1 // SPDX-License-Identifier: GPL-2.0-only 2 /* Unstable Conntrack Helpers for XDP and TC-BPF hook 3 * 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. 7 */ 8 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> 19 20 /* bpf_ct_opts - Options for CT lookup helpers 21 * 22 * Members: 23 * @netns_id - Specify the network namespace for lookup 24 * Values: 25 * BPF_F_CURRENT_NETNS (-1) 26 * Use namespace associated with ctx (xdp_md, __sk_buff) 27 * [0, S32_MAX] 28 * Network Namespace ID 29 * @error - Out parameter, set for any errors encountered 30 * Values: 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 41 * Values: 42 * IPPROTO_TCP, IPPROTO_UDP 43 * @dir: - connection tracking tuple direction. 44 * @reserved - Reserved member, will be reused for more options in future 45 * Values: 46 * 0 47 */ 48 struct bpf_ct_opts { 49 s32 netns_id; 50 s32 error; 51 u8 l4proto; 52 u8 dir; 53 u8 reserved[2]; 54 }; 55 56 enum { 57 NF_BPF_CT_OPTS_SZ = 12, 58 }; 59 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) 63 { 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 67 : &tuple->src.u; 68 union nf_conntrack_man_proto *dport = dir ? &tuple->src.u 69 : (void *)&tuple->dst.u; 70 71 if (unlikely(protonum != IPPROTO_TCP && protonum != IPPROTO_UDP)) 72 return -EPROTO; 73 74 memset(tuple, 0, sizeof(*tuple)); 75 76 switch (tuple_len) { 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; 83 break; 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; 90 break; 91 default: 92 return -EAFNOSUPPORT; 93 } 94 tuple->dst.protonum = protonum; 95 tuple->dst.dir = dir; 96 97 return 0; 98 } 99 100 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, 103 u32 timeout) 104 { 105 struct nf_conntrack_tuple otuple, rtuple; 106 struct nf_conn *ct; 107 int err; 108 109 if (!opts || !bpf_tuple || opts->reserved[0] || opts->reserved[1] || 110 opts_len != NF_BPF_CT_OPTS_SZ) 111 return ERR_PTR(-EINVAL); 112 113 if (unlikely(opts->netns_id < BPF_F_CURRENT_NETNS)) 114 return ERR_PTR(-EINVAL); 115 116 err = bpf_nf_ct_tuple_parse(bpf_tuple, tuple_len, opts->l4proto, 117 IP_CT_DIR_ORIGINAL, &otuple); 118 if (err < 0) 119 return ERR_PTR(err); 120 121 err = bpf_nf_ct_tuple_parse(bpf_tuple, tuple_len, opts->l4proto, 122 IP_CT_DIR_REPLY, &rtuple); 123 if (err < 0) 124 return ERR_PTR(err); 125 126 if (opts->netns_id >= 0) { 127 net = get_net_ns_by_id(net, opts->netns_id); 128 if (unlikely(!net)) 129 return ERR_PTR(-ENONET); 130 } 131 132 ct = nf_conntrack_alloc(net, &nf_ct_zone_dflt, &otuple, &rtuple, 133 GFP_ATOMIC); 134 if (IS_ERR(ct)) 135 goto out; 136 137 memset(&ct->proto, 0, sizeof(ct->proto)); 138 __nf_ct_set_timeout(ct, timeout * HZ); 139 140 out: 141 if (opts->netns_id >= 0) 142 put_net(net); 143 144 return ct; 145 } 146 147 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, 150 u32 opts_len) 151 { 152 struct nf_conntrack_tuple_hash *hash; 153 struct nf_conntrack_tuple tuple; 154 struct nf_conn *ct; 155 int err; 156 157 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); 164 165 err = bpf_nf_ct_tuple_parse(bpf_tuple, tuple_len, opts->l4proto, 166 IP_CT_DIR_ORIGINAL, &tuple); 167 if (err < 0) 168 return ERR_PTR(err); 169 170 if (opts->netns_id >= 0) { 171 net = get_net_ns_by_id(net, opts->netns_id); 172 if (unlikely(!net)) 173 return ERR_PTR(-ENONET); 174 } 175 176 hash = nf_conntrack_find_get(net, &nf_ct_zone_dflt, &tuple); 177 if (opts->netns_id >= 0) 178 put_net(net); 179 if (!hash) 180 return ERR_PTR(-ENOENT); 181 182 ct = nf_ct_tuplehash_to_ctrack(hash); 183 opts->dir = NF_CT_DIRECTION(hash); 184 185 return ct; 186 } 187 