Linux kernel mirror (for testing)
git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git
tjh.dev
kernel
os
linux
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/bpf-cgroup-defs.h>
7#include <linux/errno.h>
8#include <linux/jump_label.h>
9#include <linux/percpu.h>
10#include <linux/rbtree.h>
11#include <net/sock.h>
12#include <uapi/linux/bpf.h>
13
14struct sock;
15struct sockaddr;
16struct cgroup;
17struct sk_buff;
18struct bpf_map;
19struct bpf_prog;
20struct bpf_sock_ops_kern;
21struct bpf_cgroup_storage;
22struct ctl_table;
23struct ctl_table_header;
24struct task_struct;
25
26#ifdef CONFIG_CGROUP_BPF
27
28#define CGROUP_ATYPE(type) \
29 case BPF_##type: return type
30
31static inline enum cgroup_bpf_attach_type
32to_cgroup_bpf_attach_type(enum bpf_attach_type attach_type)
33{
34 switch (attach_type) {
35 CGROUP_ATYPE(CGROUP_INET_INGRESS);
36 CGROUP_ATYPE(CGROUP_INET_EGRESS);
37 CGROUP_ATYPE(CGROUP_INET_SOCK_CREATE);
38 CGROUP_ATYPE(CGROUP_SOCK_OPS);
39 CGROUP_ATYPE(CGROUP_DEVICE);
40 CGROUP_ATYPE(CGROUP_INET4_BIND);
41 CGROUP_ATYPE(CGROUP_INET6_BIND);
42 CGROUP_ATYPE(CGROUP_INET4_CONNECT);
43 CGROUP_ATYPE(CGROUP_INET6_CONNECT);
44 CGROUP_ATYPE(CGROUP_INET4_POST_BIND);
45 CGROUP_ATYPE(CGROUP_INET6_POST_BIND);
46 CGROUP_ATYPE(CGROUP_UDP4_SENDMSG);
47 CGROUP_ATYPE(CGROUP_UDP6_SENDMSG);
48 CGROUP_ATYPE(CGROUP_SYSCTL);
49 CGROUP_ATYPE(CGROUP_UDP4_RECVMSG);
50 CGROUP_ATYPE(CGROUP_UDP6_RECVMSG);
51 CGROUP_ATYPE(CGROUP_GETSOCKOPT);
52 CGROUP_ATYPE(CGROUP_SETSOCKOPT);
53 CGROUP_ATYPE(CGROUP_INET4_GETPEERNAME);
54 CGROUP_ATYPE(CGROUP_INET6_GETPEERNAME);
55 CGROUP_ATYPE(CGROUP_INET4_GETSOCKNAME);
56 CGROUP_ATYPE(CGROUP_INET6_GETSOCKNAME);
57 CGROUP_ATYPE(CGROUP_INET_SOCK_RELEASE);
58 default:
59 return CGROUP_BPF_ATTACH_TYPE_INVALID;
60 }
61}
62
63#undef CGROUP_ATYPE
64
65extern struct static_key_false cgroup_bpf_enabled_key[MAX_CGROUP_BPF_ATTACH_TYPE];
66#define cgroup_bpf_enabled(atype) static_branch_unlikely(&cgroup_bpf_enabled_key[atype])
67
68#define for_each_cgroup_storage_type(stype) \
69 for (stype = 0; stype < MAX_BPF_CGROUP_STORAGE_TYPE; stype++)
70
71struct bpf_cgroup_storage_map;
72
73struct bpf_storage_buffer {
74 struct rcu_head rcu;
75 char data[];
76};
77
78struct bpf_cgroup_storage {
79 union {
80 struct bpf_storage_buffer *buf;
81 void __percpu *percpu_buf;
82 };
83 struct bpf_cgroup_storage_map *map;
84 struct bpf_cgroup_storage_key key;
85 struct list_head list_map;
86 struct list_head list_cg;
87 struct rb_node node;
88 struct rcu_head rcu;
89};
90
91struct bpf_cgroup_link {
92 struct bpf_link link;
93 struct cgroup *cgroup;
94 enum bpf_attach_type type;
95};
96
97struct bpf_prog_list {
98 struct list_head node;
99 struct bpf_prog *prog;
100 struct bpf_cgroup_link *link;
101 struct bpf_cgroup_storage *storage[MAX_BPF_CGROUP_STORAGE_TYPE];
102};
103
104int cgroup_bpf_inherit(struct cgroup *cgrp);
105void cgroup_bpf_offline(struct cgroup *cgrp);
106
107int __cgroup_bpf_run_filter_skb(struct sock *sk,
108 struct sk_buff *skb,
109 enum cgroup_bpf_attach_type atype);
110
111int __cgroup_bpf_run_filter_sk(struct sock *sk,
112 enum cgroup_bpf_attach_type atype);
113
114int __cgroup_bpf_run_filter_sock_addr(struct sock *sk,
115 struct sockaddr *uaddr,
116 enum cgroup_bpf_attach_type atype,
117 void *t_ctx,
118 u32 *flags);
119
120int __cgroup_bpf_run_filter_sock_ops(struct sock *sk,
