1 // SPDX-License-Identifier: GPL-2.0-only
2 /* Unstable Conntrack Helpers for XDP and TC-BPF hook
4 * These are called from the XDP and SCHED_CLS BPF programs. Note that it is
5 * allowed to break compatibility for these functions since the interface they
6 * are exposed through to BPF programs is explicitly unstable.
9 #include <linux/bpf_verifier.h>
10 #include <linux/bpf.h>
11 #include <linux/btf.h>
12 #include <linux/filter.h>
13 #include <linux/mutex.h>
14 #include <linux/types.h>
15 #include <linux/btf_ids.h>
16 #include <linux/net_namespace.h>
17 #include <net/netfilter/nf_conntrack_bpf.h>
18 #include <net/netfilter/nf_conntrack_core.h>
20 /* bpf_ct_opts - Options for CT lookup helpers
23 * @netns_id - Specify the network namespace for lookup
25 * BPF_F_CURRENT_NETNS (-1)
26 * Use namespace associated with ctx (xdp_md, __sk_buff)
28 * Network Namespace ID
29 * @error - Out parameter, set for any errors encountered
31 * -EINVAL - Passed NULL for bpf_tuple pointer
32 * -EINVAL - opts->reserved is not 0
33 * -EINVAL - netns_id is less than -1
34 * -EINVAL - opts__sz isn't NF_BPF_CT_OPTS_SZ (12)
35 * -EPROTO - l4proto isn't one of IPPROTO_TCP or IPPROTO_UDP
36 * -ENONET - No network namespace found for netns_id
37 * -ENOENT - Conntrack lookup could not find entry for tuple
38 * -EAFNOSUPPORT - tuple__sz isn't one of sizeof(tuple->ipv4)
39 * or sizeof(tuple->ipv6)
40 * @l4proto - Layer 4 protocol
42 * IPPROTO_TCP, IPPROTO_UDP
43 * @dir: - connection tracking tuple direction.
44 * @reserved - Reserved member, will be reused for more options in future
57 NF_BPF_CT_OPTS_SZ = 12,
60 static int bpf_nf_ct_tuple_parse(struct bpf_sock_tuple *bpf_tuple,
61 u32 tuple_len, u8 protonum, u8 dir,
62 struct nf_conntrack_tuple *tuple)
64 union nf_inet_addr *src = dir ? &tuple->dst.u3 : &tuple->src.u3;
65 union nf_inet_addr *dst = dir ? &tuple->src.u3 : &tuple->dst.u3;
66 union nf_conntrack_man_proto *sport = dir ? (void *)&tuple->dst.u
68 union nf_conntrack_man_proto *dport = dir ? &tuple->src.u
69 : (void *)&tuple->dst.u;
71 if (unlikely(protonum != IPPROTO_TCP && protonum != IPPROTO_UDP))
74 memset(tuple, 0, sizeof(*tuple));
77 case sizeof(bpf_tuple->ipv4):
78 tuple->src.l3num = AF_INET;
79 src->ip = bpf_tuple->ipv4.saddr;
80 sport->tcp.port = bpf_tuple->ipv4.sport;
81 dst->ip = bpf_tuple->ipv4.daddr;
82 dport->tcp.port = bpf_tuple->ipv4.dport;
84 case sizeof(bpf_tuple->ipv6):
85 tuple->src.l3num = AF_INET6;
86 memcpy(src->ip6, bpf_tuple->ipv6.saddr, sizeof(bpf_tuple->ipv6.saddr));
87 sport->tcp.port = bpf_tuple->ipv6.sport;
88 memcpy(dst->ip6, bpf_tuple->ipv6.daddr, sizeof(bpf_tuple->ipv6.daddr));
89 dport->tcp.port = bpf_tuple->ipv6.dport;
94 tuple->dst.protonum = protonum;
100 static struct nf_conn *
101 __bpf_nf_ct_alloc_entry(struct net *net, struct bpf_sock_tuple *bpf_tuple,
102 u32 tuple_len, struct bpf_ct_opts *opts, u32 opts_len,
105 struct nf_conntrack_tuple otuple, rtuple;
109 if (!opts || !bpf_tuple || opts->reserved[0] || opts->reserved[1] ||
110 opts_len != NF_BPF_CT_OPTS_SZ)
111 return ERR_PTR(-EINVAL);
113 if (unlikely(opts->netns_id < BPF_F_CURRENT_NETNS))
114 return ERR_PTR(-EINVAL);
116 err = bpf_nf_ct_tuple_parse(bpf_tuple, tuple_len, opts->l4proto,
117 IP_CT_DIR_ORIGINAL, &otuple);
121 err = bpf_nf_ct_tuple_parse(bpf_tuple, tuple_len, opts->l4proto,
122 IP_CT_DIR_REPLY, &rtuple);
126 if (opts->netns_id >= 0) {
127 net = get_net_ns_by_id(net, opts->netns_id);
