• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /* SPDX-License-Identifier: GPL-2.0 */
2 #ifndef _BPF_CGROUP_H
3 #define _BPF_CGROUP_H
4 
5 #include <linux/bpf.h>
6 #include <linux/errno.h>
7 #include <linux/jump_label.h>
8 #include <linux/percpu.h>
9 #include <linux/percpu-refcount.h>
10 #include <linux/rbtree.h>
11 #include <uapi/linux/bpf.h>
12 
13 struct sock;
14 struct sockaddr;
15 struct cgroup;
16 struct sk_buff;
17 struct bpf_map;
18 struct bpf_prog;
19 struct bpf_sock_ops_kern;
20 struct bpf_cgroup_storage;
21 struct ctl_table;
22 struct ctl_table_header;
23 struct task_struct;
24 
25 #ifdef CONFIG_CGROUP_BPF
26 
27 extern struct static_key_false cgroup_bpf_enabled_key;
28 #define cgroup_bpf_enabled static_branch_unlikely(&cgroup_bpf_enabled_key)
29 
30 #define for_each_cgroup_storage_type(stype) \
31 	for (stype = 0; stype < MAX_BPF_CGROUP_STORAGE_TYPE; stype++)
32 
33 struct bpf_cgroup_storage_map;
34 
35 struct bpf_storage_buffer {
36 	struct rcu_head rcu;
37 	char data[];
38 };
39 
40 struct bpf_cgroup_storage {
41 	union {
42 		struct bpf_storage_buffer *buf;
43 		void __percpu *percpu_buf;
44 	};
45 	struct bpf_cgroup_storage_map *map;
46 	struct bpf_cgroup_storage_key key;
47 	struct list_head list_map;
48 	struct list_head list_cg;
49 	struct rb_node node;
50 	struct rcu_head rcu;
51 };
52 
53 struct bpf_cgroup_link {
54 	struct bpf_link link;
55 	struct cgroup *cgroup;
56 	enum bpf_attach_type type;
57 };
58 
59 struct bpf_prog_list {
60 	struct list_head node;
61 	struct bpf_prog *prog;
62 	struct bpf_cgroup_link *link;
63 	struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE];
64 };
65 
66 struct bpf_prog_array;
67 
68 struct cgroup_bpf {
69 	/* array of effective progs in this cgroup */
70 	struct bpf_prog_array __rcu *effective[MAX_BPF_ATTACH_TYPE];
71 
72 	/* attached progs to this cgroup and attach flags
73 	 * when flags == 0 or BPF_F_ALLOW_OVERRIDE the progs list will
74 	 * have either zero or one element
75 	 * when BPF_F_ALLOW_MULTI the list can have up to BPF_CGROUP_MAX_PROGS
76 	 */
77 	struct list_head progs[MAX_BPF_ATTACH_TYPE];
78 	u32 flags[MAX_BPF_ATTACH_TYPE];
79 
80 	/* list of cgroup shared storages */
81 	struct list_head storages;
82 
83 	/* temp storage for effective prog array used by prog_attach/detach */
84 	struct bpf_prog_array *inactive;
85 
86 	/* reference counter used to detach bpf programs after cgroup removal */
87 	struct percpu_ref refcnt;
88 
89 	/* cgroup_bpf is released using a work queue */
90 	struct work_struct release_work;
91 };
92 
93 int cgroup_bpf_inherit(struct cgroup *cgrp);
94 void cgroup_bpf_offline(struct cgroup *cgrp);
95 
96 int __cgroup_bpf_attach(struct cgroup *cgrp,
97 			struct bpf_prog *prog, struct bpf_prog *replace_prog,
98 			struct bpf_cgroup_link *link,
99 			enum bpf_attach_type type, u32 flags);
100 int __cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
101 			struct bpf_cgroup_link *link,
102 			enum bpf_attach_type type);
103 int __cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
104 		       union bpf_attr __user *uattr);
105 
106 /* Wrapper for __cgroup_bpf_*() protected by cgroup_mutex */
107 int cgroup_bpf_attach(struct cgroup *cgrp,
108 		      struct bpf_prog *prog, struct bpf_prog *replace_prog,
109 		      struct bpf_cgroup_link *link, enum bpf_attach_type type,
110 		      u32 flags);
111 int cgroup_bpf_detach(struct cgroup *cgrp, struct bpf_prog *prog,
112 		      enum bpf_attach_type type);
113 int cgroup_bpf_query(struct cgroup *cgrp, const union bpf_attr *attr,
114 		     union bpf_attr __user *uattr);
115 
116 int __cgroup_bpf_run_filter_skb(struct sock *sk,