121 struct bpf_sock_ops_kern *sock_ops,
122 enum cgroup_bpf_attach_type atype);
123
124int __cgroup_bpf_check_dev_permission(short dev_type, u32 major, u32 minor,
125 short access, enum cgroup_bpf_attach_type atype);
126
127int __cgroup_bpf_run_filter_sysctl(struct ctl_table_header *head,
128 struct ctl_table *table, int write,
129 char **buf, size_t *pcount, loff_t *ppos,
130 enum cgroup_bpf_attach_type atype);
131
132int __cgroup_bpf_run_filter_setsockopt(struct sock *sock, int *level,
133 int *optname, char __user *optval,
134 int *optlen, char **kernel_optval);
135int __cgroup_bpf_run_filter_getsockopt(struct sock *sk, int level,
136 int optname, char __user *optval,
137 int __user *optlen, int max_optlen,
138 int retval);
139
140int __cgroup_bpf_run_filter_getsockopt_kern(struct sock *sk, int level,
141 int optname, void *optval,
142 int *optlen, int retval);
143
144static inline enum bpf_cgroup_storage_type cgroup_storage_type(
145 struct bpf_map *map)
146{
147 if (map->map_type == BPF_MAP_TYPE_PERCPU_CGROUP_STORAGE)
148 return BPF_CGROUP_STORAGE_PERCPU;
149
150 return BPF_CGROUP_STORAGE_SHARED;
151}
152
153struct bpf_cgroup_storage *
154cgroup_storage_lookup(struct bpf_cgroup_storage_map *map,
155 void *key, bool locked);
156struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(struct bpf_prog *prog,
157 enum bpf_cgroup_storage_type stype);
158void bpf_cgroup_storage_free(struct bpf_cgroup_storage *storage);
159void bpf_cgroup_storage_link(struct bpf_cgroup_storage *storage,
160 struct cgroup *cgroup,
161 enum bpf_attach_type type);
162void bpf_cgroup_storage_unlink(struct bpf_cgroup_storage *storage);
163int bpf_cgroup_storage_assign(struct bpf_prog_aux *aux, struct bpf_map *map);
164
165int bpf_percpu_cgroup_storage_copy(struct bpf_map *map, void *key, void *value);
166int bpf_percpu_cgroup_storage_update(struct bpf_map *map, void *key,
167 void *value, u64 flags);
168
169/* Opportunistic check to see whether we have any BPF program attached*/
170static inline bool cgroup_bpf_sock_enabled(struct sock *sk,
171 enum cgroup_bpf_attach_type type)
172{
173 struct cgroup *cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data);
174 struct bpf_prog_array *array;
175
176 array = rcu_access_pointer(cgrp->bpf.effective[type]);
177 return array != &bpf_empty_prog_array.hdr;
178}
179
180/* Wrappers for __cgroup_bpf_run_filter_skb() guarded by cgroup_bpf_enabled. */
181#define BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb) \
182({ \
183 int __ret = 0; \
184 if (cgroup_bpf_enabled(CGROUP_INET_INGRESS) && \
185 cgroup_bpf_sock_enabled(sk, CGROUP_INET_INGRESS)) \
186 __ret = __cgroup_bpf_run_filter_skb(sk, skb, \
187 CGROUP_INET_INGRESS); \
188 \
189 __ret; \
190})
191
192#define BPF_CGROUP_RUN_PROG_INET_EGRESS(sk, skb) \
193({ \
194 int __ret = 0; \
195 if (cgroup_bpf_enabled(CGROUP_INET_EGRESS) && sk && sk == skb->sk) { \
196 typeof(sk) __sk = sk_to_full_sk(sk); \
197 if (sk_fullsock(__sk) && \
198 cgroup_bpf_sock_enabled(__sk, CGROUP_INET_EGRESS)) \
199 __ret = __cgroup_bpf_run_filter_skb(__sk, skb, \
200 CGROUP_INET_EGRESS); \
201 } \
202 __ret; \
203})
204
205#define BPF_CGROUP_RUN_SK_PROG(sk, atype) \
206({ \
207 int __ret = 0; \
208 if (cgroup_bpf_enabled(atype)) { \
209 __ret = __cgroup_bpf_run_filter_sk(sk, atype); \
210 } \
211 __ret; \
212})
213
214#define BPF_CGROUP_RUN_PROG_INET_SOCK(sk) \
