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 static 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 err = -EINVAL; 4494 goto err_clear; 4495 } 4496 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { 4497 err = -EPROTO; 4498 goto err_clear; 4499 } 4500 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4501 err = -EINVAL; 4502 switch (size) { 4503 case offsetof(struct bpf_tunnel_key, local_ipv6[0]): 4504 case offsetof(struct bpf_tunnel_key, tunnel_label): 4505 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4506 goto set_compat; 4507 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4508 /* Fixup deprecated structure layouts here, so we have 4509 * a common path later on. 4510 */ 4511 if (ip_tunnel_info_af(info) != AF_INET) 4512 goto err_clear; 4513 set_compat: 4514 to = (struct bpf_tunnel_key *)compat; 4515 break; 4516 default: 4517 goto err_clear; 4518 } 4519 } 4520 4521 to->tunnel_id = be64_to_cpu(info->key.tun_id); 4522 to->tunnel_tos = info->key.tos; 4523 to->tunnel_ttl = info->key.ttl; 4524 to->tunnel_ext = 0; 4525 4526 if (flags & BPF_F_TUNINFO_IPV6) { 4527 memcpy(to->remote_ipv6, &info->key.u.ipv6.src, 4528 sizeof(to->remote_ipv6)); 4529 memcpy(to->local_ipv6, &info->key.u.ipv6.dst, 4530 sizeof(to->local_ipv6)); 4531 to->tunnel_label = be32_to_cpu(info->key.label); 4532 } else { 4533 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); 4534 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 4535 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst); 4536 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3); 4537 to->tunnel_label = 0; 4538 } 4539 4540 if (unlikely(size != sizeof(struct bpf_tunnel_key))) 4541 memcpy(to_orig, to, size); 4542 4543 return 0; 4544 err_clear: 4545 memset(to_orig, 0, size); 4546 return err; 4547 } 4548 4549 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { 4550 .func = bpf_skb_get_tunnel_key, 4551 .gpl_only = false, 4552 .ret_type = RET_INTEGER, 4553 .arg1_type = ARG_PTR_TO_CTX, 4554 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4555 .arg3_type = ARG_CONST_SIZE, 4556 .arg4_type = ARG_ANYTHING, 4557 }; 4558 4559 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size) 4560 { 4561 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4562 int err; 4563 4564 if (unlikely(!info || 4565 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) { 4566 err = -ENOENT; 4567 goto err_clear; 4568 } 4569 if (unlikely(size < info->options_len)) { 4570 err = -ENOMEM; 4571 goto err_clear; 4572 } 4573 4574 ip_tunnel_info_opts_get(to, info); 4575 if (size > info->options_len) 4576 memset(to + info->options_len, 0, size - info->options_len); 4577 4578 return info->options_len; 4579 err_clear: 4580 memset(to, 0, size); 4581 return err; 4582 } 4583 4584 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { 4585 .func = bpf_skb_get_tunnel_opt, 4586 .gpl_only = false, 4587 .ret_type = RET_INTEGER, 4588 .arg1_type = ARG_PTR_TO_CTX, 4589 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4590 .arg3_type = ARG_CONST_SIZE, 4591 }; 4592 4593 static struct metadata_dst __percpu *md_dst; 4594 4595 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb, 4596 const struct bpf_tunnel_key *, from, u32, size, u64, flags) 4597 { 4598 struct metadata_dst *md = this_cpu_ptr(md_dst); 4599 u8 compat[sizeof(struct bpf_tunnel_key)]; 4600 struct ip_tunnel_info *info; 4601 4602 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | 4603 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER))) 4604 return -EINVAL; 4605 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4606 switch (size) { 4607 case offsetof(struct bpf_tunnel_key, local_ipv6[0]): 4608 case offsetof(struct bpf_tunnel_key, tunnel_label): 4609 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4610 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4611 /* Fixup deprecated structure layouts here, so we have 4612 * a common path later on. 4613 */ 4614 memcpy(compat, from, size); 4615 memset(compat + size, 0, sizeof(compat) - size); 4616 from = (const struct bpf_tunnel_key *) compat; 4617 break; 4618 default: 4619 return -EINVAL; 4620 } 4621 } 4622 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || 4623 from->tunnel_ext)) 4624 return -EINVAL; 4625 4626 skb_dst_drop(skb); 4627 dst_hold((struct dst_entry *) md); 4628 skb_dst_set(skb, (struct dst_entry *) md); 4629 4630 info = &md->u.tun_info; 4631 memset(info, 0, sizeof(*info)); 4632 info->mode = IP_TUNNEL_INFO_TX; 4633 4634 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE; 4635 if (flags & BPF_F_DONT_FRAGMENT) 4636 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT; 4637 if (flags & BPF_F_ZERO_CSUM_TX) 4638 info->key.tun_flags &= ~TUNNEL_CSUM; 4639 if (flags & BPF_F_SEQ_NUMBER) 4640 info->key.tun_flags |= TUNNEL_SEQ; 4641 4642 info->key.tun_id = cpu_to_be64(from->tunnel_id); 4643 info->key.tos = from->tunnel_tos; 4644 info->key.ttl = from->tunnel_ttl; 4645 4646 if (flags & BPF_F_TUNINFO_IPV6) { 4647 info->mode |= IP_TUNNEL_INFO_IPV6; 4648 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, 4649 sizeof(from->remote_ipv6)); 4650 memcpy(&info->key.u.ipv6.src, from->local_ipv6, 4651 sizeof(from->local_ipv6)); 4652 info->key.label = cpu_to_be32(from->tunnel_label) & 4653 IPV6_FLOWLABEL_MASK; 4654 } else { 4655 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); 4656 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4); 4657 info->key.flow_flags = FLOWI_FLAG_ANYSRC; 4658 } 4659 4660 return 0; 4661 } 4662 4663 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { 4664 .func = bpf_skb_set_tunnel_key, 4665 .gpl_only = false, 4666 .ret_type = RET_INTEGER, 4667 .arg1_type = ARG_PTR_TO_CTX, 4668 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4669 .arg3_type = ARG_CONST_SIZE, 4670 .arg4_type = ARG_ANYTHING, 4671 }; 4672 4673 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb, 4674 const u8 *, from, u32, size) 4675 { 4676 struct ip_tunnel_info *info = skb_tunnel_info(skb); 4677 const struct metadata_dst *md = this_cpu_ptr(md_dst); 4678 4679 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) 4680 return -EINVAL; 4681 if (unlikely(size > IP_TUNNEL_OPTS_MAX)) 4682 return -ENOMEM; 4683 4684 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT); 4685 4686 return 0; 4687 } 4688 4689 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { 4690 .func = bpf_skb_set_tunnel_opt, 4691 .gpl_only = false, 4692 .ret_type = RET_INTEGER, 4693 .arg1_type = ARG_PTR_TO_CTX, 4694 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4695 .arg3_type = ARG_CONST_SIZE, 4696 }; 4697 4698 static const struct bpf_func_proto * 4699 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) 4700 { 4701 if (!md_dst) { 4702 struct metadata_dst __percpu *tmp; 4703 4704 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, 4705 METADATA_IP_TUNNEL, 4706 GFP_KERNEL); 4707 if (!tmp) 4708 return NULL; 4709 if (cmpxchg(&md_dst, NULL, tmp)) 4710 metadata_dst_free_percpu(tmp); 4711 } 4712 4713 switch (which) { 4714 case BPF_FUNC_skb_set_tunnel_key: 4715 return &bpf_skb_set_tunnel_key_proto; 4716 case BPF_FUNC_skb_set_tunnel_opt: 4717 return &bpf_skb_set_tunnel_opt_proto; 4718 default: 4719 return NULL; 4720 } 4721 } 4722 4723 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map, 4724 u32, idx) 4725 { 4726 struct bpf_array *array = container_of(map, struct bpf_array, map); 4727 struct cgroup *cgrp; 4728 struct sock *sk; 4729 4730 sk = skb_to_full_sk(skb); 4731 if (!sk || !sk_fullsock(sk)) 4732 return -ENOENT; 4733 if (unlikely(idx >= array->map.max_entries)) 4734 return -E2BIG; 4735 4736 cgrp = READ_ONCE(array->ptrs[idx]); 4737 if (unlikely(!cgrp)) 4738 return -EAGAIN; 4739 4740 return sk_under_cgroup_hierarchy(sk, cgrp); 4741 } 4742 4743 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = { 4744 .func = bpf_skb_under_cgroup, 4745 .gpl_only = false, 4746 .ret_type = RET_INTEGER, 4747 .arg1_type = ARG_PTR_TO_CTX, 4748 .arg2_type = ARG_CONST_MAP_PTR, 4749 .arg3_type = ARG_ANYTHING, 4750 }; 4751 4752 #ifdef CONFIG_SOCK_CGROUP_DATA 4753 static inline u64 __bpf_sk_cgroup_id(struct sock *sk) 4754 { 4755 struct cgroup *cgrp; 4756 4757 sk = sk_to_full_sk(sk); 4758 if (!sk || !sk_fullsock(sk)) 4759 return 0; 4760 4761 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4762 return cgroup_id(cgrp); 4763 } 4764 4765 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb) 4766 { 4767 return __bpf_sk_cgroup_id(skb->sk); 4768 } 4769 4770 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = { 4771 .func = bpf_skb_cgroup_id, 4772 .gpl_only = false, 4773 .ret_type = RET_INTEGER, 4774 .arg1_type = ARG_PTR_TO_CTX, 4775 }; 4776 4777 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk, 4778 int ancestor_level) 4779 { 4780 struct cgroup *ancestor; 4781 struct cgroup *cgrp; 4782 4783 sk = sk_to_full_sk(sk); 4784 if (!sk || !sk_fullsock(sk)) 4785 return 0; 4786 4787 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4788 ancestor = cgroup_ancestor(cgrp, ancestor_level); 4789 if (!ancestor) 4790 return 0; 4791 4792 return cgroup_id(ancestor); 4793 } 4794 4795 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int, 4796 ancestor_level) 4797 { 4798 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level); 4799 } 4800 4801 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = { 4802 .func = bpf_skb_ancestor_cgroup_id, 4803 .gpl_only = false, 4804 .ret_type = RET_INTEGER, 4805 .arg1_type = ARG_PTR_TO_CTX, 4806 .arg2_type = ARG_ANYTHING, 4807 }; 4808 4809 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk) 4810 { 4811 return __bpf_sk_cgroup_id(sk); 4812 } 4813 4814 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = { 4815 .func = bpf_sk_cgroup_id, 4816 .gpl_only = false, 4817 .ret_type = RET_INTEGER, 4818 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4819 }; 4820 4821 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level) 4822 { 4823 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level); 4824 } 4825 4826 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = { 4827 .func = bpf_sk_ancestor_cgroup_id, 4828 .gpl_only = false, 4829 .ret_type = RET_INTEGER, 4830 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4831 .arg2_type = ARG_ANYTHING, 4832 }; 4833 #endif 4834 4835 static unsigned long bpf_xdp_copy(void *dst, const void *ctx, 4836 unsigned long off, unsigned long len) 4837 { 4838 struct xdp_buff *xdp = (struct xdp_buff *)ctx; 4839 4840 bpf_xdp_copy_buf(xdp, off, dst, len, false); 4841 return 0; 4842 } 4843 4844 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map, 4845 u64, flags, void *, meta, u64, meta_size) 4846 { 4847 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4848 4849 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4850 return -EINVAL; 4851 4852 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp))) 4853 return -EFAULT; 4854 4855 return bpf_event_output(map, flags, meta, meta_size, xdp, 4856 xdp_size, bpf_xdp_copy); 4857 } 4858 4859 static const struct bpf_func_proto bpf_xdp_event_output_proto = { 4860 .func = bpf_xdp_event_output, 4861 .gpl_only = true, 4862 .ret_type = RET_INTEGER, 4863 .arg1_type = ARG_PTR_TO_CTX, 4864 .arg2_type = ARG_CONST_MAP_PTR, 4865 .arg3_type = ARG_ANYTHING, 4866 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4867 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4868 }; 4869 4870 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff) 4871 4872 const struct bpf_func_proto bpf_xdp_output_proto = { 4873 .func = bpf_xdp_event_output, 4874 .gpl_only = true, 4875 .ret_type = RET_INTEGER, 4876 .arg1_type = ARG_PTR_TO_BTF_ID, 4877 .arg1_btf_id = &bpf_xdp_output_btf_ids[0], 4878 .arg2_type = ARG_CONST_MAP_PTR, 4879 .arg3_type = ARG_ANYTHING, 4880 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4881 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4882 }; 4883 4884 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb) 4885 { 4886 return skb->sk ? __sock_gen_cookie(skb->sk) : 0; 4887 } 4888 4889 static const struct bpf_func_proto bpf_get_socket_cookie_proto = { 4890 .func = bpf_get_socket_cookie, 4891 .gpl_only = false, 4892 .ret_type = RET_INTEGER, 4893 .arg1_type = ARG_PTR_TO_CTX, 4894 }; 4895 4896 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4897 { 4898 return __sock_gen_cookie(ctx->sk); 4899 } 4900 4901 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = { 4902 .func = bpf_get_socket_cookie_sock_addr, 4903 .gpl_only = false, 4904 .ret_type = RET_INTEGER, 4905 .arg1_type = ARG_PTR_TO_CTX, 4906 }; 4907 4908 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx) 4909 { 4910 return __sock_gen_cookie(ctx); 4911 } 4912 4913 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = { 4914 .func = bpf_get_socket_cookie_sock, 4915 .gpl_only = false, 4916 .ret_type = RET_INTEGER, 4917 .arg1_type = ARG_PTR_TO_CTX, 4918 }; 4919 4920 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk) 4921 { 4922 return sk ? sock_gen_cookie(sk) : 0; 4923 } 4924 4925 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = { 4926 .func = bpf_get_socket_ptr_cookie, 4927 .gpl_only = false, 4928 .ret_type = RET_INTEGER, 4929 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4930 }; 4931 4932 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 4933 { 4934 return __sock_gen_cookie(ctx->sk); 4935 } 4936 4937 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = { 4938 .func = bpf_get_socket_cookie_sock_ops, 4939 .gpl_only = false, 4940 .ret_type = RET_INTEGER, 4941 .arg1_type = ARG_PTR_TO_CTX, 4942 }; 4943 4944 static u64 __bpf_get_netns_cookie(struct sock *sk) 4945 { 4946 const struct net *net = sk ? sock_net(sk) : &init_net; 4947 4948 return net->net_cookie; 4949 } 4950 4951 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx) 4952 { 4953 return __bpf_get_netns_cookie(ctx); 4954 } 4955 4956 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = { 4957 .func = bpf_get_netns_cookie_sock, 4958 .gpl_only = false, 4959 .ret_type = RET_INTEGER, 4960 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4961 }; 4962 4963 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4964 { 4965 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4966 } 4967 4968 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = { 4969 .func = bpf_get_netns_cookie_sock_addr, 4970 .gpl_only = false, 4971 .ret_type = RET_INTEGER, 4972 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4973 }; 4974 4975 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 4976 { 4977 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4978 } 4979 4980 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = { 4981 .func = bpf_get_netns_cookie_sock_ops, 4982 .gpl_only = false, 4983 .ret_type = RET_INTEGER, 4984 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4985 }; 4986 4987 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx) 4988 { 4989 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4990 } 4991 4992 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = { 4993 .func = bpf_get_netns_cookie_sk_msg, 4994 .gpl_only = false, 4995 .ret_type = RET_INTEGER, 4996 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4997 }; 4998 4999 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb) 5000 { 5001 struct sock *sk = sk_to_full_sk(skb->sk); 5002 kuid_t kuid; 5003 5004 if (!sk || !sk_fullsock(sk)) 5005 return overflowuid; 5006 kuid = sock_net_uid(sock_net(sk), sk); 5007 return from_kuid_munged(sock_net(sk)->user_ns, kuid); 5008 } 5009 5010 static const struct bpf_func_proto bpf_get_socket_uid_proto = { 5011 .func = bpf_get_socket_uid, 5012 .gpl_only = false, 5013 .ret_type = RET_INTEGER, 5014 .arg1_type = ARG_PTR_TO_CTX, 5015 }; 5016 5017 static int __bpf_setsockopt(struct sock *sk, int level, int optname, 5018 char *optval, int optlen) 5019 { 5020 char devname[IFNAMSIZ]; 5021 int val, valbool; 5022 struct net *net; 5023 int ifindex; 5024 int ret = 0; 5025 5026 if (!sk_fullsock(sk)) 5027 return -EINVAL; 5028 5029 if (level == SOL_SOCKET) { 5030 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE) 5031 return -EINVAL; 5032 val = *((int *)optval); 5033 valbool = val ? 1 : 0; 5034 5035 /* Only some socketops are supported */ 5036 switch (optname) { 5037 case SO_RCVBUF: 5038 val = min_t(u32, val, READ_ONCE(sysctl_rmem_max)); 5039 val = min_t(int, val, INT_MAX / 2); 5040 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 5041 WRITE_ONCE(sk->sk_rcvbuf, 5042 max_t(int, val * 2, SOCK_MIN_RCVBUF)); 5043 break; 5044 case SO_SNDBUF: 5045 val = min_t(u32, val, READ_ONCE(sysctl_wmem_max)); 5046 val = min_t(int, val, INT_MAX / 2); 5047 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 5048 WRITE_ONCE(sk->sk_sndbuf, 5049 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 5050 break; 5051 case SO_MAX_PACING_RATE: /* 32bit version */ 5052 if (val != ~0U) 5053 cmpxchg(&sk->sk_pacing_status, 5054 SK_PACING_NONE, 5055 SK_PACING_NEEDED); 5056 sk->sk_max_pacing_rate = (val == ~0U) ? 5057 ~0UL : (unsigned int)val; 5058 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 5059 sk->sk_max_pacing_rate); 5060 break; 5061 case SO_PRIORITY: 5062 sk->sk_priority = val; 5063 break; 5064 case SO_RCVLOWAT: 5065 if (val < 0) 5066 val = INT_MAX; 5067 if (sk->sk_socket && sk->sk_socket->ops->set_rcvlowat) 5068 ret = sk->sk_socket->ops->set_rcvlowat(sk, val); 5069 else 5070 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 5071 break; 5072 case SO_MARK: 5073 if (sk->sk_mark != val) { 5074 sk->sk_mark = val; 5075 sk_dst_reset(sk); 5076 } 5077 break; 5078 case SO_BINDTODEVICE: 5079 optlen = min_t(long, optlen, IFNAMSIZ - 1); 5080 strncpy(devname, optval, optlen); 5081 devname[optlen] = 0; 5082 5083 ifindex = 0; 5084 if (devname[0] != '\0') { 5085 struct net_device *dev; 5086 5087 ret = -ENODEV; 5088 5089 net = sock_net(sk); 5090 dev = dev_get_by_name(net, devname); 5091 if (!dev) 5092 break; 5093 ifindex = dev->ifindex; 5094 dev_put(dev); 5095 } 5096 fallthrough; 5097 case SO_BINDTOIFINDEX: 5098 if (optname == SO_BINDTOIFINDEX) 5099 ifindex = val; 5100 ret = sock_bindtoindex(sk, ifindex, false); 5101 break; 5102 case SO_KEEPALIVE: 5103 if (sk->sk_prot->keepalive) 5104 sk->sk_prot->keepalive(sk, valbool); 5105 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 5106 break; 5107 case SO_REUSEPORT: 5108 sk->sk_reuseport = valbool; 5109 break; 5110 case SO_TXREHASH: 5111 if (val < -1 || val > 1) { 5112 ret = -EINVAL; 5113 break; 5114 } 5115 sk->sk_txrehash = (u8)val; 5116 break; 5117 default: 5118 ret = -EINVAL; 5119 } 5120 #ifdef CONFIG_INET 5121 } else if (level == SOL_IP) { 5122 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 5123 return -EINVAL; 5124 5125 val = *((int *)optval); 5126 /* Only some options are supported */ 5127 switch (optname) { 5128 case IP_TOS: 5129 if (val < -1 || val > 0xff) { 5130 ret = -EINVAL; 5131 } else { 5132 struct inet_sock *inet = inet_sk(sk); 5133 5134 if (val == -1) 5135 val = 0; 5136 inet->tos = val; 5137 } 5138 break; 5139 default: 5140 ret = -EINVAL; 5141 } 5142 #if IS_ENABLED(CONFIG_IPV6) 5143 } else if (level == SOL_IPV6) { 5144 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 5145 return -EINVAL; 5146 5147 val = *((int *)optval); 5148 /* Only some options are supported */ 5149 switch (optname) { 5150 case IPV6_TCLASS: 5151 if (val < -1 || val > 0xff) { 5152 ret = -EINVAL; 5153 } else { 5154 struct ipv6_pinfo *np = inet6_sk(sk); 5155 5156 if (val == -1) 5157 val = 0; 5158 np->tclass = val; 5159 } 5160 break; 5161 default: 5162 ret = -EINVAL; 5163 } 5164 #endif 5165 } else if (level == SOL_TCP && 5166 sk->sk_prot->setsockopt == tcp_setsockopt) { 5167 if (optname == TCP_CONGESTION) { 5168 char name[TCP_CA_NAME_MAX]; 5169 5170 strncpy(name, optval, min_t(long, optlen, 5171 TCP_CA_NAME_MAX-1)); 5172 name[TCP_CA_NAME_MAX-1] = 0; 5173 ret = tcp_set_congestion_control(sk, name, false, true); 5174 } else { 5175 struct inet_connection_sock *icsk = inet_csk(sk); 5176 struct tcp_sock *tp = tcp_sk(sk); 5177 unsigned long timeout; 5178 5179 if (optlen != sizeof(int)) 5180 return -EINVAL; 5181 5182 val = *((int *)optval); 5183 /* Only some options are supported */ 5184 switch (optname) { 5185 case TCP_BPF_IW: 5186 if (val <= 0 || tp->data_segs_out > tp->syn_data) 5187 ret = -EINVAL; 5188 else 5189 tcp_snd_cwnd_set(tp, val); 5190 break; 5191 case TCP_BPF_SNDCWND_CLAMP: 5192 if (val <= 0) { 5193 ret = -EINVAL; 5194 } else { 5195 tp->snd_cwnd_clamp = val; 5196 tp->snd_ssthresh = val; 5197 } 5198 break; 5199 case TCP_BPF_DELACK_MAX: 5200 timeout = usecs_to_jiffies(val); 5201 if (timeout > TCP_DELACK_MAX || 5202 timeout < TCP_TIMEOUT_MIN) 5203 return -EINVAL; 5204 inet_csk(sk)->icsk_delack_max = timeout; 5205 break; 5206 case TCP_BPF_RTO_MIN: 5207 timeout = usecs_to_jiffies(val); 5208 if (timeout > TCP_RTO_MIN || 5209 timeout < TCP_TIMEOUT_MIN) 5210 return -EINVAL; 5211 inet_csk(sk)->icsk_rto_min = timeout; 5212 break; 5213 case TCP_SAVE_SYN: 5214 if (val < 0 || val > 1) 5215 ret = -EINVAL; 5216 else 5217 tp->save_syn = val; 5218 break; 5219 case TCP_KEEPIDLE: 5220 ret = tcp_sock_set_keepidle_locked(sk, val); 5221 break; 5222 case TCP_KEEPINTVL: 5223 if (val < 1 || val > MAX_TCP_KEEPINTVL) 5224 ret = -EINVAL; 5225 else 5226 tp->keepalive_intvl = val * HZ; 5227 break; 5228 case TCP_KEEPCNT: 5229 if (val < 1 || val > MAX_TCP_KEEPCNT) 5230 ret = -EINVAL; 5231 else 5232 tp->keepalive_probes = val; 5233 break; 5234 case TCP_SYNCNT: 5235 if (val < 1 || val > MAX_TCP_SYNCNT) 5236 ret = -EINVAL; 5237 else 5238 icsk->icsk_syn_retries = val; 5239 break; 5240 case TCP_USER_TIMEOUT: 5241 if (val < 0) 5242 ret = -EINVAL; 5243 else 5244 icsk->icsk_user_timeout = val; 5245 break; 5246 case TCP_NOTSENT_LOWAT: 5247 tp->notsent_lowat = val; 5248 sk->sk_write_space(sk); 5249 break; 5250 case TCP_WINDOW_CLAMP: 5251 ret = tcp_set_window_clamp(sk, val); 5252 break; 5253 default: 5254 ret = -EINVAL; 5255 } 5256 } 5257 #endif 5258 } else { 5259 ret = -EINVAL; 5260 } 5261 return ret; 5262 } 5263 5264 static int _bpf_setsockopt(struct sock *sk, int level, int optname, 5265 char *optval, int optlen) 5266 { 5267 if (sk_fullsock(sk)) 5268 sock_owned_by_me(sk); 5269 return __bpf_setsockopt(sk, level, optname, optval, optlen); 5270 } 5271 5272 static int __bpf_getsockopt(struct sock *sk, int level, int optname, 5273 char *optval, int optlen) 5274 { 5275 if (!sk_fullsock(sk)) 5276 goto err_clear; 5277 5278 if (level == SOL_SOCKET) { 5279 if (optlen != sizeof(int)) 5280 goto err_clear; 5281 5282 switch (optname) { 5283 case SO_RCVBUF: 5284 *((int *)optval) = sk->sk_rcvbuf; 5285 break; 5286 case SO_SNDBUF: 5287 *((int *)optval) = sk->sk_sndbuf; 5288 break; 5289 case SO_MARK: 5290 *((int *)optval) = sk->sk_mark; 5291 break; 5292 case SO_PRIORITY: 5293 *((int *)optval) = sk->sk_priority; 5294 break; 5295 case SO_BINDTOIFINDEX: 5296 *((int *)optval) = sk->sk_bound_dev_if; 5297 break; 5298 case SO_REUSEPORT: 5299 *((int *)optval) = sk->sk_reuseport; 5300 break; 5301 case SO_TXREHASH: 5302 *((int *)optval) = sk->sk_txrehash; 5303 break; 5304 default: 5305 goto err_clear; 5306 } 5307 #ifdef CONFIG_INET 5308 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) { 5309 struct inet_connection_sock *icsk; 5310 struct tcp_sock *tp; 5311 5312 switch (optname) { 5313 case TCP_CONGESTION: 5314 icsk = inet_csk(sk); 5315 5316 if (!icsk->icsk_ca_ops || optlen <= 1) 5317 goto err_clear; 5318 strncpy(optval, icsk->icsk_ca_ops->name, optlen); 5319 optval[optlen - 1] = 0; 5320 break; 5321 case TCP_SAVED_SYN: 5322 tp = tcp_sk(sk); 5323 5324 if (optlen <= 0 || !tp->saved_syn || 5325 optlen > tcp_saved_syn_len(tp->saved_syn)) 5326 goto err_clear; 5327 memcpy(optval, tp->saved_syn->data, optlen); 5328 break; 5329 default: 5330 goto err_clear; 5331 } 5332 } else if (level == SOL_IP) { 5333 struct inet_sock *inet = inet_sk(sk); 5334 5335 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 5336 goto err_clear; 5337 5338 /* Only some options are supported */ 5339 switch (optname) { 5340 case IP_TOS: 5341 *((int *)optval) = (int)inet->tos; 5342 break; 5343 default: 5344 goto err_clear; 5345 } 5346 #if IS_ENABLED(CONFIG_IPV6) 5347 } else if (level == SOL_IPV6) { 5348 struct ipv6_pinfo *np = inet6_sk(sk); 5349 5350 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 5351 goto err_clear; 5352 5353 /* Only some options are supported */ 5354 switch (optname) { 5355 case IPV6_TCLASS: 5356 *((int *)optval) = (int)np->tclass; 5357 break; 5358 default: 5359 goto err_clear; 5360 } 5361 #endif 5362 #endif 5363 } else { 5364 goto err_clear; 5365 } 5366 return 0; 5367 err_clear: 5368 memset(optval, 0, optlen); 5369 return -EINVAL; 5370 } 5371 5372 static int _bpf_getsockopt(struct sock *sk, int level, int optname, 5373 char *optval, int optlen) 5374 { 5375 if (sk_fullsock(sk)) 5376 sock_owned_by_me(sk); 5377 return __bpf_getsockopt(sk, level, optname, optval, optlen); 5378 } 5379 5380 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level, 5381 int, optname, char *, optval, int, optlen) 5382 { 5383 if (level == SOL_TCP && optname == TCP_CONGESTION) { 5384 if (optlen >= sizeof("cdg") - 1 && 5385 !strncmp("cdg", optval, optlen)) 5386 return -ENOTSUPP; 5387 } 5388 5389 return _bpf_setsockopt(sk, level, optname, optval, optlen); 5390 } 5391 5392 const struct bpf_func_proto bpf_sk_setsockopt_proto = { 5393 .func = bpf_sk_setsockopt, 5394 .gpl_only = false, 5395 .ret_type = RET_INTEGER, 5396 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5397 .arg2_type = ARG_ANYTHING, 5398 .arg3_type = ARG_ANYTHING, 5399 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5400 .arg5_type = ARG_CONST_SIZE, 5401 }; 5402 5403 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level, 5404 int, optname, char *, optval, int, optlen) 5405 { 5406 return _bpf_getsockopt(sk, level, optname, optval, optlen); 5407 } 5408 5409 const struct bpf_func_proto bpf_sk_getsockopt_proto = { 5410 .func = bpf_sk_getsockopt, 5411 .gpl_only = false, 5412 .ret_type = RET_INTEGER, 5413 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5414 .arg2_type = ARG_ANYTHING, 5415 .arg3_type = ARG_ANYTHING, 5416 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5417 .arg5_type = ARG_CONST_SIZE, 5418 }; 5419 5420 BPF_CALL_5(bpf_unlocked_sk_setsockopt, struct sock *, sk, int, level, 5421 int, optname, char *, optval, int, optlen) 5422 { 5423 return __bpf_setsockopt(sk, level, optname, optval, optlen); 5424 } 5425 5426 const struct bpf_func_proto bpf_unlocked_sk_setsockopt_proto = { 5427 .func = bpf_unlocked_sk_setsockopt, 5428 .gpl_only = false, 5429 .ret_type = RET_INTEGER, 5430 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5431 .arg2_type = ARG_ANYTHING, 5432 .arg3_type = ARG_ANYTHING, 5433 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5434 .arg5_type = ARG_CONST_SIZE, 5435 }; 5436 5437 BPF_CALL_5(bpf_unlocked_sk_getsockopt, struct sock *, sk, int, level, 5438 int, optname, char *, optval, int, optlen) 5439 { 5440 return __bpf_getsockopt(sk, level, optname, optval, optlen); 5441 } 5442 5443 const struct bpf_func_proto bpf_unlocked_sk_getsockopt_proto = { 5444 .func = bpf_unlocked_sk_getsockopt, 5445 .gpl_only = false, 5446 .ret_type = RET_INTEGER, 5447 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5448 .arg2_type = ARG_ANYTHING, 5449 .arg3_type = ARG_ANYTHING, 5450 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5451 .arg5_type = ARG_CONST_SIZE, 5452 }; 5453 5454 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx, 5455 int, level, int, optname, char *, optval, int, optlen) 5456 { 5457 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen); 5458 } 5459 5460 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = { 5461 .func = bpf_sock_addr_setsockopt, 5462 .gpl_only = false, 5463 .ret_type = RET_INTEGER, 5464 .arg1_type = ARG_PTR_TO_CTX, 5465 .arg2_type = ARG_ANYTHING, 5466 .arg3_type = ARG_ANYTHING, 5467 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5468 .arg5_type = ARG_CONST_SIZE, 5469 }; 5470 5471 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx, 5472 int, level, int, optname, char *, optval, int, optlen) 5473 { 5474 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen); 5475 } 5476 5477 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = { 5478 .func = bpf_sock_addr_getsockopt, 5479 .gpl_only = false, 5480 .ret_type = RET_INTEGER, 5481 .arg1_type = ARG_PTR_TO_CTX, 5482 .arg2_type = ARG_ANYTHING, 5483 .arg3_type = ARG_ANYTHING, 5484 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5485 .arg5_type = ARG_CONST_SIZE, 5486 }; 5487 5488 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5489 int, level, int, optname, char *, optval, int, optlen) 5490 { 5491 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen); 5492 } 5493 5494 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = { 5495 .func = bpf_sock_ops_setsockopt, 5496 .gpl_only = false, 5497 .ret_type = RET_INTEGER, 5498 .arg1_type = ARG_PTR_TO_CTX, 5499 .arg2_type = ARG_ANYTHING, 5500 .arg3_type = ARG_ANYTHING, 5501 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5502 .arg5_type = ARG_CONST_SIZE, 5503 }; 5504 5505 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock, 5506 int optname, const u8 **start) 5507 { 5508 struct sk_buff *syn_skb = bpf_sock->syn_skb; 5509 const u8 *hdr_start; 5510 int ret; 5511 5512 if (syn_skb) { 5513 /* sk is a request_sock here */ 5514 5515 if (optname == TCP_BPF_SYN) { 5516 hdr_start = syn_skb->data; 5517 ret = tcp_hdrlen(syn_skb); 5518 } else if (optname == TCP_BPF_SYN_IP) { 5519 hdr_start = skb_network_header(syn_skb); 5520 ret = skb_network_header_len(syn_skb) + 5521 tcp_hdrlen(syn_skb); 5522 } else { 5523 /* optname == TCP_BPF_SYN_MAC */ 5524 hdr_start = skb_mac_header(syn_skb); 5525 ret = skb_mac_header_len(syn_skb) + 5526 skb_network_header_len(syn_skb) + 5527 tcp_hdrlen(syn_skb); 5528 } 5529 } else { 5530 struct sock *sk = bpf_sock->sk; 5531 struct saved_syn *saved_syn; 5532 5533 if (sk->sk_state == TCP_NEW_SYN_RECV) 5534 /* synack retransmit. bpf_sock->syn_skb will 5535 * not be available. It has to resort to 5536 * saved_syn (if it is saved). 