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> | |
16 | #include <linux/percpu-refcount.h> | |
7d7efec3 | 17 | #include <linux/percpu-rwsem.h> |
b4a04ab7 TH |
18 | #include <linux/workqueue.h> |
19 | ||
20 | #ifdef CONFIG_CGROUPS | |
21 | ||
22 | struct cgroup; | |
23 | struct cgroup_root; | |
24 | struct cgroup_subsys; | |
25 | struct cgroup_taskset; | |
26 | struct kernfs_node; | |
27 | struct kernfs_ops; | |
28 | struct kernfs_open_file; | |
c80ef9e0 | 29 | struct seq_file; |
b4a04ab7 TH |
30 | |
31 | #define MAX_CGROUP_TYPE_NAMELEN 32 | |
32 | #define MAX_CGROUP_ROOT_NAMELEN 64 | |
33 | #define MAX_CFTYPE_NAME 64 | |
34 | ||
35 | /* define the enumeration of all cgroup subsystems */ | |
36 | #define SUBSYS(_x) _x ## _cgrp_id, | |
37 | enum cgroup_subsys_id { | |
38 | #include <linux/cgroup_subsys.h> | |
39 | CGROUP_SUBSYS_COUNT, | |
40 | }; | |
41 | #undef SUBSYS | |
42 | ||
43 | /* bits in struct cgroup_subsys_state flags field */ | |
44 | enum { | |
45 | CSS_NO_REF = (1 << 0), /* no reference counting for this css */ | |
46 | CSS_ONLINE = (1 << 1), /* between ->css_online() and ->css_offline() */ | |
47 | CSS_RELEASED = (1 << 2), /* refcnt reached zero, released */ | |
48 | }; | |
49 | ||
50 | /* bits in struct cgroup flags field */ | |
51 | enum { | |
52 | /* Control Group requires release notifications to userspace */ | |
53 | CGRP_NOTIFY_ON_RELEASE, | |
54 | /* | |
55 | * Clone the parent's configuration when creating a new child | |
56 | * cpuset cgroup. For historical reasons, this option can be | |
57 | * specified at mount time and thus is implemented here. | |
58 | */ | |
59 | CGRP_CPUSET_CLONE_CHILDREN, | |
60 | }; | |
61 | ||
62 | /* cgroup_root->flags */ | |
63 | enum { | |
b4a04ab7 TH |
64 | CGRP_ROOT_NOPREFIX = (1 << 1), /* mounted subsystems have no named prefix */ |
65 | CGRP_ROOT_XATTR = (1 << 2), /* supports extended attributes */ | |
66 | }; | |
67 | ||
68 | /* cftype->flags */ | |
69 | enum { | |
70 | CFTYPE_ONLY_ON_ROOT = (1 << 0), /* only create on root cgrp */ | |
71 | CFTYPE_NOT_ON_ROOT = (1 << 1), /* don't create on root cgrp */ | |
72 | CFTYPE_NO_PREFIX = (1 << 3), /* (DON'T USE FOR NEW FILES) no subsys prefix */ | |
7dbdb199 | 73 | CFTYPE_WORLD_WRITABLE = (1 << 4), /* (DON'T USE FOR NEW FILES) S_IWUGO */ |
b4a04ab7 TH |
74 | |
75 | /* internal flags, do not use outside cgroup core proper */ | |
76 | __CFTYPE_ONLY_ON_DFL = (1 << 16), /* only on default hierarchy */ | |
77 | __CFTYPE_NOT_ON_DFL = (1 << 17), /* not on default hierarchy */ | |
78 | }; | |
79 | ||
6f60eade TH |
80 | /* |
81 | * cgroup_file is the handle for a file instance created in a cgroup which | |
82 | * is used, for example, to generate file changed notifications. This can | |
83 | * be obtained by setting cftype->file_offset. | |
84 | */ | |
85 | struct cgroup_file { | |
86 | /* do not access any fields from outside cgroup core */ | |
6f60eade TH |
87 | struct kernfs_node *kn; |
88 | }; | |
89 | ||
b4a04ab7 TH |
90 | /* |
91 | * Per-subsystem/per-cgroup state maintained by the system. This is the | |
92 | * fundamental structural building block that controllers deal with. | |
93 | * | |
94 | * Fields marked with "PI:" are public and immutable and may be accessed | |
95 | * directly without synchronization. | |
96 | */ | |
97 | struct cgroup_subsys_state { | |
98 | /* PI: the cgroup that this css is attached to */ | |
99 | struct cgroup *cgroup; | |
100 | ||
101 | /* PI: the cgroup subsystem that this css is attached to */ | |
102 | struct cgroup_subsys *ss; | |
103 | ||
104 | /* reference count - access via css_[try]get() and css_put() */ | |
105 | struct percpu_ref refcnt; | |
