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/module.h> 21 #include <linux/types.h> 22 #include <linux/mm.h> 23 #include <linux/fcntl.h> 24 #include <linux/socket.h> 25 #include <linux/sock_diag.h> 26 #include <linux/in.h> 27 #include <linux/inet.h> 28 #include <linux/netdevice.h> 29 #include <linux/if_packet.h> 30 #include <linux/if_arp.h> 31 #include <linux/gfp.h> 32 #include <net/inet_common.h> 33 #include <net/ip.h> 34 #include <net/protocol.h> 35 #include <net/netlink.h> 36 #include <linux/skbuff.h> 37 #include <linux/skmsg.h> 38 #include <net/sock.h> 39 #include <net/flow_dissector.h> 40 #include <linux/errno.h> 41 #include <linux/timer.h> 42 #include <linux/uaccess.h> 43 #include <asm/unaligned.h> 44 #include <asm/cmpxchg.h> 45 #include <linux/filter.h> 46 #include <linux/ratelimit.h> 47 #include <linux/seccomp.h> 48 #include <linux/if_vlan.h> 49 #include <linux/bpf.h> 50 #include <linux/btf.h> 51 #include <net/sch_generic.h> 52 #include <net/cls_cgroup.h> 53 #include <net/dst_metadata.h> 54 #include <net/dst.h> 55 #include <net/sock_reuseport.h> 56 #include <net/busy_poll.h> 57 #include <net/tcp.h> 58 #include <net/xfrm.h> 59 #include <net/udp.h> 60 #include <linux/bpf_trace.h> 61 #include <net/xdp_sock.h> 62 #include <linux/inetdevice.h> 63 #include <net/inet_hashtables.h> 64 #include <net/inet6_hashtables.h> 65 #include <net/ip_fib.h> 66 #include <net/nexthop.h> 67 #include <net/flow.h> 68 #include <net/arp.h> 69 #include <net/ipv6.h> 70 #include <net/net_namespace.h> 71 #include <linux/seg6_local.h> 72 #include <net/seg6.h> 73 #include <net/seg6_local.h> 74 #include <net/lwtunnel.h> 75 #include <net/ipv6_stubs.h> 76 #include <net/bpf_sk_storage.h> 77 #include <net/transp_v6.h> 78 #include <linux/btf_ids.h> 79 #include <net/tls.h> 80 81 static const struct bpf_func_proto * 82 bpf_sk_base_func_proto(enum bpf_func_id func_id); 83 84 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len) 85 { 86 if (in_compat_syscall()) { 87 struct compat_sock_fprog f32; 88 89 if (len != sizeof(f32)) 90 return -EINVAL; 91 if (copy_from_sockptr(&f32, src, sizeof(f32))) 92 return -EFAULT; 93 memset(dst, 0, sizeof(*dst)); 94 dst->len = f32.len; 95 dst->filter = compat_ptr(f32.filter); 96 } else { 97 if (len != sizeof(*dst)) 98 return -EINVAL; 99 if (copy_from_sockptr(dst, src, sizeof(*dst))) 100 return -EFAULT; 101 } 102 103 return 0; 104 } 105 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user); 106 107 /** 108 * sk_filter_trim_cap - run a packet through a socket filter 109 * @sk: sock associated with &sk_buff 110 * @skb: buffer to filter 111 * @cap: limit on how short the eBPF program may trim the packet 112 * 113 * Run the eBPF program and then cut skb->data to correct size returned by 114 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller 115 * than pkt_len we keep whole skb->data. This is the socket level 116 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should 117 * be accepted or -EPERM if the packet should be tossed. 118 * 119 */ 120 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap) 121 { 122 int err; 123 struct sk_filter *filter; 124 125 /* 126 * If the skb was allocated from pfmemalloc reserves, only 127 * allow SOCK_MEMALLOC sockets to use it as this socket is 128 * helping free memory 129 */ 130 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) { 131 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP); 132 return -ENOMEM; 133 } 134 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb); 135 if (err) 136 return err; 137 138 err = security_sock_rcv_skb(sk, skb); 139 if (err) 140 return err; 141 142 rcu_read_lock(); 143 filter = rcu_dereference(sk->sk_filter); 144 if (filter) { 145 struct sock *save_sk = skb->sk; 146 unsigned int pkt_len; 147 148 skb->sk = sk; 149 pkt_len = bpf_prog_run_save_cb(filter->prog, skb); 150 skb->sk = save_sk; 151 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM; 152 } 153 rcu_read_unlock(); 154 155 return err; 156 } 157 EXPORT_SYMBOL(sk_filter_trim_cap); 158 159 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb) 160 { 161 return skb_get_poff(skb); 162 } 163 164 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x) 165 { 166 struct nlattr *nla; 167 168 if (skb_is_nonlinear(skb)) 169 return 0; 170 171 if (skb->len < sizeof(struct nlattr)) 172 return 0; 173 174 if (a > skb->len - sizeof(struct nlattr)) 175 return 0; 176 177 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x); 178 if (nla) 179 return (void *) nla - (void *) skb->data; 180 181 return 0; 182 } 183 184 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x) 185 { 186 struct nlattr *nla; 187 188 if (skb_is_nonlinear(skb)) 189 return 0; 190 191 if (skb->len < sizeof(struct nlattr)) 192 return 0; 193 194 if (a > skb->len - sizeof(struct nlattr)) 195 return 0; 196 197 nla = (struct nlattr *) &skb->data[a]; 198 if (nla->nla_len > skb->len - a) 199 return 0; 200 201 nla = nla_find_nested(nla, x); 202 if (nla) 203 return (void *) nla - (void *) skb->data; 204 205 return 0; 206 } 207 208 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *, 209 data, int, headlen, int, offset) 210 { 211 u8 tmp, *ptr; 212 const int len = sizeof(tmp); 213 214 if (offset >= 0) { 215 if (headlen - offset >= len) 216 return *(u8 *)(data + offset); 217 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 218 return tmp; 219 } else { 220 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 221 if (likely(ptr)) 222 return *(u8 *)ptr; 223 } 224 225 return -EFAULT; 226 } 227 228 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb, 229 int, offset) 230 { 231 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len, 232 offset); 233 } 234 235 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *, 236 data, int, headlen, int, offset) 237 { 238 u16 tmp, *ptr; 239 const int len = sizeof(tmp); 240 241 if (offset >= 0) { 242 if (headlen - offset >= len) 243 return get_unaligned_be16(data + offset); 244 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 245 return be16_to_cpu(tmp); 246 } else { 247 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 248 if (likely(ptr)) 249 return get_unaligned_be16(ptr); 250 } 251 252 return -EFAULT; 253 } 254 255 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb, 256 int, offset) 257 { 258 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len, 259 offset); 260 } 261 262 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *, 263 data, int, headlen, int, offset) 264 { 265 u32 tmp, *ptr; 266 const int len = sizeof(tmp); 267 268 if (likely(offset >= 0)) { 269 if (headlen - offset >= len) 270 return get_unaligned_be32(data + offset); 271 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 272 return be32_to_cpu(tmp); 273 } else { 274 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 275 if (likely(ptr)) 276 return get_unaligned_be32(ptr); 277 } 278 279 return -EFAULT; 280 } 281 282 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb, 283 int, offset) 284 { 285 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len, 286 offset); 287 } 288 289 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg, 290 struct bpf_insn *insn_buf) 291 { 292 struct bpf_insn *insn = insn_buf; 293 294 switch (skb_field) { 295 case SKF_AD_MARK: 296 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4); 297 298 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, 299 offsetof(struct sk_buff, mark)); 300 break; 301 302 case SKF_AD_PKTTYPE: 303 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET()); 304 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX); 305 #ifdef __BIG_ENDIAN_BITFIELD 306 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5); 307 #endif 308 break; 309 310 case SKF_AD_QUEUE: 311 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2); 312 313 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 314 offsetof(struct sk_buff, queue_mapping)); 315 break; 316 317 case SKF_AD_VLAN_TAG: 318 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2); 319 320 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */ 321 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 322 offsetof(struct sk_buff, vlan_tci)); 323 break; 324 case SKF_AD_VLAN_TAG_PRESENT: 325 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET()); 326 if (PKT_VLAN_PRESENT_BIT) 327 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT); 328 if (PKT_VLAN_PRESENT_BIT < 7) 329 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1); 330 break; 331 } 332 333 return insn - insn_buf; 334 } 335 336 static bool convert_bpf_extensions(struct sock_filter *fp, 337 struct bpf_insn **insnp) 338 { 339 struct bpf_insn *insn = *insnp; 340 u32 cnt; 341 342 switch (fp->k) { 343 case SKF_AD_OFF + SKF_AD_PROTOCOL: 344 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2); 345 346 /* A = *(u16 *) (CTX + offsetof(protocol)) */ 347 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 348 offsetof(struct sk_buff, protocol)); 349 /* A = ntohs(A) [emitting a nop or swap16] */ 350 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 351 break; 352 353 case SKF_AD_OFF + SKF_AD_PKTTYPE: 354 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn); 355 insn += cnt - 1; 356 break; 357 358 case SKF_AD_OFF + SKF_AD_IFINDEX: 359 case SKF_AD_OFF + SKF_AD_HATYPE: 360 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4); 361 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2); 362 363 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 364 BPF_REG_TMP, BPF_REG_CTX, 365 offsetof(struct sk_buff, dev)); 366 /* if (tmp != 0) goto pc + 1 */ 367 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1); 368 *insn++ = BPF_EXIT_INSN(); 369 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) 370 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP, 371 offsetof(struct net_device, ifindex)); 372 else 373 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP, 374 offsetof(struct net_device, type)); 375 break; 376 377 case SKF_AD_OFF + SKF_AD_MARK: 378 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn); 379 insn += cnt - 1; 380 break; 381 382 case SKF_AD_OFF + SKF_AD_RXHASH: 383 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4); 384 385 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, 386 offsetof(struct sk_buff, hash)); 387 break; 388 389 case SKF_AD_OFF + SKF_AD_QUEUE: 390 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn); 391 insn += cnt - 1; 392 break; 393 394 case SKF_AD_OFF + SKF_AD_VLAN_TAG: 395 cnt = convert_skb_access(SKF_AD_VLAN_TAG, 396 BPF_REG_A, BPF_REG_CTX, insn); 397 insn += cnt - 1; 398 break; 399 400 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT: 401 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT, 402 BPF_REG_A, BPF_REG_CTX, insn); 403 insn += cnt - 1; 404 break; 405 406 case SKF_AD_OFF + SKF_AD_VLAN_TPID: 407 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2); 408 409 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */ 410 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 411 offsetof(struct sk_buff, vlan_proto)); 412 /* A = ntohs(A) [emitting a nop or swap16] */ 413 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 414 break; 415 416 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 417 case SKF_AD_OFF + SKF_AD_NLATTR: 418 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 419 case SKF_AD_OFF + SKF_AD_CPU: 420 case SKF_AD_OFF + SKF_AD_RANDOM: 421 /* arg1 = CTX */ 422 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 423 /* arg2 = A */ 424 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A); 425 /* arg3 = X */ 426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X); 427 /* Emit call(arg1=CTX, arg2=A, arg3=X) */ 428 switch (fp->k) { 429 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 430 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset); 431 break; 432 case SKF_AD_OFF + SKF_AD_NLATTR: 433 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr); 434 break; 435 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 436 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest); 437 break; 438 case SKF_AD_OFF + SKF_AD_CPU: 439 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id); 440 break; 441 case SKF_AD_OFF + SKF_AD_RANDOM: 442 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32); 443 bpf_user_rnd_init_once(); 444 break; 445 } 446 break; 447 448 case SKF_AD_OFF + SKF_AD_ALU_XOR_X: 449 /* A ^= X */ 450 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X); 451 break; 452 453 default: 454 /* This is just a dummy call to avoid letting the compiler 455 * evict __bpf_call_base() as an optimization. Placed here 456 * where no-one bothers. 457 */ 458 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0); 459 return false; 460 } 461 462 *insnp = insn; 463 return true; 464 } 465 466 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp) 467 { 468 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS); 469 int size = bpf_size_to_bytes(BPF_SIZE(fp->code)); 470 bool endian = BPF_SIZE(fp->code) == BPF_H || 471 BPF_SIZE(fp->code) == BPF_W; 472 bool indirect = BPF_MODE(fp->code) == BPF_IND; 473 const int ip_align = NET_IP_ALIGN; 474 struct bpf_insn *insn = *insnp; 475 int offset = fp->k; 476 477 if (!indirect && 478 ((unaligned_ok && offset >= 0) || 479 (!unaligned_ok && offset >= 0 && 480 offset + ip_align >= 0 && 481 offset + ip_align % size == 0))) { 482 bool ldx_off_ok = offset <= S16_MAX; 483 484 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H); 485 if (offset) 486 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset); 487 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP, 488 size, 2 + endian + (!ldx_off_ok * 2)); 489 if (ldx_off_ok) { 490 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 491 BPF_REG_D, offset); 492 } else { 493 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D); 494 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset); 495 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 496 BPF_REG_TMP, 0); 497 } 498 if (endian) 499 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8); 500 *insn++ = BPF_JMP_A(8); 501 } 502 503 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 504 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D); 505 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H); 506 if (!indirect) { 507 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset); 508 } else { 509 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X); 510 if (fp->k) 511 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset); 512 } 513 514 switch (BPF_SIZE(fp->code)) { 515 case BPF_B: 516 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8); 517 break; 518 case BPF_H: 519 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16); 520 break; 521 case BPF_W: 522 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32); 523 break; 524 default: 525 return false; 526 } 527 528 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2); 529 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 530 *insn = BPF_EXIT_INSN(); 531 532 *insnp = insn; 533 return true; 534 } 535 536 /** 537 * bpf_convert_filter - convert filter program 538 * @prog: the user passed filter program 539 * @len: the length of the user passed filter program 540 * @new_prog: allocated 'struct bpf_prog' or NULL 541 * @new_len: pointer to store length of converted program 542 * @seen_ld_abs: bool whether we've seen ld_abs/ind 543 * 544 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn' 545 * style extended BPF (eBPF). 546 * Conversion workflow: 547 * 548 * 1) First pass for calculating the new program length: 549 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs) 550 * 551 * 2) 2nd pass to remap in two passes: 1st pass finds new 552 * jump offsets, 2nd pass remapping: 553 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs) 554 */ 555 static int bpf_convert_filter(struct sock_filter *prog, int len, 556 struct bpf_prog *new_prog, int *new_len, 557 bool *seen_ld_abs) 558 { 559 int new_flen = 0, pass = 0, target, i, stack_off; 560 struct bpf_insn *new_insn, *first_insn = NULL; 561 struct sock_filter *fp; 562 int *addrs = NULL; 563 u8 bpf_src; 564 565 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK); 566 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); 567 568 if (len <= 0 || len > BPF_MAXINSNS) 569 return -EINVAL; 570 571 if (new_prog) { 572 first_insn = new_prog->insnsi; 573 addrs = kcalloc(len, sizeof(*addrs), 574 GFP_KERNEL | __GFP_NOWARN); 575 if (!addrs) 576 return -ENOMEM; 577 } 578 579 do_pass: 580 new_insn = first_insn; 581 fp = prog; 582 583 /* Classic BPF related prologue emission. */ 584 if (new_prog) { 585 /* Classic BPF expects A and X to be reset first. These need 586 * to be guaranteed to be the first two instructions. 587 */ 588 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 589 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X); 590 591 /* All programs must keep CTX in callee saved BPF_REG_CTX. 592 * In eBPF case it's done by the compiler, here we need to 593 * do this ourself. Initial CTX is present in BPF_REG_ARG1. 594 */ 595 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1); 596 if (*seen_ld_abs) { 597 /* For packet access in classic BPF, cache skb->data 598 * in callee-saved BPF R8 and skb->len - skb->data_len 599 * (headlen) in BPF R9. Since classic BPF is read-only 600 * on CTX, we only need to cache it once. 601 */ 602 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 603 BPF_REG_D, BPF_REG_CTX, 604 offsetof(struct sk_buff, data)); 605 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX, 606 offsetof(struct sk_buff, len)); 607 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX, 608 offsetof(struct sk_buff, data_len)); 609 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP); 610 } 611 } else { 612 new_insn += 3; 613 } 614 615 for (i = 0; i < len; fp++, i++) { 616 struct bpf_insn tmp_insns[32] = { }; 617 struct bpf_insn *insn = tmp_insns; 618 619 if (addrs) 620 addrs[i] = new_insn - first_insn; 621 622 switch (fp->code) { 623 /* All arithmetic insns and skb loads map as-is. */ 624 case BPF_ALU | BPF_ADD | BPF_X: 625 case BPF_ALU | BPF_ADD | BPF_K: 626 case BPF_ALU | BPF_SUB | BPF_X: 627 case BPF_ALU | BPF_SUB | BPF_K: 628 case BPF_ALU | BPF_AND | BPF_X: 629 case BPF_ALU | BPF_AND | BPF_K: 630 case BPF_ALU | BPF_OR | BPF_X: 631 case BPF_ALU | BPF_OR | BPF_K: 632 case BPF_ALU | BPF_LSH | BPF_X: 633 case BPF_ALU | BPF_LSH | BPF_K: 634 case BPF_ALU | BPF_RSH | BPF_X: 635 case BPF_ALU | BPF_RSH | BPF_K: 636 case BPF_ALU | BPF_XOR | BPF_X: 637 case BPF_ALU | BPF_XOR | BPF_K: 638 case BPF_ALU | BPF_MUL | BPF_X: 639 case BPF_ALU | BPF_MUL | BPF_K: 640 case BPF_ALU | BPF_DIV | BPF_X: 641 case BPF_ALU | BPF_DIV | BPF_K: 642 case BPF_ALU | BPF_MOD | BPF_X: 643 case BPF_ALU | BPF_MOD | BPF_K: 644 case BPF_ALU | BPF_NEG: 645 case BPF_LD | BPF_ABS | BPF_W: 646 case BPF_LD | BPF_ABS | BPF_H: 647 case BPF_LD | BPF_ABS | BPF_B: 648 case BPF_LD | BPF_IND | BPF_W: 649 case BPF_LD | BPF_IND | BPF_H: 650 case BPF_LD | BPF_IND | BPF_B: 651 /* Check for overloaded BPF extension and 652 * directly convert it if found, otherwise 653 * just move on with mapping. 654 */ 655 if (BPF_CLASS(fp->code) == BPF_LD && 656 BPF_MODE(fp->code) == BPF_ABS && 657 convert_bpf_extensions(fp, &insn)) 658 break; 659 if (BPF_CLASS(fp->code) == BPF_LD && 660 convert_bpf_ld_abs(fp, &insn)) { 661 *seen_ld_abs = true; 662 break; 663 } 664 665 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) || 666 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) { 667 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X); 668 /* Error with exception code on div/mod by 0. 669 * For cBPF programs, this was always return 0. 670 */ 671 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2); 672 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 673 *insn++ = BPF_EXIT_INSN(); 674 } 675 676 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k); 677 break; 678 679 /* Jump transformation cannot use BPF block macros 680 * everywhere as offset calculation and target updates 681 * require a bit more work than the rest, i.e. jump 682 * opcodes map as-is, but offsets need adjustment. 683 */ 684 685 #define BPF_EMIT_JMP \ 686 do { \ 687 const s32 off_min = S16_MIN, off_max = S16_MAX; \ 688 s32 off; \ 689 \ 690 if (target >= len || target < 0) \ 691 goto err; \ 692 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \ 693 /* Adjust pc relative offset for 2nd or 3rd insn. */ \ 694 off -= insn - tmp_insns; \ 695 /* Reject anything not fitting into insn->off. */ \ 696 if (off < off_min || off > off_max) \ 697 goto err; \ 698 insn->off = off; \ 699 } while (0) 700 701 case BPF_JMP | BPF_JA: 702 target = i + fp->k + 1; 703 insn->code = fp->code; 704 BPF_EMIT_JMP; 705 break; 706 707 case BPF_JMP | BPF_JEQ | BPF_K: 708 case BPF_JMP | BPF_JEQ | BPF_X: 709 case BPF_JMP | BPF_JSET | BPF_K: 710 case BPF_JMP | BPF_JSET | BPF_X: 711 case BPF_JMP | BPF_JGT | BPF_K: 712 case BPF_JMP | BPF_JGT | BPF_X: 713 case BPF_JMP | BPF_JGE | BPF_K: 714 case BPF_JMP | BPF_JGE | BPF_X: 715 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) { 716 /* BPF immediates are signed, zero extend 717 * immediate into tmp register and use it 718 * in compare insn. 719 */ 720 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k); 721 722 insn->dst_reg = BPF_REG_A; 723 insn->src_reg = BPF_REG_TMP; 724 bpf_src = BPF_X; 725 } else { 726 insn->dst_reg = BPF_REG_A; 727 insn->imm = fp->k; 728 bpf_src = BPF_SRC(fp->code); 729 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0; 730 } 731 732 /* Common case where 'jump_false' is next insn. */ 733 if (fp->jf == 0) { 734 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 735 target = i + fp->jt + 1; 736 BPF_EMIT_JMP; 737 break; 738 } 739 740 /* Convert some jumps when 'jump_true' is next insn. */ 741 if (fp->jt == 0) { 742 switch (BPF_OP(fp->code)) { 743 case BPF_JEQ: 744 insn->code = BPF_JMP | BPF_JNE | bpf_src; 745 break; 746 case BPF_JGT: 747 insn->code = BPF_JMP | BPF_JLE | bpf_src; 748 break; 749 case BPF_JGE: 750 insn->code = BPF_JMP | BPF_JLT | bpf_src; 751 break; 752 default: 753 goto jmp_rest; 754 } 755 756 target = i + fp->jf + 1; 757 BPF_EMIT_JMP; 758 break; 759 } 760 jmp_rest: 761 /* Other jumps are mapped into two insns: Jxx and JA. */ 762 target = i + fp->jt + 1; 763 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 764 BPF_EMIT_JMP; 765 insn++; 766 767 insn->code = BPF_JMP | BPF_JA; 768 target = i + fp->jf + 1; 769 BPF_EMIT_JMP; 770 break; 771 772 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */ 773 case BPF_LDX | BPF_MSH | BPF_B: { 774 struct sock_filter tmp = { 775 .code = BPF_LD | BPF_ABS | BPF_B, 776 .k = fp->k, 777 }; 778 779 *seen_ld_abs = true; 780 781 /* X = A */ 782 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 783 /* A = BPF_R0 = *(u8 *) (skb->data + K) */ 784 convert_bpf_ld_abs(&tmp, &insn); 785 insn++; 786 /* A &= 0xf */ 787 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf); 788 /* A <<= 2 */ 789 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2); 790 /* tmp = X */ 791 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X); 792 /* X = A */ 793 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 794 /* A = tmp */ 795 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP); 796 break; 797 } 798 /* RET_K is remaped into 2 insns. RET_A case doesn't need an 799 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A. 800 */ 801 case BPF_RET | BPF_A: 802 case BPF_RET | BPF_K: 803 if (BPF_RVAL(fp->code) == BPF_K) 804 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0, 805 0, fp->k); 806 *insn = BPF_EXIT_INSN(); 807 break; 808 809 /* Store to stack. */ 810 case BPF_ST: 811 case BPF_STX: 812 stack_off = fp->k * 4 + 4; 813 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) == 814 BPF_ST ? BPF_REG_A : BPF_REG_X, 815 -stack_off); 816 /* check_load_and_stores() verifies that classic BPF can 817 * load from stack only after write, so tracking 818 * stack_depth for ST|STX insns is enough 819 */ 820 if (new_prog && new_prog->aux->stack_depth < stack_off) 821 new_prog->aux->stack_depth = stack_off; 822 break; 823 824 /* Load from stack. */ 825 case BPF_LD | BPF_MEM: 826 case BPF_LDX | BPF_MEM: 827 stack_off = fp->k * 4 + 4; 828 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 829 BPF_REG_A : BPF_REG_X, BPF_REG_FP, 830 -stack_off); 831 break; 832 833 /* A = K or X = K */ 834 case BPF_LD | BPF_IMM: 835 case BPF_LDX | BPF_IMM: 836 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ? 837 BPF_REG_A : BPF_REG_X, fp->k); 838 break; 839 840 /* X = A */ 841 case BPF_MISC | BPF_TAX: 842 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 843 break; 844 845 /* A = X */ 846 case BPF_MISC | BPF_TXA: 847 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X); 848 break; 849 850 /* A = skb->len or X = skb->len */ 851 case BPF_LD | BPF_W | BPF_LEN: 852 case BPF_LDX | BPF_W | BPF_LEN: 853 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 854 BPF_REG_A : BPF_REG_X, BPF_REG_CTX, 855 offsetof(struct sk_buff, len)); 856 break; 857 858 /* Access seccomp_data fields. */ 859 case BPF_LDX | BPF_ABS | BPF_W: 860 /* A = *(u32 *) (ctx + K) */ 861 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k); 862 break; 863 864 /* Unknown instruction. */ 865 default: 866 goto err; 867 } 868 869 insn++; 870 if (new_prog) 871 memcpy(new_insn, tmp_insns, 872 sizeof(*insn) * (insn - tmp_insns)); 873 new_insn += insn - tmp_insns; 874 } 875 876 if (!new_prog) { 877 /* Only calculating new length. */ 878 *new_len = new_insn - first_insn; 879 if (*seen_ld_abs) 880 *new_len += 4; /* Prologue bits. */ 881 return 0; 882 } 883 884 pass++; 885 if (new_flen != new_insn - first_insn) { 886 new_flen = new_insn - first_insn; 887 if (pass > 2) 888 goto err; 889 goto do_pass; 890 } 891 892 kfree(addrs); 893 BUG_ON(*new_len != new_flen); 894 return 0; 895 err: 896 kfree(addrs); 897 return -EINVAL; 898 } 899 900 /* Security: 901 * 902 * As we dont want to clear mem[] array for each packet going through 903 * __bpf_prog_run(), we check that filter loaded by user never try to read 904 * a cell if not previously written, and we check all branches to be sure 905 * a malicious user doesn't try to abuse us. 906 */ 907 static int check_load_and_stores(const struct sock_filter *filter, int flen) 908 { 909 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */ 910 int pc, ret = 0; 911 912 BUILD_BUG_ON(BPF_MEMWORDS > 16); 913 914 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL); 915 if (!masks) 916 return -ENOMEM; 917 918 memset(masks, 0xff, flen * sizeof(*masks)); 919 920 for (pc = 0; pc < flen; pc++) { 921 memvalid &= masks[pc]; 922 923 switch (filter[pc].code) { 924 case BPF_ST: 925 case BPF_STX: 926 memvalid |= (1 << filter[pc].k); 927 break; 928 case BPF_LD | BPF_MEM: 929 case BPF_LDX | BPF_MEM: 930 if (!(memvalid & (1 << filter[pc].k))) { 931 ret = -EINVAL; 932 goto error; 933 } 934 break; 935 case BPF_JMP | BPF_JA: 936 /* A jump must set masks on target */ 937 masks[pc + 1 + filter[pc].k] &= memvalid; 938 memvalid = ~0; 939 break; 940 case BPF_JMP | BPF_JEQ | BPF_K: 941 case BPF_JMP | BPF_JEQ | BPF_X: 942 case BPF_JMP | BPF_JGE | BPF_K: 943 case BPF_JMP | BPF_JGE | BPF_X: 944 case BPF_JMP | BPF_JGT | BPF_K: 945 case BPF_JMP | BPF_JGT | BPF_X: 946 case BPF_JMP | BPF_JSET | BPF_K: 947 case BPF_JMP | BPF_JSET | BPF_X: 948 /* A jump must set masks on targets */ 949 masks[pc + 1 + filter[pc].jt] &= memvalid; 950 masks[pc + 1 + filter[pc].jf] &= memvalid; 951 memvalid = ~0; 952 break; 953 } 954 } 955 error: 956 kfree(masks); 957 return ret; 958 } 959 960 static bool chk_code_allowed(u16 code_to_probe) 961 { 962 static const bool codes[] = { 963 /* 32 bit ALU operations */ 964 [BPF_ALU | BPF_ADD | BPF_K] = true, 965 [BPF_ALU | BPF_ADD | BPF_X] = true, 966 [BPF_ALU | BPF_SUB | BPF_K] = true, 967 [BPF_ALU | BPF_SUB | BPF_X] = true, 968 [BPF_ALU | BPF_MUL | BPF_K] = true, 969 [BPF_ALU | BPF_MUL | BPF_X] = true, 970 [BPF_ALU | BPF_DIV | BPF_K] = true, 971 [BPF_ALU | BPF_DIV | BPF_X] = true, 972 [BPF_ALU | BPF_MOD | BPF_K] = true, 973 [BPF_ALU | BPF_MOD | BPF_X] = true, 974 [BPF_ALU | BPF_AND | BPF_K] = true, 975 [BPF_ALU | BPF_AND | BPF_X] = true, 976 [BPF_ALU | BPF_OR | BPF_K] = true, 977 [BPF_ALU | BPF_OR | BPF_X] = true, 978 [BPF_ALU | BPF_XOR | BPF_K] = true, 979 [BPF_ALU | BPF_XOR | BPF_X] = true, 980 [BPF_ALU | BPF_LSH | BPF_K] = true, 981 [BPF_ALU | BPF_LSH | BPF_X] = true, 982 [BPF_ALU | BPF_RSH | BPF_K] = true, 983 [BPF_ALU | BPF_RSH | BPF_X] = true, 984 [BPF_ALU | BPF_NEG] = true, 985 /* Load instructions */ 986 [BPF_LD | BPF_W | BPF_ABS] = true, 987 [BPF_LD | BPF_H | BPF_ABS] = true, 988 [BPF_LD | BPF_B | BPF_ABS] = true, 989 [BPF_LD | BPF_W | BPF_LEN] = true, 990 [BPF_LD | BPF_W | BPF_IND] = true, 991 [BPF_LD | BPF_H | BPF_IND] = true, 992 [BPF_LD | BPF_B | BPF_IND] = true, 993 [BPF_LD | BPF_IMM] = true, 994 [BPF_LD | BPF_MEM] = true, 995 [BPF_LDX | BPF_W | BPF_LEN] = true, 996 [BPF_LDX | BPF_B | BPF_MSH] = true, 997 [BPF_LDX | BPF_IMM] = true, 998 [BPF_LDX | BPF_MEM] = true, 999 /* Store instructions */ 1000 [BPF_ST] = true, 1001 [BPF_STX] = true, 1002 /* Misc instructions */ 1003 [BPF_MISC | BPF_TAX] = true, 1004 [BPF_MISC | BPF_TXA] = true, 1005 /* Return instructions */ 1006 [BPF_RET | BPF_K] = true, 1007 [BPF_RET | BPF_A] = true, 1008 /* Jump instructions */ 1009 [BPF_JMP | BPF_JA] = true, 1010 [BPF_JMP | BPF_JEQ | BPF_K] = true, 1011 [BPF_JMP | BPF_JEQ | BPF_X] = true, 1012 [BPF_JMP | BPF_JGE | BPF_K] = true, 1013 [BPF_JMP | BPF_JGE | BPF_X] = true, 1014 [BPF_JMP | BPF_JGT | BPF_K] = true, 1015 [BPF_JMP | BPF_JGT | BPF_X] = true, 1016 [BPF_JMP | BPF_JSET | BPF_K] = true, 1017 [BPF_JMP | BPF_JSET | BPF_X] = true, 1018 }; 1019 1020 if (code_to_probe >= ARRAY_SIZE(codes)) 1021 return false; 1022 1023 return codes[code_to_probe]; 1024 } 1025 1026 static bool bpf_check_basics_ok(const struct sock_filter *filter, 1027 unsigned int flen) 1028 { 1029 if (filter == NULL) 1030 return false; 1031 if (flen == 0 || flen > BPF_MAXINSNS) 1032 return false; 1033 1034 return true; 1035 } 1036 1037 /** 1038 * bpf_check_classic - verify socket filter code 1039 * @filter: filter to verify 1040 * @flen: length of filter 1041 * 1042 * Check the user's filter code. If we let some ugly 1043 * filter code slip through kaboom! The filter must contain 1044 * no references or jumps that are out of range, no illegal 1045 * instructions, and must end with a RET instruction. 1046 * 1047 * All jumps are forward as they are not signed. 1048 * 1049 * Returns 0 if the rule set is legal or -EINVAL if not. 1050 */ 1051 static int bpf_check_classic(const struct sock_filter *filter, 1052 unsigned int flen) 1053 { 1054 bool anc_found; 1055 int pc; 1056 1057 /* Check the filter code now */ 1058 for (pc = 0; pc < flen; pc++) { 1059 const struct sock_filter *ftest = &filter[pc]; 1060 1061 /* May we actually operate on this code? */ 1062 if (!chk_code_allowed(ftest->code)) 1063 return -EINVAL; 1064 1065 /* Some instructions need special checks */ 1066 switch (ftest->code) { 1067 case BPF_ALU | BPF_DIV | BPF_K: 1068 case BPF_ALU | BPF_MOD | BPF_K: 1069 /* Check for division by zero */ 1070 if (ftest->k == 0) 1071 return -EINVAL; 1072 break; 1073 case BPF_ALU | BPF_LSH | BPF_K: 1074 case BPF_ALU | BPF_RSH | BPF_K: 1075 if (ftest->k >= 32) 1076 return -EINVAL; 1077 break; 1078 case BPF_LD | BPF_MEM: 1079 case BPF_LDX | BPF_MEM: 1080 case BPF_ST: 1081 case BPF_STX: 1082 /* Check for invalid memory addresses */ 1083 if (ftest->k >= BPF_MEMWORDS) 1084 return -EINVAL; 1085 break; 1086 case BPF_JMP | BPF_JA: 1087 /* Note, the large ftest->k might cause loops. 1088 * Compare this with conditional jumps below, 1089 * where offsets are limited. --ANK (981016) 1090 */ 1091 if (ftest->k >= (unsigned int)(flen - pc - 1)) 1092 return -EINVAL; 1093 break; 1094 case BPF_JMP | BPF_JEQ | BPF_K: 1095 case BPF_JMP | BPF_JEQ | BPF_X: 1096 case BPF_JMP | BPF_JGE | BPF_K: 1097 case BPF_JMP | BPF_JGE | BPF_X: 1098 case BPF_JMP | BPF_JGT | BPF_K: 1099 case BPF_JMP | BPF_JGT | BPF_X: 1100 case BPF_JMP | BPF_JSET | BPF_K: 1101 case BPF_JMP | BPF_JSET | BPF_X: 1102 /* Both conditionals must be safe */ 1103 if (pc + ftest->jt + 1 >= flen || 1104 pc + ftest->jf + 1 >= flen) 1105 return -EINVAL; 1106 break; 1107 case BPF_LD | BPF_W | BPF_ABS: 1108 case BPF_LD | BPF_H | BPF_ABS: 1109 case BPF_LD | BPF_B | BPF_ABS: 1110 anc_found = false; 1111 if (bpf_anc_helper(ftest) & BPF_ANC) 1112 anc_found = true; 1113 /* Ancillary operation unknown or unsupported */ 1114 if (anc_found == false && ftest->k >= SKF_AD_OFF) 1115 return -EINVAL; 1116 } 1117 } 1118 1119 /* Last instruction must be a RET code */ 1120 switch (filter[flen - 1].code) { 1121 case BPF_RET | BPF_K: 1122 case BPF_RET | BPF_A: 1123 return check_load_and_stores(filter, flen); 1124 } 1125 1126 return -EINVAL; 1127 } 1128 1129 static int bpf_prog_store_orig_filter(struct bpf_prog *fp, 1130 const struct sock_fprog *fprog) 1131 { 1132 unsigned int fsize = bpf_classic_proglen(fprog); 1133 struct sock_fprog_kern *fkprog; 1134 1135 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL); 1136 if (!fp->orig_prog) 1137 return -ENOMEM; 1138 1139 fkprog = fp->orig_prog; 1140 fkprog->len = fprog->len; 1141 1142 fkprog->filter = kmemdup(fp->insns, fsize, 1143 GFP_KERNEL | __GFP_NOWARN); 1144 if (!fkprog->filter) { 1145 kfree(fp->orig_prog); 1146 return -ENOMEM; 1147 } 1148 1149 return 0; 1150 } 1151 1152 static void bpf_release_orig_filter(struct bpf_prog *fp) 1153 { 1154 struct sock_fprog_kern *fprog = fp->orig_prog; 1155 1156 if (fprog) { 1157 kfree(fprog->filter); 1158 kfree(fprog); 1159 } 1160 } 1161 1162 static void __bpf_prog_release(struct bpf_prog *prog) 1163 { 1164 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) { 1165 bpf_prog_put(prog); 1166 } else { 1167 bpf_release_orig_filter(prog); 1168 bpf_prog_free(prog); 1169 } 1170 } 1171 1172 static void __sk_filter_release(struct sk_filter *fp) 1173 { 1174 __bpf_prog_release(fp->prog); 1175 kfree(fp); 1176 } 1177 1178 /** 1179 * sk_filter_release_rcu - Release a socket filter by rcu_head 1180 * @rcu: rcu_head that contains the sk_filter to free 1181 */ 1182 static void sk_filter_release_rcu(struct rcu_head *rcu) 1183 { 1184 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); 1185 1186 __sk_filter_release(fp); 1187 } 1188 1189 /** 1190 * sk_filter_release - release a socket filter 1191 * @fp: filter to remove 1192 * 1193 * Remove a filter from a socket and release its resources. 1194 */ 1195 static void sk_filter_release(struct sk_filter *fp) 1196 { 1197 if (refcount_dec_and_test(&fp->refcnt)) 1198 call_rcu(&fp->rcu, sk_filter_release_rcu); 1199 } 1200 1201 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1202 { 1203 u32 filter_size = bpf_prog_size(fp->prog->len); 1204 1205 atomic_sub(filter_size, &sk->sk_omem_alloc); 1206 sk_filter_release(fp); 1207 } 1208 1209 /* try to charge the socket memory if there is space available 1210 * return true on success 1211 */ 1212 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1213 { 1214 u32 filter_size = bpf_prog_size(fp->prog->len); 1215 1216 /* same check as in sock_kmalloc() */ 1217 if (filter_size <= sysctl_optmem_max && 1218 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) { 1219 atomic_add(filter_size, &sk->sk_omem_alloc); 1220 return true; 1221 } 1222 return false; 1223 } 1224 1225 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1226 { 1227 if (!refcount_inc_not_zero(&fp->refcnt)) 1228 return false; 1229 1230 if (!__sk_filter_charge(sk, fp)) { 1231 sk_filter_release(fp); 1232 return false; 1233 } 1234 return true; 1235 } 1236 1237 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp) 1238 { 1239 struct sock_filter *old_prog; 1240 struct bpf_prog *old_fp; 1241 int err, new_len, old_len = fp->len; 1242 bool seen_ld_abs = false; 1243 1244 /* We are free to overwrite insns et al right here as it 1245 * won't be used at this point in time anymore internally 1246 * after the migration to the internal BPF instruction 1247 * representation. 1248 */ 1249 BUILD_BUG_ON(sizeof(struct sock_filter) != 1250 sizeof(struct bpf_insn)); 1251 1252 /* Conversion cannot happen on overlapping memory areas, 1253 * so we need to keep the user BPF around until the 2nd 1254 * pass. At this time, the user BPF is stored in fp->insns. 1255 */ 1256 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter), 1257 GFP_KERNEL | __GFP_NOWARN); 1258 if (!old_prog) { 1259 err = -ENOMEM; 1260 goto out_err; 1261 } 1262 1263 /* 1st pass: calculate the new program length. */ 1264 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len, 1265 &seen_ld_abs); 1266 if (err) 1267 goto out_err_free; 1268 1269 /* Expand fp for appending the new filter representation. */ 1270 old_fp = fp; 1271 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0); 1272 if (!fp) { 1273 /* The old_fp is still around in case we couldn't 1274 * allocate new memory, so uncharge on that one. 1275 */ 1276 fp = old_fp; 1277 err = -ENOMEM; 1278 goto out_err_free; 1279 } 1280 1281 fp->len = new_len; 1282 1283 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */ 1284 err = bpf_convert_filter(old_prog, old_len, fp, &new_len, 1285 &seen_ld_abs); 1286 if (err) 1287 /* 2nd bpf_convert_filter() can fail only if it fails 1288 * to allocate memory, remapping must succeed. Note, 1289 * that at this time old_fp has already been released 1290 * by krealloc(). 1291 */ 1292 goto out_err_free; 1293 1294 fp = bpf_prog_select_runtime(fp, &err); 1295 if (err) 1296 goto out_err_free; 1297 1298 kfree(old_prog); 1299 return fp; 1300 1301 out_err_free: 1302 kfree(old_prog); 1303 out_err: 1304 __bpf_prog_release(fp); 1305 return ERR_PTR(err); 1306 } 1307 1308 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp, 1309 bpf_aux_classic_check_t trans) 1310 { 1311 int err; 1312 1313 fp->bpf_func = NULL; 1314 fp->jited = 0; 1315 1316 err = bpf_check_classic(fp->insns, fp->len); 1317 if (err) { 1318 __bpf_prog_release(fp); 1319 return ERR_PTR(err); 1320 } 1321 1322 /* There might be additional checks and transformations 1323 * needed on classic filters, f.e. in case of seccomp. 1324 */ 1325 if (trans) { 1326 err = trans(fp->insns, fp->len); 1327 if (err) { 1328 __bpf_prog_release(fp); 1329 return ERR_PTR(err); 1330 } 1331 } 1332 1333 /* Probe if we can JIT compile the filter and if so, do 1334 * the compilation of the filter. 1335 */ 1336 bpf_jit_compile(fp); 1337 1338 /* JIT compiler couldn't process this filter, so do the 1339 * internal BPF translation for the optimized interpreter. 1340 */ 1341 if (!fp->jited) 1342 fp = bpf_migrate_filter(fp); 1343 1344 return fp; 1345 } 1346 1347 /** 1348 * bpf_prog_create - create an unattached filter 1349 * @pfp: the unattached filter that is created 1350 * @fprog: the filter program 1351 * 1352 * Create a filter independent of any socket. We first run some 1353 * sanity checks on it to make sure it does not explode on us later. 