117 				struct sk_buff *skb,
118 				enum bpf_attach_type type);
119 
120 int __cgroup_bpf_run_filter_sk(struct sock *sk,
121 			       enum bpf_attach_type type);
122 
123 int __cgroup_bpf_run_filter_sock_addr(struct sock *sk,
124 				      struct sockaddr *uaddr,
125 				      enum bpf_attach_type type,
126 				      void *t_ctx);
127 
128 int __cgroup_bpf_run_filter_sock_ops(struct sock *sk,
129 				     struct bpf_sock_ops_kern *sock_ops,
130 				     enum bpf_attach_type type);
131 
132 int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor,
133 				      short access, enum bpf_attach_type type);
134 
135 int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head,
136 				   struct ctl_table *table, int write,
137 				   char **buf, size_t *pcount, loff_t *ppos,
138 				   enum bpf_attach_type type);
139 
140 int __cgroup_bpf_run_filter_setsockopt(struct sock *sock, int *level,
141 				       int *optname, char __user *optval,
142 				       int *optlen, char **kernel_optval);
143 int __cgroup_bpf_run_filter_getsockopt(struct sock *sk, int level,
144 				       int optname, char __user *optval,
145 				       int __user *optlen, int max_optlen,
146 				       int retval);
147 
cgroup_storage_type(struct bpf_map * map)148 static inline enum bpf_cgroup_storage_type cgroup_storage_type(
149 	struct bpf_map *map)
150 {
151 	if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
152 		return BPF_CGROUP_STORAGE_PERCPU;
153 
154 	return BPF_CGROUP_STORAGE_SHARED;
155 }
156 
157 struct bpf_cgroup_storage *
158 cgroup_storage_lookup(struct bpf_cgroup_storage_map *map,
159 		      void *key, bool locked);
160 struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(struct bpf_prog *prog,
161 					enum bpf_cgroup_storage_type stype);
162 void bpf_cgroup_storage_free(struct bpf_cgroup_storage *storage);
163 void bpf_cgroup_storage_link(struct bpf_cgroup_storage *storage,
164 			     struct cgroup *cgroup,
165 			     enum bpf_attach_type type);
166 void bpf_cgroup_storage_unlink(struct bpf_cgroup_storage *storage);
167 int bpf_cgroup_storage_assign(struct bpf_prog_aux *aux, struct bpf_map *map);
168 
169 int bpf_percpu_cgroup_storage_copy(struct bpf_map *map, void *key, void *value);
170 int bpf_percpu_cgroup_storage_update(struct bpf_map *map, void *key,
171 				     void *value, u64 flags);
172 
173 /* Wrappers for __cgroup_bpf_run_filter_skb() guarded by cgroup_bpf_enabled. */
174 #define BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb)			      \
175 ({									      \
176 	int __ret = 0;							      \
177 	if (cgroup_bpf_enabled)						      \
178 		__ret = __cgroup_bpf_run_filter_skb(sk, skb,		      \
179 						    BPF_CGROUP_INET_INGRESS); \
180 									      \
181 	__ret;								      \
182 })
183 
184 #define BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb)			       \
185 ({									       \
186 	int __ret = 0;							       \
187 	if (cgroup_bpf_enabled && sk && sk == skb->sk) {		       \
188 		typeof(sk) __sk = sk_to_full_sk(sk);			       \
189 		if (sk_fullsock(__sk))					       \
190 			__ret = __cgroup_bpf_run_filter_skb(__sk, skb,	       \
191 						      BPF_CGROUP_INET_EGRESS); \
192 	}								       \
193 	__ret;								       \
194 })
195 
196 #define BPF_CGROUP_RUN_SK_PROG(sk, type)				       \
197 ({									       \
198 	int __ret = 0;							       \
199 	if (cgroup_bpf_enabled) {					       \
200 		__ret = __cgroup_bpf_run_filter_sk(sk, type);		       \
201 	}								       \
202 	__ret;								       \
203 })
204 
205 #define BPF_CGROUP_RUN_PROG_INET_SOCK(sk)				       \
206 	BPF_CGROUP_RUN_SK_PROG(sk, BPF_CGROUP_INET_SOCK_CREATE)