5537 */ 5538 saved_syn = inet_reqsk(sk)->saved_syn; 5539 else 5540 saved_syn = tcp_sk(sk)->saved_syn; 5541 5542 if (!saved_syn) 5543 return -ENOENT; 5544 5545 if (optname == TCP_BPF_SYN) { 5546 hdr_start = saved_syn->data + 5547 saved_syn->mac_hdrlen + 5548 saved_syn->network_hdrlen; 5549 ret = saved_syn->tcp_hdrlen; 5550 } else if (optname == TCP_BPF_SYN_IP) { 5551 hdr_start = saved_syn->data + 5552 saved_syn->mac_hdrlen; 5553 ret = saved_syn->network_hdrlen + 5554 saved_syn->tcp_hdrlen; 5555 } else { 5556 /* optname == TCP_BPF_SYN_MAC */ 5557 5558 /* TCP_SAVE_SYN may not have saved the mac hdr */ 5559 if (!saved_syn->mac_hdrlen) 5560 return -ENOENT; 5561 5562 hdr_start = saved_syn->data; 5563 ret = saved_syn->mac_hdrlen + 5564 saved_syn->network_hdrlen + 5565 saved_syn->tcp_hdrlen; 5566 } 5567 } 5568 5569 *start = hdr_start; 5570 return ret; 5571 } 5572 5573 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5574 int, level, int, optname, char *, optval, int, optlen) 5575 { 5576 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP && 5577 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) { 5578 int ret, copy_len = 0; 5579 const u8 *start; 5580 5581 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start); 5582 if (ret > 0) { 5583 copy_len = ret; 5584 if (optlen < copy_len) { 5585 copy_len = optlen; 5586 ret = -ENOSPC; 5587 } 5588 5589 memcpy(optval, start, copy_len); 5590 } 5591 5592 /* Zero out unused buffer at the end */ 5593 memset(optval + copy_len, 0, optlen - copy_len); 5594 5595 return ret; 5596 } 5597 5598 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen); 5599 } 5600 5601 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = { 5602 .func = bpf_sock_ops_getsockopt, 5603 .gpl_only = false, 5604 .ret_type = RET_INTEGER, 5605 .arg1_type = ARG_PTR_TO_CTX, 5606 .arg2_type = ARG_ANYTHING, 5607 .arg3_type = ARG_ANYTHING, 5608 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5609 .arg5_type = ARG_CONST_SIZE, 5610 }; 5611 5612 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock, 5613 int, argval) 5614 { 5615 struct sock *sk = bpf_sock->sk; 5616 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS; 5617 5618 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk)) 5619 return -EINVAL; 5620 5621 tcp_sk(sk)->bpf_sock_ops_cb_flags = val; 5622 5623 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS); 5624 } 5625 5626 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = { 5627 .func = bpf_sock_ops_cb_flags_set, 5628 .gpl_only = false, 5629 .ret_type = RET_INTEGER, 5630 .arg1_type = ARG_PTR_TO_CTX, 5631 .arg2_type = ARG_ANYTHING, 5632 }; 5633 5634 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly; 5635 EXPORT_SYMBOL_GPL(ipv6_bpf_stub); 5636 5637 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr, 5638 int, addr_len) 5639 { 5640 #ifdef CONFIG_INET 5641 struct sock *sk = ctx->sk; 5642 u32 flags = BIND_FROM_BPF; 5643 int err; 5644 5645 err = -EINVAL; 5646 if (addr_len < offsetofend(struct sockaddr, sa_family)) 5647 return err; 5648 if (addr->sa_family == AF_INET) { 5649 if (addr_len < sizeof(struct sockaddr_in)) 5650 return err; 5651 if (((struct sockaddr_in *)addr)->sin_port == htons(0)) 5652 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5653 return __inet_bind(sk, addr, addr_len, flags); 5654 #if IS_ENABLED(CONFIG_IPV6) 5655 } else if (addr->sa_family == AF_INET6) { 5656 if (addr_len < SIN6_LEN_RFC2133) 5657 return err; 5658 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0)) 5659 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5660 /* ipv6_bpf_stub cannot be NULL, since it's called from 5661 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded 5662 */ 5663 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags); 5664 #endif /* CONFIG_IPV6 */ 5665 } 5666 #endif /* CONFIG_INET */ 5667 5668 return -EAFNOSUPPORT; 5669 } 5670 5671 static const struct bpf_func_proto bpf_bind_proto = { 5672 .func = bpf_bind, 5673 .gpl_only = false, 5674 .ret_type = RET_INTEGER, 5675 .arg1_type = ARG_PTR_TO_CTX, 5676 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5677 .arg3_type = ARG_CONST_SIZE, 5678 }; 5679 5680 #ifdef CONFIG_XFRM 5681 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index, 5682 struct bpf_xfrm_state *, to, u32, size, u64, flags) 5683 { 5684 const struct sec_path *sp = skb_sec_path(skb); 5685 const struct xfrm_state *x; 5686 5687 if (!sp || unlikely(index >= sp->len || flags)) 5688 goto err_clear; 5689 5690 x = sp->xvec[index]; 5691 5692 if (unlikely(size != sizeof(struct bpf_xfrm_state))) 5693 goto err_clear; 5694 5695 to->reqid = x->props.reqid; 5696 to->spi = x->id.spi; 5697 to->family = x->props.family; 5698 to->ext = 0; 5699 5700 if (to->family == AF_INET6) { 5701 memcpy(to->remote_ipv6, x->props.saddr.a6, 5702 sizeof(to->remote_ipv6)); 5703 } else { 5704 to->remote_ipv4 = x->props.saddr.a4; 5705 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 5706 } 5707 5708 return 0; 5709 err_clear: 5710 memset(to, 0, size); 5711 return -EINVAL; 5712 } 5713 5714 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = { 5715 .func = bpf_skb_get_xfrm_state, 5716 .gpl_only = false, 5717 .ret_type = RET_INTEGER, 5718 .arg1_type = ARG_PTR_TO_CTX, 5719 .arg2_type = ARG_ANYTHING, 5720 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 5721 .arg4_type = ARG_CONST_SIZE, 5722 .arg5_type = ARG_ANYTHING, 5723 }; 5724 #endif 5725 5726 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6) 5727 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, 5728 const struct neighbour *neigh, 5729 const struct net_device *dev, u32 mtu) 5730 { 5731 memcpy(params->dmac, neigh->ha, ETH_ALEN); 5732 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 5733 params->h_vlan_TCI = 0; 5734 params->h_vlan_proto = 0; 5735 if (mtu) 5736 params->mtu_result = mtu; /* union with tot_len */ 5737 5738 return 0; 5739 } 5740 #endif 5741 5742 #if IS_ENABLED(CONFIG_INET) 5743 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5744 u32 flags, bool check_mtu) 5745 { 5746 struct fib_nh_common *nhc; 5747 struct in_device *in_dev; 5748 struct neighbour *neigh; 5749 struct net_device *dev; 5750 struct fib_result res; 5751 struct flowi4 fl4; 5752 u32 mtu = 0; 5753 int err; 5754 5755 dev = dev_get_by_index_rcu(net, params->ifindex); 5756 if (unlikely(!dev)) 5757 return -ENODEV; 5758 5759 /* verify forwarding is enabled on this interface */ 5760 in_dev = __in_dev_get_rcu(dev); 5761 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev))) 5762 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5763 5764 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5765 fl4.flowi4_iif = 1; 5766 fl4.flowi4_oif = params->ifindex; 5767 } else { 5768 fl4.flowi4_iif = params->ifindex; 5769 fl4.flowi4_oif = 0; 5770 } 5771 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK; 5772 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 5773 fl4.flowi4_flags = 0; 5774 5775 fl4.flowi4_proto = params->l4_protocol; 5776 fl4.daddr = params->ipv4_dst; 5777 fl4.saddr = params->ipv4_src; 5778 fl4.fl4_sport = params->sport; 5779 fl4.fl4_dport = params->dport; 5780 fl4.flowi4_multipath_hash = 0; 5781 5782 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5783 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5784 struct fib_table *tb; 5785 5786 tb = fib_get_table(net, tbid); 5787 if (unlikely(!tb)) 5788 return BPF_FIB_LKUP_RET_NOT_FWDED; 5789 5790 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 5791 } else { 5792 fl4.flowi4_mark = 0; 5793 fl4.flowi4_secid = 0; 5794 fl4.flowi4_tun_key.tun_id = 0; 5795 fl4.flowi4_uid = sock_net_uid(net, NULL); 5796 5797 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF); 5798 } 5799 5800 if (err) { 5801 /* map fib lookup errors to RTN_ type */ 5802 if (err == -EINVAL) 5803 return BPF_FIB_LKUP_RET_BLACKHOLE; 5804 if (err == -EHOSTUNREACH) 5805 return BPF_FIB_LKUP_RET_UNREACHABLE; 5806 if (err == -EACCES) 5807 return BPF_FIB_LKUP_RET_PROHIBIT; 5808 5809 return BPF_FIB_LKUP_RET_NOT_FWDED; 5810 } 5811 5812 if (res.type != RTN_UNICAST) 5813 return BPF_FIB_LKUP_RET_NOT_FWDED; 5814 5815 if (fib_info_num_path(res.fi) > 1) 5816 fib_select_path(net, &res, &fl4, NULL); 5817 5818 if (check_mtu) { 5819 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst); 5820 if (params->tot_len > mtu) { 5821 params->mtu_result = mtu; /* union with tot_len */ 5822 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5823 } 5824 } 5825 5826 nhc = res.nhc; 5827 5828 /* do not handle lwt encaps right now */ 5829 if (nhc->nhc_lwtstate) 5830 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5831 5832 dev = nhc->nhc_dev; 5833 5834 params->rt_metric = res.fi->fib_priority; 5835 params->ifindex = dev->ifindex; 5836 5837 /* xdp and cls_bpf programs are run in RCU-bh so 5838 * rcu_read_lock_bh is not needed here 5839 */ 5840 if (likely(nhc->nhc_gw_family != AF_INET6)) { 5841 if (nhc->nhc_gw_family) 5842 params->ipv4_dst = nhc->nhc_gw.ipv4; 5843 5844 neigh = __ipv4_neigh_lookup_noref(dev, 5845 (__force u32)params->ipv4_dst); 5846 } else { 5847 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst; 5848 5849 params->family = AF_INET6; 5850 *dst = nhc->nhc_gw.ipv6; 5851 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5852 } 5853 5854 if (!neigh) 5855 return BPF_FIB_LKUP_RET_NO_NEIGH; 5856 5857 return bpf_fib_set_fwd_params(params, neigh, dev, mtu); 5858 } 5859 #endif 5860 5861 #if IS_ENABLED(CONFIG_IPV6) 5862 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5863 u32 flags, bool check_mtu) 5864 { 5865 struct in6_addr *src = (struct in6_addr *) params->ipv6_src; 5866 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst; 5867 struct fib6_result res = {}; 5868 struct neighbour *neigh; 5869 struct net_device *dev; 5870 struct inet6_dev *idev; 5871 struct flowi6 fl6; 5872 int strict = 0; 5873 int oif, err; 5874 u32 mtu = 0; 5875 5876 /* link local addresses are never forwarded */ 5877 if (rt6_need_strict(dst) || rt6_need_strict(src)) 5878 return BPF_FIB_LKUP_RET_NOT_FWDED; 5879 5880 dev = dev_get_by_index_rcu(net, params->ifindex); 5881 if (unlikely(!dev)) 5882 return -ENODEV; 5883 5884 idev = __in6_dev_get_safely(dev); 5885 if (unlikely(!idev || !idev->cnf.forwarding)) 5886 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5887 5888 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5889 fl6.flowi6_iif = 1; 5890 oif = fl6.flowi6_oif = params->ifindex; 5891 } else { 5892 oif = fl6.flowi6_iif = params->ifindex; 5893 fl6.flowi6_oif = 0; 5894 strict = RT6_LOOKUP_F_HAS_SADDR; 5895 } 5896 fl6.flowlabel = params->flowinfo; 5897 fl6.flowi6_scope = 0; 5898 fl6.flowi6_flags = 0; 5899 fl6.mp_hash = 0; 5900 5901 fl6.flowi6_proto = params->l4_protocol; 5902 fl6.daddr = *dst; 5903 fl6.saddr = *src; 5904 fl6.fl6_sport = params->sport; 5905 fl6.fl6_dport = params->dport; 5906 5907 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5908 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5909 struct fib6_table *tb; 5910 5911 tb = ipv6_stub->fib6_get_table(net, tbid); 5912 if (unlikely(!tb)) 5913 return BPF_FIB_LKUP_RET_NOT_FWDED; 5914 5915 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res, 5916 strict); 5917 } else { 5918 fl6.flowi6_mark = 0; 5919 fl6.flowi6_secid = 0; 5920 fl6.flowi6_tun_key.tun_id = 0; 5921 fl6.flowi6_uid = sock_net_uid(net, NULL); 5922 5923 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict); 5924 } 5925 5926 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) || 5927 res.f6i == net->ipv6.fib6_null_entry)) 5928 return BPF_FIB_LKUP_RET_NOT_FWDED; 5929 5930 switch (res.fib6_type) { 5931 /* only unicast is forwarded */ 5932 case RTN_UNICAST: 5933 break; 5934 case RTN_BLACKHOLE: 5935 return BPF_FIB_LKUP_RET_BLACKHOLE; 5936 case RTN_UNREACHABLE: 5937 return BPF_FIB_LKUP_RET_UNREACHABLE; 5938 case RTN_PROHIBIT: 5939 return BPF_FIB_LKUP_RET_PROHIBIT; 5940 default: 5941 return BPF_FIB_LKUP_RET_NOT_FWDED; 5942 } 5943 5944 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif, 5945 fl6.flowi6_oif != 0, NULL, strict); 5946 5947 if (check_mtu) { 5948 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src); 5949 if (params->tot_len > mtu) { 5950 params->mtu_result = mtu; /* union with tot_len */ 5951 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5952 } 5953 } 5954 5955 if (res.nh->fib_nh_lws) 5956 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5957 5958 if (res.nh->fib_nh_gw_family) 5959 *dst = res.nh->fib_nh_gw6; 5960 5961 dev = res.nh->fib_nh_dev; 5962 params->rt_metric = res.f6i->fib6_metric; 5963 params->ifindex = dev->ifindex; 5964 5965 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is 5966 * not needed here. 5967 */ 5968 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5969 if (!neigh) 5970 return BPF_FIB_LKUP_RET_NO_NEIGH; 5971 5972 return bpf_fib_set_fwd_params(params, neigh, dev, mtu); 5973 } 5974 #endif 5975 5976 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx, 5977 struct bpf_fib_lookup *, params, int, plen, u32, flags) 5978 { 5979 if (plen < sizeof(*params)) 5980 return -EINVAL; 5981 5982 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 5983 return -EINVAL; 5984 5985 switch (params->family) { 5986 #if IS_ENABLED(CONFIG_INET) 5987 case AF_INET: 5988 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params, 5989 flags, true); 5990 #endif 5991 #if IS_ENABLED(CONFIG_IPV6) 5992 case AF_INET6: 5993 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params, 5994 flags, true); 5995 #endif 5996 } 5997 return -EAFNOSUPPORT; 5998 } 5999 6000 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = { 6001 .func = bpf_xdp_fib_lookup, 6002 .gpl_only = true, 6003 .ret_type = RET_INTEGER, 6004 .arg1_type = ARG_PTR_TO_CTX, 6005 .arg2_type = ARG_PTR_TO_MEM, 6006 .arg3_type = ARG_CONST_SIZE, 6007 .arg4_type = ARG_ANYTHING, 6008 }; 6009 6010 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb, 6011 struct bpf_fib_lookup *, params, int, plen, u32, flags) 6012 { 6013 struct net *net = dev_net(skb->dev); 6014 int rc = -EAFNOSUPPORT; 6015 bool check_mtu = false; 6016 6017 if (plen < sizeof(*params)) 6018 return -EINVAL; 6019 6020 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 6021 return -EINVAL; 6022 6023 if (params->tot_len) 6024 check_mtu = true; 6025 6026 switch (params->family) { 6027 #if IS_ENABLED(CONFIG_INET) 6028 case AF_INET: 6029 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu); 6030 break; 6031 #endif 6032 #if IS_ENABLED(CONFIG_IPV6) 6033 case AF_INET6: 6034 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu); 6035 break; 6036 #endif 6037 } 6038 6039 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) { 6040 struct net_device *dev; 6041 6042 /* When tot_len isn't provided by user, check skb 6043 * against MTU of FIB lookup resulting net_device 6044 */ 6045 dev = dev_get_by_index_rcu(net, params->ifindex); 6046 if (!is_skb_forwardable(dev, skb)) 6047 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED; 6048 6049 params->mtu_result = dev->mtu; /* union with tot_len */ 6050 } 6051 6052 return rc; 6053 } 6054 6055 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = { 6056 .func = bpf_skb_fib_lookup, 6057 .gpl_only = true, 6058 .ret_type = RET_INTEGER, 6059 .arg1_type = ARG_PTR_TO_CTX, 6060 .arg2_type = ARG_PTR_TO_MEM, 6061 .arg3_type = ARG_CONST_SIZE, 6062 .arg4_type = ARG_ANYTHING, 6063 }; 6064 6065 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr, 6066 u32 ifindex) 6067 { 6068 struct net *netns = dev_net(dev_curr); 6069 6070 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */ 6071 if (ifindex == 0) 6072 return dev_curr; 6073 6074 return dev_get_by_index_rcu(netns, ifindex); 6075 } 6076 6077 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb, 6078 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 6079 { 6080 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 6081 struct net_device *dev = skb->dev; 6082 int skb_len, dev_len; 6083 int mtu; 6084 6085 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS))) 6086 return -EINVAL; 6087 6088 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len))) 6089 return -EINVAL; 6090 6091 dev = __dev_via_ifindex(dev, ifindex); 6092 if (unlikely(!dev)) 6093 return -ENODEV; 6094 6095 mtu = READ_ONCE(dev->mtu); 6096 6097 dev_len = mtu + dev->hard_header_len; 6098 6099 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 6100 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len; 6101 6102 skb_len += len_diff; /* minus result pass check */ 6103 if (skb_len <= dev_len) { 6104 ret = BPF_MTU_CHK_RET_SUCCESS; 6105 goto out; 6106 } 6107 /* At this point, skb->len exceed MTU, but as it include length of all 6108 * segments, it can still be below MTU. The SKB can possibly get 6109 * re-segmented in transmit path (see validate_xmit_skb). Thus, user 6110 * must choose if segs are to be MTU checked. 6111 */ 6112 if (skb_is_gso(skb)) { 6113 ret = BPF_MTU_CHK_RET_SUCCESS; 6114 6115 if (flags & BPF_MTU_CHK_SEGS && 6116 !skb_gso_validate_network_len(skb, mtu)) 6117 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG; 6118 } 6119 out: 6120 /* BPF verifier guarantees valid pointer */ 6121 *mtu_len = mtu; 6122 6123 return ret; 6124 } 6125 6126 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp, 6127 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 6128 { 6129 struct net_device *dev = xdp->rxq->dev; 6130 int xdp_len = xdp->data_end - xdp->data; 6131 int ret = BPF_MTU_CHK_RET_SUCCESS; 6132 int mtu, dev_len; 6133 6134 /* XDP variant doesn't support multi-buffer segment check (yet) */ 6135 if (unlikely(flags)) 6136 return -EINVAL; 6137 6138 dev = __dev_via_ifindex(dev, ifindex); 6139 if (unlikely(!dev)) 6140 return -ENODEV; 6141 6142 mtu = READ_ONCE(dev->mtu); 6143 6144 /* Add L2-header as dev MTU is L3 size */ 6145 dev_len = mtu + dev->hard_header_len; 6146 6147 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 6148 if (*mtu_len) 6149 xdp_len = *mtu_len + dev->hard_header_len; 6150 6151 xdp_len += len_diff; /* minus result pass check */ 6152 if (xdp_len > dev_len) 6153 ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 6154 6155 /* BPF verifier guarantees valid pointer */ 6156 *mtu_len = mtu; 6157 6158 return ret; 6159 } 6160 6161 static const struct bpf_func_proto bpf_skb_check_mtu_proto = { 6162 .func = bpf_skb_check_mtu, 6163 .gpl_only = true, 6164 .ret_type = RET_INTEGER, 6165 .arg1_type = ARG_PTR_TO_CTX, 6166 .arg2_type = ARG_ANYTHING, 6167 .arg3_type = ARG_PTR_TO_INT, 6168 .arg4_type = ARG_ANYTHING, 6169 .arg5_type = ARG_ANYTHING, 6170 }; 6171 6172 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = { 6173 .func = bpf_xdp_check_mtu, 6174 .gpl_only = true, 6175 .ret_type = RET_INTEGER, 6176 .arg1_type = ARG_PTR_TO_CTX, 6177 .arg2_type = ARG_ANYTHING, 6178 .arg3_type = ARG_PTR_TO_INT, 6179 .arg4_type = ARG_ANYTHING, 6180 .arg5_type = ARG_ANYTHING, 6181 }; 6182 6183 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6184 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len) 6185 { 6186 int err; 6187 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr; 6188 6189 if (!seg6_validate_srh(srh, len, false)) 6190 return -EINVAL; 6191 6192 switch (type) { 6193 case BPF_LWT_ENCAP_SEG6_INLINE: 6194 if (skb->protocol != htons(ETH_P_IPV6)) 6195 return -EBADMSG; 6196 6197 err = seg6_do_srh_inline(skb, srh); 6198 break; 6199 case BPF_LWT_ENCAP_SEG6: 6200 skb_reset_inner_headers(skb); 6201 skb->encapsulation = 1; 6202 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6); 6203 break; 6204 default: 6205 return -EINVAL; 6206 } 6207 6208 bpf_compute_data_pointers(skb); 6209 if (err) 6210 return err; 6211 6212 skb_set_transport_header(skb, sizeof(struct ipv6hdr)); 6213 6214 return seg6_lookup_nexthop(skb, NULL, 0); 6215 } 6216 #endif /* CONFIG_IPV6_SEG6_BPF */ 6217 6218 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6219 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, 6220 bool ingress) 6221 { 6222 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress); 6223 } 6224 #endif 6225 6226 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr, 6227 u32, len) 6228 { 6229 switch (type) { 6230 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6231 case BPF_LWT_ENCAP_SEG6: 6232 case BPF_LWT_ENCAP_SEG6_INLINE: 6233 return bpf_push_seg6_encap(skb, type, hdr, len); 6234 #endif 6235 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6236 case BPF_LWT_ENCAP_IP: 6237 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */); 6238 #endif 6239 default: 6240 return -EINVAL; 6241 } 6242 } 6243 6244 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type, 6245 void *, hdr, u32, len) 6246 { 6247 switch (type) { 6248 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6249 case BPF_LWT_ENCAP_IP: 6250 return bpf_push_ip_encap(skb, hdr, len, false /* egress */); 6251 #endif 6252 default: 6253 return -EINVAL; 6254 } 6255 } 6256 6257 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = { 6258 .func = bpf_lwt_in_push_encap, 6259 .gpl_only = false, 6260 .ret_type = RET_INTEGER, 6261 .arg1_type = ARG_PTR_TO_CTX, 6262 .arg2_type = ARG_ANYTHING, 6263 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6264 .arg4_type = ARG_CONST_SIZE 6265 }; 6266 6267 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = { 6268 .func = bpf_lwt_xmit_push_encap, 6269 .gpl_only = false, 6270 .ret_type = RET_INTEGER, 6271 .arg1_type = ARG_PTR_TO_CTX, 6272 .arg2_type = ARG_ANYTHING, 6273 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6274 .arg4_type = ARG_CONST_SIZE 6275 }; 6276 6277 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6278 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset, 6279 const void *, from, u32, len) 6280 { 6281 struct seg6_bpf_srh_state *srh_state = 6282 this_cpu_ptr(&seg6_bpf_srh_states); 6283 struct ipv6_sr_hdr *srh = srh_state->srh; 6284 void *srh_tlvs, *srh_end, *ptr; 6285 int srhoff = 0; 6286 6287 if (srh == NULL) 6288 return -EINVAL; 6289 6290 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4)); 6291 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen); 6292 6293 ptr = skb->data + offset; 6294 if (ptr >= srh_tlvs && ptr + len <= srh_end) 6295 srh_state->valid = false; 6296 else if (ptr < (void *)&srh->flags || 6297 ptr + len > (void *)&srh->segments) 6298 return -EFAULT; 6299 6300 if (unlikely(bpf_try_make_writable(skb, offset + len))) 6301 return -EFAULT; 6302 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 6303 return -EINVAL; 6304 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6305 6306 memcpy(skb->data + offset, from, len); 6307 return 0; 6308 } 6309 6310 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = { 6311 .func = bpf_lwt_seg6_store_bytes, 6312 .gpl_only = false, 6313 .ret_type = RET_INTEGER, 6314 .arg1_type = ARG_PTR_TO_CTX, 6315 .arg2_type = ARG_ANYTHING, 6316 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6317 .arg4_type = ARG_CONST_SIZE 6318 }; 6319 6320 static void bpf_update_srh_state(struct sk_buff *skb) 6321 { 6322 struct seg6_bpf_srh_state *srh_state = 6323 this_cpu_ptr(&seg6_bpf_srh_states); 6324 int srhoff = 0; 6325 6326 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) { 6327 srh_state->srh = NULL; 6328 } else { 6329 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6330 srh_state->hdrlen = srh_state->srh->hdrlen << 3; 6331 srh_state->valid = true; 6332 } 6333 } 6334 6335 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb, 6336 u32, action, void *, param, u32, param_len) 6337 { 6338 struct seg6_bpf_srh_state *srh_state = 6339 this_cpu_ptr(&seg6_bpf_srh_states); 6340 int hdroff = 0; 6341 int err; 6342 6343 switch (action) { 6344 case SEG6_LOCAL_ACTION_END_X: 6345 if (!seg6_bpf_has_valid_srh(skb)) 6346 return -EBADMSG; 6347 if (param_len != sizeof(struct in6_addr)) 6348 return -EINVAL; 6349 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0); 6350 case SEG6_LOCAL_ACTION_END_T: 6351 if (!seg6_bpf_has_valid_srh(skb)) 6352 return -EBADMSG; 6353 if (param_len != sizeof(int)) 6354 return -EINVAL; 6355 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 6356 case SEG6_LOCAL_ACTION_END_DT6: 6357 if (!seg6_bpf_has_valid_srh(skb)) 6358 return -EBADMSG; 6359 if (param_len != sizeof(int)) 6360 return -EINVAL; 6361 6362 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0) 6363 return -EBADMSG; 6364 if (!pskb_pull(skb, hdroff)) 6365 return -EBADMSG; 6366 6367 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff); 6368 skb_reset_network_header(skb); 6369 skb_reset_transport_header(skb); 6370 skb->encapsulation = 0; 6371 6372 bpf_compute_data_pointers(skb); 6373 bpf_update_srh_state(skb); 6374 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 6375 case SEG6_LOCAL_ACTION_END_B6: 6376 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 6377 return -EBADMSG; 6378 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE, 6379 param, param_len); 6380 if (!err) 6381 bpf_update_srh_state(skb); 6382 6383 return err; 6384 case SEG6_LOCAL_ACTION_END_B6_ENCAP: 6385 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 6386 return -EBADMSG; 6387 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6, 6388 param, param_len); 6389 if (!err) 6390 bpf_update_srh_state(skb); 6391 6392 return err; 6393 default: 6394 return -EINVAL; 6395 } 6396 } 6397 6398 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = { 6399 .func = bpf_lwt_seg6_action, 6400 .gpl_only = false, 6401 .ret_type = RET_INTEGER, 6402 .arg1_type = ARG_PTR_TO_CTX, 6403 .arg2_type = ARG_ANYTHING, 6404 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6405 .arg4_type = ARG_CONST_SIZE 6406 }; 6407 6408 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset, 6409 s32, len) 6410 { 6411 struct seg6_bpf_srh_state *srh_state = 6412 this_cpu_ptr(&seg6_bpf_srh_states); 6413 struct ipv6_sr_hdr *srh = srh_state->srh; 6414 void *srh_end, *srh_tlvs, *ptr; 6415 struct ipv6hdr *hdr; 6416 int srhoff = 0; 6417 int ret; 6418 6419 if (unlikely(srh == NULL)) 6420 return -EINVAL; 6421 6422 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) + 6423 ((srh->first_segment + 1) << 4)); 6424 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) + 6425 srh_state->hdrlen); 6426 ptr = skb->data + offset; 6427 6428 if (unlikely(ptr < srh_tlvs || ptr > srh_end)) 6429 return -EFAULT; 6430 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end)) 6431 return -EFAULT; 6432 6433 if (len > 0) { 6434 ret = skb_cow_head(skb, len); 6435 if (unlikely(ret < 0)) 6436 return ret; 6437 6438 ret = bpf_skb_net_hdr_push(skb, offset, len); 6439 } else { 6440 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len); 6441 } 6442 6443 bpf_compute_data_pointers(skb); 6444 if (unlikely(ret < 0)) 6445 return ret; 6446 6447 hdr = (struct ipv6hdr *)skb->data; 6448 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 6449 6450 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 6451 return -EINVAL; 6452 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6453 srh_state->hdrlen += len; 6454 srh_state->valid = false; 6455 return 0; 6456 } 6457 6458 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = { 6459 .func = bpf_lwt_seg6_adjust_srh, 6460 .gpl_only = false, 6461 .ret_type = RET_INTEGER, 6462 .arg1_type = ARG_PTR_TO_CTX, 6463 .arg2_type = ARG_ANYTHING, 6464 .arg3_type = ARG_ANYTHING, 6465 }; 6466 #endif /* CONFIG_IPV6_SEG6_BPF */ 6467 6468 #ifdef CONFIG_INET 6469 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple, 6470 int dif, int sdif, u8 family, u8 proto) 6471 { 6472 bool refcounted = false; 6473 struct sock *sk = NULL; 6474 6475 if (family == AF_INET) { 6476 __be32 src4 = tuple->ipv4.saddr; 6477 __be32 dst4 = tuple->ipv4.daddr; 6478 6479 if (proto == IPPROTO_TCP) 6480 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0, 6481 src4, tuple->ipv4.sport, 6482 dst4, tuple->ipv4.dport, 6483 dif, sdif, &refcounted); 6484 else 6485 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport, 6486 dst4, tuple->ipv4.dport, 6487 dif, sdif, &udp_table, NULL); 6488 #if IS_ENABLED(CONFIG_IPV6) 6489 } else { 6490 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr; 6491 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr; 6492 6493 if (proto == IPPROTO_TCP) 6494 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0, 6495 src6, tuple->ipv6.sport, 6496 dst6, ntohs(tuple->ipv6.dport), 6497 dif, sdif, &refcounted); 6498 else if (likely(ipv6_bpf_stub)) 6499 sk = ipv6_bpf_stub->udp6_lib_lookup(net, 6500 src6, tuple->ipv6.sport, 6501 dst6, tuple->ipv6.dport, 6502 dif, sdif, 6503 &udp_table, NULL); 6504 #endif 6505 } 6506 6507 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) { 6508 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6509 sk = NULL; 6510 } 6511 return sk; 6512 } 6513 6514 /* bpf_skc_lookup performs the core lookup for different types of sockets, 6515 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE. 6516 */ 6517 static struct sock * 6518 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6519 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 6520 u64 flags) 6521 { 6522 struct sock *sk = NULL; 6523 struct net *net; 6524 u8 family; 6525 int sdif; 6526 6527 if (len == sizeof(tuple->ipv4)) 6528 family = AF_INET; 6529 else if (len == sizeof(tuple->ipv6)) 6530 family = AF_INET6; 6531 else 6532 return NULL; 6533 6534 if (unlikely(flags || !