1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Linux Socket Filter - Kernel level socket filtering 4 * 5 * Based on the design of the Berkeley Packet Filter. The new 6 * internal format has been designed by PLUMgrid: 7 * 8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com 9 * 10 * Authors: 11 * 12 * Jay Schulist <jschlst@samba.org> 13 * Alexei Starovoitov <ast@plumgrid.com> 14 * Daniel Borkmann <dborkman@redhat.com> 15 * 16 * Andi Kleen - Fix a few bad bugs and races. 17 * Kris Katterjohn - Added many additional checks in bpf_check_classic() 18 */ 19 20 #include <linux/atomic.h> 21 #include <linux/bpf_verifier.h> 22 #include <linux/module.h> 23 #include <linux/types.h> 24 #include <linux/mm.h> 25 #include <linux/fcntl.h> 26 #include <linux/socket.h> 27 #include <linux/sock_diag.h> 28 #include <linux/in.h> 29 #include <linux/inet.h> 30 #include <linux/netdevice.h> 31 #include <linux/if_packet.h> 32 #include <linux/if_arp.h> 33 #include <linux/gfp.h> 34 #include <net/inet_common.h> 35 #include <net/ip.h> 36 #include <net/protocol.h> 37 #include <net/netlink.h> 38 #include <linux/skbuff.h> 39 #include <linux/skmsg.h> 40 #include <net/sock.h> 41 #include <net/flow_dissector.h> 42 #include <linux/errno.h> 43 #include <linux/timer.h> 44 #include <linux/uaccess.h> 45 #include <asm/unaligned.h> 46 #include <linux/filter.h> 47 #include <linux/ratelimit.h> 48 #include <linux/seccomp.h> 49 #include <linux/if_vlan.h> 50 #include <linux/bpf.h> 51 #include <linux/btf.h> 52 #include <net/sch_generic.h> 53 #include <net/cls_cgroup.h> 54 #include <net/dst_metadata.h> 55 #include <net/dst.h> 56 #include <net/sock_reuseport.h> 57 #include <net/busy_poll.h> 58 #include <net/tcp.h> 59 #include <net/xfrm.h> 60 #include <net/udp.h> 61 #include <linux/bpf_trace.h> 62 #include <net/xdp_sock.h> 63 #include <linux/inetdevice.h> 64 #include <net/inet_hashtables.h> 65 #include <net/inet6_hashtables.h> 66 #include <net/ip_fib.h> 67 #include <net/nexthop.h> 68 #include <net/flow.h> 69 #include <net/arp.h> 70 #include <net/ipv6.h> 71 #include <net/net_namespace.h> 72 #include <linux/seg6_local.h> 73 #include <net/seg6.h> 74 #include <net/seg6_local.h> 75 #include <net/lwtunnel.h> 76 #include <net/ipv6_stubs.h> 77 #include <net/bpf_sk_storage.h> 78 #include <net/transp_v6.h> 79 #include <linux/btf_ids.h> 80 #include <net/tls.h> 81 #include <net/xdp.h> 82 #include <net/mptcp.h> 83 #include <net/netfilter/nf_conntrack_bpf.h> 84 85 static const struct bpf_func_proto * 86 bpf_sk_base_func_proto(enum bpf_func_id func_id); 87 88 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len) 89 { 90 if (in_compat_syscall()) { 91 struct compat_sock_fprog f32; 92 93 if (len != sizeof(f32)) 94 return -EINVAL; 95 if (copy_from_sockptr(&f32, src, sizeof(f32))) 96 return -EFAULT; 97 memset(dst, 0, sizeof(*dst)); 98 dst->len = f32.len; 99 dst->filter = compat_ptr(f32.filter); 100 } else { 101 if (len != sizeof(*dst)) 102 return -EINVAL; 103 if (copy_from_sockptr(dst, src, sizeof(*dst))) 104 return -EFAULT; 105 } 106 107 return 0; 108 } 109 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user); 110 111 /** 112 * sk_filter_trim_cap - run a packet through a socket filter 113 * @sk: sock associated with &sk_buff 114 * @skb: buffer to filter 115 * @cap: limit on how short the eBPF program may trim the packet 116 * 117 * Run the eBPF program and then cut skb->data to correct size returned by 118 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller 119 * than pkt_len we keep whole skb->data. This is the socket level 120 * wrapper to bpf_prog_run. It returns 0 if the packet should 121 * be accepted or -EPERM if the packet should be tossed. 122 * 123 */ 124 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap) 125 { 126 int err; 127 struct sk_filter *filter; 128 129 /* 130 * If the skb was allocated from pfmemalloc reserves, only 131 * allow SOCK_MEMALLOC sockets to use it as this socket is 132 * helping free memory 133 */ 134 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) { 135 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP); 136 return -ENOMEM; 137 } 138 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb); 139 if (err) 140 return err; 141 142 err = security_sock_rcv_skb(sk, skb); 143 if (err) 144 return err; 145 146 rcu_read_lock(); 147 filter = rcu_dereference(sk->sk_filter); 148 if (filter) { 149 struct sock *save_sk = skb->sk; 150 unsigned int pkt_len; 151 152 skb->sk = sk; 153 pkt_len = bpf_prog_run_save_cb(filter->prog, skb); 154 skb->sk = save_sk; 155 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM; 156 } 157 rcu_read_unlock(); 158 159 return err; 160 } 161 EXPORT_SYMBOL(sk_filter_trim_cap); 162 163 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb) 164 { 165 return skb_get_poff(skb); 166 } 167 168 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x) 169 { 170 struct nlattr *nla; 171 172 if (skb_is_nonlinear(skb)) 173 return 0; 174 175 if (skb->len < sizeof(struct nlattr)) 176 return 0; 177 178 if (a > skb->len - sizeof(struct nlattr)) 179 return 0; 180 181 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x); 182 if (nla) 183 return (void *) nla - (void *) skb->data; 184 185 return 0; 186 } 187 188 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x) 189 { 190 struct nlattr *nla; 191 192 if (skb_is_nonlinear(skb)) 193 return 0; 194 195 if (skb->len < sizeof(struct nlattr)) 196 return 0; 197 198 if (a > skb->len - sizeof(struct nlattr)) 199 return 0; 200 201 nla = (struct nlattr *) &skb->data[a]; 202 if (nla->nla_len > skb->len - a) 203 return 0; 204 205 nla = nla_find_nested(nla, x); 206 if (nla) 207 return (void *) nla - (void *) skb->data; 208 209 return 0; 210 } 211 212 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *, 213 data, int, headlen, int, offset) 214 { 215 u8 tmp, *ptr; 216 const int len = sizeof(tmp); 217 218 if (offset >= 0) { 219 if (headlen - offset >= len) 220 return *(u8 *)(data + offset); 221 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 222 return tmp; 223 } else { 224 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 225 if (likely(ptr)) 226 return *(u8 *)ptr; 227 } 228 229 return -EFAULT; 230 } 231 232 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb, 233 int, offset) 234 { 235 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len, 236 offset); 237 } 238 239 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *, 240 data, int, headlen, int, offset) 241 { 242 __be16 tmp, *ptr; 243 const int len = sizeof(tmp); 244 245 if (offset >= 0) { 246 if (headlen - offset >= len) 247 return get_unaligned_be16(data + offset); 248 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 249 return be16_to_cpu(tmp); 250 } else { 251 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 252 if (likely(ptr)) 253 return get_unaligned_be16(ptr); 254 } 255 256 return -EFAULT; 257 } 258 259 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb, 260 int, offset) 261 { 262 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len, 263 offset); 264 } 265 266 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *, 267 data, int, headlen, int, offset) 268 { 269 __be32 tmp, *ptr; 270 const int len = sizeof(tmp); 271 272 if (likely(offset >= 0)) { 273 if (headlen - offset >= len) 274 return get_unaligned_be32(data + offset); 275 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 276 return be32_to_cpu(tmp); 277 } else { 278 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 279 if (likely(ptr)) 280 return get_unaligned_be32(ptr); 281 } 282 283 return -EFAULT; 284 } 285 286 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb, 287 int, offset) 288 { 289 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len, 290 offset); 291 } 292 293 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg, 294 struct bpf_insn *insn_buf) 295 { 296 struct bpf_insn *insn = insn_buf; 297 298 switch (skb_field) { 299 case SKF_AD_MARK: 300 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4); 301 302 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, 303 offsetof(struct sk_buff, mark)); 304 break; 305 306 case SKF_AD_PKTTYPE: 307 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET); 308 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX); 309 #ifdef __BIG_ENDIAN_BITFIELD 310 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5); 311 #endif 312 break; 313 314 case SKF_AD_QUEUE: 315 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2); 316 317 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 318 offsetof(struct sk_buff, queue_mapping)); 319 break; 320 321 case SKF_AD_VLAN_TAG: 322 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2); 323 324 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */ 325 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 326 offsetof(struct sk_buff, vlan_tci)); 327 break; 328 case SKF_AD_VLAN_TAG_PRESENT: 329 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_all) != 4); 330 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, 331 offsetof(struct sk_buff, vlan_all)); 332 *insn++ = BPF_JMP_IMM(BPF_JEQ, dst_reg, 0, 1); 333 *insn++ = BPF_ALU32_IMM(BPF_MOV, dst_reg, 1); 334 break; 335 } 336 337 return insn - insn_buf; 338 } 339 340 static bool convert_bpf_extensions(struct sock_filter *fp, 341 struct bpf_insn **insnp) 342 { 343 struct bpf_insn *insn = *insnp; 344 u32 cnt; 345 346 switch (fp->k) { 347 case SKF_AD_OFF + SKF_AD_PROTOCOL: 348 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2); 349 350 /* A = *(u16 *) (CTX + offsetof(protocol)) */ 351 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 352 offsetof(struct sk_buff, protocol)); 353 /* A = ntohs(A) [emitting a nop or swap16] */ 354 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 355 break; 356 357 case SKF_AD_OFF + SKF_AD_PKTTYPE: 358 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn); 359 insn += cnt - 1; 360 break; 361 362 case SKF_AD_OFF + SKF_AD_IFINDEX: 363 case SKF_AD_OFF + SKF_AD_HATYPE: 364 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4); 365 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2); 366 367 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 368 BPF_REG_TMP, BPF_REG_CTX, 369 offsetof(struct sk_buff, dev)); 370 /* if (tmp != 0) goto pc + 1 */ 371 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1); 372 *insn++ = BPF_EXIT_INSN(); 373 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) 374 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP, 375 offsetof(struct net_device, ifindex)); 376 else 377 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP, 378 offsetof(struct net_device, type)); 379 break; 380 381 case SKF_AD_OFF + SKF_AD_MARK: 382 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn); 383 insn += cnt - 1; 384 break; 385 386 case SKF_AD_OFF + SKF_AD_RXHASH: 387 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4); 388 389 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, 390 offsetof(struct sk_buff, hash)); 391 break; 392 393 case SKF_AD_OFF + SKF_AD_QUEUE: 394 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn); 395 insn += cnt - 1; 396 break; 397 398 case SKF_AD_OFF + SKF_AD_VLAN_TAG: 399 cnt = convert_skb_access(SKF_AD_VLAN_TAG, 400 BPF_REG_A, BPF_REG_CTX, insn); 401 insn += cnt - 1; 402 break; 403 404 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT: 405 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT, 406 BPF_REG_A, BPF_REG_CTX, insn); 407 insn += cnt - 1; 408 break; 409 410 case SKF_AD_OFF + SKF_AD_VLAN_TPID: 411 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2); 412 413 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */ 414 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 415 offsetof(struct sk_buff, vlan_proto)); 416 /* A = ntohs(A) [emitting a nop or swap16] */ 417 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 418 break; 419 420 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 421 case SKF_AD_OFF + SKF_AD_NLATTR: 422 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 423 case SKF_AD_OFF + SKF_AD_CPU: 424 case SKF_AD_OFF + SKF_AD_RANDOM: 425 /* arg1 = CTX */ 426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 427 /* arg2 = A */ 428 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A); 429 /* arg3 = X */ 430 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X); 431 /* Emit call(arg1=CTX, arg2=A, arg3=X) */ 432 switch (fp->k) { 433 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 434 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset); 435 break; 436 case SKF_AD_OFF + SKF_AD_NLATTR: 437 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr); 438 break; 439 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 440 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest); 441 break; 442 case SKF_AD_OFF + SKF_AD_CPU: 443 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id); 444 break; 445 case SKF_AD_OFF + SKF_AD_RANDOM: 446 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32); 447 bpf_user_rnd_init_once(); 448 break; 449 } 450 break; 451 452 case SKF_AD_OFF + SKF_AD_ALU_XOR_X: 453 /* A ^= X */ 454 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X); 455 break; 456 457 default: 458 /* This is just a dummy call to avoid letting the compiler 459 * evict __bpf_call_base() as an optimization. Placed here 460 * where no-one bothers. 461 */ 462 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0); 463 return false; 464 } 465 466 *insnp = insn; 467 return true; 468 } 469 470 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp) 471 { 472 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS); 473 int size = bpf_size_to_bytes(BPF_SIZE(fp->code)); 474 bool endian = BPF_SIZE(fp->code) == BPF_H || 475 BPF_SIZE(fp->code) == BPF_W; 476 bool indirect = BPF_MODE(fp->code) == BPF_IND; 477 const int ip_align = NET_IP_ALIGN; 478 struct bpf_insn *insn = *insnp; 479 int offset = fp->k; 480 481 if (!indirect && 482 ((unaligned_ok && offset >= 0) || 483 (!unaligned_ok && offset >= 0 && 484 offset + ip_align >= 0 && 485 offset + ip_align % size == 0))) { 486 bool ldx_off_ok = offset <= S16_MAX; 487 488 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H); 489 if (offset) 490 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset); 491 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP, 492 size, 2 + endian + (!ldx_off_ok * 2)); 493 if (ldx_off_ok) { 494 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 495 BPF_REG_D, offset); 496 } else { 497 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D); 498 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset); 499 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 500 BPF_REG_TMP, 0); 501 } 502 if (endian) 503 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8); 504 *insn++ = BPF_JMP_A(8); 505 } 506 507 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 508 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D); 509 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H); 510 if (!indirect) { 511 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset); 512 } else { 513 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X); 514 if (fp->k) 515 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset); 516 } 517 518 switch (BPF_SIZE(fp->code)) { 519 case BPF_B: 520 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8); 521 break; 522 case BPF_H: 523 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16); 524 break; 525 case BPF_W: 526 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32); 527 break; 528 default: 529 return false; 530 } 531 532 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2); 533 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 534 *insn = BPF_EXIT_INSN(); 535 536 *insnp = insn; 537 return true; 538 } 539 540 /** 541 * bpf_convert_filter - convert filter program 542 * @prog: the user passed filter program 543 * @len: the length of the user passed filter program 544 * @new_prog: allocated 'struct bpf_prog' or NULL 545 * @new_len: pointer to store length of converted program 546 * @seen_ld_abs: bool whether we've seen ld_abs/ind 547 * 548 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn' 549 * style extended BPF (eBPF). 550 * Conversion workflow: 551 * 552 * 1) First pass for calculating the new program length: 553 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs) 554 * 555 * 2) 2nd pass to remap in two passes: 1st pass finds new 556 * jump offsets, 2nd pass remapping: 557 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs) 558 */ 559 static int bpf_convert_filter(struct sock_filter *prog, int len, 560 struct bpf_prog *new_prog, int *new_len, 561 bool *seen_ld_abs) 562 { 563 int new_flen = 0, pass = 0, target, i, stack_off; 564 struct bpf_insn *new_insn, *first_insn = NULL; 565 struct sock_filter *fp; 566 int *addrs = NULL; 567 u8 bpf_src; 568 569 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK); 570 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); 571 572 if (len <= 0 || len > BPF_MAXINSNS) 573 return -EINVAL; 574 575 if (new_prog) { 576 first_insn = new_prog->insnsi; 577 addrs = kcalloc(len, sizeof(*addrs), 578 GFP_KERNEL | __GFP_NOWARN); 579 if (!addrs) 580 return -ENOMEM; 581 } 582 583 do_pass: 584 new_insn = first_insn; 585 fp = prog; 586 587 /* Classic BPF related prologue emission. */ 588 if (new_prog) { 589 /* Classic BPF expects A and X to be reset first. These need 590 * to be guaranteed to be the first two instructions. 591 */ 592 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 593 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X); 594 595 /* All programs must keep CTX in callee saved BPF_REG_CTX. 596 * In eBPF case it's done by the compiler, here we need to 597 * do this ourself. Initial CTX is present in BPF_REG_ARG1. 598 */ 599 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1); 600 if (*seen_ld_abs) { 601 /* For packet access in classic BPF, cache skb->data 602 * in callee-saved BPF R8 and skb->len - skb->data_len 603 * (headlen) in BPF R9. Since classic BPF is read-only 604 * on CTX, we only need to cache it once. 605 */ 606 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 607 BPF_REG_D, BPF_REG_CTX, 608 offsetof(struct sk_buff, data)); 609 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX, 610 offsetof(struct sk_buff, len)); 611 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX, 612 offsetof(struct sk_buff, data_len)); 613 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP); 614 } 615 } else { 616 new_insn += 3; 617 } 618 619 for (i = 0; i < len; fp++, i++) { 620 struct bpf_insn tmp_insns[32] = { }; 621 struct bpf_insn *insn = tmp_insns; 622 623 if (addrs) 624 addrs[i] = new_insn - first_insn; 625 626 switch (fp->code) { 627 /* All arithmetic insns and skb loads map as-is. */ 628 case BPF_ALU | BPF_ADD | BPF_X: 629 case BPF_ALU | BPF_ADD | BPF_K: 630 case BPF_ALU | BPF_SUB | BPF_X: 631 case BPF_ALU | BPF_SUB | BPF_K: 632 case BPF_ALU | BPF_AND | BPF_X: 633 case BPF_ALU | BPF_AND | BPF_K: 634 case BPF_ALU | BPF_OR | BPF_X: 635 case BPF_ALU | BPF_OR | BPF_K: 636 case BPF_ALU | BPF_LSH | BPF_X: 637 case BPF_ALU | BPF_LSH | BPF_K: 638 case BPF_ALU | BPF_RSH | BPF_X: 639 case BPF_ALU | BPF_RSH | BPF_K: 640 case BPF_ALU | BPF_XOR | BPF_X: 641 case BPF_ALU | BPF_XOR | BPF_K: 642 case BPF_ALU | BPF_MUL | BPF_X: 643 case BPF_ALU | BPF_MUL | BPF_K: 644 case BPF_ALU | BPF_DIV | BPF_X: 645 case BPF_ALU | BPF_DIV | BPF_K: 646 case BPF_ALU | BPF_MOD | BPF_X: 647 case BPF_ALU | BPF_MOD | BPF_K: 648 case BPF_ALU | BPF_NEG: 649 case BPF_LD | BPF_ABS | BPF_W: 650 case BPF_LD | BPF_ABS | BPF_H: 651 case BPF_LD | BPF_ABS | BPF_B: 652 case BPF_LD | BPF_IND | BPF_W: 653 case BPF_LD | BPF_IND | BPF_H: 654 case BPF_LD | BPF_IND | BPF_B: 655 /* Check for overloaded BPF extension and 656 * directly convert it if found, otherwise 657 * just move on with mapping. 658 */ 659 if (BPF_CLASS(fp->code) == BPF_LD && 660 BPF_MODE(fp->code) == BPF_ABS && 661 convert_bpf_extensions(fp, &insn)) 662 break; 663 if (BPF_CLASS(fp->code) == BPF_LD && 664 convert_bpf_ld_abs(fp, &insn)) { 665 *seen_ld_abs = true; 666 break; 667 } 668 669 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) || 670 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) { 671 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X); 672 /* Error with exception code on div/mod by 0. 673 * For cBPF programs, this was always return 0. 674 */ 675 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2); 676 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 677 *insn++ = BPF_EXIT_INSN(); 678 } 679 680 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k); 681 break; 682 683 /* Jump transformation cannot use BPF block macros 684 * everywhere as offset calculation and target updates 685 * require a bit more work than the rest, i.e. jump 686 * opcodes map as-is, but offsets need adjustment. 687 */ 688 689 #define BPF_EMIT_JMP \ 690 do { \ 691 const s32 off_min = S16_MIN, off_max = S16_MAX; \ 692 s32 off; \ 693 \ 694 if (target >= len || target < 0) \ 695 goto err; \ 696 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \ 697 /* Adjust pc relative offset for 2nd or 3rd insn. */ \ 698 off -= insn - tmp_insns; \ 699 /* Reject anything not fitting into insn->off. */ \ 700 if (off < off_min || off > off_max) \ 701 goto err; \ 702 insn->off = off; \ 703 } while (0) 704 705 case BPF_JMP | BPF_JA: 706 target = i + fp->k + 1; 707 insn->code = fp->code; 708 BPF_EMIT_JMP; 709 break; 710 711 case BPF_JMP | BPF_JEQ | BPF_K: 712 case BPF_JMP | BPF_JEQ | BPF_X: 713 case BPF_JMP | BPF_JSET | BPF_K: 714 case BPF_JMP | BPF_JSET | BPF_X: 715 case BPF_JMP | BPF_JGT | BPF_K: 716 case BPF_JMP | BPF_JGT | BPF_X: 717 case BPF_JMP | BPF_JGE | BPF_K: 718 case BPF_JMP | BPF_JGE | BPF_X: 719 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) { 720 /* BPF immediates are signed, zero extend 721 * immediate into tmp register and use it 722 * in compare insn. 723 */ 724 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k); 725 726 insn->dst_reg = BPF_REG_A; 727 insn->src_reg = BPF_REG_TMP; 728 bpf_src = BPF_X; 729 } else { 730 insn->dst_reg = BPF_REG_A; 731 insn->imm = fp->k; 732 bpf_src = BPF_SRC(fp->code); 733 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0; 734 } 735 736 /* Common case where 'jump_false' is next insn. */ 737 if (fp->jf == 0) { 738 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 739 target = i + fp->jt + 1; 740 BPF_EMIT_JMP; 741 break; 742 } 743 744 /* Convert some jumps when 'jump_true' is next insn. */ 745 if (fp->jt == 0) { 746 switch (BPF_OP(fp->code)) { 747 case BPF_JEQ: 748 insn->code = BPF_JMP | BPF_JNE | bpf_src; 749 break; 750 case BPF_JGT: 751 insn->code = BPF_JMP | BPF_JLE | bpf_src; 752 break; 753 case BPF_JGE: 754 insn->code = BPF_JMP | BPF_JLT | bpf_src; 755 break; 756 default: 757 goto jmp_rest; 758 } 759 760 target = i + fp->jf + 1; 761 BPF_EMIT_JMP; 762 break; 763 } 764 jmp_rest: 765 /* Other jumps are mapped into two insns: Jxx and JA. */ 766 target = i + fp->jt + 1; 767 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 768 BPF_EMIT_JMP; 769 insn++; 770 771 insn->code = BPF_JMP | BPF_JA; 772 target = i + fp->jf + 1; 773 BPF_EMIT_JMP; 774 break; 775 776 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */ 777 case BPF_LDX | BPF_MSH | BPF_B: { 778 struct sock_filter tmp = { 779 .code = BPF_LD | BPF_ABS | BPF_B, 780 .k = fp->k, 781 }; 782 783 *seen_ld_abs = true; 784 785 /* X = A */ 786 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 787 /* A = BPF_R0 = *(u8 *) (skb->data + K) */ 788 convert_bpf_ld_abs(&tmp, &insn); 789 insn++; 790 /* A &= 0xf */ 791 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf); 792 /* A <<= 2 */ 793 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2); 794 /* tmp = X */ 795 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X); 796 /* X = A */ 797 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 798 /* A = tmp */ 799 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP); 800 break; 801 } 802 /* RET_K is remaped into 2 insns. RET_A case doesn't need an 803 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A. 804 */ 805 case BPF_RET | BPF_A: 806 case BPF_RET | BPF_K: 807 if (BPF_RVAL(fp->code) == BPF_K) 808 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0, 809 0, fp->k); 810 *insn = BPF_EXIT_INSN(); 811 break; 812 813 /* Store to stack. */ 814 case BPF_ST: 815 case BPF_STX: 816 stack_off = fp->k * 4 + 4; 817 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) == 818 BPF_ST ? BPF_REG_A : BPF_REG_X, 819 -stack_off); 820 /* check_load_and_stores() verifies that classic BPF can 821 * load from stack only after write, so tracking 822 * stack_depth for ST|STX insns is enough 823 */ 824 if (new_prog && new_prog->aux->stack_depth < stack_off) 825 new_prog->aux->stack_depth = stack_off; 826 break; 827 828 /* Load from stack. */ 829 case BPF_LD | BPF_MEM: 830 case BPF_LDX | BPF_MEM: 831 stack_off = fp->k * 4 + 4; 832 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 833 BPF_REG_A : BPF_REG_X, BPF_REG_FP, 834 -stack_off); 835 break; 836 837 /* A = K or X = K */ 838 case BPF_LD | BPF_IMM: 839 case BPF_LDX | BPF_IMM: 840 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ? 841 BPF_REG_A : BPF_REG_X, fp->k); 842 break; 843 844 /* X = A */ 845 case BPF_MISC | BPF_TAX: 846 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 847 break; 848 849 /* A = X */ 850 case BPF_MISC | BPF_TXA: 851 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X); 852 break; 853 854 /* A = skb->len or X = skb->len */ 855 case BPF_LD | BPF_W | BPF_LEN: 856 case BPF_LDX | BPF_W | BPF_LEN: 857 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 858 BPF_REG_A : BPF_REG_X, BPF_REG_CTX, 859 offsetof(struct sk_buff, len)); 860 break; 861 862 /* Access seccomp_data fields. */ 863 case BPF_LDX | BPF_ABS | BPF_W: 864 /* A = *(u32 *) (ctx + K) */ 865 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k); 866 break; 867 868 /* Unknown instruction. */ 869 default: 870 goto err; 871 } 872 873 insn++; 874 if (new_prog) 875 memcpy(new_insn, tmp_insns, 876 sizeof(*insn) * (insn - tmp_insns)); 877 new_insn += insn - tmp_insns; 878 } 879 880 if (!new_prog) { 881 /* Only calculating new length. */ 882 *new_len = new_insn - first_insn; 883 if (*seen_ld_abs) 884 *new_len += 4; /* Prologue bits. */ 885 return 0; 886 } 887 888 pass++; 889 if (new_flen != new_insn - first_insn) { 890 new_flen = new_insn - first_insn; 891 if (pass > 2) 892 goto err; 893 goto do_pass; 894 } 895 896 kfree(addrs); 897 BUG_ON(*new_len != new_flen); 898 return 0; 899 err: 900 kfree(addrs); 901 return -EINVAL; 902 } 903 904 /* Security: 905 * 906 * As we dont want to clear mem[] array for each packet going through 907 * __bpf_prog_run(), we check that filter loaded by user never try to read 908 * a cell if not previously written, and we check all branches to be sure 909 * a malicious user doesn't try to abuse us. 910 */ 911 static int check_load_and_stores(const struct sock_filter *filter, int flen) 912 { 913 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */ 914 int pc, ret = 0; 915 916 BUILD_BUG_ON(BPF_MEMWORDS > 16); 917 918 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL); 919 if (!masks) 920 return -ENOMEM; 921 922 memset(masks, 0xff, flen * sizeof(*masks)); 923 924 for (pc = 0; pc < flen; pc++) { 925 memvalid &= masks[pc]; 926 927 switch (filter[pc].code) { 928 case BPF_ST: 929 case BPF_STX: 930 memvalid |= (1 << filter[pc].k); 931 break; 932 case BPF_LD | BPF_MEM: 933 case BPF_LDX | BPF_MEM: 934 if (!(memvalid & (1 << filter[pc].k))) { 935 ret = -EINVAL; 936 goto error; 937 } 938 break; 939 case BPF_JMP | BPF_JA: 940 /* A jump must set masks on target */ 941 masks[pc + 1 + filter[pc].k] &= memvalid; 942 memvalid = ~0; 943 break; 944 case BPF_JMP | BPF_JEQ | BPF_K: 945 case BPF_JMP | BPF_JEQ | BPF_X: 946 case BPF_JMP | BPF_JGE | BPF_K: 947 case BPF_JMP | BPF_JGE | BPF_X: 948 case BPF_JMP | BPF_JGT | BPF_K: 949 case BPF_JMP | BPF_JGT | BPF_X: 950 case BPF_JMP | BPF_JSET | BPF_K: 951 case BPF_JMP | BPF_JSET | BPF_X: 952 /* A jump must set masks on targets */ 953 masks[pc + 1 + filter[pc].jt] &= memvalid; 954 masks[pc + 1 + filter[pc].jf] &= memvalid; 955 memvalid = ~0; 956 break; 957 } 958 } 959 error: 960 kfree(masks); 961 return ret; 962 } 963 964 static bool chk_code_allowed(u16 code_to_probe) 965 { 966 static const bool codes[] = { 967 /* 32 bit ALU operations */ 968 [BPF_ALU | BPF_ADD | BPF_K] = true, 969 [BPF_ALU | BPF_ADD | BPF_X] = true, 970 [BPF_ALU | BPF_SUB | BPF_K] = true, 971 [BPF_ALU | BPF_SUB | BPF_X] = true, 972 [BPF_ALU | BPF_MUL | BPF_K] = true, 973 [BPF_ALU | BPF_MUL | BPF_X] = true, 974 [BPF_ALU | BPF_DIV | BPF_K] = true, 975 [BPF_ALU | BPF_DIV | BPF_X] = true, 976 [BPF_ALU | BPF_MOD | BPF_K] = true, 977 [BPF_ALU | BPF_MOD | BPF_X] = true, 978 [BPF_ALU | BPF_AND | BPF_K] = true, 979 [BPF_ALU | BPF_AND | BPF_X] = true, 980 [BPF_ALU | BPF_OR | BPF_K] = true, 981 [BPF_ALU | BPF_OR | BPF_X] = true, 982 [BPF_ALU | BPF_XOR | BPF_K] = true, 983 [BPF_ALU | BPF_XOR | BPF_X] = true, 984 [BPF_ALU | BPF_LSH | BPF_K] = true, 985 [BPF_ALU | BPF_LSH | BPF_X] = true, 986 [BPF_ALU | BPF_RSH | BPF_K] = true, 987 [BPF_ALU | BPF_RSH | BPF_X] = true, 988 [BPF_ALU | BPF_NEG] = true, 989 /* Load instructions */ 990 [BPF_LD | BPF_W | BPF_ABS] = true, 991 [BPF_LD | BPF_H | BPF_ABS] = true, 992 [BPF_LD | BPF_B | BPF_ABS] = true, 993 [BPF_LD | BPF_W | BPF_LEN] = true, 994 [BPF_LD | BPF_W | BPF_IND] = true, 995 [BPF_LD | BPF_H | BPF_IND] = true, 996 [BPF_LD | BPF_B | BPF_IND] = true, 997 [BPF_LD | BPF_IMM] = true, 998 [BPF_LD | BPF_MEM] = true, 999 [BPF_LDX | BPF_W | BPF_LEN] = true, 1000 [BPF_LDX | BPF_B | BPF_MSH] = true, 1001 [BPF_LDX | BPF_IMM] = true, 1002 [BPF_LDX | BPF_MEM] = true, 1003 /* Store instructions */ 1004 [BPF_ST] = true, 1005 [BPF_STX] = true, 1006 /* Misc instructions */ 1007 [BPF_MISC | BPF_TAX] = true, 1008 [BPF_MISC | BPF_TXA] = true, 1009 /* Return instructions */ 1010 [BPF_RET | BPF_K] = true, 1011 [BPF_RET | BPF_A] = true, 1012 /* Jump instructions */ 1013 [BPF_JMP | BPF_JA] = true, 1014 [BPF_JMP | BPF_JEQ | BPF_K] = true, 1015 [BPF_JMP | BPF_JEQ | BPF_X] = true, 1016 [BPF_JMP | BPF_JGE | BPF_K] = true, 1017 [BPF_JMP | BPF_JGE | BPF_X] = true, 1018 [BPF_JMP | BPF_JGT | BPF_K] = true, 1019 [BPF_JMP | BPF_JGT | BPF_X] = true, 1020 [BPF_JMP | BPF_JSET | BPF_K] = true, 1021 [BPF_JMP | BPF_JSET | BPF_X] = true, 1022 }; 1023 1024 if (code_to_probe >= ARRAY_SIZE(codes)) 1025 return false; 1026 1027 return codes[code_to_probe]; 1028 } 1029 1030 static bool bpf_check_basics_ok(const struct sock_filter *filter, 1031 unsigned int flen) 1032 { 1033 if (filter == NULL) 1034 return false; 1035 if (flen == 0 || flen > BPF_MAXINSNS) 1036 return false; 1037 1038 return true; 1039 } 1040 1041 /** 1042 * bpf_check_classic - verify socket filter code 1043 * @filter: filter to verify 1044 * @flen: length of filter 1045 * 1046 * Check the user's filter code. If we let some ugly 1047 * filter code slip through kaboom! The filter must contain 1048 * no references or jumps that are out of range, no illegal 1049 * instructions, and must end with a RET instruction. 1050 * 1051 * All jumps are forward as they are not signed. 1052 * 1053 * Returns 0 if the rule set is legal or -EINVAL if not. 1054 */ 1055 static int bpf_check_classic(const struct sock_filter *filter, 1056 unsigned int flen) 1057 { 1058 bool anc_found; 1059 int pc; 1060 1061 /* Check the filter code now */ 1062 for (pc = 0; pc < flen; pc++) { 1063 const struct sock_filter *ftest = &filter[pc]; 1064 1065 /* May we actually operate on this code? */ 1066 if (!chk_code_allowed(ftest->code)) 1067 return -EINVAL; 1068 1069 /* Some instructions need special checks */ 1070 switch (ftest->code) { 1071 case BPF_ALU | BPF_DIV | BPF_K: 1072 case BPF_ALU | BPF_MOD | BPF_K: 1073 /* Check for division by zero */ 1074 if (ftest->k == 0) 1075 return -EINVAL; 1076 break; 1077 case BPF_ALU | BPF_LSH | BPF_K: 1078 case BPF_ALU | BPF_RSH | BPF_K: 1079 if (ftest->k >= 32) 1080 return -EINVAL; 1081 break; 1082 case BPF_LD | BPF_MEM: 1083 case BPF_LDX | BPF_MEM: 1084 case BPF_ST: 1085 case BPF_STX: 1086 /* Check for invalid memory addresses */ 1087 if (ftest->k >= BPF_MEMWORDS) 1088 return -EINVAL; 1089 break; 1090 case BPF_JMP | BPF_JA: 1091 /* Note, the large ftest->k might cause loops. 1092 * Compare this with conditional jumps below, 1093 * where offsets are limited. --ANK (981016) 1094 */ 1095 if (ftest->k >= (unsigned int)(flen - pc - 1)) 1096 return -EINVAL; 1097 break; 1098 case BPF_JMP | BPF_JEQ | BPF_K: 1099 case BPF_JMP | BPF_JEQ | BPF_X: 1100 case BPF_JMP | BPF_JGE | BPF_K: 1101 case BPF_JMP | BPF_JGE | BPF_X: 1102 case BPF_JMP | BPF_JGT | BPF_K: 1103 case BPF_JMP | BPF_JGT | BPF_X: 1104 case BPF_JMP | BPF_JSET | BPF_K: 1105 case BPF_JMP | BPF_JSET | BPF_X: 1106 /* Both conditionals must be safe */ 1107 if (pc + ftest->jt + 1 >= flen || 1108 pc + ftest->jf + 1 >= flen) 1109 return -EINVAL; 1110 break; 1111 case BPF_LD | BPF_W | BPF_ABS: 1112 case BPF_LD | BPF_H | BPF_ABS: 1113 case BPF_LD | BPF_B | BPF_ABS: 1114 anc_found = false; 1115 if (bpf_anc_helper(ftest) & BPF_ANC) 1116 anc_found = true; 1117 /* Ancillary operation unknown or unsupported */ 1118 if (anc_found == false && ftest->k >= SKF_AD_OFF) 1119 return -EINVAL; 1120 } 1121 } 1122 1123 /* Last instruction must be a RET code */ 1124 switch (filter[flen - 1].code) { 1125 case BPF_RET | BPF_K: 1126 case BPF_RET | BPF_A: 1127 return check_load_and_stores(filter, flen); 1128 } 1129 1130 return -EINVAL; 1131 } 1132 1133 static int bpf_prog_store_orig_filter(struct bpf_prog *fp, 1134 const struct sock_fprog *fprog) 1135 { 1136 unsigned int fsize = bpf_classic_proglen(fprog); 1137 struct sock_fprog_kern *fkprog; 1138 1139 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL); 1140 if (!fp->orig_prog) 1141 return -ENOMEM; 1142 1143 fkprog = fp->orig_prog; 1144 fkprog->len = fprog->len; 1145 1146 fkprog->filter = kmemdup(fp->insns, fsize, 1147 GFP_KERNEL | __GFP_NOWARN); 1148 if (!fkprog->filter) { 1149 kfree(fp->orig_prog); 1150 return -ENOMEM; 1151 } 1152 1153 return 0; 1154 } 1155 1156 static void bpf_release_orig_filter(struct bpf_prog *fp) 1157 { 1158 struct sock_fprog_kern *fprog = fp->orig_prog; 1159 1160 if (fprog) { 1161 kfree(fprog->filter); 1162 kfree(fprog); 1163 } 1164 } 1165 1166 static void __bpf_prog_release(struct bpf_prog *prog) 1167 { 1168 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) { 1169 bpf_prog_put(prog); 1170 } else { 1171 bpf_release_orig_filter(prog); 1172 bpf_prog_free(prog); 1173 } 1174 } 1175 1176 static void __sk_filter_release(struct sk_filter *fp) 1177 { 1178 __bpf_prog_release(fp->prog); 1179 kfree(fp); 1180 } 1181 1182 /** 1183 * sk_filter_release_rcu - Release a socket filter by rcu_head 1184 * @rcu: rcu_head that contains the sk_filter to free 1185 */ 1186 static void sk_filter_release_rcu(struct rcu_head *rcu) 1187 { 1188 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); 1189 1190 __sk_filter_release(fp); 1191 } 1192 1193 /** 1194 * sk_filter_release - release a socket filter 1195 * @fp: filter to remove 1196 * 1197 * Remove a filter from a socket and release its resources. 1198 */ 1199 static void sk_filter_release(struct sk_filter *fp) 1200 { 1201 if (refcount_dec_and_test(&fp->refcnt)) 1202 call_rcu(&fp->rcu, sk_filter_release_rcu); 1203 } 1204 1205 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1206 { 1207 u32 filter_size = bpf_prog_size(fp->prog->len); 1208 1209 atomic_sub(filter_size, &sk->sk_omem_alloc); 1210 sk_filter_release(fp); 1211 } 1212 1213 /* try to charge the socket memory if there is space available 1214 * return true on success 1215 */ 1216 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1217 { 1218 u32 filter_size = bpf_prog_size(fp->prog->len); 1219 int optmem_max = READ_ONCE(sysctl_optmem_max); 1220 1221 /* same check as in sock_kmalloc() */ 1222 if (filter_size <= optmem_max && 1223 atomic_read(&sk->sk_omem_alloc) + filter_size < optmem_max) { 1224 atomic_add(filter_size, &sk->sk_omem_alloc); 1225 return true; 1226 } 1227 return false; 1228 } 1229 1230 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1231 { 1232 if (!refcount_inc_not_zero(&fp->refcnt)) 1233 return false; 1234 1235 if (!__sk_filter_charge(sk, fp)) { 1236 sk_filter_release(fp); 1237 return false; 1238 } 1239 return true; 1240 } 1241 1242 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp) 1243 { 1244 struct sock_filter *old_prog; 1245 struct bpf_prog *old_fp; 1246 int err, new_len, old_len = fp->len; 1247 bool seen_ld_abs = false; 1248 1249 /* We are free to overwrite insns et al right here as it won't be used at 1250 * this point in time anymore internally after the migration to the eBPF 1251 * instruction representation. 1252 */ 1253 BUILD_BUG_ON(sizeof(struct sock_filter) != 1254 sizeof(struct bpf_insn)); 1255 1256 /* Conversion cannot happen on overlapping memory areas, 1257 * so we need to keep the user BPF around until the 2nd 1258 * pass. At this time, the user BPF is stored in fp->insns. 1259 */ 1260 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter), 1261 GFP_KERNEL | __GFP_NOWARN); 1262 if (!old_prog) { 1263 err = -ENOMEM; 1264 goto out_err; 1265 } 1266 1267 /* 1st pass: calculate the new program length. */ 1268 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len, 1269 &seen_ld_abs); 1270 if (err) 1271 goto out_err_free; 1272 1273 /* Expand fp for appending the new filter representation. */ 1274 old_fp = fp; 1275 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0); 1276 if (!fp) { 1277 /* The old_fp is still around in case we couldn't 1278 * allocate new memory, so uncharge on that one. 1279 */ 1280 fp = old_fp; 1281 err = -ENOMEM; 1282 goto out_err_free; 1283 } 1284 1285 fp->len = new_len; 1286 1287 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */ 1288 err = bpf_convert_filter(old_prog, old_len, fp, &new_len, 1289 &seen_ld_abs); 1290 if (err) 1291 /* 2nd bpf_convert_filter() can fail only if it fails 1292 * to allocate memory, remapping must succeed. Note, 1293 * that at this time old_fp has already been released 1294 * by krealloc(). 1295 */ 1296 goto out_err_free; 1297 1298 fp = bpf_prog_select_runtime(fp, &err); 1299 if (err) 1300 goto out_err_free; 1301 1302 kfree(old_prog); 1303 return fp; 1304 1305 out_err_free: 1306 kfree(old_prog); 1307 out_err: 1308 __bpf_prog_release(fp); 1309 return ERR_PTR(err); 1310 } 1311 1312 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp, 1313 bpf_aux_classic_check_t trans) 1314 { 1315 int err; 1316 1317 fp->bpf_func = NULL; 1318 fp->jited = 0; 1319 1320 err = bpf_check_classic(fp->insns, fp->len); 1321 if (err) { 1322 __bpf_prog_release(fp); 1323 return ERR_PTR(err); 1324 } 1325 1326 /* There might be additional checks and transformations 1327 * needed on classic filters, f.e. in case of seccomp. 1328 */ 1329 if (trans) { 1330 err = trans(fp->insns, fp->len); 1331 if (err) { 1332 __bpf_prog_release(fp); 1333 return ERR_PTR(err); 1334 } 1335 } 1336 1337 /* Probe if we can JIT compile the filter and if so, do 1338 * the compilation of the filter. 1339 */ 1340 bpf_jit_compile(fp); 1341 1342 /* JIT compiler couldn't process this filter, so do the eBPF translation 1343 * for the optimized interpreter. 1344 */ 1345 if (!fp->jited) 1346 fp = bpf_migrate_filter(fp); 1347 1348 return fp; 1349 } 1350 1351 /** 1352 * bpf_prog_create - create an unattached filter 1353 * @pfp: the unattached filter that is created 1354 * @fprog: the filter program 1355 * 1356 * Create a filter independent of any socket. We first run some 1357 * sanity checks on it to make sure it does not explode on us later. 1358 * If an error occurs or there is insufficient memory for the filter 1359 * a negative errno code is returned. On success the return is zero. 1360 */ 1361 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog) 1362 { 1363 unsigned int fsize = bpf_classic_proglen(fprog); 1364 struct bpf_prog *fp; 1365 1366 /* Make sure new filter is there and in the right amounts. */ 1367 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1368 return -EINVAL; 1369 1370 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1371 if (!fp) 1372 return -ENOMEM; 1373 1374 memcpy(fp->insns, fprog->filter, fsize); 1375 1376 fp->len = fprog->len; 1377 /* Since unattached filters are not copied back to user 1378 * space through sk_get_filter(), we do not need to hold 1379 * a copy here, and can spare us the work. 1380 */ 1381 fp->orig_prog = NULL; 1382 1383 /* bpf_prepare_filter() already takes care of freeing 1384 * memory in case something goes wrong. 1385 */ 1386 fp = bpf_prepare_filter(fp, NULL); 1387 if (IS_ERR(fp)) 1388 return PTR_ERR(fp); 1389 1390 *pfp = fp; 1391 return 0; 1392 } 1393 EXPORT_SYMBOL_GPL(bpf_prog_create); 1394 1395 /** 1396 * bpf_prog_create_from_user - create an unattached filter from user buffer 1397 * @pfp: the unattached filter that is created 1398 * @fprog: the filter program 1399 * @trans: post-classic verifier transformation handler 1400 * @save_orig: save classic BPF program 1401 * 1402 * This function effectively does the same as bpf_prog_create(), only 1403 * that it builds up its insns buffer from user space provided buffer. 1404 * It also allows for passing a bpf_aux_classic_check_t handler. 1405 */ 1406 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 1407 bpf_aux_classic_check_t trans, bool save_orig) 1408 { 1409 unsigned int fsize = bpf_classic_proglen(fprog); 1410 struct bpf_prog *fp; 1411 int err; 1412 1413 /* Make sure new filter is there and in the right amounts. */ 1414 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1415 return -EINVAL; 1416 1417 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1418 if (!fp) 1419 return -ENOMEM; 1420 1421 if (copy_from_user(fp->insns, fprog->filter, fsize)) { 1422 __bpf_prog_free(fp); 1423 return -EFAULT; 1424 } 1425 1426 fp->len = fprog->len; 1427 fp->orig_prog = NULL; 1428 1429 if (save_orig) { 1430 err = bpf_prog_store_orig_filter(fp, fprog); 1431 if (err) { 1432 __bpf_prog_free(fp); 1433 return -ENOMEM; 1434 } 1435 } 1436 1437 /* bpf_prepare_filter() already takes care of freeing 1438 * memory in case something goes wrong. 1439 */ 1440 fp = bpf_prepare_filter(fp, trans); 1441 if (IS_ERR(fp)) 1442 return PTR_ERR(fp); 1443 1444 *pfp = fp; 1445 return 0; 1446 } 1447 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user); 1448 1449 void bpf_prog_destroy(struct bpf_prog *fp) 1450 { 1451 __bpf_prog_release(fp); 1452 } 1453 EXPORT_SYMBOL_GPL(bpf_prog_destroy); 1454 1455 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk) 1456 { 1457 struct sk_filter *fp, *old_fp; 1458 1459 fp = kmalloc(sizeof(*fp), GFP_KERNEL); 1460 if (!fp) 1461 return -ENOMEM; 1462 1463 fp->prog = prog; 1464 1465 if (!__sk_filter_charge(sk, fp)) { 1466 kfree(fp); 1467 return -ENOMEM; 1468 } 1469 refcount_set(&fp->refcnt, 1); 1470 1471 old_fp = rcu_dereference_protected(sk->sk_filter, 1472 lockdep_sock_is_held(sk)); 1473 rcu_assign_pointer(sk->sk_filter, fp); 1474 1475 if (old_fp) 1476 sk_filter_uncharge(sk, old_fp); 1477 1478 return 0; 1479 } 1480 1481 static 1482 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk) 1483 { 1484 unsigned int fsize = bpf_classic_proglen(fprog); 1485 struct bpf_prog *prog; 1486 int err; 1487 1488 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1489 return ERR_PTR(-EPERM); 1490 1491 /* Make sure new filter is there and in the right amounts. */ 1492 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1493 return ERR_PTR(-EINVAL); 1494 1495 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1496 if (!prog) 1497 return ERR_PTR(-ENOMEM); 1498 1499 if (copy_from_user(prog->insns, fprog->filter, fsize)) { 1500 __bpf_prog_free(prog); 1501 return ERR_PTR(-EFAULT); 1502 } 1503 1504 prog->len = fprog->len; 1505 1506 err = bpf_prog_store_orig_filter(prog, fprog); 1507 if (err) { 1508 __bpf_prog_free(prog); 1509 return ERR_PTR(-ENOMEM); 1510 } 1511 1512 /* bpf_prepare_filter() already takes care of freeing 1513 * memory in case something goes wrong. 