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 default: 4914 ret = -EINVAL; 4915 } 4916 } 4917 #endif 4918 } else { 4919 ret = -EINVAL; 4920 } 4921 return ret; 4922 } 4923 4924 static int _bpf_getsockopt(struct sock *sk, int level, int optname, 4925 char *optval, int optlen) 4926 { 4927 if (!sk_fullsock(sk)) 4928 goto err_clear; 4929 4930 sock_owned_by_me(sk); 4931 4932 #ifdef CONFIG_INET 4933 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) { 4934 struct inet_connection_sock *icsk; 4935 struct tcp_sock *tp; 4936 4937 switch (optname) { 4938 case TCP_CONGESTION: 4939 icsk = inet_csk(sk); 4940 4941 if (!icsk->icsk_ca_ops || optlen <= 1) 4942 goto err_clear; 4943 strncpy(optval, icsk->icsk_ca_ops->name, optlen); 4944 optval[optlen - 1] = 0; 4945 break; 4946 case TCP_SAVED_SYN: 4947 tp = tcp_sk(sk); 4948 4949 if (optlen <= 0 || !tp->saved_syn || 4950 optlen > tcp_saved_syn_len(tp->saved_syn)) 4951 goto err_clear; 4952 memcpy(optval, tp->saved_syn->data, optlen); 4953 break; 4954 default: 4955 goto err_clear; 4956 } 4957 } else if (level == SOL_IP) { 4958 struct inet_sock *inet = inet_sk(sk); 4959 4960 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 4961 goto err_clear; 4962 4963 /* Only some options are supported */ 4964 switch (optname) { 4965 case IP_TOS: 4966 *((int *)optval) = (int)inet->tos; 4967 break; 4968 default: 4969 goto err_clear; 4970 } 4971 #if IS_ENABLED(CONFIG_IPV6) 4972 } else if (level == SOL_IPV6) { 4973 struct ipv6_pinfo *np = inet6_sk(sk); 4974 4975 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 4976 goto err_clear; 4977 4978 /* Only some options are supported */ 4979 switch (optname) { 4980 case IPV6_TCLASS: 4981 *((int *)optval) = (int)np->tclass; 4982 break; 4983 default: 4984 goto err_clear; 4985 } 4986 #endif 4987 } else { 4988 goto err_clear; 4989 } 4990 return 0; 4991 #endif 4992 err_clear: 4993 memset(optval, 0, optlen); 4994 return -EINVAL; 4995 } 4996 4997 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx, 4998 int, level, int, optname, char *, optval, int, optlen) 4999 { 5000 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen); 5001 } 5002 5003 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = { 5004 .func = bpf_sock_addr_setsockopt, 5005 .gpl_only = false, 5006 .ret_type = RET_INTEGER, 5007 .arg1_type = ARG_PTR_TO_CTX, 5008 .arg2_type = ARG_ANYTHING, 5009 .arg3_type = ARG_ANYTHING, 5010 .arg4_type = ARG_PTR_TO_MEM, 5011 .arg5_type = ARG_CONST_SIZE, 5012 }; 5013 5014 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx, 5015 int, level, int, optname, char *, optval, int, optlen) 5016 { 5017 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen); 5018 } 5019 5020 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = { 5021 .func = bpf_sock_addr_getsockopt, 5022 .gpl_only = false, 5023 .ret_type = RET_INTEGER, 5024 .arg1_type = ARG_PTR_TO_CTX, 5025 .arg2_type = ARG_ANYTHING, 5026 .arg3_type = ARG_ANYTHING, 5027 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5028 .arg5_type = ARG_CONST_SIZE, 5029 }; 5030 5031 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5032 int, level, int, optname, char *, optval, int, optlen) 5033 { 5034 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen); 5035 } 5036 5037 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = { 5038 .func = bpf_sock_ops_setsockopt, 5039 .gpl_only = false, 5040 .ret_type = RET_INTEGER, 5041 .arg1_type = ARG_PTR_TO_CTX, 5042 .arg2_type = ARG_ANYTHING, 5043 .arg3_type = ARG_ANYTHING, 5044 .arg4_type = ARG_PTR_TO_MEM, 5045 .arg5_type = ARG_CONST_SIZE, 5046 }; 5047 5048 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock, 5049 int optname, const u8 **start) 5050 { 5051 struct sk_buff *syn_skb = bpf_sock->syn_skb; 5052 const u8 *hdr_start; 5053 int ret; 5054 5055 if (syn_skb) { 5056 /* sk is a request_sock here */ 5057 5058 if (optname == TCP_BPF_SYN) { 5059 hdr_start = syn_skb->data; 5060 ret = tcp_hdrlen(syn_skb); 5061 } else if (optname == TCP_BPF_SYN_IP) { 5062 hdr_start = skb_network_header(syn_skb); 5063 ret = skb_network_header_len(syn_skb) + 5064 tcp_hdrlen(syn_skb); 5065 } else { 5066 /* optname == TCP_BPF_SYN_MAC */ 5067 hdr_start = skb_mac_header(syn_skb); 5068 ret = skb_mac_header_len(syn_skb) + 5069 skb_network_header_len(syn_skb) + 5070 tcp_hdrlen(syn_skb); 5071 } 5072 } else { 5073 struct sock *sk = bpf_sock->sk; 5074 struct saved_syn *saved_syn; 5075 5076 if (sk->sk_state == TCP_NEW_SYN_RECV) 5077 /* synack retransmit. bpf_sock->syn_skb will 5078 * not be available. It has to resort to 5079 * saved_syn (if it is saved). 5080 */ 5081 saved_syn = inet_reqsk(sk)->saved_syn; 5082 else 5083 saved_syn = tcp_sk(sk)->saved_syn; 5084 5085 if (!saved_syn) 5086 return -ENOENT; 5087 5088 if (optname == TCP_BPF_SYN) { 5089 hdr_start = saved_syn->data + 5090 saved_syn->mac_hdrlen + 5091 saved_syn->network_hdrlen; 5092 ret = saved_syn->tcp_hdrlen; 5093 } else if (optname == TCP_BPF_SYN_IP) { 5094 hdr_start = saved_syn->data + 5095 saved_syn->mac_hdrlen; 5096 ret = saved_syn->network_hdrlen + 5097 saved_syn->tcp_hdrlen; 5098 } else { 5099 /* optname == TCP_BPF_SYN_MAC */ 5100 5101 /* TCP_SAVE_SYN may not have saved the mac hdr */ 5102 if (!saved_syn->mac_hdrlen) 5103 return -ENOENT; 5104 5105 hdr_start = saved_syn->data; 5106 ret = saved_syn->mac_hdrlen + 5107 saved_syn->network_hdrlen + 5108 saved_syn->tcp_hdrlen; 5109 } 5110 } 5111 5112 *start = hdr_start; 5113 return ret; 5114 } 5115 5116 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5117 int, level, int, optname, char *, optval, int, optlen) 5118 { 5119 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP && 5120 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) { 5121 int ret, copy_len = 0; 5122 const u8 *start; 5123 5124 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start); 5125 if (ret > 0) { 5126 copy_len = ret; 5127 if (optlen < copy_len) { 5128 copy_len = optlen; 5129 ret = -ENOSPC; 5130 } 5131 5132 memcpy(optval, start, copy_len); 5133 } 5134 5135 /* Zero out unused buffer at the end */ 5136 memset(optval + copy_len, 0, optlen - copy_len); 5137 5138 return ret; 5139 } 5140 5141 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen); 5142 } 5143 5144 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = { 5145 .func = bpf_sock_ops_getsockopt, 5146 .gpl_only = false, 5147 .ret_type = RET_INTEGER, 5148 .arg1_type = ARG_PTR_TO_CTX, 5149 .arg2_type = ARG_ANYTHING, 5150 .arg3_type = ARG_ANYTHING, 5151 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5152 .arg5_type = ARG_CONST_SIZE, 5153 }; 5154 5155 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock, 5156 int, argval) 5157 { 5158 struct sock *sk = bpf_sock->sk; 5159 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS; 5160 5161 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk)) 5162 return -EINVAL; 5163 5164 tcp_sk(sk)->bpf_sock_ops_cb_flags = val; 5165 5166 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS); 5167 } 5168 5169 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = { 5170 .func = bpf_sock_ops_cb_flags_set, 5171 .gpl_only = false, 5172 .ret_type = RET_INTEGER, 5173 .arg1_type = ARG_PTR_TO_CTX, 5174 .arg2_type = ARG_ANYTHING, 5175 }; 5176 5177 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly; 5178 EXPORT_SYMBOL_GPL(ipv6_bpf_stub); 5179 5180 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr, 5181 int, addr_len) 5182 { 5183 #ifdef CONFIG_INET 5184 struct sock *sk = ctx->sk; 5185 u32 flags = BIND_FROM_BPF; 5186 int err; 5187 5188 err = -EINVAL; 5189 if (addr_len < offsetofend(struct sockaddr, sa_family)) 5190 return err; 5191 if (addr->sa_family == AF_INET) { 5192 if (addr_len < sizeof(struct sockaddr_in)) 5193 return err; 5194 if (((struct sockaddr_in *)addr)->sin_port == htons(0)) 5195 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5196 return __inet_bind(sk, addr, addr_len, flags); 5197 #if IS_ENABLED(CONFIG_IPV6) 5198 } else if (addr->sa_family == AF_INET6) { 5199 if (addr_len < SIN6_LEN_RFC2133) 5200 return err; 5201 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0)) 5202 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5203 /* ipv6_bpf_stub cannot be NULL, since it's called from 5204 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded 5205 */ 5206 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags); 5207 #endif /* CONFIG_IPV6 */ 5208 } 5209 #endif /* CONFIG_INET */ 5210 5211 return -EAFNOSUPPORT; 5212 } 5213 5214 static const struct bpf_func_proto bpf_bind_proto = { 5215 .func = bpf_bind, 5216 .gpl_only = false, 5217 .ret_type = RET_INTEGER, 5218 .arg1_type = ARG_PTR_TO_CTX, 5219 .arg2_type = ARG_PTR_TO_MEM, 5220 .arg3_type = ARG_CONST_SIZE, 5221 }; 5222 5223 #ifdef CONFIG_XFRM 5224 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index, 5225 struct bpf_xfrm_state *, to, u32, size, u64, flags) 5226 { 5227 const struct sec_path *sp = skb_sec_path(skb); 5228 const struct xfrm_state *x; 5229 5230 if (!sp || unlikely(index >= sp->len || flags)) 5231 goto err_clear; 5232 5233 x = sp->xvec[index]; 5234 5235 if (unlikely(size != sizeof(struct bpf_xfrm_state))) 5236 goto err_clear; 5237 5238 to->reqid = x->props.reqid; 5239 to->spi = x->id.spi; 5240 to->family = x->props.family; 5241 to->ext = 0; 5242 5243 if (to->family == AF_INET6) { 5244 memcpy(to->remote_ipv6, x->props.saddr.a6, 5245 sizeof(to->remote_ipv6)); 5246 } else { 5247 to->remote_ipv4 = x->props.saddr.a4; 5248 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 5249 } 5250 5251 return 0; 5252 err_clear: 5253 memset(to, 0, size); 5254 return -EINVAL; 5255 } 5256 5257 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = { 5258 .func = bpf_skb_get_xfrm_state, 5259 .gpl_only = false, 5260 .ret_type = RET_INTEGER, 5261 .arg1_type = ARG_PTR_TO_CTX, 5262 .arg2_type = ARG_ANYTHING, 5263 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 5264 .arg4_type = ARG_CONST_SIZE, 5265 .arg5_type = ARG_ANYTHING, 5266 }; 5267 #endif 5268 5269 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6) 5270 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, 5271 const struct neighbour *neigh, 5272 const struct net_device *dev) 5273 { 5274 memcpy(params->dmac, neigh->ha, ETH_ALEN); 5275 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 5276 params->h_vlan_TCI = 0; 5277 params->h_vlan_proto = 0; 5278 5279 return 0; 5280 } 5281 #endif 5282 5283 #if IS_ENABLED(CONFIG_INET) 5284 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5285 u32 flags, bool check_mtu) 5286 { 5287 struct fib_nh_common *nhc; 5288 struct in_device *in_dev; 5289 struct neighbour *neigh; 5290 struct net_device *dev; 5291 struct fib_result res; 5292 struct flowi4 fl4; 5293 int err; 5294 u32 mtu; 5295 5296 dev = dev_get_by_index_rcu(net, params->ifindex); 5297 if (unlikely(!dev)) 5298 return -ENODEV; 5299 5300 /* verify forwarding is enabled on this interface */ 5301 in_dev = __in_dev_get_rcu(dev); 5302 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev))) 5303 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5304 5305 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5306 fl4.flowi4_iif = 1; 5307 fl4.flowi4_oif = params->ifindex; 5308 } else { 5309 fl4.flowi4_iif = params->ifindex; 5310 fl4.flowi4_oif = 0; 5311 } 5312 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK; 5313 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 5314 fl4.flowi4_flags = 0; 5315 5316 fl4.flowi4_proto = params->l4_protocol; 5317 fl4.daddr = params->ipv4_dst; 5318 fl4.saddr = params->ipv4_src; 5319 fl4.fl4_sport = params->sport; 5320 fl4.fl4_dport = params->dport; 5321 fl4.flowi4_multipath_hash = 0; 5322 5323 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5324 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5325 struct fib_table *tb; 5326 5327 tb = fib_get_table(net, tbid); 5328 if (unlikely(!tb)) 5329 return BPF_FIB_LKUP_RET_NOT_FWDED; 5330 5331 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 5332 } else { 5333 fl4.flowi4_mark = 0; 5334 fl4.flowi4_secid = 0; 5335 fl4.flowi4_tun_key.tun_id = 0; 5336 fl4.flowi4_uid = sock_net_uid(net, NULL); 5337 5338 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF); 5339 } 5340 5341 if (err) { 5342 /* map fib lookup errors to RTN_ type */ 5343 if (err == -EINVAL) 5344 return BPF_FIB_LKUP_RET_BLACKHOLE; 5345 if (err == -EHOSTUNREACH) 5346 return BPF_FIB_LKUP_RET_UNREACHABLE; 5347 if (err == -EACCES) 5348 return BPF_FIB_LKUP_RET_PROHIBIT; 5349 5350 return BPF_FIB_LKUP_RET_NOT_FWDED; 5351 } 5352 5353 if (res.type != RTN_UNICAST) 5354 return BPF_FIB_LKUP_RET_NOT_FWDED; 5355 5356 if (fib_info_num_path(res.fi) > 1) 5357 fib_select_path(net, &res, &fl4, NULL); 5358 5359 if (check_mtu) { 5360 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst); 5361 if (params->tot_len > mtu) 5362 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5363 } 5364 5365 nhc = res.nhc; 5366 5367 /* do not handle lwt encaps right now */ 5368 if (nhc->nhc_lwtstate) 5369 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5370 5371 dev = nhc->nhc_dev; 5372 5373 params->rt_metric = res.fi->fib_priority; 5374 params->ifindex = dev->ifindex; 5375 5376 /* xdp and cls_bpf programs are run in RCU-bh so 5377 * rcu_read_lock_bh is not needed here 5378 */ 5379 if (likely(nhc->nhc_gw_family != AF_INET6)) { 5380 if (nhc->nhc_gw_family) 5381 params->ipv4_dst = nhc->nhc_gw.ipv4; 5382 5383 neigh = __ipv4_neigh_lookup_noref(dev, 5384 (__force u32)params->ipv4_dst); 5385 } else { 5386 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst; 5387 5388 params->family = AF_INET6; 5389 *dst = nhc->nhc_gw.ipv6; 5390 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5391 } 5392 5393 if (!neigh) 5394 return BPF_FIB_LKUP_RET_NO_NEIGH; 5395 5396 return bpf_fib_set_fwd_params(params, neigh, dev); 5397 } 5398 #endif 5399 5400 #if IS_ENABLED(CONFIG_IPV6) 5401 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5402 u32 flags, bool check_mtu) 5403 { 5404 struct in6_addr *src = (struct in6_addr *) params->ipv6_src; 5405 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst; 5406 struct fib6_result res = {}; 5407 struct neighbour *neigh; 5408 struct net_device *dev; 5409 struct inet6_dev *idev; 5410 struct flowi6 fl6; 5411 int strict = 0; 5412 int oif, err; 5413 u32 mtu; 5414 5415 /* link local addresses are never forwarded */ 5416 if (rt6_need_strict(dst) || rt6_need_strict(src)) 5417 return BPF_FIB_LKUP_RET_NOT_FWDED; 5418 5419 dev = dev_get_by_index_rcu(net, params->ifindex); 5420 if (unlikely(!dev)) 5421 return -ENODEV; 5422 5423 idev = __in6_dev_get_safely(dev); 5424 if (unlikely(!idev || !idev->cnf.forwarding)) 5425 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5426 5427 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5428 fl6.flowi6_iif = 1; 5429 oif = fl6.flowi6_oif = params->ifindex; 5430 } else { 5431 oif = fl6.flowi6_iif = params->ifindex; 5432 fl6.flowi6_oif = 0; 5433 strict = RT6_LOOKUP_F_HAS_SADDR; 5434 } 5435 fl6.flowlabel = params->flowinfo; 5436 fl6.flowi6_scope = 0; 5437 fl6.flowi6_flags = 0; 5438 fl6.mp_hash = 0; 5439 5440 fl6.flowi6_proto = params->l4_protocol; 5441 fl6.daddr = *dst; 5442 fl6.saddr = *src; 5443 fl6.fl6_sport = params->sport; 5444 fl6.fl6_dport = params->dport; 5445 5446 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5447 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5448 struct fib6_table *tb; 5449 5450 tb = ipv6_stub->fib6_get_table(net, tbid); 5451 if (unlikely(!tb)) 5452 return BPF_FIB_LKUP_RET_NOT_FWDED; 5453 5454 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res, 5455 strict); 5456 } else { 5457 fl6.flowi6_mark = 0; 5458 fl6.flowi6_secid = 0; 5459 fl6.flowi6_tun_key.tun_id = 0; 5460 fl6.flowi6_uid = sock_net_uid(net, NULL); 5461 5462 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict); 5463 } 5464 5465 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) || 5466 res.f6i == net->ipv6.fib6_null_entry)) 5467 return BPF_FIB_LKUP_RET_NOT_FWDED; 5468 5469 switch (res.fib6_type) { 5470 /* only unicast is forwarded */ 5471 case RTN_UNICAST: 5472 break; 5473 case RTN_BLACKHOLE: 5474 return BPF_FIB_LKUP_RET_BLACKHOLE; 5475 case RTN_UNREACHABLE: 5476 return BPF_FIB_LKUP_RET_UNREACHABLE; 5477 case RTN_PROHIBIT: 5478 return BPF_FIB_LKUP_RET_PROHIBIT; 5479 default: 5480 return BPF_FIB_LKUP_RET_NOT_FWDED; 5481 } 5482 5483 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif, 5484 fl6.flowi6_oif != 0, NULL, strict); 5485 5486 if (check_mtu) { 5487 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src); 5488 if (params->tot_len > mtu) 5489 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5490 } 5491 5492 if (res.nh->fib_nh_lws) 5493 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5494 5495 if (res.nh->fib_nh_gw_family) 5496 *dst = res.nh->fib_nh_gw6; 5497 5498 dev = res.nh->fib_nh_dev; 5499 params->rt_metric = res.f6i->fib6_metric; 5500 params->ifindex = dev->ifindex; 5501 5502 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is 5503 * not needed here. 5504 */ 5505 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5506 if (!neigh) 5507 return BPF_FIB_LKUP_RET_NO_NEIGH; 5508 5509 return bpf_fib_set_fwd_params(params, neigh, dev); 5510 } 5511 #endif 5512 5513 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx, 5514 struct bpf_fib_lookup *, params, int, plen, u32, flags) 5515 { 5516 if (plen < sizeof(*params)) 5517 return -EINVAL; 5518 5519 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 5520 return -EINVAL; 5521 5522 switch (params->family) { 5523 #if IS_ENABLED(CONFIG_INET) 5524 case AF_INET: 5525 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params, 5526 flags, true); 5527 #endif 5528 #if IS_ENABLED(CONFIG_IPV6) 5529 case AF_INET6: 5530 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params, 5531 flags, true); 5532 #endif 5533 } 5534 return -EAFNOSUPPORT; 5535 } 5536 5537 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = { 5538 .func = bpf_xdp_fib_lookup, 5539 .gpl_only = true, 5540 .ret_type = RET_INTEGER, 5541 .arg1_type = ARG_PTR_TO_CTX, 5542 .arg2_type = ARG_PTR_TO_MEM, 5543 .arg3_type = ARG_CONST_SIZE, 5544 .arg4_type = ARG_ANYTHING, 5545 }; 5546 5547 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb, 5548 struct bpf_fib_lookup *, params, int, plen, u32, flags) 5549 { 5550 struct net *net = dev_net(skb->dev); 5551 int rc = -EAFNOSUPPORT; 5552 5553 if (plen < sizeof(*params)) 5554 return -EINVAL; 5555 5556 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 5557 return -EINVAL; 5558 5559 switch (params->family) { 5560 #if IS_ENABLED(CONFIG_INET) 5561 case AF_INET: 5562 rc = bpf_ipv4_fib_lookup(net, params, flags, false); 5563 break; 5564 #endif 5565 #if IS_ENABLED(CONFIG_IPV6) 5566 case AF_INET6: 5567 rc = bpf_ipv6_fib_lookup(net, params, flags, false); 5568 break; 5569 #endif 5570 } 5571 5572 if (!rc) { 5573 struct net_device *dev; 5574 5575 dev = dev_get_by_index_rcu(net, params->ifindex); 5576 if (!is_skb_forwardable(dev, skb)) 5577 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED; 5578 } 5579 5580 return rc; 5581 } 5582 5583 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = { 5584 .func = bpf_skb_fib_lookup, 5585 .gpl_only = true, 5586 .ret_type = RET_INTEGER, 5587 .arg1_type = ARG_PTR_TO_CTX, 5588 .arg2_type = ARG_PTR_TO_MEM, 5589 .arg3_type = ARG_CONST_SIZE, 5590 .arg4_type = ARG_ANYTHING, 5591 }; 5592 5593 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5594 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len) 5595 { 5596 int err; 5597 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr; 5598 5599 if (!seg6_validate_srh(srh, len, false)) 5600 return -EINVAL; 5601 5602 switch (type) { 5603 case BPF_LWT_ENCAP_SEG6_INLINE: 5604 if (skb->protocol != htons(ETH_P_IPV6)) 5605 return -EBADMSG; 5606 5607 err = seg6_do_srh_inline(skb, srh); 5608 break; 5609 case BPF_LWT_ENCAP_SEG6: 5610 skb_reset_inner_headers(skb); 5611 skb->encapsulation = 1; 5612 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6); 5613 break; 5614 default: 5615 return -EINVAL; 5616 } 5617 5618 bpf_compute_data_pointers(skb); 5619 if (err) 5620 return err; 5621 5622 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 5623 skb_set_transport_header(skb, sizeof(struct ipv6hdr)); 5624 5625 return seg6_lookup_nexthop(skb, NULL, 0); 5626 } 5627 #endif /* CONFIG_IPV6_SEG6_BPF */ 5628 5629 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 5630 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, 5631 bool ingress) 5632 { 5633 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress); 5634 } 5635 #endif 5636 5637 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr, 5638 u32, len) 5639 { 5640 switch (type) { 5641 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5642 case BPF_LWT_ENCAP_SEG6: 5643 case BPF_LWT_ENCAP_SEG6_INLINE: 5644 return bpf_push_seg6_encap(skb, type, hdr, len); 5645 #endif 5646 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 5647 case BPF_LWT_ENCAP_IP: 5648 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */); 5649 #endif 5650 default: 5651 return -EINVAL; 5652 } 5653 } 5654 5655 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type, 5656 void *, hdr, u32, len) 5657 { 5658 switch (type) { 5659 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 5660 case BPF_LWT_ENCAP_IP: 5661 return bpf_push_ip_encap(skb, hdr, len, false /* egress */); 5662 #endif 5663 default: 5664 return -EINVAL; 5665 } 5666 } 5667 5668 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = { 5669 .func = bpf_lwt_in_push_encap, 5670 .gpl_only = false, 5671 .ret_type = RET_INTEGER, 5672 .arg1_type = ARG_PTR_TO_CTX, 5673 .arg2_type = ARG_ANYTHING, 5674 .arg3_type = ARG_PTR_TO_MEM, 5675 .arg4_type = ARG_CONST_SIZE 5676 }; 5677 5678 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = { 5679 .func = bpf_lwt_xmit_push_encap, 5680 .gpl_only = false, 5681 .ret_type = RET_INTEGER, 5682 .arg1_type = ARG_PTR_TO_CTX, 5683 .arg2_type = ARG_ANYTHING, 5684 .arg3_type = ARG_PTR_TO_MEM, 5685 .arg4_type = ARG_CONST_SIZE 5686 }; 5687 5688 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5689 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset, 5690 const void *, from, u32, len) 5691 { 5692 struct seg6_bpf_srh_state *srh_state = 5693 this_cpu_ptr(&seg6_bpf_srh_states); 5694 struct ipv6_sr_hdr *srh = srh_state->srh; 5695 void *srh_tlvs, *srh_end, *ptr; 5696 int srhoff = 0; 5697 5698 if (srh == NULL) 5699 return -EINVAL; 5700 5701 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4)); 5702 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen); 5703 5704 ptr = skb->data + offset; 5705 if (ptr >= srh_tlvs && ptr + len <= srh_end) 5706 srh_state->valid = false; 5707 else if (ptr < (void *)&srh->flags || 5708 ptr + len > (void *)&srh->segments) 5709 return -EFAULT; 5710 5711 if (unlikely(bpf_try_make_writable(skb, offset + len))) 5712 return -EFAULT; 5713 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 5714 return -EINVAL; 5715 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5716 5717 memcpy(skb->data + offset, from, len); 5718 return 0; 5719 } 5720 5721 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = { 5722 .func = bpf_lwt_seg6_store_bytes, 5723 .gpl_only = false, 5724 .ret_type = RET_INTEGER, 5725 .arg1_type = ARG_PTR_TO_CTX, 5726 .arg2_type = ARG_ANYTHING, 5727 .arg3_type = ARG_PTR_TO_MEM, 5728 .arg4_type = ARG_CONST_SIZE 5729 }; 5730 5731 static void bpf_update_srh_state(struct sk_buff *skb) 5732 { 5733 struct seg6_bpf_srh_state *srh_state = 5734 this_cpu_ptr(&seg6_bpf_srh_states); 5735 int srhoff = 0; 5736 5737 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) { 5738 srh_state->srh = NULL; 5739 } else { 5740 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5741 srh_state->hdrlen = srh_state->srh->hdrlen << 3; 5742 srh_state->valid = true; 5743 } 5744 } 5745 5746 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb, 5747 u32, action, void *, param, u32, param_len) 5748 { 5749 struct seg6_bpf_srh_state *srh_state = 5750 this_cpu_ptr(&seg6_bpf_srh_states); 5751 int hdroff = 0; 5752 int err; 5753 5754 switch (action) { 5755 case SEG6_LOCAL_ACTION_END_X: 5756 if (!seg6_bpf_has_valid_srh(skb)) 5757 return -EBADMSG; 5758 if (param_len != sizeof(struct in6_addr)) 5759 return -EINVAL; 5760 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0); 5761 case SEG6_LOCAL_ACTION_END_T: 5762 if (!seg6_bpf_has_valid_srh(skb)) 5763 return -EBADMSG; 5764 if (param_len != sizeof(int)) 5765 return -EINVAL; 5766 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 5767 case SEG6_LOCAL_ACTION_END_DT6: 5768 if (!seg6_bpf_has_valid_srh(skb)) 5769 return -EBADMSG; 5770 if (param_len != sizeof(int)) 5771 return -EINVAL; 5772 5773 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0) 5774 return -EBADMSG; 5775 if (!pskb_pull(skb, hdroff)) 5776 return -EBADMSG; 5777 5778 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff); 5779 skb_reset_network_header(skb); 5780 skb_reset_transport_header(skb); 5781 skb->encapsulation = 0; 5782 5783 bpf_compute_data_pointers(skb); 5784 bpf_update_srh_state(skb); 5785 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 5786 case SEG6_LOCAL_ACTION_END_B6: 5787 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 5788 return -EBADMSG; 5789 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE, 5790 param, param_len); 5791 if (!err) 5792 bpf_update_srh_state(skb); 5793 5794 return err; 5795 case SEG6_LOCAL_ACTION_END_B6_ENCAP: 5796 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 5797 return -EBADMSG; 5798 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6, 5799 param, param_len); 5800 if (!err) 5801 bpf_update_srh_state(skb); 5802 5803 return err; 5804 default: 5805 return -EINVAL; 5806 } 5807 } 5808 5809 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = { 5810 .func = bpf_lwt_seg6_action, 5811 .gpl_only = false, 5812 .ret_type = RET_INTEGER, 5813 .arg1_type = ARG_PTR_TO_CTX, 5814 .arg2_type = ARG_ANYTHING, 5815 .arg3_type = ARG_PTR_TO_MEM, 5816 .arg4_type = ARG_CONST_SIZE 5817 }; 5818 5819 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset, 5820 s32, len) 5821 { 5822 struct seg6_bpf_srh_state *srh_state = 5823 this_cpu_ptr(&seg6_bpf_srh_states); 5824 struct ipv6_sr_hdr *srh = srh_state->srh; 5825 void *srh_end, *srh_tlvs, *ptr; 5826 struct ipv6hdr *hdr; 5827 int srhoff = 0; 5828 int ret; 5829 5830 if (unlikely(srh == NULL)) 5831 return -EINVAL; 5832 5833 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) + 5834 ((srh->first_segment + 1) << 4)); 5835 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) + 5836 srh_state->hdrlen); 5837 ptr = skb->data + offset; 5838 5839 if (unlikely(ptr < srh_tlvs || ptr > srh_end)) 5840 return -EFAULT; 5841 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end)) 5842 return -EFAULT; 5843 5844 if (len > 0) { 5845 ret = skb_cow_head(skb, len); 5846 if (unlikely(ret < 0)) 5847 return ret; 5848 5849 ret = bpf_skb_net_hdr_push(skb, offset, len); 5850 } else { 5851 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len); 5852 } 5853 5854 bpf_compute_data_pointers(skb); 5855 if (unlikely(ret < 0)) 5856 return ret; 5857 5858 hdr = (struct ipv6hdr *)skb->data; 5859 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 5860 5861 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 5862 return -EINVAL; 5863 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5864 srh_state->hdrlen += len; 5865 srh_state->valid = false; 5866 return 0; 5867 } 5868 5869 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = { 5870 .func = bpf_lwt_seg6_adjust_srh, 5871 .gpl_only = false, 5872 .ret_type = RET_INTEGER, 5873 .arg1_type = ARG_PTR_TO_CTX, 5874 .arg2_type = ARG_ANYTHING, 5875 .arg3_type = ARG_ANYTHING, 5876 }; 5877 #endif /* CONFIG_IPV6_SEG6_BPF */ 5878 5879 #ifdef CONFIG_INET 5880 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple, 5881 int dif, int sdif, u8 family, u8 proto) 5882 { 5883 bool refcounted = false; 5884 struct sock *sk = NULL; 5885 5886 if (family == AF_INET) { 5887 __be32 src4 = tuple->ipv4.saddr; 5888 __be32 dst4 = tuple->ipv4.daddr; 5889 5890 if (proto == IPPROTO_TCP) 5891 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0, 5892 src4, tuple->ipv4.sport, 5893 dst4, tuple->ipv4.dport, 5894 dif, sdif, &refcounted); 5895 else 5896 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport, 5897 dst4, tuple->ipv4.dport, 5898 dif, sdif, &udp_table, NULL); 5899 #if IS_ENABLED(CONFIG_IPV6) 5900 } else { 5901 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr; 5902 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr; 5903 5904 if (proto == IPPROTO_TCP) 5905 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0, 5906 src6, tuple->ipv6.sport, 5907 dst6, ntohs(tuple->ipv6.dport), 5908 dif, sdif, &refcounted); 5909 else if (likely(ipv6_bpf_stub)) 5910 sk = ipv6_bpf_stub->udp6_lib_lookup(net, 5911 src6, tuple->ipv6.sport, 5912 dst6, tuple->ipv6.dport, 5913 dif, sdif, 5914 &udp_table, NULL); 5915 #endif 5916 } 5917 5918 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) { 5919 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 5920 sk = NULL; 5921 } 5922 return sk; 5923 } 5924 5925 /* bpf_skc_lookup performs the core lookup for different types of sockets, 5926 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE. 5927 * Returns the socket as an 'unsigned long' to simplify the casting in the 5928 * callers to satisfy BPF_CALL declarations. 5929 */ 5930 static struct sock * 5931 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5932 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 5933 u64 flags) 5934 { 5935 struct sock *sk = NULL; 5936 u8 family = AF_UNSPEC; 5937 struct net *net; 5938 int sdif; 5939 5940 if (len == sizeof(tuple->ipv4)) 5941 family = AF_INET; 5942 else if (len == sizeof(tuple->ipv6)) 5943 family = AF_INET6; 5944 else 5945 return NULL; 5946 5947 if (unlikely(family == AF_UNSPEC || flags || 5948 !((s32)netns_id < 0 || netns_id <= S32_MAX))) 5949 goto out; 5950 5951 if (family == AF_INET) 5952 sdif = inet_sdif(skb); 5953 else 5954 sdif = inet6_sdif(skb); 5955 5956 if ((s32)netns_id < 0) { 5957 net = caller_net; 5958 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 5959 } else { 5960 net = get_net_ns_by_id(caller_net, netns_id); 5961 if (unlikely(!net)) 5962 goto out; 5963 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 5964 put_net(net); 5965 } 5966 5967 out: 5968 return sk; 5969 } 5970 5971 static struct sock * 5972 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5973 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 5974 u64 flags) 5975 { 5976 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net, 5977 ifindex, proto, netns_id, flags); 5978 5979 if (sk) { 5980 sk = sk_to_full_sk(sk); 5981 if (!sk_fullsock(sk)) { 5982 sock_gen_put(sk); 5983 return NULL; 5984 } 5985 } 5986 5987 return sk; 5988 } 5989 5990 static struct sock * 5991 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5992 u8 proto, u64 netns_id, u64 flags) 5993 { 5994 struct net *caller_net; 5995 int ifindex; 5996 5997 if (skb->dev) { 5998 caller_net = dev_net(skb->dev); 5999 ifindex = skb->dev->ifindex; 6000 } else { 6001 caller_net = sock_net(skb->sk); 6002 ifindex = 0; 6003 } 6004 6005 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto, 6006 netns_id, flags); 6007 } 6008 6009 static struct sock * 6010 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6011 u8 proto, u64 netns_id, u64 flags) 6012 { 6013 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id, 6014 flags); 6015 6016 if (sk) { 6017 sk = sk_to_full_sk(sk); 6018 if (!sk_fullsock(sk)) { 6019 sock_gen_put(sk); 6020 return NULL; 6021 } 6022 } 6023 6024 return sk; 6025 } 6026 6027 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb, 6028 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6029 { 6030 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP, 6031 netns_id, flags); 6032 } 6033 6034 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = { 6035 .func = bpf_skc_lookup_tcp, 6036 .gpl_only = false, 6037 .pkt_access = true, 6038 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6039 .arg1_type = ARG_PTR_TO_CTX, 6040 .arg2_type = ARG_PTR_TO_MEM, 6041 .arg3_type = ARG_CONST_SIZE, 6042 .arg4_type = ARG_ANYTHING, 6043 .arg5_type = ARG_ANYTHING, 6044 }; 6045 6046 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb, 6047 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6048 { 6049 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, 6050 netns_id, flags); 6051 } 6052 6053 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = { 6054 .func = bpf_sk_lookup_tcp, 6055 .gpl_only = false, 6056 .pkt_access = true, 6057 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6058 .arg1_type = ARG_PTR_TO_CTX, 6059 .arg2_type = ARG_PTR_TO_MEM, 6060 .arg3_type = ARG_CONST_SIZE, 6061 .arg4_type = ARG_ANYTHING, 6062 .arg5_type = ARG_ANYTHING, 6063 }; 6064 6065 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb, 6066 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6067 { 6068 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, 6069 netns_id, flags); 6070 } 6071 6072 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = { 6073 .func = bpf_sk_lookup_udp, 6074 .gpl_only = false, 6075 .pkt_access = true, 6076 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6077 .arg1_type = ARG_PTR_TO_CTX, 6078 .arg2_type = ARG_PTR_TO_MEM, 6079 .arg3_type = ARG_CONST_SIZE, 6080 .arg4_type = ARG_ANYTHING, 6081 .arg5_type = ARG_ANYTHING, 6082 }; 6083 6084 BPF_CALL_1(bpf_sk_release, struct sock *, sk) 6085 { 6086 if (sk && sk_is_refcounted(sk)) 6087 sock_gen_put(sk); 6088 return 0; 6089 } 6090 6091 static const struct bpf_func_proto bpf_sk_release_proto = { 6092 .func = bpf_sk_release, 6093 .gpl_only = false, 6094 .ret_type = RET_INTEGER, 6095 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6096 }; 6097 6098 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx, 6099 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6100 { 6101 struct net *caller_net = dev_net(ctx->rxq->dev); 6102 int ifindex = ctx->rxq->dev->ifindex; 6103 6104 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6105 ifindex, IPPROTO_UDP, netns_id, 6106 flags); 6107 } 6108 6109 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = { 6110 .func = bpf_xdp_sk_lookup_udp, 6111 .gpl_only = false, 6112 .pkt_access = true, 6113 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6114 .arg1_type = ARG_PTR_TO_CTX, 6115 .arg2_type = ARG_PTR_TO_MEM, 6116 .arg3_type = ARG_CONST_SIZE, 6117 .arg4_type = ARG_ANYTHING, 6118 .arg5_type = ARG_ANYTHING, 6119 }; 6120 6121 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx, 6122 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6123 { 6124 struct net *caller_net = dev_net(ctx->rxq->dev); 6125 int ifindex = ctx->rxq->dev->ifindex; 6126 6127 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net, 6128 ifindex, IPPROTO_TCP, netns_id, 6129 flags); 6130 } 6131 6132 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = { 6133 .func = bpf_xdp_skc_lookup_tcp, 6134 .gpl_only = false, 6135 .pkt_access = true, 6136 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6137 .arg1_type = ARG_PTR_TO_CTX, 6138 .arg2_type = ARG_PTR_TO_MEM, 6139 .arg3_type = ARG_CONST_SIZE, 6140 .arg4_type = ARG_ANYTHING, 6141 .arg5_type = ARG_ANYTHING, 6142 }; 6143 6144 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx, 6145 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6146 { 6147 struct net *caller_net = dev_net(ctx->rxq->dev); 6148 int ifindex = ctx->rxq->dev->ifindex; 6149 6150 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6151 ifindex, IPPROTO_TCP, netns_id, 6152 flags); 6153 } 6154 6155 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = { 6156 .func = bpf_xdp_sk_lookup_tcp, 6157 .gpl_only = false, 6158 .pkt_access = true, 6159 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6160 .arg1_type = ARG_PTR_TO_CTX, 6161 .arg2_type = ARG_PTR_TO_MEM, 6162 .arg3_type = ARG_CONST_SIZE, 6163 .arg4_type = ARG_ANYTHING, 6164 .arg5_type = ARG_ANYTHING, 6165 }; 6166 6167 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6168 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6169 { 6170 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, 6171 sock_net(ctx->sk), 0, 6172 IPPROTO_TCP, netns_id, flags); 6173 } 6174 6175 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = { 6176 .func = bpf_sock_addr_skc_lookup_tcp, 6177 .gpl_only = false, 6178 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6179 .arg1_type = ARG_PTR_TO_CTX, 6180 .arg2_type = ARG_PTR_TO_MEM, 6181 .arg3_type = ARG_CONST_SIZE, 6182 .arg4_type = ARG_ANYTHING, 6183 .arg5_type = ARG_ANYTHING, 6184 }; 6185 6186 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6187 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6188 { 6189 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6190 sock_net(ctx->sk), 0, IPPROTO_TCP, 6191 netns_id, flags); 6192 } 6193 6194 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = { 6195 .func = bpf_sock_addr_sk_lookup_tcp, 6196 .gpl_only = false, 6197 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6198 .arg1_type = ARG_PTR_TO_CTX, 6199 .arg2_type = ARG_PTR_TO_MEM, 6200 .arg3_type = ARG_CONST_SIZE, 6201 .arg4_type = ARG_ANYTHING, 6202 .arg5_type = ARG_ANYTHING, 6203 }; 6204 6205 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx, 6206 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6207 { 6208 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6209 sock_net(ctx->sk), 0, IPPROTO_UDP, 6210 netns_id, flags); 6211 } 6212 6213 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = { 6214 .func = bpf_sock_addr_sk_lookup_udp, 6215 .gpl_only = false, 6216 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6217 .arg1_type = ARG_PTR_TO_CTX, 6218 .arg2_type = ARG_PTR_TO_MEM, 6219 .arg3_type = ARG_CONST_SIZE, 6220 .arg4_type = ARG_ANYTHING, 6221 .arg5_type = ARG_ANYTHING, 6222 }; 6223 6224 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6225 struct bpf_insn_access_aux *info) 6226 { 6227 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, 6228 icsk_retransmits)) 6229 return false; 6230 6231 if (off % size != 0) 6232 return false; 6233 6234 switch (off) { 6235 case offsetof(struct bpf_tcp_sock, bytes_received): 6236 case offsetof(struct bpf_tcp_sock, bytes_acked): 6237 return size == sizeof(__u64); 6238 default: 6239 return size == sizeof(__u32); 6240 } 6241 } 6242 6243 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 6244 const struct bpf_insn *si, 6245 struct bpf_insn *insn_buf, 6246 struct bpf_prog *prog, u32 *target_size) 6247 { 6248 struct bpf_insn *insn = insn_buf; 6249 6250 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \ 6251 do { \ 6252 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \ 6253 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6254 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\ 6255 si->dst_reg, si->src_reg, \ 6256 offsetof(struct tcp_sock, FIELD)); \ 6257 } while (0) 6258 6259 #define BPF_INET_SOCK_GET_COMMON(FIELD) \ 6260 do { \ 6261 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \ 6262 FIELD) > \ 6263 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6264 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 6265 struct inet_connection_sock, \ 6266 FIELD), \ 6267 si->dst_reg, si->src_reg, \ 6268 offsetof( \ 6269 struct inet_connection_sock, \ 6270 FIELD)); \ 6271 } while (0) 6272 6273 if (insn > insn_buf) 6274 return insn - insn_buf; 6275 6276 switch (si->off) { 6277 case offsetof(struct bpf_tcp_sock, rtt_min): 6278 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 6279 sizeof(struct minmax)); 6280 BUILD_BUG_ON(sizeof(struct minmax) < 6281 sizeof(struct minmax_sample)); 6282 6283 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 6284 offsetof(struct tcp_sock, rtt_min) + 6285 offsetof(struct minmax_sample, v)); 6286 break; 6287 case offsetof(struct bpf_tcp_sock, snd_cwnd): 6288 BPF_TCP_SOCK_GET_COMMON(snd_cwnd); 6289 break; 6290 case offsetof(struct bpf_tcp_sock, srtt_us): 6291 BPF_TCP_SOCK_GET_COMMON(srtt_us); 6292 break; 6293 case offsetof(struct bpf_tcp_sock, snd_ssthresh): 6294 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh); 6295 break; 6296 case offsetof(struct bpf_tcp_sock, rcv_nxt): 6297 BPF_TCP_SOCK_GET_COMMON(rcv_nxt); 6298 break; 6299 case offsetof(struct bpf_tcp_sock, snd_nxt): 6300 BPF_TCP_SOCK_GET_COMMON(snd_nxt); 6301 break; 6302 case offsetof(struct bpf_tcp_sock, snd_una): 6303 BPF_TCP_SOCK_GET_COMMON(snd_una); 6304 break; 6305 case offsetof(struct bpf_tcp_sock, mss_cache): 6306 BPF_TCP_SOCK_GET_COMMON(mss_cache); 6307 break; 6308 case offsetof(struct bpf_tcp_sock, ecn_flags): 6309 BPF_TCP_SOCK_GET_COMMON(ecn_flags); 6310 break; 6311 case offsetof(struct bpf_tcp_sock, rate_delivered): 6312 BPF_TCP_SOCK_GET_COMMON(rate_delivered); 6313 break; 6314 case offsetof(struct bpf_tcp_sock, rate_interval_us): 6315 BPF_TCP_SOCK_GET_COMMON(rate_interval_us); 6316 break; 6317 case offsetof(struct bpf_tcp_sock, packets_out): 6318 BPF_TCP_SOCK_GET_COMMON(packets_out); 6319 break; 6320 case offsetof(struct bpf_tcp_sock, retrans_out): 6321 BPF_TCP_SOCK_GET_COMMON(retrans_out); 6322 break; 6323 case offsetof(struct bpf_tcp_sock, total_retrans): 6324 BPF_TCP_SOCK_GET_COMMON(total_retrans); 6325 break; 6326 case offsetof(struct bpf_tcp_sock, segs_in): 6327 BPF_TCP_SOCK_GET_COMMON(segs_in); 6328 break; 6329 case offsetof(struct bpf_tcp_sock, data_segs_in): 6330 BPF_TCP_SOCK_GET_COMMON(data_segs_in); 6331 break; 6332 case offsetof(struct bpf_tcp_sock, segs_out): 6333 BPF_TCP_SOCK_GET_COMMON(segs_out); 6334 break; 6335 case offsetof(struct bpf_tcp_sock, data_segs_out): 6336 BPF_TCP_SOCK_GET_COMMON(data_segs_out); 6337 break; 6338 case offsetof(struct bpf_tcp_sock, lost_out): 6339 BPF_TCP_SOCK_GET_COMMON(lost_out); 6340 break; 6341 case offsetof(struct bpf_tcp_sock, sacked_out): 6342 BPF_TCP_SOCK_GET_COMMON(sacked_out); 6343 break; 6344 case offsetof(struct bpf_tcp_sock, bytes_received): 6345 BPF_TCP_SOCK_GET_COMMON(bytes_received); 6346 break; 6347 case offsetof(struct bpf_tcp_sock, bytes_acked): 6348 BPF_TCP_SOCK_GET_COMMON(bytes_acked); 6349 break; 6350 case offsetof(struct bpf_tcp_sock, dsack_dups): 6351 BPF_TCP_SOCK_GET_COMMON(dsack_dups); 6352 break; 6353 case offsetof(struct bpf_tcp_sock, delivered): 6354 BPF_TCP_SOCK_GET_COMMON(delivered); 6355 break; 6356 case offsetof(struct bpf_tcp_sock, delivered_ce): 6357 BPF_TCP_SOCK_GET_COMMON(delivered_ce); 6358 break; 6359 case offsetof(struct bpf_tcp_sock, icsk_retransmits): 6360 BPF_INET_SOCK_GET_COMMON(icsk_retransmits); 6361 break; 6362 } 6363 6364 return insn - insn_buf; 6365 } 6366 6367 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk) 6368 { 6369 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 6370 return (unsigned long)sk; 6371 6372 return (unsigned long)NULL; 6373 } 6374 6375 const struct bpf_func_proto bpf_tcp_sock_proto = { 6376 .func = bpf_tcp_sock, 6377 .gpl_only = false, 6378 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL, 6379 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 6380 }; 6381 6382 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk) 6383 { 6384 sk = sk_to_full_sk(sk); 6385 6386 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE)) 