((s32)netns_id < 0 || netns_id <= S32_MAX))) 6535 goto out; 6536 6537 if (family == AF_INET) 6538 sdif = inet_sdif(skb); 6539 else 6540 sdif = inet6_sdif(skb); 6541 6542 if ((s32)netns_id < 0) { 6543 net = caller_net; 6544 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 6545 } else { 6546 net = get_net_ns_by_id(caller_net, netns_id); 6547 if (unlikely(!net)) 6548 goto out; 6549 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 6550 put_net(net); 6551 } 6552 6553 out: 6554 return sk; 6555 } 6556 6557 static struct sock * 6558 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6559 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 6560 u64 flags) 6561 { 6562 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net, 6563 ifindex, proto, netns_id, flags); 6564 6565 if (sk) { 6566 struct sock *sk2 = sk_to_full_sk(sk); 6567 6568 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk 6569 * sock refcnt is decremented to prevent a request_sock leak. 6570 */ 6571 if (!sk_fullsock(sk2)) 6572 sk2 = NULL; 6573 if (sk2 != sk) { 6574 sock_gen_put(sk); 6575 /* Ensure there is no need to bump sk2 refcnt */ 6576 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) { 6577 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6578 return NULL; 6579 } 6580 sk = sk2; 6581 } 6582 } 6583 6584 return sk; 6585 } 6586 6587 static struct sock * 6588 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6589 u8 proto, u64 netns_id, u64 flags) 6590 { 6591 struct net *caller_net; 6592 int ifindex; 6593 6594 if (skb->dev) { 6595 caller_net = dev_net(skb->dev); 6596 ifindex = skb->dev->ifindex; 6597 } else { 6598 caller_net = sock_net(skb->sk); 6599 ifindex = 0; 6600 } 6601 6602 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto, 6603 netns_id, flags); 6604 } 6605 6606 static struct sock * 6607 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6608 u8 proto, u64 netns_id, u64 flags) 6609 { 6610 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id, 6611 flags); 6612 6613 if (sk) { 6614 struct sock *sk2 = sk_to_full_sk(sk); 6615 6616 /* sk_to_full_sk() may return (sk)->rsk_listener, so make sure the original sk 6617 * sock refcnt is decremented to prevent a request_sock leak. 6618 */ 6619 if (!sk_fullsock(sk2)) 6620 sk2 = NULL; 6621 if (sk2 != sk) { 6622 sock_gen_put(sk); 6623 /* Ensure there is no need to bump sk2 refcnt */ 6624 if (unlikely(sk2 && !sock_flag(sk2, SOCK_RCU_FREE))) { 6625 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6626 return NULL; 6627 } 6628 sk = sk2; 6629 } 6630 } 6631 6632 return sk; 6633 } 6634 6635 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb, 6636 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6637 { 6638 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP, 6639 netns_id, flags); 6640 } 6641 6642 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = { 6643 .func = bpf_skc_lookup_tcp, 6644 .gpl_only = false, 6645 .pkt_access = true, 6646 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6647 .arg1_type = ARG_PTR_TO_CTX, 6648 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6649 .arg3_type = ARG_CONST_SIZE, 6650 .arg4_type = ARG_ANYTHING, 6651 .arg5_type = ARG_ANYTHING, 6652 }; 6653 6654 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb, 6655 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6656 { 6657 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, 6658 netns_id, flags); 6659 } 6660 6661 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = { 6662 .func = bpf_sk_lookup_tcp, 6663 .gpl_only = false, 6664 .pkt_access = true, 6665 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6666 .arg1_type = ARG_PTR_TO_CTX, 6667 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6668 .arg3_type = ARG_CONST_SIZE, 6669 .arg4_type = ARG_ANYTHING, 6670 .arg5_type = ARG_ANYTHING, 6671 }; 6672 6673 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb, 6674 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6675 { 6676 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, 6677 netns_id, flags); 6678 } 6679 6680 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = { 6681 .func = bpf_sk_lookup_udp, 6682 .gpl_only = false, 6683 .pkt_access = true, 6684 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6685 .arg1_type = ARG_PTR_TO_CTX, 6686 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6687 .arg3_type = ARG_CONST_SIZE, 6688 .arg4_type = ARG_ANYTHING, 6689 .arg5_type = ARG_ANYTHING, 6690 }; 6691 6692 BPF_CALL_1(bpf_sk_release, struct sock *, sk) 6693 { 6694 if (sk && sk_is_refcounted(sk)) 6695 sock_gen_put(sk); 6696 return 0; 6697 } 6698 6699 static const struct bpf_func_proto bpf_sk_release_proto = { 6700 .func = bpf_sk_release, 6701 .gpl_only = false, 6702 .ret_type = RET_INTEGER, 6703 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE, 6704 }; 6705 6706 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx, 6707 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6708 { 6709 struct net *caller_net = dev_net(ctx->rxq->dev); 6710 int ifindex = ctx->rxq->dev->ifindex; 6711 6712 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6713 ifindex, IPPROTO_UDP, netns_id, 6714 flags); 6715 } 6716 6717 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = { 6718 .func = bpf_xdp_sk_lookup_udp, 6719 .gpl_only = false, 6720 .pkt_access = true, 6721 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6722 .arg1_type = ARG_PTR_TO_CTX, 6723 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6724 .arg3_type = ARG_CONST_SIZE, 6725 .arg4_type = ARG_ANYTHING, 6726 .arg5_type = ARG_ANYTHING, 6727 }; 6728 6729 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx, 6730 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6731 { 6732 struct net *caller_net = dev_net(ctx->rxq->dev); 6733 int ifindex = ctx->rxq->dev->ifindex; 6734 6735 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net, 6736 ifindex, IPPROTO_TCP, netns_id, 6737 flags); 6738 } 6739 6740 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = { 6741 .func = bpf_xdp_skc_lookup_tcp, 6742 .gpl_only = false, 6743 .pkt_access = true, 6744 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6745 .arg1_type = ARG_PTR_TO_CTX, 6746 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6747 .arg3_type = ARG_CONST_SIZE, 6748 .arg4_type = ARG_ANYTHING, 6749 .arg5_type = ARG_ANYTHING, 6750 }; 6751 6752 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx, 6753 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6754 { 6755 struct net *caller_net = dev_net(ctx->rxq->dev); 6756 int ifindex = ctx->rxq->dev->ifindex; 6757 6758 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6759 ifindex, IPPROTO_TCP, netns_id, 6760 flags); 6761 } 6762 6763 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = { 6764 .func = bpf_xdp_sk_lookup_tcp, 6765 .gpl_only = false, 6766 .pkt_access = true, 6767 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6768 .arg1_type = ARG_PTR_TO_CTX, 6769 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6770 .arg3_type = ARG_CONST_SIZE, 6771 .arg4_type = ARG_ANYTHING, 6772 .arg5_type = ARG_ANYTHING, 6773 }; 6774 6775 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6776 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6777 { 6778 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, 6779 sock_net(ctx->sk), 0, 6780 IPPROTO_TCP, netns_id, flags); 6781 } 6782 6783 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = { 6784 .func = bpf_sock_addr_skc_lookup_tcp, 6785 .gpl_only = false, 6786 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6787 .arg1_type = ARG_PTR_TO_CTX, 6788 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6789 .arg3_type = ARG_CONST_SIZE, 6790 .arg4_type = ARG_ANYTHING, 6791 .arg5_type = ARG_ANYTHING, 6792 }; 6793 6794 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6795 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6796 { 6797 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6798 sock_net(ctx->sk), 0, IPPROTO_TCP, 6799 netns_id, flags); 6800 } 6801 6802 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = { 6803 .func = bpf_sock_addr_sk_lookup_tcp, 6804 .gpl_only = false, 6805 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6806 .arg1_type = ARG_PTR_TO_CTX, 6807 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6808 .arg3_type = ARG_CONST_SIZE, 6809 .arg4_type = ARG_ANYTHING, 6810 .arg5_type = ARG_ANYTHING, 6811 }; 6812 6813 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx, 6814 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6815 { 6816 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6817 sock_net(ctx->sk), 0, IPPROTO_UDP, 6818 netns_id, flags); 6819 } 6820 6821 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = { 6822 .func = bpf_sock_addr_sk_lookup_udp, 6823 .gpl_only = false, 6824 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6825 .arg1_type = ARG_PTR_TO_CTX, 6826 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6827 .arg3_type = ARG_CONST_SIZE, 6828 .arg4_type = ARG_ANYTHING, 6829 .arg5_type = ARG_ANYTHING, 6830 }; 6831 6832 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6833 struct bpf_insn_access_aux *info) 6834 { 6835 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, 6836 icsk_retransmits)) 6837 return false; 6838 6839 if (off % size != 0) 6840 return false; 6841 6842 switch (off) { 6843 case offsetof(struct bpf_tcp_sock, bytes_received): 6844 case offsetof(struct bpf_tcp_sock, bytes_acked): 6845 return size == sizeof(__u64); 6846 default: 6847 return size == sizeof(__u32); 6848 } 6849 } 6850 6851 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 6852 const struct bpf_insn *si, 6853 struct bpf_insn *insn_buf, 6854 struct bpf_prog *prog, u32 *target_size) 6855 { 6856 struct bpf_insn *insn = insn_buf; 6857 6858 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \ 6859 do { \ 6860 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \ 6861 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6862 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\ 6863 si->dst_reg, si->src_reg, \ 6864 offsetof(struct tcp_sock, FIELD)); \ 6865 } while (0) 6866 6867 #define BPF_INET_SOCK_GET_COMMON(FIELD) \ 6868 do { \ 6869 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \ 6870 FIELD) > \ 6871 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6872 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 6873 struct inet_connection_sock, \ 6874 FIELD), \ 6875 si->dst_reg, si->src_reg, \ 6876 offsetof( \ 6877 struct inet_connection_sock, \ 6878 FIELD)); \ 6879 } while (0) 6880 6881 if (insn > insn_buf) 6882 return insn - insn_buf; 6883 6884 switch (si->off) { 6885 case offsetof(struct bpf_tcp_sock, rtt_min): 6886 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 6887 sizeof(struct minmax)); 6888 BUILD_BUG_ON(sizeof(struct minmax) < 6889 sizeof(struct minmax_sample)); 6890 6891 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 6892 offsetof(struct tcp_sock, rtt_min) + 6893 offsetof(struct minmax_sample, v)); 6894 break; 6895 case offsetof(struct bpf_tcp_sock, snd_cwnd): 6896 BPF_TCP_SOCK_GET_COMMON(snd_cwnd); 6897 break; 6898 case offsetof(struct bpf_tcp_sock, srtt_us): 6899 BPF_TCP_SOCK_GET_COMMON(srtt_us); 6900 break; 6901 case offsetof(struct bpf_tcp_sock, snd_ssthresh): 6902 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh); 6903 break; 6904 case offsetof(struct bpf_tcp_sock, rcv_nxt): 6905 BPF_TCP_SOCK_GET_COMMON(rcv_nxt); 6906 break; 6907 case offsetof(struct bpf_tcp_sock, snd_nxt): 6908 BPF_TCP_SOCK_GET_COMMON(snd_nxt); 6909 break; 6910 case offsetof(struct bpf_tcp_sock, snd_una): 6911 BPF_TCP_SOCK_GET_COMMON(snd_una); 6912 break; 6913 case offsetof(struct bpf_tcp_sock, mss_cache): 6914 BPF_TCP_SOCK_GET_COMMON(mss_cache); 6915 break; 6916 case offsetof(struct bpf_tcp_sock, ecn_flags): 6917 BPF_TCP_SOCK_GET_COMMON(ecn_flags); 6918 break; 6919 case offsetof(struct bpf_tcp_sock, rate_delivered): 6920 BPF_TCP_SOCK_GET_COMMON(rate_delivered); 6921 break; 6922 case offsetof(struct bpf_tcp_sock, rate_interval_us): 6923 BPF_TCP_SOCK_GET_COMMON(rate_interval_us); 6924 break; 6925 case offsetof(struct bpf_tcp_sock, packets_out): 6926 BPF_TCP_SOCK_GET_COMMON(packets_out); 6927 break; 6928 case offsetof(struct bpf_tcp_sock, retrans_out): 6929 BPF_TCP_SOCK_GET_COMMON(retrans_out); 6930 break; 6931 case offsetof(struct bpf_tcp_sock, total_retrans): 6932 BPF_TCP_SOCK_GET_COMMON(total_retrans); 6933 break; 6934 case offsetof(struct bpf_tcp_sock, segs_in): 6935 BPF_TCP_SOCK_GET_COMMON(segs_in); 6936 break; 6937 case offsetof(struct bpf_tcp_sock, data_segs_in): 6938 BPF_TCP_SOCK_GET_COMMON(data_segs_in); 6939 break; 6940 case offsetof(struct bpf_tcp_sock, segs_out): 6941 BPF_TCP_SOCK_GET_COMMON(segs_out); 6942 break; 6943 case offsetof(struct bpf_tcp_sock, data_segs_out): 6944 BPF_TCP_SOCK_GET_COMMON(data_segs_out); 6945 break; 6946 case offsetof(struct bpf_tcp_sock, lost_out): 6947 BPF_TCP_SOCK_GET_COMMON(lost_out); 6948 break; 6949 case offsetof(struct bpf_tcp_sock, sacked_out): 6950 BPF_TCP_SOCK_GET_COMMON(sacked_out); 6951 break; 6952 case offsetof(struct bpf_tcp_sock, bytes_received): 6953 BPF_TCP_SOCK_GET_COMMON(bytes_received); 6954 break; 6955 case offsetof(struct bpf_tcp_sock, bytes_acked): 6956 BPF_TCP_SOCK_GET_COMMON(bytes_acked); 6957 break; 6958 case offsetof(struct bpf_tcp_sock, dsack_dups): 6959 BPF_TCP_SOCK_GET_COMMON(dsack_dups); 6960 break; 6961 case offsetof(struct bpf_tcp_sock, delivered): 6962 BPF_TCP_SOCK_GET_COMMON(delivered); 6963 break; 6964 case offsetof(struct bpf_tcp_sock, delivered_ce): 6965 BPF_TCP_SOCK_GET_COMMON(delivered_ce); 6966 break; 6967 case offsetof(struct bpf_tcp_sock, icsk_retransmits): 6968 BPF_INET_SOCK_GET_COMMON(icsk_retransmits); 6969 break; 6970 } 6971 6972 return insn - insn_buf; 6973 } 6974 6975 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk) 6976 { 6977 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 6978 return (unsigned long)sk; 6979 6980 return (unsigned long)NULL; 6981 } 6982 6983 const struct bpf_func_proto bpf_tcp_sock_proto = { 6984 .func = bpf_tcp_sock, 6985 .gpl_only = false, 6986 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL, 6987 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 6988 }; 6989 6990 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk) 6991 { 6992 sk = sk_to_full_sk(sk); 6993 6994 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE)) 6995 return (unsigned long)sk; 6996 6997 return (unsigned long)NULL; 6998 } 6999 7000 static const struct bpf_func_proto bpf_get_listener_sock_proto = { 7001 .func = bpf_get_listener_sock, 7002 .gpl_only = false, 7003 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 7004 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 7005 }; 7006 7007 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb) 7008 { 7009 unsigned int iphdr_len; 7010 7011 switch (skb_protocol(skb, true)) { 7012 case cpu_to_be16(ETH_P_IP): 7013 iphdr_len = sizeof(struct iphdr); 7014 break; 7015 case cpu_to_be16(ETH_P_IPV6): 7016 iphdr_len = sizeof(struct ipv6hdr); 7017 break; 7018 default: 7019 return 0; 7020 } 7021 7022 if (skb_headlen(skb) < iphdr_len) 7023 return 0; 7024 7025 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len)) 7026 return 0; 7027 7028 return INET_ECN_set_ce(skb); 7029 } 7030 7031 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 7032 struct bpf_insn_access_aux *info) 7033 { 7034 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id)) 7035 return false; 7036 7037 if (off % size != 0) 7038 return false; 7039 7040 switch (off) { 7041 default: 7042 return size == sizeof(__u32); 7043 } 7044 } 7045 7046 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 7047 const struct bpf_insn *si, 7048 struct bpf_insn *insn_buf, 7049 struct bpf_prog *prog, u32 *target_size) 7050 { 7051 struct bpf_insn *insn = insn_buf; 7052 7053 #define BPF_XDP_SOCK_GET(FIELD) \ 7054 do { \ 7055 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \ 7056 sizeof_field(struct bpf_xdp_sock, FIELD)); \ 7057 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\ 7058 si->dst_reg, si->src_reg, \ 7059 offsetof(struct xdp_sock, FIELD)); \ 7060 } while (0) 7061 7062 switch (si->off) { 7063 case offsetof(struct bpf_xdp_sock, queue_id): 7064 BPF_XDP_SOCK_GET(queue_id); 7065 break; 7066 } 7067 7068 return insn - insn_buf; 7069 } 7070 7071 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = { 7072 .func = bpf_skb_ecn_set_ce, 7073 .gpl_only = false, 7074 .ret_type = RET_INTEGER, 7075 .arg1_type = ARG_PTR_TO_CTX, 7076 }; 7077 7078 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 7079 struct tcphdr *, th, u32, th_len) 7080 { 7081 #ifdef CONFIG_SYN_COOKIES 7082 u32 cookie; 7083 int ret; 7084 7085 if (unlikely(!sk || th_len < sizeof(*th))) 7086 return -EINVAL; 7087 7088 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */ 7089 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 7090 return -EINVAL; 7091 7092 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies)) 7093 return -EINVAL; 7094 7095 if (!th->ack || th->rst || th->syn) 7096 return -ENOENT; 7097 7098 if (unlikely(iph_len < sizeof(struct iphdr))) 7099 return -EINVAL; 7100 7101 if (tcp_synq_no_recent_overflow(sk)) 7102 return -ENOENT; 7103 7104 cookie = ntohl(th->ack_seq) - 1; 7105 7106 /* Both struct iphdr and struct ipv6hdr have the version field at the 7107 * same offset so we can cast to the shorter header (struct iphdr). 7108 */ 7109 switch (((struct iphdr *)iph)->version) { 7110 case 4: 7111 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk)) 7112 return -EINVAL; 7113 7114 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie); 7115 break; 7116 7117 #if IS_BUILTIN(CONFIG_IPV6) 7118 case 6: 7119 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 7120 return -EINVAL; 7121 7122 if (sk->sk_family != AF_INET6) 7123 return -EINVAL; 7124 7125 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie); 7126 break; 7127 #endif /* CONFIG_IPV6 */ 7128 7129 default: 7130 return -EPROTONOSUPPORT; 7131 } 7132 7133 if (ret > 0) 7134 return 0; 7135 7136 return -ENOENT; 7137 #else 7138 return -ENOTSUPP; 7139 #endif 7140 } 7141 7142 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = { 7143 .func = bpf_tcp_check_syncookie, 7144 .gpl_only = true, 7145 .pkt_access = true, 7146 .ret_type = RET_INTEGER, 7147 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7148 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7149 .arg3_type = ARG_CONST_SIZE, 7150 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7151 .arg5_type = ARG_CONST_SIZE, 7152 }; 7153 7154 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 7155 struct tcphdr *, th, u32, th_len) 7156 { 7157 #ifdef CONFIG_SYN_COOKIES 7158 u32 cookie; 7159 u16 mss; 7160 7161 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4)) 7162 return -EINVAL; 7163 7164 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 7165 return -EINVAL; 7166 7167 if (!READ_ONCE(sock_net(sk)->ipv4.sysctl_tcp_syncookies)) 7168 return -ENOENT; 7169 7170 if (!th->syn || th->ack || th->fin || th->rst) 7171 return -EINVAL; 7172 7173 if (unlikely(iph_len < sizeof(struct iphdr))) 7174 return -EINVAL; 7175 7176 /* Both struct iphdr and struct ipv6hdr have the version field at the 7177 * same offset so we can cast to the shorter header (struct iphdr). 7178 */ 7179 switch (((struct iphdr *)iph)->version) { 7180 case 4: 7181 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk)) 7182 return -EINVAL; 7183 7184 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie); 7185 break; 7186 7187 #if IS_BUILTIN(CONFIG_IPV6) 7188 case 6: 7189 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 7190 return -EINVAL; 7191 7192 if (sk->sk_family != AF_INET6) 7193 return -EINVAL; 7194 7195 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie); 7196 break; 7197 #endif /* CONFIG_IPV6 */ 7198 7199 default: 7200 return -EPROTONOSUPPORT; 7201 } 7202 if (mss == 0) 7203 return -ENOENT; 7204 7205 return cookie | ((u64)mss << 32); 7206 #else 7207 return -EOPNOTSUPP; 7208 #endif /* CONFIG_SYN_COOKIES */ 7209 } 7210 7211 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = { 7212 .func = bpf_tcp_gen_syncookie, 7213 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */ 7214 .pkt_access = true, 7215 .ret_type = RET_INTEGER, 7216 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7217 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7218 .arg3_type = ARG_CONST_SIZE, 7219 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7220 .arg5_type = ARG_CONST_SIZE, 7221 }; 7222 7223 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags) 7224 { 7225 if (!sk || flags != 0) 7226 return -EINVAL; 7227 if (!skb_at_tc_ingress(skb)) 7228 return -EOPNOTSUPP; 7229 if (unlikely(dev_net(skb->dev) != sock_net(sk))) 7230 return -ENETUNREACH; 7231 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport)) 7232 return -ESOCKTNOSUPPORT; 7233 if (sk_is_refcounted(sk) && 7234 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) 7235 return -ENOENT; 7236 7237 skb_orphan(skb); 7238 skb->sk = sk; 7239 skb->destructor = sock_pfree; 7240 7241 return 0; 7242 } 7243 7244 static const struct bpf_func_proto bpf_sk_assign_proto = { 7245 .func = bpf_sk_assign, 7246 .gpl_only = false, 7247 .ret_type = RET_INTEGER, 7248 .arg1_type = ARG_PTR_TO_CTX, 7249 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7250 .arg3_type = ARG_ANYTHING, 7251 }; 7252 7253 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend, 7254 u8 search_kind, const u8 *magic, 7255 u8 magic_len, bool *eol) 7256 { 7257 u8 kind, kind_len; 7258 7259 *eol = false; 7260 7261 while (op < opend) { 7262 kind = op[0]; 7263 7264 if (kind == TCPOPT_EOL) { 7265 *eol = true; 7266 return ERR_PTR(-ENOMSG); 7267 } else if (kind == TCPOPT_NOP) { 7268 op++; 7269 continue; 7270 } 7271 7272 if (opend - op < 2 || opend - op < op[1] || op[1] < 2) 7273 /* Something is wrong in the received header. 7274 * Follow the TCP stack's tcp_parse_options() 7275 * and just bail here. 7276 */ 7277 return ERR_PTR(-EFAULT); 7278 7279 kind_len = op[1]; 7280 if (search_kind == kind) { 7281 if (!magic_len) 7282 return op; 7283 7284 if (magic_len > kind_len - 2) 7285 return ERR_PTR(-ENOMSG); 7286 7287 if (!memcmp(&op[2], magic, magic_len)) 7288 return op; 7289 } 7290 7291 op += kind_len; 7292 } 7293 7294 return ERR_PTR(-ENOMSG); 7295 } 7296 7297 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7298 void *, search_res, u32, len, u64, flags) 7299 { 7300 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN; 7301 const u8 *op, *opend, *magic, *search = search_res; 7302 u8 search_kind, search_len, copy_len, magic_len; 7303 int ret; 7304 7305 /* 2 byte is the minimal option len except TCPOPT_NOP and 7306 * TCPOPT_EOL which are useless for the bpf prog to learn 7307 * and this helper disallow loading them also. 7308 */ 7309 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN) 7310 return -EINVAL; 7311 7312 search_kind = search[0]; 7313 search_len = search[1]; 7314 7315 if (search_len > len || search_kind == TCPOPT_NOP || 7316 search_kind == TCPOPT_EOL) 7317 return -EINVAL; 7318 7319 if (search_kind == TCPOPT_EXP || search_kind == 253) { 7320 /* 16 or 32 bit magic. +2 for kind and kind length */ 7321 if (search_len != 4 && search_len != 6) 7322 return -EINVAL; 7323 magic = &search[2]; 7324 magic_len = search_len - 2; 7325 } else { 7326 if (search_len) 7327 return -EINVAL; 7328 magic = NULL; 7329 magic_len = 0; 7330 } 7331 7332 if (load_syn) { 7333 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op); 7334 if (ret < 0) 7335 return ret; 7336 7337 opend = op + ret; 7338 op += sizeof(struct tcphdr); 7339 } else { 7340 if (!bpf_sock->skb || 7341 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB) 7342 /* This bpf_sock->op cannot call this helper */ 7343 return -EPERM; 7344 7345 opend = bpf_sock->skb_data_end; 7346 op = bpf_sock->skb->data + sizeof(struct tcphdr); 7347 } 7348 7349 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len, 7350 &eol); 7351 if (IS_ERR(op)) 7352 return PTR_ERR(op); 7353 7354 copy_len = op[1]; 7355 ret = copy_len; 7356 if (copy_len > len) { 7357 ret = -ENOSPC; 7358 copy_len = len; 7359 } 7360 7361 memcpy(search_res, op, copy_len); 7362 return ret; 7363 } 7364 7365 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = { 7366 .func = bpf_sock_ops_load_hdr_opt, 7367 .gpl_only = false, 7368 .ret_type = RET_INTEGER, 7369 .arg1_type = ARG_PTR_TO_CTX, 7370 .arg2_type = ARG_PTR_TO_MEM, 7371 .arg3_type = ARG_CONST_SIZE, 7372 .arg4_type = ARG_ANYTHING, 7373 }; 7374 7375 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7376 const void *, from, u32, len, u64, flags) 7377 { 7378 u8 new_kind, new_kind_len, magic_len = 0, *opend; 7379 const u8 *op, *new_op, *magic = NULL; 7380 struct sk_buff *skb; 7381 bool eol; 7382 7383 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB) 7384 return -EPERM; 7385 7386 if (len < 2 || flags) 7387 return -EINVAL; 7388 7389 new_op = from; 7390 new_kind = new_op[0]; 7391 new_kind_len = new_op[1]; 7392 7393 if (new_kind_len > len || new_kind == TCPOPT_NOP || 7394 new_kind == TCPOPT_EOL) 7395 return -EINVAL; 7396 7397 if (new_kind_len > bpf_sock->remaining_opt_len) 7398 return -ENOSPC; 7399 7400 /* 253 is another experimental kind */ 7401 if (new_kind == TCPOPT_EXP || new_kind == 253) { 7402 if (new_kind_len < 4) 7403 return -EINVAL; 7404 /* Match for the 2 byte magic also. 7405 * RFC 6994: the magic could be 2 or 4 bytes. 