1354 * If an error occurs or there is insufficient memory for the filter 1355 * a negative errno code is returned. On success the return is zero. 1356 */ 1357 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog) 1358 { 1359 unsigned int fsize = bpf_classic_proglen(fprog); 1360 struct bpf_prog *fp; 1361 1362 /* Make sure new filter is there and in the right amounts. */ 1363 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1364 return -EINVAL; 1365 1366 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1367 if (!fp) 1368 return -ENOMEM; 1369 1370 memcpy(fp->insns, fprog->filter, fsize); 1371 1372 fp->len = fprog->len; 1373 /* Since unattached filters are not copied back to user 1374 * space through sk_get_filter(), we do not need to hold 1375 * a copy here, and can spare us the work. 1376 */ 1377 fp->orig_prog = NULL; 1378 1379 /* bpf_prepare_filter() already takes care of freeing 1380 * memory in case something goes wrong. 1381 */ 1382 fp = bpf_prepare_filter(fp, NULL); 1383 if (IS_ERR(fp)) 1384 return PTR_ERR(fp); 1385 1386 *pfp = fp; 1387 return 0; 1388 } 1389 EXPORT_SYMBOL_GPL(bpf_prog_create); 1390 1391 /** 1392 * bpf_prog_create_from_user - create an unattached filter from user buffer 1393 * @pfp: the unattached filter that is created 1394 * @fprog: the filter program 1395 * @trans: post-classic verifier transformation handler 1396 * @save_orig: save classic BPF program 1397 * 1398 * This function effectively does the same as bpf_prog_create(), only 1399 * that it builds up its insns buffer from user space provided buffer. 1400 * It also allows for passing a bpf_aux_classic_check_t handler. 1401 */ 1402 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 1403 bpf_aux_classic_check_t trans, bool save_orig) 1404 { 1405 unsigned int fsize = bpf_classic_proglen(fprog); 1406 struct bpf_prog *fp; 1407 int err; 1408 1409 /* Make sure new filter is there and in the right amounts. */ 1410 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1411 return -EINVAL; 1412 1413 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1414 if (!fp) 1415 return -ENOMEM; 1416 1417 if (copy_from_user(fp->insns, fprog->filter, fsize)) { 1418 __bpf_prog_free(fp); 1419 return -EFAULT; 1420 } 1421 1422 fp->len = fprog->len; 1423 fp->orig_prog = NULL; 1424 1425 if (save_orig) { 1426 err = bpf_prog_store_orig_filter(fp, fprog); 1427 if (err) { 1428 __bpf_prog_free(fp); 1429 return -ENOMEM; 1430 } 1431 } 1432 1433 /* bpf_prepare_filter() already takes care of freeing 1434 * memory in case something goes wrong. 1435 */ 1436 fp = bpf_prepare_filter(fp, trans); 1437 if (IS_ERR(fp)) 1438 return PTR_ERR(fp); 1439 1440 *pfp = fp; 1441 return 0; 1442 } 1443 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user); 1444 1445 void bpf_prog_destroy(struct bpf_prog *fp) 1446 { 1447 __bpf_prog_release(fp); 1448 } 1449 EXPORT_SYMBOL_GPL(bpf_prog_destroy); 1450 1451 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk) 1452 { 1453 struct sk_filter *fp, *old_fp; 1454 1455 fp = kmalloc(sizeof(*fp), GFP_KERNEL); 1456 if (!fp) 1457 return -ENOMEM; 1458 1459 fp->prog = prog; 1460 1461 if (!__sk_filter_charge(sk, fp)) { 1462 kfree(fp); 1463 return -ENOMEM; 1464 } 1465 refcount_set(&fp->refcnt, 1); 1466 1467 old_fp = rcu_dereference_protected(sk->sk_filter, 1468 lockdep_sock_is_held(sk)); 1469 rcu_assign_pointer(sk->sk_filter, fp); 1470 1471 if (old_fp) 1472 sk_filter_uncharge(sk, old_fp); 1473 1474 return 0; 1475 } 1476 1477 static 1478 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk) 1479 { 1480 unsigned int fsize = bpf_classic_proglen(fprog); 1481 struct bpf_prog *prog; 1482 int err; 1483 1484 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1485 return ERR_PTR(-EPERM); 1486 1487 /* Make sure new filter is there and in the right amounts. */ 1488 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1489 return ERR_PTR(-EINVAL); 1490 1491 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1492 if (!prog) 1493 return ERR_PTR(-ENOMEM); 1494 1495 if (copy_from_user(prog->insns, fprog->filter, fsize)) { 1496 __bpf_prog_free(prog); 1497 return ERR_PTR(-EFAULT); 1498 } 1499 1500 prog->len = fprog->len; 1501 1502 err = bpf_prog_store_orig_filter(prog, fprog); 1503 if (err) { 1504 __bpf_prog_free(prog); 1505 return ERR_PTR(-ENOMEM); 1506 } 1507 1508 /* bpf_prepare_filter() already takes care of freeing 1509 * memory in case something goes wrong. 1510 */ 1511 return bpf_prepare_filter(prog, NULL); 1512 } 1513 1514 /** 1515 * sk_attach_filter - attach a socket filter 1516 * @fprog: the filter program 1517 * @sk: the socket to use 1518 * 1519 * Attach the user's filter code. We first run some sanity checks on 1520 * it to make sure it does not explode on us later. If an error 1521 * occurs or there is insufficient memory for the filter a negative 1522 * errno code is returned. On success the return is zero. 1523 */ 1524 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1525 { 1526 struct bpf_prog *prog = __get_filter(fprog, sk); 1527 int err; 1528 1529 if (IS_ERR(prog)) 1530 return PTR_ERR(prog); 1531 1532 err = __sk_attach_prog(prog, sk); 1533 if (err < 0) { 1534 __bpf_prog_release(prog); 1535 return err; 1536 } 1537 1538 return 0; 1539 } 1540 EXPORT_SYMBOL_GPL(sk_attach_filter); 1541 1542 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1543 { 1544 struct bpf_prog *prog = __get_filter(fprog, sk); 1545 int err; 1546 1547 if (IS_ERR(prog)) 1548 return PTR_ERR(prog); 1549 1550 if (bpf_prog_size(prog->len) > sysctl_optmem_max) 1551 err = -ENOMEM; 1552 else 1553 err = reuseport_attach_prog(sk, prog); 1554 1555 if (err) 1556 __bpf_prog_release(prog); 1557 1558 return err; 1559 } 1560 1561 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk) 1562 { 1563 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1564 return ERR_PTR(-EPERM); 1565 1566 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1567 } 1568 1569 int sk_attach_bpf(u32 ufd, struct sock *sk) 1570 { 1571 struct bpf_prog *prog = __get_bpf(ufd, sk); 1572 int err; 1573 1574 if (IS_ERR(prog)) 1575 return PTR_ERR(prog); 1576 1577 err = __sk_attach_prog(prog, sk); 1578 if (err < 0) { 1579 bpf_prog_put(prog); 1580 return err; 1581 } 1582 1583 return 0; 1584 } 1585 1586 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk) 1587 { 1588 struct bpf_prog *prog; 1589 int err; 1590 1591 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1592 return -EPERM; 1593 1594 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1595 if (PTR_ERR(prog) == -EINVAL) 1596 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT); 1597 if (IS_ERR(prog)) 1598 return PTR_ERR(prog); 1599 1600 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) { 1601 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER 1602 * bpf prog (e.g. sockmap). It depends on the 1603 * limitation imposed by bpf_prog_load(). 1604 * Hence, sysctl_optmem_max is not checked. 1605 */ 1606 if ((sk->sk_type != SOCK_STREAM && 1607 sk->sk_type != SOCK_DGRAM) || 1608 (sk->sk_protocol != IPPROTO_UDP && 1609 sk->sk_protocol != IPPROTO_TCP) || 1610 (sk->sk_family != AF_INET && 1611 sk->sk_family != AF_INET6)) { 1612 err = -ENOTSUPP; 1613 goto err_prog_put; 1614 } 1615 } else { 1616 /* BPF_PROG_TYPE_SOCKET_FILTER */ 1617 if (bpf_prog_size(prog->len) > sysctl_optmem_max) { 1618 err = -ENOMEM; 1619 goto err_prog_put; 1620 } 1621 } 1622 1623 err = reuseport_attach_prog(sk, prog); 1624 err_prog_put: 1625 if (err) 1626 bpf_prog_put(prog); 1627 1628 return err; 1629 } 1630 1631 void sk_reuseport_prog_free(struct bpf_prog *prog) 1632 { 1633 if (!prog) 1634 return; 1635 1636 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) 1637 bpf_prog_put(prog); 1638 else 1639 bpf_prog_destroy(prog); 1640 } 1641 1642 struct bpf_scratchpad { 1643 union { 1644 __be32 diff[MAX_BPF_STACK / sizeof(__be32)]; 1645 u8 buff[MAX_BPF_STACK]; 1646 }; 1647 }; 1648 1649 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp); 1650 1651 static inline int __bpf_try_make_writable(struct sk_buff *skb, 1652 unsigned int write_len) 1653 { 1654 return skb_ensure_writable(skb, write_len); 1655 } 1656 1657 static inline int bpf_try_make_writable(struct sk_buff *skb, 1658 unsigned int write_len) 1659 { 1660 int err = __bpf_try_make_writable(skb, write_len); 1661 1662 bpf_compute_data_pointers(skb); 1663 return err; 1664 } 1665 1666 static int bpf_try_make_head_writable(struct sk_buff *skb) 1667 { 1668 return bpf_try_make_writable(skb, skb_headlen(skb)); 1669 } 1670 1671 static inline void bpf_push_mac_rcsum(struct sk_buff *skb) 1672 { 1673 if (skb_at_tc_ingress(skb)) 1674 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1675 } 1676 1677 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb) 1678 { 1679 if (skb_at_tc_ingress(skb)) 1680 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1681 } 1682 1683 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset, 1684 const void *, from, u32, len, u64, flags) 1685 { 1686 void *ptr; 1687 1688 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH))) 1689 return -EINVAL; 1690 if (unlikely(offset > 0xffff)) 1691 return -EFAULT; 1692 if (unlikely(bpf_try_make_writable(skb, offset + len))) 1693 return -EFAULT; 1694 1695 ptr = skb->data + offset; 1696 if (flags & BPF_F_RECOMPUTE_CSUM) 1697 __skb_postpull_rcsum(skb, ptr, len, offset); 1698 1699 memcpy(ptr, from, len); 1700 1701 if (flags & BPF_F_RECOMPUTE_CSUM) 1702 __skb_postpush_rcsum(skb, ptr, len, offset); 1703 if (flags & BPF_F_INVALIDATE_HASH) 1704 skb_clear_hash(skb); 1705 1706 return 0; 1707 } 1708 1709 static const struct bpf_func_proto bpf_skb_store_bytes_proto = { 1710 .func = bpf_skb_store_bytes, 1711 .gpl_only = false, 1712 .ret_type = RET_INTEGER, 1713 .arg1_type = ARG_PTR_TO_CTX, 1714 .arg2_type = ARG_ANYTHING, 1715 .arg3_type = ARG_PTR_TO_MEM, 1716 .arg4_type = ARG_CONST_SIZE, 1717 .arg5_type = ARG_ANYTHING, 1718 }; 1719 1720 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset, 1721 void *, to, u32, len) 1722 { 1723 void *ptr; 1724 1725 if (unlikely(offset > 0xffff)) 1726 goto err_clear; 1727 1728 ptr = skb_header_pointer(skb, offset, len, to); 1729 if (unlikely(!ptr)) 1730 goto err_clear; 1731 if (ptr != to) 1732 memcpy(to, ptr, len); 1733 1734 return 0; 1735 err_clear: 1736 memset(to, 0, len); 1737 return -EFAULT; 1738 } 1739 1740 static const struct bpf_func_proto bpf_skb_load_bytes_proto = { 1741 .func = bpf_skb_load_bytes, 1742 .gpl_only = false, 1743 .ret_type = RET_INTEGER, 1744 .arg1_type = ARG_PTR_TO_CTX, 1745 .arg2_type = ARG_ANYTHING, 1746 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1747 .arg4_type = ARG_CONST_SIZE, 1748 }; 1749 1750 BPF_CALL_4(bpf_flow_dissector_load_bytes, 1751 const struct bpf_flow_dissector *, ctx, u32, offset, 1752 void *, to, u32, len) 1753 { 1754 void *ptr; 1755 1756 if (unlikely(offset > 0xffff)) 1757 goto err_clear; 1758 1759 if (unlikely(!ctx->skb)) 1760 goto err_clear; 1761 1762 ptr = skb_header_pointer(ctx->skb, offset, len, to); 1763 if (unlikely(!ptr)) 1764 goto err_clear; 1765 if (ptr != to) 1766 memcpy(to, ptr, len); 1767 1768 return 0; 1769 err_clear: 1770 memset(to, 0, len); 1771 return -EFAULT; 1772 } 1773 1774 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = { 1775 .func = bpf_flow_dissector_load_bytes, 1776 .gpl_only = false, 1777 .ret_type = RET_INTEGER, 1778 .arg1_type = ARG_PTR_TO_CTX, 1779 .arg2_type = ARG_ANYTHING, 1780 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1781 .arg4_type = ARG_CONST_SIZE, 1782 }; 1783 1784 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb, 1785 u32, offset, void *, to, u32, len, u32, start_header) 1786 { 1787 u8 *end = skb_tail_pointer(skb); 1788 u8 *start, *ptr; 1789 1790 if (unlikely(offset > 0xffff)) 1791 goto err_clear; 1792 1793 switch (start_header) { 1794 case BPF_HDR_START_MAC: 1795 if (unlikely(!skb_mac_header_was_set(skb))) 1796 goto err_clear; 1797 start = skb_mac_header(skb); 1798 break; 1799 case BPF_HDR_START_NET: 1800 start = skb_network_header(skb); 1801 break; 1802 default: 1803 goto err_clear; 1804 } 1805 1806 ptr = start + offset; 1807 1808 if (likely(ptr + len <= end)) { 1809 memcpy(to, ptr, len); 1810 return 0; 1811 } 1812 1813 err_clear: 1814 memset(to, 0, len); 1815 return -EFAULT; 1816 } 1817 1818 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = { 1819 .func = bpf_skb_load_bytes_relative, 1820 .gpl_only = false, 1821 .ret_type = RET_INTEGER, 1822 .arg1_type = ARG_PTR_TO_CTX, 1823 .arg2_type = ARG_ANYTHING, 1824 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1825 .arg4_type = ARG_CONST_SIZE, 1826 .arg5_type = ARG_ANYTHING, 1827 }; 1828 1829 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len) 1830 { 1831 /* Idea is the following: should the needed direct read/write 1832 * test fail during runtime, we can pull in more data and redo 1833 * again, since implicitly, we invalidate previous checks here. 1834 * 1835 * Or, since we know how much we need to make read/writeable, 1836 * this can be done once at the program beginning for direct 1837 * access case. By this we overcome limitations of only current 1838 * headroom being accessible. 1839 */ 1840 return bpf_try_make_writable(skb, len ? : skb_headlen(skb)); 1841 } 1842 1843 static const struct bpf_func_proto bpf_skb_pull_data_proto = { 1844 .func = bpf_skb_pull_data, 1845 .gpl_only = false, 1846 .ret_type = RET_INTEGER, 1847 .arg1_type = ARG_PTR_TO_CTX, 1848 .arg2_type = ARG_ANYTHING, 1849 }; 1850 1851 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk) 1852 { 1853 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL; 1854 } 1855 1856 static const struct bpf_func_proto bpf_sk_fullsock_proto = { 1857 .func = bpf_sk_fullsock, 1858 .gpl_only = false, 1859 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 1860 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 1861 }; 1862 1863 static inline int sk_skb_try_make_writable(struct sk_buff *skb, 1864 unsigned int write_len) 1865 { 1866 int err = __bpf_try_make_writable(skb, write_len); 1867 1868 bpf_compute_data_end_sk_skb(skb); 1869 return err; 1870 } 1871 1872 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len) 1873 { 1874 /* Idea is the following: should the needed direct read/write 1875 * test fail during runtime, we can pull in more data and redo 1876 * again, since implicitly, we invalidate previous checks here. 1877 * 1878 * Or, since we know how much we need to make read/writeable, 1879 * this can be done once at the program beginning for direct 1880 * access case. By this we overcome limitations of only current 1881 * headroom being accessible. 1882 */ 1883 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb)); 1884 } 1885 1886 static const struct bpf_func_proto sk_skb_pull_data_proto = { 1887 .func = sk_skb_pull_data, 1888 .gpl_only = false, 1889 .ret_type = RET_INTEGER, 1890 .arg1_type = ARG_PTR_TO_CTX, 1891 .arg2_type = ARG_ANYTHING, 1892 }; 1893 1894 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset, 1895 u64, from, u64, to, u64, flags) 1896 { 1897 __sum16 *ptr; 1898 1899 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK))) 1900 return -EINVAL; 1901 if (unlikely(offset > 0xffff || offset & 1)) 1902 return -EFAULT; 1903 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1904 return -EFAULT; 1905 1906 ptr = (__sum16 *)(skb->data + offset); 1907 switch (flags & BPF_F_HDR_FIELD_MASK) { 1908 case 0: 1909 if (unlikely(from != 0)) 1910 return -EINVAL; 1911 1912 csum_replace_by_diff(ptr, to); 1913 break; 1914 case 2: 1915 csum_replace2(ptr, from, to); 1916 break; 1917 case 4: 1918 csum_replace4(ptr, from, to); 1919 break; 1920 default: 1921 return -EINVAL; 1922 } 1923 1924 return 0; 1925 } 1926 1927 static const struct bpf_func_proto bpf_l3_csum_replace_proto = { 1928 .func = bpf_l3_csum_replace, 1929 .gpl_only = false, 1930 .ret_type = RET_INTEGER, 1931 .arg1_type = ARG_PTR_TO_CTX, 1932 .arg2_type = ARG_ANYTHING, 1933 .arg3_type = ARG_ANYTHING, 1934 .arg4_type = ARG_ANYTHING, 1935 .arg5_type = ARG_ANYTHING, 1936 }; 1937 1938 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset, 1939 u64, from, u64, to, u64, flags) 1940 { 1941 bool is_pseudo = flags & BPF_F_PSEUDO_HDR; 1942 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0; 1943 bool do_mforce = flags & BPF_F_MARK_ENFORCE; 1944 __sum16 *ptr; 1945 1946 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE | 1947 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK))) 1948 return -EINVAL; 1949 if (unlikely(offset > 0xffff || offset & 1)) 1950 return -EFAULT; 1951 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1952 return -EFAULT; 1953 1954 ptr = (__sum16 *)(skb->data + offset); 1955 if (is_mmzero && !do_mforce && !*ptr) 1956 return 0; 1957 1958 switch (flags & BPF_F_HDR_FIELD_MASK) { 1959 case 0: 1960 if (unlikely(from != 0)) 1961 return -EINVAL; 1962 1963 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo); 1964 break; 1965 case 2: 1966 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo); 1967 break; 1968 case 4: 1969 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo); 1970 break; 1971 default: 1972 return -EINVAL; 1973 } 1974 1975 if (is_mmzero && !*ptr) 1976 *ptr = CSUM_MANGLED_0; 1977 return 0; 1978 } 1979 1980 static const struct bpf_func_proto bpf_l4_csum_replace_proto = { 1981 .func = bpf_l4_csum_replace, 1982 .gpl_only = false, 1983 .ret_type = RET_INTEGER, 1984 .arg1_type = ARG_PTR_TO_CTX, 1985 .arg2_type = ARG_ANYTHING, 1986 .arg3_type = ARG_ANYTHING, 1987 .arg4_type = ARG_ANYTHING, 1988 .arg5_type = ARG_ANYTHING, 1989 }; 1990 1991 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size, 1992 __be32 *, to, u32, to_size, __wsum, seed) 1993 { 1994 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp); 1995 u32 diff_size = from_size + to_size; 1996 int i, j = 0; 1997 1998 /* This is quite flexible, some examples: 1999 * 2000 * from_size == 0, to_size > 0, seed := csum --> pushing data 2001 * from_size > 0, to_size == 0, seed := csum --> pulling data 2002 * from_size > 0, to_size > 0, seed := 0 --> diffing data 2003 * 2004 * Even for diffing, from_size and to_size don't need to be equal. 2005 */ 2006 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) || 2007 diff_size > sizeof(sp->diff))) 2008 return -EINVAL; 2009 2010 for (i = 0; i < from_size / sizeof(__be32); i++, j++) 2011 sp->diff[j] = ~from[i]; 2012 for (i = 0; i < to_size / sizeof(__be32); i++, j++) 2013 sp->diff[j] = to[i]; 2014 2015 return csum_partial(sp->diff, diff_size, seed); 2016 } 2017 2018 static const struct bpf_func_proto bpf_csum_diff_proto = { 2019 .func = bpf_csum_diff, 2020 .gpl_only = false, 2021 .pkt_access = true, 2022 .ret_type = RET_INTEGER, 2023 .arg1_type = ARG_PTR_TO_MEM_OR_NULL, 2024 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 2025 .arg3_type = ARG_PTR_TO_MEM_OR_NULL, 2026 .arg4_type = ARG_CONST_SIZE_OR_ZERO, 2027 .arg5_type = ARG_ANYTHING, 2028 }; 2029 2030 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum) 2031 { 2032 /* The interface is to be used in combination with bpf_csum_diff() 2033 * for direct packet writes. csum rotation for alignment as well 2034 * as emulating csum_sub() can be done from the eBPF program. 2035 */ 2036 if (skb->ip_summed == CHECKSUM_COMPLETE) 2037 return (skb->csum = csum_add(skb->csum, csum)); 2038 2039 return -ENOTSUPP; 2040 } 2041 2042 static const struct bpf_func_proto bpf_csum_update_proto = { 2043 .func = bpf_csum_update, 2044 .gpl_only = false, 2045 .ret_type = RET_INTEGER, 2046 .arg1_type = ARG_PTR_TO_CTX, 2047 .arg2_type = ARG_ANYTHING, 2048 }; 2049 2050 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level) 2051 { 2052 /* The interface is to be used in combination with bpf_skb_adjust_room() 2053 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET 2054 * is passed as flags, for example. 2055 */ 2056 switch (level) { 2057 case BPF_CSUM_LEVEL_INC: 2058 __skb_incr_checksum_unnecessary(skb); 2059 break; 2060 case BPF_CSUM_LEVEL_DEC: 2061 __skb_decr_checksum_unnecessary(skb); 2062 break; 2063 case BPF_CSUM_LEVEL_RESET: 2064 __skb_reset_checksum_unnecessary(skb); 2065 break; 2066 case BPF_CSUM_LEVEL_QUERY: 2067 return skb->ip_summed == CHECKSUM_UNNECESSARY ? 2068 skb->csum_level : -EACCES; 2069 default: 2070 return -EINVAL; 2071 } 2072 2073 return 0; 2074 } 2075 2076 static const struct bpf_func_proto bpf_csum_level_proto = { 2077 .func = bpf_csum_level, 2078 .gpl_only = false, 2079 .ret_type = RET_INTEGER, 2080 .arg1_type = ARG_PTR_TO_CTX, 2081 .arg2_type = ARG_ANYTHING, 2082 }; 2083 2084 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb) 2085 { 2086 return dev_forward_skb(dev, skb); 2087 } 2088 2089 static inline int __bpf_rx_skb_no_mac(struct net_device *dev, 2090 struct sk_buff *skb) 2091 { 2092 int ret = ____dev_forward_skb(dev, skb); 2093 2094 if (likely(!ret)) { 2095 skb->dev = dev; 2096 ret = netif_rx(skb); 2097 } 2098 2099 return ret; 2100 } 2101 2102 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb) 2103 { 2104 int ret; 2105 2106 if (dev_xmit_recursion()) { 2107 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2108 kfree_skb(skb); 2109 return -ENETDOWN; 2110 } 2111 2112 skb->dev = dev; 2113 skb->tstamp = 0; 2114 2115 dev_xmit_recursion_inc(); 2116 ret = dev_queue_xmit(skb); 2117 dev_xmit_recursion_dec(); 2118 2119 return ret; 2120 } 2121 2122 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev, 2123 u32 flags) 2124 { 2125 unsigned int mlen = skb_network_offset(skb); 2126 2127 if (mlen) { 2128 __skb_pull(skb, mlen); 2129 2130 /* At ingress, the mac header has already been pulled once. 2131 * At egress, skb_pospull_rcsum has to be done in case that 2132 * the skb is originated from ingress (i.e. a forwarded skb) 2133 * to ensure that rcsum starts at net header. 2134 */ 2135 if (!skb_at_tc_ingress(skb)) 2136 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen); 2137 } 2138 skb_pop_mac_header(skb); 2139 skb_reset_mac_len(skb); 2140 return flags & BPF_F_INGRESS ? 2141 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb); 2142 } 2143 2144 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev, 2145 u32 flags) 2146 { 2147 /* Verify that a link layer header is carried */ 2148 if (unlikely(skb->mac_header >= skb->network_header)) { 2149 kfree_skb(skb); 2150 return -ERANGE; 2151 } 2152 2153 bpf_push_mac_rcsum(skb); 2154 return flags & BPF_F_INGRESS ? 2155 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb); 2156 } 2157 2158 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev, 2159 u32 flags) 2160 { 2161 if (dev_is_mac_header_xmit(dev)) 2162 return __bpf_redirect_common(skb, dev, flags); 2163 else 2164 return __bpf_redirect_no_mac(skb, dev, flags); 2165 } 2166 2167 #if IS_ENABLED(CONFIG_IPV6) 2168 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb, 2169 struct net_device *dev, struct bpf_nh_params *nh) 2170 { 2171 u32 hh_len = LL_RESERVED_SPACE(dev); 2172 const struct in6_addr *nexthop; 2173 struct dst_entry *dst = NULL; 2174 struct neighbour *neigh; 2175 2176 if (dev_xmit_recursion()) { 2177 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2178 goto out_drop; 2179 } 2180 2181 skb->dev = dev; 2182 skb->tstamp = 0; 2183 2184 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 2185 struct sk_buff *skb2; 2186 2187 skb2 = skb_realloc_headroom(skb, hh_len); 2188 if (unlikely(!skb2)) { 2189 kfree_skb(skb); 2190 return -ENOMEM; 2191 } 2192 if (skb->sk) 2193 skb_set_owner_w(skb2, skb->sk); 2194 consume_skb(skb); 2195 skb = skb2; 2196 } 2197 2198 rcu_read_lock_bh(); 2199 if (!nh) { 2200 dst = skb_dst(skb); 2201 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst), 2202 &ipv6_hdr(skb)->daddr); 2203 } else { 2204 nexthop = &nh->ipv6_nh; 2205 } 2206 neigh = ip_neigh_gw6(dev, nexthop); 2207 if (likely(!IS_ERR(neigh))) { 2208 int ret; 2209 2210 sock_confirm_neigh(skb, neigh); 2211 dev_xmit_recursion_inc(); 2212 ret = neigh_output(neigh, skb, false); 2213 dev_xmit_recursion_dec(); 2214 rcu_read_unlock_bh(); 2215 return ret; 2216 } 2217 rcu_read_unlock_bh(); 2218 if (dst) 2219 IP6_INC_STATS(dev_net(dst->dev), 2220 ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 2221 out_drop: 2222 kfree_skb(skb); 2223 return -ENETDOWN; 2224 } 2225 2226 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev, 2227 struct bpf_nh_params *nh) 2228 { 2229 const struct ipv6hdr *ip6h = ipv6_hdr(skb); 2230 struct net *net = dev_net(dev); 2231 int err, ret = NET_XMIT_DROP; 2232 2233 if (!nh) { 2234 struct dst_entry *dst; 2235 struct flowi6 fl6 = { 2236 .flowi6_flags = FLOWI_FLAG_ANYSRC, 2237 .flowi6_mark = skb->mark, 2238 .flowlabel = ip6_flowinfo(ip6h), 2239 .flowi6_oif = dev->ifindex, 2240 .flowi6_proto = ip6h->nexthdr, 2241 .daddr = ip6h->daddr, 2242 .saddr = ip6h->saddr, 2243 }; 2244 2245 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL); 2246 if (IS_ERR(dst)) 2247 goto out_drop; 2248 2249 skb_dst_set(skb, dst); 2250 } else if (nh->nh_family != AF_INET6) { 2251 goto out_drop; 2252 } 2253 2254 err = bpf_out_neigh_v6(net, skb, dev, nh); 2255 if (unlikely(net_xmit_eval(err))) 2256 dev->stats.tx_errors++; 2257 else 2258 ret = NET_XMIT_SUCCESS; 2259 goto out_xmit; 2260 out_drop: 2261 dev->stats.tx_errors++; 2262 kfree_skb(skb); 2263 out_xmit: 2264 return ret; 2265 } 2266 #else 2267 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev, 2268 struct bpf_nh_params *nh) 2269 { 2270 kfree_skb(skb); 2271 return NET_XMIT_DROP; 2272 } 2273 #endif /* CONFIG_IPV6 */ 2274 2275 #if IS_ENABLED(CONFIG_INET) 2276 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb, 2277 struct net_device *dev, struct bpf_nh_params *nh) 2278 { 2279 u32 hh_len = LL_RESERVED_SPACE(dev); 2280 struct neighbour *neigh; 2281 bool is_v6gw = false; 2282 2283 if (dev_xmit_recursion()) { 2284 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2285 goto out_drop; 2286 } 2287 2288 skb->dev = dev; 2289 skb->tstamp = 0; 2290 2291 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 2292 struct sk_buff *skb2; 2293 2294 skb2 = skb_realloc_headroom(skb, hh_len); 2295 if (unlikely(!skb2)) { 2296 kfree_skb(skb); 2297 return -ENOMEM; 2298 } 2299 if (skb->sk) 2300 skb_set_owner_w(skb2, skb->sk); 2301 consume_skb(skb); 2302 skb = skb2; 2303 } 2304 2305 rcu_read_lock_bh(); 2306 if (!nh) { 2307 struct dst_entry *dst = skb_dst(skb); 2308 struct rtable *rt = container_of(dst, struct rtable, dst); 2309 2310 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 2311 } else if (nh->nh_family == AF_INET6) { 2312 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh); 2313 is_v6gw = true; 2314 } else if (nh->nh_family == AF_INET) { 2315 neigh = ip_neigh_gw4(dev, nh->ipv4_nh); 2316 } else { 2317 rcu_read_unlock_bh(); 2318 goto out_drop; 2319 } 2320 2321 if (likely(!IS_ERR(neigh))) { 2322 int ret; 2323 2324 sock_confirm_neigh(skb, neigh); 2325 dev_xmit_recursion_inc(); 2326 ret = neigh_output(neigh, skb, is_v6gw); 2327 dev_xmit_recursion_dec(); 2328 rcu_read_unlock_bh(); 2329 return ret; 2330 } 2331 rcu_read_unlock_bh(); 2332 out_drop: 2333 kfree_skb(skb); 2334 return -ENETDOWN; 2335 } 2336 2337 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev, 2338 struct bpf_nh_params *nh) 2339 { 2340 const struct iphdr *ip4h = ip_hdr(skb); 2341 struct net *net = dev_net(dev); 2342 int err, ret = NET_XMIT_DROP; 2343 2344 if (!nh) { 2345 struct flowi4 fl4 = { 2346 .flowi4_flags = FLOWI_FLAG_ANYSRC, 2347 .flowi4_mark = skb->mark, 2348 .flowi4_tos = RT_TOS(ip4h->tos), 2349 .flowi4_oif = dev->ifindex, 2350 .flowi4_proto = ip4h->protocol, 2351 .daddr = ip4h->daddr, 2352 .saddr = ip4h->saddr, 2353 }; 2354 struct rtable *rt; 2355 2356 rt = ip_route_output_flow(net, &fl4, NULL); 2357 if (IS_ERR(rt)) 2358 goto out_drop; 2359 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { 2360 ip_rt_put(rt); 2361 goto out_drop; 2362 } 2363 2364 skb_dst_set(skb, &rt->dst); 2365 } 2366 2367 err = bpf_out_neigh_v4(net, skb, dev, nh); 2368 if (unlikely(net_xmit_eval(err))) 2369 dev->stats.tx_errors++; 2370 else 2371 ret = NET_XMIT_SUCCESS; 2372 goto out_xmit; 2373 out_drop: 2374 dev->stats.tx_errors++; 2375 kfree_skb(skb); 2376 out_xmit: 2377 return ret; 2378 } 2379 #else 2380 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev, 2381 struct bpf_nh_params *nh) 2382 { 2383 kfree_skb(skb); 2384 return NET_XMIT_DROP; 2385 } 2386 #endif /* CONFIG_INET */ 2387 2388 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev, 2389 struct bpf_nh_params *nh) 2390 { 2391 struct ethhdr *ethh = eth_hdr(skb); 2392 2393 if (unlikely(skb->mac_header >= skb->network_header)) 2394 goto out; 2395 bpf_push_mac_rcsum(skb); 2396 if (is_multicast_ether_addr(ethh->h_dest)) 2397 goto out; 2398 2399 skb_pull(skb, sizeof(*ethh)); 2400 skb_unset_mac_header(skb); 2401 skb_reset_network_header(skb); 2402 2403 if (skb->protocol == htons(ETH_P_IP)) 2404 return __bpf_redirect_neigh_v4(skb, dev, nh); 2405 else if (skb->protocol == htons(ETH_P_IPV6)) 2406 return __bpf_redirect_neigh_v6(skb, dev, nh); 2407 out: 2408 kfree_skb(skb); 2409 return -ENOTSUPP; 2410 } 2411 2412 /* Internal, non-exposed redirect flags. */ 2413 enum { 2414 BPF_F_NEIGH = (1ULL << 1), 2415 BPF_F_PEER = (1ULL << 2), 2416 BPF_F_NEXTHOP = (1ULL << 3), 2417 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP) 2418 }; 2419 2420 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags) 2421 { 2422 struct net_device *dev; 2423 struct sk_buff *clone; 2424 int ret; 2425 2426 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL))) 2427 return -EINVAL; 2428 2429 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex); 2430 if (unlikely(!dev)) 2431 return -EINVAL; 2432 2433 clone = skb_clone(skb, GFP_ATOMIC); 2434 if (unlikely(!clone)) 2435 return -ENOMEM; 2436 2437 /* For direct write, we need to keep the invariant that the skbs 2438 * we're dealing with need to be uncloned. Should uncloning fail 2439 * here, we need to free the just generated clone to unclone once 2440 * again. 2441 */ 2442 ret = bpf_try_make_head_writable(skb); 2443 if (unlikely(ret)) { 2444 kfree_skb(clone); 2445 return -ENOMEM; 2446 } 2447 2448 return __bpf_redirect(clone, dev, flags); 2449 } 2450 2451 static const struct bpf_func_proto bpf_clone_redirect_proto = { 2452 .func = bpf_clone_redirect, 2453 .gpl_only = false, 2454 .ret_type = RET_INTEGER, 2455 .arg1_type = ARG_PTR_TO_CTX, 2456 .arg2_type = ARG_ANYTHING, 2457 .arg3_type = ARG_ANYTHING, 2458 }; 2459 2460 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 2461 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info); 2462 2463 int skb_do_redirect(struct sk_buff *skb) 2464 { 2465 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2466 struct net *net = dev_net(skb->dev); 2467 struct net_device *dev; 2468 u32 flags = ri->flags; 2469 2470 dev = dev_get_by_index_rcu(net, ri->tgt_index); 2471 ri->tgt_index = 0; 2472 ri->flags = 0; 2473 if (unlikely(!dev)) 2474 goto out_drop; 2475 if (flags & BPF_F_PEER) { 2476 const struct net_device_ops *ops = dev->netdev_ops; 2477 2478 if (unlikely(!ops->ndo_get_peer_dev || 2479 !skb_at_tc_ingress(skb))) 2480 goto out_drop; 2481 dev = ops->ndo_get_peer_dev(dev); 2482 if (unlikely(!dev || 2483 !is_skb_forwardable(dev, skb) || 2484 net_eq(net, dev_net(dev)))) 2485 goto out_drop; 2486 skb->dev = dev; 2487 return -EAGAIN; 2488 } 2489 return flags & BPF_F_NEIGH ? 2490 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ? 2491 &ri->nh : NULL) : 2492 __bpf_redirect(skb, dev, flags); 2493 out_drop: 2494 kfree_skb(skb); 2495 return -EINVAL; 2496 } 2497 2498 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags) 2499 { 2500 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2501 2502 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL))) 2503 return TC_ACT_SHOT; 2504 2505 ri->flags = flags; 2506 ri->tgt_index = ifindex; 2507 2508 return TC_ACT_REDIRECT; 2509 } 2510 2511 static const struct bpf_func_proto bpf_redirect_proto = { 2512 .func = bpf_redirect, 2513 .gpl_only = false, 2514 .ret_type = RET_INTEGER, 2515 .arg1_type = ARG_ANYTHING, 2516 .arg2_type = ARG_ANYTHING, 2517 }; 2518 2519 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags) 2520 { 2521 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2522 2523 if (unlikely(flags)) 2524 return TC_ACT_SHOT; 2525 2526 ri->flags = BPF_F_PEER; 2527 ri->tgt_index = ifindex; 2528 2529 return TC_ACT_REDIRECT; 2530 } 2531 2532 static const struct bpf_func_proto bpf_redirect_peer_proto = { 2533 .func = bpf_redirect_peer, 2534 .gpl_only = false, 2535 .ret_type = RET_INTEGER, 2536 .arg1_type = ARG_ANYTHING, 2537 .arg2_type = ARG_ANYTHING, 2538 }; 2539 2540 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params, 2541 int, plen, u64, flags) 2542 { 2543 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2544 2545 if (unlikely((plen && plen < sizeof(*params)) || flags)) 2546 return TC_ACT_SHOT; 2547 2548 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0); 2549 ri->tgt_index = ifindex; 2550 2551 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params)); 2552 if (plen) 2553 memcpy(&ri->nh, params, sizeof(ri->nh)); 2554 2555 return TC_ACT_REDIRECT; 2556 } 2557 2558 static const struct bpf_func_proto bpf_redirect_neigh_proto = { 2559 .func = bpf_redirect_neigh, 2560 .gpl_only = false, 2561 .ret_type = RET_INTEGER, 2562 .arg1_type = ARG_ANYTHING, 2563 .arg2_type = ARG_PTR_TO_MEM_OR_NULL, 2564 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 2565 .arg4_type = ARG_ANYTHING, 2566 }; 2567 2568 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes) 2569 { 2570 msg->apply_bytes = bytes; 2571 return 0; 2572 } 2573 2574 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = { 2575 .func = bpf_msg_apply_bytes, 2576 .gpl_only = false, 2577 .ret_type = RET_INTEGER, 2578 .arg1_type = ARG_PTR_TO_CTX, 2579 .arg2_type = ARG_ANYTHING, 2580 }; 2581 2582 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes) 2583 { 2584 msg->cork_bytes = bytes; 2585 return 0; 2586 } 2587 2588 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = { 2589 .func = bpf_msg_cork_bytes, 2590 .gpl_only = false, 2591 .ret_type = RET_INTEGER, 2592 .arg1_type = ARG_PTR_TO_CTX, 2593 .arg2_type = ARG_ANYTHING, 2594 }; 2595 2596 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start, 2597 u32, end, u64, flags) 2598 { 2599 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start; 2600 u32 first_sge, last_sge, i, shift, bytes_sg_total; 2601 struct scatterlist *sge; 2602 u8 *raw, *to, *from; 2603 struct page *page; 2604 2605 if (unlikely(flags || end <= start)) 2606 return -EINVAL; 2607 2608 /* First find the starting scatterlist element */ 2609 i = msg->sg.start; 2610 do { 2611 offset += len; 2612 len = sk_msg_elem(msg, i)->length; 2613 if (start < offset + len) 2614 break; 2615 sk_msg_iter_var_next(i); 2616 } while (i != msg->sg.end); 2617 2618 if (unlikely(start >= offset + len)) 2619 return -EINVAL; 2620 2621 first_sge = i; 2622 /* The start may point into the sg element so we need to also 2623 * account for the headroom. 2624 */ 2625 bytes_sg_total = start - offset + bytes; 2626 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len) 2627 goto out; 2628 2629 /* At this point we need to linearize multiple scatterlist 2630 * elements or a single shared page. Either way we need to 2631 * copy into a linear buffer exclusively owned by BPF. Then 2632 * place the buffer in the scatterlist and fixup the original 2633 * entries by removing the entries now in the linear buffer 2634 * and shifting the remaining entries. For now we do not try 2635 * to copy partial entries to avoid complexity of running out 2636 * of sg_entry slots. The downside is reading a single byte 2637 * will copy the entire sg entry. 2638 */ 2639 do { 2640 copy += sk_msg_elem(msg, i)->length; 2641 sk_msg_iter_var_next(i); 2642 if (bytes_sg_total <= copy) 2643 break; 2644 } while (i != msg->sg.end); 2645 last_sge = i; 2646 2647 if (unlikely(bytes_sg_total > copy)) 2648 return -EINVAL; 2649 2650 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2651 get_order(copy)); 2652 if (unlikely(!page)) 2653 return -ENOMEM; 2654 2655 raw = page_address(page); 2656 i = first_sge; 2657 do { 2658 sge = sk_msg_elem(msg, i); 2659 from = sg_virt(sge); 2660 len = sge->length; 2661 to = raw + poffset; 2662 2663 memcpy(to, from, len); 2664 poffset += len; 2665 sge->length = 0; 2666 put_page(sg_page(sge)); 2667 2668 sk_msg_iter_var_next(i); 2669 } while (i != last_sge); 2670 2671 sg_set_page(&msg->sg.data[first_sge], page, copy, 0); 2672 2673 /* To repair sg ring we need to shift entries. If we only 2674 * had a single entry though we can just replace it and 2675 * be done. Otherwise walk the ring and shift the entries. 2676 */ 2677 WARN_ON_ONCE(last_sge == first_sge); 2678 shift = last_sge > first_sge ? 2679 last_sge - first_sge - 1 : 2680 NR_MSG_FRAG_IDS - first_sge + last_sge - 1; 2681 if (!shift) 2682 goto out; 2683 2684 i = first_sge; 2685 sk_msg_iter_var_next(i); 2686 do { 2687 u32 move_from; 2688 2689 if (i + shift >= NR_MSG_FRAG_IDS) 2690 move_from = i + shift - NR_MSG_FRAG_IDS; 2691 else 2692 move_from = i + shift; 2693 if (move_from == msg->sg.end) 2694 break; 2695 2696 msg->sg.data[i] = msg->sg.data[move_from]; 2697 msg->sg.data[move_from].length = 0; 2698 msg->sg.data[move_from].page_link = 0; 2699 msg->sg.data[move_from].offset = 0; 2700 sk_msg_iter_var_next(i); 2701 } while (1); 2702 2703 msg->sg.end = msg->sg.end - shift > msg->sg.end ? 2704 msg->sg.end - shift + NR_MSG_FRAG_IDS : 2705 msg->sg.end - shift; 2706 out: 2707 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset; 2708 msg->data_end = msg->data + bytes; 2709 return 0; 2710 } 2711 2712 static const struct bpf_func_proto bpf_msg_pull_data_proto = { 2713 .func = bpf_msg_pull_data, 2714 .gpl_only = false, 2715 .ret_type = RET_INTEGER, 2716 .arg1_type = ARG_PTR_TO_CTX, 2717 .arg2_type = ARG_ANYTHING, 2718 .arg3_type = ARG_ANYTHING, 2719 .arg4_type = ARG_ANYTHING, 2720 }; 2721 2722 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start, 2723 u32, len, u64, flags) 2724 { 2725 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge; 2726 u32 new, i = 0, l = 0, space, copy = 0, offset = 0; 2727 u8 *raw, *to, *from; 2728 struct page *page; 2729 2730 if (unlikely(flags)) 2731 return -EINVAL; 2732 2733 /* First find the starting scatterlist element */ 2734 i = msg->sg.start; 2735 do { 2736 offset += l; 2737 l = sk_msg_elem(msg, i)->length; 2738 2739 if (start < offset + l) 2740 break; 2741 sk_msg_iter_var_next(i); 2742 } while (i != msg->sg.end); 2743 2744 if (start >= offset + l) 2745 return -EINVAL; 2746 2747 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2748 2749 /* If no space available will fallback to copy, we need at 2750 * least one scatterlist elem available to push data into 2751 * when start aligns to the beginning of an element or two 2752 * when it falls inside an element. We handle the start equals 2753 * offset case because its the common case for inserting a 2754 * header. 