215 BPF_CGROUP_RUN_SK_PROG(sk, CGROUP_INET_SOCK_CREATE)
216
217#define BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk) \
218 BPF_CGROUP_RUN_SK_PROG(sk, CGROUP_INET_SOCK_RELEASE)
219
220#define BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk) \
221 BPF_CGROUP_RUN_SK_PROG(sk, CGROUP_INET4_POST_BIND)
222
223#define BPF_CGROUP_RUN_PROG_INET6_POST_BIND(sk) \
224 BPF_CGROUP_RUN_SK_PROG(sk, CGROUP_INET6_POST_BIND)
225
226#define BPF_CGROUP_RUN_SA_PROG(sk, uaddr, atype) \
227({ \
228 int __ret = 0; \
229 if (cgroup_bpf_enabled(atype)) \
230 __ret = __cgroup_bpf_run_filter_sock_addr(sk, uaddr, atype, \
231 NULL, NULL); \
232 __ret; \
233})
234
235#define BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, atype, t_ctx) \
236({ \
237 int __ret = 0; \
238 if (cgroup_bpf_enabled(atype)) { \
239 lock_sock(sk); \
240 __ret = __cgroup_bpf_run_filter_sock_addr(sk, uaddr, atype, \
241 t_ctx, NULL); \
242 release_sock(sk); \
243 } \
244 __ret; \
245})
246
247/* BPF_CGROUP_INET4_BIND and BPF_CGROUP_INET6_BIND can return extra flags
248 * via upper bits of return code. The only flag that is supported
249 * (at bit position 0) is to indicate CAP_NET_BIND_SERVICE capability check
250 * should be bypassed (BPF_RET_BIND_NO_CAP_NET_BIND_SERVICE).
251 */
252#define BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, atype, bind_flags) \
253({ \
254 u32 __flags = 0; \
255 int __ret = 0; \
256 if (cgroup_bpf_enabled(atype)) { \
257 lock_sock(sk); \
258 __ret = __cgroup_bpf_run_filter_sock_addr(sk, uaddr, atype, \
259 NULL, &__flags); \
260 release_sock(sk); \
261 if (__flags & BPF_RET_BIND_NO_CAP_NET_BIND_SERVICE) \
262 *bind_flags |= BIND_NO_CAP_NET_BIND_SERVICE; \
263 } \
264 __ret; \
265})
266
267#define BPF_CGROUP_PRE_CONNECT_ENABLED(sk) \
268 ((cgroup_bpf_enabled(CGROUP_INET4_CONNECT) || \
269 cgroup_bpf_enabled(CGROUP_INET6_CONNECT)) && \
270 (sk)->sk_prot->pre_connect)
271
272#define BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr) \
273 BPF_CGROUP_RUN_SA_PROG(sk, uaddr, CGROUP_INET4_CONNECT)
274
275#define BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr) \
276 BPF_CGROUP_RUN_SA_PROG(sk, uaddr, CGROUP_INET6_CONNECT)
277
278#define BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr) \
279 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, CGROUP_INET4_CONNECT, NULL)
280
281#define BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr) \
282 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, CGROUP_INET6_CONNECT, NULL)
283
284#define BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, uaddr, t_ctx) \
285 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, CGROUP_UDP4_SENDMSG, t_ctx)
286
287#define BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, uaddr, t_ctx) \
288 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, CGROUP_UDP6_SENDMSG, t_ctx)
289
290#define BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, uaddr) \
291 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, CGROUP_UDP4_RECVMSG, NULL)
292
293#define BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, uaddr) \
294 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, CGROUP_UDP6_RECVMSG, NULL)
295
296/* The SOCK_OPS"_SK" macro should be used when sock_ops->sk is not a
297 * fullsock and its parent fullsock cannot be traced by
298 * sk_to_full_sk().