7406 * Hence, matching by 2 byte only is on the 7407 * conservative side but it is the right 7408 * thing to do for the 'search-for-duplication' 7409 * purpose. 7410 */ 7411 magic = &new_op[2]; 7412 magic_len = 2; 7413 } 7414 7415 /* Check for duplication */ 7416 skb = bpf_sock->skb; 7417 op = skb->data + sizeof(struct tcphdr); 7418 opend = bpf_sock->skb_data_end; 7419 7420 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len, 7421 &eol); 7422 if (!IS_ERR(op)) 7423 return -EEXIST; 7424 7425 if (PTR_ERR(op) != -ENOMSG) 7426 return PTR_ERR(op); 7427 7428 if (eol) 7429 /* The option has been ended. Treat it as no more 7430 * header option can be written. 7431 */ 7432 return -ENOSPC; 7433 7434 /* No duplication found. Store the header option. */ 7435 memcpy(opend, from, new_kind_len); 7436 7437 bpf_sock->remaining_opt_len -= new_kind_len; 7438 bpf_sock->skb_data_end += new_kind_len; 7439 7440 return 0; 7441 } 7442 7443 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = { 7444 .func = bpf_sock_ops_store_hdr_opt, 7445 .gpl_only = false, 7446 .ret_type = RET_INTEGER, 7447 .arg1_type = ARG_PTR_TO_CTX, 7448 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7449 .arg3_type = ARG_CONST_SIZE, 7450 .arg4_type = ARG_ANYTHING, 7451 }; 7452 7453 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7454 u32, len, u64, flags) 7455 { 7456 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB) 7457 return -EPERM; 7458 7459 if (flags || len < 2) 7460 return -EINVAL; 7461 7462 if (len > bpf_sock->remaining_opt_len) 7463 return -ENOSPC; 7464 7465 bpf_sock->remaining_opt_len -= len; 7466 7467 return 0; 7468 } 7469 7470 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = { 7471 .func = bpf_sock_ops_reserve_hdr_opt, 7472 .gpl_only = false, 7473 .ret_type = RET_INTEGER, 7474 .arg1_type = ARG_PTR_TO_CTX, 7475 .arg2_type = ARG_ANYTHING, 7476 .arg3_type = ARG_ANYTHING, 7477 }; 7478 7479 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, 7480 u64, tstamp, u32, tstamp_type) 7481 { 7482 /* skb_clear_delivery_time() is done for inet protocol */ 7483 if (skb->protocol != htons(ETH_P_IP) && 7484 skb->protocol != htons(ETH_P_IPV6)) 7485 return -EOPNOTSUPP; 7486 7487 switch (tstamp_type) { 7488 case BPF_SKB_TSTAMP_DELIVERY_MONO: 7489 if (!tstamp) 7490 return -EINVAL; 7491 skb->tstamp = tstamp; 7492 skb->mono_delivery_time = 1; 7493 break; 7494 case BPF_SKB_TSTAMP_UNSPEC: 7495 if (tstamp) 7496 return -EINVAL; 7497 skb->tstamp = 0; 7498 skb->mono_delivery_time = 0; 7499 break; 7500 default: 7501 return -EINVAL; 7502 } 7503 7504 return 0; 7505 } 7506 7507 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = { 7508 .func = bpf_skb_set_tstamp, 7509 .gpl_only = false, 7510 .ret_type = RET_INTEGER, 7511 .arg1_type = ARG_PTR_TO_CTX, 7512 .arg2_type = ARG_ANYTHING, 7513 .arg3_type = ARG_ANYTHING, 7514 }; 7515 7516 #ifdef CONFIG_SYN_COOKIES 7517 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv4, struct iphdr *, iph, 7518 struct tcphdr *, th, u32, th_len) 7519 { 7520 u32 cookie; 7521 u16 mss; 7522 7523 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4)) 7524 return -EINVAL; 7525 7526 mss = tcp_parse_mss_option(th, 0) ?: TCP_MSS_DEFAULT; 7527 cookie = __cookie_v4_init_sequence(iph, th, &mss); 7528 7529 return cookie | ((u64)mss << 32); 7530 } 7531 7532 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv4_proto = { 7533 .func = bpf_tcp_raw_gen_syncookie_ipv4, 7534 .gpl_only = true, /* __cookie_v4_init_sequence() is GPL */ 7535 .pkt_access = true, 7536 .ret_type = RET_INTEGER, 7537 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7538 .arg1_size = sizeof(struct iphdr), 7539 .arg2_type = ARG_PTR_TO_MEM, 7540 .arg3_type = ARG_CONST_SIZE, 7541 }; 7542 7543 BPF_CALL_3(bpf_tcp_raw_gen_syncookie_ipv6, struct ipv6hdr *, iph, 7544 struct tcphdr *, th, u32, th_len) 7545 { 7546 #if IS_BUILTIN(CONFIG_IPV6) 7547 const u16 mss_clamp = IPV6_MIN_MTU - sizeof(struct tcphdr) - 7548 sizeof(struct ipv6hdr); 7549 u32 cookie; 7550 u16 mss; 7551 7552 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4)) 7553 return -EINVAL; 7554 7555 mss = tcp_parse_mss_option(th, 0) ?: mss_clamp; 7556 cookie = __cookie_v6_init_sequence(iph, th, &mss); 7557 7558 return cookie | ((u64)mss << 32); 7559 #else 7560 return -EPROTONOSUPPORT; 7561 #endif 7562 } 7563 7564 static const struct bpf_func_proto bpf_tcp_raw_gen_syncookie_ipv6_proto = { 7565 .func = bpf_tcp_raw_gen_syncookie_ipv6, 7566 .gpl_only = true, /* __cookie_v6_init_sequence() is GPL */ 7567 .pkt_access = true, 7568 .ret_type = RET_INTEGER, 7569 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7570 .arg1_size = sizeof(struct ipv6hdr), 7571 .arg2_type = ARG_PTR_TO_MEM, 7572 .arg3_type = ARG_CONST_SIZE, 7573 }; 7574 7575 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv4, struct iphdr *, iph, 7576 struct tcphdr *, th) 7577 { 7578 u32 cookie = ntohl(th->ack_seq) - 1; 7579 7580 if (__cookie_v4_check(iph, th, cookie) > 0) 7581 return 0; 7582 7583 return -EACCES; 7584 } 7585 7586 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv4_proto = { 7587 .func = bpf_tcp_raw_check_syncookie_ipv4, 7588 .gpl_only = true, /* __cookie_v4_check is GPL */ 7589 .pkt_access = true, 7590 .ret_type = RET_INTEGER, 7591 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7592 .arg1_size = sizeof(struct iphdr), 7593 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7594 .arg2_size = sizeof(struct tcphdr), 7595 }; 7596 7597 BPF_CALL_2(bpf_tcp_raw_check_syncookie_ipv6, struct ipv6hdr *, iph, 7598 struct tcphdr *, th) 7599 { 7600 #if IS_BUILTIN(CONFIG_IPV6) 7601 u32 cookie = ntohl(th->ack_seq) - 1; 7602 7603 if (__cookie_v6_check(iph, th, cookie) > 0) 7604 return 0; 7605 7606 return -EACCES; 7607 #else 7608 return -EPROTONOSUPPORT; 7609 #endif 7610 } 7611 7612 static const struct bpf_func_proto bpf_tcp_raw_check_syncookie_ipv6_proto = { 7613 .func = bpf_tcp_raw_check_syncookie_ipv6, 7614 .gpl_only = true, /* __cookie_v6_check is GPL */ 7615 .pkt_access = true, 7616 .ret_type = RET_INTEGER, 7617 .arg1_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7618 .arg1_size = sizeof(struct ipv6hdr), 7619 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM, 7620 .arg2_size = sizeof(struct tcphdr), 7621 }; 7622 #endif /* CONFIG_SYN_COOKIES */ 7623 7624 #endif /* CONFIG_INET */ 7625 7626 bool bpf_helper_changes_pkt_data(void *func) 7627 { 7628 if (func == bpf_skb_vlan_push || 7629 func == bpf_skb_vlan_pop || 7630 func == bpf_skb_store_bytes || 7631 func == bpf_skb_change_proto || 7632 func == bpf_skb_change_head || 7633 func == sk_skb_change_head || 7634 func == bpf_skb_change_tail || 7635 func == sk_skb_change_tail || 7636 func == bpf_skb_adjust_room || 7637 func == sk_skb_adjust_room || 7638 func == bpf_skb_pull_data || 7639 func == sk_skb_pull_data || 7640 func == bpf_clone_redirect || 7641 func == bpf_l3_csum_replace || 7642 func == bpf_l4_csum_replace || 7643 func == bpf_xdp_adjust_head || 7644 func == bpf_xdp_adjust_meta || 7645 func == bpf_msg_pull_data || 7646 func == bpf_msg_push_data || 7647 func == bpf_msg_pop_data || 7648 func == bpf_xdp_adjust_tail || 7649 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 7650 func == bpf_lwt_seg6_store_bytes || 7651 func == bpf_lwt_seg6_adjust_srh || 7652 func == bpf_lwt_seg6_action || 7653 #endif 7654 #ifdef CONFIG_INET 7655 func == bpf_sock_ops_store_hdr_opt || 7656 #endif 7657 func == bpf_lwt_in_push_encap || 7658 func == bpf_lwt_xmit_push_encap) 7659 return true; 7660 7661 return false; 7662 } 7663 7664 const struct bpf_func_proto bpf_event_output_data_proto __weak; 7665 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak; 7666 7667 static const struct bpf_func_proto * 7668 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7669 { 7670 switch (func_id) { 7671 /* inet and inet6 sockets are created in a process 7672 * context so there is always a valid uid/gid 7673 */ 7674 case BPF_FUNC_get_current_uid_gid: 7675 return &bpf_get_current_uid_gid_proto; 7676 case BPF_FUNC_get_local_storage: 7677 return &bpf_get_local_storage_proto; 7678 case BPF_FUNC_get_socket_cookie: 7679 return &bpf_get_socket_cookie_sock_proto; 7680 case BPF_FUNC_get_netns_cookie: 7681 return &bpf_get_netns_cookie_sock_proto; 7682 case BPF_FUNC_perf_event_output: 7683 return &bpf_event_output_data_proto; 7684 case BPF_FUNC_get_current_pid_tgid: 7685 return &bpf_get_current_pid_tgid_proto; 7686 case BPF_FUNC_get_current_comm: 7687 return &bpf_get_current_comm_proto; 7688 #ifdef CONFIG_CGROUPS 7689 case BPF_FUNC_get_current_cgroup_id: 7690 return &bpf_get_current_cgroup_id_proto; 7691 case BPF_FUNC_get_current_ancestor_cgroup_id: 7692 return &bpf_get_current_ancestor_cgroup_id_proto; 7693 #endif 7694 #ifdef CONFIG_CGROUP_NET_CLASSID 7695 case BPF_FUNC_get_cgroup_classid: 7696 return &bpf_get_cgroup_classid_curr_proto; 7697 #endif 7698 case BPF_FUNC_sk_storage_get: 7699 return &bpf_sk_storage_get_cg_sock_proto; 7700 case BPF_FUNC_ktime_get_coarse_ns: 7701 return &bpf_ktime_get_coarse_ns_proto; 7702 default: 7703 return bpf_base_func_proto(func_id); 7704 } 7705 } 7706 7707 static const struct bpf_func_proto * 7708 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7709 { 7710 switch (func_id) { 7711 /* inet and inet6 sockets are created in a process 7712 * context so there is always a valid uid/gid 7713 */ 7714 case BPF_FUNC_get_current_uid_gid: 7715 return &bpf_get_current_uid_gid_proto; 7716 case BPF_FUNC_bind: 7717 switch (prog->expected_attach_type) { 7718 case BPF_CGROUP_INET4_CONNECT: 7719 case BPF_CGROUP_INET6_CONNECT: 7720 return &bpf_bind_proto; 7721 default: 7722 return NULL; 7723 } 7724 case BPF_FUNC_get_socket_cookie: 7725 return &bpf_get_socket_cookie_sock_addr_proto; 7726 case BPF_FUNC_get_netns_cookie: 7727 return &bpf_get_netns_cookie_sock_addr_proto; 7728 case BPF_FUNC_get_local_storage: 7729 return &bpf_get_local_storage_proto; 7730 case BPF_FUNC_perf_event_output: 7731 return &bpf_event_output_data_proto; 7732 case BPF_FUNC_get_current_pid_tgid: 7733 return &bpf_get_current_pid_tgid_proto; 7734 case BPF_FUNC_get_current_comm: 7735 return &bpf_get_current_comm_proto; 7736 #ifdef CONFIG_CGROUPS 7737 case BPF_FUNC_get_current_cgroup_id: 7738 return &bpf_get_current_cgroup_id_proto; 7739 case BPF_FUNC_get_current_ancestor_cgroup_id: 7740 return &bpf_get_current_ancestor_cgroup_id_proto; 7741 #endif 7742 #ifdef CONFIG_CGROUP_NET_CLASSID 7743 case BPF_FUNC_get_cgroup_classid: 7744 return &bpf_get_cgroup_classid_curr_proto; 7745 #endif 7746 #ifdef CONFIG_INET 7747 case BPF_FUNC_sk_lookup_tcp: 7748 return &bpf_sock_addr_sk_lookup_tcp_proto; 7749 case BPF_FUNC_sk_lookup_udp: 7750 return &bpf_sock_addr_sk_lookup_udp_proto; 7751 case BPF_FUNC_sk_release: 7752 return &bpf_sk_release_proto; 7753 case BPF_FUNC_skc_lookup_tcp: 7754 return &bpf_sock_addr_skc_lookup_tcp_proto; 7755 #endif /* CONFIG_INET */ 7756 case BPF_FUNC_sk_storage_get: 7757 return &bpf_sk_storage_get_proto; 7758 case BPF_FUNC_sk_storage_delete: 7759 return &bpf_sk_storage_delete_proto; 7760 case BPF_FUNC_setsockopt: 7761 switch (prog->expected_attach_type) { 7762 case BPF_CGROUP_INET4_BIND: 7763 case BPF_CGROUP_INET6_BIND: 7764 case BPF_CGROUP_INET4_CONNECT: 7765 case BPF_CGROUP_INET6_CONNECT: 7766 case BPF_CGROUP_UDP4_RECVMSG: 7767 case BPF_CGROUP_UDP6_RECVMSG: 7768 case BPF_CGROUP_UDP4_SENDMSG: 7769 case BPF_CGROUP_UDP6_SENDMSG: 7770 case BPF_CGROUP_INET4_GETPEERNAME: 7771 case BPF_CGROUP_INET6_GETPEERNAME: 7772 case BPF_CGROUP_INET4_GETSOCKNAME: 7773 case BPF_CGROUP_INET6_GETSOCKNAME: 7774 return &bpf_sock_addr_setsockopt_proto; 7775 default: 7776 return NULL; 7777 } 7778 case BPF_FUNC_getsockopt: 7779 switch (prog->expected_attach_type) { 7780 case BPF_CGROUP_INET4_BIND: 7781 case BPF_CGROUP_INET6_BIND: 7782 case BPF_CGROUP_INET4_CONNECT: 7783 case BPF_CGROUP_INET6_CONNECT: 7784 case BPF_CGROUP_UDP4_RECVMSG: 7785 case BPF_CGROUP_UDP6_RECVMSG: 7786 case BPF_CGROUP_UDP4_SENDMSG: 7787 case BPF_CGROUP_UDP6_SENDMSG: 7788 case BPF_CGROUP_INET4_GETPEERNAME: 7789 case BPF_CGROUP_INET6_GETPEERNAME: 7790 case BPF_CGROUP_INET4_GETSOCKNAME: 7791 case BPF_CGROUP_INET6_GETSOCKNAME: 7792 return &bpf_sock_addr_getsockopt_proto; 7793 default: 7794 return NULL; 7795 } 7796 default: 7797 return bpf_sk_base_func_proto(func_id); 7798 } 7799 } 7800 7801 static const struct bpf_func_proto * 7802 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7803 { 7804 switch (func_id) { 7805 case BPF_FUNC_skb_load_bytes: 7806 return &bpf_skb_load_bytes_proto; 7807 case BPF_FUNC_skb_load_bytes_relative: 7808 return &bpf_skb_load_bytes_relative_proto; 7809 case BPF_FUNC_get_socket_cookie: 7810 return &bpf_get_socket_cookie_proto; 7811 case BPF_FUNC_get_socket_uid: 7812 return &bpf_get_socket_uid_proto; 7813 case BPF_FUNC_perf_event_output: 7814 return &bpf_skb_event_output_proto; 7815 default: 7816 return bpf_sk_base_func_proto(func_id); 7817 } 7818 } 7819 7820 const struct bpf_func_proto bpf_sk_storage_get_proto __weak; 7821 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak; 7822 7823 static const struct bpf_func_proto * 7824 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7825 { 7826 switch (func_id) { 7827 case BPF_FUNC_get_local_storage: 7828 return &bpf_get_local_storage_proto; 7829 case BPF_FUNC_sk_fullsock: 7830 return &bpf_sk_fullsock_proto; 7831 case BPF_FUNC_sk_storage_get: 7832 return &bpf_sk_storage_get_proto; 7833 case BPF_FUNC_sk_storage_delete: 7834 return &bpf_sk_storage_delete_proto; 7835 case BPF_FUNC_perf_event_output: 7836 return &bpf_skb_event_output_proto; 7837 #ifdef CONFIG_SOCK_CGROUP_DATA 7838 case BPF_FUNC_skb_cgroup_id: 7839 return &bpf_skb_cgroup_id_proto; 7840 case BPF_FUNC_skb_ancestor_cgroup_id: 7841 return &bpf_skb_ancestor_cgroup_id_proto; 7842 case BPF_FUNC_sk_cgroup_id: 7843 return &bpf_sk_cgroup_id_proto; 7844 case BPF_FUNC_sk_ancestor_cgroup_id: 7845 return &bpf_sk_ancestor_cgroup_id_proto; 7846 #endif 7847 #ifdef CONFIG_INET 7848 case BPF_FUNC_sk_lookup_tcp: 7849 return &bpf_sk_lookup_tcp_proto; 7850 case BPF_FUNC_sk_lookup_udp: 7851 return &bpf_sk_lookup_udp_proto; 7852 case BPF_FUNC_sk_release: 7853 return &bpf_sk_release_proto; 7854 case BPF_FUNC_skc_lookup_tcp: 7855 return &bpf_skc_lookup_tcp_proto; 7856 case BPF_FUNC_tcp_sock: 7857 return &bpf_tcp_sock_proto; 7858 case BPF_FUNC_get_listener_sock: 7859 return &bpf_get_listener_sock_proto; 7860 case BPF_FUNC_skb_ecn_set_ce: 7861 return &bpf_skb_ecn_set_ce_proto; 7862 #endif 7863 default: 7864 return sk_filter_func_proto(func_id, prog); 7865 } 7866 } 7867 7868 static const struct bpf_func_proto * 7869 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7870 { 7871 switch (func_id) { 7872 case BPF_FUNC_skb_store_bytes: 7873 return &bpf_skb_store_bytes_proto; 7874 case BPF_FUNC_skb_load_bytes: 7875 return &bpf_skb_load_bytes_proto; 7876 case BPF_FUNC_skb_load_bytes_relative: 7877 return &bpf_skb_load_bytes_relative_proto; 7878 case BPF_FUNC_skb_pull_data: 7879 return &bpf_skb_pull_data_proto; 7880 case BPF_FUNC_csum_diff: 7881 return &bpf_csum_diff_proto; 7882 case BPF_FUNC_csum_update: 7883 return &bpf_csum_update_proto; 7884 case BPF_FUNC_csum_level: 7885 return &bpf_csum_level_proto; 7886 case BPF_FUNC_l3_csum_replace: 7887 return &bpf_l3_csum_replace_proto; 7888 case BPF_FUNC_l4_csum_replace: 7889 return &bpf_l4_csum_replace_proto; 7890 case BPF_FUNC_clone_redirect: 7891 return &bpf_clone_redirect_proto; 7892 case BPF_FUNC_get_cgroup_classid: 7893 return &bpf_get_cgroup_classid_proto; 7894 case BPF_FUNC_skb_vlan_push: 7895 return &bpf_skb_vlan_push_proto; 7896 case BPF_FUNC_skb_vlan_pop: 7897 return &bpf_skb_vlan_pop_proto; 7898 case BPF_FUNC_skb_change_proto: 7899 return &bpf_skb_change_proto_proto; 7900 case BPF_FUNC_skb_change_type: 7901 return &bpf_skb_change_type_proto; 7902 case BPF_FUNC_skb_adjust_room: 7903 return &bpf_skb_adjust_room_proto; 7904 case BPF_FUNC_skb_change_tail: 7905 return &bpf_skb_change_tail_proto; 7906 case BPF_FUNC_skb_change_head: 7907 return &bpf_skb_change_head_proto; 7908 case BPF_FUNC_skb_get_tunnel_key: 7909 return &bpf_skb_get_tunnel_key_proto; 7910 case BPF_FUNC_skb_set_tunnel_key: 7911 return bpf_get_skb_set_tunnel_proto(func_id); 7912 case BPF_FUNC_skb_get_tunnel_opt: 7913 return &bpf_skb_get_tunnel_opt_proto; 7914 case BPF_FUNC_skb_set_tunnel_opt: 7915 return bpf_get_skb_set_tunnel_proto(func_id); 7916 case BPF_FUNC_redirect: 7917 return &bpf_redirect_proto; 7918 case BPF_FUNC_redirect_neigh: 7919 return &bpf_redirect_neigh_proto; 7920 case BPF_FUNC_redirect_peer: 7921 return &bpf_redirect_peer_proto; 7922 case BPF_FUNC_get_route_realm: 7923 return &bpf_get_route_realm_proto; 7924 case BPF_FUNC_get_hash_recalc: 7925 return &bpf_get_hash_recalc_proto; 7926 case BPF_FUNC_set_hash_invalid: 7927 return &bpf_set_hash_invalid_proto; 7928 case BPF_FUNC_set_hash: 7929 return &bpf_set_hash_proto; 7930 case BPF_FUNC_perf_event_output: 7931 return &bpf_skb_event_output_proto; 7932 case BPF_FUNC_get_smp_processor_id: 7933 return &bpf_get_smp_processor_id_proto; 7934 case BPF_FUNC_skb_under_cgroup: 7935 return &bpf_skb_under_cgroup_proto; 7936 case BPF_FUNC_get_socket_cookie: 7937 return &bpf_get_socket_cookie_proto; 7938 case BPF_FUNC_get_socket_uid: 7939 return &bpf_get_socket_uid_proto; 7940 case BPF_FUNC_fib_lookup: 7941 return &bpf_skb_fib_lookup_proto; 7942 case BPF_FUNC_check_mtu: 7943 return &bpf_skb_check_mtu_proto; 7944 case BPF_FUNC_sk_fullsock: 7945 return &bpf_sk_fullsock_proto; 7946 case BPF_FUNC_sk_storage_get: 7947 return &bpf_sk_storage_get_proto; 7948 case BPF_FUNC_sk_storage_delete: 7949 return &bpf_sk_storage_delete_proto; 7950 #ifdef CONFIG_XFRM 7951 case BPF_FUNC_skb_get_xfrm_state: 7952 return &bpf_skb_get_xfrm_state_proto; 7953 #endif 7954 #ifdef CONFIG_CGROUP_NET_CLASSID 7955 case BPF_FUNC_skb_cgroup_classid: 7956 return &bpf_skb_cgroup_classid_proto; 7957 #endif 7958 #ifdef CONFIG_SOCK_CGROUP_DATA 7959 case BPF_FUNC_skb_cgroup_id: 7960 return &bpf_skb_cgroup_id_proto; 7961 case BPF_FUNC_skb_ancestor_cgroup_id: 7962 return &bpf_skb_ancestor_cgroup_id_proto; 7963 #endif 7964 #ifdef CONFIG_INET 7965 case BPF_FUNC_sk_lookup_tcp: 7966 return &bpf_sk_lookup_tcp_proto; 7967 case BPF_FUNC_sk_lookup_udp: 7968 return &bpf_sk_lookup_udp_proto; 7969 case BPF_FUNC_sk_release: 7970 return &bpf_sk_release_proto; 7971 case BPF_FUNC_tcp_sock: 7972 return &bpf_tcp_sock_proto; 7973 case BPF_FUNC_get_listener_sock: 7974 return &bpf_get_listener_sock_proto; 7975 case BPF_FUNC_skc_lookup_tcp: 7976 return &bpf_skc_lookup_tcp_proto; 7977 case BPF_FUNC_tcp_check_syncookie: 7978 return &bpf_tcp_check_syncookie_proto; 7979 case BPF_FUNC_skb_ecn_set_ce: 7980 return &bpf_skb_ecn_set_ce_proto; 7981 case BPF_FUNC_tcp_gen_syncookie: 7982 return &bpf_tcp_gen_syncookie_proto; 7983 case BPF_FUNC_sk_assign: 7984 return &bpf_sk_assign_proto; 7985 case BPF_FUNC_skb_set_tstamp: 7986 return &bpf_skb_set_tstamp_proto; 7987 #ifdef CONFIG_SYN_COOKIES 7988 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4: 7989 return &bpf_tcp_raw_gen_syncookie_ipv4_proto; 7990 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6: 7991 return &bpf_tcp_raw_gen_syncookie_ipv6_proto; 7992 case BPF_FUNC_tcp_raw_check_syncookie_ipv4: 7993 return &bpf_tcp_raw_check_syncookie_ipv4_proto; 7994 case BPF_FUNC_tcp_raw_check_syncookie_ipv6: 7995 return &bpf_tcp_raw_check_syncookie_ipv6_proto; 7996 #endif 7997 #endif 7998 default: 7999 return bpf_sk_base_func_proto(func_id); 8000 } 8001 } 8002 8003 static const struct bpf_func_proto * 8004 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8005 { 8006 switch (func_id) { 8007 case BPF_FUNC_perf_event_output: 8008 return &bpf_xdp_event_output_proto; 8009 case BPF_FUNC_get_smp_processor_id: 8010 return &bpf_get_smp_processor_id_proto; 8011 case BPF_FUNC_csum_diff: 8012 return &bpf_csum_diff_proto; 8013 case BPF_FUNC_xdp_adjust_head: 8014 return &bpf_xdp_adjust_head_proto; 8015 case BPF_FUNC_xdp_adjust_meta: 8016 return &bpf_xdp_adjust_meta_proto; 8017 case BPF_FUNC_redirect: 8018 return &bpf_xdp_redirect_proto; 8019 case BPF_FUNC_redirect_map: 8020 return &bpf_xdp_redirect_map_proto; 8021 case BPF_FUNC_xdp_adjust_tail: 8022 return &bpf_xdp_adjust_tail_proto; 8023 case BPF_FUNC_xdp_get_buff_len: 8024 return &bpf_xdp_get_buff_len_proto; 8025 case BPF_FUNC_xdp_load_bytes: 8026 return &bpf_xdp_load_bytes_proto; 8027 case BPF_FUNC_xdp_store_bytes: 8028 return &bpf_xdp_store_bytes_proto; 8029 case BPF_FUNC_fib_lookup: 8030 return &bpf_xdp_fib_lookup_proto; 8031 case BPF_FUNC_check_mtu: 8032 return &bpf_xdp_check_mtu_proto; 8033 #ifdef CONFIG_INET 8034 case BPF_FUNC_sk_lookup_udp: 8035 return &bpf_xdp_sk_lookup_udp_proto; 8036 case BPF_FUNC_sk_lookup_tcp: 8037 return &bpf_xdp_sk_lookup_tcp_proto; 8038 case BPF_FUNC_sk_release: 8039 return &bpf_sk_release_proto; 8040 case BPF_FUNC_skc_lookup_tcp: 8041 return &bpf_xdp_skc_lookup_tcp_proto; 8042 case BPF_FUNC_tcp_check_syncookie: 8043 return &bpf_tcp_check_syncookie_proto; 8044 case BPF_FUNC_tcp_gen_syncookie: 8045 return &bpf_tcp_gen_syncookie_proto; 8046 #ifdef CONFIG_SYN_COOKIES 8047 case BPF_FUNC_tcp_raw_gen_syncookie_ipv4: 8048 return &bpf_tcp_raw_gen_syncookie_ipv4_proto; 8049 case BPF_FUNC_tcp_raw_gen_syncookie_ipv6: 8050 return &bpf_tcp_raw_gen_syncookie_ipv6_proto; 8051 case BPF_FUNC_tcp_raw_check_syncookie_ipv4: 8052 return &bpf_tcp_raw_check_syncookie_ipv4_proto; 8053 case BPF_FUNC_tcp_raw_check_syncookie_ipv6: 8054 return &bpf_tcp_raw_check_syncookie_ipv6_proto; 8055 #endif 8056 #endif 8057 default: 8058 return bpf_sk_base_func_proto(func_id); 8059 } 8060 } 8061 8062 const struct bpf_func_proto bpf_sock_map_update_proto __weak; 8063 const struct bpf_func_proto bpf_sock_hash_update_proto __weak; 8064 8065 static const struct bpf_func_proto * 8066 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8067 { 8068 switch (func_id) { 8069 case BPF_FUNC_setsockopt: 8070 return &bpf_sock_ops_setsockopt_proto; 8071 case BPF_FUNC_getsockopt: 8072 return &bpf_sock_ops_getsockopt_proto; 8073 case BPF_FUNC_sock_ops_cb_flags_set: 8074 return &bpf_sock_ops_cb_flags_set_proto; 8075 case BPF_FUNC_sock_map_update: 8076 return &bpf_sock_map_update_proto; 8077 case BPF_FUNC_sock_hash_update: 8078 return &bpf_sock_hash_update_proto; 8079 case BPF_FUNC_get_socket_cookie: 8080 return &bpf_get_socket_cookie_sock_ops_proto; 8081 case BPF_FUNC_get_local_storage: 8082 return &bpf_get_local_storage_proto; 8083 case BPF_FUNC_perf_event_output: 8084 return &bpf_event_output_data_proto; 8085 case BPF_FUNC_sk_storage_get: 8086 return &bpf_sk_storage_get_proto; 8087 case BPF_FUNC_sk_storage_delete: 8088 return &bpf_sk_storage_delete_proto; 8089 case BPF_FUNC_get_netns_cookie: 8090 return &bpf_get_netns_cookie_sock_ops_proto; 8091 #ifdef CONFIG_INET 8092 case BPF_FUNC_load_hdr_opt: 8093 return &bpf_sock_ops_load_hdr_opt_proto; 8094 case BPF_FUNC_store_hdr_opt: 8095 return &bpf_sock_ops_store_hdr_opt_proto; 8096 case BPF_FUNC_reserve_hdr_opt: 8097 return &bpf_sock_ops_reserve_hdr_opt_proto; 8098 case BPF_FUNC_tcp_sock: 8099 return &bpf_tcp_sock_proto; 8100 #endif /* CONFIG_INET */ 8101 default: 8102 return bpf_sk_base_func_proto(func_id); 8103 } 8104 } 8105 8106 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak; 8107 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak; 8108 8109 static const struct bpf_func_proto * 8110 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8111 { 8112 switch (func_id) { 8113 case BPF_FUNC_msg_redirect_map: 8114 return &bpf_msg_redirect_map_proto; 8115 case BPF_FUNC_msg_redirect_hash: 8116 return &bpf_msg_redirect_hash_proto; 8117 case BPF_FUNC_msg_apply_bytes: 8118 return &bpf_msg_apply_bytes_proto; 8119 case BPF_FUNC_msg_cork_bytes: 8120 return &bpf_msg_cork_bytes_proto; 8121 case BPF_FUNC_msg_pull_data: 8122 return &bpf_msg_pull_data_proto; 8123 case BPF_FUNC_msg_push_data: 8124 return &bpf_msg_push_data_proto; 8125 case BPF_FUNC_msg_pop_data: 8126 return &bpf_msg_pop_data_proto; 8127 case BPF_FUNC_perf_event_output: 8128 return &bpf_event_output_data_proto; 8129 case BPF_FUNC_get_current_uid_gid: 8130 return &bpf_get_current_uid_gid_proto; 8131 case BPF_FUNC_get_current_pid_tgid: 8132 return &bpf_get_current_pid_tgid_proto; 8133 case BPF_FUNC_sk_storage_get: 8134 return &bpf_sk_storage_get_proto; 8135 case BPF_FUNC_sk_storage_delete: 8136 return &bpf_sk_storage_delete_proto; 8137 case BPF_FUNC_get_netns_cookie: 8138 return &bpf_get_netns_cookie_sk_msg_proto; 8139 #ifdef CONFIG_CGROUPS 8140 case BPF_FUNC_get_current_cgroup_id: 8141 return &bpf_get_current_cgroup_id_proto; 8142 case BPF_FUNC_get_current_ancestor_cgroup_id: 8143 return &bpf_get_current_ancestor_cgroup_id_proto; 8144 #endif 8145 #ifdef CONFIG_CGROUP_NET_CLASSID 8146 case BPF_FUNC_get_cgroup_classid: 8147 return &bpf_get_cgroup_classid_curr_proto; 8148 #endif 8149 default: 8150 return bpf_sk_base_func_proto(func_id); 8151 } 8152 } 8153 8154 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak; 8155 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak; 8156 8157 static const struct bpf_func_proto * 8158 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8159 { 8160 switch (func_id) { 8161 case BPF_FUNC_skb_store_bytes: 8162 return &bpf_skb_store_bytes_proto; 8163 case BPF_FUNC_skb_load_bytes: 8164 return &bpf_skb_load_bytes_proto; 8165 case BPF_FUNC_skb_pull_data: 8166 return &sk_skb_pull_data_proto; 8167 case BPF_FUNC_skb_change_tail: 8168 return &sk_skb_change_tail_proto; 8169 case BPF_FUNC_skb_change_head: 8170 return &sk_skb_change_head_proto; 8171 case BPF_FUNC_skb_adjust_room: 8172 return &sk_skb_adjust_room_proto; 8173 case BPF_FUNC_get_socket_cookie: 8174 return &bpf_get_socket_cookie_proto; 8175 case BPF_FUNC_get_socket_uid: 8176 return &bpf_get_socket_uid_proto; 8177 case BPF_FUNC_sk_redirect_map: 8178 return &bpf_sk_redirect_map_proto; 8179 case BPF_FUNC_sk_redirect_hash: 8180 return &bpf_sk_redirect_hash_proto; 8181 case BPF_FUNC_perf_event_output: 8182 return &bpf_skb_event_output_proto; 8183 #ifdef CONFIG_INET 8184 case BPF_FUNC_sk_lookup_tcp: 8185 return &bpf_sk_lookup_tcp_proto; 8186 case BPF_FUNC_sk_lookup_udp: 8187 return &bpf_sk_lookup_udp_proto; 8188 case BPF_FUNC_sk_release: 8189 return &bpf_sk_release_proto; 8190 case BPF_FUNC_skc_lookup_tcp: 8191 return &bpf_skc_lookup_tcp_proto; 8192 #endif 8193 default: 8194 return bpf_sk_base_func_proto(func_id); 8195 } 8196 } 8197 8198 static const struct bpf_func_proto * 8199 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8200 { 8201 switch (func_id) { 8202 case BPF_FUNC_skb_load_bytes: 8203 return &bpf_flow_dissector_load_bytes_proto; 8204 default: 8205 return bpf_sk_base_func_proto(func_id); 8206 } 8207 } 8208 8209 static const struct bpf_func_proto * 8210 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8211 { 8212 switch (func_id) { 8213 case BPF_FUNC_skb_load_bytes: 8214 return &bpf_skb_load_bytes_proto; 8215 case BPF_FUNC_skb_pull_data: 8216 return &bpf_skb_pull_data_proto; 8217 case BPF_FUNC_csum_diff: 8218 return &bpf_csum_diff_proto; 8219 case BPF_FUNC_get_cgroup_classid: 8220 return &bpf_get_cgroup_classid_proto; 8221 case BPF_FUNC_get_route_realm: 8222 return &bpf_get_route_realm_proto; 8223 case BPF_FUNC_get_hash_recalc: 8224 return &bpf_get_hash_recalc_proto; 8225 case BPF_FUNC_perf_event_output: 8226 return &bpf_skb_event_output_proto; 8227 case BPF_FUNC_get_smp_processor_id: 8228 return &bpf_get_smp_processor_id_proto; 8229 case BPF_FUNC_skb_under_cgroup: 8230 return &bpf_skb_under_cgroup_proto; 8231 default: 8232 return bpf_sk_base_func_proto(func_id); 8233 } 8234 } 8235 8236 static const struct bpf_func_proto * 8237 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8238 { 8239 switch (func_id) { 8240 case BPF_FUNC_lwt_push_encap: 8241 return &bpf_lwt_in_push_encap_proto; 8242 default: 8243 return lwt_out_func_proto(func_id, prog); 8244 } 8245 } 8246 8247 static const struct bpf_func_proto * 8248 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8249 { 8250 switch (func_id) { 8251 case BPF_FUNC_skb_get_tunnel_key: 8252 return &bpf_skb_get_tunnel_key_proto; 8253 case BPF_FUNC_skb_set_tunnel_key: 8254 return bpf_get_skb_set_tunnel_proto(func_id); 8255 case BPF_FUNC_skb_get_tunnel_opt: 8256 return &bpf_skb_get_tunnel_opt_proto; 8257 case BPF_FUNC_skb_set_tunnel_opt: 8258 return bpf_get_skb_set_tunnel_proto(func_id); 8259 case BPF_FUNC_redirect: 8260 return &bpf_redirect_proto; 8261 case BPF_FUNC_clone_redirect: 8262 return &bpf_clone_redirect_proto; 8263 case BPF_FUNC_skb_change_tail: 8264 return &bpf_skb_change_tail_proto; 8265 case BPF_FUNC_skb_change_head: 8266 return &bpf_skb_change_head_proto; 8267 case BPF_FUNC_skb_store_bytes: 8268 return &bpf_skb_store_bytes_proto; 8269 case BPF_FUNC_csum_update: 8270 return &bpf_csum_update_proto; 8271 case BPF_FUNC_csum_level: 8272 return &bpf_csum_level_proto; 8273 case BPF_FUNC_l3_csum_replace: 8274 return &bpf_l3_csum_replace_proto; 8275 case BPF_FUNC_l4_csum_replace: 8276 return &bpf_l4_csum_replace_proto; 8277 case BPF_FUNC_set_hash_invalid: 8278 return &bpf_set_hash_invalid_proto; 8279 case BPF_FUNC_lwt_push_encap: 8280 return &bpf_lwt_xmit_push_encap_proto; 8281 default: 8282 return lwt_out_func_proto(func_id, prog); 8283 } 8284 } 8285 8286 static const struct bpf_func_proto * 8287 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8288 { 8289 switch (func_id) { 8290 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 8291 case BPF_FUNC_lwt_seg6_store_bytes: 8292 return &bpf_lwt_seg6_store_bytes_proto; 8293 case BPF_FUNC_lwt_seg6_action: 8294 return &bpf_lwt_seg6_action_proto; 8295 case BPF_FUNC_lwt_seg6_adjust_srh: 8296 return &bpf_lwt_seg6_adjust_srh_proto; 8297 #endif 8298 default: 8299 return lwt_out_func_proto(func_id, prog); 8300 } 8301 } 8302 8303 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type, 8304 const struct bpf_prog *prog, 8305 struct bpf_insn_access_aux *info) 8306 { 8307 const int size_default = sizeof(__u32); 8308 8309 if (off < 0 || off >= sizeof(struct __sk_buff)) 8310 return false; 8311 8312 /* The verifier guarantees that size > 0. */ 8313 if (off % size != 0) 8314 return false; 8315 8316 switch (off) { 8317 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8318 if (off + size > offsetofend(struct __sk_buff, cb[4])) 8319 return false; 8320 break; 8321 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]): 8322 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]): 8323 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4): 8324 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4): 8325 case bpf_ctx_range(struct __sk_buff, data): 8326 case bpf_ctx_range(struct __sk_buff, data_meta): 8327 case bpf_ctx_range(struct __sk_buff, data_end): 8328 if (size != size_default) 8329 return false; 8330 break; 8331 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 8332 return false; 8333 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8334 if (type == BPF_WRITE || size != sizeof(__u64)) 8335 return false; 8336 break; 8337 case bpf_ctx_range(struct __sk_buff, tstamp): 8338 if (size != sizeof(__u64)) 8339 return false; 8340 break; 8341 case offsetof(struct __sk_buff, sk): 8342 if (type == BPF_WRITE || size != sizeof(__u64)) 8343 return false; 8344 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 8345 break; 8346 case offsetof(struct __sk_buff, tstamp_type): 8347 return false; 8348 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1: 8349 /* Explicitly prohibit access to padding in __sk_buff. */ 8350 return false; 8351 default: 8352 /* Only narrow read access allowed for now. */ 8353 if (type == BPF_WRITE) { 8354 if (size != size_default) 8355 return false; 8356 } else { 8357 bpf_ctx_record_field_size(info, size_default); 8358 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8359 return false; 8360 } 8361 } 8362 8363 return true; 8364 } 8365 8366 static bool sk_filter_is_valid_access(int off, int size, 8367 enum bpf_access_type type, 8368 const struct bpf_prog *prog, 8369 struct bpf_insn_access_aux *info) 8370 { 8371 switch (off) { 8372 case bpf_ctx_range(struct __sk_buff, tc_classid): 8373 case bpf_ctx_range(struct __sk_buff, data): 8374 case bpf_ctx_range(struct __sk_buff, data_meta): 8375 case bpf_ctx_range(struct __sk_buff, data_end): 8376 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8377 case bpf_ctx_range(struct __sk_buff, tstamp): 8378 case bpf_ctx_range(struct __sk_buff, wire_len): 8379 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8380 return false; 8381 } 8382 8383 if (type == BPF_WRITE) { 8384 switch (off) { 8385 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8386 break; 8387 default: 8388 return false; 8389 } 8390 } 8391 8392 return bpf_skb_is_valid_access(off, size, type, prog, info); 8393 } 8394 8395 static bool cg_skb_is_valid_access(int off, int size, 8396 enum bpf_access_type type, 8397 const struct bpf_prog *prog, 8398 struct bpf_insn_access_aux *info) 8399 { 8400 switch (off) { 8401 case bpf_ctx_range(struct __sk_buff, tc_classid): 8402 case bpf_ctx_range(struct __sk_buff, data_meta): 8403 case bpf_ctx_range(struct __sk_buff, wire_len): 8404 return false; 8405 case bpf_ctx_range(struct __sk_buff, data): 8406 case bpf_ctx_range(struct __sk_buff, data_end): 8407 if (!bpf_capable()) 8408 return false; 8409 break; 8410 } 8411 8412 if (type == BPF_WRITE) { 8413 switch (off) { 8414 case bpf_ctx_range(struct __sk_buff, mark): 8415 case bpf_ctx_range(struct __sk_buff, priority): 8416 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8417 break; 8418 case bpf_ctx_range(struct __sk_buff, tstamp): 8419 if (!bpf_capable()) 8420 return false; 8421 break; 8422 default: 8423 return false; 8424 } 8425 } 8426 8427 switch (off) { 8428 case bpf_ctx_range(struct __sk_buff, data): 8429 info->reg_type = PTR_TO_PACKET; 8430 break; 8431 case bpf_ctx_range(struct __sk_buff, data_end): 8432 info->reg_type = PTR_TO_PACKET_END; 8433 break; 8434 } 8435 8436 return bpf_skb_is_valid_access(off, size, type, prog, info); 8437 } 8438 8439 static bool lwt_is_valid_access(int off, int size, 8440 enum bpf_access_type type, 8441 const struct bpf_prog *prog, 8442 struct bpf_insn_access_aux *info) 8443 { 8444 switch (off) { 8445 case bpf_ctx_range(struct __sk_buff, tc_classid): 8446 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8447 case bpf_ctx_range(struct __sk_buff, data_meta): 8448 case bpf_ctx_range(struct __sk_buff, tstamp): 8449 case bpf_ctx_range(struct __sk_buff, wire_len): 8450 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8451 return false; 8452 } 8453 8454 if (type == BPF_WRITE) { 8455 switch (off) { 8456 case bpf_ctx_range(struct __sk_buff, mark): 8457 case bpf_ctx_range(struct __sk_buff, priority): 8458 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8459 break; 8460 default: 8461 return false; 8462 } 8463 } 8464 8465 switch (off) { 8466 case bpf_ctx_range(struct __sk_buff, data): 8467 info->reg_type = PTR_TO_PACKET; 8468 break; 8469 case bpf_ctx_range(struct __sk_buff, data_end): 8470 info->reg_type = PTR_TO_PACKET_END; 8471 break; 8472 } 8473 8474 return bpf_skb_is_valid_access(off, size, type, prog, info); 8475 } 8476 8477 /* Attach type specific accesses */ 8478 static bool __sock_filter_check_attach_type(int off, 8479 enum bpf_access_type access_type, 8480 enum bpf_attach_type attach_type) 8481 { 8482 switch (off) { 8483 case offsetof(struct bpf_sock, bound_dev_if): 8484 case offsetof(struct bpf_sock, mark): 8485 case offsetof(struct bpf_sock, priority): 8486 switch (attach_type) { 8487 case BPF_CGROUP_INET_SOCK_CREATE: 8488 case BPF_CGROUP_INET_SOCK_RELEASE: 8489 goto full_access; 8490 default: 8491 return false; 8492 } 8493 case bpf_ctx_range(struct bpf_sock, src_ip4): 8494 switch (attach_type) { 8495 case BPF_CGROUP_INET4_POST_BIND: 8496 goto read_only; 8497 default: 8498 return false; 8499 } 8500 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 8501 switch (attach_type) { 8502 case BPF_CGROUP_INET6_POST_BIND: 8503 goto read_only; 8504 default: 8505 return false; 8506 } 8507 case bpf_ctx_range(struct bpf_sock, src_port): 8508 switch (attach_type) { 8509 case BPF_CGROUP_INET4_POST_BIND: 8510 case BPF_CGROUP_INET6_POST_BIND: 8511 goto read_only; 8512 default: 8513 return false; 8514 } 8515 } 8516 read_only: 8517 return access_type == BPF_READ; 8518 full_access: 8519 return true; 8520 } 8521 8522 bool bpf_sock_common_is_valid_access(int off, int size, 8523 enum bpf_access_type type, 8524 struct bpf_insn_access_aux *info) 8525 { 8526 switch (off) { 8527 case bpf_ctx_range_till(struct bpf_sock, type, priority): 8528 return false; 8529 default: 8530 return bpf_sock_is_valid_access(off, size, type, info); 8531 } 8532 } 8533 8534 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 8535 struct bpf_insn_access_aux *info) 8536 { 8537 const int size_default = sizeof(__u32); 8538 int field_size; 8539 8540 if (off < 0 || off >= sizeof(struct bpf_sock)) 8541 return false; 8542 if (off % size != 0) 8543 return false; 8544 8545 switch (off) { 8546 case offsetof(struct bpf_sock, state): 8547 case offsetof(struct bpf_sock, family): 8548 case offsetof(struct bpf_sock, type): 8549 case offsetof(struct bpf_sock, protocol): 8550 case offsetof(struct bpf_sock, src_port): 8551 case offsetof(struct bpf_sock, rx_queue_mapping): 8552 case bpf_ctx_range(struct bpf_sock, src_ip4): 8553 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 8554 case bpf_ctx_range(struct bpf_sock, dst_ip4): 8555 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 8556 bpf_ctx_record_field_size(info, size_default); 8557 return bpf_ctx_narrow_access_ok(off, size, size_default); 8558 case bpf_ctx_range(struct bpf_sock, dst_port): 8559 field_size = size == size_default ? 8560 size_default : sizeof_field(struct bpf_sock, dst_port); 8561 bpf_ctx_record_field_size(info, field_size); 8562 return bpf_ctx_narrow_access_ok(off, size, field_size); 8563 case offsetofend(struct bpf_sock, dst_port) ... 8564 offsetof(struct bpf_sock, dst_ip4) - 1: 8565 return false; 8566 } 8567 8568 return size == size_default; 8569 } 8570 8571 static bool sock_filter_is_valid_access(int off, int size, 8572 enum bpf_access_type type, 8573 const struct bpf_prog *prog, 8574 struct bpf_insn_access_aux *info) 8575 { 8576 if (!bpf_sock_is_valid_access(off, size, type, info)) 8577 return false; 8578 return __sock_filter_check_attach_type(off, type, 8579 prog->expected_attach_type); 8580 } 8581 8582 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write, 8583 const struct bpf_prog *prog) 8584 { 8585 /* Neither direct read nor direct write requires any preliminary 8586 * action. 8587 */ 8588 return 0; 8589 } 8590 8591 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write, 8592 const struct bpf_prog *prog, int drop_verdict) 8593 { 8594 struct bpf_insn *insn = insn_buf; 8595 8596 if (!direct_write) 8597 return 0; 8598 8599 /* if (!skb->cloned) 8600 * goto start; 8601 * 8602 * (Fast-path, otherwise approximation that we might be 8603 * a clone, do the rest in helper.) 8604 */ 8605 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET); 8606 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK); 8607 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7); 8608 8609 /* ret = bpf_skb_pull_data(skb, 0); */ 8610 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1); 8611 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2); 8612 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, 8613 BPF_FUNC_skb_pull_data); 8614 /* if (!ret) 8615 * goto restore; 8616 * return TC_ACT_SHOT; 8617 */ 8618 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2); 8619 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict); 8620 *insn++ = BPF_EXIT_INSN(); 8621 8622 /* restore: */ 8623 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6); 8624 /* start: */ 8625 *insn++ = prog->insnsi[0]; 8626 8627 return insn - insn_buf; 8628 } 8629 8630 static int bpf_gen_ld_abs(const struct bpf_insn *orig, 8631 struct bpf_insn *insn_buf) 8632 { 8633 bool indirect = BPF_MODE(orig->code) == BPF_IND; 8634 struct bpf_insn *insn = insn_buf; 8635 8636 if (!indirect) { 8637 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm); 8638 } else { 8639 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg); 8640 if (orig->imm) 8641 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm); 8642 } 8643 /* We're guaranteed here that CTX is in R6. */ 8644 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX); 8645 8646 switch (BPF_SIZE(orig->code)) { 8647 case BPF_B: 8648 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache); 8649 break; 8650 case BPF_H: 8651 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache); 8652 break; 8653 case BPF_W: 8654 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache); 8655 break; 8656 } 8657 8658 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2); 8659 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0); 8660 *insn++ = BPF_EXIT_INSN(); 8661 8662 return insn - insn_buf; 8663 } 8664 8665 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write, 8666 const struct bpf_prog *prog) 8667 { 8668 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT); 8669 } 8670 8671 static bool tc_cls_act_is_valid_access(int off, int size, 8672 enum bpf_access_type type, 8673 const struct bpf_prog *prog, 8674 struct bpf_insn_access_aux *info) 8675 { 8676 if (type == BPF_WRITE) { 8677 switch (off) { 8678 case bpf_ctx_range(struct __sk_buff, mark): 8679 case bpf_ctx_range(struct __sk_buff, tc_index): 8680 case bpf_ctx_range(struct __sk_buff, priority): 8681 case bpf_ctx_range(struct __sk_buff, tc_classid): 8682 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8683 case bpf_ctx_range(struct __sk_buff, tstamp): 8684 case bpf_ctx_range(struct __sk_buff, queue_mapping): 8685 break; 8686 default: 8687 return false; 8688 } 8689 } 8690 8691 switch (off) { 8692 case bpf_ctx_range(struct __sk_buff, data): 8693 info->reg_type = PTR_TO_PACKET; 8694 break; 8695 case bpf_ctx_range(struct __sk_buff, data_meta): 8696 info->reg_type = PTR_TO_PACKET_META; 8697 break; 8698 case bpf_ctx_range(struct __sk_buff, data_end): 8699 info->reg_type = PTR_TO_PACKET_END; 8700 break; 8701 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8702 return false; 8703 case offsetof(struct __sk_buff, tstamp_type): 8704 /* The convert_ctx_access() on reading and writing 8705 * __sk_buff->tstamp depends on whether the bpf prog 8706 * has used __sk_buff->tstamp_type or not. 8707 * Thus, we need to set prog->tstamp_type_access 8708 * earlier during is_valid_access() here. 8709 */ 8710 ((struct bpf_prog *)prog)->tstamp_type_access = 1; 8711 return size == sizeof(__u8); 8712 } 8713 8714 return bpf_skb_is_valid_access(off, size, type, prog, info); 8715 } 8716 8717 static bool __is_valid_xdp_access(int off, int size) 8718 { 8719 if (off < 0 || off >= sizeof(struct xdp_md)) 8720 return false; 8721 if (off % size != 0) 8722 return false; 8723 if (size != sizeof(__u32)) 8724 return false; 8725 8726 return true; 8727 } 8728 8729 static bool xdp_is_valid_access(int off, int size, 8730 enum bpf_access_type type, 8731 const struct bpf_prog *prog, 8732 struct bpf_insn_access_aux *info) 8733 { 8734 if (prog->expected_attach_type != BPF_XDP_DEVMAP) { 8735 switch (off) { 8736 case offsetof(struct xdp_md, egress_ifindex): 8737 return false; 8738 } 8739 } 8740 8741 if (type == BPF_WRITE) { 8742 if (bpf_prog_is_dev_bound(prog->aux)) { 8743 switch (off) { 8744 case offsetof(struct xdp_md, rx_queue_index): 8745 return __is_valid_xdp_access(off, size); 8746 } 8747 } 8748 return false; 8749 } 8750 8751 switch (off) { 8752 case offsetof(struct xdp_md, data): 8753 info->reg_type = PTR_TO_PACKET; 8754 break; 8755 case offsetof(struct xdp_md, data_meta): 8756 info->reg_type = PTR_TO_PACKET_META; 8757 break; 8758 case offsetof(struct xdp_md, data_end): 8759 info->reg_type = PTR_TO_PACKET_END; 8760 break; 8761 } 8762 8763 return __is_valid_xdp_access(off, size); 8764 } 8765 8766 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act) 8767 { 8768 const u32 act_max = XDP_REDIRECT; 8769 8770 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n", 8771 act > act_max ? "Illegal" : "Driver unsupported", 8772 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A"); 8773 } 8774 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); 8775 8776 static bool sock_addr_is_valid_access(int off, int size, 8777 enum bpf_access_type type, 8778 const struct bpf_prog *prog, 8779 struct bpf_insn_access_aux *info) 8780 { 8781 const int size_default = sizeof(__u32); 8782 8783 if (off < 0 || off >= sizeof(struct bpf_sock_addr)) 8784 return false; 8785 if (off % size != 0) 8786 return false; 8787 8788 /* Disallow access to IPv6 fields from IPv4 contex and vise 8789 * versa. 8790 */ 8791 switch (off) { 8792 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 8793 switch (prog->expected_attach_type) { 8794 case BPF_CGROUP_INET4_BIND: 8795 case BPF_CGROUP_INET4_CONNECT: 8796 case BPF_CGROUP_INET4_GETPEERNAME: 8797 case BPF_CGROUP_INET4_GETSOCKNAME: 8798 case BPF_CGROUP_UDP4_SENDMSG: 8799 case BPF_CGROUP_UDP4_RECVMSG: 8800 break; 8801 default: 8802 return false; 8803 } 8804 break; 8805 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8806 switch (prog->expected_attach_type) { 8807 case BPF_CGROUP_INET6_BIND: 8808 case BPF_CGROUP_INET6_CONNECT: 8809 case BPF_CGROUP_INET6_GETPEERNAME: 8810 case BPF_CGROUP_INET6_GETSOCKNAME: 8811 case BPF_CGROUP_UDP6_SENDMSG: 8812 case BPF_CGROUP_UDP6_RECVMSG: 8813 break; 8814 default: 8815 return false; 8816 } 8817 break; 8818 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8819 switch (prog->expected_attach_type) { 8820 case BPF_CGROUP_UDP4_SENDMSG: 8821 break; 8822 default: 8823 return false; 8824 } 8825 break; 8826 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8827 msg_src_ip6[3]): 8828 switch (prog->expected_attach_type) { 8829 case BPF_CGROUP_UDP6_SENDMSG: 8830 break; 8831 default: 8832 return false; 8833 } 8834 break; 8835 } 8836 8837 switch (off) { 8838 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 8839 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8840 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8841 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8842 msg_src_ip6[3]): 8843 case bpf_ctx_range(struct bpf_sock_addr, user_port): 8844 if (type == BPF_READ) { 8845 bpf_ctx_record_field_size(info, size_default); 8846 8847 if (bpf_ctx_wide_access_ok(off, size, 8848 struct bpf_sock_addr, 8849 user_ip6)) 8850 return true; 8851 8852 if (bpf_ctx_wide_access_ok(off, size, 8853 struct bpf_sock_addr, 8854 msg_src_ip6)) 8855 return true; 8856 8857 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8858 return false; 8859 } else { 8860 if (bpf_ctx_wide_access_ok(off, size, 8861 struct bpf_sock_addr, 8862 user_ip6)) 8863 return true; 8864 8865 if (bpf_ctx_wide_access_ok(off, size, 8866 struct bpf_sock_addr, 8867 msg_src_ip6)) 8868 return true; 8869 8870 if (size != size_default) 8871 return false; 8872 } 8873 break; 8874 case offsetof(struct bpf_sock_addr, sk): 8875 if (type != BPF_READ) 8876 return false; 8877 if (size != sizeof(__u64)) 8878 return false; 8879 info->reg_type = PTR_TO_SOCKET; 8880 break; 8881 default: 8882 if (type == BPF_READ) { 8883 if (size != size_default) 8884 return false; 8885 } else { 8886 return false; 8887 } 8888 } 8889 8890 return true; 8891 } 8892 8893 static bool sock_ops_is_valid_access(int off, int size, 8894 enum bpf_access_type type, 8895 const struct bpf_prog *prog, 8896 struct bpf_insn_access_aux *info) 8897 { 8898 const int size_default = sizeof(__u32); 8899 8900 if (off < 0 || off >= sizeof(struct bpf_sock_ops)) 8901 return false; 8902 8903 /* The verifier guarantees that size > 0. */ 8904 if (off % size != 0) 8905 return false; 8906 8907 if (type == BPF_WRITE) { 8908 switch (off) { 8909 case offsetof(struct bpf_sock_ops, reply): 8910 case offsetof(struct bpf_sock_ops, sk_txhash): 8911 if (size != size_default) 8912 return false; 8913 break; 8914 default: 8915 return false; 8916 } 8917 } else { 8918 switch (off) { 8919 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received, 8920 bytes_acked): 8921 if (size != sizeof(__u64)) 8922 return false; 8923 break; 8924 case offsetof(struct bpf_sock_ops, sk): 8925 if (size != sizeof(__u64)) 8926 return false; 8927 info->reg_type = PTR_TO_SOCKET_OR_NULL; 8928 break; 8929 case offsetof(struct bpf_sock_ops, skb_data): 8930 if (size != sizeof(__u64)) 8931 return false; 8932 info->reg_type = PTR_TO_PACKET; 8933 break; 8934 case offsetof(struct bpf_sock_ops, skb_data_end): 8935 if (size != sizeof(__u64)) 8936 return false; 8937 info->reg_type = PTR_TO_PACKET_END; 8938 break; 8939 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 8940 bpf_ctx_record_field_size(info, size_default); 8941 return bpf_ctx_narrow_access_ok(off, size, 8942 size_default); 8943 default: 8944 if (size != size_default) 8945 return false; 8946 break; 8947 } 8948 } 8949 8950 return true; 8951 } 8952 8953 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write, 8954 const struct bpf_prog *prog) 8955 { 8956 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP); 8957 } 8958 8959 static bool sk_skb_is_valid_access(int off, int size, 8960 enum bpf_access_type type, 8961 const struct bpf_prog *prog, 8962 struct bpf_insn_access_aux *info) 8963 { 8964 switch (off) { 8965 case bpf_ctx_range(struct __sk_buff, tc_classid): 8966 case bpf_ctx_range(struct __sk_buff, data_meta): 8967 case bpf_ctx_range(struct __sk_buff, tstamp): 8968 case bpf_ctx_range(struct __sk_buff, wire_len): 8969 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8970 return false; 8971 } 8972 8973 if (type == BPF_WRITE) { 8974 switch (off) { 8975 case bpf_ctx_range(struct __sk_buff, tc_index): 8976 case bpf_ctx_range(struct __sk_buff, priority): 8977 break; 8978 default: 8979 return false; 8980 } 8981 } 8982 8983 switch (off) { 8984 case bpf_ctx_range(struct __sk_buff, mark): 8985 return false; 8986 case bpf_ctx_range(struct __sk_buff, data): 8987 info->reg_type = PTR_TO_PACKET; 8988 break; 8989 case bpf_ctx_range(struct __sk_buff, data_end): 8990 info->reg_type = PTR_TO_PACKET_END; 8991 break; 8992 } 8993 8994 return bpf_skb_is_valid_access(off, size, type, prog, info); 8995 } 8996 8997 static bool sk_msg_is_valid_access(int off, int size, 8998 enum bpf_access_type type, 8999 const struct bpf_prog *prog, 9000 struct bpf_insn_access_aux *info) 9001 { 9002 if (type == BPF_WRITE) 9003 return false; 9004 9005 if (off % size != 0) 9006 return false; 9007 9008 switch (off) { 9009 case offsetof(struct sk_msg_md, data): 9010 info->reg_type = PTR_TO_PACKET; 9011 if (size != sizeof(__u64)) 9012 return false; 9013 break; 9014 case offsetof(struct sk_msg_md, data_end): 9015 info->reg_type = PTR_TO_PACKET_END; 9016 if (size != sizeof(__u64)) 9017 return false; 9018 break; 9019 case offsetof(struct sk_msg_md, sk): 9020 if (size != sizeof(__u64)) 9021 return false; 9022 info->reg_type = PTR_TO_SOCKET; 9023 break; 9024 case bpf_ctx_range(struct sk_msg_md, family): 9025 case bpf_ctx_range(struct sk_msg_md, remote_ip4): 9026 case bpf_ctx_range(struct sk_msg_md, local_ip4): 9027 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]): 9028 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]): 9029 case bpf_ctx_range(struct sk_msg_md, remote_port): 9030 case bpf_ctx_range(struct sk_msg_md, local_port): 9031 case bpf_ctx_range(struct sk_msg_md, size): 9032 if (size != sizeof(__u32)) 9033 return false; 9034 break; 9035 default: 9036 return false; 9037 } 9038 return true; 9039 } 9040 9041 static bool flow_dissector_is_valid_access(int off, int size, 9042 enum bpf_access_type type, 9043 const struct bpf_prog *prog, 9044 struct bpf_insn_access_aux *info) 9045 { 9046 const int size_default = sizeof(__u32); 9047 9048 if (off < 0 || off >= sizeof(struct __sk_buff)) 9049 return false; 9050 9051 if (type == BPF_WRITE) 9052 return false; 9053 9054 switch (off) { 9055 case bpf_ctx_range(struct __sk_buff, data): 9056 if (size != size_default) 9057 return false; 9058 info->reg_type = PTR_TO_PACKET; 9059 return true; 9060 case bpf_ctx_range(struct __sk_buff, data_end): 9061 if (size != size_default) 9062 return false; 9063 info->reg_type = PTR_TO_PACKET_END; 9064 return true; 9065 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 9066 if (size != sizeof(__u64)) 9067 return false; 9068 info->reg_type = PTR_TO_FLOW_KEYS; 9069 return true; 9070 default: 9071 return false; 9072 } 9073 } 9074 9075 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type, 9076 const struct bpf_insn *si, 9077 struct bpf_insn *insn_buf, 9078 struct bpf_prog *prog, 9079 u32 *target_size) 9080 9081 { 9082 struct bpf_insn *insn = insn_buf; 9083 9084 switch (si->off) { 9085 case offsetof(struct __sk_buff, data): 9086 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data), 9087 si->dst_reg, si->src_reg, 9088 offsetof(struct bpf_flow_dissector, data)); 9089 break; 9090 9091 case offsetof(struct __sk_buff, data_end): 9092 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end), 9093 si->dst_reg, si->src_reg, 9094 offsetof(struct bpf_flow_dissector, data_end)); 9095 break; 9096 9097 case offsetof(struct __sk_buff, flow_keys): 9098 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys), 9099 si->dst_reg, si->src_reg, 9100 offsetof(struct bpf_flow_dissector, flow_keys)); 9101 break; 9102 } 9103 9104 return insn - insn_buf; 9105 } 9106 9107 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si, 9108 struct bpf_insn *insn) 9109 { 9110 __u8 value_reg = si->dst_reg; 9111 __u8 skb_reg = si->src_reg; 9112 /* AX is needed because src_reg and dst_reg could be the same */ 9113 __u8 tmp_reg = BPF_REG_AX; 9114 9115 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, 9116 PKT_VLAN_PRESENT_OFFSET); 9117 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, 9118 SKB_MONO_DELIVERY_TIME_MASK, 2); 9119 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC); 9120 *insn++ = BPF_JMP_A(1); 9121 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO); 9122 9123 return insn; 9124 } 9125 9126 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si, 9127 struct bpf_insn *insn) 9128 { 9129 /* si->dst_reg = skb_shinfo(SKB); */ 9130 #ifdef NET_SKBUFF_DATA_USES_OFFSET 9131 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 9132 BPF_REG_AX, si->src_reg, 9133 offsetof(struct sk_buff, end)); 9134 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head), 9135 si->dst_reg, si->src_reg, 9136 offsetof(struct sk_buff, head)); 9137 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX); 9138 #else 9139 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 9140 si->dst_reg, si->src_reg, 9141 offsetof(struct sk_buff, end)); 9142 #endif 9143 9144 return insn; 9145 } 9146 9147 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog, 9148 const struct bpf_insn *si, 9149 struct bpf_insn *insn) 9150 { 9151 __u8 value_reg = si->dst_reg; 9152 __u8 skb_reg = si->src_reg; 9153 9154 #ifdef CONFIG_NET_CLS_ACT 9155 /* If the tstamp_type is read, 9156 * the bpf prog is aware the tstamp could have delivery time. 9157 * Thus, read skb->tstamp as is if tstamp_type_access is true. 9158 */ 9159 if (!prog->tstamp_type_access) { 9160 /* AX is needed because src_reg and dst_reg could be the same */ 9161 __u8 tmp_reg = BPF_REG_AX; 9162 9163 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET); 9164 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, 9165 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK); 9166 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg, 9167 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2); 9168 /* skb->tc_at_ingress && skb->mono_delivery_time, 9169 * read 0 as the (rcv) timestamp. 