106 | ||
107 | /* PI: the parent css */ | |
108 | struct cgroup_subsys_state *parent; | |
109 | ||
110 | /* siblings list anchored at the parent's ->children */ | |
111 | struct list_head sibling; | |
112 | struct list_head children; | |
113 | ||
114 | /* | |
115 | * PI: Subsys-unique ID. 0 is unused and root is always 1. The | |
116 | * matching css can be looked up using css_from_id(). | |
117 | */ | |
118 | int id; | |
119 | ||
120 | unsigned int flags; | |
121 | ||
122 | /* | |
123 | * Monotonically increasing unique serial number which defines a | |
124 | * uniform order among all csses. It's guaranteed that all | |
125 | * ->children lists are in the ascending order of ->serial_nr and | |
126 | * used to allow interrupting and resuming iterations. | |
127 | */ | |
128 | u64 serial_nr; | |
129 | ||
aa226ff4 TH |
130 | /* |
131 | * Incremented by online self and children. Used to guarantee that | |
132 | * parents are not offlined before their children. | |
133 | */ | |
134 | atomic_t online_cnt; | |
135 | ||
b4a04ab7 TH |
136 | /* percpu_ref killing and RCU release */ |
137 | struct rcu_head rcu_head; | |
138 | struct work_struct destroy_work; | |
139 | }; | |
140 | ||
141 | /* | |
142 | * A css_set is a structure holding pointers to a set of | |
143 | * cgroup_subsys_state objects. This saves space in the task struct | |
144 | * object and speeds up fork()/exit(), since a single inc/dec and a | |
145 | * list_add()/del() can bump the reference count on the entire cgroup | |
146 | * set for a task. | |
147 | */ | |
148 | struct css_set { | |
149 | /* Reference count */ | |
150 | atomic_t refcount; | |
151 | ||
152 | /* | |
153 | * List running through all cgroup groups in the same hash | |
154 | * slot. Protected by css_set_lock | |
155 | */ | |
156 | struct hlist_node hlist; | |
157 | ||
158 | /* | |
159 | * Lists running through all tasks using this cgroup group. | |
160 | * mg_tasks lists tasks which belong to this cset but are in the | |
161 | * process of being migrated out or in. Protected by | |
162 | * css_set_rwsem, but, during migration, once tasks are moved to | |
163 | * mg_tasks, it can be read safely while holding cgroup_mutex. | |
164 | */ | |
165 | struct list_head tasks; | |
166 | struct list_head mg_tasks; | |
167 | ||
168 | /* | |
169 | * List of cgrp_cset_links pointing at cgroups referenced from this | |
170 | * css_set. Protected by css_set_lock. | |
171 | */ | |
172 | struct list_head cgrp_links; | |
173 | ||
174 | /* the default cgroup associated with this css_set */ | |
175 | struct cgroup *dfl_cgrp; | |
176 | ||
177 | /* | |
178 | * Set of subsystem states, one for each subsystem. This array is | |
179 | * immutable after creation apart from the init_css_set during | |
180 | * subsystem registration (at boot time). | |
181 | */ | |
182 | struct cgroup_subsys_state *subsys[CGROUP_SUBSYS_COUNT]; | |
183 | ||
184 | /* | |
185 | * List of csets participating in the on-going migration either as | |
186 | * source or destination. Protected by cgroup_mutex. | |
187 | */ | |
188 | struct list_head mg_preload_node; | |
189 | struct list_head mg_node; | |
190 | ||
191 | /* | |
192 | * If this cset is acting as the source of migration the following | |
193 | * two fields are set. mg_src_cgrp is the source cgroup of the | |
194 | * on-going migration and mg_dst_cset is the destination cset the | |
195 | * target tasks on this cset should be migrated to. Protected by | |
196 | * cgroup_mutex. | |
197 | */ | |
198 | struct cgroup *mg_src_cgrp; | |
199 | struct css_set *mg_dst_cset; | |
200 | ||
201 | /* | |
202 | * On the default hierarhcy, ->subsys[ssid] may point to a css | |