2755 */ 2756 if (!space || (space == 1 && start != offset)) 2757 copy = msg->sg.data[i].length; 2758 2759 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2760 get_order(copy + len)); 2761 if (unlikely(!page)) 2762 return -ENOMEM; 2763 2764 if (copy) { 2765 int front, back; 2766 2767 raw = page_address(page); 2768 2769 psge = sk_msg_elem(msg, i); 2770 front = start - offset; 2771 back = psge->length - front; 2772 from = sg_virt(psge); 2773 2774 if (front) 2775 memcpy(raw, from, front); 2776 2777 if (back) { 2778 from += front; 2779 to = raw + front + len; 2780 2781 memcpy(to, from, back); 2782 } 2783 2784 put_page(sg_page(psge)); 2785 } else if (start - offset) { 2786 psge = sk_msg_elem(msg, i); 2787 rsge = sk_msg_elem_cpy(msg, i); 2788 2789 psge->length = start - offset; 2790 rsge.length -= psge->length; 2791 rsge.offset += start; 2792 2793 sk_msg_iter_var_next(i); 2794 sg_unmark_end(psge); 2795 sg_unmark_end(&rsge); 2796 sk_msg_iter_next(msg, end); 2797 } 2798 2799 /* Slot(s) to place newly allocated data */ 2800 new = i; 2801 2802 /* Shift one or two slots as needed */ 2803 if (!copy) { 2804 sge = sk_msg_elem_cpy(msg, i); 2805 2806 sk_msg_iter_var_next(i); 2807 sg_unmark_end(&sge); 2808 sk_msg_iter_next(msg, end); 2809 2810 nsge = sk_msg_elem_cpy(msg, i); 2811 if (rsge.length) { 2812 sk_msg_iter_var_next(i); 2813 nnsge = sk_msg_elem_cpy(msg, i); 2814 } 2815 2816 while (i != msg->sg.end) { 2817 msg->sg.data[i] = sge; 2818 sge = nsge; 2819 sk_msg_iter_var_next(i); 2820 if (rsge.length) { 2821 nsge = nnsge; 2822 nnsge = sk_msg_elem_cpy(msg, i); 2823 } else { 2824 nsge = sk_msg_elem_cpy(msg, i); 2825 } 2826 } 2827 } 2828 2829 /* Place newly allocated data buffer */ 2830 sk_mem_charge(msg->sk, len); 2831 msg->sg.size += len; 2832 __clear_bit(new, &msg->sg.copy); 2833 sg_set_page(&msg->sg.data[new], page, len + copy, 0); 2834 if (rsge.length) { 2835 get_page(sg_page(&rsge)); 2836 sk_msg_iter_var_next(new); 2837 msg->sg.data[new] = rsge; 2838 } 2839 2840 sk_msg_compute_data_pointers(msg); 2841 return 0; 2842 } 2843 2844 static const struct bpf_func_proto bpf_msg_push_data_proto = { 2845 .func = bpf_msg_push_data, 2846 .gpl_only = false, 2847 .ret_type = RET_INTEGER, 2848 .arg1_type = ARG_PTR_TO_CTX, 2849 .arg2_type = ARG_ANYTHING, 2850 .arg3_type = ARG_ANYTHING, 2851 .arg4_type = ARG_ANYTHING, 2852 }; 2853 2854 static void sk_msg_shift_left(struct sk_msg *msg, int i) 2855 { 2856 int prev; 2857 2858 do { 2859 prev = i; 2860 sk_msg_iter_var_next(i); 2861 msg->sg.data[prev] = msg->sg.data[i]; 2862 } while (i != msg->sg.end); 2863 2864 sk_msg_iter_prev(msg, end); 2865 } 2866 2867 static void sk_msg_shift_right(struct sk_msg *msg, int i) 2868 { 2869 struct scatterlist tmp, sge; 2870 2871 sk_msg_iter_next(msg, end); 2872 sge = sk_msg_elem_cpy(msg, i); 2873 sk_msg_iter_var_next(i); 2874 tmp = sk_msg_elem_cpy(msg, i); 2875 2876 while (i != msg->sg.end) { 2877 msg->sg.data[i] = sge; 2878 sk_msg_iter_var_next(i); 2879 sge = tmp; 2880 tmp = sk_msg_elem_cpy(msg, i); 2881 } 2882 } 2883 2884 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start, 2885 u32, len, u64, flags) 2886 { 2887 u32 i = 0, l = 0, space, offset = 0; 2888 u64 last = start + len; 2889 int pop; 2890 2891 if (unlikely(flags)) 2892 return -EINVAL; 2893 2894 /* First find the starting scatterlist element */ 2895 i = msg->sg.start; 2896 do { 2897 offset += l; 2898 l = sk_msg_elem(msg, i)->length; 2899 2900 if (start < offset + l) 2901 break; 2902 sk_msg_iter_var_next(i); 2903 } while (i != msg->sg.end); 2904 2905 /* Bounds checks: start and pop must be inside message */ 2906 if (start >= offset + l || last >= msg->sg.size) 2907 return -EINVAL; 2908 2909 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2910 2911 pop = len; 2912 /* --------------| offset 2913 * -| start |-------- len -------| 2914 * 2915 * |----- a ----|-------- pop -------|----- b ----| 2916 * |______________________________________________| length 2917 * 2918 * 2919 * a: region at front of scatter element to save 2920 * b: region at back of scatter element to save when length > A + pop 2921 * pop: region to pop from element, same as input 'pop' here will be 2922 * decremented below per iteration. 2923 * 2924 * Two top-level cases to handle when start != offset, first B is non 2925 * zero and second B is zero corresponding to when a pop includes more 2926 * than one element. 2927 * 2928 * Then if B is non-zero AND there is no space allocate space and 2929 * compact A, B regions into page. If there is space shift ring to 2930 * the rigth free'ing the next element in ring to place B, leaving 2931 * A untouched except to reduce length. 2932 */ 2933 if (start != offset) { 2934 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i); 2935 int a = start; 2936 int b = sge->length - pop - a; 2937 2938 sk_msg_iter_var_next(i); 2939 2940 if (pop < sge->length - a) { 2941 if (space) { 2942 sge->length = a; 2943 sk_msg_shift_right(msg, i); 2944 nsge = sk_msg_elem(msg, i); 2945 get_page(sg_page(sge)); 2946 sg_set_page(nsge, 2947 sg_page(sge), 2948 b, sge->offset + pop + a); 2949 } else { 2950 struct page *page, *orig; 2951 u8 *to, *from; 2952 2953 page = alloc_pages(__GFP_NOWARN | 2954 __GFP_COMP | GFP_ATOMIC, 2955 get_order(a + b)); 2956 if (unlikely(!page)) 2957 return -ENOMEM; 2958 2959 sge->length = a; 2960 orig = sg_page(sge); 2961 from = sg_virt(sge); 2962 to = page_address(page); 2963 memcpy(to, from, a); 2964 memcpy(to + a, from + a + pop, b); 2965 sg_set_page(sge, page, a + b, 0); 2966 put_page(orig); 2967 } 2968 pop = 0; 2969 } else if (pop >= sge->length - a) { 2970 pop -= (sge->length - a); 2971 sge->length = a; 2972 } 2973 } 2974 2975 /* From above the current layout _must_ be as follows, 2976 * 2977 * -| offset 2978 * -| start 2979 * 2980 * |---- pop ---|---------------- b ------------| 2981 * |____________________________________________| length 2982 * 2983 * Offset and start of the current msg elem are equal because in the 2984 * previous case we handled offset != start and either consumed the 2985 * entire element and advanced to the next element OR pop == 0. 2986 * 2987 * Two cases to handle here are first pop is less than the length 2988 * leaving some remainder b above. Simply adjust the element's layout 2989 * in this case. Or pop >= length of the element so that b = 0. In this 2990 * case advance to next element decrementing pop. 2991 */ 2992 while (pop) { 2993 struct scatterlist *sge = sk_msg_elem(msg, i); 2994 2995 if (pop < sge->length) { 2996 sge->length -= pop; 2997 sge->offset += pop; 2998 pop = 0; 2999 } else { 3000 pop -= sge->length; 3001 sk_msg_shift_left(msg, i); 3002 } 3003 sk_msg_iter_var_next(i); 3004 } 3005 3006 sk_mem_uncharge(msg->sk, len - pop); 3007 msg->sg.size -= (len - pop); 3008 sk_msg_compute_data_pointers(msg); 3009 return 0; 3010 } 3011 3012 static const struct bpf_func_proto bpf_msg_pop_data_proto = { 3013 .func = bpf_msg_pop_data, 3014 .gpl_only = false, 3015 .ret_type = RET_INTEGER, 3016 .arg1_type = ARG_PTR_TO_CTX, 3017 .arg2_type = ARG_ANYTHING, 3018 .arg3_type = ARG_ANYTHING, 3019 .arg4_type = ARG_ANYTHING, 3020 }; 3021 3022 #ifdef CONFIG_CGROUP_NET_CLASSID 3023 BPF_CALL_0(bpf_get_cgroup_classid_curr) 3024 { 3025 return __task_get_classid(current); 3026 } 3027 3028 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = { 3029 .func = bpf_get_cgroup_classid_curr, 3030 .gpl_only = false, 3031 .ret_type = RET_INTEGER, 3032 }; 3033 3034 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb) 3035 { 3036 struct sock *sk = skb_to_full_sk(skb); 3037 3038 if (!sk || !sk_fullsock(sk)) 3039 return 0; 3040 3041 return sock_cgroup_classid(&sk->sk_cgrp_data); 3042 } 3043 3044 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = { 3045 .func = bpf_skb_cgroup_classid, 3046 .gpl_only = false, 3047 .ret_type = RET_INTEGER, 3048 .arg1_type = ARG_PTR_TO_CTX, 3049 }; 3050 #endif 3051 3052 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb) 3053 { 3054 return task_get_classid(skb); 3055 } 3056 3057 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = { 3058 .func = bpf_get_cgroup_classid, 3059 .gpl_only = false, 3060 .ret_type = RET_INTEGER, 3061 .arg1_type = ARG_PTR_TO_CTX, 3062 }; 3063 3064 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb) 3065 { 3066 return dst_tclassid(skb); 3067 } 3068 3069 static const struct bpf_func_proto bpf_get_route_realm_proto = { 3070 .func = bpf_get_route_realm, 3071 .gpl_only = false, 3072 .ret_type = RET_INTEGER, 3073 .arg1_type = ARG_PTR_TO_CTX, 3074 }; 3075 3076 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb) 3077 { 3078 /* If skb_clear_hash() was called due to mangling, we can 3079 * trigger SW recalculation here. Later access to hash 3080 * can then use the inline skb->hash via context directly 3081 * instead of calling this helper again. 3082 */ 3083 return skb_get_hash(skb); 3084 } 3085 3086 static const struct bpf_func_proto bpf_get_hash_recalc_proto = { 3087 .func = bpf_get_hash_recalc, 3088 .gpl_only = false, 3089 .ret_type = RET_INTEGER, 3090 .arg1_type = ARG_PTR_TO_CTX, 3091 }; 3092 3093 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb) 3094 { 3095 /* After all direct packet write, this can be used once for 3096 * triggering a lazy recalc on next skb_get_hash() invocation. 3097 */ 3098 skb_clear_hash(skb); 3099 return 0; 3100 } 3101 3102 static const struct bpf_func_proto bpf_set_hash_invalid_proto = { 3103 .func = bpf_set_hash_invalid, 3104 .gpl_only = false, 3105 .ret_type = RET_INTEGER, 3106 .arg1_type = ARG_PTR_TO_CTX, 3107 }; 3108 3109 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash) 3110 { 3111 /* Set user specified hash as L4(+), so that it gets returned 3112 * on skb_get_hash() call unless BPF prog later on triggers a 3113 * skb_clear_hash(). 3114 */ 3115 __skb_set_sw_hash(skb, hash, true); 3116 return 0; 3117 } 3118 3119 static const struct bpf_func_proto bpf_set_hash_proto = { 3120 .func = bpf_set_hash, 3121 .gpl_only = false, 3122 .ret_type = RET_INTEGER, 3123 .arg1_type = ARG_PTR_TO_CTX, 3124 .arg2_type = ARG_ANYTHING, 3125 }; 3126 3127 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto, 3128 u16, vlan_tci) 3129 { 3130 int ret; 3131 3132 if (unlikely(vlan_proto != htons(ETH_P_8021Q) && 3133 vlan_proto != htons(ETH_P_8021AD))) 3134 vlan_proto = htons(ETH_P_8021Q); 3135 3136 bpf_push_mac_rcsum(skb); 3137 ret = skb_vlan_push(skb, vlan_proto, vlan_tci); 3138 bpf_pull_mac_rcsum(skb); 3139 3140 bpf_compute_data_pointers(skb); 3141 return ret; 3142 } 3143 3144 static const struct bpf_func_proto bpf_skb_vlan_push_proto = { 3145 .func = bpf_skb_vlan_push, 3146 .gpl_only = false, 3147 .ret_type = RET_INTEGER, 3148 .arg1_type = ARG_PTR_TO_CTX, 3149 .arg2_type = ARG_ANYTHING, 3150 .arg3_type = ARG_ANYTHING, 3151 }; 3152 3153 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb) 3154 { 3155 int ret; 3156 3157 bpf_push_mac_rcsum(skb); 3158 ret = skb_vlan_pop(skb); 3159 bpf_pull_mac_rcsum(skb); 3160 3161 bpf_compute_data_pointers(skb); 3162 return ret; 3163 } 3164 3165 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = { 3166 .func = bpf_skb_vlan_pop, 3167 .gpl_only = false, 3168 .ret_type = RET_INTEGER, 3169 .arg1_type = ARG_PTR_TO_CTX, 3170 }; 3171 3172 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len) 3173 { 3174 /* Caller already did skb_cow() with len as headroom, 3175 * so no need to do it here. 3176 */ 3177 skb_push(skb, len); 3178 memmove(skb->data, skb->data + len, off); 3179 memset(skb->data + off, 0, len); 3180 3181 /* No skb_postpush_rcsum(skb, skb->data + off, len) 3182 * needed here as it does not change the skb->csum 3183 * result for checksum complete when summing over 3184 * zeroed blocks. 3185 */ 3186 return 0; 3187 } 3188 3189 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len) 3190 { 3191 /* skb_ensure_writable() is not needed here, as we're 3192 * already working on an uncloned skb. 3193 */ 3194 if (unlikely(!pskb_may_pull(skb, off + len))) 3195 return -ENOMEM; 3196 3197 skb_postpull_rcsum(skb, skb->data + off, len); 3198 memmove(skb->data + len, skb->data, off); 3199 __skb_pull(skb, len); 3200 3201 return 0; 3202 } 3203 3204 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len) 3205 { 3206 bool trans_same = skb->transport_header == skb->network_header; 3207 int ret; 3208 3209 /* There's no need for __skb_push()/__skb_pull() pair to 3210 * get to the start of the mac header as we're guaranteed 3211 * to always start from here under eBPF. 3212 */ 3213 ret = bpf_skb_generic_push(skb, off, len); 3214 if (likely(!ret)) { 3215 skb->mac_header -= len; 3216 skb->network_header -= len; 3217 if (trans_same) 3218 skb->transport_header = skb->network_header; 3219 } 3220 3221 return ret; 3222 } 3223 3224 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len) 3225 { 3226 bool trans_same = skb->transport_header == skb->network_header; 3227 int ret; 3228 3229 /* Same here, __skb_push()/__skb_pull() pair not needed. */ 3230 ret = bpf_skb_generic_pop(skb, off, len); 3231 if (likely(!ret)) { 3232 skb->mac_header += len; 3233 skb->network_header += len; 3234 if (trans_same) 3235 skb->transport_header = skb->network_header; 3236 } 3237 3238 return ret; 3239 } 3240 3241 static int bpf_skb_proto_4_to_6(struct sk_buff *skb) 3242 { 3243 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3244 u32 off = skb_mac_header_len(skb); 3245 int ret; 3246 3247 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) 3248 return -ENOTSUPP; 3249 3250 ret = skb_cow(skb, len_diff); 3251 if (unlikely(ret < 0)) 3252 return ret; 3253 3254 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3255 if (unlikely(ret < 0)) 3256 return ret; 3257 3258 if (skb_is_gso(skb)) { 3259 struct skb_shared_info *shinfo = skb_shinfo(skb); 3260 3261 /* SKB_GSO_TCPV4 needs to be changed into 3262 * SKB_GSO_TCPV6. 3263 */ 3264 if (shinfo->gso_type & SKB_GSO_TCPV4) { 3265 shinfo->gso_type &= ~SKB_GSO_TCPV4; 3266 shinfo->gso_type |= SKB_GSO_TCPV6; 3267 } 3268 3269 /* Due to IPv6 header, MSS needs to be downgraded. */ 3270 skb_decrease_gso_size(shinfo, len_diff); 3271 /* Header must be checked, and gso_segs recomputed. */ 3272 shinfo->gso_type |= SKB_GSO_DODGY; 3273 shinfo->gso_segs = 0; 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 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) 3289 return -ENOTSUPP; 3290 3291 ret = skb_unclone(skb, GFP_ATOMIC); 3292 if (unlikely(ret < 0)) 3293 return ret; 3294 3295 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3296 if (unlikely(ret < 0)) 3297 return ret; 3298 3299 if (skb_is_gso(skb)) { 3300 struct skb_shared_info *shinfo = skb_shinfo(skb); 3301 3302 /* SKB_GSO_TCPV6 needs to be changed into 3303 * SKB_GSO_TCPV4. 3304 */ 3305 if (shinfo->gso_type & SKB_GSO_TCPV6) { 3306 shinfo->gso_type &= ~SKB_GSO_TCPV6; 3307 shinfo->gso_type |= SKB_GSO_TCPV4; 3308 } 3309 3310 /* Due to IPv4 header, MSS can be upgraded. */ 3311 skb_increase_gso_size(shinfo, len_diff); 3312 /* Header must be checked, and gso_segs recomputed. */ 3313 shinfo->gso_type |= SKB_GSO_DODGY; 3314 shinfo->gso_segs = 0; 3315 } 3316 3317 skb->protocol = htons(ETH_P_IP); 3318 skb_clear_hash(skb); 3319 3320 return 0; 3321 } 3322 3323 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto) 3324 { 3325 __be16 from_proto = skb->protocol; 3326 3327 if (from_proto == htons(ETH_P_IP) && 3328 to_proto == htons(ETH_P_IPV6)) 3329 return bpf_skb_proto_4_to_6(skb); 3330 3331 if (from_proto == htons(ETH_P_IPV6) && 3332 to_proto == htons(ETH_P_IP)) 3333 return bpf_skb_proto_6_to_4(skb); 3334 3335 return -ENOTSUPP; 3336 } 3337 3338 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto, 3339 u64, flags) 3340 { 3341 int ret; 3342 3343 if (unlikely(flags)) 3344 return -EINVAL; 3345 3346 /* General idea is that this helper does the basic groundwork 3347 * needed for changing the protocol, and eBPF program fills the 3348 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace() 3349 * and other helpers, rather than passing a raw buffer here. 3350 * 3351 * The rationale is to keep this minimal and without a need to 3352 * deal with raw packet data. F.e. even if we would pass buffers 3353 * here, the program still needs to call the bpf_lX_csum_replace() 3354 * helpers anyway. Plus, this way we keep also separation of 3355 * concerns, since f.e. bpf_skb_store_bytes() should only take 3356 * care of stores. 3357 * 3358 * Currently, additional options and extension header space are 3359 * not supported, but flags register is reserved so we can adapt 3360 * that. For offloads, we mark packet as dodgy, so that headers 3361 * need to be verified first. 3362 */ 3363 ret = bpf_skb_proto_xlat(skb, proto); 3364 bpf_compute_data_pointers(skb); 3365 return ret; 3366 } 3367 3368 static const struct bpf_func_proto bpf_skb_change_proto_proto = { 3369 .func = bpf_skb_change_proto, 3370 .gpl_only = false, 3371 .ret_type = RET_INTEGER, 3372 .arg1_type = ARG_PTR_TO_CTX, 3373 .arg2_type = ARG_ANYTHING, 3374 .arg3_type = ARG_ANYTHING, 3375 }; 3376 3377 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type) 3378 { 3379 /* We only allow a restricted subset to be changed for now. */ 3380 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) || 3381 !skb_pkt_type_ok(pkt_type))) 3382 return -EINVAL; 3383 3384 skb->pkt_type = pkt_type; 3385 return 0; 3386 } 3387 3388 static const struct bpf_func_proto bpf_skb_change_type_proto = { 3389 .func = bpf_skb_change_type, 3390 .gpl_only = false, 3391 .ret_type = RET_INTEGER, 3392 .arg1_type = ARG_PTR_TO_CTX, 3393 .arg2_type = ARG_ANYTHING, 3394 }; 3395 3396 static u32 bpf_skb_net_base_len(const struct sk_buff *skb) 3397 { 3398 switch (skb->protocol) { 3399 case htons(ETH_P_IP): 3400 return sizeof(struct iphdr); 3401 case htons(ETH_P_IPV6): 3402 return sizeof(struct ipv6hdr); 3403 default: 3404 return ~0U; 3405 } 3406 } 3407 3408 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \ 3409 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3410 3411 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \ 3412 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \ 3413 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \ 3414 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \ 3415 BPF_F_ADJ_ROOM_ENCAP_L2( \ 3416 BPF_ADJ_ROOM_ENCAP_L2_MASK)) 3417 3418 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff, 3419 u64 flags) 3420 { 3421 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT; 3422 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK; 3423 u16 mac_len = 0, inner_net = 0, inner_trans = 0; 3424 unsigned int gso_type = SKB_GSO_DODGY; 3425 int ret; 3426 3427 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3428 /* udp gso_size delineates datagrams, only allow if fixed */ 3429 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3430 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3431 return -ENOTSUPP; 3432 } 3433 3434 ret = skb_cow_head(skb, len_diff); 3435 if (unlikely(ret < 0)) 3436 return ret; 3437 3438 if (encap) { 3439 if (skb->protocol != htons(ETH_P_IP) && 3440 skb->protocol != htons(ETH_P_IPV6)) 3441 return -ENOTSUPP; 3442 3443 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 && 3444 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3445 return -EINVAL; 3446 3447 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE && 3448 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3449 return -EINVAL; 3450 3451 if (skb->encapsulation) 3452 return -EALREADY; 3453 3454 mac_len = skb->network_header - skb->mac_header; 3455 inner_net = skb->network_header; 3456 if (inner_mac_len > len_diff) 3457 return -EINVAL; 3458 inner_trans = skb->transport_header; 3459 } 3460 3461 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3462 if (unlikely(ret < 0)) 3463 return ret; 3464 3465 if (encap) { 3466 skb->inner_mac_header = inner_net - inner_mac_len; 3467 skb->inner_network_header = inner_net; 3468 skb->inner_transport_header = inner_trans; 3469 skb_set_inner_protocol(skb, skb->protocol); 3470 3471 skb->encapsulation = 1; 3472 skb_set_network_header(skb, mac_len); 3473 3474 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3475 gso_type |= SKB_GSO_UDP_TUNNEL; 3476 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE) 3477 gso_type |= SKB_GSO_GRE; 3478 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3479 gso_type |= SKB_GSO_IPXIP6; 3480 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3481 gso_type |= SKB_GSO_IPXIP4; 3482 3483 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE || 3484 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) { 3485 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ? 3486 sizeof(struct ipv6hdr) : 3487 sizeof(struct iphdr); 3488 3489 skb_set_transport_header(skb, mac_len + nh_len); 3490 } 3491 3492 /* Match skb->protocol to new outer l3 protocol */ 3493 if (skb->protocol == htons(ETH_P_IP) && 3494 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3495 skb->protocol = htons(ETH_P_IPV6); 3496 else if (skb->protocol == htons(ETH_P_IPV6) && 3497 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3498 skb->protocol = htons(ETH_P_IP); 3499 } 3500 3501 if (skb_is_gso(skb)) { 3502 struct skb_shared_info *shinfo = skb_shinfo(skb); 3503 3504 /* Due to header grow, MSS needs to be downgraded. */ 3505 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3506 skb_decrease_gso_size(shinfo, len_diff); 3507 3508 /* Header must be checked, and gso_segs recomputed. */ 3509 shinfo->gso_type |= gso_type; 3510 shinfo->gso_segs = 0; 3511 } 3512 3513 return 0; 3514 } 3515 3516 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff, 3517 u64 flags) 3518 { 3519 int ret; 3520 3521 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO | 3522 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3523 return -EINVAL; 3524 3525 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3526 /* udp gso_size delineates datagrams, only allow if fixed */ 3527 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3528 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3529 return -ENOTSUPP; 3530 } 3531 3532 ret = skb_unclone(skb, GFP_ATOMIC); 3533 if (unlikely(ret < 0)) 3534 return ret; 3535 3536 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3537 if (unlikely(ret < 0)) 3538 return ret; 3539 3540 if (skb_is_gso(skb)) { 3541 struct skb_shared_info *shinfo = skb_shinfo(skb); 3542 3543 /* Due to header shrink, MSS can be upgraded. */ 3544 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3545 skb_increase_gso_size(shinfo, len_diff); 3546 3547 /* Header must be checked, and gso_segs recomputed. */ 3548 shinfo->gso_type |= SKB_GSO_DODGY; 3549 shinfo->gso_segs = 0; 3550 } 3551 3552 return 0; 3553 } 3554 3555 static u32 __bpf_skb_max_len(const struct sk_buff *skb) 3556 { 3557 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len : 3558 SKB_MAX_ALLOC; 3559 } 3560 3561 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3562 u32, mode, u64, flags) 3563 { 3564 u32 len_diff_abs = abs(len_diff); 3565 bool shrink = len_diff < 0; 3566 int ret = 0; 3567 3568 if (unlikely(flags || mode)) 3569 return -EINVAL; 3570 if (unlikely(len_diff_abs > 0xfffU)) 3571 return -EFAULT; 3572 3573 if (!shrink) { 3574 ret = skb_cow(skb, len_diff); 3575 if (unlikely(ret < 0)) 3576 return ret; 3577 __skb_push(skb, len_diff_abs); 3578 memset(skb->data, 0, len_diff_abs); 3579 } else { 3580 if (unlikely(!pskb_may_pull(skb, len_diff_abs))) 3581 return -ENOMEM; 3582 __skb_pull(skb, len_diff_abs); 3583 } 3584 bpf_compute_data_end_sk_skb(skb); 3585 if (tls_sw_has_ctx_rx(skb->sk)) { 3586 struct strp_msg *rxm = strp_msg(skb); 3587 3588 rxm->full_len += len_diff; 3589 } 3590 return ret; 3591 } 3592 3593 static const struct bpf_func_proto sk_skb_adjust_room_proto = { 3594 .func = sk_skb_adjust_room, 3595 .gpl_only = false, 3596 .ret_type = RET_INTEGER, 3597 .arg1_type = ARG_PTR_TO_CTX, 3598 .arg2_type = ARG_ANYTHING, 3599 .arg3_type = ARG_ANYTHING, 3600 .arg4_type = ARG_ANYTHING, 3601 }; 3602 3603 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3604 u32, mode, u64, flags) 3605 { 3606 u32 len_cur, len_diff_abs = abs(len_diff); 3607 u32 len_min = bpf_skb_net_base_len(skb); 3608 u32 len_max = __bpf_skb_max_len(skb); 3609 __be16 proto = skb->protocol; 3610 bool shrink = len_diff < 0; 3611 u32 off; 3612 int ret; 3613 3614 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK | 3615 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3616 return -EINVAL; 3617 if (unlikely(len_diff_abs > 0xfffU)) 3618 return -EFAULT; 3619 if (unlikely(proto != htons(ETH_P_IP) && 3620 proto != htons(ETH_P_IPV6))) 3621 return -ENOTSUPP; 3622 3623 off = skb_mac_header_len(skb); 3624 switch (mode) { 3625 case BPF_ADJ_ROOM_NET: 3626 off += bpf_skb_net_base_len(skb); 3627 break; 3628 case BPF_ADJ_ROOM_MAC: 3629 break; 3630 default: 3631 return -ENOTSUPP; 3632 } 3633 3634 len_cur = skb->len - skb_network_offset(skb); 3635 if ((shrink && (len_diff_abs >= len_cur || 3636 len_cur - len_diff_abs < len_min)) || 3637 (!shrink && (skb->len + len_diff_abs > len_max && 3638 !skb_is_gso(skb)))) 3639 return -ENOTSUPP; 3640 3641 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) : 3642 bpf_skb_net_grow(skb, off, len_diff_abs, flags); 3643 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET)) 3644 __skb_reset_checksum_unnecessary(skb); 3645 3646 bpf_compute_data_pointers(skb); 3647 return ret; 3648 } 3649 3650 static const struct bpf_func_proto bpf_skb_adjust_room_proto = { 3651 .func = bpf_skb_adjust_room, 3652 .gpl_only = false, 3653 .ret_type = RET_INTEGER, 3654 .arg1_type = ARG_PTR_TO_CTX, 3655 .arg2_type = ARG_ANYTHING, 3656 .arg3_type = ARG_ANYTHING, 3657 .arg4_type = ARG_ANYTHING, 3658 }; 3659 3660 static u32 __bpf_skb_min_len(const struct sk_buff *skb) 3661 { 3662 u32 min_len = skb_network_offset(skb); 3663 3664 if (skb_transport_header_was_set(skb)) 3665 min_len = skb_transport_offset(skb); 3666 if (skb->ip_summed == CHECKSUM_PARTIAL) 3667 min_len = skb_checksum_start_offset(skb) + 3668 skb->csum_offset + sizeof(__sum16); 3669 return min_len; 3670 } 3671 3672 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len) 3673 { 3674 unsigned int old_len = skb->len; 3675 int ret; 3676 3677 ret = __skb_grow_rcsum(skb, new_len); 3678 if (!ret) 3679 memset(skb->data + old_len, 0, new_len - old_len); 3680 return ret; 3681 } 3682 3683 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len) 3684 { 3685 return __skb_trim_rcsum(skb, new_len); 3686 } 3687 3688 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len, 3689 u64 flags) 3690 { 3691 u32 max_len = __bpf_skb_max_len(skb); 3692 u32 min_len = __bpf_skb_min_len(skb); 3693 int ret; 3694 3695 if (unlikely(flags || new_len > max_len || new_len < min_len)) 3696 return -EINVAL; 3697 if (skb->encapsulation) 3698 return -ENOTSUPP; 3699 3700 /* The basic idea of this helper is that it's performing the 3701 * needed work to either grow or trim an skb, and eBPF program 3702 * rewrites the rest via helpers like bpf_skb_store_bytes(), 3703 * bpf_lX_csum_replace() and others rather than passing a raw 3704 * buffer here. This one is a slow path helper and intended 3705 * for replies with control messages. 3706 * 3707 * Like in bpf_skb_change_proto(), we want to keep this rather 3708 * minimal and without protocol specifics so that we are able 3709 * to separate concerns as in bpf_skb_store_bytes() should only 3710 * be the one responsible for writing buffers. 3711 * 3712 * It's really expected to be a slow path operation here for 3713 * control message replies, so we're implicitly linearizing, 3714 * uncloning and drop offloads from the skb by this. 3715 */ 3716 ret = __bpf_try_make_writable(skb, skb->len); 3717 if (!ret) { 3718 if (new_len > skb->len) 3719 ret = bpf_skb_grow_rcsum(skb, new_len); 3720 else if (new_len < skb->len) 3721 ret = bpf_skb_trim_rcsum(skb, new_len); 3722 if (!ret && skb_is_gso(skb)) 3723 skb_gso_reset(skb); 3724 } 3725 return ret; 3726 } 3727 3728 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3729 u64, flags) 3730 { 3731 int ret = __bpf_skb_change_tail(skb, new_len, flags); 3732 3733 bpf_compute_data_pointers(skb); 3734 return ret; 3735 } 3736 3737 static const struct bpf_func_proto bpf_skb_change_tail_proto = { 3738 .func = bpf_skb_change_tail, 3739 .gpl_only = false, 3740 .ret_type = RET_INTEGER, 3741 .arg1_type = ARG_PTR_TO_CTX, 3742 .arg2_type = ARG_ANYTHING, 3743 .arg3_type = ARG_ANYTHING, 3744 }; 3745 3746 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3747 u64, flags) 3748 { 3749 int ret = __bpf_skb_change_tail(skb, new_len, flags); 3750 3751 bpf_compute_data_end_sk_skb(skb); 3752 return ret; 3753 } 3754 3755 static const struct bpf_func_proto sk_skb_change_tail_proto = { 3756 .func = sk_skb_change_tail, 3757 .gpl_only = false, 3758 .ret_type = RET_INTEGER, 3759 .arg1_type = ARG_PTR_TO_CTX, 3760 .arg2_type = ARG_ANYTHING, 3761 .arg3_type = ARG_ANYTHING, 3762 }; 3763 3764 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room, 3765 u64 flags) 3766 { 3767 u32 max_len = __bpf_skb_max_len(skb); 3768 u32 new_len = skb->len + head_room; 3769 int ret; 3770 3771 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) || 3772 new_len < skb->len)) 3773 return -EINVAL; 3774 3775 ret = skb_cow(skb, head_room); 3776 if (likely(!ret)) { 3777 /* Idea for this helper is that we currently only 3778 * allow to expand on mac header. This means that 3779 * skb->protocol network header, etc, stay as is. 3780 * Compared to bpf_skb_change_tail(), we're more 3781 * flexible due to not needing to linearize or 3782 * reset GSO. Intention for this helper is to be 3783 * used by an L3 skb that needs to push mac header 3784 * for redirection into L2 device. 3785 */ 3786 __skb_push(skb, head_room); 3787 memset(skb->data, 0, head_room); 3788 skb_reset_mac_header(skb); 3789 } 3790 3791 return ret; 3792 } 3793 3794 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room, 3795 u64, flags) 3796 { 3797 int ret = __bpf_skb_change_head(skb, head_room, flags); 3798 3799 bpf_compute_data_pointers(skb); 3800 return ret; 3801 } 3802 3803 static const struct bpf_func_proto bpf_skb_change_head_proto = { 3804 .func = bpf_skb_change_head, 3805 .gpl_only = false, 3806 .ret_type = RET_INTEGER, 3807 .arg1_type = ARG_PTR_TO_CTX, 3808 .arg2_type = ARG_ANYTHING, 3809 .arg3_type = ARG_ANYTHING, 3810 }; 3811 3812 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room, 3813 u64, flags) 3814 { 3815 int ret = __bpf_skb_change_head(skb, head_room, flags); 3816 3817 bpf_compute_data_end_sk_skb(skb); 3818 return ret; 3819 } 3820 3821 static const struct bpf_func_proto sk_skb_change_head_proto = { 3822 .func = sk_skb_change_head, 3823 .gpl_only = false, 3824 .ret_type = RET_INTEGER, 3825 .arg1_type = ARG_PTR_TO_CTX, 3826 .arg2_type = ARG_ANYTHING, 3827 .arg3_type = ARG_ANYTHING, 3828 }; 3829 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp) 3830 { 3831 return xdp_data_meta_unsupported(xdp) ? 0 : 3832 xdp->data - xdp->data_meta; 3833 } 3834 3835 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset) 3836 { 3837 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 3838 unsigned long metalen = xdp_get_metalen(xdp); 3839 void *data_start = xdp_frame_end + metalen; 3840 void *data = xdp->data + offset; 3841 3842 if (unlikely(data < data_start || 3843 data > xdp->data_end - ETH_HLEN)) 3844 return -EINVAL; 3845 3846 if (metalen) 3847 memmove(xdp->data_meta + offset, 3848 xdp->data_meta, metalen); 3849 xdp->data_meta += offset; 3850 xdp->data = data; 3851 3852 return 0; 3853 } 3854 3855 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = { 3856 .func = bpf_xdp_adjust_head, 3857 .gpl_only = false, 3858 .ret_type = RET_INTEGER, 3859 .arg1_type = ARG_PTR_TO_CTX, 3860 .arg2_type = ARG_ANYTHING, 3861 }; 3862 3863 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset) 3864 { 3865 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */ 3866 void *data_end = xdp->data_end + offset; 3867 3868 /* Notice that xdp_data_hard_end have reserved some tailroom */ 3869 if (unlikely(data_end > data_hard_end)) 3870 return -EINVAL; 3871 3872 /* ALL drivers MUST init xdp->frame_sz, chicken check below */ 3873 if (unlikely(xdp->frame_sz > PAGE_SIZE)) { 3874 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz); 3875 return -EINVAL; 3876 } 3877 3878 if (unlikely(data_end < xdp->data + ETH_HLEN)) 3879 return -EINVAL; 3880 3881 /* Clear memory area on grow, can contain uninit kernel memory */ 3882 if (offset > 0) 3883 memset(xdp->data_end, 0, offset); 3884 3885 xdp->data_end = data_end; 3886 3887 return 0; 3888 } 3889 3890 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = { 3891 .func = bpf_xdp_adjust_tail, 3892 .gpl_only = false, 3893 .ret_type = RET_INTEGER, 3894 .arg1_type = ARG_PTR_TO_CTX, 3895 .arg2_type = ARG_ANYTHING, 3896 }; 3897 3898 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset) 3899 { 3900 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 3901 void *meta = xdp->data_meta + offset; 3902 unsigned long metalen = xdp->data - meta; 3903 3904 if (xdp_data_meta_unsupported(xdp)) 3905 return -ENOTSUPP; 3906 if (unlikely(meta < xdp_frame_end || 3907 meta > xdp->data)) 3908 return -EINVAL; 3909 if (unlikely((metalen & (sizeof(__u32) - 1)) || 3910 (metalen > 32))) 3911 return -EACCES; 3912 3913 xdp->data_meta = meta; 3914 3915 return 0; 3916 } 3917 3918 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = { 3919 .func = bpf_xdp_adjust_meta, 3920 .gpl_only = false, 3921 .ret_type = RET_INTEGER, 3922 .arg1_type = ARG_PTR_TO_CTX, 3923 .arg2_type = ARG_ANYTHING, 3924 }; 3925 3926 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd, 3927 struct bpf_map *map, struct xdp_buff *xdp) 3928 { 3929 switch (map->map_type) { 3930 case BPF_MAP_TYPE_DEVMAP: 3931 case BPF_MAP_TYPE_DEVMAP_HASH: 3932 return dev_map_enqueue(fwd, xdp, dev_rx); 3933 case BPF_MAP_TYPE_CPUMAP: 3934 return cpu_map_enqueue(fwd, xdp, dev_rx); 3935 case BPF_MAP_TYPE_XSKMAP: 3936 return __xsk_map_redirect(fwd, xdp); 3937 default: 3938 return -EBADRQC; 3939 } 3940 return 0; 3941 } 3942 3943 void xdp_do_flush(void) 3944 { 3945 __dev_flush(); 3946 __cpu_map_flush(); 3947 __xsk_map_flush(); 3948 } 3949 EXPORT_SYMBOL_GPL(xdp_do_flush); 3950 3951 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index) 3952 { 3953 switch (map->map_type) { 3954 case BPF_MAP_TYPE_DEVMAP: 3955 return __dev_map_lookup_elem(map, index); 3956 case BPF_MAP_TYPE_DEVMAP_HASH: 3957 return __dev_map_hash_lookup_elem(map, index); 3958 case BPF_MAP_TYPE_CPUMAP: 3959 return __cpu_map_lookup_elem(map, index); 3960 case BPF_MAP_TYPE_XSKMAP: 3961 return __xsk_map_lookup_elem(map, index); 3962 default: 3963 return NULL; 3964 } 3965 } 3966 3967 void bpf_clear_redirect_map(struct bpf_map *map) 3968 { 3969 struct bpf_redirect_info *ri; 3970 int cpu; 3971 3972 for_each_possible_cpu(cpu) { 3973 ri = per_cpu_ptr(&bpf_redirect_info, cpu); 3974 /* Avoid polluting remote cacheline due to writes if 3975 * not needed. Once we pass this test, we need the 3976 * cmpxchg() to make sure it hasn't been changed in 3977 * the meantime by remote CPU. 3978 */ 3979 if (unlikely(READ_ONCE(ri->map) == map)) 3980 cmpxchg(&ri->map, map, NULL); 3981 } 3982 } 3983 3984 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp, 3985 struct bpf_prog *xdp_prog) 3986 { 3987 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3988 struct bpf_map *map = READ_ONCE(ri->map); 3989 u32 index = ri->tgt_index; 3990 void *fwd = ri->tgt_value; 3991 int err; 3992 3993 ri->tgt_index = 0; 3994 ri->tgt_value = NULL; 3995 WRITE_ONCE(ri->map, NULL); 3996 3997 if (unlikely(!map)) { 3998 fwd = dev_get_by_index_rcu(dev_net(dev), index); 3999 if (unlikely(!fwd)) { 4000 err = -EINVAL; 4001 goto err; 4002 } 4003 4004 err = dev_xdp_enqueue(fwd, xdp, dev); 4005 } else { 4006 err = __bpf_tx_xdp_map(dev, fwd, map, xdp); 4007 } 4008 4009 if (unlikely(err)) 4010 goto err; 4011 4012 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index); 4013 return 0; 4014 err: 4015 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err); 4016 return err; 4017 } 4018 EXPORT_SYMBOL_GPL(xdp_do_redirect); 4019 4020 static int xdp_do_generic_redirect_map(struct net_device *dev, 4021 struct sk_buff *skb, 4022 struct xdp_buff *xdp, 4023 struct bpf_prog *xdp_prog, 4024 struct bpf_map *map) 4025 { 4026 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4027 u32 index = ri->tgt_index; 4028 void *fwd = ri->tgt_value; 4029 int err = 0; 4030 4031 ri->tgt_index = 0; 4032 ri->tgt_value = NULL; 4033 WRITE_ONCE(ri->map, NULL); 4034 4035 if (map->map_type == BPF_MAP_TYPE_DEVMAP || 4036 map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) { 4037 struct bpf_dtab_netdev *dst = fwd; 4038 4039 err = dev_map_generic_redirect(dst, skb, xdp_prog); 4040 if (unlikely(err)) 4041 goto err; 4042 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) { 4043 struct xdp_sock *xs = fwd; 4044 4045 err = xsk_generic_rcv(xs, xdp); 4046 if (err) 4047 goto err; 4048 consume_skb(skb); 4049 } else { 4050 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */ 4051 err = -EBADRQC; 4052 goto err; 4053 } 4054 4055 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index); 4056 return 0; 4057 err: 4058 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err); 4059 return err; 4060 } 4061 4062 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 4063 struct xdp_buff *xdp, struct bpf_prog *xdp_prog) 4064 { 4065 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4066 struct bpf_map *map = READ_ONCE(ri->map); 4067 u32 index = ri->tgt_index; 4068 struct net_device *fwd; 4069 int err = 0; 4070 4071 if (map) 4072 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, 4073 map); 4074 ri->tgt_index = 0; 4075 fwd = dev_get_by_index_rcu(dev_net(dev), index); 4076 if (unlikely(!fwd)) { 4077 err = -EINVAL; 4078 goto err; 4079 } 4080 4081 err = xdp_ok_fwd_dev(fwd, skb->len); 4082 if (unlikely(err)) 4083 goto err; 4084 4085 skb->dev = fwd; 4086 _trace_xdp_redirect(dev, xdp_prog, index); 4087 generic_xdp_tx(skb, xdp_prog); 4088 return 0; 4089 err: 4090 _trace_xdp_redirect_err(dev, xdp_prog, index, err); 4091 return err; 4092 } 4093 4094 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags) 4095 { 4096 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4097 4098 if (unlikely(flags)) 4099 return XDP_ABORTED; 4100 4101 ri->flags = flags; 4102 ri->tgt_index = ifindex; 4103 ri->tgt_value = NULL; 4104 WRITE_ONCE(ri->map, NULL); 4105 4106 return XDP_REDIRECT; 4107 } 4108 4109 static const struct bpf_func_proto bpf_xdp_redirect_proto = { 4110 .func = bpf_xdp_redirect, 4111 .gpl_only = false, 4112 .ret_type = RET_INTEGER, 4113 .arg1_type = ARG_ANYTHING, 4114 .arg2_type = ARG_ANYTHING, 4115 }; 4116 4117 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, 4118 u64, flags) 4119 { 4120 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4121 4122 /* Lower bits of the flags are used as return code on lookup failure */ 4123 if (unlikely(flags > XDP_TX)) 4124 return XDP_ABORTED; 4125 4126 ri->tgt_value = __xdp_map_lookup_elem(map, ifindex); 4127 if (unlikely(!ri->tgt_value)) { 4128 /* If the lookup fails we want to clear out the state in the 4129 * redirect_info struct completely, so that if an eBPF program 4130 * performs multiple lookups, the last one always takes 4131 * precedence. 