9170 */ 9171 *insn++ = BPF_MOV64_IMM(value_reg, 0); 9172 *insn++ = BPF_JMP_A(1); 9173 } 9174 #endif 9175 9176 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg, 9177 offsetof(struct sk_buff, tstamp)); 9178 return insn; 9179 } 9180 9181 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog, 9182 const struct bpf_insn *si, 9183 struct bpf_insn *insn) 9184 { 9185 __u8 value_reg = si->src_reg; 9186 __u8 skb_reg = si->dst_reg; 9187 9188 #ifdef CONFIG_NET_CLS_ACT 9189 /* If the tstamp_type is read, 9190 * the bpf prog is aware the tstamp could have delivery time. 9191 * Thus, write skb->tstamp as is if tstamp_type_access is true. 9192 * Otherwise, writing at ingress will have to clear the 9193 * mono_delivery_time bit also. 9194 */ 9195 if (!prog->tstamp_type_access) { 9196 __u8 tmp_reg = BPF_REG_AX; 9197 9198 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET); 9199 /* Writing __sk_buff->tstamp as ingress, goto <clear> */ 9200 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1); 9201 /* goto <store> */ 9202 *insn++ = BPF_JMP_A(2); 9203 /* <clear>: mono_delivery_time */ 9204 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK); 9205 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET); 9206 } 9207 #endif 9208 9209 /* <store>: skb->tstamp = tstamp */ 9210 *insn++ = BPF_STX_MEM(BPF_DW, skb_reg, value_reg, 9211 offsetof(struct sk_buff, tstamp)); 9212 return insn; 9213 } 9214 9215 static u32 bpf_convert_ctx_access(enum bpf_access_type type, 9216 const struct bpf_insn *si, 9217 struct bpf_insn *insn_buf, 9218 struct bpf_prog *prog, u32 *target_size) 9219 { 9220 struct bpf_insn *insn = insn_buf; 9221 int off; 9222 9223 switch (si->off) { 9224 case offsetof(struct __sk_buff, len): 9225 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9226 bpf_target_off(struct sk_buff, len, 4, 9227 target_size)); 9228 break; 9229 9230 case offsetof(struct __sk_buff, protocol): 9231 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9232 bpf_target_off(struct sk_buff, protocol, 2, 9233 target_size)); 9234 break; 9235 9236 case offsetof(struct __sk_buff, vlan_proto): 9237 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9238 bpf_target_off(struct sk_buff, vlan_proto, 2, 9239 target_size)); 9240 break; 9241 9242 case offsetof(struct __sk_buff, priority): 9243 if (type == BPF_WRITE) 9244 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9245 bpf_target_off(struct sk_buff, priority, 4, 9246 target_size)); 9247 else 9248 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9249 bpf_target_off(struct sk_buff, priority, 4, 9250 target_size)); 9251 break; 9252 9253 case offsetof(struct __sk_buff, ingress_ifindex): 9254 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9255 bpf_target_off(struct sk_buff, skb_iif, 4, 9256 target_size)); 9257 break; 9258 9259 case offsetof(struct __sk_buff, ifindex): 9260 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 9261 si->dst_reg, si->src_reg, 9262 offsetof(struct sk_buff, dev)); 9263 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9264 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9265 bpf_target_off(struct net_device, ifindex, 4, 9266 target_size)); 9267 break; 9268 9269 case offsetof(struct __sk_buff, hash): 9270 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9271 bpf_target_off(struct sk_buff, hash, 4, 9272 target_size)); 9273 break; 9274 9275 case offsetof(struct __sk_buff, mark): 9276 if (type == BPF_WRITE) 9277 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9278 bpf_target_off(struct sk_buff, mark, 4, 9279 target_size)); 9280 else 9281 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9282 bpf_target_off(struct sk_buff, mark, 4, 9283 target_size)); 9284 break; 9285 9286 case offsetof(struct __sk_buff, pkt_type): 9287 *target_size = 1; 9288 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 9289 PKT_TYPE_OFFSET); 9290 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX); 9291 #ifdef __BIG_ENDIAN_BITFIELD 9292 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5); 9293 #endif 9294 break; 9295 9296 case offsetof(struct __sk_buff, queue_mapping): 9297 if (type == BPF_WRITE) { 9298 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1); 9299 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 9300 bpf_target_off(struct sk_buff, 9301 queue_mapping, 9302 2, target_size)); 9303 } else { 9304 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9305 bpf_target_off(struct sk_buff, 9306 queue_mapping, 9307 2, target_size)); 9308 } 9309 break; 9310 9311 case offsetof(struct __sk_buff, vlan_present): 9312 *target_size = 1; 9313 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 9314 PKT_VLAN_PRESENT_OFFSET); 9315 if (PKT_VLAN_PRESENT_BIT) 9316 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT); 9317 if (PKT_VLAN_PRESENT_BIT < 7) 9318 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1); 9319 break; 9320 9321 case offsetof(struct __sk_buff, vlan_tci): 9322 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9323 bpf_target_off(struct sk_buff, vlan_tci, 2, 9324 target_size)); 9325 break; 9326 9327 case offsetof(struct __sk_buff, cb[0]) ... 9328 offsetofend(struct __sk_buff, cb[4]) - 1: 9329 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20); 9330 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 9331 offsetof(struct qdisc_skb_cb, data)) % 9332 sizeof(__u64)); 9333 9334 prog->cb_access = 1; 9335 off = si->off; 9336 off -= offsetof(struct __sk_buff, cb[0]); 9337 off += offsetof(struct sk_buff, cb); 9338 off += offsetof(struct qdisc_skb_cb, data); 9339 if (type == BPF_WRITE) 9340 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 9341 si->src_reg, off); 9342 else 9343 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 9344 si->src_reg, off); 9345 break; 9346 9347 case offsetof(struct __sk_buff, tc_classid): 9348 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2); 9349 9350 off = si->off; 9351 off -= offsetof(struct __sk_buff, tc_classid); 9352 off += offsetof(struct sk_buff, cb); 9353 off += offsetof(struct qdisc_skb_cb, tc_classid); 9354 *target_size = 2; 9355 if (type == BPF_WRITE) 9356 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, 9357 si->src_reg, off); 9358 else 9359 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, 9360 si->src_reg, off); 9361 break; 9362 9363 case offsetof(struct __sk_buff, data): 9364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 9365 si->dst_reg, si->src_reg, 9366 offsetof(struct sk_buff, data)); 9367 break; 9368 9369 case offsetof(struct __sk_buff, data_meta): 9370 off = si->off; 9371 off -= offsetof(struct __sk_buff, data_meta); 9372 off += offsetof(struct sk_buff, cb); 9373 off += offsetof(struct bpf_skb_data_end, data_meta); 9374 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 9375 si->src_reg, off); 9376 break; 9377 9378 case offsetof(struct __sk_buff, data_end): 9379 off = si->off; 9380 off -= offsetof(struct __sk_buff, data_end); 9381 off += offsetof(struct sk_buff, cb); 9382 off += offsetof(struct bpf_skb_data_end, data_end); 9383 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 9384 si->src_reg, off); 9385 break; 9386 9387 case offsetof(struct __sk_buff, tc_index): 9388 #ifdef CONFIG_NET_SCHED 9389 if (type == BPF_WRITE) 9390 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 9391 bpf_target_off(struct sk_buff, tc_index, 2, 9392 target_size)); 9393 else 9394 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9395 bpf_target_off(struct sk_buff, tc_index, 2, 9396 target_size)); 9397 #else 9398 *target_size = 2; 9399 if (type == BPF_WRITE) 9400 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg); 9401 else 9402 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9403 #endif 9404 break; 9405 9406 case offsetof(struct __sk_buff, napi_id): 9407 #if defined(CONFIG_NET_RX_BUSY_POLL) 9408 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9409 bpf_target_off(struct sk_buff, napi_id, 4, 9410 target_size)); 9411 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1); 9412 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9413 #else 9414 *target_size = 4; 9415 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9416 #endif 9417 break; 9418 case offsetof(struct __sk_buff, family): 9419 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 9420 9421 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9422 si->dst_reg, si->src_reg, 9423 offsetof(struct sk_buff, sk)); 9424 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9425 bpf_target_off(struct sock_common, 9426 skc_family, 9427 2, target_size)); 9428 break; 9429 case offsetof(struct __sk_buff, remote_ip4): 9430 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 9431 9432 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9433 si->dst_reg, si->src_reg, 9434 offsetof(struct sk_buff, sk)); 9435 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9436 bpf_target_off(struct sock_common, 9437 skc_daddr, 9438 4, target_size)); 9439 break; 9440 case offsetof(struct __sk_buff, local_ip4): 9441 BUILD_BUG_ON(sizeof_field(struct sock_common, 9442 skc_rcv_saddr) != 4); 9443 9444 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9445 si->dst_reg, si->src_reg, 9446 offsetof(struct sk_buff, sk)); 9447 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9448 bpf_target_off(struct sock_common, 9449 skc_rcv_saddr, 9450 4, target_size)); 9451 break; 9452 case offsetof(struct __sk_buff, remote_ip6[0]) ... 9453 offsetof(struct __sk_buff, remote_ip6[3]): 9454 #if IS_ENABLED(CONFIG_IPV6) 9455 BUILD_BUG_ON(sizeof_field(struct sock_common, 9456 skc_v6_daddr.s6_addr32[0]) != 4); 9457 9458 off = si->off; 9459 off -= offsetof(struct __sk_buff, remote_ip6[0]); 9460 9461 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9462 si->dst_reg, si->src_reg, 9463 offsetof(struct sk_buff, sk)); 9464 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9465 offsetof(struct sock_common, 9466 skc_v6_daddr.s6_addr32[0]) + 9467 off); 9468 #else 9469 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9470 #endif 9471 break; 9472 case offsetof(struct __sk_buff, local_ip6[0]) ... 9473 offsetof(struct __sk_buff, local_ip6[3]): 9474 #if IS_ENABLED(CONFIG_IPV6) 9475 BUILD_BUG_ON(sizeof_field(struct sock_common, 9476 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 9477 9478 off = si->off; 9479 off -= offsetof(struct __sk_buff, local_ip6[0]); 9480 9481 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9482 si->dst_reg, si->src_reg, 9483 offsetof(struct sk_buff, sk)); 9484 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9485 offsetof(struct sock_common, 9486 skc_v6_rcv_saddr.s6_addr32[0]) + 9487 off); 9488 #else 9489 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9490 #endif 9491 break; 9492 9493 case offsetof(struct __sk_buff, remote_port): 9494 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 9495 9496 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9497 si->dst_reg, si->src_reg, 9498 offsetof(struct sk_buff, sk)); 9499 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9500 bpf_target_off(struct sock_common, 9501 skc_dport, 9502 2, target_size)); 9503 #ifndef __BIG_ENDIAN_BITFIELD 9504 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 9505 #endif 9506 break; 9507 9508 case offsetof(struct __sk_buff, local_port): 9509 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 9510 9511 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9512 si->dst_reg, si->src_reg, 9513 offsetof(struct sk_buff, sk)); 9514 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9515 bpf_target_off(struct sock_common, 9516 skc_num, 2, target_size)); 9517 break; 9518 9519 case offsetof(struct __sk_buff, tstamp): 9520 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8); 9521 9522 if (type == BPF_WRITE) 9523 insn = bpf_convert_tstamp_write(prog, si, insn); 9524 else 9525 insn = bpf_convert_tstamp_read(prog, si, insn); 9526 break; 9527 9528 case offsetof(struct __sk_buff, tstamp_type): 9529 insn = bpf_convert_tstamp_type_read(si, insn); 9530 break; 9531 9532 case offsetof(struct __sk_buff, gso_segs): 9533 insn = bpf_convert_shinfo_access(si, insn); 9534 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs), 9535 si->dst_reg, si->dst_reg, 9536 bpf_target_off(struct skb_shared_info, 9537 gso_segs, 2, 9538 target_size)); 9539 break; 9540 case offsetof(struct __sk_buff, gso_size): 9541 insn = bpf_convert_shinfo_access(si, insn); 9542 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size), 9543 si->dst_reg, si->dst_reg, 9544 bpf_target_off(struct skb_shared_info, 9545 gso_size, 2, 9546 target_size)); 9547 break; 9548 case offsetof(struct __sk_buff, wire_len): 9549 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4); 9550 9551 off = si->off; 9552 off -= offsetof(struct __sk_buff, wire_len); 9553 off += offsetof(struct sk_buff, cb); 9554 off += offsetof(struct qdisc_skb_cb, pkt_len); 9555 *target_size = 4; 9556 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off); 9557 break; 9558 9559 case offsetof(struct __sk_buff, sk): 9560 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9561 si->dst_reg, si->src_reg, 9562 offsetof(struct sk_buff, sk)); 9563 break; 9564 case offsetof(struct __sk_buff, hwtstamp): 9565 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8); 9566 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0); 9567 9568 insn = bpf_convert_shinfo_access(si, insn); 9569 *insn++ = BPF_LDX_MEM(BPF_DW, 9570 si->dst_reg, si->dst_reg, 9571 bpf_target_off(struct skb_shared_info, 9572 hwtstamps, 8, 9573 target_size)); 9574 break; 9575 } 9576 9577 return insn - insn_buf; 9578 } 9579 9580 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 9581 const struct bpf_insn *si, 9582 struct bpf_insn *insn_buf, 9583 struct bpf_prog *prog, u32 *target_size) 9584 { 9585 struct bpf_insn *insn = insn_buf; 9586 int off; 9587 9588 switch (si->off) { 9589 case offsetof(struct bpf_sock, bound_dev_if): 9590 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4); 9591 9592 if (type == BPF_WRITE) 9593 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9594 offsetof(struct sock, sk_bound_dev_if)); 9595 else 9596 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9597 offsetof(struct sock, sk_bound_dev_if)); 9598 break; 9599 9600 case offsetof(struct bpf_sock, mark): 9601 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4); 9602 9603 if (type == BPF_WRITE) 9604 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9605 offsetof(struct sock, sk_mark)); 9606 else 9607 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9608 offsetof(struct sock, sk_mark)); 9609 break; 9610 9611 case offsetof(struct bpf_sock, priority): 9612 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4); 9613 9614 if (type == BPF_WRITE) 9615 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9616 offsetof(struct sock, sk_priority)); 9617 else 9618 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9619 offsetof(struct sock, sk_priority)); 9620 break; 9621 9622 case offsetof(struct bpf_sock, family): 9623 *insn++ = BPF_LDX_MEM( 9624 BPF_FIELD_SIZEOF(struct sock_common, skc_family), 9625 si->dst_reg, si->src_reg, 9626 bpf_target_off(struct sock_common, 9627 skc_family, 9628 sizeof_field(struct sock_common, 9629 skc_family), 9630 target_size)); 9631 break; 9632 9633 case offsetof(struct bpf_sock, type): 9634 *insn++ = BPF_LDX_MEM( 9635 BPF_FIELD_SIZEOF(struct sock, sk_type), 9636 si->dst_reg, si->src_reg, 9637 bpf_target_off(struct sock, sk_type, 9638 sizeof_field(struct sock, sk_type), 9639 target_size)); 9640 break; 9641 9642 case offsetof(struct bpf_sock, protocol): 9643 *insn++ = BPF_LDX_MEM( 9644 BPF_FIELD_SIZEOF(struct sock, sk_protocol), 9645 si->dst_reg, si->src_reg, 9646 bpf_target_off(struct sock, sk_protocol, 9647 sizeof_field(struct sock, sk_protocol), 9648 target_size)); 9649 break; 9650 9651 case offsetof(struct bpf_sock, src_ip4): 9652 *insn++ = BPF_LDX_MEM( 9653 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9654 bpf_target_off(struct sock_common, skc_rcv_saddr, 9655 sizeof_field(struct sock_common, 9656 skc_rcv_saddr), 9657 target_size)); 9658 break; 9659 9660 case offsetof(struct bpf_sock, dst_ip4): 9661 *insn++ = BPF_LDX_MEM( 9662 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9663 bpf_target_off(struct sock_common, skc_daddr, 9664 sizeof_field(struct sock_common, 9665 skc_daddr), 9666 target_size)); 9667 break; 9668 9669 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 9670 #if IS_ENABLED(CONFIG_IPV6) 9671 off = si->off; 9672 off -= offsetof(struct bpf_sock, src_ip6[0]); 9673 *insn++ = BPF_LDX_MEM( 9674 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9675 bpf_target_off( 9676 struct sock_common, 9677 skc_v6_rcv_saddr.s6_addr32[0], 9678 sizeof_field(struct sock_common, 9679 skc_v6_rcv_saddr.s6_addr32[0]), 9680 target_size) + off); 9681 #else 9682 (void)off; 9683 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9684 #endif 9685 break; 9686 9687 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 9688 #if IS_ENABLED(CONFIG_IPV6) 9689 off = si->off; 9690 off -= offsetof(struct bpf_sock, dst_ip6[0]); 9691 *insn++ = BPF_LDX_MEM( 9692 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9693 bpf_target_off(struct sock_common, 9694 skc_v6_daddr.s6_addr32[0], 9695 sizeof_field(struct sock_common, 9696 skc_v6_daddr.s6_addr32[0]), 9697 target_size) + off); 9698 #else 9699 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9700 *target_size = 4; 9701 #endif 9702 break; 9703 9704 case offsetof(struct bpf_sock, src_port): 9705 *insn++ = BPF_LDX_MEM( 9706 BPF_FIELD_SIZEOF(struct sock_common, skc_num), 9707 si->dst_reg, si->src_reg, 9708 bpf_target_off(struct sock_common, skc_num, 9709 sizeof_field(struct sock_common, 9710 skc_num), 9711 target_size)); 9712 break; 9713 9714 case offsetof(struct bpf_sock, dst_port): 9715 *insn++ = BPF_LDX_MEM( 9716 BPF_FIELD_SIZEOF(struct sock_common, skc_dport), 9717 si->dst_reg, si->src_reg, 9718 bpf_target_off(struct sock_common, skc_dport, 9719 sizeof_field(struct sock_common, 9720 skc_dport), 9721 target_size)); 9722 break; 9723 9724 case offsetof(struct bpf_sock, state): 9725 *insn++ = BPF_LDX_MEM( 9726 BPF_FIELD_SIZEOF(struct sock_common, skc_state), 9727 si->dst_reg, si->src_reg, 9728 bpf_target_off(struct sock_common, skc_state, 9729 sizeof_field(struct sock_common, 9730 skc_state), 9731 target_size)); 9732 break; 9733 case offsetof(struct bpf_sock, rx_queue_mapping): 9734 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 9735 *insn++ = BPF_LDX_MEM( 9736 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping), 9737 si->dst_reg, si->src_reg, 9738 bpf_target_off(struct sock, sk_rx_queue_mapping, 9739 sizeof_field(struct sock, 9740 sk_rx_queue_mapping), 9741 target_size)); 9742 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING, 9743 1); 9744 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 9745 #else 9746 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 9747 *target_size = 2; 9748 #endif 9749 break; 9750 } 9751 9752 return insn - insn_buf; 9753 } 9754 9755 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, 9756 const struct bpf_insn *si, 9757 struct bpf_insn *insn_buf, 9758 struct bpf_prog *prog, u32 *target_size) 9759 { 9760 struct bpf_insn *insn = insn_buf; 9761 9762 switch (si->off) { 9763 case offsetof(struct __sk_buff, ifindex): 9764 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 9765 si->dst_reg, si->src_reg, 9766 offsetof(struct sk_buff, dev)); 9767 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9768 bpf_target_off(struct net_device, ifindex, 4, 9769 target_size)); 9770 break; 9771 default: 9772 return bpf_convert_ctx_access(type, si, insn_buf, prog, 9773 target_size); 9774 } 9775 9776 return insn - insn_buf; 9777 } 9778 9779 static u32 xdp_convert_ctx_access(enum bpf_access_type type, 9780 const struct bpf_insn *si, 9781 struct bpf_insn *insn_buf, 9782 struct bpf_prog *prog, u32 *target_size) 9783 { 9784 struct bpf_insn *insn = insn_buf; 9785 9786 switch (si->off) { 9787 case offsetof(struct xdp_md, data): 9788 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data), 9789 si->dst_reg, si->src_reg, 9790 offsetof(struct xdp_buff, data)); 9791 break; 9792 case offsetof(struct xdp_md, data_meta): 9793 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta), 9794 si->dst_reg, si->src_reg, 9795 offsetof(struct xdp_buff, data_meta)); 9796 break; 9797 case offsetof(struct xdp_md, data_end): 9798 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end), 9799 si->dst_reg, si->src_reg, 9800 offsetof(struct xdp_buff, data_end)); 9801 break; 9802 case offsetof(struct xdp_md, ingress_ifindex): 9803 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 9804 si->dst_reg, si->src_reg, 9805 offsetof(struct xdp_buff, rxq)); 9806 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev), 9807 si->dst_reg, si->dst_reg, 9808 offsetof(struct xdp_rxq_info, dev)); 9809 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9810 offsetof(struct net_device, ifindex)); 9811 break; 9812 case offsetof(struct xdp_md, rx_queue_index): 9813 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 9814 si->dst_reg, si->src_reg, 9815 offsetof(struct xdp_buff, rxq)); 9816 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9817 offsetof(struct xdp_rxq_info, 9818 queue_index)); 9819 break; 9820 case offsetof(struct xdp_md, egress_ifindex): 9821 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq), 9822 si->dst_reg, si->src_reg, 9823 offsetof(struct xdp_buff, txq)); 9824 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev), 9825 si->dst_reg, si->dst_reg, 9826 offsetof(struct xdp_txq_info, dev)); 9827 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9828 offsetof(struct net_device, ifindex)); 9829 break; 9830 } 9831 9832 return insn - insn_buf; 9833 } 9834 9835 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of 9836 * context Structure, F is Field in context structure that contains a pointer 9837 * to Nested Structure of type NS that has the field NF. 9838 * 9839 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make 9840 * sure that SIZE is not greater than actual size of S.F.NF. 9841 * 9842 * If offset OFF is provided, the load happens from that offset relative to 9843 * offset of NF. 9844 */ 9845 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \ 9846 do { \ 9847 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \ 9848 si->src_reg, offsetof(S, F)); \ 9849 *insn++ = BPF_LDX_MEM( \ 9850 SIZE, si->dst_reg, si->dst_reg, \ 9851 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 9852 target_size) \ 9853 + OFF); \ 9854 } while (0) 9855 9856 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \ 9857 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \ 9858 BPF_FIELD_SIZEOF(NS, NF), 0) 9859 9860 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to 9861 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation. 9862 * 9863 * In addition it uses Temporary Field TF (member of struct S) as the 3rd 9864 * "register" since two registers available in convert_ctx_access are not 9865 * enough: we can't override neither SRC, since it contains value to store, nor 9866 * DST since it contains pointer to context that may be used by later 9867 * instructions. But we need a temporary place to save pointer to nested 9868 * structure whose field we want to store to. 9869 */ 9870 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \ 9871 do { \ 9872 int tmp_reg = BPF_REG_9; \ 9873 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 9874 --tmp_reg; \ 9875 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 9876 --tmp_reg; \ 9877 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \ 9878 offsetof(S, TF)); \ 9879 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \ 9880 si->dst_reg, offsetof(S, F)); \ 9881 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \ 9882 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 9883 target_size) \ 9884 + OFF); \ 9885 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \ 9886 offsetof(S, TF)); \ 9887 } while (0) 9888 9889 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \ 9890 TF) \ 9891 do { \ 9892 if (type == BPF_WRITE) { \ 9893 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \ 9894 OFF, TF); \ 9895 } else { \ 9896 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \ 9897 S, NS, F, NF, SIZE, OFF); \ 9898 } \ 9899 } while (0) 9900 9901 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \ 9902 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \ 9903 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF) 9904 9905 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type, 9906 const struct bpf_insn *si, 9907 struct bpf_insn *insn_buf, 9908 struct bpf_prog *prog, u32 *target_size) 9909 { 9910 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port); 9911 struct bpf_insn *insn = insn_buf; 9912 9913 switch (si->off) { 9914 case offsetof(struct bpf_sock_addr, user_family): 9915 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9916 struct sockaddr, uaddr, sa_family); 9917 break; 9918 9919 case offsetof(struct bpf_sock_addr, user_ip4): 9920 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9921 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr, 9922 sin_addr, BPF_SIZE(si->code), 0, tmp_reg); 9923 break; 9924 9925 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 9926 off = si->off; 9927 off -= offsetof(struct bpf_sock_addr, user_ip6[0]); 9928 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9929 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 9930 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off, 9931 tmp_reg); 9932 break; 9933 9934 case offsetof(struct bpf_sock_addr, user_port): 9935 /* To get port we need to know sa_family first and then treat 9936 * sockaddr as either sockaddr_in or sockaddr_in6. 9937 * Though we can simplify since port field has same offset and 9938 * size in both structures. 