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 82 static const struct bpf_func_proto * 83 bpf_sk_base_func_proto(enum bpf_func_id func_id); 84 85 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len) 86 { 87 if (in_compat_syscall()) { 88 struct compat_sock_fprog f32; 89 90 if (len != sizeof(f32)) 91 return -EINVAL; 92 if (copy_from_sockptr(&f32, src, sizeof(f32))) 93 return -EFAULT; 94 memset(dst, 0, sizeof(*dst)); 95 dst->len = f32.len; 96 dst->filter = compat_ptr(f32.filter); 97 } else { 98 if (len != sizeof(*dst)) 99 return -EINVAL; 100 if (copy_from_sockptr(dst, src, sizeof(*dst))) 101 return -EFAULT; 102 } 103 104 return 0; 105 } 106 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user); 107 108 /** 109 * sk_filter_trim_cap - run a packet through a socket filter 110 * @sk: sock associated with &sk_buff 111 * @skb: buffer to filter 112 * @cap: limit on how short the eBPF program may trim the packet 113 * 114 * Run the eBPF program and then cut skb->data to correct size returned by 115 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller 116 * than pkt_len we keep whole skb->data. This is the socket level 117 * wrapper to bpf_prog_run. It returns 0 if the packet should 118 * be accepted or -EPERM if the packet should be tossed. 119 * 120 */ 121 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap) 122 { 123 int err; 124 struct sk_filter *filter; 125 126 /* 127 * If the skb was allocated from pfmemalloc reserves, only 128 * allow SOCK_MEMALLOC sockets to use it as this socket is 129 * helping free memory 130 */ 131 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) { 132 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP); 133 return -ENOMEM; 134 } 135 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb); 136 if (err) 137 return err; 138 139 err = security_sock_rcv_skb(sk, skb); 140 if (err) 141 return err; 142 143 rcu_read_lock(); 144 filter = rcu_dereference(sk->sk_filter); 145 if (filter) { 146 struct sock *save_sk = skb->sk; 147 unsigned int pkt_len; 148 149 skb->sk = sk; 150 pkt_len = bpf_prog_run_save_cb(filter->prog, skb); 151 skb->sk = save_sk; 152 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM; 153 } 154 rcu_read_unlock(); 155 156 return err; 157 } 158 EXPORT_SYMBOL(sk_filter_trim_cap); 159 160 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb) 161 { 162 return skb_get_poff(skb); 163 } 164 165 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x) 166 { 167 struct nlattr *nla; 168 169 if (skb_is_nonlinear(skb)) 170 return 0; 171 172 if (skb->len < sizeof(struct nlattr)) 173 return 0; 174 175 if (a > skb->len - sizeof(struct nlattr)) 176 return 0; 177 178 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x); 179 if (nla) 180 return (void *) nla - (void *) skb->data; 181 182 return 0; 183 } 184 185 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x) 186 { 187 struct nlattr *nla; 188 189 if (skb_is_nonlinear(skb)) 190 return 0; 191 192 if (skb->len < sizeof(struct nlattr)) 193 return 0; 194 195 if (a > skb->len - sizeof(struct nlattr)) 196 return 0; 197 198 nla = (struct nlattr *) &skb->data[a]; 199 if (nla->nla_len > skb->len - a) 200 return 0; 201 202 nla = nla_find_nested(nla, x); 203 if (nla) 204 return (void *) nla - (void *) skb->data; 205 206 return 0; 207 } 208 209 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *, 210 data, int, headlen, int, offset) 211 { 212 u8 tmp, *ptr; 213 const int len = sizeof(tmp); 214 215 if (offset >= 0) { 216 if (headlen - offset >= len) 217 return *(u8 *)(data + offset); 218 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 219 return tmp; 220 } else { 221 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 222 if (likely(ptr)) 223 return *(u8 *)ptr; 224 } 225 226 return -EFAULT; 227 } 228 229 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb, 230 int, offset) 231 { 232 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len, 233 offset); 234 } 235 236 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *, 237 data, int, headlen, int, offset) 238 { 239 u16 tmp, *ptr; 240 const int len = sizeof(tmp); 241 242 if (offset >= 0) { 243 if (headlen - offset >= len) 244 return get_unaligned_be16(data + offset); 245 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 246 return be16_to_cpu(tmp); 247 } else { 248 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 249 if (likely(ptr)) 250 return get_unaligned_be16(ptr); 251 } 252 253 return -EFAULT; 254 } 255 256 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb, 257 int, offset) 258 { 259 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len, 260 offset); 261 } 262 263 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *, 264 data, int, headlen, int, offset) 265 { 266 u32 tmp, *ptr; 267 const int len = sizeof(tmp); 268 269 if (likely(offset >= 0)) { 270 if (headlen - offset >= len) 271 return get_unaligned_be32(data + offset); 272 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 273 return be32_to_cpu(tmp); 274 } else { 275 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 276 if (likely(ptr)) 277 return get_unaligned_be32(ptr); 278 } 279 280 return -EFAULT; 281 } 282 283 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb, 284 int, offset) 285 { 286 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len, 287 offset); 288 } 289 290 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg, 291 struct bpf_insn *insn_buf) 292 { 293 struct bpf_insn *insn = insn_buf; 294 295 switch (skb_field) { 296 case SKF_AD_MARK: 297 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4); 298 299 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, 300 offsetof(struct sk_buff, mark)); 301 break; 302 303 case SKF_AD_PKTTYPE: 304 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET()); 305 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX); 306 #ifdef __BIG_ENDIAN_BITFIELD 307 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5); 308 #endif 309 break; 310 311 case SKF_AD_QUEUE: 312 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2); 313 314 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 315 offsetof(struct sk_buff, queue_mapping)); 316 break; 317 318 case SKF_AD_VLAN_TAG: 319 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2); 320 321 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */ 322 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 323 offsetof(struct sk_buff, vlan_tci)); 324 break; 325 case SKF_AD_VLAN_TAG_PRESENT: 326 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET()); 327 if (PKT_VLAN_PRESENT_BIT) 328 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT); 329 if (PKT_VLAN_PRESENT_BIT < 7) 330 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1); 331 break; 332 } 333 334 return insn - insn_buf; 335 } 336 337 static bool convert_bpf_extensions(struct sock_filter *fp, 338 struct bpf_insn **insnp) 339 { 340 struct bpf_insn *insn = *insnp; 341 u32 cnt; 342 343 switch (fp->k) { 344 case SKF_AD_OFF + SKF_AD_PROTOCOL: 345 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2); 346 347 /* A = *(u16 *) (CTX + offsetof(protocol)) */ 348 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 349 offsetof(struct sk_buff, protocol)); 350 /* A = ntohs(A) [emitting a nop or swap16] */ 351 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 352 break; 353 354 case SKF_AD_OFF + SKF_AD_PKTTYPE: 355 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn); 356 insn += cnt - 1; 357 break; 358 359 case SKF_AD_OFF + SKF_AD_IFINDEX: 360 case SKF_AD_OFF + SKF_AD_HATYPE: 361 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4); 362 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2); 363 364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 365 BPF_REG_TMP, BPF_REG_CTX, 366 offsetof(struct sk_buff, dev)); 367 /* if (tmp != 0) goto pc + 1 */ 368 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1); 369 *insn++ = BPF_EXIT_INSN(); 370 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) 371 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP, 372 offsetof(struct net_device, ifindex)); 373 else 374 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP, 375 offsetof(struct net_device, type)); 376 break; 377 378 case SKF_AD_OFF + SKF_AD_MARK: 379 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn); 380 insn += cnt - 1; 381 break; 382 383 case SKF_AD_OFF + SKF_AD_RXHASH: 384 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4); 385 386 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, 387 offsetof(struct sk_buff, hash)); 388 break; 389 390 case SKF_AD_OFF + SKF_AD_QUEUE: 391 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn); 392 insn += cnt - 1; 393 break; 394 395 case SKF_AD_OFF + SKF_AD_VLAN_TAG: 396 cnt = convert_skb_access(SKF_AD_VLAN_TAG, 397 BPF_REG_A, BPF_REG_CTX, insn); 398 insn += cnt - 1; 399 break; 400 401 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT: 402 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT, 403 BPF_REG_A, BPF_REG_CTX, insn); 404 insn += cnt - 1; 405 break; 406 407 case SKF_AD_OFF + SKF_AD_VLAN_TPID: 408 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2); 409 410 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */ 411 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 412 offsetof(struct sk_buff, vlan_proto)); 413 /* A = ntohs(A) [emitting a nop or swap16] */ 414 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 415 break; 416 417 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 418 case SKF_AD_OFF + SKF_AD_NLATTR: 419 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 420 case SKF_AD_OFF + SKF_AD_CPU: 421 case SKF_AD_OFF + SKF_AD_RANDOM: 422 /* arg1 = CTX */ 423 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 424 /* arg2 = A */ 425 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A); 426 /* arg3 = X */ 427 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X); 428 /* Emit call(arg1=CTX, arg2=A, arg3=X) */ 429 switch (fp->k) { 430 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 431 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset); 432 break; 433 case SKF_AD_OFF + SKF_AD_NLATTR: 434 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr); 435 break; 436 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 437 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest); 438 break; 439 case SKF_AD_OFF + SKF_AD_CPU: 440 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id); 441 break; 442 case SKF_AD_OFF + SKF_AD_RANDOM: 443 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32); 444 bpf_user_rnd_init_once(); 445 break; 446 } 447 break; 448 449 case SKF_AD_OFF + SKF_AD_ALU_XOR_X: 450 /* A ^= X */ 451 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X); 452 break; 453 454 default: 455 /* This is just a dummy call to avoid letting the compiler 456 * evict __bpf_call_base() as an optimization. Placed here 457 * where no-one bothers. 458 */ 459 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0); 460 return false; 461 } 462 463 *insnp = insn; 464 return true; 465 } 466 467 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp) 468 { 469 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS); 470 int size = bpf_size_to_bytes(BPF_SIZE(fp->code)); 471 bool endian = BPF_SIZE(fp->code) == BPF_H || 472 BPF_SIZE(fp->code) == BPF_W; 473 bool indirect = BPF_MODE(fp->code) == BPF_IND; 474 const int ip_align = NET_IP_ALIGN; 475 struct bpf_insn *insn = *insnp; 476 int offset = fp->k; 477 478 if (!indirect && 479 ((unaligned_ok && offset >= 0) || 480 (!unaligned_ok && offset >= 0 && 481 offset + ip_align >= 0 && 482 offset + ip_align % size == 0))) { 483 bool ldx_off_ok = offset <= S16_MAX; 484 485 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H); 486 if (offset) 487 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset); 488 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP, 489 size, 2 + endian + (!ldx_off_ok * 2)); 490 if (ldx_off_ok) { 491 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 492 BPF_REG_D, offset); 493 } else { 494 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D); 495 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset); 496 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 497 BPF_REG_TMP, 0); 498 } 499 if (endian) 500 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8); 501 *insn++ = BPF_JMP_A(8); 502 } 503 504 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 505 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D); 506 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H); 507 if (!indirect) { 508 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset); 509 } else { 510 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X); 511 if (fp->k) 512 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset); 513 } 514 515 switch (BPF_SIZE(fp->code)) { 516 case BPF_B: 517 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8); 518 break; 519 case BPF_H: 520 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16); 521 break; 522 case BPF_W: 523 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32); 524 break; 525 default: 526 return false; 527 } 528 529 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2); 530 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 531 *insn = BPF_EXIT_INSN(); 532 533 *insnp = insn; 534 return true; 535 } 536 537 /** 538 * bpf_convert_filter - convert filter program 539 * @prog: the user passed filter program 540 * @len: the length of the user passed filter program 541 * @new_prog: allocated 'struct bpf_prog' or NULL 542 * @new_len: pointer to store length of converted program 543 * @seen_ld_abs: bool whether we've seen ld_abs/ind 544 * 545 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn' 546 * style extended BPF (eBPF). 547 * Conversion workflow: 548 * 549 * 1) First pass for calculating the new program length: 550 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs) 551 * 552 * 2) 2nd pass to remap in two passes: 1st pass finds new 553 * jump offsets, 2nd pass remapping: 554 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs) 555 */ 556 static int bpf_convert_filter(struct sock_filter *prog, int len, 557 struct bpf_prog *new_prog, int *new_len, 558 bool *seen_ld_abs) 559 { 560 int new_flen = 0, pass = 0, target, i, stack_off; 561 struct bpf_insn *new_insn, *first_insn = NULL; 562 struct sock_filter *fp; 563 int *addrs = NULL; 564 u8 bpf_src; 565 566 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK); 567 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); 568 569 if (len <= 0 || len > BPF_MAXINSNS) 570 return -EINVAL; 571 572 if (new_prog) { 573 first_insn = new_prog->insnsi; 574 addrs = kcalloc(len, sizeof(*addrs), 575 GFP_KERNEL | __GFP_NOWARN); 576 if (!addrs) 577 return -ENOMEM; 578 } 579 580 do_pass: 581 new_insn = first_insn; 582 fp = prog; 583 584 /* Classic BPF related prologue emission. */ 585 if (new_prog) { 586 /* Classic BPF expects A and X to be reset first. These need 587 * to be guaranteed to be the first two instructions. 588 */ 589 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 590 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X); 591 592 /* All programs must keep CTX in callee saved BPF_REG_CTX. 593 * In eBPF case it's done by the compiler, here we need to 594 * do this ourself. Initial CTX is present in BPF_REG_ARG1. 595 */ 596 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1); 597 if (*seen_ld_abs) { 598 /* For packet access in classic BPF, cache skb->data 599 * in callee-saved BPF R8 and skb->len - skb->data_len 600 * (headlen) in BPF R9. Since classic BPF is read-only 601 * on CTX, we only need to cache it once. 602 */ 603 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 604 BPF_REG_D, BPF_REG_CTX, 605 offsetof(struct sk_buff, data)); 606 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX, 607 offsetof(struct sk_buff, len)); 608 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX, 609 offsetof(struct sk_buff, data_len)); 610 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP); 611 } 612 } else { 613 new_insn += 3; 614 } 615 616 for (i = 0; i < len; fp++, i++) { 617 struct bpf_insn tmp_insns[32] = { }; 618 struct bpf_insn *insn = tmp_insns; 619 620 if (addrs) 621 addrs[i] = new_insn - first_insn; 622 623 switch (fp->code) { 624 /* All arithmetic insns and skb loads map as-is. */ 625 case BPF_ALU | BPF_ADD | BPF_X: 626 case BPF_ALU | BPF_ADD | BPF_K: 627 case BPF_ALU | BPF_SUB | BPF_X: 628 case BPF_ALU | BPF_SUB | BPF_K: 629 case BPF_ALU | BPF_AND | BPF_X: 630 case BPF_ALU | BPF_AND | BPF_K: 631 case BPF_ALU | BPF_OR | BPF_X: 632 case BPF_ALU | BPF_OR | BPF_K: 633 case BPF_ALU | BPF_LSH | BPF_X: 634 case BPF_ALU | BPF_LSH | BPF_K: 635 case BPF_ALU | BPF_RSH | BPF_X: 636 case BPF_ALU | BPF_RSH | BPF_K: 637 case BPF_ALU | BPF_XOR | BPF_X: 638 case BPF_ALU | BPF_XOR | BPF_K: 639 case BPF_ALU | BPF_MUL | BPF_X: 640 case BPF_ALU | BPF_MUL | BPF_K: 641 case BPF_ALU | BPF_DIV | BPF_X: 642 case BPF_ALU | BPF_DIV | BPF_K: 643 case BPF_ALU | BPF_MOD | BPF_X: 644 case BPF_ALU | BPF_MOD | BPF_K: 645 case BPF_ALU | BPF_NEG: 646 case BPF_LD | BPF_ABS | BPF_W: 647 case BPF_LD | BPF_ABS | BPF_H: 648 case BPF_LD | BPF_ABS | BPF_B: 649 case BPF_LD | BPF_IND | BPF_W: 650 case BPF_LD | BPF_IND | BPF_H: 651 case BPF_LD | BPF_IND | BPF_B: 652 /* Check for overloaded BPF extension and 653 * directly convert it if found, otherwise 654 * just move on with mapping. 655 */ 656 if (BPF_CLASS(fp->code) == BPF_LD && 657 BPF_MODE(fp->code) == BPF_ABS && 658 convert_bpf_extensions(fp, &insn)) 659 break; 660 if (BPF_CLASS(fp->code) == BPF_LD && 661 convert_bpf_ld_abs(fp, &insn)) { 662 *seen_ld_abs = true; 663 break; 664 } 665 666 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) || 667 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) { 668 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X); 669 /* Error with exception code on div/mod by 0. 670 * For cBPF programs, this was always return 0. 671 */ 672 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2); 673 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 674 *insn++ = BPF_EXIT_INSN(); 675 } 676 677 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k); 678 break; 679 680 /* Jump transformation cannot use BPF block macros 681 * everywhere as offset calculation and target updates 682 * require a bit more work than the rest, i.e. jump 683 * opcodes map as-is, but offsets need adjustment. 684 */ 685 686 #define BPF_EMIT_JMP \ 687 do { \ 688 const s32 off_min = S16_MIN, off_max = S16_MAX; \ 689 s32 off; \ 690 \ 691 if (target >= len || target < 0) \ 692 goto err; \ 693 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \ 694 /* Adjust pc relative offset for 2nd or 3rd insn. */ \ 695 off -= insn - tmp_insns; \ 696 /* Reject anything not fitting into insn->off. */ \ 697 if (off < off_min || off > off_max) \ 698 goto err; \ 699 insn->off = off; \ 700 } while (0) 701 702 case BPF_JMP | BPF_JA: 703 target = i + fp->k + 1; 704 insn->code = fp->code; 705 BPF_EMIT_JMP; 706 break; 707 708 case BPF_JMP | BPF_JEQ | BPF_K: 709 case BPF_JMP | BPF_JEQ | BPF_X: 710 case BPF_JMP | BPF_JSET | BPF_K: 711 case BPF_JMP | BPF_JSET | BPF_X: 712 case BPF_JMP | BPF_JGT | BPF_K: 713 case BPF_JMP | BPF_JGT | BPF_X: 714 case BPF_JMP | BPF_JGE | BPF_K: 715 case BPF_JMP | BPF_JGE | BPF_X: 716 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) { 717 /* BPF immediates are signed, zero extend 718 * immediate into tmp register and use it 719 * in compare insn. 720 */ 721 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k); 722 723 insn->dst_reg = BPF_REG_A; 724 insn->src_reg = BPF_REG_TMP; 725 bpf_src = BPF_X; 726 } else { 727 insn->dst_reg = BPF_REG_A; 728 insn->imm = fp->k; 729 bpf_src = BPF_SRC(fp->code); 730 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0; 731 } 732 733 /* Common case where 'jump_false' is next insn. */ 734 if (fp->jf == 0) { 735 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 736 target = i + fp->jt + 1; 737 BPF_EMIT_JMP; 738 break; 739 } 740 741 /* Convert some jumps when 'jump_true' is next insn. */ 742 if (fp->jt == 0) { 743 switch (BPF_OP(fp->code)) { 744 case BPF_JEQ: 745 insn->code = BPF_JMP | BPF_JNE | bpf_src; 746 break; 747 case BPF_JGT: 748 insn->code = BPF_JMP | BPF_JLE | bpf_src; 749 break; 750 case BPF_JGE: 751 insn->code = BPF_JMP | BPF_JLT | bpf_src; 752 break; 753 default: 754 goto jmp_rest; 755 } 756 757 target = i + fp->jf + 1; 758 BPF_EMIT_JMP; 759 break; 760 } 761 jmp_rest: 762 /* Other jumps are mapped into two insns: Jxx and JA. */ 763 target = i + fp->jt + 1; 764 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 765 BPF_EMIT_JMP; 766 insn++; 767 768 insn->code = BPF_JMP | BPF_JA; 769 target = i + fp->jf + 1; 770 BPF_EMIT_JMP; 771 break; 772 773 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */ 774 case BPF_LDX | BPF_MSH | BPF_B: { 775 struct sock_filter tmp = { 776 .code = BPF_LD | BPF_ABS | BPF_B, 777 .k = fp->k, 778 }; 779 780 *seen_ld_abs = true; 781 782 /* X = A */ 783 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 784 /* A = BPF_R0 = *(u8 *) (skb->data + K) */ 785 convert_bpf_ld_abs(&tmp, &insn); 786 insn++; 787 /* A &= 0xf */ 788 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf); 789 /* A <<= 2 */ 790 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2); 791 /* tmp = X */ 792 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X); 793 /* X = A */ 794 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 795 /* A = tmp */ 796 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP); 797 break; 798 } 799 /* RET_K is remaped into 2 insns. RET_A case doesn't need an 800 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A. 801 */ 802 case BPF_RET | BPF_A: 803 case BPF_RET | BPF_K: 804 if (BPF_RVAL(fp->code) == BPF_K) 805 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0, 806 0, fp->k); 807 *insn = BPF_EXIT_INSN(); 808 break; 809 810 /* Store to stack. */ 811 case BPF_ST: 812 case BPF_STX: 813 stack_off = fp->k * 4 + 4; 814 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) == 815 BPF_ST ? BPF_REG_A : BPF_REG_X, 816 -stack_off); 817 /* check_load_and_stores() verifies that classic BPF can 818 * load from stack only after write, so tracking 819 * stack_depth for ST|STX insns is enough 820 */ 821 if (new_prog && new_prog->aux->stack_depth < stack_off) 822 new_prog->aux->stack_depth = stack_off; 823 break; 824 825 /* Load from stack. */ 826 case BPF_LD | BPF_MEM: 827 case BPF_LDX | BPF_MEM: 828 stack_off = fp->k * 4 + 4; 829 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 830 BPF_REG_A : BPF_REG_X, BPF_REG_FP, 831 -stack_off); 832 break; 833 834 /* A = K or X = K */ 835 case BPF_LD | BPF_IMM: 836 case BPF_LDX | BPF_IMM: 837 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ? 838 BPF_REG_A : BPF_REG_X, fp->k); 839 break; 840 841 /* X = A */ 842 case BPF_MISC | BPF_TAX: 843 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 844 break; 845 846 /* A = X */ 847 case BPF_MISC | BPF_TXA: 848 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X); 849 break; 850 851 /* A = skb->len or X = skb->len */ 852 case BPF_LD | BPF_W | BPF_LEN: 853 case BPF_LDX | BPF_W | BPF_LEN: 854 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 855 BPF_REG_A : BPF_REG_X, BPF_REG_CTX, 856 offsetof(struct sk_buff, len)); 857 break; 858 859 /* Access seccomp_data fields. */ 860 case BPF_LDX | BPF_ABS | BPF_W: 861 /* A = *(u32 *) (ctx + K) */ 862 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k); 863 break; 864 865 /* Unknown instruction. */ 866 default: 867 goto err; 868 } 869 870 insn++; 871 if (new_prog) 872 memcpy(new_insn, tmp_insns, 873 sizeof(*insn) * (insn - tmp_insns)); 874 new_insn += insn - tmp_insns; 875 } 876 877 if (!new_prog) { 878 /* Only calculating new length. */ 879 *new_len = new_insn - first_insn; 880 if (*seen_ld_abs) 881 *new_len += 4; /* Prologue bits. */ 882 return 0; 883 } 884 885 pass++; 886 if (new_flen != new_insn - first_insn) { 887 new_flen = new_insn - first_insn; 888 if (pass > 2) 889 goto err; 890 goto do_pass; 891 } 892 893 kfree(addrs); 894 BUG_ON(*new_len != new_flen); 895 return 0; 896 err: 897 kfree(addrs); 898 return -EINVAL; 899 } 900 901 /* Security: 902 * 903 * As we dont want to clear mem[] array for each packet going through 904 * __bpf_prog_run(), we check that filter loaded by user never try to read 905 * a cell if not previously written, and we check all branches to be sure 906 * a malicious user doesn't try to abuse us. 907 */ 908 static int check_load_and_stores(const struct sock_filter *filter, int flen) 909 { 910 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */ 911 int pc, ret = 0; 912 913 BUILD_BUG_ON(BPF_MEMWORDS > 16); 914 915 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL); 916 if (!masks) 917 return -ENOMEM; 918 919 memset(masks, 0xff, flen * sizeof(*masks)); 920 921 for (pc = 0; pc < flen; pc++) { 922 memvalid &= masks[pc]; 923 924 switch (filter[pc].code) { 925 case BPF_ST: 926 case BPF_STX: 927 memvalid |= (1 << filter[pc].k); 928 break; 929 case BPF_LD | BPF_MEM: 930 case BPF_LDX | BPF_MEM: 931 if (!(memvalid & (1 << filter[pc].k))) { 932 ret = -EINVAL; 933 goto error; 934 } 935 break; 936 case BPF_JMP | BPF_JA: 937 /* A jump must set masks on target */ 938 masks[pc + 1 + filter[pc].k] &= memvalid; 939 memvalid = ~0; 940 break; 941 case BPF_JMP | BPF_JEQ | BPF_K: 942 case BPF_JMP | BPF_JEQ | BPF_X: 943 case BPF_JMP | BPF_JGE | BPF_K: 944 case BPF_JMP | BPF_JGE | BPF_X: 945 case BPF_JMP | BPF_JGT | BPF_K: 946 case BPF_JMP | BPF_JGT | BPF_X: 947 case BPF_JMP | BPF_JSET | BPF_K: 948 case BPF_JMP | BPF_JSET | BPF_X: 949 /* A jump must set masks on targets */ 950 masks[pc + 1 + filter[pc].jt] &= memvalid; 951 masks[pc + 1 + filter[pc].jf] &= memvalid; 952 memvalid = ~0; 953 break; 954 } 955 } 956 error: 957 kfree(masks); 958 return ret; 959 } 960 961 static bool chk_code_allowed(u16 code_to_probe) 962 { 963 static const bool codes[] = { 964 /* 32 bit ALU operations */ 965 [BPF_ALU | BPF_ADD | BPF_K] = true, 966 [BPF_ALU | BPF_ADD | BPF_X] = true, 967 [BPF_ALU | BPF_SUB | BPF_K] = true, 968 [BPF_ALU | BPF_SUB | BPF_X] = true, 969 [BPF_ALU | BPF_MUL | BPF_K] = true, 970 [BPF_ALU | BPF_MUL | BPF_X] = true, 971 [BPF_ALU | BPF_DIV | BPF_K] = true, 972 [BPF_ALU | BPF_DIV | BPF_X] = true, 973 [BPF_ALU | BPF_MOD | BPF_K] = true, 974 [BPF_ALU | BPF_MOD | BPF_X] = true, 975 [BPF_ALU | BPF_AND | BPF_K] = true, 976 [BPF_ALU | BPF_AND | BPF_X] = true, 977 [BPF_ALU | BPF_OR | BPF_K] = true, 978 [BPF_ALU | BPF_OR | BPF_X] = true, 979 [BPF_ALU | BPF_XOR | BPF_K] = true, 980 [BPF_ALU | BPF_XOR | BPF_X] = true, 981 [BPF_ALU | BPF_LSH | BPF_K] = true, 982 [BPF_ALU | BPF_LSH | BPF_X] = true, 983 [BPF_ALU | BPF_RSH | BPF_K] = true, 984 [BPF_ALU | BPF_RSH | BPF_X] = true, 985 [BPF_ALU | BPF_NEG] = true, 986 /* Load instructions */ 987 [BPF_LD | BPF_W | BPF_ABS] = true, 988 [BPF_LD | BPF_H | BPF_ABS] = true, 989 [BPF_LD | BPF_B | BPF_ABS] = true, 990 [BPF_LD | BPF_W | BPF_LEN] = true, 991 [BPF_LD | BPF_W | BPF_IND] = true, 992 [BPF_LD | BPF_H | BPF_IND] = true, 993 [BPF_LD | BPF_B | BPF_IND] = true, 994 [BPF_LD | BPF_IMM] = true, 995 [BPF_LD | BPF_MEM] = true, 996 [BPF_LDX | BPF_W | BPF_LEN] = true, 997 [BPF_LDX | BPF_B | BPF_MSH] = true, 998 [BPF_LDX | BPF_IMM] = true, 999 [BPF_LDX | BPF_MEM] = true, 1000 /* Store instructions */ 1001 [BPF_ST] = true, 1002 [BPF_STX] = true, 1003 /* Misc instructions */ 1004 [BPF_MISC | BPF_TAX] = true, 1005 [BPF_MISC | BPF_TXA] = true, 1006 /* Return instructions */ 1007 [BPF_RET | BPF_K] = true, 1008 [BPF_RET | BPF_A] = true, 1009 /* Jump instructions */ 1010 [BPF_JMP | BPF_JA] = true, 1011 [BPF_JMP | BPF_JEQ | BPF_K] = true, 1012 [BPF_JMP | BPF_JEQ | BPF_X] = true, 1013 [BPF_JMP | BPF_JGE | BPF_K] = true, 1014 [BPF_JMP | BPF_JGE | BPF_X] = true, 1015 [BPF_JMP | BPF_JGT | BPF_K] = true, 1016 [BPF_JMP | BPF_JGT | BPF_X] = true, 1017 [BPF_JMP | BPF_JSET | BPF_K] = true, 1018 [BPF_JMP | BPF_JSET | BPF_X] = true, 1019 }; 1020 1021 if (code_to_probe >= ARRAY_SIZE(codes)) 1022 return false; 1023 1024 return codes[code_to_probe]; 1025 } 1026 1027 static bool bpf_check_basics_ok(const struct sock_filter *filter, 1028 unsigned int flen) 1029 { 1030 if (filter == NULL) 1031 return false; 1032 if (flen == 0 || flen > BPF_MAXINSNS) 1033 return false; 1034 1035 return true; 1036 } 1037 1038 /** 1039 * bpf_check_classic - verify socket filter code 1040 * @filter: filter to verify 1041 * @flen: length of filter 1042 * 1043 * Check the user's filter code. If we let some ugly 1044 * filter code slip through kaboom! The filter must contain 1045 * no references or jumps that are out of range, no illegal 1046 * instructions, and must end with a RET instruction. 1047 * 1048 * All jumps are forward as they are not signed. 1049 * 1050 * Returns 0 if the rule set is legal or -EINVAL if not. 1051 */ 1052 static int bpf_check_classic(const struct sock_filter *filter, 1053 unsigned int flen) 1054 { 1055 bool anc_found; 1056 int pc; 1057 1058 /* Check the filter code now */ 1059 for (pc = 0; pc < flen; pc++) { 1060 const struct sock_filter *ftest = &filter[pc]; 1061 1062 /* May we actually operate on this code? */ 1063 if (!chk_code_allowed(ftest->code)) 1064 return -EINVAL; 1065 1066 /* Some instructions need special checks */ 1067 switch (ftest->code) { 1068 case BPF_ALU | BPF_DIV | BPF_K: 1069 case BPF_ALU | BPF_MOD | BPF_K: 1070 /* Check for division by zero */ 1071 if (ftest->k == 0) 1072 return -EINVAL; 1073 break; 1074 case BPF_ALU | BPF_LSH | BPF_K: 1075 case BPF_ALU | BPF_RSH | BPF_K: 1076 if (ftest->k >= 32) 1077 return -EINVAL; 1078 break; 1079 case BPF_LD | BPF_MEM: 1080 case BPF_LDX | BPF_MEM: 1081 case BPF_ST: 1082 case BPF_STX: 1083 /* Check for invalid memory addresses */ 1084 if (ftest->k >= BPF_MEMWORDS) 1085 return -EINVAL; 1086 break; 1087 case BPF_JMP | BPF_JA: 1088 /* Note, the large ftest->k might cause loops. 1089 * Compare this with conditional jumps below, 1090 * where offsets are limited. --ANK (981016) 1091 */ 1092 if (ftest->k >= (unsigned int)(flen - pc - 1)) 1093 return -EINVAL; 1094 break; 1095 case BPF_JMP | BPF_JEQ | BPF_K: 1096 case BPF_JMP | BPF_JEQ | BPF_X: 1097 case BPF_JMP | BPF_JGE | BPF_K: 1098 case BPF_JMP | BPF_JGE | BPF_X: 1099 case BPF_JMP | BPF_JGT | BPF_K: 1100 case BPF_JMP | BPF_JGT | BPF_X: 1101 case BPF_JMP | BPF_JSET | BPF_K: 1102 case BPF_JMP | BPF_JSET | BPF_X: 1103 /* Both conditionals must be safe */ 1104 if (pc + ftest->jt + 1 >= flen || 1105 pc + ftest->jf + 1 >= flen) 1106 return -EINVAL; 1107 break; 1108 case BPF_LD | BPF_W | BPF_ABS: 1109 case BPF_LD | BPF_H | BPF_ABS: 1110 case BPF_LD | BPF_B | BPF_ABS: 1111 anc_found = false; 1112 if (bpf_anc_helper(ftest) & BPF_ANC) 1113 anc_found = true; 1114 /* Ancillary operation unknown or unsupported */ 1115 if (anc_found == false && ftest->k >= SKF_AD_OFF) 1116 return -EINVAL; 1117 } 1118 } 1119 1120 /* Last instruction must be a RET code */ 1121 switch (filter[flen - 1].code) { 1122 case BPF_RET | BPF_K: 1123 case BPF_RET | BPF_A: 1124 return check_load_and_stores(filter, flen); 1125 } 1126 1127 return -EINVAL; 1128 } 1129 1130 static int bpf_prog_store_orig_filter(struct bpf_prog *fp, 1131 const struct sock_fprog *fprog) 1132 { 1133 unsigned int fsize = bpf_classic_proglen(fprog); 1134 struct sock_fprog_kern *fkprog; 1135 1136 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL); 1137 if (!fp->orig_prog) 1138 return -ENOMEM; 1139 1140 fkprog = fp->orig_prog; 1141 fkprog->len = fprog->len; 1142 1143 fkprog->filter = kmemdup(fp->insns, fsize, 1144 GFP_KERNEL | __GFP_NOWARN); 1145 if (!fkprog->filter) { 1146 kfree(fp->orig_prog); 1147 return -ENOMEM; 1148 } 1149 1150 return 0; 1151 } 1152 1153 static void bpf_release_orig_filter(struct bpf_prog *fp) 1154 { 1155 struct sock_fprog_kern *fprog = fp->orig_prog; 1156 1157 if (fprog) { 1158 kfree(fprog->filter); 1159 kfree(fprog); 1160 } 1161 } 1162 1163 static void __bpf_prog_release(struct bpf_prog *prog) 1164 { 1165 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) { 1166 bpf_prog_put(prog); 1167 } else { 1168 bpf_release_orig_filter(prog); 1169 bpf_prog_free(prog); 1170 } 1171 } 1172 1173 static void __sk_filter_release(struct sk_filter *fp) 1174 { 1175 __bpf_prog_release(fp->prog); 1176 kfree(fp); 1177 } 1178 1179 /** 1180 * sk_filter_release_rcu - Release a socket filter by rcu_head 1181 * @rcu: rcu_head that contains the sk_filter to free 1182 */ 1183 static void sk_filter_release_rcu(struct rcu_head *rcu) 1184 { 1185 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); 1186 1187 __sk_filter_release(fp); 1188 } 1189 1190 /** 1191 * sk_filter_release - release a socket filter 1192 * @fp: filter to remove 1193 * 1194 * Remove a filter from a socket and release its resources. 1195 */ 1196 static void sk_filter_release(struct sk_filter *fp) 1197 { 1198 if (refcount_dec_and_test(&fp->refcnt)) 1199 call_rcu(&fp->rcu, sk_filter_release_rcu); 1200 } 1201 1202 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1203 { 1204 u32 filter_size = bpf_prog_size(fp->prog->len); 1205 1206 atomic_sub(filter_size, &sk->sk_omem_alloc); 1207 sk_filter_release(fp); 1208 } 1209 1210 /* try to charge the socket memory if there is space available 1211 * return true on success 1212 */ 1213 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1214 { 1215 u32 filter_size = bpf_prog_size(fp->prog->len); 1216 1217 /* same check as in sock_kmalloc() */ 1218 if (filter_size <= sysctl_optmem_max && 1219 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) { 1220 atomic_add(filter_size, &sk->sk_omem_alloc); 1221 return true; 1222 } 1223 return false; 1224 } 1225 1226 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1227 { 1228 if (!refcount_inc_not_zero(&fp->refcnt)) 1229 return false; 1230 1231 if (!__sk_filter_charge(sk, fp)) { 1232 sk_filter_release(fp); 1233 return false; 1234 } 1235 return true; 1236 } 1237 1238 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp) 1239 { 1240 struct sock_filter *old_prog; 1241 struct bpf_prog *old_fp; 1242 int err, new_len, old_len = fp->len; 1243 bool seen_ld_abs = false; 1244 1245 /* We are free to overwrite insns et al right here as it 1246 * won't be used at this point in time anymore internally 1247 * after the migration to the internal BPF instruction 1248 * representation. 1249 */ 1250 BUILD_BUG_ON(sizeof(struct sock_filter) != 1251 sizeof(struct bpf_insn)); 1252 1253 /* Conversion cannot happen on overlapping memory areas, 1254 * so we need to keep the user BPF around until the 2nd 1255 * pass. At this time, the user BPF is stored in fp->insns. 1256 */ 1257 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter), 1258 GFP_KERNEL | __GFP_NOWARN); 1259 if (!old_prog) { 1260 err = -ENOMEM; 1261 goto out_err; 1262 } 1263 1264 /* 1st pass: calculate the new program length. */ 1265 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len, 1266 &seen_ld_abs); 1267 if (err) 1268 goto out_err_free; 1269 1270 /* Expand fp for appending the new filter representation. */ 1271 old_fp = fp; 1272 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0); 1273 if (!fp) { 1274 /* The old_fp is still around in case we couldn't 1275 * allocate new memory, so uncharge on that one. 1276 */ 1277 fp = old_fp; 1278 err = -ENOMEM; 1279 goto out_err_free; 1280 } 1281 1282 fp->len = new_len; 1283 1284 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */ 1285 err = bpf_convert_filter(old_prog, old_len, fp, &new_len, 1286 &seen_ld_abs); 1287 if (err) 1288 /* 2nd bpf_convert_filter() can fail only if it fails 1289 * to allocate memory, remapping must succeed. Note, 1290 * that at this time old_fp has already been released 1291 * by krealloc(). 1292 */ 1293 goto out_err_free; 1294 1295 fp = bpf_prog_select_runtime(fp, &err); 1296 if (err) 1297 goto out_err_free; 1298 1299 kfree(old_prog); 1300 return fp; 1301 1302 out_err_free: 1303 kfree(old_prog); 1304 out_err: 1305 __bpf_prog_release(fp); 1306 return ERR_PTR(err); 1307 } 1308 1309 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp, 1310 bpf_aux_classic_check_t trans) 1311 { 1312 int err; 1313 1314 fp->bpf_func = NULL; 1315 fp->jited = 0; 1316 1317 err = bpf_check_classic(fp->insns, fp->len); 1318 if (err) { 1319 __bpf_prog_release(fp); 1320 return ERR_PTR(err); 1321 } 1322 1323 /* There might be additional checks and transformations 1324 * needed on classic filters, f.e. in case of seccomp. 1325 */ 1326 if (trans) { 1327 err = trans(fp->insns, fp->len); 1328 if (err) { 1329 __bpf_prog_release(fp); 1330 return ERR_PTR(err); 1331 } 1332 } 1333 1334 /* Probe if we can JIT compile the filter and if so, do 1335 * the compilation of the filter. 1336 */ 1337 bpf_jit_compile(fp); 1338 1339 /* JIT compiler couldn't process this filter, so do the 1340 * internal BPF translation for the optimized interpreter. 1341 */ 1342 if (!fp->jited) 1343 fp = bpf_migrate_filter(fp); 1344 1345 return fp; 1346 } 1347 1348 /** 1349 * bpf_prog_create - create an unattached filter 1350 * @pfp: the unattached filter that is created 1351 * @fprog: the filter program 1352 * 1353 * Create a filter independent of any socket. We first run some 1354 * sanity checks on it to make sure it does not explode on us later. 1355 * If an error occurs or there is insufficient memory for the filter 1356 * a negative errno code is returned. On success the return is zero. 1357 */ 1358 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog) 1359 { 1360 unsigned int fsize = bpf_classic_proglen(fprog); 1361 struct bpf_prog *fp; 1362 1363 /* Make sure new filter is there and in the right amounts. */ 1364 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1365 return -EINVAL; 1366 1367 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1368 if (!fp) 1369 return -ENOMEM; 1370 1371 memcpy(fp->insns, fprog->filter, fsize); 1372 1373 fp->len = fprog->len; 1374 /* Since unattached filters are not copied back to user 1375 * space through sk_get_filter(), we do not need to hold 1376 * a copy here, and can spare us the work. 1377 */ 1378 fp->orig_prog = NULL; 1379 1380 /* bpf_prepare_filter() already takes care of freeing 1381 * memory in case something goes wrong. 1382 */ 1383 fp = bpf_prepare_filter(fp, NULL); 1384 if (IS_ERR(fp)) 1385 return PTR_ERR(fp); 1386 1387 *pfp = fp; 1388 return 0; 1389 } 1390 EXPORT_SYMBOL_GPL(bpf_prog_create); 1391 1392 /** 1393 * bpf_prog_create_from_user - create an unattached filter from user buffer 1394 * @pfp: the unattached filter that is created 1395 * @fprog: the filter program 1396 * @trans: post-classic verifier transformation handler 1397 * @save_orig: save classic BPF program 1398 * 1399 * This function effectively does the same as bpf_prog_create(), only 1400 * that it builds up its insns buffer from user space provided buffer. 1401 * It also allows for passing a bpf_aux_classic_check_t handler. 1402 */ 1403 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 1404 bpf_aux_classic_check_t trans, bool save_orig) 1405 { 1406 unsigned int fsize = bpf_classic_proglen(fprog); 1407 struct bpf_prog *fp; 1408 int err; 1409 1410 /* Make sure new filter is there and in the right amounts. */ 1411 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1412 return -EINVAL; 1413 1414 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1415 if (!fp) 1416 return -ENOMEM; 1417 1418 if (copy_from_user(fp->insns, fprog->filter, fsize)) { 1419 __bpf_prog_free(fp); 1420 return -EFAULT; 1421 } 1422 1423 fp->len = fprog->len; 1424 fp->orig_prog = NULL; 1425 1426 if (save_orig) { 1427 err = bpf_prog_store_orig_filter(fp, fprog); 1428 if (err) { 1429 __bpf_prog_free(fp); 1430 return -ENOMEM; 1431 } 1432 } 1433 1434 /* bpf_prepare_filter() already takes care of freeing 1435 * memory in case something goes wrong. 1436 */ 1437 fp = bpf_prepare_filter(fp, trans); 1438 if (IS_ERR(fp)) 1439 return PTR_ERR(fp); 1440 1441 *pfp = fp; 1442 return 0; 1443 } 1444 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user); 1445 1446 void bpf_prog_destroy(struct bpf_prog *fp) 1447 { 1448 __bpf_prog_release(fp); 1449 } 1450 EXPORT_SYMBOL_GPL(bpf_prog_destroy); 1451 1452 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk) 1453 { 1454 struct sk_filter *fp, *old_fp; 1455 1456 fp = kmalloc(sizeof(*fp), GFP_KERNEL); 1457 if (!fp) 1458 return -ENOMEM; 1459 1460 fp->prog = prog; 1461 1462 if (!