4132 */ 4133 WRITE_ONCE(ri->map, NULL); 4134 return flags; 4135 } 4136 4137 ri->flags = flags; 4138 ri->tgt_index = ifindex; 4139 WRITE_ONCE(ri->map, map); 4140 4141 return XDP_REDIRECT; 4142 } 4143 4144 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = { 4145 .func = bpf_xdp_redirect_map, 4146 .gpl_only = false, 4147 .ret_type = RET_INTEGER, 4148 .arg1_type = ARG_CONST_MAP_PTR, 4149 .arg2_type = ARG_ANYTHING, 4150 .arg3_type = ARG_ANYTHING, 4151 }; 4152 4153 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb, 4154 unsigned long off, unsigned long len) 4155 { 4156 void *ptr = skb_header_pointer(skb, off, len, dst_buff); 4157 4158 if (unlikely(!ptr)) 4159 return len; 4160 if (ptr != dst_buff) 4161 memcpy(dst_buff, ptr, len); 4162 4163 return 0; 4164 } 4165 4166 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map, 4167 u64, flags, void *, meta, u64, meta_size) 4168 { 4169 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4170 4171 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4172 return -EINVAL; 4173 if (unlikely(!skb || skb_size > skb->len)) 4174 return -EFAULT; 4175 4176 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size, 4177 bpf_skb_copy); 4178 } 4179 4180 static const struct bpf_func_proto bpf_skb_event_output_proto = { 4181 .func = bpf_skb_event_output, 4182 .gpl_only = true, 4183 .ret_type = RET_INTEGER, 4184 .arg1_type = ARG_PTR_TO_CTX, 4185 .arg2_type = ARG_CONST_MAP_PTR, 4186 .arg3_type = ARG_ANYTHING, 4187 .arg4_type = ARG_PTR_TO_MEM, 4188 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4189 }; 4190 4191 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff) 4192 4193 const struct bpf_func_proto bpf_skb_output_proto = { 4194 .func = bpf_skb_event_output, 4195 .gpl_only = true, 4196 .ret_type = RET_INTEGER, 4197 .arg1_type = ARG_PTR_TO_BTF_ID, 4198 .arg1_btf_id = &bpf_skb_output_btf_ids[0], 4199 .arg2_type = ARG_CONST_MAP_PTR, 4200 .arg3_type = ARG_ANYTHING, 4201 .arg4_type = ARG_PTR_TO_MEM, 4202 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4203 }; 4204 4205 static unsigned short bpf_tunnel_key_af(u64 flags) 4206 { 4207 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET; 4208 } 4209 4210 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to, 4211 u32, size, u64, flags) 4212 { 4213 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4214 u8 compat[sizeof(struct bpf_tunnel_key)]; 4215 void *to_orig = to; 4216 int err; 4217 4218 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) { 4219 err = -EINVAL; 4220 goto err_clear; 4221 } 4222 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { 4223 err = -EPROTO; 4224 goto err_clear; 4225 } 4226 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4227 err = -EINVAL; 4228 switch (size) { 4229 case offsetof(struct bpf_tunnel_key, tunnel_label): 4230 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4231 goto set_compat; 4232 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4233 /* Fixup deprecated structure layouts here, so we have 4234 * a common path later on. 4235 */ 4236 if (ip_tunnel_info_af(info) != AF_INET) 4237 goto err_clear; 4238 set_compat: 4239 to = (struct bpf_tunnel_key *)compat; 4240 break; 4241 default: 4242 goto err_clear; 4243 } 4244 } 4245 4246 to->tunnel_id = be64_to_cpu(info->key.tun_id); 4247 to->tunnel_tos = info->key.tos; 4248 to->tunnel_ttl = info->key.ttl; 4249 to->tunnel_ext = 0; 4250 4251 if (flags & BPF_F_TUNINFO_IPV6) { 4252 memcpy(to->remote_ipv6, &info->key.u.ipv6.src, 4253 sizeof(to->remote_ipv6)); 4254 to->tunnel_label = be32_to_cpu(info->key.label); 4255 } else { 4256 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); 4257 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 4258 to->tunnel_label = 0; 4259 } 4260 4261 if (unlikely(size != sizeof(struct bpf_tunnel_key))) 4262 memcpy(to_orig, to, size); 4263 4264 return 0; 4265 err_clear: 4266 memset(to_orig, 0, size); 4267 return err; 4268 } 4269 4270 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { 4271 .func = bpf_skb_get_tunnel_key, 4272 .gpl_only = false, 4273 .ret_type = RET_INTEGER, 4274 .arg1_type = ARG_PTR_TO_CTX, 4275 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4276 .arg3_type = ARG_CONST_SIZE, 4277 .arg4_type = ARG_ANYTHING, 4278 }; 4279 4280 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size) 4281 { 4282 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4283 int err; 4284 4285 if (unlikely(!info || 4286 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) { 4287 err = -ENOENT; 4288 goto err_clear; 4289 } 4290 if (unlikely(size < info->options_len)) { 4291 err = -ENOMEM; 4292 goto err_clear; 4293 } 4294 4295 ip_tunnel_info_opts_get(to, info); 4296 if (size > info->options_len) 4297 memset(to + info->options_len, 0, size - info->options_len); 4298 4299 return info->options_len; 4300 err_clear: 4301 memset(to, 0, size); 4302 return err; 4303 } 4304 4305 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { 4306 .func = bpf_skb_get_tunnel_opt, 4307 .gpl_only = false, 4308 .ret_type = RET_INTEGER, 4309 .arg1_type = ARG_PTR_TO_CTX, 4310 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4311 .arg3_type = ARG_CONST_SIZE, 4312 }; 4313 4314 static struct metadata_dst __percpu *md_dst; 4315 4316 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb, 4317 const struct bpf_tunnel_key *, from, u32, size, u64, flags) 4318 { 4319 struct metadata_dst *md = this_cpu_ptr(md_dst); 4320 u8 compat[sizeof(struct bpf_tunnel_key)]; 4321 struct ip_tunnel_info *info; 4322 4323 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | 4324 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER))) 4325 return -EINVAL; 4326 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4327 switch (size) { 4328 case offsetof(struct bpf_tunnel_key, tunnel_label): 4329 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4330 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4331 /* Fixup deprecated structure layouts here, so we have 4332 * a common path later on. 4333 */ 4334 memcpy(compat, from, size); 4335 memset(compat + size, 0, sizeof(compat) - size); 4336 from = (const struct bpf_tunnel_key *) compat; 4337 break; 4338 default: 4339 return -EINVAL; 4340 } 4341 } 4342 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || 4343 from->tunnel_ext)) 4344 return -EINVAL; 4345 4346 skb_dst_drop(skb); 4347 dst_hold((struct dst_entry *) md); 4348 skb_dst_set(skb, (struct dst_entry *) md); 4349 4350 info = &md->u.tun_info; 4351 memset(info, 0, sizeof(*info)); 4352 info->mode = IP_TUNNEL_INFO_TX; 4353 4354 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE; 4355 if (flags & BPF_F_DONT_FRAGMENT) 4356 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT; 4357 if (flags & BPF_F_ZERO_CSUM_TX) 4358 info->key.tun_flags &= ~TUNNEL_CSUM; 4359 if (flags & BPF_F_SEQ_NUMBER) 4360 info->key.tun_flags |= TUNNEL_SEQ; 4361 4362 info->key.tun_id = cpu_to_be64(from->tunnel_id); 4363 info->key.tos = from->tunnel_tos; 4364 info->key.ttl = from->tunnel_ttl; 4365 4366 if (flags & BPF_F_TUNINFO_IPV6) { 4367 info->mode |= IP_TUNNEL_INFO_IPV6; 4368 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, 4369 sizeof(from->remote_ipv6)); 4370 info->key.label = cpu_to_be32(from->tunnel_label) & 4371 IPV6_FLOWLABEL_MASK; 4372 } else { 4373 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); 4374 } 4375 4376 return 0; 4377 } 4378 4379 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { 4380 .func = bpf_skb_set_tunnel_key, 4381 .gpl_only = false, 4382 .ret_type = RET_INTEGER, 4383 .arg1_type = ARG_PTR_TO_CTX, 4384 .arg2_type = ARG_PTR_TO_MEM, 4385 .arg3_type = ARG_CONST_SIZE, 4386 .arg4_type = ARG_ANYTHING, 4387 }; 4388 4389 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb, 4390 const u8 *, from, u32, size) 4391 { 4392 struct ip_tunnel_info *info = skb_tunnel_info(skb); 4393 const struct metadata_dst *md = this_cpu_ptr(md_dst); 4394 4395 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) 4396 return -EINVAL; 4397 if (unlikely(size > IP_TUNNEL_OPTS_MAX)) 4398 return -ENOMEM; 4399 4400 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT); 4401 4402 return 0; 4403 } 4404 4405 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { 4406 .func = bpf_skb_set_tunnel_opt, 4407 .gpl_only = false, 4408 .ret_type = RET_INTEGER, 4409 .arg1_type = ARG_PTR_TO_CTX, 4410 .arg2_type = ARG_PTR_TO_MEM, 4411 .arg3_type = ARG_CONST_SIZE, 4412 }; 4413 4414 static const struct bpf_func_proto * 4415 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) 4416 { 4417 if (!md_dst) { 4418 struct metadata_dst __percpu *tmp; 4419 4420 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, 4421 METADATA_IP_TUNNEL, 4422 GFP_KERNEL); 4423 if (!tmp) 4424 return NULL; 4425 if (cmpxchg(&md_dst, NULL, tmp)) 4426 metadata_dst_free_percpu(tmp); 4427 } 4428 4429 switch (which) { 4430 case BPF_FUNC_skb_set_tunnel_key: 4431 return &bpf_skb_set_tunnel_key_proto; 4432 case BPF_FUNC_skb_set_tunnel_opt: 4433 return &bpf_skb_set_tunnel_opt_proto; 4434 default: 4435 return NULL; 4436 } 4437 } 4438 4439 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map, 4440 u32, idx) 4441 { 4442 struct bpf_array *array = container_of(map, struct bpf_array, map); 4443 struct cgroup *cgrp; 4444 struct sock *sk; 4445 4446 sk = skb_to_full_sk(skb); 4447 if (!sk || !sk_fullsock(sk)) 4448 return -ENOENT; 4449 if (unlikely(idx >= array->map.max_entries)) 4450 return -E2BIG; 4451 4452 cgrp = READ_ONCE(array->ptrs[idx]); 4453 if (unlikely(!cgrp)) 4454 return -EAGAIN; 4455 4456 return sk_under_cgroup_hierarchy(sk, cgrp); 4457 } 4458 4459 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = { 4460 .func = bpf_skb_under_cgroup, 4461 .gpl_only = false, 4462 .ret_type = RET_INTEGER, 4463 .arg1_type = ARG_PTR_TO_CTX, 4464 .arg2_type = ARG_CONST_MAP_PTR, 4465 .arg3_type = ARG_ANYTHING, 4466 }; 4467 4468 #ifdef CONFIG_SOCK_CGROUP_DATA 4469 static inline u64 __bpf_sk_cgroup_id(struct sock *sk) 4470 { 4471 struct cgroup *cgrp; 4472 4473 sk = sk_to_full_sk(sk); 4474 if (!sk || !sk_fullsock(sk)) 4475 return 0; 4476 4477 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4478 return cgroup_id(cgrp); 4479 } 4480 4481 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb) 4482 { 4483 return __bpf_sk_cgroup_id(skb->sk); 4484 } 4485 4486 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = { 4487 .func = bpf_skb_cgroup_id, 4488 .gpl_only = false, 4489 .ret_type = RET_INTEGER, 4490 .arg1_type = ARG_PTR_TO_CTX, 4491 }; 4492 4493 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk, 4494 int ancestor_level) 4495 { 4496 struct cgroup *ancestor; 4497 struct cgroup *cgrp; 4498 4499 sk = sk_to_full_sk(sk); 4500 if (!sk || !sk_fullsock(sk)) 4501 return 0; 4502 4503 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4504 ancestor = cgroup_ancestor(cgrp, ancestor_level); 4505 if (!ancestor) 4506 return 0; 4507 4508 return cgroup_id(ancestor); 4509 } 4510 4511 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int, 4512 ancestor_level) 4513 { 4514 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level); 4515 } 4516 4517 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = { 4518 .func = bpf_skb_ancestor_cgroup_id, 4519 .gpl_only = false, 4520 .ret_type = RET_INTEGER, 4521 .arg1_type = ARG_PTR_TO_CTX, 4522 .arg2_type = ARG_ANYTHING, 4523 }; 4524 4525 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk) 4526 { 4527 return __bpf_sk_cgroup_id(sk); 4528 } 4529 4530 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = { 4531 .func = bpf_sk_cgroup_id, 4532 .gpl_only = false, 4533 .ret_type = RET_INTEGER, 4534 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4535 }; 4536 4537 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level) 4538 { 4539 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level); 4540 } 4541 4542 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = { 4543 .func = bpf_sk_ancestor_cgroup_id, 4544 .gpl_only = false, 4545 .ret_type = RET_INTEGER, 4546 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4547 .arg2_type = ARG_ANYTHING, 4548 }; 4549 #endif 4550 4551 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff, 4552 unsigned long off, unsigned long len) 4553 { 4554 memcpy(dst_buff, src_buff + off, len); 4555 return 0; 4556 } 4557 4558 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map, 4559 u64, flags, void *, meta, u64, meta_size) 4560 { 4561 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4562 4563 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4564 return -EINVAL; 4565 if (unlikely(!xdp || 4566 xdp_size > (unsigned long)(xdp->data_end - xdp->data))) 4567 return -EFAULT; 4568 4569 return bpf_event_output(map, flags, meta, meta_size, xdp->data, 4570 xdp_size, bpf_xdp_copy); 4571 } 4572 4573 static const struct bpf_func_proto bpf_xdp_event_output_proto = { 4574 .func = bpf_xdp_event_output, 4575 .gpl_only = true, 4576 .ret_type = RET_INTEGER, 4577 .arg1_type = ARG_PTR_TO_CTX, 4578 .arg2_type = ARG_CONST_MAP_PTR, 4579 .arg3_type = ARG_ANYTHING, 4580 .arg4_type = ARG_PTR_TO_MEM, 4581 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4582 }; 4583 4584 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff) 4585 4586 const struct bpf_func_proto bpf_xdp_output_proto = { 4587 .func = bpf_xdp_event_output, 4588 .gpl_only = true, 4589 .ret_type = RET_INTEGER, 4590 .arg1_type = ARG_PTR_TO_BTF_ID, 4591 .arg1_btf_id = &bpf_xdp_output_btf_ids[0], 4592 .arg2_type = ARG_CONST_MAP_PTR, 4593 .arg3_type = ARG_ANYTHING, 4594 .arg4_type = ARG_PTR_TO_MEM, 4595 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4596 }; 4597 4598 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb) 4599 { 4600 return skb->sk ? __sock_gen_cookie(skb->sk) : 0; 4601 } 4602 4603 static const struct bpf_func_proto bpf_get_socket_cookie_proto = { 4604 .func = bpf_get_socket_cookie, 4605 .gpl_only = false, 4606 .ret_type = RET_INTEGER, 4607 .arg1_type = ARG_PTR_TO_CTX, 4608 }; 4609 4610 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4611 { 4612 return __sock_gen_cookie(ctx->sk); 4613 } 4614 4615 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = { 4616 .func = bpf_get_socket_cookie_sock_addr, 4617 .gpl_only = false, 4618 .ret_type = RET_INTEGER, 4619 .arg1_type = ARG_PTR_TO_CTX, 4620 }; 4621 4622 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx) 4623 { 4624 return __sock_gen_cookie(ctx); 4625 } 4626 4627 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = { 4628 .func = bpf_get_socket_cookie_sock, 4629 .gpl_only = false, 4630 .ret_type = RET_INTEGER, 4631 .arg1_type = ARG_PTR_TO_CTX, 4632 }; 4633 4634 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 4635 { 4636 return __sock_gen_cookie(ctx->sk); 4637 } 4638 4639 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = { 4640 .func = bpf_get_socket_cookie_sock_ops, 4641 .gpl_only = false, 4642 .ret_type = RET_INTEGER, 4643 .arg1_type = ARG_PTR_TO_CTX, 4644 }; 4645 4646 static u64 __bpf_get_netns_cookie(struct sock *sk) 4647 { 4648 #ifdef CONFIG_NET_NS 4649 return __net_gen_cookie(sk ? sk->sk_net.net : &init_net); 4650 #else 4651 return 0; 4652 #endif 4653 } 4654 4655 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx) 4656 { 4657 return __bpf_get_netns_cookie(ctx); 4658 } 4659 4660 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = { 4661 .func = bpf_get_netns_cookie_sock, 4662 .gpl_only = false, 4663 .ret_type = RET_INTEGER, 4664 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4665 }; 4666 4667 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4668 { 4669 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4670 } 4671 4672 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = { 4673 .func = bpf_get_netns_cookie_sock_addr, 4674 .gpl_only = false, 4675 .ret_type = RET_INTEGER, 4676 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4677 }; 4678 4679 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb) 4680 { 4681 struct sock *sk = sk_to_full_sk(skb->sk); 4682 kuid_t kuid; 4683 4684 if (!sk || !sk_fullsock(sk)) 4685 return overflowuid; 4686 kuid = sock_net_uid(sock_net(sk), sk); 4687 return from_kuid_munged(sock_net(sk)->user_ns, kuid); 4688 } 4689 4690 static const struct bpf_func_proto bpf_get_socket_uid_proto = { 4691 .func = bpf_get_socket_uid, 4692 .gpl_only = false, 4693 .ret_type = RET_INTEGER, 4694 .arg1_type = ARG_PTR_TO_CTX, 4695 }; 4696 4697 static int _bpf_setsockopt(struct sock *sk, int level, int optname, 4698 char *optval, int optlen) 4699 { 4700 char devname[IFNAMSIZ]; 4701 int val, valbool; 4702 struct net *net; 4703 int ifindex; 4704 int ret = 0; 4705 4706 if (!sk_fullsock(sk)) 4707 return -EINVAL; 4708 4709 sock_owned_by_me(sk); 4710 4711 if (level == SOL_SOCKET) { 4712 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE) 4713 return -EINVAL; 4714 val = *((int *)optval); 4715 valbool = val ? 1 : 0; 4716 4717 /* Only some socketops are supported */ 4718 switch (optname) { 4719 case SO_RCVBUF: 4720 val = min_t(u32, val, sysctl_rmem_max); 4721 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 4722 WRITE_ONCE(sk->sk_rcvbuf, 4723 max_t(int, val * 2, SOCK_MIN_RCVBUF)); 4724 break; 4725 case SO_SNDBUF: 4726 val = min_t(u32, val, sysctl_wmem_max); 4727 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 4728 WRITE_ONCE(sk->sk_sndbuf, 4729 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 4730 break; 4731 case SO_MAX_PACING_RATE: /* 32bit version */ 4732 if (val != ~0U) 4733 cmpxchg(&sk->sk_pacing_status, 4734 SK_PACING_NONE, 4735 SK_PACING_NEEDED); 4736 sk->sk_max_pacing_rate = (val == ~0U) ? 4737 ~0UL : (unsigned int)val; 4738 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 4739 sk->sk_max_pacing_rate); 4740 break; 4741 case SO_PRIORITY: 4742 sk->sk_priority = val; 4743 break; 4744 case SO_RCVLOWAT: 4745 if (val < 0) 4746 val = INT_MAX; 4747 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 4748 break; 4749 case SO_MARK: 4750 if (sk->sk_mark != val) { 4751 sk->sk_mark = val; 4752 sk_dst_reset(sk); 4753 } 4754 break; 4755 case SO_BINDTODEVICE: 4756 optlen = min_t(long, optlen, IFNAMSIZ - 1); 4757 strncpy(devname, optval, optlen); 4758 devname[optlen] = 0; 4759 4760 ifindex = 0; 4761 if (devname[0] != '\0') { 4762 struct net_device *dev; 4763 4764 ret = -ENODEV; 4765 4766 net = sock_net(sk); 4767 dev = dev_get_by_name(net, devname); 4768 if (!dev) 4769 break; 4770 ifindex = dev->ifindex; 4771 dev_put(dev); 4772 } 4773 ret = sock_bindtoindex(sk, ifindex, false); 4774 break; 4775 case SO_KEEPALIVE: 4776 if (sk->sk_prot->keepalive) 4777 sk->sk_prot->keepalive(sk, valbool); 4778 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 4779 break; 4780 default: 4781 ret = -EINVAL; 4782 } 4783 #ifdef CONFIG_INET 4784 } else if (level == SOL_IP) { 4785 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 4786 return -EINVAL; 4787 4788 val = *((int *)optval); 4789 /* Only some options are supported */ 4790 switch (optname) { 4791 case IP_TOS: 4792 if (val < -1 || val > 0xff) { 4793 ret = -EINVAL; 4794 } else { 4795 struct inet_sock *inet = inet_sk(sk); 4796 4797 if (val == -1) 4798 val = 0; 4799 inet->tos = val; 4800 } 4801 break; 4802 default: 4803 ret = -EINVAL; 4804 } 4805 #if IS_ENABLED(CONFIG_IPV6) 4806 } else if (level == SOL_IPV6) { 4807 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 4808 return -EINVAL; 4809 4810 val = *((int *)optval); 4811 /* Only some options are supported */ 4812 switch (optname) { 4813 case IPV6_TCLASS: 4814 if (val < -1 || val > 0xff) { 4815 ret = -EINVAL; 4816 } else { 4817 struct ipv6_pinfo *np = inet6_sk(sk); 4818 4819 if (val == -1) 4820 val = 0; 4821 np->tclass = val; 4822 } 4823 break; 4824 default: 4825 ret = -EINVAL; 4826 } 4827 #endif 4828 } else if (level == SOL_TCP && 4829 sk->sk_prot->setsockopt == tcp_setsockopt) { 4830 if (optname == TCP_CONGESTION) { 4831 char name[TCP_CA_NAME_MAX]; 4832 4833 strncpy(name, optval, min_t(long, optlen, 4834 TCP_CA_NAME_MAX-1)); 4835 name[TCP_CA_NAME_MAX-1] = 0; 4836 ret = tcp_set_congestion_control(sk, name, false, true); 4837 } else { 4838 struct inet_connection_sock *icsk = inet_csk(sk); 4839 struct tcp_sock *tp = tcp_sk(sk); 4840 unsigned long timeout; 4841 4842 if (optlen != sizeof(int)) 4843 return -EINVAL; 4844 4845 val = *((int *)optval); 4846 /* Only some options are supported */ 4847 switch (optname) { 4848 case TCP_BPF_IW: 4849 if (val <= 0 || tp->data_segs_out > tp->syn_data) 4850 ret = -EINVAL; 4851 else 4852 tp->snd_cwnd = val; 4853 break; 4854 case TCP_BPF_SNDCWND_CLAMP: 4855 if (val <= 0) { 4856 ret = -EINVAL; 4857 } else { 4858 tp->snd_cwnd_clamp = val; 4859 tp->snd_ssthresh = val; 4860 } 4861 break; 4862 case TCP_BPF_DELACK_MAX: 4863 timeout = usecs_to_jiffies(val); 4864 if (timeout > TCP_DELACK_MAX || 4865 timeout < TCP_TIMEOUT_MIN) 4866 return -EINVAL; 4867 inet_csk(sk)->icsk_delack_max = timeout; 4868 break; 4869 case TCP_BPF_RTO_MIN: 4870 timeout = usecs_to_jiffies(val); 4871 if (timeout > TCP_RTO_MIN || 4872 timeout < TCP_TIMEOUT_MIN) 4873 return -EINVAL; 4874 inet_csk(sk)->icsk_rto_min = timeout; 4875 break; 4876 case TCP_SAVE_SYN: 4877 if (val < 0 || val > 1) 4878 ret = -EINVAL; 4879 else 4880 tp->save_syn = val; 4881 break; 4882 case TCP_KEEPIDLE: 4883 ret = tcp_sock_set_keepidle_locked(sk, val); 4884 break; 4885 case TCP_KEEPINTVL: 4886 if (val < 1 || val > MAX_TCP_KEEPINTVL) 4887 ret = -EINVAL; 4888 else 4889 tp->keepalive_intvl = val * HZ; 4890 break; 4891 case TCP_KEEPCNT: 4892 if (val < 1 || val > MAX_TCP_KEEPCNT) 4893 ret = -EINVAL; 4894 else 4895 tp->keepalive_probes = val; 4896 break; 4897 case TCP_SYNCNT: 4898 if (val < 1 || val > MAX_TCP_SYNCNT) 4899 ret = -EINVAL; 4900 else 4901 icsk->icsk_syn_retries = val; 4902 break; 4903 case TCP_USER_TIMEOUT: 4904 if (val < 0) 4905 ret = -EINVAL; 4906 else 4907 icsk->icsk_user_timeout = val; 4908 break; 4909 case TCP_NOTSENT_LOWAT: 4910 tp->notsent_lowat = val; 4911 sk->sk_write_space(sk); 4912 break; 4913 case TCP_WINDOW_CLAMP: 4914 ret = tcp_set_window_clamp(sk, val); 4915 break; 4916 default: 4917 ret = -EINVAL; 4918 } 4919 } 4920 #endif 4921 } else { 4922 ret = -EINVAL; 4923 } 4924 return ret; 4925 } 4926 4927 static int _bpf_getsockopt(struct sock *sk, int level, int optname, 4928 char *optval, int optlen) 4929 { 4930 if (!sk_fullsock(sk)) 4931 goto err_clear; 4932 4933 sock_owned_by_me(sk); 4934 4935 #ifdef CONFIG_INET 4936 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) { 4937 struct inet_connection_sock *icsk; 4938 struct tcp_sock *tp; 4939 4940 switch (optname) { 4941 case TCP_CONGESTION: 4942 icsk = inet_csk(sk); 4943 4944 if (!icsk->icsk_ca_ops || optlen <= 1) 4945 goto err_clear; 4946 strncpy(optval, icsk->icsk_ca_ops->name, optlen); 4947 optval[optlen - 1] = 0; 4948 break; 4949 case TCP_SAVED_SYN: 4950 tp = tcp_sk(sk); 4951 4952 if (optlen <= 0 || !tp->saved_syn || 4953 optlen > tcp_saved_syn_len(tp->saved_syn)) 4954 goto err_clear; 4955 memcpy(optval, tp->saved_syn->data, optlen); 4956 break; 4957 default: 4958 goto err_clear; 4959 } 4960 } else if (level == SOL_IP) { 4961 struct inet_sock *inet = inet_sk(sk); 4962 4963 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 4964 goto err_clear; 4965 4966 /* Only some options are supported */ 4967 switch (optname) { 4968 case IP_TOS: 4969 *((int *)optval) = (int)inet->tos; 4970 break; 4971 default: 4972 goto err_clear; 4973 } 4974 #if IS_ENABLED(CONFIG_IPV6) 4975 } else if (level == SOL_IPV6) { 4976 struct ipv6_pinfo *np = inet6_sk(sk); 4977 4978 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 4979 goto err_clear; 4980 4981 /* Only some options are supported */ 4982 switch (optname) { 4983 case IPV6_TCLASS: 4984 *((int *)optval) = (int)np->tclass; 4985 break; 4986 default: 4987 goto err_clear; 4988 } 4989 #endif 4990 } else { 4991 goto err_clear; 4992 } 4993 return 0; 4994 #endif 4995 err_clear: 4996 memset(optval, 0, optlen); 4997 return -EINVAL; 4998 } 4999 5000 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx, 5001 int, level, int, optname, char *, optval, int, optlen) 5002 { 5003 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen); 5004 } 5005 5006 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = { 5007 .func = bpf_sock_addr_setsockopt, 5008 .gpl_only = false, 5009 .ret_type = RET_INTEGER, 5010 .arg1_type = ARG_PTR_TO_CTX, 5011 .arg2_type = ARG_ANYTHING, 5012 .arg3_type = ARG_ANYTHING, 5013 .arg4_type = ARG_PTR_TO_MEM, 5014 .arg5_type = ARG_CONST_SIZE, 5015 }; 5016 5017 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx, 5018 int, level, int, optname, char *, optval, int, optlen) 5019 { 5020 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen); 5021 } 5022 5023 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = { 5024 .func = bpf_sock_addr_getsockopt, 5025 .gpl_only = false, 5026 .ret_type = RET_INTEGER, 5027 .arg1_type = ARG_PTR_TO_CTX, 5028 .arg2_type = ARG_ANYTHING, 5029 .arg3_type = ARG_ANYTHING, 5030 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5031 .arg5_type = ARG_CONST_SIZE, 5032 }; 5033 5034 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5035 int, level, int, optname, char *, optval, int, optlen) 5036 { 5037 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen); 5038 } 5039 5040 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = { 5041 .func = bpf_sock_ops_setsockopt, 5042 .gpl_only = false, 5043 .ret_type = RET_INTEGER, 5044 .arg1_type = ARG_PTR_TO_CTX, 5045 .arg2_type = ARG_ANYTHING, 5046 .arg3_type = ARG_ANYTHING, 5047 .arg4_type = ARG_PTR_TO_MEM, 5048 .arg5_type = ARG_CONST_SIZE, 5049 }; 5050 5051 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock, 5052 int optname, const u8 **start) 5053 { 5054 struct sk_buff *syn_skb = bpf_sock->syn_skb; 5055 const u8 *hdr_start; 5056 int ret; 5057 5058 if (syn_skb) { 5059 /* sk is a request_sock here */ 5060 5061 if (optname == TCP_BPF_SYN) { 5062 hdr_start = syn_skb->data; 5063 ret = tcp_hdrlen(syn_skb); 5064 } else if (optname == TCP_BPF_SYN_IP) { 5065 hdr_start = skb_network_header(syn_skb); 5066 ret = skb_network_header_len(syn_skb) + 5067 tcp_hdrlen(syn_skb); 5068 } else { 5069 /* optname == TCP_BPF_SYN_MAC */ 5070 hdr_start = skb_mac_header(syn_skb); 5071 ret = skb_mac_header_len(syn_skb) + 5072 skb_network_header_len(syn_skb) + 5073 tcp_hdrlen(syn_skb); 5074 } 5075 } else { 5076 struct sock *sk = bpf_sock->sk; 5077 struct saved_syn *saved_syn; 5078 5079 if (sk->sk_state == TCP_NEW_SYN_RECV) 5080 /* synack retransmit. bpf_sock->syn_skb will 5081 * not be available. It has to resort to 5082 * saved_syn (if it is saved). 5083 */ 5084 saved_syn = inet_reqsk(sk)->saved_syn; 5085 else 5086 saved_syn = tcp_sk(sk)->saved_syn; 5087 5088 if (!saved_syn) 5089 return -ENOENT; 5090 5091 if (optname == TCP_BPF_SYN) { 5092 hdr_start = saved_syn->data + 5093 saved_syn->mac_hdrlen + 5094 saved_syn->network_hdrlen; 5095 ret = saved_syn->tcp_hdrlen; 5096 } else if (optname == TCP_BPF_SYN_IP) { 5097 hdr_start = saved_syn->data + 5098 saved_syn->mac_hdrlen; 5099 ret = saved_syn->network_hdrlen + 5100 saved_syn->tcp_hdrlen; 5101 } else { 5102 /* optname == TCP_BPF_SYN_MAC */ 5103 5104 /* TCP_SAVE_SYN may not have saved the mac hdr */ 5105 if (!saved_syn->mac_hdrlen) 5106 return -ENOENT; 5107 5108 hdr_start = saved_syn->data; 5109 ret = saved_syn->mac_hdrlen + 5110 saved_syn->network_hdrlen + 5111 saved_syn->tcp_hdrlen; 5112 } 5113 } 5114 5115 *start = hdr_start; 5116 return ret; 5117 } 5118 5119 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5120 int, level, int, optname, char *, optval, int, optlen) 5121 { 5122 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP && 5123 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) { 5124 int ret, copy_len = 0; 5125 const u8 *start; 5126 5127 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start); 5128 if (ret > 0) { 5129 copy_len = ret; 5130 if (optlen < copy_len) { 5131 copy_len = optlen; 5132 ret = -ENOSPC; 5133 } 5134 5135 memcpy(optval, start, copy_len); 5136 } 5137 5138 /* Zero out unused buffer at the end */ 5139 memset(optval + copy_len, 0, optlen - copy_len); 5140 5141 return ret; 5142 } 5143 5144 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen); 5145 } 5146 5147 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = { 5148 .func = bpf_sock_ops_getsockopt, 5149 .gpl_only = false, 5150 .ret_type = RET_INTEGER, 5151 .arg1_type = ARG_PTR_TO_CTX, 5152 .arg2_type = ARG_ANYTHING, 5153 .arg3_type = ARG_ANYTHING, 5154 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5155 .arg5_type = ARG_CONST_SIZE, 5156 }; 5157 5158 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock, 5159 int, argval) 5160 { 5161 struct sock *sk = bpf_sock->sk; 5162 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS; 5163 5164 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk)) 5165 return -EINVAL; 5166 5167 tcp_sk(sk)->bpf_sock_ops_cb_flags = val; 5168 5169 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS); 5170 } 5171 5172 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = { 5173 .func = bpf_sock_ops_cb_flags_set, 5174 .gpl_only = false, 5175 .ret_type = RET_INTEGER, 5176 .arg1_type = ARG_PTR_TO_CTX, 5177 .arg2_type = ARG_ANYTHING, 5178 }; 5179 5180 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly; 5181 EXPORT_SYMBOL_GPL(ipv6_bpf_stub); 5182 5183 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr, 5184 int, addr_len) 5185 { 5186 #ifdef CONFIG_INET 5187 struct sock *sk = ctx->sk; 5188 u32 flags = BIND_FROM_BPF; 5189 int err; 5190 5191 err = -EINVAL; 5192 if (addr_len < offsetofend(struct sockaddr, sa_family)) 5193 return err; 5194 if (addr->sa_family == AF_INET) { 5195 if (addr_len < sizeof(struct sockaddr_in)) 5196 return err; 5197 if (((struct sockaddr_in *)addr)->sin_port == htons(0)) 5198 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5199 return __inet_bind(sk, addr, addr_len, flags); 5200 #if IS_ENABLED(CONFIG_IPV6) 5201 } else if (addr->sa_family == AF_INET6) { 5202 if (addr_len < SIN6_LEN_RFC2133) 5203 return err; 5204 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0)) 5205 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5206 /* ipv6_bpf_stub cannot be NULL, since it's called from 5207 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded 5208 */ 5209 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags); 5210 #endif /* CONFIG_IPV6 */ 5211 } 5212 #endif /* CONFIG_INET */ 5213 5214 return -EAFNOSUPPORT; 5215 } 5216 5217 static const struct bpf_func_proto bpf_bind_proto = { 5218 .func = bpf_bind, 5219 .gpl_only = false, 5220 .ret_type = RET_INTEGER, 5221 .arg1_type = ARG_PTR_TO_CTX, 5222 .arg2_type = ARG_PTR_TO_MEM, 5223 .arg3_type = ARG_CONST_SIZE, 5224 }; 5225 5226 #ifdef CONFIG_XFRM 5227 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index, 5228 struct bpf_xfrm_state *, to, u32, size, u64, flags) 5229 { 5230 const struct sec_path *sp = skb_sec_path(skb); 5231 const struct xfrm_state *x; 5232 5233 if (!sp || unlikely(index >= sp->len || flags)) 5234 goto err_clear; 5235 5236 x = sp->xvec[index]; 5237 5238 if (unlikely(size != sizeof(struct bpf_xfrm_state))) 5239 goto err_clear; 5240 5241 to->reqid = x->props.reqid; 5242 to->spi = x->id.spi; 5243 to->family = x->props.family; 5244 to->ext = 0; 5245 5246 if (to->family == AF_INET6) { 5247 memcpy(to->remote_ipv6, x->props.saddr.a6, 5248 sizeof(to->remote_ipv6)); 5249 } else { 5250 to->remote_ipv4 = x->props.saddr.a4; 5251 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 5252 } 5253 5254 return 0; 5255 err_clear: 5256 memset(to, 0, size); 5257 return -EINVAL; 5258 } 5259 5260 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = { 5261 .func = bpf_skb_get_xfrm_state, 5262 .gpl_only = false, 5263 .ret_type = RET_INTEGER, 5264 .arg1_type = ARG_PTR_TO_CTX, 5265 .arg2_type = ARG_ANYTHING, 5266 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 5267 .arg4_type = ARG_CONST_SIZE, 5268 .arg5_type = ARG_ANYTHING, 5269 }; 5270 #endif 5271 5272 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6) 5273 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, 5274 const struct neighbour *neigh, 5275 const struct net_device *dev) 5276 { 5277 memcpy(params->dmac, neigh->ha, ETH_ALEN); 5278 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 5279 params->h_vlan_TCI = 0; 5280 params->h_vlan_proto = 0; 5281 5282 return 0; 5283 } 5284 #endif 5285 5286 #if IS_ENABLED(CONFIG_INET) 5287 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5288 u32 flags, bool check_mtu) 5289 { 5290 struct fib_nh_common *nhc; 5291 struct in_device *in_dev; 5292 struct neighbour *neigh; 5293 struct net_device *dev; 5294 struct fib_result res; 5295 struct flowi4 fl4; 5296 int err; 5297 u32 mtu; 5298 5299 dev = dev_get_by_index_rcu(net, params->ifindex); 5300 if (unlikely(!dev)) 5301 return -ENODEV; 5302 5303 /* verify forwarding is enabled on this interface */ 5304 in_dev = __in_dev_get_rcu(dev); 5305 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev))) 5306 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5307 5308 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5309 fl4.flowi4_iif = 1; 5310 fl4.flowi4_oif = params->ifindex; 5311 } else { 5312 fl4.flowi4_iif = params->ifindex; 5313 fl4.flowi4_oif = 0; 5314 } 5315 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK; 5316 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 5317 fl4.flowi4_flags = 0; 5318 5319 fl4.flowi4_proto = params->l4_protocol; 5320 fl4.daddr = params->ipv4_dst; 5321 fl4.saddr = params->ipv4_src; 5322 fl4.fl4_sport = params->sport; 5323 fl4.fl4_dport = params->dport; 5324 fl4.flowi4_multipath_hash = 0; 5325 5326 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5327 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5328 struct fib_table *tb; 5329 5330 tb = fib_get_table(net, tbid); 5331 if (unlikely(!tb)) 5332 return BPF_FIB_LKUP_RET_NOT_FWDED; 5333 5334 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 5335 } else { 5336 fl4.flowi4_mark = 0; 5337 fl4.flowi4_secid = 0; 5338 fl4.flowi4_tun_key.tun_id = 0; 5339 fl4.flowi4_uid = sock_net_uid(net, NULL); 5340 5341 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF); 5342 } 5343 5344 if (err) { 5345 /* map fib lookup errors to RTN_ type */ 5346 if (err == -EINVAL) 5347 return BPF_FIB_LKUP_RET_BLACKHOLE; 5348 if (err == -EHOSTUNREACH) 5349 return BPF_FIB_LKUP_RET_UNREACHABLE; 5350 if (err == -EACCES) 5351 return BPF_FIB_LKUP_RET_PROHIBIT; 5352 5353 return BPF_FIB_LKUP_RET_NOT_FWDED; 5354 } 5355 5356 if (res.type != RTN_UNICAST) 5357 return BPF_FIB_LKUP_RET_NOT_FWDED; 5358 5359 if (fib_info_num_path(res.fi) > 1) 5360 fib_select_path(net, &res, &fl4, NULL); 5361 5362 if (check_mtu) { 5363 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst); 5364 if (params->tot_len > mtu) 5365 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5366 } 5367 5368 nhc = res.nhc; 5369 5370 /* do not handle lwt encaps right now */ 5371 if (nhc->nhc_lwtstate) 5372 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5373 5374 dev = nhc->nhc_dev; 5375 5376 params->rt_metric = res.fi->fib_priority; 5377 params->ifindex = dev->ifindex; 5378 5379 /* xdp and cls_bpf programs are run in RCU-bh so 5380 * rcu_read_lock_bh is not needed here 5381 */ 5382 if (likely(nhc->nhc_gw_family != AF_INET6)) { 5383 if (nhc->nhc_gw_family) 5384 params->ipv4_dst = nhc->nhc_gw.ipv4; 5385 5386 neigh = __ipv4_neigh_lookup_noref(dev, 5387 (__force u32)params->ipv4_dst); 5388 } else { 5389 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst; 5390 5391 params->family = AF_INET6; 5392 *dst = nhc->nhc_gw.ipv6; 5393 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5394 } 5395 5396 if (!neigh) 5397 return BPF_FIB_LKUP_RET_NO_NEIGH; 5398 5399 return bpf_fib_set_fwd_params(params, neigh, dev); 5400 } 5401 #endif 5402 5403 #if IS_ENABLED(CONFIG_IPV6) 5404 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5405 u32 flags, bool check_mtu) 5406 { 5407 struct in6_addr *src = (struct in6_addr *) params->ipv6_src; 5408 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst; 5409 struct fib6_result res = {}; 5410 struct neighbour *neigh; 5411 struct net_device *dev; 5412 struct inet6_dev *idev; 5413 struct flowi6 fl6; 5414 int strict = 0; 5415 int oif, err; 5416 u32 mtu; 5417 5418 /* link local addresses are never forwarded */ 5419 if (rt6_need_strict(dst) || rt6_need_strict(src)) 5420 return BPF_FIB_LKUP_RET_NOT_FWDED; 5421 5422 dev = dev_get_by_index_rcu(net, params->ifindex); 5423 if (unlikely(!dev)) 5424 return -ENODEV; 5425 5426 idev = __in6_dev_get_safely(dev); 5427 if (unlikely(!idev || !idev->cnf.forwarding)) 5428 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5429 5430 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5431 fl6.flowi6_iif = 1; 5432 oif = fl6.flowi6_oif = params->ifindex; 5433 } else { 5434 oif = fl6.flowi6_iif = params->ifindex; 5435 fl6.flowi6_oif = 0; 5436 strict = RT6_LOOKUP_F_HAS_SADDR; 5437 } 5438 fl6.flowlabel = params->flowinfo; 5439 fl6.flowi6_scope = 0; 5440 fl6.flowi6_flags = 0; 5441 fl6.mp_hash = 0; 5442 5443 fl6.flowi6_proto = params->l4_protocol; 5444 fl6.daddr = *dst; 5445 fl6.saddr = *src; 5446 fl6.fl6_sport = params->sport; 5447 fl6.fl6_dport = params->dport; 5448 5449 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5450 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5451 struct fib6_table *tb; 5452 5453 tb = ipv6_stub->fib6_get_table(net, tbid); 5454 if (unlikely(!tb)) 5455 return BPF_FIB_LKUP_RET_NOT_FWDED; 5456 5457 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res, 5458 strict); 5459 } else { 5460 fl6.flowi6_mark = 0; 5461 fl6.flowi6_secid = 0; 5462 fl6.flowi6_tun_key.tun_id = 0; 5463 fl6.flowi6_uid = sock_net_uid(net, NULL); 5464 5465 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict); 5466 } 5467 5468 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) || 5469 res.f6i == net->ipv6.fib6_null_entry)) 5470 return BPF_FIB_LKUP_RET_NOT_FWDED; 5471 5472 switch (res.fib6_type) { 5473 /* only unicast is forwarded */ 5474 case RTN_UNICAST: 5475 break; 5476 case RTN_BLACKHOLE: 5477 return BPF_FIB_LKUP_RET_BLACKHOLE; 5478 case RTN_UNREACHABLE: 5479 return BPF_FIB_LKUP_RET_UNREACHABLE; 5480 case RTN_PROHIBIT: 5481 return BPF_FIB_LKUP_RET_PROHIBIT; 5482 default: 5483 return BPF_FIB_LKUP_RET_NOT_FWDED; 5484 } 5485 5486 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif, 5487 fl6.flowi6_oif != 0, NULL, strict); 5488 5489 if (check_mtu) { 5490 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src); 5491 if (params->tot_len > mtu) 5492 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5493 } 5494 5495 if (res.nh->fib_nh_lws) 5496 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5497 5498 if (res.nh->fib_nh_gw_family) 5499 *dst = res.nh->fib_nh_gw6; 5500 5501 dev = res.nh->fib_nh_dev; 5502 params->rt_metric = res.f6i->fib6_metric; 5503 params->ifindex = dev->ifindex; 5504 5505 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is 5506 * not needed here. 5507 */ 5508 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5509 if (!neigh) 5510 return BPF_FIB_LKUP_RET_NO_NEIGH; 5511 5512 return bpf_fib_set_fwd_params(params, neigh, dev); 5513 } 5514 #endif 5515 5516 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx, 5517 struct bpf_fib_lookup *, params, int, plen, u32, flags) 