1514 */ 1515 return bpf_prepare_filter(prog, NULL); 1516 } 1517 1518 /** 1519 * sk_attach_filter - attach a socket filter 1520 * @fprog: the filter program 1521 * @sk: the socket to use 1522 * 1523 * Attach the user's filter code. We first run some sanity checks on 1524 * it to make sure it does not explode on us later. If an error 1525 * occurs or there is insufficient memory for the filter a negative 1526 * errno code is returned. On success the return is zero. 1527 */ 1528 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1529 { 1530 struct bpf_prog *prog = __get_filter(fprog, sk); 1531 int err; 1532 1533 if (IS_ERR(prog)) 1534 return PTR_ERR(prog); 1535 1536 err = __sk_attach_prog(prog, sk); 1537 if (err < 0) { 1538 __bpf_prog_release(prog); 1539 return err; 1540 } 1541 1542 return 0; 1543 } 1544 EXPORT_SYMBOL_GPL(sk_attach_filter); 1545 1546 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1547 { 1548 struct bpf_prog *prog = __get_filter(fprog, sk); 1549 int err; 1550 1551 if (IS_ERR(prog)) 1552 return PTR_ERR(prog); 1553 1554 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) 1555 err = -ENOMEM; 1556 else 1557 err = reuseport_attach_prog(sk, prog); 1558 1559 if (err) 1560 __bpf_prog_release(prog); 1561 1562 return err; 1563 } 1564 1565 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk) 1566 { 1567 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1568 return ERR_PTR(-EPERM); 1569 1570 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1571 } 1572 1573 int sk_attach_bpf(u32 ufd, struct sock *sk) 1574 { 1575 struct bpf_prog *prog = __get_bpf(ufd, sk); 1576 int err; 1577 1578 if (IS_ERR(prog)) 1579 return PTR_ERR(prog); 1580 1581 err = __sk_attach_prog(prog, sk); 1582 if (err < 0) { 1583 bpf_prog_put(prog); 1584 return err; 1585 } 1586 1587 return 0; 1588 } 1589 1590 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk) 1591 { 1592 struct bpf_prog *prog; 1593 int err; 1594 1595 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1596 return -EPERM; 1597 1598 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1599 if (PTR_ERR(prog) == -EINVAL) 1600 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT); 1601 if (IS_ERR(prog)) 1602 return PTR_ERR(prog); 1603 1604 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) { 1605 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER 1606 * bpf prog (e.g. sockmap). It depends on the 1607 * limitation imposed by bpf_prog_load(). 1608 * Hence, sysctl_optmem_max is not checked. 1609 */ 1610 if ((sk->sk_type != SOCK_STREAM && 1611 sk->sk_type != SOCK_DGRAM) || 1612 (sk->sk_protocol != IPPROTO_UDP && 1613 sk->sk_protocol != IPPROTO_TCP) || 1614 (sk->sk_family != AF_INET && 1615 sk->sk_family != AF_INET6)) { 1616 err = -ENOTSUPP; 1617 goto err_prog_put; 1618 } 1619 } else { 1620 /* BPF_PROG_TYPE_SOCKET_FILTER */ 1621 if (bpf_prog_size(prog->len) > READ_ONCE(sysctl_optmem_max)) { 1622 err = -ENOMEM; 1623 goto err_prog_put; 1624 } 1625 } 1626 1627 err = reuseport_attach_prog(sk, prog); 1628 err_prog_put: 1629 if (err) 1630 bpf_prog_put(prog); 1631 1632 return err; 1633 } 1634 1635 void sk_reuseport_prog_free(struct bpf_prog *prog) 1636 { 1637 if (!prog) 1638 return; 1639 1640 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) 1641 bpf_prog_put(prog); 1642 else 1643 bpf_prog_destroy(prog); 1644 } 1645 1646 struct bpf_scratchpad { 1647 union { 1648 __be32 diff[MAX_BPF_STACK / sizeof(__be32)]; 1649 u8 buff[MAX_BPF_STACK]; 1650 }; 1651 }; 1652 1653 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp); 1654 1655 static inline int __bpf_try_make_writable(struct sk_buff *skb, 1656 unsigned int write_len) 1657 { 1658 return skb_ensure_writable(skb, write_len); 1659 } 1660 1661 static inline int bpf_try_make_writable(struct sk_buff *skb, 1662 unsigned int write_len) 1663 { 1664 int err = __bpf_try_make_writable(skb, write_len); 1665 1666 bpf_compute_data_pointers(skb); 1667 return err; 1668 } 1669 1670 static int bpf_try_make_head_writable(struct sk_buff *skb) 1671 { 1672 return bpf_try_make_writable(skb, skb_headlen(skb)); 1673 } 1674 1675 static inline void bpf_push_mac_rcsum(struct sk_buff *skb) 1676 { 1677 if (skb_at_tc_ingress(skb)) 1678 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1679 } 1680 1681 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb) 1682 { 1683 if (skb_at_tc_ingress(skb)) 1684 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1685 } 1686 1687 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset, 1688 const void *, from, u32, len, u64, flags) 1689 { 1690 void *ptr; 1691 1692 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH))) 1693 return -EINVAL; 1694 if (unlikely(offset > INT_MAX)) 1695 return -EFAULT; 1696 if (unlikely(bpf_try_make_writable(skb, offset + len))) 1697 return -EFAULT; 1698 1699 ptr = skb->data + offset; 1700 if (flags & BPF_F_RECOMPUTE_CSUM) 1701 __skb_postpull_rcsum(skb, ptr, len, offset); 1702 1703 memcpy(ptr, from, len); 1704 1705 if (flags & BPF_F_RECOMPUTE_CSUM) 1706 __skb_postpush_rcsum(skb, ptr, len, offset); 1707 if (flags & BPF_F_INVALIDATE_HASH) 1708 skb_clear_hash(skb); 1709 1710 return 0; 1711 } 1712 1713 static const struct bpf_func_proto bpf_skb_store_bytes_proto = { 1714 .func = bpf_skb_store_bytes, 1715 .gpl_only = false, 1716 .ret_type = RET_INTEGER, 1717 .arg1_type = ARG_PTR_TO_CTX, 1718 .arg2_type = ARG_ANYTHING, 1719 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 1720 .arg4_type = ARG_CONST_SIZE, 1721 .arg5_type = ARG_ANYTHING, 1722 }; 1723 1724 int __bpf_skb_store_bytes(struct sk_buff *skb, u32 offset, const void *from, 1725 u32 len, u64 flags) 1726 { 1727 return ____bpf_skb_store_bytes(skb, offset, from, len, flags); 1728 } 1729 1730 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset, 1731 void *, to, u32, len) 1732 { 1733 void *ptr; 1734 1735 if (unlikely(offset > INT_MAX)) 1736 goto err_clear; 1737 1738 ptr = skb_header_pointer(skb, offset, len, to); 1739 if (unlikely(!ptr)) 1740 goto err_clear; 1741 if (ptr != to) 1742 memcpy(to, ptr, len); 1743 1744 return 0; 1745 err_clear: 1746 memset(to, 0, len); 1747 return -EFAULT; 1748 } 1749 1750 static const struct bpf_func_proto bpf_skb_load_bytes_proto = { 1751 .func = bpf_skb_load_bytes, 1752 .gpl_only = false, 1753 .ret_type = RET_INTEGER, 1754 .arg1_type = ARG_PTR_TO_CTX, 1755 .arg2_type = ARG_ANYTHING, 1756 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1757 .arg4_type = ARG_CONST_SIZE, 1758 }; 1759 1760 int __bpf_skb_load_bytes(const struct sk_buff *skb, u32 offset, void *to, u32 len) 1761 { 1762 return ____bpf_skb_load_bytes(skb, offset, to, len); 1763 } 1764 1765 BPF_CALL_4(bpf_flow_dissector_load_bytes, 1766 const struct bpf_flow_dissector *, ctx, u32, offset, 1767 void *, to, u32, len) 1768 { 1769 void *ptr; 1770 1771 if (unlikely(offset > 0xffff)) 1772 goto err_clear; 1773 1774 if (unlikely(!ctx->skb)) 1775 goto err_clear; 1776 1777 ptr = skb_header_pointer(ctx->skb, offset, len, to); 1778 if (unlikely(!ptr)) 1779 goto err_clear; 1780 if (ptr != to) 1781 memcpy(to, ptr, len); 1782 1783 return 0; 1784 err_clear: 1785 memset(to, 0, len); 1786 return -EFAULT; 1787 } 1788 1789 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = { 1790 .func = bpf_flow_dissector_load_bytes, 1791 .gpl_only = false, 1792 .ret_type = RET_INTEGER, 1793 .arg1_type = ARG_PTR_TO_CTX, 1794 .arg2_type = ARG_ANYTHING, 1795 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1796 .arg4_type = ARG_CONST_SIZE, 1797 }; 1798 1799 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb, 1800 u32, offset, void *, to, u32, len, u32, start_header) 1801 { 1802 u8 *end = skb_tail_pointer(skb); 1803 u8 *start, *ptr; 1804 1805 if (unlikely(offset > 0xffff)) 1806 goto err_clear; 1807 1808 switch (start_header) { 1809 case BPF_HDR_START_MAC: 1810 if (unlikely(!skb_mac_header_was_set(skb))) 1811 goto err_clear; 1812 start = skb_mac_header(skb); 1813 break; 1814 case BPF_HDR_START_NET: 1815 start = skb_network_header(skb); 1816 break; 1817 default: 1818 goto err_clear; 1819 } 1820 1821 ptr = start + offset; 1822 1823 if (likely(ptr + len <= end)) { 1824 memcpy(to, ptr, len); 1825 return 0; 1826 } 1827 1828 err_clear: 1829 memset(to, 0, len); 1830 return -EFAULT; 1831 } 1832 1833 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = { 1834 .func = bpf_skb_load_bytes_relative, 1835 .gpl_only = false, 1836 .ret_type = RET_INTEGER, 1837 .arg1_type = ARG_PTR_TO_CTX, 1838 .arg2_type = ARG_ANYTHING, 1839 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1840 .arg4_type = ARG_CONST_SIZE, 1841 .arg5_type = ARG_ANYTHING, 1842 }; 1843 1844 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len) 1845 { 1846 /* Idea is the following: should the needed direct read/write 1847 * test fail during runtime, we can pull in more data and redo 1848 * again, since implicitly, we invalidate previous checks here. 1849 * 1850 * Or, since we know how much we need to make read/writeable, 1851 * this can be done once at the program beginning for direct 1852 * access case. By this we overcome limitations of only current 1853 * headroom being accessible. 1854 */ 1855 return bpf_try_make_writable(skb, len ? : skb_headlen(skb)); 1856 } 1857 1858 static const struct bpf_func_proto bpf_skb_pull_data_proto = { 1859 .func = bpf_skb_pull_data, 1860 .gpl_only = false, 1861 .ret_type = RET_INTEGER, 1862 .arg1_type = ARG_PTR_TO_CTX, 1863 .arg2_type = ARG_ANYTHING, 1864 }; 1865 1866 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk) 1867 { 1868 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL; 1869 } 1870 1871 static const struct bpf_func_proto bpf_sk_fullsock_proto = { 1872 .func = bpf_sk_fullsock, 1873 .gpl_only = false, 1874 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 1875 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 1876 }; 1877 1878 static inline int sk_skb_try_make_writable(struct sk_buff *skb, 1879 unsigned int write_len) 1880 { 1881 return __bpf_try_make_writable(skb, write_len); 1882 } 1883 1884 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len) 1885 { 1886 /* Idea is the following: should the needed direct read/write 1887 * test fail during runtime, we can pull in more data and redo 1888 * again, since implicitly, we invalidate previous checks here. 1889 * 1890 * Or, since we know how much we need to make read/writeable, 1891 * this can be done once at the program beginning for direct 1892 * access case. By this we overcome limitations of only current 1893 * headroom being accessible. 1894 */ 1895 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb)); 1896 } 1897 1898 static const struct bpf_func_proto sk_skb_pull_data_proto = { 1899 .func = sk_skb_pull_data, 1900 .gpl_only = false, 1901 .ret_type = RET_INTEGER, 1902 .arg1_type = ARG_PTR_TO_CTX, 1903 .arg2_type = ARG_ANYTHING, 1904 }; 1905 1906 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset, 1907 u64, from, u64, to, u64, flags) 1908 { 1909 __sum16 *ptr; 1910 1911 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK))) 1912 return -EINVAL; 1913 if (unlikely(offset > 0xffff || offset & 1)) 1914 return -EFAULT; 1915 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1916 return -EFAULT; 1917 1918 ptr = (__sum16 *)(skb->data + offset); 1919 switch (flags & BPF_F_HDR_FIELD_MASK) { 1920 case 0: 1921 if (unlikely(from != 0)) 1922 return -EINVAL; 1923 1924 csum_replace_by_diff(ptr, to); 1925 break; 1926 case 2: 1927 csum_replace2(ptr, from, to); 1928 break; 1929 case 4: 1930 csum_replace4(ptr, from, to); 1931 break; 1932 default: 1933 return -EINVAL; 1934 } 1935 1936 return 0; 1937 } 1938 1939 static const struct bpf_func_proto bpf_l3_csum_replace_proto = { 1940 .func = bpf_l3_csum_replace, 1941 .gpl_only = false, 1942 .ret_type = RET_INTEGER, 1943 .arg1_type = ARG_PTR_TO_CTX, 1944 .arg2_type = ARG_ANYTHING, 1945 .arg3_type = ARG_ANYTHING, 1946 .arg4_type = ARG_ANYTHING, 1947 .arg5_type = ARG_ANYTHING, 1948 }; 1949 1950 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset, 1951 u64, from, u64, to, u64, flags) 1952 { 1953 bool is_pseudo = flags & BPF_F_PSEUDO_HDR; 1954 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0; 1955 bool do_mforce = flags & BPF_F_MARK_ENFORCE; 1956 __sum16 *ptr; 1957 1958 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE | 1959 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK))) 1960 return -EINVAL; 1961 if (unlikely(offset > 0xffff || offset & 1)) 1962 return -EFAULT; 1963 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1964 return -EFAULT; 1965 1966 ptr = (__sum16 *)(skb->data + offset); 1967 if (is_mmzero && !do_mforce && !*ptr) 1968 return 0; 1969 1970 switch (flags & BPF_F_HDR_FIELD_MASK) { 1971 case 0: 1972 if (unlikely(from != 0)) 1973 return -EINVAL; 1974 1975 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo); 1976 break; 1977 case 2: 1978 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo); 1979 break; 1980 case 4: 1981 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo); 1982 break; 1983 default: 1984 return -EINVAL; 1985 } 1986 1987 if (is_mmzero && !*ptr) 1988 *ptr = CSUM_MANGLED_0; 1989 return 0; 1990 } 1991 1992 static const struct bpf_func_proto bpf_l4_csum_replace_proto = { 1993 .func = bpf_l4_csum_replace, 1994 .gpl_only = false, 1995 .ret_type = RET_INTEGER, 1996 .arg1_type = ARG_PTR_TO_CTX, 1997 .arg2_type = ARG_ANYTHING, 1998 .arg3_type = ARG_ANYTHING, 1999 .arg4_type = ARG_ANYTHING, 2000 .arg5_type = ARG_ANYTHING, 2001 }; 2002 2003 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size, 2004 __be32 *, to, u32, to_size, __wsum, seed) 2005 { 2006 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp); 2007 u32 diff_size = from_size + to_size; 2008 int i, j = 0; 2009 2010 /* This is quite flexible, some examples: 2011 * 2012 * from_size == 0, to_size > 0, seed := csum --> pushing data 2013 * from_size > 0, to_size == 0, seed := csum --> pulling data 2014 * from_size > 0, to_size > 0, seed := 0 --> diffing data 2015 * 2016 * Even for diffing, from_size and to_size don't need to be equal. 2017 */ 2018 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) || 2019 diff_size > sizeof(sp->diff))) 2020 return -EINVAL; 2021 2022 for (i = 0; i < from_size / sizeof(__be32); i++, j++) 2023 sp->diff[j] = ~from[i]; 2024 for (i = 0; i < to_size / sizeof(__be32); i++, j++) 2025 sp->diff[j] = to[i]; 2026 2027 return csum_partial(sp->diff, diff_size, seed); 2028 } 2029 2030 static const struct bpf_func_proto bpf_csum_diff_proto = { 2031 .func = bpf_csum_diff, 2032 .gpl_only = false, 2033 .pkt_access = true, 2034 .ret_type = RET_INTEGER, 2035 .arg1_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 2036 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 2037 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 2038 .arg4_type = ARG_CONST_SIZE_OR_ZERO, 2039 .arg5_type = ARG_ANYTHING, 2040 }; 2041 2042 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum) 2043 { 2044 /* The interface is to be used in combination with bpf_csum_diff() 2045 * for direct packet writes. csum rotation for alignment as well 2046 * as emulating csum_sub() can be done from the eBPF program. 2047 */ 2048 if (skb->ip_summed == CHECKSUM_COMPLETE) 2049 return (skb->csum = csum_add(skb->csum, csum)); 2050 2051 return -ENOTSUPP; 2052 } 2053 2054 static const struct bpf_func_proto bpf_csum_update_proto = { 2055 .func = bpf_csum_update, 2056 .gpl_only = false, 2057 .ret_type = RET_INTEGER, 2058 .arg1_type = ARG_PTR_TO_CTX, 2059 .arg2_type = ARG_ANYTHING, 2060 }; 2061 2062 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level) 2063 { 2064 /* The interface is to be used in combination with bpf_skb_adjust_room() 2065 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET 2066 * is passed as flags, for example. 2067 */ 2068 switch (level) { 2069 case BPF_CSUM_LEVEL_INC: 2070 __skb_incr_checksum_unnecessary(skb); 2071 break; 2072 case BPF_CSUM_LEVEL_DEC: 2073 __skb_decr_checksum_unnecessary(skb); 2074 break; 2075 case BPF_CSUM_LEVEL_RESET: 2076 __skb_reset_checksum_unnecessary(skb); 2077 break; 2078 case BPF_CSUM_LEVEL_QUERY: 2079 return skb->ip_summed == CHECKSUM_UNNECESSARY ? 2080 skb->csum_level : -EACCES; 2081 default: 2082 return -EINVAL; 2083 } 2084 2085 return 0; 2086 } 2087 2088 static const struct bpf_func_proto bpf_csum_level_proto = { 2089 .func = bpf_csum_level, 2090 .gpl_only = false, 2091 .ret_type = RET_INTEGER, 2092 .arg1_type = ARG_PTR_TO_CTX, 2093 .arg2_type = ARG_ANYTHING, 2094 }; 2095 2096 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb) 2097 { 2098 return dev_forward_skb_nomtu(dev, skb); 2099 } 2100 2101 static inline int __bpf_rx_skb_no_mac(struct net_device *dev, 2102 struct sk_buff *skb) 2103 { 2104 int ret = ____dev_forward_skb(dev, skb, false); 2105 2106 if (likely(!ret)) { 2107 skb->dev = dev; 2108 ret = netif_rx(skb); 2109 } 2110 2111 return ret; 2112 } 2113 2114 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb) 2115 { 2116 int ret; 2117 2118 if (dev_xmit_recursion()) { 2119 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2120 kfree_skb(skb); 2121 return -ENETDOWN; 2122 } 2123 2124 skb->dev = dev; 2125 skb_set_redirected_noclear(skb, skb_at_tc_ingress(skb)); 2126 skb_clear_tstamp(skb); 2127 2128 dev_xmit_recursion_inc(); 2129 ret = dev_queue_xmit(skb); 2130 dev_xmit_recursion_dec(); 2131 2132 return ret; 2133 } 2134 2135 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev, 2136 u32 flags) 2137 { 2138 unsigned int mlen = skb_network_offset(skb); 2139 2140 if (unlikely(skb->len <= mlen)) { 2141 kfree_skb(skb); 2142 return -ERANGE; 2143 } 2144 2145 if (mlen) { 2146 __skb_pull(skb, mlen); 2147 2148 /* At ingress, the mac header has already been pulled once. 2149 * At egress, skb_pospull_rcsum has to be done in case that 2150 * the skb is originated from ingress (i.e. a forwarded skb) 2151 * to ensure that rcsum starts at net header. 2152 */ 2153 if (!skb_at_tc_ingress(skb)) 2154 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen); 2155 } 2156 skb_pop_mac_header(skb); 2157 skb_reset_mac_len(skb); 2158 return flags & BPF_F_INGRESS ? 2159 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb); 2160 } 2161 2162 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev, 2163 u32 flags) 2164 { 2165 /* Verify that a link layer header is carried */ 2166 if (unlikely(skb->mac_header >= skb->network_header || skb->len == 0)) { 2167 kfree_skb(skb); 2168 return -ERANGE; 2169 } 2170 2171 bpf_push_mac_rcsum(skb); 2172 return flags & BPF_F_INGRESS ? 2173 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb); 2174 } 2175 2176 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev, 2177 u32 flags) 2178 { 2179 if (dev_is_mac_header_xmit(dev)) 2180 return __bpf_redirect_common(skb, dev, flags); 2181 else 2182 return __bpf_redirect_no_mac(skb, dev, flags); 2183 } 2184 2185 #if IS_ENABLED(CONFIG_IPV6) 2186 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb, 2187 struct net_device *dev, struct bpf_nh_params *nh) 2188 { 2189 u32 hh_len = LL_RESERVED_SPACE(dev); 2190 const struct in6_addr *nexthop; 2191 struct dst_entry *dst = NULL; 2192 struct neighbour *neigh; 2193 2194 if (dev_xmit_recursion()) { 2195 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2196 goto out_drop; 2197 } 2198 2199 skb->dev = dev; 2200 skb_clear_tstamp(skb); 2201 2202 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 2203 skb = skb_expand_head(skb, hh_len); 2204 if (!skb) 2205 return -ENOMEM; 2206 } 2207 2208 rcu_read_lock(); 2209 if (!nh) { 2210 dst = skb_dst(skb); 2211 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst), 2212 &ipv6_hdr(skb)->daddr); 2213 } else { 2214 nexthop = &nh->ipv6_nh; 2215 } 2216 neigh = ip_neigh_gw6(dev, nexthop); 2217 if (likely(!IS_ERR(neigh))) { 2218 int ret; 2219 2220 sock_confirm_neigh(skb, neigh); 2221 local_bh_disable(); 2222 dev_xmit_recursion_inc(); 2223 ret = neigh_output(neigh, skb, false); 2224 dev_xmit_recursion_dec(); 2225 local_bh_enable(); 2226 rcu_read_unlock(); 2227 return ret; 2228 } 2229 rcu_read_unlock_bh(); 2230 if (dst) 2231 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 2232 out_drop: 2233 kfree_skb(skb); 2234 return -ENETDOWN; 2235 } 2236 2237 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev, 2238 struct bpf_nh_params *nh) 2239 { 2240 const struct ipv6hdr *ip6h = ipv6_hdr(skb); 2241 struct net *net = dev_net(dev); 2242 int err, ret = NET_XMIT_DROP; 2243 2244 if (!nh) { 2245 struct dst_entry *dst; 2246 struct flowi6 fl6 = { 2247 .flowi6_flags = FLOWI_FLAG_ANYSRC, 2248 .flowi6_mark = skb->mark, 2249 .flowlabel = ip6_flowinfo(ip6h), 2250 .flowi6_oif = dev->ifindex, 2251 .flowi6_proto = ip6h->nexthdr, 2252 .daddr = ip6h->daddr, 2253 .saddr = ip6h->saddr, 2254 }; 2255 2256 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL); 2257 if (IS_ERR(dst)) 2258 goto out_drop; 2259 2260 skb_dst_set(skb, dst); 2261 } else if (nh->nh_family != AF_INET6) { 2262 goto out_drop; 2263 } 2264 2265 err = bpf_out_neigh_v6(net, skb, dev, nh); 2266 if (unlikely(net_xmit_eval(err))) 2267 dev->stats.tx_errors++; 2268 else 2269 ret = NET_XMIT_SUCCESS; 2270 goto out_xmit; 2271 out_drop: 2272 dev->stats.tx_errors++; 2273 kfree_skb(skb); 2274 out_xmit: 2275 return ret; 2276 } 2277 #else 2278 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev, 2279 struct bpf_nh_params *nh) 2280 { 2281 kfree_skb(skb); 2282 return NET_XMIT_DROP; 2283 } 2284 #endif /* CONFIG_IPV6 */ 2285 2286 #if IS_ENABLED(CONFIG_INET) 2287 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb, 2288 struct net_device *dev, struct bpf_nh_params *nh) 2289 { 2290 u32 hh_len = LL_RESERVED_SPACE(dev); 2291 struct neighbour *neigh; 2292 bool is_v6gw = false; 2293 2294 if (dev_xmit_recursion()) { 2295 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2296 goto out_drop; 2297 } 2298 2299 skb->dev = dev; 2300 skb_clear_tstamp(skb); 2301 2302 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 2303 skb = skb_expand_head(skb, hh_len); 2304 if (!skb) 2305 return -ENOMEM; 2306 } 2307 2308 rcu_read_lock(); 2309 if (!nh) { 2310 struct dst_entry *dst = skb_dst(skb); 2311 struct rtable *rt = container_of(dst, struct rtable, dst); 2312 2313 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 2314 } else if (nh->nh_family == AF_INET6) { 2315 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh); 2316 is_v6gw = true; 2317 } else if (nh->nh_family == AF_INET) { 2318 neigh = ip_neigh_gw4(dev, nh->ipv4_nh); 2319 } else { 2320 rcu_read_unlock(); 2321 goto out_drop; 2322 } 2323 2324 if (likely(!IS_ERR(neigh))) { 2325 int ret; 2326 2327 sock_confirm_neigh(skb, neigh); 2328 local_bh_disable(); 2329 dev_xmit_recursion_inc(); 2330 ret = neigh_output(neigh, skb, is_v6gw); 2331 dev_xmit_recursion_dec(); 2332 local_bh_enable(); 2333 rcu_read_unlock(); 2334 return ret; 2335 } 2336 rcu_read_unlock(); 2337 out_drop: 2338 kfree_skb(skb); 2339 return -ENETDOWN; 2340 } 2341 2342 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev, 2343 struct bpf_nh_params *nh) 2344 { 2345 const struct iphdr *ip4h = ip_hdr(skb); 2346 struct net *net = dev_net(dev); 2347 int err, ret = NET_XMIT_DROP; 2348 2349 if (!nh) { 2350 struct flowi4 fl4 = { 2351 .flowi4_flags = FLOWI_FLAG_ANYSRC, 2352 .flowi4_mark = skb->mark, 2353 .flowi4_tos = RT_TOS(ip4h->tos), 2354 .flowi4_oif = dev->ifindex, 2355 .flowi4_proto = ip4h->protocol, 2356 .daddr = ip4h->daddr, 2357 .saddr = ip4h->saddr, 2358 }; 2359 struct rtable *rt; 2360 2361 rt = ip_route_output_flow(net, &fl4, NULL); 2362 if (IS_ERR(rt)) 2363 goto out_drop; 2364 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { 2365 ip_rt_put(rt); 2366 goto out_drop; 2367 } 2368 2369 skb_dst_set(skb, &rt->dst); 2370 } 2371 2372 err = bpf_out_neigh_v4(net, skb, dev, nh); 2373 if (unlikely(net_xmit_eval(err))) 2374 dev->stats.tx_errors++; 2375 else 2376 ret = NET_XMIT_SUCCESS; 2377 goto out_xmit; 2378 out_drop: 2379 dev->stats.tx_errors++; 2380 kfree_skb(skb); 2381 out_xmit: 2382 return ret; 2383 } 2384 #else 2385 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev, 2386 struct bpf_nh_params *nh) 2387 { 2388 kfree_skb(skb); 2389 return NET_XMIT_DROP; 2390 } 2391 #endif /* CONFIG_INET */ 2392 2393 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev, 2394 struct bpf_nh_params *nh) 2395 { 2396 struct ethhdr *ethh = eth_hdr(skb); 2397 2398 if (unlikely(skb->mac_header >= skb->network_header)) 2399 goto out; 2400 bpf_push_mac_rcsum(skb); 2401 if (is_multicast_ether_addr(ethh->h_dest)) 2402 goto out; 2403 2404 skb_pull(skb, sizeof(*ethh)); 2405 skb_unset_mac_header(skb); 2406 skb_reset_network_header(skb); 2407 2408 if (skb->protocol == htons(ETH_P_IP)) 2409 return __bpf_redirect_neigh_v4(skb, dev, nh); 2410 else if (skb->protocol == htons(ETH_P_IPV6)) 2411 return __bpf_redirect_neigh_v6(skb, dev, nh); 2412 out: 2413 kfree_skb(skb); 2414 return -ENOTSUPP; 2415 } 2416 2417 /* Internal, non-exposed redirect flags. */ 2418 enum { 2419 BPF_F_NEIGH = (1ULL << 1), 2420 BPF_F_PEER = (1ULL << 2), 2421 BPF_F_NEXTHOP = (1ULL << 3), 2422 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP) 2423 }; 2424 2425 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags) 2426 { 2427 struct net_device *dev; 2428 struct sk_buff *clone; 2429 int ret; 2430 2431 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL))) 2432 return -EINVAL; 2433 2434 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex); 2435 if (unlikely(!dev)) 2436 return -EINVAL; 2437 2438 clone = skb_clone(skb, GFP_ATOMIC); 2439 if (unlikely(!clone)) 2440 return -ENOMEM; 2441 2442 /* For direct write, we need to keep the invariant that the skbs 2443 * we're dealing with need to be uncloned. Should uncloning fail 2444 * here, we need to free the just generated clone to unclone once 2445 * again. 2446 */ 2447 ret = bpf_try_make_head_writable(skb); 2448 if (unlikely(ret)) { 2449 kfree_skb(clone); 2450 return -ENOMEM; 2451 } 2452 2453 return __bpf_redirect(clone, dev, flags); 2454 } 2455 2456 static const struct bpf_func_proto bpf_clone_redirect_proto = { 2457 .func = bpf_clone_redirect, 2458 .gpl_only = false, 2459 .ret_type = RET_INTEGER, 2460 .arg1_type = ARG_PTR_TO_CTX, 2461 .arg2_type = ARG_ANYTHING, 2462 .arg3_type = ARG_ANYTHING, 2463 }; 2464 2465 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 2466 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info); 2467 2468 int skb_do_redirect(struct sk_buff *skb) 2469 { 2470 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2471 struct net *net = dev_net(skb->dev); 2472 struct net_device *dev; 2473 u32 flags = ri->flags; 2474 2475 dev = dev_get_by_index_rcu(net, ri->tgt_index); 2476 ri->tgt_index = 0; 2477 ri->flags = 0; 2478 if (unlikely(!dev)) 2479 goto out_drop; 2480 if (flags & BPF_F_PEER) { 2481 const struct net_device_ops *ops = dev->netdev_ops; 2482 2483 if (unlikely(!ops->ndo_get_peer_dev || 2484 !skb_at_tc_ingress(skb))) 2485 goto out_drop; 2486 dev = ops->ndo_get_peer_dev(dev); 2487 if (unlikely(!dev || 2488 !(dev->flags & IFF_UP) || 2489 net_eq(net, dev_net(dev)))) 2490 goto out_drop; 2491 skb->dev = dev; 2492 return -EAGAIN; 2493 } 2494 return flags & BPF_F_NEIGH ? 2495 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ? 2496 &ri->nh : NULL) : 2497 __bpf_redirect(skb, dev, flags); 2498 out_drop: 2499 kfree_skb(skb); 2500 return -EINVAL; 2501 } 2502 2503 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags) 2504 { 2505 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2506 2507 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL))) 2508 return TC_ACT_SHOT; 2509 2510 ri->flags = flags; 2511 ri->tgt_index = ifindex; 2512 2513 return TC_ACT_REDIRECT; 2514 } 2515 2516 static const struct bpf_func_proto bpf_redirect_proto = { 2517 .func = bpf_redirect, 2518 .gpl_only = false, 2519 .ret_type = RET_INTEGER, 2520 .arg1_type = ARG_ANYTHING, 2521 .arg2_type = ARG_ANYTHING, 2522 }; 2523 2524 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags) 2525 { 2526 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2527 2528 if (unlikely(flags)) 2529 return TC_ACT_SHOT; 2530 2531 ri->flags = BPF_F_PEER; 2532 ri->tgt_index = ifindex; 2533 2534 return TC_ACT_REDIRECT; 2535 } 2536 2537 static const struct bpf_func_proto bpf_redirect_peer_proto = { 2538 .func = bpf_redirect_peer, 2539 .gpl_only = false, 2540 .ret_type = RET_INTEGER, 2541 .arg1_type = ARG_ANYTHING, 2542 .arg2_type = ARG_ANYTHING, 2543 }; 2544 2545 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params, 2546 int, plen, u64, flags) 2547 { 2548 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2549 2550 if (unlikely((plen && plen < sizeof(*params)) || flags)) 2551 return TC_ACT_SHOT; 2552 2553 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0); 2554 ri->tgt_index = ifindex; 2555 2556 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params)); 2557 if (plen) 2558 memcpy(&ri->nh, params, sizeof(ri->nh)); 2559 2560 return TC_ACT_REDIRECT; 2561 } 2562 2563 static const struct bpf_func_proto bpf_redirect_neigh_proto = { 2564 .func = bpf_redirect_neigh, 2565 .gpl_only = false, 2566 .ret_type = RET_INTEGER, 2567 .arg1_type = ARG_ANYTHING, 2568 .arg2_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 2569 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 2570 .arg4_type = ARG_ANYTHING, 2571 }; 2572 2573 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes) 2574 { 2575 msg->apply_bytes = bytes; 2576 return 0; 2577 } 2578 2579 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = { 2580 .func = bpf_msg_apply_bytes, 2581 .gpl_only = false, 2582 .ret_type = RET_INTEGER, 2583 .arg1_type = ARG_PTR_TO_CTX, 2584 .arg2_type = ARG_ANYTHING, 2585 }; 2586 2587 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes) 2588 { 2589 msg->cork_bytes = bytes; 2590 return 0; 2591 } 2592 2593 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = { 2594 .func = bpf_msg_cork_bytes, 2595 .gpl_only = false, 2596 .ret_type = RET_INTEGER, 2597 .arg1_type = ARG_PTR_TO_CTX, 2598 .arg2_type = ARG_ANYTHING, 2599 }; 2600 2601 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start, 2602 u32, end, u64, flags) 2603 { 2604 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start; 2605 u32 first_sge, last_sge, i, shift, bytes_sg_total; 2606 struct scatterlist *sge; 2607 u8 *raw, *to, *from; 2608 struct page *page; 2609 2610 if (unlikely(flags || end <= start)) 2611 return -EINVAL; 2612 2613 /* First find the starting scatterlist element */ 2614 i = msg->sg.start; 2615 do { 2616 offset += len; 2617 len = sk_msg_elem(msg, i)->length; 2618 if (start < offset + len) 2619 break; 2620 sk_msg_iter_var_next(i); 2621 } while (i != msg->sg.end); 2622 2623 if (unlikely(start >= offset + len)) 2624 return -EINVAL; 2625 2626 first_sge = i; 2627 /* The start may point into the sg element so we need to also 2628 * account for the headroom. 2629 */ 2630 bytes_sg_total = start - offset + bytes; 2631 if (!test_bit(i, msg->sg.copy) && bytes_sg_total <= len) 2632 goto out; 2633 2634 /* At this point we need to linearize multiple scatterlist 2635 * elements or a single shared page. Either way we need to 2636 * copy into a linear buffer exclusively owned by BPF. Then 2637 * place the buffer in the scatterlist and fixup the original 2638 * entries by removing the entries now in the linear buffer 2639 * and shifting the remaining entries. For now we do not try 2640 * to copy partial entries to avoid complexity of running out 2641 * of sg_entry slots. The downside is reading a single byte 2642 * will copy the entire sg entry. 2643 */ 2644 do { 2645 copy += sk_msg_elem(msg, i)->length; 2646 sk_msg_iter_var_next(i); 2647 if (bytes_sg_total <= copy) 2648 break; 2649 } while (i != msg->sg.end); 2650 last_sge = i; 2651 2652 if (unlikely(bytes_sg_total > copy)) 2653 return -EINVAL; 2654 2655 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2656 get_order(copy)); 2657 if (unlikely(!page)) 2658 return -ENOMEM; 2659 2660 raw = page_address(page); 2661 i = first_sge; 2662 do { 2663 sge = sk_msg_elem(msg, i); 2664 from = sg_virt(sge); 2665 len = sge->length; 2666 to = raw + poffset; 2667 2668 memcpy(to, from, len); 2669 poffset += len; 2670 sge->length = 0; 2671 put_page(sg_page(sge)); 2672 2673 sk_msg_iter_var_next(i); 2674 } while (i != last_sge); 2675 2676 sg_set_page(&msg->sg.data[first_sge], page, copy, 0); 2677 2678 /* To repair sg ring we need to shift entries. If we only 2679 * had a single entry though we can just replace it and 2680 * be done. Otherwise walk the ring and shift the entries. 2681 */ 2682 WARN_ON_ONCE(last_sge == first_sge); 2683 shift = last_sge > first_sge ? 2684 last_sge - first_sge - 1 : 2685 NR_MSG_FRAG_IDS - first_sge + last_sge - 1; 2686 if (!shift) 2687 goto out; 2688 2689 i = first_sge; 2690 sk_msg_iter_var_next(i); 2691 do { 2692 u32 move_from; 2693 2694 if (i + shift >= NR_MSG_FRAG_IDS) 2695 move_from = i + shift - NR_MSG_FRAG_IDS; 2696 else 2697 move_from = i + shift; 2698 if (move_from == msg->sg.end) 2699 break; 2700 2701 msg->sg.data[i] = msg->sg.data[move_from]; 2702 msg->sg.data[move_from].length = 0; 2703 msg->sg.data[move_from].page_link = 0; 2704 msg->sg.data[move_from].offset = 0; 2705 sk_msg_iter_var_next(i); 2706 } while (1); 2707 2708 msg->sg.end = msg->sg.end - shift > msg->sg.end ? 2709 msg->sg.end - shift + NR_MSG_FRAG_IDS : 2710 msg->sg.end - shift; 2711 out: 2712 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset; 2713 msg->data_end = msg->data + bytes; 2714 return 0; 2715 } 2716 2717 static const struct bpf_func_proto bpf_msg_pull_data_proto = { 2718 .func = bpf_msg_pull_data, 2719 .gpl_only = false, 2720 .ret_type = RET_INTEGER, 2721 .arg1_type = ARG_PTR_TO_CTX, 2722 .arg2_type = ARG_ANYTHING, 2723 .arg3_type = ARG_ANYTHING, 2724 .arg4_type = ARG_ANYTHING, 2725 }; 2726 2727 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start, 2728 u32, len, u64, flags) 2729 { 2730 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge; 2731 u32 new, i = 0, l = 0, space, copy = 0, offset = 0; 2732 u8 *raw, *to, *from; 2733 struct page *page; 2734 2735 if (unlikely(flags)) 2736 return -EINVAL; 2737 2738 if (unlikely(len == 0)) 2739 return 0; 2740 2741 /* First find the starting scatterlist element */ 2742 i = msg->sg.start; 2743 do { 2744 offset += l; 2745 l = sk_msg_elem(msg, i)->length; 2746 2747 if (start < offset + l) 2748 break; 2749 sk_msg_iter_var_next(i); 2750 } while (i != msg->sg.end); 2751 2752 if (start >= offset + l) 2753 return -EINVAL; 2754 2755 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2756 2757 /* If no space available will fallback to copy, we need at 2758 * least one scatterlist elem available to push data into 2759 * when start aligns to the beginning of an element or two 2760 * when it falls inside an element. We handle the start equals 2761 * offset case because its the common case for inserting a 2762 * header. 2763 */ 2764 if (!space || (space == 1 && start != offset)) 2765 copy = msg->sg.data[i].length; 2766 2767 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2768 get_order(copy + len)); 2769 if (unlikely(!page)) 2770 return -ENOMEM; 2771 2772 if (copy) { 2773 int front, back; 2774 2775 raw = page_address(page); 2776 2777 psge = sk_msg_elem(msg, i); 2778 front = start - offset; 2779 back = psge->length - front; 2780 from = sg_virt(psge); 2781 2782 if (front) 2783 memcpy(raw, from, front); 2784 2785 if (back) { 2786 from += front; 2787 to = raw + front + len; 2788 2789 memcpy(to, from, back); 2790 } 2791 2792 put_page(sg_page(psge)); 2793 } else if (start - offset) { 2794 psge = sk_msg_elem(msg, i); 2795 rsge = sk_msg_elem_cpy(msg, i); 2796 2797 psge->length = start - offset; 2798 rsge.length -= psge->length; 2799 rsge.offset += start; 2800 2801 sk_msg_iter_var_next(i); 2802 sg_unmark_end(psge); 2803 sg_unmark_end(&rsge); 2804 sk_msg_iter_next(msg, end); 2805 } 2806 2807 /* Slot(s) to place newly allocated data */ 2808 new = i; 2809 2810 /* Shift one or two slots as needed */ 2811 if (!copy) { 2812 sge = sk_msg_elem_cpy(msg, i); 2813 2814 sk_msg_iter_var_next(i); 2815 sg_unmark_end(&sge); 2816 sk_msg_iter_next(msg, end); 2817 2818 nsge = sk_msg_elem_cpy(msg, i); 2819 if (rsge.length) { 2820 sk_msg_iter_var_next(i); 2821 nnsge = sk_msg_elem_cpy(msg, i); 2822 } 2823 2824 while (i != msg->sg.end) { 2825 msg->sg.data[i] = sge; 2826 sge = nsge; 2827 sk_msg_iter_var_next(i); 2828 if (rsge.length) { 2829 nsge = nnsge; 2830 nnsge = sk_msg_elem_cpy(msg, i); 2831 } else { 2832 nsge = sk_msg_elem_cpy(msg, i); 2833 } 2834 } 2835 } 2836 2837 /* Place newly allocated data buffer */ 2838 sk_mem_charge(msg->sk, len); 2839 msg->sg.size += len; 2840 __clear_bit(new, msg->sg.copy); 2841 sg_set_page(&msg->sg.data[new], page, len + copy, 0); 2842 if (rsge.length) { 2843 get_page(sg_page(&rsge)); 2844 sk_msg_iter_var_next(new); 2845 msg->sg.data[new] = rsge; 2846 } 2847 2848 sk_msg_compute_data_pointers(msg); 2849 return 0; 2850 } 2851 2852 static const struct bpf_func_proto bpf_msg_push_data_proto = { 2853 .func = bpf_msg_push_data, 2854 .gpl_only = false, 2855 .ret_type = RET_INTEGER, 2856 .arg1_type = ARG_PTR_TO_CTX, 2857 .arg2_type = ARG_ANYTHING, 2858 .arg3_type = ARG_ANYTHING, 2859 .arg4_type = ARG_ANYTHING, 2860 }; 2861 2862 static void sk_msg_shift_left(struct sk_msg *msg, int i) 2863 { 2864 int prev; 2865 2866 do { 2867 prev = i; 2868 sk_msg_iter_var_next(i); 2869 msg->sg.data[prev] = msg->sg.data[i]; 2870 } while (i != msg->sg.end); 2871 2872 sk_msg_iter_prev(msg, end); 2873 } 2874 2875 static void sk_msg_shift_right(struct sk_msg *msg, int i) 2876 { 2877 struct scatterlist tmp, sge; 2878 2879 sk_msg_iter_next(msg, end); 2880 sge = sk_msg_elem_cpy(msg, i); 2881 sk_msg_iter_var_next(i); 2882 tmp = sk_msg_elem_cpy(msg, i); 2883 2884 while (i != msg->sg.end) { 2885 msg->sg.data[i] = sge; 2886 sk_msg_iter_var_next(i); 2887 sge = tmp; 2888 tmp = sk_msg_elem_cpy(msg, i); 2889 } 2890 } 2891 2892 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start, 2893 u32, len, u64, flags) 2894 { 2895 u32 i = 0, l = 0, space, offset = 0; 2896 u64 last = start + len; 2897 int pop; 2898 2899 if (unlikely(flags)) 2900 return -EINVAL; 2901 2902 /* First find the starting scatterlist element */ 2903 i = msg->sg.start; 2904 do { 2905 offset += l; 2906 l = sk_msg_elem(msg, i)->length; 2907 2908 if (start < offset + l) 2909 break; 2910 sk_msg_iter_var_next(i); 2911 } while (i != msg->sg.end); 2912 2913 /* Bounds checks: start and pop must be inside message */ 2914 if (start >= offset + l || last >= msg->sg.size) 2915 return -EINVAL; 2916 2917 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2918 2919 pop = len; 2920 /* --------------| offset 2921 * -| start |-------- len -------| 2922 * 2923 * |----- a ----|-------- pop -------|----- b ----| 2924 * |______________________________________________| length 2925 * 2926 * 2927 * a: region at front of scatter element to save 2928 * b: region at back of scatter element to save when length > A + pop 2929 * pop: region to pop from element, same as input 'pop' here will be 2930 * decremented below per iteration. 2931 * 2932 * Two top-level cases to handle when start != offset, first B is non 2933 * zero and second B is zero corresponding to when a pop includes more 2934 * than one element. 2935 * 2936 * Then if B is non-zero AND there is no space allocate space and 2937 * compact A, B regions into page. If there is space shift ring to 2938 * the rigth free'ing the next element in ring to place B, leaving 2939 * A untouched except to reduce length. 2940 */ 2941 if (start != offset) { 2942 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i); 2943 int a = start; 2944 int b = sge->length - pop - a; 2945 2946 sk_msg_iter_var_next(i); 2947 2948 if (pop < sge->length - a) { 2949 if (space) { 2950 sge->length = a; 2951 sk_msg_shift_right(msg, i); 2952 nsge = sk_msg_elem(msg, i); 2953 get_page(sg_page(sge)); 2954 sg_set_page(nsge, 2955 sg_page(sge), 2956 b, sge->offset + pop + a); 2957 } else { 2958 struct page *page, *orig; 2959 u8 *to, *from; 2960 2961 page = alloc_pages(__GFP_NOWARN | 2962 __GFP_COMP | GFP_ATOMIC, 2963 get_order(a + b)); 2964 if (unlikely(!page)) 2965 return -ENOMEM; 2966 2967 sge->length = a; 2968 orig = sg_page(sge); 2969 from = sg_virt(sge); 2970 to = page_address(page); 2971 memcpy(to, from, a); 2972 memcpy(to + a, from + a + pop, b); 2973 sg_set_page(sge, page, a + b, 0); 2974 put_page(orig); 2975 } 2976 pop = 0; 2977 } else if (pop >= sge->length - a) { 2978 pop -= (sge->length - a); 2979 sge->length = a; 2980 } 2981 } 2982 2983 /* From above the current layout _must_ be as follows, 2984 * 2985 * -| offset 2986 * -| start 2987 * 2988 * |---- pop ---|---------------- b ------------| 2989 * |____________________________________________| length 2990 * 2991 * Offset and start of the current msg elem are equal because in the 2992 * previous case we handled offset != start and either consumed the 2993 * entire element and advanced to the next element OR pop == 0. 2994 * 2995 * Two cases to handle here are first pop is less than the length 2996 * leaving some remainder b above. Simply adjust the element's layout 2997 * in this case. Or pop >= length of the element so that b = 0. In this 2998 * case advance to next element decrementing pop. 2999 */ 3000 while (pop) { 3001 struct scatterlist *sge = sk_msg_elem(msg, i); 3002 3003 if (pop < sge->length) { 3004 sge->length -= pop; 3005 sge->offset += pop; 3006 pop = 0; 3007 } else { 3008 pop -= sge->length; 3009 sk_msg_shift_left(msg, i); 3010 } 3011 sk_msg_iter_var_next(i); 3012 } 3013 3014 sk_mem_uncharge(msg->sk, len - pop); 3015 msg->sg.size -= (len - pop); 3016 sk_msg_compute_data_pointers(msg); 3017 return 0; 3018 } 3019 3020 static const struct bpf_func_proto bpf_msg_pop_data_proto = { 3021 .func = bpf_msg_pop_data, 3022 .gpl_only = false, 3023 .ret_type = RET_INTEGER, 3024 .arg1_type = ARG_PTR_TO_CTX, 3025 .arg2_type = ARG_ANYTHING, 3026 .arg3_type = ARG_ANYTHING, 3027 .arg4_type = ARG_ANYTHING, 3028 }; 3029 3030 #ifdef CONFIG_CGROUP_NET_CLASSID 3031 BPF_CALL_0(bpf_get_cgroup_classid_curr) 3032 { 3033 return __task_get_classid(current); 3034 } 3035 3036 const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = { 3037 .func = bpf_get_cgroup_classid_curr, 3038 .gpl_only = false, 3039 .ret_type = RET_INTEGER, 3040 }; 3041 3042 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb) 3043 { 3044 struct sock *sk = skb_to_full_sk(skb); 3045 3046 if (!sk || !sk_fullsock(sk)) 3047 return 0; 3048 3049 return sock_cgroup_classid(&sk->sk_cgrp_data); 3050 } 3051 3052 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = { 3053 .func = bpf_skb_cgroup_classid, 3054 .gpl_only = false, 3055 .ret_type = RET_INTEGER, 3056 .arg1_type = ARG_PTR_TO_CTX, 3057 }; 3058 #endif 3059 3060 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb) 3061 { 3062 return task_get_classid(skb); 3063 } 3064 3065 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = { 3066 .func = bpf_get_cgroup_classid, 3067 .gpl_only = false, 3068 .ret_type = RET_INTEGER, 3069 .arg1_type = ARG_PTR_TO_CTX, 3070 }; 3071 3072 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb) 3073 { 3074 return dst_tclassid(skb); 3075 } 3076 3077 static const struct bpf_func_proto bpf_get_route_realm_proto = { 3078 .func = bpf_get_route_realm, 3079 .gpl_only = false, 3080 .ret_type = RET_INTEGER, 3081 .arg1_type = ARG_PTR_TO_CTX, 3082 }; 3083 3084 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb) 3085 { 3086 /* If skb_clear_hash() was called due to mangling, we can 3087 * trigger SW recalculation here. Later access to hash 3088 * can then use the inline skb->hash via context directly 3089 * instead of calling this helper again. 3090 */ 3091 return skb_get_hash(skb); 3092 } 3093 3094 static const struct bpf_func_proto bpf_get_hash_recalc_proto = { 3095 .func = bpf_get_hash_recalc, 3096 .gpl_only = false, 3097 .ret_type = RET_INTEGER, 3098 .arg1_type = ARG_PTR_TO_CTX, 3099 }; 3100 3101 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb) 3102 { 3103 /* After all direct packet write, this can be used once for 3104 * triggering a lazy recalc on next skb_get_hash() invocation. 3105 */ 3106 skb_clear_hash(skb); 3107 return 0; 3108 } 3109 3110 static const struct bpf_func_proto bpf_set_hash_invalid_proto = { 3111 .func = bpf_set_hash_invalid, 3112 .gpl_only = false, 3113 .ret_type = RET_INTEGER, 3114 .arg1_type = ARG_PTR_TO_CTX, 3115 }; 3116 3117 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash) 3118 { 3119 /* Set user specified hash as L4(+), so that it gets returned 3120 * on skb_get_hash() call unless BPF prog later on triggers a 3121 * skb_clear_hash(). 