6387 return (unsigned long)sk; 6388 6389 return (unsigned long)NULL; 6390 } 6391 6392 static const struct bpf_func_proto bpf_get_listener_sock_proto = { 6393 .func = bpf_get_listener_sock, 6394 .gpl_only = false, 6395 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6396 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 6397 }; 6398 6399 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb) 6400 { 6401 unsigned int iphdr_len; 6402 6403 switch (skb_protocol(skb, true)) { 6404 case cpu_to_be16(ETH_P_IP): 6405 iphdr_len = sizeof(struct iphdr); 6406 break; 6407 case cpu_to_be16(ETH_P_IPV6): 6408 iphdr_len = sizeof(struct ipv6hdr); 6409 break; 6410 default: 6411 return 0; 6412 } 6413 6414 if (skb_headlen(skb) < iphdr_len) 6415 return 0; 6416 6417 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len)) 6418 return 0; 6419 6420 return INET_ECN_set_ce(skb); 6421 } 6422 6423 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6424 struct bpf_insn_access_aux *info) 6425 { 6426 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id)) 6427 return false; 6428 6429 if (off % size != 0) 6430 return false; 6431 6432 switch (off) { 6433 default: 6434 return size == sizeof(__u32); 6435 } 6436 } 6437 6438 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 6439 const struct bpf_insn *si, 6440 struct bpf_insn *insn_buf, 6441 struct bpf_prog *prog, u32 *target_size) 6442 { 6443 struct bpf_insn *insn = insn_buf; 6444 6445 #define BPF_XDP_SOCK_GET(FIELD) \ 6446 do { \ 6447 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \ 6448 sizeof_field(struct bpf_xdp_sock, FIELD)); \ 6449 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\ 6450 si->dst_reg, si->src_reg, \ 6451 offsetof(struct xdp_sock, FIELD)); \ 6452 } while (0) 6453 6454 switch (si->off) { 6455 case offsetof(struct bpf_xdp_sock, queue_id): 6456 BPF_XDP_SOCK_GET(queue_id); 6457 break; 6458 } 6459 6460 return insn - insn_buf; 6461 } 6462 6463 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = { 6464 .func = bpf_skb_ecn_set_ce, 6465 .gpl_only = false, 6466 .ret_type = RET_INTEGER, 6467 .arg1_type = ARG_PTR_TO_CTX, 6468 }; 6469 6470 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 6471 struct tcphdr *, th, u32, th_len) 6472 { 6473 #ifdef CONFIG_SYN_COOKIES 6474 u32 cookie; 6475 int ret; 6476 6477 if (unlikely(!sk || th_len < sizeof(*th))) 6478 return -EINVAL; 6479 6480 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */ 6481 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 6482 return -EINVAL; 6483 6484 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies) 6485 return -EINVAL; 6486 6487 if (!th->ack || th->rst || th->syn) 6488 return -ENOENT; 6489 6490 if (tcp_synq_no_recent_overflow(sk)) 6491 return -ENOENT; 6492 6493 cookie = ntohl(th->ack_seq) - 1; 6494 6495 switch (sk->sk_family) { 6496 case AF_INET: 6497 if (unlikely(iph_len < sizeof(struct iphdr))) 6498 return -EINVAL; 6499 6500 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie); 6501 break; 6502 6503 #if IS_BUILTIN(CONFIG_IPV6) 6504 case AF_INET6: 6505 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 6506 return -EINVAL; 6507 6508 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie); 6509 break; 6510 #endif /* CONFIG_IPV6 */ 6511 6512 default: 6513 return -EPROTONOSUPPORT; 6514 } 6515 6516 if (ret > 0) 6517 return 0; 6518 6519 return -ENOENT; 6520 #else 6521 return -ENOTSUPP; 6522 #endif 6523 } 6524 6525 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = { 6526 .func = bpf_tcp_check_syncookie, 6527 .gpl_only = true, 6528 .pkt_access = true, 6529 .ret_type = RET_INTEGER, 6530 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6531 .arg2_type = ARG_PTR_TO_MEM, 6532 .arg3_type = ARG_CONST_SIZE, 6533 .arg4_type = ARG_PTR_TO_MEM, 6534 .arg5_type = ARG_CONST_SIZE, 6535 }; 6536 6537 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 6538 struct tcphdr *, th, u32, th_len) 6539 { 6540 #ifdef CONFIG_SYN_COOKIES 6541 u32 cookie; 6542 u16 mss; 6543 6544 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4)) 6545 return -EINVAL; 6546 6547 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 6548 return -EINVAL; 6549 6550 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies) 6551 return -ENOENT; 6552 6553 if (!th->syn || th->ack || th->fin || th->rst) 6554 return -EINVAL; 6555 6556 if (unlikely(iph_len < sizeof(struct iphdr))) 6557 return -EINVAL; 6558 6559 /* Both struct iphdr and struct ipv6hdr have the version field at the 6560 * same offset so we can cast to the shorter header (struct iphdr). 6561 */ 6562 switch (((struct iphdr *)iph)->version) { 6563 case 4: 6564 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only) 6565 return -EINVAL; 6566 6567 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie); 6568 break; 6569 6570 #if IS_BUILTIN(CONFIG_IPV6) 6571 case 6: 6572 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 6573 return -EINVAL; 6574 6575 if (sk->sk_family != AF_INET6) 6576 return -EINVAL; 6577 6578 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie); 6579 break; 6580 #endif /* CONFIG_IPV6 */ 6581 6582 default: 6583 return -EPROTONOSUPPORT; 6584 } 6585 if (mss == 0) 6586 return -ENOENT; 6587 6588 return cookie | ((u64)mss << 32); 6589 #else 6590 return -EOPNOTSUPP; 6591 #endif /* CONFIG_SYN_COOKIES */ 6592 } 6593 6594 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = { 6595 .func = bpf_tcp_gen_syncookie, 6596 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */ 6597 .pkt_access = true, 6598 .ret_type = RET_INTEGER, 6599 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6600 .arg2_type = ARG_PTR_TO_MEM, 6601 .arg3_type = ARG_CONST_SIZE, 6602 .arg4_type = ARG_PTR_TO_MEM, 6603 .arg5_type = ARG_CONST_SIZE, 6604 }; 6605 6606 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags) 6607 { 6608 if (!sk || flags != 0) 6609 return -EINVAL; 6610 if (!skb_at_tc_ingress(skb)) 6611 return -EOPNOTSUPP; 6612 if (unlikely(dev_net(skb->dev) != sock_net(sk))) 6613 return -ENETUNREACH; 6614 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport)) 6615 return -ESOCKTNOSUPPORT; 6616 if (sk_is_refcounted(sk) && 6617 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) 6618 return -ENOENT; 6619 6620 skb_orphan(skb); 6621 skb->sk = sk; 6622 skb->destructor = sock_pfree; 6623 6624 return 0; 6625 } 6626 6627 static const struct bpf_func_proto bpf_sk_assign_proto = { 6628 .func = bpf_sk_assign, 6629 .gpl_only = false, 6630 .ret_type = RET_INTEGER, 6631 .arg1_type = ARG_PTR_TO_CTX, 6632 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6633 .arg3_type = ARG_ANYTHING, 6634 }; 6635 6636 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend, 6637 u8 search_kind, const u8 *magic, 6638 u8 magic_len, bool *eol) 6639 { 6640 u8 kind, kind_len; 6641 6642 *eol = false; 6643 6644 while (op < opend) { 6645 kind = op[0]; 6646 6647 if (kind == TCPOPT_EOL) { 6648 *eol = true; 6649 return ERR_PTR(-ENOMSG); 6650 } else if (kind == TCPOPT_NOP) { 6651 op++; 6652 continue; 6653 } 6654 6655 if (opend - op < 2 || opend - op < op[1] || op[1] < 2) 6656 /* Something is wrong in the received header. 6657 * Follow the TCP stack's tcp_parse_options() 6658 * and just bail here. 6659 */ 6660 return ERR_PTR(-EFAULT); 6661 6662 kind_len = op[1]; 6663 if (search_kind == kind) { 6664 if (!magic_len) 6665 return op; 6666 6667 if (magic_len > kind_len - 2) 6668 return ERR_PTR(-ENOMSG); 6669 6670 if (!memcmp(&op[2], magic, magic_len)) 6671 return op; 6672 } 6673 6674 op += kind_len; 6675 } 6676 6677 return ERR_PTR(-ENOMSG); 6678 } 6679 6680 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 6681 void *, search_res, u32, len, u64, flags) 6682 { 6683 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN; 6684 const u8 *op, *opend, *magic, *search = search_res; 6685 u8 search_kind, search_len, copy_len, magic_len; 6686 int ret; 6687 6688 /* 2 byte is the minimal option len except TCPOPT_NOP and 6689 * TCPOPT_EOL which are useless for the bpf prog to learn 6690 * and this helper disallow loading them also. 6691 */ 6692 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN) 6693 return -EINVAL; 6694 6695 search_kind = search[0]; 6696 search_len = search[1]; 6697 6698 if (search_len > len || search_kind == TCPOPT_NOP || 6699 search_kind == TCPOPT_EOL) 6700 return -EINVAL; 6701 6702 if (search_kind == TCPOPT_EXP || search_kind == 253) { 6703 /* 16 or 32 bit magic. +2 for kind and kind length */ 6704 if (search_len != 4 && search_len != 6) 6705 return -EINVAL; 6706 magic = &search[2]; 6707 magic_len = search_len - 2; 6708 } else { 6709 if (search_len) 6710 return -EINVAL; 6711 magic = NULL; 6712 magic_len = 0; 6713 } 6714 6715 if (load_syn) { 6716 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op); 6717 if (ret < 0) 6718 return ret; 6719 6720 opend = op + ret; 6721 op += sizeof(struct tcphdr); 6722 } else { 6723 if (!bpf_sock->skb || 6724 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB) 6725 /* This bpf_sock->op cannot call this helper */ 6726 return -EPERM; 6727 6728 opend = bpf_sock->skb_data_end; 6729 op = bpf_sock->skb->data + sizeof(struct tcphdr); 6730 } 6731 6732 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len, 6733 &eol); 6734 if (IS_ERR(op)) 6735 return PTR_ERR(op); 6736 6737 copy_len = op[1]; 6738 ret = copy_len; 6739 if (copy_len > len) { 6740 ret = -ENOSPC; 6741 copy_len = len; 6742 } 6743 6744 memcpy(search_res, op, copy_len); 6745 return ret; 6746 } 6747 6748 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = { 6749 .func = bpf_sock_ops_load_hdr_opt, 6750 .gpl_only = false, 6751 .ret_type = RET_INTEGER, 6752 .arg1_type = ARG_PTR_TO_CTX, 6753 .arg2_type = ARG_PTR_TO_MEM, 6754 .arg3_type = ARG_CONST_SIZE, 6755 .arg4_type = ARG_ANYTHING, 6756 }; 6757 6758 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 6759 const void *, from, u32, len, u64, flags) 6760 { 6761 u8 new_kind, new_kind_len, magic_len = 0, *opend; 6762 const u8 *op, *new_op, *magic = NULL; 6763 struct sk_buff *skb; 6764 bool eol; 6765 6766 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB) 6767 return -EPERM; 6768 6769 if (len < 2 || flags) 6770 return -EINVAL; 6771 6772 new_op = from; 6773 new_kind = new_op[0]; 6774 new_kind_len = new_op[1]; 6775 6776 if (new_kind_len > len || new_kind == TCPOPT_NOP || 6777 new_kind == TCPOPT_EOL) 6778 return -EINVAL; 6779 6780 if (new_kind_len > bpf_sock->remaining_opt_len) 6781 return -ENOSPC; 6782 6783 /* 253 is another experimental kind */ 6784 if (new_kind == TCPOPT_EXP || new_kind == 253) { 6785 if (new_kind_len < 4) 6786 return -EINVAL; 6787 /* Match for the 2 byte magic also. 6788 * RFC 6994: the magic could be 2 or 4 bytes. 6789 * Hence, matching by 2 byte only is on the 6790 * conservative side but it is the right 6791 * thing to do for the 'search-for-duplication' 6792 * purpose. 6793 */ 6794 magic = &new_op[2]; 6795 magic_len = 2; 6796 } 6797 6798 /* Check for duplication */ 6799 skb = bpf_sock->skb; 6800 op = skb->data + sizeof(struct tcphdr); 6801 opend = bpf_sock->skb_data_end; 6802 6803 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len, 6804 &eol); 6805 if (!IS_ERR(op)) 6806 return -EEXIST; 6807 6808 if (PTR_ERR(op) != -ENOMSG) 6809 return PTR_ERR(op); 6810 6811 if (eol) 6812 /* The option has been ended. Treat it as no more 6813 * header option can be written. 6814 */ 6815 return -ENOSPC; 6816 6817 /* No duplication found. Store the header option. */ 6818 memcpy(opend, from, new_kind_len); 6819 6820 bpf_sock->remaining_opt_len -= new_kind_len; 6821 bpf_sock->skb_data_end += new_kind_len; 6822 6823 return 0; 6824 } 6825 6826 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = { 6827 .func = bpf_sock_ops_store_hdr_opt, 6828 .gpl_only = false, 6829 .ret_type = RET_INTEGER, 6830 .arg1_type = ARG_PTR_TO_CTX, 6831 .arg2_type = ARG_PTR_TO_MEM, 6832 .arg3_type = ARG_CONST_SIZE, 6833 .arg4_type = ARG_ANYTHING, 6834 }; 6835 6836 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 6837 u32, len, u64, flags) 6838 { 6839 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB) 6840 return -EPERM; 6841 6842 if (flags || len < 2) 6843 return -EINVAL; 6844 6845 if (len > bpf_sock->remaining_opt_len) 6846 return -ENOSPC; 6847 6848 bpf_sock->remaining_opt_len -= len; 6849 6850 return 0; 6851 } 6852 6853 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = { 6854 .func = bpf_sock_ops_reserve_hdr_opt, 6855 .gpl_only = false, 6856 .ret_type = RET_INTEGER, 6857 .arg1_type = ARG_PTR_TO_CTX, 6858 .arg2_type = ARG_ANYTHING, 6859 .arg3_type = ARG_ANYTHING, 6860 }; 6861 6862 #endif /* CONFIG_INET */ 6863 6864 bool bpf_helper_changes_pkt_data(void *func) 6865 { 6866 if (func == bpf_skb_vlan_push || 6867 func == bpf_skb_vlan_pop || 6868 func == bpf_skb_store_bytes || 6869 func == bpf_skb_change_proto || 6870 func == bpf_skb_change_head || 6871 func == sk_skb_change_head || 6872 func == bpf_skb_change_tail || 6873 func == sk_skb_change_tail || 6874 func == bpf_skb_adjust_room || 6875 func == sk_skb_adjust_room || 6876 func == bpf_skb_pull_data || 6877 func == sk_skb_pull_data || 6878 func == bpf_clone_redirect || 6879 func == bpf_l3_csum_replace || 6880 func == bpf_l4_csum_replace || 6881 func == bpf_xdp_adjust_head || 6882 func == bpf_xdp_adjust_meta || 6883 func == bpf_msg_pull_data || 6884 func == bpf_msg_push_data || 6885 func == bpf_msg_pop_data || 6886 func == bpf_xdp_adjust_tail || 6887 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6888 func == bpf_lwt_seg6_store_bytes || 6889 func == bpf_lwt_seg6_adjust_srh || 6890 func == bpf_lwt_seg6_action || 6891 #endif 6892 #ifdef CONFIG_INET 6893 func == bpf_sock_ops_store_hdr_opt || 6894 #endif 6895 func == bpf_lwt_in_push_encap || 6896 func == bpf_lwt_xmit_push_encap) 6897 return true; 6898 6899 return false; 6900 } 6901 6902 const struct bpf_func_proto bpf_event_output_data_proto __weak; 6903 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak; 6904 6905 static const struct bpf_func_proto * 6906 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6907 { 6908 switch (func_id) { 6909 /* inet and inet6 sockets are created in a process 6910 * context so there is always a valid uid/gid 6911 */ 6912 case BPF_FUNC_get_current_uid_gid: 6913 return &bpf_get_current_uid_gid_proto; 6914 case BPF_FUNC_get_local_storage: 6915 return &bpf_get_local_storage_proto; 6916 case BPF_FUNC_get_socket_cookie: 6917 return &bpf_get_socket_cookie_sock_proto; 6918 case BPF_FUNC_get_netns_cookie: 6919 return &bpf_get_netns_cookie_sock_proto; 6920 case BPF_FUNC_perf_event_output: 6921 return &bpf_event_output_data_proto; 6922 case BPF_FUNC_get_current_pid_tgid: 6923 return &bpf_get_current_pid_tgid_proto; 6924 case BPF_FUNC_get_current_comm: 6925 return &bpf_get_current_comm_proto; 6926 #ifdef CONFIG_CGROUPS 6927 case BPF_FUNC_get_current_cgroup_id: 6928 return &bpf_get_current_cgroup_id_proto; 6929 case BPF_FUNC_get_current_ancestor_cgroup_id: 6930 return &bpf_get_current_ancestor_cgroup_id_proto; 6931 #endif 6932 #ifdef CONFIG_CGROUP_NET_CLASSID 6933 case BPF_FUNC_get_cgroup_classid: 6934 return &bpf_get_cgroup_classid_curr_proto; 6935 #endif 6936 case BPF_FUNC_sk_storage_get: 6937 return &bpf_sk_storage_get_cg_sock_proto; 6938 default: 6939 return bpf_base_func_proto(func_id); 6940 } 6941 } 6942 6943 static const struct bpf_func_proto * 6944 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6945 { 6946 switch (func_id) { 6947 /* inet and inet6 sockets are created in a process 6948 * context so there is always a valid uid/gid 6949 */ 6950 case BPF_FUNC_get_current_uid_gid: 6951 return &bpf_get_current_uid_gid_proto; 6952 case BPF_FUNC_bind: 6953 switch (prog->expected_attach_type) { 6954 case BPF_CGROUP_INET4_CONNECT: 6955 case BPF_CGROUP_INET6_CONNECT: 6956 return &bpf_bind_proto; 6957 default: 6958 return NULL; 6959 } 6960 case BPF_FUNC_get_socket_cookie: 6961 return &bpf_get_socket_cookie_sock_addr_proto; 6962 case BPF_FUNC_get_netns_cookie: 6963 return &bpf_get_netns_cookie_sock_addr_proto; 6964 case BPF_FUNC_get_local_storage: 6965 return &bpf_get_local_storage_proto; 6966 case BPF_FUNC_perf_event_output: 6967 return &bpf_event_output_data_proto; 6968 case BPF_FUNC_get_current_pid_tgid: 6969 return &bpf_get_current_pid_tgid_proto; 6970 case BPF_FUNC_get_current_comm: 6971 return &bpf_get_current_comm_proto; 6972 #ifdef CONFIG_CGROUPS 6973 case BPF_FUNC_get_current_cgroup_id: 6974 return &bpf_get_current_cgroup_id_proto; 6975 case BPF_FUNC_get_current_ancestor_cgroup_id: 6976 return &bpf_get_current_ancestor_cgroup_id_proto; 6977 #endif 6978 #ifdef CONFIG_CGROUP_NET_CLASSID 6979 case BPF_FUNC_get_cgroup_classid: 6980 return &bpf_get_cgroup_classid_curr_proto; 6981 #endif 6982 #ifdef CONFIG_INET 6983 case BPF_FUNC_sk_lookup_tcp: 6984 return &bpf_sock_addr_sk_lookup_tcp_proto; 6985 case BPF_FUNC_sk_lookup_udp: 6986 return &bpf_sock_addr_sk_lookup_udp_proto; 6987 case BPF_FUNC_sk_release: 6988 return &bpf_sk_release_proto; 6989 case BPF_FUNC_skc_lookup_tcp: 6990 return &bpf_sock_addr_skc_lookup_tcp_proto; 6991 #endif /* CONFIG_INET */ 6992 case BPF_FUNC_sk_storage_get: 6993 return &bpf_sk_storage_get_proto; 6994 case BPF_FUNC_sk_storage_delete: 6995 return &bpf_sk_storage_delete_proto; 6996 case BPF_FUNC_setsockopt: 6997 switch (prog->expected_attach_type) { 6998 case BPF_CGROUP_INET4_CONNECT: 6999 case BPF_CGROUP_INET6_CONNECT: 7000 return &bpf_sock_addr_setsockopt_proto; 7001 default: 7002 return NULL; 7003 } 7004 case BPF_FUNC_getsockopt: 7005 switch (prog->expected_attach_type) { 7006 case BPF_CGROUP_INET4_CONNECT: 7007 case BPF_CGROUP_INET6_CONNECT: 7008 return &bpf_sock_addr_getsockopt_proto; 7009 default: 7010 return NULL; 7011 } 7012 default: 7013 return bpf_sk_base_func_proto(func_id); 7014 } 7015 } 7016 7017 static const struct bpf_func_proto * 7018 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7019 { 7020 switch (func_id) { 7021 case BPF_FUNC_skb_load_bytes: 7022 return &bpf_skb_load_bytes_proto; 7023 case BPF_FUNC_skb_load_bytes_relative: 7024 return &bpf_skb_load_bytes_relative_proto; 7025 case BPF_FUNC_get_socket_cookie: 7026 return &bpf_get_socket_cookie_proto; 7027 case BPF_FUNC_get_socket_uid: 7028 return &bpf_get_socket_uid_proto; 7029 case BPF_FUNC_perf_event_output: 7030 return &bpf_skb_event_output_proto; 7031 default: 7032 return bpf_sk_base_func_proto(func_id); 7033 } 7034 } 7035 7036 const struct bpf_func_proto bpf_sk_storage_get_proto __weak; 7037 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak; 7038 7039 static const struct bpf_func_proto * 7040 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7041 { 7042 switch (func_id) { 7043 case BPF_FUNC_get_local_storage: 7044 return &bpf_get_local_storage_proto; 7045 case BPF_FUNC_sk_fullsock: 7046 return &bpf_sk_fullsock_proto; 7047 case BPF_FUNC_sk_storage_get: 7048 return &bpf_sk_storage_get_proto; 7049 case BPF_FUNC_sk_storage_delete: 7050 return &bpf_sk_storage_delete_proto; 7051 case BPF_FUNC_perf_event_output: 7052 return &bpf_skb_event_output_proto; 7053 #ifdef CONFIG_SOCK_CGROUP_DATA 7054 case BPF_FUNC_skb_cgroup_id: 7055 return &bpf_skb_cgroup_id_proto; 7056 case BPF_FUNC_skb_ancestor_cgroup_id: 7057 return &bpf_skb_ancestor_cgroup_id_proto; 7058 case BPF_FUNC_sk_cgroup_id: 7059 return &bpf_sk_cgroup_id_proto; 7060 case BPF_FUNC_sk_ancestor_cgroup_id: 7061 return &bpf_sk_ancestor_cgroup_id_proto; 7062 #endif 7063 #ifdef CONFIG_INET 7064 case BPF_FUNC_sk_lookup_tcp: 7065 return &bpf_sk_lookup_tcp_proto; 7066 case BPF_FUNC_sk_lookup_udp: 7067 return &bpf_sk_lookup_udp_proto; 7068 case BPF_FUNC_sk_release: 7069 return &bpf_sk_release_proto; 7070 case BPF_FUNC_skc_lookup_tcp: 7071 return &bpf_skc_lookup_tcp_proto; 7072 case BPF_FUNC_tcp_sock: 7073 return &bpf_tcp_sock_proto; 7074 case BPF_FUNC_get_listener_sock: 7075 return &bpf_get_listener_sock_proto; 7076 case BPF_FUNC_skb_ecn_set_ce: 7077 return &bpf_skb_ecn_set_ce_proto; 7078 #endif 7079 default: 7080 return sk_filter_func_proto(func_id, prog); 7081 } 7082 } 7083 7084 static const struct bpf_func_proto * 7085 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7086 { 7087 switch (func_id) { 7088 case BPF_FUNC_skb_store_bytes: 7089 return &bpf_skb_store_bytes_proto; 7090 case BPF_FUNC_skb_load_bytes: 7091 return &bpf_skb_load_bytes_proto; 7092 case BPF_FUNC_skb_load_bytes_relative: 7093 return &bpf_skb_load_bytes_relative_proto; 7094 case BPF_FUNC_skb_pull_data: 7095 return &bpf_skb_pull_data_proto; 7096 case BPF_FUNC_csum_diff: 7097 return &bpf_csum_diff_proto; 7098 case BPF_FUNC_csum_update: 7099 return &bpf_csum_update_proto; 7100 case BPF_FUNC_csum_level: 7101 return &bpf_csum_level_proto; 7102 case BPF_FUNC_l3_csum_replace: 7103 return &bpf_l3_csum_replace_proto; 7104 case BPF_FUNC_l4_csum_replace: 7105 return &bpf_l4_csum_replace_proto; 7106 case BPF_FUNC_clone_redirect: 7107 return &bpf_clone_redirect_proto; 7108 case BPF_FUNC_get_cgroup_classid: 7109 return &bpf_get_cgroup_classid_proto; 7110 case BPF_FUNC_skb_vlan_push: 7111 return &bpf_skb_vlan_push_proto; 7112 case BPF_FUNC_skb_vlan_pop: 7113 return &bpf_skb_vlan_pop_proto; 7114 case BPF_FUNC_skb_change_proto: 7115 return &bpf_skb_change_proto_proto; 7116 case BPF_FUNC_skb_change_type: 7117 return &bpf_skb_change_type_proto; 7118 case BPF_FUNC_skb_adjust_room: 7119 return &bpf_skb_adjust_room_proto; 7120 case BPF_FUNC_skb_change_tail: 7121 return &bpf_skb_change_tail_proto; 7122 case BPF_FUNC_skb_change_head: 7123 return &bpf_skb_change_head_proto; 7124 case BPF_FUNC_skb_get_tunnel_key: 7125 return &bpf_skb_get_tunnel_key_proto; 7126 case BPF_FUNC_skb_set_tunnel_key: 7127 return