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