__sk_filter_charge(sk, fp)) { 1463 kfree(fp); 1464 return -ENOMEM; 1465 } 1466 refcount_set(&fp->refcnt, 1); 1467 1468 old_fp = rcu_dereference_protected(sk->sk_filter, 1469 lockdep_sock_is_held(sk)); 1470 rcu_assign_pointer(sk->sk_filter, fp); 1471 1472 if (old_fp) 1473 sk_filter_uncharge(sk, old_fp); 1474 1475 return 0; 1476 } 1477 1478 static 1479 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk) 1480 { 1481 unsigned int fsize = bpf_classic_proglen(fprog); 1482 struct bpf_prog *prog; 1483 int err; 1484 1485 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1486 return ERR_PTR(-EPERM); 1487 1488 /* Make sure new filter is there and in the right amounts. */ 1489 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1490 return ERR_PTR(-EINVAL); 1491 1492 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1493 if (!prog) 1494 return ERR_PTR(-ENOMEM); 1495 1496 if (copy_from_user(prog->insns, fprog->filter, fsize)) { 1497 __bpf_prog_free(prog); 1498 return ERR_PTR(-EFAULT); 1499 } 1500 1501 prog->len = fprog->len; 1502 1503 err = bpf_prog_store_orig_filter(prog, fprog); 1504 if (err) { 1505 __bpf_prog_free(prog); 1506 return ERR_PTR(-ENOMEM); 1507 } 1508 1509 /* bpf_prepare_filter() already takes care of freeing 1510 * memory in case something goes wrong. 1511 */ 1512 return bpf_prepare_filter(prog, NULL); 1513 } 1514 1515 /** 1516 * sk_attach_filter - attach a socket filter 1517 * @fprog: the filter program 1518 * @sk: the socket to use 1519 * 1520 * Attach the user's filter code. We first run some sanity checks on 1521 * it to make sure it does not explode on us later. If an error 1522 * occurs or there is insufficient memory for the filter a negative 1523 * errno code is returned. On success the return is zero. 1524 */ 1525 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1526 { 1527 struct bpf_prog *prog = __get_filter(fprog, sk); 1528 int err; 1529 1530 if (IS_ERR(prog)) 1531 return PTR_ERR(prog); 1532 1533 err = __sk_attach_prog(prog, sk); 1534 if (err < 0) { 1535 __bpf_prog_release(prog); 1536 return err; 1537 } 1538 1539 return 0; 1540 } 1541 EXPORT_SYMBOL_GPL(sk_attach_filter); 1542 1543 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1544 { 1545 struct bpf_prog *prog = __get_filter(fprog, sk); 1546 int err; 1547 1548 if (IS_ERR(prog)) 1549 return PTR_ERR(prog); 1550 1551 if (bpf_prog_size(prog->len) > sysctl_optmem_max) 1552 err = -ENOMEM; 1553 else 1554 err = reuseport_attach_prog(sk, prog); 1555 1556 if (err) 1557 __bpf_prog_release(prog); 1558 1559 return err; 1560 } 1561 1562 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk) 1563 { 1564 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1565 return ERR_PTR(-EPERM); 1566 1567 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1568 } 1569 1570 int sk_attach_bpf(u32 ufd, struct sock *sk) 1571 { 1572 struct bpf_prog *prog = __get_bpf(ufd, sk); 1573 int err; 1574 1575 if (IS_ERR(prog)) 1576 return PTR_ERR(prog); 1577 1578 err = __sk_attach_prog(prog, sk); 1579 if (err < 0) { 1580 bpf_prog_put(prog); 1581 return err; 1582 } 1583 1584 return 0; 1585 } 1586 1587 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk) 1588 { 1589 struct bpf_prog *prog; 1590 int err; 1591 1592 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1593 return -EPERM; 1594 1595 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1596 if (PTR_ERR(prog) == -EINVAL) 1597 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT); 1598 if (IS_ERR(prog)) 1599 return PTR_ERR(prog); 1600 1601 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) { 1602 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER 1603 * bpf prog (e.g. sockmap). It depends on the 1604 * limitation imposed by bpf_prog_load(). 1605 * Hence, sysctl_optmem_max is not checked. 1606 */ 1607 if ((sk->sk_type != SOCK_STREAM && 1608 sk->sk_type != SOCK_DGRAM) || 1609 (sk->sk_protocol != IPPROTO_UDP && 1610 sk->sk_protocol != IPPROTO_TCP) || 1611 (sk->sk_family != AF_INET && 1612 sk->sk_family != AF_INET6)) { 1613 err = -ENOTSUPP; 1614 goto err_prog_put; 1615 } 1616 } else { 1617 /* BPF_PROG_TYPE_SOCKET_FILTER */ 1618 if (bpf_prog_size(prog->len) > sysctl_optmem_max) { 1619 err = -ENOMEM; 1620 goto err_prog_put; 1621 } 1622 } 1623 1624 err = reuseport_attach_prog(sk, prog); 1625 err_prog_put: 1626 if (err) 1627 bpf_prog_put(prog); 1628 1629 return err; 1630 } 1631 1632 void sk_reuseport_prog_free(struct bpf_prog *prog) 1633 { 1634 if (!prog) 1635 return; 1636 1637 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) 1638 bpf_prog_put(prog); 1639 else 1640 bpf_prog_destroy(prog); 1641 } 1642 1643 struct bpf_scratchpad { 1644 union { 1645 __be32 diff[MAX_BPF_STACK / sizeof(__be32)]; 1646 u8 buff[MAX_BPF_STACK]; 1647 }; 1648 }; 1649 1650 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp); 1651 1652 static inline int __bpf_try_make_writable(struct sk_buff *skb, 1653 unsigned int write_len) 1654 { 1655 return skb_ensure_writable(skb, write_len); 1656 } 1657 1658 static inline int bpf_try_make_writable(struct sk_buff *skb, 1659 unsigned int write_len) 1660 { 1661 int err = __bpf_try_make_writable(skb, write_len); 1662 1663 bpf_compute_data_pointers(skb); 1664 return err; 1665 } 1666 1667 static int bpf_try_make_head_writable(struct sk_buff *skb) 1668 { 1669 return bpf_try_make_writable(skb, skb_headlen(skb)); 1670 } 1671 1672 static inline void bpf_push_mac_rcsum(struct sk_buff *skb) 1673 { 1674 if (skb_at_tc_ingress(skb)) 1675 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1676 } 1677 1678 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb) 1679 { 1680 if (skb_at_tc_ingress(skb)) 1681 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1682 } 1683 1684 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset, 1685 const void *, from, u32, len, u64, flags) 1686 { 1687 void *ptr; 1688 1689 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH))) 1690 return -EINVAL; 1691 if (unlikely(offset > 0xffff)) 1692 return -EFAULT; 1693 if (unlikely(bpf_try_make_writable(skb, offset + len))) 1694 return -EFAULT; 1695 1696 ptr = skb->data + offset; 1697 if (flags & BPF_F_RECOMPUTE_CSUM) 1698 __skb_postpull_rcsum(skb, ptr, len, offset); 1699 1700 memcpy(ptr, from, len); 1701 1702 if (flags & BPF_F_RECOMPUTE_CSUM) 1703 __skb_postpush_rcsum(skb, ptr, len, offset); 1704 if (flags & BPF_F_INVALIDATE_HASH) 1705 skb_clear_hash(skb); 1706 1707 return 0; 1708 } 1709 1710 static const struct bpf_func_proto bpf_skb_store_bytes_proto = { 1711 .func = bpf_skb_store_bytes, 1712 .gpl_only = false, 1713 .ret_type = RET_INTEGER, 1714 .arg1_type = ARG_PTR_TO_CTX, 1715 .arg2_type = ARG_ANYTHING, 1716 .arg3_type = ARG_PTR_TO_MEM, 1717 .arg4_type = ARG_CONST_SIZE, 1718 .arg5_type = ARG_ANYTHING, 1719 }; 1720 1721 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset, 1722 void *, to, u32, len) 1723 { 1724 void *ptr; 1725 1726 if (unlikely(offset > 0xffff)) 1727 goto err_clear; 1728 1729 ptr = skb_header_pointer(skb, offset, len, to); 1730 if (unlikely(!ptr)) 1731 goto err_clear; 1732 if (ptr != to) 1733 memcpy(to, ptr, len); 1734 1735 return 0; 1736 err_clear: 1737 memset(to, 0, len); 1738 return -EFAULT; 1739 } 1740 1741 static const struct bpf_func_proto bpf_skb_load_bytes_proto = { 1742 .func = bpf_skb_load_bytes, 1743 .gpl_only = false, 1744 .ret_type = RET_INTEGER, 1745 .arg1_type = ARG_PTR_TO_CTX, 1746 .arg2_type = ARG_ANYTHING, 1747 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1748 .arg4_type = ARG_CONST_SIZE, 1749 }; 1750 1751 BPF_CALL_4(bpf_flow_dissector_load_bytes, 1752 const struct bpf_flow_dissector *, ctx, u32, offset, 1753 void *, to, u32, len) 1754 { 1755 void *ptr; 1756 1757 if (unlikely(offset > 0xffff)) 1758 goto err_clear; 1759 1760 if (unlikely(!ctx->skb)) 1761 goto err_clear; 1762 1763 ptr = skb_header_pointer(ctx->skb, offset, len, to); 1764 if (unlikely(!ptr)) 1765 goto err_clear; 1766 if (ptr != to) 1767 memcpy(to, ptr, len); 1768 1769 return 0; 1770 err_clear: 1771 memset(to, 0, len); 1772 return -EFAULT; 1773 } 1774 1775 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = { 1776 .func = bpf_flow_dissector_load_bytes, 1777 .gpl_only = false, 1778 .ret_type = RET_INTEGER, 1779 .arg1_type = ARG_PTR_TO_CTX, 1780 .arg2_type = ARG_ANYTHING, 1781 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1782 .arg4_type = ARG_CONST_SIZE, 1783 }; 1784 1785 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb, 1786 u32, offset, void *, to, u32, len, u32, start_header) 1787 { 1788 u8 *end = skb_tail_pointer(skb); 1789 u8 *start, *ptr; 1790 1791 if (unlikely(offset > 0xffff)) 1792 goto err_clear; 1793 1794 switch (start_header) { 1795 case BPF_HDR_START_MAC: 1796 if (unlikely(!skb_mac_header_was_set(skb))) 1797 goto err_clear; 1798 start = skb_mac_header(skb); 1799 break; 1800 case BPF_HDR_START_NET: 1801 start = skb_network_header(skb); 1802 break; 1803 default: 1804 goto err_clear; 1805 } 1806 1807 ptr = start + offset; 1808 1809 if (likely(ptr + len <= end)) { 1810 memcpy(to, ptr, len); 1811 return 0; 1812 } 1813 1814 err_clear: 1815 memset(to, 0, len); 1816 return -EFAULT; 1817 } 1818 1819 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = { 1820 .func = bpf_skb_load_bytes_relative, 1821 .gpl_only = false, 1822 .ret_type = RET_INTEGER, 1823 .arg1_type = ARG_PTR_TO_CTX, 1824 .arg2_type = ARG_ANYTHING, 1825 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1826 .arg4_type = ARG_CONST_SIZE, 1827 .arg5_type = ARG_ANYTHING, 1828 }; 1829 1830 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len) 1831 { 1832 /* Idea is the following: should the needed direct read/write 1833 * test fail during runtime, we can pull in more data and redo 1834 * again, since implicitly, we invalidate previous checks here. 1835 * 1836 * Or, since we know how much we need to make read/writeable, 1837 * this can be done once at the program beginning for direct 1838 * access case. By this we overcome limitations of only current 1839 * headroom being accessible. 1840 */ 1841 return bpf_try_make_writable(skb, len ? : skb_headlen(skb)); 1842 } 1843 1844 static const struct bpf_func_proto bpf_skb_pull_data_proto = { 1845 .func = bpf_skb_pull_data, 1846 .gpl_only = false, 1847 .ret_type = RET_INTEGER, 1848 .arg1_type = ARG_PTR_TO_CTX, 1849 .arg2_type = ARG_ANYTHING, 1850 }; 1851 1852 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk) 1853 { 1854 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL; 1855 } 1856 1857 static const struct bpf_func_proto bpf_sk_fullsock_proto = { 1858 .func = bpf_sk_fullsock, 1859 .gpl_only = false, 1860 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 1861 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 1862 }; 1863 1864 static inline int sk_skb_try_make_writable(struct sk_buff *skb, 1865 unsigned int write_len) 1866 { 1867 return __bpf_try_make_writable(skb, write_len); 1868 } 1869 1870 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len) 1871 { 1872 /* Idea is the following: should the needed direct read/write 1873 * test fail during runtime, we can pull in more data and redo 1874 * again, since implicitly, we invalidate previous checks here. 1875 * 1876 * Or, since we know how much we need to make read/writeable, 1877 * this can be done once at the program beginning for direct 1878 * access case. By this we overcome limitations of only current 1879 * headroom being accessible. 1880 */ 1881 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb)); 1882 } 1883 1884 static const struct bpf_func_proto sk_skb_pull_data_proto = { 1885 .func = sk_skb_pull_data, 1886 .gpl_only = false, 1887 .ret_type = RET_INTEGER, 1888 .arg1_type = ARG_PTR_TO_CTX, 1889 .arg2_type = ARG_ANYTHING, 1890 }; 1891 1892 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset, 1893 u64, from, u64, to, u64, flags) 1894 { 1895 __sum16 *ptr; 1896 1897 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK))) 1898 return -EINVAL; 1899 if (unlikely(offset > 0xffff || offset & 1)) 1900 return -EFAULT; 1901 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1902 return -EFAULT; 1903 1904 ptr = (__sum16 *)(skb->data + offset); 1905 switch (flags & BPF_F_HDR_FIELD_MASK) { 1906 case 0: 1907 if (unlikely(from != 0)) 1908 return -EINVAL; 1909 1910 csum_replace_by_diff(ptr, to); 1911 break; 1912 case 2: 1913 csum_replace2(ptr, from, to); 1914 break; 1915 case 4: 1916 csum_replace4(ptr, from, to); 1917 break; 1918 default: 1919 return -EINVAL; 1920 } 1921 1922 return 0; 1923 } 1924 1925 static const struct bpf_func_proto bpf_l3_csum_replace_proto = { 1926 .func = bpf_l3_csum_replace, 1927 .gpl_only = false, 1928 .ret_type = RET_INTEGER, 1929 .arg1_type = ARG_PTR_TO_CTX, 1930 .arg2_type = ARG_ANYTHING, 1931 .arg3_type = ARG_ANYTHING, 1932 .arg4_type = ARG_ANYTHING, 1933 .arg5_type = ARG_ANYTHING, 1934 }; 1935 1936 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset, 1937 u64, from, u64, to, u64, flags) 1938 { 1939 bool is_pseudo = flags & BPF_F_PSEUDO_HDR; 1940 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0; 1941 bool do_mforce = flags & BPF_F_MARK_ENFORCE; 1942 __sum16 *ptr; 1943 1944 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE | 1945 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK))) 1946 return -EINVAL; 1947 if (unlikely(offset > 0xffff || offset & 1)) 1948 return -EFAULT; 1949 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1950 return -EFAULT; 1951 1952 ptr = (__sum16 *)(skb->data + offset); 1953 if (is_mmzero && !do_mforce && !*ptr) 1954 return 0; 1955 1956 switch (flags & BPF_F_HDR_FIELD_MASK) { 1957 case 0: 1958 if (unlikely(from != 0)) 1959 return -EINVAL; 1960 1961 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo); 1962 break; 1963 case 2: 1964 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo); 1965 break; 1966 case 4: 1967 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo); 1968 break; 1969 default: 1970 return -EINVAL; 1971 } 1972 1973 if (is_mmzero && !*ptr) 1974 *ptr = CSUM_MANGLED_0; 1975 return 0; 1976 } 1977 1978 static const struct bpf_func_proto bpf_l4_csum_replace_proto = { 1979 .func = bpf_l4_csum_replace, 1980 .gpl_only = false, 1981 .ret_type = RET_INTEGER, 1982 .arg1_type = ARG_PTR_TO_CTX, 1983 .arg2_type = ARG_ANYTHING, 1984 .arg3_type = ARG_ANYTHING, 1985 .arg4_type = ARG_ANYTHING, 1986 .arg5_type = ARG_ANYTHING, 1987 }; 1988 1989 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size, 1990 __be32 *, to, u32, to_size, __wsum, seed) 1991 { 1992 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp); 1993 u32 diff_size = from_size + to_size; 1994 int i, j = 0; 1995 1996 /* This is quite flexible, some examples: 1997 * 1998 * from_size == 0, to_size > 0, seed := csum --> pushing data 1999 * from_size > 0, to_size == 0, seed := csum --> pulling data 2000 * from_size > 0, to_size > 0, seed := 0 --> diffing data 2001 * 2002 * Even for diffing, from_size and to_size don't need to be equal. 2003 */ 2004 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) || 2005 diff_size > sizeof(sp->diff))) 2006 return -EINVAL; 2007 2008 for (i = 0; i < from_size / sizeof(__be32); i++, j++) 2009 sp->diff[j] = ~from[i]; 2010 for (i = 0; i < to_size / sizeof(__be32); i++, j++) 2011 sp->diff[j] = to[i]; 2012 2013 return csum_partial(sp->diff, diff_size, seed); 2014 } 2015 2016 static const struct bpf_func_proto bpf_csum_diff_proto = { 2017 .func = bpf_csum_diff, 2018 .gpl_only = false, 2019 .pkt_access = true, 2020 .ret_type = RET_INTEGER, 2021 .arg1_type = ARG_PTR_TO_MEM_OR_NULL, 2022 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 2023 .arg3_type = ARG_PTR_TO_MEM_OR_NULL, 2024 .arg4_type = ARG_CONST_SIZE_OR_ZERO, 2025 .arg5_type = ARG_ANYTHING, 2026 }; 2027 2028 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum) 2029 { 2030 /* The interface is to be used in combination with bpf_csum_diff() 2031 * for direct packet writes. csum rotation for alignment as well 2032 * as emulating csum_sub() can be done from the eBPF program. 2033 */ 2034 if (skb->ip_summed == CHECKSUM_COMPLETE) 2035 return (skb->csum = csum_add(skb->csum, csum)); 2036 2037 return -ENOTSUPP; 2038 } 2039 2040 static const struct bpf_func_proto bpf_csum_update_proto = { 2041 .func = bpf_csum_update, 2042 .gpl_only = false, 2043 .ret_type = RET_INTEGER, 2044 .arg1_type = ARG_PTR_TO_CTX, 2045 .arg2_type = ARG_ANYTHING, 2046 }; 2047 2048 BPF_CALL_2(bpf_csum_level, struct sk_buff *, skb, u64, level) 2049 { 2050 /* The interface is to be used in combination with bpf_skb_adjust_room() 2051 * for encap/decap of packet headers when BPF_F_ADJ_ROOM_NO_CSUM_RESET 2052 * is passed as flags, for example. 2053 */ 2054 switch (level) { 2055 case BPF_CSUM_LEVEL_INC: 2056 __skb_incr_checksum_unnecessary(skb); 2057 break; 2058 case BPF_CSUM_LEVEL_DEC: 2059 __skb_decr_checksum_unnecessary(skb); 2060 break; 2061 case BPF_CSUM_LEVEL_RESET: 2062 __skb_reset_checksum_unnecessary(skb); 2063 break; 2064 case BPF_CSUM_LEVEL_QUERY: 2065 return skb->ip_summed == CHECKSUM_UNNECESSARY ? 2066 skb->csum_level : -EACCES; 2067 default: 2068 return -EINVAL; 2069 } 2070 2071 return 0; 2072 } 2073 2074 static const struct bpf_func_proto bpf_csum_level_proto = { 2075 .func = bpf_csum_level, 2076 .gpl_only = false, 2077 .ret_type = RET_INTEGER, 2078 .arg1_type = ARG_PTR_TO_CTX, 2079 .arg2_type = ARG_ANYTHING, 2080 }; 2081 2082 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb) 2083 { 2084 return dev_forward_skb_nomtu(dev, skb); 2085 } 2086 2087 static inline int __bpf_rx_skb_no_mac(struct net_device *dev, 2088 struct sk_buff *skb) 2089 { 2090 int ret = ____dev_forward_skb(dev, skb, false); 2091 2092 if (likely(!ret)) { 2093 skb->dev = dev; 2094 ret = netif_rx(skb); 2095 } 2096 2097 return ret; 2098 } 2099 2100 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb) 2101 { 2102 int ret; 2103 2104 if (dev_xmit_recursion()) { 2105 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2106 kfree_skb(skb); 2107 return -ENETDOWN; 2108 } 2109 2110 skb->dev = dev; 2111 skb->tstamp = 0; 2112 2113 dev_xmit_recursion_inc(); 2114 ret = dev_queue_xmit(skb); 2115 dev_xmit_recursion_dec(); 2116 2117 return ret; 2118 } 2119 2120 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev, 2121 u32 flags) 2122 { 2123 unsigned int mlen = skb_network_offset(skb); 2124 2125 if (mlen) { 2126 __skb_pull(skb, mlen); 2127 2128 /* At ingress, the mac header has already been pulled once. 2129 * At egress, skb_pospull_rcsum has to be done in case that 2130 * the skb is originated from ingress (i.e. a forwarded skb) 2131 * to ensure that rcsum starts at net header. 2132 */ 2133 if (!skb_at_tc_ingress(skb)) 2134 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen); 2135 } 2136 skb_pop_mac_header(skb); 2137 skb_reset_mac_len(skb); 2138 return flags & BPF_F_INGRESS ? 2139 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb); 2140 } 2141 2142 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev, 2143 u32 flags) 2144 { 2145 /* Verify that a link layer header is carried */ 2146 if (unlikely(skb->mac_header >= skb->network_header)) { 2147 kfree_skb(skb); 2148 return -ERANGE; 2149 } 2150 2151 bpf_push_mac_rcsum(skb); 2152 return flags & BPF_F_INGRESS ? 2153 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb); 2154 } 2155 2156 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev, 2157 u32 flags) 2158 { 2159 if (dev_is_mac_header_xmit(dev)) 2160 return __bpf_redirect_common(skb, dev, flags); 2161 else 2162 return __bpf_redirect_no_mac(skb, dev, flags); 2163 } 2164 2165 #if IS_ENABLED(CONFIG_IPV6) 2166 static int bpf_out_neigh_v6(struct net *net, struct sk_buff *skb, 2167 struct net_device *dev, struct bpf_nh_params *nh) 2168 { 2169 u32 hh_len = LL_RESERVED_SPACE(dev); 2170 const struct in6_addr *nexthop; 2171 struct dst_entry *dst = NULL; 2172 struct neighbour *neigh; 2173 2174 if (dev_xmit_recursion()) { 2175 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2176 goto out_drop; 2177 } 2178 2179 skb->dev = dev; 2180 skb->tstamp = 0; 2181 2182 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 2183 skb = skb_expand_head(skb, hh_len); 2184 if (!skb) 2185 return -ENOMEM; 2186 } 2187 2188 rcu_read_lock_bh(); 2189 if (!nh) { 2190 dst = skb_dst(skb); 2191 nexthop = rt6_nexthop(container_of(dst, struct rt6_info, dst), 2192 &ipv6_hdr(skb)->daddr); 2193 } else { 2194 nexthop = &nh->ipv6_nh; 2195 } 2196 neigh = ip_neigh_gw6(dev, nexthop); 2197 if (likely(!IS_ERR(neigh))) { 2198 int ret; 2199 2200 sock_confirm_neigh(skb, neigh); 2201 dev_xmit_recursion_inc(); 2202 ret = neigh_output(neigh, skb, false); 2203 dev_xmit_recursion_dec(); 2204 rcu_read_unlock_bh(); 2205 return ret; 2206 } 2207 rcu_read_unlock_bh(); 2208 if (dst) 2209 IP6_INC_STATS(net, ip6_dst_idev(dst), IPSTATS_MIB_OUTNOROUTES); 2210 out_drop: 2211 kfree_skb(skb); 2212 return -ENETDOWN; 2213 } 2214 2215 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev, 2216 struct bpf_nh_params *nh) 2217 { 2218 const struct ipv6hdr *ip6h = ipv6_hdr(skb); 2219 struct net *net = dev_net(dev); 2220 int err, ret = NET_XMIT_DROP; 2221 2222 if (!nh) { 2223 struct dst_entry *dst; 2224 struct flowi6 fl6 = { 2225 .flowi6_flags = FLOWI_FLAG_ANYSRC, 2226 .flowi6_mark = skb->mark, 2227 .flowlabel = ip6_flowinfo(ip6h), 2228 .flowi6_oif = dev->ifindex, 2229 .flowi6_proto = ip6h->nexthdr, 2230 .daddr = ip6h->daddr, 2231 .saddr = ip6h->saddr, 2232 }; 2233 2234 dst = ipv6_stub->ipv6_dst_lookup_flow(net, NULL, &fl6, NULL); 2235 if (IS_ERR(dst)) 2236 goto out_drop; 2237 2238 skb_dst_set(skb, dst); 2239 } else if (nh->nh_family != AF_INET6) { 2240 goto out_drop; 2241 } 2242 2243 err = bpf_out_neigh_v6(net, skb, dev, nh); 2244 if (unlikely(net_xmit_eval(err))) 2245 dev->stats.tx_errors++; 2246 else 2247 ret = NET_XMIT_SUCCESS; 2248 goto out_xmit; 2249 out_drop: 2250 dev->stats.tx_errors++; 2251 kfree_skb(skb); 2252 out_xmit: 2253 return ret; 2254 } 2255 #else 2256 static int __bpf_redirect_neigh_v6(struct sk_buff *skb, struct net_device *dev, 2257 struct bpf_nh_params *nh) 2258 { 2259 kfree_skb(skb); 2260 return NET_XMIT_DROP; 2261 } 2262 #endif /* CONFIG_IPV6 */ 2263 2264 #if IS_ENABLED(CONFIG_INET) 2265 static int bpf_out_neigh_v4(struct net *net, struct sk_buff *skb, 2266 struct net_device *dev, struct bpf_nh_params *nh) 2267 { 2268 u32 hh_len = LL_RESERVED_SPACE(dev); 2269 struct neighbour *neigh; 2270 bool is_v6gw = false; 2271 2272 if (dev_xmit_recursion()) { 2273 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2274 goto out_drop; 2275 } 2276 2277 skb->dev = dev; 2278 skb->tstamp = 0; 2279 2280 if (unlikely(skb_headroom(skb) < hh_len && dev->header_ops)) { 2281 skb = skb_expand_head(skb, hh_len); 2282 if (!skb) 2283 return -ENOMEM; 2284 } 2285 2286 rcu_read_lock_bh(); 2287 if (!nh) { 2288 struct dst_entry *dst = skb_dst(skb); 2289 struct rtable *rt = container_of(dst, struct rtable, dst); 2290 2291 neigh = ip_neigh_for_gw(rt, skb, &is_v6gw); 2292 } else if (nh->nh_family == AF_INET6) { 2293 neigh = ip_neigh_gw6(dev, &nh->ipv6_nh); 2294 is_v6gw = true; 2295 } else if (nh->nh_family == AF_INET) { 2296 neigh = ip_neigh_gw4(dev, nh->ipv4_nh); 2297 } else { 2298 rcu_read_unlock_bh(); 2299 goto out_drop; 2300 } 2301 2302 if (likely(!IS_ERR(neigh))) { 2303 int ret; 2304 2305 sock_confirm_neigh(skb, neigh); 2306 dev_xmit_recursion_inc(); 2307 ret = neigh_output(neigh, skb, is_v6gw); 2308 dev_xmit_recursion_dec(); 2309 rcu_read_unlock_bh(); 2310 return ret; 2311 } 2312 rcu_read_unlock_bh(); 2313 out_drop: 2314 kfree_skb(skb); 2315 return -ENETDOWN; 2316 } 2317 2318 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev, 2319 struct bpf_nh_params *nh) 2320 { 2321 const struct iphdr *ip4h = ip_hdr(skb); 2322 struct net *net = dev_net(dev); 2323 int err, ret = NET_XMIT_DROP; 2324 2325 if (!nh) { 2326 struct flowi4 fl4 = { 2327 .flowi4_flags = FLOWI_FLAG_ANYSRC, 2328 .flowi4_mark = skb->mark, 2329 .flowi4_tos = RT_TOS(ip4h->tos), 2330 .flowi4_oif = dev->ifindex, 2331 .flowi4_proto = ip4h->protocol, 2332 .daddr = ip4h->daddr, 2333 .saddr = ip4h->saddr, 2334 }; 2335 struct rtable *rt; 2336 2337 rt = ip_route_output_flow(net, &fl4, NULL); 2338 if (IS_ERR(rt)) 2339 goto out_drop; 2340 if (rt->rt_type != RTN_UNICAST && rt->rt_type != RTN_LOCAL) { 2341 ip_rt_put(rt); 2342 goto out_drop; 2343 } 2344 2345 skb_dst_set(skb, &rt->dst); 2346 } 2347 2348 err = bpf_out_neigh_v4(net, skb, dev, nh); 2349 if (unlikely(net_xmit_eval(err))) 2350 dev->stats.tx_errors++; 2351 else 2352 ret = NET_XMIT_SUCCESS; 2353 goto out_xmit; 2354 out_drop: 2355 dev->stats.tx_errors++; 2356 kfree_skb(skb); 2357 out_xmit: 2358 return ret; 2359 } 2360 #else 2361 static int __bpf_redirect_neigh_v4(struct sk_buff *skb, struct net_device *dev, 2362 struct bpf_nh_params *nh) 2363 { 2364 kfree_skb(skb); 2365 return NET_XMIT_DROP; 2366 } 2367 #endif /* CONFIG_INET */ 2368 2369 static int __bpf_redirect_neigh(struct sk_buff *skb, struct net_device *dev, 2370 struct bpf_nh_params *nh) 2371 { 2372 struct ethhdr *ethh = eth_hdr(skb); 2373 2374 if (unlikely(skb->mac_header >= skb->network_header)) 2375 goto out; 2376 bpf_push_mac_rcsum(skb); 2377 if (is_multicast_ether_addr(ethh->h_dest)) 2378 goto out; 2379 2380 skb_pull(skb, sizeof(*ethh)); 2381 skb_unset_mac_header(skb); 2382 skb_reset_network_header(skb); 2383 2384 if (skb->protocol == htons(ETH_P_IP)) 2385 return __bpf_redirect_neigh_v4(skb, dev, nh); 2386 else if (skb->protocol == htons(ETH_P_IPV6)) 2387 return __bpf_redirect_neigh_v6(skb, dev, nh); 2388 out: 2389 kfree_skb(skb); 2390 return -ENOTSUPP; 2391 } 2392 2393 /* Internal, non-exposed redirect flags. */ 2394 enum { 2395 BPF_F_NEIGH = (1ULL << 1), 2396 BPF_F_PEER = (1ULL << 2), 2397 BPF_F_NEXTHOP = (1ULL << 3), 2398 #define BPF_F_REDIRECT_INTERNAL (BPF_F_NEIGH | BPF_F_PEER | BPF_F_NEXTHOP) 2399 }; 2400 2401 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags) 2402 { 2403 struct net_device *dev; 2404 struct sk_buff *clone; 2405 int ret; 2406 2407 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL))) 2408 return -EINVAL; 2409 2410 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex); 2411 if (unlikely(!dev)) 2412 return -EINVAL; 2413 2414 clone = skb_clone(skb, GFP_ATOMIC); 2415 if (unlikely(!clone)) 2416 return -ENOMEM; 2417 2418 /* For direct write, we need to keep the invariant that the skbs 2419 * we're dealing with need to be uncloned. Should uncloning fail 2420 * here, we need to free the just generated clone to unclone once 2421 * again. 2422 */ 2423 ret = bpf_try_make_head_writable(skb); 2424 if (unlikely(ret)) { 2425 kfree_skb(clone); 2426 return -ENOMEM; 2427 } 2428 2429 return __bpf_redirect(clone, dev, flags); 2430 } 2431 2432 static const struct bpf_func_proto bpf_clone_redirect_proto = { 2433 .func = bpf_clone_redirect, 2434 .gpl_only = false, 2435 .ret_type = RET_INTEGER, 2436 .arg1_type = ARG_PTR_TO_CTX, 2437 .arg2_type = ARG_ANYTHING, 2438 .arg3_type = ARG_ANYTHING, 2439 }; 2440 2441 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 2442 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info); 2443 2444 int skb_do_redirect(struct sk_buff *skb) 2445 { 2446 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2447 struct net *net = dev_net(skb->dev); 2448 struct net_device *dev; 2449 u32 flags = ri->flags; 2450 2451 dev = dev_get_by_index_rcu(net, ri->tgt_index); 2452 ri->tgt_index = 0; 2453 ri->flags = 0; 2454 if (unlikely(!dev)) 2455 goto out_drop; 2456 if (flags & BPF_F_PEER) { 2457 const struct net_device_ops *ops = dev->netdev_ops; 2458 2459 if (unlikely(!ops->ndo_get_peer_dev || 2460 !skb_at_tc_ingress(skb))) 2461 goto out_drop; 2462 dev = ops->ndo_get_peer_dev(dev); 2463 if (unlikely(!dev || 2464 !(dev->flags & IFF_UP) || 2465 net_eq(net, dev_net(dev)))) 2466 goto out_drop; 2467 skb->dev = dev; 2468 return -EAGAIN; 2469 } 2470 return flags & BPF_F_NEIGH ? 2471 __bpf_redirect_neigh(skb, dev, flags & BPF_F_NEXTHOP ? 2472 &ri->nh : NULL) : 2473 __bpf_redirect(skb, dev, flags); 2474 out_drop: 2475 kfree_skb(skb); 2476 return -EINVAL; 2477 } 2478 2479 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags) 2480 { 2481 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2482 2483 if (unlikely(flags & (~(BPF_F_INGRESS) | BPF_F_REDIRECT_INTERNAL))) 2484 return TC_ACT_SHOT; 2485 2486 ri->flags = flags; 2487 ri->tgt_index = ifindex; 2488 2489 return TC_ACT_REDIRECT; 2490 } 2491 2492 static const struct bpf_func_proto bpf_redirect_proto = { 2493 .func = bpf_redirect, 2494 .gpl_only = false, 2495 .ret_type = RET_INTEGER, 2496 .arg1_type = ARG_ANYTHING, 2497 .arg2_type = ARG_ANYTHING, 2498 }; 2499 2500 BPF_CALL_2(bpf_redirect_peer, u32, ifindex, u64, flags) 2501 { 2502 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2503 2504 if (unlikely(flags)) 2505 return TC_ACT_SHOT; 2506 2507 ri->flags = BPF_F_PEER; 2508 ri->tgt_index = ifindex; 2509 2510 return TC_ACT_REDIRECT; 2511 } 2512 2513 static const struct bpf_func_proto bpf_redirect_peer_proto = { 2514 .func = bpf_redirect_peer, 2515 .gpl_only = false, 2516 .ret_type = RET_INTEGER, 2517 .arg1_type = ARG_ANYTHING, 2518 .arg2_type = ARG_ANYTHING, 2519 }; 2520 2521 BPF_CALL_4(bpf_redirect_neigh, u32, ifindex, struct bpf_redir_neigh *, params, 2522 int, plen, u64, flags) 2523 { 2524 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2525 2526 if (unlikely((plen && plen < sizeof(*params)) || flags)) 2527 return TC_ACT_SHOT; 2528 2529 ri->flags = BPF_F_NEIGH | (plen ? BPF_F_NEXTHOP : 0); 2530 ri->tgt_index = ifindex; 2531 2532 BUILD_BUG_ON(sizeof(struct bpf_redir_neigh) != sizeof(struct bpf_nh_params)); 2533 if (plen) 2534 memcpy(&ri->nh, params, sizeof(ri->nh)); 2535 2536 return TC_ACT_REDIRECT; 2537 } 2538 2539 static const struct bpf_func_proto bpf_redirect_neigh_proto = { 2540 .func = bpf_redirect_neigh, 2541 .gpl_only = false, 2542 .ret_type = RET_INTEGER, 2543 .arg1_type = ARG_ANYTHING, 2544 .arg2_type = ARG_PTR_TO_MEM_OR_NULL, 2545 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 2546 .arg4_type = ARG_ANYTHING, 2547 }; 2548 2549 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes) 2550 { 2551 msg->apply_bytes = bytes; 2552 return 0; 2553 } 2554 2555 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = { 2556 .func = bpf_msg_apply_bytes, 2557 .gpl_only = false, 2558 .ret_type = RET_INTEGER, 2559 .arg1_type = ARG_PTR_TO_CTX, 2560 .arg2_type = ARG_ANYTHING, 2561 }; 2562 2563 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes) 2564 { 2565 msg->cork_bytes = bytes; 2566 return 0; 2567 } 2568 2569 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = { 2570 .func = bpf_msg_cork_bytes, 2571 .gpl_only = false, 2572 .ret_type = RET_INTEGER, 2573 .arg1_type = ARG_PTR_TO_CTX, 2574 .arg2_type = ARG_ANYTHING, 2575 }; 2576 2577 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start, 2578 u32, end, u64, flags) 2579 { 2580 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start; 2581 u32 first_sge, last_sge, i, shift, bytes_sg_total; 2582 struct scatterlist *sge; 2583 u8 *raw, *to, *from; 2584 struct page *page; 2585 2586 if (unlikely(flags || end <= start)) 2587 return -EINVAL; 2588 2589 /* First find the starting scatterlist element */ 2590 i = msg->sg.start; 2591 do { 2592 offset += len; 2593 len = sk_msg_elem(msg, i)->length; 2594 if (start < offset + len) 2595 break; 2596 sk_msg_iter_var_next(i); 2597 } while (i != msg->sg.end); 2598 2599 if (unlikely(start >= offset + len)) 2600 return -EINVAL; 2601 2602 first_sge = i; 2603 /* The start may point into the sg element so we need to also 2604 * account for the headroom. 2605 */ 2606 bytes_sg_total = start - offset + bytes; 2607 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len) 2608 goto out; 2609 2610 /* At this point we need to linearize multiple scatterlist 2611 * elements or a single shared page. Either way we need to 2612 * copy into a linear buffer exclusively owned by BPF. Then 2613 * place the buffer in the scatterlist and fixup the original 2614 * entries by removing the entries now in the linear buffer 2615 * and shifting the remaining entries. For now we do not try 2616 * to copy partial entries to avoid complexity of running out 2617 * of sg_entry slots. The downside is reading a single byte 2618 * will copy the entire sg entry. 2619 */ 2620 do { 2621 copy += sk_msg_elem(msg, i)->length; 2622 sk_msg_iter_var_next(i); 2623 if (bytes_sg_total <= copy) 2624 break; 2625 } while (i != msg->sg.end); 2626 last_sge = i; 2627 2628 if (unlikely(bytes_sg_total > copy)) 2629 return -EINVAL; 2630 2631 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2632 get_order(copy)); 2633 if (unlikely(!page)) 2634 return -ENOMEM; 2635 2636 raw = page_address(page); 2637 i = first_sge; 2638 do { 2639 sge = sk_msg_elem(msg, i); 2640 from = sg_virt(sge); 2641 len = sge->length; 2642 to = raw + poffset; 2643 2644 memcpy(to, from, len); 2645 poffset += len; 2646 sge->length = 0; 2647 put_page(sg_page(sge)); 2648 2649 sk_msg_iter_var_next(i); 2650 } while (i != last_sge); 2651 2652 sg_set_page(&msg->sg.data[first_sge], page, copy, 0); 2653 2654 /* To repair sg ring we need to shift entries. If we only 2655 * had a single entry though we can just replace it and 2656 * be done. Otherwise walk the ring and shift the entries. 2657 */ 2658 WARN_ON_ONCE(last_sge == first_sge); 2659 shift = last_sge > first_sge ? 2660 last_sge - first_sge - 1 : 2661 NR_MSG_FRAG_IDS - first_sge + last_sge - 1; 2662 if (!shift) 2663 goto out; 2664 2665 i = first_sge; 2666 sk_msg_iter_var_next(i); 2667 do { 2668 u32 move_from; 2669 2670 if (i + shift >= NR_MSG_FRAG_IDS) 2671 move_from = i + shift - NR_MSG_FRAG_IDS; 2672 else 2673 move_from = i + shift; 2674 if (move_from == msg->sg.end) 2675 break; 2676 2677 msg->sg.data[i] = msg->sg.data[move_from]; 2678 msg->sg.data[move_from].length = 0; 2679 msg->sg.data[move_from].page_link = 0; 2680 msg->sg.data[move_from].offset = 0; 2681 sk_msg_iter_var_next(i); 2682 } while (1); 2683 2684 msg->sg.end = msg->sg.end - shift > msg->sg.end ? 2685 msg->sg.end - shift + NR_MSG_FRAG_IDS : 2686 msg->sg.end - shift; 2687 out: 2688 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset; 2689 msg->data_end = msg->data + bytes; 2690 return 0; 2691 } 2692 2693 static const struct bpf_func_proto bpf_msg_pull_data_proto = { 2694 .func = bpf_msg_pull_data, 2695 .gpl_only = false, 2696 .ret_type = RET_INTEGER, 2697 .arg1_type = ARG_PTR_TO_CTX, 2698 .arg2_type = ARG_ANYTHING, 2699 .arg3_type = ARG_ANYTHING, 2700 .arg4_type = ARG_ANYTHING, 2701 }; 2702 2703 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start, 2704 u32, len, u64, flags) 2705 { 2706 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge; 2707 u32 new, i = 0, l = 0, space, copy = 0, offset = 0; 2708 u8 *raw, *to, *from; 2709 struct page *page; 2710 2711 if (unlikely(flags)) 2712 return -EINVAL; 2713 2714 /* First find the starting scatterlist element */ 2715 i = msg->sg.start; 2716 do { 2717 offset += l; 2718 l = sk_msg_elem(msg, i)->length; 2719 2720 if (start < offset + l) 2721 break; 2722 sk_msg_iter_var_next(i); 2723 } while (i != msg->sg.end); 2724 2725 if (start >= offset + l) 2726 return -EINVAL; 2727 2728 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2729 2730 /* If no space available will fallback to copy, we need at 2731 * least one scatterlist elem available to push data into 2732 * when start aligns to the beginning of an element or two 2733 * when it falls inside an element. We handle the start equals 2734 * offset case because its the common case for inserting a 2735 * header. 2736 */ 2737 if (!space || (space == 1 && start != offset)) 2738 copy = msg->sg.data[i].length; 2739 2740 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2741 get_order(copy + len)); 2742 if (unlikely(!page)) 2743 return -ENOMEM; 2744 2745 if (copy) { 2746 int front, back; 2747 2748 raw = page_address(page); 2749 2750 psge = sk_msg_elem(msg, i); 2751 front = start - offset; 2752 back = psge->length - front; 2753 from = sg_virt(psge); 2754 2755 if (front) 2756 memcpy(raw, from, front); 2757 2758 if (back) { 2759 from += front; 2760 to = raw + front + len; 2761 2762 memcpy(to, from, back); 2763 } 2764 2765 put_page(sg_page(psge)); 2766 } else if (start - offset) { 2767 psge = sk_msg_elem(msg, i); 2768 rsge = sk_msg_elem_cpy(msg, i); 2769 2770 psge->length = start - offset; 2771 rsge.length -= psge->length; 2772 rsge.offset += start; 2773 2774 sk_msg_iter_var_next(i); 2775 sg_unmark_end(psge); 2776 sg_unmark_end(&rsge); 2777 sk_msg_iter_next(msg, end); 2778 } 2779 2780 /* Slot(s) to place newly allocated data */ 2781 new = i; 2782 2783 /* Shift one or two slots as needed */ 2784 if (!copy) { 2785 sge = sk_msg_elem_cpy(msg, i); 2786 2787 sk_msg_iter_var_next(i); 2788 sg_unmark_end(&sge); 2789 sk_msg_iter_next(msg, end); 2790 2791 nsge = sk_msg_elem_cpy(msg, i); 2792 if (rsge.length) { 2793 sk_msg_iter_var_next(i); 2794 nnsge = sk_msg_elem_cpy(msg, i); 2795 } 2796 2797 while (i != msg->sg.end) { 2798 msg->sg.data[i] = sge; 2799 sge = nsge; 2800 sk_msg_iter_var_next(i); 2801 if (rsge.length) { 2802 nsge = nnsge; 2803 nnsge = sk_msg_elem_cpy(msg, i); 2804 } else { 2805 nsge = sk_msg_elem_cpy(msg, i); 2806 } 2807 } 2808 } 2809 2810 /* Place newly allocated data buffer */ 2811 sk_mem_charge(msg->sk, len); 2812 msg->sg.size += len; 2813 __clear_bit(new, &msg->sg.copy); 2814 sg_set_page(&msg->sg.data[new], page, len + copy, 0); 2815 if (rsge.length) { 2816 get_page(sg_page(&rsge)); 2817 sk_msg_iter_var_next(new); 2818 msg->sg.data[new] = rsge; 2819 } 2820 2821 sk_msg_compute_data_pointers(msg); 2822 return 0; 2823 } 2824 2825 static const struct bpf_func_proto bpf_msg_push_data_proto = { 2826 .func = bpf_msg_push_data, 2827 .gpl_only = false, 2828 .ret_type = RET_INTEGER, 2829 .arg1_type = ARG_PTR_TO_CTX, 2830 .arg2_type = ARG_ANYTHING, 2831 .arg3_type = ARG_ANYTHING, 2832 .arg4_type = ARG_ANYTHING, 2833 }; 2834 2835 static void sk_msg_shift_left(struct sk_msg *msg, int i) 2836 { 2837 int prev; 2838 2839 do { 2840 prev = i; 2841 sk_msg_iter_var_next(i); 2842 msg->sg.data[prev] = msg->sg.data[i]; 2843 } while (i != msg->sg.end); 2844 2845 sk_msg_iter_prev(msg, end); 2846 } 2847 2848 static void sk_msg_shift_right(struct sk_msg *msg, int i) 2849 { 2850 struct scatterlist tmp, sge; 2851 2852 sk_msg_iter_next(msg, end); 2853 sge = sk_msg_elem_cpy(msg, i); 2854 sk_msg_iter_var_next(i); 2855 tmp = sk_msg_elem_cpy(msg, i); 2856 2857 while (i != msg->sg.end) { 2858 msg->sg.data[i] = sge; 2859 sk_msg_iter_var_next(i); 2860 sge = tmp; 2861 tmp = sk_msg_elem_cpy(msg, i); 2862 } 2863 } 2864 2865 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start, 2866 u32, len, u64, flags) 2867 { 2868 u32 i = 0, l = 0, space, offset = 0; 2869 u64 last = start + len; 2870 int pop; 2871 2872 if (unlikely(flags)) 2873 return -EINVAL; 2874 2875 /* First find the starting scatterlist element */ 2876 i = msg->sg.start; 2877 do { 2878 offset += l; 2879 l = sk_msg_elem(msg, i)->length; 2880 2881 if (start < offset + l) 2882 break; 2883 sk_msg_iter_var_next(i); 2884 } while (i != msg->sg.end); 2885 2886 /* Bounds checks: start and pop must be inside message */ 2887 if (start >= offset + l || last >= msg->sg.size) 2888 return -EINVAL; 2889 2890 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2891 2892 pop = len; 2893 /* --------------| offset 2894 * -| start |-------- len -------| 2895 * 2896 * |----- a ----|-------- pop -------|----- b ----| 2897 * |______________________________________________| length 2898 * 2899 * 2900 * a: region at front of scatter element to save 2901 * b: region at back of scatter element to save when length > A + pop 2902 * pop: region to pop from element, same as input 'pop' here will be 2903 * decremented below per iteration. 2904 * 2905 * Two top-level cases to handle when start != offset, first B is non 2906 * zero and second B is zero corresponding to when a pop includes more 2907 * than one element. 2908 * 2909 * Then if B is non-zero AND there is no space allocate space and 2910 * compact A, B regions into page. If there is space shift ring to 2911 * the rigth free'ing the next element in ring to place B, leaving 2912 * A untouched except to reduce length. 2913 */ 2914 if (start != offset) { 2915 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i); 2916 int a = start; 2917 int b = sge->length - pop - a; 2918 2919 sk_msg_iter_var_next(i); 2920 2921 if (pop < sge->length - a) { 2922 if (space) { 2923 sge->length = a; 2924 sk_msg_shift_right(msg, i); 2925 nsge = sk_msg_elem(msg, i); 2926 get_page(sg_page(sge)); 2927 sg_set_page(nsge, 2928 sg_page(sge), 2929 b, sge->offset + pop + a); 2930 } else { 2931 struct page *page, *orig; 2932 u8 *to, *from; 2933 2934 page = alloc_pages(__GFP_NOWARN | 2935 __GFP_COMP | GFP_ATOMIC, 2936 get_order(a + b)); 2937 if (unlikely(!page)) 2938 return -ENOMEM; 2939 2940 sge->length = a; 2941 orig = sg_page(sge); 2942 from = sg_virt(sge); 2943 to = page_address(page); 2944 memcpy(to, from, a); 2945 memcpy(to + a, from + a + pop, b); 2946 sg_set_page(sge, page, a + b, 0); 2947 put_page(orig); 2948 } 2949 pop = 0; 2950 } else if (pop >= sge->length - a) { 2951 pop -= (sge->length - a); 2952 sge->length = a; 2953 } 2954 } 2955 2956 /* From above the current layout _must_ be as follows, 2957 * 2958 * -| offset 2959 * -| start 2960 * 2961 * |---- pop ---|---------------- b ------------| 2962 * |____________________________________________| length 2963 * 2964 * Offset and start of the current msg elem are equal because in the 2965 * previous case we handled offset != start and either consumed the 2966 * entire element and advanced to the next element OR pop == 0. 