188 BTF_ID_LIST(btf_nf_conn_ids) 189 BTF_ID(struct, nf_conn) 190 BTF_ID(struct, nf_conn___init) 191 192 /* Check writes into `struct nf_conn` */ 193 static int _nf_conntrack_btf_struct_access(struct bpf_verifier_log *log, 194 const struct bpf_reg_state *reg, 195 int off, int size, enum bpf_access_type atype, 196 u32 *next_btf_id, enum bpf_type_flag *flag) 197 { 198 const struct btf_type *ncit, *nct, *t; 199 size_t end; 200 201 ncit = btf_type_by_id(reg->btf, btf_nf_conn_ids[1]); 202 nct = btf_type_by_id(reg->btf, btf_nf_conn_ids[0]); 203 t = btf_type_by_id(reg->btf, reg->btf_id); 204 if (t != nct && t != ncit) { 205 bpf_log(log, "only read is supported\n"); 206 return -EACCES; 207 } 208 209 /* `struct nf_conn` and `struct nf_conn___init` have the same layout 210 * so we are safe to simply merge offset checks here 211 */ 212 switch (off) { 213 #if defined(CONFIG_NF_CONNTRACK_MARK) 214 case offsetof(struct nf_conn, mark): 215 end = offsetofend(struct nf_conn, mark); 216 break; 217 #endif 218 default: 219 bpf_log(log, "no write support to nf_conn at off %d\n", off); 220 return -EACCES; 221 } 222 223 if (off + size > end) { 224 bpf_log(log, 225 "write access at off %d with size %d beyond the member of nf_conn ended at %zu\n", 226 off, size, end); 227 return -EACCES; 228 } 229 230 return 0; 231 } 232 233 __diag_push(); 234 __diag_ignore_all("-Wmissing-prototypes", 235 "Global functions as their definitions will be in nf_conntrack BTF"); 236 237 /* bpf_xdp_ct_alloc - Allocate a new CT entry 238 * 239 * Parameters: 240 * @xdp_ctx - Pointer to ctx (xdp_md) in XDP program 241 * Cannot be NULL 242 * @bpf_tuple - Pointer to memory representing the tuple to look up 243 * Cannot be NULL 244 * @tuple__sz - Length of the tuple structure 245 * Must be one of sizeof(bpf_tuple->ipv4) or 246 * sizeof(bpf_tuple->ipv6) 247 * @opts - Additional options for allocation (documented above) 248 * Cannot be NULL 249 * @opts__sz - Length of the bpf_ct_opts structure 250 * Must be NF_BPF_CT_OPTS_SZ (12) 251 */ 252 struct nf_conn___init * 253 bpf_xdp_ct_alloc(struct xdp_md *xdp_ctx, struct bpf_sock_tuple *bpf_tuple, 254 u32 tuple__sz, struct bpf_ct_opts *opts, u32 opts__sz) 255 { 256 struct xdp_buff *ctx = (struct xdp_buff *)xdp_ctx; 257 struct nf_conn *nfct; 258 259 nfct = __bpf_nf_ct_alloc_entry(dev_net(ctx->rxq->dev), bpf_tuple, tuple__sz, 260 opts, opts__sz, 10); 261 if (IS_ERR(nfct)) { 262 if (opts) 263 opts->error = PTR_ERR(nfct); 264 return NULL; 265 } 266 267 return (struct nf_conn___init *)nfct; 268 } 269 270 /* bpf_xdp_ct_lookup - Lookup CT entry for the given tuple, and acquire a 271 * reference to it 272 * 273 * Parameters: 274 * @xdp_ctx - Pointer to ctx (xdp_md) in XDP program 275 * Cannot be NULL 276 * @bpf_tuple - Pointer to memory representing the tuple to look up 277 * Cannot be NULL 278 * @tuple__sz - Length of the tuple structure 279 * Must be one of sizeof(bpf_tuple->ipv4) or 280 * sizeof(bpf_tuple->ipv6) 281 * @opts - Additional options for lookup (documented above) 282 * Cannot be NULL 283 * @opts__sz - Length of the bpf_ct_opts structure 284 * Must be NF_BPF_CT_OPTS_SZ (12) 285 */ 286 struct nf_conn * 287 bpf_xdp_ct_lookup(struct xdp_md *xdp_ctx, struct bpf_sock_tuple *bpf_tuple, 288 u32 tuple__sz, struct bpf_ct_opts *opts, u32 opts__sz) 289 { 290 struct xdp_buff *ctx = (struct xdp_buff *)xdp_ctx; 291 struct net *caller_net; 292 struct nf_conn *nfct; 293 294 caller_net = dev_net(ctx->rxq->dev); 295 nfct = __bpf_nf_ct_lookup(caller_net, bpf_tuple, tuple__sz, opts, opts__sz); 296 if (IS_ERR(nfct)) { 297 if (opts) 298 