3122 */ 3123 __skb_set_sw_hash(skb, hash, true); 3124 return 0; 3125 } 3126 3127 static const struct bpf_func_proto bpf_set_hash_proto = { 3128 .func = bpf_set_hash, 3129 .gpl_only = false, 3130 .ret_type = RET_INTEGER, 3131 .arg1_type = ARG_PTR_TO_CTX, 3132 .arg2_type = ARG_ANYTHING, 3133 }; 3134 3135 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto, 3136 u16, vlan_tci) 3137 { 3138 int ret; 3139 3140 if (unlikely(vlan_proto != htons(ETH_P_8021Q) && 3141 vlan_proto != htons(ETH_P_8021AD))) 3142 vlan_proto = htons(ETH_P_8021Q); 3143 3144 bpf_push_mac_rcsum(skb); 3145 ret = skb_vlan_push(skb, vlan_proto, vlan_tci); 3146 bpf_pull_mac_rcsum(skb); 3147 3148 bpf_compute_data_pointers(skb); 3149 return ret; 3150 } 3151 3152 static const struct bpf_func_proto bpf_skb_vlan_push_proto = { 3153 .func = bpf_skb_vlan_push, 3154 .gpl_only = false, 3155 .ret_type = RET_INTEGER, 3156 .arg1_type = ARG_PTR_TO_CTX, 3157 .arg2_type = ARG_ANYTHING, 3158 .arg3_type = ARG_ANYTHING, 3159 }; 3160 3161 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb) 3162 { 3163 int ret; 3164 3165 bpf_push_mac_rcsum(skb); 3166 ret = skb_vlan_pop(skb); 3167 bpf_pull_mac_rcsum(skb); 3168 3169 bpf_compute_data_pointers(skb); 3170 return ret; 3171 } 3172 3173 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = { 3174 .func = bpf_skb_vlan_pop, 3175 .gpl_only = false, 3176 .ret_type = RET_INTEGER, 3177 .arg1_type = ARG_PTR_TO_CTX, 3178 }; 3179 3180 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len) 3181 { 3182 /* Caller already did skb_cow() with len as headroom, 3183 * so no need to do it here. 3184 */ 3185 skb_push(skb, len); 3186 memmove(skb->data, skb->data + len, off); 3187 memset(skb->data + off, 0, len); 3188 3189 /* No skb_postpush_rcsum(skb, skb->data + off, len) 3190 * needed here as it does not change the skb->csum 3191 * result for checksum complete when summing over 3192 * zeroed blocks. 3193 */ 3194 return 0; 3195 } 3196 3197 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len) 3198 { 3199 void *old_data; 3200 3201 /* skb_ensure_writable() is not needed here, as we're 3202 * already working on an uncloned skb. 3203 */ 3204 if (unlikely(!pskb_may_pull(skb, off + len))) 3205 return -ENOMEM; 3206 3207 old_data = skb->data; 3208 __skb_pull(skb, len); 3209 skb_postpull_rcsum(skb, old_data + off, len); 3210 memmove(skb->data, old_data, off); 3211 3212 return 0; 3213 } 3214 3215 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len) 3216 { 3217 bool trans_same = skb->transport_header == skb->network_header; 3218 int ret; 3219 3220 /* There's no need for __skb_push()/__skb_pull() pair to 3221 * get to the start of the mac header as we're guaranteed 3222 * to always start from here under eBPF. 3223 */ 3224 ret = bpf_skb_generic_push(skb, off, len); 3225 if (likely(!ret)) { 3226 skb->mac_header -= len; 3227 skb->network_header -= len; 3228 if (trans_same) 3229 skb->transport_header = skb->network_header; 3230 } 3231 3232 return ret; 3233 } 3234 3235 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len) 3236 { 3237 bool trans_same = skb->transport_header == skb->network_header; 3238 int ret; 3239 3240 /* Same here, __skb_push()/__skb_pull() pair not needed. */ 3241 ret = bpf_skb_generic_pop(skb, off, len); 3242 if (likely(!ret)) { 3243 skb->mac_header += len; 3244 skb->network_header += len; 3245 if (trans_same) 3246 skb->transport_header = skb->network_header; 3247 } 3248 3249 return ret; 3250 } 3251 3252 static int bpf_skb_proto_4_to_6(struct sk_buff *skb) 3253 { 3254 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3255 u32 off = skb_mac_header_len(skb); 3256 int ret; 3257 3258 ret = skb_cow(skb, len_diff); 3259 if (unlikely(ret < 0)) 3260 return ret; 3261 3262 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3263 if (unlikely(ret < 0)) 3264 return ret; 3265 3266 if (skb_is_gso(skb)) { 3267 struct skb_shared_info *shinfo = skb_shinfo(skb); 3268 3269 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */ 3270 if (shinfo->gso_type & SKB_GSO_TCPV4) { 3271 shinfo->gso_type &= ~SKB_GSO_TCPV4; 3272 shinfo->gso_type |= SKB_GSO_TCPV6; 3273 } 3274 } 3275 3276 skb->protocol = htons(ETH_P_IPV6); 3277 skb_clear_hash(skb); 3278 3279 return 0; 3280 } 3281 3282 static int bpf_skb_proto_6_to_4(struct sk_buff *skb) 3283 { 3284 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3285 u32 off = skb_mac_header_len(skb); 3286 int ret; 3287 3288 ret = skb_unclone(skb, GFP_ATOMIC); 3289 if (unlikely(ret < 0)) 3290 return ret; 3291 3292 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3293 if (unlikely(ret < 0)) 3294 return ret; 3295 3296 if (skb_is_gso(skb)) { 3297 struct skb_shared_info *shinfo = skb_shinfo(skb); 3298 3299 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */ 3300 if (shinfo->gso_type & SKB_GSO_TCPV6) { 3301 shinfo->gso_type &= ~SKB_GSO_TCPV6; 3302 shinfo->gso_type |= SKB_GSO_TCPV4; 3303 } 3304 } 3305 3306 skb->protocol = htons(ETH_P_IP); 3307 skb_clear_hash(skb); 3308 3309 return 0; 3310 } 3311 3312 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto) 3313 { 3314 __be16 from_proto = skb->protocol; 3315 3316 if (from_proto == htons(ETH_P_IP) && 3317 to_proto == htons(ETH_P_IPV6)) 3318 return bpf_skb_proto_4_to_6(skb); 3319 3320 if (from_proto == htons(ETH_P_IPV6) && 3321 to_proto == htons(ETH_P_IP)) 3322 return bpf_skb_proto_6_to_4(skb); 3323 3324 return -ENOTSUPP; 3325 } 3326 3327 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto, 3328 u64, flags) 3329 { 3330 int ret; 3331 3332 if (unlikely(flags)) 3333 return -EINVAL; 3334 3335 /* General idea is that this helper does the basic groundwork 3336 * needed for changing the protocol, and eBPF program fills the 3337 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace() 3338 * and other helpers, rather than passing a raw buffer here. 3339 * 3340 * The rationale is to keep this minimal and without a need to 3341 * deal with raw packet data. F.e. even if we would pass buffers 3342 * here, the program still needs to call the bpf_lX_csum_replace() 3343 * helpers anyway. Plus, this way we keep also separation of 3344 * concerns, since f.e. bpf_skb_store_bytes() should only take 3345 * care of stores. 3346 * 3347 * Currently, additional options and extension header space are 3348 * not supported, but flags register is reserved so we can adapt 3349 * that. For offloads, we mark packet as dodgy, so that headers 3350 * need to be verified first. 3351 */ 3352 ret = bpf_skb_proto_xlat(skb, proto); 3353 bpf_compute_data_pointers(skb); 3354 return ret; 3355 } 3356 3357 static const struct bpf_func_proto bpf_skb_change_proto_proto = { 3358 .func = bpf_skb_change_proto, 3359 .gpl_only = false, 3360 .ret_type = RET_INTEGER, 3361 .arg1_type = ARG_PTR_TO_CTX, 3362 .arg2_type = ARG_ANYTHING, 3363 .arg3_type = ARG_ANYTHING, 3364 }; 3365 3366 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type) 3367 { 3368 /* We only allow a restricted subset to be changed for now. */ 3369 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) || 3370 !skb_pkt_type_ok(pkt_type))) 3371 return -EINVAL; 3372 3373 skb->pkt_type = pkt_type; 3374 return 0; 3375 } 3376 3377 static const struct bpf_func_proto bpf_skb_change_type_proto = { 3378 .func = bpf_skb_change_type, 3379 .gpl_only = false, 3380 .ret_type = RET_INTEGER, 3381 .arg1_type = ARG_PTR_TO_CTX, 3382 .arg2_type = ARG_ANYTHING, 3383 }; 3384 3385 static u32 bpf_skb_net_base_len(const struct sk_buff *skb) 3386 { 3387 switch (skb->protocol) { 3388 case htons(ETH_P_IP): 3389 return sizeof(struct iphdr); 3390 case htons(ETH_P_IPV6): 3391 return sizeof(struct ipv6hdr); 3392 default: 3393 return ~0U; 3394 } 3395 } 3396 3397 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \ 3398 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3399 3400 #define BPF_F_ADJ_ROOM_DECAP_L3_MASK (BPF_F_ADJ_ROOM_DECAP_L3_IPV4 | \ 3401 BPF_F_ADJ_ROOM_DECAP_L3_IPV6) 3402 3403 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \ 3404 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \ 3405 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \ 3406 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \ 3407 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \ 3408 BPF_F_ADJ_ROOM_ENCAP_L2( \ 3409 BPF_ADJ_ROOM_ENCAP_L2_MASK) | \ 3410 BPF_F_ADJ_ROOM_DECAP_L3_MASK) 3411 3412 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff, 3413 u64 flags) 3414 { 3415 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT; 3416 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK; 3417 u16 mac_len = 0, inner_net = 0, inner_trans = 0; 3418 unsigned int gso_type = SKB_GSO_DODGY; 3419 int ret; 3420 3421 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3422 /* udp gso_size delineates datagrams, only allow if fixed */ 3423 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3424 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3425 return -ENOTSUPP; 3426 } 3427 3428 ret = skb_cow_head(skb, len_diff); 3429 if (unlikely(ret < 0)) 3430 return ret; 3431 3432 if (encap) { 3433 if (skb->protocol != htons(ETH_P_IP) && 3434 skb->protocol != htons(ETH_P_IPV6)) 3435 return -ENOTSUPP; 3436 3437 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 && 3438 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3439 return -EINVAL; 3440 3441 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE && 3442 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3443 return -EINVAL; 3444 3445 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH && 3446 inner_mac_len < ETH_HLEN) 3447 return -EINVAL; 3448 3449 if (skb->encapsulation) 3450 return -EALREADY; 3451 3452 mac_len = skb->network_header - skb->mac_header; 3453 inner_net = skb->network_header; 3454 if (inner_mac_len > len_diff) 3455 return -EINVAL; 3456 inner_trans = skb->transport_header; 3457 } 3458 3459 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3460 if (unlikely(ret < 0)) 3461 return ret; 3462 3463 if (encap) { 3464 skb->inner_mac_header = inner_net - inner_mac_len; 3465 skb->inner_network_header = inner_net; 3466 skb->inner_transport_header = inner_trans; 3467 3468 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH) 3469 skb_set_inner_protocol(skb, htons(ETH_P_TEB)); 3470 else 3471 skb_set_inner_protocol(skb, skb->protocol); 3472 3473 skb->encapsulation = 1; 3474 skb_set_network_header(skb, mac_len); 3475 3476 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3477 gso_type |= SKB_GSO_UDP_TUNNEL; 3478 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE) 3479 gso_type |= SKB_GSO_GRE; 3480 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3481 gso_type |= SKB_GSO_IPXIP6; 3482 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3483 gso_type |= SKB_GSO_IPXIP4; 3484 3485 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE || 3486 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) { 3487 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ? 3488 sizeof(struct ipv6hdr) : 3489 sizeof(struct iphdr); 3490 3491 skb_set_transport_header(skb, mac_len + nh_len); 3492 } 3493 3494 /* Match skb->protocol to new outer l3 protocol */ 3495 if (skb->protocol == htons(ETH_P_IP) && 3496 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3497 skb->protocol = htons(ETH_P_IPV6); 3498 else if (skb->protocol == htons(ETH_P_IPV6) && 3499 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3500 skb->protocol = htons(ETH_P_IP); 3501 } 3502 3503 if (skb_is_gso(skb)) { 3504 struct skb_shared_info *shinfo = skb_shinfo(skb); 3505 3506 /* Due to header grow, MSS needs to be downgraded. */ 3507 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3508 skb_decrease_gso_size(shinfo, len_diff); 3509 3510 /* Header must be checked, and gso_segs recomputed. */ 3511 shinfo->gso_type |= gso_type; 3512 shinfo->gso_segs = 0; 3513 } 3514 3515 return 0; 3516 } 3517 3518 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff, 3519 u64 flags) 3520 { 3521 int ret; 3522 3523 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO | 3524 BPF_F_ADJ_ROOM_DECAP_L3_MASK | 3525 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3526 return -EINVAL; 3527 3528 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3529 /* udp gso_size delineates datagrams, only allow if fixed */ 3530 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3531 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3532 return -ENOTSUPP; 3533 } 3534 3535 ret = skb_unclone(skb, GFP_ATOMIC); 3536 if (unlikely(ret < 0)) 3537 return ret; 3538 3539 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3540 if (unlikely(ret < 0)) 3541 return ret; 3542 3543 /* Match skb->protocol to new outer l3 protocol */ 3544 if (skb->protocol == htons(ETH_P_IP) && 3545 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV6) 3546 skb->protocol = htons(ETH_P_IPV6); 3547 else if (skb->protocol == htons(ETH_P_IPV6) && 3548 flags & BPF_F_ADJ_ROOM_DECAP_L3_IPV4) 3549 skb->protocol = htons(ETH_P_IP); 3550 3551 if (skb_is_gso(skb)) { 3552 struct skb_shared_info *shinfo = skb_shinfo(skb); 3553 3554 /* Due to header shrink, MSS can be upgraded. */ 3555 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3556 skb_increase_gso_size(shinfo, len_diff); 3557 3558 /* Header must be checked, and gso_segs recomputed. */ 3559 shinfo->gso_type |= SKB_GSO_DODGY; 3560 shinfo->gso_segs = 0; 3561 } 3562 3563 return 0; 3564 } 3565 3566 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC 3567 3568 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3569 u32, mode, u64, flags) 3570 { 3571 u32 len_diff_abs = abs(len_diff); 3572 bool shrink = len_diff < 0; 3573 int ret = 0; 3574 3575 if (unlikely(flags || mode)) 3576 return -EINVAL; 3577 if (unlikely(len_diff_abs > 0xfffU)) 3578 return -EFAULT; 3579 3580 if (!shrink) { 3581 ret = skb_cow(skb, len_diff); 3582 if (unlikely(ret < 0)) 3583 return ret; 3584 __skb_push(skb, len_diff_abs); 3585 memset(skb->data, 0, len_diff_abs); 3586 } else { 3587 if (unlikely(!pskb_may_pull(skb, len_diff_abs))) 3588 return -ENOMEM; 3589 __skb_pull(skb, len_diff_abs); 3590 } 3591 if (tls_sw_has_ctx_rx(skb->sk)) { 3592 struct strp_msg *rxm = strp_msg(skb); 3593 3594 rxm->full_len += len_diff; 3595 } 3596 return ret; 3597 } 3598 3599 static const struct bpf_func_proto sk_skb_adjust_room_proto = { 3600 .func = sk_skb_adjust_room, 3601 .gpl_only = false, 3602 .ret_type = RET_INTEGER, 3603 .arg1_type = ARG_PTR_TO_CTX, 3604 .arg2_type = ARG_ANYTHING, 3605 .arg3_type = ARG_ANYTHING, 3606 .arg4_type = ARG_ANYTHING, 3607 }; 3608 3609 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3610 u32, mode, u64, flags) 3611 { 3612 u32 len_cur, len_diff_abs = abs(len_diff); 3613 u32 len_min = bpf_skb_net_base_len(skb); 3614 u32 len_max = BPF_SKB_MAX_LEN; 3615 __be16 proto = skb->protocol; 3616 bool shrink = len_diff < 0; 3617 u32 off; 3618 int ret; 3619 3620 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK | 3621 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3622 return -EINVAL; 3623 if (unlikely(len_diff_abs > 0xfffU)) 3624 return -EFAULT; 3625 if (unlikely(proto != htons(ETH_P_IP) && 3626 proto != htons(ETH_P_IPV6))) 3627 return -ENOTSUPP; 3628 3629 off = skb_mac_header_len(skb); 3630 switch (mode) { 3631 case BPF_ADJ_ROOM_NET: 3632 off += bpf_skb_net_base_len(skb); 3633 break; 3634 case BPF_ADJ_ROOM_MAC: 3635 break; 3636 default: 3637 return -ENOTSUPP; 3638 } 3639 3640 if (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) { 3641 if (!shrink) 3642 return -EINVAL; 3643 3644 switch (flags & BPF_F_ADJ_ROOM_DECAP_L3_MASK) { 3645 case BPF_F_ADJ_ROOM_DECAP_L3_IPV4: 3646 len_min = sizeof(struct iphdr); 3647 break; 3648 case BPF_F_ADJ_ROOM_DECAP_L3_IPV6: 3649 len_min = sizeof(struct ipv6hdr); 3650 break; 3651 default: 3652 return -EINVAL; 3653 } 3654 } 3655 3656 len_cur = skb->len - skb_network_offset(skb); 3657 if ((shrink && (len_diff_abs >= len_cur || 3658 len_cur - len_diff_abs < len_min)) || 3659 (!shrink && (skb->len + len_diff_abs > len_max && 3660 !skb_is_gso(skb)))) 3661 return -ENOTSUPP; 3662 3663 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) : 3664 bpf_skb_net_grow(skb, off, len_diff_abs, flags); 3665 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET)) 3666 __skb_reset_checksum_unnecessary(skb); 3667 3668 bpf_compute_data_pointers(skb); 3669 return ret; 3670 } 3671 3672 static const struct bpf_func_proto bpf_skb_adjust_room_proto = { 3673 .func = bpf_skb_adjust_room, 3674 .gpl_only = false, 3675 .ret_type = RET_INTEGER, 3676 .arg1_type = ARG_PTR_TO_CTX, 3677 .arg2_type = ARG_ANYTHING, 3678 .arg3_type = ARG_ANYTHING, 3679 .arg4_type = ARG_ANYTHING, 3680 }; 3681 3682 static u32 __bpf_skb_min_len(const struct sk_buff *skb) 3683 { 3684 u32 min_len = skb_network_offset(skb); 3685 3686 if (skb_transport_header_was_set(skb)) 3687 min_len = skb_transport_offset(skb); 3688 if (skb->ip_summed == CHECKSUM_PARTIAL) 3689 min_len = skb_checksum_start_offset(skb) + 3690 skb->csum_offset + sizeof(__sum16); 3691 return min_len; 3692 } 3693 3694 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len) 3695 { 3696 unsigned int old_len = skb->len; 3697 int ret; 3698 3699 ret = __skb_grow_rcsum(skb, new_len); 3700 if (!ret) 3701 memset(skb->data + old_len, 0, new_len - old_len); 3702 return ret; 3703 } 3704 3705 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len) 3706 { 3707 return __skb_trim_rcsum(skb, new_len); 3708 } 3709 3710 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len, 3711 u64 flags) 3712 { 3713 u32 max_len = BPF_SKB_MAX_LEN; 3714 u32 min_len = __bpf_skb_min_len(skb); 3715 int ret; 3716 3717 if (unlikely(flags || new_len > max_len || new_len < min_len)) 3718 return -EINVAL; 3719 if (skb->encapsulation) 3720 return -ENOTSUPP; 3721 3722 /* The basic idea of this helper is that it's performing the 3723 * needed work to either grow or trim an skb, and eBPF program 3724 * rewrites the rest via helpers like bpf_skb_store_bytes(), 3725 * bpf_lX_csum_replace() and others rather than passing a raw 3726 * buffer here. This one is a slow path helper and intended 3727 * for replies with control messages. 3728 * 3729 * Like in bpf_skb_change_proto(), we want to keep this rather 3730 * minimal and without protocol specifics so that we are able 3731 * to separate concerns as in bpf_skb_store_bytes() should only 3732 * be the one responsible for writing buffers. 3733 * 3734 * It's really expected to be a slow path operation here for 3735 * control message replies, so we're implicitly linearizing, 3736 * uncloning and drop offloads from the skb by this. 3737 */ 3738 ret = __bpf_try_make_writable(skb, skb->len); 3739 if (!ret) { 3740 if (new_len > skb->len) 3741 ret = bpf_skb_grow_rcsum(skb, new_len); 3742 else if (new_len < skb->len) 3743 ret = bpf_skb_trim_rcsum(skb, new_len); 3744 if (!ret && skb_is_gso(skb)) 3745 skb_gso_reset(skb); 3746 } 3747 return ret; 3748 } 3749 3750 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3751 u64, flags) 3752 { 3753 int ret = __bpf_skb_change_tail(skb, new_len, flags); 3754 3755 bpf_compute_data_pointers(skb); 3756 return ret; 3757 } 3758 3759 static const struct bpf_func_proto bpf_skb_change_tail_proto = { 3760 .func = bpf_skb_change_tail, 3761 .gpl_only = false, 3762 .ret_type = RET_INTEGER, 3763 .arg1_type = ARG_PTR_TO_CTX, 3764 .arg2_type = ARG_ANYTHING, 3765 .arg3_type = ARG_ANYTHING, 3766 }; 3767 3768 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3769 u64, flags) 3770 { 3771 return __bpf_skb_change_tail(skb, new_len, flags); 3772 } 3773 3774 static const struct bpf_func_proto sk_skb_change_tail_proto = { 3775 .func = sk_skb_change_tail, 3776 .gpl_only = false, 3777 .ret_type = RET_INTEGER, 3778 .arg1_type = ARG_PTR_TO_CTX, 3779 .arg2_type = ARG_ANYTHING, 3780 .arg3_type = ARG_ANYTHING, 3781 }; 3782 3783 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room, 3784 u64 flags) 3785 { 3786 u32 max_len = BPF_SKB_MAX_LEN; 3787 u32 new_len = skb->len + head_room; 3788 int ret; 3789 3790 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) || 3791 new_len < skb->len)) 3792 return -EINVAL; 3793 3794 ret = skb_cow(skb, head_room); 3795 if (likely(!ret)) { 3796 /* Idea for this helper is that we currently only 3797 * allow to expand on mac header. This means that 3798 * skb->protocol network header, etc, stay as is. 3799 * Compared to bpf_skb_change_tail(), we're more 3800 * flexible due to not needing to linearize or 3801 * reset GSO. Intention for this helper is to be 3802 * used by an L3 skb that needs to push mac header 3803 * for redirection into L2 device. 3804 */ 3805 __skb_push(skb, head_room); 3806 memset(skb->data, 0, head_room); 3807 skb_reset_mac_header(skb); 3808 skb_reset_mac_len(skb); 3809 } 3810 3811 return ret; 3812 } 3813 3814 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room, 3815 u64, flags) 3816 { 3817 int ret = __bpf_skb_change_head(skb, head_room, flags); 3818 3819 bpf_compute_data_pointers(skb); 3820 return ret; 3821 } 3822 3823 static const struct bpf_func_proto bpf_skb_change_head_proto = { 3824 .func = bpf_skb_change_head, 3825 .gpl_only = false, 3826 .ret_type = RET_INTEGER, 3827 .arg1_type = ARG_PTR_TO_CTX, 3828 .arg2_type = ARG_ANYTHING, 3829 .arg3_type = ARG_ANYTHING, 3830 }; 3831 3832 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room, 3833 u64, flags) 3834 { 3835 return __bpf_skb_change_head(skb, head_room, flags); 3836 } 3837 3838 static const struct bpf_func_proto sk_skb_change_head_proto = { 3839 .func = sk_skb_change_head, 3840 .gpl_only = false, 3841 .ret_type = RET_INTEGER, 3842 .arg1_type = ARG_PTR_TO_CTX, 3843 .arg2_type = ARG_ANYTHING, 3844 .arg3_type = ARG_ANYTHING, 3845 }; 3846 3847 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp) 3848 { 3849 return xdp_get_buff_len(xdp); 3850 } 3851 3852 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = { 3853 .func = bpf_xdp_get_buff_len, 3854 .gpl_only = false, 3855 .ret_type = RET_INTEGER, 3856 .arg1_type = ARG_PTR_TO_CTX, 3857 }; 3858 3859 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff) 3860 3861 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = { 3862 .func = bpf_xdp_get_buff_len, 3863 .gpl_only = false, 3864 .arg1_type = ARG_PTR_TO_BTF_ID, 3865 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0], 3866 }; 3867 3868 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp) 3869 { 3870 return xdp_data_meta_unsupported(xdp) ? 0 : 3871 xdp->data - xdp->data_meta; 3872 } 3873 3874 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset) 3875 { 3876 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 3877 unsigned long metalen = xdp_get_metalen(xdp); 3878 void *data_start = xdp_frame_end + metalen; 3879 void *data = xdp->data + offset; 3880 3881 if (unlikely(data < data_start || 3882 data > xdp->data_end - ETH_HLEN)) 3883 return -EINVAL; 3884 3885 if (metalen) 3886 memmove(xdp->data_meta + offset, 3887 xdp->data_meta, metalen); 3888 xdp->data_meta += offset; 3889 xdp->data = data; 3890 3891 return 0; 3892 } 3893 3894 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = { 3895 .func = bpf_xdp_adjust_head, 3896 .gpl_only = false, 3897 .ret_type = RET_INTEGER, 3898 .arg1_type = ARG_PTR_TO_CTX, 3899 .arg2_type = ARG_ANYTHING, 3900 }; 3901 3902 void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 3903 void *buf, unsigned long len, bool flush) 3904 { 3905 unsigned long ptr_len, ptr_off = 0; 3906 skb_frag_t *next_frag, *end_frag; 3907 struct skb_shared_info *sinfo; 3908 void *src, *dst; 3909 u8 *ptr_buf; 3910 3911 if (likely(xdp->data_end - xdp->data >= off + len)) { 3912 src = flush ? buf : xdp->data + off; 3913 dst = flush ? xdp->data + off : buf; 3914 memcpy(dst, src, len); 3915 return; 3916 } 3917 3918 sinfo = xdp_get_shared_info_from_buff(xdp); 3919 end_frag = &sinfo->frags[sinfo->nr_frags]; 3920 next_frag = &sinfo->frags[0]; 3921 3922 ptr_len = xdp->data_end - xdp->data; 3923 ptr_buf = xdp->data; 3924 3925 while (true) { 3926 if (off < ptr_off + ptr_len) { 3927 unsigned long copy_off = off - ptr_off; 3928 unsigned long copy_len = min(len, ptr_len - copy_off); 3929 3930 src = flush ? buf : ptr_buf + copy_off; 3931 dst = flush ? ptr_buf + copy_off : buf; 3932 memcpy(dst, src, copy_len); 3933 3934 off += copy_len; 3935 len -= copy_len; 3936 buf += copy_len; 3937 } 3938 3939 if (!len || next_frag == end_frag) 3940 break; 3941 3942 ptr_off += ptr_len; 3943 ptr_buf = skb_frag_address(next_frag); 3944 ptr_len = skb_frag_size(next_frag); 3945 next_frag++; 3946 } 3947 } 3948 3949 void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 3950 { 3951 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 3952 u32 size = xdp->data_end - xdp->data; 3953 void *addr = xdp->data; 3954 int i; 3955 3956 if (unlikely(offset > 0xffff || len > 0xffff)) 3957 return ERR_PTR(-EFAULT); 3958 3959 if (offset + len > xdp_get_buff_len(xdp)) 3960 return ERR_PTR(-EINVAL); 3961 3962 if (offset < size) /* linear area */ 3963 goto out; 3964 3965 offset -= size; 3966 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */ 3967 u32 frag_size = skb_frag_size(&sinfo->frags[i]); 3968 3969 if (offset < frag_size) { 3970 addr = skb_frag_address(&sinfo->frags[i]); 3971 size = frag_size; 3972 break; 3973 } 3974 offset -= frag_size; 3975 } 3976 out: 3977 return offset + len <= size ? addr + offset : NULL; 3978 } 3979 3980 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset, 3981 void *, buf, u32, len) 3982 { 3983 void *ptr; 3984 3985 ptr = bpf_xdp_pointer(xdp, offset, len); 3986 if (IS_ERR(ptr)) 3987 return PTR_ERR(ptr); 3988 3989 if (!ptr) 3990 bpf_xdp_copy_buf(xdp, offset, buf, len, false); 3991 else 3992 memcpy(buf, ptr, len); 3993 3994 return 0; 3995 } 3996 3997 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = { 3998 .func = bpf_xdp_load_bytes, 3999 .gpl_only = false, 4000 .ret_type = RET_INTEGER, 4001 .arg1_type = ARG_PTR_TO_CTX, 4002 .arg2_type = ARG_ANYTHING, 4003 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 4004 .arg4_type = ARG_CONST_SIZE, 4005 }; 4006 4007 int __bpf_xdp_load_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len) 4008 { 4009 return ____bpf_xdp_load_bytes(xdp, offset, buf, len); 4010 } 4011 4012 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset, 4013 void *, buf, u32, len) 4014 { 4015 void *ptr; 4016 4017 ptr = bpf_xdp_pointer(xdp, offset, len); 4018 if (IS_ERR(ptr)) 4019 return PTR_ERR(ptr); 4020 4021 if (!ptr) 4022 bpf_xdp_copy_buf(xdp, offset, buf, len, true); 4023 else 4024 memcpy(ptr, buf, len); 4025 4026 return 0; 4027 } 4028 4029 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = { 4030 .func = bpf_xdp_store_bytes, 4031 .gpl_only = false, 4032 .ret_type = RET_INTEGER, 4033 .arg1_type = ARG_PTR_TO_CTX, 4034 .arg2_type = ARG_ANYTHING, 4035 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 4036 .arg4_type = ARG_CONST_SIZE, 4037 }; 4038 4039 int __bpf_xdp_store_bytes(struct xdp_buff *xdp, u32 offset, void *buf, u32 len) 4040 { 4041 return ____bpf_xdp_store_bytes(xdp, offset, buf, len); 4042 } 4043 4044 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset) 4045 { 4046 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 4047 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1]; 4048 struct xdp_rxq_info *rxq = xdp->rxq; 4049 unsigned int tailroom; 4050 4051 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz) 4052 return -EOPNOTSUPP; 4053 4054 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag); 4055 if (unlikely(offset > tailroom)) 4056 return -EINVAL; 4057 4058 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset); 4059 skb_frag_size_add(frag, offset); 4060 sinfo->xdp_frags_size += offset; 4061 4062 return 0; 4063 } 4064 4065 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset) 4066 { 4067 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 4068 int i, n_frags_free = 0, len_free = 0; 4069 4070 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN)) 4071 return -EINVAL; 4072 4073 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) { 4074 skb_frag_t *frag = &sinfo->frags[i]; 4075 int shrink = min_t(int, offset, skb_frag_size(frag)); 4076 4077 len_free += shrink; 4078 offset -= shrink; 4079 4080 if (skb_frag_size(frag) == shrink) { 4081 struct page *page = skb_frag_page(frag); 4082 4083 __xdp_return(page_address(page), &xdp->rxq->mem, 4084 false, NULL); 4085 n_frags_free++; 4086 } else { 4087 skb_frag_size_sub(frag, shrink); 4088 break; 4089 } 4090 } 4091 sinfo->nr_frags -= n_frags_free; 4092 sinfo->xdp_frags_size -= len_free; 4093 4094 if (unlikely(!sinfo->nr_frags)) { 4095 xdp_buff_clear_frags_flag(xdp); 4096 xdp->data_end -= offset; 4097 } 4098 4099 return 0; 4100 } 4101 4102 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset) 4103 { 4104 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */ 4105 void *data_end = xdp->data_end + offset; 4106 4107 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */ 4108 if (offset < 0) 4109 return bpf_xdp_frags_shrink_tail(xdp, -offset); 4110 4111 return bpf_xdp_frags_increase_tail(xdp, offset); 4112 } 4113 4114 /* Notice that xdp_data_hard_end have reserved some tailroom */ 4115 if (unlikely(data_end > data_hard_end)) 4116 return -EINVAL; 4117 4118 /* ALL drivers MUST init xdp->frame_sz, chicken check below */ 4119 if (unlikely(xdp->frame_sz > PAGE_SIZE)) { 4120 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz); 4121 return -EINVAL; 4122 } 4123 4124 if (unlikely(data_end < xdp->data + ETH_HLEN)) 4125 return -EINVAL; 4126 4127 /* Clear memory area on grow, can contain uninit kernel memory */ 4128 if (offset > 0) 4129 memset(xdp->data_end, 0, offset); 4130 4131 xdp->data_end = data_end; 4132 4133 return 0; 4134 } 4135 4136 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = { 4137 .func = bpf_xdp_adjust_tail, 4138 .gpl_only = false, 4139 .ret_type = RET_INTEGER, 4140 .arg1_type = ARG_PTR_TO_CTX, 4141 .arg2_type = ARG_ANYTHING, 4142 }; 4143 4144 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset) 4145 { 4146 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 4147 void *meta = xdp->data_meta + offset; 4148 unsigned long metalen = xdp->data - meta; 4149 4150 if (xdp_data_meta_unsupported(xdp)) 4151 return -ENOTSUPP; 4152 if (unlikely(meta < xdp_frame_end || 4153 meta > xdp->data)) 4154 return -EINVAL; 4155 if (unlikely(xdp_metalen_invalid(metalen))) 4156 return -EACCES; 4157 4158 xdp->data_meta = meta; 4159 4160 return 0; 4161 } 4162 4163 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = { 4164 .func = bpf_xdp_adjust_meta, 4165 .gpl_only = false, 4166 .ret_type = RET_INTEGER, 4167 .arg1_type = ARG_PTR_TO_CTX, 4168 .arg2_type = ARG_ANYTHING, 4169 }; 4170 4171 /** 4172 * DOC: xdp redirect 4173 * 4174 * XDP_REDIRECT works by a three-step process, implemented in the functions 4175 * below: 4176 * 4177 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target 4178 * of the redirect and store it (along with some other metadata) in a per-CPU 4179 * struct bpf_redirect_info. 4180 * 4181 * 2. When the program returns the XDP_REDIRECT return code, the driver will 4182 * call xdp_do_redirect() which will use the information in struct 4183 * bpf_redirect_info to actually enqueue the frame into a map type-specific 4184 * bulk queue structure. 4185 * 4186 * 3. Before exiting its NAPI poll loop, the driver will call 4187 * xdp_do_flush(), which will flush all the different bulk queues, 4188 * thus completing the redirect. Note that xdp_do_flush() must be 4189 * called before napi_complete_done() in the driver, as the 4190 * XDP_REDIRECT logic relies on being inside a single NAPI instance 4191 * through to the xdp_do_flush() call for RCU protection of all 4192 * in-kernel data structures. 4193 */ 4194 /* 4195 * Pointers to the map entries will be kept around for this whole sequence of 4196 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in 4197 * the core code; instead, the RCU protection relies on everything happening 4198 * inside a single NAPI poll sequence, which means it's between a pair of calls 4199 * to local_bh_disable()/local_bh_enable(). 4200 * 4201 * The map entries are marked as __rcu and the map code makes sure to 4202 * dereference those pointers with rcu_dereference_check() in a way that works 4203 * for both sections that to hold an rcu_read_lock() and sections that are 4204 * called from NAPI without a separate rcu_read_lock(). The code below does not 4205 * use RCU annotations, but relies on those in the map code. 4206 */ 4207 void xdp_do_flush(void) 4208 { 4209 __dev_flush(); 4210 __cpu_map_flush(); 4211 __xsk_map_flush(); 4212 } 4213 EXPORT_SYMBOL_GPL(xdp_do_flush); 4214 4215 void bpf_clear_redirect_map(struct bpf_map *map) 4216 { 4217 struct bpf_redirect_info *ri; 4218 int cpu; 4219 4220 for_each_possible_cpu(cpu) { 4221 ri = per_cpu_ptr(&bpf_redirect_info, cpu); 4222 /* Avoid polluting remote cacheline due to writes if 4223 * not needed. Once we pass this test, we need the 4224 * cmpxchg() to make sure it hasn't been changed in 4225 * the meantime by remote CPU. 4226 */ 4227 if (unlikely(READ_ONCE(ri->map) == map)) 4228 cmpxchg(&ri->map, map, NULL); 4229 } 4230 } 4231 4232 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 4233 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key); 4234 4235 u32 xdp_master_redirect(struct xdp_buff *xdp) 4236 { 4237 struct net_device *master, *slave; 4238 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4239 4240 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev); 4241 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp); 4242 if (slave && slave != xdp->rxq->dev) { 4243 /* The target device is different from the receiving device, so 4244 * redirect it to the new device. 4245 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled 4246 * drivers to unmap the packet from their rx ring. 4247 */ 4248 ri->tgt_index = slave->ifindex; 4249 ri->map_id = INT_MAX; 4250 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4251 return XDP_REDIRECT; 4252 } 4253 return XDP_TX; 4254 } 4255 EXPORT_SYMBOL_GPL(xdp_master_redirect); 4256 4257 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri, 4258 struct net_device *dev, 4259 struct xdp_buff *xdp, 4260 struct bpf_prog *xdp_prog) 4261 { 4262 enum bpf_map_type map_type = ri->map_type; 4263 void *fwd = ri->tgt_value; 4264 u32 map_id = ri->map_id; 4265 int err; 4266 4267 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4268 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4269 4270 err = __xsk_map_redirect(fwd, xdp); 4271 if (unlikely(err)) 4272 goto err; 4273 4274 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4275 return 0; 4276 err: 4277 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4278 return err; 4279 } 4280 4281 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri, 4282 struct net_device *dev, 4283 struct xdp_frame *xdpf, 4284 struct bpf_prog *xdp_prog) 4285 { 4286 enum bpf_map_type map_type = ri->map_type; 4287 void *fwd = ri->tgt_value; 4288 u32 map_id = ri->map_id; 4289 struct bpf_map *map; 4290 int err; 4291 4292 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4293 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4294 4295 if (unlikely(!xdpf)) { 4296 err = -EOVERFLOW; 4297 goto err; 4298 } 4299 4300 switch (map_type) { 4301 case BPF_MAP_TYPE_DEVMAP: 4302 fallthrough; 4303 case BPF_MAP_TYPE_DEVMAP_HASH: 4304 map = READ_ONCE(ri->map); 4305 if (unlikely(map)) { 4306 WRITE_ONCE(ri->map, NULL); 4307 err = dev_map_enqueue_multi(xdpf, dev, map, 4308 ri->flags & BPF_F_EXCLUDE_INGRESS); 4309 } else { 4310 err = dev_map_enqueue(fwd, xdpf, dev); 4311 } 4312 break; 4313 case BPF_MAP_TYPE_CPUMAP: 4314 err = cpu_map_enqueue(fwd, xdpf, dev); 4315 break; 4316 case BPF_MAP_TYPE_UNSPEC: 4317 if (map_id == INT_MAX) { 4318 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index); 4319 if (unlikely(!fwd)) { 4320 err = -EINVAL; 4321 break; 4322 } 4323 err = dev_xdp_enqueue(fwd, xdpf, dev); 4324 break; 4325 } 4326 fallthrough; 4327 default: 4328 err = -EBADRQC; 4329 } 4330 4331 if (unlikely(err)) 4332 goto err; 4333 4334 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4335 return 0; 4336 err: 4337 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4338 return err; 4339 } 4340 4341 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp, 4342 struct bpf_prog *xdp_prog) 4343 { 4344 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4345 enum bpf_map_type map_type = ri->map_type; 4346 4347 if (map_type == BPF_MAP_TYPE_XSKMAP) { 4348 /* XDP_REDIRECT is not supported AF_XDP yet. */ 4349 if (unlikely(xdp_buff_has_frags(xdp))) 4350 return -EOPNOTSUPP; 4351 4352 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog); 4353 } 4354 4355 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp), 4356 xdp_prog); 4357 } 4358 EXPORT_SYMBOL_GPL(xdp_do_redirect); 4359 4360 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp, 4361 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog) 4362 { 4363 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4364 enum bpf_map_type map_type = ri->map_type; 4365 4366 if (map_type == BPF_MAP_TYPE_XSKMAP) 4367 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog); 4368 4369 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog); 4370 } 4371 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame); 4372 4373 static int xdp_do_generic_redirect_map(struct net_device *dev, 4374 struct sk_buff *skb, 4375 struct xdp_buff *xdp, 4376 struct bpf_prog *xdp_prog, 4377 void *fwd, 4378 enum bpf_map_type map_type, u32 map_id) 4379 { 4380 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4381 struct bpf_map *map; 4382 int err; 4383 4384 switch (map_type) { 4385 case BPF_MAP_TYPE_DEVMAP: 4386 fallthrough; 4387 case BPF_MAP_TYPE_DEVMAP_HASH: 4388 map = READ_ONCE(ri->map); 4389 if (unlikely(map)) { 4390 WRITE_ONCE(ri->map, NULL); 4391 err = dev_map_redirect_multi(dev, skb, xdp_prog, map, 4392 ri->flags & BPF_F_EXCLUDE_INGRESS); 4393 } else { 4394 err = dev_map_generic_redirect(fwd, skb, xdp_prog); 4395 } 4396 if (unlikely(err)) 4397 goto err; 4398 break; 4399 case BPF_MAP_TYPE_XSKMAP: 4400 err = xsk_generic_rcv(fwd, xdp); 4401 if (err) 4402 goto err; 4403 consume_skb(skb); 4404 break; 4405 case BPF_MAP_TYPE_CPUMAP: 4406 err = cpu_map_generic_redirect(fwd, skb); 4407 if (unlikely(err)) 4408 goto err; 4409 break; 4410 default: 4411 err = -EBADRQC; 4412 goto err; 4413 } 4414 4415 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4416 return 0; 4417 err: 4418 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4419 return err; 4420 } 4421 4422 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 4423 struct xdp_buff *xdp, struct bpf_prog *xdp_prog) 4424 { 4425 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4426 enum bpf_map_type map_type = ri->map_type; 4427 void *fwd = ri->tgt_value; 4428 u32 map_id = ri->map_id; 4429 int err; 4430 4431 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4432 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4433 4434 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) { 4435 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index); 4436 if (unlikely(!fwd)) { 4437 err = -EINVAL; 4438 goto err; 4439 } 4440 4441 err = xdp_ok_fwd_dev(fwd, skb->len); 4442 if (unlikely(err)) 4443 goto err; 4444 4445 skb->dev = fwd; 4446 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index); 4447 generic_xdp_tx(skb, xdp_prog); 4448 return 0; 4449 } 4450 4451 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id); 4452 err: 4453 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err); 4454 return err; 4455 } 4456 4457 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags) 4458 { 4459 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4460 4461 if (unlikely(flags)) 4462 return XDP_ABORTED; 4463 4464 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated 4465 * by map_idr) is used for ifindex based XDP redirect. 