207 
208 #define BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk)			       \
209 	BPF_CGROUP_RUN_SK_PROG(sk, BPF_CGROUP_INET_SOCK_RELEASE)
210 
211 #define BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk)				       \
212 	BPF_CGROUP_RUN_SK_PROG(sk, BPF_CGROUP_INET4_POST_BIND)
213 
214 #define BPF_CGROUP_RUN_PROG_INET6_POST_BIND(sk)				       \
215 	BPF_CGROUP_RUN_SK_PROG(sk, BPF_CGROUP_INET6_POST_BIND)
216 
217 #define BPF_CGROUP_RUN_SA_PROG(sk, uaddr, type)				       \
218 ({									       \
219 	int __ret = 0;							       \
220 	if (cgroup_bpf_enabled)						       \
221 		__ret = __cgroup_bpf_run_filter_sock_addr(sk, uaddr, type,     \
222 							  NULL);	       \
223 	__ret;								       \
224 })
225 
226 #define BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, type, t_ctx)		       \
227 ({									       \
228 	int __ret = 0;							       \
229 	if (cgroup_bpf_enabled)	{					       \
230 		lock_sock(sk);						       \
231 		__ret = __cgroup_bpf_run_filter_sock_addr(sk, uaddr, type,     \
232 							  t_ctx);	       \
233 		release_sock(sk);					       \
234 	}								       \
235 	__ret;								       \
236 })
237 
238 #define BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr)			       \
239 	BPF_CGROUP_RUN_SA_PROG(sk, uaddr, BPF_CGROUP_INET4_BIND)
240 
241 #define BPF_CGROUP_RUN_PROG_INET6_BIND(sk, uaddr)			       \
242 	BPF_CGROUP_RUN_SA_PROG(sk, uaddr, BPF_CGROUP_INET6_BIND)
243 
244 #define BPF_CGROUP_PRE_CONNECT_ENABLED(sk) (cgroup_bpf_enabled && \
245 					    sk->sk_prot->pre_connect)
246 
247 #define BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr)			       \
248 	BPF_CGROUP_RUN_SA_PROG(sk, uaddr, BPF_CGROUP_INET4_CONNECT)
249 
250 #define BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr)			       \
251 	BPF_CGROUP_RUN_SA_PROG(sk, uaddr, BPF_CGROUP_INET6_CONNECT)
252 
253 #define BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr)		       \
254 	BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_INET4_CONNECT, NULL)
255 
256 #define BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr)		       \
257 	BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_INET6_CONNECT, NULL)
258 
259 #define BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, uaddr, t_ctx)		       \
260 	BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_UDP4_SENDMSG, t_ctx)
261 
262 #define BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, uaddr, t_ctx)		       \
263 	BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_UDP6_SENDMSG, t_ctx)
264 
265 #define BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, uaddr)			\
266 	BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_UDP4_RECVMSG, NULL)
267 
268 #define BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, uaddr)			\
269 	BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, BPF_CGROUP_UDP6_RECVMSG, NULL)
270 
271 /* The SOCK_OPS"_SK" macro should be used when sock_ops->sk is not a
272  * fullsock and its parent fullsock cannot be traced by
273  * sk_to_full_sk().
274  *
275  * e.g. sock_ops->sk is a request_sock and it is under syncookie mode.
276  * Its listener-sk is not attached to the rsk_listener.
277  * In this case, the caller holds the listener-sk (unlocked),
278  * set its sock_ops->sk to req_sk, and call this SOCK_OPS"_SK" with
279  * the listener-sk such that the cgroup-bpf-progs of the
280  * listener-sk will be run.
281  *
282  * Regardless of syncookie mode or not,
283  * calling bpf_setsockopt on listener-sk will not make sense anyway,
284  * so passing 'sock_ops->sk == req_sk' to the bpf prog is appropriate here.