bpf_get_skb_set_tunnel_proto(func_id); 7128 case BPF_FUNC_skb_get_tunnel_opt: 7129 return &bpf_skb_get_tunnel_opt_proto; 7130 case BPF_FUNC_skb_set_tunnel_opt: 7131 return bpf_get_skb_set_tunnel_proto(func_id); 7132 case BPF_FUNC_redirect: 7133 return &bpf_redirect_proto; 7134 case BPF_FUNC_redirect_neigh: 7135 return &bpf_redirect_neigh_proto; 7136 case BPF_FUNC_redirect_peer: 7137 return &bpf_redirect_peer_proto; 7138 case BPF_FUNC_get_route_realm: 7139 return &bpf_get_route_realm_proto; 7140 case BPF_FUNC_get_hash_recalc: 7141 return &bpf_get_hash_recalc_proto; 7142 case BPF_FUNC_set_hash_invalid: 7143 return &bpf_set_hash_invalid_proto; 7144 case BPF_FUNC_set_hash: 7145 return &bpf_set_hash_proto; 7146 case BPF_FUNC_perf_event_output: 7147 return &bpf_skb_event_output_proto; 7148 case BPF_FUNC_get_smp_processor_id: 7149 return &bpf_get_smp_processor_id_proto; 7150 case BPF_FUNC_skb_under_cgroup: 7151 return &bpf_skb_under_cgroup_proto; 7152 case BPF_FUNC_get_socket_cookie: 7153 return &bpf_get_socket_cookie_proto; 7154 case BPF_FUNC_get_socket_uid: 7155 return &bpf_get_socket_uid_proto; 7156 case BPF_FUNC_fib_lookup: 7157 return &bpf_skb_fib_lookup_proto; 7158 case BPF_FUNC_sk_fullsock: 7159 return &bpf_sk_fullsock_proto; 7160 case BPF_FUNC_sk_storage_get: 7161 return &bpf_sk_storage_get_proto; 7162 case BPF_FUNC_sk_storage_delete: 7163 return &bpf_sk_storage_delete_proto; 7164 #ifdef CONFIG_XFRM 7165 case BPF_FUNC_skb_get_xfrm_state: 7166 return &bpf_skb_get_xfrm_state_proto; 7167 #endif 7168 #ifdef CONFIG_CGROUP_NET_CLASSID 7169 case BPF_FUNC_skb_cgroup_classid: 7170 return &bpf_skb_cgroup_classid_proto; 7171 #endif 7172 #ifdef CONFIG_SOCK_CGROUP_DATA 7173 case BPF_FUNC_skb_cgroup_id: 7174 return &bpf_skb_cgroup_id_proto; 7175 case BPF_FUNC_skb_ancestor_cgroup_id: 7176 return &bpf_skb_ancestor_cgroup_id_proto; 7177 #endif 7178 #ifdef CONFIG_INET 7179 case BPF_FUNC_sk_lookup_tcp: 7180 return &bpf_sk_lookup_tcp_proto; 7181 case BPF_FUNC_sk_lookup_udp: 7182 return &bpf_sk_lookup_udp_proto; 7183 case BPF_FUNC_sk_release: 7184 return &bpf_sk_release_proto; 7185 case BPF_FUNC_tcp_sock: 7186 return &bpf_tcp_sock_proto; 7187 case BPF_FUNC_get_listener_sock: 7188 return &bpf_get_listener_sock_proto; 7189 case BPF_FUNC_skc_lookup_tcp: 7190 return &bpf_skc_lookup_tcp_proto; 7191 case BPF_FUNC_tcp_check_syncookie: 7192 return &bpf_tcp_check_syncookie_proto; 7193 case BPF_FUNC_skb_ecn_set_ce: 7194 return &bpf_skb_ecn_set_ce_proto; 7195 case BPF_FUNC_tcp_gen_syncookie: 7196 return &bpf_tcp_gen_syncookie_proto; 7197 case BPF_FUNC_sk_assign: 7198 return &bpf_sk_assign_proto; 7199 #endif 7200 default: 7201 return bpf_sk_base_func_proto(func_id); 7202 } 7203 } 7204 7205 static const struct bpf_func_proto * 7206 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7207 { 7208 switch (func_id) { 7209 case BPF_FUNC_perf_event_output: 7210 return &bpf_xdp_event_output_proto; 7211 case BPF_FUNC_get_smp_processor_id: 7212 return &bpf_get_smp_processor_id_proto; 7213 case BPF_FUNC_csum_diff: 7214 return &bpf_csum_diff_proto; 7215 case BPF_FUNC_xdp_adjust_head: 7216 return &bpf_xdp_adjust_head_proto; 7217 case BPF_FUNC_xdp_adjust_meta: 7218 return &bpf_xdp_adjust_meta_proto; 7219 case BPF_FUNC_redirect: 7220 return &bpf_xdp_redirect_proto; 7221 case BPF_FUNC_redirect_map: 7222 return &bpf_xdp_redirect_map_proto; 7223 case BPF_FUNC_xdp_adjust_tail: 7224 return &bpf_xdp_adjust_tail_proto; 7225 case BPF_FUNC_fib_lookup: 7226 return &bpf_xdp_fib_lookup_proto; 7227 #ifdef CONFIG_INET 7228 case BPF_FUNC_sk_lookup_udp: 7229 return &bpf_xdp_sk_lookup_udp_proto; 7230 case BPF_FUNC_sk_lookup_tcp: 7231 return &bpf_xdp_sk_lookup_tcp_proto; 7232 case BPF_FUNC_sk_release: 7233 return &bpf_sk_release_proto; 7234 case BPF_FUNC_skc_lookup_tcp: 7235 return &bpf_xdp_skc_lookup_tcp_proto; 7236 case BPF_FUNC_tcp_check_syncookie: 7237 return &bpf_tcp_check_syncookie_proto; 7238 case BPF_FUNC_tcp_gen_syncookie: 7239 return &bpf_tcp_gen_syncookie_proto; 7240 #endif 7241 default: 7242 return bpf_sk_base_func_proto(func_id); 7243 } 7244 } 7245 7246 const struct bpf_func_proto bpf_sock_map_update_proto __weak; 7247 const struct bpf_func_proto bpf_sock_hash_update_proto __weak; 7248 7249 static const struct bpf_func_proto * 7250 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7251 { 7252 switch (func_id) { 7253 case BPF_FUNC_setsockopt: 7254 return &bpf_sock_ops_setsockopt_proto; 7255 case BPF_FUNC_getsockopt: 7256 return &bpf_sock_ops_getsockopt_proto; 7257 case BPF_FUNC_sock_ops_cb_flags_set: 7258 return &bpf_sock_ops_cb_flags_set_proto; 7259 case BPF_FUNC_sock_map_update: 7260 return &bpf_sock_map_update_proto; 7261 case BPF_FUNC_sock_hash_update: 7262 return &bpf_sock_hash_update_proto; 7263 case BPF_FUNC_get_socket_cookie: 7264 return &bpf_get_socket_cookie_sock_ops_proto; 7265 case BPF_FUNC_get_local_storage: 7266 return &bpf_get_local_storage_proto; 7267 case BPF_FUNC_perf_event_output: 7268 return &bpf_event_output_data_proto; 7269 case BPF_FUNC_sk_storage_get: 7270 return &bpf_sk_storage_get_proto; 7271 case BPF_FUNC_sk_storage_delete: 7272 return &bpf_sk_storage_delete_proto; 7273 #ifdef CONFIG_INET 7274 case BPF_FUNC_load_hdr_opt: 7275 return &bpf_sock_ops_load_hdr_opt_proto; 7276 case BPF_FUNC_store_hdr_opt: 7277 return &bpf_sock_ops_store_hdr_opt_proto; 7278 case BPF_FUNC_reserve_hdr_opt: 7279 return &bpf_sock_ops_reserve_hdr_opt_proto; 7280 case BPF_FUNC_tcp_sock: 7281 return &bpf_tcp_sock_proto; 7282 #endif /* CONFIG_INET */ 7283 default: 7284 return bpf_sk_base_func_proto(func_id); 7285 } 7286 } 7287 7288 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak; 7289 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak; 7290 7291 static const struct bpf_func_proto * 7292 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7293 { 7294 switch (func_id) { 7295 case BPF_FUNC_msg_redirect_map: 7296 return &bpf_msg_redirect_map_proto; 7297 case BPF_FUNC_msg_redirect_hash: 7298 return &bpf_msg_redirect_hash_proto; 7299 case BPF_FUNC_msg_apply_bytes: 7300 return &bpf_msg_apply_bytes_proto; 7301 case BPF_FUNC_msg_cork_bytes: 7302 return &bpf_msg_cork_bytes_proto; 7303 case BPF_FUNC_msg_pull_data: 7304 return &bpf_msg_pull_data_proto; 7305 case BPF_FUNC_msg_push_data: 7306 return &bpf_msg_push_data_proto; 7307 case BPF_FUNC_msg_pop_data: 7308 return &bpf_msg_pop_data_proto; 7309 case BPF_FUNC_perf_event_output: 7310 return &bpf_event_output_data_proto; 7311 case BPF_FUNC_get_current_uid_gid: 7312 return &bpf_get_current_uid_gid_proto; 7313 case BPF_FUNC_get_current_pid_tgid: 7314 return &bpf_get_current_pid_tgid_proto; 7315 case BPF_FUNC_sk_storage_get: 7316 return &bpf_sk_storage_get_proto; 7317 case BPF_FUNC_sk_storage_delete: 7318 return &bpf_sk_storage_delete_proto; 7319 #ifdef CONFIG_CGROUPS 7320 case BPF_FUNC_get_current_cgroup_id: 7321 return &bpf_get_current_cgroup_id_proto; 7322 case BPF_FUNC_get_current_ancestor_cgroup_id: 7323 return &bpf_get_current_ancestor_cgroup_id_proto; 7324 #endif 7325 #ifdef CONFIG_CGROUP_NET_CLASSID 7326 case BPF_FUNC_get_cgroup_classid: 7327 return &bpf_get_cgroup_classid_curr_proto; 7328 #endif 7329 default: 7330 return bpf_sk_base_func_proto(func_id); 7331 } 7332 } 7333 7334 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak; 7335 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak; 7336 7337 static const struct bpf_func_proto * 7338 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7339 { 7340 switch (func_id) { 7341 case BPF_FUNC_skb_store_bytes: 7342 return &bpf_skb_store_bytes_proto; 7343 case BPF_FUNC_skb_load_bytes: 7344 return &bpf_skb_load_bytes_proto; 7345 case BPF_FUNC_skb_pull_data: 7346 return &sk_skb_pull_data_proto; 7347 case BPF_FUNC_skb_change_tail: 7348 return &sk_skb_change_tail_proto; 7349 case BPF_FUNC_skb_change_head: 7350 return &sk_skb_change_head_proto; 7351 case BPF_FUNC_skb_adjust_room: 7352 return &sk_skb_adjust_room_proto; 7353 case BPF_FUNC_get_socket_cookie: 7354 return &bpf_get_socket_cookie_proto; 7355 case BPF_FUNC_get_socket_uid: 7356 return &bpf_get_socket_uid_proto; 7357 case BPF_FUNC_sk_redirect_map: 7358 return &bpf_sk_redirect_map_proto; 7359 case BPF_FUNC_sk_redirect_hash: 7360 return &bpf_sk_redirect_hash_proto; 7361 case BPF_FUNC_perf_event_output: 7362 return &bpf_skb_event_output_proto; 7363 #ifdef CONFIG_INET 7364 case BPF_FUNC_sk_lookup_tcp: 7365 return &bpf_sk_lookup_tcp_proto; 7366 case BPF_FUNC_sk_lookup_udp: 7367 return &bpf_sk_lookup_udp_proto; 7368 case BPF_FUNC_sk_release: 7369 return &bpf_sk_release_proto; 7370 case BPF_FUNC_skc_lookup_tcp: 7371 return &bpf_skc_lookup_tcp_proto; 7372 #endif 7373 default: 7374 return bpf_sk_base_func_proto(func_id); 7375 } 7376 } 7377 7378 static const struct bpf_func_proto * 7379 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7380 { 7381 switch (func_id) { 7382 case BPF_FUNC_skb_load_bytes: 7383 return &bpf_flow_dissector_load_bytes_proto; 7384 default: 7385 return bpf_sk_base_func_proto(func_id); 7386 } 7387 } 7388 7389 static const struct bpf_func_proto * 7390 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7391 { 7392 switch (func_id) { 7393 case BPF_FUNC_skb_load_bytes: 7394 return &bpf_skb_load_bytes_proto; 7395 case BPF_FUNC_skb_pull_data: 7396 return &bpf_skb_pull_data_proto; 7397 case BPF_FUNC_csum_diff: 7398 return &bpf_csum_diff_proto; 7399 case BPF_FUNC_get_cgroup_classid: 7400 return &bpf_get_cgroup_classid_proto; 7401 case BPF_FUNC_get_route_realm: 7402 return &bpf_get_route_realm_proto; 7403 case BPF_FUNC_get_hash_recalc: 7404 return &bpf_get_hash_recalc_proto; 7405 case BPF_FUNC_perf_event_output: 7406 return &bpf_skb_event_output_proto; 7407 case BPF_FUNC_get_smp_processor_id: 7408 return &bpf_get_smp_processor_id_proto; 7409 case BPF_FUNC_skb_under_cgroup: 7410 return &bpf_skb_under_cgroup_proto; 7411 default: 7412 return bpf_sk_base_func_proto(func_id); 7413 } 7414 } 7415 7416 static const struct bpf_func_proto * 7417 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7418 { 7419 switch (func_id) { 7420 case BPF_FUNC_lwt_push_encap: 7421 return &bpf_lwt_in_push_encap_proto; 7422 default: 7423 return lwt_out_func_proto(func_id, prog); 7424 } 7425 } 7426 7427 static const struct bpf_func_proto * 7428 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7429 { 7430 switch (func_id) { 7431 case BPF_FUNC_skb_get_tunnel_key: 7432 return &bpf_skb_get_tunnel_key_proto; 7433 case BPF_FUNC_skb_set_tunnel_key: 7434 return bpf_get_skb_set_tunnel_proto(func_id); 7435 case BPF_FUNC_skb_get_tunnel_opt: 7436 return &bpf_skb_get_tunnel_opt_proto; 7437 case BPF_FUNC_skb_set_tunnel_opt: 7438 return bpf_get_skb_set_tunnel_proto(func_id); 7439 case BPF_FUNC_redirect: 7440 return &bpf_redirect_proto; 7441 case BPF_FUNC_clone_redirect: 7442 return &bpf_clone_redirect_proto; 7443 case BPF_FUNC_skb_change_tail: 7444 return &bpf_skb_change_tail_proto; 7445 case BPF_FUNC_skb_change_head: 7446 return &bpf_skb_change_head_proto; 7447 case BPF_FUNC_skb_store_bytes: 7448 return &bpf_skb_store_bytes_proto; 7449 case BPF_FUNC_csum_update: 7450 return &bpf_csum_update_proto; 7451 case BPF_FUNC_csum_level: 7452 return &bpf_csum_level_proto; 7453 case BPF_FUNC_l3_csum_replace: 7454 return &bpf_l3_csum_replace_proto; 7455 case BPF_FUNC_l4_csum_replace: 7456 return &bpf_l4_csum_replace_proto; 7457 case BPF_FUNC_set_hash_invalid: 7458 return &bpf_set_hash_invalid_proto; 7459 case BPF_FUNC_lwt_push_encap: 7460 return &bpf_lwt_xmit_push_encap_proto; 7461 default: 7462 return lwt_out_func_proto(func_id, prog); 7463 } 7464 } 7465 7466 static const struct bpf_func_proto * 7467 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7468 { 7469 switch (func_id) { 7470 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 7471 case BPF_FUNC_lwt_seg6_store_bytes: 7472 return &bpf_lwt_seg6_store_bytes_proto; 7473 case BPF_FUNC_lwt_seg6_action: 7474 return &bpf_lwt_seg6_action_proto; 7475 case BPF_FUNC_lwt_seg6_adjust_srh: 7476 return &bpf_lwt_seg6_adjust_srh_proto; 7477 #endif 7478 default: 7479 return lwt_out_func_proto(func_id, prog); 7480 } 7481 } 7482 7483 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type, 7484 const struct bpf_prog *prog, 7485 struct bpf_insn_access_aux *info) 7486 { 7487 const int size_default = sizeof(__u32); 7488 7489 if (off < 0 || off >= sizeof(struct __sk_buff)) 7490 return false; 7491 7492 /* The verifier guarantees that size > 0. */ 7493 if (off % size != 0) 7494 return false; 7495 7496 switch (off) { 7497 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7498 if (off + size > offsetofend(struct __sk_buff, cb[4])) 7499 return false; 7500 break; 7501 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]): 7502 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]): 7503 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4): 7504 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4): 7505 case bpf_ctx_range(struct __sk_buff, data): 7506 case bpf_ctx_range(struct __sk_buff, data_meta): 7507 case bpf_ctx_range(struct __sk_buff, data_end): 7508 if (size != size_default) 7509 return false; 7510 break; 7511 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 7512 return false; 7513 case bpf_ctx_range(struct __sk_buff, tstamp): 7514 if (size != sizeof(__u64)) 7515 return false; 7516 break; 7517 case offsetof(struct __sk_buff, sk): 7518 if (type == BPF_WRITE || size != sizeof(__u64)) 7519 return false; 7520 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 7521 break; 7522 default: 7523 /* Only narrow read access allowed for now. */ 7524 if (type == BPF_WRITE) { 7525 if (size != size_default) 7526 return false; 7527 } else { 7528 bpf_ctx_record_field_size(info, size_default); 7529 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 7530 return false; 7531 } 7532 } 7533 7534 return true; 7535 } 7536 7537 static bool sk_filter_is_valid_access(int off, int size, 7538 enum bpf_access_type type, 7539 const struct bpf_prog *prog, 7540 struct bpf_insn_access_aux *info) 7541 { 7542 switch (off) { 7543 case bpf_ctx_range(struct __sk_buff, tc_classid): 7544 case bpf_ctx_range(struct __sk_buff, data): 7545 case bpf_ctx_range(struct __sk_buff, data_meta): 7546 case bpf_ctx_range(struct __sk_buff, data_end): 7547 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 7548 case bpf_ctx_range(struct __sk_buff, tstamp): 7549 case bpf_ctx_range(struct __sk_buff, wire_len): 7550 return false; 7551 } 7552 7553 if (type == BPF_WRITE) { 7554 switch (off) { 7555 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7556 break; 7557 default: 7558 return false; 7559 } 7560 } 7561 7562 return bpf_skb_is_valid_access(off, size, type, prog, info); 7563 } 7564 7565 static bool cg_skb_is_valid_access(int off, int size, 7566 enum bpf_access_type type, 7567 const struct bpf_prog *prog, 7568 struct bpf_insn_access_aux *info) 7569 { 7570 switch (off) { 7571 case bpf_ctx_range(struct __sk_buff, tc_classid): 7572 case bpf_ctx_range(struct __sk_buff, data_meta): 7573 case bpf_ctx_range(struct __sk_buff, wire_len): 7574 return false; 7575 case bpf_ctx_range(struct __sk_buff, data): 7576 case bpf_ctx_range(struct __sk_buff, data_end): 7577 if (!bpf_capable()) 7578 return false; 7579 break; 7580 } 7581 7582 if (type == BPF_WRITE) { 7583 switch (off) { 7584 case bpf_ctx_range(struct __sk_buff, mark): 7585 case bpf_ctx_range(struct __sk_buff, priority): 7586 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7587 break; 7588 case bpf_ctx_range(struct __sk_buff, tstamp): 7589 if (!bpf_capable()) 7590 return false; 7591 break; 7592 default: 7593 return false; 7594 } 7595 } 7596 7597 switch (off) { 7598 case bpf_ctx_range(struct __sk_buff, data): 7599 info->reg_type = PTR_TO_PACKET; 7600 break; 7601 case bpf_ctx_range(struct __sk_buff, data_end): 7602 info->reg_type = PTR_TO_PACKET_END; 7603 break; 7604 } 7605 7606 return bpf_skb_is_valid_access(off, size, type, prog, info); 7607 } 7608 7609 static bool lwt_is_valid_access(int off, int size, 7610 enum bpf_access_type type, 7611 const struct bpf_prog *prog, 7612 struct bpf_insn_access_aux *info) 7613 { 7614 switch (off) { 7615 case bpf_ctx_range(struct __sk_buff, tc_classid): 7616 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 7617 case bpf_ctx_range(struct __sk_buff, data_meta): 7618 case bpf_ctx_range(struct __sk_buff, tstamp): 7619 case bpf_ctx_range(struct __sk_buff, wire_len): 7620 return false; 7621 } 7622 7623 if (type == BPF_WRITE) { 7624 switch (off) { 7625 case bpf_ctx_range(struct __sk_buff, mark): 7626 case bpf_ctx_range(struct __sk_buff, priority): 7627 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7628 break; 7629 default: 7630 return false; 7631 } 7632 } 7633 7634 switch (off) { 7635 case bpf_ctx_range(struct __sk_buff, data): 7636 info->reg_type = PTR_TO_PACKET; 7637 break; 7638 case bpf_ctx_range(struct __sk_buff, data_end): 7639 info->reg_type = PTR_TO_PACKET_END; 7640 break; 7641 } 7642 7643 return bpf_skb_is_valid_access(off, size, type, prog, info); 7644 } 7645 7646 /* Attach type specific accesses */ 7647 static bool __sock_filter_check_attach_type(int off, 7648 enum bpf_access_type access_type, 7649 enum bpf_attach_type attach_type) 7650 { 7651 switch (off) { 7652 case offsetof(struct bpf_sock, bound_dev_if): 7653 case offsetof(struct bpf_sock, mark): 7654 case offsetof(struct bpf_sock, priority): 7655 switch (attach_type) { 7656 case BPF_CGROUP_INET_SOCK_CREATE: 7657 case BPF_CGROUP_INET_SOCK_RELEASE: 7658 goto full_access; 7659 default: 7660 return false; 7661 } 7662 case bpf_ctx_range(struct bpf_sock, src_ip4): 7663 switch (attach_type) { 7664 case BPF_CGROUP_INET4_POST_BIND: 7665 goto read_only; 7666 default: 7667 return false; 7668 } 7669 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 7670 switch (attach_type) { 7671 case BPF_CGROUP_INET6_POST_BIND: 7672 goto read_only; 7673 default: 7674 return false; 7675 } 7676 case bpf_ctx_range(struct bpf_sock, src_port): 7677 switch (attach_type) { 7678 case BPF_CGROUP_INET4_POST_BIND: 7679 case BPF_CGROUP_INET6_POST_BIND: 7680 goto read_only; 7681 default: 7682 return false; 7683 } 7684 } 7685 read_only: 7686 return access_type == BPF_READ; 7687 full_access: 7688 return true; 7689 } 7690 7691 bool bpf_sock_common_is_valid_access(int off, int size, 7692 enum bpf_access_type type, 7693 struct bpf_insn_access_aux *info) 7694 { 7695 switch (off) { 7696 case bpf_ctx_range_till(struct bpf_sock, type, priority): 7697 return false; 7698 default: 7699 return bpf_sock_is_valid_access(off, size, type, info); 7700 } 7701 } 7702 7703 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 7704 struct bpf_insn_access_aux *info) 7705 { 7706 const int size_default = sizeof(__u32); 7707 7708 if (off < 0 || off >= sizeof(struct bpf_sock)) 7709 return false; 7710 if (off % size != 0) 7711 return false; 7712 7713 switch (off) { 7714 case offsetof(struct bpf_sock, state): 7715 case offsetof(struct bpf_sock, family): 7716 case offsetof(struct bpf_sock, type): 7717 case offsetof(struct bpf_sock, protocol): 7718 case offsetof(struct bpf_sock, dst_port): 7719 case offsetof(struct bpf_sock, src_port): 7720 case offsetof(struct bpf_sock, rx_queue_mapping): 7721 case bpf_ctx_range(struct bpf_sock, src_ip4): 7722 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 7723 case bpf_ctx_range(struct bpf_sock, dst_ip4): 7724 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 7725 bpf_ctx_record_field_size(info, size_default); 7726 return bpf_ctx_narrow_access_ok(off, size, size_default); 7727 } 7728 7729 return size == size_default; 7730 } 7731 7732 static bool sock_filter_is_valid_access(int off, int size, 7733 enum bpf_access_type type, 7734 const struct bpf_prog *prog, 7735 struct bpf_insn_access_aux *info) 7736 { 7737 if (!bpf_sock_is_valid_access(off, size, type, info)) 7738 return false; 7739 return __sock_filter_check_attach_type(off, type, 7740 prog->expected_attach_type); 7741 } 7742 7743 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write, 7744 const struct bpf_prog *prog) 7745 { 7746 /* Neither direct read nor direct write requires any preliminary 7747 * action. 7748 */ 7749 return 0; 7750 } 7751 7752 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write, 7753 const struct bpf_prog *prog, int drop_verdict) 7754 { 7755 struct bpf_insn *insn = insn_buf; 7756 7757 if (!direct_write) 7758 return 0; 7759 7760 /* if (!skb->cloned) 7761 * goto start; 7762 * 7763 * (Fast-path, otherwise approximation that we might be 7764 * a clone, do the rest in helper.) 