129 return ERR_PTR(-ENONET);
132 ct = nf_conntrack_alloc(net, &nf_ct_zone_dflt, &otuple, &rtuple,
137 memset(&ct->proto, 0, sizeof(ct->proto));
138 __nf_ct_set_timeout(ct, timeout * HZ);
141 if (opts->netns_id >= 0)
147 static struct nf_conn *__bpf_nf_ct_lookup(struct net *net,
148 struct bpf_sock_tuple *bpf_tuple,
149 u32 tuple_len, struct bpf_ct_opts *opts,
152 struct nf_conntrack_tuple_hash *hash;
153 struct nf_conntrack_tuple tuple;
157 if (!opts || !bpf_tuple || opts->reserved[0] || opts->reserved[1] ||
158 opts_len != NF_BPF_CT_OPTS_SZ)
159 return ERR_PTR(-EINVAL);
160 if (unlikely(opts->l4proto != IPPROTO_TCP && opts->l4proto != IPPROTO_UDP))
161 return ERR_PTR(-EPROTO);
162 if (unlikely(opts->netns_id < BPF_F_CURRENT_NETNS))
163 return ERR_PTR(-EINVAL);
165 err = bpf_nf_ct_tuple_parse(bpf_tuple, tuple_len, opts->l4proto,
166 IP_CT_DIR_ORIGINAL, &tuple);
170 if (opts->netns_id >= 0) {
171 net = get_net_ns_by_id(net, opts->netns_id);
173 return ERR_PTR(-ENONET);
176 hash = nf_conntrack_find_get(net, &nf_ct_zone_dflt, &tuple);
177 if (opts->netns_id >= 0)
180 return ERR_PTR(-ENOENT);
182 ct = nf_ct_tuplehash_to_ctrack(hash);
183 opts->dir = NF_CT_DIRECTION(hash);
188 BTF_ID_LIST(btf_nf_conn_ids)
189 BTF_ID(struct, nf_conn)
190 BTF_ID(struct, nf_conn___init)
192 /* Check writes into `struct nf_conn` */
193 static int _nf_conntrack_btf_struct_access(struct bpf_verifier_log *log,
194 const struct btf *btf,
195 const struct btf_type *t, int off,
196 int size, enum bpf_access_type atype,
198 enum bpf_type_flag *flag)
200 const struct btf_type *ncit;
201 const struct btf_type *nct;
204 ncit = btf_type_by_id(btf, btf_nf_conn_ids[1]);
205 nct = btf_type_by_id(btf, btf_nf_conn_ids[0]);
207 if (t != nct && t != ncit) {
208 bpf_log(log, "only read is supported\n");
212 /* `struct nf_conn` and `struct nf_conn___init` have the same layout
213 * so we are safe to simply merge offset checks here
216 #if defined(CONFIG_NF_CONNTRACK_MARK)
217 case offsetof(struct nf_conn, mark):
218 end = offsetofend(struct nf_conn, mark);
222 bpf_log(log, "no write support to nf_conn at off %d\n", off);
226 if (off + size > end) {
228 "write access at off %d with size %d beyond the member of nf_conn ended at %zu\n",
237 __diag_ignore_all("-Wmissing-prototypes",
238 "Global functions as their definitions will be in nf_conntrack BTF");
240 /* bpf_xdp_ct_alloc - Allocate a new CT entry
243 * @xdp_ctx - Pointer to ctx (xdp_md) in XDP program
245 * @bpf_tuple - Pointer to memory representing the tuple to look up
247 * @tuple__sz - Length of the tuple structure
248 * Must be one of sizeof(bpf_tuple->ipv4) or
249 * sizeof(bpf_tuple->ipv6)
250 * @opts - Additional options for allocation (documented above)
252 * @opts__sz - Length of the bpf_ct_opts structure
253 * Must be NF_BPF_CT_OPTS_SZ (12)
255 struct nf_conn___init *
256 bpf_xdp_ct_alloc(struct xdp_md *xdp_ctx, struct bpf_sock_tuple *bpf_tuple,
257 u32 tuple__sz, struct bpf_ct_opts *opts, u32 opts__sz)
259 struct xdp_buff *ctx = (struct xdp_buff *)xdp_ctx;
260 struct nf_conn *nfct;
262 nfct = __bpf_nf_ct_alloc_entry(dev_net(ctx->rxq->dev), bpf_tuple, tuple__sz,
266 opts->error = PTR_ERR(nfct);
270 return (struct nf_conn___init *)nfct;
273 /* bpf_xdp_ct_lookup - Lookup CT entry for the given tuple, and acquire a
277 * @xdp_ctx - Pointer to ctx (xdp_md) in XDP program
279 * @bpf_tuple - Pointer to memory representing the tuple to look up
