1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Linux Socket Filter - Kernel level socket filtering 4 * 5 * Based on the design of the Berkeley Packet Filter. The new 6 * internal format has been designed by PLUMgrid: 7 * 8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com 9 * 10 * Authors: 11 * 12 * Jay Schulist <jschlst@samba.org> 13 * Alexei Starovoitov <ast@plumgrid.com> 14 * Daniel Borkmann <dborkman@redhat.com> 15 * 16 * Andi Kleen - Fix a few bad bugs and races. 17 * Kris Katterjohn - Added many additional checks in bpf_check_classic() 18 */ 19 20 #include <linux/atomic.h> 21 #include <linux/module.h> 22 #include <linux/types.h> 23 #include <linux/mm.h> 24 #include <linux/fcntl.h> 25 #include <linux/socket.h> 26 #include <linux/sock_diag.h> 27 #include <linux/in.h> 28 #include <linux/inet.h> 29 #include <linux/netdevice.h> 30 #include <linux/if_packet.h> 31 #include <linux/if_arp.h> 32 #include <linux/gfp.h> 33 #include <net/inet_common.h> 34 #include <net/ip.h> 35 #include <net/protocol.h> 36 #include <net/netlink.h> 37 #include <linux/skbuff.h> 38 #include <linux/skmsg.h> 39 #include <net/sock.h> 40 #include <net/flow_dissector.h> 41 #include <linux/errno.h> 42 #include <linux/timer.h> 43 #include <linux/uaccess.h> 44 #include <asm/unaligned.h> 45 #include <linux/filter.h> 46 #include <linux/ratelimit.h> 47 #include <linux/seccomp.h> 48 #include <linux/if_vlan.h> 49 #include <linux/bpf.h> 50 #include <linux/btf.h> 51 #include <net/sch_generic.h> 52 #include <net/cls_cgroup.h> 53 #include <net/dst_metadata.h> 54 #include <net/dst.h> 55 #include <net/sock_reuseport.h> 56 #include <net/busy_poll.h> 57 #include <net/tcp.h> 58 #include <net/xfrm.h> 59 #include <net/udp.h> 60 #include <linux/bpf_trace.h> 61 #include <net/xdp_sock.h> 62 #include <linux/inetdevice.h> 63 #include <net/inet_hashtables.h> 64 #include <net/inet6_hashtables.h> 65 #include <net/ip_fib.h> 66 #include <net/nexthop.h> 67 #include <net/flow.h> 68 #include <net/arp.h> 69 #include <net/ipv6.h> 70 #include <net/net_namespace.h> 71 #include <linux/seg6_local.h> 72 #include <net/seg6.h> 73 #include <net/seg6_local.h> 74 #include <net/lwtunnel.h> 75 #include <net/ipv6_stubs.h> 76 #include <net/bpf_sk_storage.h> 77 #include <net/transp_v6.h> 78 #include <linux/btf_ids.h> 79 #include <net/tls.h> 80 #include <net/xdp.h> 81 #include <net/mptcp.h> 82 83 static const struct bpf_func_proto * 84 bpf_sk_base_func_proto(enum bpf_func_id func_id); 85 86 int copy_bpf_fprog_from_user(struct sock_fprog *dst, sockptr_t src, int len) 87 { 88 if (in_compat_syscall()) { 89 struct compat_sock_fprog f32; 90 91 if (len != sizeof(f32)) 92 return -EINVAL; 93 if (copy_from_sockptr(&f32, src, sizeof(f32))) 94 return -EFAULT; 95 memset(dst, 0, sizeof(*dst)); 96 dst->len = f32.len; 97 dst->filter = compat_ptr(f32.filter); 98 } else { 99 if (len != sizeof(*dst)) 100 return -EINVAL; 101 if (copy_from_sockptr(dst, src, sizeof(*dst))) 102 return -EFAULT; 103 } 104 105 return 0; 106 } 107 EXPORT_SYMBOL_GPL(copy_bpf_fprog_from_user); 108 109 /** 110 * sk_filter_trim_cap - run a packet through a socket filter 111 * @sk: sock associated with &sk_buff 112 * @skb: buffer to filter 113 * @cap: limit on how short the eBPF program may trim the packet 114 * 115 * Run the eBPF program and then cut skb->data to correct size returned by 116 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller 117 * than pkt_len we keep whole skb->data. This is the socket level 118 * wrapper to bpf_prog_run. It returns 0 if the packet should 119 * be accepted or -EPERM if the packet should be tossed. 120 * 121 */ 122 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap) 123 { 124 int err; 125 struct sk_filter *filter; 126 127 /* 128 * If the skb was allocated from pfmemalloc reserves, only 129 * allow SOCK_MEMALLOC sockets to use it as this socket is 130 * helping free memory 131 */ 132 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) { 133 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP); 134 return -ENOMEM; 135 } 136 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb); 137 if (err) 138 return err; 139 140 err = security_sock_rcv_skb(sk, skb); 141 if (err) 142 return err; 143 144 rcu_read_lock(); 145 filter = rcu_dereference(sk->sk_filter); 146 if (filter) { 147 struct sock *save_sk = skb->sk; 148 unsigned int pkt_len; 149 150 skb->sk = sk; 151 pkt_len = bpf_prog_run_save_cb(filter->prog, skb); 152 skb->sk = save_sk; 153 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM; 154 } 155 rcu_read_unlock(); 156 157 return err; 158 } 159 EXPORT_SYMBOL(sk_filter_trim_cap); 160 161 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb) 162 { 163 return skb_get_poff(skb); 164 } 165 166 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x) 167 { 168 struct nlattr *nla; 169 170 if (skb_is_nonlinear(skb)) 171 return 0; 172 173 if (skb->len < sizeof(struct nlattr)) 174 return 0; 175 176 if (a > skb->len - sizeof(struct nlattr)) 177 return 0; 178 179 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x); 180 if (nla) 181 return (void *) nla - (void *) skb->data; 182 183 return 0; 184 } 185 186 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x) 187 { 188 struct nlattr *nla; 189 190 if (skb_is_nonlinear(skb)) 191 return 0; 192 193 if (skb->len < sizeof(struct nlattr)) 194 return 0; 195 196 if (a > skb->len - sizeof(struct nlattr)) 197 return 0; 198 199 nla = (struct nlattr *) &skb->data[a]; 200 if (nla->nla_len > skb->len - a) 201 return 0; 202 203 nla = nla_find_nested(nla, x); 204 if (nla) 205 return (void *) nla - (void *) skb->data; 206 207 return 0; 208 } 209 210 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *, 211 data, int, headlen, int, offset) 212 { 213 u8 tmp, *ptr; 214 const int len = sizeof(tmp); 215 216 if (offset >= 0) { 217 if (headlen - offset >= len) 218 return *(u8 *)(data + offset); 219 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 220 return tmp; 221 } else { 222 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 223 if (likely(ptr)) 224 return *(u8 *)ptr; 225 } 226 227 return -EFAULT; 228 } 229 230 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb, 231 int, offset) 232 { 233 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len, 234 offset); 235 } 236 237 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *, 238 data, int, headlen, int, offset) 239 { 240 u16 tmp, *ptr; 241 const int len = sizeof(tmp); 242 243 if (offset >= 0) { 244 if (headlen - offset >= len) 245 return get_unaligned_be16(data + offset); 246 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 247 return be16_to_cpu(tmp); 248 } else { 249 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 250 if (likely(ptr)) 251 return get_unaligned_be16(ptr); 252 } 253 254 return -EFAULT; 255 } 256 257 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb, 258 int, offset) 259 { 260 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len, 261 offset); 262 } 263 264 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *, 265 data, int, headlen, int, offset) 266 { 267 u32 tmp, *ptr; 268 const int len = sizeof(tmp); 269 270 if (likely(offset >= 0)) { 271 if (headlen - offset >= len) 272 return get_unaligned_be32(data + offset); 273 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 274 return be32_to_cpu(tmp); 275 } else { 276 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 277 if (likely(ptr)) 278 return get_unaligned_be32(ptr); 279 } 280 281 return -EFAULT; 282 } 283 284 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb, 285 int, offset) 286 { 287 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len, 288 offset); 289 } 290 291 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg, 292 struct bpf_insn *insn_buf) 293 { 294 struct bpf_insn *insn = insn_buf; 295 296 switch (skb_field) { 297 case SKF_AD_MARK: 298 BUILD_BUG_ON(sizeof_field(struct sk_buff, mark) != 4); 299 300 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, 301 offsetof(struct sk_buff, mark)); 302 break; 303 304 case SKF_AD_PKTTYPE: 305 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET); 306 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX); 307 #ifdef __BIG_ENDIAN_BITFIELD 308 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5); 309 #endif 310 break; 311 312 case SKF_AD_QUEUE: 313 BUILD_BUG_ON(sizeof_field(struct sk_buff, queue_mapping) != 2); 314 315 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 316 offsetof(struct sk_buff, queue_mapping)); 317 break; 318 319 case SKF_AD_VLAN_TAG: 320 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_tci) != 2); 321 322 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */ 323 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 324 offsetof(struct sk_buff, vlan_tci)); 325 break; 326 case SKF_AD_VLAN_TAG_PRESENT: 327 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET); 328 if (PKT_VLAN_PRESENT_BIT) 329 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT); 330 if (PKT_VLAN_PRESENT_BIT < 7) 331 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1); 332 break; 333 } 334 335 return insn - insn_buf; 336 } 337 338 static bool convert_bpf_extensions(struct sock_filter *fp, 339 struct bpf_insn **insnp) 340 { 341 struct bpf_insn *insn = *insnp; 342 u32 cnt; 343 344 switch (fp->k) { 345 case SKF_AD_OFF + SKF_AD_PROTOCOL: 346 BUILD_BUG_ON(sizeof_field(struct sk_buff, protocol) != 2); 347 348 /* A = *(u16 *) (CTX + offsetof(protocol)) */ 349 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 350 offsetof(struct sk_buff, protocol)); 351 /* A = ntohs(A) [emitting a nop or swap16] */ 352 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 353 break; 354 355 case SKF_AD_OFF + SKF_AD_PKTTYPE: 356 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn); 357 insn += cnt - 1; 358 break; 359 360 case SKF_AD_OFF + SKF_AD_IFINDEX: 361 case SKF_AD_OFF + SKF_AD_HATYPE: 362 BUILD_BUG_ON(sizeof_field(struct net_device, ifindex) != 4); 363 BUILD_BUG_ON(sizeof_field(struct net_device, type) != 2); 364 365 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 366 BPF_REG_TMP, BPF_REG_CTX, 367 offsetof(struct sk_buff, dev)); 368 /* if (tmp != 0) goto pc + 1 */ 369 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1); 370 *insn++ = BPF_EXIT_INSN(); 371 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) 372 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP, 373 offsetof(struct net_device, ifindex)); 374 else 375 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP, 376 offsetof(struct net_device, type)); 377 break; 378 379 case SKF_AD_OFF + SKF_AD_MARK: 380 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn); 381 insn += cnt - 1; 382 break; 383 384 case SKF_AD_OFF + SKF_AD_RXHASH: 385 BUILD_BUG_ON(sizeof_field(struct sk_buff, hash) != 4); 386 387 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, 388 offsetof(struct sk_buff, hash)); 389 break; 390 391 case SKF_AD_OFF + SKF_AD_QUEUE: 392 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn); 393 insn += cnt - 1; 394 break; 395 396 case SKF_AD_OFF + SKF_AD_VLAN_TAG: 397 cnt = convert_skb_access(SKF_AD_VLAN_TAG, 398 BPF_REG_A, BPF_REG_CTX, insn); 399 insn += cnt - 1; 400 break; 401 402 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT: 403 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT, 404 BPF_REG_A, BPF_REG_CTX, insn); 405 insn += cnt - 1; 406 break; 407 408 case SKF_AD_OFF + SKF_AD_VLAN_TPID: 409 BUILD_BUG_ON(sizeof_field(struct sk_buff, vlan_proto) != 2); 410 411 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */ 412 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 413 offsetof(struct sk_buff, vlan_proto)); 414 /* A = ntohs(A) [emitting a nop or swap16] */ 415 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 416 break; 417 418 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 419 case SKF_AD_OFF + SKF_AD_NLATTR: 420 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 421 case SKF_AD_OFF + SKF_AD_CPU: 422 case SKF_AD_OFF + SKF_AD_RANDOM: 423 /* arg1 = CTX */ 424 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 425 /* arg2 = A */ 426 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A); 427 /* arg3 = X */ 428 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X); 429 /* Emit call(arg1=CTX, arg2=A, arg3=X) */ 430 switch (fp->k) { 431 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 432 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset); 433 break; 434 case SKF_AD_OFF + SKF_AD_NLATTR: 435 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr); 436 break; 437 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 438 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest); 439 break; 440 case SKF_AD_OFF + SKF_AD_CPU: 441 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id); 442 break; 443 case SKF_AD_OFF + SKF_AD_RANDOM: 444 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32); 445 bpf_user_rnd_init_once(); 446 break; 447 } 448 break; 449 450 case SKF_AD_OFF + SKF_AD_ALU_XOR_X: 451 /* A ^= X */ 452 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X); 453 break; 454 455 default: 456 /* This is just a dummy call to avoid letting the compiler 457 * evict __bpf_call_base() as an optimization. Placed here 458 * where no-one bothers. 459 */ 460 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0); 461 return false; 462 } 463 464 *insnp = insn; 465 return true; 466 } 467 468 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp) 469 { 470 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS); 471 int size = bpf_size_to_bytes(BPF_SIZE(fp->code)); 472 bool endian = BPF_SIZE(fp->code) == BPF_H || 473 BPF_SIZE(fp->code) == BPF_W; 474 bool indirect = BPF_MODE(fp->code) == BPF_IND; 475 const int ip_align = NET_IP_ALIGN; 476 struct bpf_insn *insn = *insnp; 477 int offset = fp->k; 478 479 if (!indirect && 480 ((unaligned_ok && offset >= 0) || 481 (!unaligned_ok && offset >= 0 && 482 offset + ip_align >= 0 && 483 offset + ip_align % size == 0))) { 484 bool ldx_off_ok = offset <= S16_MAX; 485 486 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H); 487 if (offset) 488 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset); 489 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP, 490 size, 2 + endian + (!ldx_off_ok * 2)); 491 if (ldx_off_ok) { 492 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 493 BPF_REG_D, offset); 494 } else { 495 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D); 496 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset); 497 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 498 BPF_REG_TMP, 0); 499 } 500 if (endian) 501 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8); 502 *insn++ = BPF_JMP_A(8); 503 } 504 505 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 506 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D); 507 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H); 508 if (!indirect) { 509 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset); 510 } else { 511 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X); 512 if (fp->k) 513 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset); 514 } 515 516 switch (BPF_SIZE(fp->code)) { 517 case BPF_B: 518 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8); 519 break; 520 case BPF_H: 521 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16); 522 break; 523 case BPF_W: 524 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32); 525 break; 526 default: 527 return false; 528 } 529 530 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2); 531 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 532 *insn = BPF_EXIT_INSN(); 533 534 *insnp = insn; 535 return true; 536 } 537 538 /** 539 * bpf_convert_filter - convert filter program 540 * @prog: the user passed filter program 541 * @len: the length of the user passed filter program 542 * @new_prog: allocated 'struct bpf_prog' or NULL 543 * @new_len: pointer to store length of converted program 544 * @seen_ld_abs: bool whether we've seen ld_abs/ind 545 * 546 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn' 547 * style extended BPF (eBPF). 548 * Conversion workflow: 549 * 550 * 1) First pass for calculating the new program length: 551 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs) 552 * 553 * 2) 2nd pass to remap in two passes: 1st pass finds new 554 * jump offsets, 2nd pass remapping: 555 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs) 556 */ 557 static int bpf_convert_filter(struct sock_filter *prog, int len, 558 struct bpf_prog *new_prog, int *new_len, 559 bool *seen_ld_abs) 560 { 561 int new_flen = 0, pass = 0, target, i, stack_off; 562 struct bpf_insn *new_insn, *first_insn = NULL; 563 struct sock_filter *fp; 564 int *addrs = NULL; 565 u8 bpf_src; 566 567 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK); 568 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); 569 570 if (len <= 0 || len > BPF_MAXINSNS) 571 return -EINVAL; 572 573 if (new_prog) { 574 first_insn = new_prog->insnsi; 575 addrs = kcalloc(len, sizeof(*addrs), 576 GFP_KERNEL | __GFP_NOWARN); 577 if (!addrs) 578 return -ENOMEM; 579 } 580 581 do_pass: 582 new_insn = first_insn; 583 fp = prog; 584 585 /* Classic BPF related prologue emission. */ 586 if (new_prog) { 587 /* Classic BPF expects A and X to be reset first. These need 588 * to be guaranteed to be the first two instructions. 589 */ 590 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 591 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X); 592 593 /* All programs must keep CTX in callee saved BPF_REG_CTX. 594 * In eBPF case it's done by the compiler, here we need to 595 * do this ourself. Initial CTX is present in BPF_REG_ARG1. 596 */ 597 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1); 598 if (*seen_ld_abs) { 599 /* For packet access in classic BPF, cache skb->data 600 * in callee-saved BPF R8 and skb->len - skb->data_len 601 * (headlen) in BPF R9. Since classic BPF is read-only 602 * on CTX, we only need to cache it once. 603 */ 604 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 605 BPF_REG_D, BPF_REG_CTX, 606 offsetof(struct sk_buff, data)); 607 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX, 608 offsetof(struct sk_buff, len)); 609 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX, 610 offsetof(struct sk_buff, data_len)); 611 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP); 612 } 613 } else { 614 new_insn += 3; 615 } 616 617 for (i = 0; i < len; fp++, i++) { 618 struct bpf_insn tmp_insns[32] = { }; 619 struct bpf_insn *insn = tmp_insns; 620 621 if (addrs) 622 addrs[i] = new_insn - first_insn; 623 624 switch (fp->code) { 625 /* All arithmetic insns and skb loads map as-is. */ 626 case BPF_ALU | BPF_ADD | BPF_X: 627 case BPF_ALU | BPF_ADD | BPF_K: 628 case BPF_ALU | BPF_SUB | BPF_X: 629 case BPF_ALU | BPF_SUB | BPF_K: 630 case BPF_ALU | BPF_AND | BPF_X: 631 case BPF_ALU | BPF_AND | BPF_K: 632 case BPF_ALU | BPF_OR | BPF_X: 633 case BPF_ALU | BPF_OR | BPF_K: 634 case BPF_ALU | BPF_LSH | BPF_X: 635 case BPF_ALU | BPF_LSH | BPF_K: 636 case BPF_ALU | BPF_RSH | BPF_X: 637 case BPF_ALU | BPF_RSH | BPF_K: 638 case BPF_ALU | BPF_XOR | BPF_X: 639 case BPF_ALU | BPF_XOR | BPF_K: 640 case BPF_ALU | BPF_MUL | BPF_X: 641 case BPF_ALU | BPF_MUL | BPF_K: 642 case BPF_ALU | BPF_DIV | BPF_X: 643 case BPF_ALU | BPF_DIV | BPF_K: 644 case BPF_ALU | BPF_MOD | BPF_X: 645 case BPF_ALU | BPF_MOD | BPF_K: 646 case BPF_ALU | BPF_NEG: 647 case BPF_LD | BPF_ABS | BPF_W: 648 case BPF_LD | BPF_ABS | BPF_H: 649 case BPF_LD | BPF_ABS | BPF_B: 650 case BPF_LD | BPF_IND | BPF_W: 651 case BPF_LD | BPF_IND | BPF_H: 652 case BPF_LD | BPF_IND | BPF_B: 653 /* Check for overloaded BPF extension and 654 * directly convert it if found, otherwise 655 * just move on with mapping. 656 */ 657 if (BPF_CLASS(fp->code) == BPF_LD && 658 BPF_MODE(fp->code) == BPF_ABS && 659 convert_bpf_extensions(fp, &insn)) 660 break; 661 if (BPF_CLASS(fp->code) == BPF_LD && 662 convert_bpf_ld_abs(fp, &insn)) { 663 *seen_ld_abs = true; 664 break; 665 } 666 667 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) || 668 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) { 669 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X); 670 /* Error with exception code on div/mod by 0. 671 * For cBPF programs, this was always return 0. 672 */ 673 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2); 674 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 675 *insn++ = BPF_EXIT_INSN(); 676 } 677 678 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k); 679 break; 680 681 /* Jump transformation cannot use BPF block macros 682 * everywhere as offset calculation and target updates 683 * require a bit more work than the rest, i.e. jump 684 * opcodes map as-is, but offsets need adjustment. 685 */ 686 687 #define BPF_EMIT_JMP \ 688 do { \ 689 const s32 off_min = S16_MIN, off_max = S16_MAX; \ 690 s32 off; \ 691 \ 692 if (target >= len || target < 0) \ 693 goto err; \ 694 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \ 695 /* Adjust pc relative offset for 2nd or 3rd insn. */ \ 696 off -= insn - tmp_insns; \ 697 /* Reject anything not fitting into insn->off. */ \ 698 if (off < off_min || off > off_max) \ 699 goto err; \ 700 insn->off = off; \ 701 } while (0) 702 703 case BPF_JMP | BPF_JA: 704 target = i + fp->k + 1; 705 insn->code = fp->code; 706 BPF_EMIT_JMP; 707 break; 708 709 case BPF_JMP | BPF_JEQ | BPF_K: 710 case BPF_JMP | BPF_JEQ | BPF_X: 711 case BPF_JMP | BPF_JSET | BPF_K: 712 case BPF_JMP | BPF_JSET | BPF_X: 713 case BPF_JMP | BPF_JGT | BPF_K: 714 case BPF_JMP | BPF_JGT | BPF_X: 715 case BPF_JMP | BPF_JGE | BPF_K: 716 case BPF_JMP | BPF_JGE | BPF_X: 717 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) { 718 /* BPF immediates are signed, zero extend 719 * immediate into tmp register and use it 720 * in compare insn. 721 */ 722 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k); 723 724 insn->dst_reg = BPF_REG_A; 725 insn->src_reg = BPF_REG_TMP; 726 bpf_src = BPF_X; 727 } else { 728 insn->dst_reg = BPF_REG_A; 729 insn->imm = fp->k; 730 bpf_src = BPF_SRC(fp->code); 731 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0; 732 } 733 734 /* Common case where 'jump_false' is next insn. */ 735 if (fp->jf == 0) { 736 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 737 target = i + fp->jt + 1; 738 BPF_EMIT_JMP; 739 break; 740 } 741 742 /* Convert some jumps when 'jump_true' is next insn. */ 743 if (fp->jt == 0) { 744 switch (BPF_OP(fp->code)) { 745 case BPF_JEQ: 746 insn->code = BPF_JMP | BPF_JNE | bpf_src; 747 break; 748 case BPF_JGT: 749 insn->code = BPF_JMP | BPF_JLE | bpf_src; 750 break; 751 case BPF_JGE: 752 insn->code = BPF_JMP | BPF_JLT | bpf_src; 753 break; 754 default: 755 goto jmp_rest; 756 } 757 758 target = i + fp->jf + 1; 759 BPF_EMIT_JMP; 760 break; 761 } 762 jmp_rest: 763 /* Other jumps are mapped into two insns: Jxx and JA. */ 764 target = i + fp->jt + 1; 765 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 766 BPF_EMIT_JMP; 767 insn++; 768 769 insn->code = BPF_JMP | BPF_JA; 770 target = i + fp->jf + 1; 771 BPF_EMIT_JMP; 772 break; 773 774 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */ 775 case BPF_LDX | BPF_MSH | BPF_B: { 776 struct sock_filter tmp = { 777 .code = BPF_LD | BPF_ABS | BPF_B, 778 .k = fp->k, 779 }; 780 781 *seen_ld_abs = true; 782 783 /* X = A */ 784 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 785 /* A = BPF_R0 = *(u8 *) (skb->data + K) */ 786 convert_bpf_ld_abs(&tmp, &insn); 787 insn++; 788 /* A &= 0xf */ 789 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf); 790 /* A <<= 2 */ 791 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2); 792 /* tmp = X */ 793 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X); 794 /* X = A */ 795 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 796 /* A = tmp */ 797 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP); 798 break; 799 } 800 /* RET_K is remaped into 2 insns. RET_A case doesn't need an 801 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A. 802 */ 803 case BPF_RET | BPF_A: 804 case BPF_RET | BPF_K: 805 if (BPF_RVAL(fp->code) == BPF_K) 806 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0, 807 0, fp->k); 808 *insn = BPF_EXIT_INSN(); 809 break; 810 811 /* Store to stack. */ 812 case BPF_ST: 813 case BPF_STX: 814 stack_off = fp->k * 4 + 4; 815 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) == 816 BPF_ST ? BPF_REG_A : BPF_REG_X, 817 -stack_off); 818 /* check_load_and_stores() verifies that classic BPF can 819 * load from stack only after write, so tracking 820 * stack_depth for ST|STX insns is enough 821 */ 822 if (new_prog && new_prog->aux->stack_depth < stack_off) 823 new_prog->aux->stack_depth = stack_off; 824 break; 825 826 /* Load from stack. */ 827 case BPF_LD | BPF_MEM: 828 case BPF_LDX | BPF_MEM: 829 stack_off = fp->k * 4 + 4; 830 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 831 BPF_REG_A : BPF_REG_X, BPF_REG_FP, 832 -stack_off); 833 break; 834 835 /* A = K or X = K */ 836 case BPF_LD | BPF_IMM: 837 case BPF_LDX | BPF_IMM: 838 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ? 839 BPF_REG_A : BPF_REG_X, fp->k); 840 break; 841 842 /* X = A */ 843 case BPF_MISC | BPF_TAX: 844 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 845 break; 846 847 /* A = X */ 848 case BPF_MISC | BPF_TXA: 849 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X); 850 break; 851 852 /* A = skb->len or X = skb->len */ 853 case BPF_LD | BPF_W | BPF_LEN: 854 case BPF_LDX | BPF_W | BPF_LEN: 855 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 856 BPF_REG_A : BPF_REG_X, BPF_REG_CTX, 857 offsetof(struct sk_buff, len)); 858 break; 859 860 /* Access seccomp_data fields. */ 861 case BPF_LDX | BPF_ABS | BPF_W: 862 /* A = *(u32 *) (ctx + K) */ 863 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k); 864 break; 865 866 /* Unknown instruction. */ 867 default: 868 goto err; 869 } 870 871 insn++; 872 if (new_prog) 873 memcpy(new_insn, tmp_insns, 874 sizeof(*insn) * (insn - tmp_insns)); 875 new_insn += insn - tmp_insns; 876 } 877 878 if (!new_prog) { 879 /* Only calculating new length. */ 880 *new_len = new_insn - first_insn; 881 if (*seen_ld_abs) 882 *new_len += 4; /* Prologue bits. */ 883 return 0; 884 } 885 886 pass++; 887 if (new_flen != new_insn - first_insn) { 888 new_flen = new_insn - first_insn; 889 if (pass > 2) 890 goto err; 891 goto do_pass; 892 } 893 894 kfree(addrs); 895 BUG_ON(*new_len != new_flen); 896 return 0; 897 err: 898 kfree(addrs); 899 return -EINVAL; 900 } 901 902 /* Security: 903 * 904 * As we dont want to clear mem[] array for each packet going through 905 * __bpf_prog_run(), we check that filter loaded by user never try to read 906 * a cell if not previously written, and we check all branches to be sure 907 * a malicious user doesn't try to abuse us. 908 */ 909 static int check_load_and_stores(const struct sock_filter *filter, int flen) 910 { 911 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */ 912 int pc, ret = 0; 913 914 BUILD_BUG_ON(BPF_MEMWORDS > 16); 915 916 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL); 917 if (!masks) 918 return -ENOMEM; 919 920 memset(masks, 0xff, flen * sizeof(*masks)); 921 922 for (pc = 0; pc < flen; pc++) { 923 memvalid &= masks[pc]; 924 925 switch (filter[pc].code) { 926 case BPF_ST: 927 case BPF_STX: 928 memvalid |= (1 << filter[pc].k); 929 break; 930 case BPF_LD | BPF_MEM: 931 case BPF_LDX | BPF_MEM: 932 if (!(memvalid & (1 << filter[pc].k))) { 933 ret = -EINVAL; 934 goto error; 935 } 936 break; 937 case BPF_JMP | BPF_JA: 938 /* A jump must set masks on target */ 939 masks[pc + 1 + filter[pc].k] &= memvalid; 940 memvalid = ~0; 941 break; 942 case BPF_JMP | BPF_JEQ | BPF_K: 943 case BPF_JMP | BPF_JEQ | BPF_X: 944 case BPF_JMP | BPF_JGE | BPF_K: 945 case BPF_JMP | BPF_JGE | BPF_X: 946 case BPF_JMP | BPF_JGT | BPF_K: 947 case BPF_JMP | BPF_JGT | BPF_X: 948 case BPF_JMP | BPF_JSET | BPF_K: 949 case BPF_JMP | BPF_JSET | BPF_X: 950 /* A jump must set masks on targets */ 951 masks[pc + 1 + filter[pc].jt] &= memvalid; 952 masks[pc + 1 + filter[pc].jf] &= memvalid; 953 memvalid = ~0; 954 break; 955 } 956 } 957 error: 958 kfree(masks); 959 return ret; 960 } 961 962 static bool chk_code_allowed(u16 code_to_probe) 963 { 964 static const bool codes[] = { 965 /* 32 bit ALU operations */ 966 [BPF_ALU | BPF_ADD | BPF_K] = true, 967 [BPF_ALU | BPF_ADD | BPF_X] = true, 968 [BPF_ALU | BPF_SUB | BPF_K] = true, 969 [BPF_ALU | BPF_SUB | BPF_X] = true, 970 [BPF_ALU | BPF_MUL | BPF_K] = true, 971 [BPF_ALU | BPF_MUL | BPF_X] = true, 972 [BPF_ALU | BPF_DIV | BPF_K] = true, 973 [BPF_ALU | BPF_DIV | BPF_X] = true, 974 [BPF_ALU | BPF_MOD | BPF_K] = true, 975 [BPF_ALU | BPF_MOD | BPF_X] = true, 976 [BPF_ALU | BPF_AND | BPF_K] = true, 977 [BPF_ALU | BPF_AND | BPF_X] = true, 978 [BPF_ALU | BPF_OR | BPF_K] = true, 979 [BPF_ALU | BPF_OR | BPF_X] = true, 980 [BPF_ALU | BPF_XOR | BPF_K] = true, 981 [BPF_ALU | BPF_XOR | BPF_X] = true, 982 [BPF_ALU | BPF_LSH | BPF_K] = true, 983 [BPF_ALU | BPF_LSH | BPF_X] = true, 984 [BPF_ALU | BPF_RSH | BPF_K] = true, 985 [BPF_ALU | BPF_RSH | BPF_X] = true, 986 [BPF_ALU | BPF_NEG] = true, 987 /* Load instructions */ 988 [BPF_LD | BPF_W | BPF_ABS] = true, 989 [BPF_LD | BPF_H | BPF_ABS] = true, 990 [BPF_LD | BPF_B | BPF_ABS] = true, 991 [BPF_LD | BPF_W | BPF_LEN] = true, 992 [BPF_LD | BPF_W | BPF_IND] = true, 993 [BPF_LD | BPF_H | BPF_IND] = true, 994 [BPF_LD | BPF_B | BPF_IND] = true, 995 [BPF_LD | BPF_IMM] = true, 996 [BPF_LD | BPF_MEM] = true, 997 [BPF_LDX | BPF_W | BPF_LEN] = true, 998 [BPF_LDX | BPF_B | BPF_MSH] = true, 999 [BPF_LDX | BPF_IMM] = true, 1000 [BPF_LDX | BPF_MEM] = true, 1001 /* Store instructions */ 1002 [BPF_ST] = true, 1003 [BPF_STX] = true, 1004 /* Misc instructions */ 1005 [BPF_MISC | BPF_TAX] = true, 1006 [BPF_MISC | BPF_TXA] = true, 1007 /* Return instructions */ 1008 [BPF_RET | BPF_K] = true, 1009 [BPF_RET | BPF_A] = true, 1010 /* Jump instructions */ 1011 [BPF_JMP | BPF_JA] = true, 1012 [BPF_JMP | BPF_JEQ | BPF_K] = true, 1013 [BPF_JMP | BPF_JEQ | BPF_X] = true, 1014 [BPF_JMP | BPF_JGE | BPF_K] = true, 1015 [BPF_JMP | BPF_JGE | BPF_X] = true, 1016 [BPF_JMP | BPF_JGT | BPF_K] = true, 1017 [BPF_JMP | BPF_JGT | BPF_X] = true, 1018 [BPF_JMP | BPF_JSET | BPF_K] = true, 1019 [BPF_JMP | BPF_JSET | BPF_X] = true, 1020 }; 1021 1022 if (code_to_probe >= ARRAY_SIZE(codes)) 1023 return false; 1024 1025 return codes[code_to_probe]; 1026 } 1027 1028 static bool bpf_check_basics_ok(const struct sock_filter *filter, 1029 unsigned int flen) 1030 { 1031 if (filter == NULL) 1032 return false; 1033 if (flen == 0 || flen > BPF_MAXINSNS) 1034 return false; 1035 1036 return true; 1037 } 1038 1039 /** 1040 * bpf_check_classic - verify socket filter code 1041 * @filter: filter to verify 1042 * @flen: length of filter 1043 * 1044 * Check the user's filter code. If we let some ugly 1045 * filter code slip through kaboom! The filter must contain 1046 * no references or jumps that are out of range, no illegal 1047 * instructions, and must end with a RET instruction. 