5518 { 5519 if (plen < sizeof(*params)) 5520 return -EINVAL; 5521 5522 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 5523 return -EINVAL; 5524 5525 switch (params->family) { 5526 #if IS_ENABLED(CONFIG_INET) 5527 case AF_INET: 5528 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params, 5529 flags, true); 5530 #endif 5531 #if IS_ENABLED(CONFIG_IPV6) 5532 case AF_INET6: 5533 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params, 5534 flags, true); 5535 #endif 5536 } 5537 return -EAFNOSUPPORT; 5538 } 5539 5540 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = { 5541 .func = bpf_xdp_fib_lookup, 5542 .gpl_only = true, 5543 .ret_type = RET_INTEGER, 5544 .arg1_type = ARG_PTR_TO_CTX, 5545 .arg2_type = ARG_PTR_TO_MEM, 5546 .arg3_type = ARG_CONST_SIZE, 5547 .arg4_type = ARG_ANYTHING, 5548 }; 5549 5550 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb, 5551 struct bpf_fib_lookup *, params, int, plen, u32, flags) 5552 { 5553 struct net *net = dev_net(skb->dev); 5554 int rc = -EAFNOSUPPORT; 5555 5556 if (plen < sizeof(*params)) 5557 return -EINVAL; 5558 5559 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 5560 return -EINVAL; 5561 5562 switch (params->family) { 5563 #if IS_ENABLED(CONFIG_INET) 5564 case AF_INET: 5565 rc = bpf_ipv4_fib_lookup(net, params, flags, false); 5566 break; 5567 #endif 5568 #if IS_ENABLED(CONFIG_IPV6) 5569 case AF_INET6: 5570 rc = bpf_ipv6_fib_lookup(net, params, flags, false); 5571 break; 5572 #endif 5573 } 5574 5575 if (!rc) { 5576 struct net_device *dev; 5577 5578 dev = dev_get_by_index_rcu(net, params->ifindex); 5579 if (!is_skb_forwardable(dev, skb)) 5580 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED; 5581 } 5582 5583 return rc; 5584 } 5585 5586 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = { 5587 .func = bpf_skb_fib_lookup, 5588 .gpl_only = true, 5589 .ret_type = RET_INTEGER, 5590 .arg1_type = ARG_PTR_TO_CTX, 5591 .arg2_type = ARG_PTR_TO_MEM, 5592 .arg3_type = ARG_CONST_SIZE, 5593 .arg4_type = ARG_ANYTHING, 5594 }; 5595 5596 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5597 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len) 5598 { 5599 int err; 5600 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr; 5601 5602 if (!seg6_validate_srh(srh, len, false)) 5603 return -EINVAL; 5604 5605 switch (type) { 5606 case BPF_LWT_ENCAP_SEG6_INLINE: 5607 if (skb->protocol != htons(ETH_P_IPV6)) 5608 return -EBADMSG; 5609 5610 err = seg6_do_srh_inline(skb, srh); 5611 break; 5612 case BPF_LWT_ENCAP_SEG6: 5613 skb_reset_inner_headers(skb); 5614 skb->encapsulation = 1; 5615 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6); 5616 break; 5617 default: 5618 return -EINVAL; 5619 } 5620 5621 bpf_compute_data_pointers(skb); 5622 if (err) 5623 return err; 5624 5625 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 5626 skb_set_transport_header(skb, sizeof(struct ipv6hdr)); 5627 5628 return seg6_lookup_nexthop(skb, NULL, 0); 5629 } 5630 #endif /* CONFIG_IPV6_SEG6_BPF */ 5631 5632 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 5633 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, 5634 bool ingress) 5635 { 5636 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress); 5637 } 5638 #endif 5639 5640 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr, 5641 u32, len) 5642 { 5643 switch (type) { 5644 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5645 case BPF_LWT_ENCAP_SEG6: 5646 case BPF_LWT_ENCAP_SEG6_INLINE: 5647 return bpf_push_seg6_encap(skb, type, hdr, len); 5648 #endif 5649 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 5650 case BPF_LWT_ENCAP_IP: 5651 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */); 5652 #endif 5653 default: 5654 return -EINVAL; 5655 } 5656 } 5657 5658 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type, 5659 void *, hdr, u32, len) 5660 { 5661 switch (type) { 5662 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 5663 case BPF_LWT_ENCAP_IP: 5664 return bpf_push_ip_encap(skb, hdr, len, false /* egress */); 5665 #endif 5666 default: 5667 return -EINVAL; 5668 } 5669 } 5670 5671 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = { 5672 .func = bpf_lwt_in_push_encap, 5673 .gpl_only = false, 5674 .ret_type = RET_INTEGER, 5675 .arg1_type = ARG_PTR_TO_CTX, 5676 .arg2_type = ARG_ANYTHING, 5677 .arg3_type = ARG_PTR_TO_MEM, 5678 .arg4_type = ARG_CONST_SIZE 5679 }; 5680 5681 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = { 5682 .func = bpf_lwt_xmit_push_encap, 5683 .gpl_only = false, 5684 .ret_type = RET_INTEGER, 5685 .arg1_type = ARG_PTR_TO_CTX, 5686 .arg2_type = ARG_ANYTHING, 5687 .arg3_type = ARG_PTR_TO_MEM, 5688 .arg4_type = ARG_CONST_SIZE 5689 }; 5690 5691 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5692 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset, 5693 const void *, from, u32, len) 5694 { 5695 struct seg6_bpf_srh_state *srh_state = 5696 this_cpu_ptr(&seg6_bpf_srh_states); 5697 struct ipv6_sr_hdr *srh = srh_state->srh; 5698 void *srh_tlvs, *srh_end, *ptr; 5699 int srhoff = 0; 5700 5701 if (srh == NULL) 5702 return -EINVAL; 5703 5704 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4)); 5705 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen); 5706 5707 ptr = skb->data + offset; 5708 if (ptr >= srh_tlvs && ptr + len <= srh_end) 5709 srh_state->valid = false; 5710 else if (ptr < (void *)&srh->flags || 5711 ptr + len > (void *)&srh->segments) 5712 return -EFAULT; 5713 5714 if (unlikely(bpf_try_make_writable(skb, offset + len))) 5715 return -EFAULT; 5716 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 5717 return -EINVAL; 5718 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5719 5720 memcpy(skb->data + offset, from, len); 5721 return 0; 5722 } 5723 5724 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = { 5725 .func = bpf_lwt_seg6_store_bytes, 5726 .gpl_only = false, 5727 .ret_type = RET_INTEGER, 5728 .arg1_type = ARG_PTR_TO_CTX, 5729 .arg2_type = ARG_ANYTHING, 5730 .arg3_type = ARG_PTR_TO_MEM, 5731 .arg4_type = ARG_CONST_SIZE 5732 }; 5733 5734 static void bpf_update_srh_state(struct sk_buff *skb) 5735 { 5736 struct seg6_bpf_srh_state *srh_state = 5737 this_cpu_ptr(&seg6_bpf_srh_states); 5738 int srhoff = 0; 5739 5740 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) { 5741 srh_state->srh = NULL; 5742 } else { 5743 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5744 srh_state->hdrlen = srh_state->srh->hdrlen << 3; 5745 srh_state->valid = true; 5746 } 5747 } 5748 5749 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb, 5750 u32, action, void *, param, u32, param_len) 5751 { 5752 struct seg6_bpf_srh_state *srh_state = 5753 this_cpu_ptr(&seg6_bpf_srh_states); 5754 int hdroff = 0; 5755 int err; 5756 5757 switch (action) { 5758 case SEG6_LOCAL_ACTION_END_X: 5759 if (!seg6_bpf_has_valid_srh(skb)) 5760 return -EBADMSG; 5761 if (param_len != sizeof(struct in6_addr)) 5762 return -EINVAL; 5763 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0); 5764 case SEG6_LOCAL_ACTION_END_T: 5765 if (!seg6_bpf_has_valid_srh(skb)) 5766 return -EBADMSG; 5767 if (param_len != sizeof(int)) 5768 return -EINVAL; 5769 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 5770 case SEG6_LOCAL_ACTION_END_DT6: 5771 if (!seg6_bpf_has_valid_srh(skb)) 5772 return -EBADMSG; 5773 if (param_len != sizeof(int)) 5774 return -EINVAL; 5775 5776 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0) 5777 return -EBADMSG; 5778 if (!pskb_pull(skb, hdroff)) 5779 return -EBADMSG; 5780 5781 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff); 5782 skb_reset_network_header(skb); 5783 skb_reset_transport_header(skb); 5784 skb->encapsulation = 0; 5785 5786 bpf_compute_data_pointers(skb); 5787 bpf_update_srh_state(skb); 5788 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 5789 case SEG6_LOCAL_ACTION_END_B6: 5790 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 5791 return -EBADMSG; 5792 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE, 5793 param, param_len); 5794 if (!err) 5795 bpf_update_srh_state(skb); 5796 5797 return err; 5798 case SEG6_LOCAL_ACTION_END_B6_ENCAP: 5799 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 5800 return -EBADMSG; 5801 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6, 5802 param, param_len); 5803 if (!err) 5804 bpf_update_srh_state(skb); 5805 5806 return err; 5807 default: 5808 return -EINVAL; 5809 } 5810 } 5811 5812 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = { 5813 .func = bpf_lwt_seg6_action, 5814 .gpl_only = false, 5815 .ret_type = RET_INTEGER, 5816 .arg1_type = ARG_PTR_TO_CTX, 5817 .arg2_type = ARG_ANYTHING, 5818 .arg3_type = ARG_PTR_TO_MEM, 5819 .arg4_type = ARG_CONST_SIZE 5820 }; 5821 5822 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset, 5823 s32, len) 5824 { 5825 struct seg6_bpf_srh_state *srh_state = 5826 this_cpu_ptr(&seg6_bpf_srh_states); 5827 struct ipv6_sr_hdr *srh = srh_state->srh; 5828 void *srh_end, *srh_tlvs, *ptr; 5829 struct ipv6hdr *hdr; 5830 int srhoff = 0; 5831 int ret; 5832 5833 if (unlikely(srh == NULL)) 5834 return -EINVAL; 5835 5836 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) + 5837 ((srh->first_segment + 1) << 4)); 5838 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) + 5839 srh_state->hdrlen); 5840 ptr = skb->data + offset; 5841 5842 if (unlikely(ptr < srh_tlvs || ptr > srh_end)) 5843 return -EFAULT; 5844 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end)) 5845 return -EFAULT; 5846 5847 if (len > 0) { 5848 ret = skb_cow_head(skb, len); 5849 if (unlikely(ret < 0)) 5850 return ret; 5851 5852 ret = bpf_skb_net_hdr_push(skb, offset, len); 5853 } else { 5854 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len); 5855 } 5856 5857 bpf_compute_data_pointers(skb); 5858 if (unlikely(ret < 0)) 5859 return ret; 5860 5861 hdr = (struct ipv6hdr *)skb->data; 5862 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 5863 5864 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 5865 return -EINVAL; 5866 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5867 srh_state->hdrlen += len; 5868 srh_state->valid = false; 5869 return 0; 5870 } 5871 5872 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = { 5873 .func = bpf_lwt_seg6_adjust_srh, 5874 .gpl_only = false, 5875 .ret_type = RET_INTEGER, 5876 .arg1_type = ARG_PTR_TO_CTX, 5877 .arg2_type = ARG_ANYTHING, 5878 .arg3_type = ARG_ANYTHING, 5879 }; 5880 #endif /* CONFIG_IPV6_SEG6_BPF */ 5881 5882 #ifdef CONFIG_INET 5883 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple, 5884 int dif, int sdif, u8 family, u8 proto) 5885 { 5886 bool refcounted = false; 5887 struct sock *sk = NULL; 5888 5889 if (family == AF_INET) { 5890 __be32 src4 = tuple->ipv4.saddr; 5891 __be32 dst4 = tuple->ipv4.daddr; 5892 5893 if (proto == IPPROTO_TCP) 5894 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0, 5895 src4, tuple->ipv4.sport, 5896 dst4, tuple->ipv4.dport, 5897 dif, sdif, &refcounted); 5898 else 5899 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport, 5900 dst4, tuple->ipv4.dport, 5901 dif, sdif, &udp_table, NULL); 5902 #if IS_ENABLED(CONFIG_IPV6) 5903 } else { 5904 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr; 5905 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr; 5906 5907 if (proto == IPPROTO_TCP) 5908 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0, 5909 src6, tuple->ipv6.sport, 5910 dst6, ntohs(tuple->ipv6.dport), 5911 dif, sdif, &refcounted); 5912 else if (likely(ipv6_bpf_stub)) 5913 sk = ipv6_bpf_stub->udp6_lib_lookup(net, 5914 src6, tuple->ipv6.sport, 5915 dst6, tuple->ipv6.dport, 5916 dif, sdif, 5917 &udp_table, NULL); 5918 #endif 5919 } 5920 5921 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) { 5922 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 5923 sk = NULL; 5924 } 5925 return sk; 5926 } 5927 5928 /* bpf_skc_lookup performs the core lookup for different types of sockets, 5929 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE. 5930 * Returns the socket as an 'unsigned long' to simplify the casting in the 5931 * callers to satisfy BPF_CALL declarations. 5932 */ 5933 static struct sock * 5934 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5935 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 5936 u64 flags) 5937 { 5938 struct sock *sk = NULL; 5939 u8 family = AF_UNSPEC; 5940 struct net *net; 5941 int sdif; 5942 5943 if (len == sizeof(tuple->ipv4)) 5944 family = AF_INET; 5945 else if (len == sizeof(tuple->ipv6)) 5946 family = AF_INET6; 5947 else 5948 return NULL; 5949 5950 if (unlikely(family == AF_UNSPEC || flags || 5951 !((s32)netns_id < 0 || netns_id <= S32_MAX))) 5952 goto out; 5953 5954 if (family == AF_INET) 5955 sdif = inet_sdif(skb); 5956 else 5957 sdif = inet6_sdif(skb); 5958 5959 if ((s32)netns_id < 0) { 5960 net = caller_net; 5961 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 5962 } else { 5963 net = get_net_ns_by_id(caller_net, netns_id); 5964 if (unlikely(!net)) 5965 goto out; 5966 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 5967 put_net(net); 5968 } 5969 5970 out: 5971 return sk; 5972 } 5973 5974 static struct sock * 5975 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5976 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 5977 u64 flags) 5978 { 5979 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net, 5980 ifindex, proto, netns_id, flags); 5981 5982 if (sk) { 5983 sk = sk_to_full_sk(sk); 5984 if (!sk_fullsock(sk)) { 5985 sock_gen_put(sk); 5986 return NULL; 5987 } 5988 } 5989 5990 return sk; 5991 } 5992 5993 static struct sock * 5994 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5995 u8 proto, u64 netns_id, u64 flags) 5996 { 5997 struct net *caller_net; 5998 int ifindex; 5999 6000 if (skb->dev) { 6001 caller_net = dev_net(skb->dev); 6002 ifindex = skb->dev->ifindex; 6003 } else { 6004 caller_net = sock_net(skb->sk); 6005 ifindex = 0; 6006 } 6007 6008 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto, 6009 netns_id, flags); 6010 } 6011 6012 static struct sock * 6013 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6014 u8 proto, u64 netns_id, u64 flags) 6015 { 6016 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id, 6017 flags); 6018 6019 if (sk) { 6020 sk = sk_to_full_sk(sk); 6021 if (!sk_fullsock(sk)) { 6022 sock_gen_put(sk); 6023 return NULL; 6024 } 6025 } 6026 6027 return sk; 6028 } 6029 6030 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb, 6031 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6032 { 6033 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP, 6034 netns_id, flags); 6035 } 6036 6037 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = { 6038 .func = bpf_skc_lookup_tcp, 6039 .gpl_only = false, 6040 .pkt_access = true, 6041 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6042 .arg1_type = ARG_PTR_TO_CTX, 6043 .arg2_type = ARG_PTR_TO_MEM, 6044 .arg3_type = ARG_CONST_SIZE, 6045 .arg4_type = ARG_ANYTHING, 6046 .arg5_type = ARG_ANYTHING, 6047 }; 6048 6049 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb, 6050 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6051 { 6052 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, 6053 netns_id, flags); 6054 } 6055 6056 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = { 6057 .func = bpf_sk_lookup_tcp, 6058 .gpl_only = false, 6059 .pkt_access = true, 6060 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6061 .arg1_type = ARG_PTR_TO_CTX, 6062 .arg2_type = ARG_PTR_TO_MEM, 6063 .arg3_type = ARG_CONST_SIZE, 6064 .arg4_type = ARG_ANYTHING, 6065 .arg5_type = ARG_ANYTHING, 6066 }; 6067 6068 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb, 6069 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6070 { 6071 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, 6072 netns_id, flags); 6073 } 6074 6075 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = { 6076 .func = bpf_sk_lookup_udp, 6077 .gpl_only = false, 6078 .pkt_access = true, 6079 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6080 .arg1_type = ARG_PTR_TO_CTX, 6081 .arg2_type = ARG_PTR_TO_MEM, 6082 .arg3_type = ARG_CONST_SIZE, 6083 .arg4_type = ARG_ANYTHING, 6084 .arg5_type = ARG_ANYTHING, 6085 }; 6086 6087 BPF_CALL_1(bpf_sk_release, struct sock *, sk) 6088 { 6089 if (sk && sk_is_refcounted(sk)) 6090 sock_gen_put(sk); 6091 return 0; 6092 } 6093 6094 static const struct bpf_func_proto bpf_sk_release_proto = { 6095 .func = bpf_sk_release, 6096 .gpl_only = false, 6097 .ret_type = RET_INTEGER, 6098 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6099 }; 6100 6101 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx, 6102 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6103 { 6104 struct net *caller_net = dev_net(ctx->rxq->dev); 6105 int ifindex = ctx->rxq->dev->ifindex; 6106 6107 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6108 ifindex, IPPROTO_UDP, netns_id, 6109 flags); 6110 } 6111 6112 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = { 6113 .func = bpf_xdp_sk_lookup_udp, 6114 .gpl_only = false, 6115 .pkt_access = true, 6116 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6117 .arg1_type = ARG_PTR_TO_CTX, 6118 .arg2_type = ARG_PTR_TO_MEM, 6119 .arg3_type = ARG_CONST_SIZE, 6120 .arg4_type = ARG_ANYTHING, 6121 .arg5_type = ARG_ANYTHING, 6122 }; 6123 6124 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx, 6125 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6126 { 6127 struct net *caller_net = dev_net(ctx->rxq->dev); 6128 int ifindex = ctx->rxq->dev->ifindex; 6129 6130 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net, 6131 ifindex, IPPROTO_TCP, netns_id, 6132 flags); 6133 } 6134 6135 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = { 6136 .func = bpf_xdp_skc_lookup_tcp, 6137 .gpl_only = false, 6138 .pkt_access = true, 6139 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6140 .arg1_type = ARG_PTR_TO_CTX, 6141 .arg2_type = ARG_PTR_TO_MEM, 6142 .arg3_type = ARG_CONST_SIZE, 6143 .arg4_type = ARG_ANYTHING, 6144 .arg5_type = ARG_ANYTHING, 6145 }; 6146 6147 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx, 6148 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6149 { 6150 struct net *caller_net = dev_net(ctx->rxq->dev); 6151 int ifindex = ctx->rxq->dev->ifindex; 6152 6153 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6154 ifindex, IPPROTO_TCP, netns_id, 6155 flags); 6156 } 6157 6158 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = { 6159 .func = bpf_xdp_sk_lookup_tcp, 6160 .gpl_only = false, 6161 .pkt_access = true, 6162 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6163 .arg1_type = ARG_PTR_TO_CTX, 6164 .arg2_type = ARG_PTR_TO_MEM, 6165 .arg3_type = ARG_CONST_SIZE, 6166 .arg4_type = ARG_ANYTHING, 6167 .arg5_type = ARG_ANYTHING, 6168 }; 6169 6170 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6171 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6172 { 6173 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, 6174 sock_net(ctx->sk), 0, 6175 IPPROTO_TCP, netns_id, flags); 6176 } 6177 6178 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = { 6179 .func = bpf_sock_addr_skc_lookup_tcp, 6180 .gpl_only = false, 6181 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6182 .arg1_type = ARG_PTR_TO_CTX, 6183 .arg2_type = ARG_PTR_TO_MEM, 6184 .arg3_type = ARG_CONST_SIZE, 6185 .arg4_type = ARG_ANYTHING, 6186 .arg5_type = ARG_ANYTHING, 6187 }; 6188 6189 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6190 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6191 { 6192 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6193 sock_net(ctx->sk), 0, IPPROTO_TCP, 6194 netns_id, flags); 6195 } 6196 6197 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = { 6198 .func = bpf_sock_addr_sk_lookup_tcp, 6199 .gpl_only = false, 6200 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6201 .arg1_type = ARG_PTR_TO_CTX, 6202 .arg2_type = ARG_PTR_TO_MEM, 6203 .arg3_type = ARG_CONST_SIZE, 6204 .arg4_type = ARG_ANYTHING, 6205 .arg5_type = ARG_ANYTHING, 6206 }; 6207 6208 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx, 6209 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6210 { 6211 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6212 sock_net(ctx->sk), 0, IPPROTO_UDP, 6213 netns_id, flags); 6214 } 6215 6216 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = { 6217 .func = bpf_sock_addr_sk_lookup_udp, 6218 .gpl_only = false, 6219 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6220 .arg1_type = ARG_PTR_TO_CTX, 6221 .arg2_type = ARG_PTR_TO_MEM, 6222 .arg3_type = ARG_CONST_SIZE, 6223 .arg4_type = ARG_ANYTHING, 6224 .arg5_type = ARG_ANYTHING, 6225 }; 6226 6227 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6228 struct bpf_insn_access_aux *info) 6229 { 6230 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, 6231 icsk_retransmits)) 6232 return false; 6233 6234 if (off % size != 0) 6235 return false; 6236 6237 switch (off) { 6238 case offsetof(struct bpf_tcp_sock, bytes_received): 6239 case offsetof(struct bpf_tcp_sock, bytes_acked): 6240 return size == sizeof(__u64); 6241 default: 6242 return size == sizeof(__u32); 6243 } 6244 } 6245 6246 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 6247 const struct bpf_insn *si, 6248 struct bpf_insn *insn_buf, 6249 struct bpf_prog *prog, u32 *target_size) 6250 { 6251 struct bpf_insn *insn = insn_buf; 6252 6253 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \ 6254 do { \ 6255 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \ 6256 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6257 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\ 6258 si->dst_reg, si->src_reg, \ 6259 offsetof(struct tcp_sock, FIELD)); \ 6260 } while (0) 6261 6262 #define BPF_INET_SOCK_GET_COMMON(FIELD) \ 6263 do { \ 6264 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \ 6265 FIELD) > \ 6266 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6267 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 6268 struct inet_connection_sock, \ 6269 FIELD), \ 6270 si->dst_reg, si->src_reg, \ 6271 offsetof( \ 6272 struct inet_connection_sock, \ 6273 FIELD)); \ 6274 } while (0) 6275 6276 if (insn > insn_buf) 6277 return insn - insn_buf; 6278 6279 switch (si->off) { 6280 case offsetof(struct bpf_tcp_sock, rtt_min): 6281 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 6282 sizeof(struct minmax)); 6283 BUILD_BUG_ON(sizeof(struct minmax) < 6284 sizeof(struct minmax_sample)); 6285 6286 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 6287 offsetof(struct tcp_sock, rtt_min) + 6288 offsetof(struct minmax_sample, v)); 6289 break; 6290 case offsetof(struct bpf_tcp_sock, snd_cwnd): 6291 BPF_TCP_SOCK_GET_COMMON(snd_cwnd); 6292 break; 6293 case offsetof(struct bpf_tcp_sock, srtt_us): 6294 BPF_TCP_SOCK_GET_COMMON(srtt_us); 6295 break; 6296 case offsetof(struct bpf_tcp_sock, snd_ssthresh): 6297 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh); 6298 break; 6299 case offsetof(struct bpf_tcp_sock, rcv_nxt): 6300 BPF_TCP_SOCK_GET_COMMON(rcv_nxt); 6301 break; 6302 case offsetof(struct bpf_tcp_sock, snd_nxt): 6303 BPF_TCP_SOCK_GET_COMMON(snd_nxt); 6304 break; 6305 case offsetof(struct bpf_tcp_sock, snd_una): 6306 BPF_TCP_SOCK_GET_COMMON(snd_una); 6307 break; 6308 case offsetof(struct bpf_tcp_sock, mss_cache): 6309 BPF_TCP_SOCK_GET_COMMON(mss_cache); 6310 break; 6311 case offsetof(struct bpf_tcp_sock, ecn_flags): 6312 BPF_TCP_SOCK_GET_COMMON(ecn_flags); 6313 break; 6314 case offsetof(struct bpf_tcp_sock, rate_delivered): 6315 BPF_TCP_SOCK_GET_COMMON(rate_delivered); 6316 break; 6317 case offsetof(struct bpf_tcp_sock, rate_interval_us): 6318 BPF_TCP_SOCK_GET_COMMON(rate_interval_us); 6319 break; 6320 case offsetof(struct bpf_tcp_sock, packets_out): 6321 BPF_TCP_SOCK_GET_COMMON(packets_out); 6322 break; 6323 case offsetof(struct bpf_tcp_sock, retrans_out): 6324 BPF_TCP_SOCK_GET_COMMON(retrans_out); 6325 break; 6326 case offsetof(struct bpf_tcp_sock, total_retrans): 6327 BPF_TCP_SOCK_GET_COMMON(total_retrans); 6328 break; 6329 case offsetof(struct bpf_tcp_sock, segs_in): 6330 BPF_TCP_SOCK_GET_COMMON(segs_in); 6331 break; 6332 case offsetof(struct bpf_tcp_sock, data_segs_in): 6333 BPF_TCP_SOCK_GET_COMMON(data_segs_in); 6334 break; 6335 case offsetof(struct bpf_tcp_sock, segs_out): 6336 BPF_TCP_SOCK_GET_COMMON(segs_out); 6337 break; 6338 case offsetof(struct bpf_tcp_sock, data_segs_out): 6339 BPF_TCP_SOCK_GET_COMMON(data_segs_out); 6340 break; 6341 case offsetof(struct bpf_tcp_sock, lost_out): 6342 BPF_TCP_SOCK_GET_COMMON(lost_out); 6343 break; 6344 case offsetof(struct bpf_tcp_sock, sacked_out): 6345 BPF_TCP_SOCK_GET_COMMON(sacked_out); 6346 break; 6347 case offsetof(struct bpf_tcp_sock, bytes_received): 6348 BPF_TCP_SOCK_GET_COMMON(bytes_received); 6349 break; 6350 case offsetof(struct bpf_tcp_sock, bytes_acked): 6351 BPF_TCP_SOCK_GET_COMMON(bytes_acked); 6352 break; 6353 case offsetof(struct bpf_tcp_sock, dsack_dups): 6354 BPF_TCP_SOCK_GET_COMMON(dsack_dups); 6355 break; 6356 case offsetof(struct bpf_tcp_sock, delivered): 6357 BPF_TCP_SOCK_GET_COMMON(delivered); 6358 break; 6359 case offsetof(struct bpf_tcp_sock, delivered_ce): 6360 BPF_TCP_SOCK_GET_COMMON(delivered_ce); 6361 break; 6362 case offsetof(struct bpf_tcp_sock, icsk_retransmits): 6363 BPF_INET_SOCK_GET_COMMON(icsk_retransmits); 6364 break; 6365 } 6366 6367 return insn - insn_buf; 6368 } 6369 6370 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk) 6371 { 6372 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 6373 return (unsigned long)sk; 6374 6375 return (unsigned long)NULL; 6376 } 6377 6378 const struct bpf_func_proto bpf_tcp_sock_proto = { 6379 .func = bpf_tcp_sock, 6380 .gpl_only = false, 6381 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL, 6382 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 6383 }; 6384 6385 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk) 6386 { 6387 sk = sk_to_full_sk(sk); 6388 6389 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE)) 6390 return (unsigned long)sk; 6391 6392 return (unsigned long)NULL; 6393 } 6394 6395 static const struct bpf_func_proto bpf_get_listener_sock_proto = { 6396 .func = bpf_get_listener_sock, 6397 .gpl_only = false, 6398 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6399 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 6400 }; 6401 6402 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb) 6403 { 6404 unsigned int iphdr_len; 6405 6406 switch (skb_protocol(skb, true)) { 6407 case cpu_to_be16(ETH_P_IP): 6408 iphdr_len = sizeof(struct iphdr); 6409 break; 6410 case cpu_to_be16(ETH_P_IPV6): 6411 iphdr_len = sizeof(struct ipv6hdr); 6412 break; 6413 default: 6414 return 0; 6415 } 6416 6417 if (skb_headlen(skb) < iphdr_len) 6418 return 0; 6419 6420 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len)) 6421 return 0; 6422 6423 return INET_ECN_set_ce(skb); 6424 } 6425 6426 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6427 struct bpf_insn_access_aux *info) 6428 { 6429 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id)) 6430 return false; 6431 6432 if (off % size != 0) 6433 return false; 6434 6435 switch (off) { 6436 default: 6437 return size == sizeof(__u32); 6438 } 6439 } 6440 6441 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 6442 const struct bpf_insn *si, 6443 struct bpf_insn *insn_buf, 6444 struct bpf_prog *prog, u32 *target_size) 6445 { 6446 struct bpf_insn *insn = insn_buf; 6447 6448 #define BPF_XDP_SOCK_GET(FIELD) \ 6449 do { \ 6450 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \ 6451 sizeof_field(struct bpf_xdp_sock, FIELD)); \ 6452 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\ 6453 si->dst_reg, si->src_reg, \ 6454 offsetof(struct xdp_sock, FIELD)); \ 6455 } while (0) 6456 6457 switch (si->off) { 6458 case offsetof(struct bpf_xdp_sock, queue_id): 6459 BPF_XDP_SOCK_GET(queue_id); 6460 break; 6461 } 6462 6463 return insn - insn_buf; 6464 } 6465 6466 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = { 6467 .func = bpf_skb_ecn_set_ce, 6468 .gpl_only = false, 6469 .ret_type = RET_INTEGER, 6470 .arg1_type = ARG_PTR_TO_CTX, 6471 }; 6472 6473 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 6474 struct tcphdr *, th, u32, th_len) 6475 { 6476 #ifdef CONFIG_SYN_COOKIES 6477 u32 cookie; 6478 int ret; 6479 6480 if (unlikely(!sk || th_len < sizeof(*th))) 6481 return -EINVAL; 6482 6483 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */ 6484 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 6485 return -EINVAL; 6486 6487 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies) 6488 return -EINVAL; 6489 6490 if (!th->ack || th->rst || th->syn) 6491 return -ENOENT; 6492 6493 if (tcp_synq_no_recent_overflow(sk)) 6494 return -ENOENT; 6495 6496 cookie = ntohl(th->ack_seq) - 1; 6497 6498 switch (sk->sk_family) { 6499 case AF_INET: 6500 if (unlikely(iph_len < sizeof(struct iphdr))) 6501 return -EINVAL; 6502 6503 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie); 6504 break; 6505 6506 #if IS_BUILTIN(CONFIG_IPV6) 6507 case AF_INET6: 6508 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 6509 return -EINVAL; 6510 6511 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie); 6512 break; 6513 #endif /* CONFIG_IPV6 */ 6514 6515 default: 6516 return -EPROTONOSUPPORT; 6517 } 6518 6519 if (ret > 0) 6520 return 0; 6521 6522 return -ENOENT; 6523 #else 6524 return -ENOTSUPP; 6525 #endif 6526 } 6527 6528 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = { 6529 .func = bpf_tcp_check_syncookie, 6530 .gpl_only = true, 6531 .pkt_access = true, 6532 .ret_type = RET_INTEGER, 6533 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6534 .arg2_type = ARG_PTR_TO_MEM, 6535 .arg3_type = ARG_CONST_SIZE, 6536 .arg4_type = ARG_PTR_TO_MEM, 6537 .arg5_type = ARG_CONST_SIZE, 6538 }; 6539 6540 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 6541 struct tcphdr *, th, u32, th_len) 6542 { 6543 #ifdef CONFIG_SYN_COOKIES 6544 u32 cookie; 6545 u16 mss; 6546 6547 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4)) 6548 return -EINVAL; 6549 6550 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 6551 return -EINVAL; 6552 6553 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies) 6554 return -ENOENT; 6555 6556 if (!th->syn || th->ack || th->fin || th->rst) 6557 return -EINVAL; 6558 6559 if (unlikely(iph_len < sizeof(struct iphdr))) 6560 return -EINVAL; 6561 6562 /* Both struct iphdr and struct ipv6hdr have the version field at the 6563 * same offset so we can cast to the shorter header (struct iphdr). 6564 */ 6565 switch (((struct iphdr *)iph)->version) { 6566 case 4: 6567 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only) 6568 return -EINVAL; 6569 6570 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie); 6571 break; 6572 6573 #if IS_BUILTIN(CONFIG_IPV6) 6574 case 6: 6575 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 6576 return -EINVAL; 6577 6578 if (sk->sk_family != AF_INET6) 6579 return -EINVAL; 6580 6581 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie); 6582 break; 6583 #endif /* CONFIG_IPV6 */ 6584 6585 default: 6586 return -EPROTONOSUPPORT; 6587 } 6588 if (mss == 0) 6589 return -ENOENT; 6590 6591 return cookie | ((u64)mss << 32); 6592 #else 6593 return -EOPNOTSUPP; 6594 #endif /* CONFIG_SYN_COOKIES */ 6595 } 6596 6597 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = { 6598 .func = bpf_tcp_gen_syncookie, 6599 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */ 6600 .pkt_access = true, 6601 .ret_type = RET_INTEGER, 6602 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6603 .arg2_type = ARG_PTR_TO_MEM, 6604 .arg3_type = ARG_CONST_SIZE, 6605 .arg4_type = ARG_PTR_TO_MEM, 6606 .arg5_type = ARG_CONST_SIZE, 6607 }; 6608 6609 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags) 6610 { 6611 if (!sk || flags != 0) 6612 return -EINVAL; 6613 if (!skb_at_tc_ingress(skb)) 6614 return -EOPNOTSUPP; 6615 if (unlikely(dev_net(skb->dev) != sock_net(sk))) 6616 return -ENETUNREACH; 6617 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport)) 6618 return -ESOCKTNOSUPPORT; 6619 if (sk_is_refcounted(sk) && 6620 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) 6621 return -ENOENT; 6622 6623 skb_orphan(skb); 6624 skb->sk = sk; 6625 skb->destructor = sock_pfree; 6626 6627 return 0; 6628 } 6629 6630 static const struct bpf_func_proto bpf_sk_assign_proto = { 6631 .func = bpf_sk_assign, 6632 .gpl_only = false, 6633 .ret_type = RET_INTEGER, 6634 .arg1_type = ARG_PTR_TO_CTX, 6635 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6636 .arg3_type = ARG_ANYTHING, 6637 }; 6638 6639 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend, 6640 u8 search_kind, const u8 *magic, 6641 u8 magic_len, bool *eol) 6642 { 6643 u8 kind, kind_len; 6644 6645 *eol = false; 6646 6647 while (op < opend) { 6648 kind = op[0]; 6649 6650 if (kind == TCPOPT_EOL) { 6651 *eol = true; 6652 return ERR_PTR(-ENOMSG); 6653 } else if (kind == TCPOPT_NOP) { 6654 op++; 6655 continue; 6656 } 6657 6658 if (opend - op < 2 || opend - op < op[1] || op[1] < 2) 6659 /* Something is wrong in the received header. 6660 * Follow the TCP stack's tcp_parse_options() 6661 * and just bail here. 6662 */ 6663 return ERR_PTR(-EFAULT); 6664 6665 kind_len = op[1]; 6666 if (search_kind == kind) { 6667 if (!magic_len) 6668 return op; 6669 6670 if (magic_len > kind_len - 2) 6671 return ERR_PTR(-ENOMSG); 6672 6673 if (!memcmp(&op[2], magic, magic_len)) 6674 return op; 6675 } 6676 6677 op += kind_len; 6678 } 6679 6680 return ERR_PTR(-ENOMSG); 6681 } 6682 6683 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 6684 void *, search_res, u32, len, u64, flags) 6685 { 6686 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN; 6687 const u8 *op, *opend, *magic, *search = search_res; 6688 u8 search_kind, search_len, copy_len, magic_len; 6689 int ret; 6690 6691 /* 2 byte is the minimal option len except TCPOPT_NOP and 6692 * TCPOPT_EOL which are useless for the bpf prog to learn 6693 * and this helper disallow loading them also. 6694 */ 6695 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN) 6696 return -EINVAL; 6697 6698 search_kind = search[0]; 6699 search_len = search[1]; 6700 6701 if (search_len > len || search_kind == TCPOPT_NOP || 6702 search_kind == TCPOPT_EOL) 6703 return -EINVAL; 6704 6705 if (search_kind == TCPOPT_EXP || search_kind == 253) { 6706 /* 16 or 32 bit magic. +2 for kind and kind length */ 6707 if (search_len != 4 && search_len != 6) 6708 return -EINVAL; 6709 magic = &search[2]; 6710 magic_len = search_len - 2; 6711 } else { 6712 if (search_len) 6713 return -EINVAL; 6714 magic = NULL; 6715 magic_len = 0; 6716 } 6717 6718 if (load_syn) { 6719 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op); 6720 if (ret < 0) 6721 return ret; 6722 6723 opend = op + ret; 6724 op += sizeof(struct tcphdr); 6725 } else { 6726 if (!bpf_sock->skb || 6727 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB) 6728 /* This bpf_sock->op cannot call this helper */ 6729 return -EPERM; 6730 6731 opend = bpf_sock->skb_data_end; 6732 op = bpf_sock->skb->data + sizeof(struct tcphdr); 6733 } 6734 6735 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len, 6736 &eol); 6737 if (IS_ERR(op)) 6738 return PTR_ERR(op); 6739 6740 copy_len = op[1]; 6741 ret = copy_len; 6742 if (copy_len > len) { 6743 ret = -ENOSPC; 6744 copy_len = len; 6745 } 6746 6747 memcpy(search_res, op, copy_len); 6748 return ret; 6749 } 6750 6751 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = { 6752 .func = bpf_sock_ops_load_hdr_opt, 6753 .gpl_only = false, 6754 .ret_type = RET_INTEGER, 6755 .arg1_type = ARG_PTR_TO_CTX, 6756 .arg2_type = ARG_PTR_TO_MEM, 6757 .arg3_type = ARG_CONST_SIZE, 6758 .arg4_type = ARG_ANYTHING, 6759 }; 6760 6761 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 6762 const void *, from, u32, len, u64, flags) 6763 { 6764 u8 new_kind, new_kind_len, magic_len = 0, *opend; 6765 const u8 *op, *new_op, *magic = NULL; 6766 struct sk_buff *skb; 6767 bool eol; 6768 6769 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB) 6770 return -EPERM; 6771 6772 if (len < 2 || flags) 6773 return -EINVAL; 6774 6775 new_op = from; 6776 new_kind = new_op[0]; 6777 new_kind_len = new_op[1]; 6778 6779 if (new_kind_len > len || new_kind == TCPOPT_NOP || 6780 new_kind == TCPOPT_EOL) 6781 return -EINVAL; 6782 6783 if (new_kind_len > bpf_sock->remaining_opt_len) 6784 return -ENOSPC; 6785 6786 /* 253 is another experimental kind */ 6787 if (new_kind == TCPOPT_EXP || new_kind == 253) { 6788 if (new_kind_len < 4) 6789 return -EINVAL; 6790 /* Match for the 2 byte magic also. 6791 * RFC 6994: the magic could be 2 or 4 bytes. 6792 * Hence, matching by 2 byte only is on the 6793 * conservative side but it is the right 6794 * thing to do for the 'search-for-duplication' 6795 * purpose. 6796 */ 6797 magic = &new_op[2]; 6798 magic_len = 2; 6799 } 6800 6801 /* Check for duplication */ 6802 skb = bpf_sock->skb; 6803 op = skb->data + sizeof(struct tcphdr); 6804 opend = bpf_sock->skb_data_end; 6805 6806 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len, 6807 &eol); 6808 if (!IS_ERR(op)) 6809 return -EEXIST; 6810 6811 if (PTR_ERR(op) != -ENOMSG) 6812 return PTR_ERR(op); 6813 6814 if (eol) 6815 /* The option has been ended. Treat it as no more 6816 * header option can be written. 