1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright (c) 2017 - 2018 Covalent IO, Inc. http://covalent.io */ 3 4 #include <linux/bpf.h> 5 #include <linux/btf_ids.h> 6 #include <linux/filter.h> 7 #include <linux/errno.h> 8 #include <linux/file.h> 9 #include <linux/net.h> 10 #include <linux/workqueue.h> 11 #include <linux/skmsg.h> 12 #include <linux/list.h> 13 #include <linux/jhash.h> 14 #include <linux/sock_diag.h> 15 #include <net/udp.h> 16 17 struct bpf_stab { 18 struct bpf_map map; 19 struct sock **sks; 20 struct sk_psock_progs progs; 21 raw_spinlock_t lock; 22 }; 23 24 #define SOCK_CREATE_FLAG_MASK \ 25 (BPF_F_NUMA_NODE | BPF_F_RDONLY | BPF_F_WRONLY) 26 27 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog, 28 struct bpf_prog *old, u32 which); 29 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map); 30 31 static struct bpf_map *sock_map_alloc(union bpf_attr *attr) 32 { 33 struct bpf_stab *stab; 34 35 if (!capable(CAP_NET_ADMIN)) 36 return ERR_PTR(-EPERM); 37 if (attr->max_entries == 0 || 38 attr->key_size != 4 || 39 (attr->value_size != sizeof(u32) && 40 attr->value_size != sizeof(u64)) || 41 attr->map_flags & ~SOCK_CREATE_FLAG_MASK) 42 return ERR_PTR(-EINVAL); 43 44 stab = kzalloc(sizeof(*stab), GFP_USER | __GFP_ACCOUNT); 45 if (!stab) 46 return ERR_PTR(-ENOMEM); 47 48 bpf_map_init_from_attr(&stab->map, attr); 49 raw_spin_lock_init(&stab->lock); 50 51 stab->sks = bpf_map_area_alloc((u64) stab->map.max_entries * 52 sizeof(struct sock *), 53 stab->map.numa_node); 54 if (!stab->sks) { 55 kfree(stab); 56 return ERR_PTR(-ENOMEM); 57 } 58 59 return &stab->map; 60 } 61 62 int sock_map_get_from_fd(const union bpf_attr *attr, struct bpf_prog *prog) 63 { 64 u32 ufd = attr->target_fd; 65 struct bpf_map *map; 66 struct fd f; 67 int ret; 68 69 if (attr->attach_flags || attr->replace_bpf_fd) 70 return -EINVAL; 71 72 f = fdget(ufd); 73 map = __bpf_map_get(f); 74 if (IS_ERR(map)) 75 return PTR_ERR(map); 76 ret = sock_map_prog_update(map, prog, NULL, attr->attach_type); 77 fdput(f); 78 return ret; 79 } 80 81 int sock_map_prog_detach(const union bpf_attr *attr, enum bpf_prog_type ptype) 82 { 83 u32 ufd = attr->target_fd; 84 struct bpf_prog *prog; 85 struct bpf_map *map; 86 struct fd f; 87 int ret; 88 89 if (attr->attach_flags || attr->replace_bpf_fd) 90 return -EINVAL; 91 92 f = fdget(ufd); 93 map = __bpf_map_get(f); 94 if (IS_ERR(map)) 95 return PTR_ERR(map); 96 97 prog = bpf_prog_get(attr->attach_bpf_fd); 98 if (IS_ERR(prog)) { 99 ret = PTR_ERR(prog); 100 goto put_map; 101 } 102 103 if (prog->type != ptype) { 104 ret = -EINVAL; 105 goto put_prog; 106 } 107 108 ret = sock_map_prog_update(map, NULL, prog, attr->attach_type); 109 put_prog: 110 bpf_prog_put(prog); 111 put_map: 112 fdput(f); 113 return ret; 114 } 115 116 static void sock_map_sk_acquire(struct sock *sk) 117 __acquires(&sk->sk_lock.slock) 118 { 119 lock_sock(sk); 120 preempt_disable(); 121 rcu_read_lock(); 122 } 123 124 static void sock_map_sk_release(struct sock *sk) 125 __releases(&sk->sk_lock.slock) 126 { 127 rcu_read_unlock(); 128 preempt_enable(); 129 release_sock(sk); 130 } 131 132 static void sock_map_add_link(struct sk_psock *psock, 133 struct sk_psock_link *link, 134 struct bpf_map *map, void *link_raw) 135 { 136 link->link_raw = link_raw; 137 link->map = map; 138 spin_lock_bh(&psock->link_lock); 139 list_add_tail(&link->list, &psock->link); 140 spin_unlock_bh(&psock->link_lock); 141 } 142 143 static void sock_map_del_link(struct sock *sk, 144 struct sk_psock *psock, void *link_raw) 145 { 146 bool strp_stop = false, verdict_stop = false; 147 struct sk_psock_link *link, *tmp; 148 149 spin_lock_bh(&psock->link_lock); 150 list_for_each_entry_safe(link, tmp, &psock->link, list) { 151 if (link->link_raw == link_raw) { 152 struct bpf_map *map = link->map; 153 struct bpf_stab *stab = container_of(map, struct bpf_stab, 154 map); 155 if (psock->saved_data_ready && stab->progs.stream_parser) 156 strp_stop = true; 157 if (psock->saved_data_ready && stab->progs.stream_verdict) 158 verdict_stop = true; 159 if (psock->saved_data_ready && stab->progs.skb_verdict) 160 verdict_stop = true; 161 list_del(&link->list); 162 sk_psock_free_link(link); 163 } 164 } 165 spin_unlock_bh(&psock->link_lock); 166 if (strp_stop || verdict_stop) { 167 write_lock_bh(&sk->sk_callback_lock); 168 if (strp_stop) 169 sk_psock_stop_strp(sk, psock); 170 if (verdict_stop) 171 sk_psock_stop_verdict(sk, psock); 172 173 if (psock->psock_update_sk_prot) 174 psock->psock_update_sk_prot(sk, psock, false); 175 write_unlock_bh(&sk->sk_callback_lock); 176 } 177 } 178 179 static void sock_map_unref(struct sock *sk, void *link_raw) 180 { 181 struct sk_psock *psock = sk_psock(sk); 182 183 if (likely(psock)) { 184 sock_map_del_link(sk, psock, link_raw); 185 sk_psock_put(sk, psock); 186 } 187 } 188 189 static int sock_map_init_proto(struct sock *sk, struct sk_psock *psock) 190 { 191 if (!sk->sk_prot->psock_update_sk_prot) 192 return -EINVAL; 193 psock->psock_update_sk_prot = sk->sk_prot->psock_update_sk_prot; 194 return sk->sk_prot->psock_update_sk_prot(sk, psock, false); 195 } 196 197 static struct sk_psock *sock_map_psock_get_checked(struct sock *sk) 198 { 199 struct sk_psock *psock; 200 201 rcu_read_lock(); 202 psock = sk_psock(sk); 203 if (psock) { 204 if (sk->sk_prot->close != sock_map_close) { 205 psock = ERR_PTR(-EBUSY); 206 goto out; 207 } 208 209 if (!refcount_inc_not_zero(&psock->refcnt)) 210 psock = ERR_PTR(-EBUSY); 211 } 212 out: 213 rcu_read_unlock(); 214 return psock; 215 } 216 217 static int sock_map_link(struct bpf_map *map, struct sock *sk) 218 { 219 struct sk_psock_progs *progs = sock_map_progs(map); 220 struct bpf_prog *stream_verdict = NULL; 221 struct bpf_prog *stream_parser = NULL; 222 struct bpf_prog *skb_verdict = NULL; 223 struct bpf_prog *msg_parser = NULL; 224 struct sk_psock *psock; 225 int ret; 226 227 stream_verdict = READ_ONCE(progs->stream_verdict); 228 if (stream_verdict) { 229 stream_verdict = bpf_prog_inc_not_zero(stream_verdict); 230 if (IS_ERR(stream_verdict)) 231 return PTR_ERR(stream_verdict); 232 } 233 234 stream_parser = READ_ONCE(progs->stream_parser); 235 if (stream_parser) { 236 stream_parser = bpf_prog_inc_not_zero(stream_parser); 237 if (IS_ERR(stream_parser)) { 238 ret = PTR_ERR(stream_parser); 239 goto out_put_stream_verdict; 240 } 241 } 242 243 msg_parser = READ_ONCE(progs->msg_parser); 244 if (msg_parser) { 245 msg_parser = bpf_prog_inc_not_zero(msg_parser); 246 if (IS_ERR(msg_parser)) { 247 ret = PTR_ERR(msg_parser); 248 goto out_put_stream_parser; 249 } 250 } 251 252 skb_verdict = READ_ONCE(progs->skb_verdict); 253 if (skb_verdict) { 254 skb_verdict = bpf_prog_inc_not_zero(skb_verdict); 255 if (IS_ERR(skb_verdict)) { 256 ret = PTR_ERR(skb_verdict); 257 goto out_put_msg_parser; 258 } 259 } 260 261 psock = sock_map_psock_get_checked(sk); 262 if (IS_ERR(psock)) { 263 ret = PTR_ERR(psock); 264 goto out_progs; 265 } 266 267 if (psock) { 268 if ((msg_parser && READ_ONCE(psock->progs.msg_parser)) || 269 (stream_parser && READ_ONCE(psock->progs.stream_parser)) || 270 (skb_verdict && READ_ONCE(psock->progs.skb_verdict)) || 271 (skb_verdict && READ_ONCE(psock->progs.stream_verdict)) || 272 (stream_verdict && READ_ONCE(psock->progs.skb_verdict)) || 273 (stream_verdict && READ_ONCE(psock->progs.stream_verdict))) { 274 sk_psock_put(sk, psock); 275 ret = -EBUSY; 276 goto out_progs; 277 } 278 } else { 279 psock = sk_psock_init(sk, map->numa_node); 280 if (IS_ERR(psock)) { 281 ret = PTR_ERR(psock); 282 goto out_progs; 283 } 284 } 285 286 if (msg_parser) 287 psock_set_prog(&psock->progs.msg_parser, msg_parser); 288 if (stream_parser) 289 psock_set_prog(&psock->progs.stream_parser, stream_parser); 290 if (stream_verdict) 291 psock_set_prog(&psock->progs.stream_verdict, stream_verdict); 292 if (skb_verdict) 293 psock_set_prog(&psock->progs.skb_verdict, skb_verdict); 294 295 /* msg_* and stream_* programs references tracked in psock after this 296 * point. Reference dec and cleanup will occur through psock destructor 297 */ 298 ret = sock_map_init_proto(sk, psock); 299 if (ret < 0) { 300 sk_psock_put(sk, psock); 301 goto out; 302 } 303 304 write_lock_bh(&sk->sk_callback_lock); 305 if (stream_parser && stream_verdict && !psock->saved_data_ready) { 306 ret = sk_psock_init_strp(sk, psock); 307 if (ret) { 308 write_unlock_bh(&sk->sk_callback_lock); 309 sk_psock_put(sk, psock); 310 goto out; 311 } 312 sk_psock_start_strp(sk, psock); 313 } else if (!stream_parser && stream_verdict && !psock->saved_data_ready) { 314 sk_psock_start_verdict(sk,psock); 315 } else if (!stream_verdict && skb_verdict && !psock->saved_data_ready) { 316 sk_psock_start_verdict(sk, psock); 317 } 318 write_unlock_bh(&sk->sk_callback_lock); 319 return 0; 320 out_progs: 321 if (skb_verdict) 322 bpf_prog_put(skb_verdict); 323 out_put_msg_parser: 324 if (msg_parser) 325 bpf_prog_put(msg_parser); 326 out_put_stream_parser: 327 if (stream_parser) 328 bpf_prog_put(stream_parser); 329 out_put_stream_verdict: 330 if (stream_verdict) 331 bpf_prog_put(stream_verdict); 332 out: 333 return ret; 334 } 335 336 static void sock_map_free(struct bpf_map *map) 337 { 338 struct bpf_stab *stab = container_of(map, struct bpf_stab, map); 339 int i; 340 341 /* After the sync no updates or deletes will be in-flight so it 342 * is safe to walk map and remove entries without risking a race 343 * in EEXIST update case. 344 */ 345 synchronize_rcu(); 346 for (i = 0; i < stab->map.max_entries; i++) { 347 struct sock **psk = &stab->sks[i]; 348 struct sock *sk; 349 350 sk = xchg(psk, NULL); 351 if (sk) { 352 lock_sock(sk); 353 rcu_read_lock(); 354 sock_map_unref(sk, psk); 355 rcu_read_unlock(); 356 release_sock(sk); 357 } 358 } 359 360 /* wait for psock readers accessing its map link */ 361 synchronize_rcu(); 362 363 bpf_map_area_free(stab->sks); 364 kfree(stab); 365 } 366 367 static void sock_map_release_progs(struct bpf_map *map) 368 { 369 psock_progs_drop(&container_of(map, struct bpf_stab, map)->progs); 370 } 371 372 static struct sock *__sock_map_lookup_elem(struct bpf_map *map, u32 key) 373 { 374 struct bpf_stab *stab = container_of(map, struct bpf_stab, map); 375 376 WARN_ON_ONCE(!rcu_read_lock_held()); 377 378 if (unlikely(key >= map->max_entries)) 379 return NULL; 380 return READ_ONCE(stab->sks[key]); 381 } 382 383 static void *sock_map_lookup(struct bpf_map *map, void *key) 384 { 385 struct sock *sk; 386 387 sk = __sock_map_lookup_elem(map, *(u32 *)key); 388 if (!sk) 389 return NULL; 390 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt)) 391 return NULL; 392 return sk; 393 } 394 395 static void *sock_map_lookup_sys(struct bpf_map *map, void *key) 396 { 397 struct sock *sk; 398 399 if (map->value_size != sizeof(u64)) 400 