203 | * attached to an ancestor instead of the cgroup this css_set is | |
204 | * associated with. The following node is anchored at | |
205 | * ->subsys[ssid]->cgroup->e_csets[ssid] and provides a way to | |
206 | * iterate through all css's attached to a given cgroup. | |
207 | */ | |
208 | struct list_head e_cset_node[CGROUP_SUBSYS_COUNT]; | |
209 | ||
ed27b9f7 TH |
210 | /* all css_task_iters currently walking this cset */ |
211 | struct list_head task_iters; | |
212 | ||
b4a04ab7 TH |
213 | /* For RCU-protected deletion */ |
214 | struct rcu_head rcu_head; | |
215 | }; | |
216 | ||
217 | struct cgroup { | |
218 | /* self css with NULL ->ss, points back to this cgroup */ | |
219 | struct cgroup_subsys_state self; | |
220 | ||
221 | unsigned long flags; /* "unsigned long" so bitops work */ | |
222 | ||
223 | /* | |
224 | * idr allocated in-hierarchy ID. | |
225 | * | |
226 | * ID 0 is not used, the ID of the root cgroup is always 1, and a | |
227 | * new cgroup will be assigned with a smallest available ID. | |
228 | * | |
229 | * Allocating/Removing ID must be protected by cgroup_mutex. | |
230 | */ | |
231 | int id; | |
232 | ||
b11cfb58 TH |
233 | /* |
234 | * The depth this cgroup is at. The root is at depth zero and each | |
235 | * step down the hierarchy increments the level. This along with | |
236 | * ancestor_ids[] can determine whether a given cgroup is a | |
237 | * descendant of another without traversing the hierarchy. | |
238 | */ | |
239 | int level; | |
240 | ||
b4a04ab7 | 241 | /* |
0de0942d TH |
242 | * Each non-empty css_set associated with this cgroup contributes |
243 | * one to populated_cnt. All children with non-zero popuplated_cnt | |
244 | * of their own contribute one. The count is zero iff there's no | |
245 | * task in this cgroup or its subtree. | |
b4a04ab7 TH |
246 | */ |
247 | int populated_cnt; | |
248 | ||
249 | struct kernfs_node *kn; /* cgroup kernfs entry */ | |
6f60eade TH |
250 | struct cgroup_file procs_file; /* handle for "cgroup.procs" */ |
251 | struct cgroup_file events_file; /* handle for "cgroup.events" */ | |
b4a04ab7 TH |
252 | |
253 | /* | |
254 | * The bitmask of subsystems enabled on the child cgroups. | |
255 | * ->subtree_control is the one configured through | |
256 | * "cgroup.subtree_control" while ->child_subsys_mask is the | |
257 | * effective one which may have more subsystems enabled. | |
258 | * Controller knobs are made available iff it's enabled in | |
259 | * ->subtree_control. | |
260 | */ | |
261 | unsigned int subtree_control; | |
262 | unsigned int child_subsys_mask; | |
263 | ||
264 | /* Private pointers for each registered subsystem */ | |
265 | struct cgroup_subsys_state __rcu *subsys[CGROUP_SUBSYS_COUNT]; | |
266 | ||
267 | struct cgroup_root *root; | |
268 | ||
269 | /* | |
270 | * List of cgrp_cset_links pointing at css_sets with tasks in this | |
271 | * cgroup. Protected by css_set_lock. | |
272 | */ | |
273 | struct list_head cset_links; | |
274 | ||
275 | /* | |
276 | * On the default hierarchy, a css_set for a cgroup with some | |
277 | * susbsys disabled will point to css's which are associated with | |
278 | * the closest ancestor which has the subsys enabled. The | |
279 | * following lists all css_sets which point to this cgroup's css | |
280 | * for the given subsystem. | |
281 | */ | |
282 | struct list_head e_csets[CGROUP_SUBSYS_COUNT]; | |
283 | ||
284 | /* | |
285 | * list of pidlists, up to two for each namespace (one for procs, one | |
286 | * for tasks); created on demand. | |
287 | */ | |
288 | struct list_head pidlists; | |
289 | struct mutex pidlist_mutex; | |
290 | ||