opts->error = PTR_ERR(nfct); 299 return NULL; 300 } 301 return nfct; 302 } 303 304 /* bpf_skb_ct_alloc - Allocate a new CT entry 305 * 306 * Parameters: 307 * @skb_ctx - Pointer to ctx (__sk_buff) in TC program 308 * Cannot be NULL 309 * @bpf_tuple - Pointer to memory representing the tuple to look up 310 * Cannot be NULL 311 * @tuple__sz - Length of the tuple structure 312 * Must be one of sizeof(bpf_tuple->ipv4) or 313 * sizeof(bpf_tuple->ipv6) 314 * @opts - Additional options for allocation (documented above) 315 * Cannot be NULL 316 * @opts__sz - Length of the bpf_ct_opts structure 317 * Must be NF_BPF_CT_OPTS_SZ (12) 318 */ 319 struct nf_conn___init * 320 bpf_skb_ct_alloc(struct __sk_buff *skb_ctx, struct bpf_sock_tuple *bpf_tuple, 321 u32 tuple__sz, struct bpf_ct_opts *opts, u32 opts__sz) 322 { 323 struct sk_buff *skb = (struct sk_buff *)skb_ctx; 324 struct nf_conn *nfct; 325 struct net *net; 326 327 net = skb->dev ? dev_net(skb->dev) : sock_net(skb->sk); 328 nfct = __bpf_nf_ct_alloc_entry(net, bpf_tuple, tuple__sz, opts, opts__sz, 10); 329 if (IS_ERR(nfct)) { 330 if (opts) 331 opts->error = PTR_ERR(nfct); 332 return NULL; 333 } 334 335 return (struct nf_conn___init *)nfct; 336 } 337 338 /* bpf_skb_ct_lookup - Lookup CT entry for the given tuple, and acquire a 339 * reference to it 340 * 341 * Parameters: 342 * @skb_ctx - Pointer to ctx (__sk_buff) in TC program 343 * Cannot be NULL 344 * @bpf_tuple - Pointer to memory representing the tuple to look up 345 * Cannot be NULL 346 * @tuple__sz - Length of the tuple structure 347 * Must be one of sizeof(bpf_tuple->ipv4) or 348 * sizeof(bpf_tuple->ipv6) 349 * @opts - Additional options for lookup (documented above) 350 * Cannot be NULL 351 * @opts__sz - Length of the bpf_ct_opts structure 352 * Must be NF_BPF_CT_OPTS_SZ (12) 353 */ 354 struct nf_conn * 355 bpf_skb_ct_lookup(struct __sk_buff *skb_ctx, struct bpf_sock_tuple *bpf_tuple, 356 u32 tuple__sz, struct bpf_ct_opts *opts, u32 opts__sz) 357 { 358 struct sk_buff *skb = (struct sk_buff *)skb_ctx; 359 struct net *caller_net; 360 struct nf_conn *nfct; 361 362 caller_net = skb->dev ? dev_net(skb->dev) : sock_net(skb->sk); 363 nfct = __bpf_nf_ct_lookup(caller_net, bpf_tuple, tuple__sz, opts, opts__sz); 364 if (IS_ERR(nfct)) { 365 if (opts) 366 opts->error = PTR_ERR(nfct); 367 return NULL; 368 } 369 return nfct; 370 } 371 372 /* bpf_ct_insert_entry - Add the provided entry into a CT map 373 * 374 * This must be invoked for referenced PTR_TO_BTF_ID. 375 * 376 * @nfct - Pointer to referenced nf_conn___init object, obtained 377 * using bpf_xdp_ct_alloc or bpf_skb_ct_alloc. 378 */ 379 struct nf_conn *bpf_ct_insert_entry(struct nf_conn___init *nfct_i) 380 { 381 struct nf_conn *nfct = (struct nf_conn *)nfct_i; 382 int err; 383 384 nfct->status |= IPS_CONFIRMED; 385 err = nf_conntrack_hash_check_insert(nfct); 386 if (err < 0) { 387 nf_conntrack_free(nfct); 388 return NULL; 389 } 390 return nfct; 391 } 392 393 /* bpf_ct_release - Release acquired nf_conn object 394 * 395 * This must be invoked for referenced PTR_TO_BTF_ID, and the verifier rejects 396 * the program if any references remain in the program in all of the explored 397 * states. 398 * 399 * Parameters: 400 * @nf_conn - Pointer to referenced nf_conn object, obtained using 401 * bpf_xdp_ct_lookup or bpf_skb_ct_lookup. 402 */ 403 void bpf_ct_release(struct nf_conn *nfct) 404 { 405 if (!nfct) 406 return; 407 nf_ct_put(nfct); 408 } 409 410 /* bpf_ct_set_timeout - Set timeout of allocated nf_conn 411 * 412 * Sets the default timeout of newly allocated nf_conn before insertion. 