4466 */ 4467 ri->tgt_index = ifindex; 4468 ri->map_id = INT_MAX; 4469 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4470 4471 return XDP_REDIRECT; 4472 } 4473 4474 static const struct bpf_func_proto bpf_xdp_redirect_proto = { 4475 .func = bpf_xdp_redirect, 4476 .gpl_only = false, 4477 .ret_type = RET_INTEGER, 4478 .arg1_type = ARG_ANYTHING, 4479 .arg2_type = ARG_ANYTHING, 4480 }; 4481 4482 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u64, key, 4483 u64, flags) 4484 { 4485 return map->ops->map_redirect(map, key, flags); 4486 } 4487 4488 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = { 4489 .func = bpf_xdp_redirect_map, 4490 .gpl_only = false, 4491 .ret_type = RET_INTEGER, 4492 .arg1_type = ARG_CONST_MAP_PTR, 4493 .arg2_type = ARG_ANYTHING, 4494 .arg3_type = ARG_ANYTHING, 4495 }; 4496 4497 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb, 4498 unsigned long off, unsigned long len) 4499 { 4500 void *ptr = skb_header_pointer(skb, off, len, dst_buff); 4501 4502 if (unlikely(!ptr)) 4503 return len; 4504 if (ptr != dst_buff) 4505 memcpy(dst_buff, ptr, len); 4506 4507 return 0; 4508 } 4509 4510 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map, 4511 u64, flags, void *, meta, u64, meta_size) 4512 { 4513 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4514 4515 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4516 return -EINVAL; 4517 if (unlikely(!skb || skb_size > skb->len)) 4518 return -EFAULT; 4519 4520 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size, 4521 bpf_skb_copy); 4522 } 4523 4524 static const struct bpf_func_proto bpf_skb_event_output_proto = { 4525 .func = bpf_skb_event_output, 4526 .gpl_only = true, 4527 .ret_type = RET_INTEGER, 4528 .arg1_type = ARG_PTR_TO_CTX, 4529 .arg2_type = ARG_CONST_MAP_PTR, 4530 .arg3_type = ARG_ANYTHING, 4531 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4532 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4533 }; 4534 4535 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff) 4536 4537 const struct bpf_func_proto bpf_skb_output_proto = { 4538 .func = bpf_skb_event_output, 4539 .gpl_only = true, 4540 .ret_type = RET_INTEGER, 4541 .arg1_type = ARG_PTR_TO_BTF_ID, 4542 .arg1_btf_id = &bpf_skb_output_btf_ids[0], 4543 .arg2_type = ARG_CONST_MAP_PTR, 4544 .arg3_type = ARG_ANYTHING, 4545 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4546 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4547 }; 4548 4549 static unsigned short bpf_tunnel_key_af(u64 flags) 4550 { 4551 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET; 4552 } 4553 4554 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to, 4555 u32, size, u64, flags) 4556 { 4557 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4558 u8 compat[sizeof(struct bpf_tunnel_key)]; 4559 void *to_orig = to; 4560 int err; 4561 4562 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6 | 4563 BPF_F_TUNINFO_FLAGS)))) { 4564 err = -EINVAL; 4565 goto err_clear; 4566 } 4567 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { 4568 err = -EPROTO; 4569 goto err_clear; 4570 } 4571 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4572 err = -EINVAL; 4573 switch (size) { 4574 case offsetof(struct bpf_tunnel_key, local_ipv6[0]): 4575 case offsetof(struct bpf_tunnel_key, tunnel_label): 4576 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4577 goto set_compat; 4578 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4579 /* Fixup deprecated structure layouts here, so we have 4580 * a common path later on. 4581 */ 4582 if (ip_tunnel_info_af(info) != AF_INET) 4583 goto err_clear; 4584 set_compat: 4585 to = (struct bpf_tunnel_key *)compat; 4586 break; 4587 default: 4588 goto err_clear; 4589 } 4590 } 4591 4592 to->tunnel_id = be64_to_cpu(info->key.tun_id); 4593 to->tunnel_tos = info->key.tos; 4594 to->tunnel_ttl = info->key.ttl; 4595 if (flags & BPF_F_TUNINFO_FLAGS) 4596 to->tunnel_flags = info->key.tun_flags; 4597 else 4598 to->tunnel_ext = 0; 4599 4600 if (flags & BPF_F_TUNINFO_IPV6) { 4601 memcpy(to->remote_ipv6, &info->key.u.ipv6.src, 4602 sizeof(to->remote_ipv6)); 4603 memcpy(to->local_ipv6, &info->key.u.ipv6.dst, 4604 sizeof(to->local_ipv6)); 4605 to->tunnel_label = be32_to_cpu(info->key.label); 4606 } else { 4607 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); 4608 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 4609 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst); 4610 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3); 4611 to->tunnel_label = 0; 4612 } 4613 4614 if (unlikely(size != sizeof(struct bpf_tunnel_key))) 4615 memcpy(to_orig, to, size); 4616 4617 return 0; 4618 err_clear: 4619 memset(to_orig, 0, size); 4620 return err; 4621 } 4622 4623 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { 4624 .func = bpf_skb_get_tunnel_key, 4625 .gpl_only = false, 4626 .ret_type = RET_INTEGER, 4627 .arg1_type = ARG_PTR_TO_CTX, 4628 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4629 .arg3_type = ARG_CONST_SIZE, 4630 .arg4_type = ARG_ANYTHING, 4631 }; 4632 4633 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size) 4634 { 4635 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4636 int err; 4637 4638 if (unlikely(!info || 4639 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) { 4640 err = -ENOENT; 4641 goto err_clear; 4642 } 4643 if (unlikely(size < info->options_len)) { 4644 err = -ENOMEM; 4645 goto err_clear; 4646 } 4647 4648 ip_tunnel_info_opts_get(to, info); 4649 if (size > info->options_len) 4650 memset(to + info->options_len, 0, size - info->options_len); 4651 4652 return info->options_len; 4653 err_clear: 4654 memset(to, 0, size); 4655 return err; 4656 } 4657 4658 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { 4659 .func = bpf_skb_get_tunnel_opt, 4660 .gpl_only = false, 4661 .ret_type = RET_INTEGER, 4662 .arg1_type = ARG_PTR_TO_CTX, 4663 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4664 .arg3_type = ARG_CONST_SIZE, 4665 }; 4666 4667 static struct metadata_dst __percpu *md_dst; 4668 4669 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb, 4670 const struct bpf_tunnel_key *, from, u32, size, u64, flags) 4671 { 4672 struct metadata_dst *md = this_cpu_ptr(md_dst); 4673 u8 compat[sizeof(struct bpf_tunnel_key)]; 4674 struct ip_tunnel_info *info; 4675 4676 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | 4677 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER | 4678 BPF_F_NO_TUNNEL_KEY))) 4679 return -EINVAL; 4680 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4681 switch (size) { 4682 case offsetof(struct bpf_tunnel_key, local_ipv6[0]): 4683 case offsetof(struct bpf_tunnel_key, tunnel_label): 4684 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4685 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4686 /* Fixup deprecated structure layouts here, so we have 4687 * a common path later on. 4688 */ 4689 memcpy(compat, from, size); 4690 memset(compat + size, 0, sizeof(compat) - size); 4691 from = (const struct bpf_tunnel_key *) compat; 4692 break; 4693 default: 4694 return -EINVAL; 4695 } 4696 } 4697 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || 4698 from->tunnel_ext)) 4699 return -EINVAL; 4700 4701 skb_dst_drop(skb); 4702 dst_hold((struct dst_entry *) md); 4703 skb_dst_set(skb, (struct dst_entry *) md); 4704 4705 info = &md->u.tun_info; 4706 memset(info, 0, sizeof(*info)); 4707 info->mode = IP_TUNNEL_INFO_TX; 4708 4709 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE; 4710 if (flags & BPF_F_DONT_FRAGMENT) 4711 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT; 4712 if (flags & BPF_F_ZERO_CSUM_TX) 4713 info->key.tun_flags &= ~TUNNEL_CSUM; 4714 if (flags & BPF_F_SEQ_NUMBER) 4715 info->key.tun_flags |= TUNNEL_SEQ; 4716 if (flags & BPF_F_NO_TUNNEL_KEY) 4717 info->key.tun_flags &= ~TUNNEL_KEY; 4718 4719 info->key.tun_id = cpu_to_be64(from->tunnel_id); 4720 info->key.tos = from->tunnel_tos; 4721 info->key.ttl = from->tunnel_ttl; 4722 4723 if (flags & BPF_F_TUNINFO_IPV6) { 4724 info->mode |= IP_TUNNEL_INFO_IPV6; 4725 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, 4726 sizeof(from->remote_ipv6)); 4727 memcpy(&info->key.u.ipv6.src, from->local_ipv6, 4728 sizeof(from->local_ipv6)); 4729 info->key.label = cpu_to_be32(from->tunnel_label) & 4730 IPV6_FLOWLABEL_MASK; 4731 } else { 4732 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); 4733 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4); 4734 info->key.flow_flags = FLOWI_FLAG_ANYSRC; 4735 } 4736 4737 return 0; 4738 } 4739 4740 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { 4741 .func = bpf_skb_set_tunnel_key, 4742 .gpl_only = false, 4743 .ret_type = RET_INTEGER, 4744 .arg1_type = ARG_PTR_TO_CTX, 4745 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4746 .arg3_type = ARG_CONST_SIZE, 4747 .arg4_type = ARG_ANYTHING, 4748 }; 4749 4750 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb, 4751 const u8 *, from, u32, size) 4752 { 4753 struct ip_tunnel_info *info = skb_tunnel_info(skb); 4754 const struct metadata_dst *md = this_cpu_ptr(md_dst); 4755 4756 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) 4757 return -EINVAL; 4758 if (unlikely(size > IP_TUNNEL_OPTS_MAX)) 4759 return -ENOMEM; 4760 4761 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT); 4762 4763 return 0; 4764 } 4765 4766 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { 4767 .func = bpf_skb_set_tunnel_opt, 4768 .gpl_only = false, 4769 .ret_type = RET_INTEGER, 4770 .arg1_type = ARG_PTR_TO_CTX, 4771 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4772 .arg3_type = ARG_CONST_SIZE, 4773 }; 4774 4775 static const struct bpf_func_proto * 4776 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) 4777 { 4778 if (!md_dst) { 4779 struct metadata_dst __percpu *tmp; 4780 4781 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, 4782 METADATA_IP_TUNNEL, 4783 GFP_KERNEL); 4784 if (!tmp) 4785 return NULL; 4786 if (cmpxchg(&md_dst, NULL, tmp)) 4787 metadata_dst_free_percpu(tmp); 4788 } 4789 4790 switch (which) { 4791 case BPF_FUNC_skb_set_tunnel_key: 4792 return &bpf_skb_set_tunnel_key_proto; 4793 case BPF_FUNC_skb_set_tunnel_opt: 4794 return &bpf_skb_set_tunnel_opt_proto; 4795 default: 4796 return NULL; 4797 } 4798 } 4799 4800 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map, 4801 u32, idx) 4802 { 4803 struct bpf_array *array = container_of(map, struct bpf_array, map); 4804 struct cgroup *cgrp; 4805 struct sock *sk; 4806 4807 sk = skb_to_full_sk(skb); 4808 if (!sk || !sk_fullsock(sk)) 4809 return -ENOENT; 4810 if (unlikely(idx >= array->map.max_entries)) 4811 return -E2BIG; 4812 4813 cgrp = READ_ONCE(array->ptrs[idx]); 4814 if (unlikely(!cgrp)) 4815 return -EAGAIN; 4816 4817 return sk_under_cgroup_hierarchy(sk, cgrp); 4818 } 4819 4820 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = { 4821 .func = bpf_skb_under_cgroup, 4822 .gpl_only = false, 4823 .ret_type = RET_INTEGER, 4824 .arg1_type = ARG_PTR_TO_CTX, 4825 .arg2_type = ARG_CONST_MAP_PTR, 4826 .arg3_type = ARG_ANYTHING, 4827 }; 4828 4829 #ifdef CONFIG_SOCK_CGROUP_DATA 4830 static inline u64 __bpf_sk_cgroup_id(struct sock *sk) 4831 { 4832 struct cgroup *cgrp; 4833 4834 sk = sk_to_full_sk(sk); 4835 if (!sk || !sk_fullsock(sk)) 4836 return 0; 4837 4838 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4839 return cgroup_id(cgrp); 4840 } 4841 4842 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb) 4843 { 4844 return __bpf_sk_cgroup_id(skb->sk); 4845 } 4846 4847 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = { 4848 .func = bpf_skb_cgroup_id, 4849 .gpl_only = false, 4850 .ret_type = RET_INTEGER, 4851 .arg1_type = ARG_PTR_TO_CTX, 4852 }; 4853 4854 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk, 4855 int ancestor_level) 4856 { 4857 struct cgroup *ancestor; 4858 struct cgroup *cgrp; 4859 4860 sk = sk_to_full_sk(sk); 4861 if (!sk || !sk_fullsock(sk)) 4862 return 0; 4863 4864 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4865 ancestor = cgroup_ancestor(cgrp, ancestor_level); 4866 if (!ancestor) 4867 return 0; 4868 4869 return cgroup_id(ancestor); 4870 } 4871 4872 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int, 4873 ancestor_level) 4874 { 4875 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level); 4876 } 4877 4878 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = { 4879 .func = bpf_skb_ancestor_cgroup_id, 4880 .gpl_only = false, 4881 .ret_type = RET_INTEGER, 4882 .arg1_type = ARG_PTR_TO_CTX, 4883 .arg2_type = ARG_ANYTHING, 4884 }; 4885 4886 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk) 4887 { 4888 return __bpf_sk_cgroup_id(sk); 4889 } 4890 4891 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = { 4892 .func = bpf_sk_cgroup_id, 4893 .gpl_only = false, 4894 .ret_type = RET_INTEGER, 4895 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4896 }; 4897 4898 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level) 4899 { 4900 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level); 4901 } 4902 4903 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = { 4904 .func = bpf_sk_ancestor_cgroup_id, 4905 .gpl_only = false, 4906 .ret_type = RET_INTEGER, 4907 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4908 .arg2_type = ARG_ANYTHING, 4909 }; 4910 #endif 4911 4912 static unsigned long bpf_xdp_copy(void *dst, const void *ctx, 4913 unsigned long off, unsigned long len) 4914 { 4915 struct xdp_buff *xdp = (struct xdp_buff *)ctx; 4916 4917 bpf_xdp_copy_buf(xdp, off, dst, len, false); 4918 return 0; 4919 } 4920 4921 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map, 4922 u64, flags, void *, meta, u64, meta_size) 4923 { 4924 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4925 4926 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4927 return -EINVAL; 4928 4929 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp))) 4930 return -EFAULT; 4931 4932 return bpf_event_output(map, flags, meta, meta_size, xdp, 4933 xdp_size, bpf_xdp_copy); 4934 } 4935 4936 static const struct bpf_func_proto bpf_xdp_event_output_proto = { 4937 .func = bpf_xdp_event_output, 4938 .gpl_only = true, 4939 .ret_type = RET_INTEGER, 4940 .arg1_type = ARG_PTR_TO_CTX, 4941 .arg2_type = ARG_CONST_MAP_PTR, 4942 .arg3_type = ARG_ANYTHING, 4943 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4944 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4945 }; 4946 4947 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff) 4948 4949 const struct bpf_func_proto bpf_xdp_output_proto = { 4950 .func = bpf_xdp_event_output, 4951 .gpl_only = true, 4952 .ret_type = RET_INTEGER, 4953 .arg1_type = ARG_PTR_TO_BTF_ID, 4954 .arg1_btf_id = &bpf_xdp_output_btf_ids[0], 4955 .arg2_type = ARG_CONST_MAP_PTR, 4956 .arg3_type = ARG_ANYTHING, 4957 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4958 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4959 }; 4960 4961 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb) 4962 { 4963 return skb->sk ? __sock_gen_cookie(skb->sk) : 0; 4964 } 4965 4966 static const struct bpf_func_proto bpf_get_socket_cookie_proto = { 4967 .func = bpf_get_socket_cookie, 4968 .gpl_only = false, 4969 .ret_type = RET_INTEGER, 4970 .arg1_type = ARG_PTR_TO_CTX, 4971 }; 4972 4973 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4974 { 4975 return __sock_gen_cookie(ctx->sk); 4976 } 4977 4978 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = { 4979 .func = bpf_get_socket_cookie_sock_addr, 4980 .gpl_only = false, 4981 .ret_type = RET_INTEGER, 4982 .arg1_type = ARG_PTR_TO_CTX, 4983 }; 4984 4985 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx) 4986 { 4987 return __sock_gen_cookie(ctx); 4988 } 4989 4990 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = { 4991 .func = bpf_get_socket_cookie_sock, 4992 .gpl_only = false, 4993 .ret_type = RET_INTEGER, 4994 .arg1_type = ARG_PTR_TO_CTX, 4995 }; 4996 4997 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk) 4998 { 4999 return sk ? sock_gen_cookie(sk) : 0; 5000 } 5001 5002 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = { 5003 .func = bpf_get_socket_ptr_cookie, 5004 .gpl_only = false, 5005 .ret_type = RET_INTEGER, 5006 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | PTR_MAYBE_NULL, 5007 }; 5008 5009 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 5010 { 5011 return __sock_gen_cookie(ctx->sk); 5012 } 5013 5014 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = { 5015 .func = bpf_get_socket_cookie_sock_ops, 5016 .gpl_only = false, 5017 .ret_type = RET_INTEGER, 5018 .arg1_type = ARG_PTR_TO_CTX, 5019 }; 5020 5021 static u64 __bpf_get_netns_cookie(struct sock *sk) 5022 { 5023 const struct net *net = sk ? sock_net(sk) : &init_net; 5024 5025 return net->net_cookie; 5026 } 5027 5028 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx) 5029 { 5030 return __bpf_get_netns_cookie(ctx); 5031 } 5032 5033 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = { 5034 .func = bpf_get_netns_cookie_sock, 5035 .gpl_only = false, 5036 .ret_type = RET_INTEGER, 5037 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5038 }; 5039 5040 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 5041 { 5042 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 5043 } 5044 5045 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = { 5046 .func = bpf_get_netns_cookie_sock_addr, 5047 .gpl_only = false, 5048 .ret_type = RET_INTEGER, 5049 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5050 }; 5051 5052 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 5053 { 5054 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 5055 } 5056 5057 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = { 5058 .func = bpf_get_netns_cookie_sock_ops, 5059 .gpl_only = false, 5060 .ret_type = RET_INTEGER, 5061 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5062 }; 5063 5064 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx) 5065 { 5066 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 5067 } 5068 5069 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = { 5070 .func = bpf_get_netns_cookie_sk_msg, 5071 .gpl_only = false, 5072 .ret_type = RET_INTEGER, 5073 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 5074 }; 5075 5076 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb) 5077 { 5078 struct sock *sk = sk_to_full_sk(skb->sk); 5079 kuid_t kuid; 5080 5081 if (!sk || !sk_fullsock(sk)) 5082 return overflowuid; 5083 kuid = sock_net_uid(sock_net(sk), sk); 5084 return from_kuid_munged(sock_net(sk)->user_ns, kuid); 5085 } 5086 5087 static const struct bpf_func_proto bpf_get_socket_uid_proto = { 5088 .func = bpf_get_socket_uid, 5089 .gpl_only = false, 5090 .ret_type = RET_INTEGER, 5091 .arg1_type = ARG_PTR_TO_CTX, 5092 }; 5093 5094 static int sol_socket_sockopt(struct sock *sk, int optname, 5095 char *optval, int *optlen, 5096 bool getopt) 5097 { 5098 switch (optname) { 5099 case SO_REUSEADDR: 5100 case SO_SNDBUF: 5101 case SO_RCVBUF: 5102 case SO_KEEPALIVE: 5103 case SO_PRIORITY: 5104 case SO_REUSEPORT: 5105 case SO_RCVLOWAT: 5106 case SO_MARK: 5107 case SO_MAX_PACING_RATE: 5108 case SO_BINDTOIFINDEX: 5109 case SO_TXREHASH: 5110 if (*optlen != sizeof(int)) 5111 return -EINVAL; 5112 break; 5113 case SO_BINDTODEVICE: 5114 break; 5115 default: 5116 return -EINVAL; 5117 } 5118 5119 if (getopt) { 5120 if (optname == SO_BINDTODEVICE) 5121 return -EINVAL; 5122 return sk_getsockopt(sk, SOL_SOCKET, optname, 5123 KERNEL_SOCKPTR(optval), 5124 KERNEL_SOCKPTR(optlen)); 5125 } 5126 5127 return sk_setsockopt(sk, SOL_SOCKET, optname, 5128 KERNEL_SOCKPTR(optval), *optlen); 5129 } 5130 5131 static int bpf_sol_tcp_setsockopt(struct sock *sk, int optname, 5132 char *optval, int optlen) 5133 { 5134 struct tcp_sock *tp = tcp_sk(sk); 5135 unsigned long timeout; 5136 int val; 5137 5138 if (optlen != sizeof(int)) 5139 return -EINVAL; 5140 5141 val = *(int *)optval; 5142 5143 /* Only some options are supported */ 5144 switch (optname) { 5145 case TCP_BPF_IW: 5146 if (val <= 0 || tp->data_segs_out > tp->syn_data) 5147 return -EINVAL; 5148 tcp_snd_cwnd_set(tp, val); 5149 break; 5150 case TCP_BPF_SNDCWND_CLAMP: 5151 if (val <= 0) 5152 return -EINVAL; 5153 tp->snd_cwnd_clamp = val; 5154 tp->snd_ssthresh = val; 5155 break; 5156 case TCP_BPF_DELACK_MAX: 5157 timeout = usecs_to_jiffies(val); 5158 if (timeout > TCP_DELACK_MAX || 5159 timeout < TCP_TIMEOUT_MIN) 5160 return -EINVAL; 5161 inet_csk(sk)->icsk_delack_max = timeout; 5162 break; 5163 case TCP_BPF_RTO_MIN: 5164 timeout = usecs_to_jiffies(val); 5165 if (timeout > TCP_RTO_MIN || 5166 timeout < TCP_TIMEOUT_MIN) 5167 return -EINVAL; 5168 inet_csk(sk)->icsk_rto_min = timeout; 5169 break; 5170 default: 5171 return -EINVAL; 5172 } 5173 5174 return 0; 5175 } 5176 5177 static int sol_tcp_sockopt_congestion(struct sock *sk, char *optval, 5178 int *optlen, bool getopt) 5179 { 5180 struct tcp_sock *tp; 5181 int ret; 5182 5183 if (*optlen < 2) 5184 return -EINVAL; 5185 5186 if (getopt) { 5187 if (!inet_csk(sk)->icsk_ca_ops) 5188 return -EINVAL; 5189 /* BPF expects NULL-terminated tcp-cc string */ 5190 optval[--(*optlen)] = '\0'; 5191 return do_tcp_getsockopt(sk, SOL_TCP, TCP_CONGESTION, 5192 KERNEL_SOCKPTR(optval), 5193 KERNEL_SOCKPTR(optlen)); 5194 } 5195 5196 /* "cdg" is the only cc that alloc a ptr 5197 * in inet_csk_ca area. The bpf-tcp-cc may 5198 * overwrite this ptr after switching to cdg. 5199 */ 5200 if (*optlen >= sizeof("cdg") - 1 && !strncmp("cdg", optval, *optlen)) 5201 return -ENOTSUPP; 5202 5203 /* It stops this looping 5204 * 5205 * .init => bpf_setsockopt(tcp_cc) => .init => 5206 * bpf_setsockopt(tcp_cc)" => .init => .... 5207 * 5208 * The second bpf_setsockopt(tcp_cc) is not allowed 5209 * in order to break the loop when both .init 5210 * are the same bpf prog. 5211 * 5212 * This applies even the second bpf_setsockopt(tcp_cc) 5213 * does not cause a loop. This limits only the first 5214 * '.init' can call bpf_setsockopt(TCP_CONGESTION) to 5215 * pick a fallback cc (eg. peer does not support ECN) 5216 * and the second '.init' cannot fallback to 5217 * another. 5218 */ 5219 tp = tcp_sk(sk); 5220 if (tp->bpf_chg_cc_inprogress) 5221 return -EBUSY; 5222 5223 tp->bpf_chg_cc_inprogress = 1; 5224 ret = do_tcp_setsockopt(sk, SOL_TCP, TCP_CONGESTION, 5225 KERNEL_SOCKPTR(optval), *optlen); 5226 tp->bpf_chg_cc_inprogress = 0; 5227 return ret; 5228 } 5229 5230 static int sol_tcp_sockopt(struct sock *sk, int optname, 5231 char *optval, int *optlen, 5232 bool getopt) 5233 { 5234 if (sk->sk_protocol != IPPROTO_TCP) 5235 return -EINVAL; 5236 5237 switch (optname) { 5238 case TCP_NODELAY: 5239 case TCP_MAXSEG: 5240 case TCP_KEEPIDLE: 5241 case TCP_KEEPINTVL: 5242 case TCP_KEEPCNT: 5243 case TCP_SYNCNT: 5244 case TCP_WINDOW_CLAMP: 5245 case TCP_THIN_LINEAR_TIMEOUTS: 5246 case TCP_USER_TIMEOUT: 5247 case TCP_NOTSENT_LOWAT: 5248 case TCP_SAVE_SYN: 5249 if (*optlen != sizeof(int)) 5250 return -EINVAL; 5251 break; 5252 case TCP_CONGESTION: 5253 return sol_tcp_sockopt_congestion(sk, optval, optlen, getopt); 5254 case TCP_SAVED_SYN: 5255 if (*optlen < 1) 5256 return -EINVAL; 5257 break; 5258 default: 5259 if (getopt) 5260 return -EINVAL; 5261 return bpf_sol_tcp_setsockopt(sk, optname, optval, *optlen); 5262 } 5263 5264 if (getopt) { 5265 if (optname == TCP_SAVED_SYN) { 5266 struct tcp_sock *tp = tcp_sk(sk); 5267 5268 if (!tp->saved_syn || 5269 *optlen > tcp_saved_syn_len(tp->saved_syn)) 5270 return -EINVAL; 5271 memcpy(optval, tp->saved_syn->data, *optlen); 5272 /* It cannot free tp->saved_syn here because it 5273 * does not know if the user space still needs it. 5274 */ 5275 return 0; 5276 } 5277 5278 return do_tcp_getsockopt(sk, SOL_TCP, optname, 5279 KERNEL_SOCKPTR(optval), 5280 KERNEL_SOCKPTR(optlen)); 5281 } 5282 5283 return do_tcp_setsockopt(sk, SOL_TCP, optname, 5284 KERNEL_SOCKPTR(optval), *optlen); 5285 } 5286 5287 static int sol_ip_sockopt(struct sock *sk, int optname, 5288 char *optval, int *optlen, 5289 bool getopt) 5290 { 5291 if (sk->sk_family != AF_INET) 5292 return -EINVAL; 5293 5294 switch (optname) { 5295 case IP_TOS: 5296 if (*optlen != sizeof(int)) 5297 return -EINVAL; 5298 break; 5299 default: 5300 return -EINVAL; 5301 } 5302 5303 if (getopt) 5304 return do_ip_getsockopt(sk, SOL_IP, optname, 5305 KERNEL_SOCKPTR(optval), 5306 KERNEL_SOCKPTR(optlen)); 5307 5308 return do_ip_setsockopt(sk, SOL_IP, optname, 5309 KERNEL_SOCKPTR(optval), *optlen); 5310 } 5311 5312 static int sol_ipv6_sockopt(struct sock *sk, int optname, 5313 char *optval, int *optlen, 5314 bool getopt) 5315 { 5316 if (sk->sk_family != AF_INET6) 5317 return -EINVAL; 5318 5319 switch (optname) { 5320 case IPV6_TCLASS: 5321 case IPV6_AUTOFLOWLABEL: 5322 if (*optlen != sizeof(int)) 5323 return -EINVAL; 5324 break; 5325 default: 5326 return -EINVAL; 5327 } 5328 5329 if (getopt) 5330 return ipv6_bpf_stub->ipv6_getsockopt(sk, SOL_IPV6, optname, 5331 KERNEL_SOCKPTR(optval), 5332 KERNEL_SOCKPTR(optlen)); 5333 5334 return ipv6_bpf_stub->ipv6_setsockopt(sk, SOL_IPV6, optname, 5335 KERNEL_SOCKPTR(optval), *optlen); 5336 } 5337 5338 static int __bpf_setsockopt(struct sock *sk, int level, int optname, 5339 char *optval, int optlen) 5340 { 5341 if (!sk_fullsock(sk)) 5342 return -EINVAL; 5343 5344 if (level == SOL_SOCKET) 5345 return sol_socket_sockopt(sk, optname, optval, &optlen, false); 5346 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP) 5347 return sol_ip_sockopt(sk, optname, optval, &optlen, false); 5348 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6) 5349 return sol_ipv6_sockopt(sk, optname, optval, &optlen, false); 5350 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP) 5351 return sol_tcp_sockopt(sk, optname, optval, &optlen, false); 5352 5353 return -EINVAL; 5354 } 5355 5356 static int _bpf_setsockopt(struct sock *sk, int level, int optname, 5357 char *optval, int optlen) 5358 { 5359 if (sk_fullsock(sk)) 5360 sock_owned_by_me(sk); 5361 return __bpf_setsockopt(sk, level, optname, optval, optlen); 5362 } 5363 5364 static int __bpf_getsockopt(struct sock *sk, int level, int optname, 5365 char *optval, int optlen) 5366 { 5367 int err, saved_optlen = optlen; 5368 5369 if (!sk_fullsock(sk)) { 5370 err = -EINVAL; 5371 goto done; 5372 } 5373 5374 if (level == SOL_SOCKET) 5375 err = sol_socket_sockopt(sk, optname, optval, &optlen, true); 5376 else if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP) 5377 err = sol_tcp_sockopt(sk, optname, optval, &optlen, true); 5378 else if (IS_ENABLED(CONFIG_INET) && level == SOL_IP) 5379 err = sol_ip_sockopt(sk, optname, optval, &optlen, true); 5380 else if (IS_ENABLED(CONFIG_IPV6) && level == SOL_IPV6) 5381 err = sol_ipv6_sockopt(sk, optname, optval, &optlen, true); 5382 else 5383 err = -EINVAL; 5384 5385 done: 5386 if (err) 5387 optlen = 0; 5388 if (optlen < saved_optlen) 5389 memset(optval + optlen, 0, saved_optlen - optlen); 5390 return err; 5391 } 5392 5393 static int _bpf_getsockopt(struct sock *sk, int level, int optname, 5394 char *optval, int optlen) 5395 { 5396 if (sk_fullsock(sk)) 5397 sock_owned_by_me(sk); 5398 return __bpf_getsockopt(sk, level, optname, optval, optlen); 5399 } 5400 5401 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level, 5402 int, optname, char *, optval, int, optlen) 5403 { 5404 return _bpf_setsockopt(sk, level, optname, optval, optlen); 5405 } 5406 5407 const struct bpf_func_proto bpf_sk_setsockopt_proto = { 5408 .func = bpf_sk_setsockopt, 5409 .gpl_only = false, 5410 .ret_type = RET_INTEGER, 5411 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5412 .arg2_type = ARG_ANYTHING, 5413 .arg3_type = ARG_ANYTHING, 5414 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5415 .arg5_type = ARG_CONST_SIZE, 5416 }; 5417 5418 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level, 5419 int, optname, char *, optval, int, optlen) 5420 { 5421 return _bpf_getsockopt(sk, level, optname, optval, optlen); 5422 } 5423 5424 const struct bpf_func_proto bpf_sk_getsockopt_proto = { 5425 .func = bpf_sk_getsockopt, 5426 .gpl_only = false, 5427 .ret_type = RET_INTEGER, 5428 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5429 .arg2_type = ARG_ANYTHING, 5430 .arg3_type = ARG_ANYTHING, 5431 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5432 .arg5_type = ARG_CONST_SIZE, 5433 }; 5434 5435 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level, 5436 int, optname, char *, optval, int, optlen) 5437 { 5438 return __bpf_setsockopt(sk, level, optname, optval, optlen); 5439 } 5440 5441 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = { 5442 .func = bpf_unlocked_sk_setsockopt, 5443 .gpl_only = false, 5444 .ret_type = RET_INTEGER, 5445 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5446 .arg2_type = ARG_ANYTHING, 5447 .arg3_type = ARG_ANYTHING, 5448 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5449 .arg5_type = ARG_CONST_SIZE, 5450 }; 5451 5452 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level, 5453 int, optname, char *, optval, int, optlen) 5454 { 5455 return __bpf_getsockopt(sk, level, optname, optval, optlen); 5456 } 5457 5458 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = { 5459 .func = bpf_unlocked_sk_getsockopt, 5460 .gpl_only = false, 5461 .ret_type = RET_INTEGER, 5462 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5463 .arg2_type = ARG_ANYTHING, 5464 .arg3_type = ARG_ANYTHING, 5465 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5466 .arg5_type = ARG_CONST_SIZE, 5467 }; 5468 5469 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx, 5470 int, level, int, optname, char *, optval, int, optlen) 5471 { 5472 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen); 5473 } 5474 5475 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = { 5476 .func = bpf_sock_addr_setsockopt, 5477 .gpl_only = false, 5478 .ret_type = RET_INTEGER, 5479 .arg1_type = ARG_PTR_TO_CTX, 5480 .arg2_type = ARG_ANYTHING, 5481 .arg3_type = ARG_ANYTHING, 5482 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5483 .arg5_type = ARG_CONST_SIZE, 5484 }; 5485 5486 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx, 5487 int, level, int, optname, char *, optval, int, optlen) 5488 { 5489 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen); 5490 } 5491 5492 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = { 5493 .func = bpf_sock_addr_getsockopt, 5494 .gpl_only = false, 5495 .ret_type = RET_INTEGER, 5496 .arg1_type = ARG_PTR_TO_CTX, 5497 .arg2_type = ARG_ANYTHING, 5498 .arg3_type = ARG_ANYTHING, 5499 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5500 .arg5_type = ARG_CONST_SIZE, 5501 }; 5502 5503 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5504 int, level, int, optname, char *, optval, int, optlen) 5505 { 5506 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen); 5507 } 5508 5509 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = { 5510 .func = bpf_sock_ops_setsockopt, 5511 .gpl_only = false, 5512 .ret_type = RET_INTEGER, 5513 .arg1_type = ARG_PTR_TO_CTX, 5514 .arg2_type = ARG_ANYTHING, 5515 .arg3_type = ARG_ANYTHING, 5516 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5517 .arg5_type = ARG_CONST_SIZE, 5518 }; 5519 5520 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock, 5521 int optname, const u8 **start) 5522 { 5523 struct sk_buff *syn_skb = bpf_sock->syn_skb; 5524 const u8 *hdr_start; 5525 int ret; 5526 5527 if (syn_skb) { 5528 /* sk is a request_sock here */ 5529 5530 if (optname == TCP_BPF_SYN) { 5531 hdr_start = syn_skb->data; 5532 ret = tcp_hdrlen(syn_skb); 5533 } else if (optname == TCP_BPF_SYN_IP) { 5534 hdr_start = skb_network_header(syn_skb); 5535 ret = skb_network_header_len(syn_skb) + 5536 tcp_hdrlen(syn_skb); 5537 } else { 5538 /* optname == TCP_BPF_SYN_MAC */ 5539 hdr_start = skb_mac_header(syn_skb); 5540 ret = skb_mac_header_len(syn_skb) + 5541 skb_network_header_len(syn_skb) + 5542 tcp_hdrlen(syn_skb); 5543 } 5544 } else { 5545 struct sock *sk = bpf_sock->sk; 5546 struct saved_syn *saved_syn; 5547 5548 if (sk->sk_state == TCP_NEW_SYN_RECV) 5549 /* synack retransmit. bpf_sock->syn_skb will 5550 * not be available. It has to resort to 5551 * saved_syn (if it is saved). 5552 */ 5553 saved_syn = inet_reqsk(sk)->saved_syn; 5554 else 5555 saved_syn = tcp_sk(sk)->saved_syn; 5556 5557 if (!saved_syn) 5558 return -ENOENT; 5559 5560 if (optname == TCP_BPF_SYN) { 5561 hdr_start = saved_syn->data + 5562 saved_syn->mac_hdrlen + 5563 saved_syn->network_hdrlen; 5564 ret = saved_syn->tcp_hdrlen; 5565 } else if (optname == TCP_BPF_SYN_IP) { 5566 hdr_start = saved_syn->data + 5567 saved_syn->mac_hdrlen; 5568 ret = saved_syn->network_hdrlen + 5569 saved_syn->tcp_hdrlen; 5570 } else { 5571 /* optname == TCP_BPF_SYN_MAC */ 5572 5573 /* TCP_SAVE_SYN may not have saved the mac hdr */ 5574 if (!saved_syn->mac_hdrlen) 5575 return -ENOENT; 5576 5577 hdr_start = saved_syn->data; 5578 ret = saved_syn->mac_hdrlen + 5579 saved_syn->network_hdrlen + 5580 saved_syn->tcp_hdrlen; 5581 } 5582 } 5583 5584 *start = hdr_start; 5585 return ret; 5586 } 5587 5588 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5589 int, level, int, optname, char *, optval, int, optlen) 5590 { 5591 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP && 5592 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) { 5593 int ret, copy_len = 0; 5594 const u8 *start; 5595 5596 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start); 5597 if (ret > 0) { 5598 copy_len = ret; 5599 if (optlen < copy_len) { 5600 copy_len = optlen; 5601 ret = -ENOSPC; 5602 } 5603 5604 memcpy(optval, start, copy_len); 5605 } 5606 5607 /* Zero out unused buffer at the end */ 5608 memset(optval + copy_len, 0, optlen - copy_len); 5609 5610 return ret; 5611 } 5612 5613 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen); 5614 } 5615 5616 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = { 5617 .func = bpf_sock_ops_getsockopt, 5618 .gpl_only = false, 5619 .ret_type = RET_INTEGER, 5620 .arg1_type = ARG_PTR_TO_CTX, 5621 .arg2_type = ARG_ANYTHING, 5622 .arg3_type = ARG_ANYTHING, 5623 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5624 .arg5_type = ARG_CONST_SIZE, 5625 }; 5626 5627 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock, 5628 int, argval) 5629 { 5630 struct sock *sk = bpf_sock->sk; 5631 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS; 5632 5633 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk)) 5634 return -EINVAL; 5635 5636 tcp_sk(sk)->bpf_sock_ops_cb_flags = val; 5637 5638 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS); 5639 } 5640 5641 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = { 5642 .func = bpf_sock_ops_cb_flags_set, 5643 .gpl_only = false, 5644 .ret_type = RET_INTEGER, 5645 .arg1_type = ARG_PTR_TO_CTX, 5646 .arg2_type = ARG_ANYTHING, 5647 }; 5648 5649 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly; 5650 EXPORT_SYMBOL_GPL(ipv6_bpf_stub); 5651 5652 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr, 5653 int, addr_len) 5654 { 5655 #ifdef CONFIG_INET 5656 struct sock *sk = ctx->sk; 5657 u32 flags = BIND_FROM_BPF; 5658 int err; 5659 5660 err = -EINVAL; 5661 if (addr_len < offsetofend(struct sockaddr, sa_family)) 5662 return err; 5663 if (addr->sa_family == AF_INET) { 5664 if (addr_len < sizeof(struct sockaddr_in)) 5665 return err; 5666 if (((struct sockaddr_in *)addr)->sin_port == htons(0)) 5667 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5668 return __inet_bind(sk, addr, addr_len, flags); 5669 #if IS_ENABLED(CONFIG_IPV6) 5670 } else if (addr->sa_family == AF_INET6) { 5671 if (addr_len < SIN6_LEN_RFC2133) 5672 return err; 5673 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0)) 5674 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5675 /* ipv6_bpf_stub cannot be NULL, since it's called from 5676 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded 5677 */ 5678 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags); 5679 #endif /* CONFIG_IPV6 */ 5680 } 5681 #endif /* CONFIG_INET */ 5682 5683 return -EAFNOSUPPORT; 5684 } 5685 5686 static const struct bpf_func_proto bpf_bind_proto = { 5687 .func = bpf_bind, 5688 .gpl_only = false, 5689 .ret_type = RET_INTEGER, 5690 .arg1_type = ARG_PTR_TO_CTX, 5691 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5692 .arg3_type = ARG_CONST_SIZE, 5693 }; 5694 5695 #ifdef CONFIG_XFRM 5696 5697 #if (IS_BUILTIN(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF)) || \ 5698 (IS_MODULE(CONFIG_XFRM_INTERFACE) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES)) 5699 5700 struct metadata_dst __percpu *xfrm_bpf_md_dst; 5701 EXPORT_SYMBOL_GPL(xfrm_bpf_md_dst); 5702 5703 #endif 5704 5705 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index, 5706 struct bpf_xfrm_state *, to, u32, size, u64, flags) 5707 { 5708 const struct sec_path *sp = skb_sec_path(skb); 5709 const struct xfrm_state *x; 5710 5711 if (!sp || unlikely(index >= sp->len || flags)) 5712 goto err_clear; 5713 5714 x = sp->xvec[index]; 5715 5716 if (unlikely(size != sizeof(struct bpf_xfrm_state))) 5717 goto err_clear; 5718 5719 to->reqid = x->props.reqid; 5720 to->spi = x->id.spi; 5721 to->family = x->props.family; 5722 to->ext = 0; 5723 5724 if (to->family == AF_INET6) { 5725 memcpy(to->remote_ipv6, x->props.saddr.a6, 5726 sizeof(to->remote_ipv6)); 5727 } else { 5728 to->remote_ipv4 = x->props.saddr.a4; 5729 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 5730 } 5731 5732 return 0; 5733 err_clear: 5734 memset(to, 0, size); 5735 return -EINVAL; 5736 } 5737 5738 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = { 5739 .func = bpf_skb_get_xfrm_state, 5740 .gpl_only = false, 5741 .ret_type = RET_INTEGER, 5742 .arg1_type = ARG_PTR_TO_CTX, 5743 .arg2_type = ARG_ANYTHING, 5744 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 5745 .arg4_type = ARG_CONST_SIZE, 5746 .arg5_type = ARG_ANYTHING, 5747 }; 5748 #endif 5749 5750 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6) 5751 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, u32 mtu) 5752 { 5753 params->h_vlan_TCI = 0; 5754 params->h_vlan_proto = 0; 5755 if (mtu) 5756 params->mtu_result = mtu; /* union with tot_len */ 5757 5758 return 0; 5759 } 5760 #endif 5761 5762 #if IS_ENABLED(CONFIG_INET) 5763 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5764 u32 flags, bool check_mtu) 5765 { 5766 struct fib_nh_common *nhc; 5767 struct in_device *in_dev; 5768 struct neighbour *neigh; 5769 struct net_device *dev; 5770 struct fib_result res; 5771 struct flowi4 fl4; 5772 u32 mtu = 0; 5773 int err; 5774 5775 dev = dev_get_by_index_rcu(net, params->ifindex); 5776 if (unlikely(!dev)) 5777 return -ENODEV; 5778 5779 /* verify forwarding is enabled on this interface */ 5780 in_dev = __in_dev_get_rcu(dev); 5781 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev))) 5782 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5783 5784 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5785 fl4.flowi4_iif = 1; 5786 fl4.flowi4_oif = params->ifindex; 5787 } else { 5788 fl4.flowi4_iif = params->ifindex; 5789 fl4.flowi4_oif = 0; 5790 } 5791 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK; 5792 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 5793 fl4.flowi4_flags = 0; 5794 5795 fl4.flowi4_proto = params->l4_protocol; 5796 fl4.daddr = params->ipv4_dst; 5797 fl4.saddr = params->ipv4_src; 5798 fl4.fl4_sport = params->sport; 5799 fl4.fl4_dport = params->dport; 5800 fl4.flowi4_multipath_hash = 0; 5801 5802 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5803 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5804 struct fib_table *tb; 5805 5806 tb = fib_get_table(net, tbid); 5807 if (unlikely(!tb)) 5808 return BPF_FIB_LKUP_RET_NOT_FWDED; 5809 5810 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 5811 } else { 5812 fl4.flowi4_mark = 0; 5813 fl4.flowi4_secid = 0; 5814 fl4.flowi4_tun_key.tun_id = 0; 5815 fl4.flowi4_uid = sock_net_uid(net, NULL); 5816 5817 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF); 5818 } 5819 5820 if (err) { 5821 /* map fib lookup errors to RTN_ type */ 5822 if (err == -EINVAL) 5823 return BPF_FIB_LKUP_RET_BLACKHOLE; 5824 if (err == -EHOSTUNREACH) 5825 return BPF_FIB_LKUP_RET_UNREACHABLE; 5826 if (err == -EACCES) 5827 return BPF_FIB_LKUP_RET_PROHIBIT; 5828 5829 return BPF_FIB_LKUP_RET_NOT_FWDED; 5830 } 5831 5832 if (res.type != RTN_UNICAST) 5833 return BPF_FIB_LKUP_RET_NOT_FWDED; 5834 5835 if (fib_info_num_path(res.fi) > 1) 5836 fib_select_path(net, &res, &fl4, NULL); 5837 5838 if (check_mtu) { 5839 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst); 5840 if (params->tot_len > mtu) { 5841 params->mtu_result = mtu; /* union with tot_len */ 5842 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5843 } 5844 } 5845 5846 nhc = res.nhc; 5847 5848 /* do not handle lwt encaps right now */ 5849 if (nhc->nhc_lwtstate) 5850 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5851 5852 dev = nhc->nhc_dev; 5853 5854 params->rt_metric = res.fi->fib_priority; 5855 params->ifindex = dev->ifindex; 5856 5857 /* xdp and cls_bpf programs are run in RCU-bh so 5858 * rcu_read_lock_bh is not needed here 5859 */ 5860 if (likely(nhc->nhc_gw_family != AF_INET6)) { 5861 if (nhc->nhc_gw_family) 5862 params->ipv4_dst = nhc->nhc_gw.ipv4; 5863 } else { 5864 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst; 5865 5866 params->family = AF_INET6; 5867 *dst = nhc->nhc_gw.ipv6; 5868 } 5869 5870 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH) 5871 goto set_fwd_params; 5872 5873 if (likely(nhc->nhc_gw_family != AF_INET6)) 5874 neigh = __ipv4_neigh_lookup_noref(dev, 5875 (__force u32)params->ipv4_dst); 5876 else 5877 neigh = __ipv6_neigh_lookup_noref_stub(dev, params->ipv6_dst); 5878 5879 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID)) 5880 return BPF_FIB_LKUP_RET_NO_NEIGH; 5881 memcpy(params->dmac, neigh->ha, ETH_ALEN); 5882 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 5883 5884 set_fwd_params: 5885 return bpf_fib_set_fwd_params(params, mtu); 5886 } 5887 #endif 5888 5889 #if IS_ENABLED(CONFIG_IPV6) 5890 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5891 u32 flags, bool check_mtu) 5892 { 5893 struct in6_addr *src = (struct in6_addr *) params->ipv6_src; 5894 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst; 5895 struct fib6_result res = {}; 5896 struct neighbour *neigh; 5897 struct net_device *dev; 5898 struct inet6_dev *idev; 5899 struct flowi6 fl6; 5900 int strict = 0; 5901 int oif, err; 5902 u32 mtu = 0; 5903 5904 /* link local addresses are never forwarded */ 5905 if (rt6_need_strict(dst) || rt6_need_strict(src)) 5906 return BPF_FIB_LKUP_RET_NOT_FWDED; 5907 5908 dev = dev_get_by_index_rcu(net, params->ifindex); 5909 if (unlikely(!dev)) 5910 return -ENODEV; 5911 5912 idev = __in6_dev_get_safely(dev); 5913 if (unlikely(!idev || !idev->cnf.forwarding)) 5914 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5915 5916 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5917 fl6.flowi6_iif = 1; 5918 oif = fl6.flowi6_oif = params->ifindex; 5919 } else { 5920 oif = fl6.flowi6_iif = params->ifindex; 5921 fl6.flowi6_oif = 0; 5922 strict = RT6_LOOKUP_F_HAS_SADDR; 5923 } 5924 fl6.flowlabel = params->flowinfo; 5925 fl6.flowi6_scope = 0; 5926 fl6.flowi6_flags = 0; 5927 fl6.mp_hash = 0; 5928 5929 fl6.flowi6_proto = params->l4_protocol; 5930 fl6.daddr = *dst; 5931 fl6.saddr = *src; 5932 fl6.fl6_sport = params->sport; 5933 fl6.fl6_dport = params->dport; 5934 5935 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5936 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5937 struct fib6_table *tb; 5938 5939 tb = ipv6_stub->fib6_get_table(net, tbid); 5940 if (unlikely(!tb)) 5941 return BPF_FIB_LKUP_RET_NOT_FWDED; 5942 5943 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res, 5944 strict); 5945 } else { 5946 fl6.flowi6_mark = 0; 5947 fl6.flowi6_secid = 0; 5948 fl6.flowi6_tun_key.tun_id = 0; 5949 fl6.flowi6_uid = sock_net_uid(net, NULL); 5950 5951 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict); 5952 } 5953 5954 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) || 5955 res.f6i == net->ipv6.fib6_null_entry)) 5956 return BPF_FIB_LKUP_RET_NOT_FWDED; 5957 5958 switch (res.fib6_type) { 5959 /* only unicast is forwarded */ 5960 case RTN_UNICAST: 5961 break; 5962 case RTN_BLACKHOLE: 5963 return BPF_FIB_LKUP_RET_BLACKHOLE; 5964 case RTN_UNREACHABLE: 5965 return BPF_FIB_LKUP_RET_UNREACHABLE; 5966 case RTN_PROHIBIT: 5967 return BPF_FIB_LKUP_RET_PROHIBIT; 5968 default: 5969 return BPF_FIB_LKUP_RET_NOT_FWDED; 5970 } 5971 5972 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif, 5973 fl6.flowi6_oif != 0, NULL, strict); 5974 5975 if (check_mtu) { 5976 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src); 5977 if (params->tot_len > mtu) { 5978 params->mtu_result = mtu; /* union with tot_len */ 5979 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5980 } 5981 } 5982 5983 if (res.nh->fib_nh_lws) 5984 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5985 5986 if (res.nh->fib_nh_gw_family) 5987 *dst = res.nh->fib_nh_gw6; 5988 5989 dev = res.nh->fib_nh_dev; 5990 params->rt_metric = res.f6i->fib6_metric; 5991 params->ifindex = dev->ifindex; 5992 5993 if (flags & BPF_FIB_LOOKUP_SKIP_NEIGH) 5994 goto set_fwd_params; 5995 5996 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is 5997 * not needed here. 5998 */ 5999 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 6000 if (!neigh || !(READ_ONCE(neigh->nud_state) & NUD_VALID)) 6001 return BPF_FIB_LKUP_RET_NO_NEIGH; 6002 memcpy(params->dmac, neigh->ha, ETH_ALEN); 6003 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 6004 6005 set_fwd_params: 6006 return bpf_fib_set_fwd_params(params, mtu); 6007 } 6008 #endif 6009 6010 #define BPF_FIB_LOOKUP_MASK (BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT | \ 6011 BPF_FIB_LOOKUP_SKIP_NEIGH) 6012 6013 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx, 6014 struct bpf_fib_lookup *, params, int, plen, u32, flags) 6015 { 6016 if (plen < sizeof(*params)) 6017 return -EINVAL; 6018 6019 if (flags & ~BPF_FIB_LOOKUP_MASK) 6020 return -EINVAL; 6021 6022 switch (params->family) { 6023 #if IS_ENABLED(CONFIG_INET) 6024 case AF_INET: 6025 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params, 6026 flags, true); 6027 #endif 6028 #if IS_ENABLED(CONFIG_IPV6) 6029 case AF_INET6: 6030 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params, 6031 flags, true); 6032 #endif 6033 } 6034 return -EAFNOSUPPORT; 6035 } 6036 6037 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = { 6038 .func = bpf_xdp_fib_lookup, 6039 .gpl_only = true, 6040 .ret_type = RET_INTEGER, 6041 .arg1_type = ARG_PTR_TO_CTX, 6042 .arg2_type = ARG_PTR_TO_MEM, 6043 .arg3_type = ARG_CONST_SIZE, 6044 .arg4_type = ARG_ANYTHING, 6045 }; 6046 6047 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb, 6048 struct bpf_fib_lookup *, params, int, plen, u32, flags) 6049 { 6050 struct net *net = dev_net(skb->dev); 6051 int rc = -EAFNOSUPPORT; 6052 bool check_mtu = false; 6053 6054 if (plen < sizeof(*params)) 6055 return -EINVAL; 6056 6057 if (flags & ~BPF_FIB_LOOKUP_MASK) 6058 return -EINVAL; 6059 6060 if (params->tot_len) 6061 check_mtu = true; 6062 6063 switch (params->family) { 6064 #if IS_ENABLED(CONFIG_INET) 6065 case AF_INET: 6066 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu); 6067 break; 6068 #endif 6069 #if IS_ENABLED(CONFIG_IPV6) 6070 case AF_INET6: 6071 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu); 6072 break; 6073 #endif 6074 } 6075 6076 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) { 6077 struct net_device *dev; 6078 6079 /* When tot_len isn't provided by user, check skb 6080 * against MTU of FIB lookup resulting net_device 6081 */ 6082 dev = dev_get_by_index_rcu(net, params->ifindex); 6083 if (!is_skb_forwardable(dev, skb)) 6084 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED; 6085 6086 params->mtu_result = dev->mtu; /* union with tot_len */ 6087 } 6088 6089 return rc; 6090 } 6091 6092 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = { 6093 .func = bpf_skb_fib_lookup, 6094 .gpl_only = true, 6095 .ret_type = RET_INTEGER, 6096 .arg1_type = ARG_PTR_TO_CTX, 6097 .arg2_type = ARG_PTR_TO_MEM, 6098 .arg3_type = ARG_CONST_SIZE, 6099 .arg4_type = ARG_ANYTHING, 6100 }; 6101 6102 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr, 6103 u32 ifindex) 6104 { 6105 struct net *netns = dev_net(dev_curr); 6106 6107 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */ 6108 if (ifindex == 0) 6109 return dev_curr; 6110 6111 return dev_get_by_index_rcu(netns, ifindex); 6112 } 6113 6114 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb, 6115 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 6116 { 6117 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 6118 struct net_device *dev = skb->dev; 6119 int skb_len, dev_len; 6120 int mtu; 6121 6122 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS))) 6123 return -EINVAL; 6124 6125 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len))) 6126 return -EINVAL; 6127 6128 dev = __dev_via_ifindex(dev, ifindex); 6129 if (unlikely(!dev)) 6130 return -ENODEV; 6131 6132 mtu = READ_ONCE(dev->mtu); 6133 6134 dev_len = mtu + dev->hard_header_len; 6135 6136 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 6137 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len; 6138 6139 skb_len += len_diff; /* minus result pass check */ 6140 if (skb_len <= dev_len) { 6141 ret = BPF_MTU_CHK_RET_SUCCESS; 6142 goto out; 6143 } 6144 /* At this point, skb->len exceed MTU, but as it include length of all 6145 * segments, it can still be below MTU. The SKB can possibly get 6146 * re-segmented in transmit path (see validate_xmit_skb). Thus, user 6147 * must choose if segs are to be MTU checked. 