2967 * 2968 * Two cases to handle here are first pop is less than the length 2969 * leaving some remainder b above. Simply adjust the element's layout 2970 * in this case. Or pop >= length of the element so that b = 0. In this 2971 * case advance to next element decrementing pop. 2972 */ 2973 while (pop) { 2974 struct scatterlist *sge = sk_msg_elem(msg, i); 2975 2976 if (pop < sge->length) { 2977 sge->length -= pop; 2978 sge->offset += pop; 2979 pop = 0; 2980 } else { 2981 pop -= sge->length; 2982 sk_msg_shift_left(msg, i); 2983 } 2984 sk_msg_iter_var_next(i); 2985 } 2986 2987 sk_mem_uncharge(msg->sk, len - pop); 2988 msg->sg.size -= (len - pop); 2989 sk_msg_compute_data_pointers(msg); 2990 return 0; 2991 } 2992 2993 static const struct bpf_func_proto bpf_msg_pop_data_proto = { 2994 .func = bpf_msg_pop_data, 2995 .gpl_only = false, 2996 .ret_type = RET_INTEGER, 2997 .arg1_type = ARG_PTR_TO_CTX, 2998 .arg2_type = ARG_ANYTHING, 2999 .arg3_type = ARG_ANYTHING, 3000 .arg4_type = ARG_ANYTHING, 3001 }; 3002 3003 #ifdef CONFIG_CGROUP_NET_CLASSID 3004 BPF_CALL_0(bpf_get_cgroup_classid_curr) 3005 { 3006 return __task_get_classid(current); 3007 } 3008 3009 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = { 3010 .func = bpf_get_cgroup_classid_curr, 3011 .gpl_only = false, 3012 .ret_type = RET_INTEGER, 3013 }; 3014 3015 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb) 3016 { 3017 struct sock *sk = skb_to_full_sk(skb); 3018 3019 if (!sk || !sk_fullsock(sk)) 3020 return 0; 3021 3022 return sock_cgroup_classid(&sk->sk_cgrp_data); 3023 } 3024 3025 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = { 3026 .func = bpf_skb_cgroup_classid, 3027 .gpl_only = false, 3028 .ret_type = RET_INTEGER, 3029 .arg1_type = ARG_PTR_TO_CTX, 3030 }; 3031 #endif 3032 3033 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb) 3034 { 3035 return task_get_classid(skb); 3036 } 3037 3038 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = { 3039 .func = bpf_get_cgroup_classid, 3040 .gpl_only = false, 3041 .ret_type = RET_INTEGER, 3042 .arg1_type = ARG_PTR_TO_CTX, 3043 }; 3044 3045 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb) 3046 { 3047 return dst_tclassid(skb); 3048 } 3049 3050 static const struct bpf_func_proto bpf_get_route_realm_proto = { 3051 .func = bpf_get_route_realm, 3052 .gpl_only = false, 3053 .ret_type = RET_INTEGER, 3054 .arg1_type = ARG_PTR_TO_CTX, 3055 }; 3056 3057 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb) 3058 { 3059 /* If skb_clear_hash() was called due to mangling, we can 3060 * trigger SW recalculation here. Later access to hash 3061 * can then use the inline skb->hash via context directly 3062 * instead of calling this helper again. 3063 */ 3064 return skb_get_hash(skb); 3065 } 3066 3067 static const struct bpf_func_proto bpf_get_hash_recalc_proto = { 3068 .func = bpf_get_hash_recalc, 3069 .gpl_only = false, 3070 .ret_type = RET_INTEGER, 3071 .arg1_type = ARG_PTR_TO_CTX, 3072 }; 3073 3074 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb) 3075 { 3076 /* After all direct packet write, this can be used once for 3077 * triggering a lazy recalc on next skb_get_hash() invocation. 3078 */ 3079 skb_clear_hash(skb); 3080 return 0; 3081 } 3082 3083 static const struct bpf_func_proto bpf_set_hash_invalid_proto = { 3084 .func = bpf_set_hash_invalid, 3085 .gpl_only = false, 3086 .ret_type = RET_INTEGER, 3087 .arg1_type = ARG_PTR_TO_CTX, 3088 }; 3089 3090 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash) 3091 { 3092 /* Set user specified hash as L4(+), so that it gets returned 3093 * on skb_get_hash() call unless BPF prog later on triggers a 3094 * skb_clear_hash(). 3095 */ 3096 __skb_set_sw_hash(skb, hash, true); 3097 return 0; 3098 } 3099 3100 static const struct bpf_func_proto bpf_set_hash_proto = { 3101 .func = bpf_set_hash, 3102 .gpl_only = false, 3103 .ret_type = RET_INTEGER, 3104 .arg1_type = ARG_PTR_TO_CTX, 3105 .arg2_type = ARG_ANYTHING, 3106 }; 3107 3108 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto, 3109 u16, vlan_tci) 3110 { 3111 int ret; 3112 3113 if (unlikely(vlan_proto != htons(ETH_P_8021Q) && 3114 vlan_proto != htons(ETH_P_8021AD))) 3115 vlan_proto = htons(ETH_P_8021Q); 3116 3117 bpf_push_mac_rcsum(skb); 3118 ret = skb_vlan_push(skb, vlan_proto, vlan_tci); 3119 bpf_pull_mac_rcsum(skb); 3120 3121 bpf_compute_data_pointers(skb); 3122 return ret; 3123 } 3124 3125 static const struct bpf_func_proto bpf_skb_vlan_push_proto = { 3126 .func = bpf_skb_vlan_push, 3127 .gpl_only = false, 3128 .ret_type = RET_INTEGER, 3129 .arg1_type = ARG_PTR_TO_CTX, 3130 .arg2_type = ARG_ANYTHING, 3131 .arg3_type = ARG_ANYTHING, 3132 }; 3133 3134 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb) 3135 { 3136 int ret; 3137 3138 bpf_push_mac_rcsum(skb); 3139 ret = skb_vlan_pop(skb); 3140 bpf_pull_mac_rcsum(skb); 3141 3142 bpf_compute_data_pointers(skb); 3143 return ret; 3144 } 3145 3146 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = { 3147 .func = bpf_skb_vlan_pop, 3148 .gpl_only = false, 3149 .ret_type = RET_INTEGER, 3150 .arg1_type = ARG_PTR_TO_CTX, 3151 }; 3152 3153 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len) 3154 { 3155 /* Caller already did skb_cow() with len as headroom, 3156 * so no need to do it here. 3157 */ 3158 skb_push(skb, len); 3159 memmove(skb->data, skb->data + len, off); 3160 memset(skb->data + off, 0, len); 3161 3162 /* No skb_postpush_rcsum(skb, skb->data + off, len) 3163 * needed here as it does not change the skb->csum 3164 * result for checksum complete when summing over 3165 * zeroed blocks. 3166 */ 3167 return 0; 3168 } 3169 3170 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len) 3171 { 3172 /* skb_ensure_writable() is not needed here, as we're 3173 * already working on an uncloned skb. 3174 */ 3175 if (unlikely(!pskb_may_pull(skb, off + len))) 3176 return -ENOMEM; 3177 3178 skb_postpull_rcsum(skb, skb->data + off, len); 3179 memmove(skb->data + len, skb->data, off); 3180 __skb_pull(skb, len); 3181 3182 return 0; 3183 } 3184 3185 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len) 3186 { 3187 bool trans_same = skb->transport_header == skb->network_header; 3188 int ret; 3189 3190 /* There's no need for __skb_push()/__skb_pull() pair to 3191 * get to the start of the mac header as we're guaranteed 3192 * to always start from here under eBPF. 3193 */ 3194 ret = bpf_skb_generic_push(skb, off, len); 3195 if (likely(!ret)) { 3196 skb->mac_header -= len; 3197 skb->network_header -= len; 3198 if (trans_same) 3199 skb->transport_header = skb->network_header; 3200 } 3201 3202 return ret; 3203 } 3204 3205 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len) 3206 { 3207 bool trans_same = skb->transport_header == skb->network_header; 3208 int ret; 3209 3210 /* Same here, __skb_push()/__skb_pull() pair not needed. */ 3211 ret = bpf_skb_generic_pop(skb, off, len); 3212 if (likely(!ret)) { 3213 skb->mac_header += len; 3214 skb->network_header += len; 3215 if (trans_same) 3216 skb->transport_header = skb->network_header; 3217 } 3218 3219 return ret; 3220 } 3221 3222 static int bpf_skb_proto_4_to_6(struct sk_buff *skb) 3223 { 3224 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3225 u32 off = skb_mac_header_len(skb); 3226 int ret; 3227 3228 ret = skb_cow(skb, len_diff); 3229 if (unlikely(ret < 0)) 3230 return ret; 3231 3232 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3233 if (unlikely(ret < 0)) 3234 return ret; 3235 3236 if (skb_is_gso(skb)) { 3237 struct skb_shared_info *shinfo = skb_shinfo(skb); 3238 3239 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */ 3240 if (shinfo->gso_type & SKB_GSO_TCPV4) { 3241 shinfo->gso_type &= ~SKB_GSO_TCPV4; 3242 shinfo->gso_type |= SKB_GSO_TCPV6; 3243 } 3244 } 3245 3246 skb->protocol = htons(ETH_P_IPV6); 3247 skb_clear_hash(skb); 3248 3249 return 0; 3250 } 3251 3252 static int bpf_skb_proto_6_to_4(struct sk_buff *skb) 3253 { 3254 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3255 u32 off = skb_mac_header_len(skb); 3256 int ret; 3257 3258 ret = skb_unclone(skb, GFP_ATOMIC); 3259 if (unlikely(ret < 0)) 3260 return ret; 3261 3262 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3263 if (unlikely(ret < 0)) 3264 return ret; 3265 3266 if (skb_is_gso(skb)) { 3267 struct skb_shared_info *shinfo = skb_shinfo(skb); 3268 3269 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */ 3270 if (shinfo->gso_type & SKB_GSO_TCPV6) { 3271 shinfo->gso_type &= ~SKB_GSO_TCPV6; 3272 shinfo->gso_type |= SKB_GSO_TCPV4; 3273 } 3274 } 3275 3276 skb->protocol = htons(ETH_P_IP); 3277 skb_clear_hash(skb); 3278 3279 return 0; 3280 } 3281 3282 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto) 3283 { 3284 __be16 from_proto = skb->protocol; 3285 3286 if (from_proto == htons(ETH_P_IP) && 3287 to_proto == htons(ETH_P_IPV6)) 3288 return bpf_skb_proto_4_to_6(skb); 3289 3290 if (from_proto == htons(ETH_P_IPV6) && 3291 to_proto == htons(ETH_P_IP)) 3292 return bpf_skb_proto_6_to_4(skb); 3293 3294 return -ENOTSUPP; 3295 } 3296 3297 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto, 3298 u64, flags) 3299 { 3300 int ret; 3301 3302 if (unlikely(flags)) 3303 return -EINVAL; 3304 3305 /* General idea is that this helper does the basic groundwork 3306 * needed for changing the protocol, and eBPF program fills the 3307 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace() 3308 * and other helpers, rather than passing a raw buffer here. 3309 * 3310 * The rationale is to keep this minimal and without a need to 3311 * deal with raw packet data. F.e. even if we would pass buffers 3312 * here, the program still needs to call the bpf_lX_csum_replace() 3313 * helpers anyway. Plus, this way we keep also separation of 3314 * concerns, since f.e. bpf_skb_store_bytes() should only take 3315 * care of stores. 3316 * 3317 * Currently, additional options and extension header space are 3318 * not supported, but flags register is reserved so we can adapt 3319 * that. For offloads, we mark packet as dodgy, so that headers 3320 * need to be verified first. 3321 */ 3322 ret = bpf_skb_proto_xlat(skb, proto); 3323 bpf_compute_data_pointers(skb); 3324 return ret; 3325 } 3326 3327 static const struct bpf_func_proto bpf_skb_change_proto_proto = { 3328 .func = bpf_skb_change_proto, 3329 .gpl_only = false, 3330 .ret_type = RET_INTEGER, 3331 .arg1_type = ARG_PTR_TO_CTX, 3332 .arg2_type = ARG_ANYTHING, 3333 .arg3_type = ARG_ANYTHING, 3334 }; 3335 3336 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type) 3337 { 3338 /* We only allow a restricted subset to be changed for now. */ 3339 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) || 3340 !skb_pkt_type_ok(pkt_type))) 3341 return -EINVAL; 3342 3343 skb->pkt_type = pkt_type; 3344 return 0; 3345 } 3346 3347 static const struct bpf_func_proto bpf_skb_change_type_proto = { 3348 .func = bpf_skb_change_type, 3349 .gpl_only = false, 3350 .ret_type = RET_INTEGER, 3351 .arg1_type = ARG_PTR_TO_CTX, 3352 .arg2_type = ARG_ANYTHING, 3353 }; 3354 3355 static u32 bpf_skb_net_base_len(const struct sk_buff *skb) 3356 { 3357 switch (skb->protocol) { 3358 case htons(ETH_P_IP): 3359 return sizeof(struct iphdr); 3360 case htons(ETH_P_IPV6): 3361 return sizeof(struct ipv6hdr); 3362 default: 3363 return ~0U; 3364 } 3365 } 3366 3367 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \ 3368 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3369 3370 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \ 3371 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \ 3372 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \ 3373 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \ 3374 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \ 3375 BPF_F_ADJ_ROOM_ENCAP_L2( \ 3376 BPF_ADJ_ROOM_ENCAP_L2_MASK)) 3377 3378 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff, 3379 u64 flags) 3380 { 3381 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT; 3382 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK; 3383 u16 mac_len = 0, inner_net = 0, inner_trans = 0; 3384 unsigned int gso_type = SKB_GSO_DODGY; 3385 int ret; 3386 3387 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3388 /* udp gso_size delineates datagrams, only allow if fixed */ 3389 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3390 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3391 return -ENOTSUPP; 3392 } 3393 3394 ret = skb_cow_head(skb, len_diff); 3395 if (unlikely(ret < 0)) 3396 return ret; 3397 3398 if (encap) { 3399 if (skb->protocol != htons(ETH_P_IP) && 3400 skb->protocol != htons(ETH_P_IPV6)) 3401 return -ENOTSUPP; 3402 3403 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 && 3404 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3405 return -EINVAL; 3406 3407 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE && 3408 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3409 return -EINVAL; 3410 3411 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH && 3412 inner_mac_len < ETH_HLEN) 3413 return -EINVAL; 3414 3415 if (skb->encapsulation) 3416 return -EALREADY; 3417 3418 mac_len = skb->network_header - skb->mac_header; 3419 inner_net = skb->network_header; 3420 if (inner_mac_len > len_diff) 3421 return -EINVAL; 3422 inner_trans = skb->transport_header; 3423 } 3424 3425 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3426 if (unlikely(ret < 0)) 3427 return ret; 3428 3429 if (encap) { 3430 skb->inner_mac_header = inner_net - inner_mac_len; 3431 skb->inner_network_header = inner_net; 3432 skb->inner_transport_header = inner_trans; 3433 3434 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH) 3435 skb_set_inner_protocol(skb, htons(ETH_P_TEB)); 3436 else 3437 skb_set_inner_protocol(skb, skb->protocol); 3438 3439 skb->encapsulation = 1; 3440 skb_set_network_header(skb, mac_len); 3441 3442 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3443 gso_type |= SKB_GSO_UDP_TUNNEL; 3444 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE) 3445 gso_type |= SKB_GSO_GRE; 3446 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3447 gso_type |= SKB_GSO_IPXIP6; 3448 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3449 gso_type |= SKB_GSO_IPXIP4; 3450 3451 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE || 3452 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) { 3453 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ? 3454 sizeof(struct ipv6hdr) : 3455 sizeof(struct iphdr); 3456 3457 skb_set_transport_header(skb, mac_len + nh_len); 3458 } 3459 3460 /* Match skb->protocol to new outer l3 protocol */ 3461 if (skb->protocol == htons(ETH_P_IP) && 3462 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3463 skb->protocol = htons(ETH_P_IPV6); 3464 else if (skb->protocol == htons(ETH_P_IPV6) && 3465 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3466 skb->protocol = htons(ETH_P_IP); 3467 } 3468 3469 if (skb_is_gso(skb)) { 3470 struct skb_shared_info *shinfo = skb_shinfo(skb); 3471 3472 /* Due to header grow, MSS needs to be downgraded. */ 3473 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3474 skb_decrease_gso_size(shinfo, len_diff); 3475 3476 /* Header must be checked, and gso_segs recomputed. */ 3477 shinfo->gso_type |= gso_type; 3478 shinfo->gso_segs = 0; 3479 } 3480 3481 return 0; 3482 } 3483 3484 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff, 3485 u64 flags) 3486 { 3487 int ret; 3488 3489 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO | 3490 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3491 return -EINVAL; 3492 3493 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3494 /* udp gso_size delineates datagrams, only allow if fixed */ 3495 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3496 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3497 return -ENOTSUPP; 3498 } 3499 3500 ret = skb_unclone(skb, GFP_ATOMIC); 3501 if (unlikely(ret < 0)) 3502 return ret; 3503 3504 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3505 if (unlikely(ret < 0)) 3506 return ret; 3507 3508 if (skb_is_gso(skb)) { 3509 struct skb_shared_info *shinfo = skb_shinfo(skb); 3510 3511 /* Due to header shrink, MSS can be upgraded. */ 3512 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3513 skb_increase_gso_size(shinfo, len_diff); 3514 3515 /* Header must be checked, and gso_segs recomputed. */ 3516 shinfo->gso_type |= SKB_GSO_DODGY; 3517 shinfo->gso_segs = 0; 3518 } 3519 3520 return 0; 3521 } 3522 3523 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC 3524 3525 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3526 u32, mode, u64, flags) 3527 { 3528 u32 len_diff_abs = abs(len_diff); 3529 bool shrink = len_diff < 0; 3530 int ret = 0; 3531 3532 if (unlikely(flags || mode)) 3533 return -EINVAL; 3534 if (unlikely(len_diff_abs > 0xfffU)) 3535 return -EFAULT; 3536 3537 if (!shrink) { 3538 ret = skb_cow(skb, len_diff); 3539 if (unlikely(ret < 0)) 3540 return ret; 3541 __skb_push(skb, len_diff_abs); 3542 memset(skb->data, 0, len_diff_abs); 3543 } else { 3544 if (unlikely(!pskb_may_pull(skb, len_diff_abs))) 3545 return -ENOMEM; 3546 __skb_pull(skb, len_diff_abs); 3547 } 3548 if (tls_sw_has_ctx_rx(skb->sk)) { 3549 struct strp_msg *rxm = strp_msg(skb); 3550 3551 rxm->full_len += len_diff; 3552 } 3553 return ret; 3554 } 3555 3556 static const struct bpf_func_proto sk_skb_adjust_room_proto = { 3557 .func = sk_skb_adjust_room, 3558 .gpl_only = false, 3559 .ret_type = RET_INTEGER, 3560 .arg1_type = ARG_PTR_TO_CTX, 3561 .arg2_type = ARG_ANYTHING, 3562 .arg3_type = ARG_ANYTHING, 3563 .arg4_type = ARG_ANYTHING, 3564 }; 3565 3566 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3567 u32, mode, u64, flags) 3568 { 3569 u32 len_cur, len_diff_abs = abs(len_diff); 3570 u32 len_min = bpf_skb_net_base_len(skb); 3571 u32 len_max = BPF_SKB_MAX_LEN; 3572 __be16 proto = skb->protocol; 3573 bool shrink = len_diff < 0; 3574 u32 off; 3575 int ret; 3576 3577 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK | 3578 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3579 return -EINVAL; 3580 if (unlikely(len_diff_abs > 0xfffU)) 3581 return -EFAULT; 3582 if (unlikely(proto != htons(ETH_P_IP) && 3583 proto != htons(ETH_P_IPV6))) 3584 return -ENOTSUPP; 3585 3586 off = skb_mac_header_len(skb); 3587 switch (mode) { 3588 case BPF_ADJ_ROOM_NET: 3589 off += bpf_skb_net_base_len(skb); 3590 break; 3591 case BPF_ADJ_ROOM_MAC: 3592 break; 3593 default: 3594 return -ENOTSUPP; 3595 } 3596 3597 len_cur = skb->len - skb_network_offset(skb); 3598 if ((shrink && (len_diff_abs >= len_cur || 3599 len_cur - len_diff_abs < len_min)) || 3600 (!shrink && (skb->len + len_diff_abs > len_max && 3601 !skb_is_gso(skb)))) 3602 return -ENOTSUPP; 3603 3604 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) : 3605 bpf_skb_net_grow(skb, off, len_diff_abs, flags); 3606 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET)) 3607 __skb_reset_checksum_unnecessary(skb); 3608 3609 bpf_compute_data_pointers(skb); 3610 return ret; 3611 } 3612 3613 static const struct bpf_func_proto bpf_skb_adjust_room_proto = { 3614 .func = bpf_skb_adjust_room, 3615 .gpl_only = false, 3616 .ret_type = RET_INTEGER, 3617 .arg1_type = ARG_PTR_TO_CTX, 3618 .arg2_type = ARG_ANYTHING, 3619 .arg3_type = ARG_ANYTHING, 3620 .arg4_type = ARG_ANYTHING, 3621 }; 3622 3623 static u32 __bpf_skb_min_len(const struct sk_buff *skb) 3624 { 3625 u32 min_len = skb_network_offset(skb); 3626 3627 if (skb_transport_header_was_set(skb)) 3628 min_len = skb_transport_offset(skb); 3629 if (skb->ip_summed == CHECKSUM_PARTIAL) 3630 min_len = skb_checksum_start_offset(skb) + 3631 skb->csum_offset + sizeof(__sum16); 3632 return min_len; 3633 } 3634 3635 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len) 3636 { 3637 unsigned int old_len = skb->len; 3638 int ret; 3639 3640 ret = __skb_grow_rcsum(skb, new_len); 3641 if (!ret) 3642 memset(skb->data + old_len, 0, new_len - old_len); 3643 return ret; 3644 } 3645 3646 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len) 3647 { 3648 return __skb_trim_rcsum(skb, new_len); 3649 } 3650 3651 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len, 3652 u64 flags) 3653 { 3654 u32 max_len = BPF_SKB_MAX_LEN; 3655 u32 min_len = __bpf_skb_min_len(skb); 3656 int ret; 3657 3658 if (unlikely(flags || new_len > max_len || new_len < min_len)) 3659 return -EINVAL; 3660 if (skb->encapsulation) 3661 return -ENOTSUPP; 3662 3663 /* The basic idea of this helper is that it's performing the 3664 * needed work to either grow or trim an skb, and eBPF program 3665 * rewrites the rest via helpers like bpf_skb_store_bytes(), 3666 * bpf_lX_csum_replace() and others rather than passing a raw 3667 * buffer here. This one is a slow path helper and intended 3668 * for replies with control messages. 3669 * 3670 * Like in bpf_skb_change_proto(), we want to keep this rather 3671 * minimal and without protocol specifics so that we are able 3672 * to separate concerns as in bpf_skb_store_bytes() should only 3673 * be the one responsible for writing buffers. 3674 * 3675 * It's really expected to be a slow path operation here for 3676 * control message replies, so we're implicitly linearizing, 3677 * uncloning and drop offloads from the skb by this. 3678 */ 3679 ret = __bpf_try_make_writable(skb, skb->len); 3680 if (!ret) { 3681 if (new_len > skb->len) 3682 ret = bpf_skb_grow_rcsum(skb, new_len); 3683 else if (new_len < skb->len) 3684 ret = bpf_skb_trim_rcsum(skb, new_len); 3685 if (!ret && skb_is_gso(skb)) 3686 skb_gso_reset(skb); 3687 } 3688 return ret; 3689 } 3690 3691 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3692 u64, flags) 3693 { 3694 int ret = __bpf_skb_change_tail(skb, new_len, flags); 3695 3696 bpf_compute_data_pointers(skb); 3697 return ret; 3698 } 3699 3700 static const struct bpf_func_proto bpf_skb_change_tail_proto = { 3701 .func = bpf_skb_change_tail, 3702 .gpl_only = false, 3703 .ret_type = RET_INTEGER, 3704 .arg1_type = ARG_PTR_TO_CTX, 3705 .arg2_type = ARG_ANYTHING, 3706 .arg3_type = ARG_ANYTHING, 3707 }; 3708 3709 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3710 u64, flags) 3711 { 3712 return __bpf_skb_change_tail(skb, new_len, flags); 3713 } 3714 3715 static const struct bpf_func_proto sk_skb_change_tail_proto = { 3716 .func = sk_skb_change_tail, 3717 .gpl_only = false, 3718 .ret_type = RET_INTEGER, 3719 .arg1_type = ARG_PTR_TO_CTX, 3720 .arg2_type = ARG_ANYTHING, 3721 .arg3_type = ARG_ANYTHING, 3722 }; 3723 3724 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room, 3725 u64 flags) 3726 { 3727 u32 max_len = BPF_SKB_MAX_LEN; 3728 u32 new_len = skb->len + head_room; 3729 int ret; 3730 3731 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) || 3732 new_len < skb->len)) 3733 return -EINVAL; 3734 3735 ret = skb_cow(skb, head_room); 3736 if (likely(!ret)) { 3737 /* Idea for this helper is that we currently only 3738 * allow to expand on mac header. This means that 3739 * skb->protocol network header, etc, stay as is. 3740 * Compared to bpf_skb_change_tail(), we're more 3741 * flexible due to not needing to linearize or 3742 * reset GSO. Intention for this helper is to be 3743 * used by an L3 skb that needs to push mac header 3744 * for redirection into L2 device. 3745 */ 3746 __skb_push(skb, head_room); 3747 memset(skb->data, 0, head_room); 3748 skb_reset_mac_header(skb); 3749 skb_reset_mac_len(skb); 3750 } 3751 3752 return ret; 3753 } 3754 3755 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room, 3756 u64, flags) 3757 { 3758 int ret = __bpf_skb_change_head(skb, head_room, flags); 3759 3760 bpf_compute_data_pointers(skb); 3761 return ret; 3762 } 3763 3764 static const struct bpf_func_proto bpf_skb_change_head_proto = { 3765 .func = bpf_skb_change_head, 3766 .gpl_only = false, 3767 .ret_type = RET_INTEGER, 3768 .arg1_type = ARG_PTR_TO_CTX, 3769 .arg2_type = ARG_ANYTHING, 3770 .arg3_type = ARG_ANYTHING, 3771 }; 3772 3773 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room, 3774 u64, flags) 3775 { 3776 return __bpf_skb_change_head(skb, head_room, flags); 3777 } 3778 3779 static const struct bpf_func_proto sk_skb_change_head_proto = { 3780 .func = sk_skb_change_head, 3781 .gpl_only = false, 3782 .ret_type = RET_INTEGER, 3783 .arg1_type = ARG_PTR_TO_CTX, 3784 .arg2_type = ARG_ANYTHING, 3785 .arg3_type = ARG_ANYTHING, 3786 }; 3787 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp) 3788 { 3789 return xdp_data_meta_unsupported(xdp) ? 0 : 3790 xdp->data - xdp->data_meta; 3791 } 3792 3793 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset) 3794 { 3795 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 3796 unsigned long metalen = xdp_get_metalen(xdp); 3797 void *data_start = xdp_frame_end + metalen; 3798 void *data = xdp->data + offset; 3799 3800 if (unlikely(data < data_start || 3801 data > xdp->data_end - ETH_HLEN)) 3802 return -EINVAL; 3803 3804 if (metalen) 3805 memmove(xdp->data_meta + offset, 3806 xdp->data_meta, metalen); 3807 xdp->data_meta += offset; 3808 xdp->data = data; 3809 3810 return 0; 3811 } 3812 3813 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = { 3814 .func = bpf_xdp_adjust_head, 3815 .gpl_only = false, 3816 .ret_type = RET_INTEGER, 3817 .arg1_type = ARG_PTR_TO_CTX, 3818 .arg2_type = ARG_ANYTHING, 3819 }; 3820 3821 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset) 3822 { 3823 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */ 3824 void *data_end = xdp->data_end + offset; 3825 3826 /* Notice that xdp_data_hard_end have reserved some tailroom */ 3827 if (unlikely(data_end > data_hard_end)) 3828 return -EINVAL; 3829 3830 /* ALL drivers MUST init xdp->frame_sz, chicken check below */ 3831 if (unlikely(xdp->frame_sz > PAGE_SIZE)) { 3832 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz); 3833 return -EINVAL; 3834 } 3835 3836 if (unlikely(data_end < xdp->data + ETH_HLEN)) 3837 return -EINVAL; 3838 3839 /* Clear memory area on grow, can contain uninit kernel memory */ 3840 if (offset > 0) 3841 memset(xdp->data_end, 0, offset); 3842 3843 xdp->data_end = data_end; 3844 3845 return 0; 3846 } 3847 3848 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = { 3849 .func = bpf_xdp_adjust_tail, 3850 .gpl_only = false, 3851 .ret_type = RET_INTEGER, 3852 .arg1_type = ARG_PTR_TO_CTX, 3853 .arg2_type = ARG_ANYTHING, 3854 }; 3855 3856 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset) 3857 { 3858 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 3859 void *meta = xdp->data_meta + offset; 3860 unsigned long metalen = xdp->data - meta; 3861 3862 if (xdp_data_meta_unsupported(xdp)) 3863 return -ENOTSUPP; 3864 if (unlikely(meta < xdp_frame_end || 3865 meta > xdp->data)) 3866 return -EINVAL; 3867 if (unlikely(xdp_metalen_invalid(metalen))) 3868 return -EACCES; 3869 3870 xdp->data_meta = meta; 3871 3872 return 0; 3873 } 3874 3875 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = { 3876 .func = bpf_xdp_adjust_meta, 3877 .gpl_only = false, 3878 .ret_type = RET_INTEGER, 3879 .arg1_type = ARG_PTR_TO_CTX, 3880 .arg2_type = ARG_ANYTHING, 3881 }; 3882 3883 /* XDP_REDIRECT works by a three-step process, implemented in the functions 3884 * below: 3885 * 3886 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target 3887 * of the redirect and store it (along with some other metadata) in a per-CPU 3888 * struct bpf_redirect_info. 3889 * 3890 * 2. When the program returns the XDP_REDIRECT return code, the driver will 3891 * call xdp_do_redirect() which will use the information in struct 3892 * bpf_redirect_info to actually enqueue the frame into a map type-specific 3893 * bulk queue structure. 3894 * 3895 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(), 3896 * which will flush all the different bulk queues, thus completing the 3897 * redirect. 3898 * 3899 * Pointers to the map entries will be kept around for this whole sequence of 3900 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in 3901 * the core code; instead, the RCU protection relies on everything happening 3902 * inside a single NAPI poll sequence, which means it's between a pair of calls 3903 * to local_bh_disable()/local_bh_enable(). 3904 * 3905 * The map entries are marked as __rcu and the map code makes sure to 3906 * dereference those pointers with rcu_dereference_check() in a way that works 3907 * for both sections that to hold an rcu_read_lock() and sections that are 3908 * called from NAPI without a separate rcu_read_lock(). The code below does not 3909 * use RCU annotations, but relies on those in the map code. 3910 */ 3911 void xdp_do_flush(void) 3912 { 3913 __dev_flush(); 3914 __cpu_map_flush(); 3915 __xsk_map_flush(); 3916 } 3917 EXPORT_SYMBOL_GPL(xdp_do_flush); 3918 3919 void bpf_clear_redirect_map(struct bpf_map *map) 3920 { 3921 struct bpf_redirect_info *ri; 3922 int cpu; 3923 3924 for_each_possible_cpu(cpu) { 3925 ri = per_cpu_ptr(&bpf_redirect_info, cpu); 3926 /* Avoid polluting remote cacheline due to writes if 3927 * not needed. Once we pass this test, we need the 3928 * cmpxchg() to make sure it hasn't been changed in 3929 * the meantime by remote CPU. 3930 */ 3931 if (unlikely(READ_ONCE(ri->map) == map)) 3932 cmpxchg(&ri->map, map, NULL); 3933 } 3934 } 3935 3936 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 3937 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key); 3938 3939 u32 xdp_master_redirect(struct xdp_buff *xdp) 3940 { 3941 struct net_device *master, *slave; 3942 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3943 3944 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev); 3945 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp); 3946 if (slave && slave != xdp->rxq->dev) { 3947 /* The target device is different from the receiving device, so 3948 * redirect it to the new device. 3949 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled 3950 * drivers to unmap the packet from their rx ring. 3951 */ 3952 ri->tgt_index = slave->ifindex; 3953 ri->map_id = INT_MAX; 3954 ri->map_type = BPF_MAP_TYPE_UNSPEC; 3955 return XDP_REDIRECT; 3956 } 3957 return XDP_TX; 3958 } 3959 EXPORT_SYMBOL_GPL(xdp_master_redirect); 3960 3961 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp, 3962 struct bpf_prog *xdp_prog) 3963 { 3964 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3965 enum bpf_map_type map_type = ri->map_type; 3966 void *fwd = ri->tgt_value; 3967 u32 map_id = ri->map_id; 3968 struct bpf_map *map; 3969 int err; 3970 3971 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 3972 ri->map_type = BPF_MAP_TYPE_UNSPEC; 3973 3974 switch (map_type) { 3975 case BPF_MAP_TYPE_DEVMAP: 3976 fallthrough; 3977 case BPF_MAP_TYPE_DEVMAP_HASH: 3978 map = READ_ONCE(ri->map); 3979 if (unlikely(map)) { 3980 WRITE_ONCE(ri->map, NULL); 3981 err = dev_map_enqueue_multi(xdp, dev, map, 3982 ri->flags & BPF_F_EXCLUDE_INGRESS); 3983 } else { 3984 err = dev_map_enqueue(fwd, xdp, dev); 3985 } 3986 break; 3987 case BPF_MAP_TYPE_CPUMAP: 3988 err = cpu_map_enqueue(fwd, xdp, dev); 3989 break; 3990 case BPF_MAP_TYPE_XSKMAP: 3991 err = __xsk_map_redirect(fwd, xdp); 3992 break; 3993 case BPF_MAP_TYPE_UNSPEC: 3994 if (map_id == INT_MAX) { 3995 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index); 3996 if (unlikely(!fwd)) { 3997 err = -EINVAL; 3998 break; 3999 } 4000 err = dev_xdp_enqueue(fwd, xdp, dev); 4001 break; 4002 } 4003 fallthrough; 4004 default: 4005 err = -EBADRQC; 4006 } 4007 4008 if (unlikely(err)) 4009 goto err; 4010 4011 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4012 return 0; 4013 err: 4014 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4015 return err; 4016 } 4017 EXPORT_SYMBOL_GPL(xdp_do_redirect); 4018 4019 static int xdp_do_generic_redirect_map(struct net_device *dev, 4020 struct sk_buff *skb, 4021 struct xdp_buff *xdp, 4022 struct bpf_prog *xdp_prog, 4023 void *fwd, 4024 enum bpf_map_type map_type, u32 map_id) 4025 { 4026 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4027 struct bpf_map *map; 4028 int err; 4029 4030 switch (map_type) { 4031 case BPF_MAP_TYPE_DEVMAP: 4032 fallthrough; 4033 case BPF_MAP_TYPE_DEVMAP_HASH: 4034 map = READ_ONCE(ri->map); 4035 if (unlikely(map)) { 4036 WRITE_ONCE(ri->map, NULL); 4037 err = dev_map_redirect_multi(dev, skb, xdp_prog, map, 4038 ri->flags & BPF_F_EXCLUDE_INGRESS); 4039 } else { 4040 err = dev_map_generic_redirect(fwd, skb, xdp_prog); 4041 } 4042 if (unlikely(err)) 4043 goto err; 4044 break; 4045 case BPF_MAP_TYPE_XSKMAP: 4046 err = xsk_generic_rcv(fwd, xdp); 4047 if (err) 4048 goto err; 4049 consume_skb(skb); 4050 break; 4051 case BPF_MAP_TYPE_CPUMAP: 4052 err = cpu_map_generic_redirect(fwd, skb); 4053 if (unlikely(err)) 4054 goto err; 4055 break; 4056 default: 4057 err = -EBADRQC; 4058 goto err; 4059 } 4060 4061 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4062 return 0; 4063 err: 4064 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4065 return err; 4066 } 4067 4068 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 4069 struct xdp_buff *xdp, struct bpf_prog *xdp_prog) 4070 { 4071 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4072 enum bpf_map_type map_type = ri->map_type; 4073 void *fwd = ri->tgt_value; 4074 u32 map_id = ri->map_id; 4075 int err; 4076 4077 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4078 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4079 4080 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) { 4081 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index); 4082 if (unlikely(!fwd)) { 4083 err = -EINVAL; 4084 goto err; 4085 } 4086 4087 err = xdp_ok_fwd_dev(fwd, skb->len); 4088 if (unlikely(err)) 4089 goto err; 4090 4091 skb->dev = fwd; 4092 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index); 4093 generic_xdp_tx(skb, xdp_prog); 4094 return 0; 4095 } 4096 4097 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id); 4098 err: 4099 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err); 4100 return err; 4101 } 4102 4103 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags) 4104 { 4105 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4106 4107 if (unlikely(flags)) 4108 return XDP_ABORTED; 4109 4110 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated 4111 * by map_idr) is used for ifindex based XDP redirect. 