7765 */ 7766 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET()); 7767 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK); 7768 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7); 7769 7770 /* ret = bpf_skb_pull_data(skb, 0); */ 7771 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1); 7772 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2); 7773 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, 7774 BPF_FUNC_skb_pull_data); 7775 /* if (!ret) 7776 * goto restore; 7777 * return TC_ACT_SHOT; 7778 */ 7779 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2); 7780 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict); 7781 *insn++ = BPF_EXIT_INSN(); 7782 7783 /* restore: */ 7784 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6); 7785 /* start: */ 7786 *insn++ = prog->insnsi[0]; 7787 7788 return insn - insn_buf; 7789 } 7790 7791 static int bpf_gen_ld_abs(const struct bpf_insn *orig, 7792 struct bpf_insn *insn_buf) 7793 { 7794 bool indirect = BPF_MODE(orig->code) == BPF_IND; 7795 struct bpf_insn *insn = insn_buf; 7796 7797 if (!indirect) { 7798 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm); 7799 } else { 7800 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg); 7801 if (orig->imm) 7802 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm); 7803 } 7804 /* We're guaranteed here that CTX is in R6. */ 7805 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX); 7806 7807 switch (BPF_SIZE(orig->code)) { 7808 case BPF_B: 7809 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache); 7810 break; 7811 case BPF_H: 7812 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache); 7813 break; 7814 case BPF_W: 7815 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache); 7816 break; 7817 } 7818 7819 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2); 7820 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0); 7821 *insn++ = BPF_EXIT_INSN(); 7822 7823 return insn - insn_buf; 7824 } 7825 7826 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write, 7827 const struct bpf_prog *prog) 7828 { 7829 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT); 7830 } 7831 7832 static bool tc_cls_act_is_valid_access(int off, int size, 7833 enum bpf_access_type type, 7834 const struct bpf_prog *prog, 7835 struct bpf_insn_access_aux *info) 7836 { 7837 if (type == BPF_WRITE) { 7838 switch (off) { 7839 case bpf_ctx_range(struct __sk_buff, mark): 7840 case bpf_ctx_range(struct __sk_buff, tc_index): 7841 case bpf_ctx_range(struct __sk_buff, priority): 7842 case bpf_ctx_range(struct __sk_buff, tc_classid): 7843 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7844 case bpf_ctx_range(struct __sk_buff, tstamp): 7845 case bpf_ctx_range(struct __sk_buff, queue_mapping): 7846 break; 7847 default: 7848 return false; 7849 } 7850 } 7851 7852 switch (off) { 7853 case bpf_ctx_range(struct __sk_buff, data): 7854 info->reg_type = PTR_TO_PACKET; 7855 break; 7856 case bpf_ctx_range(struct __sk_buff, data_meta): 7857 info->reg_type = PTR_TO_PACKET_META; 7858 break; 7859 case bpf_ctx_range(struct __sk_buff, data_end): 7860 info->reg_type = PTR_TO_PACKET_END; 7861 break; 7862 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 7863 return false; 7864 } 7865 7866 return bpf_skb_is_valid_access(off, size, type, prog, info); 7867 } 7868 7869 static bool __is_valid_xdp_access(int off, int size) 7870 { 7871 if (off < 0 || off >= sizeof(struct xdp_md)) 7872 return false; 7873 if (off % size != 0) 7874 return false; 7875 if (size != sizeof(__u32)) 7876 return false; 7877 7878 return true; 7879 } 7880 7881 static bool xdp_is_valid_access(int off, int size, 7882 enum bpf_access_type type, 7883 const struct bpf_prog *prog, 7884 struct bpf_insn_access_aux *info) 7885 { 7886 if (prog->expected_attach_type != BPF_XDP_DEVMAP) { 7887 switch (off) { 7888 case offsetof(struct xdp_md, egress_ifindex): 7889 return false; 7890 } 7891 } 7892 7893 if (type == BPF_WRITE) { 7894 if (bpf_prog_is_dev_bound(prog->aux)) { 7895 switch (off) { 7896 case offsetof(struct xdp_md, rx_queue_index): 7897 return __is_valid_xdp_access(off, size); 7898 } 7899 } 7900 return false; 7901 } 7902 7903 switch (off) { 7904 case offsetof(struct xdp_md, data): 7905 info->reg_type = PTR_TO_PACKET; 7906 break; 7907 case offsetof(struct xdp_md, data_meta): 7908 info->reg_type = PTR_TO_PACKET_META; 7909 break; 7910 case offsetof(struct xdp_md, data_end): 7911 info->reg_type = PTR_TO_PACKET_END; 7912 break; 7913 } 7914 7915 return __is_valid_xdp_access(off, size); 7916 } 7917 7918 void bpf_warn_invalid_xdp_action(u32 act) 7919 { 7920 const u32 act_max = XDP_REDIRECT; 7921 7922 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n", 7923 act > act_max ? "Illegal" : "Driver unsupported", 7924 act); 7925 } 7926 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); 7927 7928 static bool sock_addr_is_valid_access(int off, int size, 7929 enum bpf_access_type type, 7930 const struct bpf_prog *prog, 7931 struct bpf_insn_access_aux *info) 7932 { 7933 const int size_default = sizeof(__u32); 7934 7935 if (off < 0 || off >= sizeof(struct bpf_sock_addr)) 7936 return false; 7937 if (off % size != 0) 7938 return false; 7939 7940 /* Disallow access to IPv6 fields from IPv4 contex and vise 7941 * versa. 7942 */ 7943 switch (off) { 7944 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 7945 switch (prog->expected_attach_type) { 7946 case BPF_CGROUP_INET4_BIND: 7947 case BPF_CGROUP_INET4_CONNECT: 7948 case BPF_CGROUP_INET4_GETPEERNAME: 7949 case BPF_CGROUP_INET4_GETSOCKNAME: 7950 case BPF_CGROUP_UDP4_SENDMSG: 7951 case BPF_CGROUP_UDP4_RECVMSG: 7952 break; 7953 default: 7954 return false; 7955 } 7956 break; 7957 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 7958 switch (prog->expected_attach_type) { 7959 case BPF_CGROUP_INET6_BIND: 7960 case BPF_CGROUP_INET6_CONNECT: 7961 case BPF_CGROUP_INET6_GETPEERNAME: 7962 case BPF_CGROUP_INET6_GETSOCKNAME: 7963 case BPF_CGROUP_UDP6_SENDMSG: 7964 case BPF_CGROUP_UDP6_RECVMSG: 7965 break; 7966 default: 7967 return false; 7968 } 7969 break; 7970 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 7971 switch (prog->expected_attach_type) { 7972 case BPF_CGROUP_UDP4_SENDMSG: 7973 break; 7974 default: 7975 return false; 7976 } 7977 break; 7978 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 7979 msg_src_ip6[3]): 7980 switch (prog->expected_attach_type) { 7981 case BPF_CGROUP_UDP6_SENDMSG: 7982 break; 7983 default: 7984 return false; 7985 } 7986 break; 7987 } 7988 7989 switch (off) { 7990 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 7991 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 7992 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 7993 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 7994 msg_src_ip6[3]): 7995 case bpf_ctx_range(struct bpf_sock_addr, user_port): 7996 if (type == BPF_READ) { 7997 bpf_ctx_record_field_size(info, size_default); 7998 7999 if (bpf_ctx_wide_access_ok(off, size, 8000 struct bpf_sock_addr, 8001 user_ip6)) 8002 return true; 8003 8004 if (bpf_ctx_wide_access_ok(off, size, 8005 struct bpf_sock_addr, 8006 msg_src_ip6)) 8007 return true; 8008 8009 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8010 return false; 8011 } else { 8012 if (bpf_ctx_wide_access_ok(off, size, 8013 struct bpf_sock_addr, 8014 user_ip6)) 8015 return true; 8016 8017 if (bpf_ctx_wide_access_ok(off, size, 8018 struct bpf_sock_addr, 8019 msg_src_ip6)) 8020 return true; 8021 8022 if (size != size_default) 8023 return false; 8024 } 8025 break; 8026 case offsetof(struct bpf_sock_addr, sk): 8027 if (type != BPF_READ) 8028 return false; 8029 if (size != sizeof(__u64)) 8030 return false; 8031 info->reg_type = PTR_TO_SOCKET; 8032 break; 8033 default: 8034 if (type == BPF_READ) { 8035 if (size != size_default) 8036 return false; 8037 } else { 8038 return false; 8039 } 8040 } 8041 8042 return true; 8043 } 8044 8045 static bool sock_ops_is_valid_access(int off, int size, 8046 enum bpf_access_type type, 8047 const struct bpf_prog *prog, 8048 struct bpf_insn_access_aux *info) 8049 { 8050 const int size_default = sizeof(__u32); 8051 8052 if (off < 0 || off >= sizeof(struct bpf_sock_ops)) 8053 return false; 8054 8055 /* The verifier guarantees that size > 0. */ 8056 if (off % size != 0) 8057 return false; 8058 8059 if (type == BPF_WRITE) { 8060 switch (off) { 8061 case offsetof(struct bpf_sock_ops, reply): 8062 case offsetof(struct bpf_sock_ops, sk_txhash): 8063 if (size != size_default) 8064 return false; 8065 break; 8066 default: 8067 return false; 8068 } 8069 } else { 8070 switch (off) { 8071 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received, 8072 bytes_acked): 8073 if (size != sizeof(__u64)) 8074 return false; 8075 break; 8076 case offsetof(struct bpf_sock_ops, sk): 8077 if (size != sizeof(__u64)) 8078 return false; 8079 info->reg_type = PTR_TO_SOCKET_OR_NULL; 8080 break; 8081 case offsetof(struct bpf_sock_ops, skb_data): 8082 if (size != sizeof(__u64)) 8083 return false; 8084 info->reg_type = PTR_TO_PACKET; 8085 break; 8086 case offsetof(struct bpf_sock_ops, skb_data_end): 8087 if (size != sizeof(__u64)) 8088 return false; 8089 info->reg_type = PTR_TO_PACKET_END; 8090 break; 8091 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 8092 bpf_ctx_record_field_size(info, size_default); 8093 return bpf_ctx_narrow_access_ok(off, size, 8094 size_default); 8095 default: 8096 if (size != size_default) 8097 return false; 8098 break; 8099 } 8100 } 8101 8102 return true; 8103 } 8104 8105 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write, 8106 const struct bpf_prog *prog) 8107 { 8108 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP); 8109 } 8110 8111 static bool sk_skb_is_valid_access(int off, int size, 8112 enum bpf_access_type type, 8113 const struct bpf_prog *prog, 8114 struct bpf_insn_access_aux *info) 8115 { 8116 switch (off) { 8117 case bpf_ctx_range(struct __sk_buff, tc_classid): 8118 case bpf_ctx_range(struct __sk_buff, data_meta): 8119 case bpf_ctx_range(struct __sk_buff, tstamp): 8120 case bpf_ctx_range(struct __sk_buff, wire_len): 8121 return false; 8122 } 8123 8124 if (type == BPF_WRITE) { 8125 switch (off) { 8126 case bpf_ctx_range(struct __sk_buff, tc_index): 8127 case bpf_ctx_range(struct __sk_buff, priority): 8128 break; 8129 default: 8130 return false; 8131 } 8132 } 8133 8134 switch (off) { 8135 case bpf_ctx_range(struct __sk_buff, mark): 8136 return false; 8137 case bpf_ctx_range(struct __sk_buff, data): 8138 info->reg_type = PTR_TO_PACKET; 8139 break; 8140 case bpf_ctx_range(struct __sk_buff, data_end): 8141 info->reg_type = PTR_TO_PACKET_END; 8142 break; 8143 } 8144 8145 return bpf_skb_is_valid_access(off, size, type, prog, info); 8146 } 8147 8148 static bool sk_msg_is_valid_access(int off, int size, 8149 enum bpf_access_type type, 8150 const struct bpf_prog *prog, 8151 struct bpf_insn_access_aux *info) 8152 { 8153 if (type == BPF_WRITE) 8154 return false; 8155 8156 if (off % size != 0) 8157 return false; 8158 8159 switch (off) { 8160 case offsetof(struct sk_msg_md, data): 8161 info->reg_type = PTR_TO_PACKET; 8162 if (size != sizeof(__u64)) 8163 return false; 8164 break; 8165 case offsetof(struct sk_msg_md, data_end): 8166 info->reg_type = PTR_TO_PACKET_END; 8167 if (size != sizeof(__u64)) 8168 return false; 8169 break; 8170 case offsetof(struct sk_msg_md, sk): 8171 if (size != sizeof(__u64)) 8172 return false; 8173 info->reg_type = PTR_TO_SOCKET; 8174 break; 8175 case bpf_ctx_range(struct sk_msg_md, family): 8176 case bpf_ctx_range(struct sk_msg_md, remote_ip4): 8177 case bpf_ctx_range(struct sk_msg_md, local_ip4): 8178 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]): 8179 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]): 8180 case bpf_ctx_range(struct sk_msg_md, remote_port): 8181 case bpf_ctx_range(struct sk_msg_md, local_port): 8182 case bpf_ctx_range(struct sk_msg_md, size): 8183 if (size != sizeof(__u32)) 8184 return false; 8185 break; 8186 default: 8187 return false; 8188 } 8189 return true; 8190 } 8191 8192 static bool flow_dissector_is_valid_access(int off, int size, 8193 enum bpf_access_type type, 8194 const struct bpf_prog *prog, 8195 struct bpf_insn_access_aux *info) 8196 { 8197 const int size_default = sizeof(__u32); 8198 8199 if (off < 0 || off >= sizeof(struct __sk_buff)) 8200 return false; 8201 8202 if (type == BPF_WRITE) 8203 return false; 8204 8205 switch (off) { 8206 case bpf_ctx_range(struct __sk_buff, data): 8207 if (size != size_default) 8208 return false; 8209 info->reg_type = PTR_TO_PACKET; 8210 return true; 8211 case bpf_ctx_range(struct __sk_buff, data_end): 8212 if (size != size_default) 8213 return false; 8214 info->reg_type = PTR_TO_PACKET_END; 8215 return true; 8216 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 8217 if (size != sizeof(__u64)) 8218 return false; 8219 info->reg_type = PTR_TO_FLOW_KEYS; 8220 return true; 8221 default: 8222 return false; 8223 } 8224 } 8225 8226 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type, 8227 const struct bpf_insn *si, 8228 struct bpf_insn *insn_buf, 8229 struct bpf_prog *prog, 8230 u32 *target_size) 8231 8232 { 8233 struct bpf_insn *insn = insn_buf; 8234 8235 switch (si->off) { 8236 case offsetof(struct __sk_buff, data): 8237 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data), 8238 si->dst_reg, si->src_reg, 8239 offsetof(struct bpf_flow_dissector, data)); 8240 break; 8241 8242 case offsetof(struct __sk_buff, data_end): 8243 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end), 8244 si->dst_reg, si->src_reg, 8245 offsetof(struct bpf_flow_dissector, data_end)); 8246 break; 8247 8248 case offsetof(struct __sk_buff, flow_keys): 8249 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys), 8250 si->dst_reg, si->src_reg, 8251 offsetof(struct bpf_flow_dissector, flow_keys)); 8252 break; 8253 } 8254 8255 return insn - insn_buf; 8256 } 8257 8258 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si, 8259 struct bpf_insn *insn) 8260 { 8261 /* si->dst_reg = skb_shinfo(SKB); */ 8262 #ifdef NET_SKBUFF_DATA_USES_OFFSET 8263 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 8264 BPF_REG_AX, si->src_reg, 8265 offsetof(struct sk_buff, end)); 8266 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head), 8267 si->dst_reg, si->src_reg, 8268 offsetof(struct sk_buff, head)); 8269 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX); 8270 #else 8271 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 8272 si->dst_reg, si->src_reg, 8273 offsetof(struct sk_buff, end)); 8274 #endif 8275 8276 return insn; 8277 } 8278 8279 static u32 bpf_convert_ctx_access(enum bpf_access_type type, 8280 const struct bpf_insn *si, 8281 struct bpf_insn *insn_buf, 8282 struct bpf_prog *prog, u32 *target_size) 8283 { 8284 struct bpf_insn *insn = insn_buf; 8285 int off; 8286 8287 switch (si->off) { 8288 case offsetof(struct __sk_buff, len): 8289 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8290 bpf_target_off(struct sk_buff, len, 4, 8291 target_size)); 8292 break; 8293 8294 case offsetof(struct __sk_buff, protocol): 8295 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8296 bpf_target_off(struct sk_buff, protocol, 2, 8297 target_size)); 8298 break; 8299 8300 case offsetof(struct __sk_buff, vlan_proto): 8301 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8302 bpf_target_off(struct sk_buff, vlan_proto, 2, 8303 target_size)); 8304 break; 8305 8306 case offsetof(struct __sk_buff, priority): 8307 if (type == BPF_WRITE) 8308 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8309 bpf_target_off(struct sk_buff, priority, 4, 8310 target_size)); 8311 else 8312 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8313 bpf_target_off(struct sk_buff, priority, 4, 8314 target_size)); 8315 break; 8316 8317 case offsetof(struct __sk_buff, ingress_ifindex): 8318 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8319 bpf_target_off(struct sk_buff, skb_iif, 4, 8320 target_size)); 8321 break; 8322 8323 case offsetof(struct __sk_buff, ifindex): 8324 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 8325 si->dst_reg, si->src_reg, 8326 offsetof(struct sk_buff, dev)); 8327 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 8328 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8329 bpf_target_off(struct net_device, ifindex, 4, 8330 target_size)); 8331 break; 8332 8333 case offsetof(struct __sk_buff, hash): 8334 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8335 bpf_target_off(struct sk_buff, hash, 4, 8336 target_size)); 8337 break; 8338 8339 case offsetof(struct __sk_buff, mark): 8340 if (type == BPF_WRITE) 8341 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8342 bpf_target_off(struct sk_buff, mark, 4, 8343 target_size)); 8344 else 8345 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8346 bpf_target_off(struct sk_buff, mark, 4, 8347 target_size)); 8348 break; 8349 8350 case offsetof(struct __sk_buff, pkt_type): 8351 *target_size = 1; 8352 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 8353 PKT_TYPE_OFFSET()); 8354 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX); 8355 #ifdef __BIG_ENDIAN_BITFIELD 8356 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5); 8357 #endif 8358 break; 8359 8360 case offsetof(struct __sk_buff, queue_mapping): 8361 if (type == BPF_WRITE) { 8362 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1); 8363 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 8364 bpf_target_off(struct sk_buff, 8365 queue_mapping, 8366 2, target_size)); 8367 } else { 8368 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8369 bpf_target_off(struct sk_buff, 8370 queue_mapping, 8371 2, target_size)); 8372 } 8373 break; 8374 8375 case offsetof(struct __sk_buff, vlan_present): 8376 *target_size = 1; 8377 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 8378 PKT_VLAN_PRESENT_OFFSET()); 8379 if (PKT_VLAN_PRESENT_BIT) 8380 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT); 8381 if (PKT_VLAN_PRESENT_BIT < 7) 8382 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1); 8383 break; 8384 8385 case offsetof(struct __sk_buff, vlan_tci): 8386 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8387 bpf_target_off(struct sk_buff, vlan_tci, 2, 8388 target_size)); 8389 break; 8390 8391 case offsetof(struct __sk_buff, cb[0]) ... 8392 offsetofend(struct __sk_buff, cb[4]) - 1: 8393 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20); 8394 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 8395 offsetof(struct qdisc_skb_cb, data)) % 8396 sizeof(__u64)); 8397 8398 prog->cb_access = 1; 8399 off = si->off; 8400 off -= offsetof(struct __sk_buff, cb[0]); 8401 off += offsetof(struct sk_buff, cb); 8402 off += offsetof(struct qdisc_skb_cb, data); 8403 if (type == BPF_WRITE) 8404 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 8405 si->src_reg, off); 8406 else 8407 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 8408 si->src_reg, off); 8409 break; 8410 8411 case offsetof(struct __sk_buff, tc_classid): 8412 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2); 8413 8414 off = si->off; 8415 off -= offsetof(struct __sk_buff, tc_classid); 8416 off += offsetof(struct sk_buff, cb); 8417 off += offsetof(struct qdisc_skb_cb, tc_classid); 8418 *target_size = 2; 8419 if (type == BPF_WRITE) 8420 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, 8421 si->src_reg, off); 8422 else 8423 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, 8424 si->src_reg, off); 8425 break; 8426 8427 case offsetof(struct __sk_buff, data): 8428 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 8429 si->dst_reg, si->src_reg, 8430 offsetof(struct sk_buff, data)); 8431 break; 8432 8433 case offsetof(struct __sk_buff, data_meta): 8434 off = si->off; 8435 off -= offsetof(struct __sk_buff, data_meta); 8436 off += offsetof(struct sk_buff, cb); 8437 off += offsetof(struct bpf_skb_data_end, data_meta); 8438 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 8439 si->src_reg, off); 8440 break; 8441 8442 case offsetof(struct __sk_buff, data_end): 8443 off = si->off; 8444 off -= offsetof(struct __sk_buff, data_end); 8445 off += offsetof(struct sk_buff, cb); 8446 off += offsetof(struct bpf_skb_data_end, data_end); 8447 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 8448 si->src_reg, off); 8449 break; 8450 8451 case offsetof(struct __sk_buff, tc_index): 8452 #ifdef CONFIG_NET_SCHED 8453 if (type == BPF_WRITE) 8454 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 8455 bpf_target_off(struct sk_buff, tc_index, 2, 8456 target_size)); 8457 else 8458 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8459 bpf_target_off(struct sk_buff, tc_index, 2, 8460 target_size)); 8461 #else 8462 *target_size = 2; 8463 if (type == BPF_WRITE) 8464 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg); 8465 else 8466 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 8467 #endif 8468 break; 8469 8470 case offsetof(struct __sk_buff, napi_id): 8471 #if defined(CONFIG_NET_RX_BUSY_POLL) 8472 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8473 bpf_target_off(struct sk_buff, napi_id, 4, 8474 target_size)); 8475 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1); 8476 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 8477 #else 8478 *target_size = 4; 8479 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 8480 #endif 8481 break; 8482 case offsetof(struct __sk_buff, family): 8483 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 8484 8485 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8486 si->dst_reg, si->src_reg, 8487 offsetof(struct sk_buff, sk)); 8488 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8489 bpf_target_off(struct sock_common, 8490 skc_family, 8491 2, target_size)); 8492 break; 8493 case offsetof(struct __sk_buff, remote_ip4): 8494 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 8495 8496 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8497 si->dst_reg, si->src_reg, 8498 offsetof(struct sk_buff, sk)); 8499 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8500 bpf_target_off(struct sock_common, 8501 skc_daddr, 8502 4, target_size)); 8503 break; 8504 case offsetof(struct __sk_buff, local_ip4): 8505 BUILD_BUG_ON(sizeof_field(struct sock_common, 8506 skc_rcv_saddr) != 4); 8507 8508 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8509 si->dst_reg, si->src_reg, 8510 offsetof(struct sk_buff, sk)); 8511 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8512 bpf_target_off(struct sock_common, 8513 skc_rcv_saddr, 8514 4, target_size)); 8515 break; 8516 case offsetof(struct __sk_buff, remote_ip6[0]) ... 