299 *
300 * e.g. sock_ops->sk is a request_sock and it is under syncookie mode.
301 * Its listener-sk is not attached to the rsk_listener.
302 * In this case, the caller holds the listener-sk (unlocked),
303 * set its sock_ops->sk to req_sk, and call this SOCK_OPS"_SK" with
304 * the listener-sk such that the cgroup-bpf-progs of the
305 * listener-sk will be run.
306 *
307 * Regardless of syncookie mode or not,
308 * calling bpf_setsockopt on listener-sk will not make sense anyway,
309 * so passing 'sock_ops->sk == req_sk' to the bpf prog is appropriate here.
310 */
311#define BPF_CGROUP_RUN_PROG_SOCK_OPS_SK(sock_ops, sk) \
312({ \
313 int __ret = 0; \
314 if (cgroup_bpf_enabled(CGROUP_SOCK_OPS)) \
315 __ret = __cgroup_bpf_run_filter_sock_ops(sk, \
316 sock_ops, \
317 CGROUP_SOCK_OPS); \
318 __ret; \
319})
320
321#define BPF_CGROUP_RUN_PROG_SOCK_OPS(sock_ops) \
322({ \
323 int __ret = 0; \
324 if (cgroup_bpf_enabled(CGROUP_SOCK_OPS) && (sock_ops)->sk) { \
325 typeof(sk) __sk = sk_to_full_sk((sock_ops)->sk); \
326 if (__sk && sk_fullsock(__sk)) \
327 __ret = __cgroup_bpf_run_filter_sock_ops(__sk, \
328 sock_ops, \
329 CGROUP_SOCK_OPS); \
330 } \
331 __ret; \
332})
333
334#define BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(atype, major, minor, access) \
335({ \
336 int __ret = 0; \
337 if (cgroup_bpf_enabled(CGROUP_DEVICE)) \
338 __ret = __cgroup_bpf_check_dev_permission(atype, major, minor, \
339 access, \
340 CGROUP_DEVICE); \
341 \
342 __ret; \
343})
344
345
346#define BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, buf, count, pos) \
347({ \
348 int __ret = 0; \
349 if (cgroup_bpf_enabled(CGROUP_SYSCTL)) \
350 __ret = __cgroup_bpf_run_filter_sysctl(head, table, write, \
351 buf, count, pos, \
352 CGROUP_SYSCTL); \
353 __ret; \
354})
355
356#define BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock, level, optname, optval, optlen, \
357 kernel_optval) \
358({ \
359 int __ret = 0; \
360 if (cgroup_bpf_enabled(CGROUP_SETSOCKOPT) && \
361 cgroup_bpf_sock_enabled(sock, CGROUP_SETSOCKOPT)) \
362 __ret = __cgroup_bpf_run_filter_setsockopt(sock, level, \
363 optname, optval, \
364 optlen, \
365 kernel_optval); \
366 __ret; \
367})
368
369#define BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen) \
370({ \
371 int __ret = 0; \
372 if (cgroup_bpf_enabled(CGROUP_GETSOCKOPT)) \
373 get_user(__ret, optlen); \
374 __ret; \
375})
376
377#define BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock, level, optname, optval, optlen, \
378 max_optlen, retval) \
379({ \
380 int __ret = retval; \
381 if (cgroup_bpf_enabled(CGROUP_GETSOCKOPT) && \
382 cgroup_bpf_sock_enabled(sock, CGROUP_GETSOCKOPT)) \
383 if (!(sock)->sk_prot->bpf_bypass_getsockopt || \
384 !INDIRECT_CALL_INET_1((sock)->sk_prot->bpf_bypass_getsockopt, \
385 tcp_bpf_bypass_getsockopt, \
386 level, optname)) \
387 __ret = __cgroup_bpf_run_filter_getsockopt( \
388 sock, level, optname, optval, optlen, \
389 max_optlen, retval); \
390 __ret; \
391})
392
393#define BPF_CGROUP_RUN_PROG_GETSOCKOPT_KERN(sock, level, optname, optval, \
394 optlen, retval) \
395({ \
396 int __ret = retval; \
397 if (cgroup_bpf_enabled(CGROUP_GETSOCKOPT)) \
398 __ret = __cgroup_bpf_run_filter_getsockopt_kern( \
399 sock, level, optname, optval, optlen, retval); \
400 __ret; \
401})
402
403int cgroup_bpf_prog_attach(const union bpf_attr *attr,
404 enum bpf_prog_type ptype, struct bpf_prog *prog);
405int cgroup_bpf_prog_detach(const union bpf_attr *attr,
406 enum bpf_prog_type ptype);
407int cgroup_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
408int cgroup_bpf_prog_query(const union bpf_attr *attr,