281 * @tuple__sz - Length of the tuple structure
282 * Must be one of sizeof(bpf_tuple->ipv4) or
283 * sizeof(bpf_tuple->ipv6)
284 * @opts - Additional options for lookup (documented above)
286 * @opts__sz - Length of the bpf_ct_opts structure
287 * Must be NF_BPF_CT_OPTS_SZ (12)
290 bpf_xdp_ct_lookup(struct xdp_md *xdp_ctx, struct bpf_sock_tuple *bpf_tuple,
291 u32 tuple__sz, struct bpf_ct_opts *opts, u32 opts__sz)
293 struct xdp_buff *ctx = (struct xdp_buff *)xdp_ctx;
294 struct net *caller_net;
295 struct nf_conn *nfct;
297 caller_net = dev_net(ctx->rxq->dev);
298 nfct = __bpf_nf_ct_lookup(caller_net, bpf_tuple, tuple__sz, opts, opts__sz);
301 opts->error = PTR_ERR(nfct);
307 /* bpf_skb_ct_alloc - Allocate a new CT entry
310 * @skb_ctx - Pointer to ctx (__sk_buff) in TC program
312 * @bpf_tuple - Pointer to memory representing the tuple to look up
314 * @tuple__sz - Length of the tuple structure
315 * Must be one of sizeof(bpf_tuple->ipv4) or
316 * sizeof(bpf_tuple->ipv6)
317 * @opts - Additional options for allocation (documented above)
319 * @opts__sz - Length of the bpf_ct_opts structure
320 * Must be NF_BPF_CT_OPTS_SZ (12)
322 struct nf_conn___init *
323 bpf_skb_ct_alloc(struct __sk_buff *skb_ctx, struct bpf_sock_tuple *bpf_tuple,
324 u32 tuple__sz, struct bpf_ct_opts *opts, u32 opts__sz)
326 struct sk_buff *skb = (struct sk_buff *)skb_ctx;
327 struct nf_conn *nfct;
330 net = skb->dev ? dev_net(skb->dev) : sock_net(skb->sk);
331 nfct = __bpf_nf_ct_alloc_entry(net, bpf_tuple, tuple__sz, opts, opts__sz, 10);
334 opts->error = PTR_ERR(nfct);
338 return (struct nf_conn___init *)nfct;
341 /* bpf_skb_ct_lookup - Lookup CT entry for the given tuple, and acquire a
345 * @skb_ctx - Pointer to ctx (__sk_buff) in TC program
347 * @bpf_tuple - Pointer to memory representing the tuple to look up
349 * @tuple__sz - Length of the tuple structure
350 * Must be one of sizeof(bpf_tuple->ipv4) or
351 * sizeof(bpf_tuple->ipv6)
352 * @opts - Additional options for lookup (documented above)
354 * @opts__sz - Length of the bpf_ct_opts structure
355 * Must be NF_BPF_CT_OPTS_SZ (12)
358 bpf_skb_ct_lookup(struct __sk_buff *skb_ctx, struct bpf_sock_tuple *bpf_tuple,
359 u32 tuple__sz, struct bpf_ct_opts *opts, u32 opts__sz)
361 struct sk_buff *skb = (struct sk_buff *)skb_ctx;
362 struct net *caller_net;
363 struct nf_conn *nfct;
365 caller_net = skb->dev ? dev_net(skb->dev) : sock_net(skb->sk);
366 nfct = __bpf_nf_ct_lookup(caller_net, bpf_tuple, tuple__sz, opts, opts__sz);
369 opts->error = PTR_ERR(nfct);
375 /* bpf_ct_insert_entry - Add the provided entry into a CT map
377 * This must be invoked for referenced PTR_TO_BTF_ID.
379 * @nfct - Pointer to referenced nf_conn___init object, obtained
380 * using bpf_xdp_ct_alloc or bpf_skb_ct_alloc.
382 struct nf_conn *bpf_ct_insert_entry(struct nf_conn___init *nfct_i)
384 struct nf_conn *nfct = (struct nf_conn *)nfct_i;
387 if (!nf_ct_is_confirmed(nfct))
388 nfct->timeout += nfct_time_stamp;
389 nfct->status |= IPS_CONFIRMED;
390 err = nf_conntrack_hash_check_insert(nfct);
392 nf_conntrack_free(nfct);
398 /* bpf_ct_release - Release acquired nf_conn object
400 * This must be invoked for referenced PTR_TO_BTF_ID, and the verifier rejects
401 * the program if any references remain in the program in all of the explored
405 * @nf_conn - Pointer to referenced nf_conn object, obtained using
406 * bpf_xdp_ct_lookup or bpf_skb_ct_lookup.
