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 u32 __unused_flags; \
229 int __ret = 0; \
230 if (cgroup_bpf_enabled(atype)) \
231 __ret = __cgroup_bpf_run_filter_sock_addr(sk, uaddr, atype, \
232 NULL, \
233 &__unused_flags); \
234 __ret; \
235})
236
237#define BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, atype, t_ctx) \
238({ \
239 u32 __unused_flags; \
240 int __ret = 0; \
241 if (cgroup_bpf_enabled(atype)) { \
242 lock_sock(sk); \
243 __ret = __cgroup_bpf_run_filter_sock_addr(sk, uaddr, atype, \
244 t_ctx, \
245 &__unused_flags); \
246 release_sock(sk); \
247 } \
248 __ret; \
249})
250
251/* BPF_CGROUP_INET4_BIND and BPF_CGROUP_INET6_BIND can return extra flags
252 * via upper bits of return code. The only flag that is supported
253 * (at bit position 0) is to indicate CAP_NET_BIND_SERVICE capability check
254 * should be bypassed (BPF_RET_BIND_NO_CAP_NET_BIND_SERVICE).
255 */
256#define BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, atype, bind_flags) \
257({ \
258 u32 __flags = 0; \
259 int __ret = 0; \
260 if (cgroup_bpf_enabled(atype)) { \
261 lock_sock(sk); \
262 __ret = __cgroup_bpf_run_filter_sock_addr(sk, uaddr, atype, \
263 NULL, &__flags); \
264 release_sock(sk); \
265 if (__flags & BPF_RET_BIND_NO_CAP_NET_BIND_SERVICE) \
266 *bind_flags |= BIND_NO_CAP_NET_BIND_SERVICE; \
267 } \
268 __ret; \
269})
270
271#define BPF_CGROUP_PRE_CONNECT_ENABLED(sk) \
272 ((cgroup_bpf_enabled(CGROUP_INET4_CONNECT) || \
273 cgroup_bpf_enabled(CGROUP_INET6_CONNECT)) && \
274 (sk)->sk_prot->pre_connect)
275
276#define BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr) \
277 BPF_CGROUP_RUN_SA_PROG(sk, uaddr, CGROUP_INET4_CONNECT)
278
279#define BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr) \
280 BPF_CGROUP_RUN_SA_PROG(sk, uaddr, CGROUP_INET6_CONNECT)
281
282#define BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr) \
283 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, CGROUP_INET4_CONNECT, NULL)
284
285#define BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr) \
286 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, CGROUP_INET6_CONNECT, NULL)
287
288#define BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, uaddr, t_ctx) \
289 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, CGROUP_UDP4_SENDMSG, t_ctx)
290
291#define BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, uaddr, t_ctx) \
292 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, CGROUP_UDP6_SENDMSG, t_ctx)
293
294#define BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, uaddr) \
295 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, CGROUP_UDP4_RECVMSG, NULL)
296
297#define BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, uaddr) \
298 BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, CGROUP_UDP6_RECVMSG, NULL)
299
300/* The SOCK_OPS"_SK" macro should be used when sock_ops->sk is not a
301 * fullsock and its parent fullsock cannot be traced by
302 * sk_to_full_sk().
303 *
304 * e.g. sock_ops->sk is a request_sock and it is under syncookie mode.
305 * Its listener-sk is not attached to the rsk_listener.
306 * In this case, the caller holds the listener-sk (unlocked),
307 * set its sock_ops->sk to req_sk, and call this SOCK_OPS"_SK" with
308 * the listener-sk such that the cgroup-bpf-progs of the
309 * listener-sk will be run.
310 *
311 * Regardless of syncookie mode or not,
312 * calling bpf_setsockopt on listener-sk will not make sense anyway,
313 * so passing 'sock_ops->sk == req_sk' to the bpf prog is appropriate here.