6817 */ 6818 return -ENOSPC; 6819 6820 /* No duplication found. Store the header option. */ 6821 memcpy(opend, from, new_kind_len); 6822 6823 bpf_sock->remaining_opt_len -= new_kind_len; 6824 bpf_sock->skb_data_end += new_kind_len; 6825 6826 return 0; 6827 } 6828 6829 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = { 6830 .func = bpf_sock_ops_store_hdr_opt, 6831 .gpl_only = false, 6832 .ret_type = RET_INTEGER, 6833 .arg1_type = ARG_PTR_TO_CTX, 6834 .arg2_type = ARG_PTR_TO_MEM, 6835 .arg3_type = ARG_CONST_SIZE, 6836 .arg4_type = ARG_ANYTHING, 6837 }; 6838 6839 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 6840 u32, len, u64, flags) 6841 { 6842 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB) 6843 return -EPERM; 6844 6845 if (flags || len < 2) 6846 return -EINVAL; 6847 6848 if (len > bpf_sock->remaining_opt_len) 6849 return -ENOSPC; 6850 6851 bpf_sock->remaining_opt_len -= len; 6852 6853 return 0; 6854 } 6855 6856 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = { 6857 .func = bpf_sock_ops_reserve_hdr_opt, 6858 .gpl_only = false, 6859 .ret_type = RET_INTEGER, 6860 .arg1_type = ARG_PTR_TO_CTX, 6861 .arg2_type = ARG_ANYTHING, 6862 .arg3_type = ARG_ANYTHING, 6863 }; 6864 6865 #endif /* CONFIG_INET */ 6866 6867 bool bpf_helper_changes_pkt_data(void *func) 6868 { 6869 if (func == bpf_skb_vlan_push || 6870 func == bpf_skb_vlan_pop || 6871 func == bpf_skb_store_bytes || 6872 func == bpf_skb_change_proto || 6873 func == bpf_skb_change_head || 6874 func == sk_skb_change_head || 6875 func == bpf_skb_change_tail || 6876 func == sk_skb_change_tail || 6877 func == bpf_skb_adjust_room || 6878 func == sk_skb_adjust_room || 6879 func == bpf_skb_pull_data || 6880 func == sk_skb_pull_data || 6881 func == bpf_clone_redirect || 6882 func == bpf_l3_csum_replace || 6883 func == bpf_l4_csum_replace || 6884 func == bpf_xdp_adjust_head || 6885 func == bpf_xdp_adjust_meta || 6886 func == bpf_msg_pull_data || 6887 func == bpf_msg_push_data || 6888 func == bpf_msg_pop_data || 6889 func == bpf_xdp_adjust_tail || 6890 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6891 func == bpf_lwt_seg6_store_bytes || 6892 func == bpf_lwt_seg6_adjust_srh || 6893 func == bpf_lwt_seg6_action || 6894 #endif 6895 #ifdef CONFIG_INET 6896 func == bpf_sock_ops_store_hdr_opt || 6897 #endif 6898 func == bpf_lwt_in_push_encap || 6899 func == bpf_lwt_xmit_push_encap) 6900 return true; 6901 6902 return false; 6903 } 6904 6905 const struct bpf_func_proto bpf_event_output_data_proto __weak; 6906 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak; 6907 6908 static const struct bpf_func_proto * 6909 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6910 { 6911 switch (func_id) { 6912 /* inet and inet6 sockets are created in a process 6913 * context so there is always a valid uid/gid 6914 */ 6915 case BPF_FUNC_get_current_uid_gid: 6916 return &bpf_get_current_uid_gid_proto; 6917 case BPF_FUNC_get_local_storage: 6918 return &bpf_get_local_storage_proto; 6919 case BPF_FUNC_get_socket_cookie: 6920 return &bpf_get_socket_cookie_sock_proto; 6921 case BPF_FUNC_get_netns_cookie: 6922 return &bpf_get_netns_cookie_sock_proto; 6923 case BPF_FUNC_perf_event_output: 6924 return &bpf_event_output_data_proto; 6925 case BPF_FUNC_get_current_pid_tgid: 6926 return &bpf_get_current_pid_tgid_proto; 6927 case BPF_FUNC_get_current_comm: 6928 return &bpf_get_current_comm_proto; 6929 #ifdef CONFIG_CGROUPS 6930 case BPF_FUNC_get_current_cgroup_id: 6931 return &bpf_get_current_cgroup_id_proto; 6932 case BPF_FUNC_get_current_ancestor_cgroup_id: 6933 return &bpf_get_current_ancestor_cgroup_id_proto; 6934 #endif 6935 #ifdef CONFIG_CGROUP_NET_CLASSID 6936 case BPF_FUNC_get_cgroup_classid: 6937 return &bpf_get_cgroup_classid_curr_proto; 6938 #endif 6939 case BPF_FUNC_sk_storage_get: 6940 return &bpf_sk_storage_get_cg_sock_proto; 6941 default: 6942 return bpf_base_func_proto(func_id); 6943 } 6944 } 6945 6946 static const struct bpf_func_proto * 6947 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6948 { 6949 switch (func_id) { 6950 /* inet and inet6 sockets are created in a process 6951 * context so there is always a valid uid/gid 6952 */ 6953 case BPF_FUNC_get_current_uid_gid: 6954 return &bpf_get_current_uid_gid_proto; 6955 case BPF_FUNC_bind: 6956 switch (prog->expected_attach_type) { 6957 case BPF_CGROUP_INET4_CONNECT: 6958 case BPF_CGROUP_INET6_CONNECT: 6959 return &bpf_bind_proto; 6960 default: 6961 return NULL; 6962 } 6963 case BPF_FUNC_get_socket_cookie: 6964 return &bpf_get_socket_cookie_sock_addr_proto; 6965 case BPF_FUNC_get_netns_cookie: 6966 return &bpf_get_netns_cookie_sock_addr_proto; 6967 case BPF_FUNC_get_local_storage: 6968 return &bpf_get_local_storage_proto; 6969 case BPF_FUNC_perf_event_output: 6970 return &bpf_event_output_data_proto; 6971 case BPF_FUNC_get_current_pid_tgid: 6972 return &bpf_get_current_pid_tgid_proto; 6973 case BPF_FUNC_get_current_comm: 6974 return &bpf_get_current_comm_proto; 6975 #ifdef CONFIG_CGROUPS 6976 case BPF_FUNC_get_current_cgroup_id: 6977 return &bpf_get_current_cgroup_id_proto; 6978 case BPF_FUNC_get_current_ancestor_cgroup_id: 6979 return &bpf_get_current_ancestor_cgroup_id_proto; 6980 #endif 6981 #ifdef CONFIG_CGROUP_NET_CLASSID 6982 case BPF_FUNC_get_cgroup_classid: 6983 return &bpf_get_cgroup_classid_curr_proto; 6984 #endif 6985 #ifdef CONFIG_INET 6986 case BPF_FUNC_sk_lookup_tcp: 6987 return &bpf_sock_addr_sk_lookup_tcp_proto; 6988 case BPF_FUNC_sk_lookup_udp: 6989 return &bpf_sock_addr_sk_lookup_udp_proto; 6990 case BPF_FUNC_sk_release: 6991 return &bpf_sk_release_proto; 6992 case BPF_FUNC_skc_lookup_tcp: 6993 return &bpf_sock_addr_skc_lookup_tcp_proto; 6994 #endif /* CONFIG_INET */ 6995 case BPF_FUNC_sk_storage_get: 6996 return &bpf_sk_storage_get_proto; 6997 case BPF_FUNC_sk_storage_delete: 6998 return &bpf_sk_storage_delete_proto; 6999 case BPF_FUNC_setsockopt: 7000 switch (prog->expected_attach_type) { 7001 case BPF_CGROUP_INET4_BIND: 7002 case BPF_CGROUP_INET6_BIND: 7003 case BPF_CGROUP_INET4_CONNECT: 7004 case BPF_CGROUP_INET6_CONNECT: 7005 return &bpf_sock_addr_setsockopt_proto; 7006 default: 7007 return NULL; 7008 } 7009 case BPF_FUNC_getsockopt: 7010 switch (prog->expected_attach_type) { 7011 case BPF_CGROUP_INET4_BIND: 7012 case BPF_CGROUP_INET6_BIND: 7013 case BPF_CGROUP_INET4_CONNECT: 7014 case BPF_CGROUP_INET6_CONNECT: 7015 return &bpf_sock_addr_getsockopt_proto; 7016 default: 7017 return NULL; 7018 } 7019 default: 7020 return bpf_sk_base_func_proto(func_id); 7021 } 7022 } 7023 7024 static const struct bpf_func_proto * 7025 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7026 { 7027 switch (func_id) { 7028 case BPF_FUNC_skb_load_bytes: 7029 return &bpf_skb_load_bytes_proto; 7030 case BPF_FUNC_skb_load_bytes_relative: 7031 return &bpf_skb_load_bytes_relative_proto; 7032 case BPF_FUNC_get_socket_cookie: 7033 return &bpf_get_socket_cookie_proto; 7034 case BPF_FUNC_get_socket_uid: 7035 return &bpf_get_socket_uid_proto; 7036 case BPF_FUNC_perf_event_output: 7037 return &bpf_skb_event_output_proto; 7038 default: 7039 return bpf_sk_base_func_proto(func_id); 7040 } 7041 } 7042 7043 const struct bpf_func_proto bpf_sk_storage_get_proto __weak; 7044 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak; 7045 7046 static const struct bpf_func_proto * 7047 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7048 { 7049 switch (func_id) { 7050 case BPF_FUNC_get_local_storage: 7051 return &bpf_get_local_storage_proto; 7052 case BPF_FUNC_sk_fullsock: 7053 return &bpf_sk_fullsock_proto; 7054 case BPF_FUNC_sk_storage_get: 7055 return &bpf_sk_storage_get_proto; 7056 case BPF_FUNC_sk_storage_delete: 7057 return &bpf_sk_storage_delete_proto; 7058 case BPF_FUNC_perf_event_output: 7059 return &bpf_skb_event_output_proto; 7060 #ifdef CONFIG_SOCK_CGROUP_DATA 7061 case BPF_FUNC_skb_cgroup_id: 7062 return &bpf_skb_cgroup_id_proto; 7063 case BPF_FUNC_skb_ancestor_cgroup_id: 7064 return &bpf_skb_ancestor_cgroup_id_proto; 7065 case BPF_FUNC_sk_cgroup_id: 7066 return &bpf_sk_cgroup_id_proto; 7067 case BPF_FUNC_sk_ancestor_cgroup_id: 7068 return &bpf_sk_ancestor_cgroup_id_proto; 7069 #endif 7070 #ifdef CONFIG_INET 7071 case BPF_FUNC_sk_lookup_tcp: 7072 return &bpf_sk_lookup_tcp_proto; 7073 case BPF_FUNC_sk_lookup_udp: 7074 return &bpf_sk_lookup_udp_proto; 7075 case BPF_FUNC_sk_release: 7076 return &bpf_sk_release_proto; 7077 case BPF_FUNC_skc_lookup_tcp: 7078 return &bpf_skc_lookup_tcp_proto; 7079 case BPF_FUNC_tcp_sock: 7080 return &bpf_tcp_sock_proto; 7081 case BPF_FUNC_get_listener_sock: 7082 return &bpf_get_listener_sock_proto; 7083 case BPF_FUNC_skb_ecn_set_ce: 7084 return &bpf_skb_ecn_set_ce_proto; 7085 #endif 7086 default: 7087 return sk_filter_func_proto(func_id, prog); 7088 } 7089 } 7090 7091 static const struct bpf_func_proto * 7092 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7093 { 7094 switch (func_id) { 7095 case BPF_FUNC_skb_store_bytes: 7096 return &bpf_skb_store_bytes_proto; 7097 case BPF_FUNC_skb_load_bytes: 7098 return &bpf_skb_load_bytes_proto; 7099 case BPF_FUNC_skb_load_bytes_relative: 7100 return &bpf_skb_load_bytes_relative_proto; 7101 case BPF_FUNC_skb_pull_data: 7102 return &bpf_skb_pull_data_proto; 7103 case BPF_FUNC_csum_diff: 7104 return &bpf_csum_diff_proto; 7105 case BPF_FUNC_csum_update: 7106 return &bpf_csum_update_proto; 7107 case BPF_FUNC_csum_level: 7108 return &bpf_csum_level_proto; 7109 case BPF_FUNC_l3_csum_replace: 7110 return &bpf_l3_csum_replace_proto; 7111 case BPF_FUNC_l4_csum_replace: 7112 return &bpf_l4_csum_replace_proto; 7113 case BPF_FUNC_clone_redirect: 7114 return &bpf_clone_redirect_proto; 7115 case BPF_FUNC_get_cgroup_classid: 7116 return &bpf_get_cgroup_classid_proto; 7117 case BPF_FUNC_skb_vlan_push: 7118 return &bpf_skb_vlan_push_proto; 7119 case BPF_FUNC_skb_vlan_pop: 7120 return &bpf_skb_vlan_pop_proto; 7121 case BPF_FUNC_skb_change_proto: 7122 return &bpf_skb_change_proto_proto; 7123 case BPF_FUNC_skb_change_type: 7124 return &bpf_skb_change_type_proto; 7125 case BPF_FUNC_skb_adjust_room: 7126 return &bpf_skb_adjust_room_proto; 7127 case BPF_FUNC_skb_change_tail: 7128 return &bpf_skb_change_tail_proto; 7129 case BPF_FUNC_skb_change_head: 7130 return &bpf_skb_change_head_proto; 7131 case BPF_FUNC_skb_get_tunnel_key: 7132 return &bpf_skb_get_tunnel_key_proto; 7133 case BPF_FUNC_skb_set_tunnel_key: 7134 return bpf_get_skb_set_tunnel_proto(func_id); 7135 case BPF_FUNC_skb_get_tunnel_opt: 7136 return &bpf_skb_get_tunnel_opt_proto; 7137 case BPF_FUNC_skb_set_tunnel_opt: 7138 return bpf_get_skb_set_tunnel_proto(func_id); 7139 case BPF_FUNC_redirect: 7140 return &bpf_redirect_proto; 7141 case BPF_FUNC_redirect_neigh: 7142 return &bpf_redirect_neigh_proto; 7143 case BPF_FUNC_redirect_peer: 7144 return &bpf_redirect_peer_proto; 7145 case BPF_FUNC_get_route_realm: 7146 return &bpf_get_route_realm_proto; 7147 case BPF_FUNC_get_hash_recalc: 7148 return &bpf_get_hash_recalc_proto; 7149 case BPF_FUNC_set_hash_invalid: 7150 return &bpf_set_hash_invalid_proto; 7151 case BPF_FUNC_set_hash: 7152 return &bpf_set_hash_proto; 7153 case BPF_FUNC_perf_event_output: 7154 return &bpf_skb_event_output_proto; 7155 case BPF_FUNC_get_smp_processor_id: 7156 return &bpf_get_smp_processor_id_proto; 7157 case BPF_FUNC_skb_under_cgroup: 7158 return &bpf_skb_under_cgroup_proto; 7159 case BPF_FUNC_get_socket_cookie: 7160 return &bpf_get_socket_cookie_proto; 7161 case BPF_FUNC_get_socket_uid: 7162 return &bpf_get_socket_uid_proto; 7163 case BPF_FUNC_fib_lookup: 7164 return &bpf_skb_fib_lookup_proto; 7165 case BPF_FUNC_sk_fullsock: 7166 return &bpf_sk_fullsock_proto; 7167 case BPF_FUNC_sk_storage_get: 7168 return &bpf_sk_storage_get_proto; 7169 case BPF_FUNC_sk_storage_delete: 7170 return &bpf_sk_storage_delete_proto; 7171 #ifdef CONFIG_XFRM 7172 case BPF_FUNC_skb_get_xfrm_state: 7173 return &bpf_skb_get_xfrm_state_proto; 7174 #endif 7175 #ifdef CONFIG_CGROUP_NET_CLASSID 7176 case BPF_FUNC_skb_cgroup_classid: 7177 return &bpf_skb_cgroup_classid_proto; 7178 #endif 7179 #ifdef CONFIG_SOCK_CGROUP_DATA 7180 case BPF_FUNC_skb_cgroup_id: 7181 return &bpf_skb_cgroup_id_proto; 7182 case BPF_FUNC_skb_ancestor_cgroup_id: 7183 return &bpf_skb_ancestor_cgroup_id_proto; 7184 #endif 7185 #ifdef CONFIG_INET 7186 case BPF_FUNC_sk_lookup_tcp: 7187 return &bpf_sk_lookup_tcp_proto; 7188 case BPF_FUNC_sk_lookup_udp: 7189 return &bpf_sk_lookup_udp_proto; 7190 case BPF_FUNC_sk_release: 7191 return &bpf_sk_release_proto; 7192 case BPF_FUNC_tcp_sock: 7193 return &bpf_tcp_sock_proto; 7194 case BPF_FUNC_get_listener_sock: 7195 return &bpf_get_listener_sock_proto; 7196 case BPF_FUNC_skc_lookup_tcp: 7197 return &bpf_skc_lookup_tcp_proto; 7198 case BPF_FUNC_tcp_check_syncookie: 7199 return &bpf_tcp_check_syncookie_proto; 7200 case BPF_FUNC_skb_ecn_set_ce: 7201 return &bpf_skb_ecn_set_ce_proto; 7202 case BPF_FUNC_tcp_gen_syncookie: 7203 return &bpf_tcp_gen_syncookie_proto; 7204 case BPF_FUNC_sk_assign: 7205 return &bpf_sk_assign_proto; 7206 #endif 7207 default: 7208 return bpf_sk_base_func_proto(func_id); 7209 } 7210 } 7211 7212 static const struct bpf_func_proto * 7213 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7214 { 7215 switch (func_id) { 7216 case BPF_FUNC_perf_event_output: 7217 return &bpf_xdp_event_output_proto; 7218 case BPF_FUNC_get_smp_processor_id: 7219 return &bpf_get_smp_processor_id_proto; 7220 case BPF_FUNC_csum_diff: 7221 return &bpf_csum_diff_proto; 7222 case BPF_FUNC_xdp_adjust_head: 7223 return &bpf_xdp_adjust_head_proto; 7224 case BPF_FUNC_xdp_adjust_meta: 7225 return &bpf_xdp_adjust_meta_proto; 7226 case BPF_FUNC_redirect: 7227 return &bpf_xdp_redirect_proto; 7228 case BPF_FUNC_redirect_map: 7229 return &bpf_xdp_redirect_map_proto; 7230 case BPF_FUNC_xdp_adjust_tail: 7231 return &bpf_xdp_adjust_tail_proto; 7232 case BPF_FUNC_fib_lookup: 7233 return &bpf_xdp_fib_lookup_proto; 7234 #ifdef CONFIG_INET 7235 case BPF_FUNC_sk_lookup_udp: 7236 return &bpf_xdp_sk_lookup_udp_proto; 7237 case BPF_FUNC_sk_lookup_tcp: 7238 return &bpf_xdp_sk_lookup_tcp_proto; 7239 case BPF_FUNC_sk_release: 7240 return &bpf_sk_release_proto; 7241 case BPF_FUNC_skc_lookup_tcp: 7242 return &bpf_xdp_skc_lookup_tcp_proto; 7243 case BPF_FUNC_tcp_check_syncookie: 7244 return &bpf_tcp_check_syncookie_proto; 7245 case BPF_FUNC_tcp_gen_syncookie: 7246 return &bpf_tcp_gen_syncookie_proto; 7247 #endif 7248 default: 7249 return bpf_sk_base_func_proto(func_id); 7250 } 7251 } 7252 7253 const struct bpf_func_proto bpf_sock_map_update_proto __weak; 7254 const struct bpf_func_proto bpf_sock_hash_update_proto __weak; 7255 7256 static const struct bpf_func_proto * 7257 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7258 { 7259 switch (func_id) { 7260 case BPF_FUNC_setsockopt: 7261 return &bpf_sock_ops_setsockopt_proto; 7262 case BPF_FUNC_getsockopt: 7263 return &bpf_sock_ops_getsockopt_proto; 7264 case BPF_FUNC_sock_ops_cb_flags_set: 7265 return &bpf_sock_ops_cb_flags_set_proto; 7266 case BPF_FUNC_sock_map_update: 7267 return &bpf_sock_map_update_proto; 7268 case BPF_FUNC_sock_hash_update: 7269 return &bpf_sock_hash_update_proto; 7270 case BPF_FUNC_get_socket_cookie: 7271 return &bpf_get_socket_cookie_sock_ops_proto; 7272 case BPF_FUNC_get_local_storage: 7273 return &bpf_get_local_storage_proto; 7274 case BPF_FUNC_perf_event_output: 7275 return &bpf_event_output_data_proto; 7276 case BPF_FUNC_sk_storage_get: 7277 return &bpf_sk_storage_get_proto; 7278 case BPF_FUNC_sk_storage_delete: 7279 return &bpf_sk_storage_delete_proto; 7280 #ifdef CONFIG_INET 7281 case BPF_FUNC_load_hdr_opt: 7282 return &bpf_sock_ops_load_hdr_opt_proto; 7283 case BPF_FUNC_store_hdr_opt: 7284 return &bpf_sock_ops_store_hdr_opt_proto; 7285 case BPF_FUNC_reserve_hdr_opt: 7286 return &bpf_sock_ops_reserve_hdr_opt_proto; 7287 case BPF_FUNC_tcp_sock: 7288 return &bpf_tcp_sock_proto; 7289 #endif /* CONFIG_INET */ 7290 default: 7291 return bpf_sk_base_func_proto(func_id); 7292 } 7293 } 7294 7295 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak; 7296 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak; 7297 7298 static const struct bpf_func_proto * 7299 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7300 { 7301 switch (func_id) { 7302 case BPF_FUNC_msg_redirect_map: 7303 return &bpf_msg_redirect_map_proto; 7304 case BPF_FUNC_msg_redirect_hash: 7305 return &bpf_msg_redirect_hash_proto; 7306 case BPF_FUNC_msg_apply_bytes: 7307 return &bpf_msg_apply_bytes_proto; 7308 case BPF_FUNC_msg_cork_bytes: 7309 return &bpf_msg_cork_bytes_proto; 7310 case BPF_FUNC_msg_pull_data: 7311 return &bpf_msg_pull_data_proto; 7312 case BPF_FUNC_msg_push_data: 7313 return &bpf_msg_push_data_proto; 7314 case BPF_FUNC_msg_pop_data: 7315 return &bpf_msg_pop_data_proto; 7316 case BPF_FUNC_perf_event_output: 7317 return &bpf_event_output_data_proto; 7318 case BPF_FUNC_get_current_uid_gid: 7319 return &bpf_get_current_uid_gid_proto; 7320 case BPF_FUNC_get_current_pid_tgid: 7321 return &bpf_get_current_pid_tgid_proto; 7322 case BPF_FUNC_sk_storage_get: 7323 return &bpf_sk_storage_get_proto; 7324 case BPF_FUNC_sk_storage_delete: 7325 return &bpf_sk_storage_delete_proto; 7326 #ifdef CONFIG_CGROUPS 7327 case BPF_FUNC_get_current_cgroup_id: 7328 return &bpf_get_current_cgroup_id_proto; 7329 case BPF_FUNC_get_current_ancestor_cgroup_id: 7330 return &bpf_get_current_ancestor_cgroup_id_proto; 7331 #endif 7332 #ifdef CONFIG_CGROUP_NET_CLASSID 7333 case BPF_FUNC_get_cgroup_classid: 7334 return &bpf_get_cgroup_classid_curr_proto; 7335 #endif 7336 default: 7337 return bpf_sk_base_func_proto(func_id); 7338 } 7339 } 7340 7341 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak; 7342 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak; 7343 7344 static const struct bpf_func_proto * 7345 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7346 { 7347 switch (func_id) { 7348 case BPF_FUNC_skb_store_bytes: 7349 return &bpf_skb_store_bytes_proto; 7350 case BPF_FUNC_skb_load_bytes: 7351 return &bpf_skb_load_bytes_proto; 7352 case BPF_FUNC_skb_pull_data: 7353 return &sk_skb_pull_data_proto; 7354 case BPF_FUNC_skb_change_tail: 7355 return &sk_skb_change_tail_proto; 7356 case BPF_FUNC_skb_change_head: 7357 return &sk_skb_change_head_proto; 7358 case BPF_FUNC_skb_adjust_room: 7359 return &sk_skb_adjust_room_proto; 7360 case BPF_FUNC_get_socket_cookie: 7361 return &bpf_get_socket_cookie_proto; 7362 case BPF_FUNC_get_socket_uid: 7363 return &bpf_get_socket_uid_proto; 7364 case BPF_FUNC_sk_redirect_map: 7365 return &bpf_sk_redirect_map_proto; 7366 case BPF_FUNC_sk_redirect_hash: 7367 return &bpf_sk_redirect_hash_proto; 7368 case BPF_FUNC_perf_event_output: 7369 return &bpf_skb_event_output_proto; 7370 #ifdef CONFIG_INET 7371 case BPF_FUNC_sk_lookup_tcp: 7372 return &bpf_sk_lookup_tcp_proto; 7373 case BPF_FUNC_sk_lookup_udp: 7374 return &bpf_sk_lookup_udp_proto; 7375 case BPF_FUNC_sk_release: 7376 return &bpf_sk_release_proto; 7377 case BPF_FUNC_skc_lookup_tcp: 7378 return &bpf_skc_lookup_tcp_proto; 7379 #endif 7380 default: 7381 return bpf_sk_base_func_proto(func_id); 7382 } 7383 } 7384 7385 static const struct bpf_func_proto * 7386 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7387 { 7388 switch (func_id) { 7389 case BPF_FUNC_skb_load_bytes: 7390 return &bpf_flow_dissector_load_bytes_proto; 7391 default: 7392 return bpf_sk_base_func_proto(func_id); 7393 } 7394 } 7395 7396 static const struct bpf_func_proto * 7397 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7398 { 7399 switch (func_id) { 7400 case BPF_FUNC_skb_load_bytes: 7401 return &bpf_skb_load_bytes_proto; 7402 case BPF_FUNC_skb_pull_data: 7403 return &bpf_skb_pull_data_proto; 7404 case BPF_FUNC_csum_diff: 7405 return &bpf_csum_diff_proto; 7406 case BPF_FUNC_get_cgroup_classid: 7407 return &bpf_get_cgroup_classid_proto; 7408 case BPF_FUNC_get_route_realm: 7409 return &bpf_get_route_realm_proto; 7410 case BPF_FUNC_get_hash_recalc: 7411 return &bpf_get_hash_recalc_proto; 7412 case BPF_FUNC_perf_event_output: 7413 return &bpf_skb_event_output_proto; 7414 case BPF_FUNC_get_smp_processor_id: 7415 return &bpf_get_smp_processor_id_proto; 7416 case BPF_FUNC_skb_under_cgroup: 7417 return &bpf_skb_under_cgroup_proto; 7418 default: 7419 return bpf_sk_base_func_proto(func_id); 7420 } 7421 } 7422 7423 static const struct bpf_func_proto * 7424 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7425 { 7426 switch (func_id) { 7427 case BPF_FUNC_lwt_push_encap: 7428 return &bpf_lwt_in_push_encap_proto; 7429 default: 7430 return lwt_out_func_proto(func_id, prog); 7431 } 7432 } 7433 7434 static const struct bpf_func_proto * 7435 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7436 { 7437 switch (func_id) { 7438 case BPF_FUNC_skb_get_tunnel_key: 7439 return &bpf_skb_get_tunnel_key_proto; 7440 case BPF_FUNC_skb_set_tunnel_key: 7441 return bpf_get_skb_set_tunnel_proto(func_id); 7442 case BPF_FUNC_skb_get_tunnel_opt: 7443 return &bpf_skb_get_tunnel_opt_proto; 7444 case BPF_FUNC_skb_set_tunnel_opt: 7445 return bpf_get_skb_set_tunnel_proto(func_id); 7446 case BPF_FUNC_redirect: 7447 return &bpf_redirect_proto; 7448 case BPF_FUNC_clone_redirect: 7449 return &bpf_clone_redirect_proto; 7450 case BPF_FUNC_skb_change_tail: 7451 return &bpf_skb_change_tail_proto; 7452 case BPF_FUNC_skb_change_head: 7453 return &bpf_skb_change_head_proto; 7454 case BPF_FUNC_skb_store_bytes: 7455 return &bpf_skb_store_bytes_proto; 7456 case BPF_FUNC_csum_update: 7457 return &bpf_csum_update_proto; 7458 case BPF_FUNC_csum_level: 7459 return &bpf_csum_level_proto; 7460 case BPF_FUNC_l3_csum_replace: 7461 return &bpf_l3_csum_replace_proto; 7462 case BPF_FUNC_l4_csum_replace: 7463 return &bpf_l4_csum_replace_proto; 7464 case BPF_FUNC_set_hash_invalid: 7465 return &bpf_set_hash_invalid_proto; 7466 case BPF_FUNC_lwt_push_encap: 7467 return &bpf_lwt_xmit_push_encap_proto; 7468 default: 7469 return lwt_out_func_proto(func_id, prog); 7470 } 7471 } 7472 7473 static const struct bpf_func_proto * 7474 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7475 { 7476 switch (func_id) { 7477 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 7478 case BPF_FUNC_lwt_seg6_store_bytes: 7479 return &bpf_lwt_seg6_store_bytes_proto; 7480 case BPF_FUNC_lwt_seg6_action: 7481 return &bpf_lwt_seg6_action_proto; 7482 case BPF_FUNC_lwt_seg6_adjust_srh: 7483 return &bpf_lwt_seg6_adjust_srh_proto; 7484 #endif 7485 default: 7486 return lwt_out_func_proto(func_id, prog); 7487 } 7488 } 7489 7490 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type, 7491 const struct bpf_prog *prog, 7492 struct bpf_insn_access_aux *info) 7493 { 7494 const int size_default = sizeof(__u32); 7495 7496 if (off < 0 || off >= sizeof(struct __sk_buff)) 7497 return false; 7498 7499 /* The verifier guarantees that size > 0. */ 7500 if (off % size != 0) 7501 return false; 7502 7503 switch (off) { 7504 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7505 if (off + size > offsetofend(struct __sk_buff, cb[4])) 7506 return false; 7507 break; 7508 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]): 7509 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]): 7510 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4): 7511 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4): 7512 case bpf_ctx_range(struct __sk_buff, data): 7513 case bpf_ctx_range(struct __sk_buff, data_meta): 7514 case bpf_ctx_range(struct __sk_buff, data_end): 7515 if (size != size_default) 7516 return false; 7517 break; 7518 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 7519 return false; 7520 case bpf_ctx_range(struct __sk_buff, tstamp): 7521 if (size != sizeof(__u64)) 7522 return false; 7523 break; 7524 case offsetof(struct __sk_buff, sk): 7525 if (type == BPF_WRITE || size != sizeof(__u64)) 7526 return false; 7527 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 7528 break; 7529 default: 7530 /* Only narrow read access allowed for now. */ 7531 if (type == BPF_WRITE) { 7532 if (size != size_default) 7533 return false; 7534 } else { 7535 bpf_ctx_record_field_size(info, size_default); 7536 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 7537 return false; 7538 } 7539 } 7540 7541 return true; 7542 } 7543 7544 static bool sk_filter_is_valid_access(int off, int size, 7545 enum bpf_access_type type, 7546 const struct bpf_prog *prog, 7547 struct bpf_insn_access_aux *info) 7548 { 7549 switch (off) { 7550 case bpf_ctx_range(struct __sk_buff, tc_classid): 7551 case bpf_ctx_range(struct __sk_buff, data): 7552 case bpf_ctx_range(struct __sk_buff, data_meta): 7553 case bpf_ctx_range(struct __sk_buff, data_end): 7554 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 7555 case bpf_ctx_range(struct __sk_buff, tstamp): 7556 case bpf_ctx_range(struct __sk_buff, wire_len): 7557 return false; 7558 } 7559 7560 if (type == BPF_WRITE) { 7561 switch (off) { 7562 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7563 break; 7564 default: 7565 return false; 7566 } 7567 } 7568 7569 return bpf_skb_is_valid_access(off, size, type, prog, info); 7570 } 7571 7572 static bool cg_skb_is_valid_access(int off, int size, 7573 enum bpf_access_type type, 7574 const struct bpf_prog *prog, 7575 struct bpf_insn_access_aux *info) 7576 { 7577 switch (off) { 7578 case bpf_ctx_range(struct __sk_buff, tc_classid): 7579 case bpf_ctx_range(struct __sk_buff, data_meta): 7580 case bpf_ctx_range(struct __sk_buff, wire_len): 7581 return false; 7582 case bpf_ctx_range(struct __sk_buff, data): 7583 case bpf_ctx_range(struct __sk_buff, data_end): 7584 if (!bpf_capable()) 7585 return false; 7586 break; 7587 } 7588 7589 if (type == BPF_WRITE) { 7590 switch (off) { 7591 case bpf_ctx_range(struct __sk_buff, mark): 7592 case bpf_ctx_range(struct __sk_buff, priority): 7593 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7594 break; 7595 case bpf_ctx_range(struct __sk_buff, tstamp): 7596 if (!bpf_capable()) 7597 return false; 7598 break; 7599 default: 7600 return false; 7601 } 7602 } 7603 7604 switch (off) { 7605 case bpf_ctx_range(struct __sk_buff, data): 7606 info->reg_type = PTR_TO_PACKET; 7607 break; 7608 case bpf_ctx_range(struct __sk_buff, data_end): 7609 info->reg_type = PTR_TO_PACKET_END; 7610 break; 7611 } 7612 7613 return bpf_skb_is_valid_access(off, size, type, prog, info); 7614 } 7615 7616 static bool lwt_is_valid_access(int off, int size, 7617 enum bpf_access_type type, 7618 const struct bpf_prog *prog, 7619 struct bpf_insn_access_aux *info) 7620 { 7621 switch (off) { 7622 case bpf_ctx_range(struct __sk_buff, tc_classid): 7623 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 7624 case bpf_ctx_range(struct __sk_buff, data_meta): 7625 case bpf_ctx_range(struct __sk_buff, tstamp): 7626 case bpf_ctx_range(struct __sk_buff, wire_len): 7627 return false; 7628 } 7629 7630 if (type == BPF_WRITE) { 7631 switch (off) { 7632 case bpf_ctx_range(struct __sk_buff, mark): 7633 case bpf_ctx_range(struct __sk_buff, priority): 7634 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7635 break; 7636 default: 7637 return false; 7638 } 7639 } 7640 7641 switch (off) { 7642 case bpf_ctx_range(struct __sk_buff, data): 7643 info->reg_type = PTR_TO_PACKET; 7644 break; 7645 case bpf_ctx_range(struct __sk_buff, data_end): 7646 info->reg_type = PTR_TO_PACKET_END; 7647 break; 7648 } 7649 7650 return bpf_skb_is_valid_access(off, size, type, prog, info); 7651 } 7652 7653 /* Attach type specific accesses */ 7654 static bool __sock_filter_check_attach_type(int off, 7655 enum bpf_access_type access_type, 7656 enum bpf_attach_type attach_type) 7657 { 7658 switch (off) { 7659 case offsetof(struct bpf_sock, bound_dev_if): 7660 case offsetof(struct bpf_sock, mark): 7661 case offsetof(struct bpf_sock, priority): 7662 switch (attach_type) { 7663 case BPF_CGROUP_INET_SOCK_CREATE: 7664 case BPF_CGROUP_INET_SOCK_RELEASE: 7665 goto full_access; 7666 default: 7667 return false; 7668 } 7669 case bpf_ctx_range(struct bpf_sock, src_ip4): 7670 switch (attach_type) { 7671 case BPF_CGROUP_INET4_POST_BIND: 7672 goto read_only; 7673 default: 7674 return false; 7675 } 7676 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 7677 switch (attach_type) { 7678 case BPF_CGROUP_INET6_POST_BIND: 7679 goto read_only; 7680 default: 7681 return false; 7682 } 7683 case bpf_ctx_range(struct bpf_sock, src_port): 7684 switch (attach_type) { 7685 case BPF_CGROUP_INET4_POST_BIND: 7686 case BPF_CGROUP_INET6_POST_BIND: 7687 goto read_only; 7688 default: 7689 return false; 7690 } 7691 } 7692 read_only: 7693 return access_type == BPF_READ; 7694 full_access: 7695 return true; 7696 } 7697 7698 bool bpf_sock_common_is_valid_access(int off, int size, 7699 enum bpf_access_type type, 7700 struct bpf_insn_access_aux *info) 7701 { 7702 switch (off) { 7703 case bpf_ctx_range_till(struct bpf_sock, type, priority): 7704 return false; 7705 default: 7706 return bpf_sock_is_valid_access(off, size, type, info); 7707 } 7708 } 7709 7710 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 7711 struct bpf_insn_access_aux *info) 7712 { 7713 const int size_default = sizeof(__u32); 7714 7715 if (off < 0 || off >= sizeof(struct bpf_sock)) 7716 return false; 7717 if (off % size != 0) 7718 return false; 7719 7720 switch (off) { 7721 case offsetof(struct bpf_sock, state): 7722 case offsetof(struct bpf_sock, family): 7723 case offsetof(struct bpf_sock, type): 7724 case offsetof(struct bpf_sock, protocol): 7725 case offsetof(struct bpf_sock, dst_port): 7726 case offsetof(struct bpf_sock, src_port): 7727 case offsetof(struct bpf_sock, rx_queue_mapping): 7728 case bpf_ctx_range(struct bpf_sock, src_ip4): 7729 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 7730 case bpf_ctx_range(struct bpf_sock, dst_ip4): 7731 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 7732 bpf_ctx_record_field_size(info, size_default); 7733 return bpf_ctx_narrow_access_ok(off, size, size_default); 7734 } 7735 7736 return size == size_default; 7737 } 7738 7739 static bool sock_filter_is_valid_access(int off, int size, 7740 enum bpf_access_type type, 7741 const struct bpf_prog *prog, 7742 struct bpf_insn_access_aux *info) 7743 { 7744 if (!bpf_sock_is_valid_access(off, size, type, info)) 7745 return false; 7746 return __sock_filter_check_attach_type(off, type, 7747 prog->expected_attach_type); 7748 } 7749 7750 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write, 7751 const struct bpf_prog *prog) 7752 { 7753 /* Neither direct read nor direct write requires any preliminary 7754 * action. 7755 */ 7756 return 0; 7757 } 7758 7759 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write, 7760 const struct bpf_prog *prog, int drop_verdict) 7761 { 7762 struct bpf_insn *insn = insn_buf; 7763 7764 if (!direct_write) 7765 return 0; 7766 7767 /* if (!skb->cloned) 7768 * goto start; 7769 * 7770 * (Fast-path, otherwise approximation that we might be 7771 * a clone, do the rest in helper.) 7772 */ 7773 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET()); 7774 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK); 7775 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7); 7776 7777 /* ret = bpf_skb_pull_data(skb, 0); */ 7778 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1); 7779 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2); 7780 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, 7781 BPF_FUNC_skb_pull_data); 7782 /* if (!ret) 7783 * goto restore; 7784 * return TC_ACT_SHOT; 7785 */ 7786 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2); 7787 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict); 7788 *insn++ = BPF_EXIT_INSN(); 7789 7790 /* restore: */ 7791 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6); 7792 /* start: */ 7793 *insn++ = prog->insnsi[0]; 7794 7795 return insn - insn_buf; 7796 } 7797 7798 static int bpf_gen_ld_abs(const struct bpf_insn *orig, 7799 struct bpf_insn *insn_buf) 7800 { 7801 bool indirect = BPF_MODE(orig->code) == BPF_IND; 7802 struct bpf_insn *insn = insn_buf; 7803 7804 if (!indirect) { 7805 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm); 7806 } else { 7807 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg); 7808 if (orig->imm) 7809 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm); 7810 } 7811 /* We're guaranteed here that CTX is in R6. */ 7812 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX); 7813 7814 switch (BPF_SIZE(orig->code)) { 7815 case BPF_B: 7816 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache); 7817 break; 7818 case BPF_H: 7819 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache); 7820 break; 7821 case BPF_W: 7822 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache); 7823 break; 7824 } 7825 7826 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2); 7827 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0); 7828 *insn++ = BPF_EXIT_INSN(); 7829 7830 return insn - insn_buf; 7831 } 7832 7833 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write, 7834 const struct bpf_prog *prog) 7835 { 7836 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT); 7837 } 7838 7839 static bool tc_cls_act_is_valid_access(int off, int size, 7840 enum bpf_access_type type, 7841 const struct bpf_prog *prog, 7842 struct bpf_insn_access_aux *info) 7843 { 7844 if (type == BPF_WRITE) { 7845 switch (off) { 7846 case bpf_ctx_range(struct __sk_buff, mark): 7847 case bpf_ctx_range(struct __sk_buff, tc_index): 7848 case bpf_ctx_range(struct __sk_buff, priority): 7849 case bpf_ctx_range(struct __sk_buff, tc_classid): 7850 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7851 case bpf_ctx_range(struct __sk_buff, tstamp): 7852 case bpf_ctx_range(struct __sk_buff, queue_mapping): 7853 break; 7854 default: 7855 return false; 7856 } 7857 } 7858 7859 switch (off) { 7860 case bpf_ctx_range(struct __sk_buff, data): 7861 info->reg_type = PTR_TO_PACKET; 7862 break; 7863 case bpf_ctx_range(struct __sk_buff, data_meta): 7864 info->reg_type = PTR_TO_PACKET_META; 7865 break; 7866 case bpf_ctx_range(struct __sk_buff, data_end): 7867 info->reg_type = PTR_TO_PACKET_END; 7868 break; 7869 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 7870 return false; 7871 } 7872 7873 return bpf_skb_is_valid_access(off, size, type, prog, info); 7874 } 7875 7876 static bool __is_valid_xdp_access(int off, int size) 7877 { 7878 if (off < 0 || off >= sizeof(struct xdp_md)) 7879 return false; 7880 if (off % size != 0) 7881 return false; 7882 if (size != sizeof(__u32)) 7883 return false; 7884 7885 return true; 7886 } 7887 7888 static bool xdp_is_valid_access(int off, int size, 7889 enum bpf_access_type type, 7890 const struct bpf_prog *prog, 7891 struct bpf_insn_access_aux *info) 7892 { 7893 if (prog->expected_attach_type != BPF_XDP_DEVMAP) { 7894 switch (off) { 7895 case offsetof(struct xdp_md, egress_ifindex): 7896 return false; 7897 } 7898 } 7899 7900 if (type == BPF_WRITE) { 7901 if (bpf_prog_is_dev_bound(prog->aux)) { 7902 switch (off) { 7903 case offsetof(struct xdp_md, rx_queue_index): 7904 return __is_valid_xdp_access(off, size); 7905 } 7906 } 7907 return false; 7908 } 7909 7910 switch (off) { 7911 case offsetof(struct xdp_md, data): 7912 info->reg_type = PTR_TO_PACKET; 7913 break; 7914 case offsetof(struct xdp_md, data_meta): 7915 info->reg_type = PTR_TO_PACKET_META; 7916 break; 7917 case offsetof(struct xdp_md, data_end): 7918 info->reg_type = PTR_TO_PACKET_END; 7919 break; 7920 } 7921 7922 return __is_valid_xdp_access(off, size); 7923 } 7924 7925 void bpf_warn_invalid_xdp_action(u32 act) 7926 { 7927 const u32 act_max = XDP_REDIRECT; 7928 7929 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n", 7930 act > act_max ? "Illegal" : "Driver unsupported", 7931 act); 7932 } 7933 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); 7934 7935 static bool sock_addr_is_valid_access(int off, int size, 7936 enum bpf_access_type type, 7937 const struct bpf_prog *prog, 7938 struct bpf_insn_access_aux *info) 7939 { 7940 const int size_default = sizeof(__u32); 7941 7942 if (off < 0 || off >= sizeof(struct bpf_sock_addr)) 7943 return false; 7944 if (off % size != 0) 7945 return false; 7946 7947 /* Disallow access to IPv6 fields from IPv4 contex and vise 7948 * versa. 