = 1; | |
670 | } | |
671 | ||
672 | skcd_buf.prioidx = prioidx; | |
673 | WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */ | |
2a56a1fe TH |
674 | } |
675 | ||
676 | static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd, | |
677 | u32 classid) | |
678 | { | |
ad2c8c73 | 679 | struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }}; |
bd1060a1 TH |
680 | |
681 | if (sock_cgroup_classid(&skcd_buf) == classid) | |
682 | return; | |
683 | ||
684 | if (!(skcd_buf.is_data & 1)) { | |
685 | skcd_buf.val = 0; | |
686 | skcd_buf.is_data = 1; | |
687 | } | |
688 | ||
689 | skcd_buf.classid = classid; | |
690 | WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */ | |
2a56a1fe TH |
691 | } |
692 | ||
693 | #else /* CONFIG_SOCK_CGROUP_DATA */ | |
694 | ||
695 | struct sock_cgroup_data { | |
696 | }; | |
697 | ||
698 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
699 | ||
b4a04ab7 | 700 | #endif /* _LINUX_CGROUP_DEFS_H */ |