4112 */ 4113 ri->tgt_index = ifindex; 4114 ri->map_id = INT_MAX; 4115 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4116 4117 return XDP_REDIRECT; 4118 } 4119 4120 static const struct bpf_func_proto bpf_xdp_redirect_proto = { 4121 .func = bpf_xdp_redirect, 4122 .gpl_only = false, 4123 .ret_type = RET_INTEGER, 4124 .arg1_type = ARG_ANYTHING, 4125 .arg2_type = ARG_ANYTHING, 4126 }; 4127 4128 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, 4129 u64, flags) 4130 { 4131 return map->ops->map_redirect(map, ifindex, flags); 4132 } 4133 4134 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = { 4135 .func = bpf_xdp_redirect_map, 4136 .gpl_only = false, 4137 .ret_type = RET_INTEGER, 4138 .arg1_type = ARG_CONST_MAP_PTR, 4139 .arg2_type = ARG_ANYTHING, 4140 .arg3_type = ARG_ANYTHING, 4141 }; 4142 4143 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb, 4144 unsigned long off, unsigned long len) 4145 { 4146 void *ptr = skb_header_pointer(skb, off, len, dst_buff); 4147 4148 if (unlikely(!ptr)) 4149 return len; 4150 if (ptr != dst_buff) 4151 memcpy(dst_buff, ptr, len); 4152 4153 return 0; 4154 } 4155 4156 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map, 4157 u64, flags, void *, meta, u64, meta_size) 4158 { 4159 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4160 4161 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4162 return -EINVAL; 4163 if (unlikely(!skb || skb_size > skb->len)) 4164 return -EFAULT; 4165 4166 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size, 4167 bpf_skb_copy); 4168 } 4169 4170 static const struct bpf_func_proto bpf_skb_event_output_proto = { 4171 .func = bpf_skb_event_output, 4172 .gpl_only = true, 4173 .ret_type = RET_INTEGER, 4174 .arg1_type = ARG_PTR_TO_CTX, 4175 .arg2_type = ARG_CONST_MAP_PTR, 4176 .arg3_type = ARG_ANYTHING, 4177 .arg4_type = ARG_PTR_TO_MEM, 4178 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4179 }; 4180 4181 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff) 4182 4183 const struct bpf_func_proto bpf_skb_output_proto = { 4184 .func = bpf_skb_event_output, 4185 .gpl_only = true, 4186 .ret_type = RET_INTEGER, 4187 .arg1_type = ARG_PTR_TO_BTF_ID, 4188 .arg1_btf_id = &bpf_skb_output_btf_ids[0], 4189 .arg2_type = ARG_CONST_MAP_PTR, 4190 .arg3_type = ARG_ANYTHING, 4191 .arg4_type = ARG_PTR_TO_MEM, 4192 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4193 }; 4194 4195 static unsigned short bpf_tunnel_key_af(u64 flags) 4196 { 4197 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET; 4198 } 4199 4200 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to, 4201 u32, size, u64, flags) 4202 { 4203 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4204 u8 compat[sizeof(struct bpf_tunnel_key)]; 4205 void *to_orig = to; 4206 int err; 4207 4208 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) { 4209 err = -EINVAL; 4210 goto err_clear; 4211 } 4212 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { 4213 err = -EPROTO; 4214 goto err_clear; 4215 } 4216 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4217 err = -EINVAL; 4218 switch (size) { 4219 case offsetof(struct bpf_tunnel_key, tunnel_label): 4220 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4221 goto set_compat; 4222 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4223 /* Fixup deprecated structure layouts here, so we have 4224 * a common path later on. 4225 */ 4226 if (ip_tunnel_info_af(info) != AF_INET) 4227 goto err_clear; 4228 set_compat: 4229 to = (struct bpf_tunnel_key *)compat; 4230 break; 4231 default: 4232 goto err_clear; 4233 } 4234 } 4235 4236 to->tunnel_id = be64_to_cpu(info->key.tun_id); 4237 to->tunnel_tos = info->key.tos; 4238 to->tunnel_ttl = info->key.ttl; 4239 to->tunnel_ext = 0; 4240 4241 if (flags & BPF_F_TUNINFO_IPV6) { 4242 memcpy(to->remote_ipv6, &info->key.u.ipv6.src, 4243 sizeof(to->remote_ipv6)); 4244 to->tunnel_label = be32_to_cpu(info->key.label); 4245 } else { 4246 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); 4247 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 4248 to->tunnel_label = 0; 4249 } 4250 4251 if (unlikely(size != sizeof(struct bpf_tunnel_key))) 4252 memcpy(to_orig, to, size); 4253 4254 return 0; 4255 err_clear: 4256 memset(to_orig, 0, size); 4257 return err; 4258 } 4259 4260 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { 4261 .func = bpf_skb_get_tunnel_key, 4262 .gpl_only = false, 4263 .ret_type = RET_INTEGER, 4264 .arg1_type = ARG_PTR_TO_CTX, 4265 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4266 .arg3_type = ARG_CONST_SIZE, 4267 .arg4_type = ARG_ANYTHING, 4268 }; 4269 4270 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size) 4271 { 4272 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4273 int err; 4274 4275 if (unlikely(!info || 4276 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) { 4277 err = -ENOENT; 4278 goto err_clear; 4279 } 4280 if (unlikely(size < info->options_len)) { 4281 err = -ENOMEM; 4282 goto err_clear; 4283 } 4284 4285 ip_tunnel_info_opts_get(to, info); 4286 if (size > info->options_len) 4287 memset(to + info->options_len, 0, size - info->options_len); 4288 4289 return info->options_len; 4290 err_clear: 4291 memset(to, 0, size); 4292 return err; 4293 } 4294 4295 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { 4296 .func = bpf_skb_get_tunnel_opt, 4297 .gpl_only = false, 4298 .ret_type = RET_INTEGER, 4299 .arg1_type = ARG_PTR_TO_CTX, 4300 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4301 .arg3_type = ARG_CONST_SIZE, 4302 }; 4303 4304 static struct metadata_dst __percpu *md_dst; 4305 4306 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb, 4307 const struct bpf_tunnel_key *, from, u32, size, u64, flags) 4308 { 4309 struct metadata_dst *md = this_cpu_ptr(md_dst); 4310 u8 compat[sizeof(struct bpf_tunnel_key)]; 4311 struct ip_tunnel_info *info; 4312 4313 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | 4314 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER))) 4315 return -EINVAL; 4316 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4317 switch (size) { 4318 case offsetof(struct bpf_tunnel_key, tunnel_label): 4319 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4320 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4321 /* Fixup deprecated structure layouts here, so we have 4322 * a common path later on. 4323 */ 4324 memcpy(compat, from, size); 4325 memset(compat + size, 0, sizeof(compat) - size); 4326 from = (const struct bpf_tunnel_key *) compat; 4327 break; 4328 default: 4329 return -EINVAL; 4330 } 4331 } 4332 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || 4333 from->tunnel_ext)) 4334 return -EINVAL; 4335 4336 skb_dst_drop(skb); 4337 dst_hold((struct dst_entry *) md); 4338 skb_dst_set(skb, (struct dst_entry *) md); 4339 4340 info = &md->u.tun_info; 4341 memset(info, 0, sizeof(*info)); 4342 info->mode = IP_TUNNEL_INFO_TX; 4343 4344 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE; 4345 if (flags & BPF_F_DONT_FRAGMENT) 4346 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT; 4347 if (flags & BPF_F_ZERO_CSUM_TX) 4348 info->key.tun_flags &= ~TUNNEL_CSUM; 4349 if (flags & BPF_F_SEQ_NUMBER) 4350 info->key.tun_flags |= TUNNEL_SEQ; 4351 4352 info->key.tun_id = cpu_to_be64(from->tunnel_id); 4353 info->key.tos = from->tunnel_tos; 4354 info->key.ttl = from->tunnel_ttl; 4355 4356 if (flags & BPF_F_TUNINFO_IPV6) { 4357 info->mode |= IP_TUNNEL_INFO_IPV6; 4358 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, 4359 sizeof(from->remote_ipv6)); 4360 info->key.label = cpu_to_be32(from->tunnel_label) & 4361 IPV6_FLOWLABEL_MASK; 4362 } else { 4363 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); 4364 } 4365 4366 return 0; 4367 } 4368 4369 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { 4370 .func = bpf_skb_set_tunnel_key, 4371 .gpl_only = false, 4372 .ret_type = RET_INTEGER, 4373 .arg1_type = ARG_PTR_TO_CTX, 4374 .arg2_type = ARG_PTR_TO_MEM, 4375 .arg3_type = ARG_CONST_SIZE, 4376 .arg4_type = ARG_ANYTHING, 4377 }; 4378 4379 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb, 4380 const u8 *, from, u32, size) 4381 { 4382 struct ip_tunnel_info *info = skb_tunnel_info(skb); 4383 const struct metadata_dst *md = this_cpu_ptr(md_dst); 4384 4385 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) 4386 return -EINVAL; 4387 if (unlikely(size > IP_TUNNEL_OPTS_MAX)) 4388 return -ENOMEM; 4389 4390 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT); 4391 4392 return 0; 4393 } 4394 4395 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { 4396 .func = bpf_skb_set_tunnel_opt, 4397 .gpl_only = false, 4398 .ret_type = RET_INTEGER, 4399 .arg1_type = ARG_PTR_TO_CTX, 4400 .arg2_type = ARG_PTR_TO_MEM, 4401 .arg3_type = ARG_CONST_SIZE, 4402 }; 4403 4404 static const struct bpf_func_proto * 4405 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) 4406 { 4407 if (!md_dst) { 4408 struct metadata_dst __percpu *tmp; 4409 4410 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, 4411 METADATA_IP_TUNNEL, 4412 GFP_KERNEL); 4413 if (!tmp) 4414 return NULL; 4415 if (cmpxchg(&md_dst, NULL, tmp)) 4416 metadata_dst_free_percpu(tmp); 4417 } 4418 4419 switch (which) { 4420 case BPF_FUNC_skb_set_tunnel_key: 4421 return &bpf_skb_set_tunnel_key_proto; 4422 case BPF_FUNC_skb_set_tunnel_opt: 4423 return &bpf_skb_set_tunnel_opt_proto; 4424 default: 4425 return NULL; 4426 } 4427 } 4428 4429 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map, 4430 u32, idx) 4431 { 4432 struct bpf_array *array = container_of(map, struct bpf_array, map); 4433 struct cgroup *cgrp; 4434 struct sock *sk; 4435 4436 sk = skb_to_full_sk(skb); 4437 if (!sk || !sk_fullsock(sk)) 4438 return -ENOENT; 4439 if (unlikely(idx >= array->map.max_entries)) 4440 return -E2BIG; 4441 4442 cgrp = READ_ONCE(array->ptrs[idx]); 4443 if (unlikely(!cgrp)) 4444 return -EAGAIN; 4445 4446 return sk_under_cgroup_hierarchy(sk, cgrp); 4447 } 4448 4449 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = { 4450 .func = bpf_skb_under_cgroup, 4451 .gpl_only = false, 4452 .ret_type = RET_INTEGER, 4453 .arg1_type = ARG_PTR_TO_CTX, 4454 .arg2_type = ARG_CONST_MAP_PTR, 4455 .arg3_type = ARG_ANYTHING, 4456 }; 4457 4458 #ifdef CONFIG_SOCK_CGROUP_DATA 4459 static inline u64 __bpf_sk_cgroup_id(struct sock *sk) 4460 { 4461 struct cgroup *cgrp; 4462 4463 sk = sk_to_full_sk(sk); 4464 if (!sk || !sk_fullsock(sk)) 4465 return 0; 4466 4467 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4468 return cgroup_id(cgrp); 4469 } 4470 4471 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb) 4472 { 4473 return __bpf_sk_cgroup_id(skb->sk); 4474 } 4475 4476 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = { 4477 .func = bpf_skb_cgroup_id, 4478 .gpl_only = false, 4479 .ret_type = RET_INTEGER, 4480 .arg1_type = ARG_PTR_TO_CTX, 4481 }; 4482 4483 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk, 4484 int ancestor_level) 4485 { 4486 struct cgroup *ancestor; 4487 struct cgroup *cgrp; 4488 4489 sk = sk_to_full_sk(sk); 4490 if (!sk || !sk_fullsock(sk)) 4491 return 0; 4492 4493 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4494 ancestor = cgroup_ancestor(cgrp, ancestor_level); 4495 if (!ancestor) 4496 return 0; 4497 4498 return cgroup_id(ancestor); 4499 } 4500 4501 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int, 4502 ancestor_level) 4503 { 4504 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level); 4505 } 4506 4507 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = { 4508 .func = bpf_skb_ancestor_cgroup_id, 4509 .gpl_only = false, 4510 .ret_type = RET_INTEGER, 4511 .arg1_type = ARG_PTR_TO_CTX, 4512 .arg2_type = ARG_ANYTHING, 4513 }; 4514 4515 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk) 4516 { 4517 return __bpf_sk_cgroup_id(sk); 4518 } 4519 4520 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = { 4521 .func = bpf_sk_cgroup_id, 4522 .gpl_only = false, 4523 .ret_type = RET_INTEGER, 4524 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4525 }; 4526 4527 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level) 4528 { 4529 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level); 4530 } 4531 4532 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = { 4533 .func = bpf_sk_ancestor_cgroup_id, 4534 .gpl_only = false, 4535 .ret_type = RET_INTEGER, 4536 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4537 .arg2_type = ARG_ANYTHING, 4538 }; 4539 #endif 4540 4541 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff, 4542 unsigned long off, unsigned long len) 4543 { 4544 memcpy(dst_buff, src_buff + off, len); 4545 return 0; 4546 } 4547 4548 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map, 4549 u64, flags, void *, meta, u64, meta_size) 4550 { 4551 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4552 4553 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4554 return -EINVAL; 4555 if (unlikely(!xdp || 4556 xdp_size > (unsigned long)(xdp->data_end - xdp->data))) 4557 return -EFAULT; 4558 4559 return bpf_event_output(map, flags, meta, meta_size, xdp->data, 4560 xdp_size, bpf_xdp_copy); 4561 } 4562 4563 static const struct bpf_func_proto bpf_xdp_event_output_proto = { 4564 .func = bpf_xdp_event_output, 4565 .gpl_only = true, 4566 .ret_type = RET_INTEGER, 4567 .arg1_type = ARG_PTR_TO_CTX, 4568 .arg2_type = ARG_CONST_MAP_PTR, 4569 .arg3_type = ARG_ANYTHING, 4570 .arg4_type = ARG_PTR_TO_MEM, 4571 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4572 }; 4573 4574 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff) 4575 4576 const struct bpf_func_proto bpf_xdp_output_proto = { 4577 .func = bpf_xdp_event_output, 4578 .gpl_only = true, 4579 .ret_type = RET_INTEGER, 4580 .arg1_type = ARG_PTR_TO_BTF_ID, 4581 .arg1_btf_id = &bpf_xdp_output_btf_ids[0], 4582 .arg2_type = ARG_CONST_MAP_PTR, 4583 .arg3_type = ARG_ANYTHING, 4584 .arg4_type = ARG_PTR_TO_MEM, 4585 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4586 }; 4587 4588 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb) 4589 { 4590 return skb->sk ? __sock_gen_cookie(skb->sk) : 0; 4591 } 4592 4593 static const struct bpf_func_proto bpf_get_socket_cookie_proto = { 4594 .func = bpf_get_socket_cookie, 4595 .gpl_only = false, 4596 .ret_type = RET_INTEGER, 4597 .arg1_type = ARG_PTR_TO_CTX, 4598 }; 4599 4600 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4601 { 4602 return __sock_gen_cookie(ctx->sk); 4603 } 4604 4605 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = { 4606 .func = bpf_get_socket_cookie_sock_addr, 4607 .gpl_only = false, 4608 .ret_type = RET_INTEGER, 4609 .arg1_type = ARG_PTR_TO_CTX, 4610 }; 4611 4612 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx) 4613 { 4614 return __sock_gen_cookie(ctx); 4615 } 4616 4617 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = { 4618 .func = bpf_get_socket_cookie_sock, 4619 .gpl_only = false, 4620 .ret_type = RET_INTEGER, 4621 .arg1_type = ARG_PTR_TO_CTX, 4622 }; 4623 4624 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk) 4625 { 4626 return sk ? sock_gen_cookie(sk) : 0; 4627 } 4628 4629 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = { 4630 .func = bpf_get_socket_ptr_cookie, 4631 .gpl_only = false, 4632 .ret_type = RET_INTEGER, 4633 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4634 }; 4635 4636 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 4637 { 4638 return __sock_gen_cookie(ctx->sk); 4639 } 4640 4641 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = { 4642 .func = bpf_get_socket_cookie_sock_ops, 4643 .gpl_only = false, 4644 .ret_type = RET_INTEGER, 4645 .arg1_type = ARG_PTR_TO_CTX, 4646 }; 4647 4648 static u64 __bpf_get_netns_cookie(struct sock *sk) 4649 { 4650 const struct net *net = sk ? sock_net(sk) : &init_net; 4651 4652 return net->net_cookie; 4653 } 4654 4655 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx) 4656 { 4657 return __bpf_get_netns_cookie(ctx); 4658 } 4659 4660 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = { 4661 .func = bpf_get_netns_cookie_sock, 4662 .gpl_only = false, 4663 .ret_type = RET_INTEGER, 4664 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4665 }; 4666 4667 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4668 { 4669 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4670 } 4671 4672 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = { 4673 .func = bpf_get_netns_cookie_sock_addr, 4674 .gpl_only = false, 4675 .ret_type = RET_INTEGER, 4676 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4677 }; 4678 4679 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 4680 { 4681 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4682 } 4683 4684 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = { 4685 .func = bpf_get_netns_cookie_sock_ops, 4686 .gpl_only = false, 4687 .ret_type = RET_INTEGER, 4688 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4689 }; 4690 4691 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx) 4692 { 4693 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4694 } 4695 4696 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = { 4697 .func = bpf_get_netns_cookie_sk_msg, 4698 .gpl_only = false, 4699 .ret_type = RET_INTEGER, 4700 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4701 }; 4702 4703 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb) 4704 { 4705 struct sock *sk = sk_to_full_sk(skb->sk); 4706 kuid_t kuid; 4707 4708 if (!sk || !sk_fullsock(sk)) 4709 return overflowuid; 4710 kuid = sock_net_uid(sock_net(sk), sk); 4711 return from_kuid_munged(sock_net(sk)->user_ns, kuid); 4712 } 4713 4714 static const struct bpf_func_proto bpf_get_socket_uid_proto = { 4715 .func = bpf_get_socket_uid, 4716 .gpl_only = false, 4717 .ret_type = RET_INTEGER, 4718 .arg1_type = ARG_PTR_TO_CTX, 4719 }; 4720 4721 static int _bpf_setsockopt(struct sock *sk, int level, int optname, 4722 char *optval, int optlen) 4723 { 4724 char devname[IFNAMSIZ]; 4725 int val, valbool; 4726 struct net *net; 4727 int ifindex; 4728 int ret = 0; 4729 4730 if (!sk_fullsock(sk)) 4731 return -EINVAL; 4732 4733 sock_owned_by_me(sk); 4734 4735 if (level == SOL_SOCKET) { 4736 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE) 4737 return -EINVAL; 4738 val = *((int *)optval); 4739 valbool = val ? 1 : 0; 4740 4741 /* Only some socketops are supported */ 4742 switch (optname) { 4743 case SO_RCVBUF: 4744 val = min_t(u32, val, sysctl_rmem_max); 4745 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 4746 WRITE_ONCE(sk->sk_rcvbuf, 4747 max_t(int, val * 2, SOCK_MIN_RCVBUF)); 4748 break; 4749 case SO_SNDBUF: 4750 val = min_t(u32, val, sysctl_wmem_max); 4751 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 4752 WRITE_ONCE(sk->sk_sndbuf, 4753 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 4754 break; 4755 case SO_MAX_PACING_RATE: /* 32bit version */ 4756 if (val != ~0U) 4757 cmpxchg(&sk->sk_pacing_status, 4758 SK_PACING_NONE, 4759 SK_PACING_NEEDED); 4760 sk->sk_max_pacing_rate = (val == ~0U) ? 4761 ~0UL : (unsigned int)val; 4762 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 4763 sk->sk_max_pacing_rate); 4764 break; 4765 case SO_PRIORITY: 4766 sk->sk_priority = val; 4767 break; 4768 case SO_RCVLOWAT: 4769 if (val < 0) 4770 val = INT_MAX; 4771 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 4772 break; 4773 case SO_MARK: 4774 if (sk->sk_mark != val) { 4775 sk->sk_mark = val; 4776 sk_dst_reset(sk); 4777 } 4778 break; 4779 case SO_BINDTODEVICE: 4780 optlen = min_t(long, optlen, IFNAMSIZ - 1); 4781 strncpy(devname, optval, optlen); 4782 devname[optlen] = 0; 4783 4784 ifindex = 0; 4785 if (devname[0] != '\0') { 4786 struct net_device *dev; 4787 4788 ret = -ENODEV; 4789 4790 net = sock_net(sk); 4791 dev = dev_get_by_name(net, devname); 4792 if (!dev) 4793 break; 4794 ifindex = dev->ifindex; 4795 dev_put(dev); 4796 } 4797 fallthrough; 4798 case SO_BINDTOIFINDEX: 4799 if (optname == SO_BINDTOIFINDEX) 4800 ifindex = val; 4801 ret = sock_bindtoindex(sk, ifindex, false); 4802 break; 4803 case SO_KEEPALIVE: 4804 if (sk->sk_prot->keepalive) 4805 sk->sk_prot->keepalive(sk, valbool); 4806 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 4807 break; 4808 case SO_REUSEPORT: 4809 sk->sk_reuseport = valbool; 4810 break; 4811 default: 4812 ret = -EINVAL; 4813 } 4814 #ifdef CONFIG_INET 4815 } else if (level == SOL_IP) { 4816 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 4817 return -EINVAL; 4818 4819 val = *((int *)optval); 4820 /* Only some options are supported */ 4821 switch (optname) { 4822 case IP_TOS: 4823 if (val < -1 || val > 0xff) { 4824 ret = -EINVAL; 4825 } else { 4826 struct inet_sock *inet = inet_sk(sk); 4827 4828 if (val == -1) 4829 val = 0; 4830 inet->tos = val; 4831 } 4832 break; 4833 default: 4834 ret = -EINVAL; 4835 } 4836 #if IS_ENABLED(CONFIG_IPV6) 4837 } else if (level == SOL_IPV6) { 4838 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 4839 return -EINVAL; 4840 4841 val = *((int *)optval); 4842 /* Only some options are supported */ 4843 switch (optname) { 4844 case IPV6_TCLASS: 4845 if (val < -1 || val > 0xff) { 4846 ret = -EINVAL; 4847 } else { 4848 struct ipv6_pinfo *np = inet6_sk(sk); 4849 4850 if (val == -1) 4851 val = 0; 4852 np->tclass = val; 4853 } 4854 break; 4855 default: 4856 ret = -EINVAL; 4857 } 4858 #endif 4859 } else if (level == SOL_TCP && 4860 sk->sk_prot->setsockopt == tcp_setsockopt) { 4861 if (optname == TCP_CONGESTION) { 4862 char name[TCP_CA_NAME_MAX]; 4863 4864 strncpy(name, optval, min_t(long, optlen, 4865 TCP_CA_NAME_MAX-1)); 4866 name[TCP_CA_NAME_MAX-1] = 0; 4867 ret = tcp_set_congestion_control(sk, name, false, true); 4868 } else { 4869 struct inet_connection_sock *icsk = inet_csk(sk); 4870 struct tcp_sock *tp = tcp_sk(sk); 4871 unsigned long timeout; 4872 4873 if (optlen != sizeof(int)) 4874 return -EINVAL; 4875 4876 val = *((int *)optval); 4877 /* Only some options are supported */ 4878 switch (optname) { 4879 case TCP_BPF_IW: 4880 if (val <= 0 || tp->data_segs_out > tp->syn_data) 4881 ret = -EINVAL; 4882 else 4883 tp->snd_cwnd = val; 4884 break; 4885 case TCP_BPF_SNDCWND_CLAMP: 4886 if (val <= 0) { 4887 ret = -EINVAL; 4888 } else { 4889 tp->snd_cwnd_clamp = val; 4890 tp->snd_ssthresh = val; 4891 } 4892 break; 4893 case TCP_BPF_DELACK_MAX: 4894 timeout = usecs_to_jiffies(val); 4895 if (timeout > TCP_DELACK_MAX || 4896 timeout < TCP_TIMEOUT_MIN) 4897 return -EINVAL; 4898 inet_csk(sk)->icsk_delack_max = timeout; 4899 break; 4900 case TCP_BPF_RTO_MIN: 4901 timeout = usecs_to_jiffies(val); 4902 if (timeout > TCP_RTO_MIN || 4903 timeout < TCP_TIMEOUT_MIN) 4904 return -EINVAL; 4905 inet_csk(sk)->icsk_rto_min = timeout; 4906 break; 4907 case TCP_SAVE_SYN: 4908 if (val < 0 || val > 1) 4909 ret = -EINVAL; 4910 else 4911 tp->save_syn = val; 4912 break; 4913 case TCP_KEEPIDLE: 4914 ret = tcp_sock_set_keepidle_locked(sk, val); 4915 break; 4916 case TCP_KEEPINTVL: 4917 if (val < 1 || val > MAX_TCP_KEEPINTVL) 4918 ret = -EINVAL; 4919 else 4920 tp->keepalive_intvl = val * HZ; 4921 break; 4922 case TCP_KEEPCNT: 4923 if (val < 1 || val > MAX_TCP_KEEPCNT) 4924 ret = -EINVAL; 4925 else 4926 tp->keepalive_probes = val; 4927 break; 4928 case TCP_SYNCNT: 4929 if (val < 1 || val > MAX_TCP_SYNCNT) 4930 ret = -EINVAL; 4931 else 4932 icsk->icsk_syn_retries = val; 4933 break; 4934 case TCP_USER_TIMEOUT: 4935 if (val < 0) 4936 ret = -EINVAL; 4937 else 4938 icsk->icsk_user_timeout = val; 4939 break; 4940 case TCP_NOTSENT_LOWAT: 4941 tp->notsent_lowat = val; 4942 sk->sk_write_space(sk); 4943 break; 4944 case TCP_WINDOW_CLAMP: 4945 ret = tcp_set_window_clamp(sk, val); 4946 break; 4947 default: 4948 ret = -EINVAL; 4949 } 4950 } 4951 #endif 4952 } else { 4953 ret = -EINVAL; 4954 } 4955 return ret; 4956 } 4957 4958 static int _bpf_getsockopt(struct sock *sk, int level, int optname, 4959 char *optval, int optlen) 4960 { 4961 if (!sk_fullsock(sk)) 4962 goto err_clear; 4963 4964 sock_owned_by_me(sk); 4965 4966 if (level == SOL_SOCKET) { 4967 if (optlen != sizeof(int)) 4968 goto err_clear; 4969 4970 switch (optname) { 4971 case SO_MARK: 4972 *((int *)optval) = sk->sk_mark; 4973 break; 4974 case SO_PRIORITY: 4975 *((int *)optval) = sk->sk_priority; 4976 break; 4977 case SO_BINDTOIFINDEX: 4978 *((int *)optval) = sk->sk_bound_dev_if; 4979 break; 4980 case SO_REUSEPORT: 4981 *((int *)optval) = sk->sk_reuseport; 4982 break; 4983 default: 4984 goto err_clear; 4985 } 4986 #ifdef CONFIG_INET 4987 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) { 4988 struct inet_connection_sock *icsk; 4989 struct tcp_sock *tp; 4990 4991 switch (optname) { 4992 case TCP_CONGESTION: 4993 icsk = inet_csk(sk); 4994 4995 if (!icsk->icsk_ca_ops || optlen <= 1) 4996 goto err_clear; 4997 strncpy(optval, icsk->icsk_ca_ops->name, optlen); 4998 optval[optlen - 1] = 0; 4999 break; 5000 case TCP_SAVED_SYN: 5001 tp = tcp_sk(sk); 5002 5003 if (optlen <= 0 || !tp->saved_syn || 5004 optlen > tcp_saved_syn_len(tp->saved_syn)) 5005 goto err_clear; 5006 memcpy(optval, tp->saved_syn->data, optlen); 5007 break; 5008 default: 5009 goto err_clear; 5010 } 5011 } else if (level == SOL_IP) { 5012 struct inet_sock *inet = inet_sk(sk); 5013 5014 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 5015 goto err_clear; 5016 5017 /* Only some options are supported */ 5018 switch (optname) { 5019 case IP_TOS: 5020 *((int *)optval) = (int)inet->tos; 5021 break; 5022 default: 5023 goto err_clear; 5024 } 5025 #if IS_ENABLED(CONFIG_IPV6) 5026 } else if (level == SOL_IPV6) { 5027 struct ipv6_pinfo *np = inet6_sk(sk); 5028 5029 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 5030 goto err_clear; 5031 5032 /* Only some options are supported */ 5033 switch (optname) { 5034 case IPV6_TCLASS: 5035 *((int *)optval) = (int)np->tclass; 5036 break; 5037 default: 5038 goto err_clear; 5039 } 5040 #endif 5041 #endif 5042 } else { 5043 goto err_clear; 5044 } 5045 return 0; 5046 err_clear: 5047 memset(optval, 0, optlen); 5048 return -EINVAL; 5049 } 5050 5051 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level, 5052 int, optname, char *, optval, int, optlen) 5053 { 5054 if (level == SOL_TCP && optname == TCP_CONGESTION) { 5055 if (optlen >= sizeof("cdg") - 1 && 5056 !strncmp("cdg", optval, optlen)) 5057 return -ENOTSUPP; 5058 } 5059 5060 return _bpf_setsockopt(sk, level, optname, optval, optlen); 5061 } 5062 5063 const struct bpf_func_proto bpf_sk_setsockopt_proto = { 5064 .func = bpf_sk_setsockopt, 5065 .gpl_only = false, 5066 .ret_type = RET_INTEGER, 5067 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5068 .arg2_type = ARG_ANYTHING, 5069 .arg3_type = ARG_ANYTHING, 5070 .arg4_type = ARG_PTR_TO_MEM, 5071 .arg5_type = ARG_CONST_SIZE, 5072 }; 5073 5074 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level, 5075 int, optname, char *, optval, int, optlen) 5076 { 5077 return _bpf_getsockopt(sk, level, optname, optval, optlen); 5078 } 5079 5080 const struct bpf_func_proto bpf_sk_getsockopt_proto = { 5081 .func = bpf_sk_getsockopt, 5082 .gpl_only = false, 5083 .ret_type = RET_INTEGER, 5084 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5085 .arg2_type = ARG_ANYTHING, 5086 .arg3_type = ARG_ANYTHING, 5087 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5088 .arg5_type = ARG_CONST_SIZE, 5089 }; 5090 5091 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx, 5092 int, level, int, optname, char *, optval, int, optlen) 5093 { 5094 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen); 5095 } 5096 5097 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = { 5098 .func = bpf_sock_addr_setsockopt, 5099 .gpl_only = false, 5100 .ret_type = RET_INTEGER, 5101 .arg1_type = ARG_PTR_TO_CTX, 5102 .arg2_type = ARG_ANYTHING, 5103 .arg3_type = ARG_ANYTHING, 5104 .arg4_type = ARG_PTR_TO_MEM, 5105 .arg5_type = ARG_CONST_SIZE, 5106 }; 5107 5108 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx, 5109 int, level, int, optname, char *, optval, int, optlen) 5110 { 5111 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen); 5112 } 5113 5114 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = { 5115 .func = bpf_sock_addr_getsockopt, 5116 .gpl_only = false, 5117 .ret_type = RET_INTEGER, 5118 .arg1_type = ARG_PTR_TO_CTX, 5119 .arg2_type = ARG_ANYTHING, 5120 .arg3_type = ARG_ANYTHING, 5121 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5122 .arg5_type = ARG_CONST_SIZE, 5123 }; 5124 5125 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5126 int, level, int, optname, char *, optval, int, optlen) 5127 { 5128 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen); 5129 } 5130 5131 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = { 5132 .func = bpf_sock_ops_setsockopt, 5133 .gpl_only = false, 5134 .ret_type = RET_INTEGER, 5135 .arg1_type = ARG_PTR_TO_CTX, 5136 .arg2_type = ARG_ANYTHING, 5137 .arg3_type = ARG_ANYTHING, 5138 .arg4_type = ARG_PTR_TO_MEM, 5139 .arg5_type = ARG_CONST_SIZE, 5140 }; 5141 5142 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock, 5143 int optname, const u8 **start) 5144 { 5145 struct sk_buff *syn_skb = bpf_sock->syn_skb; 5146 const u8 *hdr_start; 5147 int ret; 5148 5149 if (syn_skb) { 5150 /* sk is a request_sock here */ 5151 5152 if (optname == TCP_BPF_SYN) { 5153 hdr_start = syn_skb->data; 5154 ret = tcp_hdrlen(syn_skb); 5155 } else if (optname == TCP_BPF_SYN_IP) { 5156 hdr_start = skb_network_header(syn_skb); 5157 ret = skb_network_header_len(syn_skb) + 5158 tcp_hdrlen(syn_skb); 5159 } else { 5160 /* optname == TCP_BPF_SYN_MAC */ 5161 hdr_start = skb_mac_header(syn_skb); 5162 ret = skb_mac_header_len(syn_skb) + 5163 skb_network_header_len(syn_skb) + 5164 tcp_hdrlen(syn_skb); 5165 } 5166 } else { 5167 struct sock *sk = bpf_sock->sk; 5168 struct saved_syn *saved_syn; 5169 5170 if (sk->sk_state == TCP_NEW_SYN_RECV) 5171 /* synack retransmit. bpf_sock->syn_skb will 5172 * not be available. It has to resort to 5173 * saved_syn (if it is saved). 5174 */ 5175 saved_syn = inet_reqsk(sk)->saved_syn; 5176 else 5177 saved_syn = tcp_sk(sk)->saved_syn; 5178 5179 if (!saved_syn) 5180 return -ENOENT; 5181 5182 if (optname == TCP_BPF_SYN) { 5183 hdr_start = saved_syn->data + 5184 saved_syn->mac_hdrlen + 5185 saved_syn->network_hdrlen; 5186 ret = saved_syn->tcp_hdrlen; 5187 } else if (optname == TCP_BPF_SYN_IP) { 5188 hdr_start = saved_syn->data + 5189 saved_syn->mac_hdrlen; 5190 ret = saved_syn->network_hdrlen + 5191 saved_syn->tcp_hdrlen; 5192 } else { 5193 /* optname == TCP_BPF_SYN_MAC */ 5194 5195 /* TCP_SAVE_SYN may not have saved the mac hdr */ 5196 if (!saved_syn->mac_hdrlen) 5197 return -ENOENT; 5198 5199 hdr_start = saved_syn->data; 5200 ret = saved_syn->mac_hdrlen + 5201 saved_syn->network_hdrlen + 5202 saved_syn->tcp_hdrlen; 5203 } 5204 } 5205 5206 *start = hdr_start; 5207 return ret; 5208 } 5209 5210 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5211 int, level, int, optname, char *, optval, int, optlen) 5212 { 5213 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP && 5214 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) { 5215 int ret, copy_len = 0; 5216 const u8 *start; 5217 5218 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start); 5219 if (ret > 0) { 5220 copy_len = ret; 5221 if (optlen < copy_len) { 5222 copy_len = optlen; 5223 ret = -ENOSPC; 5224 } 5225 5226 memcpy(optval, start, copy_len); 5227 } 5228 5229 /* Zero out unused buffer at the end */ 5230 memset(optval + copy_len, 0, optlen - copy_len); 5231 5232 return ret; 5233 } 5234 5235 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen); 5236 } 5237 5238 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = { 5239 .func = bpf_sock_ops_getsockopt, 5240 .gpl_only = false, 5241 .ret_type = RET_INTEGER, 5242 .arg1_type = ARG_PTR_TO_CTX, 5243 .arg2_type = ARG_ANYTHING, 5244 .arg3_type = ARG_ANYTHING, 5245 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5246 .arg5_type = ARG_CONST_SIZE, 5247 }; 5248 5249 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock, 5250 int, argval) 5251 { 5252 struct sock *sk = bpf_sock->sk; 5253 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS; 5254 5255 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk)) 5256 return -EINVAL; 5257 5258 tcp_sk(sk)->bpf_sock_ops_cb_flags = val; 5259 5260 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS); 5261 } 5262 5263 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = { 5264 .func = bpf_sock_ops_cb_flags_set, 5265 .gpl_only = false, 5266 .ret_type = RET_INTEGER, 5267 .arg1_type = ARG_PTR_TO_CTX, 5268 .arg2_type = ARG_ANYTHING, 5269 }; 5270 5271 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly; 5272 EXPORT_SYMBOL_GPL(ipv6_bpf_stub); 5273 5274 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr, 5275 int, addr_len) 5276 { 5277 #ifdef CONFIG_INET 5278 struct sock *sk = ctx->sk; 5279 u32 flags = BIND_FROM_BPF; 5280 int err; 5281 5282 err = -EINVAL; 5283 if (addr_len < offsetofend(struct sockaddr, sa_family)) 5284 return err; 5285 if (addr->sa_family == AF_INET) { 5286 if (addr_len < sizeof(struct sockaddr_in)) 5287 return err; 5288 if (((struct sockaddr_in *)addr)->sin_port == htons(0)) 5289 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5290 return __inet_bind(sk, addr, addr_len, flags); 5291 #if IS_ENABLED(CONFIG_IPV6) 5292 } else if (addr->sa_family == AF_INET6) { 5293 if (addr_len < SIN6_LEN_RFC2133) 5294 return err; 5295 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0)) 5296 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5297 /* ipv6_bpf_stub cannot be NULL, since it's called from 5298 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded 5299 */ 5300 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags); 5301 #endif /* CONFIG_IPV6 */ 5302 } 5303 #endif /* CONFIG_INET */ 5304 5305 return -EAFNOSUPPORT; 5306 } 5307 5308 static const struct bpf_func_proto bpf_bind_proto = { 5309 .func = bpf_bind, 5310 .gpl_only = false, 5311 .ret_type = RET_INTEGER, 5312 .arg1_type = ARG_PTR_TO_CTX, 5313 .arg2_type = ARG_PTR_TO_MEM, 5314 .arg3_type = ARG_CONST_SIZE, 5315 }; 5316 5317 #ifdef CONFIG_XFRM 5318 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index, 5319 struct bpf_xfrm_state *, to, u32, size, u64, flags) 5320 { 5321 const struct sec_path *sp = skb_sec_path(skb); 5322 const struct xfrm_state *x; 5323 5324 if (!sp || unlikely(index >= sp->len || flags)) 5325 goto err_clear; 5326 5327 x = sp->xvec[index]; 5328 5329 if (unlikely(size != sizeof(struct bpf_xfrm_state))) 5330 goto err_clear; 5331 5332 to->reqid = x->props.reqid; 5333 to->spi = x->id.spi; 5334 to->family = x->props.family; 5335 to->ext = 0; 5336 5337 if (to->family == AF_INET6) { 5338 memcpy(to->remote_ipv6, x->props.saddr.a6, 5339 sizeof(to->remote_ipv6)); 5340 } else { 5341 to->remote_ipv4 = x->props.saddr.a4; 5342 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 5343 } 5344 5345 return 0; 5346 err_clear: 5347 memset(to, 0, size); 5348 return -EINVAL; 5349 } 5350 5351 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = { 5352 .func = bpf_skb_get_xfrm_state, 5353 .gpl_only = false, 5354 .ret_type = RET_INTEGER, 5355 .arg1_type = ARG_PTR_TO_CTX, 5356 .arg2_type = ARG_ANYTHING, 5357 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 5358 .arg4_type = ARG_CONST_SIZE, 5359 .arg5_type = ARG_ANYTHING, 5360 }; 5361 #endif 5362 5363 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6) 5364 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, 5365 const struct neighbour *neigh, 5366 const struct net_device *dev, u32 mtu) 5367 { 5368 memcpy(params->dmac, neigh->ha, ETH_ALEN); 5369 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 5370 params->h_vlan_TCI = 0; 5371 params->h_vlan_proto = 0; 5372 if (mtu) 5373 params->mtu_result = mtu; /* union with tot_len */ 5374 5375 return 0; 5376 } 5377 #endif 5378 5379 #if IS_ENABLED(CONFIG_INET) 5380 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5381 u32 flags, bool check_mtu) 5382 { 5383 struct fib_nh_common *nhc; 5384 struct in_device *in_dev; 5385 struct neighbour *neigh; 5386 struct net_device *dev; 5387 struct fib_result res; 5388 struct flowi4 fl4; 5389 u32 mtu = 0; 5390 int err; 5391 5392 dev = dev_get_by_index_rcu(net, params->ifindex); 5393 if (unlikely(!dev)) 5394 return -ENODEV; 5395 5396 /* verify forwarding is enabled on this interface */ 5397 in_dev = __in_dev_get_rcu(dev); 5398 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev))) 5399 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5400 5401 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5402 fl4.flowi4_iif = 1; 5403 fl4.flowi4_oif = params->ifindex; 5404 } else { 5405 fl4.flowi4_iif = params->ifindex; 5406 fl4.flowi4_oif = 0; 5407 } 5408 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK; 5409 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 5410 fl4.flowi4_flags = 0; 5411 5412 fl4.flowi4_proto = params->l4_protocol; 5413 fl4.daddr = params->ipv4_dst; 5414 fl4.saddr = params->ipv4_src; 5415 fl4.fl4_sport = params->sport; 5416 fl4.fl4_dport = params->dport; 5417 fl4.flowi4_multipath_hash = 0; 5418 5419 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5420 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5421 struct fib_table *tb; 5422 5423 tb = fib_get_table(net, tbid); 5424 if (unlikely(!tb)) 5425 return BPF_FIB_LKUP_RET_NOT_FWDED; 5426 5427 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 5428 } else { 5429 fl4.flowi4_mark = 0; 5430 fl4.flowi4_secid = 0; 5431 fl4.flowi4_tun_key.tun_id = 0; 5432 fl4.flowi4_uid = sock_net_uid(net, NULL); 5433 5434 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF); 5435 } 5436 5437 if (err) { 5438 /* map fib lookup errors to RTN_ type */ 5439 if (err == -EINVAL) 5440 return BPF_FIB_LKUP_RET_BLACKHOLE; 5441 if (err == -EHOSTUNREACH) 5442 return BPF_FIB_LKUP_RET_UNREACHABLE; 5443 if (err == -EACCES) 5444 return BPF_FIB_LKUP_RET_PROHIBIT; 5445 5446 return BPF_FIB_LKUP_RET_NOT_FWDED; 5447 } 5448 5449 if (res.type != RTN_UNICAST) 5450 return BPF_FIB_LKUP_RET_NOT_FWDED; 5451 5452 if (fib_info_num_path(res.fi) > 1) 5453 fib_select_path(net, &res, &fl4, NULL); 5454 5455 if (check_mtu) { 5456 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst); 5457 if (params->tot_len > mtu) { 5458 params->mtu_result = mtu; /* union with tot_len */ 5459 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5460 } 5461 } 5462 5463 nhc = res.nhc; 5464 5465 /* do not handle lwt encaps right now */ 5466 if (nhc->nhc_lwtstate) 5467 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5468 5469 dev = nhc->nhc_dev; 5470 5471 params->rt_metric = res.fi->fib_priority; 5472 params->ifindex = dev->ifindex; 5473 5474 /* xdp and cls_bpf programs are run in RCU-bh so 5475 * rcu_read_lock_bh is not needed here 5476 */ 5477 if (likely(nhc->nhc_gw_family != AF_INET6)) { 5478 if (nhc->nhc_gw_family) 5479 params->ipv4_dst = nhc->nhc_gw.ipv4; 5480 5481 neigh = __ipv4_neigh_lookup_noref(dev, 5482 (__force u32)params->ipv4_dst); 5483 } else { 5484 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst; 5485 5486 params->family = AF_INET6; 5487 *dst = nhc->nhc_gw.ipv6; 5488 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5489 } 5490 5491 if (!neigh) 5492 return BPF_FIB_LKUP_RET_NO_NEIGH; 5493 5494 return bpf_fib_set_fwd_params(params, neigh, dev, mtu); 5495 } 5496 #endif 5497 5498 #if IS_ENABLED(CONFIG_IPV6) 5499 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5500 u32 flags, bool check_mtu) 5501 { 5502 struct in6_addr *src = (struct in6_addr *) params->ipv6_src; 5503 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst; 5504 struct fib6_result res = {}; 5505 struct neighbour *neigh; 5506 struct net_device *dev; 5507 struct inet6_dev *idev; 5508 struct flowi6 fl6; 5509 int strict = 0; 5510 int oif, err; 5511 u32 mtu = 0; 5512 5513 /* link local addresses are never forwarded */ 5514 if (rt6_need_strict(dst) || rt6_need_strict(src)) 5515 return BPF_FIB_LKUP_RET_NOT_FWDED; 5516 5517 dev = dev_get_by_index_rcu(net, params->ifindex); 5518 if (unlikely(!dev)) 5519 return -ENODEV; 5520 5521 idev = __in6_dev_get_safely(dev); 5522 if (unlikely(!idev || !idev->cnf.forwarding)) 5523 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5524 5525 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5526 fl6.flowi6_iif = 1; 5527 oif = fl6.flowi6_oif = params->ifindex; 5528 } else { 5529 oif = fl6.flowi6_iif = params->ifindex; 5530 fl6.flowi6_oif = 0; 5531 strict = RT6_LOOKUP_F_HAS_SADDR; 5532 } 5533 fl6.flowlabel = params->flowinfo; 5534 fl6.flowi6_scope = 0; 5535 fl6.flowi6_flags = 0; 5536 fl6.mp_hash = 0; 5537 5538 fl6.flowi6_proto = params->l4_protocol; 5539 fl6.daddr = *dst; 5540 fl6.saddr = *src; 5541 fl6.fl6_sport = params->sport; 5542 fl6.fl6_dport = params->dport; 5543 5544 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5545 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5546 struct fib6_table *tb; 5547 5548 tb = ipv6_stub->fib6_get_table(net, tbid); 5549 if (unlikely(!tb)) 5550 return BPF_FIB_LKUP_RET_NOT_FWDED; 5551 5552 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res, 5553 strict); 5554 } else { 5555 fl6.flowi6_mark = 0; 5556 fl6.flowi6_secid = 0; 5557 fl6.flowi6_tun_key.tun_id = 0; 5558 fl6.flowi6_uid = sock_net_uid(net, NULL); 5559 5560 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict); 5561 } 5562 5563 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) || 5564 res.f6i == net->ipv6.fib6_null_entry)) 5565 return BPF_FIB_LKUP_RET_NOT_FWDED; 5566 5567 switch (res.fib6_type) { 5568 /* only unicast is forwarded */ 5569 case RTN_UNICAST: 5570 break; 5571 case RTN_BLACKHOLE: 5572 return BPF_FIB_LKUP_RET_BLACKHOLE; 5573 case RTN_UNREACHABLE: 5574 return BPF_FIB_LKUP_RET_UNREACHABLE; 5575 case RTN_PROHIBIT: 5576 return BPF_FIB_LKUP_RET_PROHIBIT; 5577 default: 5578 return BPF_FIB_LKUP_RET_NOT_FWDED; 5579 } 5580 5581 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif, 5582 fl6.flowi6_oif != 0, NULL, strict); 5583 5584 if (check_mtu) { 5585 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src); 5586 if (params->tot_len > mtu) { 5587 params->mtu_result = mtu; /* union with tot_len */ 5588 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5589 } 5590 } 5591 5592 if (res.nh->fib_nh_lws) 5593 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5594 5595 if (res.nh->fib_nh_gw_family) 5596 *dst = res.nh->fib_nh_gw6; 5597 5598 dev = res.nh->fib_nh_dev; 5599 params->rt_metric = res.f6i->fib6_metric; 5600 params->ifindex = dev->ifindex; 5601 5602 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is 5603 * not needed here. 5604 */ 5605 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5606 if (!neigh) 5607 return BPF_FIB_LKUP_RET_NO_NEIGH; 5608 5609 return bpf_fib_set_fwd_params(params, neigh, dev, mtu); 5610 } 5611 #endif 5612 5613 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx, 5614 struct bpf_fib_lookup *, params, int, plen, u32, flags) 5615 { 5616 if (plen < sizeof(*params)) 5617 return -EINVAL; 5618 5619 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 5620 return -EINVAL; 5621 5622 switch (params->family) { 5623 #if IS_ENABLED(CONFIG_INET) 5624 case AF_INET: 5625 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params, 5626 flags, true); 5627 #endif 5628 #if IS_ENABLED(CONFIG_IPV6) 5629 case AF_INET6: 5630 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params, 5631 flags, true); 5632 #endif 5633 } 5634 return -EAFNOSUPPORT; 5635 } 5636 5637 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = { 5638 .func = bpf_xdp_fib_lookup, 5639 .gpl_only = true, 5640 .ret_type = RET_INTEGER, 5641 .arg1_type = ARG_PTR_TO_CTX, 5642 .arg2_type = ARG_PTR_TO_MEM, 5643 .arg3_type = ARG_CONST_SIZE, 5644 .arg4_type = ARG_ANYTHING, 5645 }; 5646 5647 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb, 5648 struct bpf_fib_lookup *, params, int, plen, u32, flags) 5649 { 5650 struct net *net = dev_net(skb->dev); 5651 int rc = -EAFNOSUPPORT; 5652 bool check_mtu = false; 5653 5654 if (plen < sizeof(*params)) 5655 return -EINVAL; 5656 5657 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 5658 return -EINVAL; 5659 5660 if (params->tot_len) 5661 check_mtu = true; 5662 5663 switch (params->family) { 5664 #if IS_ENABLED(CONFIG_INET) 5665 case AF_INET: 5666 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu); 5667 break; 5668 #endif 5669 #if IS_ENABLED(CONFIG_IPV6) 5670 case AF_INET6: 5671 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu); 5672 break; 5673 #endif 5674 } 5675 5676 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) { 5677 struct net_device *dev; 5678 5679 /* When tot_len isn't provided by user, check skb 5680 * against MTU of FIB lookup resulting net_device 5681 */ 5682 dev = dev_get_by_index_rcu(net, params->ifindex); 5683 if (!is_skb_forwardable(dev, skb)) 5684 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED; 5685 5686 params->mtu_result = dev->mtu; /* union with tot_len */ 5687 } 5688 5689 return rc; 5690 } 5691 5692 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = { 5693 .func = bpf_skb_fib_lookup, 5694 .gpl_only = true, 5695 .ret_type = RET_INTEGER, 5696 .arg1_type = ARG_PTR_TO_CTX, 5697 .arg2_type = ARG_PTR_TO_MEM, 5698 .arg3_type = ARG_CONST_SIZE, 5699 .arg4_type = ARG_ANYTHING, 5700 }; 5701 5702 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr, 5703 u32 ifindex) 5704 { 5705 struct net *netns = dev_net(dev_curr); 5706 5707 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */ 5708 if (ifindex == 0) 5709 return dev_curr; 5710 5711 return dev_get_by_index_rcu(netns, ifindex); 5712 } 5713 5714 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb, 5715 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 5716 { 5717 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 5718 struct net_device *dev = skb->dev; 5719 int skb_len, dev_len; 5720 int mtu; 5721 5722 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS))) 5723 return -EINVAL; 5724 5725 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len))) 5726 return -EINVAL; 5727 5728 dev = __dev_via_ifindex(dev, ifindex); 5729 if (unlikely(!dev)) 5730 return -ENODEV; 5731 5732 mtu = READ_ONCE(dev->mtu); 5733 5734 dev_len = mtu + dev->hard_header_len; 5735 5736 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 5737 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len; 5738 5739 skb_len += len_diff; /* minus result pass check */ 5740 if (skb_len <= dev_len) { 5741 ret = BPF_MTU_CHK_RET_SUCCESS; 5742 goto out; 5743 } 5744 /* At this point, skb->len exceed MTU, but as it include length of all 5745 * segments, it can still be below MTU. The SKB can possibly get 5746 * re-segmented in transmit path (see validate_xmit_skb). Thus, user 5747 * must choose if segs are to be MTU checked. 