6148 */ 6149 if (skb_is_gso(skb)) { 6150 ret = BPF_MTU_CHK_RET_SUCCESS; 6151 6152 if (flags & BPF_MTU_CHK_SEGS && 6153 !skb_gso_validate_network_len(skb, mtu)) 6154 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG; 6155 } 6156 out: 6157 /* BPF verifier guarantees valid pointer */ 6158 *mtu_len = mtu; 6159 6160 return ret; 6161 } 6162 6163 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp, 6164 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 6165 { 6166 struct net_device *dev = xdp->rxq->dev; 6167 int xdp_len = xdp->data_end - xdp->data; 6168 int ret = BPF_MTU_CHK_RET_SUCCESS; 6169 int mtu, dev_len; 6170 6171 /* XDP variant doesn't support multi-buffer segment check (yet) */ 6172 if (unlikely(flags)) 6173 return -EINVAL; 6174 6175 dev = __dev_via_ifindex(dev, ifindex); 6176 if (unlikely(!dev)) 6177 return -ENODEV; 6178 6179 mtu = READ_ONCE(dev->mtu); 6180 6181 /* Add L2-header as dev MTU is L3 size */ 6182 dev_len = mtu + dev->hard_header_len; 6183 6184 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 6185 if (*mtu_len) 6186 xdp_len = *mtu_len + dev->hard_header_len; 6187 6188 xdp_len += len_diff; /* minus result pass check */ 6189 if (xdp_len > dev_len) 6190 ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 6191 6192 /* BPF verifier guarantees valid pointer */ 6193 *mtu_len = mtu; 6194 6195 return ret; 6196 } 6197 6198 static const struct bpf_func_proto bpf_skb_check_mtu_proto = { 6199 .func = bpf_skb_check_mtu, 6200 .gpl_only = true, 6201 .ret_type = RET_INTEGER, 6202 .arg1_type = ARG_PTR_TO_CTX, 6203 .arg2_type = ARG_ANYTHING, 6204 .arg3_type = ARG_PTR_TO_INT, 6205 .arg4_type = ARG_ANYTHING, 6206 .arg5_type = ARG_ANYTHING, 6207 }; 6208 6209 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = { 6210 .func = bpf_xdp_check_mtu, 6211 .gpl_only = true, 6212 .ret_type = RET_INTEGER, 6213 .arg1_type = ARG_PTR_TO_CTX, 6214 .arg2_type = ARG_ANYTHING, 6215 .arg3_type = ARG_PTR_TO_INT, 6216 .arg4_type = ARG_ANYTHING, 6217 .arg5_type = ARG_ANYTHING, 6218 }; 6219 6220 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6221 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len) 6222 { 6223 int err; 6224 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr; 6225 6226 if (!seg6_validate_srh(srh, len, false)) 6227 return -EINVAL; 6228 6229 switch (type) { 6230 case BPF_LWT_ENCAP_SEG6_INLINE: 6231 if (skb->protocol != htons(ETH_P_IPV6)) 6232 return -EBADMSG; 6233 6234 err = seg6_do_srh_inline(skb, srh); 6235 break; 6236 case BPF_LWT_ENCAP_SEG6: 6237 skb_reset_inner_headers(skb); 6238 skb->encapsulation = 1; 6239 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6); 6240 break; 6241 default: 6242 return -EINVAL; 6243 } 6244 6245 bpf_compute_data_pointers(skb); 6246 if (err) 6247 return err; 6248 6249 skb_set_transport_header(skb, sizeof(struct ipv6hdr)); 6250 6251 return seg6_lookup_nexthop(skb, NULL, 0); 6252 } 6253 #endif /* CONFIG_IPV6_SEG6_BPF */ 6254 6255 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6256 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, 6257 bool ingress) 6258 { 6259 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress); 6260 } 6261 #endif 6262 6263 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr, 6264 u32, len) 6265 { 6266 switch (type) { 6267 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6268 case BPF_LWT_ENCAP_SEG6: 6269 case BPF_LWT_ENCAP_SEG6_INLINE: 6270 return bpf_push_seg6_encap(skb, type, hdr, len); 6271 #endif 6272 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6273 case BPF_LWT_ENCAP_IP: 6274 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */); 6275 #endif 6276 default: 6277 return -EINVAL; 6278 } 6279 } 6280 6281 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type, 6282 void *, hdr, u32, len) 6283 { 6284 switch (type) { 6285 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6286 case BPF_LWT_ENCAP_IP: 6287 return bpf_push_ip_encap(skb, hdr, len, false /* egress */); 6288 #endif 6289 default: 6290 return -EINVAL; 6291 } 6292 } 6293 6294 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = { 6295 .func = bpf_lwt_in_push_encap, 6296 .gpl_only = false, 6297 .ret_type = RET_INTEGER, 6298 .arg1_type = ARG_PTR_TO_CTX, 6299 .arg2_type = ARG_ANYTHING, 6300 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6301 .arg4_type = ARG_CONST_SIZE 6302 }; 6303 6304 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = { 6305 .func = bpf_lwt_xmit_push_encap, 6306 .gpl_only = false, 6307 .ret_type = RET_INTEGER, 6308 .arg1_type = ARG_PTR_TO_CTX, 6309 .arg2_type = ARG_ANYTHING, 6310 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6311 .arg4_type = ARG_CONST_SIZE 6312 }; 6313 6314 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6315 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset, 6316 const void *, from, u32, len) 6317 { 6318 struct seg6_bpf_srh_state *srh_state = 6319 this_cpu_ptr(&seg6_bpf_srh_states); 6320 struct ipv6_sr_hdr *srh = srh_state->srh; 6321 void *srh_tlvs, *srh_end, *ptr; 6322 int srhoff = 0; 6323 6324 if (srh == NULL) 6325 return -EINVAL; 6326 6327 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4)); 6328 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen); 6329 6330 ptr = skb->data + offset; 6331 if (ptr >= srh_tlvs && ptr + len <= srh_end) 6332 srh_state->valid = false; 6333 else if (ptr < (void *)&srh->flags || 6334 ptr + len > (void *)&srh->segments) 6335 return -EFAULT; 6336 6337 if (unlikely(bpf_try_make_writable(skb, offset + len))) 6338 return -EFAULT; 6339 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 6340 return -EINVAL; 6341 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6342 6343 memcpy(skb->data + offset, from, len); 6344 return 0; 6345 } 6346 6347 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = { 6348 .func = bpf_lwt_seg6_store_bytes, 6349 .gpl_only = false, 6350 .ret_type = RET_INTEGER, 6351 .arg1_type = ARG_PTR_TO_CTX, 6352 .arg2_type = ARG_ANYTHING, 6353 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6354 .arg4_type = ARG_CONST_SIZE 6355 }; 6356 6357 static void bpf_update_srh_state(struct sk_buff *skb) 6358 { 6359 struct seg6_bpf_srh_state *srh_state = 6360 this_cpu_ptr(&seg6_bpf_srh_states); 6361 int srhoff = 0; 6362 6363 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) { 6364 srh_state->srh = NULL; 6365 } else { 6366 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6367 srh_state->hdrlen = srh_state->srh->hdrlen << 3; 6368 srh_state->valid = true; 6369 } 6370 } 6371 6372 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb, 6373 u32, action, void *, param, u32, param_len) 6374 { 6375 struct seg6_bpf_srh_state *srh_state = 6376 this_cpu_ptr(&seg6_bpf_srh_states); 6377 int hdroff = 0; 6378 int err; 6379 6380 switch (action) { 6381 case SEG6_LOCAL_ACTION_END_X: 6382 if (!seg6_bpf_has_valid_srh(skb)) 6383 return -EBADMSG; 6384 if (param_len != sizeof(struct in6_addr)) 6385 return -EINVAL; 6386 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0); 6387 case SEG6_LOCAL_ACTION_END_T: 6388 if (!seg6_bpf_has_valid_srh(skb)) 6389 return -EBADMSG; 6390 if (param_len != sizeof(int)) 6391 return -EINVAL; 6392 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 6393 case SEG6_LOCAL_ACTION_END_DT6: 6394 if (!seg6_bpf_has_valid_srh(skb)) 6395 return -EBADMSG; 6396 if (param_len != sizeof(int)) 6397 return -EINVAL; 6398 6399 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0) 6400 return -EBADMSG; 6401 if (!pskb_pull(skb, hdroff)) 6402 return -EBADMSG; 6403 6404 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff); 6405 skb_reset_network_header(skb); 6406 skb_reset_transport_header(skb); 6407 skb->encapsulation = 0; 6408 6409 bpf_compute_data_pointers(skb); 6410 bpf_update_srh_state(skb); 6411 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 6412 case SEG6_LOCAL_ACTION_END_B6: 6413 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 6414 return -EBADMSG; 6415 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE, 6416 param, param_len); 6417 if (!err) 6418 bpf_update_srh_state(skb); 6419 6420 return err; 6421 case SEG6_LOCAL_ACTION_END_B6_ENCAP: 6422 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 6423 return -EBADMSG; 6424 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6, 6425 param, param_len); 6426 if (!err) 6427 bpf_update_srh_state(skb); 6428 6429 return err; 6430 default: 6431 return -EINVAL; 6432 } 6433 } 6434 6435 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = { 6436 .func = bpf_lwt_seg6_action, 6437 .gpl_only = false, 6438 .ret_type = RET_INTEGER, 6439 .arg1_type = ARG_PTR_TO_CTX, 6440 .arg2_type = ARG_ANYTHING, 6441 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6442 .arg4_type = ARG_CONST_SIZE 6443 }; 6444 6445 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset, 6446 s32, len) 6447 { 6448 struct seg6_bpf_srh_state *srh_state = 6449 this_cpu_ptr(&seg6_bpf_srh_states); 6450 struct ipv6_sr_hdr *srh = srh_state->srh; 6451 void *srh_end, *srh_tlvs, *ptr; 6452 struct ipv6hdr *hdr; 6453 int srhoff = 0; 6454 int ret; 6455 6456 if (unlikely(srh == NULL)) 6457 return -EINVAL; 6458 6459 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) + 6460 ((srh->first_segment + 1) << 4)); 6461 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) + 6462 srh_state->hdrlen); 6463 ptr = skb->data + offset; 6464 6465 if (unlikely(ptr < srh_tlvs || ptr > srh_end)) 6466 return -EFAULT; 6467 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end)) 6468 return -EFAULT; 6469 6470 if (len > 0) { 6471 ret = skb_cow_head(skb, len); 6472 if (unlikely(ret < 0)) 6473 return ret; 6474 6475 ret = bpf_skb_net_hdr_push(skb, offset, len); 6476 } else { 6477 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len); 6478 } 6479 6480 bpf_compute_data_pointers(skb); 6481 if (unlikely(ret < 0)) 6482 return ret; 6483 6484 hdr = (struct ipv6hdr *)skb->data; 6485 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 6486 6487 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 6488 return -EINVAL; 6489 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6490 srh_state->hdrlen += len; 6491 srh_state->valid = false; 6492 return 0; 6493 } 6494 6495 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = { 6496 .func = bpf_lwt_seg6_adjust_srh, 6497 .gpl_only = false, 6498 .ret_type = RET_INTEGER, 6499 .arg1_type = ARG_PTR_TO_CTX, 6500 .arg2_type = ARG_ANYTHING, 6501 .arg3_type = ARG_ANYTHING, 6502 }; 6503 #endif /* CONFIG_IPV6_SEG6_BPF */ 6504 6505 #ifdef CONFIG_INET 6506 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple, 6507 int dif, int sdif, u8 family, u8 proto) 6508 { 6509 struct inet_hashinfo *hinfo = net->ipv4.tcp_death_row.hashinfo; 6510 bool refcounted = false; 6511 struct sock *sk = NULL; 6512 6513 if (family == AF_INET) { 6514 __be32 src4 = tuple->ipv4.saddr; 6515 __be32 dst4 = tuple->ipv4.daddr; 6516 6517 if (proto == IPPROTO_TCP) 6518 sk = __inet_lookup(net, hinfo, NULL, 0, 6519 src4, tuple->ipv4.sport, 6520 dst4, tuple->ipv4.dport, 6521 dif, sdif, &refcounted); 6522 else 6523 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport, 6524 dst4, tuple->ipv4.dport, 6525 dif, sdif, net->ipv4.udp_table, NULL); 6526 #if IS_ENABLED(CONFIG_IPV6) 6527 } else { 6528 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr; 6529 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr; 6530 6531 if (proto == IPPROTO_TCP) 6532 sk = __inet6_lookup(net, hinfo, NULL, 0, 6533 src6, tuple->ipv6.sport, 6534 dst6, ntohs(tuple->ipv6.dport), 6535 dif, sdif, &refcounted); 6536 else if (likely(ipv6_bpf_stub)) 6537 sk = ipv6_bpf_stub->udp6_lib_lookup(net, 6538 src6, tuple->ipv6.sport, 6539 dst6, tuple->ipv6.dport, 6540 dif, sdif, 6541 net->ipv4.udp_table, NULL); 6542 #endif 6543 } 6544 6545 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) { 6546 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6547 sk = NULL; 6548 } 6549 return sk; 6550 } 6551 6552 /* bpf_skc_lookup performs the core lookup for different types of sockets, 6553 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE. 6554 */ 6555 static struct sock * 6556 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6557 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 6558 u64 flags) 6559 { 6560 struct sock *sk = NULL; 6561 struct net *net; 6562 u8 family; 6563 int sdif; 6564 6565 if (len == sizeof(tuple->ipv4)) 6566 family = AF_INET; 6567 else if (len == sizeof(tuple->ipv6)) 6568 family = AF_INET6; 6569 else 6570 return NULL; 6571 6572 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX))) 6573 goto out; 6574 6575 if (family == AF_INET) 6576 sdif = inet_sdif(skb); 6577 else 6578 sdif = inet6_sdif(skb); 6579 6580 if ((s32)netns_id < 0) { 6581 net = caller_net; 6582 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 6583 } else { 6584 net = get_net_ns_by_id(caller_net, netns_id); 6585 if (unlikely(!net)) 6586 goto out; 6587 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 6588 put_net(net); 6589 } 6590 6591 out: 6592 return sk; 6593 } 6594 6595 static struct sock * 6596 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6597 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 6598 u64 flags) 6599 { 6600 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net, 6601 ifindex, proto, netns_id, flags); 6602 6603 if (sk) { 6604 struct sock *sk2 = sk_to_full_sk(sk); 6605 6606 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk 6607 * sock refcnt is decremented to prevent a request_sock leak. 6608 */ 6609 if (!sk_fullsock(sk2)) 6610 sk2 = NULL; 6611 if (sk2 != sk) { 6612 sock_gen_put(sk); 6613 /* Ensure there is no need to bump sk2 refcnt */ 6614 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) { 6615 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6616 return NULL; 6617 } 6618 sk = sk2; 6619 } 6620 } 6621 6622 return sk; 6623 } 6624 6625 static struct sock * 6626 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6627 u8 proto, u64 netns_id, u64 flags) 6628 { 6629 struct net *caller_net; 6630 int ifindex; 6631 6632 if (skb->dev) { 6633 caller_net = dev_net(skb->dev); 6634 ifindex = skb->dev->ifindex; 6635 } else { 6636 caller_net = sock_net(skb->sk); 6637 ifindex = 0; 6638 } 6639 6640 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto, 6641 netns_id, flags); 6642 } 6643 6644 static struct sock * 6645 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6646 u8 proto, u64 netns_id, u64 flags) 6647 { 6648 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id, 6649 flags); 6650 6651 if (sk) { 6652 struct sock *sk2 = sk_to_full_sk(sk); 6653 6654 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk 6655 * sock refcnt is decremented to prevent a request_sock leak. 6656 */ 6657 if (!sk_fullsock(sk2)) 6658 sk2 = NULL; 6659 if (sk2 != sk) { 6660 sock_gen_put(sk); 6661 /* Ensure there is no need to bump sk2 refcnt */ 6662 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) { 6663 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6664 return NULL; 6665 } 6666 sk = sk2; 6667 } 6668 } 6669 6670 return sk; 6671 } 6672 6673 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb, 6674 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6675 { 6676 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP, 6677 netns_id, flags); 6678 } 6679 6680 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = { 6681 .func = bpf_skc_lookup_tcp, 6682 .gpl_only = false, 6683 .pkt_access = true, 6684 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6685 .arg1_type = ARG_PTR_TO_CTX, 6686 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6687 .arg3_type = ARG_CONST_SIZE, 6688 .arg4_type = ARG_ANYTHING, 6689 .arg5_type = ARG_ANYTHING, 6690 }; 6691 6692 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb, 6693 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6694 { 6695 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, 6696 netns_id, flags); 6697 } 6698 6699 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = { 6700 .func = bpf_sk_lookup_tcp, 6701 .gpl_only = false, 6702 .pkt_access = true, 6703 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6704 .arg1_type = ARG_PTR_TO_CTX, 6705 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6706 .arg3_type = ARG_CONST_SIZE, 6707 .arg4_type = ARG_ANYTHING, 6708 .arg5_type = ARG_ANYTHING, 6709 }; 6710 6711 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb, 6712 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6713 { 6714 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, 6715 netns_id, flags); 6716 } 6717 6718 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = { 6719 .func = bpf_sk_lookup_udp, 6720 .gpl_only = false, 6721 .pkt_access = true, 6722 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6723 .arg1_type = ARG_PTR_TO_CTX, 6724 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6725 .arg3_type = ARG_CONST_SIZE, 6726 .arg4_type = ARG_ANYTHING, 6727 .arg5_type = ARG_ANYTHING, 6728 }; 6729 6730 BPF_CALL_1(bpf_sk_release, struct sock *, sk) 6731 { 6732 if (sk && sk_is_refcounted(sk)) 6733 sock_gen_put(sk); 6734 return 0; 6735 } 6736 6737 static const struct bpf_func_proto bpf_sk_release_proto = { 6738 .func = bpf_sk_release, 6739 .gpl_only = false, 6740 .ret_type = RET_INTEGER, 6741 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE, 6742 }; 6743 6744 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx, 6745 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6746 { 6747 struct net *caller_net = dev_net(ctx->rxq->dev); 6748 int ifindex = ctx->rxq->dev->ifindex; 6749 6750 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6751 ifindex, IPPROTO_UDP, netns_id, 6752 flags); 6753 } 6754 6755 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = { 6756 .func = bpf_xdp_sk_lookup_udp, 6757 .gpl_only = false, 6758 .pkt_access = true, 6759 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6760 .arg1_type = ARG_PTR_TO_CTX, 6761 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6762 .arg3_type = ARG_CONST_SIZE, 6763 .arg4_type = ARG_ANYTHING, 6764 .arg5_type = ARG_ANYTHING, 6765 }; 6766 6767 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx, 6768 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6769 { 6770 struct net *caller_net = dev_net(ctx->rxq->dev); 6771 int ifindex = ctx->rxq->dev->ifindex; 6772 6773 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net, 6774 ifindex, IPPROTO_TCP, netns_id, 6775 flags); 6776 } 6777 6778 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = { 6779 .func = bpf_xdp_skc_lookup_tcp, 6780 .gpl_only = false, 6781 .pkt_access = true, 6782 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6783 .arg1_type = ARG_PTR_TO_CTX, 6784 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6785 .arg3_type = ARG_CONST_SIZE, 6786 .arg4_type = ARG_ANYTHING, 6787 .arg5_type = ARG_ANYTHING, 6788 }; 6789 6790 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx, 6791 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6792 { 6793 struct net *caller_net = dev_net(ctx->rxq->dev); 6794 int ifindex = ctx->rxq->dev->ifindex; 6795 6796 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6797 ifindex, IPPROTO_TCP, netns_id, 6798 flags); 6799 } 6800 6801 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = { 6802 .func = bpf_xdp_sk_lookup_tcp, 6803 .gpl_only = false, 6804 .pkt_access = true, 6805 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6806 .arg1_type = ARG_PTR_TO_CTX, 6807 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6808 .arg3_type = ARG_CONST_SIZE, 6809 .arg4_type = ARG_ANYTHING, 6810 .arg5_type = ARG_ANYTHING, 6811 }; 6812 6813 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6814 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6815 { 6816 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, 6817 sock_net(ctx->sk), 0, 6818 IPPROTO_TCP, netns_id, flags); 6819 } 6820 6821 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = { 6822 .func = bpf_sock_addr_skc_lookup_tcp, 6823 .gpl_only = false, 6824 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6825 .arg1_type = ARG_PTR_TO_CTX, 6826 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6827 .arg3_type = ARG_CONST_SIZE, 6828 .arg4_type = ARG_ANYTHING, 6829 .arg5_type = ARG_ANYTHING, 6830 }; 6831 6832 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6833 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6834 { 6835 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6836 sock_net(ctx->sk), 0, IPPROTO_TCP, 6837 netns_id, flags); 6838 } 6839 6840 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = { 6841 .func = bpf_sock_addr_sk_lookup_tcp, 6842 .gpl_only = false, 6843 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6844 .arg1_type = ARG_PTR_TO_CTX, 6845 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6846 .arg3_type = ARG_CONST_SIZE, 6847 .arg4_type = ARG_ANYTHING, 6848 .arg5_type = ARG_ANYTHING, 6849 }; 6850 6851 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx, 6852 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6853 { 6854 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6855 sock_net(ctx->sk), 0, IPPROTO_UDP, 6856 netns_id, flags); 6857 } 6858 6859 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = { 6860 .func = bpf_sock_addr_sk_lookup_udp, 6861 .gpl_only = false, 6862 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6863 .arg1_type = ARG_PTR_TO_CTX, 6864 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6865 .arg3_type = ARG_CONST_SIZE, 6866 .arg4_type = ARG_ANYTHING, 6867 .arg5_type = ARG_ANYTHING, 6868 }; 6869 6870 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6871 struct bpf_insn_access_aux *info) 6872 { 6873 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, 6874 icsk_retransmits)) 6875 return false; 6876 6877 if (off % size != 0) 6878 return false; 6879 6880 switch (off) { 6881 case offsetof(struct bpf_tcp_sock, bytes_received): 6882 case offsetof(struct bpf_tcp_sock, bytes_acked): 6883 return size == sizeof(__u64); 6884 default: 6885 return size == sizeof(__u32); 6886 } 6887 } 6888 6889 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 6890 const struct bpf_insn *si, 6891 struct bpf_insn *insn_buf, 6892 struct bpf_prog *prog, u32 *target_size) 6893 { 6894 struct bpf_insn *insn = insn_buf; 6895 6896 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \ 6897 do { \ 6898 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \ 6899 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6900 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\ 6901 si->dst_reg, si->src_reg, \ 6902 offsetof(struct tcp_sock, FIELD)); \ 6903 } while (0) 6904 6905 #define BPF_INET_SOCK_GET_COMMON(FIELD) \ 6906 do { \ 6907 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \ 6908 FIELD) > \ 6909 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6910 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 6911 struct inet_connection_sock, \ 6912 FIELD), \ 6913 si->dst_reg, si->src_reg, \ 6914 offsetof( \ 6915 struct inet_connection_sock, \ 6916 FIELD)); \ 6917 } while (0) 6918 6919 switch (si->off) { 6920 case offsetof(struct bpf_tcp_sock, rtt_min): 6921 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 6922 sizeof(struct minmax)); 6923 BUILD_BUG_ON(sizeof(struct minmax) < 6924 sizeof(struct minmax_sample)); 6925 6926 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 6927 offsetof(struct tcp_sock, rtt_min) + 6928 offsetof(struct minmax_sample, v)); 6929 break; 6930 case offsetof(struct bpf_tcp_sock, snd_cwnd): 6931 BPF_TCP_SOCK_GET_COMMON(snd_cwnd); 6932 break; 6933 case offsetof(struct bpf_tcp_sock, srtt_us): 6934 BPF_TCP_SOCK_GET_COMMON(srtt_us); 6935 break; 6936 case offsetof(struct bpf_tcp_sock, snd_ssthresh): 6937 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh); 6938 break; 6939 case offsetof(struct bpf_tcp_sock, rcv_nxt): 6940 BPF_TCP_SOCK_GET_COMMON(rcv_nxt); 6941 break; 6942 case offsetof(struct bpf_tcp_sock, snd_nxt): 6943 BPF_TCP_SOCK_GET_COMMON(snd_nxt); 6944 break; 6945 case offsetof(struct bpf_tcp_sock, snd_una): 6946 BPF_TCP_SOCK_GET_COMMON(snd_una); 6947 break; 6948 case offsetof(struct bpf_tcp_sock, mss_cache): 6949 BPF_TCP_SOCK_GET_COMMON(mss_cache); 6950 break; 6951 case offsetof(struct bpf_tcp_sock, ecn_flags): 6952 BPF_TCP_SOCK_GET_COMMON(ecn_flags); 6953 break; 6954 case offsetof(struct bpf_tcp_sock, rate_delivered): 6955 BPF_TCP_SOCK_GET_COMMON(rate_delivered); 6956 break; 6957 case offsetof(struct bpf_tcp_sock, rate_interval_us): 6958 BPF_TCP_SOCK_GET_COMMON(rate_interval_us); 6959 break; 6960 case offsetof(struct bpf_tcp_sock, packets_out): 6961 BPF_TCP_SOCK_GET_COMMON(packets_out); 6962 break; 6963 case offsetof(struct bpf_tcp_sock, retrans_out): 6964 BPF_TCP_SOCK_GET_COMMON(retrans_out); 6965 break; 6966 case offsetof(struct bpf_tcp_sock, total_retrans): 6967 BPF_TCP_SOCK_GET_COMMON(total_retrans); 6968 break; 6969 case offsetof(struct bpf_tcp_sock, segs_in): 6970 BPF_TCP_SOCK_GET_COMMON(segs_in); 6971 break; 6972 case offsetof(struct bpf_tcp_sock, data_segs_in): 6973 BPF_TCP_SOCK_GET_COMMON(data_segs_in); 6974 break; 6975 case offsetof(struct bpf_tcp_sock, segs_out): 6976 BPF_TCP_SOCK_GET_COMMON(segs_out); 6977 break; 6978 case offsetof(struct bpf_tcp_sock, data_segs_out): 6979 BPF_TCP_SOCK_GET_COMMON(data_segs_out); 6980 break; 6981 case offsetof(struct bpf_tcp_sock, lost_out): 6982 BPF_TCP_SOCK_GET_COMMON(lost_out); 6983 break; 6984 case offsetof(struct bpf_tcp_sock, sacked_out): 6985 BPF_TCP_SOCK_GET_COMMON(sacked_out); 6986 break; 6987 case offsetof(struct bpf_tcp_sock, bytes_received): 6988 BPF_TCP_SOCK_GET_COMMON(bytes_received); 6989 break; 6990 case offsetof(struct bpf_tcp_sock, bytes_acked): 6991 BPF_TCP_SOCK_GET_COMMON(bytes_acked); 6992 break; 6993 case offsetof(struct bpf_tcp_sock, dsack_dups): 6994 BPF_TCP_SOCK_GET_COMMON(dsack_dups); 6995 break; 6996 case offsetof(struct bpf_tcp_sock, delivered): 6997 BPF_TCP_SOCK_GET_COMMON(delivered); 6998 break; 6999 case offsetof(struct bpf_tcp_sock, delivered_ce): 7000 BPF_TCP_SOCK_GET_COMMON(delivered_ce); 7001 break; 7002 case offsetof(struct bpf_tcp_sock, icsk_retransmits): 7003 BPF_INET_SOCK_GET_COMMON(icsk_retransmits); 7004 break; 7005 } 7006 7007 return insn - insn_buf; 7008 } 7009 7010 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk) 7011 { 7012 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 7013 return (unsigned long)sk; 7014 7015 return (unsigned long)NULL; 7016 } 7017 7018 const struct bpf_func_proto bpf_tcp_sock_proto = { 7019 .func = bpf_tcp_sock, 7020 .gpl_only = false, 7021 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL, 7022 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 7023 }; 7024 7025 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk) 7026 { 7027 sk = sk_to_full_sk(sk); 7028 7029 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE)) 7030 return (unsigned long)sk; 7031 7032 return (unsigned long)NULL; 7033 } 7034 7035 static const struct bpf_func_proto bpf_get_listener_sock_proto = { 7036 .func = bpf_get_listener_sock, 7037 .gpl_only = false, 7038 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 7039 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 7040 }; 7041 7042 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb) 7043 { 7044 unsigned int iphdr_len; 7045 7046 switch (skb_protocol(skb, true)) { 7047 case cpu_to_be16(ETH_P_IP): 7048 iphdr_len = sizeof(struct iphdr); 7049 break; 7050 case cpu_to_be16(ETH_P_IPV6): 7051 iphdr_len = sizeof(struct ipv6hdr); 7052 break; 7053 default: 7054 return 0; 7055 } 7056 7057 if (skb_headlen(skb) < iphdr_len) 7058 return 0; 7059 7060 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len)) 7061 return 0; 7062 7063 return INET_ECN_set_ce(skb); 7064 } 7065 7066 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 7067 struct bpf_insn_access_aux *info) 7068 { 7069 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id)) 7070 return false; 7071 7072 if (off % size != 0) 7073 return false; 7074 7075 switch (off) { 7076 default: 7077 return size == sizeof(__u32); 7078 } 7079 } 7080 7081 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 7082 const struct bpf_insn *si, 7083 struct bpf_insn *insn_buf, 7084 struct bpf_prog *prog, u32 *target_size) 7085 { 7086 struct bpf_insn *insn = insn_buf; 7087 7088 #define BPF_XDP_SOCK_GET(FIELD) \ 7089 do { \ 7090 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \ 7091 sizeof_field(struct bpf_xdp_sock, FIELD)); \ 7092 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\ 7093 si->dst_reg, si->src_reg, \ 7094 offsetof(struct xdp_sock, FIELD)); \ 7095 } while (0) 7096 7097 switch (si->off) { 7098 case offsetof(struct bpf_xdp_sock, queue_id): 7099 BPF_XDP_SOCK_GET(queue_id); 7100 break; 7101 } 7102 7103 return insn - insn_buf; 7104 } 7105 7106 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = { 7107 .func = bpf_skb_ecn_set_ce, 7108 .gpl_only = false, 7109 .ret_type = RET_INTEGER, 7110 .arg1_type = ARG_PTR_TO_CTX, 7111 }; 7112 7113 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 7114 struct tcphdr *, th, u32, th_len) 7115 { 7116 #ifdef CONFIG_SYN_COOKIES 7117 u32 cookie; 7118 int ret; 7119 7120 if (unlikely(!sk || th_len < sizeof(*th))) 7121 return -EINVAL; 7122 7123 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */ 7124 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 7125 return -EINVAL; 7126 7127 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies)) 7128 return -EINVAL; 7129 7130 if (!th->ack || th->rst || th->syn) 7131 return -ENOENT; 7132 7133 if (unlikely(iph_len < sizeof(struct iphdr))) 7134 return -EINVAL; 7135 7136 if (tcp_synq_no_recent_overflow(sk)) 7137 return -ENOENT; 7138 7139 cookie = ntohl(th->ack_seq) - 1; 7140 7141 /* Both struct iphdr and struct ipv6hdr have the version field at the 7142 * same offset so we can cast to the shorter header (struct iphdr). 7143 */ 7144 switch (((struct iphdr *)iph)->version) { 7145 case 4: 7146 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk)) 7147 return -EINVAL; 7148 7149 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie); 7150 break; 7151 7152 #if IS_BUILTIN(CONFIG_IPV6) 7153 case 6: 7154 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 7155 return -EINVAL; 7156 7157 if (sk->sk_family != AF_INET6) 7158 return -EINVAL; 7159 7160 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie); 7161 break; 7162 #endif /* CONFIG_IPV6 */ 7163 7164 default: 7165 return -EPROTONOSUPPORT; 7166 } 7167 7168 if (ret > 0) 7169 return 0; 7170 7171 return -ENOENT; 7172 #else 7173 return -ENOTSUPP; 7174 #endif 7175 } 7176 7177 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = { 7178 .func = bpf_tcp_check_syncookie, 7179 .gpl_only = true, 7180 .pkt_access = true, 7181 .ret_type = RET_INTEGER, 7182 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7183 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7184 .arg3_type = ARG_CONST_SIZE, 7185 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7186 .arg5_type = ARG_CONST_SIZE, 7187 }; 7188 7189 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 7190 struct tcphdr *, th, u32, th_len) 7191 { 7192 #ifdef CONFIG_SYN_COOKIES 7193 u32 cookie; 7194 u16 mss; 7195 7196 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4)) 7197 return -EINVAL; 7198 7199 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 7200 return -EINVAL; 7201 7202 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies)) 7203 return -ENOENT; 7204 7205 if (!th->syn || th->ack || th->fin || th->rst) 7206 return -EINVAL; 7207 7208 if (unlikely(iph_len < sizeof(struct iphdr))) 7209 return -EINVAL; 7210 7211 /* Both struct iphdr and struct ipv6hdr have the version field at the 7212 * same offset so we can cast to the shorter header (struct iphdr). 7213 */ 7214 switch (((struct iphdr *)iph)->version) { 7215 case 4: 7216 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk)) 7217 return -EINVAL; 7218 7219 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie); 7220 break; 7221 7222 #if IS_BUILTIN(CONFIG_IPV6) 7223 case 6: 7224 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 7225 return -EINVAL; 7226 7227 if (sk->sk_family != AF_INET6) 7228 return -EINVAL; 7229 7230 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie); 7231 break; 7232 #endif /* CONFIG_IPV6 */ 7233 7234 default: 7235 return -EPROTONOSUPPORT; 7236 } 7237 if (mss == 0) 7238 return -ENOENT; 7239 7240 return cookie | ((u64)mss << 32); 7241 #else 7242 return -EOPNOTSUPP; 7243 #endif /* CONFIG_SYN_COOKIES */ 7244 } 7245 7246 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = { 7247 .func = bpf_tcp_gen_syncookie, 7248 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */ 7249 .pkt_access = true, 7250 .ret_type = RET_INTEGER, 7251 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7252 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7253 .arg3_type = ARG_CONST_SIZE, 7254 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7255 .arg5_type = ARG_CONST_SIZE, 7256 }; 7257 7258 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags) 7259 { 7260 if (!sk || flags != 0) 7261 return -EINVAL; 7262 if (!skb_at_tc_ingress(skb)) 7263 return -EOPNOTSUPP; 7264 if (unlikely(dev_net(skb->dev) != sock_net(sk))) 7265 return -ENETUNREACH; 7266 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport)) 7267 return -ESOCKTNOSUPPORT; 7268 if (sk_is_refcounted(sk) && 7269 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) 7270 return -ENOENT; 7271 7272 skb_orphan(skb); 7273 skb->sk = sk; 7274 skb->destructor = sock_pfree; 7275 7276 return 0; 7277 } 7278 7279 static const struct bpf_func_proto bpf_sk_assign_proto = { 7280 .func = bpf_sk_assign, 7281 .gpl_only = false, 7282 .ret_type = RET_INTEGER, 7283 .arg1_type = ARG_PTR_TO_CTX, 7284 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7285 .arg3_type = ARG_ANYTHING, 7286 }; 7287 7288 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend, 7289 u8 search_kind, const u8 *magic, 7290 u8 magic_len, bool *eol) 7291 { 7292 u8 kind, kind_len; 7293 7294 *eol = false; 7295 7296 while (op < opend) { 7297 kind = op[0]; 7298 7299 if (kind == TCPOPT_EOL) { 7300 *eol = true; 7301 return ERR_PTR(-ENOMSG); 7302 } else if (kind == TCPOPT_NOP) { 7303 op++; 7304 continue; 7305 } 7306 7307 if (opend - op < 2 || opend - op < op[1] || op[1] < 2) 7308 /* Something is wrong in the received header. 7309 * Follow the TCP stack's tcp_parse_options() 7310 * and just bail here. 7311 */ 7312 return ERR_PTR(-EFAULT); 7313 7314 kind_len = op[1]; 7315 if (search_kind == kind) { 7316 if (!magic_len) 7317 return op; 7318 7319 if (magic_len > kind_len - 2) 7320 return ERR_PTR(-ENOMSG); 7321 7322 if (!memcmp(&op[2], magic, magic_len)) 7323 return op; 7324 } 7325 7326 op += kind_len; 7327 } 7328 7329 return ERR_PTR(-ENOMSG); 7330 } 7331 7332 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7333 void *, search_res, u32, len, u64, flags) 7334 { 7335 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN; 7336 const u8 *op, *opend, *magic, *search = search_res; 7337 u8 search_kind, search_len, copy_len, magic_len; 7338 int ret; 7339 7340 /* 2 byte is the minimal option len except TCPOPT_NOP and 7341 * TCPOPT_EOL which are useless for the bpf prog to learn 7342 * and this helper disallow loading them also. 7343 */ 7344 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN) 7345 return -EINVAL; 7346 7347 search_kind = search[0]; 7348 search_len = search[1]; 7349 7350 if (search_len > len || search_kind == TCPOPT_NOP || 7351 search_kind == TCPOPT_EOL) 7352 return -EINVAL; 7353 7354 if (search_kind == TCPOPT_EXP || search_kind == 253) { 7355 /* 16 or 32 bit magic. +2 for kind and kind length */ 7356 if (search_len != 4 && search_len != 6) 7357 return -EINVAL; 7358 magic = &search[2]; 7359 magic_len = search_len - 2; 7360 } else { 7361 if (search_len) 7362 return -EINVAL; 7363 magic = NULL; 7364 magic_len = 0; 7365 } 7366 7367 if (load_syn) { 7368 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op); 7369 if (ret < 0) 7370 return ret; 7371 7372 opend = op + ret; 7373 op += sizeof(struct tcphdr); 7374 } else { 7375 if (!bpf_sock->skb || 7376 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB) 7377 /* This bpf_sock->op cannot call this helper */ 7378 return -EPERM; 7379 7380 opend = bpf_sock->skb_data_end; 7381 op = bpf_sock->skb->data + sizeof(struct tcphdr); 7382 } 7383 7384 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len, 7385 &eol); 7386 if (IS_ERR(op)) 7387 return PTR_ERR(op); 7388 7389 copy_len = op[1]; 7390 ret = copy_len; 7391 if (copy_len > len) { 7392 ret = -ENOSPC; 7393 copy_len = len; 7394 } 7395 7396 memcpy(search_res, op, copy_len); 7397 return ret; 7398 } 7399 7400 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = { 7401 .func = bpf_sock_ops_load_hdr_opt, 7402 .gpl_only = false, 7403 .ret_type = RET_INTEGER, 7404 .arg1_type = ARG_PTR_TO_CTX, 7405 .arg2_type = ARG_PTR_TO_MEM, 7406 .arg3_type = ARG_CONST_SIZE, 7407 .arg4_type = ARG_ANYTHING, 7408 }; 7409 7410 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7411 const void *, from, u32, len, u64, flags) 7412 { 7413 u8 new_kind, new_kind_len, magic_len = 0, *opend; 7414 const u8 *op, *new_op, *magic = NULL; 7415 struct sk_buff *skb; 7416 bool eol; 7417 7418 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB) 7419 return -EPERM; 7420 7421 if (len < 2 || flags) 7422 return -EINVAL; 7423 7424 new_op = from; 7425 new_kind = new_op[0]; 7426 new_kind_len = new_op[1]; 7427 7428 if (new_kind_len > len || new_kind == TCPOPT_NOP || 7429 new_kind == TCPOPT_EOL) 7430 return -EINVAL; 7431 7432 if (new_kind_len > bpf_sock->remaining_opt_len) 7433 return -ENOSPC; 7434 7435 /* 253 is another experimental kind */ 7436 if (new_kind == TCPOPT_EXP || new_kind == 253) { 7437 if (new_kind_len < 4) 7438 return -EINVAL; 7439 /* Match for the 2 byte magic also. 7440 * RFC 6994: the magic could be 2 or 4 bytes. 7441 * Hence, matching by 2 byte only is on the 7442 * conservative side but it is the right 7443 * thing to do for the 'search-for-duplication' 7444 * purpose. 7445 */ 7446 magic = &new_op[2]; 7447 magic_len = 2; 7448 } 7449 7450 /* Check for duplication */ 7451 skb = bpf_sock->skb; 7452 op = skb->data + sizeof(struct tcphdr); 7453 opend = bpf_sock->skb_data_end; 7454 7455 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len, 7456 &eol); 7457 if (!IS_ERR(op)) 7458 return -EEXIST; 7459 7460 if (PTR_ERR(op) != -ENOMSG) 7461 return PTR_ERR(op); 7462 7463 if (eol) 7464 /* The option has been ended. Treat it as no more 7465 * header option can be written. 7466 */ 7467 return -ENOSPC; 7468 7469 /* No duplication found. Store the header option. */ 7470 memcpy(opend, from, new_kind_len); 7471 7472 bpf_sock->remaining_opt_len -= new_kind_len; 7473 bpf_sock->skb_data_end += new_kind_len; 7474 7475 return 0; 7476 } 7477 7478 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = { 7479 .func = bpf_sock_ops_store_hdr_opt, 7480 .gpl_only = false, 7481 .ret_type = RET_INTEGER, 7482 .arg1_type = ARG_PTR_TO_CTX, 7483 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7484 .arg3_type = ARG_CONST_SIZE, 7485 .arg4_type = ARG_ANYTHING, 7486 }; 7487 7488 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7489 u32, len, u64, flags) 7490 { 7491 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB) 7492 return -EPERM; 7493 7494 if (flags || len < 2) 7495 return -EINVAL; 7496 7497 if (len > bpf_sock->remaining_opt_len) 7498 return -ENOSPC; 7499 7500 bpf_sock->remaining_opt_len -= len; 7501 7502 return 0; 7503 } 7504 7505 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = { 7506 .func = bpf_sock_ops_reserve_hdr_opt, 7507 .gpl_only = false, 7508 .ret_type = RET_INTEGER, 7509 .arg1_type = ARG_PTR_TO_CTX, 7510 .arg2_type = ARG_ANYTHING, 7511 .arg3_type = ARG_ANYTHING, 7512 }; 7513 7514 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, 7515 u64, tstamp, u32, tstamp_type) 7516 { 7517 /* skb_clear_delivery_time() is done for inet protocol */ 7518 if (skb->protocol != htons(ETH_P_IP) && 7519 skb->protocol != htons(ETH_P_IPV6)) 7520 return -EOPNOTSUPP; 7521 7522 switch (tstamp_type) { 7523 case BPF_SKB_TSTAMP_DELIVERY_MONO: 7524 if (!tstamp) 7525 return -EINVAL; 7526 skb->tstamp = tstamp; 7527 skb->mono_delivery_time = 1; 7528 break; 7529 case BPF_SKB_TSTAMP_UNSPEC: 7530 if (tstamp) 7531 return -EINVAL; 7532 skb->tstamp = 0; 7533 skb->mono_delivery_time = 0; 7534 break; 7535 default: 7536 return -EINVAL; 7537 } 7538 7539 return 0; 7540 } 7541 7542 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = { 7543 .func = bpf_skb_set_tstamp, 7544 .gpl_only = false, 7545 .ret_type = RET_INTEGER, 7546 .arg1_type = ARG_PTR_TO_CTX, 7547 .arg2_type = ARG_ANYTHING, 7548 .arg3_type = ARG_ANYTHING, 7549 }; 7550 7551 #ifdef CONFIG_SYN_COOKIES 7552 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph, 7553 struct tcphdr *, th, u32, th_len) 7554 { 7555 u32 cookie; 7556 u16 mss; 7557 7558 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4)) 7559 return -EINVAL; 7560 7561 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT; 7562 cookie = __cookie_v4_init_sequence(iph, th, &mss); 7563 7564 return cookie | ((u64)mss << 32); 7565 } 7566 7567 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = { 7568 .func = bpf_tcp_raw_gen_syncookie_ipv4, 7569 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */ 7570 .pkt_access = true, 7571 .ret_type = RET_INTEGER, 7572 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7573 .arg1_size = sizeof(struct iphdr), 7574 .arg2_type = ARG_PTR_TO_MEM, 7575 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 7576 }; 7577 7578 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph, 7579 struct tcphdr *, th, u32, th_len) 7580 { 7581 #if IS_BUILTIN(CONFIG_IPV6) 7582 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - 7583 sizeof(struct ipv6hdr); 7584 u32 cookie; 7585 u16 mss; 7586 7587 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4)) 7588 return -EINVAL; 7589 7590 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp; 7591 cookie = __cookie_v6_init_sequence(iph, th, &mss); 7592 7593 return cookie | ((u64)mss << 32); 7594 #else 7595 return -EPROTONOSUPPORT; 7596 #endif 7597 } 7598 7599 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = { 7600 .func = bpf_tcp_raw_gen_syncookie_ipv6, 7601 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */ 7602 .pkt_access = true, 7603 .ret_type = RET_INTEGER, 7604 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7605 .arg1_size = sizeof(struct ipv6hdr), 7606 .arg2_type = ARG_PTR_TO_MEM, 7607 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 7608 }; 7609 7610 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph, 7611 struct tcphdr *, th) 7612 { 7613 u32 cookie = ntohl(th->ack_seq) - 1; 7614 7615 if (__cookie_v4_check(iph, th, cookie) > 0) 7616 return 0; 7617 7618 return -EACCES; 7619 } 7620 7621 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = { 7622 .func = bpf_tcp_raw_check_syncookie_ipv4, 7623 .gpl_only = true, /* __cookie_v4_check is GPL */ 7624 .pkt_access = true, 7625 .ret_type = RET_INTEGER, 7626 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7627 .arg1_size = sizeof(struct iphdr), 7628 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7629 .arg2_size = sizeof(struct tcphdr), 7630 }; 7631 7632 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph, 7633 struct tcphdr *, th) 7634 { 7635 #if IS_BUILTIN(CONFIG_IPV6) 7636 u32 cookie = ntohl(th->ack_seq) - 1; 7637 7638 if (__cookie_v6_check(iph, th, cookie) > 0) 7639 return 0; 7640 7641 return -EACCES; 7642 #else 7643 return -EPROTONOSUPPORT; 7644 #endif 7645 } 7646 7647 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = { 7648 .func = bpf_tcp_raw_check_syncookie_ipv6, 7649 .gpl_only = true, /* __cookie_v6_check is GPL */ 7650 .pkt_access = true, 7651 .ret_type = RET_INTEGER, 7652 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7653 .arg1_size = sizeof(struct ipv6hdr), 7654 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7655 .arg2_size = sizeof(struct tcphdr), 7656 }; 7657 #endif /* CONFIG_SYN_COOKIES */ 7658 7659 #endif /* CONFIG_INET */ 7660 7661 bool bpf_helper_changes_pkt_data(void *func) 7662 { 7663 if (func == bpf_skb_vlan_push || 7664 func == bpf_skb_vlan_pop || 7665 func == bpf_skb_store_bytes || 7666 func == bpf_skb_change_proto || 7667 func == bpf_skb_change_head || 7668 func == sk_skb_change_head || 7669 func == bpf_skb_change_tail || 7670 func == sk_skb_change_tail || 7671 func == bpf_skb_adjust_room || 7672 func == sk_skb_adjust_room || 7673 func == bpf_skb_pull_data || 7674 func == sk_skb_pull_data || 7675 func == bpf_clone_redirect || 7676 func == bpf_l3_csum_replace || 7677 func == bpf_l4_csum_replace || 7678 func == bpf_xdp_adjust_head || 7679 func == bpf_xdp_adjust_meta || 7680 func == bpf_msg_pull_data || 7681 func == bpf_msg_push_data || 7682 func == bpf_msg_pop_data || 7683 func == bpf_xdp_adjust_tail || 7684 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 7685 func == bpf_lwt_seg6_store_bytes || 7686 func == bpf_lwt_seg6_adjust_srh || 7687 func == bpf_lwt_seg6_action || 7688 #endif 7689 #ifdef CONFIG_INET 7690 func == bpf_sock_ops_store_hdr_opt || 7691 #endif 7692 func == bpf_lwt_in_push_encap || 7693 func == bpf_lwt_xmit_push_encap) 7694 return true; 7695 7696 return false; 7697 } 7698 7699 const struct bpf_func_proto bpf_event_output_data_proto __weak; 7700 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak; 7701 7702 static const struct bpf_func_proto * 7703 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7704 { 7705 const struct bpf_func_proto *func_proto; 7706 7707 func_proto = cgroup_common_func_proto(func_id, prog); 7708 if (func_proto) 7709 return func_proto; 7710 7711 func_proto = cgroup_current_func_proto(func_id, prog); 7712 if (func_proto) 7713 