409 union bpf_attr __user *uattr);
410#else
411
412static inline int cgroup_bpf_inherit(struct cgroup *cgrp) { return 0; }
413static inline void cgroup_bpf_offline(struct cgroup *cgrp) {}
414
415static inline int cgroup_bpf_prog_attach(const union bpf_attr *attr,
416 enum bpf_prog_type ptype,
417 struct bpf_prog *prog)
418{
419 return -EINVAL;
420}
421
422static inline int cgroup_bpf_prog_detach(const union bpf_attr *attr,
423 enum bpf_prog_type ptype)
424{
425 return -EINVAL;
426}
427
428static inline int cgroup_bpf_link_attach(const union bpf_attr *attr,
429 struct bpf_prog *prog)
430{
431 return -EINVAL;
432}
433
434static inline int cgroup_bpf_prog_query(const union bpf_attr *attr,
435 union bpf_attr __user *uattr)
436{
437 return -EINVAL;
438}
439
440static inline int bpf_cgroup_storage_assign(struct bpf_prog_aux *aux,
441 struct bpf_map *map) { return 0; }
442static inline struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(
443 struct bpf_prog *prog, enum bpf_cgroup_storage_type stype) { return NULL; }
444static inline void bpf_cgroup_storage_free(
445 struct bpf_cgroup_storage *storage) {}
446static inline int bpf_percpu_cgroup_storage_copy(struct bpf_map *map, void *key,
447 void *value) {
448 return 0;
449}
450static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,
451 void *key, void *value, u64 flags) {
452 return 0;
453}
454
455#define cgroup_bpf_enabled(atype) (0)
456#define BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, atype, t_ctx) ({ 0; })
457#define BPF_CGROUP_RUN_SA_PROG(sk, uaddr, atype) ({ 0; })
458#define BPF_CGROUP_PRE_CONNECT_ENABLED(sk) (0)
459#define BPF_CGROUP_RUN_PROG_INET_INGRESS(sk,skb) ({ 0; })
460#define BPF_CGROUP_RUN_PROG_INET_EGRESS(sk,skb) ({ 0; })
461#define BPF_CGROUP_RUN_PROG_INET_SOCK(sk) ({ 0; })
462#define BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk) ({ 0; })
463#define BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, atype, flags) ({ 0; })
464#define BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk) ({ 0; })
465#define BPF_CGROUP_RUN_PROG_INET6_POST_BIND(sk) ({ 0; })
466#define BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr) ({ 0; })
467#define BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr) ({ 0; })
468#define BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr) ({ 0; })
469#define BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr) ({ 0; })
470#define BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, uaddr, t_ctx) ({ 0; })
471#define BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, uaddr, t_ctx) ({ 0; })
472#define BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, uaddr) ({ 0; })
473#define BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, uaddr) ({ 0; })
474#define BPF_CGROUP_RUN_PROG_SOCK_OPS(sock_ops) ({ 0; })
475#define BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(atype, major, minor, access) ({ 0; })
476#define BPF_CGROUP_RUN_PROG_SYSCTL(head,table,write,buf,count,pos) ({ 0; })
477#define BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen) ({ 0; })
478#define BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock, level, optname, optval, \
479 optlen, max_optlen, retval) ({ retval; })
480#define BPF_CGROUP_RUN_PROG_GETSOCKOPT_KERN(sock, level, optname, optval, \
481 optlen, retval) ({ retval; })
482#define BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock, level, optname, optval, optlen, \
483 kernel_optval) ({ 0; })
484
485#define for_each_cgroup_storage_type(stype) for (; false; )
486
487#endif /* CONFIG_CGROUP_BPF */
488
489#endif /* _BPF_CGROUP_H */