5748 */ 5749 if (skb_is_gso(skb)) { 5750 ret = BPF_MTU_CHK_RET_SUCCESS; 5751 5752 if (flags & BPF_MTU_CHK_SEGS && 5753 !skb_gso_validate_network_len(skb, mtu)) 5754 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG; 5755 } 5756 out: 5757 /* BPF verifier guarantees valid pointer */ 5758 *mtu_len = mtu; 5759 5760 return ret; 5761 } 5762 5763 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp, 5764 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 5765 { 5766 struct net_device *dev = xdp->rxq->dev; 5767 int xdp_len = xdp->data_end - xdp->data; 5768 int ret = BPF_MTU_CHK_RET_SUCCESS; 5769 int mtu, dev_len; 5770 5771 /* XDP variant doesn't support multi-buffer segment check (yet) */ 5772 if (unlikely(flags)) 5773 return -EINVAL; 5774 5775 dev = __dev_via_ifindex(dev, ifindex); 5776 if (unlikely(!dev)) 5777 return -ENODEV; 5778 5779 mtu = READ_ONCE(dev->mtu); 5780 5781 /* Add L2-header as dev MTU is L3 size */ 5782 dev_len = mtu + dev->hard_header_len; 5783 5784 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 5785 if (*mtu_len) 5786 xdp_len = *mtu_len + dev->hard_header_len; 5787 5788 xdp_len += len_diff; /* minus result pass check */ 5789 if (xdp_len > dev_len) 5790 ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 5791 5792 /* BPF verifier guarantees valid pointer */ 5793 *mtu_len = mtu; 5794 5795 return ret; 5796 } 5797 5798 static const struct bpf_func_proto bpf_skb_check_mtu_proto = { 5799 .func = bpf_skb_check_mtu, 5800 .gpl_only = true, 5801 .ret_type = RET_INTEGER, 5802 .arg1_type = ARG_PTR_TO_CTX, 5803 .arg2_type = ARG_ANYTHING, 5804 .arg3_type = ARG_PTR_TO_INT, 5805 .arg4_type = ARG_ANYTHING, 5806 .arg5_type = ARG_ANYTHING, 5807 }; 5808 5809 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = { 5810 .func = bpf_xdp_check_mtu, 5811 .gpl_only = true, 5812 .ret_type = RET_INTEGER, 5813 .arg1_type = ARG_PTR_TO_CTX, 5814 .arg2_type = ARG_ANYTHING, 5815 .arg3_type = ARG_PTR_TO_INT, 5816 .arg4_type = ARG_ANYTHING, 5817 .arg5_type = ARG_ANYTHING, 5818 }; 5819 5820 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5821 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len) 5822 { 5823 int err; 5824 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr; 5825 5826 if (!seg6_validate_srh(srh, len, false)) 5827 return -EINVAL; 5828 5829 switch (type) { 5830 case BPF_LWT_ENCAP_SEG6_INLINE: 5831 if (skb->protocol != htons(ETH_P_IPV6)) 5832 return -EBADMSG; 5833 5834 err = seg6_do_srh_inline(skb, srh); 5835 break; 5836 case BPF_LWT_ENCAP_SEG6: 5837 skb_reset_inner_headers(skb); 5838 skb->encapsulation = 1; 5839 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6); 5840 break; 5841 default: 5842 return -EINVAL; 5843 } 5844 5845 bpf_compute_data_pointers(skb); 5846 if (err) 5847 return err; 5848 5849 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 5850 skb_set_transport_header(skb, sizeof(struct ipv6hdr)); 5851 5852 return seg6_lookup_nexthop(skb, NULL, 0); 5853 } 5854 #endif /* CONFIG_IPV6_SEG6_BPF */ 5855 5856 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 5857 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, 5858 bool ingress) 5859 { 5860 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress); 5861 } 5862 #endif 5863 5864 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr, 5865 u32, len) 5866 { 5867 switch (type) { 5868 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5869 case BPF_LWT_ENCAP_SEG6: 5870 case BPF_LWT_ENCAP_SEG6_INLINE: 5871 return bpf_push_seg6_encap(skb, type, hdr, len); 5872 #endif 5873 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 5874 case BPF_LWT_ENCAP_IP: 5875 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */); 5876 #endif 5877 default: 5878 return -EINVAL; 5879 } 5880 } 5881 5882 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type, 5883 void *, hdr, u32, len) 5884 { 5885 switch (type) { 5886 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 5887 case BPF_LWT_ENCAP_IP: 5888 return bpf_push_ip_encap(skb, hdr, len, false /* egress */); 5889 #endif 5890 default: 5891 return -EINVAL; 5892 } 5893 } 5894 5895 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = { 5896 .func = bpf_lwt_in_push_encap, 5897 .gpl_only = false, 5898 .ret_type = RET_INTEGER, 5899 .arg1_type = ARG_PTR_TO_CTX, 5900 .arg2_type = ARG_ANYTHING, 5901 .arg3_type = ARG_PTR_TO_MEM, 5902 .arg4_type = ARG_CONST_SIZE 5903 }; 5904 5905 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = { 5906 .func = bpf_lwt_xmit_push_encap, 5907 .gpl_only = false, 5908 .ret_type = RET_INTEGER, 5909 .arg1_type = ARG_PTR_TO_CTX, 5910 .arg2_type = ARG_ANYTHING, 5911 .arg3_type = ARG_PTR_TO_MEM, 5912 .arg4_type = ARG_CONST_SIZE 5913 }; 5914 5915 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5916 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset, 5917 const void *, from, u32, len) 5918 { 5919 struct seg6_bpf_srh_state *srh_state = 5920 this_cpu_ptr(&seg6_bpf_srh_states); 5921 struct ipv6_sr_hdr *srh = srh_state->srh; 5922 void *srh_tlvs, *srh_end, *ptr; 5923 int srhoff = 0; 5924 5925 if (srh == NULL) 5926 return -EINVAL; 5927 5928 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4)); 5929 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen); 5930 5931 ptr = skb->data + offset; 5932 if (ptr >= srh_tlvs && ptr + len <= srh_end) 5933 srh_state->valid = false; 5934 else if (ptr < (void *)&srh->flags || 5935 ptr + len > (void *)&srh->segments) 5936 return -EFAULT; 5937 5938 if (unlikely(bpf_try_make_writable(skb, offset + len))) 5939 return -EFAULT; 5940 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 5941 return -EINVAL; 5942 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5943 5944 memcpy(skb->data + offset, from, len); 5945 return 0; 5946 } 5947 5948 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = { 5949 .func = bpf_lwt_seg6_store_bytes, 5950 .gpl_only = false, 5951 .ret_type = RET_INTEGER, 5952 .arg1_type = ARG_PTR_TO_CTX, 5953 .arg2_type = ARG_ANYTHING, 5954 .arg3_type = ARG_PTR_TO_MEM, 5955 .arg4_type = ARG_CONST_SIZE 5956 }; 5957 5958 static void bpf_update_srh_state(struct sk_buff *skb) 5959 { 5960 struct seg6_bpf_srh_state *srh_state = 5961 this_cpu_ptr(&seg6_bpf_srh_states); 5962 int srhoff = 0; 5963 5964 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) { 5965 srh_state->srh = NULL; 5966 } else { 5967 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5968 srh_state->hdrlen = srh_state->srh->hdrlen << 3; 5969 srh_state->valid = true; 5970 } 5971 } 5972 5973 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb, 5974 u32, action, void *, param, u32, param_len) 5975 { 5976 struct seg6_bpf_srh_state *srh_state = 5977 this_cpu_ptr(&seg6_bpf_srh_states); 5978 int hdroff = 0; 5979 int err; 5980 5981 switch (action) { 5982 case SEG6_LOCAL_ACTION_END_X: 5983 if (!seg6_bpf_has_valid_srh(skb)) 5984 return -EBADMSG; 5985 if (param_len != sizeof(struct in6_addr)) 5986 return -EINVAL; 5987 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0); 5988 case SEG6_LOCAL_ACTION_END_T: 5989 if (!seg6_bpf_has_valid_srh(skb)) 5990 return -EBADMSG; 5991 if (param_len != sizeof(int)) 5992 return -EINVAL; 5993 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 5994 case SEG6_LOCAL_ACTION_END_DT6: 5995 if (!seg6_bpf_has_valid_srh(skb)) 5996 return -EBADMSG; 5997 if (param_len != sizeof(int)) 5998 return -EINVAL; 5999 6000 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0) 6001 return -EBADMSG; 6002 if (!pskb_pull(skb, hdroff)) 6003 return -EBADMSG; 6004 6005 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff); 6006 skb_reset_network_header(skb); 6007 skb_reset_transport_header(skb); 6008 skb->encapsulation = 0; 6009 6010 bpf_compute_data_pointers(skb); 6011 bpf_update_srh_state(skb); 6012 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 6013 case SEG6_LOCAL_ACTION_END_B6: 6014 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 6015 return -EBADMSG; 6016 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE, 6017 param, param_len); 6018 if (!err) 6019 bpf_update_srh_state(skb); 6020 6021 return err; 6022 case SEG6_LOCAL_ACTION_END_B6_ENCAP: 6023 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 6024 return -EBADMSG; 6025 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6, 6026 param, param_len); 6027 if (!err) 6028 bpf_update_srh_state(skb); 6029 6030 return err; 6031 default: 6032 return -EINVAL; 6033 } 6034 } 6035 6036 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = { 6037 .func = bpf_lwt_seg6_action, 6038 .gpl_only = false, 6039 .ret_type = RET_INTEGER, 6040 .arg1_type = ARG_PTR_TO_CTX, 6041 .arg2_type = ARG_ANYTHING, 6042 .arg3_type = ARG_PTR_TO_MEM, 6043 .arg4_type = ARG_CONST_SIZE 6044 }; 6045 6046 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset, 6047 s32, len) 6048 { 6049 struct seg6_bpf_srh_state *srh_state = 6050 this_cpu_ptr(&seg6_bpf_srh_states); 6051 struct ipv6_sr_hdr *srh = srh_state->srh; 6052 void *srh_end, *srh_tlvs, *ptr; 6053 struct ipv6hdr *hdr; 6054 int srhoff = 0; 6055 int ret; 6056 6057 if (unlikely(srh == NULL)) 6058 return -EINVAL; 6059 6060 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) + 6061 ((srh->first_segment + 1) << 4)); 6062 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) + 6063 srh_state->hdrlen); 6064 ptr = skb->data + offset; 6065 6066 if (unlikely(ptr < srh_tlvs || ptr > srh_end)) 6067 return -EFAULT; 6068 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end)) 6069 return -EFAULT; 6070 6071 if (len > 0) { 6072 ret = skb_cow_head(skb, len); 6073 if (unlikely(ret < 0)) 6074 return ret; 6075 6076 ret = bpf_skb_net_hdr_push(skb, offset, len); 6077 } else { 6078 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len); 6079 } 6080 6081 bpf_compute_data_pointers(skb); 6082 if (unlikely(ret < 0)) 6083 return ret; 6084 6085 hdr = (struct ipv6hdr *)skb->data; 6086 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 6087 6088 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 6089 return -EINVAL; 6090 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6091 srh_state->hdrlen += len; 6092 srh_state->valid = false; 6093 return 0; 6094 } 6095 6096 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = { 6097 .func = bpf_lwt_seg6_adjust_srh, 6098 .gpl_only = false, 6099 .ret_type = RET_INTEGER, 6100 .arg1_type = ARG_PTR_TO_CTX, 6101 .arg2_type = ARG_ANYTHING, 6102 .arg3_type = ARG_ANYTHING, 6103 }; 6104 #endif /* CONFIG_IPV6_SEG6_BPF */ 6105 6106 #ifdef CONFIG_INET 6107 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple, 6108 int dif, int sdif, u8 family, u8 proto) 6109 { 6110 bool refcounted = false; 6111 struct sock *sk = NULL; 6112 6113 if (family == AF_INET) { 6114 __be32 src4 = tuple->ipv4.saddr; 6115 __be32 dst4 = tuple->ipv4.daddr; 6116 6117 if (proto == IPPROTO_TCP) 6118 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0, 6119 src4, tuple->ipv4.sport, 6120 dst4, tuple->ipv4.dport, 6121 dif, sdif, &refcounted); 6122 else 6123 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport, 6124 dst4, tuple->ipv4.dport, 6125 dif, sdif, &udp_table, NULL); 6126 #if IS_ENABLED(CONFIG_IPV6) 6127 } else { 6128 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr; 6129 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr; 6130 6131 if (proto == IPPROTO_TCP) 6132 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0, 6133 src6, tuple->ipv6.sport, 6134 dst6, ntohs(tuple->ipv6.dport), 6135 dif, sdif, &refcounted); 6136 else if (likely(ipv6_bpf_stub)) 6137 sk = ipv6_bpf_stub->udp6_lib_lookup(net, 6138 src6, tuple->ipv6.sport, 6139 dst6, tuple->ipv6.dport, 6140 dif, sdif, 6141 &udp_table, NULL); 6142 #endif 6143 } 6144 6145 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) { 6146 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6147 sk = NULL; 6148 } 6149 return sk; 6150 } 6151 6152 /* bpf_skc_lookup performs the core lookup for different types of sockets, 6153 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE. 6154 * Returns the socket as an 'unsigned long' to simplify the casting in the 6155 * callers to satisfy BPF_CALL declarations. 6156 */ 6157 static struct sock * 6158 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6159 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 6160 u64 flags) 6161 { 6162 struct sock *sk = NULL; 6163 u8 family = AF_UNSPEC; 6164 struct net *net; 6165 int sdif; 6166 6167 if (len == sizeof(tuple->ipv4)) 6168 family = AF_INET; 6169 else if (len == sizeof(tuple->ipv6)) 6170 family = AF_INET6; 6171 else 6172 return NULL; 6173 6174 if (unlikely(family == AF_UNSPEC || flags || 6175 !((s32)netns_id < 0 || netns_id <= S32_MAX))) 6176 goto out; 6177 6178 if (family == AF_INET) 6179 sdif = inet_sdif(skb); 6180 else 6181 sdif = inet6_sdif(skb); 6182 6183 if ((s32)netns_id < 0) { 6184 net = caller_net; 6185 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 6186 } else { 6187 net = get_net_ns_by_id(caller_net, netns_id); 6188 if (unlikely(!net)) 6189 goto out; 6190 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 6191 put_net(net); 6192 } 6193 6194 out: 6195 return sk; 6196 } 6197 6198 static struct sock * 6199 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6200 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 6201 u64 flags) 6202 { 6203 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net, 6204 ifindex, proto, netns_id, flags); 6205 6206 if (sk) { 6207 sk = sk_to_full_sk(sk); 6208 if (!sk_fullsock(sk)) { 6209 sock_gen_put(sk); 6210 return NULL; 6211 } 6212 } 6213 6214 return sk; 6215 } 6216 6217 static struct sock * 6218 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6219 u8 proto, u64 netns_id, u64 flags) 6220 { 6221 struct net *caller_net; 6222 int ifindex; 6223 6224 if (skb->dev) { 6225 caller_net = dev_net(skb->dev); 6226 ifindex = skb->dev->ifindex; 6227 } else { 6228 caller_net = sock_net(skb->sk); 6229 ifindex = 0; 6230 } 6231 6232 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto, 6233 netns_id, flags); 6234 } 6235 6236 static struct sock * 6237 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6238 u8 proto, u64 netns_id, u64 flags) 6239 { 6240 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id, 6241 flags); 6242 6243 if (sk) { 6244 sk = sk_to_full_sk(sk); 6245 if (!sk_fullsock(sk)) { 6246 sock_gen_put(sk); 6247 return NULL; 6248 } 6249 } 6250 6251 return sk; 6252 } 6253 6254 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb, 6255 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6256 { 6257 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP, 6258 netns_id, flags); 6259 } 6260 6261 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = { 6262 .func = bpf_skc_lookup_tcp, 6263 .gpl_only = false, 6264 .pkt_access = true, 6265 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6266 .arg1_type = ARG_PTR_TO_CTX, 6267 .arg2_type = ARG_PTR_TO_MEM, 6268 .arg3_type = ARG_CONST_SIZE, 6269 .arg4_type = ARG_ANYTHING, 6270 .arg5_type = ARG_ANYTHING, 6271 }; 6272 6273 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb, 6274 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6275 { 6276 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, 6277 netns_id, flags); 6278 } 6279 6280 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = { 6281 .func = bpf_sk_lookup_tcp, 6282 .gpl_only = false, 6283 .pkt_access = true, 6284 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6285 .arg1_type = ARG_PTR_TO_CTX, 6286 .arg2_type = ARG_PTR_TO_MEM, 6287 .arg3_type = ARG_CONST_SIZE, 6288 .arg4_type = ARG_ANYTHING, 6289 .arg5_type = ARG_ANYTHING, 6290 }; 6291 6292 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb, 6293 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6294 { 6295 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, 6296 netns_id, flags); 6297 } 6298 6299 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = { 6300 .func = bpf_sk_lookup_udp, 6301 .gpl_only = false, 6302 .pkt_access = true, 6303 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6304 .arg1_type = ARG_PTR_TO_CTX, 6305 .arg2_type = ARG_PTR_TO_MEM, 6306 .arg3_type = ARG_CONST_SIZE, 6307 .arg4_type = ARG_ANYTHING, 6308 .arg5_type = ARG_ANYTHING, 6309 }; 6310 6311 BPF_CALL_1(bpf_sk_release, struct sock *, sk) 6312 { 6313 if (sk && sk_is_refcounted(sk)) 6314 sock_gen_put(sk); 6315 return 0; 6316 } 6317 6318 static const struct bpf_func_proto bpf_sk_release_proto = { 6319 .func = bpf_sk_release, 6320 .gpl_only = false, 6321 .ret_type = RET_INTEGER, 6322 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6323 }; 6324 6325 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx, 6326 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6327 { 6328 struct net *caller_net = dev_net(ctx->rxq->dev); 6329 int ifindex = ctx->rxq->dev->ifindex; 6330 6331 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6332 ifindex, IPPROTO_UDP, netns_id, 6333 flags); 6334 } 6335 6336 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = { 6337 .func = bpf_xdp_sk_lookup_udp, 6338 .gpl_only = false, 6339 .pkt_access = true, 6340 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6341 .arg1_type = ARG_PTR_TO_CTX, 6342 .arg2_type = ARG_PTR_TO_MEM, 6343 .arg3_type = ARG_CONST_SIZE, 6344 .arg4_type = ARG_ANYTHING, 6345 .arg5_type = ARG_ANYTHING, 6346 }; 6347 6348 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx, 6349 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6350 { 6351 struct net *caller_net = dev_net(ctx->rxq->dev); 6352 int ifindex = ctx->rxq->dev->ifindex; 6353 6354 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net, 6355 ifindex, IPPROTO_TCP, netns_id, 6356 flags); 6357 } 6358 6359 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = { 6360 .func = bpf_xdp_skc_lookup_tcp, 6361 .gpl_only = false, 6362 .pkt_access = true, 6363 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6364 .arg1_type = ARG_PTR_TO_CTX, 6365 .arg2_type = ARG_PTR_TO_MEM, 6366 .arg3_type = ARG_CONST_SIZE, 6367 .arg4_type = ARG_ANYTHING, 6368 .arg5_type = ARG_ANYTHING, 6369 }; 6370 6371 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx, 6372 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6373 { 6374 struct net *caller_net = dev_net(ctx->rxq->dev); 6375 int ifindex = ctx->rxq->dev->ifindex; 6376 6377 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6378 ifindex, IPPROTO_TCP, netns_id, 6379 flags); 6380 } 6381 6382 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = { 6383 .func = bpf_xdp_sk_lookup_tcp, 6384 .gpl_only = false, 6385 .pkt_access = true, 6386 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6387 .arg1_type = ARG_PTR_TO_CTX, 6388 .arg2_type = ARG_PTR_TO_MEM, 6389 .arg3_type = ARG_CONST_SIZE, 6390 .arg4_type = ARG_ANYTHING, 6391 .arg5_type = ARG_ANYTHING, 6392 }; 6393 6394 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6395 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6396 { 6397 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, 6398 sock_net(ctx->sk), 0, 6399 IPPROTO_TCP, netns_id, flags); 6400 } 6401 6402 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = { 6403 .func = bpf_sock_addr_skc_lookup_tcp, 6404 .gpl_only = false, 6405 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6406 .arg1_type = ARG_PTR_TO_CTX, 6407 .arg2_type = ARG_PTR_TO_MEM, 6408 .arg3_type = ARG_CONST_SIZE, 6409 .arg4_type = ARG_ANYTHING, 6410 .arg5_type = ARG_ANYTHING, 6411 }; 6412 6413 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6414 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6415 { 6416 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6417 sock_net(ctx->sk), 0, IPPROTO_TCP, 6418 netns_id, flags); 6419 } 6420 6421 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = { 6422 .func = bpf_sock_addr_sk_lookup_tcp, 6423 .gpl_only = false, 6424 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6425 .arg1_type = ARG_PTR_TO_CTX, 6426 .arg2_type = ARG_PTR_TO_MEM, 6427 .arg3_type = ARG_CONST_SIZE, 6428 .arg4_type = ARG_ANYTHING, 6429 .arg5_type = ARG_ANYTHING, 6430 }; 6431 6432 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx, 6433 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6434 { 6435 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6436 sock_net(ctx->sk), 0, IPPROTO_UDP, 6437 netns_id, flags); 6438 } 6439 6440 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = { 6441 .func = bpf_sock_addr_sk_lookup_udp, 6442 .gpl_only = false, 6443 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6444 .arg1_type = ARG_PTR_TO_CTX, 6445 .arg2_type = ARG_PTR_TO_MEM, 6446 .arg3_type = ARG_CONST_SIZE, 6447 .arg4_type = ARG_ANYTHING, 6448 .arg5_type = ARG_ANYTHING, 6449 }; 6450 6451 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6452 struct bpf_insn_access_aux *info) 6453 { 6454 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, 6455 icsk_retransmits)) 6456 return false; 6457 6458 if (off % size != 0) 6459 return false; 6460 6461 switch (off) { 6462 case offsetof(struct bpf_tcp_sock, bytes_received): 6463 case offsetof(struct bpf_tcp_sock, bytes_acked): 6464 return size == sizeof(__u64); 6465 default: 6466 return size == sizeof(__u32); 6467 } 6468 } 6469 6470 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 6471 const struct bpf_insn *si, 6472 struct bpf_insn *insn_buf, 6473 struct bpf_prog *prog, u32 *target_size) 6474 { 6475 struct bpf_insn *insn = insn_buf; 6476 6477 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \ 6478 do { \ 6479 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \ 6480 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6481 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\ 6482 si->dst_reg, si->src_reg, \ 6483 offsetof(struct tcp_sock, FIELD)); \ 6484 } while (0) 6485 6486 #define BPF_INET_SOCK_GET_COMMON(FIELD) \ 6487 do { \ 6488 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \ 6489 FIELD) > \ 6490 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6491 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 6492 struct inet_connection_sock, \ 6493 FIELD), \ 6494 si->dst_reg, si->src_reg, \ 6495 offsetof( \ 6496 struct inet_connection_sock, \ 6497 FIELD)); \ 6498 } while (0) 6499 6500 if (insn > insn_buf) 6501 return insn - insn_buf; 6502 6503 switch (si->off) { 6504 case offsetof(struct bpf_tcp_sock, rtt_min): 6505 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 6506 sizeof(struct minmax)); 6507 BUILD_BUG_ON(sizeof(struct minmax) < 6508 sizeof(struct minmax_sample)); 6509 6510 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 6511 offsetof(struct tcp_sock, rtt_min) + 6512 offsetof(struct minmax_sample, v)); 6513 break; 6514 case offsetof(struct bpf_tcp_sock, snd_cwnd): 6515 BPF_TCP_SOCK_GET_COMMON(snd_cwnd); 6516 break; 6517 case offsetof(struct bpf_tcp_sock, srtt_us): 6518 BPF_TCP_SOCK_GET_COMMON(srtt_us); 6519 break; 6520 case offsetof(struct bpf_tcp_sock, snd_ssthresh): 6521 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh); 6522 break; 6523 case offsetof(struct bpf_tcp_sock, rcv_nxt): 6524 BPF_TCP_SOCK_GET_COMMON(rcv_nxt); 6525 break; 6526 case offsetof(struct bpf_tcp_sock, snd_nxt): 6527 BPF_TCP_SOCK_GET_COMMON(snd_nxt); 6528 break; 6529 case offsetof(struct bpf_tcp_sock, snd_una): 6530 BPF_TCP_SOCK_GET_COMMON(snd_una); 6531 break; 6532 case offsetof(struct bpf_tcp_sock, mss_cache): 6533 BPF_TCP_SOCK_GET_COMMON(mss_cache); 6534 break; 6535 case offsetof(struct bpf_tcp_sock, ecn_flags): 6536 BPF_TCP_SOCK_GET_COMMON(ecn_flags); 6537 break; 6538 case offsetof(struct bpf_tcp_sock, rate_delivered): 6539 BPF_TCP_SOCK_GET_COMMON(rate_delivered); 6540 break; 6541 case offsetof(struct bpf_tcp_sock, rate_interval_us): 6542 BPF_TCP_SOCK_GET_COMMON(rate_interval_us); 6543 break; 6544 case offsetof(struct bpf_tcp_sock, packets_out): 6545 BPF_TCP_SOCK_GET_COMMON(packets_out); 6546 break; 6547 case offsetof(struct bpf_tcp_sock, retrans_out): 6548 BPF_TCP_SOCK_GET_COMMON(retrans_out); 6549 break; 6550 case offsetof(struct bpf_tcp_sock, total_retrans): 6551 BPF_TCP_SOCK_GET_COMMON(total_retrans); 6552 break; 6553 case offsetof(struct bpf_tcp_sock, segs_in): 6554 BPF_TCP_SOCK_GET_COMMON(segs_in); 6555 break; 6556 case offsetof(struct bpf_tcp_sock, data_segs_in): 6557 BPF_TCP_SOCK_GET_COMMON(data_segs_in); 6558 break; 6559 case offsetof(struct bpf_tcp_sock, segs_out): 6560 BPF_TCP_SOCK_GET_COMMON(segs_out); 6561 break; 6562 case offsetof(struct bpf_tcp_sock, data_segs_out): 6563 BPF_TCP_SOCK_GET_COMMON(data_segs_out); 6564 break; 6565 case offsetof(struct bpf_tcp_sock, lost_out): 6566 BPF_TCP_SOCK_GET_COMMON(lost_out); 6567 break; 6568 case offsetof(struct bpf_tcp_sock, sacked_out): 6569 BPF_TCP_SOCK_GET_COMMON(sacked_out); 6570 break; 6571 case offsetof(struct bpf_tcp_sock, bytes_received): 6572 BPF_TCP_SOCK_GET_COMMON(bytes_received); 6573 break; 6574 case offsetof(struct bpf_tcp_sock, bytes_acked): 6575 BPF_TCP_SOCK_GET_COMMON(bytes_acked); 6576 break; 6577 case offsetof(struct bpf_tcp_sock, dsack_dups): 6578 BPF_TCP_SOCK_GET_COMMON(dsack_dups); 6579 break; 6580 case offsetof(struct bpf_tcp_sock, delivered): 6581 BPF_TCP_SOCK_GET_COMMON(delivered); 6582 break; 6583 case offsetof(struct bpf_tcp_sock, delivered_ce): 6584 BPF_TCP_SOCK_GET_COMMON(delivered_ce); 6585 break; 6586 case offsetof(struct bpf_tcp_sock, icsk_retransmits): 6587 BPF_INET_SOCK_GET_COMMON(icsk_retransmits); 6588 break; 6589 } 6590 6591 return insn - insn_buf; 6592 } 6593 6594 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk) 6595 { 6596 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 6597 return (unsigned long)sk; 6598 6599 return (unsigned long)NULL; 6600 } 6601 6602 const struct bpf_func_proto bpf_tcp_sock_proto = { 6603 .func = bpf_tcp_sock, 6604 .gpl_only = false, 6605 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL, 6606 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 6607 }; 6608 6609 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk) 6610 { 6611 sk = sk_to_full_sk(sk); 6612 6613 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE)) 6614 return (unsigned long)sk; 6615 6616 return (unsigned long)NULL; 6617 } 6618 6619 static const struct bpf_func_proto bpf_get_listener_sock_proto = { 6620 .func = bpf_get_listener_sock, 6621 .gpl_only = false, 6622 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6623 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 6624 }; 6625 6626 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb) 6627 { 6628 unsigned int iphdr_len; 6629 6630 switch (skb_protocol(skb, true)) { 6631 case cpu_to_be16(ETH_P_IP): 6632 iphdr_len = sizeof(struct iphdr); 6633 break; 6634 case cpu_to_be16(ETH_P_IPV6): 6635 iphdr_len = sizeof(struct ipv6hdr); 6636 break; 6637 default: 6638 return 0; 6639 } 6640 6641 if (skb_headlen(skb) < iphdr_len) 6642 return 0; 6643 6644 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len)) 6645 return 0; 6646 6647 return INET_ECN_set_ce(skb); 6648 } 6649 6650 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6651 struct bpf_insn_access_aux *info) 6652 { 6653 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id)) 6654 return false; 6655 6656 if (off % size != 0) 6657 return false; 6658 6659 switch (off) { 6660 default: 6661 return size == sizeof(__u32); 6662 } 6663 } 6664 6665 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 6666 const struct bpf_insn *si, 6667 struct bpf_insn *insn_buf, 6668 struct bpf_prog *prog, u32 *target_size) 6669 { 6670 struct bpf_insn *insn = insn_buf; 6671 6672 #define BPF_XDP_SOCK_GET(FIELD) \ 6673 do { \ 6674 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \ 6675 sizeof_field(struct bpf_xdp_sock, FIELD)); \ 6676 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\ 6677 si->dst_reg, si->src_reg, \ 6678 offsetof(struct xdp_sock, FIELD)); \ 6679 } while (0) 6680 6681 switch (si->off) { 6682 case offsetof(struct bpf_xdp_sock, queue_id): 6683 BPF_XDP_SOCK_GET(queue_id); 6684 break; 6685 } 6686 6687 return insn - insn_buf; 6688 } 6689 6690 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = { 6691 .func = bpf_skb_ecn_set_ce, 6692 .gpl_only = false, 6693 .ret_type = RET_INTEGER, 6694 .arg1_type = ARG_PTR_TO_CTX, 6695 }; 6696 6697 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 6698 struct tcphdr *, th, u32, th_len) 6699 { 6700 #ifdef CONFIG_SYN_COOKIES 6701 u32 cookie; 6702 int ret; 6703 6704 if (unlikely(!sk || th_len < sizeof(*th))) 6705 return -EINVAL; 6706 6707 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */ 6708 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 6709 return -EINVAL; 6710 6711 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies) 6712 return -EINVAL; 6713 6714 if (!th->ack || th->rst || th->syn) 6715 return -ENOENT; 6716 6717 if (tcp_synq_no_recent_overflow(sk)) 6718 return -ENOENT; 6719 6720 cookie = ntohl(th->ack_seq) - 1; 6721 6722 switch (sk->sk_family) { 6723 case AF_INET: 6724 if (unlikely(iph_len < sizeof(struct iphdr))) 6725 return -EINVAL; 6726 6727 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie); 6728 break; 6729 6730 #if IS_BUILTIN(CONFIG_IPV6) 6731 case AF_INET6: 6732 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 6733 return -EINVAL; 6734 6735 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie); 6736 break; 6737 #endif /* CONFIG_IPV6 */ 6738 6739 default: 6740 return -EPROTONOSUPPORT; 6741 } 6742 6743 if (ret > 0) 6744 return 0; 6745 6746 return -ENOENT; 6747 #else 6748 return -ENOTSUPP; 6749 #endif 6750 } 6751 6752 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = { 6753 .func = bpf_tcp_check_syncookie, 6754 .gpl_only = true, 6755 .pkt_access = true, 6756 .ret_type = RET_INTEGER, 6757 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6758 .arg2_type = ARG_PTR_TO_MEM, 6759 .arg3_type = ARG_CONST_SIZE, 6760 .arg4_type = ARG_PTR_TO_MEM, 6761 .arg5_type = ARG_CONST_SIZE, 6762 }; 6763 6764 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 6765 struct tcphdr *, th, u32, th_len) 6766 { 6767 #ifdef CONFIG_SYN_COOKIES 6768 u32 cookie; 6769 u16 mss; 6770 6771 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4)) 6772 return -EINVAL; 6773 6774 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 6775 return -EINVAL; 6776 6777 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies) 6778 return -ENOENT; 6779 6780 if (!th->syn || th->ack || th->fin || th->rst) 6781 return -EINVAL; 6782 6783 if (unlikely(iph_len < sizeof(struct iphdr))) 6784 return -EINVAL; 6785 6786 /* Both struct iphdr and struct ipv6hdr have the version field at the 6787 * same offset so we can cast to the shorter header (struct iphdr). 6788 */ 6789 switch (((struct iphdr *)iph)->version) { 6790 case 4: 6791 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only) 6792 return -EINVAL; 6793 6794 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie); 6795 break; 6796 6797 #if IS_BUILTIN(CONFIG_IPV6) 6798 case 6: 6799 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 6800 return -EINVAL; 6801 6802 if (sk->sk_family != AF_INET6) 6803 return -EINVAL; 6804 6805 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie); 6806 break; 6807 #endif /* CONFIG_IPV6 */ 6808 6809 default: 6810 return -EPROTONOSUPPORT; 6811 } 6812 if (mss == 0) 6813 return -ENOENT; 6814 6815 return cookie | ((u64)mss << 32); 6816 #else 6817 return -EOPNOTSUPP; 6818 #endif /* CONFIG_SYN_COOKIES */ 6819 } 6820 6821 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = { 6822 .func = bpf_tcp_gen_syncookie, 6823 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */ 6824 .pkt_access = true, 6825 .ret_type = RET_INTEGER, 6826 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6827 .arg2_type = ARG_PTR_TO_MEM, 6828 .arg3_type = ARG_CONST_SIZE, 6829 .arg4_type = ARG_PTR_TO_MEM, 6830 .arg5_type = ARG_CONST_SIZE, 6831 }; 6832 6833 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags) 6834 { 6835 if (!sk || flags != 0) 6836 return -EINVAL; 6837 if (!skb_at_tc_ingress(skb)) 6838 return -EOPNOTSUPP; 6839 if (unlikely(dev_net(skb->dev) != sock_net(sk))) 6840 return -ENETUNREACH; 6841 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport)) 6842 return -ESOCKTNOSUPPORT; 6843 if (sk_is_refcounted(sk) && 6844 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) 6845 return -ENOENT; 6846 6847 skb_orphan(skb); 6848 skb->sk = sk; 6849 skb->destructor = sock_pfree; 6850 6851 return 0; 6852 } 6853 6854 static const struct bpf_func_proto bpf_sk_assign_proto = { 6855 .func = bpf_sk_assign, 6856 .gpl_only = false, 6857 .ret_type = RET_INTEGER, 6858 .arg1_type = ARG_PTR_TO_CTX, 6859 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 6860 .arg3_type = ARG_ANYTHING, 6861 }; 6862 6863 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend, 6864 u8 search_kind, const u8 *magic, 6865 u8 magic_len, bool *eol) 6866 { 6867 u8 kind, kind_len; 6868 6869 *eol = false; 6870 6871 while (op < opend) { 6872 kind = op[0]; 6873 6874 if (kind == TCPOPT_EOL) { 6875 *eol = true; 6876 return ERR_PTR(-ENOMSG); 6877 } else if (kind == TCPOPT_NOP) { 6878 op++; 6879 continue; 6880 } 6881 6882 if (opend - op < 2 || opend - op < op[1] || op[1] < 2) 6883 /* Something is wrong in the received header. 6884 * Follow the TCP stack's tcp_parse_options() 6885 * and just bail here. 6886 */ 6887 return ERR_PTR(-EFAULT); 6888 6889 kind_len = op[1]; 6890 if (search_kind == kind) { 6891 if (!magic_len) 6892 return op; 6893 6894 if (magic_len > kind_len - 2) 6895 return ERR_PTR(-ENOMSG); 6896 6897 if (!memcmp(&op[2], magic, magic_len)) 6898 return op; 6899 } 6900 6901 op += kind_len; 6902 } 6903 6904 return ERR_PTR(-ENOMSG); 6905 } 6906 6907 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 6908 void *, search_res, u32, len, u64, flags) 6909 { 6910 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN; 6911 const u8 *op, *opend, *magic, *search = search_res; 6912 u8 search_kind, search_len, copy_len, magic_len; 6913 int ret; 6914 6915 /* 2 byte is the minimal option len except TCPOPT_NOP and 6916 * TCPOPT_EOL which are useless for the bpf prog to learn 6917 * and this helper disallow loading them also. 6918 */ 6919 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN) 6920 return -EINVAL; 6921 6922 search_kind = search[0]; 6923 search_len = search[1]; 6924 6925 if (search_len > len || search_kind == TCPOPT_NOP || 6926 search_kind == TCPOPT_EOL) 6927 return -EINVAL; 6928 6929 if (search_kind == TCPOPT_EXP || search_kind == 253) { 6930 /* 16 or 32 bit magic. +2 for kind and kind length */ 6931 if (search_len != 4 && search_len != 6) 6932 return -EINVAL; 6933 magic = &search[2]; 6934 magic_len = search_len - 2; 6935 } else { 6936 if (search_len) 6937 return -EINVAL; 6938 magic = NULL; 6939 magic_len = 0; 6940 } 6941 6942 if (load_syn) { 6943 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op); 6944 if (ret < 0) 6945 return ret; 6946 6947 opend = op + ret; 6948 op += sizeof(struct tcphdr); 6949 } else { 6950 if (!bpf_sock->skb || 6951 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB) 6952 /* This bpf_sock->op cannot call this helper */ 6953 return -EPERM; 6954 6955 opend = bpf_sock->skb_data_end; 6956 op = bpf_sock->skb->data + sizeof(struct tcphdr); 6957 } 6958 6959 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len, 6960 &eol); 6961 if (IS_ERR(op)) 6962 return PTR_ERR(op); 6963 6964 copy_len = op[1]; 6965 ret = copy_len; 6966 if (copy_len > len) { 6967 ret = -ENOSPC; 6968 copy_len = len; 6969 } 6970 6971 memcpy(search_res, op, copy_len); 6972 return ret; 6973 } 6974 6975 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = { 6976 .func = bpf_sock_ops_load_hdr_opt, 6977 .gpl_only = false, 6978 .ret_type = RET_INTEGER, 6979 .arg1_type = ARG_PTR_TO_CTX, 6980 .arg2_type = ARG_PTR_TO_MEM, 6981 .arg3_type = ARG_CONST_SIZE, 6982 .arg4_type = ARG_ANYTHING, 6983 }; 6984 6985 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 6986 const void *, from, u32, len, u64, flags) 6987 { 6988 u8 new_kind, new_kind_len, magic_len = 0, *opend; 6989 const u8 *op, *new_op, *magic = NULL; 6990 struct sk_buff *skb; 6991 bool eol; 6992 6993 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB) 6994 return -EPERM; 6995 6996 if (len < 2 || flags) 6997 return -EINVAL; 6998 6999 new_op = from; 7000 new_kind = new_op[0]; 7001 new_kind_len = new_op[1]; 7002 7003 if (new_kind_len > len || new_kind == TCPOPT_NOP || 7004 new_kind == TCPOPT_EOL) 7005 return -EINVAL; 7006 7007 if (new_kind_len > bpf_sock->remaining_opt_len) 7008 return -ENOSPC; 7009 7010 /* 253 is another experimental kind */ 7011 if (new_kind == TCPOPT_EXP || new_kind == 253) { 7012 if (new_kind_len < 4) 7013 return -EINVAL; 7014 /* Match for the 2 byte magic also. 7015 * RFC 6994: the magic could be 2 or 4 bytes. 7016 * Hence, matching by 2 byte only is on the 7017 * conservative side but it is the right 7018 * thing to do for the 'search-for-duplication' 7019 * purpose. 7020 */ 7021 magic = &new_op[2]; 7022 magic_len = 2; 7023 } 7024 7025 /* Check for duplication */ 7026 skb = bpf_sock->skb; 7027 op = skb->data + sizeof(struct tcphdr); 7028 opend = bpf_sock->skb_data_end; 7029 7030 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len, 7031 &eol); 7032 if (!IS_ERR(op)) 7033 return -EEXIST; 7034 7035 if (PTR_ERR(op) != -ENOMSG) 7036 return PTR_ERR(op); 7037 7038 if (eol) 7039 /* The option has been ended. Treat it as no more 7040 * header option can be written. 