291 | /* used to wait for offlining of csses */ | |
292 | wait_queue_head_t offline_waitq; | |
293 | ||
294 | /* used to schedule release agent */ | |
295 | struct work_struct release_agent_work; | |
b11cfb58 TH |
296 | |
297 | /* ids of the ancestors at each level including self */ | |
298 | int ancestor_ids[]; | |
b4a04ab7 TH |
299 | }; |
300 | ||
301 | /* | |
302 | * A cgroup_root represents the root of a cgroup hierarchy, and may be | |
303 | * associated with a kernfs_root to form an active hierarchy. This is | |
304 | * internal to cgroup core. Don't access directly from controllers. | |
305 | */ | |
306 | struct cgroup_root { | |
307 | struct kernfs_root *kf_root; | |
308 | ||
309 | /* The bitmask of subsystems attached to this hierarchy */ | |
310 | unsigned int subsys_mask; | |
311 | ||
312 | /* Unique id for this hierarchy. */ | |
313 | int hierarchy_id; | |
314 | ||
315 | /* The root cgroup. Root is destroyed on its release. */ | |
316 | struct cgroup cgrp; | |
317 | ||
b11cfb58 TH |
318 | /* for cgrp->ancestor_ids[0] */ |
319 | int cgrp_ancestor_id_storage; | |
320 | ||
b4a04ab7 TH |
321 | /* Number of cgroups in the hierarchy, used only for /proc/cgroups */ |
322 | atomic_t nr_cgrps; | |
323 | ||
324 | /* A list running through the active hierarchies */ | |
325 | struct list_head root_list; | |
326 | ||
327 | /* Hierarchy-specific flags */ | |
328 | unsigned int flags; | |
329 | ||
330 | /* IDs for cgroups in this hierarchy */ | |
331 | struct idr cgroup_idr; | |
332 | ||
333 | /* The path to use for release notifications. */ | |
334 | char release_agent_path[PATH_MAX]; | |
335 | ||
336 | /* The name for this hierarchy - may be empty */ | |
337 | char name[MAX_CGROUP_ROOT_NAMELEN]; | |
338 | }; | |
339 | ||
340 | /* | |
341 | * struct cftype: handler definitions for cgroup control files | |
342 | * | |
343 | * When reading/writing to a file: | |
344 | * - the cgroup to use is file->f_path.dentry->d_parent->d_fsdata | |
345 | * - the 'cftype' of the file is file->f_path.dentry->d_fsdata | |
346 | */ | |
347 | struct cftype { | |
348 | /* | |
349 | * By convention, the name should begin with the name of the | |
350 | * subsystem, followed by a period. Zero length string indicates | |
351 | * end of cftype array. | |
352 | */ | |
353 | char name[MAX_CFTYPE_NAME]; | |
731a981e | 354 | unsigned long private; |
b4a04ab7 TH |
355 | |
356 | /* | |
357 | * The maximum length of string, excluding trailing nul, that can | |
358 | * be passed to write. If < PAGE_SIZE-1, PAGE_SIZE-1 is assumed. | |
359 | */ | |
360 | size_t max_write_len; | |
361 | ||
362 | /* CFTYPE_* flags */ | |
363 | unsigned int flags; | |
364 | ||
6f60eade TH |
365 | /* |
366 | * If non-zero, should contain the offset from the start of css to | |
367 | * a struct cgroup_file field. cgroup will record the handle of | |
368 | * the created file into it. The recorded handle can be used as | |
369 | * long as the containing css remains accessible. | |
370 | */ | |
371 | unsigned int file_offset; | |
372 | ||
b4a04ab7 TH |
373 | /* |
374 | * Fields used for internal bookkeeping. Initialized automatically | |
375 | * during registration. | |
376 | */ | |
377 | struct cgroup_subsys *ss; /* NULL for cgroup core files */ | |
378 | struct list_head node; /* anchored at ss->cfts */ | |
379 | struct kernfs_ops *kf_ops; | |
380 | ||
381 | /* | |
382 | * read_u64() is a shortcut for the common case of returning a | |
383 | * single integer. Use it in place of read() | |
384 | */ | |
385 | u64 (*read_u64)(struct cgroup_subsys_state *css, struct cftype *cft); | |
386 | /* | |
387 | * read_s64() is a signed version of read_u64() | |