return func_proto; 7714 7715 switch (func_id) { 7716 case BPF_FUNC_get_socket_cookie: 7717 return &bpf_get_socket_cookie_sock_proto; 7718 case BPF_FUNC_get_netns_cookie: 7719 return &bpf_get_netns_cookie_sock_proto; 7720 case BPF_FUNC_perf_event_output: 7721 return &bpf_event_output_data_proto; 7722 case BPF_FUNC_sk_storage_get: 7723 return &bpf_sk_storage_get_cg_sock_proto; 7724 case BPF_FUNC_ktime_get_coarse_ns: 7725 return &bpf_ktime_get_coarse_ns_proto; 7726 default: 7727 return bpf_base_func_proto(func_id); 7728 } 7729 } 7730 7731 static const struct bpf_func_proto * 7732 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7733 { 7734 const struct bpf_func_proto *func_proto; 7735 7736 func_proto = cgroup_common_func_proto(func_id, prog); 7737 if (func_proto) 7738 return func_proto; 7739 7740 func_proto = cgroup_current_func_proto(func_id, prog); 7741 if (func_proto) 7742 return func_proto; 7743 7744 switch (func_id) { 7745 case BPF_FUNC_bind: 7746 switch (prog->expected_attach_type) { 7747 case BPF_CGROUP_INET4_CONNECT: 7748 case BPF_CGROUP_INET6_CONNECT: 7749 return &bpf_bind_proto; 7750 default: 7751 return NULL; 7752 } 7753 case BPF_FUNC_get_socket_cookie: 7754 return &bpf_get_socket_cookie_sock_addr_proto; 7755 case BPF_FUNC_get_netns_cookie: 7756 return &bpf_get_netns_cookie_sock_addr_proto; 7757 case BPF_FUNC_perf_event_output: 7758 return &bpf_event_output_data_proto; 7759 #ifdef CONFIG_INET 7760 case BPF_FUNC_sk_lookup_tcp: 7761 return &bpf_sock_addr_sk_lookup_tcp_proto; 7762 case BPF_FUNC_sk_lookup_udp: 7763 return &bpf_sock_addr_sk_lookup_udp_proto; 7764 case BPF_FUNC_sk_release: 7765 return &bpf_sk_release_proto; 7766 case BPF_FUNC_skc_lookup_tcp: 7767 return &bpf_sock_addr_skc_lookup_tcp_proto; 7768 #endif /* CONFIG_INET */ 7769 case BPF_FUNC_sk_storage_get: 7770 return &bpf_sk_storage_get_proto; 7771 case BPF_FUNC_sk_storage_delete: 7772 return &bpf_sk_storage_delete_proto; 7773 case BPF_FUNC_setsockopt: 7774 switch (prog->expected_attach_type) { 7775 case BPF_CGROUP_INET4_BIND: 7776 case BPF_CGROUP_INET6_BIND: 7777 case BPF_CGROUP_INET4_CONNECT: 7778 case BPF_CGROUP_INET6_CONNECT: 7779 case BPF_CGROUP_UDP4_RECVMSG: 7780 case BPF_CGROUP_UDP6_RECVMSG: 7781 case BPF_CGROUP_UDP4_SENDMSG: 7782 case BPF_CGROUP_UDP6_SENDMSG: 7783 case BPF_CGROUP_INET4_GETPEERNAME: 7784 case BPF_CGROUP_INET6_GETPEERNAME: 7785 case BPF_CGROUP_INET4_GETSOCKNAME: 7786 case BPF_CGROUP_INET6_GETSOCKNAME: 7787 return &bpf_sock_addr_setsockopt_proto; 7788 default: 7789 return NULL; 7790 } 7791 case BPF_FUNC_getsockopt: 7792 switch (prog->expected_attach_type) { 7793 case BPF_CGROUP_INET4_BIND: 7794 case BPF_CGROUP_INET6_BIND: 7795 case BPF_CGROUP_INET4_CONNECT: 7796 case BPF_CGROUP_INET6_CONNECT: 7797 case BPF_CGROUP_UDP4_RECVMSG: 7798 case BPF_CGROUP_UDP6_RECVMSG: 7799 case BPF_CGROUP_UDP4_SENDMSG: 7800 case BPF_CGROUP_UDP6_SENDMSG: 7801 case BPF_CGROUP_INET4_GETPEERNAME: 7802 case BPF_CGROUP_INET6_GETPEERNAME: 7803 case BPF_CGROUP_INET4_GETSOCKNAME: 7804 case BPF_CGROUP_INET6_GETSOCKNAME: 7805 return &bpf_sock_addr_getsockopt_proto; 7806 default: 7807 return NULL; 7808 } 7809 default: 7810 return bpf_sk_base_func_proto(func_id); 7811 } 7812 } 7813 7814 static const struct bpf_func_proto * 7815 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7816 { 7817 switch (func_id) { 7818 case BPF_FUNC_skb_load_bytes: 7819 return &bpf_skb_load_bytes_proto; 7820 case BPF_FUNC_skb_load_bytes_relative: 7821 return &bpf_skb_load_bytes_relative_proto; 7822 case BPF_FUNC_get_socket_cookie: 7823 return &bpf_get_socket_cookie_proto; 7824 case BPF_FUNC_get_socket_uid: 7825 return &bpf_get_socket_uid_proto; 7826 case BPF_FUNC_perf_event_output: 7827 return &bpf_skb_event_output_proto; 7828 default: 7829 return bpf_sk_base_func_proto(func_id); 7830 } 7831 } 7832 7833 const struct bpf_func_proto bpf_sk_storage_get_proto __weak; 7834 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak; 7835 7836 static const struct bpf_func_proto * 7837 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7838 { 7839 const struct bpf_func_proto *func_proto; 7840 7841 func_proto = cgroup_common_func_proto(func_id, prog); 7842 if (func_proto) 7843 return func_proto; 7844 7845 switch (func_id) { 7846 case BPF_FUNC_sk_fullsock: 7847 return &bpf_sk_fullsock_proto; 7848 case BPF_FUNC_sk_storage_get: 7849 return &bpf_sk_storage_get_proto; 7850 case BPF_FUNC_sk_storage_delete: 7851 return &bpf_sk_storage_delete_proto; 7852 case BPF_FUNC_perf_event_output: 7853 return &bpf_skb_event_output_proto; 7854 #ifdef CONFIG_SOCK_CGROUP_DATA 7855 case BPF_FUNC_skb_cgroup_id: 7856 return &bpf_skb_cgroup_id_proto; 7857 case BPF_FUNC_skb_ancestor_cgroup_id: 7858 return &bpf_skb_ancestor_cgroup_id_proto; 7859 case BPF_FUNC_sk_cgroup_id: 7860 return &bpf_sk_cgroup_id_proto; 7861 case BPF_FUNC_sk_ancestor_cgroup_id: 7862 return &bpf_sk_ancestor_cgroup_id_proto; 7863 #endif 7864 #ifdef CONFIG_INET 7865 case BPF_FUNC_sk_lookup_tcp: 7866 return &bpf_sk_lookup_tcp_proto; 7867 case BPF_FUNC_sk_lookup_udp: 7868 return &bpf_sk_lookup_udp_proto; 7869 case BPF_FUNC_sk_release: 7870 return &bpf_sk_release_proto; 7871 case BPF_FUNC_skc_lookup_tcp: 7872 return &bpf_skc_lookup_tcp_proto; 7873 case BPF_FUNC_tcp_sock: 7874 return &bpf_tcp_sock_proto; 7875 case BPF_FUNC_get_listener_sock: 7876 return &bpf_get_listener_sock_proto; 7877 case BPF_FUNC_skb_ecn_set_ce: 7878 return &bpf_skb_ecn_set_ce_proto; 7879 #endif 7880 default: 7881 return sk_filter_func_proto(func_id, prog); 7882 } 7883 } 7884 7885 static const struct bpf_func_proto * 7886 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7887 { 7888 switch (func_id) { 7889 case BPF_FUNC_skb_store_bytes: 7890 return &bpf_skb_store_bytes_proto; 7891 case BPF_FUNC_skb_load_bytes: 7892 return &bpf_skb_load_bytes_proto; 7893 case BPF_FUNC_skb_load_bytes_relative: 7894 return &bpf_skb_load_bytes_relative_proto; 7895 case BPF_FUNC_skb_pull_data: 7896 return &bpf_skb_pull_data_proto; 7897 case BPF_FUNC_csum_diff: 7898 return &bpf_csum_diff_proto; 7899 case BPF_FUNC_csum_update: 7900 return &bpf_csum_update_proto; 7901 case BPF_FUNC_csum_level: 7902 return &bpf_csum_level_proto; 7903 case BPF_FUNC_l3_csum_replace: 7904 return &bpf_l3_csum_replace_proto; 7905 case BPF_FUNC_l4_csum_replace: 7906 return &bpf_l4_csum_replace_proto; 7907 case BPF_FUNC_clone_redirect: 7908 return &bpf_clone_redirect_proto; 7909 case BPF_FUNC_get_cgroup_classid: 7910 return &bpf_get_cgroup_classid_proto; 7911 case BPF_FUNC_skb_vlan_push: 7912 return &bpf_skb_vlan_push_proto; 7913 case BPF_FUNC_skb_vlan_pop: 7914 return &bpf_skb_vlan_pop_proto; 7915 case BPF_FUNC_skb_change_proto: 7916 return &bpf_skb_change_proto_proto; 7917 case BPF_FUNC_skb_change_type: 7918 return &bpf_skb_change_type_proto; 7919 case BPF_FUNC_skb_adjust_room: 7920 return &bpf_skb_adjust_room_proto; 7921 case BPF_FUNC_skb_change_tail: 7922 return &bpf_skb_change_tail_proto; 7923 case BPF_FUNC_skb_change_head: 7924 return &bpf_skb_change_head_proto; 7925 case BPF_FUNC_skb_get_tunnel_key: 7926 return &bpf_skb_get_tunnel_key_proto; 7927 case BPF_FUNC_skb_set_tunnel_key: 7928 return bpf_get_skb_set_tunnel_proto(func_id); 7929 case BPF_FUNC_skb_get_tunnel_opt: 7930 return &bpf_skb_get_tunnel_opt_proto; 7931 case BPF_FUNC_skb_set_tunnel_opt: 7932 return bpf_get_skb_set_tunnel_proto(func_id); 7933 case BPF_FUNC_redirect: 7934 return &bpf_redirect_proto; 7935 case BPF_FUNC_redirect_neigh: 7936 return &bpf_redirect_neigh_proto; 7937 case BPF_FUNC_redirect_peer: 7938 return &bpf_redirect_peer_proto; 7939 case BPF_FUNC_get_route_realm: 7940 return &bpf_get_route_realm_proto; 7941 case BPF_FUNC_get_hash_recalc: 7942 return &bpf_get_hash_recalc_proto; 7943 case BPF_FUNC_set_hash_invalid: 7944 return &bpf_set_hash_invalid_proto; 7945 case BPF_FUNC_set_hash: 7946 return &bpf_set_hash_proto; 7947 case BPF_FUNC_perf_event_output: 7948 return &bpf_skb_event_output_proto; 7949 case BPF_FUNC_get_smp_processor_id: 7950 return &bpf_get_smp_processor_id_proto; 7951 case BPF_FUNC_skb_under_cgroup: 7952 return &bpf_skb_under_cgroup_proto; 7953 case BPF_FUNC_get_socket_cookie: 7954 return &bpf_get_socket_cookie_proto; 7955 case BPF_FUNC_get_socket_uid: 7956 return &bpf_get_socket_uid_proto; 7957 case BPF_FUNC_fib_lookup: 7958 return &bpf_skb_fib_lookup_proto; 7959 case BPF_FUNC_check_mtu: 7960 return &bpf_skb_check_mtu_proto; 7961 case BPF_FUNC_sk_fullsock: 7962 return &bpf_sk_fullsock_proto; 7963 case BPF_FUNC_sk_storage_get: 7964 return &bpf_sk_storage_get_proto; 7965 case BPF_FUNC_sk_storage_delete: 7966 return &bpf_sk_storage_delete_proto; 7967 #ifdef CONFIG_XFRM 7968 case BPF_FUNC_skb_get_xfrm_state: 7969 return &bpf_skb_get_xfrm_state_proto; 7970 #endif 7971 #ifdef CONFIG_CGROUP_NET_CLASSID 7972 case BPF_FUNC_skb_cgroup_classid: 7973 return &bpf_skb_cgroup_classid_proto; 7974 #endif 7975 #ifdef CONFIG_SOCK_CGROUP_DATA 7976 case BPF_FUNC_skb_cgroup_id: 7977 return &bpf_skb_cgroup_id_proto; 7978 case BPF_FUNC_skb_ancestor_cgroup_id: 7979 return &bpf_skb_ancestor_cgroup_id_proto; 7980 #endif 7981 #ifdef CONFIG_INET 7982 case BPF_FUNC_sk_lookup_tcp: 7983 return &bpf_sk_lookup_tcp_proto; 7984 case BPF_FUNC_sk_lookup_udp: 7985 return &bpf_sk_lookup_udp_proto; 7986 case BPF_FUNC_sk_release: 7987 return &bpf_sk_release_proto; 7988 case BPF_FUNC_tcp_sock: 7989 return &bpf_tcp_sock_proto; 7990 case BPF_FUNC_get_listener_sock: 7991 return &bpf_get_listener_sock_proto; 7992 case BPF_FUNC_skc_lookup_tcp: 7993 return &bpf_skc_lookup_tcp_proto; 7994 case BPF_FUNC_tcp_check_syncookie: 7995 return &bpf_tcp_check_syncookie_proto; 7996 case BPF_FUNC_skb_ecn_set_ce: 7997 return &bpf_skb_ecn_set_ce_proto; 7998 case BPF_FUNC_tcp_gen_syncookie: 7999 return &bpf_tcp_gen_syncookie_proto; 8000 case BPF_FUNC_sk_assign: 8001 return &bpf_sk_assign_proto; 8002 case BPF_FUNC_skb_set_tstamp: 8003 return &bpf_skb_set_tstamp_proto; 8004 #ifdef CONFIG_SYN_COOKIES 8005 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4: 8006 return &bpf_tcp_raw_gen_syncookie_ipv4_proto; 8007 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6: 8008 return &bpf_tcp_raw_gen_syncookie_ipv6_proto; 8009 case BPF_FUNC_tcp_raw_check_syncookie_ipv4: 8010 return &bpf_tcp_raw_check_syncookie_ipv4_proto; 8011 case BPF_FUNC_tcp_raw_check_syncookie_ipv6: 8012 return &bpf_tcp_raw_check_syncookie_ipv6_proto; 8013 #endif 8014 #endif 8015 default: 8016 return bpf_sk_base_func_proto(func_id); 8017 } 8018 } 8019 8020 static const struct bpf_func_proto * 8021 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8022 { 8023 switch (func_id) { 8024 case BPF_FUNC_perf_event_output: 8025 return &bpf_xdp_event_output_proto; 8026 case BPF_FUNC_get_smp_processor_id: 8027 return &bpf_get_smp_processor_id_proto; 8028 case BPF_FUNC_csum_diff: 8029 return &bpf_csum_diff_proto; 8030 case BPF_FUNC_xdp_adjust_head: 8031 return &bpf_xdp_adjust_head_proto; 8032 case BPF_FUNC_xdp_adjust_meta: 8033 return &bpf_xdp_adjust_meta_proto; 8034 case BPF_FUNC_redirect: 8035 return &bpf_xdp_redirect_proto; 8036 case BPF_FUNC_redirect_map: 8037 return &bpf_xdp_redirect_map_proto; 8038 case BPF_FUNC_xdp_adjust_tail: 8039 return &bpf_xdp_adjust_tail_proto; 8040 case BPF_FUNC_xdp_get_buff_len: 8041 return &bpf_xdp_get_buff_len_proto; 8042 case BPF_FUNC_xdp_load_bytes: 8043 return &bpf_xdp_load_bytes_proto; 8044 case BPF_FUNC_xdp_store_bytes: 8045 return &bpf_xdp_store_bytes_proto; 8046 case BPF_FUNC_fib_lookup: 8047 return &bpf_xdp_fib_lookup_proto; 8048 case BPF_FUNC_check_mtu: 8049 return &bpf_xdp_check_mtu_proto; 8050 #ifdef CONFIG_INET 8051 case BPF_FUNC_sk_lookup_udp: 8052 return &bpf_xdp_sk_lookup_udp_proto; 8053 case BPF_FUNC_sk_lookup_tcp: 8054 return &bpf_xdp_sk_lookup_tcp_proto; 8055 case BPF_FUNC_sk_release: 8056 return &bpf_sk_release_proto; 8057 case BPF_FUNC_skc_lookup_tcp: 8058 return &bpf_xdp_skc_lookup_tcp_proto; 8059 case BPF_FUNC_tcp_check_syncookie: 8060 return &bpf_tcp_check_syncookie_proto; 8061 case BPF_FUNC_tcp_gen_syncookie: 8062 return &bpf_tcp_gen_syncookie_proto; 8063 #ifdef CONFIG_SYN_COOKIES 8064 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4: 8065 return &bpf_tcp_raw_gen_syncookie_ipv4_proto; 8066 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6: 8067 return &bpf_tcp_raw_gen_syncookie_ipv6_proto; 8068 case BPF_FUNC_tcp_raw_check_syncookie_ipv4: 8069 return &bpf_tcp_raw_check_syncookie_ipv4_proto; 8070 case BPF_FUNC_tcp_raw_check_syncookie_ipv6: 8071 return &bpf_tcp_raw_check_syncookie_ipv6_proto; 8072 #endif 8073 #endif 8074 default: 8075 return bpf_sk_base_func_proto(func_id); 8076 } 8077 8078 #if IS_MODULE(CONFIG_NF_CONNTRACK) && IS_ENABLED(CONFIG_DEBUG_INFO_BTF_MODULES) 8079 /* The nf_conn___init type is used in the NF_CONNTRACK kfuncs. The 8080 * kfuncs are defined in two different modules, and we want to be able 8081 * to use them interchangably with the same BTF type ID. Because modules 8082 * can't de-duplicate BTF IDs between each other, we need the type to be 8083 * referenced in the vmlinux BTF or the verifier will get confused about 8084 * the different types. So we add this dummy type reference which will 8085 * be included in vmlinux BTF, allowing both modules to refer to the 8086 * same type ID. 8087 */ 8088 BTF_TYPE_EMIT(struct nf_conn___init); 8089 #endif 8090 } 8091 8092 const struct bpf_func_proto bpf_sock_map_update_proto __weak; 8093 const struct bpf_func_proto bpf_sock_hash_update_proto __weak; 8094 8095 static const struct bpf_func_proto * 8096 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8097 { 8098 const struct bpf_func_proto *func_proto; 8099 8100 func_proto = cgroup_common_func_proto(func_id, prog); 8101 if (func_proto) 8102 return func_proto; 8103 8104 switch (func_id) { 8105 case BPF_FUNC_setsockopt: 8106 return &bpf_sock_ops_setsockopt_proto; 8107 case BPF_FUNC_getsockopt: 8108 return &bpf_sock_ops_getsockopt_proto; 8109 case BPF_FUNC_sock_ops_cb_flags_set: 8110 return &bpf_sock_ops_cb_flags_set_proto; 8111 case BPF_FUNC_sock_map_update: 8112 return &bpf_sock_map_update_proto; 8113 case BPF_FUNC_sock_hash_update: 8114 return &bpf_sock_hash_update_proto; 8115 case BPF_FUNC_get_socket_cookie: 8116 return &bpf_get_socket_cookie_sock_ops_proto; 8117 case BPF_FUNC_perf_event_output: 8118 return &bpf_event_output_data_proto; 8119 case BPF_FUNC_sk_storage_get: 8120 return &bpf_sk_storage_get_proto; 8121 case BPF_FUNC_sk_storage_delete: 8122 return &bpf_sk_storage_delete_proto; 8123 case BPF_FUNC_get_netns_cookie: 8124 return &bpf_get_netns_cookie_sock_ops_proto; 8125 #ifdef CONFIG_INET 8126 case BPF_FUNC_load_hdr_opt: 8127 return &bpf_sock_ops_load_hdr_opt_proto; 8128 case BPF_FUNC_store_hdr_opt: 8129 return &bpf_sock_ops_store_hdr_opt_proto; 8130 case BPF_FUNC_reserve_hdr_opt: 8131 return &bpf_sock_ops_reserve_hdr_opt_proto; 8132 case BPF_FUNC_tcp_sock: 8133 return &bpf_tcp_sock_proto; 8134 #endif /* CONFIG_INET */ 8135 default: 8136 return bpf_sk_base_func_proto(func_id); 8137 } 8138 } 8139 8140 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak; 8141 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak; 8142 8143 static const struct bpf_func_proto * 8144 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8145 { 8146 switch (func_id) { 8147 case BPF_FUNC_msg_redirect_map: 8148 return &bpf_msg_redirect_map_proto; 8149 case BPF_FUNC_msg_redirect_hash: 8150 return &bpf_msg_redirect_hash_proto; 8151 case BPF_FUNC_msg_apply_bytes: 8152 return &bpf_msg_apply_bytes_proto; 8153 case BPF_FUNC_msg_cork_bytes: 8154 return &bpf_msg_cork_bytes_proto; 8155 case BPF_FUNC_msg_pull_data: 8156 return &bpf_msg_pull_data_proto; 8157 case BPF_FUNC_msg_push_data: 8158 return &bpf_msg_push_data_proto; 8159 case BPF_FUNC_msg_pop_data: 8160 return &bpf_msg_pop_data_proto; 8161 case BPF_FUNC_perf_event_output: 8162 return &bpf_event_output_data_proto; 8163 case BPF_FUNC_get_current_uid_gid: 8164 return &bpf_get_current_uid_gid_proto; 8165 case BPF_FUNC_get_current_pid_tgid: 8166 return &bpf_get_current_pid_tgid_proto; 8167 case BPF_FUNC_sk_storage_get: 8168 return &bpf_sk_storage_get_proto; 8169 case BPF_FUNC_sk_storage_delete: 8170 return &bpf_sk_storage_delete_proto; 8171 case BPF_FUNC_get_netns_cookie: 8172 return &bpf_get_netns_cookie_sk_msg_proto; 8173 #ifdef CONFIG_CGROUP_NET_CLASSID 8174 case BPF_FUNC_get_cgroup_classid: 8175 return &bpf_get_cgroup_classid_curr_proto; 8176 #endif 8177 default: 8178 return bpf_sk_base_func_proto(func_id); 8179 } 8180 } 8181 8182 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak; 8183 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak; 8184 8185 static const struct bpf_func_proto * 8186 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8187 { 8188 switch (func_id) { 8189 case BPF_FUNC_skb_store_bytes: 8190 return &bpf_skb_store_bytes_proto; 8191 case BPF_FUNC_skb_load_bytes: 8192 return &bpf_skb_load_bytes_proto; 8193 case BPF_FUNC_skb_pull_data: 8194 return &sk_skb_pull_data_proto; 8195 case BPF_FUNC_skb_change_tail: 8196 return &sk_skb_change_tail_proto; 8197 case BPF_FUNC_skb_change_head: 8198 return &sk_skb_change_head_proto; 8199 case BPF_FUNC_skb_adjust_room: 8200 return &sk_skb_adjust_room_proto; 8201 case BPF_FUNC_get_socket_cookie: 8202 return &bpf_get_socket_cookie_proto; 8203 case BPF_FUNC_get_socket_uid: 8204 return &bpf_get_socket_uid_proto; 8205 case BPF_FUNC_sk_redirect_map: 8206 return &bpf_sk_redirect_map_proto; 8207 case BPF_FUNC_sk_redirect_hash: 8208 return &bpf_sk_redirect_hash_proto; 8209 case BPF_FUNC_perf_event_output: 8210 return &bpf_skb_event_output_proto; 8211 #ifdef CONFIG_INET 8212 case BPF_FUNC_sk_lookup_tcp: 8213 return &bpf_sk_lookup_tcp_proto; 8214 case BPF_FUNC_sk_lookup_udp: 8215 return &bpf_sk_lookup_udp_proto; 8216 case BPF_FUNC_sk_release: 8217 return &bpf_sk_release_proto; 8218 case BPF_FUNC_skc_lookup_tcp: 8219 return &bpf_skc_lookup_tcp_proto; 8220 #endif 8221 default: 8222 return bpf_sk_base_func_proto(func_id); 8223 } 8224 } 8225 8226 static const struct bpf_func_proto * 8227 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8228 { 8229 switch (func_id) { 8230 case BPF_FUNC_skb_load_bytes: 8231 return &bpf_flow_dissector_load_bytes_proto; 8232 default: 8233 return bpf_sk_base_func_proto(func_id); 8234 } 8235 } 8236 8237 static const struct bpf_func_proto * 8238 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8239 { 8240 switch (func_id) { 8241 case BPF_FUNC_skb_load_bytes: 8242 return &bpf_skb_load_bytes_proto; 8243 case BPF_FUNC_skb_pull_data: 8244 return &bpf_skb_pull_data_proto; 8245 case BPF_FUNC_csum_diff: 8246 return &bpf_csum_diff_proto; 8247 case BPF_FUNC_get_cgroup_classid: 8248 return &bpf_get_cgroup_classid_proto; 8249 case BPF_FUNC_get_route_realm: 8250 return &bpf_get_route_realm_proto; 8251 case BPF_FUNC_get_hash_recalc: 8252 return &bpf_get_hash_recalc_proto; 8253 case BPF_FUNC_perf_event_output: 8254 return &bpf_skb_event_output_proto; 8255 case BPF_FUNC_get_smp_processor_id: 8256 return &bpf_get_smp_processor_id_proto; 8257 case BPF_FUNC_skb_under_cgroup: 8258 return &bpf_skb_under_cgroup_proto; 8259 default: 8260 return bpf_sk_base_func_proto(func_id); 8261 } 8262 } 8263 8264 static const struct bpf_func_proto * 8265 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8266 { 8267 switch (func_id) { 8268 case BPF_FUNC_lwt_push_encap: 8269 return &bpf_lwt_in_push_encap_proto; 8270 default: 8271 return lwt_out_func_proto(func_id, prog); 8272 } 8273 } 8274 8275 static const struct bpf_func_proto * 8276 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8277 { 8278 switch (func_id) { 8279 case BPF_FUNC_skb_get_tunnel_key: 8280 return &bpf_skb_get_tunnel_key_proto; 8281 case BPF_FUNC_skb_set_tunnel_key: 8282 return bpf_get_skb_set_tunnel_proto(func_id); 8283 case BPF_FUNC_skb_get_tunnel_opt: 8284 return &bpf_skb_get_tunnel_opt_proto; 8285 case BPF_FUNC_skb_set_tunnel_opt: 8286 return bpf_get_skb_set_tunnel_proto(func_id); 8287 case BPF_FUNC_redirect: 8288 return &bpf_redirect_proto; 8289 case BPF_FUNC_clone_redirect: 8290 return &bpf_clone_redirect_proto; 8291 case BPF_FUNC_skb_change_tail: 8292 return &bpf_skb_change_tail_proto; 8293 case BPF_FUNC_skb_change_head: 8294 return &bpf_skb_change_head_proto; 8295 case BPF_FUNC_skb_store_bytes: 8296 return &bpf_skb_store_bytes_proto; 8297 case BPF_FUNC_csum_update: 8298 return &bpf_csum_update_proto; 8299 case BPF_FUNC_csum_level: 8300 return &bpf_csum_level_proto; 8301 case BPF_FUNC_l3_csum_replace: 8302 return &bpf_l3_csum_replace_proto; 8303 case BPF_FUNC_l4_csum_replace: 8304 return &bpf_l4_csum_replace_proto; 8305 case BPF_FUNC_set_hash_invalid: 8306 return &bpf_set_hash_invalid_proto; 8307 case BPF_FUNC_lwt_push_encap: 8308 return &bpf_lwt_xmit_push_encap_proto; 8309 default: 8310 return lwt_out_func_proto(func_id, prog); 8311 } 8312 } 8313 8314 static const struct bpf_func_proto * 8315 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8316 { 8317 switch (func_id) { 8318 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 8319 case BPF_FUNC_lwt_seg6_store_bytes: 8320 return &bpf_lwt_seg6_store_bytes_proto; 8321 case BPF_FUNC_lwt_seg6_action: 8322 return &bpf_lwt_seg6_action_proto; 8323 case BPF_FUNC_lwt_seg6_adjust_srh: 8324 return &bpf_lwt_seg6_adjust_srh_proto; 8325 #endif 8326 default: 8327 return lwt_out_func_proto(func_id, prog); 8328 } 8329 } 8330 8331 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type, 8332 const struct bpf_prog *prog, 8333 struct bpf_insn_access_aux *info) 8334 { 8335 const int size_default = sizeof(__u32); 8336 8337 if (off < 0 || off >= sizeof(struct __sk_buff)) 8338 return false; 8339 8340 /* The verifier guarantees that size > 0. */ 8341 if (off % size != 0) 8342 return false; 8343 8344 switch (off) { 8345 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8346 if (off + size > offsetofend(struct __sk_buff, cb[4])) 8347 return false; 8348 break; 8349 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]): 8350 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]): 8351 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4): 8352 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4): 8353 case bpf_ctx_range(struct __sk_buff, data): 8354 case bpf_ctx_range(struct __sk_buff, data_meta): 8355 case bpf_ctx_range(struct __sk_buff, data_end): 8356 if (size != size_default) 8357 return false; 8358 break; 8359 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 8360 return false; 8361 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8362 if (type == BPF_WRITE || size != sizeof(__u64)) 8363 return false; 8364 break; 8365 case bpf_ctx_range(struct __sk_buff, tstamp): 8366 if (size != sizeof(__u64)) 8367 return false; 8368 break; 8369 case offsetof(struct __sk_buff, sk): 8370 if (type == BPF_WRITE || size != sizeof(__u64)) 8371 return false; 8372 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 8373 break; 8374 case offsetof(struct __sk_buff, tstamp_type): 8375 return false; 8376 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1: 8377 /* Explicitly prohibit access to padding in __sk_buff. */ 8378 return false; 8379 default: 8380 /* Only narrow read access allowed for now. */ 8381 if (type == BPF_WRITE) { 8382 if (size != size_default) 8383 return false; 8384 } else { 8385 bpf_ctx_record_field_size(info, size_default); 8386 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8387 return false; 8388 } 8389 } 8390 8391 return true; 8392 } 8393 8394 static bool sk_filter_is_valid_access(int off, int size, 8395 enum bpf_access_type type, 8396 const struct bpf_prog *prog, 8397 struct bpf_insn_access_aux *info) 8398 { 8399 switch (off) { 8400 case bpf_ctx_range(struct __sk_buff, tc_classid): 8401 case bpf_ctx_range(struct __sk_buff, data): 8402 case bpf_ctx_range(struct __sk_buff, data_meta): 8403 case bpf_ctx_range(struct __sk_buff, data_end): 8404 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8405 case bpf_ctx_range(struct __sk_buff, tstamp): 8406 case bpf_ctx_range(struct __sk_buff, wire_len): 8407 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8408 return false; 8409 } 8410 8411 if (type == BPF_WRITE) { 8412 switch (off) { 8413 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8414 break; 8415 default: 8416 return false; 8417 } 8418 } 8419 8420 return bpf_skb_is_valid_access(off, size, type, prog, info); 8421 } 8422 8423 static bool cg_skb_is_valid_access(int off, int size, 8424 enum bpf_access_type type, 8425 const struct bpf_prog *prog, 8426 struct bpf_insn_access_aux *info) 8427 { 8428 switch (off) { 8429 case bpf_ctx_range(struct __sk_buff, tc_classid): 8430 case bpf_ctx_range(struct __sk_buff, data_meta): 8431 case bpf_ctx_range(struct __sk_buff, wire_len): 8432 return false; 8433 case bpf_ctx_range(struct __sk_buff, data): 8434 case bpf_ctx_range(struct __sk_buff, data_end): 8435 if (!bpf_capable()) 8436 return false; 8437 break; 8438 } 8439 8440 if (type == BPF_WRITE) { 8441 switch (off) { 8442 case bpf_ctx_range(struct __sk_buff, mark): 8443 case bpf_ctx_range(struct __sk_buff, priority): 8444 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8445 break; 8446 case bpf_ctx_range(struct __sk_buff, tstamp): 8447 if (!bpf_capable()) 8448 return false; 8449 break; 8450 default: 8451 return false; 8452 } 8453 } 8454 8455 switch (off) { 8456 case bpf_ctx_range(struct __sk_buff, data): 8457 info->reg_type = PTR_TO_PACKET; 8458 break; 8459 case bpf_ctx_range(struct __sk_buff, data_end): 8460 info->reg_type = PTR_TO_PACKET_END; 8461 break; 8462 } 8463 8464 return bpf_skb_is_valid_access(off, size, type, prog, info); 8465 } 8466 8467 static bool lwt_is_valid_access(int off, int size, 8468 enum bpf_access_type type, 8469 const struct bpf_prog *prog, 8470 struct bpf_insn_access_aux *info) 8471 { 8472 switch (off) { 8473 case bpf_ctx_range(struct __sk_buff, tc_classid): 8474 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8475 case bpf_ctx_range(struct __sk_buff, data_meta): 8476 case bpf_ctx_range(struct __sk_buff, tstamp): 8477 case bpf_ctx_range(struct __sk_buff, wire_len): 8478 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8479 return false; 8480 } 8481 8482 if (type == BPF_WRITE) { 8483 switch (off) { 8484 case bpf_ctx_range(struct __sk_buff, mark): 8485 case bpf_ctx_range(struct __sk_buff, priority): 8486 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8487 break; 8488 default: 8489 return false; 8490 } 8491 } 8492 8493 switch (off) { 8494 case bpf_ctx_range(struct __sk_buff, data): 8495 info->reg_type = PTR_TO_PACKET; 8496 break; 8497 case bpf_ctx_range(struct __sk_buff, data_end): 8498 info->reg_type = PTR_TO_PACKET_END; 8499 break; 8500 } 8501 8502 return bpf_skb_is_valid_access(off, size, type, prog, info); 8503 } 8504 8505 /* Attach type specific accesses */ 8506 static bool __sock_filter_check_attach_type(int off, 8507 enum bpf_access_type access_type, 8508 enum bpf_attach_type attach_type) 8509 { 8510 switch (off) { 8511 case offsetof(struct bpf_sock, bound_dev_if): 8512 case offsetof(struct bpf_sock, mark): 8513 case offsetof(struct bpf_sock, priority): 8514 switch (attach_type) { 8515 case BPF_CGROUP_INET_SOCK_CREATE: 8516 case BPF_CGROUP_INET_SOCK_RELEASE: 8517 goto full_access; 8518 default: 8519 return false; 8520 } 8521 case bpf_ctx_range(struct bpf_sock, src_ip4): 8522 switch (attach_type) { 8523 case BPF_CGROUP_INET4_POST_BIND: 8524 goto read_only; 8525 default: 8526 return false; 8527 } 8528 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 8529 switch (attach_type) { 8530 case BPF_CGROUP_INET6_POST_BIND: 8531 goto read_only; 8532 default: 8533 return false; 8534 } 8535 case bpf_ctx_range(struct bpf_sock, src_port): 8536 switch (attach_type) { 8537 case BPF_CGROUP_INET4_POST_BIND: 8538 case BPF_CGROUP_INET6_POST_BIND: 8539 goto read_only; 8540 default: 8541 return false; 8542 } 8543 } 8544 read_only: 8545 return access_type == BPF_READ; 8546 full_access: 8547 return true; 8548 } 8549 8550 bool bpf_sock_common_is_valid_access(int off, int size, 8551 enum bpf_access_type type, 8552 struct bpf_insn_access_aux *info) 8553 { 8554 switch (off) { 8555 case bpf_ctx_range_till(struct bpf_sock, type, priority): 8556 return false; 8557 default: 8558 return bpf_sock_is_valid_access(off, size, type, info); 8559 } 8560 } 8561 8562 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 8563 struct bpf_insn_access_aux *info) 8564 { 8565 const int size_default = sizeof(__u32); 8566 int field_size; 8567 8568 if (off < 0 || off >= sizeof(struct bpf_sock)) 8569 return false; 8570 if (off % size != 0) 8571 return false; 8572 8573 switch (off) { 8574 case offsetof(struct bpf_sock, state): 8575 case offsetof(struct bpf_sock, family): 8576 case offsetof(struct bpf_sock, type): 8577 case offsetof(struct bpf_sock, protocol): 8578 case offsetof(struct bpf_sock, src_port): 8579 case offsetof(struct bpf_sock, rx_queue_mapping): 8580 case bpf_ctx_range(struct bpf_sock, src_ip4): 8581 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 8582 case bpf_ctx_range(struct bpf_sock, dst_ip4): 8583 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 8584 bpf_ctx_record_field_size(info, size_default); 8585 return bpf_ctx_narrow_access_ok(off, size, size_default); 8586 case bpf_ctx_range(struct bpf_sock, dst_port): 8587 field_size = size == size_default ? 8588 size_default : sizeof_field(struct bpf_sock, dst_port); 8589 bpf_ctx_record_field_size(info, field_size); 8590 return bpf_ctx_narrow_access_ok(off, size, field_size); 8591 case offsetofend(struct bpf_sock, dst_port) ... 8592 offsetof(struct bpf_sock, dst_ip4) - 1: 8593 return false; 8594 } 8595 8596 return size == size_default; 8597 } 8598 8599 static bool sock_filter_is_valid_access(int off, int size, 8600 enum bpf_access_type type, 8601 const struct bpf_prog *prog, 8602 struct bpf_insn_access_aux *info) 8603 { 8604 if (!bpf_sock_is_valid_access(off, size, type, info)) 8605 return false; 8606 return __sock_filter_check_attach_type(off, type, 8607 prog->expected_attach_type); 8608 } 8609 8610 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write, 8611 const struct bpf_prog *prog) 8612 { 8613 /* Neither direct read nor direct write requires any preliminary 8614 * action. 8615 */ 8616 return 0; 8617 } 8618 8619 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write, 8620 const struct bpf_prog *prog, int drop_verdict) 8621 { 8622 struct bpf_insn *insn = insn_buf; 8623 8624 if (!direct_write) 8625 return 0; 8626 8627 /* if (!skb->cloned) 8628 * goto start; 8629 * 8630 * (Fast-path, otherwise approximation that we might be 8631 * a clone, do the rest in helper.) 8632 */ 8633 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET); 8634 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK); 8635 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7); 8636 8637 /* ret = bpf_skb_pull_data(skb, 0); */ 8638 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1); 8639 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2); 8640 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, 8641 BPF_FUNC_skb_pull_data); 8642 /* if (!ret) 8643 * goto restore; 8644 * return TC_ACT_SHOT; 8645 */ 8646 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2); 8647 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict); 8648 *insn++ = BPF_EXIT_INSN(); 8649 8650 /* restore: */ 8651 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6); 8652 /* start: */ 8653 *insn++ = prog->insnsi[0]; 8654 8655 return insn - insn_buf; 8656 } 8657 8658 static int bpf_gen_ld_abs(const struct bpf_insn *orig, 8659 struct bpf_insn *insn_buf) 8660 { 8661 bool indirect = BPF_MODE(orig->code) == BPF_IND; 8662 struct bpf_insn *insn = insn_buf; 8663 8664 if (!indirect) { 8665 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm); 8666 } else { 8667 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg); 8668 if (orig->imm) 8669 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm); 8670 } 8671 /* We're guaranteed here that CTX is in R6. */ 8672 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX); 8673 8674 switch (BPF_SIZE(orig->code)) { 8675 case BPF_B: 8676 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache); 8677 break; 8678 case BPF_H: 8679 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache); 8680 break; 8681 case BPF_W: 8682 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache); 8683 break; 8684 } 8685 8686 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2); 8687 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0); 8688 *insn++ = BPF_EXIT_INSN(); 8689 8690 return insn - insn_buf; 8691 } 8692 8693 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write, 8694 const struct bpf_prog *prog) 8695 { 8696 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT); 8697 } 8698 8699 static bool tc_cls_act_is_valid_access(int off, int size, 8700 enum bpf_access_type type, 8701 const struct bpf_prog *prog, 8702 struct bpf_insn_access_aux *info) 8703 { 8704 if (type == BPF_WRITE) { 8705 switch (off) { 8706 case bpf_ctx_range(struct __sk_buff, mark): 8707 case bpf_ctx_range(struct __sk_buff, tc_index): 8708 case bpf_ctx_range(struct __sk_buff, priority): 8709 case bpf_ctx_range(struct __sk_buff, tc_classid): 8710 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8711 case bpf_ctx_range(struct __sk_buff, tstamp): 8712 case bpf_ctx_range(struct __sk_buff, queue_mapping): 8713 break; 8714 default: 8715 return false; 8716 } 8717 } 8718 8719 switch (off) { 8720 case bpf_ctx_range(struct __sk_buff, data): 8721 info->reg_type = PTR_TO_PACKET; 8722 break; 8723 case bpf_ctx_range(struct __sk_buff, data_meta): 8724 info->reg_type = PTR_TO_PACKET_META; 8725 break; 8726 case bpf_ctx_range(struct __sk_buff, data_end): 8727 info->reg_type = PTR_TO_PACKET_END; 8728 break; 8729 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8730 return false; 8731 case offsetof(struct __sk_buff, tstamp_type): 8732 /* The convert_ctx_access() on reading and writing 8733 * __sk_buff->tstamp depends on whether the bpf prog 8734 * has used __sk_buff->tstamp_type or not. 8735 * Thus, we need to set prog->tstamp_type_access 8736 * earlier during is_valid_access() here. 8737 */ 8738 ((struct bpf_prog *)prog)->tstamp_type_access = 1; 8739 return size == sizeof(__u8); 8740 } 8741 8742 return bpf_skb_is_valid_access(off, size, type, prog, info); 8743 } 8744 8745 DEFINE_MUTEX(nf_conn_btf_access_lock); 8746 EXPORT_SYMBOL_GPL(nf_conn_btf_access_lock); 8747 8748 int (*nfct_btf_struct_access)(struct bpf_verifier_log *log, 8749 const struct bpf_reg_state *reg, 8750 int off, int size); 8751 EXPORT_SYMBOL_GPL(nfct_btf_struct_access); 8752 8753 static int tc_cls_act_btf_struct_access(struct bpf_verifier_log *log, 8754 const struct bpf_reg_state *reg, 8755 int off, int size) 8756 { 8757 int ret = -EACCES; 8758 8759 mutex_lock(&nf_conn_btf_access_lock); 8760 if (nfct_btf_struct_access) 8761 ret = nfct_btf_struct_access(log, reg, off, size); 8762 mutex_unlock(&nf_conn_btf_access_lock); 8763 8764 return ret; 8765 } 8766 8767 static bool __is_valid_xdp_access(int off, int size) 8768 { 8769 if (off < 0 || off >= sizeof(struct xdp_md)) 8770 return false; 8771 if (off % size != 0) 8772 return false; 8773 if (size != sizeof(__u32)) 8774 return false; 8775 8776 return true; 8777 } 8778 8779 static bool xdp_is_valid_access(int off, int size, 8780 enum bpf_access_type type, 8781 const struct bpf_prog *prog, 8782 struct bpf_insn_access_aux *info) 8783 { 8784 if (prog->expected_attach_type != BPF_XDP_DEVMAP) { 8785 switch (off) { 8786 case offsetof(struct xdp_md, egress_ifindex): 8787 return false; 8788 } 8789 } 8790 8791 if (type == BPF_WRITE) { 8792 if (bpf_prog_is_offloaded(prog->aux)) { 8793 switch (off) { 8794 case offsetof(struct xdp_md, rx_queue_index): 8795 return __is_valid_xdp_access(off, size); 8796 } 8797 } 8798 return false; 8799 } 8800 8801 switch (off) { 8802 case offsetof(struct xdp_md, data): 8803 info->reg_type = PTR_TO_PACKET; 8804 break; 8805 case offsetof(struct xdp_md, data_meta): 8806 info->reg_type = PTR_TO_PACKET_META; 8807 break; 8808 case offsetof(struct xdp_md, data_end): 8809 info->reg_type = PTR_TO_PACKET_END; 8810 break; 8811 } 8812 8813 return __is_valid_xdp_access(off, size); 8814 } 8815 8816 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act) 8817 { 8818 const u32 act_max = XDP_REDIRECT; 8819 8820 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n", 8821 act > act_max ? "Illegal" : "Driver unsupported", 8822 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A"); 8823 } 8824 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); 8825 8826 static int xdp_btf_struct_access(struct bpf_verifier_log *log, 8827 const struct bpf_reg_state *reg, 8828 int off, int size) 8829 { 8830 int ret = -EACCES; 8831 8832 mutex_lock(&nf_conn_btf_access_lock); 8833 if (nfct_btf_struct_access) 8834 ret = nfct_btf_struct_access(log, reg, off, size); 8835 mutex_unlock(&nf_conn_btf_access_lock); 8836 8837 return ret; 8838 } 8839 8840 static bool sock_addr_is_valid_access(int off, int size, 8841 enum bpf_access_type type, 8842 const struct bpf_prog *prog, 8843 struct bpf_insn_access_aux *info) 8844 { 8845 const int size_default = sizeof(__u32); 8846 8847 if (off < 0 || off >= sizeof(struct bpf_sock_addr)) 8848 return false; 8849 if (off % size != 0) 8850 return false; 8851 8852 /* Disallow access to IPv6 fields from IPv4 contex and vise 8853 * versa. 8854 */ 8855 switch (off) { 8856 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 8857 switch (prog->expected_attach_type) { 8858 case BPF_CGROUP_INET4_BIND: 8859 case BPF_CGROUP_INET4_CONNECT: 8860 case BPF_CGROUP_INET4_GETPEERNAME: 8861 case BPF_CGROUP_INET4_GETSOCKNAME: 8862 case BPF_CGROUP_UDP4_SENDMSG: 8863 case BPF_CGROUP_UDP4_RECVMSG: 8864 break; 8865 default: 8866 return false; 8867 } 8868 break; 8869 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8870 switch (prog->expected_attach_type) { 8871 case BPF_CGROUP_INET6_BIND: 8872 case BPF_CGROUP_INET6_CONNECT: 8873 case BPF_CGROUP_INET6_GETPEERNAME: 8874 case BPF_CGROUP_INET6_GETSOCKNAME: 8875 case BPF_CGROUP_UDP6_SENDMSG: 8876 case BPF_CGROUP_UDP6_RECVMSG: 8877 break; 8878 default: 8879 return false; 8880 } 8881 break; 8882 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8883 switch (prog->expected_attach_type) { 8884 case BPF_CGROUP_UDP4_SENDMSG: 8885 break; 8886 default: 8887 return false; 8888 } 8889 break; 8890 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8891 msg_src_ip6[3]): 8892 switch (prog->expected_attach_type) { 8893 case BPF_CGROUP_UDP6_SENDMSG: 8894 break; 8895 default: 8896 return false; 8897 } 8898 break; 8899 } 8900 8901 switch (off) { 8902 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 8903 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8904 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8905 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8906 msg_src_ip6[3]): 8907 case bpf_ctx_range(struct bpf_sock_addr, user_port): 8908 if (type == BPF_READ) { 8909 bpf_ctx_record_field_size(info, size_default); 8910 8911 if (bpf_ctx_wide_access_ok(off, size, 8912 struct bpf_sock_addr, 8913 user_ip6)) 8914 return true; 8915 8916 if (bpf_ctx_wide_access_ok(off, size, 8917 struct bpf_sock_addr, 8918 msg_src_ip6)) 8919 return true; 8920 8921 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8922 return false; 8923 } else { 8924 if (bpf_ctx_wide_access_ok(off, size, 8925 struct bpf_sock_addr, 8926 user_ip6)) 8927 return true; 8928 8929 if (bpf_ctx_wide_access_ok(off, size, 8930 struct bpf_sock_addr, 8931 msg_src_ip6)) 8932 return true; 8933 8934 if (size != size_default) 8935 return false; 8936 } 8937 break; 8938 case offsetof(struct bpf_sock_addr, sk): 8939 if (type != BPF_READ) 8940 return false; 8941 if (size != sizeof(__u64)) 8942 return false; 8943 info->reg_type = PTR_TO_SOCKET; 8944 break; 8945 default: 8946 if (type == BPF_READ) { 8947 if (size != size_default) 8948 return false; 8949 } else { 8950 return false; 8951 } 8952 } 8953 8954 return true; 8955 } 8956 8957 static bool sock_ops_is_valid_access(int off, int size, 8958 enum bpf_access_type type, 8959 const struct bpf_prog *prog, 8960 struct bpf_insn_access_aux *info) 8961 { 8962 const int size_default = sizeof(__u32); 8963 8964 if (off < 0 || off >= sizeof(struct bpf_sock_ops)) 8965 return false; 8966 8967 /* The verifier guarantees that size > 0. */ 8968 if (off % size != 0) 8969 return false; 8970 8971 if (type == BPF_WRITE) { 8972 switch (off) { 8973 case offsetof(struct bpf_sock_ops, reply): 8974 case offsetof(struct bpf_sock_ops, sk_txhash): 8975 if (size != size_default) 8976 return false; 8977 break; 8978 default: 8979 return false; 8980 } 8981 } else { 8982 switch (off) { 8983 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received, 8984 bytes_acked): 8985 if (size != sizeof(__u64)) 8986 return false; 8987 break; 8988 case offsetof(struct bpf_sock_ops, sk): 8989 if (size != sizeof(__u64)) 8990 return false; 8991 info->reg_type = PTR_TO_SOCKET_OR_NULL; 8992 break; 8993 case offsetof(struct bpf_sock_ops, skb_data): 8994 if (size != sizeof(__u64)) 8995 return false; 8996 info->reg_type = PTR_TO_PACKET; 8997 break; 8998 case offsetof(struct bpf_sock_ops, skb_data_end): 8999 if (size != sizeof(__u64)) 9000 return false; 9001 info->reg_type = PTR_TO_PACKET_END; 9002 break; 9003 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 9004 bpf_ctx_record_field_size(info, size_default); 9005 return bpf_ctx_narrow_access_ok(off, size, 9006 size_default); 9007 case offsetof(struct bpf_sock_ops, skb_hwtstamp): 9008 if (size != sizeof(__u64)) 9009 return false; 9010 break; 9011 default: 9012 if (size != size_default) 9013 return false; 9014 break; 9015 } 9016 } 9017 9018 return true; 9019 } 9020 9021 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write, 9022 const struct bpf_prog *prog) 9023 { 9024 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP); 9025 } 9026 9027 static bool sk_skb_is_valid_access(int off, int size, 9028 enum bpf_access_type type, 9029 const struct bpf_prog *prog, 9030 struct bpf_insn_access_aux *info) 9031 { 9032 switch (off) { 9033 case bpf_ctx_range(struct __sk_buff, tc_classid): 9034 case bpf_ctx_range(struct __sk_buff, data_meta): 9035 case bpf_ctx_range(struct __sk_buff, tstamp): 9036 case bpf_ctx_range(struct __sk_buff, wire_len): 9037 case bpf_ctx_range(struct __sk_buff, hwtstamp): 9038 return false; 9039 } 9040 9041 if (type == BPF_WRITE) { 9042 switch (off) { 9043 case bpf_ctx_range(struct __sk_buff, tc_index): 9044 case bpf_ctx_range(struct __sk_buff, priority): 9045 break; 9046 default: 9047 return false; 9048 } 9049 } 9050 9051 switch (off) { 9052 case bpf_ctx_range(struct __sk_buff, mark): 9053 return false; 9054 case bpf_ctx_range(struct __sk_buff, data): 9055 info->reg_type = PTR_TO_PACKET; 9056 break; 9057 case bpf_ctx_range(struct __sk_buff, data_end): 9058 info->reg_type = PTR_TO_PACKET_END; 9059 break; 9060 } 9061 9062 return bpf_skb_is_valid_access(off, size, type, prog, info); 9063 } 9064 9065 static bool sk_msg_is_valid_access(int off, int size, 9066 enum bpf_access_type type, 9067 const struct bpf_prog *prog, 9068 struct bpf_insn_access_aux *info) 9069 { 9070 if (type == BPF_WRITE) 9071 return false; 9072 9073 if (off % size != 0) 9074 return false; 9075 9076 switch (off) { 9077 case offsetof(struct sk_msg_md, data): 9078 info->reg_type = PTR_TO_PACKET; 9079 if (size != sizeof(__u64)) 9080 return false; 9081 break; 9082 case offsetof(struct sk_msg_md, data_end): 9083 info->reg_type = PTR_TO_PACKET_END; 9084 if (size != sizeof(__u64)) 9085 return false; 9086 break; 9087 case offsetof(struct sk_msg_md, sk): 9088 if (size != sizeof(__u64)) 9089 return false; 9090 info->reg_type = PTR_TO_SOCKET; 9091 break; 9092 case bpf_ctx_range(struct sk_msg_md, family): 9093 case bpf_ctx_range(struct sk_msg_md, remote_ip4): 9094 case bpf_ctx_range(struct sk_msg_md, local_ip4): 9095 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]): 9096 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]): 9097 case bpf_ctx_range(struct sk_msg_md, remote_port): 9098 case bpf_ctx_range(struct sk_msg_md, local_port): 9099 case bpf_ctx_range(struct sk_msg_md, size): 9100 if (size != sizeof(__u32)) 9101 return false; 9102 break; 9103 default: 9104 return false; 9105 } 9106 return true; 9107 } 9108 9109 static bool flow_dissector_is_valid_access(int off, int size, 9110 enum bpf_access_type type, 9111 const struct bpf_prog *prog, 9112 struct bpf_insn_access_aux *info) 9113 { 9114 const int size_default = sizeof(__u32); 9115 9116 if (off < 0 || off >= sizeof(struct __sk_buff)) 9117 return false; 9118 9119 if (type == BPF_WRITE) 9120 return false; 9121 9122 switch (off) { 9123 case bpf_ctx_range(struct __sk_buff, data): 9124 if (size != size_default) 9125 return false; 9126 info->reg_type = PTR_TO_PACKET; 9127 return true; 9128 case bpf_ctx_range(struct __sk_buff, data_end): 9129 if (size != size_default) 9130 return false; 9131 info->reg_type = PTR_TO_PACKET_END; 9132 return true; 9133 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 9134 if (size != sizeof(__u64)) 9135 return false; 9136 info->reg_type = PTR_TO_FLOW_KEYS; 9137 return true; 9138 default: 9139 return false; 9140 } 9141 } 9142 9143 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type, 9144 const struct bpf_insn *si, 9145 struct bpf_insn *insn_buf, 9146 struct bpf_prog *prog, 9147 u32 *target_size) 9148 9149 { 9150 struct bpf_insn *insn = insn_buf; 9151 9152 switch (si->off) { 9153 case offsetof(struct __sk_buff, data): 9154 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data), 9155 si->dst_reg, si->src_reg, 9156 offsetof(struct bpf_flow_dissector, data)); 9157 break; 9158 9159 case offsetof(struct __sk_buff, data_end): 9160 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end), 9161 si->dst_reg, si->src_reg, 9162 offsetof(struct bpf_flow_dissector, data_end)); 9163 break; 9164 9165 case offsetof(struct __sk_buff, flow_keys): 9166 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys), 9167 si->dst_reg, si->src_reg, 9168 offsetof(struct bpf_flow_dissector, flow_keys)); 9169 break; 9170 } 9171 9172 return insn - insn_buf; 9173 } 9174 9175 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si, 9176 struct bpf_insn *insn) 9177 { 9178 __u8 value_reg = si->dst_reg; 9179 __u8 skb_reg = si->src_reg; 9180 /* AX is needed because src_reg and dst_reg could be the same */ 9181 __u8 tmp_reg = BPF_REG_AX; 9182 9183 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, 9184 SKB_BF_MONO_TC_OFFSET); 9185 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, 9186 SKB_MONO_DELIVERY_TIME_MASK, 2); 9187 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC); 9188 *insn++ = BPF_JMP_A(1); 9189 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO); 9190 9191 return insn; 9192 } 9193 9194 static struct bpf_insn *bpf_convert_shinfo_access(__u8 dst_reg, __u8 skb_reg, 9195 struct bpf_insn *insn) 9196 { 9197 /* si->dst_reg = skb_shinfo(SKB); */ 9198 #ifdef NET_SKBUFF_DATA_USES_OFFSET 9199 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 9200 BPF_REG_AX, skb_reg, 9201 offsetof(struct sk_buff, end)); 9202 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head), 9203 dst_reg, skb_reg, 9204 offsetof(struct sk_buff, head)); 9205 *insn++ = BPF_ALU64_REG(BPF_ADD, dst_reg, BPF_REG_AX); 9206 #else 9207 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 9208 dst_reg, skb_reg, 9209 offsetof(struct sk_buff, end)); 9210 #endif 9211 9212 return insn; 9213 } 9214 9215 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog, 9216 const struct bpf_insn *si, 9217 struct bpf_insn *insn) 9218 { 9219 __u8 value_reg = si->dst_reg; 9220 __u8 skb_reg = si->src_reg; 9221 9222 #ifdef CONFIG_NET_CLS_ACT 9223 /* If the tstamp_type is read, 9224 * the bpf prog is aware the tstamp could have delivery time. 9225 * Thus, read skb->tstamp as is if tstamp_type_access is true. 9226 */ 9227 if (!prog->tstamp_type_access) { 9228 /* AX is needed because src_reg and dst_reg could be the same */ 9229 __u8 tmp_reg = BPF_REG_AX; 9230 9231 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET); 9232 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, 9233 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK); 9234 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg, 9235 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2); 9236 /* skb->tc_at_ingress && skb->mono_delivery_time, 9237 * read 0 as the (rcv) timestamp. 