285  */
286 #define BPF_CGROUP_RUN_PROG_SOCK_OPS_SK(sock_ops, sk)			\
287 ({									\
288 	int __ret = 0;							\
289 	if (cgroup_bpf_enabled)						\
290 		__ret = __cgroup_bpf_run_filter_sock_ops(sk,		\
291 							 sock_ops,	\
292 							 BPF_CGROUP_SOCK_OPS); \
293 	__ret;								\
294 })
295 
296 #define BPF_CGROUP_RUN_PROG_SOCK_OPS(sock_ops)				       \
297 ({									       \
298 	int __ret = 0;							       \
299 	if (cgroup_bpf_enabled && (sock_ops)->sk) {	       \
300 		typeof(sk) __sk = sk_to_full_sk((sock_ops)->sk);	       \
301 		if (__sk && sk_fullsock(__sk))				       \
302 			__ret = __cgroup_bpf_run_filter_sock_ops(__sk,	       \
303 								 sock_ops,     \
304 							 BPF_CGROUP_SOCK_OPS); \
305 	}								       \
306 	__ret;								       \
307 })
308 
309 #define BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type, major, minor, access)	      \
310 ({									      \
311 	int __ret = 0;							      \
312 	if (cgroup_bpf_enabled)						      \
313 		__ret = __cgroup_bpf_check_dev_permission(type, major, minor, \
314 							  access,	      \
315 							  BPF_CGROUP_DEVICE); \
316 									      \
317 	__ret;								      \
318 })
319 
320 
321 #define BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, buf, count, pos)  \
322 ({									       \
323 	int __ret = 0;							       \
324 	if (cgroup_bpf_enabled)						       \
325 		__ret = __cgroup_bpf_run_filter_sysctl(head, table, write,     \
326 						       buf, count, pos,        \
327 						       BPF_CGROUP_SYSCTL);     \
328 	__ret;								       \
329 })
330 
331 #define BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock, level, optname, optval, optlen,   \
332 				       kernel_optval)			       \
333 ({									       \
334 	int __ret = 0;							       \
335 	if (cgroup_bpf_enabled)						       \
336 		__ret = __cgroup_bpf_run_filter_setsockopt(sock, level,	       \
337 							   optname, optval,    \
338 							   optlen,	       \
339 							   kernel_optval);     \
340 	__ret;								       \
341 })
342 
343 #define BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen)			       \
344 ({									       \
345 	int __ret = 0;							       \
346 	if (cgroup_bpf_enabled)						       \
347 		get_user(__ret, optlen);				       \
348 	__ret;								       \
349 })
350 
351 #define BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock, level, optname, optval, optlen,   \
352 				       max_optlen, retval)		       \
353 ({									       \
354 	int __ret = retval;						       \
355 	if (cgroup_bpf_enabled)						       \
356 		__ret = __cgroup_bpf_run_filter_getsockopt(sock, level,	       \
357 							   optname, optval,    \
358 							   optlen, max_optlen, \
359 							   retval);	       \
360 	__ret;								       \
361 })
362 
363 int cgroup_bpf_prog_attach(const union bpf_attr *attr,
364 			   enum bpf_prog_type ptype, struct bpf_prog *prog);
365 int cgroup_bpf_prog_detach(const union bpf_attr *attr,
366 			   enum bpf_prog_type ptype);
367 int cgroup_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
368 int cgroup_bpf_prog_query(const union bpf_attr *attr,
369 			  union bpf_attr __user *uattr);
370 #else
371 
372 struct bpf_prog;
373 struct cgroup_bpf {};
cgroup_bpf_inherit(struct cgroup * cgrp)374 static inline int cgroup_bpf_inherit(struct cgroup *cgrp) { return 0; }
cgroup_bpf_offline(struct cgroup * cgrp)375 static inline void cgroup_bpf_offline(struct cgroup *cgrp) {}
376 
cgroup_bpf_prog_attach(const union bpf_attr * attr,enum bpf_prog_type ptype,struct bpf_prog * prog)377 static inline int cgroup_bpf_prog_attach(const union bpf_attr *attr,
378 					 enum bpf_prog_type ptype,
379 					 struct bpf_prog *prog)
380 {
381 	return -EINVAL;
382 }
383 