1048 * 1049 * All jumps are forward as they are not signed. 1050 * 1051 * Returns 0 if the rule set is legal or -EINVAL if not. 1052 */ 1053 static int bpf_check_classic(const struct sock_filter *filter, 1054 unsigned int flen) 1055 { 1056 bool anc_found; 1057 int pc; 1058 1059 /* Check the filter code now */ 1060 for (pc = 0; pc < flen; pc++) { 1061 const struct sock_filter *ftest = &filter[pc]; 1062 1063 /* May we actually operate on this code? */ 1064 if (!chk_code_allowed(ftest->code)) 1065 return -EINVAL; 1066 1067 /* Some instructions need special checks */ 1068 switch (ftest->code) { 1069 case BPF_ALU | BPF_DIV | BPF_K: 1070 case BPF_ALU | BPF_MOD | BPF_K: 1071 /* Check for division by zero */ 1072 if (ftest->k == 0) 1073 return -EINVAL; 1074 break; 1075 case BPF_ALU | BPF_LSH | BPF_K: 1076 case BPF_ALU | BPF_RSH | BPF_K: 1077 if (ftest->k >= 32) 1078 return -EINVAL; 1079 break; 1080 case BPF_LD | BPF_MEM: 1081 case BPF_LDX | BPF_MEM: 1082 case BPF_ST: 1083 case BPF_STX: 1084 /* Check for invalid memory addresses */ 1085 if (ftest->k >= BPF_MEMWORDS) 1086 return -EINVAL; 1087 break; 1088 case BPF_JMP | BPF_JA: 1089 /* Note, the large ftest->k might cause loops. 1090 * Compare this with conditional jumps below, 1091 * where offsets are limited. --ANK (981016) 1092 */ 1093 if (ftest->k >= (unsigned int)(flen - pc - 1)) 1094 return -EINVAL; 1095 break; 1096 case BPF_JMP | BPF_JEQ | BPF_K: 1097 case BPF_JMP | BPF_JEQ | BPF_X: 1098 case BPF_JMP | BPF_JGE | BPF_K: 1099 case BPF_JMP | BPF_JGE | BPF_X: 1100 case BPF_JMP | BPF_JGT | BPF_K: 1101 case BPF_JMP | BPF_JGT | BPF_X: 1102 case BPF_JMP | BPF_JSET | BPF_K: 1103 case BPF_JMP | BPF_JSET | BPF_X: 1104 /* Both conditionals must be safe */ 1105 if (pc + ftest->jt + 1 >= flen || 1106 pc + ftest->jf + 1 >= flen) 1107 return -EINVAL; 1108 break; 1109 case BPF_LD | BPF_W | BPF_ABS: 1110 case BPF_LD | BPF_H | BPF_ABS: 1111 case BPF_LD | BPF_B | BPF_ABS: 1112 anc_found = false; 1113 if (bpf_anc_helper(ftest) & BPF_ANC) 1114 anc_found = true; 1115 /* Ancillary operation unknown or unsupported */ 1116 if (anc_found == false && ftest->k >= SKF_AD_OFF) 1117 return -EINVAL; 1118 } 1119 } 1120 1121 /* Last instruction must be a RET code */ 1122 switch (filter[flen - 1].code) { 1123 case BPF_RET | BPF_K: 1124 case BPF_RET | BPF_A: 1125 return check_load_and_stores(filter, flen); 1126 } 1127 1128 return -EINVAL; 1129 } 1130 1131 static int bpf_prog_store_orig_filter(struct bpf_prog *fp, 1132 const struct sock_fprog *fprog) 1133 { 1134 unsigned int fsize = bpf_classic_proglen(fprog); 1135 struct sock_fprog_kern *fkprog; 1136 1137 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL); 1138 if (!fp->orig_prog) 1139 return -ENOMEM; 1140 1141 fkprog = fp->orig_prog; 1142 fkprog->len = fprog->len; 1143 1144 fkprog->filter = kmemdup(fp->insns, fsize, 1145 GFP_KERNEL | __GFP_NOWARN); 1146 if (!fkprog->filter) { 1147 kfree(fp->orig_prog); 1148 return -ENOMEM; 1149 } 1150 1151 return 0; 1152 } 1153 1154 static void bpf_release_orig_filter(struct bpf_prog *fp) 1155 { 1156 struct sock_fprog_kern *fprog = fp->orig_prog; 1157 1158 if (fprog) { 1159 kfree(fprog->filter); 1160 kfree(fprog); 1161 } 1162 } 1163 1164 static void __bpf_prog_release(struct bpf_prog *prog) 1165 { 1166 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) { 1167 bpf_prog_put(prog); 1168 } else { 1169 bpf_release_orig_filter(prog); 1170 bpf_prog_free(prog); 1171 } 1172 } 1173 1174 static void __sk_filter_release(struct sk_filter *fp) 1175 { 1176 __bpf_prog_release(fp->prog); 1177 kfree(fp); 1178 } 1179 1180 /** 1181 * sk_filter_release_rcu - Release a socket filter by rcu_head 1182 * @rcu: rcu_head that contains the sk_filter to free 1183 */ 1184 static void sk_filter_release_rcu(struct rcu_head *rcu) 1185 { 1186 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); 1187 1188 __sk_filter_release(fp); 1189 } 1190 1191 /** 1192 * sk_filter_release - release a socket filter 1193 * @fp: filter to remove 1194 * 1195 * Remove a filter from a socket and release its resources. 1196 */ 1197 static void sk_filter_release(struct sk_filter *fp) 1198 { 1199 if (refcount_dec_and_test(&fp->refcnt)) 1200 call_rcu(&fp->rcu, sk_filter_release_rcu); 1201 } 1202 1203 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1204 { 1205 u32 filter_size = bpf_prog_size(fp->prog->len); 1206 1207 atomic_sub(filter_size, &sk->sk_omem_alloc); 1208 sk_filter_release(fp); 1209 } 1210 1211 /* try to charge the socket memory if there is space available 1212 * return true on success 1213 */ 1214 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1215 { 1216 u32 filter_size = bpf_prog_size(fp->prog->len); 1217 1218 /* same check as in sock_kmalloc() */ 1219 if (filter_size <= sysctl_optmem_max && 1220 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) { 1221 atomic_add(filter_size, &sk->sk_omem_alloc); 1222 return true; 1223 } 1224 return false; 1225 } 1226 1227 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1228 { 1229 if (!refcount_inc_not_zero(&fp->refcnt)) 1230 return false; 1231 1232 if (!__sk_filter_charge(sk, fp)) { 1233 sk_filter_release(fp); 1234 return false; 1235 } 1236 return true; 1237 } 1238 1239 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp) 1240 { 1241 struct sock_filter *old_prog; 1242 struct bpf_prog *old_fp; 1243 int err, new_len, old_len = fp->len; 1244 bool seen_ld_abs = false; 1245 1246 /* We are free to overwrite insns et al right here as it won't be used at 1247 * this point in time anymore internally after the migration to the eBPF 1248 * instruction 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 eBPF translation 1340 * 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 > INT_MAX)) 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 | MEM_RDONLY, 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 > INT_MAX)) 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 | PTR_MAYBE_NULL | MEM_RDONLY, 2022 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 2023 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 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_clear_tstamp(skb); 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_clear_tstamp(skb); 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_clear_tstamp(skb); 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 | PTR_MAYBE_NULL | MEM_RDONLY, 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 if (unlikely(len == 0)) 2715 return 0; 2716 2717 /* First find the starting scatterlist element */ 2718 i = msg->sg.start; 2719 do { 2720 offset += l; 2721 l = sk_msg_elem(msg, i)->length; 2722 2723 if (start < offset + l) 2724 break; 2725 sk_msg_iter_var_next(i); 2726 } while (i != msg->sg.end); 2727 2728 if (start >= offset + l) 2729 return -EINVAL; 2730 2731 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2732 2733 /* If no space available will fallback to copy, we need at 2734 * least one scatterlist elem available to push data into 2735 * when start aligns to the beginning of an element or two 2736 * when it falls inside an element. We handle the start equals 2737 * offset case because its the common case for inserting a 2738 * header. 2739 */ 2740 if (!space || (space == 1 && start != offset)) 2741 copy = msg->sg.data[i].length; 2742 2743 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2744 get_order(copy + len)); 2745 if (unlikely(!page)) 2746 return -ENOMEM; 2747 2748 if (copy) { 2749 int front, back; 2750 2751 raw = page_address(page); 2752 2753 psge = sk_msg_elem(msg, i); 2754 front = start - offset; 2755 back = psge->length - front; 2756 from = sg_virt(psge); 2757 2758 if (front) 2759 memcpy(raw, from, front); 2760 2761 if (back) { 2762 from += front; 2763 to = raw + front + len; 2764 2765 memcpy(to, from, back); 2766 } 2767 2768 put_page(sg_page(psge)); 2769 } else if (start - offset) { 2770 psge = sk_msg_elem(msg, i); 2771 rsge = sk_msg_elem_cpy(msg, i); 2772 2773 psge->length = start - offset; 2774 rsge.length -= psge->length; 2775 rsge.offset += start; 2776 2777 sk_msg_iter_var_next(i); 2778 sg_unmark_end(psge); 2779 sg_unmark_end(&rsge); 2780 sk_msg_iter_next(msg, end); 2781 } 2782 2783 /* Slot(s) to place newly allocated data */ 2784 new = i; 2785 2786 /* Shift one or two slots as needed */ 2787 if (!copy) { 2788 sge = sk_msg_elem_cpy(msg, i); 2789 2790 sk_msg_iter_var_next(i); 2791 sg_unmark_end(&sge); 2792 sk_msg_iter_next(msg, end); 2793 2794 nsge = sk_msg_elem_cpy(msg, i); 2795 if (rsge.length) { 2796 sk_msg_iter_var_next(i); 2797 nnsge = sk_msg_elem_cpy(msg, i); 2798 } 2799 2800 while (i != msg->sg.end) { 2801 msg->sg.data[i] = sge; 2802 sge = nsge; 2803 sk_msg_iter_var_next(i); 2804 if (rsge.length) { 2805 nsge = nnsge; 2806 nnsge = sk_msg_elem_cpy(msg, i); 2807 } else { 2808 nsge = sk_msg_elem_cpy(msg, i); 2809 } 2810 } 2811 } 2812 2813 /* Place newly allocated data buffer */ 2814 sk_mem_charge(msg->sk, len); 2815 msg->sg.size += len; 2816 __clear_bit(new, msg->sg.copy); 2817 sg_set_page(&msg->sg.data[new], page, len + copy, 0); 2818 if (rsge.length) { 2819 get_page(sg_page(&rsge)); 2820 sk_msg_iter_var_next(new); 2821 msg->sg.data[new] = rsge; 2822 } 2823 2824 sk_msg_compute_data_pointers(msg); 2825 return 0; 2826 } 2827 2828 static const struct bpf_func_proto bpf_msg_push_data_proto = { 2829 .func = bpf_msg_push_data, 2830 .gpl_only = false, 2831 .ret_type = RET_INTEGER, 2832 .arg1_type = ARG_PTR_TO_CTX, 2833 .arg2_type = ARG_ANYTHING, 2834 .arg3_type = ARG_ANYTHING, 2835 .arg4_type = ARG_ANYTHING, 2836 }; 2837 2838 static void sk_msg_shift_left(struct sk_msg *msg, int i) 2839 { 2840 int prev; 2841 2842 do { 2843 prev = i; 2844 sk_msg_iter_var_next(i); 2845 msg->sg.data[prev] = msg->sg.data[i]; 2846 } while (i != msg->sg.end); 2847 2848 sk_msg_iter_prev(msg, end); 2849 } 2850 2851 static void sk_msg_shift_right(struct sk_msg *msg, int i) 2852 { 2853 struct scatterlist tmp, sge; 2854 2855 sk_msg_iter_next(msg, end); 2856 sge = sk_msg_elem_cpy(msg, i); 2857 sk_msg_iter_var_next(i); 2858 tmp = sk_msg_elem_cpy(msg, i); 2859 2860 while (i != msg->sg.end) { 2861 msg->sg.data[i] = sge; 2862 sk_msg_iter_var_next(i); 2863 sge = tmp; 2864 tmp = sk_msg_elem_cpy(msg, i); 2865 } 2866 } 2867 2868 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start, 2869 u32, len, u64, flags) 2870 { 2871 u32 i = 0, l = 0, space, offset = 0; 2872 u64 last = start + len; 2873 int pop; 2874 2875 if (unlikely(flags)) 2876 return -EINVAL; 2877 2878 /* First find the starting scatterlist element */ 2879 i = msg->sg.start; 2880 do { 2881 offset += l; 2882 l = sk_msg_elem(msg, i)->length; 2883 2884 if (start < offset + l) 2885 break; 2886 sk_msg_iter_var_next(i); 2887 } while (i != msg->sg.end); 2888 2889 /* Bounds checks: start and pop must be inside message */ 2890 if (start >= offset + l || last >= msg->sg.size) 2891 return -EINVAL; 2892 2893 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2894 2895 pop = len; 2896 /* --------------| offset 2897 * -| start |-------- len -------| 2898 * 2899 * |----- a ----|-------- pop -------|----- b ----| 2900 * |______________________________________________| length 2901 * 2902 * 2903 * a: region at front of scatter element to save 2904 * b: region at back of scatter element to save when length > A + pop 2905 * pop: region to pop from element, same as input 'pop' here will be 2906 * decremented below per iteration. 2907 * 2908 * Two top-level cases to handle when start != offset, first B is non 2909 * zero and second B is zero corresponding to when a pop includes more 2910 * than one element. 2911 * 2912 * Then if B is non-zero AND there is no space allocate space and 2913 * compact A, B regions into page. If there is space shift ring to 2914 * the rigth free'ing the next element in ring to place B, leaving 2915 * A untouched except to reduce length. 2916 */ 2917 if (start != offset) { 2918 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i); 2919 int a = start; 2920 int b = sge->length - pop - a; 2921 2922 sk_msg_iter_var_next(i); 2923 2924 if (pop < sge->length - a) { 2925 if (space) { 2926 sge->length = a; 2927 sk_msg_shift_right(msg, i); 2928 nsge = sk_msg_elem(msg, i); 2929 get_page(sg_page(sge)); 2930 sg_set_page(nsge, 2931 sg_page(sge), 2932 b, sge->offset + pop + a); 2933 } else { 2934 struct page *page, *orig; 2935 u8 *to, *from; 2936 2937 page = alloc_pages(__GFP_NOWARN | 2938 __GFP_COMP | GFP_ATOMIC, 2939 get_order(a + b)); 2940 if (unlikely(!page)) 2941 return -ENOMEM; 2942 2943 sge->length = a; 2944 orig = sg_page(sge); 2945 from = sg_virt(sge); 2946 to = page_address(page); 2947 memcpy(to, from, a); 2948 memcpy(to + a, from + a + pop, b); 2949 sg_set_page(sge, page, a + b, 0); 2950 put_page(orig); 2951 } 2952 pop = 0; 2953 } else if (pop >= sge->length - a) { 2954 pop -= (sge->length - a); 2955 sge->length = a; 2956 } 2957 } 2958 2959 /* From above the current layout _must_ be as follows, 2960 * 2961 * -| offset 2962 * -| start 2963 * 2964 * |---- pop ---|---------------- b ------------| 2965 * |____________________________________________| length 2966 * 2967 * Offset and start of the current msg elem are equal because in the 2968 * previous case we handled offset != start and either consumed the 2969 * entire element and advanced to the next element OR pop == 0. 2970 * 2971 * Two cases to handle here are first pop is less than the length 2972 * leaving some remainder b above. Simply adjust the element's layout 2973 * in this case. Or pop >= length of the element so that b = 0. In this 2974 * case advance to next element decrementing pop. 2975 */ 2976 while (pop) { 2977 struct scatterlist *sge = sk_msg_elem(msg, i); 2978 2979 if (pop < sge->length) { 2980 sge->length -= pop; 2981 sge->offset += pop; 2982 pop = 0; 2983 } else { 2984 pop -= sge->length; 2985 sk_msg_shift_left(msg, i); 2986 } 2987 sk_msg_iter_var_next(i); 2988 } 2989 2990 sk_mem_uncharge(msg->sk, len - pop); 2991 msg->sg.size -= (len - pop); 2992 sk_msg_compute_data_pointers(msg); 2993 return 0; 2994 } 2995 2996 static const struct bpf_func_proto bpf_msg_pop_data_proto = { 2997 .func = bpf_msg_pop_data, 2998 .gpl_only = false, 2999 .ret_type = RET_INTEGER, 3000 .arg1_type = ARG_PTR_TO_CTX, 3001 .arg2_type = ARG_ANYTHING, 3002 .arg3_type = ARG_ANYTHING, 3003 .arg4_type = ARG_ANYTHING, 3004 }; 3005 3006 #ifdef CONFIG_CGROUP_NET_CLASSID 3007 BPF_CALL_0(bpf_get_cgroup_classid_curr) 3008 { 3009 return __task_get_classid(current); 3010 } 3011 3012 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = { 3013 .func = bpf_get_cgroup_classid_curr, 3014 .gpl_only = false, 3015 .ret_type = RET_INTEGER, 3016 }; 3017 3018 BPF_CALL_1(bpf_skb_cgroup_classid, const struct sk_buff *, skb) 3019 { 3020 struct sock *sk = skb_to_full_sk(skb); 3021 3022 if (!sk || !sk_fullsock(sk)) 3023 return 0; 3024 3025 return sock_cgroup_classid(&sk->sk_cgrp_data); 3026 } 3027 3028 static const struct bpf_func_proto bpf_skb_cgroup_classid_proto = { 3029 .func = bpf_skb_cgroup_classid, 3030 .gpl_only = false, 3031 .ret_type = RET_INTEGER, 3032 .arg1_type = ARG_PTR_TO_CTX, 3033 }; 3034 #endif 3035 3036 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb) 3037 { 3038 return task_get_classid(skb); 3039 } 3040 3041 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = { 3042 .func = bpf_get_cgroup_classid, 3043 .gpl_only = false, 3044 .ret_type = RET_INTEGER, 3045 .arg1_type = ARG_PTR_TO_CTX, 3046 }; 3047 3048 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb) 3049 { 3050 return dst_tclassid(skb); 3051 } 3052 3053 static const struct bpf_func_proto bpf_get_route_realm_proto = { 3054 .func = bpf_get_route_realm, 3055 .gpl_only = false, 3056 .ret_type = RET_INTEGER, 3057 .arg1_type = ARG_PTR_TO_CTX, 3058 }; 3059 3060 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb) 3061 { 3062 /* If skb_clear_hash() was called due to mangling, we can 3063 * trigger SW recalculation here. Later access to hash 3064 * can then use the inline skb->hash via context directly 3065 * instead of calling this helper again. 3066 */ 3067 return skb_get_hash(skb); 3068 } 3069 3070 static const struct bpf_func_proto bpf_get_hash_recalc_proto = { 3071 .func = bpf_get_hash_recalc, 3072 .gpl_only = false, 3073 .ret_type = RET_INTEGER, 3074 .arg1_type = ARG_PTR_TO_CTX, 3075 }; 3076 3077 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb) 3078 { 3079 /* After all direct packet write, this can be used once for 3080 * triggering a lazy recalc on next skb_get_hash() invocation. 3081 */ 3082 skb_clear_hash(skb); 3083 return 0; 3084 } 3085 3086 static const struct bpf_func_proto bpf_set_hash_invalid_proto = { 3087 .func = bpf_set_hash_invalid, 3088 .gpl_only = false, 3089 .ret_type = RET_INTEGER, 3090 .arg1_type = ARG_PTR_TO_CTX, 3091 }; 3092 3093 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash) 3094 { 3095 /* Set user specified hash as L4(+), so that it gets returned 3096 * on skb_get_hash() call unless BPF prog later on triggers a 3097 * skb_clear_hash(). 3098 */ 3099 __skb_set_sw_hash(skb, hash, true); 3100 return 0; 3101 } 3102 3103 static const struct bpf_func_proto bpf_set_hash_proto = { 3104 .func = bpf_set_hash, 3105 .gpl_only = false, 3106 .ret_type = RET_INTEGER, 3107 .arg1_type = ARG_PTR_TO_CTX, 3108 .arg2_type = ARG_ANYTHING, 3109 }; 3110 3111 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto, 3112 u16, vlan_tci) 3113 { 3114 int ret; 3115 3116 if (unlikely(vlan_proto != htons(ETH_P_8021Q) && 3117 vlan_proto != htons(ETH_P_8021AD))) 3118 vlan_proto = htons(ETH_P_8021Q); 3119 3120 bpf_push_mac_rcsum(skb); 3121 ret = skb_vlan_push(skb, vlan_proto, vlan_tci); 3122 bpf_pull_mac_rcsum(skb); 3123 3124 bpf_compute_data_pointers(skb); 3125 return ret; 3126 } 3127 3128 static const struct bpf_func_proto bpf_skb_vlan_push_proto = { 3129 .func = bpf_skb_vlan_push, 3130 .gpl_only = false, 3131 .ret_type = RET_INTEGER, 3132 .arg1_type = ARG_PTR_TO_CTX, 3133 .arg2_type = ARG_ANYTHING, 3134 .arg3_type = ARG_ANYTHING, 3135 }; 3136 3137 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb) 3138 { 3139 int ret; 3140 3141 bpf_push_mac_rcsum(skb); 3142 ret = skb_vlan_pop(skb); 3143 bpf_pull_mac_rcsum(skb); 3144 3145 bpf_compute_data_pointers(skb); 3146 return ret; 3147 } 3148 3149 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = { 3150 .func = bpf_skb_vlan_pop, 3151 .gpl_only = false, 3152 .ret_type = RET_INTEGER, 3153 .arg1_type = ARG_PTR_TO_CTX, 3154 }; 3155 3156 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len) 3157 { 3158 /* Caller already did skb_cow() with len as headroom, 3159 * so no need to do it here. 3160 */ 3161 skb_push(skb, len); 3162 memmove(skb->data, skb->data + len, off); 3163 memset(skb->data + off, 0, len); 3164 3165 /* No skb_postpush_rcsum(skb, skb->data + off, len) 3166 * needed here as it does not change the skb->csum 3167 * result for checksum complete when summing over 3168 * zeroed blocks. 3169 */ 3170 return 0; 3171 } 3172 3173 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len) 3174 { 3175 /* skb_ensure_writable() is not needed here, as we're 3176 * already working on an uncloned skb. 3177 */ 3178 if (unlikely(!pskb_may_pull(skb, off + len))) 3179 return -ENOMEM; 3180 3181 skb_postpull_rcsum(skb, skb->data + off, len); 3182 memmove(skb->data + len, skb->data, off); 3183 __skb_pull(skb, len); 3184 3185 return 0; 3186 } 3187 3188 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len) 3189 { 3190 bool trans_same = skb->transport_header == skb->network_header; 3191 int ret; 3192 3193 /* There's no need for __skb_push()/__skb_pull() pair to 3194 * get to the start of the mac header as we're guaranteed 3195 * to always start from here under eBPF. 3196 */ 3197 ret = bpf_skb_generic_push(skb, off, len); 3198 if (likely(!ret)) { 3199 skb->mac_header -= len; 3200 skb->network_header -= len; 3201 if (trans_same) 3202 skb->transport_header = skb->network_header; 3203 } 3204 3205 return ret; 3206 } 3207 3208 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len) 3209 { 3210 bool trans_same = skb->transport_header == skb->network_header; 3211 int ret; 3212 3213 /* Same here, __skb_push()/__skb_pull() pair not needed. */ 3214 ret = bpf_skb_generic_pop(skb, off, len); 3215 if (likely(!ret)) { 3216 skb->mac_header += len; 3217 skb->network_header += len; 3218 if (trans_same) 3219 skb->transport_header = skb->network_header; 3220 } 3221 3222 return ret; 3223 } 3224 3225 static int bpf_skb_proto_4_to_6(struct sk_buff *skb) 3226 { 3227 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3228 u32 off = skb_mac_header_len(skb); 3229 int ret; 3230 3231 ret = skb_cow(skb, len_diff); 3232 if (unlikely(ret < 0)) 3233 return ret; 3234 3235 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3236 if (unlikely(ret < 0)) 3237 return ret; 3238 3239 if (skb_is_gso(skb)) { 3240 struct skb_shared_info *shinfo = skb_shinfo(skb); 3241 3242 /* SKB_GSO_TCPV4 needs to be changed into SKB_GSO_TCPV6. */ 3243 if (shinfo->gso_type & SKB_GSO_TCPV4) { 3244 shinfo->gso_type &= ~SKB_GSO_TCPV4; 3245 shinfo->gso_type |= SKB_GSO_TCPV6; 3246 } 3247 } 3248 3249 skb->protocol = htons(ETH_P_IPV6); 3250 skb_clear_hash(skb); 3251 3252 return 0; 3253 } 3254 3255 static int bpf_skb_proto_6_to_4(struct sk_buff *skb) 3256 { 3257 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 3258 u32 off = skb_mac_header_len(skb); 3259 int ret; 3260 3261 ret = skb_unclone(skb, GFP_ATOMIC); 3262 if (unlikely(ret < 0)) 3263 return ret; 3264 3265 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3266 if (unlikely(ret < 0)) 3267 return ret; 3268 3269 if (skb_is_gso(skb)) { 3270 struct skb_shared_info *shinfo = skb_shinfo(skb); 3271 3272 /* SKB_GSO_TCPV6 needs to be changed into SKB_GSO_TCPV4. */ 3273 if (shinfo->gso_type & SKB_GSO_TCPV6) { 3274 shinfo->gso_type &= ~SKB_GSO_TCPV6; 3275 shinfo->gso_type |= SKB_GSO_TCPV4; 3276 } 3277 } 3278 3279 skb->protocol = htons(ETH_P_IP); 3280 skb_clear_hash(skb); 3281 3282 return 0; 3283 } 3284 3285 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto) 3286 { 3287 __be16 from_proto = skb->protocol; 3288 3289 if (from_proto == htons(ETH_P_IP) && 3290 to_proto == htons(ETH_P_IPV6)) 3291 return bpf_skb_proto_4_to_6(skb); 3292 3293 if (from_proto == htons(ETH_P_IPV6) && 3294 to_proto == htons(ETH_P_IP)) 3295 return bpf_skb_proto_6_to_4(skb); 3296 3297 return -ENOTSUPP; 3298 } 3299 3300 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto, 3301 u64, flags) 3302 { 3303 int ret; 3304 3305 if (unlikely(flags)) 3306 return -EINVAL; 3307 3308 /* General idea is that this helper does the basic groundwork 3309 * needed for changing the protocol, and eBPF program fills the 3310 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace() 3311 * and other helpers, rather than passing a raw buffer here. 3312 * 3313 * The rationale is to keep this minimal and without a need to 3314 * deal with raw packet data. F.e. even if we would pass buffers 3315 * here, the program still needs to call the bpf_lX_csum_replace() 3316 * helpers anyway. Plus, this way we keep also separation of 3317 * concerns, since f.e. bpf_skb_store_bytes() should only take 3318 * care of stores. 3319 * 3320 * Currently, additional options and extension header space are 3321 * not supported, but flags register is reserved so we can adapt 3322 * that. For offloads, we mark packet as dodgy, so that headers 3323 * need to be verified first. 3324 */ 3325 ret = bpf_skb_proto_xlat(skb, proto); 3326 bpf_compute_data_pointers(skb); 3327 return ret; 3328 } 3329 3330 static const struct bpf_func_proto bpf_skb_change_proto_proto = { 3331 .func = bpf_skb_change_proto, 3332 .gpl_only = false, 3333 .ret_type = RET_INTEGER, 3334 .arg1_type = ARG_PTR_TO_CTX, 3335 .arg2_type = ARG_ANYTHING, 3336 .arg3_type = ARG_ANYTHING, 3337 }; 3338 3339 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type) 3340 { 3341 /* We only allow a restricted subset to be changed for now. */ 3342 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) || 3343 !skb_pkt_type_ok(pkt_type))) 3344 return -EINVAL; 3345 3346 skb->pkt_type = pkt_type; 3347 return 0; 3348 } 3349 3350 static const struct bpf_func_proto bpf_skb_change_type_proto = { 3351 .func = bpf_skb_change_type, 3352 .gpl_only = false, 3353 .ret_type = RET_INTEGER, 3354 .arg1_type = ARG_PTR_TO_CTX, 3355 .arg2_type = ARG_ANYTHING, 3356 }; 3357 3358 static u32 bpf_skb_net_base_len(const struct sk_buff *skb) 3359 { 3360 switch (skb->protocol) { 3361 case htons(ETH_P_IP): 3362 return sizeof(struct iphdr); 3363 case htons(ETH_P_IPV6): 3364 return sizeof(struct ipv6hdr); 3365 default: 3366 return ~0U; 3367 } 3368 } 3369 3370 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \ 3371 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3372 3373 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \ 3374 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \ 3375 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \ 3376 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \ 3377 BPF_F_ADJ_ROOM_ENCAP_L2_ETH | \ 3378 BPF_F_ADJ_ROOM_ENCAP_L2( \ 3379 BPF_ADJ_ROOM_ENCAP_L2_MASK)) 3380 3381 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff, 3382 u64 flags) 3383 { 3384 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT; 3385 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK; 3386 u16 mac_len = 0, inner_net = 0, inner_trans = 0; 3387 unsigned int gso_type = SKB_GSO_DODGY; 3388 int ret; 3389 3390 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3391 /* udp gso_size delineates datagrams, only allow if fixed */ 3392 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3393 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3394 return -ENOTSUPP; 3395 } 3396 3397 ret = skb_cow_head(skb, len_diff); 3398 if (unlikely(ret < 0)) 3399 return ret; 3400 3401 if (encap) { 3402 if (skb->protocol != htons(ETH_P_IP) && 3403 skb->protocol != htons(ETH_P_IPV6)) 3404 return -ENOTSUPP; 3405 3406 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 && 3407 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3408 return -EINVAL; 3409 3410 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE && 3411 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3412 return -EINVAL; 3413 3414 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH && 3415 inner_mac_len < ETH_HLEN) 3416 return -EINVAL; 3417 3418 if (skb->encapsulation) 3419 return -EALREADY; 3420 3421 mac_len = skb->network_header - skb->mac_header; 3422 inner_net = skb->network_header; 3423 if (inner_mac_len > len_diff) 3424 return -EINVAL; 3425 inner_trans = skb->transport_header; 3426 } 3427 3428 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3429 if (unlikely(ret < 0)) 3430 return ret; 3431 3432 if (encap) { 3433 skb->inner_mac_header = inner_net - inner_mac_len; 3434 skb->inner_network_header = inner_net; 3435 skb->inner_transport_header = inner_trans; 3436 3437 if (flags & BPF_F_ADJ_ROOM_ENCAP_L2_ETH) 3438 skb_set_inner_protocol(skb, htons(ETH_P_TEB)); 3439 else 3440 skb_set_inner_protocol(skb, skb->protocol); 3441 3442 skb->encapsulation = 1; 3443 skb_set_network_header(skb, mac_len); 3444 3445 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3446 gso_type |= SKB_GSO_UDP_TUNNEL; 3447 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE) 3448 gso_type |= SKB_GSO_GRE; 3449 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3450 gso_type |= SKB_GSO_IPXIP6; 3451 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3452 gso_type |= SKB_GSO_IPXIP4; 3453 3454 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE || 3455 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) { 3456 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ? 