9238 */ 9239 *insn++ = BPF_MOV64_IMM(value_reg, 0); 9240 *insn++ = BPF_JMP_A(1); 9241 } 9242 #endif 9243 9244 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg, 9245 offsetof(struct sk_buff, tstamp)); 9246 return insn; 9247 } 9248 9249 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog, 9250 const struct bpf_insn *si, 9251 struct bpf_insn *insn) 9252 { 9253 __u8 value_reg = si->src_reg; 9254 __u8 skb_reg = si->dst_reg; 9255 9256 #ifdef CONFIG_NET_CLS_ACT 9257 /* If the tstamp_type is read, 9258 * the bpf prog is aware the tstamp could have delivery time. 9259 * Thus, write skb->tstamp as is if tstamp_type_access is true. 9260 * Otherwise, writing at ingress will have to clear the 9261 * mono_delivery_time bit also. 9262 */ 9263 if (!prog->tstamp_type_access) { 9264 __u8 tmp_reg = BPF_REG_AX; 9265 9266 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, SKB_BF_MONO_TC_OFFSET); 9267 /* Writing __sk_buff->tstamp as ingress, goto <clear> */ 9268 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1); 9269 /* goto <store> */ 9270 *insn++ = BPF_JMP_A(2); 9271 /* <clear>: mono_delivery_time */ 9272 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK); 9273 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, SKB_BF_MONO_TC_OFFSET); 9274 } 9275 #endif 9276 9277 /* <store>: skb->tstamp = tstamp */ 9278 *insn++ = BPF_RAW_INSN(BPF_CLASS(si->code) | BPF_DW | BPF_MEM, 9279 skb_reg, value_reg, offsetof(struct sk_buff, tstamp), si->imm); 9280 return insn; 9281 } 9282 9283 #define BPF_EMIT_STORE(size, si, off) \ 9284 BPF_RAW_INSN(BPF_CLASS((si)->code) | (size) | BPF_MEM, \ 9285 (si)->dst_reg, (si)->src_reg, (off), (si)->imm) 9286 9287 static u32 bpf_convert_ctx_access(enum bpf_access_type type, 9288 const struct bpf_insn *si, 9289 struct bpf_insn *insn_buf, 9290 struct bpf_prog *prog, u32 *target_size) 9291 { 9292 struct bpf_insn *insn = insn_buf; 9293 int off; 9294 9295 switch (si->off) { 9296 case offsetof(struct __sk_buff, len): 9297 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9298 bpf_target_off(struct sk_buff, len, 4, 9299 target_size)); 9300 break; 9301 9302 case offsetof(struct __sk_buff, protocol): 9303 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9304 bpf_target_off(struct sk_buff, protocol, 2, 9305 target_size)); 9306 break; 9307 9308 case offsetof(struct __sk_buff, vlan_proto): 9309 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9310 bpf_target_off(struct sk_buff, vlan_proto, 2, 9311 target_size)); 9312 break; 9313 9314 case offsetof(struct __sk_buff, priority): 9315 if (type == BPF_WRITE) 9316 *insn++ = BPF_EMIT_STORE(BPF_W, si, 9317 bpf_target_off(struct sk_buff, priority, 4, 9318 target_size)); 9319 else 9320 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9321 bpf_target_off(struct sk_buff, priority, 4, 9322 target_size)); 9323 break; 9324 9325 case offsetof(struct __sk_buff, ingress_ifindex): 9326 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9327 bpf_target_off(struct sk_buff, skb_iif, 4, 9328 target_size)); 9329 break; 9330 9331 case offsetof(struct __sk_buff, ifindex): 9332 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 9333 si->dst_reg, si->src_reg, 9334 offsetof(struct sk_buff, dev)); 9335 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9336 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9337 bpf_target_off(struct net_device, ifindex, 4, 9338 target_size)); 9339 break; 9340 9341 case offsetof(struct __sk_buff, hash): 9342 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9343 bpf_target_off(struct sk_buff, hash, 4, 9344 target_size)); 9345 break; 9346 9347 case offsetof(struct __sk_buff, mark): 9348 if (type == BPF_WRITE) 9349 *insn++ = BPF_EMIT_STORE(BPF_W, si, 9350 bpf_target_off(struct sk_buff, mark, 4, 9351 target_size)); 9352 else 9353 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9354 bpf_target_off(struct sk_buff, mark, 4, 9355 target_size)); 9356 break; 9357 9358 case offsetof(struct __sk_buff, pkt_type): 9359 *target_size = 1; 9360 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 9361 PKT_TYPE_OFFSET); 9362 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX); 9363 #ifdef __BIG_ENDIAN_BITFIELD 9364 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5); 9365 #endif 9366 break; 9367 9368 case offsetof(struct __sk_buff, queue_mapping): 9369 if (type == BPF_WRITE) { 9370 u32 off = bpf_target_off(struct sk_buff, queue_mapping, 2, target_size); 9371 9372 if (BPF_CLASS(si->code) == BPF_ST && si->imm >= NO_QUEUE_MAPPING) { 9373 *insn++ = BPF_JMP_A(0); /* noop */ 9374 break; 9375 } 9376 9377 if (BPF_CLASS(si->code) == BPF_STX) 9378 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1); 9379 *insn++ = BPF_EMIT_STORE(BPF_H, si, off); 9380 } else { 9381 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9382 bpf_target_off(struct sk_buff, 9383 queue_mapping, 9384 2, target_size)); 9385 } 9386 break; 9387 9388 case offsetof(struct __sk_buff, vlan_present): 9389 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9390 bpf_target_off(struct sk_buff, 9391 vlan_all, 4, target_size)); 9392 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9393 *insn++ = BPF_ALU32_IMM(BPF_MOV, si->dst_reg, 1); 9394 break; 9395 9396 case offsetof(struct __sk_buff, vlan_tci): 9397 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9398 bpf_target_off(struct sk_buff, vlan_tci, 2, 9399 target_size)); 9400 break; 9401 9402 case offsetof(struct __sk_buff, cb[0]) ... 9403 offsetofend(struct __sk_buff, cb[4]) - 1: 9404 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20); 9405 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 9406 offsetof(struct qdisc_skb_cb, data)) % 9407 sizeof(__u64)); 9408 9409 prog->cb_access = 1; 9410 off = si->off; 9411 off -= offsetof(struct __sk_buff, cb[0]); 9412 off += offsetof(struct sk_buff, cb); 9413 off += offsetof(struct qdisc_skb_cb, data); 9414 if (type == BPF_WRITE) 9415 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off); 9416 else 9417 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 9418 si->src_reg, off); 9419 break; 9420 9421 case offsetof(struct __sk_buff, tc_classid): 9422 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2); 9423 9424 off = si->off; 9425 off -= offsetof(struct __sk_buff, tc_classid); 9426 off += offsetof(struct sk_buff, cb); 9427 off += offsetof(struct qdisc_skb_cb, tc_classid); 9428 *target_size = 2; 9429 if (type == BPF_WRITE) 9430 *insn++ = BPF_EMIT_STORE(BPF_H, si, off); 9431 else 9432 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, 9433 si->src_reg, off); 9434 break; 9435 9436 case offsetof(struct __sk_buff, data): 9437 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 9438 si->dst_reg, si->src_reg, 9439 offsetof(struct sk_buff, data)); 9440 break; 9441 9442 case offsetof(struct __sk_buff, data_meta): 9443 off = si->off; 9444 off -= offsetof(struct __sk_buff, data_meta); 9445 off += offsetof(struct sk_buff, cb); 9446 off += offsetof(struct bpf_skb_data_end, data_meta); 9447 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 9448 si->src_reg, off); 9449 break; 9450 9451 case offsetof(struct __sk_buff, data_end): 9452 off = si->off; 9453 off -= offsetof(struct __sk_buff, data_end); 9454 off += offsetof(struct sk_buff, cb); 9455 off += offsetof(struct bpf_skb_data_end, data_end); 9456 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 9457 si->src_reg, off); 9458 break; 9459 9460 case offsetof(struct __sk_buff, tc_index): 9461 #ifdef CONFIG_NET_SCHED 9462 if (type == BPF_WRITE) 9463 *insn++ = BPF_EMIT_STORE(BPF_H, si, 9464 bpf_target_off(struct sk_buff, tc_index, 2, 9465 target_size)); 9466 else 9467 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9468 bpf_target_off(struct sk_buff, tc_index, 2, 9469 target_size)); 9470 #else 9471 *target_size = 2; 9472 if (type == BPF_WRITE) 9473 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg); 9474 else 9475 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9476 #endif 9477 break; 9478 9479 case offsetof(struct __sk_buff, napi_id): 9480 #if defined(CONFIG_NET_RX_BUSY_POLL) 9481 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9482 bpf_target_off(struct sk_buff, napi_id, 4, 9483 target_size)); 9484 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1); 9485 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9486 #else 9487 *target_size = 4; 9488 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9489 #endif 9490 break; 9491 case offsetof(struct __sk_buff, family): 9492 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 9493 9494 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9495 si->dst_reg, si->src_reg, 9496 offsetof(struct sk_buff, sk)); 9497 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9498 bpf_target_off(struct sock_common, 9499 skc_family, 9500 2, target_size)); 9501 break; 9502 case offsetof(struct __sk_buff, remote_ip4): 9503 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 9504 9505 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9506 si->dst_reg, si->src_reg, 9507 offsetof(struct sk_buff, sk)); 9508 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9509 bpf_target_off(struct sock_common, 9510 skc_daddr, 9511 4, target_size)); 9512 break; 9513 case offsetof(struct __sk_buff, local_ip4): 9514 BUILD_BUG_ON(sizeof_field(struct sock_common, 9515 skc_rcv_saddr) != 4); 9516 9517 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9518 si->dst_reg, si->src_reg, 9519 offsetof(struct sk_buff, sk)); 9520 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9521 bpf_target_off(struct sock_common, 9522 skc_rcv_saddr, 9523 4, target_size)); 9524 break; 9525 case offsetof(struct __sk_buff, remote_ip6[0]) ... 9526 offsetof(struct __sk_buff, remote_ip6[3]): 9527 #if IS_ENABLED(CONFIG_IPV6) 9528 BUILD_BUG_ON(sizeof_field(struct sock_common, 9529 skc_v6_daddr.s6_addr32[0]) != 4); 9530 9531 off = si->off; 9532 off -= offsetof(struct __sk_buff, remote_ip6[0]); 9533 9534 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9535 si->dst_reg, si->src_reg, 9536 offsetof(struct sk_buff, sk)); 9537 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9538 offsetof(struct sock_common, 9539 skc_v6_daddr.s6_addr32[0]) + 9540 off); 9541 #else 9542 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9543 #endif 9544 break; 9545 case offsetof(struct __sk_buff, local_ip6[0]) ... 9546 offsetof(struct __sk_buff, local_ip6[3]): 9547 #if IS_ENABLED(CONFIG_IPV6) 9548 BUILD_BUG_ON(sizeof_field(struct sock_common, 9549 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 9550 9551 off = si->off; 9552 off -= offsetof(struct __sk_buff, local_ip6[0]); 9553 9554 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9555 si->dst_reg, si->src_reg, 9556 offsetof(struct sk_buff, sk)); 9557 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9558 offsetof(struct sock_common, 9559 skc_v6_rcv_saddr.s6_addr32[0]) + 9560 off); 9561 #else 9562 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9563 #endif 9564 break; 9565 9566 case offsetof(struct __sk_buff, remote_port): 9567 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 9568 9569 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9570 si->dst_reg, si->src_reg, 9571 offsetof(struct sk_buff, sk)); 9572 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9573 bpf_target_off(struct sock_common, 9574 skc_dport, 9575 2, target_size)); 9576 #ifndef __BIG_ENDIAN_BITFIELD 9577 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 9578 #endif 9579 break; 9580 9581 case offsetof(struct __sk_buff, local_port): 9582 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 9583 9584 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9585 si->dst_reg, si->src_reg, 9586 offsetof(struct sk_buff, sk)); 9587 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9588 bpf_target_off(struct sock_common, 9589 skc_num, 2, target_size)); 9590 break; 9591 9592 case offsetof(struct __sk_buff, tstamp): 9593 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8); 9594 9595 if (type == BPF_WRITE) 9596 insn = bpf_convert_tstamp_write(prog, si, insn); 9597 else 9598 insn = bpf_convert_tstamp_read(prog, si, insn); 9599 break; 9600 9601 case offsetof(struct __sk_buff, tstamp_type): 9602 insn = bpf_convert_tstamp_type_read(si, insn); 9603 break; 9604 9605 case offsetof(struct __sk_buff, gso_segs): 9606 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn); 9607 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs), 9608 si->dst_reg, si->dst_reg, 9609 bpf_target_off(struct skb_shared_info, 9610 gso_segs, 2, 9611 target_size)); 9612 break; 9613 case offsetof(struct __sk_buff, gso_size): 9614 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn); 9615 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size), 9616 si->dst_reg, si->dst_reg, 9617 bpf_target_off(struct skb_shared_info, 9618 gso_size, 2, 9619 target_size)); 9620 break; 9621 case offsetof(struct __sk_buff, wire_len): 9622 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4); 9623 9624 off = si->off; 9625 off -= offsetof(struct __sk_buff, wire_len); 9626 off += offsetof(struct sk_buff, cb); 9627 off += offsetof(struct qdisc_skb_cb, pkt_len); 9628 *target_size = 4; 9629 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off); 9630 break; 9631 9632 case offsetof(struct __sk_buff, sk): 9633 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9634 si->dst_reg, si->src_reg, 9635 offsetof(struct sk_buff, sk)); 9636 break; 9637 case offsetof(struct __sk_buff, hwtstamp): 9638 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8); 9639 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0); 9640 9641 insn = bpf_convert_shinfo_access(si->dst_reg, si->src_reg, insn); 9642 *insn++ = BPF_LDX_MEM(BPF_DW, 9643 si->dst_reg, si->dst_reg, 9644 bpf_target_off(struct skb_shared_info, 9645 hwtstamps, 8, 9646 target_size)); 9647 break; 9648 } 9649 9650 return insn - insn_buf; 9651 } 9652 9653 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 9654 const struct bpf_insn *si, 9655 struct bpf_insn *insn_buf, 9656 struct bpf_prog *prog, u32 *target_size) 9657 { 9658 struct bpf_insn *insn = insn_buf; 9659 int off; 9660 9661 switch (si->off) { 9662 case offsetof(struct bpf_sock, bound_dev_if): 9663 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4); 9664 9665 if (type == BPF_WRITE) 9666 *insn++ = BPF_EMIT_STORE(BPF_W, si, 9667 offsetof(struct sock, sk_bound_dev_if)); 9668 else 9669 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9670 offsetof(struct sock, sk_bound_dev_if)); 9671 break; 9672 9673 case offsetof(struct bpf_sock, mark): 9674 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4); 9675 9676 if (type == BPF_WRITE) 9677 *insn++ = BPF_EMIT_STORE(BPF_W, si, 9678 offsetof(struct sock, sk_mark)); 9679 else 9680 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9681 offsetof(struct sock, sk_mark)); 9682 break; 9683 9684 case offsetof(struct bpf_sock, priority): 9685 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4); 9686 9687 if (type == BPF_WRITE) 9688 *insn++ = BPF_EMIT_STORE(BPF_W, si, 9689 offsetof(struct sock, sk_priority)); 9690 else 9691 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9692 offsetof(struct sock, sk_priority)); 9693 break; 9694 9695 case offsetof(struct bpf_sock, family): 9696 *insn++ = BPF_LDX_MEM( 9697 BPF_FIELD_SIZEOF(struct sock_common, skc_family), 9698 si->dst_reg, si->src_reg, 9699 bpf_target_off(struct sock_common, 9700 skc_family, 9701 sizeof_field(struct sock_common, 9702 skc_family), 9703 target_size)); 9704 break; 9705 9706 case offsetof(struct bpf_sock, type): 9707 *insn++ = BPF_LDX_MEM( 9708 BPF_FIELD_SIZEOF(struct sock, sk_type), 9709 si->dst_reg, si->src_reg, 9710 bpf_target_off(struct sock, sk_type, 9711 sizeof_field(struct sock, sk_type), 9712 target_size)); 9713 break; 9714 9715 case offsetof(struct bpf_sock, protocol): 9716 *insn++ = BPF_LDX_MEM( 9717 BPF_FIELD_SIZEOF(struct sock, sk_protocol), 9718 si->dst_reg, si->src_reg, 9719 bpf_target_off(struct sock, sk_protocol, 9720 sizeof_field(struct sock, sk_protocol), 9721 target_size)); 9722 break; 9723 9724 case offsetof(struct bpf_sock, src_ip4): 9725 *insn++ = BPF_LDX_MEM( 9726 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9727 bpf_target_off(struct sock_common, skc_rcv_saddr, 9728 sizeof_field(struct sock_common, 9729 skc_rcv_saddr), 9730 target_size)); 9731 break; 9732 9733 case offsetof(struct bpf_sock, dst_ip4): 9734 *insn++ = BPF_LDX_MEM( 9735 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9736 bpf_target_off(struct sock_common, skc_daddr, 9737 sizeof_field(struct sock_common, 9738 skc_daddr), 9739 target_size)); 9740 break; 9741 9742 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 9743 #if IS_ENABLED(CONFIG_IPV6) 9744 off = si->off; 9745 off -= offsetof(struct bpf_sock, src_ip6[0]); 9746 *insn++ = BPF_LDX_MEM( 9747 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9748 bpf_target_off( 9749 struct sock_common, 9750 skc_v6_rcv_saddr.s6_addr32[0], 9751 sizeof_field(struct sock_common, 9752 skc_v6_rcv_saddr.s6_addr32[0]), 9753 target_size) + off); 9754 #else 9755 (void)off; 9756 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9757 #endif 9758 break; 9759 9760 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 9761 #if IS_ENABLED(CONFIG_IPV6) 9762 off = si->off; 9763 off -= offsetof(struct bpf_sock, dst_ip6[0]); 9764 *insn++ = BPF_LDX_MEM( 9765 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9766 bpf_target_off(struct sock_common, 9767 skc_v6_daddr.s6_addr32[0], 9768 sizeof_field(struct sock_common, 9769 skc_v6_daddr.s6_addr32[0]), 9770 target_size) + off); 9771 #else 9772 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9773 *target_size = 4; 9774 #endif 9775 break; 9776 9777 case offsetof(struct bpf_sock, src_port): 9778 *insn++ = BPF_LDX_MEM( 9779 BPF_FIELD_SIZEOF(struct sock_common, skc_num), 9780 si->dst_reg, si->src_reg, 9781 bpf_target_off(struct sock_common, skc_num, 9782 sizeof_field(struct sock_common, 9783 skc_num), 9784 target_size)); 9785 break; 9786 9787 case offsetof(struct bpf_sock, dst_port): 9788 *insn++ = BPF_LDX_MEM( 9789 BPF_FIELD_SIZEOF(struct sock_common, skc_dport), 9790 si->dst_reg, si->src_reg, 9791 bpf_target_off(struct sock_common, skc_dport, 9792 sizeof_field(struct sock_common, 9793 skc_dport), 9794 target_size)); 9795 break; 9796 9797 case offsetof(struct bpf_sock, state): 9798 *insn++ = BPF_LDX_MEM( 9799 BPF_FIELD_SIZEOF(struct sock_common, skc_state), 9800 si->dst_reg, si->src_reg, 9801 bpf_target_off(struct sock_common, skc_state, 9802 sizeof_field(struct sock_common, 9803 skc_state), 9804 target_size)); 9805 break; 9806 case offsetof(struct bpf_sock, rx_queue_mapping): 9807 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 9808 *insn++ = BPF_LDX_MEM( 9809 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping), 9810 si->dst_reg, si->src_reg, 9811 bpf_target_off(struct sock, sk_rx_queue_mapping, 9812 sizeof_field(struct sock, 9813 sk_rx_queue_mapping), 9814 target_size)); 9815 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING, 9816 1); 9817 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 9818 #else 9819 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 9820 *target_size = 2; 9821 #endif 9822 break; 9823 } 9824 9825 return insn - insn_buf; 9826 } 9827 9828 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, 9829 const struct bpf_insn *si, 9830 struct bpf_insn *insn_buf, 9831 struct bpf_prog *prog, u32 *target_size) 9832 { 9833 struct bpf_insn *insn = insn_buf; 9834 9835 switch (si->off) { 9836 case offsetof(struct __sk_buff, ifindex): 9837 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 9838 si->dst_reg, si->src_reg, 9839 offsetof(struct sk_buff, dev)); 9840 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9841 bpf_target_off(struct net_device, ifindex, 4, 9842 target_size)); 9843 break; 9844 default: 9845 return bpf_convert_ctx_access(type, si, insn_buf, prog, 9846 target_size); 9847 } 9848 9849 return insn - insn_buf; 9850 } 9851 9852 static u32 xdp_convert_ctx_access(enum bpf_access_type type, 9853 const struct bpf_insn *si, 9854 struct bpf_insn *insn_buf, 9855 struct bpf_prog *prog, u32 *target_size) 9856 { 9857 struct bpf_insn *insn = insn_buf; 9858 9859 switch (si->off) { 9860 case offsetof(struct xdp_md, data): 9861 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data), 9862 si->dst_reg, si->src_reg, 9863 offsetof(struct xdp_buff, data)); 9864 break; 9865 case offsetof(struct xdp_md, data_meta): 9866 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta), 9867 si->dst_reg, si->src_reg, 9868 offsetof(struct xdp_buff, data_meta)); 9869 break; 9870 case offsetof(struct xdp_md, data_end): 9871 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end), 9872 si->dst_reg, si->src_reg, 9873 offsetof(struct xdp_buff, data_end)); 9874 break; 9875 case offsetof(struct xdp_md, ingress_ifindex): 9876 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 9877 si->dst_reg, si->src_reg, 9878 offsetof(struct xdp_buff, rxq)); 9879 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev), 9880 si->dst_reg, si->dst_reg, 9881 offsetof(struct xdp_rxq_info, dev)); 9882 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9883 offsetof(struct net_device, ifindex)); 9884 break; 9885 case offsetof(struct xdp_md, rx_queue_index): 9886 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 9887 si->dst_reg, si->src_reg, 9888 offsetof(struct xdp_buff, rxq)); 9889 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9890 offsetof(struct xdp_rxq_info, 9891 queue_index)); 9892 break; 9893 case offsetof(struct xdp_md, egress_ifindex): 9894 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq), 9895 si->dst_reg, si->src_reg, 9896 offsetof(struct xdp_buff, txq)); 9897 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev), 9898 si->dst_reg, si->dst_reg, 9899 offsetof(struct xdp_txq_info, dev)); 9900 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9901 offsetof(struct net_device, ifindex)); 9902 break; 9903 } 9904 9905 return insn - insn_buf; 9906 } 9907 9908 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of 9909 * context Structure, F is Field in context structure that contains a pointer 9910 * to Nested Structure of type NS that has the field NF. 9911 * 9912 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make 9913 * sure that SIZE is not greater than actual size of S.F.NF. 9914 * 9915 * If offset OFF is provided, the load happens from that offset relative to 9916 * offset of NF. 9917 */ 9918 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \ 9919 do { \ 9920 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \ 9921 si->src_reg, offsetof(S, F)); \ 9922 *insn++ = BPF_LDX_MEM( \ 9923 SIZE, si->dst_reg, si->dst_reg, \ 9924 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 9925 target_size) \ 9926 + OFF); \ 9927 } while (0) 9928 9929 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \ 9930 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \ 9931 BPF_FIELD_SIZEOF(NS, NF), 0) 9932 9933 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to 9934 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation. 9935 * 9936 * In addition it uses Temporary Field TF (member of struct S) as the 3rd 9937 * "register" since two registers available in convert_ctx_access are not 9938 * enough: we can't override neither SRC, since it contains value to store, nor 9939 * DST since it contains pointer to context that may be used by later 9940 * instructions. But we need a temporary place to save pointer to nested 9941 * structure whose field we want to store to. 9942 */ 9943 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \ 9944 do { \ 9945 int tmp_reg = BPF_REG_9; \ 9946 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 9947 --tmp_reg; \ 9948 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 9949 --tmp_reg; \ 9950 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \ 9951 offsetof(S, TF)); \ 9952 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \ 9953 si->dst_reg, offsetof(S, F)); \ 9954 *insn++ = BPF_RAW_INSN(SIZE | BPF_MEM | BPF_CLASS(si->code), \ 9955 tmp_reg, si->src_reg, \ 9956 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 9957 target_size) \ 9958 + OFF, \ 9959 si->imm); \ 9960 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \ 9961 offsetof(S, TF)); \ 9962 } while (0) 9963 9964 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \ 9965 TF) \ 9966 do { \ 9967 if (type == BPF_WRITE) { \ 9968 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \ 9969 OFF, TF); \ 9970 } else { \ 9971 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \ 9972 S, NS, F, NF, SIZE, OFF); \ 9973 } \ 9974 } while (0) 9975 9976 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \ 9977 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \ 9978 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF) 9979 9980 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type, 9981 const struct bpf_insn *si, 9982 struct bpf_insn *insn_buf, 9983 struct bpf_prog *prog, u32 *target_size) 9984 { 9985 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port); 9986 struct bpf_insn *insn = insn_buf; 9987 9988 switch (si->off) { 9989 case offsetof(struct bpf_sock_addr, user_family): 9990 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9991 struct sockaddr, uaddr, sa_family); 9992 break; 9993 9994 case offsetof(struct bpf_sock_addr, user_ip4): 9995 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9996 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr, 9997 sin_addr, BPF_SIZE(si->code), 0, tmp_reg); 9998 break; 9999 10000 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 10001 off = si->off; 10002 off -= offsetof(struct bpf_sock_addr, user_ip6[0]); 10003 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 10004 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 10005 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off, 10006 tmp_reg); 10007 break; 10008 10009 case offsetof(struct bpf_sock_addr, user_port): 10010 /* To get port we need to know sa_family first and then treat 10011 * sockaddr as either sockaddr_in or sockaddr_in6. 10012 * Though we can simplify since port field has same offset and 10013 * size in both structures. 10014 * Here we check this invariant and use just one of the 10015 * structures if it's true. 10016 */ 10017 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) != 10018 offsetof(struct sockaddr_in6, sin6_port)); 10019 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) != 10020 sizeof_field(struct sockaddr_in6, sin6_port)); 10021 /* Account for sin6_port being smaller than user_port. */ 10022 port_size = min(port_size, BPF_LDST_BYTES(si)); 10023 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 10024 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 10025 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg); 10026 break; 10027 10028 case offsetof(struct bpf_sock_addr, family): 10029 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 10030 struct sock, sk, sk_family); 10031 break; 10032 10033 case offsetof(struct bpf_sock_addr, type): 10034 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 10035 struct sock, sk, sk_type); 10036 break; 10037 10038 case offsetof(struct bpf_sock_addr, protocol): 10039 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 10040 struct sock, sk, sk_protocol); 10041 break; 10042 10043 case offsetof(struct bpf_sock_addr, msg_src_ip4): 10044 /* Treat t_ctx as struct in_addr for msg_src_ip4. */ 10045 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 10046 struct bpf_sock_addr_kern, struct in_addr, t_ctx, 10047 s_addr, BPF_SIZE(si->code), 0, tmp_reg); 10048 break; 10049 10050 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 10051 msg_src_ip6[3]): 10052 off = si->off; 10053 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]); 10054 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */ 10055 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 10056 struct bpf_sock_addr_kern, struct in6_addr, t_ctx, 10057 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg); 10058 break; 10059 case offsetof(struct bpf_sock_addr, sk): 10060 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk), 10061 si->dst_reg, si->src_reg, 10062 offsetof(struct bpf_sock_addr_kern, sk)); 10063 break; 10064 } 10065 10066 return insn - insn_buf; 10067 } 10068 10069 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type, 10070 const struct bpf_insn *si, 10071 struct bpf_insn *insn_buf, 10072 struct bpf_prog *prog, 10073 u32 *target_size) 10074 { 10075 struct bpf_insn *insn = insn_buf; 10076 int off; 10077 10078 /* Helper macro for adding read access to tcp_sock or sock fields. */ 10079 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 10080 do { \ 10081 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \ 10082 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 10083 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 10084 if (si->dst_reg == reg || si->src_reg == reg) \ 10085 reg--; \ 10086 if (si->dst_reg == reg || si->src_reg == reg) \ 10087 reg--; \ 10088 if (si->dst_reg == si->src_reg) { \ 10089 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 10090 offsetof(struct bpf_sock_ops_kern, \ 10091 temp)); \ 10092 fullsock_reg = reg; \ 10093 jmp += 2; \ 10094 } \ 10095 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10096 struct bpf_sock_ops_kern, \ 10097 is_fullsock), \ 10098 fullsock_reg, si->src_reg, \ 10099 offsetof(struct bpf_sock_ops_kern, \ 10100 is_fullsock)); \ 10101 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 10102 if (si->dst_reg == si->src_reg) \ 10103 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10104 offsetof(struct bpf_sock_ops_kern, \ 10105 temp)); \ 10106 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10107 struct bpf_sock_ops_kern, sk),\ 10108 si->dst_reg, si->src_reg, \ 10109 offsetof(struct bpf_sock_ops_kern, sk));\ 10110 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \ 10111 OBJ_FIELD), \ 10112 si->dst_reg, si->dst_reg, \ 10113 offsetof(OBJ, OBJ_FIELD)); \ 10114 if (si->dst_reg == si->src_reg) { \ 10115 *insn++ = BPF_JMP_A(1); \ 10116 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10117 offsetof(struct bpf_sock_ops_kern, \ 10118 temp)); \ 10119 } \ 10120 } while (0) 10121 10122 #define SOCK_OPS_GET_SK() \ 10123 do { \ 10124 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \ 10125 if (si->dst_reg == reg || si->src_reg == reg) \ 10126 reg--; \ 10127 if (si->dst_reg == reg || si->src_reg == reg) \ 10128 reg--; \ 10129 if (si->dst_reg == si->src_reg) { \ 10130 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 10131 offsetof(struct bpf_sock_ops_kern, \ 10132 temp)); \ 10133 fullsock_reg = reg; \ 10134 jmp += 2; \ 10135 } \ 10136 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10137 struct bpf_sock_ops_kern, \ 10138 is_fullsock), \ 10139 fullsock_reg, si->src_reg, \ 10140 offsetof(struct bpf_sock_ops_kern, \ 10141 is_fullsock)); \ 10142 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 10143 if (si->dst_reg == si->src_reg) \ 10144 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10145 offsetof(struct bpf_sock_ops_kern, \ 10146 temp)); \ 10147 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10148 struct bpf_sock_ops_kern, sk),\ 10149 si->dst_reg, si->src_reg, \ 10150 offsetof(struct bpf_sock_ops_kern, sk));\ 10151 if (si->dst_reg == si->src_reg) { \ 10152 *insn++ = BPF_JMP_A(1); \ 10153 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10154 offsetof(struct bpf_sock_ops_kern, \ 10155 temp)); \ 10156 } \ 10157 } while (0) 10158 10159 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \ 10160 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock) 10161 10162 /* Helper macro for adding write access to tcp_sock or sock fields. 10163 * The macro is called with two registers, dst_reg which contains a pointer 10164 * to ctx (context) and src_reg which contains the value that should be 10165 * stored. However, we need an additional register since we cannot overwrite 10166 * dst_reg because it may be used later in the program. 10167 * Instead we "borrow" one of the other register. We first save its value 10168 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore 10169 * it at the end of the macro. 10170 */ 10171 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 10172 do { \ 10173 int reg = BPF_REG_9; \ 10174 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 10175 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 10176 if (si->dst_reg == reg || si->src_reg == reg) \ 10177 reg--; \ 10178 if (si->dst_reg == reg || si->src_reg == reg) \ 10179 reg--; \ 10180 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \ 10181 offsetof(struct bpf_sock_ops_kern, \ 10182 temp)); \ 10183 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10184 struct bpf_sock_ops_kern, \ 10185 is_fullsock), \ 10186 reg, si->dst_reg, \ 10187 offsetof(struct bpf_sock_ops_kern, \ 10188 is_fullsock)); \ 10189 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \ 10190 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10191 struct bpf_sock_ops_kern, sk),\ 10192 reg, si->dst_reg, \ 10193 offsetof(struct bpf_sock_ops_kern, sk));\ 10194 *insn++ = BPF_RAW_INSN(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD) | \ 10195 BPF_MEM | BPF_CLASS(si->code), \ 10196 reg, si->src_reg, \ 10197 offsetof(OBJ, OBJ_FIELD), \ 10198 si->imm); \ 10199 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \ 10200 offsetof(struct bpf_sock_ops_kern, \ 10201 temp)); \ 10202 } while (0) 10203 10204 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \ 10205 do { \ 10206 if (TYPE == BPF_WRITE) \ 10207 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 10208 else \ 10209 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 10210 } while (0) 10211 10212 switch (si->off) { 10213 case offsetof(struct bpf_sock_ops, op): 10214 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10215 op), 10216 si->dst_reg, si->src_reg, 10217 offsetof(struct bpf_sock_ops_kern, op)); 10218 break; 10219 10220 case offsetof(struct bpf_sock_ops, replylong[0]) ... 10221 offsetof(struct bpf_sock_ops, replylong[3]): 10222 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) != 10223 sizeof_field(struct bpf_sock_ops_kern, reply)); 10224 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) != 10225 sizeof_field(struct bpf_sock_ops_kern, replylong)); 10226 off = si->off; 10227 off -= offsetof(struct bpf_sock_ops, replylong[0]); 10228 off += offsetof(struct bpf_sock_ops_kern, replylong[0]); 10229 if (type == BPF_WRITE) 10230 *insn++ = BPF_EMIT_STORE(BPF_W, si, off); 10231 else 10232 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10233 off); 10234 break; 10235 10236 case offsetof(struct bpf_sock_ops, family): 10237 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 10238 10239 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10240 struct bpf_sock_ops_kern, sk), 10241 si->dst_reg, si->src_reg, 10242 offsetof(struct bpf_sock_ops_kern, sk)); 10243 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10244 offsetof(struct sock_common, skc_family)); 10245 break; 10246 10247 case offsetof(struct bpf_sock_ops, remote_ip4): 10248 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 10249 10250 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10251 struct bpf_sock_ops_kern, sk), 10252 si->dst_reg, si->src_reg, 10253 offsetof(struct bpf_sock_ops_kern, sk)); 10254 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10255 offsetof(struct sock_common, skc_daddr)); 10256 break; 10257 10258 case offsetof(struct bpf_sock_ops, local_ip4): 10259 BUILD_BUG_ON(sizeof_field(struct sock_common, 10260 skc_rcv_saddr) != 4); 10261 10262 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10263 struct bpf_sock_ops_kern, sk), 10264 si->dst_reg, si->src_reg, 10265 offsetof(struct bpf_sock_ops_kern, sk)); 10266 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10267 offsetof(struct sock_common, 10268 skc_rcv_saddr)); 10269 break; 10270 10271 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ... 10272 offsetof(struct bpf_sock_ops, remote_ip6[3]): 10273 #if IS_ENABLED(CONFIG_IPV6) 10274 BUILD_BUG_ON(sizeof_field(struct sock_common, 10275 skc_v6_daddr.s6_addr32[0]) != 4); 10276 10277 off = si->off; 10278 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]); 10279 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10280 struct bpf_sock_ops_kern, sk), 10281 si->dst_reg, si->src_reg, 10282 offsetof(struct bpf_sock_ops_kern, sk)); 10283 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10284 offsetof(struct sock_common, 10285 skc_v6_daddr.s6_addr32[0]) + 10286 off); 10287 #else 10288 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10289 #endif 10290 break; 10291 10292 case offsetof(struct bpf_sock_ops, local_ip6[0]) ... 