408 void bpf_ct_release(struct nf_conn *nfct)
415 /* bpf_ct_set_timeout - Set timeout of allocated nf_conn
417 * Sets the default timeout of newly allocated nf_conn before insertion.
418 * This helper must be invoked for refcounted pointer to nf_conn___init.
421 * @nfct - Pointer to referenced nf_conn object, obtained using
422 * bpf_xdp_ct_alloc or bpf_skb_ct_alloc.
423 * @timeout - Timeout in msecs.
425 void bpf_ct_set_timeout(struct nf_conn___init *nfct, u32 timeout)
427 __nf_ct_set_timeout((struct nf_conn *)nfct, msecs_to_jiffies(timeout));
430 /* bpf_ct_change_timeout - Change timeout of inserted nf_conn
432 * Change timeout associated of the inserted or looked up nf_conn.
433 * This helper must be invoked for refcounted pointer to nf_conn.
436 * @nfct - Pointer to referenced nf_conn object, obtained using
437 * bpf_ct_insert_entry, bpf_xdp_ct_lookup, or bpf_skb_ct_lookup.
438 * @timeout - New timeout in msecs.
440 int bpf_ct_change_timeout(struct nf_conn *nfct, u32 timeout)
442 return __nf_ct_change_timeout(nfct, msecs_to_jiffies(timeout));
445 /* bpf_ct_set_status - Set status field of allocated nf_conn
447 * Set the status field of the newly allocated nf_conn before insertion.
448 * This must be invoked for referenced PTR_TO_BTF_ID to nf_conn___init.
451 * @nfct - Pointer to referenced nf_conn object, obtained using
452 * bpf_xdp_ct_alloc or bpf_skb_ct_alloc.
453 * @status - New status value.
455 int bpf_ct_set_status(const struct nf_conn___init *nfct, u32 status)
457 return nf_ct_change_status_common((struct nf_conn *)nfct, status);
460 /* bpf_ct_change_status - Change status of inserted nf_conn
462 * Change the status field of the provided connection tracking entry.
463 * This must be invoked for referenced PTR_TO_BTF_ID to nf_conn.
466 * @nfct - Pointer to referenced nf_conn object, obtained using
467 * bpf_ct_insert_entry, bpf_xdp_ct_lookup or bpf_skb_ct_lookup.
468 * @status - New status value.
470 int bpf_ct_change_status(struct nf_conn *nfct, u32 status)
472 return nf_ct_change_status_common(nfct, status);
477 BTF_SET8_START(nf_ct_kfunc_set)
478 BTF_ID_FLAGS(func, bpf_xdp_ct_alloc, KF_ACQUIRE | KF_RET_NULL)
479 BTF_ID_FLAGS(func, bpf_xdp_ct_lookup, KF_ACQUIRE | KF_RET_NULL)
480 BTF_ID_FLAGS(func, bpf_skb_ct_alloc, KF_ACQUIRE | KF_RET_NULL)
481 BTF_ID_FLAGS(func, bpf_skb_ct_lookup, KF_ACQUIRE | KF_RET_NULL)
482 BTF_ID_FLAGS(func, bpf_ct_insert_entry, KF_ACQUIRE | KF_RET_NULL | KF_RELEASE)
483 BTF_ID_FLAGS(func, bpf_ct_release, KF_RELEASE)
484 BTF_ID_FLAGS(func, bpf_ct_set_timeout, KF_TRUSTED_ARGS)
485 BTF_ID_FLAGS(func, bpf_ct_change_timeout, KF_TRUSTED_ARGS)
486 BTF_ID_FLAGS(func, bpf_ct_set_status, KF_TRUSTED_ARGS)
487 BTF_ID_FLAGS(func, bpf_ct_change_status, KF_TRUSTED_ARGS)
488 BTF_SET8_END(nf_ct_kfunc_set)
490 static const struct btf_kfunc_id_set nf_conntrack_kfunc_set = {
491 .owner = THIS_MODULE,
492 .set = &nf_ct_kfunc_set,
495 int register_nf_conntrack_bpf(void)
499 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &nf_conntrack_kfunc_set);
500 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &nf_conntrack_kfunc_set);
502 mutex_lock(&nf_conn_btf_access_lock);
503 nfct_btf_struct_access = _nf_conntrack_btf_struct_access;
504 mutex_unlock(&nf_conn_btf_access_lock);
510 void cleanup_nf_conntrack_bpf(void)
512 mutex_lock(&nf_conn_btf_access_lock);
513 nfct_btf_struct_access = NULL;
514 mutex_unlock(&nf_conn_btf_access_lock);