8517 offsetof(struct __sk_buff, remote_ip6[3]): 8518 #if IS_ENABLED(CONFIG_IPV6) 8519 BUILD_BUG_ON(sizeof_field(struct sock_common, 8520 skc_v6_daddr.s6_addr32[0]) != 4); 8521 8522 off = si->off; 8523 off -= offsetof(struct __sk_buff, remote_ip6[0]); 8524 8525 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8526 si->dst_reg, si->src_reg, 8527 offsetof(struct sk_buff, sk)); 8528 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8529 offsetof(struct sock_common, 8530 skc_v6_daddr.s6_addr32[0]) + 8531 off); 8532 #else 8533 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8534 #endif 8535 break; 8536 case offsetof(struct __sk_buff, local_ip6[0]) ... 8537 offsetof(struct __sk_buff, local_ip6[3]): 8538 #if IS_ENABLED(CONFIG_IPV6) 8539 BUILD_BUG_ON(sizeof_field(struct sock_common, 8540 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 8541 8542 off = si->off; 8543 off -= offsetof(struct __sk_buff, local_ip6[0]); 8544 8545 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8546 si->dst_reg, si->src_reg, 8547 offsetof(struct sk_buff, sk)); 8548 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8549 offsetof(struct sock_common, 8550 skc_v6_rcv_saddr.s6_addr32[0]) + 8551 off); 8552 #else 8553 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8554 #endif 8555 break; 8556 8557 case offsetof(struct __sk_buff, remote_port): 8558 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 8559 8560 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8561 si->dst_reg, si->src_reg, 8562 offsetof(struct sk_buff, sk)); 8563 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8564 bpf_target_off(struct sock_common, 8565 skc_dport, 8566 2, target_size)); 8567 #ifndef __BIG_ENDIAN_BITFIELD 8568 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 8569 #endif 8570 break; 8571 8572 case offsetof(struct __sk_buff, local_port): 8573 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 8574 8575 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8576 si->dst_reg, si->src_reg, 8577 offsetof(struct sk_buff, sk)); 8578 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8579 bpf_target_off(struct sock_common, 8580 skc_num, 2, target_size)); 8581 break; 8582 8583 case offsetof(struct __sk_buff, tstamp): 8584 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8); 8585 8586 if (type == BPF_WRITE) 8587 *insn++ = BPF_STX_MEM(BPF_DW, 8588 si->dst_reg, si->src_reg, 8589 bpf_target_off(struct sk_buff, 8590 tstamp, 8, 8591 target_size)); 8592 else 8593 *insn++ = BPF_LDX_MEM(BPF_DW, 8594 si->dst_reg, si->src_reg, 8595 bpf_target_off(struct sk_buff, 8596 tstamp, 8, 8597 target_size)); 8598 break; 8599 8600 case offsetof(struct __sk_buff, gso_segs): 8601 insn = bpf_convert_shinfo_access(si, insn); 8602 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs), 8603 si->dst_reg, si->dst_reg, 8604 bpf_target_off(struct skb_shared_info, 8605 gso_segs, 2, 8606 target_size)); 8607 break; 8608 case offsetof(struct __sk_buff, gso_size): 8609 insn = bpf_convert_shinfo_access(si, insn); 8610 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size), 8611 si->dst_reg, si->dst_reg, 8612 bpf_target_off(struct skb_shared_info, 8613 gso_size, 2, 8614 target_size)); 8615 break; 8616 case offsetof(struct __sk_buff, wire_len): 8617 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4); 8618 8619 off = si->off; 8620 off -= offsetof(struct __sk_buff, wire_len); 8621 off += offsetof(struct sk_buff, cb); 8622 off += offsetof(struct qdisc_skb_cb, pkt_len); 8623 *target_size = 4; 8624 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off); 8625 break; 8626 8627 case offsetof(struct __sk_buff, sk): 8628 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8629 si->dst_reg, si->src_reg, 8630 offsetof(struct sk_buff, sk)); 8631 break; 8632 } 8633 8634 return insn - insn_buf; 8635 } 8636 8637 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 8638 const struct bpf_insn *si, 8639 struct bpf_insn *insn_buf, 8640 struct bpf_prog *prog, u32 *target_size) 8641 { 8642 struct bpf_insn *insn = insn_buf; 8643 int off; 8644 8645 switch (si->off) { 8646 case offsetof(struct bpf_sock, bound_dev_if): 8647 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4); 8648 8649 if (type == BPF_WRITE) 8650 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8651 offsetof(struct sock, sk_bound_dev_if)); 8652 else 8653 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8654 offsetof(struct sock, sk_bound_dev_if)); 8655 break; 8656 8657 case offsetof(struct bpf_sock, mark): 8658 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4); 8659 8660 if (type == BPF_WRITE) 8661 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8662 offsetof(struct sock, sk_mark)); 8663 else 8664 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8665 offsetof(struct sock, sk_mark)); 8666 break; 8667 8668 case offsetof(struct bpf_sock, priority): 8669 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4); 8670 8671 if (type == BPF_WRITE) 8672 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8673 offsetof(struct sock, sk_priority)); 8674 else 8675 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8676 offsetof(struct sock, sk_priority)); 8677 break; 8678 8679 case offsetof(struct bpf_sock, family): 8680 *insn++ = BPF_LDX_MEM( 8681 BPF_FIELD_SIZEOF(struct sock_common, skc_family), 8682 si->dst_reg, si->src_reg, 8683 bpf_target_off(struct sock_common, 8684 skc_family, 8685 sizeof_field(struct sock_common, 8686 skc_family), 8687 target_size)); 8688 break; 8689 8690 case offsetof(struct bpf_sock, type): 8691 *insn++ = BPF_LDX_MEM( 8692 BPF_FIELD_SIZEOF(struct sock, sk_type), 8693 si->dst_reg, si->src_reg, 8694 bpf_target_off(struct sock, sk_type, 8695 sizeof_field(struct sock, sk_type), 8696 target_size)); 8697 break; 8698 8699 case offsetof(struct bpf_sock, protocol): 8700 *insn++ = BPF_LDX_MEM( 8701 BPF_FIELD_SIZEOF(struct sock, sk_protocol), 8702 si->dst_reg, si->src_reg, 8703 bpf_target_off(struct sock, sk_protocol, 8704 sizeof_field(struct sock, sk_protocol), 8705 target_size)); 8706 break; 8707 8708 case offsetof(struct bpf_sock, src_ip4): 8709 *insn++ = BPF_LDX_MEM( 8710 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 8711 bpf_target_off(struct sock_common, skc_rcv_saddr, 8712 sizeof_field(struct sock_common, 8713 skc_rcv_saddr), 8714 target_size)); 8715 break; 8716 8717 case offsetof(struct bpf_sock, dst_ip4): 8718 *insn++ = BPF_LDX_MEM( 8719 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 8720 bpf_target_off(struct sock_common, skc_daddr, 8721 sizeof_field(struct sock_common, 8722 skc_daddr), 8723 target_size)); 8724 break; 8725 8726 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 8727 #if IS_ENABLED(CONFIG_IPV6) 8728 off = si->off; 8729 off -= offsetof(struct bpf_sock, src_ip6[0]); 8730 *insn++ = BPF_LDX_MEM( 8731 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 8732 bpf_target_off( 8733 struct sock_common, 8734 skc_v6_rcv_saddr.s6_addr32[0], 8735 sizeof_field(struct sock_common, 8736 skc_v6_rcv_saddr.s6_addr32[0]), 8737 target_size) + off); 8738 #else 8739 (void)off; 8740 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8741 #endif 8742 break; 8743 8744 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 8745 #if IS_ENABLED(CONFIG_IPV6) 8746 off = si->off; 8747 off -= offsetof(struct bpf_sock, dst_ip6[0]); 8748 *insn++ = BPF_LDX_MEM( 8749 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 8750 bpf_target_off(struct sock_common, 8751 skc_v6_daddr.s6_addr32[0], 8752 sizeof_field(struct sock_common, 8753 skc_v6_daddr.s6_addr32[0]), 8754 target_size) + off); 8755 #else 8756 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8757 *target_size = 4; 8758 #endif 8759 break; 8760 8761 case offsetof(struct bpf_sock, src_port): 8762 *insn++ = BPF_LDX_MEM( 8763 BPF_FIELD_SIZEOF(struct sock_common, skc_num), 8764 si->dst_reg, si->src_reg, 8765 bpf_target_off(struct sock_common, skc_num, 8766 sizeof_field(struct sock_common, 8767 skc_num), 8768 target_size)); 8769 break; 8770 8771 case offsetof(struct bpf_sock, dst_port): 8772 *insn++ = BPF_LDX_MEM( 8773 BPF_FIELD_SIZEOF(struct sock_common, skc_dport), 8774 si->dst_reg, si->src_reg, 8775 bpf_target_off(struct sock_common, skc_dport, 8776 sizeof_field(struct sock_common, 8777 skc_dport), 8778 target_size)); 8779 break; 8780 8781 case offsetof(struct bpf_sock, state): 8782 *insn++ = BPF_LDX_MEM( 8783 BPF_FIELD_SIZEOF(struct sock_common, skc_state), 8784 si->dst_reg, si->src_reg, 8785 bpf_target_off(struct sock_common, skc_state, 8786 sizeof_field(struct sock_common, 8787 skc_state), 8788 target_size)); 8789 break; 8790 case offsetof(struct bpf_sock, rx_queue_mapping): 8791 #ifdef CONFIG_XPS 8792 *insn++ = BPF_LDX_MEM( 8793 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping), 8794 si->dst_reg, si->src_reg, 8795 bpf_target_off(struct sock, sk_rx_queue_mapping, 8796 sizeof_field(struct sock, 8797 sk_rx_queue_mapping), 8798 target_size)); 8799 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING, 8800 1); 8801 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 8802 #else 8803 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 8804 *target_size = 2; 8805 #endif 8806 break; 8807 } 8808 8809 return insn - insn_buf; 8810 } 8811 8812 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, 8813 const struct bpf_insn *si, 8814 struct bpf_insn *insn_buf, 8815 struct bpf_prog *prog, u32 *target_size) 8816 { 8817 struct bpf_insn *insn = insn_buf; 8818 8819 switch (si->off) { 8820 case offsetof(struct __sk_buff, ifindex): 8821 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 8822 si->dst_reg, si->src_reg, 8823 offsetof(struct sk_buff, dev)); 8824 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8825 bpf_target_off(struct net_device, ifindex, 4, 8826 target_size)); 8827 break; 8828 default: 8829 return bpf_convert_ctx_access(type, si, insn_buf, prog, 8830 target_size); 8831 } 8832 8833 return insn - insn_buf; 8834 } 8835 8836 static u32 xdp_convert_ctx_access(enum bpf_access_type type, 8837 const struct bpf_insn *si, 8838 struct bpf_insn *insn_buf, 8839 struct bpf_prog *prog, u32 *target_size) 8840 { 8841 struct bpf_insn *insn = insn_buf; 8842 8843 switch (si->off) { 8844 case offsetof(struct xdp_md, data): 8845 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data), 8846 si->dst_reg, si->src_reg, 8847 offsetof(struct xdp_buff, data)); 8848 break; 8849 case offsetof(struct xdp_md, data_meta): 8850 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta), 8851 si->dst_reg, si->src_reg, 8852 offsetof(struct xdp_buff, data_meta)); 8853 break; 8854 case offsetof(struct xdp_md, data_end): 8855 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end), 8856 si->dst_reg, si->src_reg, 8857 offsetof(struct xdp_buff, data_end)); 8858 break; 8859 case offsetof(struct xdp_md, ingress_ifindex): 8860 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 8861 si->dst_reg, si->src_reg, 8862 offsetof(struct xdp_buff, rxq)); 8863 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev), 8864 si->dst_reg, si->dst_reg, 8865 offsetof(struct xdp_rxq_info, dev)); 8866 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8867 offsetof(struct net_device, ifindex)); 8868 break; 8869 case offsetof(struct xdp_md, rx_queue_index): 8870 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 8871 si->dst_reg, si->src_reg, 8872 offsetof(struct xdp_buff, rxq)); 8873 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8874 offsetof(struct xdp_rxq_info, 8875 queue_index)); 8876 break; 8877 case offsetof(struct xdp_md, egress_ifindex): 8878 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq), 8879 si->dst_reg, si->src_reg, 8880 offsetof(struct xdp_buff, txq)); 8881 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev), 8882 si->dst_reg, si->dst_reg, 8883 offsetof(struct xdp_txq_info, dev)); 8884 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8885 offsetof(struct net_device, ifindex)); 8886 break; 8887 } 8888 8889 return insn - insn_buf; 8890 } 8891 8892 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of 8893 * context Structure, F is Field in context structure that contains a pointer 8894 * to Nested Structure of type NS that has the field NF. 8895 * 8896 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make 8897 * sure that SIZE is not greater than actual size of S.F.NF. 8898 * 8899 * If offset OFF is provided, the load happens from that offset relative to 8900 * offset of NF. 8901 */ 8902 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \ 8903 do { \ 8904 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \ 8905 si->src_reg, offsetof(S, F)); \ 8906 *insn++ = BPF_LDX_MEM( \ 8907 SIZE, si->dst_reg, si->dst_reg, \ 8908 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 8909 target_size) \ 8910 + OFF); \ 8911 } while (0) 8912 8913 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \ 8914 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \ 8915 BPF_FIELD_SIZEOF(NS, NF), 0) 8916 8917 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to 8918 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation. 8919 * 8920 * In addition it uses Temporary Field TF (member of struct S) as the 3rd 8921 * "register" since two registers available in convert_ctx_access are not 8922 * enough: we can't override neither SRC, since it contains value to store, nor 8923 * DST since it contains pointer to context that may be used by later 8924 * instructions. But we need a temporary place to save pointer to nested 8925 * structure whose field we want to store to. 8926 */ 8927 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \ 8928 do { \ 8929 int tmp_reg = BPF_REG_9; \ 8930 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 8931 --tmp_reg; \ 8932 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 8933 --tmp_reg; \ 8934 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \ 8935 offsetof(S, TF)); \ 8936 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \ 8937 si->dst_reg, offsetof(S, F)); \ 8938 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \ 8939 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 8940 target_size) \ 8941 + OFF); \ 8942 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \ 8943 offsetof(S, TF)); \ 8944 } while (0) 8945 8946 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \ 8947 TF) \ 8948 do { \ 8949 if (type == BPF_WRITE) { \ 8950 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \ 8951 OFF, TF); \ 8952 } else { \ 8953 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \ 8954 S, NS, F, NF, SIZE, OFF); \ 8955 } \ 8956 } while (0) 8957 8958 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \ 8959 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \ 8960 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF) 8961 8962 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type, 8963 const struct bpf_insn *si, 8964 struct bpf_insn *insn_buf, 8965 struct bpf_prog *prog, u32 *target_size) 8966 { 8967 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port); 8968 struct bpf_insn *insn = insn_buf; 8969 8970 switch (si->off) { 8971 case offsetof(struct bpf_sock_addr, user_family): 8972 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 8973 struct sockaddr, uaddr, sa_family); 8974 break; 8975 8976 case offsetof(struct bpf_sock_addr, user_ip4): 8977 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 8978 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr, 8979 sin_addr, BPF_SIZE(si->code), 0, tmp_reg); 8980 break; 8981 8982 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8983 off = si->off; 8984 off -= offsetof(struct bpf_sock_addr, user_ip6[0]); 8985 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 8986 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 8987 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off, 8988 tmp_reg); 8989 break; 8990 8991 case offsetof(struct bpf_sock_addr, user_port): 8992 /* To get port we need to know sa_family first and then treat 8993 * sockaddr as either sockaddr_in or sockaddr_in6. 8994 * Though we can simplify since port field has same offset and 8995 * size in both structures. 8996 * Here we check this invariant and use just one of the 8997 * structures if it's true. 8998 */ 8999 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) != 9000 offsetof(struct sockaddr_in6, sin6_port)); 9001 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) != 9002 sizeof_field(struct sockaddr_in6, sin6_port)); 9003 /* Account for sin6_port being smaller than user_port. */ 9004 port_size = min(port_size, BPF_LDST_BYTES(si)); 9005 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9006 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 9007 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg); 9008 break; 9009 9010 case offsetof(struct bpf_sock_addr, family): 9011 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9012 struct sock, sk, sk_family); 9013 break; 9014 9015 case offsetof(struct bpf_sock_addr, type): 9016 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9017 struct sock, sk, sk_type); 9018 break; 9019 9020 case offsetof(struct bpf_sock_addr, protocol): 9021 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9022 struct sock, sk, sk_protocol); 9023 break; 9024 9025 case offsetof(struct bpf_sock_addr, msg_src_ip4): 9026 /* Treat t_ctx as struct in_addr for msg_src_ip4. */ 9027 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9028 struct bpf_sock_addr_kern, struct in_addr, t_ctx, 9029 s_addr, BPF_SIZE(si->code), 0, tmp_reg); 9030 break; 9031 9032 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 9033 msg_src_ip6[3]): 9034 off = si->off; 9035 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]); 9036 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */ 9037 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9038 struct bpf_sock_addr_kern, struct in6_addr, t_ctx, 9039 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg); 9040 break; 9041 case offsetof(struct bpf_sock_addr, sk): 9042 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk), 9043 si->dst_reg, si->src_reg, 9044 offsetof(struct bpf_sock_addr_kern, sk)); 9045 break; 9046 } 9047 9048 return insn - insn_buf; 9049 } 9050 9051 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type, 9052 const struct bpf_insn *si, 9053 struct bpf_insn *insn_buf, 9054 struct bpf_prog *prog, 9055 u32 *target_size) 9056 { 9057 struct bpf_insn *insn = insn_buf; 9058 int off; 9059 9060 /* Helper macro for adding read access to tcp_sock or sock fields. */ 9061 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 9062 do { \ 9063 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \ 9064 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 9065 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 9066 if (si->dst_reg == reg || si->src_reg == reg) \ 9067 reg--; \ 9068 if (si->dst_reg == reg || si->src_reg == reg) \ 9069 reg--; \ 9070 if (si->dst_reg == si->src_reg) { \ 9071 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 9072 offsetof(struct bpf_sock_ops_kern, \ 9073 temp)); \ 9074 fullsock_reg = reg; \ 9075 jmp += 2; \ 9076 } \ 9077 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9078 struct bpf_sock_ops_kern, \ 9079 is_fullsock), \ 9080 fullsock_reg, si->src_reg, \ 9081 offsetof(struct bpf_sock_ops_kern, \ 9082 is_fullsock)); \ 9083 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 9084 if (si->dst_reg == si->src_reg) \ 9085 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9086 offsetof(struct bpf_sock_ops_kern, \ 9087 temp)); \ 9088 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9089 struct bpf_sock_ops_kern, sk),\ 9090 si->dst_reg, si->src_reg, \ 9091 offsetof(struct bpf_sock_ops_kern, sk));\ 9092 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \ 9093 OBJ_FIELD), \ 9094 si->dst_reg, si->dst_reg, \ 9095 offsetof(OBJ, OBJ_FIELD)); \ 9096 if (si->dst_reg == si->src_reg) { \ 9097 *insn++ = BPF_JMP_A(1); \ 9098 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9099 offsetof(struct bpf_sock_ops_kern, \ 9100 temp)); \ 9101 } \ 9102 } while (0) 9103 9104 #define SOCK_OPS_GET_SK() \ 9105 do { \ 9106 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \ 9107 if (si->dst_reg == reg || si->src_reg == reg) \ 9108 reg--; \ 9109 if (si->dst_reg == reg || si->src_reg == reg) \ 9110 reg--; \ 9111 if (si->dst_reg == si->src_reg) { \ 9112 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 9113 offsetof(struct bpf_sock_ops_kern, \ 9114 temp)); \ 9115 fullsock_reg = reg; \ 9116 jmp += 2; \ 9117 } \ 9118 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9119 struct bpf_sock_ops_kern, \ 9120 is_fullsock), \ 9121 fullsock_reg, si->src_reg, \ 9122 offsetof(struct bpf_sock_ops_kern, \ 9123 is_fullsock)); \ 9124 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 9125 if (si->dst_reg == si->src_reg) \ 9126 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9127 offsetof(struct bpf_sock_ops_kern, \ 9128 temp)); \ 9129 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9130 struct bpf_sock_ops_kern, sk),\ 9131 si->dst_reg, si->src_reg, \ 9132 offsetof(struct bpf_sock_ops_kern, sk));\ 9133 if (si->dst_reg == si->src_reg) { \ 9134 *insn++ = BPF_JMP_A(1); \ 9135 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9136 offsetof(struct bpf_sock_ops_kern, \ 9137 temp)); \ 9138 } \ 9139 } while (0) 9140 9141 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \ 9142 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock) 9143 9144 /* Helper macro for adding write access to tcp_sock or sock fields. 9145 * The macro is called with two registers, dst_reg which contains a pointer 9146 * to ctx (context) and src_reg which contains the value that should be 9147 * stored. However, we need an additional register since we cannot overwrite 9148 * dst_reg because it may be used later in the program. 