3457 sizeof(struct ipv6hdr) : 3458 sizeof(struct iphdr); 3459 3460 skb_set_transport_header(skb, mac_len + nh_len); 3461 } 3462 3463 /* Match skb->protocol to new outer l3 protocol */ 3464 if (skb->protocol == htons(ETH_P_IP) && 3465 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3466 skb->protocol = htons(ETH_P_IPV6); 3467 else if (skb->protocol == htons(ETH_P_IPV6) && 3468 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3469 skb->protocol = htons(ETH_P_IP); 3470 } 3471 3472 if (skb_is_gso(skb)) { 3473 struct skb_shared_info *shinfo = skb_shinfo(skb); 3474 3475 /* Due to header grow, MSS needs to be downgraded. */ 3476 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3477 skb_decrease_gso_size(shinfo, len_diff); 3478 3479 /* Header must be checked, and gso_segs recomputed. */ 3480 shinfo->gso_type |= gso_type; 3481 shinfo->gso_segs = 0; 3482 } 3483 3484 return 0; 3485 } 3486 3487 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff, 3488 u64 flags) 3489 { 3490 int ret; 3491 3492 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_FIXED_GSO | 3493 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3494 return -EINVAL; 3495 3496 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3497 /* udp gso_size delineates datagrams, only allow if fixed */ 3498 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3499 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3500 return -ENOTSUPP; 3501 } 3502 3503 ret = skb_unclone(skb, GFP_ATOMIC); 3504 if (unlikely(ret < 0)) 3505 return ret; 3506 3507 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3508 if (unlikely(ret < 0)) 3509 return ret; 3510 3511 if (skb_is_gso(skb)) { 3512 struct skb_shared_info *shinfo = skb_shinfo(skb); 3513 3514 /* Due to header shrink, MSS can be upgraded. */ 3515 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3516 skb_increase_gso_size(shinfo, len_diff); 3517 3518 /* Header must be checked, and gso_segs recomputed. */ 3519 shinfo->gso_type |= SKB_GSO_DODGY; 3520 shinfo->gso_segs = 0; 3521 } 3522 3523 return 0; 3524 } 3525 3526 #define BPF_SKB_MAX_LEN SKB_MAX_ALLOC 3527 3528 BPF_CALL_4(sk_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3529 u32, mode, u64, flags) 3530 { 3531 u32 len_diff_abs = abs(len_diff); 3532 bool shrink = len_diff < 0; 3533 int ret = 0; 3534 3535 if (unlikely(flags || mode)) 3536 return -EINVAL; 3537 if (unlikely(len_diff_abs > 0xfffU)) 3538 return -EFAULT; 3539 3540 if (!shrink) { 3541 ret = skb_cow(skb, len_diff); 3542 if (unlikely(ret < 0)) 3543 return ret; 3544 __skb_push(skb, len_diff_abs); 3545 memset(skb->data, 0, len_diff_abs); 3546 } else { 3547 if (unlikely(!pskb_may_pull(skb, len_diff_abs))) 3548 return -ENOMEM; 3549 __skb_pull(skb, len_diff_abs); 3550 } 3551 if (tls_sw_has_ctx_rx(skb->sk)) { 3552 struct strp_msg *rxm = strp_msg(skb); 3553 3554 rxm->full_len += len_diff; 3555 } 3556 return ret; 3557 } 3558 3559 static const struct bpf_func_proto sk_skb_adjust_room_proto = { 3560 .func = sk_skb_adjust_room, 3561 .gpl_only = false, 3562 .ret_type = RET_INTEGER, 3563 .arg1_type = ARG_PTR_TO_CTX, 3564 .arg2_type = ARG_ANYTHING, 3565 .arg3_type = ARG_ANYTHING, 3566 .arg4_type = ARG_ANYTHING, 3567 }; 3568 3569 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3570 u32, mode, u64, flags) 3571 { 3572 u32 len_cur, len_diff_abs = abs(len_diff); 3573 u32 len_min = bpf_skb_net_base_len(skb); 3574 u32 len_max = BPF_SKB_MAX_LEN; 3575 __be16 proto = skb->protocol; 3576 bool shrink = len_diff < 0; 3577 u32 off; 3578 int ret; 3579 3580 if (unlikely(flags & ~(BPF_F_ADJ_ROOM_MASK | 3581 BPF_F_ADJ_ROOM_NO_CSUM_RESET))) 3582 return -EINVAL; 3583 if (unlikely(len_diff_abs > 0xfffU)) 3584 return -EFAULT; 3585 if (unlikely(proto != htons(ETH_P_IP) && 3586 proto != htons(ETH_P_IPV6))) 3587 return -ENOTSUPP; 3588 3589 off = skb_mac_header_len(skb); 3590 switch (mode) { 3591 case BPF_ADJ_ROOM_NET: 3592 off += bpf_skb_net_base_len(skb); 3593 break; 3594 case BPF_ADJ_ROOM_MAC: 3595 break; 3596 default: 3597 return -ENOTSUPP; 3598 } 3599 3600 len_cur = skb->len - skb_network_offset(skb); 3601 if ((shrink && (len_diff_abs >= len_cur || 3602 len_cur - len_diff_abs < len_min)) || 3603 (!shrink && (skb->len + len_diff_abs > len_max && 3604 !skb_is_gso(skb)))) 3605 return -ENOTSUPP; 3606 3607 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) : 3608 bpf_skb_net_grow(skb, off, len_diff_abs, flags); 3609 if (!ret && !(flags & BPF_F_ADJ_ROOM_NO_CSUM_RESET)) 3610 __skb_reset_checksum_unnecessary(skb); 3611 3612 bpf_compute_data_pointers(skb); 3613 return ret; 3614 } 3615 3616 static const struct bpf_func_proto bpf_skb_adjust_room_proto = { 3617 .func = bpf_skb_adjust_room, 3618 .gpl_only = false, 3619 .ret_type = RET_INTEGER, 3620 .arg1_type = ARG_PTR_TO_CTX, 3621 .arg2_type = ARG_ANYTHING, 3622 .arg3_type = ARG_ANYTHING, 3623 .arg4_type = ARG_ANYTHING, 3624 }; 3625 3626 static u32 __bpf_skb_min_len(const struct sk_buff *skb) 3627 { 3628 u32 min_len = skb_network_offset(skb); 3629 3630 if (skb_transport_header_was_set(skb)) 3631 min_len = skb_transport_offset(skb); 3632 if (skb->ip_summed == CHECKSUM_PARTIAL) 3633 min_len = skb_checksum_start_offset(skb) + 3634 skb->csum_offset + sizeof(__sum16); 3635 return min_len; 3636 } 3637 3638 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len) 3639 { 3640 unsigned int old_len = skb->len; 3641 int ret; 3642 3643 ret = __skb_grow_rcsum(skb, new_len); 3644 if (!ret) 3645 memset(skb->data + old_len, 0, new_len - old_len); 3646 return ret; 3647 } 3648 3649 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len) 3650 { 3651 return __skb_trim_rcsum(skb, new_len); 3652 } 3653 3654 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len, 3655 u64 flags) 3656 { 3657 u32 max_len = BPF_SKB_MAX_LEN; 3658 u32 min_len = __bpf_skb_min_len(skb); 3659 int ret; 3660 3661 if (unlikely(flags || new_len > max_len || new_len < min_len)) 3662 return -EINVAL; 3663 if (skb->encapsulation) 3664 return -ENOTSUPP; 3665 3666 /* The basic idea of this helper is that it's performing the 3667 * needed work to either grow or trim an skb, and eBPF program 3668 * rewrites the rest via helpers like bpf_skb_store_bytes(), 3669 * bpf_lX_csum_replace() and others rather than passing a raw 3670 * buffer here. This one is a slow path helper and intended 3671 * for replies with control messages. 3672 * 3673 * Like in bpf_skb_change_proto(), we want to keep this rather 3674 * minimal and without protocol specifics so that we are able 3675 * to separate concerns as in bpf_skb_store_bytes() should only 3676 * be the one responsible for writing buffers. 3677 * 3678 * It's really expected to be a slow path operation here for 3679 * control message replies, so we're implicitly linearizing, 3680 * uncloning and drop offloads from the skb by this. 3681 */ 3682 ret = __bpf_try_make_writable(skb, skb->len); 3683 if (!ret) { 3684 if (new_len > skb->len) 3685 ret = bpf_skb_grow_rcsum(skb, new_len); 3686 else if (new_len < skb->len) 3687 ret = bpf_skb_trim_rcsum(skb, new_len); 3688 if (!ret && skb_is_gso(skb)) 3689 skb_gso_reset(skb); 3690 } 3691 return ret; 3692 } 3693 3694 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3695 u64, flags) 3696 { 3697 int ret = __bpf_skb_change_tail(skb, new_len, flags); 3698 3699 bpf_compute_data_pointers(skb); 3700 return ret; 3701 } 3702 3703 static const struct bpf_func_proto bpf_skb_change_tail_proto = { 3704 .func = bpf_skb_change_tail, 3705 .gpl_only = false, 3706 .ret_type = RET_INTEGER, 3707 .arg1_type = ARG_PTR_TO_CTX, 3708 .arg2_type = ARG_ANYTHING, 3709 .arg3_type = ARG_ANYTHING, 3710 }; 3711 3712 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3713 u64, flags) 3714 { 3715 return __bpf_skb_change_tail(skb, new_len, flags); 3716 } 3717 3718 static const struct bpf_func_proto sk_skb_change_tail_proto = { 3719 .func = sk_skb_change_tail, 3720 .gpl_only = false, 3721 .ret_type = RET_INTEGER, 3722 .arg1_type = ARG_PTR_TO_CTX, 3723 .arg2_type = ARG_ANYTHING, 3724 .arg3_type = ARG_ANYTHING, 3725 }; 3726 3727 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room, 3728 u64 flags) 3729 { 3730 u32 max_len = BPF_SKB_MAX_LEN; 3731 u32 new_len = skb->len + head_room; 3732 int ret; 3733 3734 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) || 3735 new_len < skb->len)) 3736 return -EINVAL; 3737 3738 ret = skb_cow(skb, head_room); 3739 if (likely(!ret)) { 3740 /* Idea for this helper is that we currently only 3741 * allow to expand on mac header. This means that 3742 * skb->protocol network header, etc, stay as is. 3743 * Compared to bpf_skb_change_tail(), we're more 3744 * flexible due to not needing to linearize or 3745 * reset GSO. Intention for this helper is to be 3746 * used by an L3 skb that needs to push mac header 3747 * for redirection into L2 device. 3748 */ 3749 __skb_push(skb, head_room); 3750 memset(skb->data, 0, head_room); 3751 skb_reset_mac_header(skb); 3752 skb_reset_mac_len(skb); 3753 } 3754 3755 return ret; 3756 } 3757 3758 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room, 3759 u64, flags) 3760 { 3761 int ret = __bpf_skb_change_head(skb, head_room, flags); 3762 3763 bpf_compute_data_pointers(skb); 3764 return ret; 3765 } 3766 3767 static const struct bpf_func_proto bpf_skb_change_head_proto = { 3768 .func = bpf_skb_change_head, 3769 .gpl_only = false, 3770 .ret_type = RET_INTEGER, 3771 .arg1_type = ARG_PTR_TO_CTX, 3772 .arg2_type = ARG_ANYTHING, 3773 .arg3_type = ARG_ANYTHING, 3774 }; 3775 3776 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room, 3777 u64, flags) 3778 { 3779 return __bpf_skb_change_head(skb, head_room, flags); 3780 } 3781 3782 static const struct bpf_func_proto sk_skb_change_head_proto = { 3783 .func = sk_skb_change_head, 3784 .gpl_only = false, 3785 .ret_type = RET_INTEGER, 3786 .arg1_type = ARG_PTR_TO_CTX, 3787 .arg2_type = ARG_ANYTHING, 3788 .arg3_type = ARG_ANYTHING, 3789 }; 3790 3791 BPF_CALL_1(bpf_xdp_get_buff_len, struct xdp_buff*, xdp) 3792 { 3793 return xdp_get_buff_len(xdp); 3794 } 3795 3796 static const struct bpf_func_proto bpf_xdp_get_buff_len_proto = { 3797 .func = bpf_xdp_get_buff_len, 3798 .gpl_only = false, 3799 .ret_type = RET_INTEGER, 3800 .arg1_type = ARG_PTR_TO_CTX, 3801 }; 3802 3803 BTF_ID_LIST_SINGLE(bpf_xdp_get_buff_len_bpf_ids, struct, xdp_buff) 3804 3805 const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto = { 3806 .func = bpf_xdp_get_buff_len, 3807 .gpl_only = false, 3808 .arg1_type = ARG_PTR_TO_BTF_ID, 3809 .arg1_btf_id = &bpf_xdp_get_buff_len_bpf_ids[0], 3810 }; 3811 3812 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp) 3813 { 3814 return xdp_data_meta_unsupported(xdp) ? 0 : 3815 xdp->data - xdp->data_meta; 3816 } 3817 3818 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset) 3819 { 3820 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 3821 unsigned long metalen = xdp_get_metalen(xdp); 3822 void *data_start = xdp_frame_end + metalen; 3823 void *data = xdp->data + offset; 3824 3825 if (unlikely(data < data_start || 3826 data > xdp->data_end - ETH_HLEN)) 3827 return -EINVAL; 3828 3829 if (metalen) 3830 memmove(xdp->data_meta + offset, 3831 xdp->data_meta, metalen); 3832 xdp->data_meta += offset; 3833 xdp->data = data; 3834 3835 return 0; 3836 } 3837 3838 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = { 3839 .func = bpf_xdp_adjust_head, 3840 .gpl_only = false, 3841 .ret_type = RET_INTEGER, 3842 .arg1_type = ARG_PTR_TO_CTX, 3843 .arg2_type = ARG_ANYTHING, 3844 }; 3845 3846 static void bpf_xdp_copy_buf(struct xdp_buff *xdp, unsigned long off, 3847 void *buf, unsigned long len, bool flush) 3848 { 3849 unsigned long ptr_len, ptr_off = 0; 3850 skb_frag_t *next_frag, *end_frag; 3851 struct skb_shared_info *sinfo; 3852 void *src, *dst; 3853 u8 *ptr_buf; 3854 3855 if (likely(xdp->data_end - xdp->data >= off + len)) { 3856 src = flush ? buf : xdp->data + off; 3857 dst = flush ? xdp->data + off : buf; 3858 memcpy(dst, src, len); 3859 return; 3860 } 3861 3862 sinfo = xdp_get_shared_info_from_buff(xdp); 3863 end_frag = &sinfo->frags[sinfo->nr_frags]; 3864 next_frag = &sinfo->frags[0]; 3865 3866 ptr_len = xdp->data_end - xdp->data; 3867 ptr_buf = xdp->data; 3868 3869 while (true) { 3870 if (off < ptr_off + ptr_len) { 3871 unsigned long copy_off = off - ptr_off; 3872 unsigned long copy_len = min(len, ptr_len - copy_off); 3873 3874 src = flush ? buf : ptr_buf + copy_off; 3875 dst = flush ? ptr_buf + copy_off : buf; 3876 memcpy(dst, src, copy_len); 3877 3878 off += copy_len; 3879 len -= copy_len; 3880 buf += copy_len; 3881 } 3882 3883 if (!len || next_frag == end_frag) 3884 break; 3885 3886 ptr_off += ptr_len; 3887 ptr_buf = skb_frag_address(next_frag); 3888 ptr_len = skb_frag_size(next_frag); 3889 next_frag++; 3890 } 3891 } 3892 3893 static void *bpf_xdp_pointer(struct xdp_buff *xdp, u32 offset, u32 len) 3894 { 3895 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 3896 u32 size = xdp->data_end - xdp->data; 3897 void *addr = xdp->data; 3898 int i; 3899 3900 if (unlikely(offset > 0xffff || len > 0xffff)) 3901 return ERR_PTR(-EFAULT); 3902 3903 if (offset + len > xdp_get_buff_len(xdp)) 3904 return ERR_PTR(-EINVAL); 3905 3906 if (offset < size) /* linear area */ 3907 goto out; 3908 3909 offset -= size; 3910 for (i = 0; i < sinfo->nr_frags; i++) { /* paged area */ 3911 u32 frag_size = skb_frag_size(&sinfo->frags[i]); 3912 3913 if (offset < frag_size) { 3914 addr = skb_frag_address(&sinfo->frags[i]); 3915 size = frag_size; 3916 break; 3917 } 3918 offset -= frag_size; 3919 } 3920 out: 3921 return offset + len < size ? addr + offset : NULL; 3922 } 3923 3924 BPF_CALL_4(bpf_xdp_load_bytes, struct xdp_buff *, xdp, u32, offset, 3925 void *, buf, u32, len) 3926 { 3927 void *ptr; 3928 3929 ptr = bpf_xdp_pointer(xdp, offset, len); 3930 if (IS_ERR(ptr)) 3931 return PTR_ERR(ptr); 3932 3933 if (!ptr) 3934 bpf_xdp_copy_buf(xdp, offset, buf, len, false); 3935 else 3936 memcpy(buf, ptr, len); 3937 3938 return 0; 3939 } 3940 3941 static const struct bpf_func_proto bpf_xdp_load_bytes_proto = { 3942 .func = bpf_xdp_load_bytes, 3943 .gpl_only = false, 3944 .ret_type = RET_INTEGER, 3945 .arg1_type = ARG_PTR_TO_CTX, 3946 .arg2_type = ARG_ANYTHING, 3947 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 3948 .arg4_type = ARG_CONST_SIZE, 3949 }; 3950 3951 BPF_CALL_4(bpf_xdp_store_bytes, struct xdp_buff *, xdp, u32, offset, 3952 void *, buf, u32, len) 3953 { 3954 void *ptr; 3955 3956 ptr = bpf_xdp_pointer(xdp, offset, len); 3957 if (IS_ERR(ptr)) 3958 return PTR_ERR(ptr); 3959 3960 if (!ptr) 3961 bpf_xdp_copy_buf(xdp, offset, buf, len, true); 3962 else 3963 memcpy(ptr, buf, len); 3964 3965 return 0; 3966 } 3967 3968 static const struct bpf_func_proto bpf_xdp_store_bytes_proto = { 3969 .func = bpf_xdp_store_bytes, 3970 .gpl_only = false, 3971 .ret_type = RET_INTEGER, 3972 .arg1_type = ARG_PTR_TO_CTX, 3973 .arg2_type = ARG_ANYTHING, 3974 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 3975 .arg4_type = ARG_CONST_SIZE, 3976 }; 3977 3978 static int bpf_xdp_frags_increase_tail(struct xdp_buff *xdp, int offset) 3979 { 3980 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 3981 skb_frag_t *frag = &sinfo->frags[sinfo->nr_frags - 1]; 3982 struct xdp_rxq_info *rxq = xdp->rxq; 3983 unsigned int tailroom; 3984 3985 if (!rxq->frag_size || rxq->frag_size > xdp->frame_sz) 3986 return -EOPNOTSUPP; 3987 3988 tailroom = rxq->frag_size - skb_frag_size(frag) - skb_frag_off(frag); 3989 if (unlikely(offset > tailroom)) 3990 return -EINVAL; 3991 3992 memset(skb_frag_address(frag) + skb_frag_size(frag), 0, offset); 3993 skb_frag_size_add(frag, offset); 3994 sinfo->xdp_frags_size += offset; 3995 3996 return 0; 3997 } 3998 3999 static int bpf_xdp_frags_shrink_tail(struct xdp_buff *xdp, int offset) 4000 { 4001 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 4002 int i, n_frags_free = 0, len_free = 0; 4003 4004 if (unlikely(offset > (int)xdp_get_buff_len(xdp) - ETH_HLEN)) 4005 return -EINVAL; 4006 4007 for (i = sinfo->nr_frags - 1; i >= 0 && offset > 0; i--) { 4008 skb_frag_t *frag = &sinfo->frags[i]; 4009 int shrink = min_t(int, offset, skb_frag_size(frag)); 4010 4011 len_free += shrink; 4012 offset -= shrink; 4013 4014 if (skb_frag_size(frag) == shrink) { 4015 struct page *page = skb_frag_page(frag); 4016 4017 __xdp_return(page_address(page), &xdp->rxq->mem, 4018 false, NULL); 4019 n_frags_free++; 4020 } else { 4021 skb_frag_size_sub(frag, shrink); 4022 break; 4023 } 4024 } 4025 sinfo->nr_frags -= n_frags_free; 4026 sinfo->xdp_frags_size -= len_free; 4027 4028 if (unlikely(!sinfo->nr_frags)) { 4029 xdp_buff_clear_frags_flag(xdp); 4030 xdp->data_end -= offset; 4031 } 4032 4033 return 0; 4034 } 4035 4036 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset) 4037 { 4038 void *data_hard_end = xdp_data_hard_end(xdp); /* use xdp->frame_sz */ 4039 void *data_end = xdp->data_end + offset; 4040 4041 if (unlikely(xdp_buff_has_frags(xdp))) { /* non-linear xdp buff */ 4042 if (offset < 0) 4043 return bpf_xdp_frags_shrink_tail(xdp, -offset); 4044 4045 return bpf_xdp_frags_increase_tail(xdp, offset); 4046 } 4047 4048 /* Notice that xdp_data_hard_end have reserved some tailroom */ 4049 if (unlikely(data_end > data_hard_end)) 4050 return -EINVAL; 4051 4052 /* ALL drivers MUST init xdp->frame_sz, chicken check below */ 4053 if (unlikely(xdp->frame_sz > PAGE_SIZE)) { 4054 WARN_ONCE(1, "Too BIG xdp->frame_sz = %d\n", xdp->frame_sz); 4055 return -EINVAL; 4056 } 4057 4058 if (unlikely(data_end < xdp->data + ETH_HLEN)) 4059 return -EINVAL; 4060 4061 /* Clear memory area on grow, can contain uninit kernel memory */ 4062 if (offset > 0) 4063 memset(xdp->data_end, 0, offset); 4064 4065 xdp->data_end = data_end; 4066 4067 return 0; 4068 } 4069 4070 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = { 4071 .func = bpf_xdp_adjust_tail, 4072 .gpl_only = false, 4073 .ret_type = RET_INTEGER, 4074 .arg1_type = ARG_PTR_TO_CTX, 4075 .arg2_type = ARG_ANYTHING, 4076 }; 4077 4078 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset) 4079 { 4080 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 4081 void *meta = xdp->data_meta + offset; 4082 unsigned long metalen = xdp->data - meta; 4083 4084 if (xdp_data_meta_unsupported(xdp)) 4085 return -ENOTSUPP; 4086 if (unlikely(meta < xdp_frame_end || 4087 meta > xdp->data)) 4088 return -EINVAL; 4089 if (unlikely(xdp_metalen_invalid(metalen))) 4090 return -EACCES; 4091 4092 xdp->data_meta = meta; 4093 4094 return 0; 4095 } 4096 4097 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = { 4098 .func = bpf_xdp_adjust_meta, 4099 .gpl_only = false, 4100 .ret_type = RET_INTEGER, 4101 .arg1_type = ARG_PTR_TO_CTX, 4102 .arg2_type = ARG_ANYTHING, 4103 }; 4104 4105 /* XDP_REDIRECT works by a three-step process, implemented in the functions 4106 * below: 4107 * 4108 * 1. The bpf_redirect() and bpf_redirect_map() helpers will lookup the target 4109 * of the redirect and store it (along with some other metadata) in a per-CPU 4110 * struct bpf_redirect_info. 4111 * 4112 * 2. When the program returns the XDP_REDIRECT return code, the driver will 4113 * call xdp_do_redirect() which will use the information in struct 4114 * bpf_redirect_info to actually enqueue the frame into a map type-specific 4115 * bulk queue structure. 4116 * 4117 * 3. Before exiting its NAPI poll loop, the driver will call xdp_do_flush(), 4118 * which will flush all the different bulk queues, thus completing the 4119 * redirect. 4120 * 4121 * Pointers to the map entries will be kept around for this whole sequence of 4122 * steps, protected by RCU. However, there is no top-level rcu_read_lock() in 4123 * the core code; instead, the RCU protection relies on everything happening 4124 * inside a single NAPI poll sequence, which means it's between a pair of calls 4125 * to local_bh_disable()/local_bh_enable(). 4126 * 4127 * The map entries are marked as __rcu and the map code makes sure to 4128 * dereference those pointers with rcu_dereference_check() in a way that works 4129 * for both sections that to hold an rcu_read_lock() and sections that are 4130 * called from NAPI without a separate rcu_read_lock(). The code below does not 4131 * use RCU annotations, but relies on those in the map code. 4132 */ 4133 void xdp_do_flush(void) 4134 { 4135 __dev_flush(); 4136 __cpu_map_flush(); 4137 __xsk_map_flush(); 4138 } 4139 EXPORT_SYMBOL_GPL(xdp_do_flush); 4140 4141 void bpf_clear_redirect_map(struct bpf_map *map) 4142 { 4143 struct bpf_redirect_info *ri; 4144 int cpu; 4145 4146 for_each_possible_cpu(cpu) { 4147 ri = per_cpu_ptr(&bpf_redirect_info, cpu); 4148 /* Avoid polluting remote cacheline due to writes if 4149 * not needed. Once we pass this test, we need the 4150 * cmpxchg() to make sure it hasn't been changed in 4151 * the meantime by remote CPU. 4152 */ 4153 if (unlikely(READ_ONCE(ri->map) == map)) 4154 cmpxchg(&ri->map, map, NULL); 4155 } 4156 } 4157 4158 DEFINE_STATIC_KEY_FALSE(bpf_master_redirect_enabled_key); 4159 EXPORT_SYMBOL_GPL(bpf_master_redirect_enabled_key); 4160 4161 u32 xdp_master_redirect(struct xdp_buff *xdp) 4162 { 4163 struct net_device *master, *slave; 4164 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4165 4166 master = netdev_master_upper_dev_get_rcu(xdp->rxq->dev); 4167 slave = master->netdev_ops->ndo_xdp_get_xmit_slave(master, xdp); 4168 if (slave && slave != xdp->rxq->dev) { 4169 /* The target device is different from the receiving device, so 4170 * redirect it to the new device. 4171 * Using XDP_REDIRECT gets the correct behaviour from XDP enabled 4172 * drivers to unmap the packet from their rx ring. 4173 */ 4174 ri->tgt_index = slave->ifindex; 4175 ri->map_id = INT_MAX; 4176 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4177 return XDP_REDIRECT; 4178 } 4179 return XDP_TX; 4180 } 4181 EXPORT_SYMBOL_GPL(xdp_master_redirect); 4182 4183 static inline int __xdp_do_redirect_xsk(struct bpf_redirect_info *ri, 4184 struct net_device *dev, 4185 struct xdp_buff *xdp, 4186 struct bpf_prog *xdp_prog) 4187 { 4188 enum bpf_map_type map_type = ri->map_type; 4189 void *fwd = ri->tgt_value; 4190 u32 map_id = ri->map_id; 4191 int err; 4192 4193 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4194 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4195 4196 err = __xsk_map_redirect(fwd, xdp); 4197 if (unlikely(err)) 4198 goto err; 4199 4200 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4201 return 0; 4202 err: 4203 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4204 return err; 4205 } 4206 4207 static __always_inline int __xdp_do_redirect_frame(struct bpf_redirect_info *ri, 4208 struct net_device *dev, 4209 struct xdp_frame *xdpf, 4210 struct bpf_prog *xdp_prog) 4211 { 4212 enum bpf_map_type map_type = ri->map_type; 4213 void *fwd = ri->tgt_value; 4214 u32 map_id = ri->map_id; 4215 struct bpf_map *map; 4216 int err; 4217 4218 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4219 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4220 4221 if (unlikely(!xdpf)) { 4222 err = -EOVERFLOW; 4223 goto err; 4224 } 4225 4226 switch (map_type) { 4227 case BPF_MAP_TYPE_DEVMAP: 4228 fallthrough; 4229 case BPF_MAP_TYPE_DEVMAP_HASH: 4230 map = READ_ONCE(ri->map); 4231 if (unlikely(map)) { 4232 WRITE_ONCE(ri->map, NULL); 4233 err = dev_map_enqueue_multi(xdpf, dev, map, 4234 ri->flags & BPF_F_EXCLUDE_INGRESS); 4235 } else { 4236 err = dev_map_enqueue(fwd, xdpf, dev); 4237 } 4238 break; 4239 case BPF_MAP_TYPE_CPUMAP: 4240 err = cpu_map_enqueue(fwd, xdpf, dev); 4241 break; 4242 case BPF_MAP_TYPE_UNSPEC: 4243 if (map_id == INT_MAX) { 4244 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index); 4245 if (unlikely(!fwd)) { 4246 err = -EINVAL; 4247 break; 4248 } 4249 err = dev_xdp_enqueue(fwd, xdpf, dev); 4250 break; 4251 } 4252 fallthrough; 4253 default: 4254 err = -EBADRQC; 4255 } 4256 4257 if (unlikely(err)) 4258 goto err; 4259 4260 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4261 return 0; 4262 err: 4263 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4264 return err; 4265 } 4266 4267 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp, 4268 struct bpf_prog *xdp_prog) 4269 { 4270 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4271 enum bpf_map_type map_type = ri->map_type; 4272 4273 /* XDP_REDIRECT is not fully supported yet for xdp frags since 4274 * not all XDP capable drivers can map non-linear xdp_frame in 4275 * ndo_xdp_xmit. 4276 */ 4277 if (unlikely(xdp_buff_has_frags(xdp) && 4278 map_type != BPF_MAP_TYPE_CPUMAP)) 4279 return -EOPNOTSUPP; 4280 4281 if (map_type == BPF_MAP_TYPE_XSKMAP) 4282 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog); 4283 4284 return __xdp_do_redirect_frame(ri, dev, xdp_convert_buff_to_frame(xdp), 4285 xdp_prog); 4286 } 4287 EXPORT_SYMBOL_GPL(xdp_do_redirect); 4288 4289 int xdp_do_redirect_frame(struct net_device *dev, struct xdp_buff *xdp, 4290 struct xdp_frame *xdpf, struct bpf_prog *xdp_prog) 4291 { 4292 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4293 enum bpf_map_type map_type = ri->map_type; 4294 4295 if (map_type == BPF_MAP_TYPE_XSKMAP) 4296 return __xdp_do_redirect_xsk(ri, dev, xdp, xdp_prog); 4297 4298 return __xdp_do_redirect_frame(ri, dev, xdpf, xdp_prog); 4299 } 4300 EXPORT_SYMBOL_GPL(xdp_do_redirect_frame); 4301 4302 static int xdp_do_generic_redirect_map(struct net_device *dev, 4303 struct sk_buff *skb, 4304 struct xdp_buff *xdp, 4305 struct bpf_prog *xdp_prog, 4306 void *fwd, 4307 enum bpf_map_type map_type, u32 map_id) 4308 { 4309 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4310 struct bpf_map *map; 4311 int err; 4312 4313 switch (map_type) { 4314 case BPF_MAP_TYPE_DEVMAP: 4315 fallthrough; 4316 case BPF_MAP_TYPE_DEVMAP_HASH: 4317 map = READ_ONCE(ri->map); 4318 if (unlikely(map)) { 4319 WRITE_ONCE(ri->map, NULL); 4320 err = dev_map_redirect_multi(dev, skb, xdp_prog, map, 4321 ri->flags & BPF_F_EXCLUDE_INGRESS); 4322 } else { 4323 err = dev_map_generic_redirect(fwd, skb, xdp_prog); 4324 } 4325 if (unlikely(err)) 4326 goto err; 4327 break; 4328 case BPF_MAP_TYPE_XSKMAP: 4329 err = xsk_generic_rcv(fwd, xdp); 4330 if (err) 4331 goto err; 4332 consume_skb(skb); 4333 break; 4334 case BPF_MAP_TYPE_CPUMAP: 4335 err = cpu_map_generic_redirect(fwd, skb); 4336 if (unlikely(err)) 4337 goto err; 4338 break; 4339 default: 4340 err = -EBADRQC; 4341 goto err; 4342 } 4343 4344 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index); 4345 return 0; 4346 err: 4347 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map_type, map_id, ri->tgt_index, err); 4348 return err; 4349 } 4350 4351 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 4352 struct xdp_buff *xdp, struct bpf_prog *xdp_prog) 4353 { 4354 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4355 enum bpf_map_type map_type = ri->map_type; 4356 void *fwd = ri->tgt_value; 4357 u32 map_id = ri->map_id; 4358 int err; 4359 4360 ri->map_id = 0; /* Valid map id idr range: [1,INT_MAX[ */ 4361 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4362 4363 if (map_type == BPF_MAP_TYPE_UNSPEC && map_id == INT_MAX) { 4364 fwd = dev_get_by_index_rcu(dev_net(dev), ri->tgt_index); 4365 if (unlikely(!fwd)) { 4366 err = -EINVAL; 4367 goto err; 4368 } 4369 4370 err = xdp_ok_fwd_dev(fwd, skb->len); 4371 if (unlikely(err)) 4372 goto err; 4373 4374 skb->dev = fwd; 4375 _trace_xdp_redirect(dev, xdp_prog, ri->tgt_index); 4376 generic_xdp_tx(skb, xdp_prog); 4377 return 0; 4378 } 4379 4380 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, fwd, map_type, map_id); 4381 err: 4382 _trace_xdp_redirect_err(dev, xdp_prog, ri->tgt_index, err); 4383 return err; 4384 } 4385 4386 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags) 4387 { 4388 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 4389 4390 if (unlikely(flags)) 4391 return XDP_ABORTED; 4392 4393 /* NB! Map type UNSPEC and map_id == INT_MAX (never generated 4394 * by map_idr) is used for ifindex based XDP redirect. 