return ERR_PTR(-ENOSPC); 401 402 sk = __sock_map_lookup_elem(map, *(u32 *)key); 403 if (!sk) 404 return ERR_PTR(-ENOENT); 405 406 __sock_gen_cookie(sk); 407 return &sk->sk_cookie; 408 } 409 410 static int __sock_map_delete(struct bpf_stab *stab, struct sock *sk_test, 411 struct sock **psk) 412 { 413 struct sock *sk; 414 int err = 0; 415 416 raw_spin_lock_bh(&stab->lock); 417 sk = *psk; 418 if (!sk_test || sk_test == sk) 419 sk = xchg(psk, NULL); 420 421 if (likely(sk)) 422 sock_map_unref(sk, psk); 423 else 424 err = -EINVAL; 425 426 raw_spin_unlock_bh(&stab->lock); 427 return err; 428 } 429 430 static void sock_map_delete_from_link(struct bpf_map *map, struct sock *sk, 431 void *link_raw) 432 { 433 struct bpf_stab *stab = container_of(map, struct bpf_stab, map); 434 435 __sock_map_delete(stab, sk, link_raw); 436 } 437 438 static int sock_map_delete_elem(struct bpf_map *map, void *key) 439 { 440 struct bpf_stab *stab = container_of(map, struct bpf_stab, map); 441 u32 i = *(u32 *)key; 442 struct sock **psk; 443 444 if (unlikely(i >= map->max_entries)) 445 return -EINVAL; 446 447 psk = &stab->sks[i]; 448 return __sock_map_delete(stab, NULL, psk); 449 } 450 451 static int sock_map_get_next_key(struct bpf_map *map, void *key, void *next) 452 { 453 struct bpf_stab *stab = container_of(map, struct bpf_stab, map); 454 u32 i = key ? *(u32 *)key : U32_MAX; 455 u32 *key_next = next; 456 457 if (i == stab->map.max_entries - 1) 458 return -ENOENT; 459 if (i >= stab->map.max_entries) 460 *key_next = 0; 461 else 462 *key_next = i + 1; 463 return 0; 464 } 465 466 static int sock_map_update_common(struct bpf_map *map, u32 idx, 467 struct sock *sk, u64 flags) 468 { 469 struct bpf_stab *stab = container_of(map, struct bpf_stab, map); 470 struct sk_psock_link *link; 471 struct sk_psock *psock; 472 struct sock *osk; 473 int ret; 474 475 WARN_ON_ONCE(!rcu_read_lock_held()); 476 if (unlikely(flags > BPF_EXIST)) 477 return -EINVAL; 478 if (unlikely(idx >= map->max_entries)) 479 return -E2BIG; 480 481 link = sk_psock_init_link(); 482 if (!link) 483 return -ENOMEM; 484 485 ret = sock_map_link(map, sk); 486 if (ret < 0) 487 goto out_free; 488 489 psock = sk_psock(sk); 490 WARN_ON_ONCE(!psock); 491 492 raw_spin_lock_bh(&stab->lock); 493 osk = stab->sks[idx]; 494 if (osk && flags == BPF_NOEXIST) { 495 ret = -EEXIST; 496 goto out_unlock; 497 } else if (!osk && flags == BPF_EXIST) { 498 ret = -ENOENT; 499 goto out_unlock; 500 } 501 502 sock_map_add_link(psock, link, map, &stab->sks[idx]); 503 stab->sks[idx] = sk; 504 if (osk) 505 sock_map_unref(osk, &stab->sks[idx]); 506 raw_spin_unlock_bh(&stab->lock); 507 return 0; 508 out_unlock: 509 raw_spin_unlock_bh(&stab->lock); 510 if (psock) 511 sk_psock_put(sk, psock); 512 out_free: 513 sk_psock_free_link(link); 514 return ret; 515 } 516 517 static bool sock_map_op_okay(const struct bpf_sock_ops_kern *ops) 518 { 519 return ops->op == BPF_SOCK_OPS_PASSIVE_ESTABLISHED_CB || 520 ops->op == BPF_SOCK_OPS_ACTIVE_ESTABLISHED_CB || 521 ops->op == BPF_SOCK_OPS_TCP_LISTEN_CB; 522 } 523 524 static bool sock_map_redirect_allowed(const struct sock *sk) 525 { 526 if (sk_is_tcp(sk)) 527 return sk->sk_state != TCP_LISTEN; 528 else 529 return sk->sk_state == TCP_ESTABLISHED; 530 } 531 532 static bool sock_map_sk_is_suitable(const struct sock *sk) 533 { 534 return !!sk->sk_prot->psock_update_sk_prot; 535 } 536 537 static bool sock_map_sk_state_allowed(const struct sock *sk) 538 { 539 if (sk_is_tcp(sk)) 540 return (1 << sk->sk_state) & (TCPF_ESTABLISHED | TCPF_LISTEN); 541 return true; 542 } 543 544 static int sock_hash_update_common(struct bpf_map *map, void *key, 545 struct sock *sk, u64 flags); 546 547 int sock_map_update_elem_sys(struct bpf_map *map, void *key, void *value, 548 u64 flags) 549 { 550 struct socket *sock; 551 struct sock *sk; 552 int ret; 553 u64 ufd; 554 555 if (map->value_size == sizeof(u64)) 556 ufd = *(u64 *)value; 557 else 558 ufd = *(u32 *)value; 559 if (ufd > S32_MAX) 560 return -EINVAL; 561 562 sock = sockfd_lookup(ufd, &ret); 563 if (!sock) 564 return ret; 565 sk = sock->sk; 566 if (!sk) { 567 ret = -EINVAL; 568 goto out; 569 } 570 if (!sock_map_sk_is_suitable(sk)) { 571 ret = -EOPNOTSUPP; 572 goto out; 573 } 574 575 sock_map_sk_acquire(sk); 576 if (!sock_map_sk_state_allowed(sk)) 577 ret = -EOPNOTSUPP; 578 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP) 579 ret = sock_map_update_common(map, *(u32 *)key, sk, flags); 580 else 581 ret = sock_hash_update_common(map, key, sk, flags); 582 sock_map_sk_release(sk); 583 out: 584 sockfd_put(sock); 585 return ret; 586 } 587 588 static int sock_map_update_elem(struct bpf_map *map, void *key, 589 void *value, u64 flags) 590 { 591 struct sock *sk = (struct sock *)value; 592 int ret; 593 594 if (unlikely(!sk || !sk_fullsock(sk))) 595 return -EINVAL; 596 597 if (!sock_map_sk_is_suitable(sk)) 598 return -EOPNOTSUPP; 599 600 local_bh_disable(); 601 bh_lock_sock(sk); 602 if (!sock_map_sk_state_allowed(sk)) 603 ret = -EOPNOTSUPP; 604 else if (map->map_type == BPF_MAP_TYPE_SOCKMAP) 605 ret = sock_map_update_common(map, *(u32 *)key, sk, flags); 606 else 607 ret = sock_hash_update_common(map, key, sk, flags); 608 bh_unlock_sock(sk); 609 local_bh_enable(); 610 return ret; 611 } 612 613 BPF_CALL_4(bpf_sock_map_update, struct bpf_sock_ops_kern *, sops, 614 struct bpf_map *, map, void *, key, u64, flags) 615 { 616 