9939 * Here we check this invariant and use just one of the 9940 * structures if it's true. 9941 */ 9942 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) != 9943 offsetof(struct sockaddr_in6, sin6_port)); 9944 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) != 9945 sizeof_field(struct sockaddr_in6, sin6_port)); 9946 /* Account for sin6_port being smaller than user_port. */ 9947 port_size = min(port_size, BPF_LDST_BYTES(si)); 9948 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9949 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 9950 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg); 9951 break; 9952 9953 case offsetof(struct bpf_sock_addr, family): 9954 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9955 struct sock, sk, sk_family); 9956 break; 9957 9958 case offsetof(struct bpf_sock_addr, type): 9959 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9960 struct sock, sk, sk_type); 9961 break; 9962 9963 case offsetof(struct bpf_sock_addr, protocol): 9964 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9965 struct sock, sk, sk_protocol); 9966 break; 9967 9968 case offsetof(struct bpf_sock_addr, msg_src_ip4): 9969 /* Treat t_ctx as struct in_addr for msg_src_ip4. */ 9970 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9971 struct bpf_sock_addr_kern, struct in_addr, t_ctx, 9972 s_addr, BPF_SIZE(si->code), 0, tmp_reg); 9973 break; 9974 9975 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 9976 msg_src_ip6[3]): 9977 off = si->off; 9978 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]); 9979 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */ 9980 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9981 struct bpf_sock_addr_kern, struct in6_addr, t_ctx, 9982 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg); 9983 break; 9984 case offsetof(struct bpf_sock_addr, sk): 9985 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk), 9986 si->dst_reg, si->src_reg, 9987 offsetof(struct bpf_sock_addr_kern, sk)); 9988 break; 9989 } 9990 9991 return insn - insn_buf; 9992 } 9993 9994 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type, 9995 const struct bpf_insn *si, 9996 struct bpf_insn *insn_buf, 9997 struct bpf_prog *prog, 9998 u32 *target_size) 9999 { 10000 struct bpf_insn *insn = insn_buf; 10001 int off; 10002 10003 /* Helper macro for adding read access to tcp_sock or sock fields. */ 10004 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 10005 do { \ 10006 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \ 10007 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 10008 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 10009 if (si->dst_reg == reg || si->src_reg == reg) \ 10010 reg--; \ 10011 if (si->dst_reg == reg || si->src_reg == reg) \ 10012 reg--; \ 10013 if (si->dst_reg == si->src_reg) { \ 10014 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 10015 offsetof(struct bpf_sock_ops_kern, \ 10016 temp)); \ 10017 fullsock_reg = reg; \ 10018 jmp += 2; \ 10019 } \ 10020 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10021 struct bpf_sock_ops_kern, \ 10022 is_fullsock), \ 10023 fullsock_reg, si->src_reg, \ 10024 offsetof(struct bpf_sock_ops_kern, \ 10025 is_fullsock)); \ 10026 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 10027 if (si->dst_reg == si->src_reg) \ 10028 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10029 offsetof(struct bpf_sock_ops_kern, \ 10030 temp)); \ 10031 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10032 struct bpf_sock_ops_kern, sk),\ 10033 si->dst_reg, si->src_reg, \ 10034 offsetof(struct bpf_sock_ops_kern, sk));\ 10035 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \ 10036 OBJ_FIELD), \ 10037 si->dst_reg, si->dst_reg, \ 10038 offsetof(OBJ, OBJ_FIELD)); \ 10039 if (si->dst_reg == si->src_reg) { \ 10040 *insn++ = BPF_JMP_A(1); \ 10041 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10042 offsetof(struct bpf_sock_ops_kern, \ 10043 temp)); \ 10044 } \ 10045 } while (0) 10046 10047 #define SOCK_OPS_GET_SK() \ 10048 do { \ 10049 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \ 10050 if (si->dst_reg == reg || si->src_reg == reg) \ 10051 reg--; \ 10052 if (si->dst_reg == reg || si->src_reg == reg) \ 10053 reg--; \ 10054 if (si->dst_reg == si->src_reg) { \ 10055 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 10056 offsetof(struct bpf_sock_ops_kern, \ 10057 temp)); \ 10058 fullsock_reg = reg; \ 10059 jmp += 2; \ 10060 } \ 10061 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10062 struct bpf_sock_ops_kern, \ 10063 is_fullsock), \ 10064 fullsock_reg, si->src_reg, \ 10065 offsetof(struct bpf_sock_ops_kern, \ 10066 is_fullsock)); \ 10067 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 10068 if (si->dst_reg == si->src_reg) \ 10069 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10070 offsetof(struct bpf_sock_ops_kern, \ 10071 temp)); \ 10072 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10073 struct bpf_sock_ops_kern, sk),\ 10074 si->dst_reg, si->src_reg, \ 10075 offsetof(struct bpf_sock_ops_kern, sk));\ 10076 if (si->dst_reg == si->src_reg) { \ 10077 *insn++ = BPF_JMP_A(1); \ 10078 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 10079 offsetof(struct bpf_sock_ops_kern, \ 10080 temp)); \ 10081 } \ 10082 } while (0) 10083 10084 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \ 10085 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock) 10086 10087 /* Helper macro for adding write access to tcp_sock or sock fields. 10088 * The macro is called with two registers, dst_reg which contains a pointer 10089 * to ctx (context) and src_reg which contains the value that should be 10090 * stored. However, we need an additional register since we cannot overwrite 10091 * dst_reg because it may be used later in the program. 10092 * Instead we "borrow" one of the other register. We first save its value 10093 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore 10094 * it at the end of the macro. 10095 */ 10096 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 10097 do { \ 10098 int reg = BPF_REG_9; \ 10099 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 10100 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 10101 if (si->dst_reg == reg || si->src_reg == reg) \ 10102 reg--; \ 10103 if (si->dst_reg == reg || si->src_reg == reg) \ 10104 reg--; \ 10105 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \ 10106 offsetof(struct bpf_sock_ops_kern, \ 10107 temp)); \ 10108 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10109 struct bpf_sock_ops_kern, \ 10110 is_fullsock), \ 10111 reg, si->dst_reg, \ 10112 offsetof(struct bpf_sock_ops_kern, \ 10113 is_fullsock)); \ 10114 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \ 10115 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 10116 struct bpf_sock_ops_kern, sk),\ 10117 reg, si->dst_reg, \ 10118 offsetof(struct bpf_sock_ops_kern, sk));\ 10119 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \ 10120 reg, si->src_reg, \ 10121 offsetof(OBJ, OBJ_FIELD)); \ 10122 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \ 10123 offsetof(struct bpf_sock_ops_kern, \ 10124 temp)); \ 10125 } while (0) 10126 10127 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \ 10128 do { \ 10129 if (TYPE == BPF_WRITE) \ 10130 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 10131 else \ 10132 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 10133 } while (0) 10134 10135 if (insn > insn_buf) 10136 return insn - insn_buf; 10137 10138 switch (si->off) { 10139 case offsetof(struct bpf_sock_ops, op): 10140 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10141 op), 10142 si->dst_reg, si->src_reg, 10143 offsetof(struct bpf_sock_ops_kern, op)); 10144 break; 10145 10146 case offsetof(struct bpf_sock_ops, replylong[0]) ... 10147 offsetof(struct bpf_sock_ops, replylong[3]): 10148 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) != 10149 sizeof_field(struct bpf_sock_ops_kern, reply)); 10150 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) != 10151 sizeof_field(struct bpf_sock_ops_kern, replylong)); 10152 off = si->off; 10153 off -= offsetof(struct bpf_sock_ops, replylong[0]); 10154 off += offsetof(struct bpf_sock_ops_kern, replylong[0]); 10155 if (type == BPF_WRITE) 10156 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 10157 off); 10158 else 10159 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10160 off); 10161 break; 10162 10163 case offsetof(struct bpf_sock_ops, family): 10164 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 10165 10166 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10167 struct bpf_sock_ops_kern, sk), 10168 si->dst_reg, si->src_reg, 10169 offsetof(struct bpf_sock_ops_kern, sk)); 10170 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10171 offsetof(struct sock_common, skc_family)); 10172 break; 10173 10174 case offsetof(struct bpf_sock_ops, remote_ip4): 10175 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 10176 10177 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10178 struct bpf_sock_ops_kern, sk), 10179 si->dst_reg, si->src_reg, 10180 offsetof(struct bpf_sock_ops_kern, sk)); 10181 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10182 offsetof(struct sock_common, skc_daddr)); 10183 break; 10184 10185 case offsetof(struct bpf_sock_ops, local_ip4): 10186 BUILD_BUG_ON(sizeof_field(struct sock_common, 10187 skc_rcv_saddr) != 4); 10188 10189 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10190 struct bpf_sock_ops_kern, sk), 10191 si->dst_reg, si->src_reg, 10192 offsetof(struct bpf_sock_ops_kern, sk)); 10193 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10194 offsetof(struct sock_common, 10195 skc_rcv_saddr)); 10196 break; 10197 10198 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ... 10199 offsetof(struct bpf_sock_ops, remote_ip6[3]): 10200 #if IS_ENABLED(CONFIG_IPV6) 10201 BUILD_BUG_ON(sizeof_field(struct sock_common, 10202 skc_v6_daddr.s6_addr32[0]) != 4); 10203 10204 off = si->off; 10205 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]); 10206 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10207 struct bpf_sock_ops_kern, sk), 10208 si->dst_reg, si->src_reg, 10209 offsetof(struct bpf_sock_ops_kern, sk)); 10210 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10211 offsetof(struct sock_common, 10212 skc_v6_daddr.s6_addr32[0]) + 10213 off); 10214 #else 10215 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10216 #endif 10217 break; 10218 10219 case offsetof(struct bpf_sock_ops, local_ip6[0]) ... 10220 offsetof(struct bpf_sock_ops, local_ip6[3]): 10221 #if IS_ENABLED(CONFIG_IPV6) 10222 BUILD_BUG_ON(sizeof_field(struct sock_common, 10223 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 10224 10225 off = si->off; 10226 off -= offsetof(struct bpf_sock_ops, local_ip6[0]); 10227 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10228 struct bpf_sock_ops_kern, sk), 10229 si->dst_reg, si->src_reg, 10230 offsetof(struct bpf_sock_ops_kern, sk)); 10231 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10232 offsetof(struct sock_common, 10233 skc_v6_rcv_saddr.s6_addr32[0]) + 10234 off); 10235 #else 10236 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10237 #endif 10238 break; 10239 10240 case offsetof(struct bpf_sock_ops, remote_port): 10241 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 10242 10243 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10244 struct bpf_sock_ops_kern, sk), 10245 si->dst_reg, si->src_reg, 10246 offsetof(struct bpf_sock_ops_kern, sk)); 10247 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10248 offsetof(struct sock_common, skc_dport)); 10249 #ifndef __BIG_ENDIAN_BITFIELD 10250 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 10251 #endif 10252 break; 10253 10254 case offsetof(struct bpf_sock_ops, local_port): 10255 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 10256 10257 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10258 struct bpf_sock_ops_kern, sk), 10259 si->dst_reg, si->src_reg, 10260 offsetof(struct bpf_sock_ops_kern, sk)); 10261 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10262 offsetof(struct sock_common, skc_num)); 10263 break; 10264 10265 case offsetof(struct bpf_sock_ops, is_fullsock): 10266 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10267 struct bpf_sock_ops_kern, 10268 is_fullsock), 10269 si->dst_reg, si->src_reg, 10270 offsetof(struct bpf_sock_ops_kern, 10271 is_fullsock)); 10272 break; 10273 10274 case offsetof(struct bpf_sock_ops, state): 10275 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1); 10276 10277 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10278 struct bpf_sock_ops_kern, sk), 10279 si->dst_reg, si->src_reg, 10280 offsetof(struct bpf_sock_ops_kern, sk)); 10281 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg, 10282 offsetof(struct sock_common, skc_state)); 10283 break; 10284 10285 case offsetof(struct bpf_sock_ops, rtt_min): 10286 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 10287 sizeof(struct minmax)); 10288 BUILD_BUG_ON(sizeof(struct minmax) < 10289 sizeof(struct minmax_sample)); 10290 10291 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10292 struct bpf_sock_ops_kern, sk), 10293 si->dst_reg, si->src_reg, 10294 offsetof(struct bpf_sock_ops_kern, sk)); 10295 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10296 offsetof(struct tcp_sock, rtt_min) + 10297 sizeof_field(struct minmax_sample, t)); 10298 break; 10299 10300 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags): 10301 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags, 10302 struct tcp_sock); 10303 break; 10304 10305 case offsetof(struct bpf_sock_ops, sk_txhash): 10306 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash, 10307 struct sock, type); 10308 break; 10309 case offsetof(struct bpf_sock_ops, snd_cwnd): 10310 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd); 10311 break; 10312 case offsetof(struct bpf_sock_ops, srtt_us): 10313 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us); 10314 break; 10315 case offsetof(struct bpf_sock_ops, snd_ssthresh): 10316 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh); 10317 break; 10318 case offsetof(struct bpf_sock_ops, rcv_nxt): 10319 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt); 10320 break; 10321 case offsetof(struct bpf_sock_ops, snd_nxt): 10322 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt); 10323 break; 10324 case offsetof(struct bpf_sock_ops, snd_una): 10325 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una); 10326 break; 10327 case offsetof(struct bpf_sock_ops, mss_cache): 10328 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache); 10329 break; 10330 case offsetof(struct bpf_sock_ops, ecn_flags): 10331 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags); 10332 break; 10333 case offsetof(struct bpf_sock_ops, rate_delivered): 10334 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered); 10335 break; 10336 case offsetof(struct bpf_sock_ops, rate_interval_us): 10337 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us); 10338 break; 10339 case offsetof(struct bpf_sock_ops, packets_out): 10340 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out); 10341 break; 10342 case offsetof(struct bpf_sock_ops, retrans_out): 10343 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out); 10344 break; 10345 case offsetof(struct bpf_sock_ops, total_retrans): 10346 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans); 10347 break; 10348 case offsetof(struct bpf_sock_ops, segs_in): 10349 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in); 10350 break; 10351 case offsetof(struct bpf_sock_ops, data_segs_in): 10352 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in); 10353 break; 10354 case offsetof(struct bpf_sock_ops, segs_out): 10355 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out); 10356 break; 10357 case offsetof(struct bpf_sock_ops, data_segs_out): 10358 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out); 10359 break; 10360 case offsetof(struct bpf_sock_ops, lost_out): 10361 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out); 10362 break; 10363 case offsetof(struct bpf_sock_ops, sacked_out): 10364 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out); 10365 break; 10366 case offsetof(struct bpf_sock_ops, bytes_received): 10367 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received); 10368 break; 10369 case offsetof(struct bpf_sock_ops, bytes_acked): 10370 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked); 10371 break; 10372 case offsetof(struct bpf_sock_ops, sk): 10373 SOCK_OPS_GET_SK(); 10374 break; 10375 case offsetof(struct bpf_sock_ops, skb_data_end): 10376 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10377 skb_data_end), 10378 si->dst_reg, si->src_reg, 10379 offsetof(struct bpf_sock_ops_kern, 10380 skb_data_end)); 10381 break; 10382 case offsetof(struct bpf_sock_ops, skb_data): 10383 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10384 skb), 10385 si->dst_reg, si->src_reg, 10386 offsetof(struct bpf_sock_ops_kern, 10387 skb)); 10388 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10389 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 10390 si->dst_reg, si->dst_reg, 10391 offsetof(struct sk_buff, data)); 10392 break; 10393 case offsetof(struct bpf_sock_ops, skb_len): 10394 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10395 skb), 10396 si->dst_reg, si->src_reg, 10397 offsetof(struct bpf_sock_ops_kern, 10398 skb)); 10399 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10400 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 10401 si->dst_reg, si->dst_reg, 10402 offsetof(struct sk_buff, len)); 10403 break; 10404 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 10405 off = offsetof(struct sk_buff, cb); 10406 off += offsetof(struct tcp_skb_cb, tcp_flags); 10407 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags); 10408 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10409 skb), 10410 si->dst_reg, si->src_reg, 10411 offsetof(struct bpf_sock_ops_kern, 10412 skb)); 10413 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10414 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb, 10415 tcp_flags), 10416 si->dst_reg, si->dst_reg, off); 10417 break; 10418 } 10419 return insn - insn_buf; 10420 } 10421 10422 /* data_end = skb->data + skb_headlen() */ 10423 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si, 10424 struct bpf_insn *insn) 10425 { 10426 int reg; 10427 int temp_reg_off = offsetof(struct sk_buff, cb) + 10428 offsetof(struct sk_skb_cb, temp_reg); 10429 10430 if (si->src_reg == si->dst_reg) { 10431 /* We need an extra register, choose and save a register. */ 10432 reg = BPF_REG_9; 10433 if (si->src_reg == reg || si->dst_reg == reg) 10434 reg--; 10435 if (si->src_reg == reg || si->dst_reg == reg) 10436 reg--; 10437 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off); 10438 } else { 10439 reg = si->dst_reg; 10440 } 10441 10442 /* reg = skb->data */ 10443 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 10444 reg, si->src_reg, 10445 offsetof(struct sk_buff, data)); 10446 /* AX = skb->len */ 10447 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 10448 BPF_REG_AX, si->src_reg, 10449 offsetof(struct sk_buff, len)); 10450 /* reg = skb->data + skb->len */ 10451 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX); 10452 /* AX = skb->data_len */ 10453 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len), 10454 BPF_REG_AX, si->src_reg, 10455 offsetof(struct sk_buff, data_len)); 10456 10457 /* reg = skb->data + skb->len - skb->data_len */ 10458 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX); 10459 10460 if (si->src_reg == si->dst_reg) { 10461 /* Restore the saved register */ 10462 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg); 10463 *insn++ = BPF_MOV64_REG(si->dst_reg, reg); 10464 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off); 10465 } 10466 10467 return insn; 10468 } 10469 10470 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type, 10471 const struct bpf_insn *si, 10472 struct bpf_insn *insn_buf, 10473 struct bpf_prog *prog, u32 *target_size) 10474 { 10475 struct bpf_insn *insn = insn_buf; 10476 int off; 10477 10478 switch (si->off) { 10479 case offsetof(struct __sk_buff, data_end): 10480 insn = bpf_convert_data_end_access(si, insn); 10481 break; 10482 case offsetof(struct __sk_buff, cb[0]) ... 10483 offsetofend(struct __sk_buff, cb[4]) - 1: 10484 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20); 10485 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 10486 offsetof(struct sk_skb_cb, data)) % 10487 sizeof(__u64)); 10488 10489 prog->cb_access = 1; 10490 off = si->off; 10491 off -= offsetof(struct __sk_buff, cb[0]); 10492 off += offsetof(struct sk_buff, cb); 10493 off += offsetof(struct sk_skb_cb, data); 10494 if (type == BPF_WRITE) 10495 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 10496 si->src_reg, off); 10497 else 10498 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 10499 si->src_reg, off); 10500 break; 10501 10502 10503 default: 10504 return bpf_convert_ctx_access(type, si, insn_buf, prog, 10505 target_size); 10506 } 10507 10508 return insn - insn_buf; 10509 } 10510 10511 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type, 10512 const struct bpf_insn *si, 10513 struct bpf_insn *insn_buf, 10514 struct bpf_prog *prog, u32 *target_size) 10515 { 10516 struct bpf_insn *insn = insn_buf; 10517 #if IS_ENABLED(CONFIG_IPV6) 10518 int off; 10519 #endif 10520 10521 /* convert ctx uses the fact sg element is first in struct */ 10522 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0); 10523 10524 switch (si->off) { 10525 case offsetof(struct sk_msg_md, data): 10526 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data), 10527 si->dst_reg, si->src_reg, 10528 offsetof(struct sk_msg, data)); 10529 break; 10530 case offsetof(struct sk_msg_md, data_end): 10531 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end), 10532 si->dst_reg, si->src_reg, 10533 offsetof(struct sk_msg, data_end)); 10534 break; 10535 case offsetof(struct sk_msg_md, family): 10536 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 10537 10538 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10539 struct sk_msg, sk), 10540 si->dst_reg, si->src_reg, 10541 offsetof(struct sk_msg, sk)); 10542 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10543 offsetof(struct sock_common, skc_family)); 10544 break; 10545 10546 case offsetof(struct sk_msg_md, remote_ip4): 10547 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 10548 10549 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10550 struct sk_msg, sk), 10551 si->dst_reg, si->src_reg, 10552 offsetof(struct sk_msg, sk)); 10553 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10554 offsetof(struct sock_common, skc_daddr)); 10555 break; 10556 10557 case offsetof(struct sk_msg_md, local_ip4): 10558 BUILD_BUG_ON(sizeof_field(struct sock_common, 10559 skc_rcv_saddr) != 4); 10560 10561 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10562 struct sk_msg, sk), 10563 si->dst_reg, si->src_reg, 10564 offsetof(struct sk_msg, sk)); 10565 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10566 offsetof(struct sock_common, 10567 skc_rcv_saddr)); 10568 break; 10569 10570 case offsetof(struct sk_msg_md, remote_ip6[0]) ... 10571 offsetof(struct sk_msg_md, remote_ip6[3]): 10572 #if IS_ENABLED(CONFIG_IPV6) 10573 BUILD_BUG_ON(sizeof_field(struct sock_common, 10574 skc_v6_daddr.s6_addr32[0]) != 4); 10575 10576 off = si->off; 10577 off -= offsetof(struct sk_msg_md, remote_ip6[0]); 10578 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10579 struct sk_msg, sk), 10580 si->dst_reg, si->src_reg, 10581 offsetof(struct sk_msg, sk)); 10582 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10583 offsetof(struct sock_common, 10584 skc_v6_daddr.s6_addr32[0]) + 10585 off); 10586 #else 10587 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10588 #endif 10589 break; 10590 10591 case offsetof(struct sk_msg_md, local_ip6[0]) ... 