4395 */ 4396 ri->tgt_index = ifindex; 4397 ri->map_id = INT_MAX; 4398 ri->map_type = BPF_MAP_TYPE_UNSPEC; 4399 4400 return XDP_REDIRECT; 4401 } 4402 4403 static const struct bpf_func_proto bpf_xdp_redirect_proto = { 4404 .func = bpf_xdp_redirect, 4405 .gpl_only = false, 4406 .ret_type = RET_INTEGER, 4407 .arg1_type = ARG_ANYTHING, 4408 .arg2_type = ARG_ANYTHING, 4409 }; 4410 4411 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, 4412 u64, flags) 4413 { 4414 return map->ops->map_redirect(map, ifindex, flags); 4415 } 4416 4417 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = { 4418 .func = bpf_xdp_redirect_map, 4419 .gpl_only = false, 4420 .ret_type = RET_INTEGER, 4421 .arg1_type = ARG_CONST_MAP_PTR, 4422 .arg2_type = ARG_ANYTHING, 4423 .arg3_type = ARG_ANYTHING, 4424 }; 4425 4426 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb, 4427 unsigned long off, unsigned long len) 4428 { 4429 void *ptr = skb_header_pointer(skb, off, len, dst_buff); 4430 4431 if (unlikely(!ptr)) 4432 return len; 4433 if (ptr != dst_buff) 4434 memcpy(dst_buff, ptr, len); 4435 4436 return 0; 4437 } 4438 4439 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map, 4440 u64, flags, void *, meta, u64, meta_size) 4441 { 4442 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4443 4444 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4445 return -EINVAL; 4446 if (unlikely(!skb || skb_size > skb->len)) 4447 return -EFAULT; 4448 4449 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size, 4450 bpf_skb_copy); 4451 } 4452 4453 static const struct bpf_func_proto bpf_skb_event_output_proto = { 4454 .func = bpf_skb_event_output, 4455 .gpl_only = true, 4456 .ret_type = RET_INTEGER, 4457 .arg1_type = ARG_PTR_TO_CTX, 4458 .arg2_type = ARG_CONST_MAP_PTR, 4459 .arg3_type = ARG_ANYTHING, 4460 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4461 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4462 }; 4463 4464 BTF_ID_LIST_SINGLE(bpf_skb_output_btf_ids, struct, sk_buff) 4465 4466 const struct bpf_func_proto bpf_skb_output_proto = { 4467 .func = bpf_skb_event_output, 4468 .gpl_only = true, 4469 .ret_type = RET_INTEGER, 4470 .arg1_type = ARG_PTR_TO_BTF_ID, 4471 .arg1_btf_id = &bpf_skb_output_btf_ids[0], 4472 .arg2_type = ARG_CONST_MAP_PTR, 4473 .arg3_type = ARG_ANYTHING, 4474 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4475 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4476 }; 4477 4478 static unsigned short bpf_tunnel_key_af(u64 flags) 4479 { 4480 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET; 4481 } 4482 4483 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to, 4484 u32, size, u64, flags) 4485 { 4486 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4487 u8 compat[sizeof(struct bpf_tunnel_key)]; 4488 void *to_orig = to; 4489 int err; 4490 4491 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) { 4492 err = -EINVAL; 4493 goto err_clear; 4494 } 4495 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { 4496 err = -EPROTO; 4497 goto err_clear; 4498 } 4499 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4500 err = -EINVAL; 4501 switch (size) { 4502 case offsetof(struct bpf_tunnel_key, local_ipv6[0]): 4503 case offsetof(struct bpf_tunnel_key, tunnel_label): 4504 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4505 goto set_compat; 4506 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4507 /* Fixup deprecated structure layouts here, so we have 4508 * a common path later on. 4509 */ 4510 if (ip_tunnel_info_af(info) != AF_INET) 4511 goto err_clear; 4512 set_compat: 4513 to = (struct bpf_tunnel_key *)compat; 4514 break; 4515 default: 4516 goto err_clear; 4517 } 4518 } 4519 4520 to->tunnel_id = be64_to_cpu(info->key.tun_id); 4521 to->tunnel_tos = info->key.tos; 4522 to->tunnel_ttl = info->key.ttl; 4523 to->tunnel_ext = 0; 4524 4525 if (flags & BPF_F_TUNINFO_IPV6) { 4526 memcpy(to->remote_ipv6, &info->key.u.ipv6.src, 4527 sizeof(to->remote_ipv6)); 4528 memcpy(to->local_ipv6, &info->key.u.ipv6.dst, 4529 sizeof(to->local_ipv6)); 4530 to->tunnel_label = be32_to_cpu(info->key.label); 4531 } else { 4532 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); 4533 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 4534 to->local_ipv4 = be32_to_cpu(info->key.u.ipv4.dst); 4535 memset(&to->local_ipv6[1], 0, sizeof(__u32) * 3); 4536 to->tunnel_label = 0; 4537 } 4538 4539 if (unlikely(size != sizeof(struct bpf_tunnel_key))) 4540 memcpy(to_orig, to, size); 4541 4542 return 0; 4543 err_clear: 4544 memset(to_orig, 0, size); 4545 return err; 4546 } 4547 4548 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { 4549 .func = bpf_skb_get_tunnel_key, 4550 .gpl_only = false, 4551 .ret_type = RET_INTEGER, 4552 .arg1_type = ARG_PTR_TO_CTX, 4553 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4554 .arg3_type = ARG_CONST_SIZE, 4555 .arg4_type = ARG_ANYTHING, 4556 }; 4557 4558 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size) 4559 { 4560 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 4561 int err; 4562 4563 if (unlikely(!info || 4564 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) { 4565 err = -ENOENT; 4566 goto err_clear; 4567 } 4568 if (unlikely(size < info->options_len)) { 4569 err = -ENOMEM; 4570 goto err_clear; 4571 } 4572 4573 ip_tunnel_info_opts_get(to, info); 4574 if (size > info->options_len) 4575 memset(to + info->options_len, 0, size - info->options_len); 4576 4577 return info->options_len; 4578 err_clear: 4579 memset(to, 0, size); 4580 return err; 4581 } 4582 4583 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { 4584 .func = bpf_skb_get_tunnel_opt, 4585 .gpl_only = false, 4586 .ret_type = RET_INTEGER, 4587 .arg1_type = ARG_PTR_TO_CTX, 4588 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 4589 .arg3_type = ARG_CONST_SIZE, 4590 }; 4591 4592 static struct metadata_dst __percpu *md_dst; 4593 4594 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb, 4595 const struct bpf_tunnel_key *, from, u32, size, u64, flags) 4596 { 4597 struct metadata_dst *md = this_cpu_ptr(md_dst); 4598 u8 compat[sizeof(struct bpf_tunnel_key)]; 4599 struct ip_tunnel_info *info; 4600 4601 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | 4602 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER))) 4603 return -EINVAL; 4604 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 4605 switch (size) { 4606 case offsetof(struct bpf_tunnel_key, local_ipv6[0]): 4607 case offsetof(struct bpf_tunnel_key, tunnel_label): 4608 case offsetof(struct bpf_tunnel_key, tunnel_ext): 4609 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 4610 /* Fixup deprecated structure layouts here, so we have 4611 * a common path later on. 4612 */ 4613 memcpy(compat, from, size); 4614 memset(compat + size, 0, sizeof(compat) - size); 4615 from = (const struct bpf_tunnel_key *) compat; 4616 break; 4617 default: 4618 return -EINVAL; 4619 } 4620 } 4621 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || 4622 from->tunnel_ext)) 4623 return -EINVAL; 4624 4625 skb_dst_drop(skb); 4626 dst_hold((struct dst_entry *) md); 4627 skb_dst_set(skb, (struct dst_entry *) md); 4628 4629 info = &md->u.tun_info; 4630 memset(info, 0, sizeof(*info)); 4631 info->mode = IP_TUNNEL_INFO_TX; 4632 4633 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE; 4634 if (flags & BPF_F_DONT_FRAGMENT) 4635 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT; 4636 if (flags & BPF_F_ZERO_CSUM_TX) 4637 info->key.tun_flags &= ~TUNNEL_CSUM; 4638 if (flags & BPF_F_SEQ_NUMBER) 4639 info->key.tun_flags |= TUNNEL_SEQ; 4640 4641 info->key.tun_id = cpu_to_be64(from->tunnel_id); 4642 info->key.tos = from->tunnel_tos; 4643 info->key.ttl = from->tunnel_ttl; 4644 4645 if (flags & BPF_F_TUNINFO_IPV6) { 4646 info->mode |= IP_TUNNEL_INFO_IPV6; 4647 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, 4648 sizeof(from->remote_ipv6)); 4649 memcpy(&info->key.u.ipv6.src, from->local_ipv6, 4650 sizeof(from->local_ipv6)); 4651 info->key.label = cpu_to_be32(from->tunnel_label) & 4652 IPV6_FLOWLABEL_MASK; 4653 } else { 4654 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); 4655 info->key.u.ipv4.src = cpu_to_be32(from->local_ipv4); 4656 } 4657 4658 return 0; 4659 } 4660 4661 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { 4662 .func = bpf_skb_set_tunnel_key, 4663 .gpl_only = false, 4664 .ret_type = RET_INTEGER, 4665 .arg1_type = ARG_PTR_TO_CTX, 4666 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4667 .arg3_type = ARG_CONST_SIZE, 4668 .arg4_type = ARG_ANYTHING, 4669 }; 4670 4671 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb, 4672 const u8 *, from, u32, size) 4673 { 4674 struct ip_tunnel_info *info = skb_tunnel_info(skb); 4675 const struct metadata_dst *md = this_cpu_ptr(md_dst); 4676 4677 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) 4678 return -EINVAL; 4679 if (unlikely(size > IP_TUNNEL_OPTS_MAX)) 4680 return -ENOMEM; 4681 4682 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT); 4683 4684 return 0; 4685 } 4686 4687 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { 4688 .func = bpf_skb_set_tunnel_opt, 4689 .gpl_only = false, 4690 .ret_type = RET_INTEGER, 4691 .arg1_type = ARG_PTR_TO_CTX, 4692 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4693 .arg3_type = ARG_CONST_SIZE, 4694 }; 4695 4696 static const struct bpf_func_proto * 4697 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) 4698 { 4699 if (!md_dst) { 4700 struct metadata_dst __percpu *tmp; 4701 4702 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, 4703 METADATA_IP_TUNNEL, 4704 GFP_KERNEL); 4705 if (!tmp) 4706 return NULL; 4707 if (cmpxchg(&md_dst, NULL, tmp)) 4708 metadata_dst_free_percpu(tmp); 4709 } 4710 4711 switch (which) { 4712 case BPF_FUNC_skb_set_tunnel_key: 4713 return &bpf_skb_set_tunnel_key_proto; 4714 case BPF_FUNC_skb_set_tunnel_opt: 4715 return &bpf_skb_set_tunnel_opt_proto; 4716 default: 4717 return NULL; 4718 } 4719 } 4720 4721 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map, 4722 u32, idx) 4723 { 4724 struct bpf_array *array = container_of(map, struct bpf_array, map); 4725 struct cgroup *cgrp; 4726 struct sock *sk; 4727 4728 sk = skb_to_full_sk(skb); 4729 if (!sk || !sk_fullsock(sk)) 4730 return -ENOENT; 4731 if (unlikely(idx >= array->map.max_entries)) 4732 return -E2BIG; 4733 4734 cgrp = READ_ONCE(array->ptrs[idx]); 4735 if (unlikely(!cgrp)) 4736 return -EAGAIN; 4737 4738 return sk_under_cgroup_hierarchy(sk, cgrp); 4739 } 4740 4741 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = { 4742 .func = bpf_skb_under_cgroup, 4743 .gpl_only = false, 4744 .ret_type = RET_INTEGER, 4745 .arg1_type = ARG_PTR_TO_CTX, 4746 .arg2_type = ARG_CONST_MAP_PTR, 4747 .arg3_type = ARG_ANYTHING, 4748 }; 4749 4750 #ifdef CONFIG_SOCK_CGROUP_DATA 4751 static inline u64 __bpf_sk_cgroup_id(struct sock *sk) 4752 { 4753 struct cgroup *cgrp; 4754 4755 sk = sk_to_full_sk(sk); 4756 if (!sk || !sk_fullsock(sk)) 4757 return 0; 4758 4759 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4760 return cgroup_id(cgrp); 4761 } 4762 4763 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb) 4764 { 4765 return __bpf_sk_cgroup_id(skb->sk); 4766 } 4767 4768 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = { 4769 .func = bpf_skb_cgroup_id, 4770 .gpl_only = false, 4771 .ret_type = RET_INTEGER, 4772 .arg1_type = ARG_PTR_TO_CTX, 4773 }; 4774 4775 static inline u64 __bpf_sk_ancestor_cgroup_id(struct sock *sk, 4776 int ancestor_level) 4777 { 4778 struct cgroup *ancestor; 4779 struct cgroup *cgrp; 4780 4781 sk = sk_to_full_sk(sk); 4782 if (!sk || !sk_fullsock(sk)) 4783 return 0; 4784 4785 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4786 ancestor = cgroup_ancestor(cgrp, ancestor_level); 4787 if (!ancestor) 4788 return 0; 4789 4790 return cgroup_id(ancestor); 4791 } 4792 4793 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int, 4794 ancestor_level) 4795 { 4796 return __bpf_sk_ancestor_cgroup_id(skb->sk, ancestor_level); 4797 } 4798 4799 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = { 4800 .func = bpf_skb_ancestor_cgroup_id, 4801 .gpl_only = false, 4802 .ret_type = RET_INTEGER, 4803 .arg1_type = ARG_PTR_TO_CTX, 4804 .arg2_type = ARG_ANYTHING, 4805 }; 4806 4807 BPF_CALL_1(bpf_sk_cgroup_id, struct sock *, sk) 4808 { 4809 return __bpf_sk_cgroup_id(sk); 4810 } 4811 4812 static const struct bpf_func_proto bpf_sk_cgroup_id_proto = { 4813 .func = bpf_sk_cgroup_id, 4814 .gpl_only = false, 4815 .ret_type = RET_INTEGER, 4816 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4817 }; 4818 4819 BPF_CALL_2(bpf_sk_ancestor_cgroup_id, struct sock *, sk, int, ancestor_level) 4820 { 4821 return __bpf_sk_ancestor_cgroup_id(sk, ancestor_level); 4822 } 4823 4824 static const struct bpf_func_proto bpf_sk_ancestor_cgroup_id_proto = { 4825 .func = bpf_sk_ancestor_cgroup_id, 4826 .gpl_only = false, 4827 .ret_type = RET_INTEGER, 4828 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4829 .arg2_type = ARG_ANYTHING, 4830 }; 4831 #endif 4832 4833 static unsigned long bpf_xdp_copy(void *dst, const void *ctx, 4834 unsigned long off, unsigned long len) 4835 { 4836 struct xdp_buff *xdp = (struct xdp_buff *)ctx; 4837 4838 bpf_xdp_copy_buf(xdp, off, dst, len, false); 4839 return 0; 4840 } 4841 4842 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map, 4843 u64, flags, void *, meta, u64, meta_size) 4844 { 4845 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4846 4847 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4848 return -EINVAL; 4849 4850 if (unlikely(!xdp || xdp_size > xdp_get_buff_len(xdp))) 4851 return -EFAULT; 4852 4853 return bpf_event_output(map, flags, meta, meta_size, xdp, 4854 xdp_size, bpf_xdp_copy); 4855 } 4856 4857 static const struct bpf_func_proto bpf_xdp_event_output_proto = { 4858 .func = bpf_xdp_event_output, 4859 .gpl_only = true, 4860 .ret_type = RET_INTEGER, 4861 .arg1_type = ARG_PTR_TO_CTX, 4862 .arg2_type = ARG_CONST_MAP_PTR, 4863 .arg3_type = ARG_ANYTHING, 4864 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4865 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4866 }; 4867 4868 BTF_ID_LIST_SINGLE(bpf_xdp_output_btf_ids, struct, xdp_buff) 4869 4870 const struct bpf_func_proto bpf_xdp_output_proto = { 4871 .func = bpf_xdp_event_output, 4872 .gpl_only = true, 4873 .ret_type = RET_INTEGER, 4874 .arg1_type = ARG_PTR_TO_BTF_ID, 4875 .arg1_btf_id = &bpf_xdp_output_btf_ids[0], 4876 .arg2_type = ARG_CONST_MAP_PTR, 4877 .arg3_type = ARG_ANYTHING, 4878 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 4879 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4880 }; 4881 4882 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb) 4883 { 4884 return skb->sk ? __sock_gen_cookie(skb->sk) : 0; 4885 } 4886 4887 static const struct bpf_func_proto bpf_get_socket_cookie_proto = { 4888 .func = bpf_get_socket_cookie, 4889 .gpl_only = false, 4890 .ret_type = RET_INTEGER, 4891 .arg1_type = ARG_PTR_TO_CTX, 4892 }; 4893 4894 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4895 { 4896 return __sock_gen_cookie(ctx->sk); 4897 } 4898 4899 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = { 4900 .func = bpf_get_socket_cookie_sock_addr, 4901 .gpl_only = false, 4902 .ret_type = RET_INTEGER, 4903 .arg1_type = ARG_PTR_TO_CTX, 4904 }; 4905 4906 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx) 4907 { 4908 return __sock_gen_cookie(ctx); 4909 } 4910 4911 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = { 4912 .func = bpf_get_socket_cookie_sock, 4913 .gpl_only = false, 4914 .ret_type = RET_INTEGER, 4915 .arg1_type = ARG_PTR_TO_CTX, 4916 }; 4917 4918 BPF_CALL_1(bpf_get_socket_ptr_cookie, struct sock *, sk) 4919 { 4920 return sk ? sock_gen_cookie(sk) : 0; 4921 } 4922 4923 const struct bpf_func_proto bpf_get_socket_ptr_cookie_proto = { 4924 .func = bpf_get_socket_ptr_cookie, 4925 .gpl_only = false, 4926 .ret_type = RET_INTEGER, 4927 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 4928 }; 4929 4930 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 4931 { 4932 return __sock_gen_cookie(ctx->sk); 4933 } 4934 4935 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = { 4936 .func = bpf_get_socket_cookie_sock_ops, 4937 .gpl_only = false, 4938 .ret_type = RET_INTEGER, 4939 .arg1_type = ARG_PTR_TO_CTX, 4940 }; 4941 4942 static u64 __bpf_get_netns_cookie(struct sock *sk) 4943 { 4944 const struct net *net = sk ? sock_net(sk) : &init_net; 4945 4946 return net->net_cookie; 4947 } 4948 4949 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx) 4950 { 4951 return __bpf_get_netns_cookie(ctx); 4952 } 4953 4954 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = { 4955 .func = bpf_get_netns_cookie_sock, 4956 .gpl_only = false, 4957 .ret_type = RET_INTEGER, 4958 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4959 }; 4960 4961 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4962 { 4963 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4964 } 4965 4966 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = { 4967 .func = bpf_get_netns_cookie_sock_addr, 4968 .gpl_only = false, 4969 .ret_type = RET_INTEGER, 4970 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4971 }; 4972 4973 BPF_CALL_1(bpf_get_netns_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 4974 { 4975 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4976 } 4977 4978 static const struct bpf_func_proto bpf_get_netns_cookie_sock_ops_proto = { 4979 .func = bpf_get_netns_cookie_sock_ops, 4980 .gpl_only = false, 4981 .ret_type = RET_INTEGER, 4982 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4983 }; 4984 4985 BPF_CALL_1(bpf_get_netns_cookie_sk_msg, struct sk_msg *, ctx) 4986 { 4987 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4988 } 4989 4990 static const struct bpf_func_proto bpf_get_netns_cookie_sk_msg_proto = { 4991 .func = bpf_get_netns_cookie_sk_msg, 4992 .gpl_only = false, 4993 .ret_type = RET_INTEGER, 4994 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4995 }; 4996 4997 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb) 4998 { 4999 struct sock *sk = sk_to_full_sk(skb->sk); 5000 kuid_t kuid; 5001 5002 if (!sk || !sk_fullsock(sk)) 5003 return overflowuid; 5004 kuid = sock_net_uid(sock_net(sk), sk); 5005 return from_kuid_munged(sock_net(sk)->user_ns, kuid); 5006 } 5007 5008 static const struct bpf_func_proto bpf_get_socket_uid_proto = { 5009 .func = bpf_get_socket_uid, 5010 .gpl_only = false, 5011 .ret_type = RET_INTEGER, 5012 .arg1_type = ARG_PTR_TO_CTX, 5013 }; 5014 5015 static int _bpf_setsockopt(struct sock *sk, int level, int optname, 5016 char *optval, int optlen) 5017 { 5018 char devname[IFNAMSIZ]; 5019 int val, valbool; 5020 struct net *net; 5021 int ifindex; 5022 int ret = 0; 5023 5024 if (!sk_fullsock(sk)) 5025 return -EINVAL; 5026 5027 sock_owned_by_me(sk); 5028 5029 if (level == SOL_SOCKET) { 5030 if (optlen != sizeof(int) && optname != SO_BINDTODEVICE) 5031 return -EINVAL; 5032 val = *((int *)optval); 5033 valbool = val ? 1 : 0; 5034 5035 /* Only some socketops are supported */ 5036 switch (optname) { 5037 case SO_RCVBUF: 5038 val = min_t(u32, val, sysctl_rmem_max); 5039 val = min_t(int, val, INT_MAX / 2); 5040 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 5041 WRITE_ONCE(sk->sk_rcvbuf, 5042 max_t(int, val * 2, SOCK_MIN_RCVBUF)); 5043 break; 5044 case SO_SNDBUF: 5045 val = min_t(u32, val, sysctl_wmem_max); 5046 val = min_t(int, val, INT_MAX / 2); 5047 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 5048 WRITE_ONCE(sk->sk_sndbuf, 5049 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 5050 break; 5051 case SO_MAX_PACING_RATE: /* 32bit version */ 5052 if (val != ~0U) 5053 cmpxchg(&sk->sk_pacing_status, 5054 SK_PACING_NONE, 5055 SK_PACING_NEEDED); 5056 sk->sk_max_pacing_rate = (val == ~0U) ? 5057 ~0UL : (unsigned int)val; 5058 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 5059 sk->sk_max_pacing_rate); 5060 break; 5061 case SO_PRIORITY: 5062 sk->sk_priority = val; 5063 break; 5064 case SO_RCVLOWAT: 5065 if (val < 0) 5066 val = INT_MAX; 5067 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 5068 break; 5069 case SO_MARK: 5070 if (sk->sk_mark != val) { 5071 sk->sk_mark = val; 5072 sk_dst_reset(sk); 5073 } 5074 break; 5075 case SO_BINDTODEVICE: 5076 optlen = min_t(long, optlen, IFNAMSIZ - 1); 5077 strncpy(devname, optval, optlen); 5078 devname[optlen] = 0; 5079 5080 ifindex = 0; 5081 if (devname[0] != '\0') { 5082 struct net_device *dev; 5083 5084 ret = -ENODEV; 5085 5086 net = sock_net(sk); 5087 dev = dev_get_by_name(net, devname); 5088 if (!dev) 5089 break; 5090 ifindex = dev->ifindex; 5091 dev_put(dev); 5092 } 5093 fallthrough; 5094 case SO_BINDTOIFINDEX: 5095 if (optname == SO_BINDTOIFINDEX) 5096 ifindex = val; 5097 ret = sock_bindtoindex(sk, ifindex, false); 5098 break; 5099 case SO_KEEPALIVE: 5100 if (sk->sk_prot->keepalive) 5101 sk->sk_prot->keepalive(sk, valbool); 5102 sock_valbool_flag(sk, SOCK_KEEPOPEN, valbool); 5103 break; 5104 case SO_REUSEPORT: 5105 sk->sk_reuseport = valbool; 5106 break; 5107 case SO_TXREHASH: 5108 if (val < -1 || val > 1) { 5109 ret = -EINVAL; 5110 break; 5111 } 5112 sk->sk_txrehash = (u8)val; 5113 break; 5114 default: 5115 ret = -EINVAL; 5116 } 5117 #ifdef CONFIG_INET 5118 } else if (level == SOL_IP) { 5119 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 5120 return -EINVAL; 5121 5122 val = *((int *)optval); 5123 /* Only some options are supported */ 5124 switch (optname) { 5125 case IP_TOS: 5126 if (val < -1 || val > 0xff) { 5127 ret = -EINVAL; 5128 } else { 5129 struct inet_sock *inet = inet_sk(sk); 5130 5131 if (val == -1) 5132 val = 0; 5133 inet->tos = val; 5134 } 5135 break; 5136 default: 5137 ret = -EINVAL; 5138 } 5139 #if IS_ENABLED(CONFIG_IPV6) 5140 } else if (level == SOL_IPV6) { 5141 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 5142 return -EINVAL; 5143 5144 val = *((int *)optval); 5145 /* Only some options are supported */ 5146 switch (optname) { 5147 case IPV6_TCLASS: 5148 if (val < -1 || val > 0xff) { 5149 ret = -EINVAL; 5150 } else { 5151 struct ipv6_pinfo *np = inet6_sk(sk); 5152 5153 if (val == -1) 5154 val = 0; 5155 np->tclass = val; 5156 } 5157 break; 5158 default: 5159 ret = -EINVAL; 5160 } 5161 #endif 5162 } else if (level == SOL_TCP && 5163 sk->sk_prot->setsockopt == tcp_setsockopt) { 5164 if (optname == TCP_CONGESTION) { 5165 char name[TCP_CA_NAME_MAX]; 5166 5167 strncpy(name, optval, min_t(long, optlen, 5168 TCP_CA_NAME_MAX-1)); 5169 name[TCP_CA_NAME_MAX-1] = 0; 5170 ret = tcp_set_congestion_control(sk, name, false, true); 5171 } else { 5172 struct inet_connection_sock *icsk = inet_csk(sk); 5173 struct tcp_sock *tp = tcp_sk(sk); 5174 unsigned long timeout; 5175 5176 if (optlen != sizeof(int)) 5177 return -EINVAL; 5178 5179 val = *((int *)optval); 5180 /* Only some options are supported */ 5181 switch (optname) { 5182 case TCP_BPF_IW: 5183 if (val <= 0 || tp->data_segs_out > tp->syn_data) 5184 ret = -EINVAL; 5185 else 5186 tcp_snd_cwnd_set(tp, val); 5187 break; 5188 case TCP_BPF_SNDCWND_CLAMP: 5189 if (val <= 0) { 5190 ret = -EINVAL; 5191 } else { 5192 tp->snd_cwnd_clamp = val; 5193 tp->snd_ssthresh = val; 5194 } 5195 break; 5196 case TCP_BPF_DELACK_MAX: 5197 timeout = usecs_to_jiffies(val); 5198 if (timeout > TCP_DELACK_MAX || 5199 timeout < TCP_TIMEOUT_MIN) 5200 return -EINVAL; 5201 inet_csk(sk)->icsk_delack_max = timeout; 5202 break; 5203 case TCP_BPF_RTO_MIN: 5204 timeout = usecs_to_jiffies(val); 5205 if (timeout > TCP_RTO_MIN || 5206 timeout < TCP_TIMEOUT_MIN) 5207 return -EINVAL; 5208 inet_csk(sk)->icsk_rto_min = timeout; 5209 break; 5210 case TCP_SAVE_SYN: 5211 if (val < 0 || val > 1) 5212 ret = -EINVAL; 5213 else 5214 tp->save_syn = val; 5215 break; 5216 case TCP_KEEPIDLE: 5217 ret = tcp_sock_set_keepidle_locked(sk, val); 5218 break; 5219 case TCP_KEEPINTVL: 5220 if (val < 1 || val > MAX_TCP_KEEPINTVL) 5221 ret = -EINVAL; 5222 else 5223 tp->keepalive_intvl = val * HZ; 5224 break; 5225 case TCP_KEEPCNT: 5226 if (val < 1 || val > MAX_TCP_KEEPCNT) 5227 ret = -EINVAL; 5228 else 5229 tp->keepalive_probes = val; 5230 break; 5231 case TCP_SYNCNT: 5232 if (val < 1 || val > MAX_TCP_SYNCNT) 5233 ret = -EINVAL; 5234 else 5235 icsk->icsk_syn_retries = val; 5236 break; 5237 case TCP_USER_TIMEOUT: 5238 if (val < 0) 5239 ret = -EINVAL; 5240 else 5241 icsk->icsk_user_timeout = val; 5242 break; 5243 case TCP_NOTSENT_LOWAT: 5244 tp->notsent_lowat = val; 5245 sk->sk_write_space(sk); 5246 break; 5247 case TCP_WINDOW_CLAMP: 5248 ret = tcp_set_window_clamp(sk, val); 5249 break; 5250 default: 5251 ret = -EINVAL; 5252 } 5253 } 5254 #endif 5255 } else { 5256 ret = -EINVAL; 5257 } 5258 return ret; 5259 } 5260 5261 static int _bpf_getsockopt(struct sock *sk, int level, int optname, 5262 char *optval, int optlen) 5263 { 5264 if (!sk_fullsock(sk)) 5265 goto err_clear; 5266 5267 sock_owned_by_me(sk); 5268 5269 if (level == SOL_SOCKET) { 5270 if (optlen != sizeof(int)) 5271 goto err_clear; 5272 5273 switch (optname) { 5274 case SO_RCVBUF: 5275 *((int *)optval) = sk->sk_rcvbuf; 5276 break; 5277 case SO_SNDBUF: 5278 *((int *)optval) = sk->sk_sndbuf; 5279 break; 5280 case SO_MARK: 5281 *((int *)optval) = sk->sk_mark; 5282 break; 5283 case SO_PRIORITY: 5284 *((int *)optval) = sk->sk_priority; 5285 break; 5286 case SO_BINDTOIFINDEX: 5287 *((int *)optval) = sk->sk_bound_dev_if; 5288 break; 5289 case SO_REUSEPORT: 5290 *((int *)optval) = sk->sk_reuseport; 5291 break; 5292 case SO_TXREHASH: 5293 *((int *)optval) = sk->sk_txrehash; 5294 break; 5295 default: 5296 goto err_clear; 5297 } 5298 #ifdef CONFIG_INET 5299 } else if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) { 5300 struct inet_connection_sock *icsk; 5301 struct tcp_sock *tp; 5302 5303 switch (optname) { 5304 case TCP_CONGESTION: 5305 icsk = inet_csk(sk); 5306 5307 if (!icsk->icsk_ca_ops || optlen <= 1) 5308 goto err_clear; 5309 strncpy(optval, icsk->icsk_ca_ops->name, optlen); 5310 optval[optlen - 1] = 0; 5311 break; 5312 case TCP_SAVED_SYN: 5313 tp = tcp_sk(sk); 5314 5315 if (optlen <= 0 || !tp->saved_syn || 5316 optlen > tcp_saved_syn_len(tp->saved_syn)) 5317 goto err_clear; 5318 memcpy(optval, tp->saved_syn->data, optlen); 5319 break; 5320 default: 5321 goto err_clear; 5322 } 5323 } else if (level == SOL_IP) { 5324 struct inet_sock *inet = inet_sk(sk); 5325 5326 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 5327 goto err_clear; 5328 5329 /* Only some options are supported */ 5330 switch (optname) { 5331 case IP_TOS: 5332 *((int *)optval) = (int)inet->tos; 5333 break; 5334 default: 5335 goto err_clear; 5336 } 5337 #if IS_ENABLED(CONFIG_IPV6) 5338 } else if (level == SOL_IPV6) { 5339 struct ipv6_pinfo *np = inet6_sk(sk); 5340 5341 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 5342 goto err_clear; 5343 5344 /* Only some options are supported */ 5345 switch (optname) { 5346 case IPV6_TCLASS: 5347 *((int *)optval) = (int)np->tclass; 5348 break; 5349 default: 5350 goto err_clear; 5351 } 5352 #endif 5353 #endif 5354 } else { 5355 goto err_clear; 5356 } 5357 return 0; 5358 err_clear: 5359 memset(optval, 0, optlen); 5360 return -EINVAL; 5361 } 5362 5363 BPF_CALL_5(bpf_sk_setsockopt, struct sock *, sk, int, level, 5364 int, optname, char *, optval, int, optlen) 5365 { 5366 if (level == SOL_TCP && optname == TCP_CONGESTION) { 5367 if (optlen >= sizeof("cdg") - 1 && 5368 !strncmp("cdg", optval, optlen)) 5369 return -ENOTSUPP; 5370 } 5371 5372 return _bpf_setsockopt(sk, level, optname, optval, optlen); 5373 } 5374 5375 const struct bpf_func_proto bpf_sk_setsockopt_proto = { 5376 .func = bpf_sk_setsockopt, 5377 .gpl_only = false, 5378 .ret_type = RET_INTEGER, 5379 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5380 .arg2_type = ARG_ANYTHING, 5381 .arg3_type = ARG_ANYTHING, 5382 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5383 .arg5_type = ARG_CONST_SIZE, 5384 }; 5385 5386 BPF_CALL_5(bpf_sk_getsockopt, struct sock *, sk, int, level, 5387 int, optname, char *, optval, int, optlen) 5388 { 5389 return _bpf_getsockopt(sk, level, optname, optval, optlen); 5390 } 5391 5392 const struct bpf_func_proto bpf_sk_getsockopt_proto = { 5393 .func = bpf_sk_getsockopt, 5394 .gpl_only = false, 5395 .ret_type = RET_INTEGER, 5396 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 5397 .arg2_type = ARG_ANYTHING, 5398 .arg3_type = ARG_ANYTHING, 5399 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5400 .arg5_type = ARG_CONST_SIZE, 5401 }; 5402 5403 BPF_CALL_5(bpf_sock_addr_setsockopt, struct bpf_sock_addr_kern *, ctx, 5404 int, level, int, optname, char *, optval, int, optlen) 5405 { 5406 return _bpf_setsockopt(ctx->sk, level, optname, optval, optlen); 5407 } 5408 5409 static const struct bpf_func_proto bpf_sock_addr_setsockopt_proto = { 5410 .func = bpf_sock_addr_setsockopt, 5411 .gpl_only = false, 5412 .ret_type = RET_INTEGER, 5413 .arg1_type = ARG_PTR_TO_CTX, 5414 .arg2_type = ARG_ANYTHING, 5415 .arg3_type = ARG_ANYTHING, 5416 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5417 .arg5_type = ARG_CONST_SIZE, 5418 }; 5419 5420 BPF_CALL_5(bpf_sock_addr_getsockopt, struct bpf_sock_addr_kern *, ctx, 5421 int, level, int, optname, char *, optval, int, optlen) 5422 { 5423 return _bpf_getsockopt(ctx->sk, level, optname, optval, optlen); 5424 } 5425 5426 static const struct bpf_func_proto bpf_sock_addr_getsockopt_proto = { 5427 .func = bpf_sock_addr_getsockopt, 5428 .gpl_only = false, 5429 .ret_type = RET_INTEGER, 5430 .arg1_type = ARG_PTR_TO_CTX, 5431 .arg2_type = ARG_ANYTHING, 5432 .arg3_type = ARG_ANYTHING, 5433 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5434 .arg5_type = ARG_CONST_SIZE, 5435 }; 5436 5437 BPF_CALL_5(bpf_sock_ops_setsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5438 int, level, int, optname, char *, optval, int, optlen) 5439 { 5440 return _bpf_setsockopt(bpf_sock->sk, level, optname, optval, optlen); 5441 } 5442 5443 static const struct bpf_func_proto bpf_sock_ops_setsockopt_proto = { 5444 .func = bpf_sock_ops_setsockopt, 5445 .gpl_only = false, 5446 .ret_type = RET_INTEGER, 5447 .arg1_type = ARG_PTR_TO_CTX, 5448 .arg2_type = ARG_ANYTHING, 5449 .arg3_type = ARG_ANYTHING, 5450 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5451 .arg5_type = ARG_CONST_SIZE, 5452 }; 5453 5454 static int bpf_sock_ops_get_syn(struct bpf_sock_ops_kern *bpf_sock, 5455 int optname, const u8 **start) 5456 { 5457 struct sk_buff *syn_skb = bpf_sock->syn_skb; 5458 const u8 *hdr_start; 5459 int ret; 5460 5461 if (syn_skb) { 5462 /* sk is a request_sock here */ 5463 5464 if (optname == TCP_BPF_SYN) { 5465 hdr_start = syn_skb->data; 5466 ret = tcp_hdrlen(syn_skb); 5467 } else if (optname == TCP_BPF_SYN_IP) { 5468 hdr_start = skb_network_header(syn_skb); 5469 ret = skb_network_header_len(syn_skb) + 5470 tcp_hdrlen(syn_skb); 5471 } else { 5472 /* optname == TCP_BPF_SYN_MAC */ 5473 hdr_start = skb_mac_header(syn_skb); 5474 ret = skb_mac_header_len(syn_skb) + 5475 skb_network_header_len(syn_skb) + 5476 tcp_hdrlen(syn_skb); 5477 } 5478 } else { 5479 struct sock *sk = bpf_sock->sk; 5480 struct saved_syn *saved_syn; 5481 5482 if (sk->sk_state == TCP_NEW_SYN_RECV) 5483 /* synack retransmit. bpf_sock->syn_skb will 5484 * not be available. It has to resort to 5485 * saved_syn (if it is saved). 