WARN_ON_ONCE(!rcu_read_lock_held()); 617 618 if (likely(sock_map_sk_is_suitable(sops->sk) && 619 sock_map_op_okay(sops))) 620 return sock_map_update_common(map, *(u32 *)key, sops->sk, 621 flags); 622 return -EOPNOTSUPP; 623 } 624 625 const struct bpf_func_proto bpf_sock_map_update_proto = { 626 .func = bpf_sock_map_update, 627 .gpl_only = false, 628 .pkt_access = true, 629 .ret_type = RET_INTEGER, 630 .arg1_type = ARG_PTR_TO_CTX, 631 .arg2_type = ARG_CONST_MAP_PTR, 632 .arg3_type = ARG_PTR_TO_MAP_KEY, 633 .arg4_type = ARG_ANYTHING, 634 }; 635 636 BPF_CALL_4(bpf_sk_redirect_map, struct sk_buff *, skb, 637 struct bpf_map *, map, u32, key, u64, flags) 638 { 639 struct sock *sk; 640 641 if (unlikely(flags & ~(BPF_F_INGRESS))) 642 return SK_DROP; 643 644 sk = __sock_map_lookup_elem(map, key); 645 if (unlikely(!sk || !sock_map_redirect_allowed(sk))) 646 return SK_DROP; 647 648 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS); 649 return SK_PASS; 650 } 651 652 const struct bpf_func_proto bpf_sk_redirect_map_proto = { 653 .func = bpf_sk_redirect_map, 654 .gpl_only = false, 655 .ret_type = RET_INTEGER, 656 .arg1_type = ARG_PTR_TO_CTX, 657 .arg2_type = ARG_CONST_MAP_PTR, 658 .arg3_type = ARG_ANYTHING, 659 .arg4_type = ARG_ANYTHING, 660 }; 661 662 BPF_CALL_4(bpf_msg_redirect_map, struct sk_msg *, msg, 663 struct bpf_map *, map, u32, key, u64, flags) 664 { 665 struct sock *sk; 666 667 if (unlikely(flags & ~(BPF_F_INGRESS))) 668 return SK_DROP; 669 670 sk = __sock_map_lookup_elem(map, key); 671 if (unlikely(!sk || !sock_map_redirect_allowed(sk))) 672 return SK_DROP; 673 674 msg->flags = flags; 675 msg->sk_redir = sk; 676 return SK_PASS; 677 } 678 679 const struct bpf_func_proto bpf_msg_redirect_map_proto = { 680 .func = bpf_msg_redirect_map, 681 .gpl_only = false, 682 .ret_type = RET_INTEGER, 683 .arg1_type = ARG_PTR_TO_CTX, 684 .arg2_type = ARG_CONST_MAP_PTR, 685 .arg3_type = ARG_ANYTHING, 686 .arg4_type = ARG_ANYTHING, 687 }; 688 689 struct sock_map_seq_info { 690 struct bpf_map *map; 691 struct sock *sk; 692 u32 index; 693 }; 694 695 struct bpf_iter__sockmap { 696 __bpf_md_ptr(struct bpf_iter_meta *, meta); 697 __bpf_md_ptr(struct bpf_map *, map); 698 __bpf_md_ptr(void *, key); 699 __bpf_md_ptr(struct sock *, sk); 700 }; 701 702 DEFINE_BPF_ITER_FUNC(sockmap, struct bpf_iter_meta *meta, 703 struct bpf_map *map, void *key, 704 struct sock *sk) 705 706 static void *sock_map_seq_lookup_elem(struct sock_map_seq_info *info) 707 { 708 if (unlikely(info->index >= info->map->max_entries)) 709 return NULL; 710 711 info->sk = __sock_map_lookup_elem(info->map, info->index); 712 713 /* can't return sk directly, since that might be NULL */ 714 return info; 715 } 716 717 static void *sock_map_seq_start(struct seq_file *seq, loff_t *pos) 718 __acquires(rcu) 719 { 720 struct sock_map_seq_info *info = seq->private; 721 722 if (*pos == 0) 723 ++*pos; 724 725 /* pairs with sock_map_seq_stop */ 726 rcu_read_lock(); 727 return sock_map_seq_lookup_elem(info); 728 } 729 730 static void *sock_map_seq_next(struct seq_file *seq, void *v, loff_t *pos) 731 __must_hold(rcu) 732 { 733 struct sock_map_seq_info *info = seq->private; 734 735 ++*pos; 736 ++info->index; 737 738 return sock_map_seq_lookup_elem(info); 739 } 740 741 static int sock_map_seq_show(struct seq_file *seq, void *v) 742 __must_hold(rcu) 743 { 744 struct sock_map_seq_info *info = seq->private; 745 struct bpf_iter__sockmap ctx = {}; 746 struct bpf_iter_meta meta; 747 struct bpf_prog *prog; 748 749 meta.seq = seq; 750 prog = bpf_iter_get_info(&meta, !v); 751 if (!prog) 752 return 0; 753 754 ctx.meta = &meta; 755 ctx.map = info->map; 756 if (v) { 757 ctx.key = &info->index; 758 ctx.sk = info->sk; 759 } 760 761 return bpf_iter_run_prog(prog, &ctx); 762 } 763 764 static void sock_map_seq_stop(struct seq_file *seq, void *v) 765 __releases(rcu) 766 { 767 if (!v) 768 (void)sock_map_seq_show(seq, NULL); 769 770 /* pairs with sock_map_seq_start */ 771 rcu_read_unlock(); 772 } 773 774 static const struct seq_operations sock_map_seq_ops = { 775 .start = sock_map_seq_start, 776 .next = sock_map_seq_next, 777 .stop = sock_map_seq_stop, 778 .show = sock_map_seq_show, 779 }; 780 781 static int sock_map_init_seq_private(void *priv_data, 782 struct bpf_iter_aux_info *aux) 783 { 784 struct sock_map_seq_info *info = priv_data; 785 786 info->map = aux->map; 787 return 0; 788 } 789 790 static const struct bpf_iter_seq_info sock_map_iter_seq_info = { 791 .seq_ops = &sock_map_seq_ops, 792 .init_seq_private = sock_map_init_seq_private, 793 .seq_priv_size = sizeof(struct sock_map_seq_info), 794 }; 795 796 static int sock_map_btf_id; 797 const struct bpf_map_ops sock_map_ops = { 798 .map_meta_equal = bpf_map_meta_equal, 799 .map_alloc = sock_map_alloc, 800 .map_free = sock_map_free, 801 .map_get_next_key = sock_map_get_next_key, 802 .map_lookup_elem_sys_only = sock_map_lookup_sys, 803 .map_update_elem = sock_map_update_elem, 804 .map_delete_elem = sock_map_delete_elem, 805 .map_lookup_elem = sock_map_lookup, 806 .map_release_uref = sock_map_release_progs, 807 .map_check_btf = map_check_no_btf, 808 .map_btf_name = "bpf_stab", 809 .map_btf_id = &sock_map_btf_id, 810 .iter_seq_info = &sock_map_iter_seq_info, 811 }; 812 813 struct bpf_shtab_elem { 814 struct rcu_head rcu; 815 u32 hash; 816 struct sock *sk; 817 struct hlist_node node; 818 u8 key[]; 819 }; 820 