7949 */ 7950 switch (off) { 7951 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 7952 switch (prog->expected_attach_type) { 7953 case BPF_CGROUP_INET4_BIND: 7954 case BPF_CGROUP_INET4_CONNECT: 7955 case BPF_CGROUP_INET4_GETPEERNAME: 7956 case BPF_CGROUP_INET4_GETSOCKNAME: 7957 case BPF_CGROUP_UDP4_SENDMSG: 7958 case BPF_CGROUP_UDP4_RECVMSG: 7959 break; 7960 default: 7961 return false; 7962 } 7963 break; 7964 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 7965 switch (prog->expected_attach_type) { 7966 case BPF_CGROUP_INET6_BIND: 7967 case BPF_CGROUP_INET6_CONNECT: 7968 case BPF_CGROUP_INET6_GETPEERNAME: 7969 case BPF_CGROUP_INET6_GETSOCKNAME: 7970 case BPF_CGROUP_UDP6_SENDMSG: 7971 case BPF_CGROUP_UDP6_RECVMSG: 7972 break; 7973 default: 7974 return false; 7975 } 7976 break; 7977 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 7978 switch (prog->expected_attach_type) { 7979 case BPF_CGROUP_UDP4_SENDMSG: 7980 break; 7981 default: 7982 return false; 7983 } 7984 break; 7985 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 7986 msg_src_ip6[3]): 7987 switch (prog->expected_attach_type) { 7988 case BPF_CGROUP_UDP6_SENDMSG: 7989 break; 7990 default: 7991 return false; 7992 } 7993 break; 7994 } 7995 7996 switch (off) { 7997 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 7998 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 7999 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8000 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8001 msg_src_ip6[3]): 8002 case bpf_ctx_range(struct bpf_sock_addr, user_port): 8003 if (type == BPF_READ) { 8004 bpf_ctx_record_field_size(info, size_default); 8005 8006 if (bpf_ctx_wide_access_ok(off, size, 8007 struct bpf_sock_addr, 8008 user_ip6)) 8009 return true; 8010 8011 if (bpf_ctx_wide_access_ok(off, size, 8012 struct bpf_sock_addr, 8013 msg_src_ip6)) 8014 return true; 8015 8016 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8017 return false; 8018 } else { 8019 if (bpf_ctx_wide_access_ok(off, size, 8020 struct bpf_sock_addr, 8021 user_ip6)) 8022 return true; 8023 8024 if (bpf_ctx_wide_access_ok(off, size, 8025 struct bpf_sock_addr, 8026 msg_src_ip6)) 8027 return true; 8028 8029 if (size != size_default) 8030 return false; 8031 } 8032 break; 8033 case offsetof(struct bpf_sock_addr, sk): 8034 if (type != BPF_READ) 8035 return false; 8036 if (size != sizeof(__u64)) 8037 return false; 8038 info->reg_type = PTR_TO_SOCKET; 8039 break; 8040 default: 8041 if (type == BPF_READ) { 8042 if (size != size_default) 8043 return false; 8044 } else { 8045 return false; 8046 } 8047 } 8048 8049 return true; 8050 } 8051 8052 static bool sock_ops_is_valid_access(int off, int size, 8053 enum bpf_access_type type, 8054 const struct bpf_prog *prog, 8055 struct bpf_insn_access_aux *info) 8056 { 8057 const int size_default = sizeof(__u32); 8058 8059 if (off < 0 || off >= sizeof(struct bpf_sock_ops)) 8060 return false; 8061 8062 /* The verifier guarantees that size > 0. */ 8063 if (off % size != 0) 8064 return false; 8065 8066 if (type == BPF_WRITE) { 8067 switch (off) { 8068 case offsetof(struct bpf_sock_ops, reply): 8069 case offsetof(struct bpf_sock_ops, sk_txhash): 8070 if (size != size_default) 8071 return false; 8072 break; 8073 default: 8074 return false; 8075 } 8076 } else { 8077 switch (off) { 8078 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received, 8079 bytes_acked): 8080 if (size != sizeof(__u64)) 8081 return false; 8082 break; 8083 case offsetof(struct bpf_sock_ops, sk): 8084 if (size != sizeof(__u64)) 8085 return false; 8086 info->reg_type = PTR_TO_SOCKET_OR_NULL; 8087 break; 8088 case offsetof(struct bpf_sock_ops, skb_data): 8089 if (size != sizeof(__u64)) 8090 return false; 8091 info->reg_type = PTR_TO_PACKET; 8092 break; 8093 case offsetof(struct bpf_sock_ops, skb_data_end): 8094 if (size != sizeof(__u64)) 8095 return false; 8096 info->reg_type = PTR_TO_PACKET_END; 8097 break; 8098 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 8099 bpf_ctx_record_field_size(info, size_default); 8100 return bpf_ctx_narrow_access_ok(off, size, 8101 size_default); 8102 default: 8103 if (size != size_default) 8104 return false; 8105 break; 8106 } 8107 } 8108 8109 return true; 8110 } 8111 8112 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write, 8113 const struct bpf_prog *prog) 8114 { 8115 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP); 8116 } 8117 8118 static bool sk_skb_is_valid_access(int off, int size, 8119 enum bpf_access_type type, 8120 const struct bpf_prog *prog, 8121 struct bpf_insn_access_aux *info) 8122 { 8123 switch (off) { 8124 case bpf_ctx_range(struct __sk_buff, tc_classid): 8125 case bpf_ctx_range(struct __sk_buff, data_meta): 8126 case bpf_ctx_range(struct __sk_buff, tstamp): 8127 case bpf_ctx_range(struct __sk_buff, wire_len): 8128 return false; 8129 } 8130 8131 if (type == BPF_WRITE) { 8132 switch (off) { 8133 case bpf_ctx_range(struct __sk_buff, tc_index): 8134 case bpf_ctx_range(struct __sk_buff, priority): 8135 break; 8136 default: 8137 return false; 8138 } 8139 } 8140 8141 switch (off) { 8142 case bpf_ctx_range(struct __sk_buff, mark): 8143 return false; 8144 case bpf_ctx_range(struct __sk_buff, data): 8145 info->reg_type = PTR_TO_PACKET; 8146 break; 8147 case bpf_ctx_range(struct __sk_buff, data_end): 8148 info->reg_type = PTR_TO_PACKET_END; 8149 break; 8150 } 8151 8152 return bpf_skb_is_valid_access(off, size, type, prog, info); 8153 } 8154 8155 static bool sk_msg_is_valid_access(int off, int size, 8156 enum bpf_access_type type, 8157 const struct bpf_prog *prog, 8158 struct bpf_insn_access_aux *info) 8159 { 8160 if (type == BPF_WRITE) 8161 return false; 8162 8163 if (off % size != 0) 8164 return false; 8165 8166 switch (off) { 8167 case offsetof(struct sk_msg_md, data): 8168 info->reg_type = PTR_TO_PACKET; 8169 if (size != sizeof(__u64)) 8170 return false; 8171 break; 8172 case offsetof(struct sk_msg_md, data_end): 8173 info->reg_type = PTR_TO_PACKET_END; 8174 if (size != sizeof(__u64)) 8175 return false; 8176 break; 8177 case offsetof(struct sk_msg_md, sk): 8178 if (size != sizeof(__u64)) 8179 return false; 8180 info->reg_type = PTR_TO_SOCKET; 8181 break; 8182 case bpf_ctx_range(struct sk_msg_md, family): 8183 case bpf_ctx_range(struct sk_msg_md, remote_ip4): 8184 case bpf_ctx_range(struct sk_msg_md, local_ip4): 8185 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]): 8186 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]): 8187 case bpf_ctx_range(struct sk_msg_md, remote_port): 8188 case bpf_ctx_range(struct sk_msg_md, local_port): 8189 case bpf_ctx_range(struct sk_msg_md, size): 8190 if (size != sizeof(__u32)) 8191 return false; 8192 break; 8193 default: 8194 return false; 8195 } 8196 return true; 8197 } 8198 8199 static bool flow_dissector_is_valid_access(int off, int size, 8200 enum bpf_access_type type, 8201 const struct bpf_prog *prog, 8202 struct bpf_insn_access_aux *info) 8203 { 8204 const int size_default = sizeof(__u32); 8205 8206 if (off < 0 || off >= sizeof(struct __sk_buff)) 8207 return false; 8208 8209 if (type == BPF_WRITE) 8210 return false; 8211 8212 switch (off) { 8213 case bpf_ctx_range(struct __sk_buff, data): 8214 if (size != size_default) 8215 return false; 8216 info->reg_type = PTR_TO_PACKET; 8217 return true; 8218 case bpf_ctx_range(struct __sk_buff, data_end): 8219 if (size != size_default) 8220 return false; 8221 info->reg_type = PTR_TO_PACKET_END; 8222 return true; 8223 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 8224 if (size != sizeof(__u64)) 8225 return false; 8226 info->reg_type = PTR_TO_FLOW_KEYS; 8227 return true; 8228 default: 8229 return false; 8230 } 8231 } 8232 8233 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type, 8234 const struct bpf_insn *si, 8235 struct bpf_insn *insn_buf, 8236 struct bpf_prog *prog, 8237 u32 *target_size) 8238 8239 { 8240 struct bpf_insn *insn = insn_buf; 8241 8242 switch (si->off) { 8243 case offsetof(struct __sk_buff, data): 8244 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data), 8245 si->dst_reg, si->src_reg, 8246 offsetof(struct bpf_flow_dissector, data)); 8247 break; 8248 8249 case offsetof(struct __sk_buff, data_end): 8250 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end), 8251 si->dst_reg, si->src_reg, 8252 offsetof(struct bpf_flow_dissector, data_end)); 8253 break; 8254 8255 case offsetof(struct __sk_buff, flow_keys): 8256 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys), 8257 si->dst_reg, si->src_reg, 8258 offsetof(struct bpf_flow_dissector, flow_keys)); 8259 break; 8260 } 8261 8262 return insn - insn_buf; 8263 } 8264 8265 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si, 8266 struct bpf_insn *insn) 8267 { 8268 /* si->dst_reg = skb_shinfo(SKB); */ 8269 #ifdef NET_SKBUFF_DATA_USES_OFFSET 8270 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 8271 BPF_REG_AX, si->src_reg, 8272 offsetof(struct sk_buff, end)); 8273 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head), 8274 si->dst_reg, si->src_reg, 8275 offsetof(struct sk_buff, head)); 8276 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX); 8277 #else 8278 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 8279 si->dst_reg, si->src_reg, 8280 offsetof(struct sk_buff, end)); 8281 #endif 8282 8283 return insn; 8284 } 8285 8286 static u32 bpf_convert_ctx_access(enum bpf_access_type type, 8287 const struct bpf_insn *si, 8288 struct bpf_insn *insn_buf, 8289 struct bpf_prog *prog, u32 *target_size) 8290 { 8291 struct bpf_insn *insn = insn_buf; 8292 int off; 8293 8294 switch (si->off) { 8295 case offsetof(struct __sk_buff, len): 8296 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8297 bpf_target_off(struct sk_buff, len, 4, 8298 target_size)); 8299 break; 8300 8301 case offsetof(struct __sk_buff, protocol): 8302 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8303 bpf_target_off(struct sk_buff, protocol, 2, 8304 target_size)); 8305 break; 8306 8307 case offsetof(struct __sk_buff, vlan_proto): 8308 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8309 bpf_target_off(struct sk_buff, vlan_proto, 2, 8310 target_size)); 8311 break; 8312 8313 case offsetof(struct __sk_buff, priority): 8314 if (type == BPF_WRITE) 8315 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8316 bpf_target_off(struct sk_buff, priority, 4, 8317 target_size)); 8318 else 8319 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8320 bpf_target_off(struct sk_buff, priority, 4, 8321 target_size)); 8322 break; 8323 8324 case offsetof(struct __sk_buff, ingress_ifindex): 8325 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8326 bpf_target_off(struct sk_buff, skb_iif, 4, 8327 target_size)); 8328 break; 8329 8330 case offsetof(struct __sk_buff, ifindex): 8331 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 8332 si->dst_reg, si->src_reg, 8333 offsetof(struct sk_buff, dev)); 8334 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 8335 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8336 bpf_target_off(struct net_device, ifindex, 4, 8337 target_size)); 8338 break; 8339 8340 case offsetof(struct __sk_buff, hash): 8341 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8342 bpf_target_off(struct sk_buff, hash, 4, 8343 target_size)); 8344 break; 8345 8346 case offsetof(struct __sk_buff, mark): 8347 if (type == BPF_WRITE) 8348 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8349 bpf_target_off(struct sk_buff, mark, 4, 8350 target_size)); 8351 else 8352 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8353 bpf_target_off(struct sk_buff, mark, 4, 8354 target_size)); 8355 break; 8356 8357 case offsetof(struct __sk_buff, pkt_type): 8358 *target_size = 1; 8359 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 8360 PKT_TYPE_OFFSET()); 8361 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX); 8362 #ifdef __BIG_ENDIAN_BITFIELD 8363 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5); 8364 #endif 8365 break; 8366 8367 case offsetof(struct __sk_buff, queue_mapping): 8368 if (type == BPF_WRITE) { 8369 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1); 8370 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 8371 bpf_target_off(struct sk_buff, 8372 queue_mapping, 8373 2, target_size)); 8374 } else { 8375 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8376 bpf_target_off(struct sk_buff, 8377 queue_mapping, 8378 2, target_size)); 8379 } 8380 break; 8381 8382 case offsetof(struct __sk_buff, vlan_present): 8383 *target_size = 1; 8384 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 8385 PKT_VLAN_PRESENT_OFFSET()); 8386 if (PKT_VLAN_PRESENT_BIT) 8387 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT); 8388 if (PKT_VLAN_PRESENT_BIT < 7) 8389 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1); 8390 break; 8391 8392 case offsetof(struct __sk_buff, vlan_tci): 8393 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8394 bpf_target_off(struct sk_buff, vlan_tci, 2, 8395 target_size)); 8396 break; 8397 8398 case offsetof(struct __sk_buff, cb[0]) ... 8399 offsetofend(struct __sk_buff, cb[4]) - 1: 8400 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20); 8401 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 8402 offsetof(struct qdisc_skb_cb, data)) % 8403 sizeof(__u64)); 8404 8405 prog->cb_access = 1; 8406 off = si->off; 8407 off -= offsetof(struct __sk_buff, cb[0]); 8408 off += offsetof(struct sk_buff, cb); 8409 off += offsetof(struct qdisc_skb_cb, data); 8410 if (type == BPF_WRITE) 8411 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 8412 si->src_reg, off); 8413 else 8414 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 8415 si->src_reg, off); 8416 break; 8417 8418 case offsetof(struct __sk_buff, tc_classid): 8419 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2); 8420 8421 off = si->off; 8422 off -= offsetof(struct __sk_buff, tc_classid); 8423 off += offsetof(struct sk_buff, cb); 8424 off += offsetof(struct qdisc_skb_cb, tc_classid); 8425 *target_size = 2; 8426 if (type == BPF_WRITE) 8427 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, 8428 si->src_reg, off); 8429 else 8430 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, 8431 si->src_reg, off); 8432 break; 8433 8434 case offsetof(struct __sk_buff, data): 8435 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 8436 si->dst_reg, si->src_reg, 8437 offsetof(struct sk_buff, data)); 8438 break; 8439 8440 case offsetof(struct __sk_buff, data_meta): 8441 off = si->off; 8442 off -= offsetof(struct __sk_buff, data_meta); 8443 off += offsetof(struct sk_buff, cb); 8444 off += offsetof(struct bpf_skb_data_end, data_meta); 8445 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 8446 si->src_reg, off); 8447 break; 8448 8449 case offsetof(struct __sk_buff, data_end): 8450 off = si->off; 8451 off -= offsetof(struct __sk_buff, data_end); 8452 off += offsetof(struct sk_buff, cb); 8453 off += offsetof(struct bpf_skb_data_end, data_end); 8454 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 8455 si->src_reg, off); 8456 break; 8457 8458 case offsetof(struct __sk_buff, tc_index): 8459 #ifdef CONFIG_NET_SCHED 8460 if (type == BPF_WRITE) 8461 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 8462 bpf_target_off(struct sk_buff, tc_index, 2, 8463 target_size)); 8464 else 8465 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8466 bpf_target_off(struct sk_buff, tc_index, 2, 8467 target_size)); 8468 #else 8469 *target_size = 2; 8470 if (type == BPF_WRITE) 8471 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg); 8472 else 8473 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 8474 #endif 8475 break; 8476 8477 case offsetof(struct __sk_buff, napi_id): 8478 #if defined(CONFIG_NET_RX_BUSY_POLL) 8479 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8480 bpf_target_off(struct sk_buff, napi_id, 4, 8481 target_size)); 8482 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1); 8483 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 8484 #else 8485 *target_size = 4; 8486 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 8487 #endif 8488 break; 8489 case offsetof(struct __sk_buff, family): 8490 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 8491 8492 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8493 si->dst_reg, si->src_reg, 8494 offsetof(struct sk_buff, sk)); 8495 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8496 bpf_target_off(struct sock_common, 8497 skc_family, 8498 2, target_size)); 8499 break; 8500 case offsetof(struct __sk_buff, remote_ip4): 8501 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 8502 8503 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8504 si->dst_reg, si->src_reg, 8505 offsetof(struct sk_buff, sk)); 8506 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8507 bpf_target_off(struct sock_common, 8508 skc_daddr, 8509 4, target_size)); 8510 break; 8511 case offsetof(struct __sk_buff, local_ip4): 8512 BUILD_BUG_ON(sizeof_field(struct sock_common, 8513 skc_rcv_saddr) != 4); 8514 8515 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8516 si->dst_reg, si->src_reg, 8517 offsetof(struct sk_buff, sk)); 8518 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8519 bpf_target_off(struct sock_common, 8520 skc_rcv_saddr, 8521 4, target_size)); 8522 break; 8523 case offsetof(struct __sk_buff, remote_ip6[0]) ... 8524 offsetof(struct __sk_buff, remote_ip6[3]): 8525 #if IS_ENABLED(CONFIG_IPV6) 8526 BUILD_BUG_ON(sizeof_field(struct sock_common, 8527 skc_v6_daddr.s6_addr32[0]) != 4); 8528 8529 off = si->off; 8530 off -= offsetof(struct __sk_buff, remote_ip6[0]); 8531 8532 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8533 si->dst_reg, si->src_reg, 8534 offsetof(struct sk_buff, sk)); 8535 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8536 offsetof(struct sock_common, 8537 skc_v6_daddr.s6_addr32[0]) + 8538 off); 8539 #else 8540 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8541 #endif 8542 break; 8543 case offsetof(struct __sk_buff, local_ip6[0]) ... 8544 offsetof(struct __sk_buff, local_ip6[3]): 8545 #if IS_ENABLED(CONFIG_IPV6) 8546 BUILD_BUG_ON(sizeof_field(struct sock_common, 8547 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 8548 8549 off = si->off; 8550 off -= offsetof(struct __sk_buff, local_ip6[0]); 8551 8552 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8553 si->dst_reg, si->src_reg, 8554 offsetof(struct sk_buff, sk)); 8555 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8556 offsetof(struct sock_common, 8557 skc_v6_rcv_saddr.s6_addr32[0]) + 8558 off); 8559 #else 8560 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8561 #endif 8562 break; 8563 8564 case offsetof(struct __sk_buff, remote_port): 8565 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 8566 8567 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8568 si->dst_reg, si->src_reg, 8569 offsetof(struct sk_buff, sk)); 8570 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8571 bpf_target_off(struct sock_common, 8572 skc_dport, 8573 2, target_size)); 8574 #ifndef __BIG_ENDIAN_BITFIELD 8575 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 8576 #endif 8577 break; 8578 8579 case offsetof(struct __sk_buff, local_port): 8580 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 8581 8582 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8583 si->dst_reg, si->src_reg, 8584 offsetof(struct sk_buff, sk)); 8585 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8586 bpf_target_off(struct sock_common, 8587 skc_num, 2, target_size)); 8588 break; 8589 8590 case offsetof(struct __sk_buff, tstamp): 8591 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8); 8592 8593 if (type == BPF_WRITE) 8594 *insn++ = BPF_STX_MEM(BPF_DW, 8595 si->dst_reg, si->src_reg, 8596 bpf_target_off(struct sk_buff, 8597 tstamp, 8, 8598 target_size)); 8599 else 8600 *insn++ = BPF_LDX_MEM(BPF_DW, 8601 si->dst_reg, si->src_reg, 8602 bpf_target_off(struct sk_buff, 8603 tstamp, 8, 8604 target_size)); 8605 break; 8606 8607 case offsetof(struct __sk_buff, gso_segs): 8608 insn = bpf_convert_shinfo_access(si, insn); 8609 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs), 8610 si->dst_reg, si->dst_reg, 8611 bpf_target_off(struct skb_shared_info, 8612 gso_segs, 2, 8613 target_size)); 8614 break; 8615 case offsetof(struct __sk_buff, gso_size): 8616 insn = bpf_convert_shinfo_access(si, insn); 8617 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size), 8618 si->dst_reg, si->dst_reg, 8619 bpf_target_off(struct skb_shared_info, 8620 gso_size, 2, 8621 target_size)); 8622 break; 8623 case offsetof(struct __sk_buff, wire_len): 8624 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4); 8625 8626 off = si->off; 8627 off -= offsetof(struct __sk_buff, wire_len); 8628 off += offsetof(struct sk_buff, cb); 8629 off += offsetof(struct qdisc_skb_cb, pkt_len); 8630 *target_size = 4; 8631 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off); 8632 break; 8633 8634 case offsetof(struct __sk_buff, sk): 8635 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8636 si->dst_reg, si->src_reg, 8637 offsetof(struct sk_buff, sk)); 8638 break; 8639 } 8640 8641 return insn - insn_buf; 8642 } 8643 8644 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 8645 const struct bpf_insn *si, 8646 struct bpf_insn *insn_buf, 8647 struct bpf_prog *prog, u32 *target_size) 8648 { 8649 struct bpf_insn *insn = insn_buf; 8650 int off; 8651 8652 switch (si->off) { 8653 case offsetof(struct bpf_sock, bound_dev_if): 8654 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4); 8655 8656 if (type == BPF_WRITE) 8657 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8658 offsetof(struct sock, sk_bound_dev_if)); 8659 else 8660 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8661 offsetof(struct sock, sk_bound_dev_if)); 8662 break; 8663 8664 case offsetof(struct bpf_sock, mark): 8665 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4); 8666 8667 if (type == BPF_WRITE) 8668 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8669 offsetof(struct sock, sk_mark)); 8670 else 8671 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8672 offsetof(struct sock, sk_mark)); 8673 break; 8674 8675 case offsetof(struct bpf_sock, priority): 8676 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4); 8677 8678 if (type == BPF_WRITE) 8679 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8680 offsetof(struct sock, sk_priority)); 8681 else 8682 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8683 offsetof(struct sock, sk_priority)); 8684 break; 8685 8686 case offsetof(struct bpf_sock, family): 8687 *insn++ = BPF_LDX_MEM( 8688 BPF_FIELD_SIZEOF(struct sock_common, skc_family), 8689 si->dst_reg, si->src_reg, 8690 bpf_target_off(struct sock_common, 8691 skc_family, 8692 sizeof_field(struct sock_common, 8693 skc_family), 8694 target_size)); 8695 break; 8696 8697 case offsetof(struct bpf_sock, type): 8698 *insn++ = BPF_LDX_MEM( 8699 BPF_FIELD_SIZEOF(struct sock, sk_type), 8700 si->dst_reg, si->src_reg, 8701 bpf_target_off(struct sock, sk_type, 8702 sizeof_field(struct sock, sk_type), 8703 target_size)); 8704 break; 8705 8706 case offsetof(struct bpf_sock, protocol): 8707 *insn++ = BPF_LDX_MEM( 8708 BPF_FIELD_SIZEOF(struct sock, sk_protocol), 8709 si->dst_reg, si->src_reg, 8710 bpf_target_off(struct sock, sk_protocol, 8711 sizeof_field(struct sock, sk_protocol), 8712 target_size)); 8713 break; 8714 8715 case offsetof(struct bpf_sock, src_ip4): 8716 *insn++ = BPF_LDX_MEM( 8717 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 8718 bpf_target_off(struct sock_common, skc_rcv_saddr, 8719 sizeof_field(struct sock_common, 8720 skc_rcv_saddr), 8721 target_size)); 8722 break; 8723 8724 case offsetof(struct bpf_sock, dst_ip4): 8725 *insn++ = BPF_LDX_MEM( 8726 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 8727 bpf_target_off(struct sock_common, skc_daddr, 8728 sizeof_field(struct sock_common, 8729 skc_daddr), 8730 target_size)); 8731 break; 8732 8733 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 8734 #if IS_ENABLED(CONFIG_IPV6) 8735 off = si->off; 8736 off -= offsetof(struct bpf_sock, src_ip6[0]); 8737 *insn++ = BPF_LDX_MEM( 8738 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 8739 bpf_target_off( 8740 struct sock_common, 8741 skc_v6_rcv_saddr.s6_addr32[0], 8742 sizeof_field(struct sock_common, 8743 skc_v6_rcv_saddr.s6_addr32[0]), 8744 target_size) + off); 8745 #else 8746 (void)off; 8747 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8748 #endif 8749 break; 8750 8751 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 8752 #if IS_ENABLED(CONFIG_IPV6) 8753 off = si->off; 8754 off -= offsetof(struct bpf_sock, dst_ip6[0]); 8755 *insn++ = BPF_LDX_MEM( 8756 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 8757 bpf_target_off(struct sock_common, 8758 skc_v6_daddr.s6_addr32[0], 8759 sizeof_field(struct sock_common, 8760 skc_v6_daddr.s6_addr32[0]), 8761 target_size) + off); 8762 #else 8763 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8764 *target_size = 4; 8765 #endif 8766 break; 8767 8768 case offsetof(struct bpf_sock, src_port): 8769 *insn++ = BPF_LDX_MEM( 8770 BPF_FIELD_SIZEOF(struct sock_common, skc_num), 8771 si->dst_reg, si->src_reg, 8772 bpf_target_off(struct sock_common, skc_num, 8773 sizeof_field(struct sock_common, 8774 skc_num), 8775 target_size)); 8776 break; 8777 8778 case offsetof(struct bpf_sock, dst_port): 8779 *insn++ = BPF_LDX_MEM( 8780 BPF_FIELD_SIZEOF(struct sock_common, skc_dport), 8781 si->dst_reg, si->src_reg, 8782 bpf_target_off(struct sock_common, skc_dport, 8783 sizeof_field(struct sock_common, 8784 skc_dport), 8785 target_size)); 8786 break; 8787 8788 case offsetof(struct bpf_sock, state): 8789 *insn++ = BPF_LDX_MEM( 8790 BPF_FIELD_SIZEOF(struct sock_common, skc_state), 8791 si->dst_reg, si->src_reg, 8792 bpf_target_off(struct sock_common, skc_state, 8793 sizeof_field(struct sock_common, 8794 skc_state), 8795 target_size)); 8796 break; 8797 case offsetof(struct bpf_sock, rx_queue_mapping): 8798 #ifdef CONFIG_XPS 8799 *insn++ = BPF_LDX_MEM( 8800 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping), 8801 si->dst_reg, si->src_reg, 8802 bpf_target_off(struct sock, sk_rx_queue_mapping, 8803 sizeof_field(struct sock, 8804 sk_rx_queue_mapping), 8805 target_size)); 8806 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING, 8807 1); 8808 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 8809 #else 8810 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 8811 *target_size = 2; 8812 #endif 8813 break; 8814 } 8815 8816 return insn - insn_buf; 8817 } 8818 8819 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, 8820 const struct bpf_insn *si, 8821 struct bpf_insn *insn_buf, 8822 struct bpf_prog *prog, u32 *target_size) 8823 { 8824 struct bpf_insn *insn = insn_buf; 8825 8826 switch (si->off) { 8827 case offsetof(struct __sk_buff, ifindex): 8828 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 8829 si->dst_reg, si->src_reg, 8830 offsetof(struct sk_buff, dev)); 8831 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8832 bpf_target_off(struct net_device, ifindex, 4, 8833 target_size)); 8834 break; 8835 default: 8836 return bpf_convert_ctx_access(type, si, insn_buf, prog, 8837 target_size); 8838 } 8839 8840 return insn - insn_buf; 8841 } 8842 8843 static u32 xdp_convert_ctx_access(enum bpf_access_type type, 8844 const struct bpf_insn *si, 8845 struct bpf_insn *insn_buf, 8846 struct bpf_prog *prog, u32 *target_size) 8847 { 8848 struct bpf_insn *insn = insn_buf; 8849 8850 switch (si->off) { 8851 case offsetof(struct xdp_md, data): 8852 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data), 8853 si->dst_reg, si->src_reg, 8854 offsetof(struct xdp_buff, data)); 8855 break; 8856 case offsetof(struct xdp_md, data_meta): 8857 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta), 8858 si->dst_reg, si->src_reg, 8859 offsetof(struct xdp_buff, data_meta)); 8860 break; 8861 case offsetof(struct xdp_md, data_end): 8862 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end), 8863 si->dst_reg, si->src_reg, 8864 offsetof(struct xdp_buff, data_end)); 8865 break; 8866 case offsetof(struct xdp_md, ingress_ifindex): 8867 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 8868 si->dst_reg, si->src_reg, 8869 offsetof(struct xdp_buff, rxq)); 8870 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev), 8871 si->dst_reg, si->dst_reg, 8872 offsetof(struct xdp_rxq_info, dev)); 8873 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8874 offsetof(struct net_device, ifindex)); 8875 break; 8876 case offsetof(struct xdp_md, rx_queue_index): 8877 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 8878 si->dst_reg, si->src_reg, 8879 offsetof(struct xdp_buff, rxq)); 8880 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8881 offsetof(struct xdp_rxq_info, 8882 queue_index)); 8883 break; 8884 case offsetof(struct xdp_md, egress_ifindex): 8885 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq), 8886 si->dst_reg, si->src_reg, 8887 offsetof(struct xdp_buff, txq)); 8888 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev), 8889 si->dst_reg, si->dst_reg, 8890 offsetof(struct xdp_txq_info, dev)); 8891 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8892 offsetof(struct net_device, ifindex)); 8893 break; 8894 } 8895 8896 return insn - insn_buf; 8897 } 8898 8899 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of 8900 * context Structure, F is Field in context structure that contains a pointer 8901 * to Nested Structure of type NS that has the field NF. 8902 * 8903 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make 8904 * sure that SIZE is not greater than actual size of S.F.NF. 8905 * 8906 * If offset OFF is provided, the load happens from that offset relative to 8907 * offset of NF. 8908 */ 8909 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \ 8910 do { \ 8911 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \ 8912 si->src_reg, offsetof(S, F)); \ 8913 *insn++ = BPF_LDX_MEM( \ 8914 SIZE, si->dst_reg, si->dst_reg, \ 8915 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 8916 target_size) \ 8917 + OFF); \ 8918 } while (0) 8919 8920 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \ 8921 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \ 8922 BPF_FIELD_SIZEOF(NS, NF), 0) 8923 8924 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to 8925 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation. 8926 * 8927 * In addition it uses Temporary Field TF (member of struct S) as the 3rd 8928 * "register" since two registers available in convert_ctx_access are not 8929 * enough: we can't override neither SRC, since it contains value to store, nor 8930 * DST since it contains pointer to context that may be used by later 8931 * instructions. But we need a temporary place to save pointer to nested 8932 * structure whose field we want to store to. 8933 */ 8934 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \ 8935 do { \ 8936 int tmp_reg = BPF_REG_9; \ 8937 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 8938 --tmp_reg; \ 8939 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 8940 --tmp_reg; \ 8941 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \ 8942 offsetof(S, TF)); \ 8943 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \ 8944 si->dst_reg, offsetof(S, F)); \ 8945 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \ 8946 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 8947 target_size) \ 8948 + OFF); \ 8949 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \ 8950 offsetof(S, TF)); \ 8951 } while (0) 8952 8953 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \ 8954 TF) \ 8955 do { \ 8956 if (type == BPF_WRITE) { \ 8957 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \ 8958 OFF, TF); \ 8959 } else { \ 8960 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \ 8961 S, NS, F, NF, SIZE, OFF); \ 8962 } \ 8963 } while (0) 8964 8965 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \ 8966 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \ 8967 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF) 8968 8969 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type, 8970 const struct bpf_insn *si, 8971 struct bpf_insn *insn_buf, 8972 struct bpf_prog *prog, u32 *target_size) 8973 { 8974 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port); 8975 struct bpf_insn *insn = insn_buf; 8976 8977 switch (si->off) { 8978 case offsetof(struct bpf_sock_addr, user_family): 8979 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 8980 struct sockaddr, uaddr, sa_family); 8981 break; 8982 8983 case offsetof(struct bpf_sock_addr, user_ip4): 8984 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 8985 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr, 8986 sin_addr, BPF_SIZE(si->code), 0, tmp_reg); 8987 break; 8988 8989 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8990 off = si->off; 8991 off -= offsetof(struct bpf_sock_addr, user_ip6[0]); 8992 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 8993 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 8994 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off, 8995 tmp_reg); 8996 break; 8997 8998 case offsetof(struct bpf_sock_addr, user_port): 8999 /* To get port we need to know sa_family first and then treat 9000 * sockaddr as either sockaddr_in or sockaddr_in6. 9001 * Though we can simplify since port field has same offset and 9002 * size in both structures. 9003 * Here we check this invariant and use just one of the 9004 * structures if it's true. 9005 */ 9006 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) != 9007 offsetof(struct sockaddr_in6, sin6_port)); 9008 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) != 9009 sizeof_field(struct sockaddr_in6, sin6_port)); 9010 /* Account for sin6_port being smaller than user_port. */ 9011 port_size = min(port_size, BPF_LDST_BYTES(si)); 9012 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9013 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 9014 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg); 9015 break; 9016 9017 case offsetof(struct bpf_sock_addr, family): 9018 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9019 struct sock, sk, sk_family); 9020 break; 9021 9022 case offsetof(struct bpf_sock_addr, type): 9023 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9024 struct sock, sk, sk_type); 9025 break; 9026 9027 case offsetof(struct bpf_sock_addr, protocol): 9028 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9029 struct sock, sk, sk_protocol); 9030 break; 9031 9032 case offsetof(struct bpf_sock_addr, msg_src_ip4): 9033 /* Treat t_ctx as struct in_addr for msg_src_ip4. */ 9034 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9035 struct bpf_sock_addr_kern, struct in_addr, t_ctx, 9036 s_addr, BPF_SIZE(si->code), 0, tmp_reg); 9037 break; 9038 9039 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 9040 msg_src_ip6[3]): 9041 off = si->off; 9042 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]); 9043 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */ 9044 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9045 struct bpf_sock_addr_kern, struct in6_addr, t_ctx, 9046 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg); 9047 break; 9048 case offsetof(struct bpf_sock_addr, sk): 9049 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk), 9050 si->dst_reg, si->src_reg, 9051 offsetof(struct bpf_sock_addr_kern, sk)); 9052 break; 9053 } 9054 9055 return insn - insn_buf; 9056 } 9057 9058 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type, 9059 const struct bpf_insn *si, 9060 struct bpf_insn *insn_buf, 9061 struct bpf_prog *prog, 9062 u32 *target_size) 9063 { 9064 struct bpf_insn *insn = insn_buf; 9065 int off; 9066 9067 /* Helper macro for adding read access to tcp_sock or sock fields. */ 9068 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 9069 do { \ 9070 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \ 9071 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 9072 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 9073 if (si->dst_reg == reg || si->src_reg == reg) \ 9074 reg--; \ 9075 if (si->dst_reg == reg || si->src_reg == reg) \ 9076 reg--; \ 9077 if (si->dst_reg == si->src_reg) { \ 9078 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 9079 offsetof(struct bpf_sock_ops_kern, \ 9080 temp)); \ 9081 fullsock_reg = reg; \ 9082 jmp += 2; \ 9083 } \ 9084 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9085 struct bpf_sock_ops_kern, \ 9086 is_fullsock), \ 9087 fullsock_reg, si->src_reg, \ 9088 offsetof(struct bpf_sock_ops_kern, \ 9089 is_fullsock)); \ 9090 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 9091 if (si->dst_reg == si->src_reg) \ 9092 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9093 offsetof(struct bpf_sock_ops_kern, \ 9094 temp)); \ 9095 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9096 struct bpf_sock_ops_kern, sk),\ 9097 si->dst_reg, si->src_reg, \ 9098 offsetof(struct bpf_sock_ops_kern, sk));\ 9099 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \ 9100 OBJ_FIELD), \ 9101 si->dst_reg, si->dst_reg, \ 9102 offsetof(OBJ, OBJ_FIELD)); \ 9103 if (si->dst_reg == si->src_reg) { \ 9104 *insn++ = BPF_JMP_A(1); \ 9105 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9106 offsetof(struct bpf_sock_ops_kern, \ 9107 temp)); \ 9108 } \ 9109 } while (0) 9110 9111 #define SOCK_OPS_GET_SK() \ 9112 do { \ 9113 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \ 9114 if (si->dst_reg == reg || si->src_reg == reg) \ 9115 reg--; \ 9116 if (si->dst_reg == reg || si->src_reg == reg) \ 9117 reg--; \ 9118 if (si->dst_reg == si->src_reg) { \ 9119 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 9120 offsetof(struct bpf_sock_ops_kern, \ 9121 temp)); \ 9122 fullsock_reg = reg; \ 9123 jmp += 2; \ 9124 } \ 9125 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9126 struct bpf_sock_ops_kern, \ 9127 is_fullsock), \ 9128 fullsock_reg, si->src_reg, \ 9129 offsetof(struct bpf_sock_ops_kern, \ 9130 is_fullsock)); \ 9131 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 9132 if (si->dst_reg == si->src_reg) \ 9133 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9134 offsetof(struct bpf_sock_ops_kern, \ 9135 temp)); \ 9136 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9137 struct bpf_sock_ops_kern, sk),\ 9138 si->dst_reg, si->src_reg, \ 9139 offsetof(struct bpf_sock_ops_kern, sk));\ 9140 if (si->dst_reg == si->src_reg) { \ 9141 *insn++ = BPF_JMP_A(1); \ 9142 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9143 offsetof(struct bpf_sock_ops_kern, \ 9144 temp)); \ 9145 } \ 9146 } while (0) 9147 9148 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \ 9149 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock) 9150 9151 /* Helper macro for adding write access to tcp_sock or sock fields. 9152 * The macro is called with two registers, dst_reg which contains a pointer 9153 * to ctx (context) and src_reg which contains the value that should be 9154 * stored. However, we need an additional register since we cannot overwrite 9155 * dst_reg because it may be used later in the program. 9156 * Instead we "borrow" one of the other register. We first save its value 9157 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore 9158 * it at the end of the macro. 9159 */ 9160 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 9161 do { \ 9162 int reg = BPF_REG_9; \ 9163 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 9164 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 9165 if (si->dst_reg == reg || si->src_reg == reg) \ 9166 reg--; \ 9167 if (si->dst_reg == reg || si->src_reg == reg) \ 9168 reg--; \ 9169 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \ 9170 offsetof(struct bpf_sock_ops_kern, \ 9171 temp)); \ 9172 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9173 struct bpf_sock_ops_kern, \ 9174 is_fullsock), \ 9175 reg, si->dst_reg, \ 9176 offsetof(struct bpf_sock_ops_kern, \ 9177 is_fullsock)); \ 9178 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \ 9179 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9180 struct bpf_sock_ops_kern, sk),\ 9181 reg, si->dst_reg, \ 9182 offsetof(struct bpf_sock_ops_kern, sk));\ 9183 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \ 9184 reg, si->src_reg, \ 9185 offsetof(OBJ, OBJ_FIELD)); \ 9186 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \ 9187 offsetof(struct bpf_sock_ops_kern, \ 9188 temp)); \ 9189 } while (0) 9190 9191 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \ 9192 do { \ 9193 if (TYPE == BPF_WRITE) \ 9194 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 9195 else \ 9196 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 9197 } while (0) 9198 9199 if (insn > insn_buf) 9200 return insn - insn_buf; 9201 9202 switch (si->off) { 9203 case offsetof(struct bpf_sock_ops, op): 9204 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9205 op), 9206 si->dst_reg, si->src_reg, 9207 offsetof(struct bpf_sock_ops_kern, op)); 9208 break; 9209 9210 case offsetof(struct bpf_sock_ops, replylong[0]) ... 9211 offsetof(struct bpf_sock_ops, replylong[3]): 9212 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) != 9213 sizeof_field(struct bpf_sock_ops_kern, reply)); 9214 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) != 9215 sizeof_field(struct bpf_sock_ops_kern, replylong)); 9216 off = si->off; 9217 off -= offsetof(struct bpf_sock_ops, replylong[0]); 9218 off += offsetof(struct bpf_sock_ops_kern, replylong[0]); 9219 if (type == BPF_WRITE) 9220 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9221 off); 9222 else 9223 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9224 off); 9225 break; 9226 9227 case offsetof(struct bpf_sock_ops, family): 9228 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 9229 9230 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9231 struct bpf_sock_ops_kern, sk), 9232 si->dst_reg, si->src_reg, 9233 offsetof(struct bpf_sock_ops_kern, sk)); 9234 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9235 offsetof(struct sock_common, skc_family)); 9236 break; 9237 9238 case offsetof(struct bpf_sock_ops, remote_ip4): 9239 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 9240 9241 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9242 struct bpf_sock_ops_kern, sk), 9243 si->dst_reg, si->src_reg, 9244 offsetof(struct bpf_sock_ops_kern, sk)); 9245 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9246 offsetof(struct sock_common, skc_daddr)); 9247 break; 9248 9249 case offsetof(struct bpf_sock_ops, local_ip4): 9250 BUILD_BUG_ON(sizeof_field(struct sock_common, 9251 skc_rcv_saddr) != 4); 9252 9253 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9254 struct bpf_sock_ops_kern, sk), 9255 si->dst_reg, si->src_reg, 9256 offsetof(struct bpf_sock_ops_kern, sk)); 9257 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9258 offsetof(struct sock_common, 9259 skc_rcv_saddr)); 9260 break; 9261 9262 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ... 9263 offsetof(struct bpf_sock_ops, remote_ip6[3]): 9264 #if IS_ENABLED(CONFIG_IPV6) 9265 BUILD_BUG_ON(sizeof_field(struct sock_common, 9266 skc_v6_daddr.s6_addr32[0]) != 4); 9267 9268 off = si->off; 9269 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]); 9270 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9271 struct bpf_sock_ops_kern, sk), 9272 si->dst_reg, si->src_reg, 9273 offsetof(struct bpf_sock_ops_kern, sk)); 9274 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9275 offsetof(struct sock_common, 9276 skc_v6_daddr.s6_addr32[0]) + 9277 off); 9278 #else 9279 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9280 #endif 9281 break; 9282 9283 case offsetof(struct bpf_sock_ops, local_ip6[0]) ... 