9149 * Instead we "borrow" one of the other register. We first save its value 9150 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore 9151 * it at the end of the macro. 9152 */ 9153 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 9154 do { \ 9155 int reg = BPF_REG_9; \ 9156 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 9157 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 9158 if (si->dst_reg == reg || si->src_reg == reg) \ 9159 reg--; \ 9160 if (si->dst_reg == reg || si->src_reg == reg) \ 9161 reg--; \ 9162 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \ 9163 offsetof(struct bpf_sock_ops_kern, \ 9164 temp)); \ 9165 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9166 struct bpf_sock_ops_kern, \ 9167 is_fullsock), \ 9168 reg, si->dst_reg, \ 9169 offsetof(struct bpf_sock_ops_kern, \ 9170 is_fullsock)); \ 9171 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \ 9172 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9173 struct bpf_sock_ops_kern, sk),\ 9174 reg, si->dst_reg, \ 9175 offsetof(struct bpf_sock_ops_kern, sk));\ 9176 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \ 9177 reg, si->src_reg, \ 9178 offsetof(OBJ, OBJ_FIELD)); \ 9179 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \ 9180 offsetof(struct bpf_sock_ops_kern, \ 9181 temp)); \ 9182 } while (0) 9183 9184 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \ 9185 do { \ 9186 if (TYPE == BPF_WRITE) \ 9187 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 9188 else \ 9189 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 9190 } while (0) 9191 9192 if (insn > insn_buf) 9193 return insn - insn_buf; 9194 9195 switch (si->off) { 9196 case offsetof(struct bpf_sock_ops, op): 9197 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9198 op), 9199 si->dst_reg, si->src_reg, 9200 offsetof(struct bpf_sock_ops_kern, op)); 9201 break; 9202 9203 case offsetof(struct bpf_sock_ops, replylong[0]) ... 9204 offsetof(struct bpf_sock_ops, replylong[3]): 9205 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) != 9206 sizeof_field(struct bpf_sock_ops_kern, reply)); 9207 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) != 9208 sizeof_field(struct bpf_sock_ops_kern, replylong)); 9209 off = si->off; 9210 off -= offsetof(struct bpf_sock_ops, replylong[0]); 9211 off += offsetof(struct bpf_sock_ops_kern, replylong[0]); 9212 if (type == BPF_WRITE) 9213 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9214 off); 9215 else 9216 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9217 off); 9218 break; 9219 9220 case offsetof(struct bpf_sock_ops, family): 9221 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 9222 9223 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9224 struct bpf_sock_ops_kern, sk), 9225 si->dst_reg, si->src_reg, 9226 offsetof(struct bpf_sock_ops_kern, sk)); 9227 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9228 offsetof(struct sock_common, skc_family)); 9229 break; 9230 9231 case offsetof(struct bpf_sock_ops, remote_ip4): 9232 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 9233 9234 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9235 struct bpf_sock_ops_kern, sk), 9236 si->dst_reg, si->src_reg, 9237 offsetof(struct bpf_sock_ops_kern, sk)); 9238 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9239 offsetof(struct sock_common, skc_daddr)); 9240 break; 9241 9242 case offsetof(struct bpf_sock_ops, local_ip4): 9243 BUILD_BUG_ON(sizeof_field(struct sock_common, 9244 skc_rcv_saddr) != 4); 9245 9246 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9247 struct bpf_sock_ops_kern, sk), 9248 si->dst_reg, si->src_reg, 9249 offsetof(struct bpf_sock_ops_kern, sk)); 9250 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9251 offsetof(struct sock_common, 9252 skc_rcv_saddr)); 9253 break; 9254 9255 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ... 9256 offsetof(struct bpf_sock_ops, remote_ip6[3]): 9257 #if IS_ENABLED(CONFIG_IPV6) 9258 BUILD_BUG_ON(sizeof_field(struct sock_common, 9259 skc_v6_daddr.s6_addr32[0]) != 4); 9260 9261 off = si->off; 9262 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]); 9263 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9264 struct bpf_sock_ops_kern, sk), 9265 si->dst_reg, si->src_reg, 9266 offsetof(struct bpf_sock_ops_kern, sk)); 9267 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9268 offsetof(struct sock_common, 9269 skc_v6_daddr.s6_addr32[0]) + 9270 off); 9271 #else 9272 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9273 #endif 9274 break; 9275 9276 case offsetof(struct bpf_sock_ops, local_ip6[0]) ... 9277 offsetof(struct bpf_sock_ops, local_ip6[3]): 9278 #if IS_ENABLED(CONFIG_IPV6) 9279 BUILD_BUG_ON(sizeof_field(struct sock_common, 9280 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 9281 9282 off = si->off; 9283 off -= offsetof(struct bpf_sock_ops, local_ip6[0]); 9284 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9285 struct bpf_sock_ops_kern, sk), 9286 si->dst_reg, si->src_reg, 9287 offsetof(struct bpf_sock_ops_kern, sk)); 9288 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9289 offsetof(struct sock_common, 9290 skc_v6_rcv_saddr.s6_addr32[0]) + 9291 off); 9292 #else 9293 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9294 #endif 9295 break; 9296 9297 case offsetof(struct bpf_sock_ops, remote_port): 9298 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 9299 9300 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9301 struct bpf_sock_ops_kern, sk), 9302 si->dst_reg, si->src_reg, 9303 offsetof(struct bpf_sock_ops_kern, sk)); 9304 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9305 offsetof(struct sock_common, skc_dport)); 9306 #ifndef __BIG_ENDIAN_BITFIELD 9307 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 9308 #endif 9309 break; 9310 9311 case offsetof(struct bpf_sock_ops, local_port): 9312 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 9313 9314 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9315 struct bpf_sock_ops_kern, sk), 9316 si->dst_reg, si->src_reg, 9317 offsetof(struct bpf_sock_ops_kern, sk)); 9318 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9319 offsetof(struct sock_common, skc_num)); 9320 break; 9321 9322 case offsetof(struct bpf_sock_ops, is_fullsock): 9323 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9324 struct bpf_sock_ops_kern, 9325 is_fullsock), 9326 si->dst_reg, si->src_reg, 9327 offsetof(struct bpf_sock_ops_kern, 9328 is_fullsock)); 9329 break; 9330 9331 case offsetof(struct bpf_sock_ops, state): 9332 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1); 9333 9334 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9335 struct bpf_sock_ops_kern, sk), 9336 si->dst_reg, si->src_reg, 9337 offsetof(struct bpf_sock_ops_kern, sk)); 9338 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg, 9339 offsetof(struct sock_common, skc_state)); 9340 break; 9341 9342 case offsetof(struct bpf_sock_ops, rtt_min): 9343 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 9344 sizeof(struct minmax)); 9345 BUILD_BUG_ON(sizeof(struct minmax) < 9346 sizeof(struct minmax_sample)); 9347 9348 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9349 struct bpf_sock_ops_kern, sk), 9350 si->dst_reg, si->src_reg, 9351 offsetof(struct bpf_sock_ops_kern, sk)); 9352 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9353 offsetof(struct tcp_sock, rtt_min) + 9354 sizeof_field(struct minmax_sample, t)); 9355 break; 9356 9357 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags): 9358 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags, 9359 struct tcp_sock); 9360 break; 9361 9362 case offsetof(struct bpf_sock_ops, sk_txhash): 9363 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash, 9364 struct sock, type); 9365 break; 9366 case offsetof(struct bpf_sock_ops, snd_cwnd): 9367 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd); 9368 break; 9369 case offsetof(struct bpf_sock_ops, srtt_us): 9370 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us); 9371 break; 9372 case offsetof(struct bpf_sock_ops, snd_ssthresh): 9373 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh); 9374 break; 9375 case offsetof(struct bpf_sock_ops, rcv_nxt): 9376 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt); 9377 break; 9378 case offsetof(struct bpf_sock_ops, snd_nxt): 9379 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt); 9380 break; 9381 case offsetof(struct bpf_sock_ops, snd_una): 9382 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una); 9383 break; 9384 case offsetof(struct bpf_sock_ops, mss_cache): 9385 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache); 9386 break; 9387 case offsetof(struct bpf_sock_ops, ecn_flags): 9388 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags); 9389 break; 9390 case offsetof(struct bpf_sock_ops, rate_delivered): 9391 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered); 9392 break; 9393 case offsetof(struct bpf_sock_ops, rate_interval_us): 9394 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us); 9395 break; 9396 case offsetof(struct bpf_sock_ops, packets_out): 9397 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out); 9398 break; 9399 case offsetof(struct bpf_sock_ops, retrans_out): 9400 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out); 9401 break; 9402 case offsetof(struct bpf_sock_ops, total_retrans): 9403 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans); 9404 break; 9405 case offsetof(struct bpf_sock_ops, segs_in): 9406 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in); 9407 break; 9408 case offsetof(struct bpf_sock_ops, data_segs_in): 9409 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in); 9410 break; 9411 case offsetof(struct bpf_sock_ops, segs_out): 9412 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out); 9413 break; 9414 case offsetof(struct bpf_sock_ops, data_segs_out): 9415 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out); 9416 break; 9417 case offsetof(struct bpf_sock_ops, lost_out): 9418 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out); 9419 break; 9420 case offsetof(struct bpf_sock_ops, sacked_out): 9421 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out); 9422 break; 9423 case offsetof(struct bpf_sock_ops, bytes_received): 9424 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received); 9425 break; 9426 case offsetof(struct bpf_sock_ops, bytes_acked): 9427 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked); 9428 break; 9429 case offsetof(struct bpf_sock_ops, sk): 9430 SOCK_OPS_GET_SK(); 9431 break; 9432 case offsetof(struct bpf_sock_ops, skb_data_end): 9433 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9434 skb_data_end), 9435 si->dst_reg, si->src_reg, 9436 offsetof(struct bpf_sock_ops_kern, 9437 skb_data_end)); 9438 break; 9439 case offsetof(struct bpf_sock_ops, skb_data): 9440 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9441 skb), 9442 si->dst_reg, si->src_reg, 9443 offsetof(struct bpf_sock_ops_kern, 9444 skb)); 9445 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9446 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 9447 si->dst_reg, si->dst_reg, 9448 offsetof(struct sk_buff, data)); 9449 break; 9450 case offsetof(struct bpf_sock_ops, skb_len): 9451 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9452 skb), 9453 si->dst_reg, si->src_reg, 9454 offsetof(struct bpf_sock_ops_kern, 9455 skb)); 9456 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9457 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 9458 si->dst_reg, si->dst_reg, 9459 offsetof(struct sk_buff, len)); 9460 break; 9461 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 9462 off = offsetof(struct sk_buff, cb); 9463 off += offsetof(struct tcp_skb_cb, tcp_flags); 9464 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags); 9465 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9466 skb), 9467 si->dst_reg, si->src_reg, 9468 offsetof(struct bpf_sock_ops_kern, 9469 skb)); 9470 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9471 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb, 9472 tcp_flags), 9473 si->dst_reg, si->dst_reg, off); 9474 break; 9475 } 9476 return insn - insn_buf; 9477 } 9478 9479 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type, 9480 const struct bpf_insn *si, 9481 struct bpf_insn *insn_buf, 9482 struct bpf_prog *prog, u32 *target_size) 9483 { 9484 struct bpf_insn *insn = insn_buf; 9485 int off; 9486 9487 switch (si->off) { 9488 case offsetof(struct __sk_buff, data_end): 9489 off = si->off; 9490 off -= offsetof(struct __sk_buff, data_end); 9491 off += offsetof(struct sk_buff, cb); 9492 off += offsetof(struct tcp_skb_cb, bpf.data_end); 9493 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 9494 si->src_reg, off); 9495 break; 9496 default: 9497 return bpf_convert_ctx_access(type, si, insn_buf, prog, 9498 target_size); 9499 } 9500 9501 return insn - insn_buf; 9502 } 9503 9504 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type, 9505 const struct bpf_insn *si, 9506 struct bpf_insn *insn_buf, 9507 struct bpf_prog *prog, u32 *target_size) 9508 { 9509 struct bpf_insn *insn = insn_buf; 9510 #if IS_ENABLED(CONFIG_IPV6) 9511 int off; 9512 #endif 9513 9514 /* convert ctx uses the fact sg element is first in struct */ 9515 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0); 9516 9517 switch (si->off) { 9518 case offsetof(struct sk_msg_md, data): 9519 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data), 9520 si->dst_reg, si->src_reg, 9521 offsetof(struct sk_msg, data)); 9522 break; 9523 case offsetof(struct sk_msg_md, data_end): 9524 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end), 9525 si->dst_reg, si->src_reg, 9526 offsetof(struct sk_msg, data_end)); 9527 break; 9528 case offsetof(struct sk_msg_md, family): 9529 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 9530 9531 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9532 struct sk_msg, sk), 9533 si->dst_reg, si->src_reg, 9534 offsetof(struct sk_msg, sk)); 9535 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9536 offsetof(struct sock_common, skc_family)); 9537 break; 9538 9539 case offsetof(struct sk_msg_md, remote_ip4): 9540 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 9541 9542 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9543 struct sk_msg, sk), 9544 si->dst_reg, si->src_reg, 9545 offsetof(struct sk_msg, sk)); 9546 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9547 offsetof(struct sock_common, skc_daddr)); 9548 break; 9549 9550 case offsetof(struct sk_msg_md, local_ip4): 9551 BUILD_BUG_ON(sizeof_field(struct sock_common, 9552 skc_rcv_saddr) != 4); 9553 9554 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9555 struct sk_msg, sk), 9556 si->dst_reg, si->src_reg, 9557 offsetof(struct sk_msg, sk)); 9558 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9559 offsetof(struct sock_common, 9560 skc_rcv_saddr)); 9561 break; 9562 9563 case offsetof(struct sk_msg_md, remote_ip6[0]) ... 9564 offsetof(struct sk_msg_md, remote_ip6[3]): 9565 #if IS_ENABLED(CONFIG_IPV6) 9566 BUILD_BUG_ON(sizeof_field(struct sock_common, 9567 skc_v6_daddr.s6_addr32[0]) != 4); 9568 9569 off = si->off; 9570 off -= offsetof(struct sk_msg_md, remote_ip6[0]); 9571 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9572 struct sk_msg, sk), 9573 si->dst_reg, si->src_reg, 9574 offsetof(struct sk_msg, sk)); 9575 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9576 offsetof(struct sock_common, 9577 skc_v6_daddr.s6_addr32[0]) + 9578 off); 9579 #else 9580 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9581 #endif 9582 break; 9583 9584 case offsetof(struct sk_msg_md, local_ip6[0]) ... 9585 offsetof(struct sk_msg_md, local_ip6[3]): 9586 #if IS_ENABLED(CONFIG_IPV6) 9587 BUILD_BUG_ON(sizeof_field(struct sock_common, 9588 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 9589 9590 off = si->off; 9591 off -= offsetof(struct sk_msg_md, local_ip6[0]); 9592 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9593 struct sk_msg, sk), 9594 si->dst_reg, si->src_reg, 9595 offsetof(struct sk_msg, sk)); 9596 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9597 offsetof(struct sock_common, 9598 skc_v6_rcv_saddr.s6_addr32[0]) + 9599 off); 9600 #else 9601 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9602 #endif 9603 break; 9604 9605 case offsetof(struct sk_msg_md, remote_port): 9606 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 9607 9608 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9609 struct sk_msg, sk), 9610 si->dst_reg, si->src_reg, 9611 offsetof(struct sk_msg, sk)); 9612 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9613 offsetof(struct sock_common, skc_dport)); 9614 #ifndef __BIG_ENDIAN_BITFIELD 9615 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 9616 #endif 9617 break; 9618 9619 case offsetof(struct sk_msg_md, local_port): 9620 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 9621 9622 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9623 struct sk_msg, sk), 9624 si->dst_reg, si->src_reg, 9625 offsetof(struct sk_msg, sk)); 9626 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9627 offsetof(struct sock_common, skc_num)); 9628 break; 9629 9630 case offsetof(struct sk_msg_md, size): 9631 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size), 9632 si->dst_reg, si->src_reg, 9633 offsetof(struct sk_msg_sg, size)); 9634 break; 9635 9636 case offsetof(struct sk_msg_md, sk): 9637 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk), 9638 si->dst_reg, si->src_reg, 9639 offsetof(struct sk_msg, sk)); 9640 break; 9641 } 9642 9643 return insn - insn_buf; 9644 } 9645 9646 const struct bpf_verifier_ops sk_filter_verifier_ops = { 9647 .get_func_proto = sk_filter_func_proto, 9648 .is_valid_access = sk_filter_is_valid_access, 9649 .convert_ctx_access = bpf_convert_ctx_access, 9650 .gen_ld_abs = bpf_gen_ld_abs, 9651 }; 9652 9653 const struct bpf_prog_ops sk_filter_prog_ops = { 9654 .test_run = bpf_prog_test_run_skb, 9655 }; 9656 9657 const struct bpf_verifier_ops tc_cls_act_verifier_ops = { 9658 .get_func_proto = tc_cls_act_func_proto, 9659 .is_valid_access = tc_cls_act_is_valid_access, 9660 .convert_ctx_access = tc_cls_act_convert_ctx_access, 9661 .gen_prologue = tc_cls_act_prologue, 9662 .gen_ld_abs = bpf_gen_ld_abs, 9663 }; 9664 9665 const struct bpf_prog_ops tc_cls_act_prog_ops = { 9666 .test_run = bpf_prog_test_run_skb, 9667 }; 9668 9669 const struct bpf_verifier_ops xdp_verifier_ops = { 9670 .get_func_proto = xdp_func_proto, 9671 .is_valid_access = xdp_is_valid_access, 9672 .convert_ctx_access = xdp_convert_ctx_access, 9673 .gen_prologue = bpf_noop_prologue, 9674 }; 9675 9676 const struct bpf_prog_ops xdp_prog_ops = { 9677 .test_run = bpf_prog_test_run_xdp, 9678 }; 9679 9680 const struct bpf_verifier_ops cg_skb_verifier_ops = { 9681 .get_func_proto = cg_skb_func_proto, 9682 .is_valid_access = cg_skb_is_valid_access, 9683 .convert_ctx_access = bpf_convert_ctx_access, 9684 }; 9685 9686 const struct bpf_prog_ops cg_skb_prog_ops = { 9687 .test_run = bpf_prog_test_run_skb, 9688 }; 9689 9690 const struct bpf_verifier_ops lwt_in_verifier_ops = { 9691 .get_func_proto = lwt_in_func_proto, 9692 .is_valid_access = lwt_is_valid_access, 9693 .convert_ctx_access = bpf_convert_ctx_access, 9694 }; 9695 9696 const struct bpf_prog_ops lwt_in_prog_ops = { 9697 .test_run = bpf_prog_test_run_skb, 9698 }; 9699 9700 const struct bpf_verifier_ops lwt_out_verifier_ops = { 9701 .get_func_proto = lwt_out_func_proto, 9702 .is_valid_access = lwt_is_valid_access, 9703 .convert_ctx_access = bpf_convert_ctx_access, 9704 }; 9705 9706 const struct bpf_prog_ops lwt_out_prog_ops = { 9707 .test_run = bpf_prog_test_run_skb, 9708 }; 9709 9710 const struct bpf_verifier_ops lwt_xmit_verifier_ops = { 9711 .get_func_proto = lwt_xmit_func_proto, 9712 .is_valid_access = lwt_is_valid_access, 9713 .convert_ctx_access = bpf_convert_ctx_access, 9714 .gen_prologue = tc_cls_act_prologue, 9715 }; 9716 9717 const struct bpf_prog_ops lwt_xmit_prog_ops = { 9718 .test_run = bpf_prog_test_run_skb, 9719 }; 9720 9721 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = { 9722 .get_func_proto = lwt_seg6local_func_proto, 9723 .is_valid_access = lwt_is_valid_access, 9724 .convert_ctx_access = bpf_convert_ctx_access, 9725 }; 9726 9727 const struct bpf_prog_ops lwt_seg6local_prog_ops = { 9728 .test_run = bpf_prog_test_run_skb, 9729 }; 9730 9731 const struct bpf_verifier_ops cg_sock_verifier_ops = { 9732 .get_func_proto = sock_filter_func_proto, 9733 .is_valid_access = sock_filter_is_valid_access, 9734 .convert_ctx_access = bpf_sock_convert_ctx_access, 9735 }; 9736 9737 const struct bpf_prog_ops cg_sock_prog_ops = { 9738 }; 9739 9740 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = { 9741 .get_func_proto = sock_addr_func_proto, 9742 .is_valid_access = sock_addr_is_valid_access, 9743 .convert_ctx_access = sock_addr_convert_ctx_access, 9744 }; 9745 9746 const struct bpf_prog_ops cg_sock_addr_prog_ops = { 9747 }; 9748 9749 const struct bpf_verifier_ops sock_ops_verifier_ops = { 9750 .get_func_proto = sock_ops_func_proto, 9751 .is_valid_access = sock_ops_is_valid_access, 9752 .convert_ctx_access = sock_ops_convert_ctx_access, 9753 }; 9754 9755 const struct bpf_prog_ops sock_ops_prog_ops = { 9756 }; 9757 9758 const struct bpf_verifier_ops sk_skb_verifier_ops = { 9759 .get_func_proto = sk_skb_func_proto, 9760 .is_valid_access = sk_skb_is_valid_access, 9761 .convert_ctx_access = sk_skb_convert_ctx_access, 9762 .gen_prologue = sk_skb_prologue, 9763 }; 9764 9765 const struct bpf_prog_ops sk_skb_prog_ops = { 9766 }; 9767 9768 const struct bpf_verifier_ops sk_msg_verifier_ops = { 9769 .get_func_proto = sk_msg_func_proto, 9770 .is_valid_access = sk_msg_is_valid_access, 9771 .convert_ctx_access = sk_msg_convert_ctx_access, 9772 .gen_prologue = bpf_noop_prologue, 9773 }; 9774 9775 const struct bpf_prog_ops sk_msg_prog_ops = { 9776 }; 9777 9778 const struct bpf_verifier_ops flow_dissector_verifier_ops = { 9779 .get_func_proto = flow_dissector_func_proto, 9780 .is_valid_access = flow_dissector_is_valid_access, 9781 .convert_ctx_access = flow_dissector_convert_ctx_access, 9782 }; 9783 9784 const struct bpf_prog_ops flow_dissector_prog_ops = { 9785 .test_run = bpf_prog_test_run_flow_dissector, 9786 }; 9787 9788 int sk_detach_filter(struct sock *sk) 9789 { 9790 int ret = -ENOENT; 9791 struct sk_filter *filter; 9792 9793 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 9794 return -EPERM; 9795 9796 filter = rcu_dereference_protected(sk->sk_filter, 9797 lockdep_sock_is_held(sk)); 9798 if (filter) { 9799 RCU_INIT_POINTER(sk->sk_filter, NULL); 9800 sk_filter_uncharge(sk, filter); 9801 ret = 0; 9802 } 9803 9804 return ret; 9805 } 9806 EXPORT_SYMBOL_GPL(sk_detach_filter); 9807 9808 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, 9809 unsigned int len) 9810 { 9811 struct sock_fprog_kern *fprog; 9812 struct sk_filter *filter; 9813 int ret = 0; 9814 9815 lock_sock(sk); 9816 filter = rcu_dereference_protected(sk->sk_filter, 9817 lockdep_sock_is_held(sk)); 9818 if (!filter) 9819 goto out; 9820 9821 /* We're copying the filter that has been originally attached, 9822 * so no conversion/decode needed anymore. eBPF programs that 9823 * have no original program cannot be dumped through this. 9824 */ 9825 ret = -EACCES; 9826 fprog = filter->prog->orig_prog; 9827 if (!fprog) 9828 goto out; 9829 9830 ret = fprog->len; 9831 if (!len) 9832 /* User space only enquires number of filter blocks. */ 9833 goto out; 9834 9835 ret = -EINVAL; 9836 if (len < fprog->len) 9837 goto out; 9838 9839 ret = -EFAULT; 9840 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog))) 9841 goto out; 9842 9843 /* Instead of bytes, the API requests to return the number 9844 * of filter blocks. 