7041 */ 7042 return -ENOSPC; 7043 7044 /* No duplication found. Store the header option. */ 7045 memcpy(opend, from, new_kind_len); 7046 7047 bpf_sock->remaining_opt_len -= new_kind_len; 7048 bpf_sock->skb_data_end += new_kind_len; 7049 7050 return 0; 7051 } 7052 7053 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = { 7054 .func = bpf_sock_ops_store_hdr_opt, 7055 .gpl_only = false, 7056 .ret_type = RET_INTEGER, 7057 .arg1_type = ARG_PTR_TO_CTX, 7058 .arg2_type = ARG_PTR_TO_MEM, 7059 .arg3_type = ARG_CONST_SIZE, 7060 .arg4_type = ARG_ANYTHING, 7061 }; 7062 7063 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7064 u32, len, u64, flags) 7065 { 7066 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB) 7067 return -EPERM; 7068 7069 if (flags || len < 2) 7070 return -EINVAL; 7071 7072 if (len > bpf_sock->remaining_opt_len) 7073 return -ENOSPC; 7074 7075 bpf_sock->remaining_opt_len -= len; 7076 7077 return 0; 7078 } 7079 7080 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = { 7081 .func = bpf_sock_ops_reserve_hdr_opt, 7082 .gpl_only = false, 7083 .ret_type = RET_INTEGER, 7084 .arg1_type = ARG_PTR_TO_CTX, 7085 .arg2_type = ARG_ANYTHING, 7086 .arg3_type = ARG_ANYTHING, 7087 }; 7088 7089 #endif /* CONFIG_INET */ 7090 7091 bool bpf_helper_changes_pkt_data(void *func) 7092 { 7093 if (func == bpf_skb_vlan_push || 7094 func == bpf_skb_vlan_pop || 7095 func == bpf_skb_store_bytes || 7096 func == bpf_skb_change_proto || 7097 func == bpf_skb_change_head || 7098 func == sk_skb_change_head || 7099 func == bpf_skb_change_tail || 7100 func == sk_skb_change_tail || 7101 func == bpf_skb_adjust_room || 7102 func == sk_skb_adjust_room || 7103 func == bpf_skb_pull_data || 7104 func == sk_skb_pull_data || 7105 func == bpf_clone_redirect || 7106 func == bpf_l3_csum_replace || 7107 func == bpf_l4_csum_replace || 7108 func == bpf_xdp_adjust_head || 7109 func == bpf_xdp_adjust_meta || 7110 func == bpf_msg_pull_data || 7111 func == bpf_msg_push_data || 7112 func == bpf_msg_pop_data || 7113 func == bpf_xdp_adjust_tail || 7114 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 7115 func == bpf_lwt_seg6_store_bytes || 7116 func == bpf_lwt_seg6_adjust_srh || 7117 func == bpf_lwt_seg6_action || 7118 #endif 7119 #ifdef CONFIG_INET 7120 func == bpf_sock_ops_store_hdr_opt || 7121 #endif 7122 func == bpf_lwt_in_push_encap || 7123 func == bpf_lwt_xmit_push_encap) 7124 return true; 7125 7126 return false; 7127 } 7128 7129 const struct bpf_func_proto bpf_event_output_data_proto __weak; 7130 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak; 7131 7132 static const struct bpf_func_proto * 7133 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7134 { 7135 switch (func_id) { 7136 /* inet and inet6 sockets are created in a process 7137 * context so there is always a valid uid/gid 7138 */ 7139 case BPF_FUNC_get_current_uid_gid: 7140 return &bpf_get_current_uid_gid_proto; 7141 case BPF_FUNC_get_local_storage: 7142 return &bpf_get_local_storage_proto; 7143 case BPF_FUNC_get_socket_cookie: 7144 return &bpf_get_socket_cookie_sock_proto; 7145 case BPF_FUNC_get_netns_cookie: 7146 return &bpf_get_netns_cookie_sock_proto; 7147 case BPF_FUNC_perf_event_output: 7148 return &bpf_event_output_data_proto; 7149 case BPF_FUNC_get_current_pid_tgid: 7150 return &bpf_get_current_pid_tgid_proto; 7151 case BPF_FUNC_get_current_comm: 7152 return &bpf_get_current_comm_proto; 7153 #ifdef CONFIG_CGROUPS 7154 case BPF_FUNC_get_current_cgroup_id: 7155 return &bpf_get_current_cgroup_id_proto; 7156 case BPF_FUNC_get_current_ancestor_cgroup_id: 7157 return &bpf_get_current_ancestor_cgroup_id_proto; 7158 #endif 7159 #ifdef CONFIG_CGROUP_NET_CLASSID 7160 case BPF_FUNC_get_cgroup_classid: 7161 return &bpf_get_cgroup_classid_curr_proto; 7162 #endif 7163 case BPF_FUNC_sk_storage_get: 7164 return &bpf_sk_storage_get_cg_sock_proto; 7165 case BPF_FUNC_ktime_get_coarse_ns: 7166 return &bpf_ktime_get_coarse_ns_proto; 7167 default: 7168 return bpf_base_func_proto(func_id); 7169 } 7170 } 7171 7172 static const struct bpf_func_proto * 7173 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7174 { 7175 switch (func_id) { 7176 /* inet and inet6 sockets are created in a process 7177 * context so there is always a valid uid/gid 7178 */ 7179 case BPF_FUNC_get_current_uid_gid: 7180 return &bpf_get_current_uid_gid_proto; 7181 case BPF_FUNC_bind: 7182 switch (prog->expected_attach_type) { 7183 case BPF_CGROUP_INET4_CONNECT: 7184 case BPF_CGROUP_INET6_CONNECT: 7185 return &bpf_bind_proto; 7186 default: 7187 return NULL; 7188 } 7189 case BPF_FUNC_get_socket_cookie: 7190 return &bpf_get_socket_cookie_sock_addr_proto; 7191 case BPF_FUNC_get_netns_cookie: 7192 return &bpf_get_netns_cookie_sock_addr_proto; 7193 case BPF_FUNC_get_local_storage: 7194 return &bpf_get_local_storage_proto; 7195 case BPF_FUNC_perf_event_output: 7196 return &bpf_event_output_data_proto; 7197 case BPF_FUNC_get_current_pid_tgid: 7198 return &bpf_get_current_pid_tgid_proto; 7199 case BPF_FUNC_get_current_comm: 7200 return &bpf_get_current_comm_proto; 7201 #ifdef CONFIG_CGROUPS 7202 case BPF_FUNC_get_current_cgroup_id: 7203 return &bpf_get_current_cgroup_id_proto; 7204 case BPF_FUNC_get_current_ancestor_cgroup_id: 7205 return &bpf_get_current_ancestor_cgroup_id_proto; 7206 #endif 7207 #ifdef CONFIG_CGROUP_NET_CLASSID 7208 case BPF_FUNC_get_cgroup_classid: 7209 return &bpf_get_cgroup_classid_curr_proto; 7210 #endif 7211 #ifdef CONFIG_INET 7212 case BPF_FUNC_sk_lookup_tcp: 7213 return &bpf_sock_addr_sk_lookup_tcp_proto; 7214 case BPF_FUNC_sk_lookup_udp: 7215 return &bpf_sock_addr_sk_lookup_udp_proto; 7216 case BPF_FUNC_sk_release: 7217 return &bpf_sk_release_proto; 7218 case BPF_FUNC_skc_lookup_tcp: 7219 return &bpf_sock_addr_skc_lookup_tcp_proto; 7220 #endif /* CONFIG_INET */ 7221 case BPF_FUNC_sk_storage_get: 7222 return &bpf_sk_storage_get_proto; 7223 case BPF_FUNC_sk_storage_delete: 7224 return &bpf_sk_storage_delete_proto; 7225 case BPF_FUNC_setsockopt: 7226 switch (prog->expected_attach_type) { 7227 case BPF_CGROUP_INET4_BIND: 7228 case BPF_CGROUP_INET6_BIND: 7229 case BPF_CGROUP_INET4_CONNECT: 7230 case BPF_CGROUP_INET6_CONNECT: 7231 case BPF_CGROUP_UDP4_RECVMSG: 7232 case BPF_CGROUP_UDP6_RECVMSG: 7233 case BPF_CGROUP_UDP4_SENDMSG: 7234 case BPF_CGROUP_UDP6_SENDMSG: 7235 case BPF_CGROUP_INET4_GETPEERNAME: 7236 case BPF_CGROUP_INET6_GETPEERNAME: 7237 case BPF_CGROUP_INET4_GETSOCKNAME: 7238 case BPF_CGROUP_INET6_GETSOCKNAME: 7239 return &bpf_sock_addr_setsockopt_proto; 7240 default: 7241 return NULL; 7242 } 7243 case BPF_FUNC_getsockopt: 7244 switch (prog->expected_attach_type) { 7245 case BPF_CGROUP_INET4_BIND: 7246 case BPF_CGROUP_INET6_BIND: 7247 case BPF_CGROUP_INET4_CONNECT: 7248 case BPF_CGROUP_INET6_CONNECT: 7249 case BPF_CGROUP_UDP4_RECVMSG: 7250 case BPF_CGROUP_UDP6_RECVMSG: 7251 case BPF_CGROUP_UDP4_SENDMSG: 7252 case BPF_CGROUP_UDP6_SENDMSG: 7253 case BPF_CGROUP_INET4_GETPEERNAME: 7254 case BPF_CGROUP_INET6_GETPEERNAME: 7255 case BPF_CGROUP_INET4_GETSOCKNAME: 7256 case BPF_CGROUP_INET6_GETSOCKNAME: 7257 return &bpf_sock_addr_getsockopt_proto; 7258 default: 7259 return NULL; 7260 } 7261 default: 7262 return bpf_sk_base_func_proto(func_id); 7263 } 7264 } 7265 7266 static const struct bpf_func_proto * 7267 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7268 { 7269 switch (func_id) { 7270 case BPF_FUNC_skb_load_bytes: 7271 return &bpf_skb_load_bytes_proto; 7272 case BPF_FUNC_skb_load_bytes_relative: 7273 return &bpf_skb_load_bytes_relative_proto; 7274 case BPF_FUNC_get_socket_cookie: 7275 return &bpf_get_socket_cookie_proto; 7276 case BPF_FUNC_get_socket_uid: 7277 return &bpf_get_socket_uid_proto; 7278 case BPF_FUNC_perf_event_output: 7279 return &bpf_skb_event_output_proto; 7280 default: 7281 return bpf_sk_base_func_proto(func_id); 7282 } 7283 } 7284 7285 const struct bpf_func_proto bpf_sk_storage_get_proto __weak; 7286 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak; 7287 7288 static const struct bpf_func_proto * 7289 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7290 { 7291 switch (func_id) { 7292 case BPF_FUNC_get_local_storage: 7293 return &bpf_get_local_storage_proto; 7294 case BPF_FUNC_sk_fullsock: 7295 return &bpf_sk_fullsock_proto; 7296 case BPF_FUNC_sk_storage_get: 7297 return &bpf_sk_storage_get_proto; 7298 case BPF_FUNC_sk_storage_delete: 7299 return &bpf_sk_storage_delete_proto; 7300 case BPF_FUNC_perf_event_output: 7301 return &bpf_skb_event_output_proto; 7302 #ifdef CONFIG_SOCK_CGROUP_DATA 7303 case BPF_FUNC_skb_cgroup_id: 7304 return &bpf_skb_cgroup_id_proto; 7305 case BPF_FUNC_skb_ancestor_cgroup_id: 7306 return &bpf_skb_ancestor_cgroup_id_proto; 7307 case BPF_FUNC_sk_cgroup_id: 7308 return &bpf_sk_cgroup_id_proto; 7309 case BPF_FUNC_sk_ancestor_cgroup_id: 7310 return &bpf_sk_ancestor_cgroup_id_proto; 7311 #endif 7312 #ifdef CONFIG_INET 7313 case BPF_FUNC_sk_lookup_tcp: 7314 return &bpf_sk_lookup_tcp_proto; 7315 case BPF_FUNC_sk_lookup_udp: 7316 return &bpf_sk_lookup_udp_proto; 7317 case BPF_FUNC_sk_release: 7318 return &bpf_sk_release_proto; 7319 case BPF_FUNC_skc_lookup_tcp: 7320 return &bpf_skc_lookup_tcp_proto; 7321 case BPF_FUNC_tcp_sock: 7322 return &bpf_tcp_sock_proto; 7323 case BPF_FUNC_get_listener_sock: 7324 return &bpf_get_listener_sock_proto; 7325 case BPF_FUNC_skb_ecn_set_ce: 7326 return &bpf_skb_ecn_set_ce_proto; 7327 #endif 7328 default: 7329 return sk_filter_func_proto(func_id, prog); 7330 } 7331 } 7332 7333 static const struct bpf_func_proto * 7334 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7335 { 7336 switch (func_id) { 7337 case BPF_FUNC_skb_store_bytes: 7338 return &bpf_skb_store_bytes_proto; 7339 case BPF_FUNC_skb_load_bytes: 7340 return &bpf_skb_load_bytes_proto; 7341 case BPF_FUNC_skb_load_bytes_relative: 7342 return &bpf_skb_load_bytes_relative_proto; 7343 case BPF_FUNC_skb_pull_data: 7344 return &bpf_skb_pull_data_proto; 7345 case BPF_FUNC_csum_diff: 7346 return &bpf_csum_diff_proto; 7347 case BPF_FUNC_csum_update: 7348 return &bpf_csum_update_proto; 7349 case BPF_FUNC_csum_level: 7350 return &bpf_csum_level_proto; 7351 case BPF_FUNC_l3_csum_replace: 7352 return &bpf_l3_csum_replace_proto; 7353 case BPF_FUNC_l4_csum_replace: 7354 return &bpf_l4_csum_replace_proto; 7355 case BPF_FUNC_clone_redirect: 7356 return &bpf_clone_redirect_proto; 7357 case BPF_FUNC_get_cgroup_classid: 7358 return &bpf_get_cgroup_classid_proto; 7359 case BPF_FUNC_skb_vlan_push: 7360 return &bpf_skb_vlan_push_proto; 7361 case BPF_FUNC_skb_vlan_pop: 7362 return &bpf_skb_vlan_pop_proto; 7363 case BPF_FUNC_skb_change_proto: 7364 return &bpf_skb_change_proto_proto; 7365 case BPF_FUNC_skb_change_type: 7366 return &bpf_skb_change_type_proto; 7367 case BPF_FUNC_skb_adjust_room: 7368 return &bpf_skb_adjust_room_proto; 7369 case BPF_FUNC_skb_change_tail: 7370 return &bpf_skb_change_tail_proto; 7371 case BPF_FUNC_skb_change_head: 7372 return &bpf_skb_change_head_proto; 7373 case BPF_FUNC_skb_get_tunnel_key: 7374 return &bpf_skb_get_tunnel_key_proto; 7375 case BPF_FUNC_skb_set_tunnel_key: 7376 return bpf_get_skb_set_tunnel_proto(func_id); 7377 case BPF_FUNC_skb_get_tunnel_opt: 7378 return &bpf_skb_get_tunnel_opt_proto; 7379 case BPF_FUNC_skb_set_tunnel_opt: 7380 return bpf_get_skb_set_tunnel_proto(func_id); 7381 case BPF_FUNC_redirect: 7382 return &bpf_redirect_proto; 7383 case BPF_FUNC_redirect_neigh: 7384 return &bpf_redirect_neigh_proto; 7385 case BPF_FUNC_redirect_peer: 7386 return &bpf_redirect_peer_proto; 7387 case BPF_FUNC_get_route_realm: 7388 return &bpf_get_route_realm_proto; 7389 case BPF_FUNC_get_hash_recalc: 7390 return &bpf_get_hash_recalc_proto; 7391 case BPF_FUNC_set_hash_invalid: 7392 return &bpf_set_hash_invalid_proto; 7393 case BPF_FUNC_set_hash: 7394 return &bpf_set_hash_proto; 7395 case BPF_FUNC_perf_event_output: 7396 return &bpf_skb_event_output_proto; 7397 case BPF_FUNC_get_smp_processor_id: 7398 return &bpf_get_smp_processor_id_proto; 7399 case BPF_FUNC_skb_under_cgroup: 7400 return &bpf_skb_under_cgroup_proto; 7401 case BPF_FUNC_get_socket_cookie: 7402 return &bpf_get_socket_cookie_proto; 7403 case BPF_FUNC_get_socket_uid: 7404 return &bpf_get_socket_uid_proto; 7405 case BPF_FUNC_fib_lookup: 7406 return &bpf_skb_fib_lookup_proto; 7407 case BPF_FUNC_check_mtu: 7408 return &bpf_skb_check_mtu_proto; 7409 case BPF_FUNC_sk_fullsock: 7410 return &bpf_sk_fullsock_proto; 7411 case BPF_FUNC_sk_storage_get: 7412 return &bpf_sk_storage_get_proto; 7413 case BPF_FUNC_sk_storage_delete: 7414 return &bpf_sk_storage_delete_proto; 7415 #ifdef CONFIG_XFRM 7416 case BPF_FUNC_skb_get_xfrm_state: 7417 return &bpf_skb_get_xfrm_state_proto; 7418 #endif 7419 #ifdef CONFIG_CGROUP_NET_CLASSID 7420 case BPF_FUNC_skb_cgroup_classid: 7421 return &bpf_skb_cgroup_classid_proto; 7422 #endif 7423 #ifdef CONFIG_SOCK_CGROUP_DATA 7424 case BPF_FUNC_skb_cgroup_id: 7425 return &bpf_skb_cgroup_id_proto; 7426 case BPF_FUNC_skb_ancestor_cgroup_id: 7427 return &bpf_skb_ancestor_cgroup_id_proto; 7428 #endif 7429 #ifdef CONFIG_INET 7430 case BPF_FUNC_sk_lookup_tcp: 7431 return &bpf_sk_lookup_tcp_proto; 7432 case BPF_FUNC_sk_lookup_udp: 7433 return &bpf_sk_lookup_udp_proto; 7434 case BPF_FUNC_sk_release: 7435 return &bpf_sk_release_proto; 7436 case BPF_FUNC_tcp_sock: 7437 return &bpf_tcp_sock_proto; 7438 case BPF_FUNC_get_listener_sock: 7439 return &bpf_get_listener_sock_proto; 7440 case BPF_FUNC_skc_lookup_tcp: 7441 return &bpf_skc_lookup_tcp_proto; 7442 case BPF_FUNC_tcp_check_syncookie: 7443 return &bpf_tcp_check_syncookie_proto; 7444 case BPF_FUNC_skb_ecn_set_ce: 7445 return &bpf_skb_ecn_set_ce_proto; 7446 case BPF_FUNC_tcp_gen_syncookie: 7447 return &bpf_tcp_gen_syncookie_proto; 7448 case BPF_FUNC_sk_assign: 7449 return &bpf_sk_assign_proto; 7450 #endif 7451 default: 7452 return bpf_sk_base_func_proto(func_id); 7453 } 7454 } 7455 7456 static const struct bpf_func_proto * 7457 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7458 { 7459 switch (func_id) { 7460 case BPF_FUNC_perf_event_output: 7461 return &bpf_xdp_event_output_proto; 7462 case BPF_FUNC_get_smp_processor_id: 7463 return &bpf_get_smp_processor_id_proto; 7464 case BPF_FUNC_csum_diff: 7465 return &bpf_csum_diff_proto; 7466 case BPF_FUNC_xdp_adjust_head: 7467 return &bpf_xdp_adjust_head_proto; 7468 case BPF_FUNC_xdp_adjust_meta: 7469 return &bpf_xdp_adjust_meta_proto; 7470 case BPF_FUNC_redirect: 7471 return &bpf_xdp_redirect_proto; 7472 case BPF_FUNC_redirect_map: 7473 return &bpf_xdp_redirect_map_proto; 7474 case BPF_FUNC_xdp_adjust_tail: 7475 return &bpf_xdp_adjust_tail_proto; 7476 case BPF_FUNC_fib_lookup: 7477 return &bpf_xdp_fib_lookup_proto; 7478 case BPF_FUNC_check_mtu: 7479 return &bpf_xdp_check_mtu_proto; 7480 #ifdef CONFIG_INET 7481 case BPF_FUNC_sk_lookup_udp: 7482 return &bpf_xdp_sk_lookup_udp_proto; 7483 case BPF_FUNC_sk_lookup_tcp: 7484 return &bpf_xdp_sk_lookup_tcp_proto; 7485 case BPF_FUNC_sk_release: 7486 return &bpf_sk_release_proto; 7487 case BPF_FUNC_skc_lookup_tcp: 7488 return &bpf_xdp_skc_lookup_tcp_proto; 7489 case BPF_FUNC_tcp_check_syncookie: 7490 return &bpf_tcp_check_syncookie_proto; 7491 case BPF_FUNC_tcp_gen_syncookie: 7492 return &bpf_tcp_gen_syncookie_proto; 7493 #endif 7494 default: 7495 return bpf_sk_base_func_proto(func_id); 7496 } 7497 } 7498 7499 const struct bpf_func_proto bpf_sock_map_update_proto __weak; 7500 const struct bpf_func_proto bpf_sock_hash_update_proto __weak; 7501 7502 static const struct bpf_func_proto * 7503 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7504 { 7505 switch (func_id) { 7506 case BPF_FUNC_setsockopt: 7507 return &bpf_sock_ops_setsockopt_proto; 7508 case BPF_FUNC_getsockopt: 7509 return &bpf_sock_ops_getsockopt_proto; 7510 case BPF_FUNC_sock_ops_cb_flags_set: 7511 return &bpf_sock_ops_cb_flags_set_proto; 7512 case BPF_FUNC_sock_map_update: 7513 return &bpf_sock_map_update_proto; 7514 case BPF_FUNC_sock_hash_update: 7515 return &bpf_sock_hash_update_proto; 7516 case BPF_FUNC_get_socket_cookie: 7517 return &bpf_get_socket_cookie_sock_ops_proto; 7518 case BPF_FUNC_get_local_storage: 7519 return &bpf_get_local_storage_proto; 7520 case BPF_FUNC_perf_event_output: 7521 return &bpf_event_output_data_proto; 7522 case BPF_FUNC_sk_storage_get: 7523 return &bpf_sk_storage_get_proto; 7524 case BPF_FUNC_sk_storage_delete: 7525 return &bpf_sk_storage_delete_proto; 7526 case BPF_FUNC_get_netns_cookie: 7527 return &bpf_get_netns_cookie_sock_ops_proto; 7528 #ifdef CONFIG_INET 7529 case BPF_FUNC_load_hdr_opt: 7530 return &bpf_sock_ops_load_hdr_opt_proto; 7531 case BPF_FUNC_store_hdr_opt: 7532 return &bpf_sock_ops_store_hdr_opt_proto; 7533 case BPF_FUNC_reserve_hdr_opt: 7534 return &bpf_sock_ops_reserve_hdr_opt_proto; 7535 case BPF_FUNC_tcp_sock: 7536 return &bpf_tcp_sock_proto; 7537 #endif /* CONFIG_INET */ 7538 default: 7539 return bpf_sk_base_func_proto(func_id); 7540 } 7541 } 7542 7543 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak; 7544 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak; 7545 7546 static const struct bpf_func_proto * 7547 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7548 { 7549 switch (func_id) { 7550 case BPF_FUNC_msg_redirect_map: 7551 return &bpf_msg_redirect_map_proto; 7552 case BPF_FUNC_msg_redirect_hash: 7553 return &bpf_msg_redirect_hash_proto; 7554 case BPF_FUNC_msg_apply_bytes: 7555 return &bpf_msg_apply_bytes_proto; 7556 case BPF_FUNC_msg_cork_bytes: 7557 return &bpf_msg_cork_bytes_proto; 7558 case BPF_FUNC_msg_pull_data: 7559 return &bpf_msg_pull_data_proto; 7560 case BPF_FUNC_msg_push_data: 7561 return &bpf_msg_push_data_proto; 7562 case BPF_FUNC_msg_pop_data: 7563 return &bpf_msg_pop_data_proto; 7564 case BPF_FUNC_perf_event_output: 7565 return &bpf_event_output_data_proto; 7566 case BPF_FUNC_get_current_uid_gid: 7567 return &bpf_get_current_uid_gid_proto; 7568 case BPF_FUNC_get_current_pid_tgid: 7569 return &bpf_get_current_pid_tgid_proto; 7570 case BPF_FUNC_sk_storage_get: 7571 return &bpf_sk_storage_get_proto; 7572 case BPF_FUNC_sk_storage_delete: 7573 return &bpf_sk_storage_delete_proto; 7574 case BPF_FUNC_get_netns_cookie: 7575 return &bpf_get_netns_cookie_sk_msg_proto; 7576 #ifdef CONFIG_CGROUPS 7577 case BPF_FUNC_get_current_cgroup_id: 7578 return &bpf_get_current_cgroup_id_proto; 7579 case BPF_FUNC_get_current_ancestor_cgroup_id: 7580 return &bpf_get_current_ancestor_cgroup_id_proto; 7581 #endif 7582 #ifdef CONFIG_CGROUP_NET_CLASSID 7583 case BPF_FUNC_get_cgroup_classid: 7584 return &bpf_get_cgroup_classid_curr_proto; 7585 #endif 7586 default: 7587 return bpf_sk_base_func_proto(func_id); 7588 } 7589 } 7590 7591 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak; 7592 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak; 7593 7594 static const struct bpf_func_proto * 7595 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7596 { 7597 switch (func_id) { 7598 case BPF_FUNC_skb_store_bytes: 7599 return &bpf_skb_store_bytes_proto; 7600 case BPF_FUNC_skb_load_bytes: 7601 return &bpf_skb_load_bytes_proto; 7602 case BPF_FUNC_skb_pull_data: 7603 return &sk_skb_pull_data_proto; 7604 case BPF_FUNC_skb_change_tail: 7605 return &sk_skb_change_tail_proto; 7606 case BPF_FUNC_skb_change_head: 7607 return &sk_skb_change_head_proto; 7608 case BPF_FUNC_skb_adjust_room: 7609 return &sk_skb_adjust_room_proto; 7610 case BPF_FUNC_get_socket_cookie: 7611 return &bpf_get_socket_cookie_proto; 7612 case BPF_FUNC_get_socket_uid: 7613 return &bpf_get_socket_uid_proto; 7614 case BPF_FUNC_sk_redirect_map: 7615 return &bpf_sk_redirect_map_proto; 7616 case BPF_FUNC_sk_redirect_hash: 7617 return &bpf_sk_redirect_hash_proto; 7618 case BPF_FUNC_perf_event_output: 7619 return &bpf_skb_event_output_proto; 7620 #ifdef CONFIG_INET 7621 case BPF_FUNC_sk_lookup_tcp: 7622 return &bpf_sk_lookup_tcp_proto; 7623 case BPF_FUNC_sk_lookup_udp: 7624 return &bpf_sk_lookup_udp_proto; 7625 case BPF_FUNC_sk_release: 7626 return &bpf_sk_release_proto; 7627 case BPF_FUNC_skc_lookup_tcp: 7628 return &bpf_skc_lookup_tcp_proto; 7629 #endif 7630 default: 7631 return bpf_sk_base_func_proto(func_id); 7632 } 7633 } 7634 7635 static const struct bpf_func_proto * 7636 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7637 { 7638 switch (func_id) { 7639 case BPF_FUNC_skb_load_bytes: 7640 return &bpf_flow_dissector_load_bytes_proto; 7641 default: 7642 return bpf_sk_base_func_proto(func_id); 7643 } 7644 } 7645 7646 static const struct bpf_func_proto * 7647 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7648 { 7649 switch (func_id) { 7650 case BPF_FUNC_skb_load_bytes: 7651 return &bpf_skb_load_bytes_proto; 7652 case BPF_FUNC_skb_pull_data: 7653 return &bpf_skb_pull_data_proto; 7654 case BPF_FUNC_csum_diff: 7655 return &bpf_csum_diff_proto; 7656 case BPF_FUNC_get_cgroup_classid: 7657 return &bpf_get_cgroup_classid_proto; 7658 case BPF_FUNC_get_route_realm: 7659 return &bpf_get_route_realm_proto; 7660 case BPF_FUNC_get_hash_recalc: 7661 return &bpf_get_hash_recalc_proto; 7662 case BPF_FUNC_perf_event_output: 7663 return &bpf_skb_event_output_proto; 7664 case BPF_FUNC_get_smp_processor_id: 7665 return &bpf_get_smp_processor_id_proto; 7666 case BPF_FUNC_skb_under_cgroup: 7667 return &bpf_skb_under_cgroup_proto; 7668 default: 7669 return bpf_sk_base_func_proto(func_id); 7670 } 7671 } 7672 7673 static const struct bpf_func_proto * 7674 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7675 { 7676 switch (func_id) { 7677 case BPF_FUNC_lwt_push_encap: 7678 return &bpf_lwt_in_push_encap_proto; 7679 default: 7680 return lwt_out_func_proto(func_id, prog); 7681 } 7682 } 7683 7684 static const struct bpf_func_proto * 7685 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7686 { 7687 switch (func_id) { 7688 case BPF_FUNC_skb_get_tunnel_key: 7689 return &bpf_skb_get_tunnel_key_proto; 7690 case BPF_FUNC_skb_set_tunnel_key: 7691 return bpf_get_skb_set_tunnel_proto(func_id); 7692 case BPF_FUNC_skb_get_tunnel_opt: 7693 return &bpf_skb_get_tunnel_opt_proto; 7694 case BPF_FUNC_skb_set_tunnel_opt: 7695 return bpf_get_skb_set_tunnel_proto(func_id); 7696 case BPF_FUNC_redirect: 7697 return &bpf_redirect_proto; 7698 case BPF_FUNC_clone_redirect: 7699 return &bpf_clone_redirect_proto; 7700 case BPF_FUNC_skb_change_tail: 7701 return &bpf_skb_change_tail_proto; 7702 case BPF_FUNC_skb_change_head: 7703 return &bpf_skb_change_head_proto; 7704 case BPF_FUNC_skb_store_bytes: 7705 return &bpf_skb_store_bytes_proto; 7706 case BPF_FUNC_csum_update: 7707 return &bpf_csum_update_proto; 7708 case BPF_FUNC_csum_level: 7709 return &bpf_csum_level_proto; 7710 case BPF_FUNC_l3_csum_replace: 7711 return &bpf_l3_csum_replace_proto; 7712 case BPF_FUNC_l4_csum_replace: 7713 return &bpf_l4_csum_replace_proto; 7714 case BPF_FUNC_set_hash_invalid: 7715 return &bpf_set_hash_invalid_proto; 7716 case BPF_FUNC_lwt_push_encap: 7717 return &bpf_lwt_xmit_push_encap_proto; 7718 default: 7719 return lwt_out_func_proto(func_id, prog); 7720 } 7721 } 7722 7723 static const struct bpf_func_proto * 7724 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7725 { 7726 switch (func_id) { 7727 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 7728 case BPF_FUNC_lwt_seg6_store_bytes: 7729 return &bpf_lwt_seg6_store_bytes_proto; 7730 case BPF_FUNC_lwt_seg6_action: 7731 return &bpf_lwt_seg6_action_proto; 7732 case BPF_FUNC_lwt_seg6_adjust_srh: 7733 return &bpf_lwt_seg6_adjust_srh_proto; 7734 #endif 7735 default: 7736 return lwt_out_func_proto(func_id, prog); 7737 } 7738 } 7739 7740 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type, 7741 const struct bpf_prog *prog, 7742 struct bpf_insn_access_aux *info) 7743 { 7744 const int size_default = sizeof(__u32); 7745 7746 if (off < 0 || off >= sizeof(struct __sk_buff)) 7747 return false; 7748 7749 /* The verifier guarantees that size > 0. */ 7750 if (off % size != 0) 7751 return false; 7752 7753 switch (off) { 7754 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7755 if (off + size > offsetofend(struct __sk_buff, cb[4])) 7756 return false; 7757 break; 7758 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]): 7759 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]): 7760 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4): 7761 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4): 7762 case bpf_ctx_range(struct __sk_buff, data): 7763 case bpf_ctx_range(struct __sk_buff, data_meta): 7764 case bpf_ctx_range(struct __sk_buff, data_end): 7765 if (size != size_default) 7766 return false; 7767 break; 7768 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 7769 return false; 7770 case bpf_ctx_range(struct __sk_buff, hwtstamp): 7771 if (type == BPF_WRITE || size != sizeof(__u64)) 7772 return false; 7773 break; 7774 case bpf_ctx_range(struct __sk_buff, tstamp): 7775 if (size != sizeof(__u64)) 7776 return false; 7777 break; 7778 case offsetof(struct __sk_buff, sk): 7779 if (type == BPF_WRITE || size != sizeof(__u64)) 7780 return false; 7781 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 7782 break; 7783 case offsetofend(struct __sk_buff, gso_size) ... offsetof(struct __sk_buff, hwtstamp) - 1: 7784 /* Explicitly prohibit access to padding in __sk_buff. */ 7785 return false; 7786 default: 7787 /* Only narrow read access allowed for now. */ 7788 if (type == BPF_WRITE) { 7789 if (size != size_default) 7790 return false; 7791 } else { 7792 bpf_ctx_record_field_size(info, size_default); 7793 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 7794 return false; 7795 } 7796 } 7797 7798 return true; 7799 } 7800 7801 static bool sk_filter_is_valid_access(int off, int size, 7802 enum bpf_access_type type, 7803 const struct bpf_prog *prog, 7804 struct bpf_insn_access_aux *info) 7805 { 7806 switch (off) { 7807 case bpf_ctx_range(struct __sk_buff, tc_classid): 7808 case bpf_ctx_range(struct __sk_buff, data): 7809 case bpf_ctx_range(struct __sk_buff, data_meta): 7810 case bpf_ctx_range(struct __sk_buff, data_end): 7811 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 7812 case bpf_ctx_range(struct __sk_buff, tstamp): 7813 case bpf_ctx_range(struct __sk_buff, wire_len): 7814 case bpf_ctx_range(struct __sk_buff, hwtstamp): 7815 return false; 7816 } 7817 7818 if (type == BPF_WRITE) { 7819 switch (off) { 7820 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7821 break; 7822 default: 7823 return false; 7824 } 7825 } 7826 7827 return bpf_skb_is_valid_access(off, size, type, prog, info); 7828 } 7829 7830 static bool cg_skb_is_valid_access(int off, int size, 7831 enum bpf_access_type type, 7832 const struct bpf_prog *prog, 7833 struct bpf_insn_access_aux *info) 7834 { 7835 switch (off) { 7836 case bpf_ctx_range(struct __sk_buff, tc_classid): 7837 case bpf_ctx_range(struct __sk_buff, data_meta): 7838 case bpf_ctx_range(struct __sk_buff, wire_len): 7839 return false; 7840 case bpf_ctx_range(struct __sk_buff, data): 7841 case bpf_ctx_range(struct __sk_buff, data_end): 7842 if (!bpf_capable()) 7843 return false; 7844 break; 7845 } 7846 7847 if (type == BPF_WRITE) { 7848 switch (off) { 7849 case bpf_ctx_range(struct __sk_buff, mark): 7850 case bpf_ctx_range(struct __sk_buff, priority): 7851 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7852 break; 7853 case bpf_ctx_range(struct __sk_buff, tstamp): 7854 if (!bpf_capable()) 7855 return false; 7856 break; 7857 default: 7858 return false; 7859 } 7860 } 7861 7862 switch (off) { 7863 case bpf_ctx_range(struct __sk_buff, data): 7864 info->reg_type = PTR_TO_PACKET; 7865 break; 7866 case bpf_ctx_range(struct __sk_buff, data_end): 7867 info->reg_type = PTR_TO_PACKET_END; 7868 break; 7869 } 7870 7871 return bpf_skb_is_valid_access(off, size, type, prog, info); 7872 } 7873 7874 static bool lwt_is_valid_access(int off, int size, 7875 enum bpf_access_type type, 7876 const struct bpf_prog *prog, 7877 struct bpf_insn_access_aux *info) 7878 { 7879 switch (off) { 7880 case bpf_ctx_range(struct __sk_buff, tc_classid): 7881 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 7882 case bpf_ctx_range(struct __sk_buff, data_meta): 7883 case bpf_ctx_range(struct __sk_buff, tstamp): 7884 case bpf_ctx_range(struct __sk_buff, wire_len): 7885 case bpf_ctx_range(struct __sk_buff, hwtstamp): 7886 return false; 7887 } 7888 7889 if (type == BPF_WRITE) { 7890 switch (off) { 7891 case bpf_ctx_range(struct __sk_buff, mark): 7892 case bpf_ctx_range(struct __sk_buff, priority): 7893 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 7894 break; 7895 default: 7896 return false; 7897 } 7898 } 7899 7900 switch (off) { 7901 case bpf_ctx_range(struct __sk_buff, data): 7902 info->reg_type = PTR_TO_PACKET; 7903 break; 7904 case bpf_ctx_range(struct __sk_buff, data_end): 7905 info->reg_type = PTR_TO_PACKET_END; 7906 break; 7907 } 7908 7909 return bpf_skb_is_valid_access(off, size, type, prog, info); 7910 } 7911 7912 /* Attach type specific accesses */ 7913 static bool __sock_filter_check_attach_type(int off, 7914 enum bpf_access_type access_type, 7915 enum bpf_attach_type attach_type) 7916 { 7917 switch (off) { 7918 case offsetof(struct bpf_sock, bound_dev_if): 7919 case offsetof(struct bpf_sock, mark): 7920 case offsetof(struct bpf_sock, priority): 7921 switch (attach_type) { 7922 case BPF_CGROUP_INET_SOCK_CREATE: 7923 case BPF_CGROUP_INET_SOCK_RELEASE: 7924 goto full_access; 7925 default: 7926 return false; 7927 } 7928 case bpf_ctx_range(struct bpf_sock, src_ip4): 7929 switch (attach_type) { 7930 case BPF_CGROUP_INET4_POST_BIND: 7931 goto read_only; 7932 default: 7933 return false; 7934 } 7935 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 7936 switch (attach_type) { 7937 case BPF_CGROUP_INET6_POST_BIND: 7938 goto read_only; 7939 default: 7940 return false; 7941 } 7942 case bpf_ctx_range(struct bpf_sock, src_port): 7943 switch (attach_type) { 7944 case BPF_CGROUP_INET4_POST_BIND: 7945 case BPF_CGROUP_INET6_POST_BIND: 7946 goto read_only; 7947 default: 7948 return false; 7949 } 7950 } 7951 read_only: 7952 return access_type == BPF_READ; 7953 full_access: 7954 return true; 7955 } 7956 7957 bool bpf_sock_common_is_valid_access(int off, int size, 7958 enum bpf_access_type type, 7959 struct bpf_insn_access_aux *info) 7960 { 7961 switch (off) { 7962 case bpf_ctx_range_till(struct bpf_sock, type, priority): 7963 return false; 7964 default: 7965 return bpf_sock_is_valid_access(off, size, type, info); 7966 } 7967 } 7968 7969 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 7970 struct bpf_insn_access_aux *info) 7971 { 7972 const int size_default = sizeof(__u32); 7973 7974 if (off < 0 || off >= sizeof(struct bpf_sock)) 7975 return false; 7976 if (off % size != 0) 7977 return false; 7978 7979 switch (off) { 7980 case offsetof(struct bpf_sock, state): 7981 case offsetof(struct bpf_sock, family): 7982 case offsetof(struct bpf_sock, type): 7983 case offsetof(struct bpf_sock, protocol): 7984 case offsetof(struct bpf_sock, dst_port): 7985 case offsetof(struct bpf_sock, src_port): 7986 case offsetof(struct bpf_sock, rx_queue_mapping): 7987 case bpf_ctx_range(struct bpf_sock, src_ip4): 7988 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 7989 case bpf_ctx_range(struct bpf_sock, dst_ip4): 7990 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 7991 bpf_ctx_record_field_size(info, size_default); 7992 return bpf_ctx_narrow_access_ok(off, size, size_default); 7993 } 7994 7995 return size == size_default; 7996 } 7997 7998 static bool sock_filter_is_valid_access(int off, int size, 7999 enum bpf_access_type type, 8000 const struct bpf_prog *prog, 8001 struct bpf_insn_access_aux *info) 8002 { 8003 if (!bpf_sock_is_valid_access(off, size, type, info)) 8004 return false; 8005 return __sock_filter_check_attach_type(off, type, 8006 prog->expected_attach_type); 8007 } 8008 8009 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write, 8010 const struct bpf_prog *prog) 8011 { 8012 /* Neither direct read nor direct write requires any preliminary 8013 * action. 8014 */ 8015 return 0; 8016 } 8017 8018 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write, 8019 const struct bpf_prog *prog, int drop_verdict) 8020 { 8021 struct bpf_insn *insn = insn_buf; 8022 8023 if (!direct_write) 8024 return 0; 8025 8026 /* if (!skb->cloned) 8027 * goto start; 8028 * 8029 * (Fast-path, otherwise approximation that we might be 8030 * a clone, do the rest in helper.) 8031 */ 8032 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET()); 8033 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK); 8034 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7); 8035 8036 /* ret = bpf_skb_pull_data(skb, 0); */ 8037 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1); 8038 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2); 8039 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, 8040 BPF_FUNC_skb_pull_data); 8041 /* if (!ret) 8042 * goto restore; 8043 * return TC_ACT_SHOT; 8044 */ 8045 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2); 8046 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict); 8047 *insn++ = BPF_EXIT_INSN(); 8048 8049 /* restore: */ 8050 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6); 8051 /* start: */ 8052 *insn++ = prog->insnsi[0]; 8053 8054 return insn - insn_buf; 8055 } 8056 8057 static int bpf_gen_ld_abs(const struct bpf_insn *orig, 8058 struct bpf_insn *insn_buf) 8059 { 8060 bool indirect = BPF_MODE(orig->code) == BPF_IND; 8061 struct bpf_insn *insn = insn_buf; 8062 8063 if (!indirect) { 8064 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm); 8065 } else { 8066 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg); 8067 if (orig->imm) 8068 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm); 8069 } 8070 /* We're guaranteed here that CTX is in R6. */ 8071 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX); 8072 8073 switch (BPF_SIZE(orig->code)) { 8074 case BPF_B: 8075 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache); 8076 break; 8077 case BPF_H: 8078 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache); 8079 break; 8080 case BPF_W: 8081 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache); 8082 break; 8083 } 8084 8085 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2); 8086 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0); 8087 *insn++ = BPF_EXIT_INSN(); 8088 8089 return insn - insn_buf; 8090 } 8091 8092 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write, 8093 const struct bpf_prog *prog) 8094 { 8095 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT); 8096 } 8097 8098 static bool tc_cls_act_is_valid_access(int off, int size, 8099 enum bpf_access_type type, 8100 const struct bpf_prog *prog, 8101 struct bpf_insn_access_aux *info) 8102 { 8103 if (type == BPF_WRITE) { 8104 switch (off) { 8105 case bpf_ctx_range(struct __sk_buff, mark): 8106 case bpf_ctx_range(struct __sk_buff, tc_index): 8107 case bpf_ctx_range(struct __sk_buff, priority): 8108 case bpf_ctx_range(struct __sk_buff, tc_classid): 8109 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8110 case bpf_ctx_range(struct __sk_buff, tstamp): 8111 case bpf_ctx_range(struct __sk_buff, queue_mapping): 8112 break; 8113 default: 8114 return false; 8115 } 8116 } 8117 8118 switch (off) { 8119 case bpf_ctx_range(struct __sk_buff, data): 8120 info->reg_type = PTR_TO_PACKET; 8121 break; 8122 case bpf_ctx_range(struct __sk_buff, data_meta): 8123 info->reg_type = PTR_TO_PACKET_META; 8124 break; 8125 case bpf_ctx_range(struct __sk_buff, data_end): 8126 info->reg_type = PTR_TO_PACKET_END; 8127 break; 8128 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8129 return false; 8130 } 8131 8132 return bpf_skb_is_valid_access(off, size, type, prog, info); 8133 } 8134 8135 static bool __is_valid_xdp_access(int off, int size) 8136 { 8137 if (off < 0 || off >= sizeof(struct xdp_md)) 8138 return false; 8139 if (off % size != 0) 8140 return false; 8141 if (size != sizeof(__u32)) 8142 return false; 8143 8144 return true; 8145 } 8146 8147 static bool xdp_is_valid_access(int off, int size, 8148 enum bpf_access_type type, 8149 const struct bpf_prog *prog, 8150 struct bpf_insn_access_aux *info) 8151 { 8152 if (prog->expected_attach_type != BPF_XDP_DEVMAP) { 8153 switch (off) { 8154 case offsetof(struct xdp_md, egress_ifindex): 8155 return false; 8156 } 8157 } 8158 8159 if (type == BPF_WRITE) { 8160 if (bpf_prog_is_dev_bound(prog->aux)) { 8161 switch (off) { 8162 case offsetof(struct xdp_md, rx_queue_index): 8163 return __is_valid_xdp_access(off, size); 8164 } 8165 } 8166 return false; 8167 } 8168 8169 switch (off) { 8170 case offsetof(struct xdp_md, data): 8171 info->reg_type = PTR_TO_PACKET; 8172 break; 8173 case offsetof(struct xdp_md, data_meta): 8174 info->reg_type = PTR_TO_PACKET_META; 8175 break; 8176 case offsetof(struct xdp_md, data_end): 8177 info->reg_type = PTR_TO_PACKET_END; 8178 break; 8179 } 8180 8181 return __is_valid_xdp_access(off, size); 8182 } 8183 8184 void bpf_warn_invalid_xdp_action(u32 act) 8185 { 8186 const u32 act_max = XDP_REDIRECT; 8187 8188 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n", 8189 act > act_max ? "Illegal" : "Driver unsupported", 8190 act); 8191 } 8192 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); 8193 8194 static bool sock_addr_is_valid_access(int off, int size, 8195 enum bpf_access_type type, 8196 const struct bpf_prog *prog, 8197 struct bpf_insn_access_aux *info) 8198 { 8199 const int size_default = sizeof(__u32); 8200 8201 if (off < 0 || off >= sizeof(struct bpf_sock_addr)) 8202 return false; 8203 if (off % size != 0) 8204 return false; 8205 8206 /* Disallow access to IPv6 fields from IPv4 contex and vise 8207 * versa. 8208 */ 8209 switch (off) { 8210 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 8211 switch (prog->expected_attach_type) { 8212 case BPF_CGROUP_INET4_BIND: 8213 case BPF_CGROUP_INET4_CONNECT: 8214 case BPF_CGROUP_INET4_GETPEERNAME: 8215 case BPF_CGROUP_INET4_GETSOCKNAME: 8216 case BPF_CGROUP_UDP4_SENDMSG: 8217 case BPF_CGROUP_UDP4_RECVMSG: 8218 break; 8219 default: 8220 return false; 8221 } 8222 break; 8223 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8224 switch (prog->expected_attach_type) { 8225 case BPF_CGROUP_INET6_BIND: 8226 case BPF_CGROUP_INET6_CONNECT: 8227 case BPF_CGROUP_INET6_GETPEERNAME: 8228 case BPF_CGROUP_INET6_GETSOCKNAME: 8229 case BPF_CGROUP_UDP6_SENDMSG: 8230 case BPF_CGROUP_UDP6_RECVMSG: 8231 break; 8232 default: 8233 return false; 8234 } 8235 break; 8236 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8237 switch (prog->expected_attach_type) { 8238 case BPF_CGROUP_UDP4_SENDMSG: 8239 break; 8240 default: 8241 return false; 8242 } 8243 break; 8244 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8245 msg_src_ip6[3]): 8246 switch (prog->expected_attach_type) { 8247 case BPF_CGROUP_UDP6_SENDMSG: 8248 break; 8249 default: 8250 return false; 8251 } 8252 break; 8253 } 8254 8255 switch (off) { 8256 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 8257 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8258 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8259 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8260 msg_src_ip6[3]): 8261 case bpf_ctx_range(struct bpf_sock_addr, user_port): 8262 if (type == BPF_READ) { 8263 bpf_ctx_record_field_size(info, size_default); 8264 8265 if (bpf_ctx_wide_access_ok(off, size, 8266 struct bpf_sock_addr, 8267 user_ip6)) 8268 return true; 8269 8270 if (bpf_ctx_wide_access_ok(off, size, 8271 struct bpf_sock_addr, 8272 msg_src_ip6)) 8273 return true; 8274 8275 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8276 return false; 8277 } else { 8278 if (bpf_ctx_wide_access_ok(off, size, 8279 struct bpf_sock_addr, 8280 user_ip6)) 8281 return true; 8282 8283 if (bpf_ctx_wide_access_ok(off, size, 8284 struct bpf_sock_addr, 8285 msg_src_ip6)) 8286 return true; 8287 8288 if (size != size_default) 8289 return false; 8290 } 8291 break; 8292 case offsetof(struct bpf_sock_addr, sk): 8293 if (type != BPF_READ) 8294 return false; 8295 if (size != sizeof(__u64)) 8296 return false; 8297 info->reg_type = PTR_TO_SOCKET; 8298 break; 8299 default: 8300 if (type == BPF_READ) { 8301 if (size != size_default) 8302 return false; 8303 } else { 8304 return false; 8305 } 8306 } 8307 8308 return true; 8309 } 8310 8311 static bool sock_ops_is_valid_access(int off, int size, 8312 enum bpf_access_type type, 8313 const struct bpf_prog *prog, 8314 struct bpf_insn_access_aux *info) 8315 { 8316 const int size_default = sizeof(__u32); 8317 8318 if (off < 0 || off >= sizeof(struct bpf_sock_ops)) 8319 return false; 8320 8321 /* The verifier guarantees that size > 0. */ 8322 if (off % size != 0) 8323 return false; 8324 8325 if (type == BPF_WRITE) { 8326 switch (off) { 8327 case offsetof(struct bpf_sock_ops, reply): 8328 case offsetof(struct bpf_sock_ops, sk_txhash): 8329 if (size != size_default) 8330 return false; 8331 break; 8332 default: 8333 return false; 8334 } 8335 } else { 8336 switch (off) { 8337 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received, 8338 bytes_acked): 8339 if (size != sizeof(__u64)) 8340 return false; 8341 break; 8342 case offsetof(struct bpf_sock_ops, sk): 8343 if (size != sizeof(__u64)) 8344 return false; 8345 info->reg_type = PTR_TO_SOCKET_OR_NULL; 8346 break; 8347 case offsetof(struct bpf_sock_ops, skb_data): 8348 if (size != sizeof(__u64)) 8349 return false; 8350 info->reg_type = PTR_TO_PACKET; 8351 break; 8352 case offsetof(struct bpf_sock_ops, skb_data_end): 8353 if (size != sizeof(__u64)) 8354 return false; 8355 info->reg_type = PTR_TO_PACKET_END; 8356 break; 8357 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 8358 bpf_ctx_record_field_size(info, size_default); 8359 return bpf_ctx_narrow_access_ok(off, size, 8360 size_default); 8361 default: 8362 if (size != size_default) 8363 return false; 8364 break; 8365 } 8366 } 8367 8368 return true; 8369 } 8370 8371 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write, 8372 const struct bpf_prog *prog) 8373 { 8374 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP); 8375 } 8376 8377 static bool sk_skb_is_valid_access(int off, int size, 8378 enum bpf_access_type type, 8379 const struct bpf_prog *prog, 8380 struct bpf_insn_access_aux *info) 8381 { 8382 switch (off) { 8383 case bpf_ctx_range(struct __sk_buff, tc_classid): 8384 case bpf_ctx_range(struct __sk_buff, data_meta): 8385 case bpf_ctx_range(struct __sk_buff, tstamp): 8386 case bpf_ctx_range(struct __sk_buff, wire_len): 8387 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8388 return false; 8389 } 8390 8391 if (type == BPF_WRITE) { 8392 switch (off) { 8393 case bpf_ctx_range(struct __sk_buff, tc_index): 8394 case bpf_ctx_range(struct __sk_buff, priority): 8395 break; 8396 default: 8397 return false; 8398 } 8399 } 8400 8401 switch (off) { 8402 case bpf_ctx_range(struct __sk_buff, mark): 8403 return false; 8404 case bpf_ctx_range(struct __sk_buff, data): 8405 info->reg_type = PTR_TO_PACKET; 8406 break; 8407 case bpf_ctx_range(struct __sk_buff, data_end): 8408 info->reg_type = PTR_TO_PACKET_END; 8409 break; 8410 } 8411 8412 return bpf_skb_is_valid_access(off, size, type, prog, info); 8413 } 8414 8415 static bool sk_msg_is_valid_access(int off, int size, 8416 enum bpf_access_type type, 8417 const struct bpf_prog *prog, 8418 struct bpf_insn_access_aux *info) 8419 { 8420 if (type == BPF_WRITE) 8421 return false; 8422 8423 if (off % size != 0) 8424 return false; 8425 8426 switch (off) { 8427 case offsetof(struct sk_msg_md, data): 8428 info->reg_type = PTR_TO_PACKET; 8429 if (size != sizeof(__u64)) 8430 return false; 8431 break; 8432 case offsetof(struct sk_msg_md, data_end): 8433 info->reg_type = PTR_TO_PACKET_END; 8434 if (size != sizeof(__u64)) 8435 return false; 8436 break; 8437 case offsetof(struct sk_msg_md, sk): 8438 if (size != sizeof(__u64)) 8439 return false; 8440 info->reg_type = PTR_TO_SOCKET; 8441 break; 8442 case bpf_ctx_range(struct sk_msg_md, family): 8443 case bpf_ctx_range(struct sk_msg_md, remote_ip4): 8444 case bpf_ctx_range(struct