10592 offsetof(struct sk_msg_md, local_ip6[3]): 10593 #if IS_ENABLED(CONFIG_IPV6) 10594 BUILD_BUG_ON(sizeof_field(struct sock_common, 10595 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 10596 10597 off = si->off; 10598 off -= offsetof(struct sk_msg_md, local_ip6[0]); 10599 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10600 struct sk_msg, sk), 10601 si->dst_reg, si->src_reg, 10602 offsetof(struct sk_msg, sk)); 10603 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10604 offsetof(struct sock_common, 10605 skc_v6_rcv_saddr.s6_addr32[0]) + 10606 off); 10607 #else 10608 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10609 #endif 10610 break; 10611 10612 case offsetof(struct sk_msg_md, remote_port): 10613 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 10614 10615 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10616 struct sk_msg, sk), 10617 si->dst_reg, si->src_reg, 10618 offsetof(struct sk_msg, sk)); 10619 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10620 offsetof(struct sock_common, skc_dport)); 10621 #ifndef __BIG_ENDIAN_BITFIELD 10622 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 10623 #endif 10624 break; 10625 10626 case offsetof(struct sk_msg_md, local_port): 10627 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 10628 10629 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10630 struct sk_msg, sk), 10631 si->dst_reg, si->src_reg, 10632 offsetof(struct sk_msg, sk)); 10633 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10634 offsetof(struct sock_common, skc_num)); 10635 break; 10636 10637 case offsetof(struct sk_msg_md, size): 10638 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size), 10639 si->dst_reg, si->src_reg, 10640 offsetof(struct sk_msg_sg, size)); 10641 break; 10642 10643 case offsetof(struct sk_msg_md, sk): 10644 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk), 10645 si->dst_reg, si->src_reg, 10646 offsetof(struct sk_msg, sk)); 10647 break; 10648 } 10649 10650 return insn - insn_buf; 10651 } 10652 10653 const struct bpf_verifier_ops sk_filter_verifier_ops = { 10654 .get_func_proto = sk_filter_func_proto, 10655 .is_valid_access = sk_filter_is_valid_access, 10656 .convert_ctx_access = bpf_convert_ctx_access, 10657 .gen_ld_abs = bpf_gen_ld_abs, 10658 }; 10659 10660 const struct bpf_prog_ops sk_filter_prog_ops = { 10661 .test_run = bpf_prog_test_run_skb, 10662 }; 10663 10664 const struct bpf_verifier_ops tc_cls_act_verifier_ops = { 10665 .get_func_proto = tc_cls_act_func_proto, 10666 .is_valid_access = tc_cls_act_is_valid_access, 10667 .convert_ctx_access = tc_cls_act_convert_ctx_access, 10668 .gen_prologue = tc_cls_act_prologue, 10669 .gen_ld_abs = bpf_gen_ld_abs, 10670 }; 10671 10672 const struct bpf_prog_ops tc_cls_act_prog_ops = { 10673 .test_run = bpf_prog_test_run_skb, 10674 }; 10675 10676 const struct bpf_verifier_ops xdp_verifier_ops = { 10677 .get_func_proto = xdp_func_proto, 10678 .is_valid_access = xdp_is_valid_access, 10679 .convert_ctx_access = xdp_convert_ctx_access, 10680 .gen_prologue = bpf_noop_prologue, 10681 }; 10682 10683 const struct bpf_prog_ops xdp_prog_ops = { 10684 .test_run = bpf_prog_test_run_xdp, 10685 }; 10686 10687 const struct bpf_verifier_ops cg_skb_verifier_ops = { 10688 .get_func_proto = cg_skb_func_proto, 10689 .is_valid_access = cg_skb_is_valid_access, 10690 .convert_ctx_access = bpf_convert_ctx_access, 10691 }; 10692 10693 const struct bpf_prog_ops cg_skb_prog_ops = { 10694 .test_run = bpf_prog_test_run_skb, 10695 }; 10696 10697 const struct bpf_verifier_ops lwt_in_verifier_ops = { 10698 .get_func_proto = lwt_in_func_proto, 10699 .is_valid_access = lwt_is_valid_access, 10700 .convert_ctx_access = bpf_convert_ctx_access, 10701 }; 10702 10703 const struct bpf_prog_ops lwt_in_prog_ops = { 10704 .test_run = bpf_prog_test_run_skb, 10705 }; 10706 10707 const struct bpf_verifier_ops lwt_out_verifier_ops = { 10708 .get_func_proto = lwt_out_func_proto, 10709 .is_valid_access = lwt_is_valid_access, 10710 .convert_ctx_access = bpf_convert_ctx_access, 10711 }; 10712 10713 const struct bpf_prog_ops lwt_out_prog_ops = { 10714 .test_run = bpf_prog_test_run_skb, 10715 }; 10716 10717 const struct bpf_verifier_ops lwt_xmit_verifier_ops = { 10718 .get_func_proto = lwt_xmit_func_proto, 10719 .is_valid_access = lwt_is_valid_access, 10720 .convert_ctx_access = bpf_convert_ctx_access, 10721 .gen_prologue = tc_cls_act_prologue, 10722 }; 10723 10724 const struct bpf_prog_ops lwt_xmit_prog_ops = { 10725 .test_run = bpf_prog_test_run_skb, 10726 }; 10727 10728 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = { 10729 .get_func_proto = lwt_seg6local_func_proto, 10730 .is_valid_access = lwt_is_valid_access, 10731 .convert_ctx_access = bpf_convert_ctx_access, 10732 }; 10733 10734 const struct bpf_prog_ops lwt_seg6local_prog_ops = { 10735 .test_run = bpf_prog_test_run_skb, 10736 }; 10737 10738 const struct bpf_verifier_ops cg_sock_verifier_ops = { 10739 .get_func_proto = sock_filter_func_proto, 10740 .is_valid_access = sock_filter_is_valid_access, 10741 .convert_ctx_access = bpf_sock_convert_ctx_access, 10742 }; 10743 10744 const struct bpf_prog_ops cg_sock_prog_ops = { 10745 }; 10746 10747 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = { 10748 .get_func_proto = sock_addr_func_proto, 10749 .is_valid_access = sock_addr_is_valid_access, 10750 .convert_ctx_access = sock_addr_convert_ctx_access, 10751 }; 10752 10753 const struct bpf_prog_ops cg_sock_addr_prog_ops = { 10754 }; 10755 10756 const struct bpf_verifier_ops sock_ops_verifier_ops = { 10757 .get_func_proto = sock_ops_func_proto, 10758 .is_valid_access = sock_ops_is_valid_access, 10759 .convert_ctx_access = sock_ops_convert_ctx_access, 10760 }; 10761 10762 const struct bpf_prog_ops sock_ops_prog_ops = { 10763 }; 10764 10765 const struct bpf_verifier_ops sk_skb_verifier_ops = { 10766 .get_func_proto = sk_skb_func_proto, 10767 .is_valid_access = sk_skb_is_valid_access, 10768 .convert_ctx_access = sk_skb_convert_ctx_access, 10769 .gen_prologue = sk_skb_prologue, 10770 }; 10771 10772 const struct bpf_prog_ops sk_skb_prog_ops = { 10773 }; 10774 10775 const struct bpf_verifier_ops sk_msg_verifier_ops = { 10776 .get_func_proto = sk_msg_func_proto, 10777 .is_valid_access = sk_msg_is_valid_access, 10778 .convert_ctx_access = sk_msg_convert_ctx_access, 10779 .gen_prologue = bpf_noop_prologue, 10780 }; 10781 10782 const struct bpf_prog_ops sk_msg_prog_ops = { 10783 }; 10784 10785 const struct bpf_verifier_ops flow_dissector_verifier_ops = { 10786 .get_func_proto = flow_dissector_func_proto, 10787 .is_valid_access = flow_dissector_is_valid_access, 10788 .convert_ctx_access = flow_dissector_convert_ctx_access, 10789 }; 10790 10791 const struct bpf_prog_ops flow_dissector_prog_ops = { 10792 .test_run = bpf_prog_test_run_flow_dissector, 10793 }; 10794 10795 int sk_detach_filter(struct sock *sk) 10796 { 10797 int ret = -ENOENT; 10798 struct sk_filter *filter; 10799 10800 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 10801 return -EPERM; 10802 10803 filter = rcu_dereference_protected(sk->sk_filter, 10804 lockdep_sock_is_held(sk)); 10805 if (filter) { 10806 RCU_INIT_POINTER(sk->sk_filter, NULL); 10807 sk_filter_uncharge(sk, filter); 10808 ret = 0; 10809 } 10810 10811 return ret; 10812 } 10813 EXPORT_SYMBOL_GPL(sk_detach_filter); 10814 10815 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, 10816 unsigned int len) 10817 { 10818 struct sock_fprog_kern *fprog; 10819 struct sk_filter *filter; 10820 int ret = 0; 10821 10822 lock_sock(sk); 10823 filter = rcu_dereference_protected(sk->sk_filter, 10824 lockdep_sock_is_held(sk)); 10825 if (!filter) 10826 goto out; 10827 10828 /* We're copying the filter that has been originally attached, 10829 * so no conversion/decode needed anymore. eBPF programs that 10830 * have no original program cannot be dumped through this. 10831 */ 10832 ret = -EACCES; 10833 fprog = filter->prog->orig_prog; 10834 if (!fprog) 10835 goto out; 10836 10837 ret = fprog->len; 10838 if (!len) 10839 /* User space only enquires number of filter blocks. */ 10840 goto out; 10841 10842 ret = -EINVAL; 10843 if (len < fprog->len) 10844 goto out; 10845 10846 ret = -EFAULT; 10847 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog))) 10848 goto out; 10849 10850 /* Instead of bytes, the API requests to return the number 10851 * of filter blocks. 10852 */ 10853 ret = fprog->len; 10854 out: 10855 release_sock(sk); 10856 return ret; 10857 } 10858 10859 #ifdef CONFIG_INET 10860 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern, 10861 struct sock_reuseport *reuse, 10862 struct sock *sk, struct sk_buff *skb, 10863 struct sock *migrating_sk, 10864 u32 hash) 10865 { 10866 reuse_kern->skb = skb; 10867 reuse_kern->sk = sk; 10868 reuse_kern->selected_sk = NULL; 10869 reuse_kern->migrating_sk = migrating_sk; 10870 reuse_kern->data_end = skb->data + skb_headlen(skb); 10871 reuse_kern->hash = hash; 10872 reuse_kern->reuseport_id = reuse->reuseport_id; 10873 reuse_kern->bind_inany = reuse->bind_inany; 10874 } 10875 10876 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 10877 struct bpf_prog *prog, struct sk_buff *skb, 10878 struct sock *migrating_sk, 10879 u32 hash) 10880 { 10881 struct sk_reuseport_kern reuse_kern; 10882 enum sk_action action; 10883 10884 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash); 10885 action = bpf_prog_run(prog, &reuse_kern); 10886 10887 if (action == SK_PASS) 10888 return reuse_kern.selected_sk; 10889 else 10890 return ERR_PTR(-ECONNREFUSED); 10891 } 10892 10893 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern, 10894 struct bpf_map *, map, void *, key, u32, flags) 10895 { 10896 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY; 10897 struct sock_reuseport *reuse; 10898 struct sock *selected_sk; 10899 10900 selected_sk = map->ops->map_lookup_elem(map, key); 10901 if (!selected_sk) 10902 return -ENOENT; 10903 10904 reuse = rcu_dereference(selected_sk->sk_reuseport_cb); 10905 if (!reuse) { 10906 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */ 10907 if (sk_is_refcounted(selected_sk)) 10908 sock_put(selected_sk); 10909 10910 /* reuseport_array has only sk with non NULL sk_reuseport_cb. 10911 * The only (!reuse) case here is - the sk has already been 10912 * unhashed (e.g. by close()), so treat it as -ENOENT. 10913 * 10914 * Other maps (e.g. sock_map) do not provide this guarantee and 10915 * the sk may never be in the reuseport group to begin with. 10916 */ 10917 return is_sockarray ? -ENOENT : -EINVAL; 10918 } 10919 10920 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) { 10921 struct sock *sk = reuse_kern->sk; 10922 10923 if (sk->sk_protocol != selected_sk->sk_protocol) 10924 return -EPROTOTYPE; 10925 else if (sk->sk_family != selected_sk->sk_family) 10926 return -EAFNOSUPPORT; 10927 10928 /* Catch all. Likely bound to a different sockaddr. */ 10929 return -EBADFD; 10930 } 10931 10932 reuse_kern->selected_sk = selected_sk; 10933 10934 return 0; 10935 } 10936 10937 static const struct bpf_func_proto sk_select_reuseport_proto = { 10938 .func = sk_select_reuseport, 10939 .gpl_only = false, 10940 .ret_type = RET_INTEGER, 10941 .arg1_type = ARG_PTR_TO_CTX, 10942 .arg2_type = ARG_CONST_MAP_PTR, 10943 .arg3_type = ARG_PTR_TO_MAP_KEY, 10944 .arg4_type = ARG_ANYTHING, 10945 }; 10946 10947 BPF_CALL_4(sk_reuseport_load_bytes, 10948 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 10949 void *, to, u32, len) 10950 { 10951 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len); 10952 } 10953 10954 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = { 10955 .func = sk_reuseport_load_bytes, 10956 .gpl_only = false, 10957 .ret_type = RET_INTEGER, 10958 .arg1_type = ARG_PTR_TO_CTX, 10959 .arg2_type = ARG_ANYTHING, 10960 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 10961 .arg4_type = ARG_CONST_SIZE, 10962 }; 10963 10964 BPF_CALL_5(sk_reuseport_load_bytes_relative, 10965 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 10966 void *, to, u32, len, u32, start_header) 10967 { 10968 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to, 10969 len, start_header); 10970 } 10971 10972 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = { 10973 .func = sk_reuseport_load_bytes_relative, 10974 .gpl_only = false, 10975 .ret_type = RET_INTEGER, 10976 .arg1_type = ARG_PTR_TO_CTX, 10977 .arg2_type = ARG_ANYTHING, 10978 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 10979 .arg4_type = ARG_CONST_SIZE, 10980 .arg5_type = ARG_ANYTHING, 10981 }; 10982 10983 static const struct bpf_func_proto * 10984 sk_reuseport_func_proto(enum bpf_func_id func_id, 10985 const struct bpf_prog *prog) 10986 { 10987 switch (func_id) { 10988 case BPF_FUNC_sk_select_reuseport: 10989 return &sk_select_reuseport_proto; 10990 case BPF_FUNC_skb_load_bytes: 10991 return &sk_reuseport_load_bytes_proto; 10992 case BPF_FUNC_skb_load_bytes_relative: 10993 return &sk_reuseport_load_bytes_relative_proto; 10994 case BPF_FUNC_get_socket_cookie: 10995 return &bpf_get_socket_ptr_cookie_proto; 10996 case BPF_FUNC_ktime_get_coarse_ns: 10997 return &bpf_ktime_get_coarse_ns_proto; 10998 default: 10999 return bpf_base_func_proto(func_id); 11000 } 11001 } 11002 11003 static bool 11004 sk_reuseport_is_valid_access(int off, int size, 11005 enum bpf_access_type type, 11006 const struct bpf_prog *prog, 11007 struct bpf_insn_access_aux *info) 11008 { 11009 const u32 size_default = sizeof(__u32); 11010 11011 if (off < 0 || off >= sizeof(struct sk_reuseport_md) || 11012 off % size || type != BPF_READ) 11013 return false; 11014 11015 switch (off) { 11016 case offsetof(struct sk_reuseport_md, data): 11017 info->reg_type = PTR_TO_PACKET; 11018 return size == sizeof(__u64); 11019 11020 case offsetof(struct sk_reuseport_md, data_end): 11021 info->reg_type = PTR_TO_PACKET_END; 11022 return size == sizeof(__u64); 11023 11024 case offsetof(struct sk_reuseport_md, hash): 11025 return size == size_default; 11026 11027 case offsetof(struct sk_reuseport_md, sk): 11028 info->reg_type = PTR_TO_SOCKET; 11029 return size == sizeof(__u64); 11030 11031 case offsetof(struct sk_reuseport_md, migrating_sk): 11032 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 11033 return size == sizeof(__u64); 11034 11035 /* Fields that allow narrowing */ 11036 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol): 11037 if (size < sizeof_field(struct sk_buff, protocol)) 11038 return false; 11039 fallthrough; 11040 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol): 11041 case bpf_ctx_range(struct sk_reuseport_md, bind_inany): 11042 case bpf_ctx_range(struct sk_reuseport_md, len): 11043 bpf_ctx_record_field_size(info, size_default); 11044 return bpf_ctx_narrow_access_ok(off, size, size_default); 11045 11046 default: 11047 return false; 11048 } 11049 } 11050 11051 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \ 11052 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \ 11053 si->dst_reg, si->src_reg, \ 11054 bpf_target_off(struct sk_reuseport_kern, F, \ 11055 sizeof_field(struct sk_reuseport_kern, F), \ 11056 target_size)); \ 11057 }) 11058 11059 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \ 11060 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 11061 struct sk_buff, \ 11062 skb, \ 11063 SKB_FIELD) 11064 11065 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \ 11066 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 11067 struct sock, \ 11068 sk, \ 11069 SK_FIELD) 11070 11071 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type, 11072 const struct bpf_insn *si, 11073 struct bpf_insn *insn_buf, 11074 struct bpf_prog *prog, 11075 u32 *target_size) 11076 { 11077 struct bpf_insn *insn = insn_buf; 11078 11079 switch (si->off) { 11080 case offsetof(struct sk_reuseport_md, data): 11081 SK_REUSEPORT_LOAD_SKB_FIELD(data); 11082 break; 11083 11084 case offsetof(struct sk_reuseport_md, len): 11085 SK_REUSEPORT_LOAD_SKB_FIELD(len); 11086 break; 11087 11088 case offsetof(struct sk_reuseport_md, eth_protocol): 11089 SK_REUSEPORT_LOAD_SKB_FIELD(protocol); 11090 break; 11091 11092 case offsetof(struct sk_reuseport_md, ip_protocol): 11093 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol); 11094 break; 11095 11096 case offsetof(struct sk_reuseport_md, data_end): 11097 SK_REUSEPORT_LOAD_FIELD(data_end); 11098 break; 11099 11100 case offsetof(struct sk_reuseport_md, hash): 11101 SK_REUSEPORT_LOAD_FIELD(hash); 11102 break; 11103 11104 case offsetof(struct sk_reuseport_md, bind_inany): 11105 SK_REUSEPORT_LOAD_FIELD(bind_inany); 11106 break; 11107 11108 case offsetof(struct sk_reuseport_md, sk): 11109 SK_REUSEPORT_LOAD_FIELD(sk); 11110 break; 11111 11112 case offsetof(struct sk_reuseport_md, migrating_sk): 11113 SK_REUSEPORT_LOAD_FIELD(migrating_sk); 11114 break; 11115 } 11116 11117 return insn - insn_buf; 11118 } 11119 11120 const struct bpf_verifier_ops sk_reuseport_verifier_ops = { 11121 .get_func_proto = sk_reuseport_func_proto, 11122 .is_valid_access = sk_reuseport_is_valid_access, 11123 .convert_ctx_access = sk_reuseport_convert_ctx_access, 11124 }; 11125 11126 const struct bpf_prog_ops sk_reuseport_prog_ops = { 11127 }; 11128 11129 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled); 11130 EXPORT_SYMBOL(bpf_sk_lookup_enabled); 11131 11132 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx, 11133 struct sock *, sk, u64, flags) 11134 { 11135 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE | 11136 BPF_SK_LOOKUP_F_NO_REUSEPORT))) 11137 return -EINVAL; 11138 if (unlikely(sk && sk_is_refcounted(sk))) 11139 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */ 11140 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN)) 11141 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */ 11142 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE)) 11143 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */ 11144 11145 /* Check if socket is suitable for packet L3/L4 protocol */ 11146 if (sk && sk->sk_protocol != ctx->protocol) 11147 return -EPROTOTYPE; 11148 if (sk && sk->sk_family != ctx->family && 11149 (sk->sk_family == AF_INET || ipv6_only_sock(sk))) 11150 return -EAFNOSUPPORT; 11151 11152 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE)) 11153 return -EEXIST; 11154 11155 /* Select socket as lookup result */ 11156 ctx->selected_sk = sk; 11157 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT; 11158 return 0; 11159 } 11160 11161 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = { 11162 .func = bpf_sk_lookup_assign, 11163 .gpl_only = false, 11164 .ret_type = RET_INTEGER, 11165 .arg1_type = ARG_PTR_TO_CTX, 11166 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL, 11167 .arg3_type = ARG_ANYTHING, 11168 }; 11169 11170 static const struct bpf_func_proto * 11171 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 11172 { 11173 switch (func_id) { 11174 case BPF_FUNC_perf_event_output: 11175 return &bpf_event_output_data_proto; 11176 case BPF_FUNC_sk_assign: 11177 return &bpf_sk_lookup_assign_proto; 11178 case BPF_FUNC_sk_release: 11179 return &bpf_sk_release_proto; 11180 default: 11181 return bpf_sk_base_func_proto(func_id); 11182 } 11183 } 11184 11185 static bool sk_lookup_is_valid_access(int off, int size, 11186 enum bpf_access_type type, 11187 const struct bpf_prog *prog, 11188 struct bpf_insn_access_aux *info) 11189 { 11190 if (off < 0 || off >= sizeof(struct bpf_sk_lookup)) 11191 return false; 11192 if (off % size != 0) 11193 return false; 11194 if (type != BPF_READ) 11195 return false; 11196 11197 switch (off) { 11198 case offsetof(struct bpf_sk_lookup, sk): 11199 info->reg_type = PTR_TO_SOCKET_OR_NULL; 11200 return size == sizeof(__u64); 11201 11202 case bpf_ctx_range(struct bpf_sk_lookup, family): 11203 case bpf_ctx_range(struct bpf_sk_lookup, protocol): 11204 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4): 11205 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4): 11206 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]): 11207 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]): 11208 case bpf_ctx_range(struct bpf_sk_lookup, local_port): 11209 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex): 11210 bpf_ctx_record_field_size(info, sizeof(__u32)); 11211 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32)); 11212 11213 case bpf_ctx_range(struct bpf_sk_lookup, remote_port): 11214 /* Allow 4-byte access to 2-byte field for backward compatibility */ 11215 if (size == sizeof(__u32)) 11216 return true; 11217 bpf_ctx_record_field_size(info, sizeof(__be16)); 11218 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16)); 11219 11220 case offsetofend(struct bpf_sk_lookup, remote_port) ... 11221 offsetof(struct bpf_sk_lookup, local_ip4) - 1: 11222 /* Allow access to zero padding for backward compatibility */ 11223 bpf_ctx_record_field_size(info, sizeof(__u16)); 11224 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16)); 11225 11226 default: 11227 return false; 11228 } 11229 } 11230 11231 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type, 11232 const struct bpf_insn *si, 11233 struct bpf_insn *insn_buf, 11234 struct bpf_prog *prog, 11235 u32 *target_size) 11236 { 11237 struct bpf_insn *insn = insn_buf; 11238 11239 switch (si->off) { 11240 case offsetof(struct bpf_sk_lookup, sk): 11241 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11242 offsetof(struct bpf_sk_lookup_kern, selected_sk)); 11243 break; 11244 11245 case offsetof(struct bpf_sk_lookup, family): 11246 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11247 bpf_target_off(struct bpf_sk_lookup_kern, 11248 family, 2, target_size)); 11249 break; 11250 11251 case offsetof(struct bpf_sk_lookup, protocol): 11252 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11253 bpf_target_off(struct bpf_sk_lookup_kern, 11254 protocol, 2, target_size)); 11255 break; 11256 11257 case offsetof(struct bpf_sk_lookup, remote_ip4): 11258 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11259 bpf_target_off(struct bpf_sk_lookup_kern, 11260 v4.saddr, 4, target_size)); 11261 break; 11262 11263 case offsetof(struct bpf_sk_lookup, local_ip4): 11264 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11265 bpf_target_off(struct bpf_sk_lookup_kern, 11266 v4.daddr, 4, target_size)); 11267 break; 11268 11269 case bpf_ctx_range_till(struct bpf_sk_lookup, 11270 remote_ip6[0], remote_ip6[3]): { 11271 #if IS_ENABLED(CONFIG_IPV6) 11272 int off = si->off; 11273 11274 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]); 11275 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 11276 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11277 offsetof(struct bpf_sk_lookup_kern, v6.saddr)); 11278 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11279 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 11280 #else 11281 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11282 #endif 11283 break; 11284 } 11285 case bpf_ctx_range_till(struct bpf_sk_lookup, 11286 local_ip6[0], local_ip6[3]): { 11287 #if IS_ENABLED(CONFIG_IPV6) 11288 int off = si->off; 11289 11290 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]); 11291 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 11292 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11293 offsetof(struct bpf_sk_lookup_kern, v6.daddr)); 11294 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11295 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 11296 #else 11297 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11298 #endif 11299 break; 11300 } 11301 case offsetof(struct bpf_sk_lookup, remote_port): 11302 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11303 bpf_target_off(struct bpf_sk_lookup_kern, 11304 sport, 2, target_size)); 11305 break; 11306 11307 case offsetofend(struct bpf_sk_lookup, remote_port): 11308 *target_size = 2; 11309 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11310 break; 11311 11312 case offsetof(struct bpf_sk_lookup, local_port): 11313 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11314 bpf_target_off(struct bpf_sk_lookup_kern, 11315 dport, 2, target_size)); 11316 break; 11317 11318 case offsetof(struct bpf_sk_lookup, ingress_ifindex): 11319 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11320 bpf_target_off(struct bpf_sk_lookup_kern, 11321 ingress_ifindex, 4, target_size)); 11322 break; 11323 } 11324 11325 return insn - insn_buf; 11326 } 11327 11328 const struct bpf_prog_ops sk_lookup_prog_ops = { 11329 .test_run = bpf_prog_test_run_sk_lookup, 11330 }; 11331 11332 const struct bpf_verifier_ops sk_lookup_verifier_ops = { 11333 .get_func_proto = sk_lookup_func_proto, 11334 .is_valid_access = sk_lookup_is_valid_access, 11335 .convert_ctx_access = sk_lookup_convert_ctx_access, 11336 }; 11337 11338 #endif /* CONFIG_INET */ 11339 11340 DEFINE_BPF_DISPATCHER(xdp) 11341 11342 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog) 11343 { 11344 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog); 11345 } 11346 11347 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE) 11348 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type) 11349 BTF_SOCK_TYPE_xxx 11350 #undef BTF_SOCK_TYPE 11351 11352 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk) 11353 { 11354 /* tcp6_sock type is not generated in dwarf and hence btf, 11355 * trigger an explicit type generation here. 11356 */ 11357 BTF_TYPE_EMIT(struct tcp6_sock); 11358 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && 11359 sk->sk_family == AF_INET6) 11360 return (unsigned long)sk; 11361 11362 return (unsigned long)NULL; 11363 } 11364 11365 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = { 11366 .func = bpf_skc_to_tcp6_sock, 11367 .gpl_only = false, 11368 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11369 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11370 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6], 11371 }; 11372 11373 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk) 11374 { 11375 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 11376 return (unsigned long)sk; 11377 11378 return (unsigned long)NULL; 11379 } 11380 11381 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = { 11382 .func = bpf_skc_to_tcp_sock, 11383 .gpl_only = false, 11384 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11385 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11386 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP], 11387 }; 11388 11389 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk) 11390 { 11391 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not 11392 * generated if CONFIG_INET=n. Trigger an explicit generation here. 11393 */ 11394 BTF_TYPE_EMIT(struct inet_timewait_sock); 11395 BTF_TYPE_EMIT(struct tcp_timewait_sock); 11396 11397 #ifdef CONFIG_INET 11398 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT) 11399 return (unsigned long)sk; 11400 #endif 11401 11402 #if IS_BUILTIN(CONFIG_IPV6) 11403 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT) 11404 return (unsigned long)sk; 11405 #endif 11406 11407 return (unsigned long)NULL; 11408 } 11409 11410 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = { 11411 .func = bpf_skc_to_tcp_timewait_sock, 11412 .gpl_only = false, 11413 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11414 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11415 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW], 11416 }; 11417 11418 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk) 11419 { 11420 #ifdef CONFIG_INET 11421 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV) 11422 return (unsigned long)sk; 11423 #endif 11424 11425 #if IS_BUILTIN(CONFIG_IPV6) 11426 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV) 11427 return (unsigned long)sk; 11428 #endif 11429 11430 return (unsigned long)NULL; 11431 } 11432 11433 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = { 11434 .func = bpf_skc_to_tcp_request_sock, 11435 .gpl_only = false, 11436 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11437 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11438 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ], 11439 }; 11440 11441 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk) 11442 { 11443 /* udp6_sock type is not generated in dwarf and hence btf, 11444 * trigger an explicit type generation here. 11445 */ 11446 BTF_TYPE_EMIT(struct udp6_sock); 11447 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP && 11448 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6) 11449 return (unsigned long)sk; 11450 11451 return (unsigned long)NULL; 11452 } 11453 11454 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = { 11455 .func = bpf_skc_to_udp6_sock, 11456 .gpl_only = false, 11457 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11458 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11459 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6], 11460 }; 11461 11462 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk) 11463 { 11464 /* unix_sock type is not generated in dwarf and hence btf, 11465 * trigger an explicit type generation here. 11466 */ 11467 BTF_TYPE_EMIT(struct unix_sock); 11468 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX) 11469 return (unsigned long)sk; 11470 11471 return (unsigned long)NULL; 11472 } 11473 11474 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = { 11475 .func = bpf_skc_to_unix_sock, 11476 .gpl_only = false, 11477 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11478 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11479 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX], 11480 }; 11481 11482 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk) 11483 { 11484 BTF_TYPE_EMIT(struct mptcp_sock); 11485 return (unsigned long)bpf_mptcp_sock_from_subflow(sk); 11486 } 11487 11488 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = { 11489 .func = bpf_skc_to_mptcp_sock, 11490 .gpl_only = false, 11491 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11492 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 11493 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP], 11494 }; 11495 11496 BPF_CALL_1(bpf_sock_from_file, struct file *, file) 11497 { 11498 return (unsigned long)sock_from_file(file); 11499 } 11500 11501 BTF_ID_LIST(bpf_sock_from_file_btf_ids) 11502 BTF_ID(struct, socket) 11503 BTF_ID(struct, file) 11504 11505 const struct bpf_func_proto bpf_sock_from_file_proto = { 11506 .func = bpf_sock_from_file, 11507 .gpl_only = false, 11508 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11509 .ret_btf_id = &bpf_sock_from_file_btf_ids[0], 11510 .arg1_type = ARG_PTR_TO_BTF_ID, 11511 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1], 11512 }; 11513 11514 static const struct bpf_func_proto * 11515 bpf_sk_base_func_proto(enum bpf_func_id func_id) 11516 { 11517 const struct bpf_func_proto *func; 11518 11519 switch (func_id) { 11520 case BPF_FUNC_skc_to_tcp6_sock: 11521 func = &bpf_skc_to_tcp6_sock_proto; 11522 break; 11523 case BPF_FUNC_skc_to_tcp_sock: 11524 func = &bpf_skc_to_tcp_sock_proto; 11525 break; 11526 case BPF_FUNC_skc_to_tcp_timewait_sock: 11527 func = &bpf_skc_to_tcp_timewait_sock_proto; 11528 break; 11529 case BPF_FUNC_skc_to_tcp_request_sock: 11530 func = &bpf_skc_to_tcp_request_sock_proto; 11531 break; 11532 case BPF_FUNC_skc_to_udp6_sock: 11533 func = &bpf_skc_to_udp6_sock_proto; 11534 break; 11535 case BPF_FUNC_skc_to_unix_sock: 11536 func = &bpf_skc_to_unix_sock_proto; 11537 break; 11538 case BPF_FUNC_skc_to_mptcp_sock: 11539 func = &bpf_skc_to_mptcp_sock_proto; 11540 break; 11541 case BPF_FUNC_ktime_get_coarse_ns: 11542 return &bpf_ktime_get_coarse_ns_proto; 11543 default: 11544 return bpf_base_func_proto(func_id); 11545 } 11546 11547 if (!perfmon_capable()) 11548 return NULL; 11549 11550 return func; 11551 } 11552