5486 */ 5487 saved_syn = inet_reqsk(sk)->saved_syn; 5488 else 5489 saved_syn = tcp_sk(sk)->saved_syn; 5490 5491 if (!saved_syn) 5492 return -ENOENT; 5493 5494 if (optname == TCP_BPF_SYN) { 5495 hdr_start = saved_syn->data + 5496 saved_syn->mac_hdrlen + 5497 saved_syn->network_hdrlen; 5498 ret = saved_syn->tcp_hdrlen; 5499 } else if (optname == TCP_BPF_SYN_IP) { 5500 hdr_start = saved_syn->data + 5501 saved_syn->mac_hdrlen; 5502 ret = saved_syn->network_hdrlen + 5503 saved_syn->tcp_hdrlen; 5504 } else { 5505 /* optname == TCP_BPF_SYN_MAC */ 5506 5507 /* TCP_SAVE_SYN may not have saved the mac hdr */ 5508 if (!saved_syn->mac_hdrlen) 5509 return -ENOENT; 5510 5511 hdr_start = saved_syn->data; 5512 ret = saved_syn->mac_hdrlen + 5513 saved_syn->network_hdrlen + 5514 saved_syn->tcp_hdrlen; 5515 } 5516 } 5517 5518 *start = hdr_start; 5519 return ret; 5520 } 5521 5522 BPF_CALL_5(bpf_sock_ops_getsockopt, struct bpf_sock_ops_kern *, bpf_sock, 5523 int, level, int, optname, char *, optval, int, optlen) 5524 { 5525 if (IS_ENABLED(CONFIG_INET) && level == SOL_TCP && 5526 optname >= TCP_BPF_SYN && optname <= TCP_BPF_SYN_MAC) { 5527 int ret, copy_len = 0; 5528 const u8 *start; 5529 5530 ret = bpf_sock_ops_get_syn(bpf_sock, optname, &start); 5531 if (ret > 0) { 5532 copy_len = ret; 5533 if (optlen < copy_len) { 5534 copy_len = optlen; 5535 ret = -ENOSPC; 5536 } 5537 5538 memcpy(optval, start, copy_len); 5539 } 5540 5541 /* Zero out unused buffer at the end */ 5542 memset(optval + copy_len, 0, optlen - copy_len); 5543 5544 return ret; 5545 } 5546 5547 return _bpf_getsockopt(bpf_sock->sk, level, optname, optval, optlen); 5548 } 5549 5550 static const struct bpf_func_proto bpf_sock_ops_getsockopt_proto = { 5551 .func = bpf_sock_ops_getsockopt, 5552 .gpl_only = false, 5553 .ret_type = RET_INTEGER, 5554 .arg1_type = ARG_PTR_TO_CTX, 5555 .arg2_type = ARG_ANYTHING, 5556 .arg3_type = ARG_ANYTHING, 5557 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 5558 .arg5_type = ARG_CONST_SIZE, 5559 }; 5560 5561 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock, 5562 int, argval) 5563 { 5564 struct sock *sk = bpf_sock->sk; 5565 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS; 5566 5567 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk)) 5568 return -EINVAL; 5569 5570 tcp_sk(sk)->bpf_sock_ops_cb_flags = val; 5571 5572 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS); 5573 } 5574 5575 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = { 5576 .func = bpf_sock_ops_cb_flags_set, 5577 .gpl_only = false, 5578 .ret_type = RET_INTEGER, 5579 .arg1_type = ARG_PTR_TO_CTX, 5580 .arg2_type = ARG_ANYTHING, 5581 }; 5582 5583 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly; 5584 EXPORT_SYMBOL_GPL(ipv6_bpf_stub); 5585 5586 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr, 5587 int, addr_len) 5588 { 5589 #ifdef CONFIG_INET 5590 struct sock *sk = ctx->sk; 5591 u32 flags = BIND_FROM_BPF; 5592 int err; 5593 5594 err = -EINVAL; 5595 if (addr_len < offsetofend(struct sockaddr, sa_family)) 5596 return err; 5597 if (addr->sa_family == AF_INET) { 5598 if (addr_len < sizeof(struct sockaddr_in)) 5599 return err; 5600 if (((struct sockaddr_in *)addr)->sin_port == htons(0)) 5601 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5602 return __inet_bind(sk, addr, addr_len, flags); 5603 #if IS_ENABLED(CONFIG_IPV6) 5604 } else if (addr->sa_family == AF_INET6) { 5605 if (addr_len < SIN6_LEN_RFC2133) 5606 return err; 5607 if (((struct sockaddr_in6 *)addr)->sin6_port == htons(0)) 5608 flags |= BIND_FORCE_ADDRESS_NO_PORT; 5609 /* ipv6_bpf_stub cannot be NULL, since it's called from 5610 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded 5611 */ 5612 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, flags); 5613 #endif /* CONFIG_IPV6 */ 5614 } 5615 #endif /* CONFIG_INET */ 5616 5617 return -EAFNOSUPPORT; 5618 } 5619 5620 static const struct bpf_func_proto bpf_bind_proto = { 5621 .func = bpf_bind, 5622 .gpl_only = false, 5623 .ret_type = RET_INTEGER, 5624 .arg1_type = ARG_PTR_TO_CTX, 5625 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 5626 .arg3_type = ARG_CONST_SIZE, 5627 }; 5628 5629 #ifdef CONFIG_XFRM 5630 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index, 5631 struct bpf_xfrm_state *, to, u32, size, u64, flags) 5632 { 5633 const struct sec_path *sp = skb_sec_path(skb); 5634 const struct xfrm_state *x; 5635 5636 if (!sp || unlikely(index >= sp->len || flags)) 5637 goto err_clear; 5638 5639 x = sp->xvec[index]; 5640 5641 if (unlikely(size != sizeof(struct bpf_xfrm_state))) 5642 goto err_clear; 5643 5644 to->reqid = x->props.reqid; 5645 to->spi = x->id.spi; 5646 to->family = x->props.family; 5647 to->ext = 0; 5648 5649 if (to->family == AF_INET6) { 5650 memcpy(to->remote_ipv6, x->props.saddr.a6, 5651 sizeof(to->remote_ipv6)); 5652 } else { 5653 to->remote_ipv4 = x->props.saddr.a4; 5654 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 5655 } 5656 5657 return 0; 5658 err_clear: 5659 memset(to, 0, size); 5660 return -EINVAL; 5661 } 5662 5663 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = { 5664 .func = bpf_skb_get_xfrm_state, 5665 .gpl_only = false, 5666 .ret_type = RET_INTEGER, 5667 .arg1_type = ARG_PTR_TO_CTX, 5668 .arg2_type = ARG_ANYTHING, 5669 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 5670 .arg4_type = ARG_CONST_SIZE, 5671 .arg5_type = ARG_ANYTHING, 5672 }; 5673 #endif 5674 5675 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6) 5676 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, 5677 const struct neighbour *neigh, 5678 const struct net_device *dev, u32 mtu) 5679 { 5680 memcpy(params->dmac, neigh->ha, ETH_ALEN); 5681 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 5682 params->h_vlan_TCI = 0; 5683 params->h_vlan_proto = 0; 5684 if (mtu) 5685 params->mtu_result = mtu; /* union with tot_len */ 5686 5687 return 0; 5688 } 5689 #endif 5690 5691 #if IS_ENABLED(CONFIG_INET) 5692 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5693 u32 flags, bool check_mtu) 5694 { 5695 struct fib_nh_common *nhc; 5696 struct in_device *in_dev; 5697 struct neighbour *neigh; 5698 struct net_device *dev; 5699 struct fib_result res; 5700 struct flowi4 fl4; 5701 u32 mtu = 0; 5702 int err; 5703 5704 dev = dev_get_by_index_rcu(net, params->ifindex); 5705 if (unlikely(!dev)) 5706 return -ENODEV; 5707 5708 /* verify forwarding is enabled on this interface */ 5709 in_dev = __in_dev_get_rcu(dev); 5710 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev))) 5711 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5712 5713 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5714 fl4.flowi4_iif = 1; 5715 fl4.flowi4_oif = params->ifindex; 5716 } else { 5717 fl4.flowi4_iif = params->ifindex; 5718 fl4.flowi4_oif = 0; 5719 } 5720 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK; 5721 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 5722 fl4.flowi4_flags = 0; 5723 5724 fl4.flowi4_proto = params->l4_protocol; 5725 fl4.daddr = params->ipv4_dst; 5726 fl4.saddr = params->ipv4_src; 5727 fl4.fl4_sport = params->sport; 5728 fl4.fl4_dport = params->dport; 5729 fl4.flowi4_multipath_hash = 0; 5730 5731 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5732 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5733 struct fib_table *tb; 5734 5735 tb = fib_get_table(net, tbid); 5736 if (unlikely(!tb)) 5737 return BPF_FIB_LKUP_RET_NOT_FWDED; 5738 5739 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 5740 } else { 5741 fl4.flowi4_mark = 0; 5742 fl4.flowi4_secid = 0; 5743 fl4.flowi4_tun_key.tun_id = 0; 5744 fl4.flowi4_uid = sock_net_uid(net, NULL); 5745 5746 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF); 5747 } 5748 5749 if (err) { 5750 /* map fib lookup errors to RTN_ type */ 5751 if (err == -EINVAL) 5752 return BPF_FIB_LKUP_RET_BLACKHOLE; 5753 if (err == -EHOSTUNREACH) 5754 return BPF_FIB_LKUP_RET_UNREACHABLE; 5755 if (err == -EACCES) 5756 return BPF_FIB_LKUP_RET_PROHIBIT; 5757 5758 return BPF_FIB_LKUP_RET_NOT_FWDED; 5759 } 5760 5761 if (res.type != RTN_UNICAST) 5762 return BPF_FIB_LKUP_RET_NOT_FWDED; 5763 5764 if (fib_info_num_path(res.fi) > 1) 5765 fib_select_path(net, &res, &fl4, NULL); 5766 5767 if (check_mtu) { 5768 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst); 5769 if (params->tot_len > mtu) { 5770 params->mtu_result = mtu; /* union with tot_len */ 5771 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5772 } 5773 } 5774 5775 nhc = res.nhc; 5776 5777 /* do not handle lwt encaps right now */ 5778 if (nhc->nhc_lwtstate) 5779 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5780 5781 dev = nhc->nhc_dev; 5782 5783 params->rt_metric = res.fi->fib_priority; 5784 params->ifindex = dev->ifindex; 5785 5786 /* xdp and cls_bpf programs are run in RCU-bh so 5787 * rcu_read_lock_bh is not needed here 5788 */ 5789 if (likely(nhc->nhc_gw_family != AF_INET6)) { 5790 if (nhc->nhc_gw_family) 5791 params->ipv4_dst = nhc->nhc_gw.ipv4; 5792 5793 neigh = __ipv4_neigh_lookup_noref(dev, 5794 (__force u32)params->ipv4_dst); 5795 } else { 5796 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst; 5797 5798 params->family = AF_INET6; 5799 *dst = nhc->nhc_gw.ipv6; 5800 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5801 } 5802 5803 if (!neigh) 5804 return BPF_FIB_LKUP_RET_NO_NEIGH; 5805 5806 return bpf_fib_set_fwd_params(params, neigh, dev, mtu); 5807 } 5808 #endif 5809 5810 #if IS_ENABLED(CONFIG_IPV6) 5811 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 5812 u32 flags, bool check_mtu) 5813 { 5814 struct in6_addr *src = (struct in6_addr *) params->ipv6_src; 5815 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst; 5816 struct fib6_result res = {}; 5817 struct neighbour *neigh; 5818 struct net_device *dev; 5819 struct inet6_dev *idev; 5820 struct flowi6 fl6; 5821 int strict = 0; 5822 int oif, err; 5823 u32 mtu = 0; 5824 5825 /* link local addresses are never forwarded */ 5826 if (rt6_need_strict(dst) || rt6_need_strict(src)) 5827 return BPF_FIB_LKUP_RET_NOT_FWDED; 5828 5829 dev = dev_get_by_index_rcu(net, params->ifindex); 5830 if (unlikely(!dev)) 5831 return -ENODEV; 5832 5833 idev = __in6_dev_get_safely(dev); 5834 if (unlikely(!idev || !idev->cnf.forwarding)) 5835 return BPF_FIB_LKUP_RET_FWD_DISABLED; 5836 5837 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 5838 fl6.flowi6_iif = 1; 5839 oif = fl6.flowi6_oif = params->ifindex; 5840 } else { 5841 oif = fl6.flowi6_iif = params->ifindex; 5842 fl6.flowi6_oif = 0; 5843 strict = RT6_LOOKUP_F_HAS_SADDR; 5844 } 5845 fl6.flowlabel = params->flowinfo; 5846 fl6.flowi6_scope = 0; 5847 fl6.flowi6_flags = 0; 5848 fl6.mp_hash = 0; 5849 5850 fl6.flowi6_proto = params->l4_protocol; 5851 fl6.daddr = *dst; 5852 fl6.saddr = *src; 5853 fl6.fl6_sport = params->sport; 5854 fl6.fl6_dport = params->dport; 5855 5856 if (flags & BPF_FIB_LOOKUP_DIRECT) { 5857 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 5858 struct fib6_table *tb; 5859 5860 tb = ipv6_stub->fib6_get_table(net, tbid); 5861 if (unlikely(!tb)) 5862 return BPF_FIB_LKUP_RET_NOT_FWDED; 5863 5864 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res, 5865 strict); 5866 } else { 5867 fl6.flowi6_mark = 0; 5868 fl6.flowi6_secid = 0; 5869 fl6.flowi6_tun_key.tun_id = 0; 5870 fl6.flowi6_uid = sock_net_uid(net, NULL); 5871 5872 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict); 5873 } 5874 5875 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) || 5876 res.f6i == net->ipv6.fib6_null_entry)) 5877 return BPF_FIB_LKUP_RET_NOT_FWDED; 5878 5879 switch (res.fib6_type) { 5880 /* only unicast is forwarded */ 5881 case RTN_UNICAST: 5882 break; 5883 case RTN_BLACKHOLE: 5884 return BPF_FIB_LKUP_RET_BLACKHOLE; 5885 case RTN_UNREACHABLE: 5886 return BPF_FIB_LKUP_RET_UNREACHABLE; 5887 case RTN_PROHIBIT: 5888 return BPF_FIB_LKUP_RET_PROHIBIT; 5889 default: 5890 return BPF_FIB_LKUP_RET_NOT_FWDED; 5891 } 5892 5893 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif, 5894 fl6.flowi6_oif != 0, NULL, strict); 5895 5896 if (check_mtu) { 5897 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src); 5898 if (params->tot_len > mtu) { 5899 params->mtu_result = mtu; /* union with tot_len */ 5900 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 5901 } 5902 } 5903 5904 if (res.nh->fib_nh_lws) 5905 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 5906 5907 if (res.nh->fib_nh_gw_family) 5908 *dst = res.nh->fib_nh_gw6; 5909 5910 dev = res.nh->fib_nh_dev; 5911 params->rt_metric = res.f6i->fib6_metric; 5912 params->ifindex = dev->ifindex; 5913 5914 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is 5915 * not needed here. 5916 */ 5917 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 5918 if (!neigh) 5919 return BPF_FIB_LKUP_RET_NO_NEIGH; 5920 5921 return bpf_fib_set_fwd_params(params, neigh, dev, mtu); 5922 } 5923 #endif 5924 5925 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx, 5926 struct bpf_fib_lookup *, params, int, plen, u32, flags) 5927 { 5928 if (plen < sizeof(*params)) 5929 return -EINVAL; 5930 5931 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 5932 return -EINVAL; 5933 5934 switch (params->family) { 5935 #if IS_ENABLED(CONFIG_INET) 5936 case AF_INET: 5937 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params, 5938 flags, true); 5939 #endif 5940 #if IS_ENABLED(CONFIG_IPV6) 5941 case AF_INET6: 5942 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params, 5943 flags, true); 5944 #endif 5945 } 5946 return -EAFNOSUPPORT; 5947 } 5948 5949 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = { 5950 .func = bpf_xdp_fib_lookup, 5951 .gpl_only = true, 5952 .ret_type = RET_INTEGER, 5953 .arg1_type = ARG_PTR_TO_CTX, 5954 .arg2_type = ARG_PTR_TO_MEM, 5955 .arg3_type = ARG_CONST_SIZE, 5956 .arg4_type = ARG_ANYTHING, 5957 }; 5958 5959 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb, 5960 struct bpf_fib_lookup *, params, int, plen, u32, flags) 5961 { 5962 struct net *net = dev_net(skb->dev); 5963 int rc = -EAFNOSUPPORT; 5964 bool check_mtu = false; 5965 5966 if (plen < sizeof(*params)) 5967 return -EINVAL; 5968 5969 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 5970 return -EINVAL; 5971 5972 if (params->tot_len) 5973 check_mtu = true; 5974 5975 switch (params->family) { 5976 #if IS_ENABLED(CONFIG_INET) 5977 case AF_INET: 5978 rc = bpf_ipv4_fib_lookup(net, params, flags, check_mtu); 5979 break; 5980 #endif 5981 #if IS_ENABLED(CONFIG_IPV6) 5982 case AF_INET6: 5983 rc = bpf_ipv6_fib_lookup(net, params, flags, check_mtu); 5984 break; 5985 #endif 5986 } 5987 5988 if (rc == BPF_FIB_LKUP_RET_SUCCESS && !check_mtu) { 5989 struct net_device *dev; 5990 5991 /* When tot_len isn't provided by user, check skb 5992 * against MTU of FIB lookup resulting net_device 5993 */ 5994 dev = dev_get_by_index_rcu(net, params->ifindex); 5995 if (!is_skb_forwardable(dev, skb)) 5996 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED; 5997 5998 params->mtu_result = dev->mtu; /* union with tot_len */ 5999 } 6000 6001 return rc; 6002 } 6003 6004 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = { 6005 .func = bpf_skb_fib_lookup, 6006 .gpl_only = true, 6007 .ret_type = RET_INTEGER, 6008 .arg1_type = ARG_PTR_TO_CTX, 6009 .arg2_type = ARG_PTR_TO_MEM, 6010 .arg3_type = ARG_CONST_SIZE, 6011 .arg4_type = ARG_ANYTHING, 6012 }; 6013 6014 static struct net_device *__dev_via_ifindex(struct net_device *dev_curr, 6015 u32 ifindex) 6016 { 6017 struct net *netns = dev_net(dev_curr); 6018 6019 /* Non-redirect use-cases can use ifindex=0 and save ifindex lookup */ 6020 if (ifindex == 0) 6021 return dev_curr; 6022 6023 return dev_get_by_index_rcu(netns, ifindex); 6024 } 6025 6026 BPF_CALL_5(bpf_skb_check_mtu, struct sk_buff *, skb, 6027 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 6028 { 6029 int ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 6030 struct net_device *dev = skb->dev; 6031 int skb_len, dev_len; 6032 int mtu; 6033 6034 if (unlikely(flags & ~(BPF_MTU_CHK_SEGS))) 6035 return -EINVAL; 6036 6037 if (unlikely(flags & BPF_MTU_CHK_SEGS && (len_diff || *mtu_len))) 6038 return -EINVAL; 6039 6040 dev = __dev_via_ifindex(dev, ifindex); 6041 if (unlikely(!dev)) 6042 return -ENODEV; 6043 6044 mtu = READ_ONCE(dev->mtu); 6045 6046 dev_len = mtu + dev->hard_header_len; 6047 6048 /* If set use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 6049 skb_len = *mtu_len ? *mtu_len + dev->hard_header_len : skb->len; 6050 6051 skb_len += len_diff; /* minus result pass check */ 6052 if (skb_len <= dev_len) { 6053 ret = BPF_MTU_CHK_RET_SUCCESS; 6054 goto out; 6055 } 6056 /* At this point, skb->len exceed MTU, but as it include length of all 6057 * segments, it can still be below MTU. The SKB can possibly get 6058 * re-segmented in transmit path (see validate_xmit_skb). Thus, user 6059 * must choose if segs are to be MTU checked. 6060 */ 6061 if (skb_is_gso(skb)) { 6062 ret = BPF_MTU_CHK_RET_SUCCESS; 6063 6064 if (flags & BPF_MTU_CHK_SEGS && 6065 !skb_gso_validate_network_len(skb, mtu)) 6066 ret = BPF_MTU_CHK_RET_SEGS_TOOBIG; 6067 } 6068 out: 6069 /* BPF verifier guarantees valid pointer */ 6070 *mtu_len = mtu; 6071 6072 return ret; 6073 } 6074 6075 BPF_CALL_5(bpf_xdp_check_mtu, struct xdp_buff *, xdp, 6076 u32, ifindex, u32 *, mtu_len, s32, len_diff, u64, flags) 6077 { 6078 struct net_device *dev = xdp->rxq->dev; 6079 int xdp_len = xdp->data_end - xdp->data; 6080 int ret = BPF_MTU_CHK_RET_SUCCESS; 6081 int mtu, dev_len; 6082 6083 /* XDP variant doesn't support multi-buffer segment check (yet) */ 6084 if (unlikely(flags)) 6085 return -EINVAL; 6086 6087 dev = __dev_via_ifindex(dev, ifindex); 6088 if (unlikely(!dev)) 6089 return -ENODEV; 6090 6091 mtu = READ_ONCE(dev->mtu); 6092 6093 /* Add L2-header as dev MTU is L3 size */ 6094 dev_len = mtu + dev->hard_header_len; 6095 6096 /* Use *mtu_len as input, L3 as iph->tot_len (like fib_lookup) */ 6097 if (*mtu_len) 6098 xdp_len = *mtu_len + dev->hard_header_len; 6099 6100 xdp_len += len_diff; /* minus result pass check */ 6101 if (xdp_len > dev_len) 6102 ret = BPF_MTU_CHK_RET_FRAG_NEEDED; 6103 6104 /* BPF verifier guarantees valid pointer */ 6105 *mtu_len = mtu; 6106 6107 return ret; 6108 } 6109 6110 static const struct bpf_func_proto bpf_skb_check_mtu_proto = { 6111 .func = bpf_skb_check_mtu, 6112 .gpl_only = true, 6113 .ret_type = RET_INTEGER, 6114 .arg1_type = ARG_PTR_TO_CTX, 6115 .arg2_type = ARG_ANYTHING, 6116 .arg3_type = ARG_PTR_TO_INT, 6117 .arg4_type = ARG_ANYTHING, 6118 .arg5_type = ARG_ANYTHING, 6119 }; 6120 6121 static const struct bpf_func_proto bpf_xdp_check_mtu_proto = { 6122 .func = bpf_xdp_check_mtu, 6123 .gpl_only = true, 6124 .ret_type = RET_INTEGER, 6125 .arg1_type = ARG_PTR_TO_CTX, 6126 .arg2_type = ARG_ANYTHING, 6127 .arg3_type = ARG_PTR_TO_INT, 6128 .arg4_type = ARG_ANYTHING, 6129 .arg5_type = ARG_ANYTHING, 6130 }; 6131 6132 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6133 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len) 6134 { 6135 int err; 6136 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr; 6137 6138 if (!seg6_validate_srh(srh, len, false)) 6139 return -EINVAL; 6140 6141 switch (type) { 6142 case BPF_LWT_ENCAP_SEG6_INLINE: 6143 if (skb->protocol != htons(ETH_P_IPV6)) 6144 return -EBADMSG; 6145 6146 err = seg6_do_srh_inline(skb, srh); 6147 break; 6148 case BPF_LWT_ENCAP_SEG6: 6149 skb_reset_inner_headers(skb); 6150 skb->encapsulation = 1; 6151 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6); 6152 break; 6153 default: 6154 return -EINVAL; 6155 } 6156 6157 bpf_compute_data_pointers(skb); 6158 if (err) 6159 return err; 6160 6161 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 6162 skb_set_transport_header(skb, sizeof(struct ipv6hdr)); 6163 6164 return seg6_lookup_nexthop(skb, NULL, 0); 6165 } 6166 #endif /* CONFIG_IPV6_SEG6_BPF */ 6167 6168 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6169 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, 6170 bool ingress) 6171 { 6172 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress); 6173 } 6174 #endif 6175 6176 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr, 6177 u32, len) 6178 { 6179 switch (type) { 6180 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6181 case BPF_LWT_ENCAP_SEG6: 6182 case BPF_LWT_ENCAP_SEG6_INLINE: 6183 return bpf_push_seg6_encap(skb, type, hdr, len); 6184 #endif 6185 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6186 case BPF_LWT_ENCAP_IP: 6187 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */); 6188 #endif 6189 default: 6190 return -EINVAL; 6191 } 6192 } 6193 6194 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type, 6195 void *, hdr, u32, len) 6196 { 6197 switch (type) { 6198 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 6199 case BPF_LWT_ENCAP_IP: 6200 return bpf_push_ip_encap(skb, hdr, len, false /* egress */); 6201 #endif 6202 default: 6203 return -EINVAL; 6204 } 6205 } 6206 6207 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = { 6208 .func = bpf_lwt_in_push_encap, 6209 .gpl_only = false, 6210 .ret_type = RET_INTEGER, 6211 .arg1_type = ARG_PTR_TO_CTX, 6212 .arg2_type = ARG_ANYTHING, 6213 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6214 .arg4_type = ARG_CONST_SIZE 6215 }; 6216 6217 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = { 6218 .func = bpf_lwt_xmit_push_encap, 6219 .gpl_only = false, 6220 .ret_type = RET_INTEGER, 6221 .arg1_type = ARG_PTR_TO_CTX, 6222 .arg2_type = ARG_ANYTHING, 6223 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6224 .arg4_type = ARG_CONST_SIZE 6225 }; 6226 6227 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6228 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset, 6229 const void *, from, u32, len) 6230 { 6231 struct seg6_bpf_srh_state *srh_state = 6232 this_cpu_ptr(&seg6_bpf_srh_states); 6233 struct ipv6_sr_hdr *srh = srh_state->srh; 6234 void *srh_tlvs, *srh_end, *ptr; 6235 int srhoff = 0; 6236 6237 if (srh == NULL) 6238 return -EINVAL; 6239 6240 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4)); 6241 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen); 6242 6243 ptr = skb->data + offset; 6244 if (ptr >= srh_tlvs && ptr + len <= srh_end) 6245 srh_state->valid = false; 6246 else if (ptr < (void *)&srh->flags || 6247 ptr + len > (void *)&srh->segments) 6248 return -EFAULT; 6249 6250 if (unlikely(bpf_try_make_writable(skb, offset + len))) 6251 return -EFAULT; 6252 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 6253 return -EINVAL; 6254 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6255 6256 memcpy(skb->data + offset, from, len); 6257 return 0; 6258 } 6259 6260 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = { 6261 .func = bpf_lwt_seg6_store_bytes, 6262 .gpl_only = false, 6263 .ret_type = RET_INTEGER, 6264 .arg1_type = ARG_PTR_TO_CTX, 6265 .arg2_type = ARG_ANYTHING, 6266 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6267 .arg4_type = ARG_CONST_SIZE 6268 }; 6269 6270 static void bpf_update_srh_state(struct sk_buff *skb) 6271 { 6272 struct seg6_bpf_srh_state *srh_state = 6273 this_cpu_ptr(&seg6_bpf_srh_states); 6274 int srhoff = 0; 6275 6276 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) { 6277 srh_state->srh = NULL; 6278 } else { 6279 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6280 srh_state->hdrlen = srh_state->srh->hdrlen << 3; 6281 srh_state->valid = true; 6282 } 6283 } 6284 6285 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb, 6286 u32, action, void *, param, u32, param_len) 6287 { 6288 struct seg6_bpf_srh_state *srh_state = 6289 this_cpu_ptr(&seg6_bpf_srh_states); 6290 int hdroff = 0; 6291 int err; 6292 6293 switch (action) { 6294 case SEG6_LOCAL_ACTION_END_X: 6295 if (!seg6_bpf_has_valid_srh(skb)) 6296 return -EBADMSG; 6297 if (param_len != sizeof(struct in6_addr)) 6298 return -EINVAL; 6299 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0); 6300 case SEG6_LOCAL_ACTION_END_T: 6301 if (!seg6_bpf_has_valid_srh(skb)) 6302 return -EBADMSG; 6303 if (param_len != sizeof(int)) 6304 return -EINVAL; 6305 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 6306 case SEG6_LOCAL_ACTION_END_DT6: 6307 if (!seg6_bpf_has_valid_srh(skb)) 6308 return -EBADMSG; 6309 if (param_len != sizeof(int)) 6310 return -EINVAL; 6311 6312 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0) 6313 return -EBADMSG; 6314 if (!pskb_pull(skb, hdroff)) 6315 return -EBADMSG; 6316 6317 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff); 6318 skb_reset_network_header(skb); 6319 skb_reset_transport_header(skb); 6320 skb->encapsulation = 0; 6321 6322 bpf_compute_data_pointers(skb); 6323 bpf_update_srh_state(skb); 6324 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 6325 case SEG6_LOCAL_ACTION_END_B6: 6326 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 6327 return -EBADMSG; 6328 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE, 6329 param, param_len); 6330 if (!err) 6331 bpf_update_srh_state(skb); 6332 6333 return err; 6334 case SEG6_LOCAL_ACTION_END_B6_ENCAP: 6335 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 6336 return -EBADMSG; 6337 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6, 6338 param, param_len); 6339 if (!err) 6340 bpf_update_srh_state(skb); 6341 6342 return err; 6343 default: 6344 return -EINVAL; 6345 } 6346 } 6347 6348 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = { 6349 .func = bpf_lwt_seg6_action, 6350 .gpl_only = false, 6351 .ret_type = RET_INTEGER, 6352 .arg1_type = ARG_PTR_TO_CTX, 6353 .arg2_type = ARG_ANYTHING, 6354 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6355 .arg4_type = ARG_CONST_SIZE 6356 }; 6357 6358 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset, 6359 s32, len) 6360 { 6361 struct seg6_bpf_srh_state *srh_state = 6362 this_cpu_ptr(&seg6_bpf_srh_states); 6363 struct ipv6_sr_hdr *srh = srh_state->srh; 6364 void *srh_end, *srh_tlvs, *ptr; 6365 struct ipv6hdr *hdr; 6366 int srhoff = 0; 6367 int ret; 6368 6369 if (unlikely(srh == NULL)) 6370 return -EINVAL; 6371 6372 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) + 6373 ((srh->first_segment + 1) << 4)); 6374 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) + 6375 srh_state->hdrlen); 6376 ptr = skb->data + offset; 6377 6378 if (unlikely(ptr < srh_tlvs || ptr > srh_end)) 6379 return -EFAULT; 6380 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end)) 6381 return -EFAULT; 6382 6383 if (len > 0) { 6384 ret = skb_cow_head(skb, len); 6385 if (unlikely(ret < 0)) 6386 return ret; 6387 6388 ret = bpf_skb_net_hdr_push(skb, offset, len); 6389 } else { 6390 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len); 6391 } 6392 6393 bpf_compute_data_pointers(skb); 6394 if (unlikely(ret < 0)) 6395 return ret; 6396 6397 hdr = (struct ipv6hdr *)skb->data; 6398 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 6399 6400 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 6401 return -EINVAL; 6402 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 6403 srh_state->hdrlen += len; 6404 srh_state->valid = false; 6405 return 0; 6406 } 6407 6408 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = { 6409 .func = bpf_lwt_seg6_adjust_srh, 6410 .gpl_only = false, 6411 .ret_type = RET_INTEGER, 6412 .arg1_type = ARG_PTR_TO_CTX, 6413 .arg2_type = ARG_ANYTHING, 6414 .arg3_type = ARG_ANYTHING, 6415 }; 6416 #endif /* CONFIG_IPV6_SEG6_BPF */ 6417 6418 #ifdef CONFIG_INET 6419 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple, 6420 int dif, int sdif, u8 family, u8 proto) 6421 { 6422 bool refcounted = false; 6423 struct sock *sk = NULL; 6424 6425 if (family == AF_INET) { 6426 __be32 src4 = tuple->ipv4.saddr; 6427 __be32 dst4 = tuple->ipv4.daddr; 6428 6429 if (proto == IPPROTO_TCP) 6430 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0, 6431 src4, tuple->ipv4.sport, 6432 dst4, tuple->ipv4.dport, 6433 dif, sdif, &refcounted); 6434 else 6435 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport, 6436 dst4, tuple->ipv4.dport, 6437 dif, sdif, &udp_table, NULL); 6438 #if IS_ENABLED(CONFIG_IPV6) 6439 } else { 6440 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr; 6441 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr; 6442 6443 if (proto == IPPROTO_TCP) 6444 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0, 6445 src6, tuple->ipv6.sport, 6446 dst6, ntohs(tuple->ipv6.dport), 6447 dif, sdif, &refcounted); 6448 else if (likely(ipv6_bpf_stub)) 6449 sk = ipv6_bpf_stub->udp6_lib_lookup(net, 6450 src6, tuple->ipv6.sport, 6451 dst6, tuple->ipv6.dport, 6452 dif, sdif, 6453 &udp_table, NULL); 6454 #endif 6455 } 6456 6457 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) { 6458 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 6459 sk = NULL; 6460 } 6461 return sk; 6462 } 6463 6464 /* bpf_skc_lookup performs the core lookup for different types of sockets, 6465 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE. 6466 * Returns the socket as an 'unsigned long' to simplify the casting in the 6467 * callers to satisfy BPF_CALL declarations. 6468 */ 6469 static struct sock * 6470 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6471 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 6472 u64 flags) 6473 { 6474 struct sock *sk = NULL; 6475 u8 family = AF_UNSPEC; 6476 struct net *net; 6477 int sdif; 6478 6479 if (len == sizeof(tuple->ipv4)) 6480 family = AF_INET; 6481 else if (len == sizeof(tuple->ipv6)) 6482 family = AF_INET6; 6483 else 6484 return NULL; 6485 6486 if (unlikely(family == AF_UNSPEC || flags || 6487 !