cgroup_bpf_prog_detach(const union bpf_attr * attr,enum bpf_prog_type ptype)384 static inline int cgroup_bpf_prog_detach(const union bpf_attr *attr,
385 					 enum bpf_prog_type ptype)
386 {
387 	return -EINVAL;
388 }
389 
cgroup_bpf_link_attach(const union bpf_attr * attr,struct bpf_prog * prog)390 static inline int cgroup_bpf_link_attach(const union bpf_attr *attr,
391 					 struct bpf_prog *prog)
392 {
393 	return -EINVAL;
394 }
395 
cgroup_bpf_prog_query(const union bpf_attr * attr,union bpf_attr __user * uattr)396 static inline int cgroup_bpf_prog_query(const union bpf_attr *attr,
397 					union bpf_attr __user *uattr)
398 {
399 	return -EINVAL;
400 }
401 
bpf_cgroup_storage_assign(struct bpf_prog_aux * aux,struct bpf_map * map)402 static inline int bpf_cgroup_storage_assign(struct bpf_prog_aux *aux,
403 					    struct bpf_map *map) { return 0; }
bpf_cgroup_storage_alloc(struct bpf_prog * prog,enum bpf_cgroup_storage_type stype)404 static inline struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(
405 	struct bpf_prog *prog, enum bpf_cgroup_storage_type stype) { return NULL; }
bpf_cgroup_storage_free(struct bpf_cgroup_storage * storage)406 static inline void bpf_cgroup_storage_free(
407 	struct bpf_cgroup_storage *storage) {}
bpf_percpu_cgroup_storage_copy(struct bpf_map * map,void * key,void * value)408 static inline int bpf_percpu_cgroup_storage_copy(struct bpf_map *map, void *key,
409 						 void *value) {
410 	return 0;
411 }
bpf_percpu_cgroup_storage_update(struct bpf_map * map,void * key,void * value,u64 flags)412 static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,
413 					void *key, void *value, u64 flags) {
414 	return 0;
415 }
416 
417 #define cgroup_bpf_enabled (0)
418 #define BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, type, t_ctx) ({ 0; })
419 #define BPF_CGROUP_PRE_CONNECT_ENABLED(sk) (0)
420 #define BPF_CGROUP_RUN_PROG_INET_INGRESS(sk,skb) ({ 0; })
421 #define BPF_CGROUP_RUN_PROG_INET_EGRESS(sk,skb) ({ 0; })
422 #define BPF_CGROUP_RUN_PROG_INET_SOCK(sk) ({ 0; })
423 #define BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk) ({ 0; })
424 #define BPF_CGROUP_RUN_PROG_INET4_BIND(sk, uaddr) ({ 0; })
425 #define BPF_CGROUP_RUN_PROG_INET6_BIND(sk, uaddr) ({ 0; })
426 #define BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk) ({ 0; })
427 #define BPF_CGROUP_RUN_PROG_INET6_POST_BIND(sk) ({ 0; })
428 #define BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr) ({ 0; })
429 #define BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr) ({ 0; })
430 #define BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr) ({ 0; })
431 #define BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr) ({ 0; })
432 #define BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, uaddr, t_ctx) ({ 0; })
433 #define BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, uaddr, t_ctx) ({ 0; })
434 #define BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, uaddr) ({ 0; })
435 #define BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, uaddr) ({ 0; })
436 #define BPF_CGROUP_RUN_PROG_SOCK_OPS(sock_ops) ({ 0; })
437 #define BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(type,major,minor,access) ({ 0; })
438 #define BPF_CGROUP_RUN_PROG_SYSCTL(head,table,write,buf,count,pos) ({ 0; })
439 #define BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen) ({ 0; })
440 #define BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock, level, optname, optval, \
441 				       optlen, max_optlen, retval) ({ retval; })
442 #define BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock, level, optname, optval, optlen, \
443 				       kernel_optval) ({ 0; })
444 
445 #define for_each_cgroup_storage_type(stype) for (; false; )
446 
447 #endif /* CONFIG_CGROUP_BPF */
448 
449 #endif /* _BPF_CGROUP_H */
450