9845 */ 9846 ret = fprog->len; 9847 out: 9848 release_sock(sk); 9849 return ret; 9850 } 9851 9852 #ifdef CONFIG_INET 9853 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern, 9854 struct sock_reuseport *reuse, 9855 struct sock *sk, struct sk_buff *skb, 9856 u32 hash) 9857 { 9858 reuse_kern->skb = skb; 9859 reuse_kern->sk = sk; 9860 reuse_kern->selected_sk = NULL; 9861 reuse_kern->data_end = skb->data + skb_headlen(skb); 9862 reuse_kern->hash = hash; 9863 reuse_kern->reuseport_id = reuse->reuseport_id; 9864 reuse_kern->bind_inany = reuse->bind_inany; 9865 } 9866 9867 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 9868 struct bpf_prog *prog, struct sk_buff *skb, 9869 u32 hash) 9870 { 9871 struct sk_reuseport_kern reuse_kern; 9872 enum sk_action action; 9873 9874 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash); 9875 action = BPF_PROG_RUN(prog, &reuse_kern); 9876 9877 if (action == SK_PASS) 9878 return reuse_kern.selected_sk; 9879 else 9880 return ERR_PTR(-ECONNREFUSED); 9881 } 9882 9883 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern, 9884 struct bpf_map *, map, void *, key, u32, flags) 9885 { 9886 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY; 9887 struct sock_reuseport *reuse; 9888 struct sock *selected_sk; 9889 9890 selected_sk = map->ops->map_lookup_elem(map, key); 9891 if (!selected_sk) 9892 return -ENOENT; 9893 9894 reuse = rcu_dereference(selected_sk->sk_reuseport_cb); 9895 if (!reuse) { 9896 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */ 9897 if (sk_is_refcounted(selected_sk)) 9898 sock_put(selected_sk); 9899 9900 /* reuseport_array has only sk with non NULL sk_reuseport_cb. 9901 * The only (!reuse) case here is - the sk has already been 9902 * unhashed (e.g. by close()), so treat it as -ENOENT. 9903 * 9904 * Other maps (e.g. sock_map) do not provide this guarantee and 9905 * the sk may never be in the reuseport group to begin with. 9906 */ 9907 return is_sockarray ? -ENOENT : -EINVAL; 9908 } 9909 9910 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) { 9911 struct sock *sk = reuse_kern->sk; 9912 9913 if (sk->sk_protocol != selected_sk->sk_protocol) 9914 return -EPROTOTYPE; 9915 else if (sk->sk_family != selected_sk->sk_family) 9916 return -EAFNOSUPPORT; 9917 9918 /* Catch all. Likely bound to a different sockaddr. */ 9919 return -EBADFD; 9920 } 9921 9922 reuse_kern->selected_sk = selected_sk; 9923 9924 return 0; 9925 } 9926 9927 static const struct bpf_func_proto sk_select_reuseport_proto = { 9928 .func = sk_select_reuseport, 9929 .gpl_only = false, 9930 .ret_type = RET_INTEGER, 9931 .arg1_type = ARG_PTR_TO_CTX, 9932 .arg2_type = ARG_CONST_MAP_PTR, 9933 .arg3_type = ARG_PTR_TO_MAP_KEY, 9934 .arg4_type = ARG_ANYTHING, 9935 }; 9936 9937 BPF_CALL_4(sk_reuseport_load_bytes, 9938 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 9939 void *, to, u32, len) 9940 { 9941 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len); 9942 } 9943 9944 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = { 9945 .func = sk_reuseport_load_bytes, 9946 .gpl_only = false, 9947 .ret_type = RET_INTEGER, 9948 .arg1_type = ARG_PTR_TO_CTX, 9949 .arg2_type = ARG_ANYTHING, 9950 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 9951 .arg4_type = ARG_CONST_SIZE, 9952 }; 9953 9954 BPF_CALL_5(sk_reuseport_load_bytes_relative, 9955 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 9956 void *, to, u32, len, u32, start_header) 9957 { 9958 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to, 9959 len, start_header); 9960 } 9961 9962 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = { 9963 .func = sk_reuseport_load_bytes_relative, 9964 .gpl_only = false, 9965 .ret_type = RET_INTEGER, 9966 .arg1_type = ARG_PTR_TO_CTX, 9967 .arg2_type = ARG_ANYTHING, 9968 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 9969 .arg4_type = ARG_CONST_SIZE, 9970 .arg5_type = ARG_ANYTHING, 9971 }; 9972 9973 static const struct bpf_func_proto * 9974 sk_reuseport_func_proto(enum bpf_func_id func_id, 9975 const struct bpf_prog *prog) 9976 { 9977 switch (func_id) { 9978 case BPF_FUNC_sk_select_reuseport: 9979 return &sk_select_reuseport_proto; 9980 case BPF_FUNC_skb_load_bytes: 9981 return &sk_reuseport_load_bytes_proto; 9982 case BPF_FUNC_skb_load_bytes_relative: 9983 return &sk_reuseport_load_bytes_relative_proto; 9984 default: 9985 return bpf_base_func_proto(func_id); 9986 } 9987 } 9988 9989 static bool 9990 sk_reuseport_is_valid_access(int off, int size, 9991 enum bpf_access_type type, 9992 const struct bpf_prog *prog, 9993 struct bpf_insn_access_aux *info) 9994 { 9995 const u32 size_default = sizeof(__u32); 9996 9997 if (off < 0 || off >= sizeof(struct sk_reuseport_md) || 9998 off % size || type != BPF_READ) 9999 return false; 10000 10001 switch (off) { 10002 case offsetof(struct sk_reuseport_md, data): 10003 info->reg_type = PTR_TO_PACKET; 10004 return size == sizeof(__u64); 10005 10006 case offsetof(struct sk_reuseport_md, data_end): 10007 info->reg_type = PTR_TO_PACKET_END; 10008 return size == sizeof(__u64); 10009 10010 case offsetof(struct sk_reuseport_md, hash): 10011 return size == size_default; 10012 10013 /* Fields that allow narrowing */ 10014 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol): 10015 if (size < sizeof_field(struct sk_buff, protocol)) 10016 return false; 10017 fallthrough; 10018 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol): 10019 case bpf_ctx_range(struct sk_reuseport_md, bind_inany): 10020 case bpf_ctx_range(struct sk_reuseport_md, len): 10021 bpf_ctx_record_field_size(info, size_default); 10022 return bpf_ctx_narrow_access_ok(off, size, size_default); 10023 10024 default: 10025 return false; 10026 } 10027 } 10028 10029 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \ 10030 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \ 10031 si->dst_reg, si->src_reg, \ 10032 bpf_target_off(struct sk_reuseport_kern, F, \ 10033 sizeof_field(struct sk_reuseport_kern, F), \ 10034 target_size)); \ 10035 }) 10036 10037 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \ 10038 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 10039 struct sk_buff, \ 10040 skb, \ 10041 SKB_FIELD) 10042 10043 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \ 10044 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 10045 struct sock, \ 10046 sk, \ 10047 SK_FIELD) 10048 10049 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type, 10050 const struct bpf_insn *si, 10051 struct bpf_insn *insn_buf, 10052 struct bpf_prog *prog, 10053 u32 *target_size) 10054 { 10055 struct bpf_insn *insn = insn_buf; 10056 10057 switch (si->off) { 10058 case offsetof(struct sk_reuseport_md, data): 10059 SK_REUSEPORT_LOAD_SKB_FIELD(data); 10060 break; 10061 10062 case offsetof(struct sk_reuseport_md, len): 10063 SK_REUSEPORT_LOAD_SKB_FIELD(len); 10064 break; 10065 10066 case offsetof(struct sk_reuseport_md, eth_protocol): 10067 SK_REUSEPORT_LOAD_SKB_FIELD(protocol); 10068 break; 10069 10070 case offsetof(struct sk_reuseport_md, ip_protocol): 10071 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol); 10072 break; 10073 10074 case offsetof(struct sk_reuseport_md, data_end): 10075 SK_REUSEPORT_LOAD_FIELD(data_end); 10076 break; 10077 10078 case offsetof(struct sk_reuseport_md, hash): 10079 SK_REUSEPORT_LOAD_FIELD(hash); 10080 break; 10081 10082 case offsetof(struct sk_reuseport_md, bind_inany): 10083 SK_REUSEPORT_LOAD_FIELD(bind_inany); 10084 break; 10085 } 10086 10087 return insn - insn_buf; 10088 } 10089 10090 const struct bpf_verifier_ops sk_reuseport_verifier_ops = { 10091 .get_func_proto = sk_reuseport_func_proto, 10092 .is_valid_access = sk_reuseport_is_valid_access, 10093 .convert_ctx_access = sk_reuseport_convert_ctx_access, 10094 }; 10095 10096 const struct bpf_prog_ops sk_reuseport_prog_ops = { 10097 }; 10098 10099 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled); 10100 EXPORT_SYMBOL(bpf_sk_lookup_enabled); 10101 10102 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx, 10103 struct sock *, sk, u64, flags) 10104 { 10105 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE | 10106 BPF_SK_LOOKUP_F_NO_REUSEPORT))) 10107 return -EINVAL; 10108 if (unlikely(sk && sk_is_refcounted(sk))) 10109 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */ 10110 if (unlikely(sk && sk->sk_state == TCP_ESTABLISHED)) 10111 return -ESOCKTNOSUPPORT; /* reject connected sockets */ 10112 10113 /* Check if socket is suitable for packet L3/L4 protocol */ 10114 if (sk && sk->sk_protocol != ctx->protocol) 10115 return -EPROTOTYPE; 10116 if (sk && sk->sk_family != ctx->family && 10117 (sk->sk_family == AF_INET || ipv6_only_sock(sk))) 10118 return -EAFNOSUPPORT; 10119 10120 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE)) 10121 return -EEXIST; 10122 10123 /* Select socket as lookup result */ 10124 ctx->selected_sk = sk; 10125 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT; 10126 return 0; 10127 } 10128 10129 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = { 10130 .func = bpf_sk_lookup_assign, 10131 .gpl_only = false, 10132 .ret_type = RET_INTEGER, 10133 .arg1_type = ARG_PTR_TO_CTX, 10134 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL, 10135 .arg3_type = ARG_ANYTHING, 10136 }; 10137 10138 static const struct bpf_func_proto * 10139 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 10140 { 10141 switch (func_id) { 10142 case BPF_FUNC_perf_event_output: 10143 return &bpf_event_output_data_proto; 10144 case BPF_FUNC_sk_assign: 10145 return &bpf_sk_lookup_assign_proto; 10146 case BPF_FUNC_sk_release: 10147 return &bpf_sk_release_proto; 10148 default: 10149 return bpf_sk_base_func_proto(func_id); 10150 } 10151 } 10152 10153 static bool sk_lookup_is_valid_access(int off, int size, 10154 enum bpf_access_type type, 10155 const struct bpf_prog *prog, 10156 struct bpf_insn_access_aux *info) 10157 { 10158 if (off < 0 || off >= sizeof(struct bpf_sk_lookup)) 10159 return false; 10160 if (off % size != 0) 10161 return false; 10162 if (type != BPF_READ) 10163 return false; 10164 10165 switch (off) { 10166 case offsetof(struct bpf_sk_lookup, sk): 10167 info->reg_type = PTR_TO_SOCKET_OR_NULL; 10168 return size == sizeof(__u64); 10169 10170 case bpf_ctx_range(struct bpf_sk_lookup, family): 10171 case bpf_ctx_range(struct bpf_sk_lookup, protocol): 10172 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4): 10173 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4): 10174 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]): 10175 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]): 10176 case bpf_ctx_range(struct bpf_sk_lookup, remote_port): 10177 case bpf_ctx_range(struct bpf_sk_lookup, local_port): 10178 bpf_ctx_record_field_size(info, sizeof(__u32)); 10179 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32)); 10180 10181 default: 10182 return false; 10183 } 10184 } 10185 10186 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type, 10187 const struct bpf_insn *si, 10188 struct bpf_insn *insn_buf, 10189 struct bpf_prog *prog, 10190 u32 *target_size) 10191 { 10192 struct bpf_insn *insn = insn_buf; 10193 10194 switch (si->off) { 10195 case offsetof(struct bpf_sk_lookup, sk): 10196 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 10197 offsetof(struct bpf_sk_lookup_kern, selected_sk)); 10198 break; 10199 10200 case offsetof(struct bpf_sk_lookup, family): 10201 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10202 bpf_target_off(struct bpf_sk_lookup_kern, 10203 family, 2, target_size)); 10204 break; 10205 10206 case offsetof(struct bpf_sk_lookup, protocol): 10207 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10208 bpf_target_off(struct bpf_sk_lookup_kern, 10209 protocol, 2, target_size)); 10210 break; 10211 10212 case offsetof(struct bpf_sk_lookup, remote_ip4): 10213 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10214 bpf_target_off(struct bpf_sk_lookup_kern, 10215 v4.saddr, 4, target_size)); 10216 break; 10217 10218 case offsetof(struct bpf_sk_lookup, local_ip4): 10219 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10220 bpf_target_off(struct bpf_sk_lookup_kern, 10221 v4.daddr, 4, target_size)); 10222 break; 10223 10224 case bpf_ctx_range_till(struct bpf_sk_lookup, 10225 remote_ip6[0], remote_ip6[3]): { 10226 #if IS_ENABLED(CONFIG_IPV6) 10227 int off = si->off; 10228 10229 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]); 10230 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 10231 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 10232 offsetof(struct bpf_sk_lookup_kern, v6.saddr)); 10233 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10234 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 10235 #else 10236 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10237 #endif 10238 break; 10239 } 10240 case bpf_ctx_range_till(struct bpf_sk_lookup, 10241 local_ip6[0], local_ip6[3]): { 10242 #if IS_ENABLED(CONFIG_IPV6) 10243 int off = si->off; 10244 10245 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]); 10246 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 10247 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 10248 offsetof(struct bpf_sk_lookup_kern, v6.daddr)); 10249 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10250 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 10251 #else 10252 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10253 #endif 10254 break; 10255 } 10256 case offsetof(struct bpf_sk_lookup, remote_port): 10257 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10258 bpf_target_off(struct bpf_sk_lookup_kern, 10259 sport, 2, target_size)); 10260 break; 10261 10262 case offsetof(struct bpf_sk_lookup, local_port): 10263 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10264 bpf_target_off(struct bpf_sk_lookup_kern, 10265 dport, 2, target_size)); 10266 break; 10267 } 10268 10269 return insn - insn_buf; 10270 } 10271 10272 const struct bpf_prog_ops sk_lookup_prog_ops = { 10273 }; 10274 10275 const struct bpf_verifier_ops sk_lookup_verifier_ops = { 10276 .get_func_proto = sk_lookup_func_proto, 10277 .is_valid_access = sk_lookup_is_valid_access, 10278 .convert_ctx_access = sk_lookup_convert_ctx_access, 10279 }; 10280 10281 #endif /* CONFIG_INET */ 10282 10283 DEFINE_BPF_DISPATCHER(xdp) 10284 10285 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog) 10286 { 10287 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog); 10288 } 10289 10290 #ifdef CONFIG_DEBUG_INFO_BTF 10291 BTF_ID_LIST_GLOBAL(btf_sock_ids) 10292 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type) 10293 BTF_SOCK_TYPE_xxx 10294 #undef BTF_SOCK_TYPE 10295 #else 10296 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE]; 10297 #endif 10298 10299 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk) 10300 { 10301 /* tcp6_sock type is not generated in dwarf and hence btf, 10302 * trigger an explicit type generation here. 10303 */ 10304 BTF_TYPE_EMIT(struct tcp6_sock); 10305 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && 10306 sk->sk_family == AF_INET6) 10307 return (unsigned long)sk; 10308 10309 return (unsigned long)NULL; 10310 } 10311 10312 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = { 10313 .func = bpf_skc_to_tcp6_sock, 10314 .gpl_only = false, 10315 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10316 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10317 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6], 10318 }; 10319 10320 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk) 10321 { 10322 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 10323 return (unsigned long)sk; 10324 10325 return (unsigned long)NULL; 10326 } 10327 10328 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = { 10329 .func = bpf_skc_to_tcp_sock, 10330 .gpl_only = false, 10331 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10332 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10333 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP], 10334 }; 10335 10336 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk) 10337 { 10338 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not 10339 * generated if CONFIG_INET=n. Trigger an explicit generation here. 10340 */ 10341 BTF_TYPE_EMIT(struct inet_timewait_sock); 10342 BTF_TYPE_EMIT(struct tcp_timewait_sock); 10343 10344 #ifdef CONFIG_INET 10345 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT) 10346 return (unsigned long)sk; 10347 #endif 10348 10349 #if IS_BUILTIN(CONFIG_IPV6) 10350 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT) 10351 return (unsigned long)sk; 10352 #endif 10353 10354 return (unsigned long)NULL; 10355 } 10356 10357 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = { 10358 .func = bpf_skc_to_tcp_timewait_sock, 10359 .gpl_only = false, 10360 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10361 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10362 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW], 10363 }; 10364 10365 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk) 10366 { 10367 #ifdef CONFIG_INET 10368 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV) 10369 return (unsigned long)sk; 10370 #endif 10371 10372 #if IS_BUILTIN(CONFIG_IPV6) 10373 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV) 10374 return (unsigned long)sk; 10375 #endif 10376 10377 return (unsigned long)NULL; 10378 } 10379 10380 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = { 10381 .func = bpf_skc_to_tcp_request_sock, 10382 .gpl_only = false, 10383 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10384 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10385 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ], 10386 }; 10387 10388 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk) 10389 { 10390 /* udp6_sock type is not generated in dwarf and hence btf, 10391 * trigger an explicit type generation here. 10392 */ 10393 BTF_TYPE_EMIT(struct udp6_sock); 10394 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP && 10395 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6) 10396 return (unsigned long)sk; 10397 10398 return (unsigned long)NULL; 10399 } 10400 10401 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = { 10402 .func = bpf_skc_to_udp6_sock, 10403 .gpl_only = false, 10404 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10405 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10406 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6], 10407 }; 10408 10409 static const struct bpf_func_proto * 10410 bpf_sk_base_func_proto(enum bpf_func_id func_id) 10411 { 10412 const struct bpf_func_proto *func; 10413 10414 switch (func_id) { 10415 case BPF_FUNC_skc_to_tcp6_sock: 10416 func = &bpf_skc_to_tcp6_sock_proto; 10417 break; 10418 case BPF_FUNC_skc_to_tcp_sock: 10419 func = &bpf_skc_to_tcp_sock_proto; 10420 break; 10421 case BPF_FUNC_skc_to_tcp_timewait_sock: 10422 func = &bpf_skc_to_tcp_timewait_sock_proto; 10423 break; 10424 case BPF_FUNC_skc_to_tcp_request_sock: 10425 func = &bpf_skc_to_tcp_request_sock_proto; 10426 break; 10427 case BPF_FUNC_skc_to_udp6_sock: 10428 func = &bpf_skc_to_udp6_sock_proto; 10429 break; 10430 default: 10431 return bpf_base_func_proto(func_id); 10432 } 10433 10434 if (!perfmon_capable()) 10435 return NULL; 10436 10437 return func; 10438 } 10439