388 | */ | |
389 | s64 (*read_s64)(struct cgroup_subsys_state *css, struct cftype *cft); | |
390 | ||
391 | /* generic seq_file read interface */ | |
392 | int (*seq_show)(struct seq_file *sf, void *v); | |
393 | ||
394 | /* optional ops, implement all or none */ | |
395 | void *(*seq_start)(struct seq_file *sf, loff_t *ppos); | |
396 | void *(*seq_next)(struct seq_file *sf, void *v, loff_t *ppos); | |
397 | void (*seq_stop)(struct seq_file *sf, void *v); | |
398 | ||
399 | /* | |
400 | * write_u64() is a shortcut for the common case of accepting | |
401 | * a single integer (as parsed by simple_strtoull) from | |
402 | * userspace. Use in place of write(); return 0 or error. | |
403 | */ | |
404 | int (*write_u64)(struct cgroup_subsys_state *css, struct cftype *cft, | |
405 | u64 val); | |
406 | /* | |
407 | * write_s64() is a signed version of write_u64() | |
408 | */ | |
409 | int (*write_s64)(struct cgroup_subsys_state *css, struct cftype *cft, | |
410 | s64 val); | |
411 | ||
412 | /* | |
413 | * write() is the generic write callback which maps directly to | |
414 | * kernfs write operation and overrides all other operations. | |
415 | * Maximum write size is determined by ->max_write_len. Use | |
416 | * of_css/cft() to access the associated css and cft. | |
417 | */ | |
418 | ssize_t (*write)(struct kernfs_open_file *of, | |
419 | char *buf, size_t nbytes, loff_t off); | |
420 | ||
421 | #ifdef CONFIG_DEBUG_LOCK_ALLOC | |
422 | struct lock_class_key lockdep_key; | |
423 | #endif | |
424 | }; | |
425 | ||
426 | /* | |
427 | * Control Group subsystem type. | |
428 | * See Documentation/cgroups/cgroups.txt for details | |
429 | */ | |
430 | struct cgroup_subsys { | |
431 | struct cgroup_subsys_state *(*css_alloc)(struct cgroup_subsys_state *parent_css); | |
432 | int (*css_online)(struct cgroup_subsys_state *css); | |
433 | void (*css_offline)(struct cgroup_subsys_state *css); | |
434 | void (*css_released)(struct cgroup_subsys_state *css); | |
435 | void (*css_free)(struct cgroup_subsys_state *css); | |
436 | void (*css_reset)(struct cgroup_subsys_state *css); | |
437 | void (*css_e_css_changed)(struct cgroup_subsys_state *css); | |
438 | ||
1f7dd3e5 TH |
439 | int (*can_attach)(struct cgroup_taskset *tset); |
440 | void (*cancel_attach)(struct cgroup_taskset *tset); | |
441 | void (*attach)(struct cgroup_taskset *tset); | |
b53202e6 ON |
442 | int (*can_fork)(struct task_struct *task); |
443 | void (*cancel_fork)(struct task_struct *task); | |
444 | void (*fork)(struct task_struct *task); | |
2e91fa7f | 445 | void (*exit)(struct task_struct *task); |
afcf6c8b | 446 | void (*free)(struct task_struct *task); |
b4a04ab7 TH |
447 | void (*bind)(struct cgroup_subsys_state *root_css); |
448 | ||
b4a04ab7 TH |
449 | int early_init; |
450 | ||
451 | /* | |
452 | * If %false, this subsystem is properly hierarchical - | |
453 | * configuration, resource accounting and restriction on a parent | |
454 | * cgroup cover those of its children. If %true, hierarchy support | |
455 | * is broken in some ways - some subsystems ignore hierarchy | |
456 | * completely while others are only implemented half-way. | |
457 | * | |
458 | * It's now disallowed to create nested cgroups if the subsystem is | |
459 | * broken and cgroup core will emit a warning message on such | |
460 | * cases. Eventually, all subsystems will be made properly | |
461 | * hierarchical and this will go away. | |
462 | */ | |
463 | bool broken_hierarchy; | |
464 | bool warned_broken_hierarchy; | |
465 | ||
466 | /* the following two fields are initialized automtically during boot */ | |
467 | int id; | |