((s32)netns_id < 0 || netns_id <= S32_MAX))) 6488 goto out; 6489 6490 if (family == AF_INET) 6491 sdif = inet_sdif(skb); 6492 else 6493 sdif = inet6_sdif(skb); 6494 6495 if ((s32)netns_id < 0) { 6496 net = caller_net; 6497 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 6498 } else { 6499 net = get_net_ns_by_id(caller_net, netns_id); 6500 if (unlikely(!net)) 6501 goto out; 6502 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 6503 put_net(net); 6504 } 6505 6506 out: 6507 return sk; 6508 } 6509 6510 static struct sock * 6511 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6512 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 6513 u64 flags) 6514 { 6515 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net, 6516 ifindex, proto, netns_id, flags); 6517 6518 if (sk) { 6519 sk = sk_to_full_sk(sk); 6520 if (!sk_fullsock(sk)) { 6521 sock_gen_put(sk); 6522 return NULL; 6523 } 6524 } 6525 6526 return sk; 6527 } 6528 6529 static struct sock * 6530 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6531 u8 proto, u64 netns_id, u64 flags) 6532 { 6533 struct net *caller_net; 6534 int ifindex; 6535 6536 if (skb->dev) { 6537 caller_net = dev_net(skb->dev); 6538 ifindex = skb->dev->ifindex; 6539 } else { 6540 caller_net = sock_net(skb->sk); 6541 ifindex = 0; 6542 } 6543 6544 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto, 6545 netns_id, flags); 6546 } 6547 6548 static struct sock * 6549 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 6550 u8 proto, u64 netns_id, u64 flags) 6551 { 6552 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id, 6553 flags); 6554 6555 if (sk) { 6556 sk = sk_to_full_sk(sk); 6557 if (!sk_fullsock(sk)) { 6558 sock_gen_put(sk); 6559 return NULL; 6560 } 6561 } 6562 6563 return sk; 6564 } 6565 6566 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb, 6567 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6568 { 6569 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP, 6570 netns_id, flags); 6571 } 6572 6573 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = { 6574 .func = bpf_skc_lookup_tcp, 6575 .gpl_only = false, 6576 .pkt_access = true, 6577 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6578 .arg1_type = ARG_PTR_TO_CTX, 6579 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6580 .arg3_type = ARG_CONST_SIZE, 6581 .arg4_type = ARG_ANYTHING, 6582 .arg5_type = ARG_ANYTHING, 6583 }; 6584 6585 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb, 6586 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6587 { 6588 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, 6589 netns_id, flags); 6590 } 6591 6592 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = { 6593 .func = bpf_sk_lookup_tcp, 6594 .gpl_only = false, 6595 .pkt_access = true, 6596 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6597 .arg1_type = ARG_PTR_TO_CTX, 6598 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6599 .arg3_type = ARG_CONST_SIZE, 6600 .arg4_type = ARG_ANYTHING, 6601 .arg5_type = ARG_ANYTHING, 6602 }; 6603 6604 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb, 6605 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6606 { 6607 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, 6608 netns_id, flags); 6609 } 6610 6611 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = { 6612 .func = bpf_sk_lookup_udp, 6613 .gpl_only = false, 6614 .pkt_access = true, 6615 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6616 .arg1_type = ARG_PTR_TO_CTX, 6617 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6618 .arg3_type = ARG_CONST_SIZE, 6619 .arg4_type = ARG_ANYTHING, 6620 .arg5_type = ARG_ANYTHING, 6621 }; 6622 6623 BPF_CALL_1(bpf_sk_release, struct sock *, sk) 6624 { 6625 if (sk && sk_is_refcounted(sk)) 6626 sock_gen_put(sk); 6627 return 0; 6628 } 6629 6630 static const struct bpf_func_proto bpf_sk_release_proto = { 6631 .func = bpf_sk_release, 6632 .gpl_only = false, 6633 .ret_type = RET_INTEGER, 6634 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON | OBJ_RELEASE, 6635 }; 6636 6637 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx, 6638 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6639 { 6640 struct net *caller_net = dev_net(ctx->rxq->dev); 6641 int ifindex = ctx->rxq->dev->ifindex; 6642 6643 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6644 ifindex, IPPROTO_UDP, netns_id, 6645 flags); 6646 } 6647 6648 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = { 6649 .func = bpf_xdp_sk_lookup_udp, 6650 .gpl_only = false, 6651 .pkt_access = true, 6652 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6653 .arg1_type = ARG_PTR_TO_CTX, 6654 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6655 .arg3_type = ARG_CONST_SIZE, 6656 .arg4_type = ARG_ANYTHING, 6657 .arg5_type = ARG_ANYTHING, 6658 }; 6659 6660 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx, 6661 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6662 { 6663 struct net *caller_net = dev_net(ctx->rxq->dev); 6664 int ifindex = ctx->rxq->dev->ifindex; 6665 6666 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net, 6667 ifindex, IPPROTO_TCP, netns_id, 6668 flags); 6669 } 6670 6671 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = { 6672 .func = bpf_xdp_skc_lookup_tcp, 6673 .gpl_only = false, 6674 .pkt_access = true, 6675 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6676 .arg1_type = ARG_PTR_TO_CTX, 6677 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6678 .arg3_type = ARG_CONST_SIZE, 6679 .arg4_type = ARG_ANYTHING, 6680 .arg5_type = ARG_ANYTHING, 6681 }; 6682 6683 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx, 6684 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 6685 { 6686 struct net *caller_net = dev_net(ctx->rxq->dev); 6687 int ifindex = ctx->rxq->dev->ifindex; 6688 6689 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 6690 ifindex, IPPROTO_TCP, netns_id, 6691 flags); 6692 } 6693 6694 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = { 6695 .func = bpf_xdp_sk_lookup_tcp, 6696 .gpl_only = false, 6697 .pkt_access = true, 6698 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6699 .arg1_type = ARG_PTR_TO_CTX, 6700 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6701 .arg3_type = ARG_CONST_SIZE, 6702 .arg4_type = ARG_ANYTHING, 6703 .arg5_type = ARG_ANYTHING, 6704 }; 6705 6706 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6707 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6708 { 6709 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, 6710 sock_net(ctx->sk), 0, 6711 IPPROTO_TCP, netns_id, flags); 6712 } 6713 6714 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = { 6715 .func = bpf_sock_addr_skc_lookup_tcp, 6716 .gpl_only = false, 6717 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 6718 .arg1_type = ARG_PTR_TO_CTX, 6719 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6720 .arg3_type = ARG_CONST_SIZE, 6721 .arg4_type = ARG_ANYTHING, 6722 .arg5_type = ARG_ANYTHING, 6723 }; 6724 6725 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 6726 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6727 { 6728 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6729 sock_net(ctx->sk), 0, IPPROTO_TCP, 6730 netns_id, flags); 6731 } 6732 6733 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = { 6734 .func = bpf_sock_addr_sk_lookup_tcp, 6735 .gpl_only = false, 6736 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6737 .arg1_type = ARG_PTR_TO_CTX, 6738 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6739 .arg3_type = ARG_CONST_SIZE, 6740 .arg4_type = ARG_ANYTHING, 6741 .arg5_type = ARG_ANYTHING, 6742 }; 6743 6744 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx, 6745 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 6746 { 6747 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 6748 sock_net(ctx->sk), 0, IPPROTO_UDP, 6749 netns_id, flags); 6750 } 6751 6752 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = { 6753 .func = bpf_sock_addr_sk_lookup_udp, 6754 .gpl_only = false, 6755 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6756 .arg1_type = ARG_PTR_TO_CTX, 6757 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 6758 .arg3_type = ARG_CONST_SIZE, 6759 .arg4_type = ARG_ANYTHING, 6760 .arg5_type = ARG_ANYTHING, 6761 }; 6762 6763 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6764 struct bpf_insn_access_aux *info) 6765 { 6766 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, 6767 icsk_retransmits)) 6768 return false; 6769 6770 if (off % size != 0) 6771 return false; 6772 6773 switch (off) { 6774 case offsetof(struct bpf_tcp_sock, bytes_received): 6775 case offsetof(struct bpf_tcp_sock, bytes_acked): 6776 return size == sizeof(__u64); 6777 default: 6778 return size == sizeof(__u32); 6779 } 6780 } 6781 6782 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 6783 const struct bpf_insn *si, 6784 struct bpf_insn *insn_buf, 6785 struct bpf_prog *prog, u32 *target_size) 6786 { 6787 struct bpf_insn *insn = insn_buf; 6788 6789 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \ 6790 do { \ 6791 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \ 6792 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6793 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\ 6794 si->dst_reg, si->src_reg, \ 6795 offsetof(struct tcp_sock, FIELD)); \ 6796 } while (0) 6797 6798 #define BPF_INET_SOCK_GET_COMMON(FIELD) \ 6799 do { \ 6800 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \ 6801 FIELD) > \ 6802 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 6803 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 6804 struct inet_connection_sock, \ 6805 FIELD), \ 6806 si->dst_reg, si->src_reg, \ 6807 offsetof( \ 6808 struct inet_connection_sock, \ 6809 FIELD)); \ 6810 } while (0) 6811 6812 if (insn > insn_buf) 6813 return insn - insn_buf; 6814 6815 switch (si->off) { 6816 case offsetof(struct bpf_tcp_sock, rtt_min): 6817 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 6818 sizeof(struct minmax)); 6819 BUILD_BUG_ON(sizeof(struct minmax) < 6820 sizeof(struct minmax_sample)); 6821 6822 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 6823 offsetof(struct tcp_sock, rtt_min) + 6824 offsetof(struct minmax_sample, v)); 6825 break; 6826 case offsetof(struct bpf_tcp_sock, snd_cwnd): 6827 BPF_TCP_SOCK_GET_COMMON(snd_cwnd); 6828 break; 6829 case offsetof(struct bpf_tcp_sock, srtt_us): 6830 BPF_TCP_SOCK_GET_COMMON(srtt_us); 6831 break; 6832 case offsetof(struct bpf_tcp_sock, snd_ssthresh): 6833 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh); 6834 break; 6835 case offsetof(struct bpf_tcp_sock, rcv_nxt): 6836 BPF_TCP_SOCK_GET_COMMON(rcv_nxt); 6837 break; 6838 case offsetof(struct bpf_tcp_sock, snd_nxt): 6839 BPF_TCP_SOCK_GET_COMMON(snd_nxt); 6840 break; 6841 case offsetof(struct bpf_tcp_sock, snd_una): 6842 BPF_TCP_SOCK_GET_COMMON(snd_una); 6843 break; 6844 case offsetof(struct bpf_tcp_sock, mss_cache): 6845 BPF_TCP_SOCK_GET_COMMON(mss_cache); 6846 break; 6847 case offsetof(struct bpf_tcp_sock, ecn_flags): 6848 BPF_TCP_SOCK_GET_COMMON(ecn_flags); 6849 break; 6850 case offsetof(struct bpf_tcp_sock, rate_delivered): 6851 BPF_TCP_SOCK_GET_COMMON(rate_delivered); 6852 break; 6853 case offsetof(struct bpf_tcp_sock, rate_interval_us): 6854 BPF_TCP_SOCK_GET_COMMON(rate_interval_us); 6855 break; 6856 case offsetof(struct bpf_tcp_sock, packets_out): 6857 BPF_TCP_SOCK_GET_COMMON(packets_out); 6858 break; 6859 case offsetof(struct bpf_tcp_sock, retrans_out): 6860 BPF_TCP_SOCK_GET_COMMON(retrans_out); 6861 break; 6862 case offsetof(struct bpf_tcp_sock, total_retrans): 6863 BPF_TCP_SOCK_GET_COMMON(total_retrans); 6864 break; 6865 case offsetof(struct bpf_tcp_sock, segs_in): 6866 BPF_TCP_SOCK_GET_COMMON(segs_in); 6867 break; 6868 case offsetof(struct bpf_tcp_sock, data_segs_in): 6869 BPF_TCP_SOCK_GET_COMMON(data_segs_in); 6870 break; 6871 case offsetof(struct bpf_tcp_sock, segs_out): 6872 BPF_TCP_SOCK_GET_COMMON(segs_out); 6873 break; 6874 case offsetof(struct bpf_tcp_sock, data_segs_out): 6875 BPF_TCP_SOCK_GET_COMMON(data_segs_out); 6876 break; 6877 case offsetof(struct bpf_tcp_sock, lost_out): 6878 BPF_TCP_SOCK_GET_COMMON(lost_out); 6879 break; 6880 case offsetof(struct bpf_tcp_sock, sacked_out): 6881 BPF_TCP_SOCK_GET_COMMON(sacked_out); 6882 break; 6883 case offsetof(struct bpf_tcp_sock, bytes_received): 6884 BPF_TCP_SOCK_GET_COMMON(bytes_received); 6885 break; 6886 case offsetof(struct bpf_tcp_sock, bytes_acked): 6887 BPF_TCP_SOCK_GET_COMMON(bytes_acked); 6888 break; 6889 case offsetof(struct bpf_tcp_sock, dsack_dups): 6890 BPF_TCP_SOCK_GET_COMMON(dsack_dups); 6891 break; 6892 case offsetof(struct bpf_tcp_sock, delivered): 6893 BPF_TCP_SOCK_GET_COMMON(delivered); 6894 break; 6895 case offsetof(struct bpf_tcp_sock, delivered_ce): 6896 BPF_TCP_SOCK_GET_COMMON(delivered_ce); 6897 break; 6898 case offsetof(struct bpf_tcp_sock, icsk_retransmits): 6899 BPF_INET_SOCK_GET_COMMON(icsk_retransmits); 6900 break; 6901 } 6902 6903 return insn - insn_buf; 6904 } 6905 6906 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk) 6907 { 6908 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 6909 return (unsigned long)sk; 6910 6911 return (unsigned long)NULL; 6912 } 6913 6914 const struct bpf_func_proto bpf_tcp_sock_proto = { 6915 .func = bpf_tcp_sock, 6916 .gpl_only = false, 6917 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL, 6918 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 6919 }; 6920 6921 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk) 6922 { 6923 sk = sk_to_full_sk(sk); 6924 6925 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE)) 6926 return (unsigned long)sk; 6927 6928 return (unsigned long)NULL; 6929 } 6930 6931 static const struct bpf_func_proto bpf_get_listener_sock_proto = { 6932 .func = bpf_get_listener_sock, 6933 .gpl_only = false, 6934 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 6935 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 6936 }; 6937 6938 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb) 6939 { 6940 unsigned int iphdr_len; 6941 6942 switch (skb_protocol(skb, true)) { 6943 case cpu_to_be16(ETH_P_IP): 6944 iphdr_len = sizeof(struct iphdr); 6945 break; 6946 case cpu_to_be16(ETH_P_IPV6): 6947 iphdr_len = sizeof(struct ipv6hdr); 6948 break; 6949 default: 6950 return 0; 6951 } 6952 6953 if (skb_headlen(skb) < iphdr_len) 6954 return 0; 6955 6956 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len)) 6957 return 0; 6958 6959 return INET_ECN_set_ce(skb); 6960 } 6961 6962 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6963 struct bpf_insn_access_aux *info) 6964 { 6965 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id)) 6966 return false; 6967 6968 if (off % size != 0) 6969 return false; 6970 6971 switch (off) { 6972 default: 6973 return size == sizeof(__u32); 6974 } 6975 } 6976 6977 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 6978 const struct bpf_insn *si, 6979 struct bpf_insn *insn_buf, 6980 struct bpf_prog *prog, u32 *target_size) 6981 { 6982 struct bpf_insn *insn = insn_buf; 6983 6984 #define BPF_XDP_SOCK_GET(FIELD) \ 6985 do { \ 6986 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \ 6987 sizeof_field(struct bpf_xdp_sock, FIELD)); \ 6988 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\ 6989 si->dst_reg, si->src_reg, \ 6990 offsetof(struct xdp_sock, FIELD)); \ 6991 } while (0) 6992 6993 switch (si->off) { 6994 case offsetof(struct bpf_xdp_sock, queue_id): 6995 BPF_XDP_SOCK_GET(queue_id); 6996 break; 6997 } 6998 6999 return insn - insn_buf; 7000 } 7001 7002 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = { 7003 .func = bpf_skb_ecn_set_ce, 7004 .gpl_only = false, 7005 .ret_type = RET_INTEGER, 7006 .arg1_type = ARG_PTR_TO_CTX, 7007 }; 7008 7009 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 7010 struct tcphdr *, th, u32, th_len) 7011 { 7012 #ifdef CONFIG_SYN_COOKIES 7013 u32 cookie; 7014 int ret; 7015 7016 if (unlikely(!sk || th_len < sizeof(*th))) 7017 return -EINVAL; 7018 7019 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */ 7020 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 7021 return -EINVAL; 7022 7023 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies) 7024 return -EINVAL; 7025 7026 if (!th->ack || th->rst || th->syn) 7027 return -ENOENT; 7028 7029 if (unlikely(iph_len < sizeof(struct iphdr))) 7030 return -EINVAL; 7031 7032 if (tcp_synq_no_recent_overflow(sk)) 7033 return -ENOENT; 7034 7035 cookie = ntohl(th->ack_seq) - 1; 7036 7037 /* Both struct iphdr and struct ipv6hdr have the version field at the 7038 * same offset so we can cast to the shorter header (struct iphdr). 7039 */ 7040 switch (((struct iphdr *)iph)->version) { 7041 case 4: 7042 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk)) 7043 return -EINVAL; 7044 7045 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie); 7046 break; 7047 7048 #if IS_BUILTIN(CONFIG_IPV6) 7049 case 6: 7050 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 7051 return -EINVAL; 7052 7053 if (sk->sk_family != AF_INET6) 7054 return -EINVAL; 7055 7056 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie); 7057 break; 7058 #endif /* CONFIG_IPV6 */ 7059 7060 default: 7061 return -EPROTONOSUPPORT; 7062 } 7063 7064 if (ret > 0) 7065 return 0; 7066 7067 return -ENOENT; 7068 #else 7069 return -ENOTSUPP; 7070 #endif 7071 } 7072 7073 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = { 7074 .func = bpf_tcp_check_syncookie, 7075 .gpl_only = true, 7076 .pkt_access = true, 7077 .ret_type = RET_INTEGER, 7078 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7079 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7080 .arg3_type = ARG_CONST_SIZE, 7081 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7082 .arg5_type = ARG_CONST_SIZE, 7083 }; 7084 7085 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 7086 struct tcphdr *, th, u32, th_len) 7087 { 7088 #ifdef CONFIG_SYN_COOKIES 7089 u32 cookie; 7090 u16 mss; 7091 7092 if (unlikely(!sk || th_len < sizeof(*th) || th_len != th->doff * 4)) 7093 return -EINVAL; 7094 7095 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 7096 return -EINVAL; 7097 7098 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies) 7099 return -ENOENT; 7100 7101 if (!th->syn || th->ack || th->fin || th->rst) 7102 return -EINVAL; 7103 7104 if (unlikely(iph_len < sizeof(struct iphdr))) 7105 return -EINVAL; 7106 7107 /* Both struct iphdr and struct ipv6hdr have the version field at the 7108 * same offset so we can cast to the shorter header (struct iphdr). 7109 */ 7110 switch (((struct iphdr *)iph)->version) { 7111 case 4: 7112 if (sk->sk_family == AF_INET6 && ipv6_only_sock(sk)) 7113 return -EINVAL; 7114 7115 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie); 7116 break; 7117 7118 #if IS_BUILTIN(CONFIG_IPV6) 7119 case 6: 7120 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 7121 return -EINVAL; 7122 7123 if (sk->sk_family != AF_INET6) 7124 return -EINVAL; 7125 7126 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie); 7127 break; 7128 #endif /* CONFIG_IPV6 */ 7129 7130 default: 7131 return -EPROTONOSUPPORT; 7132 } 7133 if (mss == 0) 7134 return -ENOENT; 7135 7136 return cookie | ((u64)mss << 32); 7137 #else 7138 return -EOPNOTSUPP; 7139 #endif /* CONFIG_SYN_COOKIES */ 7140 } 7141 7142 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = { 7143 .func = bpf_tcp_gen_syncookie, 7144 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */ 7145 .pkt_access = true, 7146 .ret_type = RET_INTEGER, 7147 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7148 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7149 .arg3_type = ARG_CONST_SIZE, 7150 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7151 .arg5_type = ARG_CONST_SIZE, 7152 }; 7153 7154 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags) 7155 { 7156 if (!sk || flags != 0) 7157 return -EINVAL; 7158 if (!skb_at_tc_ingress(skb)) 7159 return -EOPNOTSUPP; 7160 if (unlikely(dev_net(skb->dev) != sock_net(sk))) 7161 return -ENETUNREACH; 7162 if (unlikely(sk_fullsock(sk) && sk->sk_reuseport)) 7163 return -ESOCKTNOSUPPORT; 7164 if (sk_is_refcounted(sk) && 7165 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) 7166 return -ENOENT; 7167 7168 skb_orphan(skb); 7169 skb->sk = sk; 7170 skb->destructor = sock_pfree; 7171 7172 return 0; 7173 } 7174 7175 static const struct bpf_func_proto bpf_sk_assign_proto = { 7176 .func = bpf_sk_assign, 7177 .gpl_only = false, 7178 .ret_type = RET_INTEGER, 7179 .arg1_type = ARG_PTR_TO_CTX, 7180 .arg2_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 7181 .arg3_type = ARG_ANYTHING, 7182 }; 7183 7184 static const u8 *bpf_search_tcp_opt(const u8 *op, const u8 *opend, 7185 u8 search_kind, const u8 *magic, 7186 u8 magic_len, bool *eol) 7187 { 7188 u8 kind, kind_len; 7189 7190 *eol = false; 7191 7192 while (op < opend) { 7193 kind = op[0]; 7194 7195 if (kind == TCPOPT_EOL) { 7196 *eol = true; 7197 return ERR_PTR(-ENOMSG); 7198 } else if (kind == TCPOPT_NOP) { 7199 op++; 7200 continue; 7201 } 7202 7203 if (opend - op < 2 || opend - op < op[1] || op[1] < 2) 7204 /* Something is wrong in the received header. 7205 * Follow the TCP stack's tcp_parse_options() 7206 * and just bail here. 7207 */ 7208 return ERR_PTR(-EFAULT); 7209 7210 kind_len = op[1]; 7211 if (search_kind == kind) { 7212 if (!magic_len) 7213 return op; 7214 7215 if (magic_len > kind_len - 2) 7216 return ERR_PTR(-ENOMSG); 7217 7218 if (!memcmp(&op[2], magic, magic_len)) 7219 return op; 7220 } 7221 7222 op += kind_len; 7223 } 7224 7225 return ERR_PTR(-ENOMSG); 7226 } 7227 7228 BPF_CALL_4(bpf_sock_ops_load_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7229 void *, search_res, u32, len, u64, flags) 7230 { 7231 bool eol, load_syn = flags & BPF_LOAD_HDR_OPT_TCP_SYN; 7232 const u8 *op, *opend, *magic, *search = search_res; 7233 u8 search_kind, search_len, copy_len, magic_len; 7234 int ret; 7235 7236 /* 2 byte is the minimal option len except TCPOPT_NOP and 7237 * TCPOPT_EOL which are useless for the bpf prog to learn 7238 * and this helper disallow loading them also. 7239 */ 7240 if (len < 2 || flags & ~BPF_LOAD_HDR_OPT_TCP_SYN) 7241 return -EINVAL; 7242 7243 search_kind = search[0]; 7244 search_len = search[1]; 7245 7246 if (search_len > len || search_kind == TCPOPT_NOP || 7247 search_kind == TCPOPT_EOL) 7248 return -EINVAL; 7249 7250 if (search_kind == TCPOPT_EXP || search_kind == 253) { 7251 /* 16 or 32 bit magic. +2 for kind and kind length */ 7252 if (search_len != 4 && search_len != 6) 7253 return -EINVAL; 7254 magic = &search[2]; 7255 magic_len = search_len - 2; 7256 } else { 7257 if (search_len) 7258 return -EINVAL; 7259 magic = NULL; 7260 magic_len = 0; 7261 } 7262 7263 if (load_syn) { 7264 ret = bpf_sock_ops_get_syn(bpf_sock, TCP_BPF_SYN, &op); 7265 if (ret < 0) 7266 return ret; 7267 7268 opend = op + ret; 7269 op += sizeof(struct tcphdr); 7270 } else { 7271 if (!bpf_sock->skb || 7272 bpf_sock->op == BPF_SOCK_OPS_HDR_OPT_LEN_CB) 7273 /* This bpf_sock->op cannot call this helper */ 7274 return -EPERM; 7275 7276 opend = bpf_sock->skb_data_end; 7277 op = bpf_sock->skb->data + sizeof(struct tcphdr); 7278 } 7279 7280 op = bpf_search_tcp_opt(op, opend, search_kind, magic, magic_len, 7281 &eol); 7282 if (IS_ERR(op)) 7283 return PTR_ERR(op); 7284 7285 copy_len = op[1]; 7286 ret = copy_len; 7287 if (copy_len > len) { 7288 ret = -ENOSPC; 7289 copy_len = len; 7290 } 7291 7292 memcpy(search_res, op, copy_len); 7293 return ret; 7294 } 7295 7296 static const struct bpf_func_proto bpf_sock_ops_load_hdr_opt_proto = { 7297 .func = bpf_sock_ops_load_hdr_opt, 7298 .gpl_only = false, 7299 .ret_type = RET_INTEGER, 7300 .arg1_type = ARG_PTR_TO_CTX, 7301 .arg2_type = ARG_PTR_TO_MEM, 7302 .arg3_type = ARG_CONST_SIZE, 7303 .arg4_type = ARG_ANYTHING, 7304 }; 7305 7306 BPF_CALL_4(bpf_sock_ops_store_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7307 const void *, from, u32, len, u64, flags) 7308 { 7309 u8 new_kind, new_kind_len, magic_len = 0, *opend; 7310 const u8 *op, *new_op, *magic = NULL; 7311 struct sk_buff *skb; 7312 bool eol; 7313 7314 if (bpf_sock->op != BPF_SOCK_OPS_WRITE_HDR_OPT_CB) 7315 return -EPERM; 7316 7317 if (len < 2 || flags) 7318 return -EINVAL; 7319 7320 new_op = from; 7321 new_kind = new_op[0]; 7322 new_kind_len = new_op[1]; 7323 7324 if (new_kind_len > len || new_kind == TCPOPT_NOP || 7325 new_kind == TCPOPT_EOL) 7326 return -EINVAL; 7327 7328 if (new_kind_len > bpf_sock->remaining_opt_len) 7329 return -ENOSPC; 7330 7331 /* 253 is another experimental kind */ 7332 if (new_kind == TCPOPT_EXP || new_kind == 253) { 7333 if (new_kind_len < 4) 7334 return -EINVAL; 7335 /* Match for the 2 byte magic also. 7336 * RFC 6994: the magic could be 2 or 4 bytes. 7337 * Hence, matching by 2 byte only is on the 7338 * conservative side but it is the right 7339 * thing to do for the 'search-for-duplication' 7340 * purpose. 7341 */ 7342 magic = &new_op[2]; 7343 magic_len = 2; 7344 } 7345 7346 /* Check for duplication */ 7347 skb = bpf_sock->skb; 7348 op = skb->data + sizeof(struct tcphdr); 7349 opend = bpf_sock->skb_data_end; 7350 7351 op = bpf_search_tcp_opt(op, opend, new_kind, magic, magic_len, 7352 &eol); 7353 if (!IS_ERR(op)) 7354 return -EEXIST; 7355 7356 if (PTR_ERR(op) != -ENOMSG) 7357 return PTR_ERR(op); 7358 7359 if (eol) 7360 /* The option has been ended. Treat it as no more 7361 * header option can be written. 7362 */ 7363 return -ENOSPC; 7364 7365 /* No duplication found. Store the header option. */ 7366 memcpy(opend, from, new_kind_len); 7367 7368 bpf_sock->remaining_opt_len -= new_kind_len; 7369 bpf_sock->skb_data_end += new_kind_len; 7370 7371 return 0; 7372 } 7373 7374 static const struct bpf_func_proto bpf_sock_ops_store_hdr_opt_proto = { 7375 .func = bpf_sock_ops_store_hdr_opt, 7376 .gpl_only = false, 7377 .ret_type = RET_INTEGER, 7378 .arg1_type = ARG_PTR_TO_CTX, 7379 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 7380 .arg3_type = ARG_CONST_SIZE, 7381 .arg4_type = ARG_ANYTHING, 7382 }; 7383 7384 BPF_CALL_3(bpf_sock_ops_reserve_hdr_opt, struct bpf_sock_ops_kern *, bpf_sock, 7385 u32, len, u64, flags) 7386 { 7387 if (bpf_sock->op != BPF_SOCK_OPS_HDR_OPT_LEN_CB) 7388 return -EPERM; 7389 7390 if (flags || len < 2) 7391 return -EINVAL; 7392 7393 if (len > bpf_sock->remaining_opt_len) 7394 return -ENOSPC; 7395 7396 bpf_sock->remaining_opt_len -= len; 7397 7398 return 0; 7399 } 7400 7401 static const struct bpf_func_proto bpf_sock_ops_reserve_hdr_opt_proto = { 7402 .func = bpf_sock_ops_reserve_hdr_opt, 7403 .gpl_only = false, 7404 .ret_type = RET_INTEGER, 7405 .arg1_type = ARG_PTR_TO_CTX, 7406 .arg2_type = ARG_ANYTHING, 7407 .arg3_type = ARG_ANYTHING, 7408 }; 7409 7410 BPF_CALL_3(bpf_skb_set_tstamp, struct sk_buff *, skb, 7411 u64, tstamp, u32, tstamp_type) 7412 { 7413 /* skb_clear_delivery_time() is done for inet protocol */ 7414 if (skb->protocol != htons(ETH_P_IP) && 7415 skb->protocol != htons(ETH_P_IPV6)) 7416 return -EOPNOTSUPP; 7417 7418 switch (tstamp_type) { 7419 case BPF_SKB_TSTAMP_DELIVERY_MONO: 7420 if (!tstamp) 7421 return -EINVAL; 7422 skb->tstamp = tstamp; 7423 skb->mono_delivery_time = 1; 7424 break; 7425 case BPF_SKB_TSTAMP_UNSPEC: 7426 if (tstamp) 7427 return -EINVAL; 7428 skb->tstamp = 0; 7429 skb->mono_delivery_time = 0; 7430 break; 7431 default: 7432 return -EINVAL; 7433 } 7434 7435 return 0; 7436 } 7437 7438 static const struct bpf_func_proto bpf_skb_set_tstamp_proto = { 7439 .func = bpf_skb_set_tstamp, 7440 .gpl_only = false, 7441 .ret_type = RET_INTEGER, 7442 .arg1_type = ARG_PTR_TO_CTX, 7443 .arg2_type = ARG_ANYTHING, 7444 .arg3_type = ARG_ANYTHING, 7445 }; 7446 7447 #endif /* CONFIG_INET */ 7448 7449 bool bpf_helper_changes_pkt_data(void *func) 7450 { 7451 if (func == bpf_skb_vlan_push || 7452 func == bpf_skb_vlan_pop || 7453 func == bpf_skb_store_bytes || 7454 func == bpf_skb_change_proto || 7455 func == bpf_skb_change_head || 7456 func == sk_skb_change_head || 7457 func == bpf_skb_change_tail || 7458 func == sk_skb_change_tail || 7459 func == bpf_skb_adjust_room || 7460 func == sk_skb_adjust_room || 7461 func == bpf_skb_pull_data || 7462 func == sk_skb_pull_data || 7463 func == bpf_clone_redirect || 7464 func == bpf_l3_csum_replace || 7465 func == bpf_l4_csum_replace || 7466 func == bpf_xdp_adjust_head || 7467 func == bpf_xdp_adjust_meta || 7468 func == bpf_msg_pull_data || 7469 func == bpf_msg_push_data || 7470 func == bpf_msg_pop_data || 7471 func == bpf_xdp_adjust_tail || 7472 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 7473 func == bpf_lwt_seg6_store_bytes || 7474 func == bpf_lwt_seg6_adjust_srh || 7475 func == bpf_lwt_seg6_action || 7476 #endif 7477 #ifdef CONFIG_INET 7478 func == bpf_sock_ops_store_hdr_opt || 7479 #endif 7480 func == bpf_lwt_in_push_encap || 7481 func == bpf_lwt_xmit_push_encap) 7482 return true; 7483 7484 return false; 7485 } 7486 7487 const struct bpf_func_proto bpf_event_output_data_proto __weak; 7488 const struct bpf_func_proto bpf_sk_storage_get_cg_sock_proto __weak; 7489 7490 static const struct bpf_func_proto * 7491 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7492 { 7493 switch (func_id) { 7494 /* inet and inet6 sockets are created in a process 7495 * context so there is always a valid uid/gid 7496 */ 7497 case BPF_FUNC_get_current_uid_gid: 7498 return &bpf_get_current_uid_gid_proto; 7499 case BPF_FUNC_get_local_storage: 7500 return &bpf_get_local_storage_proto; 7501 case BPF_FUNC_get_socket_cookie: 7502 return &bpf_get_socket_cookie_sock_proto; 7503 case BPF_FUNC_get_netns_cookie: 7504 return &bpf_get_netns_cookie_sock_proto; 7505 case BPF_FUNC_perf_event_output: 7506 return &bpf_event_output_data_proto; 7507 case BPF_FUNC_get_current_pid_tgid: 7508 return &bpf_get_current_pid_tgid_proto; 7509 case BPF_FUNC_get_current_comm: 7510 return &bpf_get_current_comm_proto; 7511 #ifdef CONFIG_CGROUPS 7512 case BPF_FUNC_get_current_cgroup_id: 7513 return &bpf_get_current_cgroup_id_proto; 7514 case BPF_FUNC_get_current_ancestor_cgroup_id: 7515 return &bpf_get_current_ancestor_cgroup_id_proto; 7516 #endif 7517 #ifdef CONFIG_CGROUP_NET_CLASSID 7518 case BPF_FUNC_get_cgroup_classid: 7519 return &bpf_get_cgroup_classid_curr_proto; 7520 #endif 7521 case BPF_FUNC_sk_storage_get: 7522 return &bpf_sk_storage_get_cg_sock_proto; 7523 case BPF_FUNC_ktime_get_coarse_ns: 7524 return &bpf_ktime_get_coarse_ns_proto; 7525 default: 7526 return bpf_base_func_proto(func_id); 7527 } 7528 } 7529 7530 static const struct bpf_func_proto * 7531 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7532 { 7533 switch (func_id) { 7534 /* inet and inet6 sockets are created in a process 7535 * context so there is always a valid uid/gid 7536 */ 7537 case BPF_FUNC_get_current_uid_gid: 7538 return &bpf_get_current_uid_gid_proto; 7539 case BPF_FUNC_bind: 7540 switch (prog->expected_attach_type) { 7541 case BPF_CGROUP_INET4_CONNECT: 7542 case BPF_CGROUP_INET6_CONNECT: 7543 return &bpf_bind_proto; 7544 default: 7545 return NULL; 7546 } 7547 case BPF_FUNC_get_socket_cookie: 7548 return &bpf_get_socket_cookie_sock_addr_proto; 7549 case BPF_FUNC_get_netns_cookie: 7550 return &bpf_get_netns_cookie_sock_addr_proto; 7551 case BPF_FUNC_get_local_storage: 7552 return &bpf_get_local_storage_proto; 7553 case BPF_FUNC_perf_event_output: 7554 return &bpf_event_output_data_proto; 7555 case BPF_FUNC_get_current_pid_tgid: 7556 return &bpf_get_current_pid_tgid_proto; 7557 case BPF_FUNC_get_current_comm: 7558 return &bpf_get_current_comm_proto; 7559 #ifdef CONFIG_CGROUPS 7560 case BPF_FUNC_get_current_cgroup_id: 7561 return &bpf_get_current_cgroup_id_proto; 7562 case BPF_FUNC_get_current_ancestor_cgroup_id: 7563 return &bpf_get_current_ancestor_cgroup_id_proto; 7564 #endif 7565 #ifdef CONFIG_CGROUP_NET_CLASSID 7566 case BPF_FUNC_get_cgroup_classid: 7567 return &bpf_get_cgroup_classid_curr_proto; 7568 #endif 7569 #ifdef CONFIG_INET 7570 case BPF_FUNC_sk_lookup_tcp: 7571 return &bpf_sock_addr_sk_lookup_tcp_proto; 7572 case BPF_FUNC_sk_lookup_udp: 7573 return &bpf_sock_addr_sk_lookup_udp_proto; 7574 case BPF_FUNC_sk_release: 7575 return &bpf_sk_release_proto; 7576 case BPF_FUNC_skc_lookup_tcp: 7577 return &bpf_sock_addr_skc_lookup_tcp_proto; 7578 #endif /* CONFIG_INET */ 7579 case BPF_FUNC_sk_storage_get: 7580 return &bpf_sk_storage_get_proto; 7581 case BPF_FUNC_sk_storage_delete: 7582 return &bpf_sk_storage_delete_proto; 7583 case BPF_FUNC_setsockopt: 7584 switch (prog->expected_attach_type) { 7585 case BPF_CGROUP_INET4_BIND: 7586 case BPF_CGROUP_INET6_BIND: 7587 case BPF_CGROUP_INET4_CONNECT: 7588 case BPF_CGROUP_INET6_CONNECT: 7589 case BPF_CGROUP_UDP4_RECVMSG: 7590 case BPF_CGROUP_UDP6_RECVMSG: 7591 case BPF_CGROUP_UDP4_SENDMSG: 7592 case BPF_CGROUP_UDP6_SENDMSG: 7593 case BPF_CGROUP_INET4_GETPEERNAME: 7594 case BPF_CGROUP_INET6_GETPEERNAME: 7595 case BPF_CGROUP_INET4_GETSOCKNAME: 7596 case BPF_CGROUP_INET6_GETSOCKNAME: 7597 