sk_msg_md, local_ip4): 8445 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]): 8446 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]): 8447 case bpf_ctx_range(struct sk_msg_md, remote_port): 8448 case bpf_ctx_range(struct sk_msg_md, local_port): 8449 case bpf_ctx_range(struct sk_msg_md, size): 8450 if (size != sizeof(__u32)) 8451 return false; 8452 break; 8453 default: 8454 return false; 8455 } 8456 return true; 8457 } 8458 8459 static bool flow_dissector_is_valid_access(int off, int size, 8460 enum bpf_access_type type, 8461 const struct bpf_prog *prog, 8462 struct bpf_insn_access_aux *info) 8463 { 8464 const int size_default = sizeof(__u32); 8465 8466 if (off < 0 || off >= sizeof(struct __sk_buff)) 8467 return false; 8468 8469 if (type == BPF_WRITE) 8470 return false; 8471 8472 switch (off) { 8473 case bpf_ctx_range(struct __sk_buff, data): 8474 if (size != size_default) 8475 return false; 8476 info->reg_type = PTR_TO_PACKET; 8477 return true; 8478 case bpf_ctx_range(struct __sk_buff, data_end): 8479 if (size != size_default) 8480 return false; 8481 info->reg_type = PTR_TO_PACKET_END; 8482 return true; 8483 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 8484 if (size != sizeof(__u64)) 8485 return false; 8486 info->reg_type = PTR_TO_FLOW_KEYS; 8487 return true; 8488 default: 8489 return false; 8490 } 8491 } 8492 8493 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type, 8494 const struct bpf_insn *si, 8495 struct bpf_insn *insn_buf, 8496 struct bpf_prog *prog, 8497 u32 *target_size) 8498 8499 { 8500 struct bpf_insn *insn = insn_buf; 8501 8502 switch (si->off) { 8503 case offsetof(struct __sk_buff, data): 8504 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data), 8505 si->dst_reg, si->src_reg, 8506 offsetof(struct bpf_flow_dissector, data)); 8507 break; 8508 8509 case offsetof(struct __sk_buff, data_end): 8510 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end), 8511 si->dst_reg, si->src_reg, 8512 offsetof(struct bpf_flow_dissector, data_end)); 8513 break; 8514 8515 case offsetof(struct __sk_buff, flow_keys): 8516 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys), 8517 si->dst_reg, si->src_reg, 8518 offsetof(struct bpf_flow_dissector, flow_keys)); 8519 break; 8520 } 8521 8522 return insn - insn_buf; 8523 } 8524 8525 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si, 8526 struct bpf_insn *insn) 8527 { 8528 /* si->dst_reg = skb_shinfo(SKB); */ 8529 #ifdef NET_SKBUFF_DATA_USES_OFFSET 8530 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 8531 BPF_REG_AX, si->src_reg, 8532 offsetof(struct sk_buff, end)); 8533 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head), 8534 si->dst_reg, si->src_reg, 8535 offsetof(struct sk_buff, head)); 8536 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX); 8537 #else 8538 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 8539 si->dst_reg, si->src_reg, 8540 offsetof(struct sk_buff, end)); 8541 #endif 8542 8543 return insn; 8544 } 8545 8546 static u32 bpf_convert_ctx_access(enum bpf_access_type type, 8547 const struct bpf_insn *si, 8548 struct bpf_insn *insn_buf, 8549 struct bpf_prog *prog, u32 *target_size) 8550 { 8551 struct bpf_insn *insn = insn_buf; 8552 int off; 8553 8554 switch (si->off) { 8555 case offsetof(struct __sk_buff, len): 8556 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8557 bpf_target_off(struct sk_buff, len, 4, 8558 target_size)); 8559 break; 8560 8561 case offsetof(struct __sk_buff, protocol): 8562 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8563 bpf_target_off(struct sk_buff, protocol, 2, 8564 target_size)); 8565 break; 8566 8567 case offsetof(struct __sk_buff, vlan_proto): 8568 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8569 bpf_target_off(struct sk_buff, vlan_proto, 2, 8570 target_size)); 8571 break; 8572 8573 case offsetof(struct __sk_buff, priority): 8574 if (type == BPF_WRITE) 8575 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8576 bpf_target_off(struct sk_buff, priority, 4, 8577 target_size)); 8578 else 8579 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8580 bpf_target_off(struct sk_buff, priority, 4, 8581 target_size)); 8582 break; 8583 8584 case offsetof(struct __sk_buff, ingress_ifindex): 8585 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8586 bpf_target_off(struct sk_buff, skb_iif, 4, 8587 target_size)); 8588 break; 8589 8590 case offsetof(struct __sk_buff, ifindex): 8591 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 8592 si->dst_reg, si->src_reg, 8593 offsetof(struct sk_buff, dev)); 8594 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 8595 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8596 bpf_target_off(struct net_device, ifindex, 4, 8597 target_size)); 8598 break; 8599 8600 case offsetof(struct __sk_buff, hash): 8601 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8602 bpf_target_off(struct sk_buff, hash, 4, 8603 target_size)); 8604 break; 8605 8606 case offsetof(struct __sk_buff, mark): 8607 if (type == BPF_WRITE) 8608 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8609 bpf_target_off(struct sk_buff, mark, 4, 8610 target_size)); 8611 else 8612 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8613 bpf_target_off(struct sk_buff, mark, 4, 8614 target_size)); 8615 break; 8616 8617 case offsetof(struct __sk_buff, pkt_type): 8618 *target_size = 1; 8619 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 8620 PKT_TYPE_OFFSET()); 8621 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX); 8622 #ifdef __BIG_ENDIAN_BITFIELD 8623 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5); 8624 #endif 8625 break; 8626 8627 case offsetof(struct __sk_buff, queue_mapping): 8628 if (type == BPF_WRITE) { 8629 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1); 8630 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 8631 bpf_target_off(struct sk_buff, 8632 queue_mapping, 8633 2, target_size)); 8634 } else { 8635 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8636 bpf_target_off(struct sk_buff, 8637 queue_mapping, 8638 2, target_size)); 8639 } 8640 break; 8641 8642 case offsetof(struct __sk_buff, vlan_present): 8643 *target_size = 1; 8644 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 8645 PKT_VLAN_PRESENT_OFFSET()); 8646 if (PKT_VLAN_PRESENT_BIT) 8647 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT); 8648 if (PKT_VLAN_PRESENT_BIT < 7) 8649 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1); 8650 break; 8651 8652 case offsetof(struct __sk_buff, vlan_tci): 8653 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8654 bpf_target_off(struct sk_buff, vlan_tci, 2, 8655 target_size)); 8656 break; 8657 8658 case offsetof(struct __sk_buff, cb[0]) ... 8659 offsetofend(struct __sk_buff, cb[4]) - 1: 8660 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20); 8661 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 8662 offsetof(struct qdisc_skb_cb, data)) % 8663 sizeof(__u64)); 8664 8665 prog->cb_access = 1; 8666 off = si->off; 8667 off -= offsetof(struct __sk_buff, cb[0]); 8668 off += offsetof(struct sk_buff, cb); 8669 off += offsetof(struct qdisc_skb_cb, data); 8670 if (type == BPF_WRITE) 8671 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 8672 si->src_reg, off); 8673 else 8674 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 8675 si->src_reg, off); 8676 break; 8677 8678 case offsetof(struct __sk_buff, tc_classid): 8679 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2); 8680 8681 off = si->off; 8682 off -= offsetof(struct __sk_buff, tc_classid); 8683 off += offsetof(struct sk_buff, cb); 8684 off += offsetof(struct qdisc_skb_cb, tc_classid); 8685 *target_size = 2; 8686 if (type == BPF_WRITE) 8687 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, 8688 si->src_reg, off); 8689 else 8690 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, 8691 si->src_reg, off); 8692 break; 8693 8694 case offsetof(struct __sk_buff, data): 8695 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 8696 si->dst_reg, si->src_reg, 8697 offsetof(struct sk_buff, data)); 8698 break; 8699 8700 case offsetof(struct __sk_buff, data_meta): 8701 off = si->off; 8702 off -= offsetof(struct __sk_buff, data_meta); 8703 off += offsetof(struct sk_buff, cb); 8704 off += offsetof(struct bpf_skb_data_end, data_meta); 8705 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 8706 si->src_reg, off); 8707 break; 8708 8709 case offsetof(struct __sk_buff, data_end): 8710 off = si->off; 8711 off -= offsetof(struct __sk_buff, data_end); 8712 off += offsetof(struct sk_buff, cb); 8713 off += offsetof(struct bpf_skb_data_end, data_end); 8714 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 8715 si->src_reg, off); 8716 break; 8717 8718 case offsetof(struct __sk_buff, tc_index): 8719 #ifdef CONFIG_NET_SCHED 8720 if (type == BPF_WRITE) 8721 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 8722 bpf_target_off(struct sk_buff, tc_index, 2, 8723 target_size)); 8724 else 8725 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 8726 bpf_target_off(struct sk_buff, tc_index, 2, 8727 target_size)); 8728 #else 8729 *target_size = 2; 8730 if (type == BPF_WRITE) 8731 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg); 8732 else 8733 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 8734 #endif 8735 break; 8736 8737 case offsetof(struct __sk_buff, napi_id): 8738 #if defined(CONFIG_NET_RX_BUSY_POLL) 8739 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8740 bpf_target_off(struct sk_buff, napi_id, 4, 8741 target_size)); 8742 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1); 8743 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 8744 #else 8745 *target_size = 4; 8746 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 8747 #endif 8748 break; 8749 case offsetof(struct __sk_buff, family): 8750 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 8751 8752 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8753 si->dst_reg, si->src_reg, 8754 offsetof(struct sk_buff, sk)); 8755 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8756 bpf_target_off(struct sock_common, 8757 skc_family, 8758 2, target_size)); 8759 break; 8760 case offsetof(struct __sk_buff, remote_ip4): 8761 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 8762 8763 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8764 si->dst_reg, si->src_reg, 8765 offsetof(struct sk_buff, sk)); 8766 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8767 bpf_target_off(struct sock_common, 8768 skc_daddr, 8769 4, target_size)); 8770 break; 8771 case offsetof(struct __sk_buff, local_ip4): 8772 BUILD_BUG_ON(sizeof_field(struct sock_common, 8773 skc_rcv_saddr) != 4); 8774 8775 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8776 si->dst_reg, si->src_reg, 8777 offsetof(struct sk_buff, sk)); 8778 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8779 bpf_target_off(struct sock_common, 8780 skc_rcv_saddr, 8781 4, target_size)); 8782 break; 8783 case offsetof(struct __sk_buff, remote_ip6[0]) ... 8784 offsetof(struct __sk_buff, remote_ip6[3]): 8785 #if IS_ENABLED(CONFIG_IPV6) 8786 BUILD_BUG_ON(sizeof_field(struct sock_common, 8787 skc_v6_daddr.s6_addr32[0]) != 4); 8788 8789 off = si->off; 8790 off -= offsetof(struct __sk_buff, remote_ip6[0]); 8791 8792 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8793 si->dst_reg, si->src_reg, 8794 offsetof(struct sk_buff, sk)); 8795 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8796 offsetof(struct sock_common, 8797 skc_v6_daddr.s6_addr32[0]) + 8798 off); 8799 #else 8800 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8801 #endif 8802 break; 8803 case offsetof(struct __sk_buff, local_ip6[0]) ... 8804 offsetof(struct __sk_buff, local_ip6[3]): 8805 #if IS_ENABLED(CONFIG_IPV6) 8806 BUILD_BUG_ON(sizeof_field(struct sock_common, 8807 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 8808 8809 off = si->off; 8810 off -= offsetof(struct __sk_buff, local_ip6[0]); 8811 8812 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8813 si->dst_reg, si->src_reg, 8814 offsetof(struct sk_buff, sk)); 8815 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8816 offsetof(struct sock_common, 8817 skc_v6_rcv_saddr.s6_addr32[0]) + 8818 off); 8819 #else 8820 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8821 #endif 8822 break; 8823 8824 case offsetof(struct __sk_buff, remote_port): 8825 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 8826 8827 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8828 si->dst_reg, si->src_reg, 8829 offsetof(struct sk_buff, sk)); 8830 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8831 bpf_target_off(struct sock_common, 8832 skc_dport, 8833 2, target_size)); 8834 #ifndef __BIG_ENDIAN_BITFIELD 8835 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 8836 #endif 8837 break; 8838 8839 case offsetof(struct __sk_buff, local_port): 8840 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 8841 8842 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8843 si->dst_reg, si->src_reg, 8844 offsetof(struct sk_buff, sk)); 8845 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8846 bpf_target_off(struct sock_common, 8847 skc_num, 2, target_size)); 8848 break; 8849 8850 case offsetof(struct __sk_buff, tstamp): 8851 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8); 8852 8853 if (type == BPF_WRITE) 8854 *insn++ = BPF_STX_MEM(BPF_DW, 8855 si->dst_reg, si->src_reg, 8856 bpf_target_off(struct sk_buff, 8857 tstamp, 8, 8858 target_size)); 8859 else 8860 *insn++ = BPF_LDX_MEM(BPF_DW, 8861 si->dst_reg, si->src_reg, 8862 bpf_target_off(struct sk_buff, 8863 tstamp, 8, 8864 target_size)); 8865 break; 8866 8867 case offsetof(struct __sk_buff, gso_segs): 8868 insn = bpf_convert_shinfo_access(si, insn); 8869 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs), 8870 si->dst_reg, si->dst_reg, 8871 bpf_target_off(struct skb_shared_info, 8872 gso_segs, 2, 8873 target_size)); 8874 break; 8875 case offsetof(struct __sk_buff, gso_size): 8876 insn = bpf_convert_shinfo_access(si, insn); 8877 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size), 8878 si->dst_reg, si->dst_reg, 8879 bpf_target_off(struct skb_shared_info, 8880 gso_size, 2, 8881 target_size)); 8882 break; 8883 case offsetof(struct __sk_buff, wire_len): 8884 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4); 8885 8886 off = si->off; 8887 off -= offsetof(struct __sk_buff, wire_len); 8888 off += offsetof(struct sk_buff, cb); 8889 off += offsetof(struct qdisc_skb_cb, pkt_len); 8890 *target_size = 4; 8891 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off); 8892 break; 8893 8894 case offsetof(struct __sk_buff, sk): 8895 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 8896 si->dst_reg, si->src_reg, 8897 offsetof(struct sk_buff, sk)); 8898 break; 8899 case offsetof(struct __sk_buff, hwtstamp): 8900 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8); 8901 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0); 8902 8903 insn = bpf_convert_shinfo_access(si, insn); 8904 *insn++ = BPF_LDX_MEM(BPF_DW, 8905 si->dst_reg, si->dst_reg, 8906 bpf_target_off(struct skb_shared_info, 8907 hwtstamps, 8, 8908 target_size)); 8909 break; 8910 } 8911 8912 return insn - insn_buf; 8913 } 8914 8915 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 8916 const struct bpf_insn *si, 8917 struct bpf_insn *insn_buf, 8918 struct bpf_prog *prog, u32 *target_size) 8919 { 8920 struct bpf_insn *insn = insn_buf; 8921 int off; 8922 8923 switch (si->off) { 8924 case offsetof(struct bpf_sock, bound_dev_if): 8925 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4); 8926 8927 if (type == BPF_WRITE) 8928 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8929 offsetof(struct sock, sk_bound_dev_if)); 8930 else 8931 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8932 offsetof(struct sock, sk_bound_dev_if)); 8933 break; 8934 8935 case offsetof(struct bpf_sock, mark): 8936 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4); 8937 8938 if (type == BPF_WRITE) 8939 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8940 offsetof(struct sock, sk_mark)); 8941 else 8942 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8943 offsetof(struct sock, sk_mark)); 8944 break; 8945 8946 case offsetof(struct bpf_sock, priority): 8947 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4); 8948 8949 if (type == BPF_WRITE) 8950 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8951 offsetof(struct sock, sk_priority)); 8952 else 8953 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8954 offsetof(struct sock, sk_priority)); 8955 break; 8956 8957 case offsetof(struct bpf_sock, family): 8958 *insn++ = BPF_LDX_MEM( 8959 BPF_FIELD_SIZEOF(struct sock_common, skc_family), 8960 si->dst_reg, si->src_reg, 8961 bpf_target_off(struct sock_common, 8962 skc_family, 8963 sizeof_field(struct sock_common, 8964 skc_family), 8965 target_size)); 8966 break; 8967 8968 case offsetof(struct bpf_sock, type): 8969 *insn++ = BPF_LDX_MEM( 8970 BPF_FIELD_SIZEOF(struct sock, sk_type), 8971 si->dst_reg, si->src_reg, 8972 bpf_target_off(struct sock, sk_type, 8973 sizeof_field(struct sock, sk_type), 8974 target_size)); 8975 break; 8976 8977 case offsetof(struct bpf_sock, protocol): 8978 *insn++ = BPF_LDX_MEM( 8979 BPF_FIELD_SIZEOF(struct sock, sk_protocol), 8980 si->dst_reg, si->src_reg, 8981 bpf_target_off(struct sock, sk_protocol, 8982 sizeof_field(struct sock, sk_protocol), 8983 target_size)); 8984 break; 8985 8986 case offsetof(struct bpf_sock, src_ip4): 8987 *insn++ = BPF_LDX_MEM( 8988 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 8989 bpf_target_off(struct sock_common, skc_rcv_saddr, 8990 sizeof_field(struct sock_common, 8991 skc_rcv_saddr), 8992 target_size)); 8993 break; 8994 8995 case offsetof(struct bpf_sock, dst_ip4): 8996 *insn++ = BPF_LDX_MEM( 8997 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 8998 bpf_target_off(struct sock_common, skc_daddr, 8999 sizeof_field(struct sock_common, 9000 skc_daddr), 9001 target_size)); 9002 break; 9003 9004 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 9005 #if IS_ENABLED(CONFIG_IPV6) 9006 off = si->off; 9007 off -= offsetof(struct bpf_sock, src_ip6[0]); 9008 *insn++ = BPF_LDX_MEM( 9009 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9010 bpf_target_off( 9011 struct sock_common, 9012 skc_v6_rcv_saddr.s6_addr32[0], 9013 sizeof_field(struct sock_common, 9014 skc_v6_rcv_saddr.s6_addr32[0]), 9015 target_size) + off); 9016 #else 9017 (void)off; 9018 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9019 #endif 9020 break; 9021 9022 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 9023 #if IS_ENABLED(CONFIG_IPV6) 9024 off = si->off; 9025 off -= offsetof(struct bpf_sock, dst_ip6[0]); 9026 *insn++ = BPF_LDX_MEM( 9027 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9028 bpf_target_off(struct sock_common, 9029 skc_v6_daddr.s6_addr32[0], 9030 sizeof_field(struct sock_common, 9031 skc_v6_daddr.s6_addr32[0]), 9032 target_size) + off); 9033 #else 9034 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9035 *target_size = 4; 9036 #endif 9037 break; 9038 9039 case offsetof(struct bpf_sock, src_port): 9040 *insn++ = BPF_LDX_MEM( 9041 BPF_FIELD_SIZEOF(struct sock_common, skc_num), 9042 si->dst_reg, si->src_reg, 9043 bpf_target_off(struct sock_common, skc_num, 9044 sizeof_field(struct sock_common, 9045 skc_num), 9046 target_size)); 9047 break; 9048 9049 case offsetof(struct bpf_sock, dst_port): 9050 *insn++ = BPF_LDX_MEM( 9051 BPF_FIELD_SIZEOF(struct sock_common, skc_dport), 9052 si->dst_reg, si->src_reg, 9053 bpf_target_off(struct sock_common, skc_dport, 9054 sizeof_field(struct sock_common, 9055 skc_dport), 9056 target_size)); 9057 break; 9058 9059 case offsetof(struct bpf_sock, state): 9060 *insn++ = BPF_LDX_MEM( 9061 BPF_FIELD_SIZEOF(struct sock_common, skc_state), 9062 si->dst_reg, si->src_reg, 9063 bpf_target_off(struct sock_common, skc_state, 9064 sizeof_field(struct sock_common, 9065 skc_state), 9066 target_size)); 9067 break; 9068 case offsetof(struct bpf_sock, rx_queue_mapping): 9069 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 9070 *insn++ = BPF_LDX_MEM( 9071 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping), 9072 si->dst_reg, si->src_reg, 9073 bpf_target_off(struct sock, sk_rx_queue_mapping, 9074 sizeof_field(struct sock, 9075 sk_rx_queue_mapping), 9076 target_size)); 9077 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING, 9078 1); 9079 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 9080 #else 9081 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 9082 *target_size = 2; 9083 #endif 9084 break; 9085 } 9086 9087 return insn - insn_buf; 9088 } 9089 9090 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, 9091 const struct bpf_insn *si, 9092 struct bpf_insn *insn_buf, 9093 struct bpf_prog *prog, u32 *target_size) 9094 { 9095 struct bpf_insn *insn = insn_buf; 9096 9097 switch (si->off) { 9098 case offsetof(struct __sk_buff, ifindex): 9099 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 9100 si->dst_reg, si->src_reg, 9101 offsetof(struct sk_buff, dev)); 9102 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9103 bpf_target_off(struct net_device, ifindex, 4, 9104 target_size)); 9105 break; 9106 default: 9107 return bpf_convert_ctx_access(type, si, insn_buf, prog, 9108 target_size); 9109 } 9110 9111 return insn - insn_buf; 9112 } 9113 9114 static u32 xdp_convert_ctx_access(enum bpf_access_type type, 9115 const struct bpf_insn *si, 9116 struct bpf_insn *insn_buf, 9117 struct bpf_prog *prog, u32 *target_size) 9118 { 9119 struct bpf_insn *insn = insn_buf; 9120 9121 switch (si->off) { 9122 case offsetof(struct xdp_md, data): 9123 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data), 9124 si->dst_reg, si->src_reg, 9125 offsetof(struct xdp_buff, data)); 9126 break; 9127 case offsetof(struct xdp_md, data_meta): 9128 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta), 9129 si->dst_reg, si->src_reg, 9130 offsetof(struct xdp_buff, data_meta)); 9131 break; 9132 case offsetof(struct xdp_md, data_end): 9133 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end), 9134 si->dst_reg, si->src_reg, 9135 offsetof(struct xdp_buff, data_end)); 9136 break; 9137 case offsetof(struct xdp_md, ingress_ifindex): 9138 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 9139 si->dst_reg, si->src_reg, 9140 offsetof(struct xdp_buff, rxq)); 9141 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev), 9142 si->dst_reg, si->dst_reg, 9143 offsetof(struct xdp_rxq_info, dev)); 9144 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9145 offsetof(struct net_device, ifindex)); 9146 break; 9147 case offsetof(struct xdp_md, rx_queue_index): 9148 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 9149 si->dst_reg, si->src_reg, 9150 offsetof(struct xdp_buff, rxq)); 9151 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9152 offsetof(struct xdp_rxq_info, 9153 queue_index)); 9154 break; 9155 case offsetof(struct xdp_md, egress_ifindex): 9156 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq), 9157 si->dst_reg, si->src_reg, 9158 offsetof(struct xdp_buff, txq)); 9159 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev), 9160 si->dst_reg, si->dst_reg, 9161 offsetof(struct xdp_txq_info, dev)); 9162 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9163 offsetof(struct net_device, ifindex)); 9164 break; 9165 } 9166 9167 return insn - insn_buf; 9168 } 9169 9170 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of 9171 * context Structure, F is Field in context structure that contains a pointer 9172 * to Nested Structure of type NS that has the field NF. 9173 * 9174 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make 9175 * sure that SIZE is not greater than actual size of S.F.NF. 9176 * 9177 * If offset OFF is provided, the load happens from that offset relative to 9178 * offset of NF. 9179 */ 9180 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \ 9181 do { \ 9182 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \ 9183 si->src_reg, offsetof(S, F)); \ 9184 *insn++ = BPF_LDX_MEM( \ 9185 SIZE, si->dst_reg, si->dst_reg, \ 9186 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 9187 target_size) \ 9188 + OFF); \ 9189 } while (0) 9190 9191 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \ 9192 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \ 9193 BPF_FIELD_SIZEOF(NS, NF), 0) 9194 9195 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to 9196 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation. 9197 * 9198 * In addition it uses Temporary Field TF (member of struct S) as the 3rd 9199 * "register" since two registers available in convert_ctx_access are not 9200 * enough: we can't override neither SRC, since it contains value to store, nor 9201 * DST since it contains pointer to context that may be used by later 9202 * instructions. But we need a temporary place to save pointer to nested 9203 * structure whose field we want to store to. 9204 */ 9205 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \ 9206 do { \ 9207 int tmp_reg = BPF_REG_9; \ 9208 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 9209 --tmp_reg; \ 9210 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 9211 --tmp_reg; \ 9212 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \ 9213 offsetof(S, TF)); \ 9214 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \ 9215 si->dst_reg, offsetof(S, F)); \ 9216 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \ 9217 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 9218 target_size) \ 9219 + OFF); \ 9220 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \ 9221 offsetof(S, TF)); \ 9222 } while (0) 9223 9224 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \ 9225 TF) \ 9226 do { \ 9227 if (type == BPF_WRITE) { \ 9228 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \ 9229 OFF, TF); \ 9230 } else { \ 9231 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \ 9232 S, NS, F, NF, SIZE, OFF); \ 9233 } \ 9234 } while (0) 9235 9236 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \ 9237 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \ 9238 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF) 9239 9240 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type, 9241 const struct bpf_insn *si, 9242 struct bpf_insn *insn_buf, 9243 struct bpf_prog *prog, u32 *target_size) 9244 { 9245 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port); 9246 struct bpf_insn *insn = insn_buf; 9247 9248 switch (si->off) { 9249 case offsetof(struct bpf_sock_addr, user_family): 9250 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9251 struct sockaddr, uaddr, sa_family); 9252 break; 9253 9254 case offsetof(struct bpf_sock_addr, user_ip4): 9255 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9256 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr, 9257 sin_addr, BPF_SIZE(si->code), 0, tmp_reg); 9258 break; 9259 9260 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 9261 off = si->off; 9262 off -= offsetof(struct bpf_sock_addr, user_ip6[0]); 9263 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9264 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 9265 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off, 9266 tmp_reg); 9267 break; 9268 9269 case offsetof(struct bpf_sock_addr, user_port): 9270 /* To get port we need to know sa_family first and then treat 9271 * sockaddr as either sockaddr_in or sockaddr_in6. 9272 * Though we can simplify since port field has same offset and 9273 * size in both structures. 9274 * Here we check this invariant and use just one of the 9275 * structures if it's true. 9276 */ 9277 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) != 9278 offsetof(struct sockaddr_in6, sin6_port)); 9279 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) != 9280 sizeof_field(struct sockaddr_in6, sin6_port)); 9281 /* Account for sin6_port being smaller than user_port. */ 9282 port_size = min(port_size, BPF_LDST_BYTES(si)); 9283 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9284 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 9285 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg); 9286 break; 9287 9288 case offsetof(struct bpf_sock_addr, family): 9289 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9290 struct sock, sk, sk_family); 9291 break; 9292 9293 case offsetof(struct bpf_sock_addr, type): 9294 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9295 struct sock, sk, sk_type); 9296 break; 9297 9298 case offsetof(struct bpf_sock_addr, protocol): 9299 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9300 struct sock, sk, sk_protocol); 9301 break; 9302 9303 case offsetof(struct bpf_sock_addr, msg_src_ip4): 9304 /* Treat t_ctx as struct in_addr for msg_src_ip4. */ 9305 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9306 struct bpf_sock_addr_kern, struct in_addr, t_ctx, 9307 s_addr, BPF_SIZE(si->code), 0, tmp_reg); 9308 break; 9309 9310 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 9311 msg_src_ip6[3]): 9312 off = si->off; 9313 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]); 9314 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */ 9315 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9316 struct bpf_sock_addr_kern, struct in6_addr, t_ctx, 9317 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg); 9318 break; 9319 case offsetof(struct bpf_sock_addr, sk): 9320 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk), 9321 si->dst_reg, si->src_reg, 9322 offsetof(struct bpf_sock_addr_kern, sk)); 9323 break; 9324 } 9325 9326 return insn - insn_buf; 9327 } 9328 9329 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type, 9330 const struct bpf_insn *si, 9331 struct bpf_insn *insn_buf, 9332 struct bpf_prog *prog, 9333 u32 *target_size) 9334 { 9335 struct bpf_insn *insn = insn_buf; 9336 int off; 9337 9338 /* Helper macro for adding read access to tcp_sock or sock fields. */ 9339 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 9340 do { \ 9341 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \ 9342 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 9343 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 9344 if (si->dst_reg == reg || si->src_reg == reg) \ 9345 reg--; \ 9346 if (si->dst_reg == reg || si->src_reg == reg) \ 9347 reg--; \ 9348 if (si->dst_reg == si->src_reg) { \ 9349 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 9350 offsetof(struct bpf_sock_ops_kern, \ 9351 temp)); \ 9352 fullsock_reg = reg; \ 9353 jmp += 2; \ 9354 } \ 9355 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9356 struct bpf_sock_ops_kern, \ 9357 is_fullsock), \ 9358 fullsock_reg, si->src_reg, \ 9359 offsetof(struct bpf_sock_ops_kern, \ 9360 is_fullsock)); \ 9361 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 9362 if (si->dst_reg == si->src_reg) \ 9363 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9364 offsetof(struct bpf_sock_ops_kern, \ 9365 temp)); \ 9366 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9367 struct bpf_sock_ops_kern, sk),\ 9368 si->dst_reg, si->src_reg, \ 9369 offsetof(struct bpf_sock_ops_kern, sk));\ 9370 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \ 9371 OBJ_FIELD), \ 9372 si->dst_reg, si->dst_reg, \ 9373 offsetof(OBJ, OBJ_FIELD)); \ 9374 if (si->dst_reg == si->src_reg) { \ 9375 *insn++ = BPF_JMP_A(1); \ 9376 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9377 offsetof(struct bpf_sock_ops_kern, \ 9378 temp)); \ 9379 } \ 9380 } while (0) 9381 9382 #define SOCK_OPS_GET_SK() \ 9383 do { \ 9384 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \ 9385 if (si->dst_reg == reg || si->src_reg == reg) \ 9386 reg--; \ 9387 if (si->dst_reg == reg || si->src_reg == reg) \ 9388 reg--; \ 9389 if (si->dst_reg == si->src_reg) { \ 9390 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 9391 offsetof(struct bpf_sock_ops_kern, \ 9392 temp)); \ 9393 fullsock_reg = reg; \ 9394 jmp += 2; \ 9395 } \ 9396 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9397 struct bpf_sock_ops_kern, \ 9398 is_fullsock), \ 9399 fullsock_reg, si->src_reg, \ 9400 offsetof(struct bpf_sock_ops_kern, \ 9401 is_fullsock)); \ 9402 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 9403 if (si->dst_reg == si->src_reg) \ 9404 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9405 offsetof(struct bpf_sock_ops_kern, \ 9406 temp)); \ 9407 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9408 struct bpf_sock_ops_kern, sk),\ 9409 si->dst_reg, si->src_reg, \ 9410 offsetof(struct bpf_sock_ops_kern, sk));\ 9411 if (si->dst_reg == si->src_reg) { \ 9412 *insn++ = BPF_JMP_A(1); \ 9413 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9414 offsetof(struct bpf_sock_ops_kern, \ 9415 temp)); \ 9416 } \ 9417 } while (0) 9418 9419 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \ 9420 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock) 9421 9422 /* Helper macro for adding write access to tcp_sock or sock fields. 9423 * The macro is called with two registers, dst_reg which contains a pointer 9424 * to ctx (context) and src_reg which contains the value that should be 9425 * stored. However, we need an additional register since we cannot overwrite 9426 * dst_reg because it may be used later in the program. 9427 * Instead we "borrow" one of the other register. We first save its value 9428 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore 9429 * it at the end of the macro. 9430 */ 9431 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 9432 do { \ 9433 int reg = BPF_REG_9; \ 9434 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 9435 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 9436 if (si->dst_reg == reg || si->src_reg == reg) \ 9437 reg--; \ 9438 if (si->dst_reg == reg || si->src_reg == reg) \ 9439 reg--; \ 9440 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \ 9441 offsetof(struct bpf_sock_ops_kern, \ 9442 temp)); \ 9443 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9444 struct bpf_sock_ops_kern, \ 9445 is_fullsock), \ 9446 reg, si->dst_reg, \ 9447 offsetof(struct bpf_sock_ops_kern, \ 9448 is_fullsock)); \ 9449 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \ 9450 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9451 struct bpf_sock_ops_kern, sk),\ 9452 reg, si->dst_reg, \ 9453 offsetof(struct bpf_sock_ops_kern, sk));\ 9454 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \ 9455 reg, si->src_reg, \ 9456 offsetof(OBJ, OBJ_FIELD)); \ 9457 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \ 9458 offsetof(struct bpf_sock_ops_kern, \ 9459 temp)); \ 9460 } while (0) 9461 9462 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \ 9463 do { \ 9464 if (TYPE == BPF_WRITE) \ 9465 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 9466 else \ 9467 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 9468 } while (0) 9469 9470 if (insn > insn_buf) 9471 return insn - insn_buf; 9472 9473 switch (si->off) { 9474 case offsetof(struct bpf_sock_ops, op): 9475 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9476 op), 9477 si->dst_reg, si->src_reg, 9478 offsetof(struct bpf_sock_ops_kern, op)); 9479 break; 9480 9481 case offsetof(struct bpf_sock_ops, replylong[0]) ... 9482 offsetof(struct bpf_sock_ops, replylong[3]): 9483 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) != 9484 sizeof_field(struct bpf_sock_ops_kern, reply)); 9485 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) != 9486 sizeof_field(struct bpf_sock_ops_kern, replylong)); 9487 off = si->off; 9488 off -= offsetof(struct bpf_sock_ops, replylong[0]); 9489 off += offsetof(struct bpf_sock_ops_kern, replylong[0]); 9490 if (type == BPF_WRITE) 9491 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9492 off); 9493 else 9494 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9495 off); 9496 break; 9497 9498 case offsetof(struct bpf_sock_ops, family): 9499 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 9500 9501 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9502 struct bpf_sock_ops_kern, sk), 9503 si->dst_reg, si->src_reg, 9504 offsetof(struct bpf_sock_ops_kern, sk)); 9505 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9506 offsetof(struct sock_common, skc_family)); 9507 break; 9508 9509 case offsetof(struct bpf_sock_ops, remote_ip4): 9510 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 9511 9512 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9513 struct bpf_sock_ops_kern, sk), 9514 si->dst_reg, si->src_reg, 9515 offsetof(struct bpf_sock_ops_kern, sk)); 9516 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9517 offsetof(struct sock_common, skc_daddr)); 9518 break; 9519 9520 case offsetof(struct bpf_sock_ops, local_ip4): 9521 BUILD_BUG_ON(sizeof_field(struct sock_common, 9522 skc_rcv_saddr) != 4); 9523 9524 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9525 struct bpf_sock_ops_kern, sk), 9526 si->dst_reg, si->src_reg, 9527 offsetof(struct bpf_sock_ops_kern, sk)); 9528 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9529 offsetof(struct sock_common, 9530 skc_rcv_saddr)); 9531 break; 9532 9533 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ... 9534 offsetof(struct bpf_sock_ops, remote_ip6[3]): 9535 #if IS_ENABLED(CONFIG_IPV6) 9536 BUILD_BUG_ON(sizeof_field(struct sock_common, 9537 skc_v6_daddr.s6_addr32[0]) != 4); 9538 9539 off = si->off; 9540 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]); 9541 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9542 struct bpf_sock_ops_kern, sk), 9543 si->dst_reg, si->src_reg, 9544 offsetof(struct bpf_sock_ops_kern, sk)); 9545 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9546 offsetof(struct sock_common, 9547 skc_v6_daddr.s6_addr32[0]) + 9548 off); 9549 #else 9550 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9551 #endif 9552 break; 9553 9554 case offsetof(struct bpf_sock_ops, local_ip6[0]) ... 