10293 offsetof(struct bpf_sock_ops, local_ip6[3]): 10294 #if IS_ENABLED(CONFIG_IPV6) 10295 BUILD_BUG_ON(sizeof_field(struct sock_common, 10296 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 10297 10298 off = si->off; 10299 off -= offsetof(struct bpf_sock_ops, local_ip6[0]); 10300 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10301 struct bpf_sock_ops_kern, sk), 10302 si->dst_reg, si->src_reg, 10303 offsetof(struct bpf_sock_ops_kern, sk)); 10304 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10305 offsetof(struct sock_common, 10306 skc_v6_rcv_saddr.s6_addr32[0]) + 10307 off); 10308 #else 10309 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10310 #endif 10311 break; 10312 10313 case offsetof(struct bpf_sock_ops, remote_port): 10314 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 10315 10316 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10317 struct bpf_sock_ops_kern, sk), 10318 si->dst_reg, si->src_reg, 10319 offsetof(struct bpf_sock_ops_kern, sk)); 10320 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10321 offsetof(struct sock_common, skc_dport)); 10322 #ifndef __BIG_ENDIAN_BITFIELD 10323 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 10324 #endif 10325 break; 10326 10327 case offsetof(struct bpf_sock_ops, local_port): 10328 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 10329 10330 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10331 struct bpf_sock_ops_kern, sk), 10332 si->dst_reg, si->src_reg, 10333 offsetof(struct bpf_sock_ops_kern, sk)); 10334 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10335 offsetof(struct sock_common, skc_num)); 10336 break; 10337 10338 case offsetof(struct bpf_sock_ops, is_fullsock): 10339 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10340 struct bpf_sock_ops_kern, 10341 is_fullsock), 10342 si->dst_reg, si->src_reg, 10343 offsetof(struct bpf_sock_ops_kern, 10344 is_fullsock)); 10345 break; 10346 10347 case offsetof(struct bpf_sock_ops, state): 10348 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1); 10349 10350 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10351 struct bpf_sock_ops_kern, sk), 10352 si->dst_reg, si->src_reg, 10353 offsetof(struct bpf_sock_ops_kern, sk)); 10354 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg, 10355 offsetof(struct sock_common, skc_state)); 10356 break; 10357 10358 case offsetof(struct bpf_sock_ops, rtt_min): 10359 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 10360 sizeof(struct minmax)); 10361 BUILD_BUG_ON(sizeof(struct minmax) < 10362 sizeof(struct minmax_sample)); 10363 10364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10365 struct bpf_sock_ops_kern, sk), 10366 si->dst_reg, si->src_reg, 10367 offsetof(struct bpf_sock_ops_kern, sk)); 10368 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10369 offsetof(struct tcp_sock, rtt_min) + 10370 sizeof_field(struct minmax_sample, t)); 10371 break; 10372 10373 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags): 10374 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags, 10375 struct tcp_sock); 10376 break; 10377 10378 case offsetof(struct bpf_sock_ops, sk_txhash): 10379 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash, 10380 struct sock, type); 10381 break; 10382 case offsetof(struct bpf_sock_ops, snd_cwnd): 10383 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd); 10384 break; 10385 case offsetof(struct bpf_sock_ops, srtt_us): 10386 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us); 10387 break; 10388 case offsetof(struct bpf_sock_ops, snd_ssthresh): 10389 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh); 10390 break; 10391 case offsetof(struct bpf_sock_ops, rcv_nxt): 10392 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt); 10393 break; 10394 case offsetof(struct bpf_sock_ops, snd_nxt): 10395 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt); 10396 break; 10397 case offsetof(struct bpf_sock_ops, snd_una): 10398 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una); 10399 break; 10400 case offsetof(struct bpf_sock_ops, mss_cache): 10401 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache); 10402 break; 10403 case offsetof(struct bpf_sock_ops, ecn_flags): 10404 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags); 10405 break; 10406 case offsetof(struct bpf_sock_ops, rate_delivered): 10407 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered); 10408 break; 10409 case offsetof(struct bpf_sock_ops, rate_interval_us): 10410 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us); 10411 break; 10412 case offsetof(struct bpf_sock_ops, packets_out): 10413 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out); 10414 break; 10415 case offsetof(struct bpf_sock_ops, retrans_out): 10416 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out); 10417 break; 10418 case offsetof(struct bpf_sock_ops, total_retrans): 10419 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans); 10420 break; 10421 case offsetof(struct bpf_sock_ops, segs_in): 10422 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in); 10423 break; 10424 case offsetof(struct bpf_sock_ops, data_segs_in): 10425 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in); 10426 break; 10427 case offsetof(struct bpf_sock_ops, segs_out): 10428 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out); 10429 break; 10430 case offsetof(struct bpf_sock_ops, data_segs_out): 10431 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out); 10432 break; 10433 case offsetof(struct bpf_sock_ops, lost_out): 10434 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out); 10435 break; 10436 case offsetof(struct bpf_sock_ops, sacked_out): 10437 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out); 10438 break; 10439 case offsetof(struct bpf_sock_ops, bytes_received): 10440 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received); 10441 break; 10442 case offsetof(struct bpf_sock_ops, bytes_acked): 10443 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked); 10444 break; 10445 case offsetof(struct bpf_sock_ops, sk): 10446 SOCK_OPS_GET_SK(); 10447 break; 10448 case offsetof(struct bpf_sock_ops, skb_data_end): 10449 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10450 skb_data_end), 10451 si->dst_reg, si->src_reg, 10452 offsetof(struct bpf_sock_ops_kern, 10453 skb_data_end)); 10454 break; 10455 case offsetof(struct bpf_sock_ops, skb_data): 10456 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10457 skb), 10458 si->dst_reg, si->src_reg, 10459 offsetof(struct bpf_sock_ops_kern, 10460 skb)); 10461 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10462 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 10463 si->dst_reg, si->dst_reg, 10464 offsetof(struct sk_buff, data)); 10465 break; 10466 case offsetof(struct bpf_sock_ops, skb_len): 10467 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10468 skb), 10469 si->dst_reg, si->src_reg, 10470 offsetof(struct bpf_sock_ops_kern, 10471 skb)); 10472 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10473 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 10474 si->dst_reg, si->dst_reg, 10475 offsetof(struct sk_buff, len)); 10476 break; 10477 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 10478 off = offsetof(struct sk_buff, cb); 10479 off += offsetof(struct tcp_skb_cb, tcp_flags); 10480 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags); 10481 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10482 skb), 10483 si->dst_reg, si->src_reg, 10484 offsetof(struct bpf_sock_ops_kern, 10485 skb)); 10486 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10487 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb, 10488 tcp_flags), 10489 si->dst_reg, si->dst_reg, off); 10490 break; 10491 case offsetof(struct bpf_sock_ops, skb_hwtstamp): { 10492 struct bpf_insn *jmp_on_null_skb; 10493 10494 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10495 skb), 10496 si->dst_reg, si->src_reg, 10497 offsetof(struct bpf_sock_ops_kern, 10498 skb)); 10499 /* Reserve one insn to test skb == NULL */ 10500 jmp_on_null_skb = insn++; 10501 insn = bpf_convert_shinfo_access(si->dst_reg, si->dst_reg, insn); 10502 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, 10503 bpf_target_off(struct skb_shared_info, 10504 hwtstamps, 8, 10505 target_size)); 10506 *jmp_on_null_skb = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 10507 insn - jmp_on_null_skb - 1); 10508 break; 10509 } 10510 } 10511 return insn - insn_buf; 10512 } 10513 10514 /* data_end = skb->data + skb_headlen() */ 10515 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si, 10516 struct bpf_insn *insn) 10517 { 10518 int reg; 10519 int temp_reg_off = offsetof(struct sk_buff, cb) + 10520 offsetof(struct sk_skb_cb, temp_reg); 10521 10522 if (si->src_reg == si->dst_reg) { 10523 /* We need an extra register, choose and save a register. */ 10524 reg = BPF_REG_9; 10525 if (si->src_reg == reg || si->dst_reg == reg) 10526 reg--; 10527 if (si->src_reg == reg || si->dst_reg == reg) 10528 reg--; 10529 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off); 10530 } else { 10531 reg = si->dst_reg; 10532 } 10533 10534 /* reg = skb->data */ 10535 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 10536 reg, si->src_reg, 10537 offsetof(struct sk_buff, data)); 10538 /* AX = skb->len */ 10539 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 10540 BPF_REG_AX, si->src_reg, 10541 offsetof(struct sk_buff, len)); 10542 /* reg = skb->data + skb->len */ 10543 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX); 10544 /* AX = skb->data_len */ 10545 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len), 10546 BPF_REG_AX, si->src_reg, 10547 offsetof(struct sk_buff, data_len)); 10548 10549 /* reg = skb->data + skb->len - skb->data_len */ 10550 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX); 10551 10552 if (si->src_reg == si->dst_reg) { 10553 /* Restore the saved register */ 10554 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg); 10555 *insn++ = BPF_MOV64_REG(si->dst_reg, reg); 10556 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off); 10557 } 10558 10559 return insn; 10560 } 10561 10562 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type, 10563 const struct bpf_insn *si, 10564 struct bpf_insn *insn_buf, 10565 struct bpf_prog *prog, u32 *target_size) 10566 { 10567 struct bpf_insn *insn = insn_buf; 10568 int off; 10569 10570 switch (si->off) { 10571 case offsetof(struct __sk_buff, data_end): 10572 insn = bpf_convert_data_end_access(si, insn); 10573 break; 10574 case offsetof(struct __sk_buff, cb[0]) ... 10575 offsetofend(struct __sk_buff, cb[4]) - 1: 10576 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20); 10577 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 10578 offsetof(struct sk_skb_cb, data)) % 10579 sizeof(__u64)); 10580 10581 prog->cb_access = 1; 10582 off = si->off; 10583 off -= offsetof(struct __sk_buff, cb[0]); 10584 off += offsetof(struct sk_buff, cb); 10585 off += offsetof(struct sk_skb_cb, data); 10586 if (type == BPF_WRITE) 10587 *insn++ = BPF_EMIT_STORE(BPF_SIZE(si->code), si, off); 10588 else 10589 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 10590 si->src_reg, off); 10591 break; 10592 10593 10594 default: 10595 return bpf_convert_ctx_access(type, si, insn_buf, prog, 10596 target_size); 10597 } 10598 10599 return insn - insn_buf; 10600 } 10601 10602 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type, 10603 const struct bpf_insn *si, 10604 struct bpf_insn *insn_buf, 10605 struct bpf_prog *prog, u32 *target_size) 10606 { 10607 struct bpf_insn *insn = insn_buf; 10608 #if IS_ENABLED(CONFIG_IPV6) 10609 int off; 10610 #endif 10611 10612 /* convert ctx uses the fact sg element is first in struct */ 10613 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0); 10614 10615 switch (si->off) { 10616 case offsetof(struct sk_msg_md, data): 10617 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data), 10618 si->dst_reg, si->src_reg, 10619 offsetof(struct sk_msg, data)); 10620 break; 10621 case offsetof(struct sk_msg_md, data_end): 10622 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end), 10623 si->dst_reg, si->src_reg, 10624 offsetof(struct sk_msg, data_end)); 10625 break; 10626 case offsetof(struct sk_msg_md, family): 10627 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 10628 10629 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10630 struct sk_msg, sk), 10631 si->dst_reg, si->src_reg, 10632 offsetof(struct sk_msg, sk)); 10633 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10634 offsetof(struct sock_common, skc_family)); 10635 break; 10636 10637 case offsetof(struct sk_msg_md, remote_ip4): 10638 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 10639 10640 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10641 struct sk_msg, sk), 10642 si->dst_reg, si->src_reg, 10643 offsetof(struct sk_msg, sk)); 10644 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10645 offsetof(struct sock_common, skc_daddr)); 10646 break; 10647 10648 case offsetof(struct sk_msg_md, local_ip4): 10649 BUILD_BUG_ON(sizeof_field(struct sock_common, 10650 skc_rcv_saddr) != 4); 10651 10652 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10653 struct sk_msg, sk), 10654 si->dst_reg, si->src_reg, 10655 offsetof(struct sk_msg, sk)); 10656 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10657 offsetof(struct sock_common, 10658 skc_rcv_saddr)); 10659 break; 10660 10661 case offsetof(struct sk_msg_md, remote_ip6[0]) ... 10662 offsetof(struct sk_msg_md, remote_ip6[3]): 10663 #if IS_ENABLED(CONFIG_IPV6) 10664 BUILD_BUG_ON(sizeof_field(struct sock_common, 10665 skc_v6_daddr.s6_addr32[0]) != 4); 10666 10667 off = si->off; 10668 off -= offsetof(struct sk_msg_md, remote_ip6[0]); 10669 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10670 struct sk_msg, sk), 10671 si->dst_reg, si->src_reg, 10672 offsetof(struct sk_msg, sk)); 10673 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10674 offsetof(struct sock_common, 10675 skc_v6_daddr.s6_addr32[0]) + 10676 off); 10677 #else 10678 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10679 #endif 10680 break; 10681 10682 case offsetof(struct sk_msg_md, local_ip6[0]) ... 10683 offsetof(struct sk_msg_md, local_ip6[3]): 10684 #if IS_ENABLED(CONFIG_IPV6) 10685 BUILD_BUG_ON(sizeof_field(struct sock_common, 10686 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 10687 10688 off = si->off; 10689 off -= offsetof(struct sk_msg_md, local_ip6[0]); 10690 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10691 struct sk_msg, sk), 10692 si->dst_reg, si->src_reg, 10693 offsetof(struct sk_msg, sk)); 10694 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10695 offsetof(struct sock_common, 10696 skc_v6_rcv_saddr.s6_addr32[0]) + 10697 off); 10698 #else 10699 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10700 #endif 10701 break; 10702 10703 case offsetof(struct sk_msg_md, remote_port): 10704 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 10705 10706 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10707 struct sk_msg, sk), 10708 si->dst_reg, si->src_reg, 10709 offsetof(struct sk_msg, sk)); 10710 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10711 offsetof(struct sock_common, skc_dport)); 10712 #ifndef __BIG_ENDIAN_BITFIELD 10713 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 10714 #endif 10715 break; 10716 10717 case offsetof(struct sk_msg_md, local_port): 10718 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 10719 10720 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10721 struct sk_msg, sk), 10722 si->dst_reg, si->src_reg, 10723 offsetof(struct sk_msg, sk)); 10724 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10725 offsetof(struct sock_common, skc_num)); 10726 break; 10727 10728 case offsetof(struct sk_msg_md, size): 10729 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size), 10730 si->dst_reg, si->src_reg, 10731 offsetof(struct sk_msg_sg, size)); 10732 break; 10733 10734 case offsetof(struct sk_msg_md, sk): 10735 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk), 10736 si->dst_reg, si->src_reg, 10737 offsetof(struct sk_msg, sk)); 10738 break; 10739 } 10740 10741 return insn - insn_buf; 10742 } 10743 10744 const struct bpf_verifier_ops sk_filter_verifier_ops = { 10745 .get_func_proto = sk_filter_func_proto, 10746 .is_valid_access = sk_filter_is_valid_access, 10747 .convert_ctx_access = bpf_convert_ctx_access, 10748 .gen_ld_abs = bpf_gen_ld_abs, 10749 }; 10750 10751 const struct bpf_prog_ops sk_filter_prog_ops = { 10752 .test_run = bpf_prog_test_run_skb, 10753 }; 10754 10755 const struct bpf_verifier_ops tc_cls_act_verifier_ops = { 10756 .get_func_proto = tc_cls_act_func_proto, 10757 .is_valid_access = tc_cls_act_is_valid_access, 10758 .convert_ctx_access = tc_cls_act_convert_ctx_access, 10759 .gen_prologue = tc_cls_act_prologue, 10760 .gen_ld_abs = bpf_gen_ld_abs, 10761 .btf_struct_access = tc_cls_act_btf_struct_access, 10762 }; 10763 10764 const struct bpf_prog_ops tc_cls_act_prog_ops = { 10765 .test_run = bpf_prog_test_run_skb, 10766 }; 10767 10768 const struct bpf_verifier_ops xdp_verifier_ops = { 10769 .get_func_proto = xdp_func_proto, 10770 .is_valid_access = xdp_is_valid_access, 10771 .convert_ctx_access = xdp_convert_ctx_access, 10772 .gen_prologue = bpf_noop_prologue, 10773 .btf_struct_access = xdp_btf_struct_access, 10774 }; 10775 10776 const struct bpf_prog_ops xdp_prog_ops = { 10777 .test_run = bpf_prog_test_run_xdp, 10778 }; 10779 10780 const struct bpf_verifier_ops cg_skb_verifier_ops = { 10781 .get_func_proto = cg_skb_func_proto, 10782 .is_valid_access = cg_skb_is_valid_access, 10783 .convert_ctx_access = bpf_convert_ctx_access, 10784 }; 10785 10786 const struct bpf_prog_ops cg_skb_prog_ops = { 10787 .test_run = bpf_prog_test_run_skb, 10788 }; 10789 10790 const struct bpf_verifier_ops lwt_in_verifier_ops = { 10791 .get_func_proto = lwt_in_func_proto, 10792 .is_valid_access = lwt_is_valid_access, 10793 .convert_ctx_access = bpf_convert_ctx_access, 10794 }; 10795 10796 const struct bpf_prog_ops lwt_in_prog_ops = { 10797 .test_run = bpf_prog_test_run_skb, 10798 }; 10799 10800 const struct bpf_verifier_ops lwt_out_verifier_ops = { 10801 .get_func_proto = lwt_out_func_proto, 10802 .is_valid_access = lwt_is_valid_access, 10803 .convert_ctx_access = bpf_convert_ctx_access, 10804 }; 10805 10806 const struct bpf_prog_ops lwt_out_prog_ops = { 10807 .test_run = bpf_prog_test_run_skb, 10808 }; 10809 10810 const struct bpf_verifier_ops lwt_xmit_verifier_ops = { 10811 .get_func_proto = lwt_xmit_func_proto, 10812 .is_valid_access = lwt_is_valid_access, 10813 .convert_ctx_access = bpf_convert_ctx_access, 10814 .gen_prologue = tc_cls_act_prologue, 10815 }; 10816 10817 const struct bpf_prog_ops lwt_xmit_prog_ops = { 10818 .test_run = bpf_prog_test_run_skb, 10819 }; 10820 10821 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = { 10822 .get_func_proto = lwt_seg6local_func_proto, 10823 .is_valid_access = lwt_is_valid_access, 10824 .convert_ctx_access = bpf_convert_ctx_access, 10825 }; 10826 10827 const struct bpf_prog_ops lwt_seg6local_prog_ops = { 10828 .test_run = bpf_prog_test_run_skb, 10829 }; 10830 10831 const struct bpf_verifier_ops cg_sock_verifier_ops = { 10832 .get_func_proto = sock_filter_func_proto, 10833 .is_valid_access = sock_filter_is_valid_access, 10834 .convert_ctx_access = bpf_sock_convert_ctx_access, 10835 }; 10836 10837 const struct bpf_prog_ops cg_sock_prog_ops = { 10838 }; 10839 10840 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = { 10841 .get_func_proto = sock_addr_func_proto, 10842 .is_valid_access = sock_addr_is_valid_access, 10843 .convert_ctx_access = sock_addr_convert_ctx_access, 10844 }; 10845 10846 const struct bpf_prog_ops cg_sock_addr_prog_ops = { 10847 }; 10848 10849 const struct bpf_verifier_ops sock_ops_verifier_ops = { 10850 .get_func_proto = sock_ops_func_proto, 10851 .is_valid_access = sock_ops_is_valid_access, 10852 .convert_ctx_access = sock_ops_convert_ctx_access, 10853 }; 10854 10855 const struct bpf_prog_ops sock_ops_prog_ops = { 10856 }; 10857 10858 const struct bpf_verifier_ops sk_skb_verifier_ops = { 10859 .get_func_proto = sk_skb_func_proto, 10860 .is_valid_access = sk_skb_is_valid_access, 10861 .convert_ctx_access = sk_skb_convert_ctx_access, 10862 .gen_prologue = sk_skb_prologue, 10863 }; 10864 10865 const struct bpf_prog_ops sk_skb_prog_ops = { 10866 }; 10867 10868 const struct bpf_verifier_ops sk_msg_verifier_ops = { 10869 .get_func_proto = sk_msg_func_proto, 10870 .is_valid_access = sk_msg_is_valid_access, 10871 .convert_ctx_access = sk_msg_convert_ctx_access, 10872 .gen_prologue = bpf_noop_prologue, 10873 }; 10874 10875 const struct bpf_prog_ops sk_msg_prog_ops = { 10876 }; 10877 10878 const struct bpf_verifier_ops flow_dissector_verifier_ops = { 10879 .get_func_proto = flow_dissector_func_proto, 10880 .is_valid_access = flow_dissector_is_valid_access, 10881 .convert_ctx_access = flow_dissector_convert_ctx_access, 10882 }; 10883 10884 const struct bpf_prog_ops flow_dissector_prog_ops = { 10885 .test_run = bpf_prog_test_run_flow_dissector, 10886 }; 10887 10888 int sk_detach_filter(struct sock *sk) 10889 { 10890 int ret = -ENOENT; 10891 struct sk_filter *filter; 10892 10893 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 10894 return -EPERM; 10895 10896 filter = rcu_dereference_protected(sk->sk_filter, 10897 lockdep_sock_is_held(sk)); 10898 if (filter) { 10899 RCU_INIT_POINTER(sk->sk_filter, NULL); 10900 sk_filter_uncharge(sk, filter); 10901 ret = 0; 10902 } 10903 10904 return ret; 10905 } 10906 EXPORT_SYMBOL_GPL(sk_detach_filter); 10907 10908 int sk_get_filter(struct sock *sk, sockptr_t optval, unsigned int len) 10909 { 10910 struct sock_fprog_kern *fprog; 10911 struct sk_filter *filter; 10912 int ret = 0; 10913 10914 sockopt_lock_sock(sk); 10915 filter = rcu_dereference_protected(sk->sk_filter, 10916 lockdep_sock_is_held(sk)); 10917 if (!filter) 10918 goto out; 10919 10920 /* We're copying the filter that has been originally attached, 10921 * so no conversion/decode needed anymore. eBPF programs that 10922 * have no original program cannot be dumped through this. 10923 */ 10924 ret = -EACCES; 10925 fprog = filter->prog->orig_prog; 10926 if (!fprog) 10927 goto out; 10928 10929 ret = fprog->len; 10930 if (!len) 10931 /* User space only enquires number of filter blocks. */ 10932 goto out; 10933 10934 ret = -EINVAL; 10935 if (len < fprog->len) 10936 goto out; 10937 10938 ret = -EFAULT; 10939 if (copy_to_sockptr(optval, fprog->filter, bpf_classic_proglen(fprog))) 10940 goto out; 10941 10942 /* Instead of bytes, the API requests to return the number 10943 * of filter blocks. 10944 */ 10945 ret = fprog->len; 10946 out: 10947 sockopt_release_sock(sk); 10948 return ret; 10949 } 10950 10951 #ifdef CONFIG_INET 10952 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern, 10953 struct sock_reuseport *reuse, 10954 struct sock *sk, struct sk_buff *skb, 10955 struct sock *migrating_sk, 10956 u32 hash) 10957 { 10958 reuse_kern->skb = skb; 10959 reuse_kern->sk = sk; 10960 reuse_kern->selected_sk = NULL; 10961 reuse_kern->migrating_sk = migrating_sk; 10962 reuse_kern->data_end = skb->data + skb_headlen(skb); 10963 reuse_kern->hash = hash; 10964 reuse_kern->reuseport_id = reuse->reuseport_id; 10965 reuse_kern->bind_inany = reuse->bind_inany; 10966 } 10967 10968 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 10969 struct bpf_prog *prog, struct sk_buff *skb, 10970 struct sock *migrating_sk, 10971 u32 hash) 10972 { 10973 struct sk_reuseport_kern reuse_kern; 10974 enum sk_action action; 10975 10976 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash); 10977 action = bpf_prog_run(prog, &reuse_kern); 10978 10979 if (action == SK_PASS) 10980 return reuse_kern.selected_sk; 10981 else 10982 return ERR_PTR(-ECONNREFUSED); 10983 } 10984 10985 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern, 10986 struct bpf_map *, map, void *, key, u32, flags) 10987 { 10988 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY; 10989 struct sock_reuseport *reuse; 10990 struct sock *selected_sk; 10991 10992 selected_sk = map->ops->map_lookup_elem(map, key); 10993 if (!selected_sk) 10994 return -ENOENT; 10995 10996 reuse = rcu_dereference(selected_sk->sk_reuseport_cb); 10997 if (!reuse) { 10998 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */ 10999 if (sk_is_refcounted(selected_sk)) 11000 sock_put(selected_sk); 11001 11002 /* reuseport_array has only sk with non NULL sk_reuseport_cb. 11003 * The only (!reuse) case here is - the sk has already been 11004 * unhashed (e.g. by close()), so treat it as -ENOENT. 11005 * 11006 * Other maps (e.g. sock_map) do not provide this guarantee and 11007 * the sk may never be in the reuseport group to begin with. 11008 */ 11009 return is_sockarray ? -ENOENT : -EINVAL; 11010 } 11011 11012 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) { 11013 struct sock *sk = reuse_kern->sk; 11014 11015 if (sk->sk_protocol != selected_sk->sk_protocol) 11016 return -EPROTOTYPE; 11017 else if (sk->sk_family != selected_sk->sk_family) 11018 return -EAFNOSUPPORT; 11019 11020 /* Catch all. Likely bound to a different sockaddr. */ 11021 return -EBADFD; 11022 } 11023 11024 reuse_kern->selected_sk = selected_sk; 11025 11026 return 0; 11027 } 11028 11029 static const struct bpf_func_proto sk_select_reuseport_proto = { 11030 .func = sk_select_reuseport, 11031 .gpl_only = false, 11032 .ret_type = RET_INTEGER, 11033 .arg1_type = ARG_PTR_TO_CTX, 11034 .arg2_type = ARG_CONST_MAP_PTR, 11035 .arg3_type = ARG_PTR_TO_MAP_KEY, 11036 .arg4_type = ARG_ANYTHING, 11037 }; 11038 11039 BPF_CALL_4(sk_reuseport_load_bytes, 11040 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 11041 void *, to, u32, len) 11042 { 11043 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len); 11044 } 11045 11046 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = { 11047 .func = sk_reuseport_load_bytes, 11048 .gpl_only = false, 11049 .ret_type = RET_INTEGER, 11050 .arg1_type = ARG_PTR_TO_CTX, 11051 .arg2_type = ARG_ANYTHING, 11052 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 11053 .arg4_type = ARG_CONST_SIZE, 11054 }; 11055 11056 BPF_CALL_5(sk_reuseport_load_bytes_relative, 11057 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 11058 void *, to, u32, len, u32, start_header) 11059 { 11060 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to, 11061 len, start_header); 11062 } 11063 11064 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = { 11065 .func = sk_reuseport_load_bytes_relative, 11066 .gpl_only = false, 11067 .ret_type = RET_INTEGER, 11068 .arg1_type = ARG_PTR_TO_CTX, 11069 .arg2_type = ARG_ANYTHING, 11070 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 11071 .arg4_type = ARG_CONST_SIZE, 11072 .arg5_type = ARG_ANYTHING, 11073 }; 11074 11075 static const struct bpf_func_proto * 11076 sk_reuseport_func_proto(enum bpf_func_id func_id, 11077 const struct bpf_prog *prog) 11078 { 11079 switch (func_id) { 11080 case BPF_FUNC_sk_select_reuseport: 11081 return &sk_select_reuseport_proto; 11082 case BPF_FUNC_skb_load_bytes: 11083 return &sk_reuseport_load_bytes_proto; 11084 case BPF_FUNC_skb_load_bytes_relative: 11085 return &sk_reuseport_load_bytes_relative_proto; 11086 case BPF_FUNC_get_socket_cookie: 11087 return &bpf_get_socket_ptr_cookie_proto; 11088 case BPF_FUNC_ktime_get_coarse_ns: 11089 return &bpf_ktime_get_coarse_ns_proto; 11090 default: 11091 return bpf_base_func_proto(func_id); 11092 } 11093 } 11094 11095 static bool 11096 sk_reuseport_is_valid_access(int off, int size, 11097 enum bpf_access_type type, 11098 const struct bpf_prog *prog, 11099 struct bpf_insn_access_aux *info) 11100 { 11101 const u32 size_default = sizeof(__u32); 11102 11103 if (off < 0 || off >= sizeof(struct sk_reuseport_md) || 11104 off % size || type != BPF_READ) 11105 return false; 11106 11107 switch (off) { 11108 case offsetof(struct sk_reuseport_md, data): 11109 info->reg_type = PTR_TO_PACKET; 11110 return size == sizeof(__u64); 11111 11112 case offsetof(struct sk_reuseport_md, data_end): 11113 info->reg_type = PTR_TO_PACKET_END; 11114 return size == sizeof(__u64); 11115 11116 case offsetof(struct sk_reuseport_md, hash): 11117 return size == size_default; 11118 11119 case offsetof(struct sk_reuseport_md, sk): 11120 info->reg_type = PTR_TO_SOCKET; 11121 return size == sizeof(__u64); 11122 11123 case offsetof(struct sk_reuseport_md, migrating_sk): 11124 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 11125 return size == sizeof(__u64); 11126 11127 /* Fields that allow narrowing */ 11128 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol): 11129 if (size < sizeof_field(struct sk_buff, protocol)) 11130 return false; 11131 fallthrough; 11132 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol): 11133 case bpf_ctx_range(struct sk_reuseport_md, bind_inany): 11134 case bpf_ctx_range(struct sk_reuseport_md, len): 11135 bpf_ctx_record_field_size(info, size_default); 11136 return bpf_ctx_narrow_access_ok(off, size, size_default); 11137 11138 default: 11139 return false; 11140 } 11141 } 11142 11143 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \ 11144 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \ 11145 si->dst_reg, si->src_reg, \ 11146 bpf_target_off(struct sk_reuseport_kern, F, \ 11147 sizeof_field(struct sk_reuseport_kern, F), \ 11148 target_size)); \ 11149 }) 11150 11151 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \ 11152 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 11153 struct sk_buff, \ 11154 skb, \ 11155 SKB_FIELD) 11156 11157 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \ 11158 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 11159 struct sock, \ 11160 sk, \ 11161 SK_FIELD) 11162 11163 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type, 11164 const struct bpf_insn *si, 11165 struct bpf_insn *insn_buf, 11166 struct bpf_prog *prog, 11167 u32 *target_size) 11168 { 11169 struct bpf_insn *insn = insn_buf; 11170 11171 switch (si->off) { 11172 case offsetof(struct sk_reuseport_md, data): 11173 SK_REUSEPORT_LOAD_SKB_FIELD(data); 11174 break; 11175 11176 case offsetof(struct sk_reuseport_md, len): 11177 SK_REUSEPORT_LOAD_SKB_FIELD(len); 11178 break; 11179 11180 case offsetof(struct sk_reuseport_md, eth_protocol): 11181 SK_REUSEPORT_LOAD_SKB_FIELD(protocol); 11182 break; 11183 11184 case offsetof(struct sk_reuseport_md, ip_protocol): 11185 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol); 11186 break; 11187 11188 case offsetof(struct sk_reuseport_md, data_end): 11189 SK_REUSEPORT_LOAD_FIELD(data_end); 11190 break; 11191 11192 case offsetof(struct sk_reuseport_md, hash): 11193 SK_REUSEPORT_LOAD_FIELD(hash); 11194 break; 11195 11196 case offsetof(struct sk_reuseport_md, bind_inany): 11197 SK_REUSEPORT_LOAD_FIELD(bind_inany); 11198 break; 11199 11200 case offsetof(struct sk_reuseport_md, sk): 11201 SK_REUSEPORT_LOAD_FIELD(sk); 11202 break; 11203 11204 case offsetof(struct sk_reuseport_md, migrating_sk): 11205 SK_REUSEPORT_LOAD_FIELD(migrating_sk); 11206 break; 11207 } 11208 11209 return insn - insn_buf; 11210 } 11211 11212 const struct bpf_verifier_ops sk_reuseport_verifier_ops = { 11213 .get_func_proto = sk_reuseport_func_proto, 11214 .is_valid_access = sk_reuseport_is_valid_access, 11215 .convert_ctx_access = sk_reuseport_convert_ctx_access, 11216 }; 11217 11218 const struct bpf_prog_ops sk_reuseport_prog_ops = { 11219 }; 11220 11221 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled); 11222 EXPORT_SYMBOL(bpf_sk_lookup_enabled); 11223 11224 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx, 11225 struct sock *, sk, u64, flags) 11226 { 11227 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE | 11228 BPF_SK_LOOKUP_F_NO_REUSEPORT))) 11229 return -EINVAL; 11230 if (unlikely(sk && sk_is_refcounted(sk))) 11231 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */ 11232 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN)) 11233 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */ 11234 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE)) 11235 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */ 11236 11237 /* Check if socket is suitable for packet L3/L4 protocol */ 11238 if (sk && sk->sk_protocol != ctx->protocol) 11239 return -EPROTOTYPE; 11240 if (sk && sk->sk_family != ctx->family && 11241 (sk->sk_family == AF_INET || ipv6_only_sock(sk))) 11242 return -EAFNOSUPPORT; 11243 11244 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE)) 11245 return -EEXIST; 11246 11247 /* Select socket as lookup result */ 11248 ctx->selected_sk = sk; 11249 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT; 11250 return 0; 11251 } 11252 11253 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = { 11254 .func = bpf_sk_lookup_assign, 11255 .gpl_only = false, 11256 .ret_type = RET_INTEGER, 11257 .arg1_type = ARG_PTR_TO_CTX, 11258 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL, 11259 .arg3_type = ARG_ANYTHING, 11260 }; 11261 11262 static const struct bpf_func_proto * 11263 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 11264 { 11265 switch (func_id) { 11266 case BPF_FUNC_perf_event_output: 11267 return &bpf_event_output_data_proto; 11268 case BPF_FUNC_sk_assign: 11269 return &bpf_sk_lookup_assign_proto; 11270 case BPF_FUNC_sk_release: 11271 return &bpf_sk_release_proto; 11272 default: 11273 return bpf_sk_base_func_proto(func_id); 11274 } 11275 } 11276 11277 static bool sk_lookup_is_valid_access(int off, int size, 11278 enum bpf_access_type type, 11279 const struct bpf_prog *prog, 11280 struct bpf_insn_access_aux *info) 11281 { 11282 if (off < 0 || off >= sizeof(struct bpf_sk_lookup)) 11283 return false; 11284 if (off % size != 0) 11285 return false; 11286 if (type != BPF_READ) 11287 return false; 11288 11289 switch (off) { 11290 case offsetof(struct bpf_sk_lookup, sk): 11291 info->reg_type = PTR_TO_SOCKET_OR_NULL; 11292 return size == sizeof(__u64); 11293 11294 case bpf_ctx_range(struct bpf_sk_lookup, family): 11295 case bpf_ctx_range(struct bpf_sk_lookup, protocol): 11296 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4): 11297 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4): 11298 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]): 11299 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]): 11300 case bpf_ctx_range(struct bpf_sk_lookup, local_port): 11301 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex): 11302 bpf_ctx_record_field_size(info, sizeof(__u32)); 11303 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32)); 11304 11305 case bpf_ctx_range(struct bpf_sk_lookup, remote_port): 11306 /* Allow 4-byte access to 2-byte field for backward compatibility */ 11307 if (size == sizeof(__u32)) 11308 return true; 11309 bpf_ctx_record_field_size(info, sizeof(__be16)); 11310 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16)); 11311 11312 case offsetofend(struct bpf_sk_lookup, remote_port) ... 11313 offsetof(struct bpf_sk_lookup, local_ip4) - 1: 11314 /* Allow access to zero padding for backward compatibility */ 11315 bpf_ctx_record_field_size(info, sizeof(__u16)); 11316 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16)); 11317 11318 default: 11319 return false; 11320 } 11321 } 11322 11323 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type, 11324 const struct bpf_insn *si, 11325 struct bpf_insn *insn_buf, 11326 struct bpf_prog *prog, 11327 u32 *target_size) 11328 { 11329 struct bpf_insn *insn = insn_buf; 11330 11331 switch (si->off) { 11332 case offsetof(struct bpf_sk_lookup, sk): 11333 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11334 offsetof(struct bpf_sk_lookup_kern, selected_sk)); 11335 break; 11336 11337 case offsetof(struct bpf_sk_lookup, family): 11338 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11339 bpf_target_off(struct bpf_sk_lookup_kern, 11340 family, 2, target_size)); 11341 break; 11342 11343 case offsetof(struct bpf_sk_lookup, protocol): 11344 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11345 bpf_target_off(struct bpf_sk_lookup_kern, 11346 protocol, 2, target_size)); 11347 break; 11348 11349 case offsetof(struct bpf_sk_lookup, remote_ip4): 11350 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11351 bpf_target_off(struct bpf_sk_lookup_kern, 11352 v4.saddr, 4, target_size)); 11353 break; 11354 11355 case offsetof(struct bpf_sk_lookup, local_ip4): 11356 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11357 bpf_target_off(struct bpf_sk_lookup_kern, 11358 v4.daddr, 4, target_size)); 11359 break; 11360 11361 case bpf_ctx_range_till(struct bpf_sk_lookup, 11362 remote_ip6[0], remote_ip6[3]): { 11363 #if IS_ENABLED(CONFIG_IPV6) 11364 int off = si->off; 11365 11366 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]); 11367 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 11368 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11369 offsetof(struct bpf_sk_lookup_kern, v6.saddr)); 11370 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11371 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 11372 #else 11373 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11374 #endif 11375 break; 11376 } 11377 case bpf_ctx_range_till(struct bpf_sk_lookup, 11378 local_ip6[0], local_ip6[3]): { 11379 #if IS_ENABLED(CONFIG_IPV6) 11380 int off = si->off; 11381 11382 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]); 11383 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 11384 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11385 offsetof(struct bpf_sk_lookup_kern, v6.daddr)); 11386 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11387 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 11388 #else 11389 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11390 #endif 11391 break; 11392 } 11393 case offsetof(struct bpf_sk_lookup, remote_port): 11394 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11395 bpf_target_off(struct bpf_sk_lookup_kern, 11396 sport, 2, target_size)); 11397 break; 11398 11399 case offsetofend(struct bpf_sk_lookup, remote_port): 11400 *target_size = 2; 11401 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11402 break; 11403 11404 case offsetof(struct bpf_sk_lookup, local_port): 11405 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11406 bpf_target_off(struct bpf_sk_lookup_kern, 11407 dport, 2, target_size)); 11408 break; 11409 11410 case offsetof(struct bpf_sk_lookup, ingress_ifindex): 11411 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11412 bpf_target_off(struct bpf_sk_lookup_kern, 11413 ingress_ifindex, 4, target_size)); 11414 break; 11415 } 11416 11417 return insn - insn_buf; 11418 } 11419 11420 const struct bpf_prog_ops sk_lookup_prog_ops = { 11421 .test_run = bpf_prog_test_run_sk_lookup, 11422 }; 11423 11424 const struct bpf_verifier_ops sk_lookup_verifier_ops = { 11425 .get_func_proto = sk_lookup_func_proto, 11426 .is_valid_access = sk_lookup_is_valid_access, 11427 .convert_ctx_access = sk_lookup_convert_ctx_access, 11428 }; 11429 11430 #endif /* CONFIG_INET */ 11431 11432 DEFINE_BPF_DISPATCHER(xdp) 11433 11434 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog) 11435 { 11436 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog); 11437 } 11438 11439 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE) 11440 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type) 11441 BTF_SOCK_TYPE_xxx 11442 #undef BTF_SOCK_TYPE 11443 11444 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk) 11445 { 11446 /* tcp6_sock type is not generated in dwarf and hence btf, 11447 * trigger an explicit type generation here. 11448 */ 11449 BTF_TYPE_EMIT(struct tcp6_sock); 11450 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && 11451 sk->sk_family == AF_INET6) 11452 return (unsigned long)sk; 11453 11454 return (unsigned long)NULL; 11455 } 11456 11457 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = { 11458 .func = bpf_skc_to_tcp6_sock, 11459 .gpl_only = false, 11460 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11461 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11462 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6], 11463 }; 11464 11465 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk) 11466 { 11467 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 11468 return (unsigned long)sk; 11469 11470 return (unsigned long)NULL; 11471 } 11472 11473 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = { 11474 .func = bpf_skc_to_tcp_sock, 11475 .gpl_only = false, 11476 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11477 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11478 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP], 11479 }; 11480 11481 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk) 11482 { 11483 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not 11484 * generated if CONFIG_INET=n. Trigger an explicit generation here. 11485 */ 11486 BTF_TYPE_EMIT(struct inet_timewait_sock); 11487 BTF_TYPE_EMIT(struct tcp_timewait_sock); 11488 11489 #ifdef CONFIG_INET 11490 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT) 11491 return (unsigned long)sk; 11492 #endif 11493 11494 #if IS_BUILTIN(CONFIG_IPV6) 11495 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT) 11496 return (unsigned long)sk; 11497 #endif 11498 11499 return (unsigned long)NULL; 11500 } 11501 11502 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = { 11503 .func = bpf_skc_to_tcp_timewait_sock, 11504 .gpl_only = false, 11505 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11506 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11507 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW], 11508 }; 11509 11510 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk) 11511 { 11512 #ifdef CONFIG_INET 11513 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV) 11514 return (unsigned long)sk; 11515 #endif 11516 11517 #if IS_BUILTIN(CONFIG_IPV6) 11518 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV) 11519 return (unsigned long)sk; 11520 #endif 11521 11522 return (unsigned long)NULL; 11523 } 11524 11525 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = { 11526 .func = bpf_skc_to_tcp_request_sock, 11527 .gpl_only = false, 11528 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11529 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11530 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ], 11531 }; 11532 11533 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk) 11534 { 11535 /* udp6_sock type is not generated in dwarf and hence btf, 11536 * trigger an explicit type generation here. 11537 */ 11538 BTF_TYPE_EMIT(struct udp6_sock); 11539 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP && 11540 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6) 11541 return (unsigned long)sk; 11542 11543 return (unsigned long)NULL; 11544 } 11545 11546 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = { 11547 .func = bpf_skc_to_udp6_sock, 11548 .gpl_only = false, 11549 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11550 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11551 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6], 11552 }; 11553 11554 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk) 11555 { 11556 /* unix_sock type is not generated in dwarf and hence btf, 11557 * trigger an explicit type generation here. 11558 */ 11559 BTF_TYPE_EMIT(struct unix_sock); 11560 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX) 11561 return (unsigned long)sk; 11562 11563 return (unsigned long)NULL; 11564 } 11565 11566 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = { 11567 .func = bpf_skc_to_unix_sock, 11568 .gpl_only = false, 11569 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11570 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11571 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX], 11572 }; 11573 11574 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk) 11575 { 11576 BTF_TYPE_EMIT(struct mptcp_sock); 11577 return (unsigned long)bpf_mptcp_sock_from_subflow(sk); 11578 } 11579 11580 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = { 11581 .func = bpf_skc_to_mptcp_sock, 11582 .gpl_only = false, 11583 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11584 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 11585 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP], 11586 }; 11587 11588 BPF_CALL_1(bpf_sock_from_file, struct file *, file) 11589 { 11590 return (unsigned long)sock_from_file(file); 11591 } 11592 11593 BTF_ID_LIST(bpf_sock_from_file_btf_ids) 11594 BTF_ID(struct, socket) 11595 BTF_ID(struct, file) 11596 11597 const struct bpf_func_proto bpf_sock_from_file_proto = { 11598 .func = bpf_sock_from_file, 11599 .gpl_only = false, 11600 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11601 .ret_btf_id = &bpf_sock_from_file_btf_ids[0], 11602 .arg1_type = ARG_PTR_TO_BTF_ID, 11603 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1], 11604 }; 11605 11606 static const struct bpf_func_proto * 11607 bpf_sk_base_func_proto(enum bpf_func_id func_id) 11608 { 11609 const struct bpf_func_proto *func; 11610 11611 switch (func_id) { 11612 case BPF_FUNC_skc_to_tcp6_sock: 11613 func = &bpf_skc_to_tcp6_sock_proto; 11614 break; 11615 case BPF_FUNC_skc_to_tcp_sock: 11616 func = &bpf_skc_to_tcp_sock_proto; 11617 break; 11618 case BPF_FUNC_skc_to_tcp_timewait_sock: 11619 func = &bpf_skc_to_tcp_timewait_sock_proto; 11620 break; 11621 case BPF_FUNC_skc_to_tcp_request_sock: 11622 func = &bpf_skc_to_tcp_request_sock_proto; 11623 break; 11624 case BPF_FUNC_skc_to_udp6_sock: 11625 func = &bpf_skc_to_udp6_sock_proto; 11626 break; 11627 case BPF_FUNC_skc_to_unix_sock: 11628 func = &bpf_skc_to_unix_sock_proto; 11629 break; 11630 case BPF_FUNC_skc_to_mptcp_sock: 11631 func = &bpf_skc_to_mptcp_sock_proto; 11632 break; 11633 case BPF_FUNC_ktime_get_coarse_ns: 11634 return &bpf_ktime_get_coarse_ns_proto; 11635 default: 11636 return bpf_base_func_proto(func_id); 11637 } 11638 11639 if (!perfmon_capable()) 11640 return NULL; 11641 11642 return func; 11643 } 11644 11645 __diag_push(); 11646 __diag_ignore_all("-Wmissing-prototypes", 11647 "Global functions as their definitions will be in vmlinux BTF"); 11648 __bpf_kfunc int bpf_dynptr_from_skb(struct sk_buff *skb, u64 flags, 11649 struct bpf_dynptr_kern *ptr__uninit) 11650 { 11651 if (flags) { 11652 bpf_dynptr_set_null(ptr__uninit); 11653 return -EINVAL; 11654 } 11655 11656 bpf_dynptr_init(ptr__uninit, skb, BPF_DYNPTR_TYPE_SKB, 0, skb->len); 11657 11658 return 0; 11659 } 11660 11661 __bpf_kfunc int bpf_dynptr_from_xdp(struct xdp_buff *xdp, u64 flags, 11662 struct bpf_dynptr_kern *ptr__uninit) 11663 { 11664 if (flags) { 11665 bpf_dynptr_set_null(ptr__uninit); 11666 return -EINVAL; 11667 } 11668 11669 bpf_dynptr_init(ptr__uninit, xdp, BPF_DYNPTR_TYPE_XDP, 0, xdp_get_buff_len(xdp)); 11670 11671 return 0; 11672 } 11673 __diag_pop(); 11674 11675 int bpf_dynptr_from_skb_rdonly(struct sk_buff *skb, u64 flags, 11676 struct bpf_dynptr_kern *ptr__uninit) 11677 { 11678 int err; 11679 11680 err = bpf_dynptr_from_skb(skb, flags, ptr__uninit); 11681 if (err) 11682 return err; 11683 11684 bpf_dynptr_set_rdonly(ptr__uninit); 11685 11686 return 0; 11687 } 11688 11689 BTF_SET8_START(bpf_kfunc_check_set_skb) 11690 BTF_ID_FLAGS(func, bpf_dynptr_from_skb) 11691 BTF_SET8_END(bpf_kfunc_check_set_skb) 11692 11693 BTF_SET8_START(bpf_kfunc_check_set_xdp) 11694 BTF_ID_FLAGS(func, bpf_dynptr_from_xdp) 11695 BTF_SET8_END(bpf_kfunc_check_set_xdp) 11696 11697 static const struct btf_kfunc_id_set bpf_kfunc_set_skb = { 11698 .owner = THIS_MODULE, 11699 .set = &bpf_kfunc_check_set_skb, 11700 }; 11701 11702 static const struct btf_kfunc_id_set bpf_kfunc_set_xdp = { 11703 .owner = THIS_MODULE, 11704 .set = &bpf_kfunc_check_set_xdp, 11705 }; 11706 11707 static int __init bpf_kfunc_init(void) 11708 { 11709 int ret; 11710 11711 ret = register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_CLS, &bpf_kfunc_set_skb); 11712 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SCHED_ACT, &bpf_kfunc_set_skb); 11713 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SK_SKB, &bpf_kfunc_set_skb); 11714 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_SOCKET_FILTER, &bpf_kfunc_set_skb); 11715 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_CGROUP_SKB, &bpf_kfunc_set_skb); 11716 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_OUT, &bpf_kfunc_set_skb); 11717 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_IN, &bpf_kfunc_set_skb); 11718 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_XMIT, &bpf_kfunc_set_skb); 11719 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_LWT_SEG6LOCAL, &bpf_kfunc_set_skb); 11720 ret = ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_NETFILTER, &bpf_kfunc_set_skb); 11721 return ret ?: register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &bpf_kfunc_set_xdp); 11722 } 11723 late_initcall(bpf_kfunc_init); 11724