return &bpf_sock_addr_setsockopt_proto; 7598 default: 7599 return NULL; 7600 } 7601 case BPF_FUNC_getsockopt: 7602 switch (prog->expected_attach_type) { 7603 case BPF_CGROUP_INET4_BIND: 7604 case BPF_CGROUP_INET6_BIND: 7605 case BPF_CGROUP_INET4_CONNECT: 7606 case BPF_CGROUP_INET6_CONNECT: 7607 case BPF_CGROUP_UDP4_RECVMSG: 7608 case BPF_CGROUP_UDP6_RECVMSG: 7609 case BPF_CGROUP_UDP4_SENDMSG: 7610 case BPF_CGROUP_UDP6_SENDMSG: 7611 case BPF_CGROUP_INET4_GETPEERNAME: 7612 case BPF_CGROUP_INET6_GETPEERNAME: 7613 case BPF_CGROUP_INET4_GETSOCKNAME: 7614 case BPF_CGROUP_INET6_GETSOCKNAME: 7615 return &bpf_sock_addr_getsockopt_proto; 7616 default: 7617 return NULL; 7618 } 7619 default: 7620 return bpf_sk_base_func_proto(func_id); 7621 } 7622 } 7623 7624 static const struct bpf_func_proto * 7625 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7626 { 7627 switch (func_id) { 7628 case BPF_FUNC_skb_load_bytes: 7629 return &bpf_skb_load_bytes_proto; 7630 case BPF_FUNC_skb_load_bytes_relative: 7631 return &bpf_skb_load_bytes_relative_proto; 7632 case BPF_FUNC_get_socket_cookie: 7633 return &bpf_get_socket_cookie_proto; 7634 case BPF_FUNC_get_socket_uid: 7635 return &bpf_get_socket_uid_proto; 7636 case BPF_FUNC_perf_event_output: 7637 return &bpf_skb_event_output_proto; 7638 default: 7639 return bpf_sk_base_func_proto(func_id); 7640 } 7641 } 7642 7643 const struct bpf_func_proto bpf_sk_storage_get_proto __weak; 7644 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak; 7645 7646 static const struct bpf_func_proto * 7647 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7648 { 7649 switch (func_id) { 7650 case BPF_FUNC_get_local_storage: 7651 return &bpf_get_local_storage_proto; 7652 case BPF_FUNC_sk_fullsock: 7653 return &bpf_sk_fullsock_proto; 7654 case BPF_FUNC_sk_storage_get: 7655 return &bpf_sk_storage_get_proto; 7656 case BPF_FUNC_sk_storage_delete: 7657 return &bpf_sk_storage_delete_proto; 7658 case BPF_FUNC_perf_event_output: 7659 return &bpf_skb_event_output_proto; 7660 #ifdef CONFIG_SOCK_CGROUP_DATA 7661 case BPF_FUNC_skb_cgroup_id: 7662 return &bpf_skb_cgroup_id_proto; 7663 case BPF_FUNC_skb_ancestor_cgroup_id: 7664 return &bpf_skb_ancestor_cgroup_id_proto; 7665 case BPF_FUNC_sk_cgroup_id: 7666 return &bpf_sk_cgroup_id_proto; 7667 case BPF_FUNC_sk_ancestor_cgroup_id: 7668 return &bpf_sk_ancestor_cgroup_id_proto; 7669 #endif 7670 #ifdef CONFIG_INET 7671 case BPF_FUNC_sk_lookup_tcp: 7672 return &bpf_sk_lookup_tcp_proto; 7673 case BPF_FUNC_sk_lookup_udp: 7674 return &bpf_sk_lookup_udp_proto; 7675 case BPF_FUNC_sk_release: 7676 return &bpf_sk_release_proto; 7677 case BPF_FUNC_skc_lookup_tcp: 7678 return &bpf_skc_lookup_tcp_proto; 7679 case BPF_FUNC_tcp_sock: 7680 return &bpf_tcp_sock_proto; 7681 case BPF_FUNC_get_listener_sock: 7682 return &bpf_get_listener_sock_proto; 7683 case BPF_FUNC_skb_ecn_set_ce: 7684 return &bpf_skb_ecn_set_ce_proto; 7685 #endif 7686 default: 7687 return sk_filter_func_proto(func_id, prog); 7688 } 7689 } 7690 7691 static const struct bpf_func_proto * 7692 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7693 { 7694 switch (func_id) { 7695 case BPF_FUNC_skb_store_bytes: 7696 return &bpf_skb_store_bytes_proto; 7697 case BPF_FUNC_skb_load_bytes: 7698 return &bpf_skb_load_bytes_proto; 7699 case BPF_FUNC_skb_load_bytes_relative: 7700 return &bpf_skb_load_bytes_relative_proto; 7701 case BPF_FUNC_skb_pull_data: 7702 return &bpf_skb_pull_data_proto; 7703 case BPF_FUNC_csum_diff: 7704 return &bpf_csum_diff_proto; 7705 case BPF_FUNC_csum_update: 7706 return &bpf_csum_update_proto; 7707 case BPF_FUNC_csum_level: 7708 return &bpf_csum_level_proto; 7709 case BPF_FUNC_l3_csum_replace: 7710 return &bpf_l3_csum_replace_proto; 7711 case BPF_FUNC_l4_csum_replace: 7712 return &bpf_l4_csum_replace_proto; 7713 case BPF_FUNC_clone_redirect: 7714 return &bpf_clone_redirect_proto; 7715 case BPF_FUNC_get_cgroup_classid: 7716 return &bpf_get_cgroup_classid_proto; 7717 case BPF_FUNC_skb_vlan_push: 7718 return &bpf_skb_vlan_push_proto; 7719 case BPF_FUNC_skb_vlan_pop: 7720 return &bpf_skb_vlan_pop_proto; 7721 case BPF_FUNC_skb_change_proto: 7722 return &bpf_skb_change_proto_proto; 7723 case BPF_FUNC_skb_change_type: 7724 return &bpf_skb_change_type_proto; 7725 case BPF_FUNC_skb_adjust_room: 7726 return &bpf_skb_adjust_room_proto; 7727 case BPF_FUNC_skb_change_tail: 7728 return &bpf_skb_change_tail_proto; 7729 case BPF_FUNC_skb_change_head: 7730 return &bpf_skb_change_head_proto; 7731 case BPF_FUNC_skb_get_tunnel_key: 7732 return &bpf_skb_get_tunnel_key_proto; 7733 case BPF_FUNC_skb_set_tunnel_key: 7734 return bpf_get_skb_set_tunnel_proto(func_id); 7735 case BPF_FUNC_skb_get_tunnel_opt: 7736 return &bpf_skb_get_tunnel_opt_proto; 7737 case BPF_FUNC_skb_set_tunnel_opt: 7738 return bpf_get_skb_set_tunnel_proto(func_id); 7739 case BPF_FUNC_redirect: 7740 return &bpf_redirect_proto; 7741 case BPF_FUNC_redirect_neigh: 7742 return &bpf_redirect_neigh_proto; 7743 case BPF_FUNC_redirect_peer: 7744 return &bpf_redirect_peer_proto; 7745 case BPF_FUNC_get_route_realm: 7746 return &bpf_get_route_realm_proto; 7747 case BPF_FUNC_get_hash_recalc: 7748 return &bpf_get_hash_recalc_proto; 7749 case BPF_FUNC_set_hash_invalid: 7750 return &bpf_set_hash_invalid_proto; 7751 case BPF_FUNC_set_hash: 7752 return &bpf_set_hash_proto; 7753 case BPF_FUNC_perf_event_output: 7754 return &bpf_skb_event_output_proto; 7755 case BPF_FUNC_get_smp_processor_id: 7756 return &bpf_get_smp_processor_id_proto; 7757 case BPF_FUNC_skb_under_cgroup: 7758 return &bpf_skb_under_cgroup_proto; 7759 case BPF_FUNC_get_socket_cookie: 7760 return &bpf_get_socket_cookie_proto; 7761 case BPF_FUNC_get_socket_uid: 7762 return &bpf_get_socket_uid_proto; 7763 case BPF_FUNC_fib_lookup: 7764 return &bpf_skb_fib_lookup_proto; 7765 case BPF_FUNC_check_mtu: 7766 return &bpf_skb_check_mtu_proto; 7767 case BPF_FUNC_sk_fullsock: 7768 return &bpf_sk_fullsock_proto; 7769 case BPF_FUNC_sk_storage_get: 7770 return &bpf_sk_storage_get_proto; 7771 case BPF_FUNC_sk_storage_delete: 7772 return &bpf_sk_storage_delete_proto; 7773 #ifdef CONFIG_XFRM 7774 case BPF_FUNC_skb_get_xfrm_state: 7775 return &bpf_skb_get_xfrm_state_proto; 7776 #endif 7777 #ifdef CONFIG_CGROUP_NET_CLASSID 7778 case BPF_FUNC_skb_cgroup_classid: 7779 return &bpf_skb_cgroup_classid_proto; 7780 #endif 7781 #ifdef CONFIG_SOCK_CGROUP_DATA 7782 case BPF_FUNC_skb_cgroup_id: 7783 return &bpf_skb_cgroup_id_proto; 7784 case BPF_FUNC_skb_ancestor_cgroup_id: 7785 return &bpf_skb_ancestor_cgroup_id_proto; 7786 #endif 7787 #ifdef CONFIG_INET 7788 case BPF_FUNC_sk_lookup_tcp: 7789 return &bpf_sk_lookup_tcp_proto; 7790 case BPF_FUNC_sk_lookup_udp: 7791 return &bpf_sk_lookup_udp_proto; 7792 case BPF_FUNC_sk_release: 7793 return &bpf_sk_release_proto; 7794 case BPF_FUNC_tcp_sock: 7795 return &bpf_tcp_sock_proto; 7796 case BPF_FUNC_get_listener_sock: 7797 return &bpf_get_listener_sock_proto; 7798 case BPF_FUNC_skc_lookup_tcp: 7799 return &bpf_skc_lookup_tcp_proto; 7800 case BPF_FUNC_tcp_check_syncookie: 7801 return &bpf_tcp_check_syncookie_proto; 7802 case BPF_FUNC_skb_ecn_set_ce: 7803 return &bpf_skb_ecn_set_ce_proto; 7804 case BPF_FUNC_tcp_gen_syncookie: 7805 return &bpf_tcp_gen_syncookie_proto; 7806 case BPF_FUNC_sk_assign: 7807 return &bpf_sk_assign_proto; 7808 case BPF_FUNC_skb_set_tstamp: 7809 return &bpf_skb_set_tstamp_proto; 7810 #endif 7811 default: 7812 return bpf_sk_base_func_proto(func_id); 7813 } 7814 } 7815 7816 static const struct bpf_func_proto * 7817 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7818 { 7819 switch (func_id) { 7820 case BPF_FUNC_perf_event_output: 7821 return &bpf_xdp_event_output_proto; 7822 case BPF_FUNC_get_smp_processor_id: 7823 return &bpf_get_smp_processor_id_proto; 7824 case BPF_FUNC_csum_diff: 7825 return &bpf_csum_diff_proto; 7826 case BPF_FUNC_xdp_adjust_head: 7827 return &bpf_xdp_adjust_head_proto; 7828 case BPF_FUNC_xdp_adjust_meta: 7829 return &bpf_xdp_adjust_meta_proto; 7830 case BPF_FUNC_redirect: 7831 return &bpf_xdp_redirect_proto; 7832 case BPF_FUNC_redirect_map: 7833 return &bpf_xdp_redirect_map_proto; 7834 case BPF_FUNC_xdp_adjust_tail: 7835 return &bpf_xdp_adjust_tail_proto; 7836 case BPF_FUNC_xdp_get_buff_len: 7837 return &bpf_xdp_get_buff_len_proto; 7838 case BPF_FUNC_xdp_load_bytes: 7839 return &bpf_xdp_load_bytes_proto; 7840 case BPF_FUNC_xdp_store_bytes: 7841 return &bpf_xdp_store_bytes_proto; 7842 case BPF_FUNC_fib_lookup: 7843 return &bpf_xdp_fib_lookup_proto; 7844 case BPF_FUNC_check_mtu: 7845 return &bpf_xdp_check_mtu_proto; 7846 #ifdef CONFIG_INET 7847 case BPF_FUNC_sk_lookup_udp: 7848 return &bpf_xdp_sk_lookup_udp_proto; 7849 case BPF_FUNC_sk_lookup_tcp: 7850 return &bpf_xdp_sk_lookup_tcp_proto; 7851 case BPF_FUNC_sk_release: 7852 return &bpf_sk_release_proto; 7853 case BPF_FUNC_skc_lookup_tcp: 7854 return &bpf_xdp_skc_lookup_tcp_proto; 7855 case BPF_FUNC_tcp_check_syncookie: 7856 return &bpf_tcp_check_syncookie_proto; 7857 case BPF_FUNC_tcp_gen_syncookie: 7858 return &bpf_tcp_gen_syncookie_proto; 7859 #endif 7860 default: 7861 return bpf_sk_base_func_proto(func_id); 7862 } 7863 } 7864 7865 const struct bpf_func_proto bpf_sock_map_update_proto __weak; 7866 const struct bpf_func_proto bpf_sock_hash_update_proto __weak; 7867 7868 static const struct bpf_func_proto * 7869 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7870 { 7871 switch (func_id) { 7872 case BPF_FUNC_setsockopt: 7873 return &bpf_sock_ops_setsockopt_proto; 7874 case BPF_FUNC_getsockopt: 7875 return &bpf_sock_ops_getsockopt_proto; 7876 case BPF_FUNC_sock_ops_cb_flags_set: 7877 return &bpf_sock_ops_cb_flags_set_proto; 7878 case BPF_FUNC_sock_map_update: 7879 return &bpf_sock_map_update_proto; 7880 case BPF_FUNC_sock_hash_update: 7881 return &bpf_sock_hash_update_proto; 7882 case BPF_FUNC_get_socket_cookie: 7883 return &bpf_get_socket_cookie_sock_ops_proto; 7884 case BPF_FUNC_get_local_storage: 7885 return &bpf_get_local_storage_proto; 7886 case BPF_FUNC_perf_event_output: 7887 return &bpf_event_output_data_proto; 7888 case BPF_FUNC_sk_storage_get: 7889 return &bpf_sk_storage_get_proto; 7890 case BPF_FUNC_sk_storage_delete: 7891 return &bpf_sk_storage_delete_proto; 7892 case BPF_FUNC_get_netns_cookie: 7893 return &bpf_get_netns_cookie_sock_ops_proto; 7894 #ifdef CONFIG_INET 7895 case BPF_FUNC_load_hdr_opt: 7896 return &bpf_sock_ops_load_hdr_opt_proto; 7897 case BPF_FUNC_store_hdr_opt: 7898 return &bpf_sock_ops_store_hdr_opt_proto; 7899 case BPF_FUNC_reserve_hdr_opt: 7900 return &bpf_sock_ops_reserve_hdr_opt_proto; 7901 case BPF_FUNC_tcp_sock: 7902 return &bpf_tcp_sock_proto; 7903 #endif /* CONFIG_INET */ 7904 default: 7905 return bpf_sk_base_func_proto(func_id); 7906 } 7907 } 7908 7909 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak; 7910 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak; 7911 7912 static const struct bpf_func_proto * 7913 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7914 { 7915 switch (func_id) { 7916 case BPF_FUNC_msg_redirect_map: 7917 return &bpf_msg_redirect_map_proto; 7918 case BPF_FUNC_msg_redirect_hash: 7919 return &bpf_msg_redirect_hash_proto; 7920 case BPF_FUNC_msg_apply_bytes: 7921 return &bpf_msg_apply_bytes_proto; 7922 case BPF_FUNC_msg_cork_bytes: 7923 return &bpf_msg_cork_bytes_proto; 7924 case BPF_FUNC_msg_pull_data: 7925 return &bpf_msg_pull_data_proto; 7926 case BPF_FUNC_msg_push_data: 7927 return &bpf_msg_push_data_proto; 7928 case BPF_FUNC_msg_pop_data: 7929 return &bpf_msg_pop_data_proto; 7930 case BPF_FUNC_perf_event_output: 7931 return &bpf_event_output_data_proto; 7932 case BPF_FUNC_get_current_uid_gid: 7933 return &bpf_get_current_uid_gid_proto; 7934 case BPF_FUNC_get_current_pid_tgid: 7935 return &bpf_get_current_pid_tgid_proto; 7936 case BPF_FUNC_sk_storage_get: 7937 return &bpf_sk_storage_get_proto; 7938 case BPF_FUNC_sk_storage_delete: 7939 return &bpf_sk_storage_delete_proto; 7940 case BPF_FUNC_get_netns_cookie: 7941 return &bpf_get_netns_cookie_sk_msg_proto; 7942 #ifdef CONFIG_CGROUPS 7943 case BPF_FUNC_get_current_cgroup_id: 7944 return &bpf_get_current_cgroup_id_proto; 7945 case BPF_FUNC_get_current_ancestor_cgroup_id: 7946 return &bpf_get_current_ancestor_cgroup_id_proto; 7947 #endif 7948 #ifdef CONFIG_CGROUP_NET_CLASSID 7949 case BPF_FUNC_get_cgroup_classid: 7950 return &bpf_get_cgroup_classid_curr_proto; 7951 #endif 7952 default: 7953 return bpf_sk_base_func_proto(func_id); 7954 } 7955 } 7956 7957 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak; 7958 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak; 7959 7960 static const struct bpf_func_proto * 7961 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 7962 { 7963 switch (func_id) { 7964 case BPF_FUNC_skb_store_bytes: 7965 return &bpf_skb_store_bytes_proto; 7966 case BPF_FUNC_skb_load_bytes: 7967 return &bpf_skb_load_bytes_proto; 7968 case BPF_FUNC_skb_pull_data: 7969 return &sk_skb_pull_data_proto; 7970 case BPF_FUNC_skb_change_tail: 7971 return &sk_skb_change_tail_proto; 7972 case BPF_FUNC_skb_change_head: 7973 return &sk_skb_change_head_proto; 7974 case BPF_FUNC_skb_adjust_room: 7975 return &sk_skb_adjust_room_proto; 7976 case BPF_FUNC_get_socket_cookie: 7977 return &bpf_get_socket_cookie_proto; 7978 case BPF_FUNC_get_socket_uid: 7979 return &bpf_get_socket_uid_proto; 7980 case BPF_FUNC_sk_redirect_map: 7981 return &bpf_sk_redirect_map_proto; 7982 case BPF_FUNC_sk_redirect_hash: 7983 return &bpf_sk_redirect_hash_proto; 7984 case BPF_FUNC_perf_event_output: 7985 return &bpf_skb_event_output_proto; 7986 #ifdef CONFIG_INET 7987 case BPF_FUNC_sk_lookup_tcp: 7988 return &bpf_sk_lookup_tcp_proto; 7989 case BPF_FUNC_sk_lookup_udp: 7990 return &bpf_sk_lookup_udp_proto; 7991 case BPF_FUNC_sk_release: 7992 return &bpf_sk_release_proto; 7993 case BPF_FUNC_skc_lookup_tcp: 7994 return &bpf_skc_lookup_tcp_proto; 7995 #endif 7996 default: 7997 return bpf_sk_base_func_proto(func_id); 7998 } 7999 } 8000 8001 static const struct bpf_func_proto * 8002 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8003 { 8004 switch (func_id) { 8005 case BPF_FUNC_skb_load_bytes: 8006 return &bpf_flow_dissector_load_bytes_proto; 8007 default: 8008 return bpf_sk_base_func_proto(func_id); 8009 } 8010 } 8011 8012 static const struct bpf_func_proto * 8013 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8014 { 8015 switch (func_id) { 8016 case BPF_FUNC_skb_load_bytes: 8017 return &bpf_skb_load_bytes_proto; 8018 case BPF_FUNC_skb_pull_data: 8019 return &bpf_skb_pull_data_proto; 8020 case BPF_FUNC_csum_diff: 8021 return &bpf_csum_diff_proto; 8022 case BPF_FUNC_get_cgroup_classid: 8023 return &bpf_get_cgroup_classid_proto; 8024 case BPF_FUNC_get_route_realm: 8025 return &bpf_get_route_realm_proto; 8026 case BPF_FUNC_get_hash_recalc: 8027 return &bpf_get_hash_recalc_proto; 8028 case BPF_FUNC_perf_event_output: 8029 return &bpf_skb_event_output_proto; 8030 case BPF_FUNC_get_smp_processor_id: 8031 return &bpf_get_smp_processor_id_proto; 8032 case BPF_FUNC_skb_under_cgroup: 8033 return &bpf_skb_under_cgroup_proto; 8034 default: 8035 return bpf_sk_base_func_proto(func_id); 8036 } 8037 } 8038 8039 static const struct bpf_func_proto * 8040 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8041 { 8042 switch (func_id) { 8043 case BPF_FUNC_lwt_push_encap: 8044 return &bpf_lwt_in_push_encap_proto; 8045 default: 8046 return lwt_out_func_proto(func_id, prog); 8047 } 8048 } 8049 8050 static const struct bpf_func_proto * 8051 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8052 { 8053 switch (func_id) { 8054 case BPF_FUNC_skb_get_tunnel_key: 8055 return &bpf_skb_get_tunnel_key_proto; 8056 case BPF_FUNC_skb_set_tunnel_key: 8057 return bpf_get_skb_set_tunnel_proto(func_id); 8058 case BPF_FUNC_skb_get_tunnel_opt: 8059 return &bpf_skb_get_tunnel_opt_proto; 8060 case BPF_FUNC_skb_set_tunnel_opt: 8061 return bpf_get_skb_set_tunnel_proto(func_id); 8062 case BPF_FUNC_redirect: 8063 return &bpf_redirect_proto; 8064 case BPF_FUNC_clone_redirect: 8065 return &bpf_clone_redirect_proto; 8066 case BPF_FUNC_skb_change_tail: 8067 return &bpf_skb_change_tail_proto; 8068 case BPF_FUNC_skb_change_head: 8069 return &bpf_skb_change_head_proto; 8070 case BPF_FUNC_skb_store_bytes: 8071 return &bpf_skb_store_bytes_proto; 8072 case BPF_FUNC_csum_update: 8073 return &bpf_csum_update_proto; 8074 case BPF_FUNC_csum_level: 8075 return &bpf_csum_level_proto; 8076 case BPF_FUNC_l3_csum_replace: 8077 return &bpf_l3_csum_replace_proto; 8078 case BPF_FUNC_l4_csum_replace: 8079 return &bpf_l4_csum_replace_proto; 8080 case BPF_FUNC_set_hash_invalid: 8081 return &bpf_set_hash_invalid_proto; 8082 case BPF_FUNC_lwt_push_encap: 8083 return &bpf_lwt_xmit_push_encap_proto; 8084 default: 8085 return lwt_out_func_proto(func_id, prog); 8086 } 8087 } 8088 8089 static const struct bpf_func_proto * 8090 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 8091 { 8092 switch (func_id) { 8093 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 8094 case BPF_FUNC_lwt_seg6_store_bytes: 8095 return &bpf_lwt_seg6_store_bytes_proto; 8096 case BPF_FUNC_lwt_seg6_action: 8097 return &bpf_lwt_seg6_action_proto; 8098 case BPF_FUNC_lwt_seg6_adjust_srh: 8099 return &bpf_lwt_seg6_adjust_srh_proto; 8100 #endif 8101 default: 8102 return lwt_out_func_proto(func_id, prog); 8103 } 8104 } 8105 8106 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type, 8107 const struct bpf_prog *prog, 8108 struct bpf_insn_access_aux *info) 8109 { 8110 const int size_default = sizeof(__u32); 8111 8112 if (off < 0 || off >= sizeof(struct __sk_buff)) 8113 return false; 8114 8115 /* The verifier guarantees that size > 0. */ 8116 if (off % size != 0) 8117 return false; 8118 8119 switch (off) { 8120 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8121 if (off + size > offsetofend(struct __sk_buff, cb[4])) 8122 return false; 8123 break; 8124 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]): 8125 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]): 8126 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4): 8127 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4): 8128 case bpf_ctx_range(struct __sk_buff, data): 8129 case bpf_ctx_range(struct __sk_buff, data_meta): 8130 case bpf_ctx_range(struct __sk_buff, data_end): 8131 if (size != size_default) 8132 return false; 8133 break; 8134 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 8135 return false; 8136 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8137 if (type == BPF_WRITE || size != sizeof(__u64)) 8138 return false; 8139 break; 8140 case bpf_ctx_range(struct __sk_buff, tstamp): 8141 if (size != sizeof(__u64)) 8142 return false; 8143 break; 8144 case offsetof(struct __sk_buff, sk): 8145 if (type == BPF_WRITE || size != sizeof(__u64)) 8146 return false; 8147 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 8148 break; 8149 case offsetof(struct __sk_buff, tstamp_type): 8150 return false; 8151 case offsetofend(struct __sk_buff, tstamp_type) ... offsetof(struct __sk_buff, hwtstamp) - 1: 8152 /* Explicitly prohibit access to padding in __sk_buff. */ 8153 return false; 8154 default: 8155 /* Only narrow read access allowed for now. */ 8156 if (type == BPF_WRITE) { 8157 if (size != size_default) 8158 return false; 8159 } else { 8160 bpf_ctx_record_field_size(info, size_default); 8161 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8162 return false; 8163 } 8164 } 8165 8166 return true; 8167 } 8168 8169 static bool sk_filter_is_valid_access(int off, int size, 8170 enum bpf_access_type type, 8171 const struct bpf_prog *prog, 8172 struct bpf_insn_access_aux *info) 8173 { 8174 switch (off) { 8175 case bpf_ctx_range(struct __sk_buff, tc_classid): 8176 case bpf_ctx_range(struct __sk_buff, data): 8177 case bpf_ctx_range(struct __sk_buff, data_meta): 8178 case bpf_ctx_range(struct __sk_buff, data_end): 8179 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8180 case bpf_ctx_range(struct __sk_buff, tstamp): 8181 case bpf_ctx_range(struct __sk_buff, wire_len): 8182 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8183 return false; 8184 } 8185 8186 if (type == BPF_WRITE) { 8187 switch (off) { 8188 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8189 break; 8190 default: 8191 return false; 8192 } 8193 } 8194 8195 return bpf_skb_is_valid_access(off, size, type, prog, info); 8196 } 8197 8198 static bool cg_skb_is_valid_access(int off, int size, 8199 enum bpf_access_type type, 8200 const struct bpf_prog *prog, 8201 struct bpf_insn_access_aux *info) 8202 { 8203 switch (off) { 8204 case bpf_ctx_range(struct __sk_buff, tc_classid): 8205 case bpf_ctx_range(struct __sk_buff, data_meta): 8206 case bpf_ctx_range(struct __sk_buff, wire_len): 8207 return false; 8208 case bpf_ctx_range(struct __sk_buff, data): 8209 case bpf_ctx_range(struct __sk_buff, data_end): 8210 if (!bpf_capable()) 8211 return false; 8212 break; 8213 } 8214 8215 if (type == BPF_WRITE) { 8216 switch (off) { 8217 case bpf_ctx_range(struct __sk_buff, mark): 8218 case bpf_ctx_range(struct __sk_buff, priority): 8219 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8220 break; 8221 case bpf_ctx_range(struct __sk_buff, tstamp): 8222 if (!bpf_capable()) 8223 return false; 8224 break; 8225 default: 8226 return false; 8227 } 8228 } 8229 8230 switch (off) { 8231 case bpf_ctx_range(struct __sk_buff, data): 8232 info->reg_type = PTR_TO_PACKET; 8233 break; 8234 case bpf_ctx_range(struct __sk_buff, data_end): 8235 info->reg_type = PTR_TO_PACKET_END; 8236 break; 8237 } 8238 8239 return bpf_skb_is_valid_access(off, size, type, prog, info); 8240 } 8241 8242 static bool lwt_is_valid_access(int off, int size, 8243 enum bpf_access_type type, 8244 const struct bpf_prog *prog, 8245 struct bpf_insn_access_aux *info) 8246 { 8247 switch (off) { 8248 case bpf_ctx_range(struct __sk_buff, tc_classid): 8249 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8250 case bpf_ctx_range(struct __sk_buff, data_meta): 8251 case bpf_ctx_range(struct __sk_buff, tstamp): 8252 case bpf_ctx_range(struct __sk_buff, wire_len): 8253 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8254 return false; 8255 } 8256 8257 if (type == BPF_WRITE) { 8258 switch (off) { 8259 case bpf_ctx_range(struct __sk_buff, mark): 8260 case bpf_ctx_range(struct __sk_buff, priority): 8261 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8262 break; 8263 default: 8264 return false; 8265 } 8266 } 8267 8268 switch (off) { 8269 case bpf_ctx_range(struct __sk_buff, data): 8270 info->reg_type = PTR_TO_PACKET; 8271 break; 8272 case bpf_ctx_range(struct __sk_buff, data_end): 8273 info->reg_type = PTR_TO_PACKET_END; 8274 break; 8275 } 8276 8277 return bpf_skb_is_valid_access(off, size, type, prog, info); 8278 } 8279 8280 /* Attach type specific accesses */ 8281 static bool __sock_filter_check_attach_type(int off, 8282 enum bpf_access_type access_type, 8283 enum bpf_attach_type attach_type) 8284 { 8285 switch (off) { 8286 case offsetof(struct bpf_sock, bound_dev_if): 8287 case offsetof(struct bpf_sock, mark): 8288 case offsetof(struct bpf_sock, priority): 8289 switch (attach_type) { 8290 case BPF_CGROUP_INET_SOCK_CREATE: 8291 case BPF_CGROUP_INET_SOCK_RELEASE: 8292 goto full_access; 8293 default: 8294 return false; 8295 } 8296 case bpf_ctx_range(struct bpf_sock, src_ip4): 8297 switch (attach_type) { 8298 case BPF_CGROUP_INET4_POST_BIND: 8299 goto read_only; 8300 default: 8301 return false; 8302 } 8303 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 8304 switch (attach_type) { 8305 case BPF_CGROUP_INET6_POST_BIND: 8306 goto read_only; 8307 default: 8308 return false; 8309 } 8310 case bpf_ctx_range(struct bpf_sock, src_port): 8311 switch (attach_type) { 8312 case BPF_CGROUP_INET4_POST_BIND: 8313 case BPF_CGROUP_INET6_POST_BIND: 8314 goto read_only; 8315 default: 8316 return false; 8317 } 8318 } 8319 read_only: 8320 return access_type == BPF_READ; 8321 full_access: 8322 return true; 8323 } 8324 8325 bool bpf_sock_common_is_valid_access(int off, int size, 8326 enum bpf_access_type type, 8327 struct bpf_insn_access_aux *info) 8328 { 8329 switch (off) { 8330 case bpf_ctx_range_till(struct bpf_sock, type, priority): 8331 return false; 8332 default: 8333 return bpf_sock_is_valid_access(off, size, type, info); 8334 } 8335 } 8336 8337 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 8338 struct bpf_insn_access_aux *info) 8339 { 8340 const int size_default = sizeof(__u32); 8341 int field_size; 8342 8343 if (off < 0 || off >= sizeof(struct bpf_sock)) 8344 return false; 8345 if (off % size != 0) 8346 return false; 8347 8348 switch (off) { 8349 case offsetof(struct bpf_sock, state): 8350 case offsetof(struct bpf_sock, family): 8351 case offsetof(struct bpf_sock, type): 8352 case offsetof(struct bpf_sock, protocol): 8353 case offsetof(struct bpf_sock, src_port): 8354 case offsetof(struct bpf_sock, rx_queue_mapping): 8355 case bpf_ctx_range(struct bpf_sock, src_ip4): 8356 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 8357 case bpf_ctx_range(struct bpf_sock, dst_ip4): 8358 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 8359 bpf_ctx_record_field_size(info, size_default); 8360 return bpf_ctx_narrow_access_ok(off, size, size_default); 8361 case bpf_ctx_range(struct bpf_sock, dst_port): 8362 field_size = size == size_default ? 8363 size_default : sizeof_field(struct bpf_sock, dst_port); 8364 bpf_ctx_record_field_size(info, field_size); 8365 return bpf_ctx_narrow_access_ok(off, size, field_size); 8366 case offsetofend(struct bpf_sock, dst_port) ... 8367 offsetof(struct bpf_sock, dst_ip4) - 1: 8368 return false; 8369 } 8370 8371 return size == size_default; 8372 } 8373 8374 static bool sock_filter_is_valid_access(int off, int size, 8375 enum bpf_access_type type, 8376 const struct bpf_prog *prog, 8377 struct bpf_insn_access_aux *info) 8378 { 8379 if (!bpf_sock_is_valid_access(off, size, type, info)) 8380 return false; 8381 return __sock_filter_check_attach_type(off, type, 8382 prog->expected_attach_type); 8383 } 8384 8385 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write, 8386 const struct bpf_prog *prog) 8387 { 8388 /* Neither direct read nor direct write requires any preliminary 8389 * action. 8390 */ 8391 return 0; 8392 } 8393 8394 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write, 8395 const struct bpf_prog *prog, int drop_verdict) 8396 { 8397 struct bpf_insn *insn = insn_buf; 8398 8399 if (!direct_write) 8400 return 0; 8401 8402 /* if (!skb->cloned) 8403 * goto start; 8404 * 8405 * (Fast-path, otherwise approximation that we might be 8406 * a clone, do the rest in helper.) 8407 */ 8408 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET); 8409 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK); 8410 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7); 8411 8412 /* ret = bpf_skb_pull_data(skb, 0); */ 8413 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1); 8414 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2); 8415 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, 8416 BPF_FUNC_skb_pull_data); 8417 /* if (!ret) 8418 * goto restore; 8419 * return TC_ACT_SHOT; 8420 */ 8421 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2); 8422 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict); 8423 *insn++ = BPF_EXIT_INSN(); 8424 8425 /* restore: */ 8426 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6); 8427 /* start: */ 8428 *insn++ = prog->insnsi[0]; 8429 8430 return insn - insn_buf; 8431 } 8432 8433 static int bpf_gen_ld_abs(const struct bpf_insn *orig, 8434 struct bpf_insn *insn_buf) 8435 { 8436 bool indirect = BPF_MODE(orig->code) == BPF_IND; 8437 struct bpf_insn *insn = insn_buf; 8438 8439 if (!indirect) { 8440 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm); 8441 } else { 8442 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg); 8443 if (orig->imm) 8444 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm); 8445 } 8446 /* We're guaranteed here that CTX is in R6. */ 8447 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX); 8448 8449 switch (BPF_SIZE(orig->code)) { 8450 case BPF_B: 8451 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache); 8452 break; 8453 case BPF_H: 8454 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache); 8455 break; 8456 case BPF_W: 8457 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache); 8458 break; 8459 } 8460 8461 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2); 8462 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0); 8463 *insn++ = BPF_EXIT_INSN(); 8464 8465 return insn - insn_buf; 8466 } 8467 8468 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write, 8469 const struct bpf_prog *prog) 8470 { 8471 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT); 8472 } 8473 8474 static bool tc_cls_act_is_valid_access(int off, int size, 8475 enum bpf_access_type type, 8476 const struct bpf_prog *prog, 8477 struct bpf_insn_access_aux *info) 8478 { 8479 if (type == BPF_WRITE) { 8480 switch (off) { 8481 case bpf_ctx_range(struct __sk_buff, mark): 8482 case bpf_ctx_range(struct __sk_buff, tc_index): 8483 case bpf_ctx_range(struct __sk_buff, priority): 8484 case bpf_ctx_range(struct __sk_buff, tc_classid): 8485 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 8486 case bpf_ctx_range(struct __sk_buff, tstamp): 8487 case bpf_ctx_range(struct __sk_buff, queue_mapping): 8488 break; 8489 default: 8490 return false; 8491 } 8492 } 8493 8494 switch (off) { 8495 case bpf_ctx_range(struct __sk_buff, data): 8496 info->reg_type = PTR_TO_PACKET; 8497 break; 8498 case bpf_ctx_range(struct __sk_buff, data_meta): 8499 info->reg_type = PTR_TO_PACKET_META; 8500 break; 8501 case bpf_ctx_range(struct __sk_buff, data_end): 8502 info->reg_type = PTR_TO_PACKET_END; 8503 break; 8504 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 8505 return false; 8506 case offsetof(struct __sk_buff, tstamp_type): 8507 /* The convert_ctx_access() on reading and writing 8508 * __sk_buff->tstamp depends on whether the bpf prog 8509 * has used __sk_buff->tstamp_type or not. 8510 * Thus, we need to set prog->tstamp_type_access 8511 * earlier during is_valid_access() here. 8512 */ 8513 ((struct bpf_prog *)prog)->tstamp_type_access = 1; 8514 return size == sizeof(__u8); 8515 } 8516 8517 return bpf_skb_is_valid_access(off, size, type, prog, info); 8518 } 8519 8520 static bool __is_valid_xdp_access(int off, int size) 8521 { 8522 if (off < 0 || off >= sizeof(struct xdp_md)) 8523 return false; 8524 if (off % size != 0) 8525 return false; 8526 if (size != sizeof(__u32)) 8527 return false; 8528 8529 return true; 8530 } 8531 8532 static bool xdp_is_valid_access(int off, int size, 8533 enum bpf_access_type type, 8534 const struct bpf_prog *prog, 8535 struct bpf_insn_access_aux *info) 8536 { 8537 if (prog->expected_attach_type != BPF_XDP_DEVMAP) { 8538 switch (off) { 8539 case offsetof(struct xdp_md, egress_ifindex): 8540 return false; 8541 } 8542 } 8543 8544 if (type == BPF_WRITE) { 8545 if (bpf_prog_is_dev_bound(prog->aux)) { 8546 switch (off) { 8547 case offsetof(struct xdp_md, rx_queue_index): 8548 return __is_valid_xdp_access(off, size); 8549 } 8550 } 8551 return false; 8552 } 8553 8554 switch (off) { 8555 case offsetof(struct xdp_md, data): 8556 info->reg_type = PTR_TO_PACKET; 8557 break; 8558 case offsetof(struct xdp_md, data_meta): 8559 info->reg_type = PTR_TO_PACKET_META; 8560 break; 8561 case offsetof(struct xdp_md, data_end): 8562 info->reg_type = PTR_TO_PACKET_END; 8563 break; 8564 } 8565 8566 return __is_valid_xdp_access(off, size); 8567 } 8568 8569 void bpf_warn_invalid_xdp_action(struct net_device *dev, struct bpf_prog *prog, u32 act) 8570 { 8571 const u32 act_max = XDP_REDIRECT; 8572 8573 pr_warn_once("%s XDP return value %u on prog %s (id %d) dev %s, expect packet loss!