9555 offsetof(struct bpf_sock_ops, local_ip6[3]): 9556 #if IS_ENABLED(CONFIG_IPV6) 9557 BUILD_BUG_ON(sizeof_field(struct sock_common, 9558 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 9559 9560 off = si->off; 9561 off -= offsetof(struct bpf_sock_ops, local_ip6[0]); 9562 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9563 struct bpf_sock_ops_kern, sk), 9564 si->dst_reg, si->src_reg, 9565 offsetof(struct bpf_sock_ops_kern, sk)); 9566 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9567 offsetof(struct sock_common, 9568 skc_v6_rcv_saddr.s6_addr32[0]) + 9569 off); 9570 #else 9571 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9572 #endif 9573 break; 9574 9575 case offsetof(struct bpf_sock_ops, remote_port): 9576 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 9577 9578 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9579 struct bpf_sock_ops_kern, sk), 9580 si->dst_reg, si->src_reg, 9581 offsetof(struct bpf_sock_ops_kern, sk)); 9582 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9583 offsetof(struct sock_common, skc_dport)); 9584 #ifndef __BIG_ENDIAN_BITFIELD 9585 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 9586 #endif 9587 break; 9588 9589 case offsetof(struct bpf_sock_ops, local_port): 9590 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 9591 9592 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9593 struct bpf_sock_ops_kern, sk), 9594 si->dst_reg, si->src_reg, 9595 offsetof(struct bpf_sock_ops_kern, sk)); 9596 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9597 offsetof(struct sock_common, skc_num)); 9598 break; 9599 9600 case offsetof(struct bpf_sock_ops, is_fullsock): 9601 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9602 struct bpf_sock_ops_kern, 9603 is_fullsock), 9604 si->dst_reg, si->src_reg, 9605 offsetof(struct bpf_sock_ops_kern, 9606 is_fullsock)); 9607 break; 9608 9609 case offsetof(struct bpf_sock_ops, state): 9610 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1); 9611 9612 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9613 struct bpf_sock_ops_kern, sk), 9614 si->dst_reg, si->src_reg, 9615 offsetof(struct bpf_sock_ops_kern, sk)); 9616 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg, 9617 offsetof(struct sock_common, skc_state)); 9618 break; 9619 9620 case offsetof(struct bpf_sock_ops, rtt_min): 9621 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 9622 sizeof(struct minmax)); 9623 BUILD_BUG_ON(sizeof(struct minmax) < 9624 sizeof(struct minmax_sample)); 9625 9626 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9627 struct bpf_sock_ops_kern, sk), 9628 si->dst_reg, si->src_reg, 9629 offsetof(struct bpf_sock_ops_kern, sk)); 9630 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9631 offsetof(struct tcp_sock, rtt_min) + 9632 sizeof_field(struct minmax_sample, t)); 9633 break; 9634 9635 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags): 9636 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags, 9637 struct tcp_sock); 9638 break; 9639 9640 case offsetof(struct bpf_sock_ops, sk_txhash): 9641 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash, 9642 struct sock, type); 9643 break; 9644 case offsetof(struct bpf_sock_ops, snd_cwnd): 9645 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd); 9646 break; 9647 case offsetof(struct bpf_sock_ops, srtt_us): 9648 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us); 9649 break; 9650 case offsetof(struct bpf_sock_ops, snd_ssthresh): 9651 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh); 9652 break; 9653 case offsetof(struct bpf_sock_ops, rcv_nxt): 9654 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt); 9655 break; 9656 case offsetof(struct bpf_sock_ops, snd_nxt): 9657 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt); 9658 break; 9659 case offsetof(struct bpf_sock_ops, snd_una): 9660 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una); 9661 break; 9662 case offsetof(struct bpf_sock_ops, mss_cache): 9663 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache); 9664 break; 9665 case offsetof(struct bpf_sock_ops, ecn_flags): 9666 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags); 9667 break; 9668 case offsetof(struct bpf_sock_ops, rate_delivered): 9669 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered); 9670 break; 9671 case offsetof(struct bpf_sock_ops, rate_interval_us): 9672 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us); 9673 break; 9674 case offsetof(struct bpf_sock_ops, packets_out): 9675 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out); 9676 break; 9677 case offsetof(struct bpf_sock_ops, retrans_out): 9678 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out); 9679 break; 9680 case offsetof(struct bpf_sock_ops, total_retrans): 9681 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans); 9682 break; 9683 case offsetof(struct bpf_sock_ops, segs_in): 9684 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in); 9685 break; 9686 case offsetof(struct bpf_sock_ops, data_segs_in): 9687 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in); 9688 break; 9689 case offsetof(struct bpf_sock_ops, segs_out): 9690 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out); 9691 break; 9692 case offsetof(struct bpf_sock_ops, data_segs_out): 9693 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out); 9694 break; 9695 case offsetof(struct bpf_sock_ops, lost_out): 9696 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out); 9697 break; 9698 case offsetof(struct bpf_sock_ops, sacked_out): 9699 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out); 9700 break; 9701 case offsetof(struct bpf_sock_ops, bytes_received): 9702 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received); 9703 break; 9704 case offsetof(struct bpf_sock_ops, bytes_acked): 9705 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked); 9706 break; 9707 case offsetof(struct bpf_sock_ops, sk): 9708 SOCK_OPS_GET_SK(); 9709 break; 9710 case offsetof(struct bpf_sock_ops, skb_data_end): 9711 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9712 skb_data_end), 9713 si->dst_reg, si->src_reg, 9714 offsetof(struct bpf_sock_ops_kern, 9715 skb_data_end)); 9716 break; 9717 case offsetof(struct bpf_sock_ops, skb_data): 9718 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9719 skb), 9720 si->dst_reg, si->src_reg, 9721 offsetof(struct bpf_sock_ops_kern, 9722 skb)); 9723 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9724 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 9725 si->dst_reg, si->dst_reg, 9726 offsetof(struct sk_buff, data)); 9727 break; 9728 case offsetof(struct bpf_sock_ops, skb_len): 9729 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9730 skb), 9731 si->dst_reg, si->src_reg, 9732 offsetof(struct bpf_sock_ops_kern, 9733 skb)); 9734 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9735 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 9736 si->dst_reg, si->dst_reg, 9737 offsetof(struct sk_buff, len)); 9738 break; 9739 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 9740 off = offsetof(struct sk_buff, cb); 9741 off += offsetof(struct tcp_skb_cb, tcp_flags); 9742 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags); 9743 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9744 skb), 9745 si->dst_reg, si->src_reg, 9746 offsetof(struct bpf_sock_ops_kern, 9747 skb)); 9748 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9749 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb, 9750 tcp_flags), 9751 si->dst_reg, si->dst_reg, off); 9752 break; 9753 } 9754 return insn - insn_buf; 9755 } 9756 9757 /* data_end = skb->data + skb_headlen() */ 9758 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si, 9759 struct bpf_insn *insn) 9760 { 9761 int reg; 9762 int temp_reg_off = offsetof(struct sk_buff, cb) + 9763 offsetof(struct sk_skb_cb, temp_reg); 9764 9765 if (si->src_reg == si->dst_reg) { 9766 /* We need an extra register, choose and save a register. */ 9767 reg = BPF_REG_9; 9768 if (si->src_reg == reg || si->dst_reg == reg) 9769 reg--; 9770 if (si->src_reg == reg || si->dst_reg == reg) 9771 reg--; 9772 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off); 9773 } else { 9774 reg = si->dst_reg; 9775 } 9776 9777 /* reg = skb->data */ 9778 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 9779 reg, si->src_reg, 9780 offsetof(struct sk_buff, data)); 9781 /* AX = skb->len */ 9782 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 9783 BPF_REG_AX, si->src_reg, 9784 offsetof(struct sk_buff, len)); 9785 /* reg = skb->data + skb->len */ 9786 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX); 9787 /* AX = skb->data_len */ 9788 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len), 9789 BPF_REG_AX, si->src_reg, 9790 offsetof(struct sk_buff, data_len)); 9791 9792 /* reg = skb->data + skb->len - skb->data_len */ 9793 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX); 9794 9795 if (si->src_reg == si->dst_reg) { 9796 /* Restore the saved register */ 9797 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg); 9798 *insn++ = BPF_MOV64_REG(si->dst_reg, reg); 9799 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off); 9800 } 9801 9802 return insn; 9803 } 9804 9805 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type, 9806 const struct bpf_insn *si, 9807 struct bpf_insn *insn_buf, 9808 struct bpf_prog *prog, u32 *target_size) 9809 { 9810 struct bpf_insn *insn = insn_buf; 9811 int off; 9812 9813 switch (si->off) { 9814 case offsetof(struct __sk_buff, data_end): 9815 insn = bpf_convert_data_end_access(si, insn); 9816 break; 9817 case offsetof(struct __sk_buff, cb[0]) ... 9818 offsetofend(struct __sk_buff, cb[4]) - 1: 9819 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20); 9820 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 9821 offsetof(struct sk_skb_cb, data)) % 9822 sizeof(__u64)); 9823 9824 prog->cb_access = 1; 9825 off = si->off; 9826 off -= offsetof(struct __sk_buff, cb[0]); 9827 off += offsetof(struct sk_buff, cb); 9828 off += offsetof(struct sk_skb_cb, data); 9829 if (type == BPF_WRITE) 9830 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 9831 si->src_reg, off); 9832 else 9833 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 9834 si->src_reg, off); 9835 break; 9836 9837 9838 default: 9839 return bpf_convert_ctx_access(type, si, insn_buf, prog, 9840 target_size); 9841 } 9842 9843 return insn - insn_buf; 9844 } 9845 9846 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type, 9847 const struct bpf_insn *si, 9848 struct bpf_insn *insn_buf, 9849 struct bpf_prog *prog, u32 *target_size) 9850 { 9851 struct bpf_insn *insn = insn_buf; 9852 #if IS_ENABLED(CONFIG_IPV6) 9853 int off; 9854 #endif 9855 9856 /* convert ctx uses the fact sg element is first in struct */ 9857 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0); 9858 9859 switch (si->off) { 9860 case offsetof(struct sk_msg_md, data): 9861 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data), 9862 si->dst_reg, si->src_reg, 9863 offsetof(struct sk_msg, data)); 9864 break; 9865 case offsetof(struct sk_msg_md, data_end): 9866 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end), 9867 si->dst_reg, si->src_reg, 9868 offsetof(struct sk_msg, data_end)); 9869 break; 9870 case offsetof(struct sk_msg_md, family): 9871 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 9872 9873 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9874 struct sk_msg, sk), 9875 si->dst_reg, si->src_reg, 9876 offsetof(struct sk_msg, sk)); 9877 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9878 offsetof(struct sock_common, skc_family)); 9879 break; 9880 9881 case offsetof(struct sk_msg_md, remote_ip4): 9882 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 9883 9884 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9885 struct sk_msg, sk), 9886 si->dst_reg, si->src_reg, 9887 offsetof(struct sk_msg, sk)); 9888 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9889 offsetof(struct sock_common, skc_daddr)); 9890 break; 9891 9892 case offsetof(struct sk_msg_md, local_ip4): 9893 BUILD_BUG_ON(sizeof_field(struct sock_common, 9894 skc_rcv_saddr) != 4); 9895 9896 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9897 struct sk_msg, sk), 9898 si->dst_reg, si->src_reg, 9899 offsetof(struct sk_msg, sk)); 9900 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9901 offsetof(struct sock_common, 9902 skc_rcv_saddr)); 9903 break; 9904 9905 case offsetof(struct sk_msg_md, remote_ip6[0]) ... 9906 offsetof(struct sk_msg_md, remote_ip6[3]): 9907 #if IS_ENABLED(CONFIG_IPV6) 9908 BUILD_BUG_ON(sizeof_field(struct sock_common, 9909 skc_v6_daddr.s6_addr32[0]) != 4); 9910 9911 off = si->off; 9912 off -= offsetof(struct sk_msg_md, remote_ip6[0]); 9913 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9914 struct sk_msg, sk), 9915 si->dst_reg, si->src_reg, 9916 offsetof(struct sk_msg, sk)); 9917 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9918 offsetof(struct sock_common, 9919 skc_v6_daddr.s6_addr32[0]) + 9920 off); 9921 #else 9922 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9923 #endif 9924 break; 9925 9926 case offsetof(struct sk_msg_md, local_ip6[0]) ... 9927 offsetof(struct sk_msg_md, local_ip6[3]): 9928 #if IS_ENABLED(CONFIG_IPV6) 9929 BUILD_BUG_ON(sizeof_field(struct sock_common, 9930 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 9931 9932 off = si->off; 9933 off -= offsetof(struct sk_msg_md, local_ip6[0]); 9934 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9935 struct sk_msg, sk), 9936 si->dst_reg, si->src_reg, 9937 offsetof(struct sk_msg, sk)); 9938 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9939 offsetof(struct sock_common, 9940 skc_v6_rcv_saddr.s6_addr32[0]) + 9941 off); 9942 #else 9943 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9944 #endif 9945 break; 9946 9947 case offsetof(struct sk_msg_md, remote_port): 9948 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 9949 9950 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9951 struct sk_msg, sk), 9952 si->dst_reg, si->src_reg, 9953 offsetof(struct sk_msg, sk)); 9954 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9955 offsetof(struct sock_common, skc_dport)); 9956 #ifndef __BIG_ENDIAN_BITFIELD 9957 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 9958 #endif 9959 break; 9960 9961 case offsetof(struct sk_msg_md, local_port): 9962 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 9963 9964 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9965 struct sk_msg, sk), 9966 si->dst_reg, si->src_reg, 9967 offsetof(struct sk_msg, sk)); 9968 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9969 offsetof(struct sock_common, skc_num)); 9970 break; 9971 9972 case offsetof(struct sk_msg_md, size): 9973 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size), 9974 si->dst_reg, si->src_reg, 9975 offsetof(struct sk_msg_sg, size)); 9976 break; 9977 9978 case offsetof(struct sk_msg_md, sk): 9979 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk), 9980 si->dst_reg, si->src_reg, 9981 offsetof(struct sk_msg, sk)); 9982 break; 9983 } 9984 9985 return insn - insn_buf; 9986 } 9987 9988 const struct bpf_verifier_ops sk_filter_verifier_ops = { 9989 .get_func_proto = sk_filter_func_proto, 9990 .is_valid_access = sk_filter_is_valid_access, 9991 .convert_ctx_access = bpf_convert_ctx_access, 9992 .gen_ld_abs = bpf_gen_ld_abs, 9993 }; 9994 9995 const struct bpf_prog_ops sk_filter_prog_ops = { 9996 .test_run = bpf_prog_test_run_skb, 9997 }; 9998 9999 const struct bpf_verifier_ops tc_cls_act_verifier_ops = { 10000 .get_func_proto = tc_cls_act_func_proto, 10001 .is_valid_access = tc_cls_act_is_valid_access, 10002 .convert_ctx_access = tc_cls_act_convert_ctx_access, 10003 .gen_prologue = tc_cls_act_prologue, 10004 .gen_ld_abs = bpf_gen_ld_abs, 10005 .check_kfunc_call = bpf_prog_test_check_kfunc_call, 10006 }; 10007 10008 const struct bpf_prog_ops tc_cls_act_prog_ops = { 10009 .test_run = bpf_prog_test_run_skb, 10010 }; 10011 10012 const struct bpf_verifier_ops xdp_verifier_ops = { 10013 .get_func_proto = xdp_func_proto, 10014 .is_valid_access = xdp_is_valid_access, 10015 .convert_ctx_access = xdp_convert_ctx_access, 10016 .gen_prologue = bpf_noop_prologue, 10017 }; 10018 10019 const struct bpf_prog_ops xdp_prog_ops = { 10020 .test_run = bpf_prog_test_run_xdp, 10021 }; 10022 10023 const struct bpf_verifier_ops cg_skb_verifier_ops = { 10024 .get_func_proto = cg_skb_func_proto, 10025 .is_valid_access = cg_skb_is_valid_access, 10026 .convert_ctx_access = bpf_convert_ctx_access, 10027 }; 10028 10029 const struct bpf_prog_ops cg_skb_prog_ops = { 10030 .test_run = bpf_prog_test_run_skb, 10031 }; 10032 10033 const struct bpf_verifier_ops lwt_in_verifier_ops = { 10034 .get_func_proto = lwt_in_func_proto, 10035 .is_valid_access = lwt_is_valid_access, 10036 .convert_ctx_access = bpf_convert_ctx_access, 10037 }; 10038 10039 const struct bpf_prog_ops lwt_in_prog_ops = { 10040 .test_run = bpf_prog_test_run_skb, 10041 }; 10042 10043 const struct bpf_verifier_ops lwt_out_verifier_ops = { 10044 .get_func_proto = lwt_out_func_proto, 10045 .is_valid_access = lwt_is_valid_access, 10046 .convert_ctx_access = bpf_convert_ctx_access, 10047 }; 10048 10049 const struct bpf_prog_ops lwt_out_prog_ops = { 10050 .test_run = bpf_prog_test_run_skb, 10051 }; 10052 10053 const struct bpf_verifier_ops lwt_xmit_verifier_ops = { 10054 .get_func_proto = lwt_xmit_func_proto, 10055 .is_valid_access = lwt_is_valid_access, 10056 .convert_ctx_access = bpf_convert_ctx_access, 10057 .gen_prologue = tc_cls_act_prologue, 10058 }; 10059 10060 const struct bpf_prog_ops lwt_xmit_prog_ops = { 10061 .test_run = bpf_prog_test_run_skb, 10062 }; 10063 10064 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = { 10065 .get_func_proto = lwt_seg6local_func_proto, 10066 .is_valid_access = lwt_is_valid_access, 10067 .convert_ctx_access = bpf_convert_ctx_access, 10068 }; 10069 10070 const struct bpf_prog_ops lwt_seg6local_prog_ops = { 10071 .test_run = bpf_prog_test_run_skb, 10072 }; 10073 10074 const struct bpf_verifier_ops cg_sock_verifier_ops = { 10075 .get_func_proto = sock_filter_func_proto, 10076 .is_valid_access = sock_filter_is_valid_access, 10077 .convert_ctx_access = bpf_sock_convert_ctx_access, 10078 }; 10079 10080 const struct bpf_prog_ops cg_sock_prog_ops = { 10081 }; 10082 10083 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = { 10084 .get_func_proto = sock_addr_func_proto, 10085 .is_valid_access = sock_addr_is_valid_access, 10086 .convert_ctx_access = sock_addr_convert_ctx_access, 10087 }; 10088 10089 const struct bpf_prog_ops cg_sock_addr_prog_ops = { 10090 }; 10091 10092 const struct bpf_verifier_ops sock_ops_verifier_ops = { 10093 .get_func_proto = sock_ops_func_proto, 10094 .is_valid_access = sock_ops_is_valid_access, 10095 .convert_ctx_access = sock_ops_convert_ctx_access, 10096 }; 10097 10098 const struct bpf_prog_ops sock_ops_prog_ops = { 10099 }; 10100 10101 const struct bpf_verifier_ops sk_skb_verifier_ops = { 10102 .get_func_proto = sk_skb_func_proto, 10103 .is_valid_access = sk_skb_is_valid_access, 10104 .convert_ctx_access = sk_skb_convert_ctx_access, 10105 .gen_prologue = sk_skb_prologue, 10106 }; 10107 10108 const struct bpf_prog_ops sk_skb_prog_ops = { 10109 }; 10110 10111 const struct bpf_verifier_ops sk_msg_verifier_ops = { 10112 .get_func_proto = sk_msg_func_proto, 10113 .is_valid_access = sk_msg_is_valid_access, 10114 .convert_ctx_access = sk_msg_convert_ctx_access, 10115 .gen_prologue = bpf_noop_prologue, 10116 }; 10117 10118 const struct bpf_prog_ops sk_msg_prog_ops = { 10119 }; 10120 10121 const struct bpf_verifier_ops flow_dissector_verifier_ops = { 10122 .get_func_proto = flow_dissector_func_proto, 10123 .is_valid_access = flow_dissector_is_valid_access, 10124 .convert_ctx_access = flow_dissector_convert_ctx_access, 10125 }; 10126 10127 const struct bpf_prog_ops flow_dissector_prog_ops = { 10128 .test_run = bpf_prog_test_run_flow_dissector, 10129 }; 10130 10131 int sk_detach_filter(struct sock *sk) 10132 { 10133 int ret = -ENOENT; 10134 struct sk_filter *filter; 10135 10136 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 10137 return -EPERM; 10138 10139 filter = rcu_dereference_protected(sk->sk_filter, 10140 lockdep_sock_is_held(sk)); 10141 if (filter) { 10142 RCU_INIT_POINTER(sk->sk_filter, NULL); 10143 sk_filter_uncharge(sk, filter); 10144 ret = 0; 10145 } 10146 10147 return ret; 10148 } 10149 EXPORT_SYMBOL_GPL(sk_detach_filter); 10150 10151 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, 10152 unsigned int len) 10153 { 10154 struct sock_fprog_kern *fprog; 10155 struct sk_filter *filter; 10156 int ret = 0; 10157 10158 lock_sock(sk); 10159 filter = rcu_dereference_protected(sk->sk_filter, 10160 lockdep_sock_is_held(sk)); 10161 if (!filter) 10162 goto out; 10163 10164 /* We're copying the filter that has been originally attached, 10165 * so no conversion/decode needed anymore. eBPF programs that 10166 * have no original program cannot be dumped through this. 10167 */ 10168 ret = -EACCES; 10169 fprog = filter->prog->orig_prog; 10170 if (!fprog) 10171 goto out; 10172 10173 ret = fprog->len; 10174 if (!len) 10175 /* User space only enquires number of filter blocks. */ 10176 goto out; 10177 10178 ret = -EINVAL; 10179 if (len < fprog->len) 10180 goto out; 10181 10182 ret = -EFAULT; 10183 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog))) 10184 goto out; 10185 10186 /* Instead of bytes, the API requests to return the number 10187 * of filter blocks. 10188 */ 10189 ret = fprog->len; 10190 out: 10191 release_sock(sk); 10192 return ret; 10193 } 10194 10195 #ifdef CONFIG_INET 10196 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern, 10197 struct sock_reuseport *reuse, 10198 struct sock *sk, struct sk_buff *skb, 10199 struct sock *migrating_sk, 10200 u32 hash) 10201 { 10202 reuse_kern->skb = skb; 10203 reuse_kern->sk = sk; 10204 reuse_kern->selected_sk = NULL; 10205 reuse_kern->migrating_sk = migrating_sk; 10206 reuse_kern->data_end = skb->data + skb_headlen(skb); 10207 reuse_kern->hash = hash; 10208 reuse_kern->reuseport_id = reuse->reuseport_id; 10209 reuse_kern->bind_inany = reuse->bind_inany; 10210 } 10211 10212 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 10213 struct bpf_prog *prog, struct sk_buff *skb, 10214 struct sock *migrating_sk, 10215 u32 hash) 10216 { 10217 struct sk_reuseport_kern reuse_kern; 10218 enum sk_action action; 10219 10220 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash); 10221 action = bpf_prog_run(prog, &reuse_kern); 10222 10223 if (action == SK_PASS) 10224 return reuse_kern.selected_sk; 10225 else 10226 return ERR_PTR(-ECONNREFUSED); 10227 } 10228 10229 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern, 10230 struct bpf_map *, map, void *, key, u32, flags) 10231 { 10232 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY; 10233 struct sock_reuseport *reuse; 10234 struct sock *selected_sk; 10235 10236 selected_sk = map->ops->map_lookup_elem(map, key); 10237 if (!selected_sk) 10238 return -ENOENT; 10239 10240 reuse = rcu_dereference(selected_sk->sk_reuseport_cb); 10241 if (!reuse) { 10242 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */ 10243 if (sk_is_refcounted(selected_sk)) 10244 sock_put(selected_sk); 10245 10246 /* reuseport_array has only sk with non NULL sk_reuseport_cb. 10247 * The only (!reuse) case here is - the sk has already been 10248 * unhashed (e.g. by close()), so treat it as -ENOENT. 10249 * 10250 * Other maps (e.g. sock_map) do not provide this guarantee and 10251 * the sk may never be in the reuseport group to begin with. 10252 */ 10253 return is_sockarray ? -ENOENT : -EINVAL; 10254 } 10255 10256 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) { 10257 struct sock *sk = reuse_kern->sk; 10258 10259 if (sk->sk_protocol != selected_sk->sk_protocol) 10260 return -EPROTOTYPE; 10261 else if (sk->sk_family != selected_sk->sk_family) 10262 return -EAFNOSUPPORT; 10263 10264 /* Catch all. Likely bound to a different sockaddr. */ 10265 return -EBADFD; 10266 } 10267 10268 reuse_kern->selected_sk = selected_sk; 10269 10270 return 0; 10271 } 10272 10273 static const struct bpf_func_proto sk_select_reuseport_proto = { 10274 .func = sk_select_reuseport, 10275 .gpl_only = false, 10276 .ret_type = RET_INTEGER, 10277 .arg1_type = ARG_PTR_TO_CTX, 10278 .arg2_type = ARG_CONST_MAP_PTR, 10279 .arg3_type = ARG_PTR_TO_MAP_KEY, 10280 .arg4_type = ARG_ANYTHING, 10281 }; 10282 10283 BPF_CALL_4(sk_reuseport_load_bytes, 10284 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 10285 void *, to, u32, len) 10286 { 10287 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len); 10288 } 10289 10290 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = { 10291 .func = sk_reuseport_load_bytes, 10292 .gpl_only = false, 10293 .ret_type = RET_INTEGER, 10294 .arg1_type = ARG_PTR_TO_CTX, 10295 .arg2_type = ARG_ANYTHING, 10296 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 10297 .arg4_type = ARG_CONST_SIZE, 10298 }; 10299 10300 BPF_CALL_5(sk_reuseport_load_bytes_relative, 10301 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 10302 void *, to, u32, len, u32, start_header) 10303 { 10304 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to, 10305 len, start_header); 10306 } 10307 10308 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = { 10309 .func = sk_reuseport_load_bytes_relative, 10310 .gpl_only = false, 10311 .ret_type = RET_INTEGER, 10312 .arg1_type = ARG_PTR_TO_CTX, 10313 .arg2_type = ARG_ANYTHING, 10314 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 10315 .arg4_type = ARG_CONST_SIZE, 10316 .arg5_type = ARG_ANYTHING, 10317 }; 10318 10319 static const struct bpf_func_proto * 10320 sk_reuseport_func_proto(enum bpf_func_id func_id, 10321 const struct bpf_prog *prog) 10322 { 10323 switch (func_id) { 10324 case BPF_FUNC_sk_select_reuseport: 10325 return &sk_select_reuseport_proto; 10326 case BPF_FUNC_skb_load_bytes: 10327 return &sk_reuseport_load_bytes_proto; 10328 case BPF_FUNC_skb_load_bytes_relative: 10329 return &sk_reuseport_load_bytes_relative_proto; 10330 case BPF_FUNC_get_socket_cookie: 10331 return &bpf_get_socket_ptr_cookie_proto; 10332 case BPF_FUNC_ktime_get_coarse_ns: 10333 return &bpf_ktime_get_coarse_ns_proto; 10334 default: 10335 return bpf_base_func_proto(func_id); 10336 } 10337 } 10338 10339 static bool 10340 sk_reuseport_is_valid_access(int off, int size, 10341 enum bpf_access_type type, 10342 const struct bpf_prog *prog, 10343 struct bpf_insn_access_aux *info) 10344 { 10345 const u32 size_default = sizeof(__u32); 10346 10347 if (off < 0 || off >= sizeof(struct sk_reuseport_md) || 10348 off % size || type != BPF_READ) 10349 return false; 10350 10351 switch (off) { 10352 case offsetof(struct sk_reuseport_md, data): 10353 info->reg_type = PTR_TO_PACKET; 10354 return size == sizeof(__u64); 10355 10356 case offsetof(struct sk_reuseport_md, data_end): 10357 info->reg_type = PTR_TO_PACKET_END; 10358 return size == sizeof(__u64); 10359 10360 case offsetof(struct sk_reuseport_md, hash): 10361 return size == size_default; 10362 10363 case offsetof(struct sk_reuseport_md, sk): 10364 info->reg_type = PTR_TO_SOCKET; 10365 return size == sizeof(__u64); 10366 10367 case offsetof(struct sk_reuseport_md, migrating_sk): 10368 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 10369 return size == sizeof(__u64); 10370 10371 /* Fields that allow narrowing */ 10372 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol): 10373 if (size < sizeof_field(struct sk_buff, protocol)) 10374 return false; 10375 fallthrough; 10376 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol): 10377 case bpf_ctx_range(struct sk_reuseport_md, bind_inany): 10378 case bpf_ctx_range(struct sk_reuseport_md, len): 10379 bpf_ctx_record_field_size(info, size_default); 10380 return bpf_ctx_narrow_access_ok(off, size, size_default); 10381 10382 default: 10383 return false; 10384 } 10385 } 10386 10387 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \ 10388 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \ 10389 si->dst_reg, si->src_reg, \ 10390 bpf_target_off(struct sk_reuseport_kern, F, \ 10391 sizeof_field(struct sk_reuseport_kern, F), \ 10392 target_size)); \ 10393 }) 10394 10395 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \ 10396 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 10397 struct sk_buff, \ 10398 skb, \ 10399 SKB_FIELD) 10400 10401 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \ 10402 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 10403 struct sock, \ 10404 sk, \ 10405 SK_FIELD) 10406 10407 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type, 10408 const struct bpf_insn *si, 10409 struct bpf_insn *insn_buf, 10410 struct bpf_prog *prog, 10411 u32 *target_size) 10412 { 10413 struct bpf_insn *insn = insn_buf; 10414 10415 switch (si->off) { 10416 case offsetof(struct sk_reuseport_md, data): 10417 SK_REUSEPORT_LOAD_SKB_FIELD(data); 10418 break; 10419 10420 case offsetof(struct sk_reuseport_md, len): 10421 SK_REUSEPORT_LOAD_SKB_FIELD(len); 10422 break; 10423 10424 case offsetof(struct sk_reuseport_md, eth_protocol): 10425 SK_REUSEPORT_LOAD_SKB_FIELD(protocol); 10426 break; 10427 10428 case offsetof(struct sk_reuseport_md, ip_protocol): 10429 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol); 10430 break; 10431 10432 case offsetof(struct sk_reuseport_md, data_end): 10433 SK_REUSEPORT_LOAD_FIELD(data_end); 10434 break; 10435 10436 case offsetof(struct sk_reuseport_md, hash): 10437 SK_REUSEPORT_LOAD_FIELD(hash); 10438 break; 10439 10440 case offsetof(struct sk_reuseport_md, bind_inany): 10441 SK_REUSEPORT_LOAD_FIELD(bind_inany); 10442 break; 10443 10444 case offsetof(struct sk_reuseport_md, sk): 10445 SK_REUSEPORT_LOAD_FIELD(sk); 10446 break; 10447 10448 case offsetof(struct sk_reuseport_md, migrating_sk): 10449 SK_REUSEPORT_LOAD_FIELD(migrating_sk); 10450 break; 10451 } 10452 10453 return insn - insn_buf; 10454 } 10455 10456 const struct bpf_verifier_ops sk_reuseport_verifier_ops = { 10457 .get_func_proto = sk_reuseport_func_proto, 10458 .is_valid_access = sk_reuseport_is_valid_access, 10459 .convert_ctx_access = sk_reuseport_convert_ctx_access, 10460 }; 10461 10462 const struct bpf_prog_ops sk_reuseport_prog_ops = { 10463 }; 10464 10465 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled); 10466 EXPORT_SYMBOL(bpf_sk_lookup_enabled); 10467 10468 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx, 10469 struct sock *, sk, u64, flags) 10470 { 10471 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE | 10472 BPF_SK_LOOKUP_F_NO_REUSEPORT))) 10473 return -EINVAL; 10474 if (unlikely(sk && sk_is_refcounted(sk))) 10475 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */ 10476 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN)) 10477 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */ 10478 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE)) 10479 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */ 10480 10481 /* Check if socket is suitable for packet L3/L4 protocol */ 10482 if (sk && sk->sk_protocol != ctx->protocol) 10483 return -EPROTOTYPE; 10484 if (sk && sk->sk_family != ctx->family && 10485 (sk->sk_family == AF_INET || ipv6_only_sock(sk))) 10486 return -EAFNOSUPPORT; 10487 10488 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE)) 10489 return -EEXIST; 10490 10491 /* Select socket as lookup result */ 10492 ctx->selected_sk = sk; 10493 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT; 10494 return 0; 10495 } 10496 10497 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = { 10498 .func = bpf_sk_lookup_assign, 10499 .gpl_only = false, 10500 .ret_type = RET_INTEGER, 10501 .arg1_type = ARG_PTR_TO_CTX, 10502 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL, 10503 .arg3_type = ARG_ANYTHING, 10504 }; 10505 10506 static const struct bpf_func_proto * 10507 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 10508 { 10509 switch (func_id) { 10510 case BPF_FUNC_perf_event_output: 10511 return &bpf_event_output_data_proto; 10512 case BPF_FUNC_sk_assign: 10513 return &bpf_sk_lookup_assign_proto; 10514 case BPF_FUNC_sk_release: 10515 return &bpf_sk_release_proto; 10516 default: 10517 return bpf_sk_base_func_proto(func_id); 10518 } 10519 } 10520 10521 static bool sk_lookup_is_valid_access(int off, int size, 10522 enum bpf_access_type type, 10523 const struct bpf_prog *prog, 10524 struct bpf_insn_access_aux *info) 10525 { 10526 if (off < 0 || off >= sizeof(struct bpf_sk_lookup)) 10527 return false; 10528 if (off % size != 0) 10529 return false; 10530 if (type != BPF_READ) 10531 return false; 10532 10533 switch (off) { 10534 case offsetof(struct bpf_sk_lookup, sk): 10535 info->reg_type = PTR_TO_SOCKET_OR_NULL; 10536 return size == sizeof(__u64); 10537 10538 case bpf_ctx_range(struct bpf_sk_lookup, family): 10539 case bpf_ctx_range(struct bpf_sk_lookup, protocol): 10540 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4): 10541 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4): 10542 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]): 10543 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]): 10544 case bpf_ctx_range(struct bpf_sk_lookup, remote_port): 10545 case bpf_ctx_range(struct bpf_sk_lookup, local_port): 10546 bpf_ctx_record_field_size(info, sizeof(__u32)); 10547 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32)); 10548 10549 default: 10550 return false; 10551 } 10552 } 10553 10554 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type, 10555 const struct bpf_insn *si, 10556 struct bpf_insn *insn_buf, 10557 struct bpf_prog *prog, 10558 u32 *target_size) 10559 { 10560 struct bpf_insn *insn = insn_buf; 10561 10562 switch (si->off) { 10563 case offsetof(struct bpf_sk_lookup, sk): 10564 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 10565 offsetof(struct bpf_sk_lookup_kern, selected_sk)); 10566 break; 10567 10568 case offsetof(struct bpf_sk_lookup, family): 10569 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10570 bpf_target_off(struct bpf_sk_lookup_kern, 10571 family, 2, target_size)); 10572 break; 10573 10574 case offsetof(struct bpf_sk_lookup, protocol): 10575 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10576 bpf_target_off(struct bpf_sk_lookup_kern, 10577 protocol, 2, target_size)); 10578 break; 10579 10580 case offsetof(struct bpf_sk_lookup, remote_ip4): 10581 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10582 bpf_target_off(struct bpf_sk_lookup_kern, 10583 v4.saddr, 4, target_size)); 10584 break; 10585 10586 case offsetof(struct bpf_sk_lookup, local_ip4): 10587 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 10588 bpf_target_off(struct bpf_sk_lookup_kern, 10589 v4.daddr, 4, target_size)); 10590 break; 10591 10592 case bpf_ctx_range_till(struct bpf_sk_lookup, 10593 remote_ip6[0], remote_ip6[3]): { 10594 #if IS_ENABLED(CONFIG_IPV6) 10595 int off = si->off; 10596 10597 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]); 10598 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 10599 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 10600 offsetof(struct bpf_sk_lookup_kern, v6.saddr)); 10601 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10602 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 10603 #else 10604 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10605 #endif 10606 break; 10607 } 10608 case bpf_ctx_range_till(struct bpf_sk_lookup, 10609 local_ip6[0], local_ip6[3]): { 10610 #if IS_ENABLED(CONFIG_IPV6) 10611 int off = si->off; 10612 10613 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]); 10614 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 10615 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 10616 offsetof(struct bpf_sk_lookup_kern, v6.daddr)); 10617 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10618 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 10619 #else 10620 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10621 #endif 10622 break; 10623 } 10624 case offsetof(struct bpf_sk_lookup, remote_port): 10625 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10626 bpf_target_off(struct bpf_sk_lookup_kern, 10627 sport, 2, target_size)); 10628 break; 10629 10630 case offsetof(struct bpf_sk_lookup, local_port): 10631 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 10632 bpf_target_off(struct bpf_sk_lookup_kern, 10633 dport, 2, target_size)); 10634 break; 10635 } 10636 10637 return insn - insn_buf; 10638 } 10639 10640 const struct bpf_prog_ops sk_lookup_prog_ops = { 10641 .test_run = bpf_prog_test_run_sk_lookup, 10642 }; 10643 10644 const struct bpf_verifier_ops sk_lookup_verifier_ops = { 10645 .get_func_proto = sk_lookup_func_proto, 10646 .is_valid_access = sk_lookup_is_valid_access, 10647 .convert_ctx_access = sk_lookup_convert_ctx_access, 10648 }; 10649 10650 #endif /* CONFIG_INET */ 10651 10652 DEFINE_BPF_DISPATCHER(xdp) 10653 10654 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog) 10655 { 10656 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog); 10657 } 10658 10659 #ifdef CONFIG_DEBUG_INFO_BTF 10660 BTF_ID_LIST_GLOBAL(btf_sock_ids) 10661 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type) 10662 BTF_SOCK_TYPE_xxx 10663 #undef BTF_SOCK_TYPE 10664 #else 10665 u32 btf_sock_ids[MAX_BTF_SOCK_TYPE]; 10666 #endif 10667 10668 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk) 10669 { 10670 /* tcp6_sock type is not generated in dwarf and hence btf, 10671 * trigger an explicit type generation here. 10672 */ 10673 BTF_TYPE_EMIT(struct tcp6_sock); 10674 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && 10675 sk->sk_family == AF_INET6) 10676 return (unsigned long)sk; 10677 10678 return (unsigned long)NULL; 10679 } 10680 10681 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = { 10682 .func = bpf_skc_to_tcp6_sock, 10683 .gpl_only = false, 10684 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10685 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10686 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6], 10687 }; 10688 10689 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk) 10690 { 10691 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 10692 return (unsigned long)sk; 10693 10694 return (unsigned long)NULL; 10695 } 10696 10697 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = { 10698 .func = bpf_skc_to_tcp_sock, 10699 .gpl_only = false, 10700 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10701 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10702 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP], 10703 }; 10704 10705 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk) 10706 { 10707 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not 10708 * generated if CONFIG_INET=n. Trigger an explicit generation here. 10709 */ 10710 BTF_TYPE_EMIT(struct inet_timewait_sock); 10711 BTF_TYPE_EMIT(struct tcp_timewait_sock); 10712 10713 #ifdef CONFIG_INET 10714 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT) 10715 return (unsigned long)sk; 10716 #endif 10717 10718 #if IS_BUILTIN(CONFIG_IPV6) 10719 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT) 10720 return (unsigned long)sk; 10721 #endif 10722 10723 return (unsigned long)NULL; 10724 } 10725 10726 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = { 10727 .func = bpf_skc_to_tcp_timewait_sock, 10728 .gpl_only = false, 10729 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10730 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10731 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW], 10732 }; 10733 10734 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk) 10735 { 10736 #ifdef CONFIG_INET 10737 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV) 10738 return (unsigned long)sk; 10739 #endif 10740 10741 #if IS_BUILTIN(CONFIG_IPV6) 10742 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV) 10743 return (unsigned long)sk; 10744 #endif 10745 10746 return (unsigned long)NULL; 10747 } 10748 10749 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = { 10750 .func = bpf_skc_to_tcp_request_sock, 10751 .gpl_only = false, 10752 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10753 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10754 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ], 10755 }; 10756 10757 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk) 10758 { 10759 /* udp6_sock type is not generated in dwarf and hence btf, 10760 * trigger an explicit type generation here. 10761 */ 10762 BTF_TYPE_EMIT(struct udp6_sock); 10763 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP && 10764 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6) 10765 return (unsigned long)sk; 10766 10767 return (unsigned long)NULL; 10768 } 10769 10770 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = { 10771 .func = bpf_skc_to_udp6_sock, 10772 .gpl_only = false, 10773 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10774 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10775 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6], 10776 }; 10777 10778 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk) 10779 { 10780 /* unix_sock type is not generated in dwarf and hence btf, 10781 * trigger an explicit type generation here. 10782 */ 10783 BTF_TYPE_EMIT(struct unix_sock); 10784 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX) 10785 return (unsigned long)sk; 10786 10787 return (unsigned long)NULL; 10788 } 10789 10790 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = { 10791 .func = bpf_skc_to_unix_sock, 10792 .gpl_only = false, 10793 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10794 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 10795 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX], 10796 }; 10797 10798 BPF_CALL_1(bpf_sock_from_file, struct file *, file) 10799 { 10800 return (unsigned long)sock_from_file(file); 10801 } 10802 10803 BTF_ID_LIST(bpf_sock_from_file_btf_ids) 10804 BTF_ID(struct, socket) 10805 BTF_ID(struct, file) 10806 10807 const struct bpf_func_proto bpf_sock_from_file_proto = { 10808 .func = bpf_sock_from_file, 10809 .gpl_only = false, 10810 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 10811 .ret_btf_id = &bpf_sock_from_file_btf_ids[0], 10812 .arg1_type = ARG_PTR_TO_BTF_ID, 10813 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1], 10814 }; 10815 10816 static const struct bpf_func_proto * 10817 bpf_sk_base_func_proto(enum bpf_func_id func_id) 10818 { 10819 const struct bpf_func_proto *func; 10820 10821 switch (func_id) { 10822 case BPF_FUNC_skc_to_tcp6_sock: 10823 func = &bpf_skc_to_tcp6_sock_proto; 10824 break; 10825 case BPF_FUNC_skc_to_tcp_sock: 10826 func = &bpf_skc_to_tcp_sock_proto; 10827 break; 10828 case BPF_FUNC_skc_to_tcp_timewait_sock: 10829 func = &bpf_skc_to_tcp_timewait_sock_proto; 10830 break; 10831 case BPF_FUNC_skc_to_tcp_request_sock: 10832 func = &bpf_skc_to_tcp_request_sock_proto; 10833 break; 10834 case BPF_FUNC_skc_to_udp6_sock: 10835 func = &bpf_skc_to_udp6_sock_proto; 10836 break; 10837 case BPF_FUNC_skc_to_unix_sock: 10838 func = &bpf_skc_to_unix_sock_proto; 10839 break; 10840 case BPF_FUNC_ktime_get_coarse_ns: 10841 return &bpf_ktime_get_coarse_ns_proto; 10842 default: 10843 return bpf_base_func_proto(func_id); 10844 } 10845 10846 if (!perfmon_capable()) 10847 return NULL; 10848 10849 return func; 10850 } 10851