9284 offsetof(struct bpf_sock_ops, local_ip6[3]): 9285 #if IS_ENABLED(CONFIG_IPV6) 9286 BUILD_BUG_ON(sizeof_field(struct sock_common, 9287 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 9288 9289 off = si->off; 9290 off -= offsetof(struct bpf_sock_ops, local_ip6[0]); 9291 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9292 struct bpf_sock_ops_kern, sk), 9293 si->dst_reg, si->src_reg, 9294 offsetof(struct bpf_sock_ops_kern, sk)); 9295 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9296 offsetof(struct sock_common, 9297 skc_v6_rcv_saddr.s6_addr32[0]) + 9298 off); 9299 #else 9300 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9301 #endif 9302 break; 9303 9304 case offsetof(struct bpf_sock_ops, remote_port): 9305 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 9306 9307 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9308 struct bpf_sock_ops_kern, sk), 9309 si->dst_reg, si->src_reg, 9310 offsetof(struct bpf_sock_ops_kern, sk)); 9311 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9312 offsetof(struct sock_common, skc_dport)); 9313 #ifndef __BIG_ENDIAN_BITFIELD 9314 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 9315 #endif 9316 break; 9317 9318 case offsetof(struct bpf_sock_ops, local_port): 9319 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 9320 9321 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9322 struct bpf_sock_ops_kern, sk), 9323 si->dst_reg, si->src_reg, 9324 offsetof(struct bpf_sock_ops_kern, sk)); 9325 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9326 offsetof(struct sock_common, skc_num)); 9327 break; 9328 9329 case offsetof(struct bpf_sock_ops, is_fullsock): 9330 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9331 struct bpf_sock_ops_kern, 9332 is_fullsock), 9333 si->dst_reg, si->src_reg, 9334 offsetof(struct bpf_sock_ops_kern, 9335 is_fullsock)); 9336 break; 9337 9338 case offsetof(struct bpf_sock_ops, state): 9339 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1); 9340 9341 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9342 struct bpf_sock_ops_kern, sk), 9343 si->dst_reg, si->src_reg, 9344 offsetof(struct bpf_sock_ops_kern, sk)); 9345 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg, 9346 offsetof(struct sock_common, skc_state)); 9347 break; 9348 9349 case offsetof(struct bpf_sock_ops, rtt_min): 9350 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 9351 sizeof(struct minmax)); 9352 BUILD_BUG_ON(sizeof(struct minmax) < 9353 sizeof(struct minmax_sample)); 9354 9355 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9356 struct bpf_sock_ops_kern, sk), 9357 si->dst_reg, si->src_reg, 9358 offsetof(struct bpf_sock_ops_kern, sk)); 9359 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9360 offsetof(struct tcp_sock, rtt_min) + 9361 sizeof_field(struct minmax_sample, t)); 9362 break; 9363 9364 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags): 9365 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags, 9366 struct tcp_sock); 9367 break; 9368 9369 case offsetof(struct bpf_sock_ops, sk_txhash): 9370 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash, 9371 struct sock, type); 9372 break; 9373 case offsetof(struct bpf_sock_ops, snd_cwnd): 9374 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd); 9375 break; 9376 case offsetof(struct bpf_sock_ops, srtt_us): 9377 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us); 9378 break; 9379 case offsetof(struct bpf_sock_ops, snd_ssthresh): 9380 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh); 9381 break; 9382 case offsetof(struct bpf_sock_ops, rcv_nxt): 9383 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt); 9384 break; 9385 case offsetof(struct bpf_sock_ops, snd_nxt): 9386 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt); 9387 break; 9388 case offsetof(struct bpf_sock_ops, snd_una): 9389 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una); 9390 break; 9391 case offsetof(struct bpf_sock_ops, mss_cache): 9392 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache); 9393 break; 9394 case offsetof(struct bpf_sock_ops, ecn_flags): 9395 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags); 9396 break; 9397 case offsetof(struct bpf_sock_ops, rate_delivered): 9398 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered); 9399 break; 9400 case offsetof(struct bpf_sock_ops, rate_interval_us): 9401 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us); 9402 break; 9403 case offsetof(struct bpf_sock_ops, packets_out): 9404 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out); 9405 break; 9406 case offsetof(struct bpf_sock_ops, retrans_out): 9407 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out); 9408 break; 9409 case offsetof(struct bpf_sock_ops, total_retrans): 9410 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans); 9411 break; 9412 case offsetof(struct bpf_sock_ops, segs_in): 9413 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in); 9414 break; 9415 case offsetof(struct bpf_sock_ops, data_segs_in): 9416 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in); 9417 break; 9418 case offsetof(struct bpf_sock_ops, segs_out): 9419 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out); 9420 break; 9421 case offsetof(struct bpf_sock_ops, data_segs_out): 9422 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out); 9423 break; 9424 case offsetof(struct bpf_sock_ops, lost_out): 9425 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out); 9426 break; 9427 case offsetof(struct bpf_sock_ops, sacked_out): 9428 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out); 9429 break; 9430 case offsetof(struct bpf_sock_ops, bytes_received): 9431 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received); 9432 break; 9433 case offsetof(struct bpf_sock_ops, bytes_acked): 9434 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked); 9435 break; 9436 case offsetof(struct bpf_sock_ops, sk): 9437 SOCK_OPS_GET_SK(); 9438 break; 9439 case offsetof(struct bpf_sock_ops, skb_data_end): 9440 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9441 skb_data_end), 9442 si->dst_reg, si->src_reg, 9443 offsetof(struct bpf_sock_ops_kern, 9444 skb_data_end)); 9445 break; 9446 case offsetof(struct bpf_sock_ops, skb_data): 9447 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9448 skb), 9449 si->dst_reg, si->src_reg, 9450 offsetof(struct bpf_sock_ops_kern, 9451 skb)); 9452 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9453 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 9454 si->dst_reg, si->dst_reg, 9455 offsetof(struct sk_buff, data)); 9456 break; 9457 case offsetof(struct bpf_sock_ops, skb_len): 9458 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9459 skb), 9460 si->dst_reg, si->src_reg, 9461 offsetof(struct bpf_sock_ops_kern, 9462 skb)); 9463 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9464 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 9465 si->dst_reg, si->dst_reg, 9466 offsetof(struct sk_buff, len)); 9467 break; 9468 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 9469 off = offsetof(struct sk_buff, cb); 9470 off += offsetof(struct tcp_skb_cb, tcp_flags); 9471 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags); 9472 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9473 skb), 9474 si->dst_reg, si->src_reg, 9475 offsetof(struct bpf_sock_ops_kern, 9476 skb)); 9477 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9478 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb, 9479 tcp_flags), 9480 si->dst_reg, si->dst_reg, off); 9481 break; 9482 } 9483 return insn - insn_buf; 9484 } 9485 9486 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type, 9487 const struct bpf_insn *si, 9488 struct bpf_insn *insn_buf, 9489 struct bpf_prog *prog, u32 *target_size) 9490 { 9491 struct bpf_insn *insn = insn_buf; 9492 int off; 9493 9494 switch (si->off) { 9495 case offsetof(struct __sk_buff, data_end): 9496 off = si->off; 9497 off -= offsetof(struct __sk_buff, data_end); 9498 off += offsetof(struct sk_buff, cb); 9499 off += offsetof(struct tcp_skb_cb, bpf.data_end); 9500 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 9501 si->src_reg, off); 9502 break; 9503 default: 9504 return bpf_convert_ctx_access(type, si, insn_buf, prog, 9505 target_size); 9506 } 9507 9508 return insn - insn_buf; 9509 } 9510 9511 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type, 9512 const struct bpf_insn *si, 9513 struct bpf_insn *insn_buf, 9514 struct bpf_prog *prog, u32 *target_size) 9515 { 9516 struct bpf_insn *insn = insn_buf; 9517 #if IS_ENABLED(CONFIG_IPV6) 9518 int off; 9519 #endif 9520 9521 /* convert ctx uses the fact sg element is first in struct */ 9522 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0); 9523 9524 switch (si->off) { 9525 case offsetof(struct sk_msg_md, data): 9526 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data), 9527 si->dst_reg, si->src_reg, 9528 offsetof(struct sk_msg, data)); 9529 break; 9530 case offsetof(struct sk_msg_md, data_end): 9531 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end), 9532 si->dst_reg, si->src_reg, 9533 offsetof(struct sk_msg, data_end)); 9534 break; 9535 case offsetof(struct sk_msg_md, family): 9536 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 9537 9538 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9539 struct sk_msg, sk), 9540 si->dst_reg, si->src_reg, 9541 offsetof(struct sk_msg, sk)); 9542 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9543 offsetof(struct sock_common, skc_family)); 9544 break; 9545 9546 case offsetof(struct sk_msg_md, remote_ip4): 9547 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 9548 9549 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9550 struct sk_msg, sk), 9551 si->dst_reg, si->src_reg, 9552 offsetof(struct sk_msg, sk)); 9553 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9554 offsetof(struct sock_common, skc_daddr)); 9555 break; 9556 9557 case offsetof(struct sk_msg_md, local_ip4): 9558 BUILD_BUG_ON(sizeof_field(struct sock_common, 9559 skc_rcv_saddr) != 4); 9560 9561 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9562 struct sk_msg, sk), 9563 si->dst_reg, si->src_reg, 9564 offsetof(struct sk_msg, sk)); 9565 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9566 offsetof(struct sock_common, 9567 skc_rcv_saddr)); 9568 break; 9569 9570 case offsetof(struct sk_msg_md, remote_ip6[0]) ... 9571 offsetof(struct sk_msg_md, remote_ip6[3]): 9572 #if IS_ENABLED(CONFIG_IPV6) 9573 BUILD_BUG_ON(sizeof_field(struct sock_common, 9574 skc_v6_daddr.s6_addr32[0]) != 4); 9575 9576 off = si->off; 9577 off -= offsetof(struct sk_msg_md, remote_ip6[0]); 9578 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9579 struct sk_msg, sk), 9580 si->dst_reg, si->src_reg, 9581 offsetof(struct sk_msg, sk)); 9582 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9583 offsetof(struct sock_common, 9584 skc_v6_daddr.s6_addr32[0]) + 9585 off); 9586 #else 9587 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9588 #endif 9589 break; 9590 9591 case offsetof(struct sk_msg_md, local_ip6[0]) ... 9592 offsetof(struct sk_msg_md, local_ip6[3]): 9593 #if IS_ENABLED(CONFIG_IPV6) 9594 BUILD_BUG_ON(sizeof_field(struct sock_common, 9595 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 9596 9597 off = si->off; 9598 off -= offsetof(struct sk_msg_md, local_ip6[0]); 9599 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9600 struct sk_msg, sk), 9601 si->dst_reg, si->src_reg, 9602 offsetof(struct sk_msg, sk)); 9603 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9604 offsetof(struct sock_common, 9605 skc_v6_rcv_saddr.s6_addr32[0]) + 9606 off); 9607 #else 9608 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9609 #endif 9610 break; 9611 9612 case offsetof(struct sk_msg_md, remote_port): 9613 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 9614 9615 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9616 struct sk_msg, sk), 9617 si->dst_reg, si->src_reg, 9618 offsetof(struct sk_msg, sk)); 9619 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9620 offsetof(struct sock_common, skc_dport)); 9621 #ifndef __BIG_ENDIAN_BITFIELD 9622 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 9623 #endif 9624 break; 9625 9626 case offsetof(struct sk_msg_md, local_port): 9627 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 9628 9629 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9630 struct sk_msg, sk), 9631 si->dst_reg, si->src_reg, 9632 offsetof(struct sk_msg, sk)); 9633 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9634 offsetof(struct sock_common, skc_num)); 9635 break; 9636 9637 case offsetof(struct sk_msg_md, size): 9638 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size), 9639 si->dst_reg, si->src_reg, 9640 offsetof(struct sk_msg_sg, size)); 9641 break; 9642 9643 case offsetof(struct sk_msg_md, sk): 9644 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk), 9645 si->dst_reg, si->src_reg, 9646 offsetof(struct sk_msg, sk)); 9647 break; 9648 } 9649 9650 return insn - insn_buf; 9651 } 9652 9653 const struct bpf_verifier_ops sk_filter_verifier_ops = { 9654 .get_func_proto = sk_filter_func_proto, 9655 .is_valid_access = sk_filter_is_valid_access, 9656 .convert_ctx_access = bpf_convert_ctx_access, 9657 .gen_ld_abs = bpf_gen_ld_abs, 9658 }; 9659 9660 const struct bpf_prog_ops sk_filter_prog_ops = { 9661 .test_run = bpf_prog_test_run_skb, 9662 }; 9663 9664 const struct bpf_verifier_ops tc_cls_act_verifier_ops = { 9665 .get_func_proto = tc_cls_act_func_proto, 9666 .is_valid_access = tc_cls_act_is_valid_access, 9667 .convert_ctx_access = tc_cls_act_convert_ctx_access, 9668 .gen_prologue = tc_cls_act_prologue, 9669 .gen_ld_abs = bpf_gen_ld_abs, 9670 }; 9671 9672 const struct bpf_prog_ops tc_cls_act_prog_ops = { 9673 .test_run = bpf_prog_test_run_skb, 9674 }; 9675 9676 const struct bpf_verifier_ops xdp_verifier_ops = { 9677 .get_func_proto = xdp_func_proto, 9678 .is_valid_access = xdp_is_valid_access, 9679 .convert_ctx_access = xdp_convert_ctx_access, 9680 .gen_prologue = bpf_noop_prologue, 9681 }; 9682 9683 const struct bpf_prog_ops xdp_prog_ops = { 9684 .test_run = bpf_prog_test_run_xdp, 9685 }; 9686 9687 const struct bpf_verifier_ops cg_skb_verifier_ops = { 9688 .get_func_proto = cg_skb_func_proto, 9689 .is_valid_access = cg_skb_is_valid_access, 9690 .convert_ctx_access = bpf_convert_ctx_access, 9691 }; 9692 9693 const struct bpf_prog_ops cg_skb_prog_ops = { 9694 .test_run = bpf_prog_test_run_skb, 9695 }; 9696 9697 const struct bpf_verifier_ops lwt_in_verifier_ops = { 9698 .get_func_proto = lwt_in_func_proto, 9699 .is_valid_access = lwt_is_valid_access, 9700 .convert_ctx_access = bpf_convert_ctx_access, 9701 }; 9702 9703 const struct bpf_prog_ops lwt_in_prog_ops = { 9704 .test_run = bpf_prog_test_run_skb, 9705 }; 9706 9707 const struct bpf_verifier_ops lwt_out_verifier_ops = { 9708 .get_func_proto = lwt_out_func_proto, 9709 .is_valid_access = lwt_is_valid_access, 9710 .convert_ctx_access = bpf_convert_ctx_access, 9711 }; 9712 9713 const struct bpf_prog_ops lwt_out_prog_ops = { 9714 .test_run = bpf_prog_test_run_skb, 9715 }; 9716 9717 const struct bpf_verifier_ops lwt_xmit_verifier_ops = { 9718 .get_func_proto = lwt_xmit_func_proto, 9719 .is_valid_access = lwt_is_valid_access, 9720 .convert_ctx_access = bpf_convert_ctx_access, 9721 .gen_prologue = tc_cls_act_prologue, 9722 }; 9723 9724 const struct bpf_prog_ops lwt_xmit_prog_ops = { 9725 .test_run = bpf_prog_test_run_skb, 9726 }; 9727 9728 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = { 9729 .get_func_proto = lwt_seg6local_func_proto, 9730 .is_valid_access = lwt_is_valid_access, 9731 .convert_ctx_access = bpf_convert_ctx_access, 9732 }; 9733 9734 const struct bpf_prog_ops lwt_seg6local_prog_ops = { 9735 .test_run = bpf_prog_test_run_skb, 9736 }; 9737 9738 const struct bpf_verifier_ops cg_sock_verifier_ops = { 9739 .get_func_proto = sock_filter_func_proto, 9740 .is_valid_access = sock_filter_is_valid_access, 9741 .convert_ctx_access = bpf_sock_convert_ctx_access, 9742 }; 9743 9744 const struct bpf_prog_ops cg_sock_prog_ops = { 9745 }; 9746 9747 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = { 9748 .get_func_proto = sock_addr_func_proto, 9749 .is_valid_access = sock_addr_is_valid_access, 9750 .convert_ctx_access = sock_addr_convert_ctx_access, 9751 }; 9752 9753 const struct bpf_prog_ops cg_sock_addr_prog_ops = { 9754 }; 9755 9756 const struct bpf_verifier_ops sock_ops_verifier_ops = { 9757 .get_func_proto = sock_ops_func_proto, 9758 .is_valid_access = sock_ops_is_valid_access, 9759 .convert_ctx_access = sock_ops_convert_ctx_access, 9760 }; 9761 9762 const struct bpf_prog_ops sock_ops_prog_ops = { 9763 }; 9764 9765 const struct bpf_verifier_ops sk_skb_verifier_ops = { 9766 .get_func_proto = sk_skb_func_proto, 9767 .is_valid_access = sk_skb_is_valid_access, 9768 .convert_ctx_access = sk_skb_convert_ctx_access, 9769 .gen_prologue = sk_skb_prologue, 9770 }; 9771 9772 const struct bpf_prog_ops sk_skb_prog_ops = { 9773 }; 9774 9775 const struct bpf_verifier_ops sk_msg_verifier_ops = { 9776 .get_func_proto = sk_msg_func_proto, 9777 .is_valid_access = sk_msg_is_valid_access, 9778 .convert_ctx_access = sk_msg_convert_ctx_access, 9779 .gen_prologue = bpf_noop_prologue, 9780 }; 9781 9782 const struct bpf_prog_ops sk_msg_prog_ops = { 9783 }; 9784 9785 const struct bpf_verifier_ops flow_dissector_verifier_ops = { 9786 .get_func_proto = flow_dissector_func_proto, 9787 .is_valid_access = flow_dissector_is_valid_access, 9788 .convert_ctx_access = flow_dissector_convert_ctx_access, 9789 }; 9790 9791 const struct bpf_prog_ops flow_dissector_prog_ops = { 9792 .test_run = bpf_prog_test_run_flow_dissector, 9793 }; 9794 9795 int sk_detach_filter(struct sock *sk) 9796 { 9797 int ret = -ENOENT; 9798 struct sk_filter *filter; 9799 9800 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 9801 return -EPERM; 9802 9803 filter = rcu_dereference_protected(sk->sk_filter, 9804 lockdep_sock_is_held(sk)); 9805 if (filter) { 9806 RCU_INIT_POINTER(sk->sk_filter, NULL); 9807 sk_filter_uncharge(sk, filter); 9808 ret = 0; 9809 } 9810 9811 return ret; 9812 } 9813 EXPORT_SYMBOL_GPL(sk_detach_filter); 9814 9815 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, 9816 unsigned int len) 9817 { 9818 struct sock_fprog_kern *fprog; 9819 struct sk_filter *filter; 9820 int ret = 0; 9821 9822 lock_sock(sk); 9823 filter = rcu_dereference_protected(sk->sk_filter, 9824 lockdep_sock_is_held(sk)); 9825 if (!filter) 9826 goto out; 9827 9828 /* We're copying the filter that has been originally attached, 9829 * so no conversion/decode needed anymore. eBPF programs that 9830 * have no original program cannot be dumped through this. 9831 */ 9832 ret = -EACCES; 9833 fprog = filter->prog->orig_prog; 9834 if (!fprog) 9835 goto out; 9836 9837 ret = fprog->len; 9838 if (!len) 9839 /* User space only enquires number of filter blocks. */ 9840 goto out; 9841 9842 ret = -EINVAL; 9843 if (len < fprog->len) 9844 goto out; 9845 9846 ret = -EFAULT; 9847 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog))) 9848 goto out; 9849 9850 /* Instead of bytes, the API requests to return the number 9851 * of filter blocks. 9852 */ 9853 ret = fprog->len; 9854 out: 9855 release_sock(sk); 9856 return ret; 9857 } 9858 9859 #ifdef CONFIG_INET 9860 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern, 9861 struct sock_reuseport *reuse, 9862 struct sock *sk, struct sk_buff *skb, 9863 u32 hash) 9864 { 9865 reuse_kern->skb = skb; 9866 reuse_kern->sk = sk; 9867 reuse_kern->selected_sk = NULL; 9868 reuse_kern->data_end = skb->data + skb_headlen(skb); 9869 reuse_kern->hash = hash; 9870 reuse_kern->reuseport_id = reuse->reuseport_id; 9871 reuse_kern->bind_inany = reuse->bind_inany; 9872 } 9873 9874 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 9875 struct bpf_prog *prog, struct sk_buff *skb, 9876 u32 hash) 9877 { 9878 struct sk_reuseport_kern reuse_kern; 9879 enum sk_action action; 9880 9881 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash); 9882 action = BPF_PROG_RUN(prog, &reuse_kern); 9883 9884 if (action == SK_PASS) 9885 return reuse_kern.selected_sk; 9886 else 9887 return ERR_PTR(-ECONNREFUSED); 9888 } 9889 9890 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern, 9891 struct bpf_map *, map, void *, key, u32, flags) 9892 { 9893 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY; 9894 struct sock_reuseport *reuse; 9895 struct sock *selected_sk; 9896 9897 selected_sk = map->ops->map_lookup_elem(map, key); 9898 if (!selected_sk) 9899 return -ENOENT; 9900 9901 reuse = rcu_dereference(selected_sk->sk_reuseport_cb); 9902 if (!reuse) { 9903 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */ 9904 if (sk_is_refcounted(selected_sk)) 9905 sock_put(selected_sk); 9906 9907 /* reuseport_array has only sk with non NULL sk_reuseport_cb. 9908 * The only (!reuse) case here is - the sk has already been 9909 * unhashed (e.g. by close()), so treat it as -ENOENT. 9910 * 9911 * Other maps (e.g. sock_map) do not provide this guarantee and 9912 * the sk may never be in the reuseport group to begin with. 9913 */ 9914 return is_sockarray ? -ENOENT : -EINVAL; 9915 } 9916 9917 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) { 9918 struct sock *sk = reuse_kern->sk; 9919 9920 if (sk->sk_protocol != selected_sk->sk_protocol) 9921 return -EPROTOTYPE; 9922 else if (sk->sk_family != selected_sk->sk_family) 9923 return -EAFNOSUPPORT; 9924 9925 /* Catch all. Likely bound to a different sockaddr. */ 9926 return -EBADFD; 9927 } 9928 9929 reuse_kern->selected_sk = selected_sk; 9930 9931 return 0; 9932 } 9933 9934 static const struct bpf_func_proto sk_select_reuseport_proto = { 9935 .func = sk_select_reuseport, 9936 .gpl_only = false, 9937 .ret_type = RET_INTEGER, 9938 .arg1_type = ARG_PTR_TO_CTX, 9939 .arg2_type = ARG_CONST_MAP_PTR, 9940 .arg3_type = ARG_PTR_TO_MAP_KEY, 9941 .arg4_type = ARG_ANYTHING, 9942 }; 9943 9944 BPF_CALL_4(sk_reuseport_load_bytes, 9945 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 9946 void *, to, u32, len) 9947 { 9948 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len); 9949 } 9950 9951 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = { 9952 .func = sk_reuseport_load_bytes, 9953 .gpl_only = false, 9954 .ret_type = RET_INTEGER, 9955 .arg1_type = ARG_PTR_TO_CTX, 9956 .arg2_type = ARG_ANYTHING, 9957 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 9958 .arg4_type = ARG_CONST_SIZE, 9959 }; 9960 9961 BPF_CALL_5(sk_reuseport_load_bytes_relative, 9962 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 9963 void *, to, u32, len, u32, start_header) 9964 { 9965 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to, 9966 len, start_header); 9967 } 9968 9969 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = { 9970 .func = sk_reuseport_load_bytes_relative, 9971 .gpl_only = false, 9972 .ret_type = RET_INTEGER, 9973 .arg1_type = ARG_PTR_TO_CTX, 9974 .arg2_type = ARG_ANYTHING, 9975 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 9976 .arg4_type = ARG_CONST_SIZE, 9977 .arg5_type = ARG_ANYTHING, 9978 }; 9979 9980 static const struct bpf_func_proto * 9981 sk_reuseport_func_proto(enum bpf_func_id func_id, 9982 const struct bpf_prog *prog) 9983 { 9984 switch (func_id) { 9985 case BPF_FUNC_sk_select_reuseport: 9986 return &sk_select_reuseport_proto; 9987 case BPF_FUNC_skb_load_bytes: 9988 return &sk_reuseport_load_bytes_proto; 9989 case BPF_FUNC_skb_load_bytes_relative: 9990 return &sk_reuseport_load_bytes_relative_proto; 9991 default: 9992 return bpf_base_func_proto(func_id); 9993 } 9994 } 9995 9996 static bool 9997 sk_reuseport_is_valid_access(int off, int size, 9998 enum bpf_access_type type, 9999 const struct bpf_prog *prog, 10000 struct bpf_insn_access_aux *info) 10001 { 10002 const u32 size_default = sizeof(__u32); 10003 10004 if (off < 0 || off >= sizeof(struct sk_reuseport_md) || 10005 off % size || type != BPF_READ) 10006 return false; 10007 10008 switch (off) { 10009 case offsetof(struct sk_reuseport_md, data): 10010 info->reg_type = PTR_TO_PACKET; 10011 return size == sizeof(__u64); 10012 10013 case offsetof(struct sk_reuseport_md, data_end): 10014 info->reg_type = PTR_TO_PACKET_END; 10015 return size == sizeof(__u64); 10016 10017 case offsetof(struct sk_reuseport_md, hash): 10018 return size == size_default; 10019 10020 /* Fields that allow narrowing */ 10021 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol): 10022 if (size < sizeof_field(struct sk_buff, protocol)) 10023 return false; 10024 fallthrough; 10025 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol): 10026 case bpf_ctx_range(struct sk_reuseport_md, bind_inany): 10027 case bpf_ctx_range(struct sk_reuseport_md, len): 10028 bpf_ctx_record_field_size(info, size_default); 10029 return bpf_ctx_narrow_access_ok(off, size, size_default); 10030 10031 default: 10032 return false; 10033 } 10034 } 10035 10036 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \ 10037 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \ 10038 si->dst_reg, si->src_reg, \ 10039 bpf_target_off(struct sk_reuseport_kern, F, \ 10040 sizeof_field(struct sk_reuseport_kern, F), \ 10041 target_size)); \ 10042 }) 10043 10044 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \ 10045 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 10046 struct sk_buff, \ 10047 skb, \ 10048 SKB_FIELD) 10049 10050 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \ 10051 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 10052 struct sock, \ 10053 sk, \ 10054 SK_FIELD) 10055 10056 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type, 10057 const struct bpf_insn *si, 10058 struct bpf_insn *insn_buf, 10059 struct bpf_prog *prog, 10060 u32 *target_size) 10061 { 10062 struct bpf_insn *insn = insn_buf; 10063 10064 switch (si->off) { 10065 case offsetof(struct sk_reuseport_md, data): 10066 SK_REUSEPORT_LOAD_SKB_FIELD(data); 10067 break; 10068 10069 case offsetof(struct sk_reuseport_md, len): 10070 SK_REUSEPORT_LOAD_SKB_FIELD(len); 10071 break; 10072 10073 case offsetof(struct sk_reuseport_md, eth_protocol): 10074 SK_REUSEPORT_LOAD_SKB_FIELD(protocol); 10075 break; 10076 10077 case offsetof(struct sk_reuseport_md, ip_protocol): 10078 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol); 10079 break; 10080 10081 case offsetof(struct sk_reuseport_md, data_end): 10082 SK_REUSEPORT_LOAD_FIELD(data_end); 10083 break; 10084 10085 case offsetof(struct sk_reuseport_md, hash): 10086 SK_REUSEPORT_LOAD_FIELD(hash); 10087 break; 10088 10089 case offsetof(struct sk_reuseport_md, bind_inany): 10090 SK_REUSEPORT_LOAD_FIELD(bind_inany); 10091 break; 10092 } 10093 10094 return insn - insn_buf; 10095 } 10096 10097 const struct bpf_verifier_ops sk_reuseport_verifier_ops = { 10098 .get_func_proto = sk_reuseport_func_proto, 10099 .is_valid_access = sk_reuseport_is_valid_access, 10100 .convert_ctx_access = sk_reuseport_convert_ctx_access, 10101 }; 10102 10103 const struct bpf_prog_ops sk_reuseport_prog_ops = { 10104 }; 10105 10106 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled); 10107 EXPORT_SYMBOL(bpf_sk_lookup_enabled); 10108 10109 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx, 10110 struct sock *, sk, u64, flags) 10111 { 10112 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE | 10113 BPF_SK_LOOKUP_F_NO_REUSEPORT))) 10114 return -EINVAL; 10115 if (unlikely(sk && sk_is_refcounted(sk))) 10116 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */ 10117 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED)) 10118 return -ESOCKTNOSUPPORT; /* reject connected sockets */ 10119 10120 /* Check if socket is suitable for packet L3/L4 protocol */ 10121 if (sk && sk->sk_protocol != ctx->protocol) 10122 return -EPROTOTYPE; 10123 if (sk && sk->sk_family != ctx->family && 10124 (sk->sk_family == AF_INET || ipv6_only_sock(sk))) 10125 return -EAFNOSUPPORT; 10126 10127 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE)) 10128 return -EEXIST; 10129 10130 /* Select socket as lookup result */ 10131 ctx->selected_sk = sk; 10132 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT; 10133 return 0; 10134 } 10135 10136 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = { 10137 .func = bpf_sk_lookup_assign, 10138 .gpl_only = false, 10139 .ret_type = RET_INTEGER, 10140 .arg1_type = ARG_PTR_TO_CTX, 10141 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL, 10142 .arg3_type = ARG_ANYTHING, 10143 }; 10144 10145 static const struct bpf_func_proto * 10146 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 10147 { 10148 switch (func_id) { 10149 case BPF_FUNC_perf_event_output: 10150 return &bpf_event_output_data_proto; 10151 case BPF_FUNC_sk_assign: 10152 return &bpf_sk_lookup_assign_proto; 10153 case BPF_FUNC_sk_release: 10154 return &bpf_sk_release_proto; 10155 default: 10156 return bpf_sk_base_func_proto(func_id); 10157 } 10158 } 10159 10160 static bool sk_lookup_is_valid_access(int off, int size, 10161 enum bpf_access_type type, 10162 const struct bpf_prog *prog, 10163 struct bpf_insn_access_aux *info) 10164 { 10165 if (off < 0 || off >= sizeof(struct bpf_sk_lookup)) 10166 return false; 10167 if (off % size != 0) 10168 return false; 10169 if (type != BPF_READ) 10170 return false; 10171 10172 switch (off) { 10173 case offsetof(struct bpf_sk_lookup, sk): 10174 info->reg_type = PTR_TO_SOCKET_OR_NULL; 10175 return size == sizeof(__u64); 10176 10177 case bpf_ctx_range(struct bpf_sk_lookup, family): 10178 case bpf_ctx_range(struct bpf_sk_lookup, protocol): 10179 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4): 10180 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4): 10181 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]): 10182 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]): 10183 case bpf_ctx_range(struct bpf_sk_lookup, remote_port): 10184 case bpf_ctx_range(struct bpf_sk_lookup, local_port): 10185 bpf_ctx_record_field_size(info, sizeof(__u32)); 10186 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32)); 10187 10188 default: 10189 return false; 10190 } 10191 } 10192 10193 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type, 10194 const struct bpf_insn *si, 10195 struct bpf_insn *insn_buf, 10196 struct bpf_prog *prog, 10197 u32 *target_size) 10198 { 10199 struct bpf_insn *insn = insn_buf; 10200 10201 switch (si->off) { 10202 case offsetof(struct bpf_sk_lookup, sk): 10203 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 10204 offsetof(struct bpf_sk_lookup_kern, selected_sk)); 10205 break; 10206 10207 case offsetof(struct bpf_sk_lookup, family): 10208 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10209 bpf_target_off(struct bpf_sk_lookup_kern, 10210 family, 2, target_size)); 10211 break; 10212 10213 case offsetof(struct bpf_sk_lookup, protocol): 10214 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10215 bpf_target_off(struct bpf_sk_lookup_kern, 10216 protocol, 2, target_size)); 10217 break; 10218 10219 case offsetof(struct bpf_sk_lookup, remote_ip4): 10220 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10221 bpf_target_off(struct bpf_sk_lookup_kern, 10222 v4.saddr, 4, target_size)); 10223 break; 10224 10225 case offsetof(struct bpf_sk_lookup, local_ip4): 10226 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10227 bpf_target_off(struct bpf_sk_lookup_kern, 10228 v4.daddr, 4, target_size)); 10229 break; 10230 10231 case bpf_ctx_range_till(struct bpf_sk_lookup, 10232 remote_ip6[0], remote_ip6[3]): { 10233 #if IS_ENABLED(CONFIG_IPV6) 10234 int off = si->off; 10235 10236 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]); 10237 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 10238 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 10239 offsetof(struct bpf_sk_lookup_kern, v6.saddr)); 10240 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10241 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 10242 #else 10243 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10244 #endif 10245 break; 10246 } 10247 case bpf_ctx_range_till(struct bpf_sk_lookup, 10248 local_ip6[0], local_ip6[3]): { 10249 #if IS_ENABLED(CONFIG_IPV6) 10250 int off = si->off; 10251 10252 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]); 10253 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 10254 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 10255 offsetof(struct bpf_sk_lookup_kern, v6.daddr)); 10256 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10257 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 10258 #else 10259 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10260 #endif 10261 break; 10262 } 10263 case offsetof(struct bpf_sk_lookup, remote_port): 10264 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10265 bpf_target_off(struct bpf_sk_lookup_kern, 10266 sport, 2, target_size)); 10267 break; 10268 10269 case offsetof(struct bpf_sk_lookup, local_port): 10270 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10271 bpf_target_off(struct bpf_sk_lookup_kern, 10272 dport, 2, target_size)); 10273 break; 10274 } 10275 10276 return insn - insn_buf; 10277 } 10278 10279 const struct bpf_prog_ops sk_lookup_prog_ops = { 10280 }; 10281 10282 const struct bpf_verifier_ops sk_lookup_verifier_ops = { 10283 .get_func_proto = sk_lookup_func_proto, 10284 .is_valid_access = sk_lookup_is_valid_access, 10285 .convert_ctx_access = sk_lookup_convert_ctx_access, 10286 }; 10287 10288 #endif /* CONFIG_INET */ 10289 10290 DEFINE_BPF_DISPATCHER(xdp) 10291 10292 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog) 10293 { 10294 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog); 10295 } 10296 10297 #ifdef CONFIG_DEBUG_INFO_BTF 10298 BTF_ID_LIST_GLOBAL(btf_sock_ids) 10299 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type) 10300 BTF_SOCK_TYPE_xxx 10301 #undef BTF_SOCK_TYPE 10302 #else 10303 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE]; 10304 #endif 10305 10306 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk) 10307 { 10308 /* tcp6_sock type is not generated in dwarf and hence btf, 10309 * trigger an explicit type generation here. 10310 */ 10311 BTF_TYPE_EMIT(struct tcp6_sock); 10312 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && 10313 sk->sk_family == AF_INET6) 10314 return (unsigned long)sk; 10315 10316 return (unsigned long)NULL; 10317 } 10318 10319 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = { 10320 .func = bpf_skc_to_tcp6_sock, 10321 .gpl_only = false, 10322 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10323 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10324 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6], 10325 }; 10326 10327 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk) 10328 { 10329 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 10330 return (unsigned long)sk; 10331 10332 return (unsigned long)NULL; 10333 } 10334 10335 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = { 10336 .func = bpf_skc_to_tcp_sock, 10337 .gpl_only = false, 10338 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10339 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10340 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP], 10341 }; 10342 10343 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk) 10344 { 10345 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not 10346 * generated if CONFIG_INET=n. Trigger an explicit generation here. 10347 */ 10348 BTF_TYPE_EMIT(struct inet_timewait_sock); 10349 BTF_TYPE_EMIT(struct tcp_timewait_sock); 10350 10351 #ifdef CONFIG_INET 10352 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT) 10353 return (unsigned long)sk; 10354 #endif 10355 10356 #if IS_BUILTIN(CONFIG_IPV6) 10357 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT) 10358 return (unsigned long)sk; 10359 #endif 10360 10361 return (unsigned long)NULL; 10362 } 10363 10364 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = { 10365 .func = bpf_skc_to_tcp_timewait_sock, 10366 .gpl_only = false, 10367 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10368 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10369 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW], 10370 }; 10371 10372 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk) 10373 { 10374 #ifdef CONFIG_INET 10375 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV) 10376 return (unsigned long)sk; 10377 #endif 10378 10379 #if IS_BUILTIN(CONFIG_IPV6) 10380 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV) 10381 return (unsigned long)sk; 10382 #endif 10383 10384 return (unsigned long)NULL; 10385 } 10386 10387 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = { 10388 .func = bpf_skc_to_tcp_request_sock, 10389 .gpl_only = false, 10390 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10391 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10392 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ], 10393 }; 10394 10395 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk) 10396 { 10397 /* udp6_sock type is not generated in dwarf and hence btf, 10398 * trigger an explicit type generation here. 10399 */ 10400 BTF_TYPE_EMIT(struct udp6_sock); 10401 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP && 10402 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6) 10403 return (unsigned long)sk; 10404 10405 return (unsigned long)NULL; 10406 } 10407 10408 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = { 10409 .func = bpf_skc_to_udp6_sock, 10410 .gpl_only = false, 10411 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10412 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10413 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6], 10414 }; 10415 10416 BPF_CALL_1(bpf_sock_from_file, struct file *, file) 10417 { 10418 return (unsigned long)sock_from_file(file); 10419 } 10420 10421 BTF_ID_LIST(bpf_sock_from_file_btf_ids) 10422 BTF_ID(struct, socket) 10423 BTF_ID(struct, file) 10424 10425 const struct bpf_func_proto bpf_sock_from_file_proto = { 10426 .func = bpf_sock_from_file, 10427 .gpl_only = false, 10428 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10429 .ret_btf_id = &bpf_sock_from_file_btf_ids[0], 10430 .arg1_type = ARG_PTR_TO_BTF_ID, 10431 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1], 10432 }; 10433 10434 static const struct bpf_func_proto * 10435 bpf_sk_base_func_proto(enum bpf_func_id func_id) 10436 { 10437 const struct bpf_func_proto *func; 10438 10439 switch (func_id) { 10440 case BPF_FUNC_skc_to_tcp6_sock: 10441 func = &bpf_skc_to_tcp6_sock_proto; 10442 break; 10443 case BPF_FUNC_skc_to_tcp_sock: 10444 func = &bpf_skc_to_tcp_sock_proto; 10445 break; 10446 case BPF_FUNC_skc_to_tcp_timewait_sock: 10447 func = &bpf_skc_to_tcp_timewait_sock_proto; 10448 break; 10449 case BPF_FUNC_skc_to_tcp_request_sock: 10450 func = &bpf_skc_to_tcp_request_sock_proto; 10451 break; 10452 case BPF_FUNC_skc_to_udp6_sock: 10453 func = &bpf_skc_to_udp6_sock_proto; 10454 break; 10455 default: 10456 return bpf_base_func_proto(func_id); 10457 } 10458 10459 if (!perfmon_capable()) 10460 return NULL; 10461 10462 return func; 10463 } 10464