314 */
315#define BPF_CGROUP_RUN_PROG_SOCK_OPS_SK(sock_ops, sk) \
316({ \
317 int __ret = 0; \
318 if (cgroup_bpf_enabled(CGROUP_SOCK_OPS)) \
319 __ret = __cgroup_bpf_run_filter_sock_ops(sk, \
320 sock_ops, \
321 CGROUP_SOCK_OPS); \
322 __ret; \
323})
324
325#define BPF_CGROUP_RUN_PROG_SOCK_OPS(sock_ops) \
326({ \
327 int __ret = 0; \
328 if (cgroup_bpf_enabled(CGROUP_SOCK_OPS) && (sock_ops)->sk) { \
329 typeof(sk) __sk = sk_to_full_sk((sock_ops)->sk); \
330 if (__sk && sk_fullsock(__sk)) \
331 __ret = __cgroup_bpf_run_filter_sock_ops(__sk, \
332 sock_ops, \
333 CGROUP_SOCK_OPS); \
334 } \
335 __ret; \
336})
337
338#define BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(atype, major, minor, access) \
339({ \
340 int __ret = 0; \
341 if (cgroup_bpf_enabled(CGROUP_DEVICE)) \
342 __ret = __cgroup_bpf_check_dev_permission(atype, major, minor, \
343 access, \
344 CGROUP_DEVICE); \
345 \
346 __ret; \
347})
348
349
350#define BPF_CGROUP_RUN_PROG_SYSCTL(head, table, write, buf, count, pos) \
351({ \
352 int __ret = 0; \
353 if (cgroup_bpf_enabled(CGROUP_SYSCTL)) \
354 __ret = __cgroup_bpf_run_filter_sysctl(head, table, write, \
355 buf, count, pos, \
356 CGROUP_SYSCTL); \
357 __ret; \
358})
359
360#define BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock, level, optname, optval, optlen, \
361 kernel_optval) \
362({ \
363 int __ret = 0; \
364 if (cgroup_bpf_enabled(CGROUP_SETSOCKOPT) && \
365 cgroup_bpf_sock_enabled(sock, CGROUP_SETSOCKOPT)) \
366 __ret = __cgroup_bpf_run_filter_setsockopt(sock, level, \
367 optname, optval, \
368 optlen, \
369 kernel_optval); \
370 __ret; \
371})
372
373#define BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen) \
374({ \
375 int __ret = 0; \
376 if (cgroup_bpf_enabled(CGROUP_GETSOCKOPT)) \
377 get_user(__ret, optlen); \
378 __ret; \
379})
380
381#define BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock, level, optname, optval, optlen, \
382 max_optlen, retval) \
383({ \
384 int __ret = retval; \
385 if (cgroup_bpf_enabled(CGROUP_GETSOCKOPT) && \
386 cgroup_bpf_sock_enabled(sock, CGROUP_GETSOCKOPT)) \
387 if (!(sock)->sk_prot->bpf_bypass_getsockopt || \
388 !INDIRECT_CALL_INET_1((sock)->sk_prot->bpf_bypass_getsockopt, \
389 tcp_bpf_bypass_getsockopt, \
390 level, optname)) \
391 __ret = __cgroup_bpf_run_filter_getsockopt( \
392 sock, level, optname, optval, optlen, \
393 max_optlen, retval); \
394 __ret; \
395})
396
397#define BPF_CGROUP_RUN_PROG_GETSOCKOPT_KERN(sock, level, optname, optval, \
398 optlen, retval) \
399({ \
400 int __ret = retval; \
401 if (cgroup_bpf_enabled(CGROUP_GETSOCKOPT)) \
402 __ret = __cgroup_bpf_run_filter_getsockopt_kern( \
403 sock, level, optname, optval, optlen, retval); \
404 __ret; \
405})
406
407int cgroup_bpf_prog_attach(const union bpf_attr *attr,
408 enum bpf_prog_type ptype, struct bpf_prog *prog);
409int cgroup_bpf_prog_detach(const union bpf_attr *attr,
410 enum bpf_prog_type ptype);
411int cgroup_bpf_link_attach(const union bpf_attr *attr, struct bpf_prog *prog);
412int cgroup_bpf_prog_query(const union bpf_attr *attr,
413 union bpf_attr __user *uattr);
414#else
415
416static inline int cgroup_bpf_inherit(struct cgroup *cgrp) { return 0; }