413 * This helper must be invoked for refcounted pointer to nf_conn___init. 414 * 415 * Parameters: 416 * @nfct - Pointer to referenced nf_conn object, obtained using 417 * bpf_xdp_ct_alloc or bpf_skb_ct_alloc. 418 * @timeout - Timeout in msecs. 419 */ 420 void bpf_ct_set_timeout(struct nf_conn___init *nfct, u32 timeout) 421 { 422 __nf_ct_set_timeout((struct nf_conn *)nfct, msecs_to_jiffies(timeout)); 423 } 424 425 /* bpf_ct_change_timeout - Change timeout of inserted nf_conn 426 * 427 * Change timeout associated of the inserted or looked up nf_conn. 428 * This helper must be invoked for refcounted pointer to nf_conn. 429 * 430 * Parameters: 431 * @nfct - Pointer to referenced nf_conn object, obtained using 432 * bpf_ct_insert_entry, bpf_xdp_ct_lookup, or bpf_skb_ct_lookup. 433 * @timeout - New timeout in msecs. 434 */ 435 int bpf_ct_change_timeout(struct nf_conn *nfct, u32 timeout) 436 { 437 return __nf_ct_change_timeout(nfct, msecs_to_jiffies(timeout)); 438 } 439 440 /* bpf_ct_set_status - Set status field of allocated nf_conn 441 * 442 * Set the status field of the newly allocated nf_conn before insertion. 443 * This must be invoked for referenced PTR_TO_BTF_ID to nf_conn___init. 444 * 445 * Parameters: 446 * @nfct - Pointer to referenced nf_conn object, obtained using 447 * bpf_xdp_ct_alloc or bpf_skb_ct_alloc. 448 * @status - New status value. 449 */ 450 int bpf_ct_set_status(const struct nf_conn___init *nfct, u32 status) 451 { 452 return nf_ct_change_status_common((struct nf_conn *)nfct, status); 453 } 454 455 /* bpf_ct_change_status - Change status of inserted nf_conn 456 * 457 * Change the status field of the provided connection tracking entry. 458 * This must be invoked for referenced PTR_TO_BTF_ID to nf_conn. 459 * 460 * Parameters: 461 * @nfct - Pointer to referenced nf_conn object, obtained using 462 * bpf_ct_insert_entry, bpf_xdp_ct_lookup or bpf_skb_ct_lookup. 463 * @status - New status value. 464 */ 465 int bpf_ct_change_status(struct nf_conn *nfct, u32 status) 466 { 467 return nf_ct_change_status_common(nfct, status); 468 } 469 470 __diag_pop() 471 472 BTF_SET8_START(nf_ct_kfunc_set) 473 BTF_ID_FLAGS(func, bpf_xdp_ct_alloc, KF_ACQUIRE | KF_RET_NULL) 474 BTF_ID_FLAGS(func, bpf_xdp_ct_lookup, KF_ACQUIRE | KF_RET_NULL) 475 BTF_ID_FLAGS(func, bpf_skb_ct_alloc, KF_ACQUIRE | KF_RET_NULL) 476 BTF_ID_FLAGS(func, bpf_skb_ct_lookup, KF_ACQUIRE | KF_RET_NULL) 477 BTF_ID_FLAGS(func, bpf_ct_insert_entry, KF_ACQUIRE | KF_RET_NULL | KF_RELEASE) 478 BTF_ID_FLAGS(func, bpf_ct_release, KF_RELEASE) 479 BTF_ID_FLAGS(func, bpf_ct_set_timeout, KF_TRUSTED_ARGS) 480 BTF_ID_FLAGS(func, bpf_ct_change_timeout, KF_TRUSTED_ARGS) 481 BTF_ID_FLAGS(func, bpf_ct_set_status, KF_TRUSTED_ARGS) 482 BTF_ID_FLAGS(func, bpf_ct_change_status, KF_TRUSTED_ARGS) 483 BTF_SET8_END(nf_ct_kfunc_set) 484 485 static const struct btf_kfunc_id_set nf_conntrack_kfunc_set = { 486 .owner = THIS_MODULE, 487 .set = &nf_ct_kfunc_set, 488 }; 489 490 int register_nf_conntrack_bpf(void) 491 { 492 int ret; 493 494 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &nf_conntrack_kfunc_set); 495 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &nf_conntrack_kfunc_set); 496 if (!ret) { 497 mutex_lock(&nf_conn_btf_access_lock); 498 nfct_btf_struct_access = _nf_conntrack_btf_struct_access; 499 mutex_unlock(&nf_conn_btf_access_lock); 500 } 501 502 return ret; 503 } 504 505 void cleanup_nf_conntrack_bpf(void) 506 { 507 mutex_lock(&nf_conn_btf_access_lock); 508 nfct_btf_struct_access = NULL; 509 mutex_unlock(&nf_conn_btf_access_lock); 510 } 511