468 | const char *name; | |
469 | ||
3e1d2eed TH |
470 | /* optional, initialized automatically during boot if not set */ |
471 | const char *legacy_name; | |
472 | ||
b4a04ab7 TH |
473 | /* link to parent, protected by cgroup_lock() */ |
474 | struct cgroup_root *root; | |
475 | ||
476 | /* idr for css->id */ | |
477 | struct idr css_idr; | |
478 | ||
479 | /* | |
480 | * List of cftypes. Each entry is the first entry of an array | |
481 | * terminated by zero length name. | |
482 | */ | |
483 | struct list_head cfts; | |
484 | ||
485 | /* | |
486 | * Base cftypes which are automatically registered. The two can | |
487 | * point to the same array. | |
488 | */ | |
489 | struct cftype *dfl_cftypes; /* for the default hierarchy */ | |
490 | struct cftype *legacy_cftypes; /* for the legacy hierarchies */ | |
491 | ||
492 | /* | |
493 | * A subsystem may depend on other subsystems. When such subsystem | |
494 | * is enabled on a cgroup, the depended-upon subsystems are enabled | |
495 | * together if available. Subsystems enabled due to dependency are | |
496 | * not visible to userland until explicitly enabled. The following | |
497 | * specifies the mask of subsystems that this one depends on. | |
498 | */ | |
499 | unsigned int depends_on; | |
500 | }; | |
501 | ||
1ed13287 TH |
502 | extern struct percpu_rw_semaphore cgroup_threadgroup_rwsem; |
503 | ||
504 | /** | |
505 | * cgroup_threadgroup_change_begin - threadgroup exclusion for cgroups | |
506 | * @tsk: target task | |
507 | * | |
508 | * Called from threadgroup_change_begin() and allows cgroup operations to | |
509 | * synchronize against threadgroup changes using a percpu_rw_semaphore. | |
510 | */ | |
511 | static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) | |
512 | { | |
513 | percpu_down_read(&cgroup_threadgroup_rwsem); | |
514 | } | |
515 | ||
516 | /** | |
517 | * cgroup_threadgroup_change_end - threadgroup exclusion for cgroups | |
518 | * @tsk: target task | |
519 | * | |
520 | * Called from threadgroup_change_end(). Counterpart of | |
521 | * cgroup_threadcgroup_change_begin(). | |
522 | */ | |
523 | static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) | |
524 | { | |
525 | percpu_up_read(&cgroup_threadgroup_rwsem); | |
526 | } | |
7d7efec3 TH |
527 | |
528 | #else /* CONFIG_CGROUPS */ | |
529 | ||
cb4a3167 AS |
530 | #define CGROUP_SUBSYS_COUNT 0 |
531 | ||
7d7efec3 TH |
532 | static inline void cgroup_threadgroup_change_begin(struct task_struct *tsk) {} |
533 | static inline void cgroup_threadgroup_change_end(struct task_struct *tsk) {} | |
534 | ||
b4a04ab7 | 535 | #endif /* CONFIG_CGROUPS */ |
7d7efec3 | 536 | |
2a56a1fe TH |
537 | #ifdef CONFIG_SOCK_CGROUP_DATA |
538 | ||
bd1060a1 TH |
539 | /* |
540 | * sock_cgroup_data is embedded at sock->sk_cgrp_data and contains | |
541 | * per-socket cgroup information except for memcg association. | |
542 | * | |
543 | * On legacy hierarchies, net_prio and net_cls controllers directly set | |
544 | * attributes on each sock which can then be tested by the network layer. | |
545 | * On the default hierarchy, each sock is associated with the cgroup it was | |
546 | * created in and the networking layer can match the cgroup directly. | |
547 | * | |
548 | * To avoid carrying all three cgroup related fields separately in sock, | |
549 | * sock_cgroup_data overloads (prioidx, classid) and the cgroup pointer. | |
550 | * On boot, sock_cgroup_data records the cgroup that the sock was created | |
551 | * in so that cgroup2 matches can be made; however, once either net_prio or | |
552 | * net_cls starts being used, the area is overriden to carry prioidx and/or | |