821 struct bpf_shtab_bucket { 822 struct hlist_head head; 823 raw_spinlock_t lock; 824 }; 825 826 struct bpf_shtab { 827 struct bpf_map map; 828 struct bpf_shtab_bucket *buckets; 829 u32 buckets_num; 830 u32 elem_size; 831 struct sk_psock_progs progs; 832 atomic_t count; 833 }; 834 835 static inline u32 sock_hash_bucket_hash(const void *key, u32 len) 836 { 837 return jhash(key, len, 0); 838 } 839 840 static struct bpf_shtab_bucket *sock_hash_select_bucket(struct bpf_shtab *htab, 841 u32 hash) 842 { 843 return &htab->buckets[hash & (htab->buckets_num - 1)]; 844 } 845 846 static struct bpf_shtab_elem * 847 sock_hash_lookup_elem_raw(struct hlist_head *head, u32 hash, void *key, 848 u32 key_size) 849 { 850 struct bpf_shtab_elem *elem; 851 852 hlist_for_each_entry_rcu(elem, head, node) { 853 if (elem->hash == hash && 854 !memcmp(&elem->key, key, key_size)) 855 return elem; 856 } 857 858 return NULL; 859 } 860 861 static struct sock *__sock_hash_lookup_elem(struct bpf_map *map, void *key) 862 { 863 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map); 864 u32 key_size = map->key_size, hash; 865 struct bpf_shtab_bucket *bucket; 866 struct bpf_shtab_elem *elem; 867 868 WARN_ON_ONCE(!rcu_read_lock_held()); 869 870 hash = sock_hash_bucket_hash(key, key_size); 871 bucket = sock_hash_select_bucket(htab, hash); 872 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size); 873 874 return elem ? elem->sk : NULL; 875 } 876 877 static void sock_hash_free_elem(struct bpf_shtab *htab, 878 struct bpf_shtab_elem *elem) 879 { 880 atomic_dec(&htab->count); 881 kfree_rcu(elem, rcu); 882 } 883 884 static void sock_hash_delete_from_link(struct bpf_map *map, struct sock *sk, 885 void *link_raw) 886 { 887 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map); 888 struct bpf_shtab_elem *elem_probe, *elem = link_raw; 889 struct bpf_shtab_bucket *bucket; 890 891 WARN_ON_ONCE(!rcu_read_lock_held()); 892 bucket = sock_hash_select_bucket(htab, elem->hash); 893 894 /* elem may be deleted in parallel from the map, but access here 895 * is okay since it's going away only after RCU grace period. 896 * However, we need to check whether it's still present. 897 */ 898 raw_spin_lock_bh(&bucket->lock); 899 elem_probe = sock_hash_lookup_elem_raw(&bucket->head, elem->hash, 900 elem->key, map->key_size); 901 if (elem_probe && elem_probe == elem) { 902 hlist_del_rcu(&elem->node); 903 sock_map_unref(elem->sk, elem); 904 sock_hash_free_elem(htab, elem); 905 } 906 raw_spin_unlock_bh(&bucket->lock); 907 } 908 909 static int sock_hash_delete_elem(struct bpf_map *map, void *key) 910 { 911 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map); 912 u32 hash, key_size = map->key_size; 913 struct bpf_shtab_bucket *bucket; 914 struct bpf_shtab_elem *elem; 915 int ret = -ENOENT; 916 917 hash = sock_hash_bucket_hash(key, key_size); 918 bucket = sock_hash_select_bucket(htab, hash); 919 920 raw_spin_lock_bh(&bucket->lock); 921 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size); 922 if (elem) { 923 hlist_del_rcu(&elem->node); 924 sock_map_unref(elem->sk, elem); 925 sock_hash_free_elem(htab, elem); 926 ret = 0; 927 } 928 raw_spin_unlock_bh(&bucket->lock); 929 return ret; 930 } 931 932 static struct bpf_shtab_elem *sock_hash_alloc_elem(struct bpf_shtab *htab, 933 void *key, u32 key_size, 934 u32 hash, struct sock *sk, 935 struct bpf_shtab_elem *old) 936 { 937 struct bpf_shtab_elem *new; 938 939 if (atomic_inc_return(&htab->count) > htab->map.max_entries) { 940 if (!old) { 941 atomic_dec(&htab->count); 942 return ERR_PTR(-E2BIG); 943 } 944 } 945 946 new = bpf_map_kmalloc_node(&htab->map, htab->elem_size, 947 GFP_ATOMIC | __GFP_NOWARN, 948 htab->map.numa_node); 949 if (!new) { 950 atomic_dec(&htab->count); 951 return ERR_PTR(-ENOMEM); 952 } 953 memcpy(new->key, key, key_size); 954 new->sk = sk; 955 new->hash = hash; 956 return new; 957 } 958 959 static int sock_hash_update_common(struct bpf_map *map, void *key, 960 struct sock *sk, u64 flags) 961 { 962 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map); 963 u32 key_size = map->key_size, hash; 964 struct bpf_shtab_elem *elem, *elem_new; 965 struct bpf_shtab_bucket *bucket; 966 struct sk_psock_link *link; 967 struct sk_psock *psock; 968 int ret; 969 970 WARN_ON_ONCE(!rcu_read_lock_held()); 971 if (unlikely(flags > BPF_EXIST)) 972 return -EINVAL; 973 974 link = sk_psock_init_link(); 975 if (!link) 976 return -ENOMEM; 977 978 ret = sock_map_link(map, sk); 979 if (ret < 0) 980 goto out_free; 981 982 psock = sk_psock(sk); 983 WARN_ON_ONCE(!psock); 984 985 hash = sock_hash_bucket_hash(key, key_size); 986 bucket = sock_hash_select_bucket(htab, hash); 987 988 raw_spin_lock_bh(&bucket->lock); 989 elem = sock_hash_lookup_elem_raw(&bucket->head, hash, key, key_size); 990 if (elem && flags == BPF_NOEXIST) { 991 ret = -EEXIST; 992 goto out_unlock; 993 } else if (!elem && flags == BPF_EXIST) { 994 ret = -ENOENT; 995 goto out_unlock; 996 } 997 998 elem_new = sock_hash_alloc_elem(htab, key, key_size, hash, sk, elem); 999 if (IS_ERR(elem_new)) { 1000 ret = PTR_ERR(elem_new); 1001 goto out_unlock; 1002 } 1003 1004 sock_map_add_link(psock, link, map, elem_new); 1005 /* Add new element to the head of the list, so that 1006 * concurrent search will find it before old elem. 1007 */ 1008 hlist_add_head_rcu(&elem_new->node, &bucket->head); 1009 if (elem) { 1010 