417static inline void cgroup_bpf_offline(struct cgroup *cgrp) {}
418
419static inline int cgroup_bpf_prog_attach(const union bpf_attr *attr,
420 enum bpf_prog_type ptype,
421 struct bpf_prog *prog)
422{
423 return -EINVAL;
424}
425
426static inline int cgroup_bpf_prog_detach(const union bpf_attr *attr,
427 enum bpf_prog_type ptype)
428{
429 return -EINVAL;
430}
431
432static inline int cgroup_bpf_link_attach(const union bpf_attr *attr,
433 struct bpf_prog *prog)
434{
435 return -EINVAL;
436}
437
438static inline int cgroup_bpf_prog_query(const union bpf_attr *attr,
439 union bpf_attr __user *uattr)
440{
441 return -EINVAL;
442}
443
444static inline int bpf_cgroup_storage_assign(struct bpf_prog_aux *aux,
445 struct bpf_map *map) { return 0; }
446static inline struct bpf_cgroup_storage *bpf_cgroup_storage_alloc(
447 struct bpf_prog *prog, enum bpf_cgroup_storage_type stype) { return NULL; }
448static inline void bpf_cgroup_storage_free(
449 struct bpf_cgroup_storage *storage) {}
450static inline int bpf_percpu_cgroup_storage_copy(struct bpf_map *map, void *key,
451 void *value) {
452 return 0;
453}
454static inline int bpf_percpu_cgroup_storage_update(struct bpf_map *map,
455 void *key, void *value, u64 flags) {
456 return 0;
457}
458
459#define cgroup_bpf_enabled(atype) (0)
460#define BPF_CGROUP_RUN_SA_PROG_LOCK(sk, uaddr, atype, t_ctx) ({ 0; })
461#define BPF_CGROUP_RUN_SA_PROG(sk, uaddr, atype) ({ 0; })
462#define BPF_CGROUP_PRE_CONNECT_ENABLED(sk) (0)
463#define BPF_CGROUP_RUN_PROG_INET_INGRESS(sk,skb) ({ 0; })
464#define BPF_CGROUP_RUN_PROG_INET_EGRESS(sk,skb) ({ 0; })
465#define BPF_CGROUP_RUN_PROG_INET_SOCK(sk) ({ 0; })
466#define BPF_CGROUP_RUN_PROG_INET_SOCK_RELEASE(sk) ({ 0; })
467#define BPF_CGROUP_RUN_PROG_INET_BIND_LOCK(sk, uaddr, atype, flags) ({ 0; })
468#define BPF_CGROUP_RUN_PROG_INET4_POST_BIND(sk) ({ 0; })
469#define BPF_CGROUP_RUN_PROG_INET6_POST_BIND(sk) ({ 0; })
470#define BPF_CGROUP_RUN_PROG_INET4_CONNECT(sk, uaddr) ({ 0; })
471#define BPF_CGROUP_RUN_PROG_INET4_CONNECT_LOCK(sk, uaddr) ({ 0; })
472#define BPF_CGROUP_RUN_PROG_INET6_CONNECT(sk, uaddr) ({ 0; })
473#define BPF_CGROUP_RUN_PROG_INET6_CONNECT_LOCK(sk, uaddr) ({ 0; })
474#define BPF_CGROUP_RUN_PROG_UDP4_SENDMSG_LOCK(sk, uaddr, t_ctx) ({ 0; })
475#define BPF_CGROUP_RUN_PROG_UDP6_SENDMSG_LOCK(sk, uaddr, t_ctx) ({ 0; })
476#define BPF_CGROUP_RUN_PROG_UDP4_RECVMSG_LOCK(sk, uaddr) ({ 0; })
477#define BPF_CGROUP_RUN_PROG_UDP6_RECVMSG_LOCK(sk, uaddr) ({ 0; })
478#define BPF_CGROUP_RUN_PROG_SOCK_OPS(sock_ops) ({ 0; })
479#define BPF_CGROUP_RUN_PROG_DEVICE_CGROUP(atype, major, minor, access) ({ 0; })
480#define BPF_CGROUP_RUN_PROG_SYSCTL(head,table,write,buf,count,pos) ({ 0; })
481#define BPF_CGROUP_GETSOCKOPT_MAX_OPTLEN(optlen) ({ 0; })
482#define BPF_CGROUP_RUN_PROG_GETSOCKOPT(sock, level, optname, optval, \
483 optlen, max_optlen, retval) ({ retval; })
484#define BPF_CGROUP_RUN_PROG_GETSOCKOPT_KERN(sock, level, optname, optval, \
485 optlen, retval) ({ retval; })
486#define BPF_CGROUP_RUN_PROG_SETSOCKOPT(sock, level, optname, optval, optlen, \
487 kernel_optval) ({ 0; })
488
489#define for_each_cgroup_storage_type(stype) for (; false; )
490
491#endif /* CONFIG_CGROUP_BPF */
492
493#endif /* _BPF_CGROUP_H */