553 | * classid. The two modes are distinguished by whether the lowest bit is | |
554 | * set. Clear bit indicates cgroup pointer while set bit prioidx and | |
555 | * classid. | |
556 | * | |
557 | * While userland may start using net_prio or net_cls at any time, once | |
558 | * either is used, cgroup2 matching no longer works. There is no reason to | |
559 | * mix the two and this is in line with how legacy and v2 compatibility is | |
560 | * handled. On mode switch, cgroup references which are already being | |
561 | * pointed to by socks may be leaked. While this can be remedied by adding | |
562 | * synchronization around sock_cgroup_data, given that the number of leaked | |
563 | * cgroups is bound and highly unlikely to be high, this seems to be the | |
564 | * better trade-off. | |
565 | */ | |
2a56a1fe | 566 | struct sock_cgroup_data { |
bd1060a1 TH |
567 | union { |
568 | #ifdef __LITTLE_ENDIAN | |
569 | struct { | |
570 | u8 is_data; | |
571 | u8 padding; | |
572 | u16 prioidx; | |
573 | u32 classid; | |
574 | } __packed; | |
575 | #else | |
576 | struct { | |
577 | u32 classid; | |
578 | u16 prioidx; | |
579 | u8 padding; | |
580 | u8 is_data; | |
581 | } __packed; | |
582 | #endif | |
583 | u64 val; | |
584 | }; | |
2a56a1fe TH |
585 | }; |
586 | ||
bd1060a1 TH |
587 | /* |
588 | * There's a theoretical window where the following accessors race with | |
589 | * updaters and return part of the previous pointer as the prioidx or | |
590 | * classid. Such races are short-lived and the result isn't critical. | |
591 | */ | |
2a56a1fe TH |
592 | static inline u16 sock_cgroup_prioidx(struct sock_cgroup_data *skcd) |
593 | { | |
bd1060a1 TH |
594 | /* fallback to 1 which is always the ID of the root cgroup */ |
595 | return (skcd->is_data & 1) ? skcd->prioidx : 1; | |
2a56a1fe TH |
596 | } |
597 | ||
598 | static inline u32 sock_cgroup_classid(struct sock_cgroup_data *skcd) | |
599 | { | |
bd1060a1 TH |
600 | /* fallback to 0 which is the unconfigured default classid */ |
601 | return (skcd->is_data & 1) ? skcd->classid : 0; | |
2a56a1fe TH |
602 | } |
603 | ||
bd1060a1 TH |
604 | /* |
605 | * If invoked concurrently, the updaters may clobber each other. The | |
606 | * caller is responsible for synchronization. | |
607 | */ | |
2a56a1fe TH |
608 | static inline void sock_cgroup_set_prioidx(struct sock_cgroup_data *skcd, |
609 | u16 prioidx) | |
610 | { | |
ad2c8c73 | 611 | struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }}; |
bd1060a1 TH |
612 | |
613 | if (sock_cgroup_prioidx(&skcd_buf) == prioidx) | |
614 | return; | |
615 | ||
616 | if (!(skcd_buf.is_data & 1)) { | |
617 | skcd_buf.val = 0; | |
618 | skcd_buf.is_data = 1; | |
619 | } | |
620 | ||
621 | skcd_buf.prioidx = prioidx; | |
622 | WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */ | |
2a56a1fe TH |
623 | } |
624 | ||
625 | static inline void sock_cgroup_set_classid(struct sock_cgroup_data *skcd, | |
626 | u32 classid) | |
627 | { | |
ad2c8c73 | 628 | struct sock_cgroup_data skcd_buf = {{ .val = READ_ONCE(skcd->val) }}; |
bd1060a1 TH |
629 | |
630 | if (sock_cgroup_classid(&skcd_buf) == classid) | |
631 | return; | |
632 | ||
633 | if (!(skcd_buf.is_data & 1)) { | |
634 | skcd_buf.val = 0; | |
635 | skcd_buf.is_data = 1; | |
636 | } | |
637 | ||
638 | skcd_buf.classid = classid; | |
639 | WRITE_ONCE(skcd->val, skcd_buf.val); /* see sock_cgroup_ptr() */ | |
2a56a1fe TH |
640 | } |
641 | ||
642 | #else /* CONFIG_SOCK_CGROUP_DATA */ | |
643 | ||
644 | struct sock_cgroup_data { | |
645 | }; | |
646 | ||
647 | #endif /* CONFIG_SOCK_CGROUP_DATA */ | |
648 | ||
b4a04ab7 | 649 | #endif /* _LINUX_CGROUP_DEFS_H */ |