hlist_del_rcu(&elem->node); 1011 sock_map_unref(elem->sk, elem); 1012 sock_hash_free_elem(htab, elem); 1013 } 1014 raw_spin_unlock_bh(&bucket->lock); 1015 return 0; 1016 out_unlock: 1017 raw_spin_unlock_bh(&bucket->lock); 1018 sk_psock_put(sk, psock); 1019 out_free: 1020 sk_psock_free_link(link); 1021 return ret; 1022 } 1023 1024 static int sock_hash_get_next_key(struct bpf_map *map, void *key, 1025 void *key_next) 1026 { 1027 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map); 1028 struct bpf_shtab_elem *elem, *elem_next; 1029 u32 hash, key_size = map->key_size; 1030 struct hlist_head *head; 1031 int i = 0; 1032 1033 if (!key) 1034 goto find_first_elem; 1035 hash = sock_hash_bucket_hash(key, key_size); 1036 head = &sock_hash_select_bucket(htab, hash)->head; 1037 elem = sock_hash_lookup_elem_raw(head, hash, key, key_size); 1038 if (!elem) 1039 goto find_first_elem; 1040 1041 elem_next = hlist_entry_safe(rcu_dereference(hlist_next_rcu(&elem->node)), 1042 struct bpf_shtab_elem, node); 1043 if (elem_next) { 1044 memcpy(key_next, elem_next->key, key_size); 1045 return 0; 1046 } 1047 1048 i = hash & (htab->buckets_num - 1); 1049 i++; 1050 find_first_elem: 1051 for (; i < htab->buckets_num; i++) { 1052 head = &sock_hash_select_bucket(htab, i)->head; 1053 elem_next = hlist_entry_safe(rcu_dereference(hlist_first_rcu(head)), 1054 struct bpf_shtab_elem, node); 1055 if (elem_next) { 1056 memcpy(key_next, elem_next->key, key_size); 1057 return 0; 1058 } 1059 } 1060 1061 return -ENOENT; 1062 } 1063 1064 static struct bpf_map *sock_hash_alloc(union bpf_attr *attr) 1065 { 1066 struct bpf_shtab *htab; 1067 int i, err; 1068 1069 if (!capable(CAP_NET_ADMIN)) 1070 return ERR_PTR(-EPERM); 1071 if (attr->max_entries == 0 || 1072 attr->key_size == 0 || 1073 (attr->value_size != sizeof(u32) && 1074 attr->value_size != sizeof(u64)) || 1075 attr->map_flags & ~SOCK_CREATE_FLAG_MASK) 1076 return ERR_PTR(-EINVAL); 1077 if (attr->key_size > MAX_BPF_STACK) 1078 return ERR_PTR(-E2BIG); 1079 1080 htab = kzalloc(sizeof(*htab), GFP_USER | __GFP_ACCOUNT); 1081 if (!htab) 1082 return ERR_PTR(-ENOMEM); 1083 1084 bpf_map_init_from_attr(&htab->map, attr); 1085 1086 htab->buckets_num = roundup_pow_of_two(htab->map.max_entries); 1087 htab->elem_size = sizeof(struct bpf_shtab_elem) + 1088 round_up(htab->map.key_size, 8); 1089 if (htab->buckets_num == 0 || 1090 htab->buckets_num > U32_MAX / sizeof(struct bpf_shtab_bucket)) { 1091 err = -EINVAL; 1092 goto free_htab; 1093 } 1094 1095 htab->buckets = bpf_map_area_alloc(htab->buckets_num * 1096 sizeof(struct bpf_shtab_bucket), 1097 htab->map.numa_node); 1098 if (!htab->buckets) { 1099 err = -ENOMEM; 1100 goto free_htab; 1101 } 1102 1103 for (i = 0; i < htab->buckets_num; i++) { 1104 INIT_HLIST_HEAD(&htab->buckets[i].head); 1105 raw_spin_lock_init(&htab->buckets[i].lock); 1106 } 1107 1108 return &htab->map; 1109 free_htab: 1110 kfree(htab); 1111 return ERR_PTR(err); 1112 } 1113 1114 static void sock_hash_free(struct bpf_map *map) 1115 { 1116 struct bpf_shtab *htab = container_of(map, struct bpf_shtab, map); 1117 struct bpf_shtab_bucket *bucket; 1118 struct hlist_head unlink_list; 1119 struct bpf_shtab_elem *elem; 1120 struct hlist_node *node; 1121 int i; 1122 1123 /* After the sync no updates or deletes will be in-flight so it 1124 * is safe to walk map and remove entries without risking a race 1125 * in EEXIST update case. 1126 */ 1127 synchronize_rcu(); 1128 for (i = 0; i < htab->buckets_num; i++) { 1129 bucket = sock_hash_select_bucket(htab, i); 1130 1131 /* We are racing with sock_hash_delete_from_link to 1132 * enter the spin-lock critical section. Every socket on 1133 * the list is still linked to sockhash. Since link 1134 * exists, psock exists and holds a ref to socket. That 1135 * lets us to grab a socket ref too. 1136 */ 1137 raw_spin_lock_bh(&bucket->lock); 1138 hlist_for_each_entry(elem, &bucket->head, node) 1139 sock_hold(elem->sk); 1140 hlist_move_list(&bucket->head, &unlink_list); 1141 raw_spin_unlock_bh(&bucket->lock); 1142 1143 /* Process removed entries out of atomic context to 1144 * block for socket lock before deleting the psock's 1145 * link to sockhash. 1146 */ 1147 hlist_for_each_entry_safe(elem, node, &unlink_list, node) { 1148 hlist_del(&elem->node); 1149 lock_sock(elem->sk); 1150 rcu_read_lock(); 1151 sock_map_unref(elem->sk, elem); 1152 rcu_read_unlock(); 1153 release_sock(elem->sk); 1154 sock_put(elem->sk); 1155 sock_hash_free_elem(htab, elem); 1156 } 1157 } 1158 1159 /* wait for psock readers accessing its map link */ 1160 synchronize_rcu(); 1161 1162 bpf_map_area_free(htab->buckets); 1163 kfree(htab); 1164 } 1165 1166 static void *sock_hash_lookup_sys(struct bpf_map *map, void *key) 1167 { 1168 struct sock *sk; 1169 1170 if (map->value_size != sizeof(u64)) 1171 return ERR_PTR(-ENOSPC); 1172 1173 sk = __sock_hash_lookup_elem(map, key); 1174 if (!sk) 1175 return ERR_PTR(-ENOENT); 1176 1177 __sock_gen_cookie(sk); 1178 return &sk->sk_cookie; 1179 } 1180 1181 static void *sock_hash_lookup(struct bpf_map *map, void *key) 1182 { 1183 struct sock *sk; 1184 1185 sk = __sock_hash_lookup_elem(map, key); 1186 if (!sk) 1187 return NULL; 1188 if (sk_is_refcounted(sk) && !refcount_inc_not_zero(&sk->sk_refcnt)) 1189 return NULL; 1190 return sk; 1191 } 1192 1193 static void sock_hash_release_progs(struct bpf_map *map) 1194 { 1195 psock_progs_drop(&container_of(map, struct bpf_shtab, map)->progs); 1196 } 1197 1198 BPF_CALL_4(bpf_sock_hash_update, struct bpf_sock_ops_kern *, sops, 1199 struct bpf_map *, map, void *, key, u64, flags) 1200 { 1201 WARN_ON_ONCE(!rcu_read_lock_held()); 1202 1203 if (likely(sock_map_sk_is_suitable(sops->sk) && 1204 sock_map_op_okay(sops))) 1205 return sock_hash_update_common(map, key, sops->sk, flags); 1206 return -EOPNOTSUPP; 1207 } 1208 1209 const struct bpf_func_proto bpf_sock_hash_update_proto = { 1210 .func = bpf_sock_hash_update, 1211 .gpl_only = false, 1212 .pkt_access = true, 1213 .ret_type = RET_INTEGER, 1214 .arg1_type = ARG_PTR_TO_CTX, 1215 .arg2_type = ARG_CONST_MAP_PTR, 1216 .arg3_type = ARG_PTR_TO_MAP_KEY, 1217 .arg4_type = ARG_ANYTHING, 1218 }; 1219 1220 BPF_CALL_4(bpf_sk_redirect_hash, struct sk_buff *, skb, 1221 struct bpf_map *, map, void *, key, u64, flags) 1222 { 1223 struct sock *sk; 1224 1225 if (unlikely(flags & ~(BPF_F_INGRESS))) 1226 return SK_DROP; 1227 1228 sk = __sock_hash_lookup_elem(map, key); 1229 if (unlikely(!sk || !sock_map_redirect_allowed(sk))) 1230 return SK_DROP; 1231 1232 skb_bpf_set_redir(skb, sk, flags & BPF_F_INGRESS); 1233 return SK_PASS; 1234 } 1235 1236 const struct bpf_func_proto bpf_sk_redirect_hash_proto = { 1237 .func = bpf_sk_redirect_hash, 1238 .gpl_only = false, 1239 .ret_type = RET_INTEGER, 1240 .arg1_type = ARG_PTR_TO_CTX, 1241 .arg2_type = ARG_CONST_MAP_PTR, 1242 .arg3_type = ARG_PTR_TO_MAP_KEY, 1243 .arg4_type = ARG_ANYTHING, 1244 }; 1245 1246 BPF_CALL_4(bpf_msg_redirect_hash, struct sk_msg *, msg, 1247 struct bpf_map *, map, void *, key, u64, flags) 1248 { 1249 struct sock *sk; 1250 1251 if (unlikely(flags & ~(BPF_F_INGRESS))) 1252 return SK_DROP; 1253 1254 sk = __sock_hash_lookup_elem(map, key); 1255 if (unlikely(!sk || !sock_map_redirect_allowed(sk))) 1256 return SK_DROP; 1257 1258 msg->flags = flags; 1259 msg->sk_redir = sk; 1260 return SK_PASS; 1261 } 1262 1263 const struct bpf_func_proto bpf_msg_redirect_hash_proto = { 1264 .func = bpf_msg_redirect_hash, 1265 .gpl_only = false, 1266 .ret_type = RET_INTEGER, 1267 .arg1_type = ARG_PTR_TO_CTX, 1268 .arg2_type = ARG_CONST_MAP_PTR, 1269 .arg3_type = ARG_PTR_TO_MAP_KEY, 1270 .arg4_type = ARG_ANYTHING, 1271 }; 1272 1273 struct sock_hash_seq_info { 1274 struct bpf_map *map; 1275 struct bpf_shtab *htab; 1276 u32 bucket_id; 1277 }; 1278 1279 static void *sock_hash_seq_find_next(struct sock_hash_seq_info *info, 1280 struct bpf_shtab_elem *prev_elem) 1281 { 1282 const struct bpf_shtab *htab = info->htab; 1283 struct bpf_shtab_bucket *bucket; 1284 struct bpf_shtab_elem *elem; 1285 struct hlist_node *node; 1286 1287 /* try to find next elem in the same bucket */ 1288 if (prev_elem) { 1289 node = rcu_dereference(hlist_next_rcu(&prev_elem->node)); 1290 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node); 1291 if (elem) 1292 return elem; 1293 1294 /* no more elements, continue in the next bucket */ 1295 info->bucket_id++; 1296 } 1297 1298 for (; info->bucket_id < htab->buckets_num; info->bucket_id++) { 1299 bucket = &htab->buckets[info->bucket_id]; 1300 node = rcu_dereference(hlist_first_rcu(&bucket->head)); 1301 elem = hlist_entry_safe(node, struct bpf_shtab_elem, node); 1302 if (elem) 1303 return elem; 1304 } 1305 1306 return NULL; 1307 } 1308 1309 static void *sock_hash_seq_start(struct seq_file *seq, loff_t *pos) 1310 __acquires(rcu) 1311 { 1312 struct sock_hash_seq_info *info = seq->private; 1313 1314 if (*pos == 0) 1315 ++*pos; 1316 1317 /* pairs with sock_hash_seq_stop */ 1318 rcu_read_lock(); 1319 return sock_hash_seq_find_next(info, NULL); 1320 } 1321 1322 static void *sock_hash_seq_next(struct seq_file *seq, void *v, loff_t *pos) 1323 __must_hold(rcu) 1324 { 1325 struct sock_hash_seq_info *info = seq->private; 1326 1327 ++*pos; 1328 return sock_hash_seq_find_next(info, v); 1329 } 1330 1331 static int sock_hash_seq_show(struct seq_file *seq, void *v) 1332 __must_hold(rcu) 1333 { 1334 struct sock_hash_seq_info *info = seq->private; 1335 struct bpf_iter__sockmap ctx = {}; 1336 struct bpf_shtab_elem *elem = v; 1337 struct bpf_iter_meta meta; 1338 struct bpf_prog *prog; 1339 1340 meta.seq = seq; 1341 prog = bpf_iter_get_info(&meta, !elem); 1342 if (!prog) 1343 return 0; 1344 1345 ctx.meta = &meta; 1346 ctx.map = info->map; 1347 if (elem) { 1348 ctx.key = elem->key; 1349 ctx.sk = elem->sk; 1350 } 1351 1352 return bpf_iter_run_prog(prog, &ctx); 1353 } 1354 1355 static void sock_hash_seq_stop(struct seq_file *seq, void *v) 1356 __releases(rcu) 1357 { 1358 if (!v) 1359 (void)sock_hash_seq_show(seq, NULL); 1360 1361 /* pairs with sock_hash_seq_start */ 1362 rcu_read_unlock(); 1363 } 1364 1365 static const struct seq_operations sock_hash_seq_ops = { 1366 .start = sock_hash_seq_start, 1367 .next = sock_hash_seq_next, 1368 .stop = sock_hash_seq_stop, 1369 .show = sock_hash_seq_show, 1370 }; 1371 1372 static int sock_hash_init_seq_private(void *priv_data, 1373 struct bpf_iter_aux_info *aux) 1374 { 1375 struct sock_hash_seq_info *info = priv_data; 1376 1377 info->map = aux->map; 1378 info->htab = container_of(aux->map, struct bpf_shtab, map); 1379 return 0; 1380 } 1381 1382 static const struct bpf_iter_seq_info sock_hash_iter_seq_info = { 1383 .seq_ops = &sock_hash_seq_ops, 1384 .init_seq_private = sock_hash_init_seq_private, 1385 .seq_priv_size = sizeof(struct sock_hash_seq_info), 1386 }; 1387 1388 static int sock_hash_map_btf_id; 1389 const struct bpf_map_ops sock_hash_ops = { 1390 .map_meta_equal = bpf_map_meta_equal, 1391 .map_alloc = sock_hash_alloc, 1392 .map_free = sock_hash_free, 1393 .map_get_next_key = sock_hash_get_next_key, 1394 .map_update_elem = sock_map_update_elem, 1395 .map_delete_elem = sock_hash_delete_elem, 1396 .map_lookup_elem = sock_hash_lookup, 1397 .map_lookup_elem_sys_only = sock_hash_lookup_sys, 1398 .map_release_uref = sock_hash_release_progs, 1399 .map_check_btf = map_check_no_btf, 1400 .map_btf_name = "bpf_shtab", 1401 .map_btf_id = &sock_hash_map_btf_id, 1402 .iter_seq_info = &sock_hash_iter_seq_info, 1403 }; 1404 1405 static struct sk_psock_progs *sock_map_progs(struct bpf_map *map) 1406 { 1407 switch (map->map_type) { 1408 case BPF_MAP_TYPE_SOCKMAP: 1409 return &container_of(map, struct bpf_stab, map)->progs; 1410 case BPF_MAP_TYPE_SOCKHASH: 1411 return &container_of(map, struct bpf_shtab, map)->progs; 1412 default: 1413 break; 1414 } 1415 1416 return NULL; 1417 } 1418 1419 static int sock_map_prog_update(struct bpf_map *map, struct bpf_prog *prog, 1420 struct bpf_prog *old, u32 which) 1421 { 1422 struct sk_psock_progs *progs = sock_map_progs(map); 1423 struct bpf_prog **pprog; 1424 1425 if (!progs) 1426 return -EOPNOTSUPP; 1427 1428 switch (which) { 1429 case BPF_SK_MSG_VERDICT: 1430 pprog = &progs->msg_parser; 1431 break; 1432 #if IS_ENABLED(CONFIG_BPF_STREAM_PARSER) 1433 case BPF_SK_SKB_STREAM_PARSER: 1434 pprog = &progs->stream_parser; 1435 break; 1436 #endif 1437 case BPF_SK_SKB_STREAM_VERDICT: 1438 if (progs->skb_verdict) 1439 return -EBUSY; 1440 pprog = &progs->stream_verdict; 1441 break; 1442 case BPF_SK_SKB_VERDICT: 1443 if (progs->stream_verdict) 1444 return -EBUSY; 1445 pprog = &progs->skb_verdict; 1446 break; 1447 default: 1448 return -EOPNOTSUPP; 1449 } 1450 1451 if (old) 1452 return psock_replace_prog(pprog, prog, old); 1453 1454 psock_set_prog(pprog, prog); 1455 return 0; 1456 } 1457 1458 static void sock_map_unlink(struct sock *sk, struct sk_psock_link *link) 1459 { 1460 switch (link->map->map_type) { 1461 case BPF_MAP_TYPE_SOCKMAP: 1462 return sock_map_delete_from_link(link->map, sk, 1463 link->link_raw); 1464 case BPF_MAP_TYPE_SOCKHASH: 1465 return sock_hash_delete_from_link(link->map, sk, 1466 link->link_raw); 1467 default: 1468 break; 1469 } 1470 } 1471 1472 static void sock_map_remove_links(struct sock *sk, struct sk_psock *psock) 1473 { 1474 struct sk_psock_link *link; 1475 1476 while ((link = sk_psock_link_pop(psock))) { 1477 sock_map_unlink(sk, link); 1478 sk_psock_free_link(link); 1479 } 1480 } 1481 1482 void sock_map_unhash(struct sock *sk) 1483 { 1484 void (*saved_unhash)(struct sock *sk); 1485 struct sk_psock *psock; 1486 1487 rcu_read_lock(); 1488 psock = sk_psock(sk); 1489 if (unlikely(!psock)) { 1490 rcu_read_unlock(); 1491 if (sk->sk_prot->unhash) 1492 sk->sk_prot->unhash(sk); 1493 return; 1494 } 1495 1496 saved_unhash = psock->saved_unhash; 1497 sock_map_remove_links(sk, psock); 1498 rcu_read_unlock(); 1499 saved_unhash(sk); 1500 } 1501 EXPORT_SYMBOL_GPL(sock_map_unhash); 1502 1503 void sock_map_close(struct sock *sk, long timeout) 1504 { 1505 void (*saved_close)(struct sock *sk, long timeout); 1506 struct sk_psock *psock; 1507 1508 lock_sock(sk); 1509 rcu_read_lock(); 1510 psock = sk_psock_get(sk); 1511 if (unlikely(!psock)) { 1512 rcu_read_unlock(); 1513 release_sock(sk); 1514 return sk->sk_prot->close(sk, timeout); 1515 } 1516 1517 saved_close = psock->saved_close; 1518 sock_map_remove_links(sk, psock); 1519 rcu_read_unlock(); 1520 sk_psock_stop(psock, true); 1521 sk_psock_put(sk, psock); 1522 release_sock(sk); 1523 saved_close(sk, timeout); 1524 } 1525 EXPORT_SYMBOL_GPL(sock_map_close); 1526 1527 static int sock_map_iter_attach_target(struct bpf_prog *prog, 1528 union bpf_iter_link_info *linfo, 1529 struct bpf_iter_aux_info *aux) 1530 { 1531 struct bpf_map *map; 1532 int err = -EINVAL; 1533 1534 if (!linfo->map.map_fd) 1535 return -EBADF; 1536 1537 map = bpf_map_get_with_uref(linfo->map.map_fd); 1538 if (IS_ERR(map)) 1539 return PTR_ERR(map); 1540 1541 if (map->map_type != BPF_MAP_TYPE_SOCKMAP && 1542 map->map_type != BPF_MAP_TYPE_SOCKHASH) 1543 goto put_map; 1544 1545 if (prog->aux->max_rdonly_access > map->key_size) { 1546 err = -EACCES; 1547 goto put_map; 1548 } 1549 1550 aux->map = map; 1551 return 0; 1552 1553 put_map: 1554 bpf_map_put_with_uref(map); 1555 return err; 1556 } 1557 1558 static void sock_map_iter_detach_target(struct bpf_iter_aux_info *aux) 1559 { 1560 bpf_map_put_with_uref(aux->map); 1561 } 1562 1563 static struct bpf_iter_reg sock_map_iter_reg = { 1564 .target = "sockmap", 1565 .attach_target = sock_map_iter_attach_target, 1566 .detach_target = sock_map_iter_detach_target, 1567 .show_fdinfo = bpf_iter_map_show_fdinfo, 1568 .fill_link_info = bpf_iter_map_fill_link_info, 1569 .ctx_arg_info_size = 2, 1570 .ctx_arg_info = { 1571 { offsetof(struct bpf_iter__sockmap, key), 1572 PTR_TO_BUF | PTR_MAYBE_NULL | MEM_RDONLY }, 1573 { offsetof(struct bpf_iter__sockmap, sk), 1574 PTR_TO_BTF_ID_OR_NULL }, 1575 }, 1576 }; 1577 1578 static int __init bpf_sockmap_iter_init(void) 1579 { 1580 sock_map_iter_reg.ctx_arg_info[1].btf_id = 1581 btf_sock_ids[BTF_SOCK_TYPE_SOCK]; 1582 return bpf_iter_reg_target(&sock_map_iter_reg); 1583 } 1584 late_initcall(bpf_sockmap_iter_init); 1585