\n", 8574 act > act_max ? "Illegal" : "Driver unsupported", 8575 act, prog->aux->name, prog->aux->id, dev ? dev->name : "N/A"); 8576 } 8577 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); 8578 8579 static bool sock_addr_is_valid_access(int off, int size, 8580 enum bpf_access_type type, 8581 const struct bpf_prog *prog, 8582 struct bpf_insn_access_aux *info) 8583 { 8584 const int size_default = sizeof(__u32); 8585 8586 if (off < 0 || off >= sizeof(struct bpf_sock_addr)) 8587 return false; 8588 if (off % size != 0) 8589 return false; 8590 8591 /* Disallow access to IPv6 fields from IPv4 contex and vise 8592 * versa. 8593 */ 8594 switch (off) { 8595 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 8596 switch (prog->expected_attach_type) { 8597 case BPF_CGROUP_INET4_BIND: 8598 case BPF_CGROUP_INET4_CONNECT: 8599 case BPF_CGROUP_INET4_GETPEERNAME: 8600 case BPF_CGROUP_INET4_GETSOCKNAME: 8601 case BPF_CGROUP_UDP4_SENDMSG: 8602 case BPF_CGROUP_UDP4_RECVMSG: 8603 break; 8604 default: 8605 return false; 8606 } 8607 break; 8608 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8609 switch (prog->expected_attach_type) { 8610 case BPF_CGROUP_INET6_BIND: 8611 case BPF_CGROUP_INET6_CONNECT: 8612 case BPF_CGROUP_INET6_GETPEERNAME: 8613 case BPF_CGROUP_INET6_GETSOCKNAME: 8614 case BPF_CGROUP_UDP6_SENDMSG: 8615 case BPF_CGROUP_UDP6_RECVMSG: 8616 break; 8617 default: 8618 return false; 8619 } 8620 break; 8621 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8622 switch (prog->expected_attach_type) { 8623 case BPF_CGROUP_UDP4_SENDMSG: 8624 break; 8625 default: 8626 return false; 8627 } 8628 break; 8629 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8630 msg_src_ip6[3]): 8631 switch (prog->expected_attach_type) { 8632 case BPF_CGROUP_UDP6_SENDMSG: 8633 break; 8634 default: 8635 return false; 8636 } 8637 break; 8638 } 8639 8640 switch (off) { 8641 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 8642 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 8643 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 8644 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8645 msg_src_ip6[3]): 8646 case bpf_ctx_range(struct bpf_sock_addr, user_port): 8647 if (type == BPF_READ) { 8648 bpf_ctx_record_field_size(info, size_default); 8649 8650 if (bpf_ctx_wide_access_ok(off, size, 8651 struct bpf_sock_addr, 8652 user_ip6)) 8653 return true; 8654 8655 if (bpf_ctx_wide_access_ok(off, size, 8656 struct bpf_sock_addr, 8657 msg_src_ip6)) 8658 return true; 8659 8660 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 8661 return false; 8662 } else { 8663 if (bpf_ctx_wide_access_ok(off, size, 8664 struct bpf_sock_addr, 8665 user_ip6)) 8666 return true; 8667 8668 if (bpf_ctx_wide_access_ok(off, size, 8669 struct bpf_sock_addr, 8670 msg_src_ip6)) 8671 return true; 8672 8673 if (size != size_default) 8674 return false; 8675 } 8676 break; 8677 case offsetof(struct bpf_sock_addr, sk): 8678 if (type != BPF_READ) 8679 return false; 8680 if (size != sizeof(__u64)) 8681 return false; 8682 info->reg_type = PTR_TO_SOCKET; 8683 break; 8684 default: 8685 if (type == BPF_READ) { 8686 if (size != size_default) 8687 return false; 8688 } else { 8689 return false; 8690 } 8691 } 8692 8693 return true; 8694 } 8695 8696 static bool sock_ops_is_valid_access(int off, int size, 8697 enum bpf_access_type type, 8698 const struct bpf_prog *prog, 8699 struct bpf_insn_access_aux *info) 8700 { 8701 const int size_default = sizeof(__u32); 8702 8703 if (off < 0 || off >= sizeof(struct bpf_sock_ops)) 8704 return false; 8705 8706 /* The verifier guarantees that size > 0. */ 8707 if (off % size != 0) 8708 return false; 8709 8710 if (type == BPF_WRITE) { 8711 switch (off) { 8712 case offsetof(struct bpf_sock_ops, reply): 8713 case offsetof(struct bpf_sock_ops, sk_txhash): 8714 if (size != size_default) 8715 return false; 8716 break; 8717 default: 8718 return false; 8719 } 8720 } else { 8721 switch (off) { 8722 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received, 8723 bytes_acked): 8724 if (size != sizeof(__u64)) 8725 return false; 8726 break; 8727 case offsetof(struct bpf_sock_ops, sk): 8728 if (size != sizeof(__u64)) 8729 return false; 8730 info->reg_type = PTR_TO_SOCKET_OR_NULL; 8731 break; 8732 case offsetof(struct bpf_sock_ops, skb_data): 8733 if (size != sizeof(__u64)) 8734 return false; 8735 info->reg_type = PTR_TO_PACKET; 8736 break; 8737 case offsetof(struct bpf_sock_ops, skb_data_end): 8738 if (size != sizeof(__u64)) 8739 return false; 8740 info->reg_type = PTR_TO_PACKET_END; 8741 break; 8742 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 8743 bpf_ctx_record_field_size(info, size_default); 8744 return bpf_ctx_narrow_access_ok(off, size, 8745 size_default); 8746 default: 8747 if (size != size_default) 8748 return false; 8749 break; 8750 } 8751 } 8752 8753 return true; 8754 } 8755 8756 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write, 8757 const struct bpf_prog *prog) 8758 { 8759 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP); 8760 } 8761 8762 static bool sk_skb_is_valid_access(int off, int size, 8763 enum bpf_access_type type, 8764 const struct bpf_prog *prog, 8765 struct bpf_insn_access_aux *info) 8766 { 8767 switch (off) { 8768 case bpf_ctx_range(struct __sk_buff, tc_classid): 8769 case bpf_ctx_range(struct __sk_buff, data_meta): 8770 case bpf_ctx_range(struct __sk_buff, tstamp): 8771 case bpf_ctx_range(struct __sk_buff, wire_len): 8772 case bpf_ctx_range(struct __sk_buff, hwtstamp): 8773 return false; 8774 } 8775 8776 if (type == BPF_WRITE) { 8777 switch (off) { 8778 case bpf_ctx_range(struct __sk_buff, tc_index): 8779 case bpf_ctx_range(struct __sk_buff, priority): 8780 break; 8781 default: 8782 return false; 8783 } 8784 } 8785 8786 switch (off) { 8787 case bpf_ctx_range(struct __sk_buff, mark): 8788 return false; 8789 case bpf_ctx_range(struct __sk_buff, data): 8790 info->reg_type = PTR_TO_PACKET; 8791 break; 8792 case bpf_ctx_range(struct __sk_buff, data_end): 8793 info->reg_type = PTR_TO_PACKET_END; 8794 break; 8795 } 8796 8797 return bpf_skb_is_valid_access(off, size, type, prog, info); 8798 } 8799 8800 static bool sk_msg_is_valid_access(int off, int size, 8801 enum bpf_access_type type, 8802 const struct bpf_prog *prog, 8803 struct bpf_insn_access_aux *info) 8804 { 8805 if (type == BPF_WRITE) 8806 return false; 8807 8808 if (off % size != 0) 8809 return false; 8810 8811 switch (off) { 8812 case offsetof(struct sk_msg_md, data): 8813 info->reg_type = PTR_TO_PACKET; 8814 if (size != sizeof(__u64)) 8815 return false; 8816 break; 8817 case offsetof(struct sk_msg_md, data_end): 8818 info->reg_type = PTR_TO_PACKET_END; 8819 if (size != sizeof(__u64)) 8820 return false; 8821 break; 8822 case offsetof(struct sk_msg_md, sk): 8823 if (size != sizeof(__u64)) 8824 return false; 8825 info->reg_type = PTR_TO_SOCKET; 8826 break; 8827 case bpf_ctx_range(struct sk_msg_md, family): 8828 case bpf_ctx_range(struct sk_msg_md, remote_ip4): 8829 case bpf_ctx_range(struct sk_msg_md, local_ip4): 8830 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]): 8831 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]): 8832 case bpf_ctx_range(struct sk_msg_md, remote_port): 8833 case bpf_ctx_range(struct sk_msg_md, local_port): 8834 case bpf_ctx_range(struct sk_msg_md, size): 8835 if (size != sizeof(__u32)) 8836 return false; 8837 break; 8838 default: 8839 return false; 8840 } 8841 return true; 8842 } 8843 8844 static bool flow_dissector_is_valid_access(int off, int size, 8845 enum bpf_access_type type, 8846 const struct bpf_prog *prog, 8847 struct bpf_insn_access_aux *info) 8848 { 8849 const int size_default = sizeof(__u32); 8850 8851 if (off < 0 || off >= sizeof(struct __sk_buff)) 8852 return false; 8853 8854 if (type == BPF_WRITE) 8855 return false; 8856 8857 switch (off) { 8858 case bpf_ctx_range(struct __sk_buff, data): 8859 if (size != size_default) 8860 return false; 8861 info->reg_type = PTR_TO_PACKET; 8862 return true; 8863 case bpf_ctx_range(struct __sk_buff, data_end): 8864 if (size != size_default) 8865 return false; 8866 info->reg_type = PTR_TO_PACKET_END; 8867 return true; 8868 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 8869 if (size != sizeof(__u64)) 8870 return false; 8871 info->reg_type = PTR_TO_FLOW_KEYS; 8872 return true; 8873 default: 8874 return false; 8875 } 8876 } 8877 8878 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type, 8879 const struct bpf_insn *si, 8880 struct bpf_insn *insn_buf, 8881 struct bpf_prog *prog, 8882 u32 *target_size) 8883 8884 { 8885 struct bpf_insn *insn = insn_buf; 8886 8887 switch (si->off) { 8888 case offsetof(struct __sk_buff, data): 8889 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data), 8890 si->dst_reg, si->src_reg, 8891 offsetof(struct bpf_flow_dissector, data)); 8892 break; 8893 8894 case offsetof(struct __sk_buff, data_end): 8895 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end), 8896 si->dst_reg, si->src_reg, 8897 offsetof(struct bpf_flow_dissector, data_end)); 8898 break; 8899 8900 case offsetof(struct __sk_buff, flow_keys): 8901 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys), 8902 si->dst_reg, si->src_reg, 8903 offsetof(struct bpf_flow_dissector, flow_keys)); 8904 break; 8905 } 8906 8907 return insn - insn_buf; 8908 } 8909 8910 static struct bpf_insn *bpf_convert_tstamp_type_read(const struct bpf_insn *si, 8911 struct bpf_insn *insn) 8912 { 8913 __u8 value_reg = si->dst_reg; 8914 __u8 skb_reg = si->src_reg; 8915 /* AX is needed because src_reg and dst_reg could be the same */ 8916 __u8 tmp_reg = BPF_REG_AX; 8917 8918 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, 8919 PKT_VLAN_PRESENT_OFFSET); 8920 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, 8921 SKB_MONO_DELIVERY_TIME_MASK, 2); 8922 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_UNSPEC); 8923 *insn++ = BPF_JMP_A(1); 8924 *insn++ = BPF_MOV32_IMM(value_reg, BPF_SKB_TSTAMP_DELIVERY_MONO); 8925 8926 return insn; 8927 } 8928 8929 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si, 8930 struct bpf_insn *insn) 8931 { 8932 /* si->dst_reg = skb_shinfo(SKB); */ 8933 #ifdef NET_SKBUFF_DATA_USES_OFFSET 8934 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 8935 BPF_REG_AX, si->src_reg, 8936 offsetof(struct sk_buff, end)); 8937 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head), 8938 si->dst_reg, si->src_reg, 8939 offsetof(struct sk_buff, head)); 8940 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX); 8941 #else 8942 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 8943 si->dst_reg, si->src_reg, 8944 offsetof(struct sk_buff, end)); 8945 #endif 8946 8947 return insn; 8948 } 8949 8950 static struct bpf_insn *bpf_convert_tstamp_read(const struct bpf_prog *prog, 8951 const struct bpf_insn *si, 8952 struct bpf_insn *insn) 8953 { 8954 __u8 value_reg = si->dst_reg; 8955 __u8 skb_reg = si->src_reg; 8956 8957 #ifdef CONFIG_NET_CLS_ACT 8958 /* If the tstamp_type is read, 8959 * the bpf prog is aware the tstamp could have delivery time. 8960 * Thus, read skb->tstamp as is if tstamp_type_access is true. 8961 */ 8962 if (!prog->tstamp_type_access) { 8963 /* AX is needed because src_reg and dst_reg could be the same */ 8964 __u8 tmp_reg = BPF_REG_AX; 8965 8966 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET); 8967 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, 8968 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK); 8969 *insn++ = BPF_JMP32_IMM(BPF_JNE, tmp_reg, 8970 TC_AT_INGRESS_MASK | SKB_MONO_DELIVERY_TIME_MASK, 2); 8971 /* skb->tc_at_ingress && skb->mono_delivery_time, 8972 * read 0 as the (rcv) timestamp. 8973 */ 8974 *insn++ = BPF_MOV64_IMM(value_reg, 0); 8975 *insn++ = BPF_JMP_A(1); 8976 } 8977 #endif 8978 8979 *insn++ = BPF_LDX_MEM(BPF_DW, value_reg, skb_reg, 8980 offsetof(struct sk_buff, tstamp)); 8981 return insn; 8982 } 8983 8984 static struct bpf_insn *bpf_convert_tstamp_write(const struct bpf_prog *prog, 8985 const struct bpf_insn *si, 8986 struct bpf_insn *insn) 8987 { 8988 __u8 value_reg = si->src_reg; 8989 __u8 skb_reg = si->dst_reg; 8990 8991 #ifdef CONFIG_NET_CLS_ACT 8992 /* If the tstamp_type is read, 8993 * the bpf prog is aware the tstamp could have delivery time. 8994 * Thus, write skb->tstamp as is if tstamp_type_access is true. 8995 * Otherwise, writing at ingress will have to clear the 8996 * mono_delivery_time bit also. 8997 */ 8998 if (!prog->tstamp_type_access) { 8999 __u8 tmp_reg = BPF_REG_AX; 9000 9001 *insn++ = BPF_LDX_MEM(BPF_B, tmp_reg, skb_reg, PKT_VLAN_PRESENT_OFFSET); 9002 /* Writing __sk_buff->tstamp as ingress, goto <clear> */ 9003 *insn++ = BPF_JMP32_IMM(BPF_JSET, tmp_reg, TC_AT_INGRESS_MASK, 1); 9004 /* goto <store> */ 9005 *insn++ = BPF_JMP_A(2); 9006 /* <clear>: mono_delivery_time */ 9007 *insn++ = BPF_ALU32_IMM(BPF_AND, tmp_reg, ~SKB_MONO_DELIVERY_TIME_MASK); 9008 *insn++ = BPF_STX_MEM(BPF_B, skb_reg, tmp_reg, PKT_VLAN_PRESENT_OFFSET); 9009 } 9010 #endif 9011 9012 /* <store>: skb->tstamp = tstamp */ 9013 *insn++ = BPF_STX_MEM(BPF_DW, skb_reg, value_reg, 9014 offsetof(struct sk_buff, tstamp)); 9015 return insn; 9016 } 9017 9018 static u32 bpf_convert_ctx_access(enum bpf_access_type type, 9019 const struct bpf_insn *si, 9020 struct bpf_insn *insn_buf, 9021 struct bpf_prog *prog, u32 *target_size) 9022 { 9023 struct bpf_insn *insn = insn_buf; 9024 int off; 9025 9026 switch (si->off) { 9027 case offsetof(struct __sk_buff, len): 9028 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9029 bpf_target_off(struct sk_buff, len, 4, 9030 target_size)); 9031 break; 9032 9033 case offsetof(struct __sk_buff, protocol): 9034 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9035 bpf_target_off(struct sk_buff, protocol, 2, 9036 target_size)); 9037 break; 9038 9039 case offsetof(struct __sk_buff, vlan_proto): 9040 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9041 bpf_target_off(struct sk_buff, vlan_proto, 2, 9042 target_size)); 9043 break; 9044 9045 case offsetof(struct __sk_buff, priority): 9046 if (type == BPF_WRITE) 9047 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9048 bpf_target_off(struct sk_buff, priority, 4, 9049 target_size)); 9050 else 9051 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9052 bpf_target_off(struct sk_buff, priority, 4, 9053 target_size)); 9054 break; 9055 9056 case offsetof(struct __sk_buff, ingress_ifindex): 9057 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9058 bpf_target_off(struct sk_buff, skb_iif, 4, 9059 target_size)); 9060 break; 9061 9062 case offsetof(struct __sk_buff, ifindex): 9063 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 9064 si->dst_reg, si->src_reg, 9065 offsetof(struct sk_buff, dev)); 9066 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 9067 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9068 bpf_target_off(struct net_device, ifindex, 4, 9069 target_size)); 9070 break; 9071 9072 case offsetof(struct __sk_buff, hash): 9073 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9074 bpf_target_off(struct sk_buff, hash, 4, 9075 target_size)); 9076 break; 9077 9078 case offsetof(struct __sk_buff, mark): 9079 if (type == BPF_WRITE) 9080 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9081 bpf_target_off(struct sk_buff, mark, 4, 9082 target_size)); 9083 else 9084 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9085 bpf_target_off(struct sk_buff, mark, 4, 9086 target_size)); 9087 break; 9088 9089 case offsetof(struct __sk_buff, pkt_type): 9090 *target_size = 1; 9091 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 9092 PKT_TYPE_OFFSET); 9093 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX); 9094 #ifdef __BIG_ENDIAN_BITFIELD 9095 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5); 9096 #endif 9097 break; 9098 9099 case offsetof(struct __sk_buff, queue_mapping): 9100 if (type == BPF_WRITE) { 9101 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1); 9102 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 9103 bpf_target_off(struct sk_buff, 9104 queue_mapping, 9105 2, target_size)); 9106 } else { 9107 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9108 bpf_target_off(struct sk_buff, 9109 queue_mapping, 9110 2, target_size)); 9111 } 9112 break; 9113 9114 case offsetof(struct __sk_buff, vlan_present): 9115 *target_size = 1; 9116 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 9117 PKT_VLAN_PRESENT_OFFSET); 9118 if (PKT_VLAN_PRESENT_BIT) 9119 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT); 9120 if (PKT_VLAN_PRESENT_BIT < 7) 9121 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1); 9122 break; 9123 9124 case offsetof(struct __sk_buff, vlan_tci): 9125 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9126 bpf_target_off(struct sk_buff, vlan_tci, 2, 9127 target_size)); 9128 break; 9129 9130 case offsetof(struct __sk_buff, cb[0]) ... 9131 offsetofend(struct __sk_buff, cb[4]) - 1: 9132 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20); 9133 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 9134 offsetof(struct qdisc_skb_cb, data)) % 9135 sizeof(__u64)); 9136 9137 prog->cb_access = 1; 9138 off = si->off; 9139 off -= offsetof(struct __sk_buff, cb[0]); 9140 off += offsetof(struct sk_buff, cb); 9141 off += offsetof(struct qdisc_skb_cb, data); 9142 if (type == BPF_WRITE) 9143 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 9144 si->src_reg, off); 9145 else 9146 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 9147 si->src_reg, off); 9148 break; 9149 9150 case offsetof(struct __sk_buff, tc_classid): 9151 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2); 9152 9153 off = si->off; 9154 off -= offsetof(struct __sk_buff, tc_classid); 9155 off += offsetof(struct sk_buff, cb); 9156 off += offsetof(struct qdisc_skb_cb, tc_classid); 9157 *target_size = 2; 9158 if (type == BPF_WRITE) 9159 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, 9160 si->src_reg, off); 9161 else 9162 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, 9163 si->src_reg, off); 9164 break; 9165 9166 case offsetof(struct __sk_buff, data): 9167 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 9168 si->dst_reg, si->src_reg, 9169 offsetof(struct sk_buff, data)); 9170 break; 9171 9172 case offsetof(struct __sk_buff, data_meta): 9173 off = si->off; 9174 off -= offsetof(struct __sk_buff, data_meta); 9175 off += offsetof(struct sk_buff, cb); 9176 off += offsetof(struct bpf_skb_data_end, data_meta); 9177 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 9178 si->src_reg, off); 9179 break; 9180 9181 case offsetof(struct __sk_buff, data_end): 9182 off = si->off; 9183 off -= offsetof(struct __sk_buff, data_end); 9184 off += offsetof(struct sk_buff, cb); 9185 off += offsetof(struct bpf_skb_data_end, data_end); 9186 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 9187 si->src_reg, off); 9188 break; 9189 9190 case offsetof(struct __sk_buff, tc_index): 9191 #ifdef CONFIG_NET_SCHED 9192 if (type == BPF_WRITE) 9193 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 9194 bpf_target_off(struct sk_buff, tc_index, 2, 9195 target_size)); 9196 else 9197 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 9198 bpf_target_off(struct sk_buff, tc_index, 2, 9199 target_size)); 9200 #else 9201 *target_size = 2; 9202 if (type == BPF_WRITE) 9203 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg); 9204 else 9205 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9206 #endif 9207 break; 9208 9209 case offsetof(struct __sk_buff, napi_id): 9210 #if defined(CONFIG_NET_RX_BUSY_POLL) 9211 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9212 bpf_target_off(struct sk_buff, napi_id, 4, 9213 target_size)); 9214 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1); 9215 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9216 #else 9217 *target_size = 4; 9218 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 9219 #endif 9220 break; 9221 case offsetof(struct __sk_buff, family): 9222 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 9223 9224 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9225 si->dst_reg, si->src_reg, 9226 offsetof(struct sk_buff, sk)); 9227 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9228 bpf_target_off(struct sock_common, 9229 skc_family, 9230 2, target_size)); 9231 break; 9232 case offsetof(struct __sk_buff, remote_ip4): 9233 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 9234 9235 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9236 si->dst_reg, si->src_reg, 9237 offsetof(struct sk_buff, sk)); 9238 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9239 bpf_target_off(struct sock_common, 9240 skc_daddr, 9241 4, target_size)); 9242 break; 9243 case offsetof(struct __sk_buff, local_ip4): 9244 BUILD_BUG_ON(sizeof_field(struct sock_common, 9245 skc_rcv_saddr) != 4); 9246 9247 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9248 si->dst_reg, si->src_reg, 9249 offsetof(struct sk_buff, sk)); 9250 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9251 bpf_target_off(struct sock_common, 9252 skc_rcv_saddr, 9253 4, target_size)); 9254 break; 9255 case offsetof(struct __sk_buff, remote_ip6[0]) ... 9256 offsetof(struct __sk_buff, remote_ip6[3]): 9257 #if IS_ENABLED(CONFIG_IPV6) 9258 BUILD_BUG_ON(sizeof_field(struct sock_common, 9259 skc_v6_daddr.s6_addr32[0]) != 4); 9260 9261 off = si->off; 9262 off -= offsetof(struct __sk_buff, remote_ip6[0]); 9263 9264 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9265 si->dst_reg, si->src_reg, 9266 offsetof(struct sk_buff, sk)); 9267 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9268 offsetof(struct sock_common, 9269 skc_v6_daddr.s6_addr32[0]) + 9270 off); 9271 #else 9272 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9273 #endif 9274 break; 9275 case offsetof(struct __sk_buff, local_ip6[0]) ... 9276 offsetof(struct __sk_buff, local_ip6[3]): 9277 #if IS_ENABLED(CONFIG_IPV6) 9278 BUILD_BUG_ON(sizeof_field(struct sock_common, 9279 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 9280 9281 off = si->off; 9282 off -= offsetof(struct __sk_buff, local_ip6[0]); 9283 9284 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9285 si->dst_reg, si->src_reg, 9286 offsetof(struct sk_buff, sk)); 9287 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9288 offsetof(struct sock_common, 9289 skc_v6_rcv_saddr.s6_addr32[0]) + 9290 off); 9291 #else 9292 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9293 #endif 9294 break; 9295 9296 case offsetof(struct __sk_buff, remote_port): 9297 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 9298 9299 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9300 si->dst_reg, si->src_reg, 9301 offsetof(struct sk_buff, sk)); 9302 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9303 bpf_target_off(struct sock_common, 9304 skc_dport, 9305 2, target_size)); 9306 #ifndef __BIG_ENDIAN_BITFIELD 9307 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 9308 #endif 9309 break; 9310 9311 case offsetof(struct __sk_buff, local_port): 9312 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 9313 9314 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9315 si->dst_reg, si->src_reg, 9316 offsetof(struct sk_buff, sk)); 9317 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9318 bpf_target_off(struct sock_common, 9319 skc_num, 2, target_size)); 9320 break; 9321 9322 case offsetof(struct __sk_buff, tstamp): 9323 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8); 9324 9325 if (type == BPF_WRITE) 9326 insn = bpf_convert_tstamp_write(prog, si, insn); 9327 else 9328 insn = bpf_convert_tstamp_read(prog, si, insn); 9329 break; 9330 9331 case offsetof(struct __sk_buff, tstamp_type): 9332 insn = bpf_convert_tstamp_type_read(si, insn); 9333 break; 9334 9335 case offsetof(struct __sk_buff, gso_segs): 9336 insn = bpf_convert_shinfo_access(si, insn); 9337 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs), 9338 si->dst_reg, si->dst_reg, 9339 bpf_target_off(struct skb_shared_info, 9340 gso_segs, 2, 9341 target_size)); 9342 break; 9343 case offsetof(struct __sk_buff, gso_size): 9344 insn = bpf_convert_shinfo_access(si, insn); 9345 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size), 9346 si->dst_reg, si->dst_reg, 9347 bpf_target_off(struct skb_shared_info, 9348 gso_size, 2, 9349 target_size)); 9350 break; 9351 case offsetof(struct __sk_buff, wire_len): 9352 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4); 9353 9354 off = si->off; 9355 off -= offsetof(struct __sk_buff, wire_len); 9356 off += offsetof(struct sk_buff, cb); 9357 off += offsetof(struct qdisc_skb_cb, pkt_len); 9358 *target_size = 4; 9359 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off); 9360 break; 9361 9362 case offsetof(struct __sk_buff, sk): 9363 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 9364 si->dst_reg, si->src_reg, 9365 offsetof(struct sk_buff, sk)); 9366 break; 9367 case offsetof(struct __sk_buff, hwtstamp): 9368 BUILD_BUG_ON(sizeof_field(struct skb_shared_hwtstamps, hwtstamp) != 8); 9369 BUILD_BUG_ON(offsetof(struct skb_shared_hwtstamps, hwtstamp) != 0); 9370 9371 insn = bpf_convert_shinfo_access(si, insn); 9372 *insn++ = BPF_LDX_MEM(BPF_DW, 9373 si->dst_reg, si->dst_reg, 9374 bpf_target_off(struct skb_shared_info, 9375 hwtstamps, 8, 9376 target_size)); 9377 break; 9378 } 9379 9380 return insn - insn_buf; 9381 } 9382 9383 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 9384 const struct bpf_insn *si, 9385 struct bpf_insn *insn_buf, 9386 struct bpf_prog *prog, u32 *target_size) 9387 { 9388 struct bpf_insn *insn = insn_buf; 9389 int off; 9390 9391 switch (si->off) { 9392 case offsetof(struct bpf_sock, bound_dev_if): 9393 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4); 9394 9395 if (type == BPF_WRITE) 9396 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9397 offsetof(struct sock, sk_bound_dev_if)); 9398 else 9399 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9400 offsetof(struct sock, sk_bound_dev_if)); 9401 break; 9402 9403 case offsetof(struct bpf_sock, mark): 9404 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4); 9405 9406 if (type == BPF_WRITE) 9407 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9408 offsetof(struct sock, sk_mark)); 9409 else 9410 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9411 offsetof(struct sock, sk_mark)); 9412 break; 9413 9414 case offsetof(struct bpf_sock, priority): 9415 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4); 9416 9417 if (type == BPF_WRITE) 9418 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9419 offsetof(struct sock, sk_priority)); 9420 else 9421 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9422 offsetof(struct sock, sk_priority)); 9423 break; 9424 9425 case offsetof(struct bpf_sock, family): 9426 *insn++ = BPF_LDX_MEM( 9427 BPF_FIELD_SIZEOF(struct sock_common, skc_family), 9428 si->dst_reg, si->src_reg, 9429 bpf_target_off(struct sock_common, 9430 skc_family, 9431 sizeof_field(struct sock_common, 9432 skc_family), 9433 target_size)); 9434 break; 9435 9436 case offsetof(struct bpf_sock, type): 9437 *insn++ = BPF_LDX_MEM( 9438 BPF_FIELD_SIZEOF(struct sock, sk_type), 9439 si->dst_reg, si->src_reg, 9440 bpf_target_off(struct sock, sk_type, 9441 sizeof_field(struct sock, sk_type), 9442 target_size)); 9443 break; 9444 9445 case offsetof(struct bpf_sock, protocol): 9446 *insn++ = BPF_LDX_MEM( 9447 BPF_FIELD_SIZEOF(struct sock, sk_protocol), 9448 si->dst_reg, si->src_reg, 9449 bpf_target_off(struct sock, sk_protocol, 9450 sizeof_field(struct sock, sk_protocol), 9451 target_size)); 9452 break; 9453 9454 case offsetof(struct bpf_sock, src_ip4): 9455 *insn++ = BPF_LDX_MEM( 9456 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9457 bpf_target_off(struct sock_common, skc_rcv_saddr, 9458 sizeof_field(struct sock_common, 9459 skc_rcv_saddr), 9460 target_size)); 9461 break; 9462 9463 case offsetof(struct bpf_sock, dst_ip4): 9464 *insn++ = BPF_LDX_MEM( 9465 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9466 bpf_target_off(struct sock_common, skc_daddr, 9467 sizeof_field(struct sock_common, 9468 skc_daddr), 9469 target_size)); 9470 break; 9471 9472 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 9473 #if IS_ENABLED(CONFIG_IPV6) 9474 off = si->off; 9475 off -= offsetof(struct bpf_sock, src_ip6[0]); 9476 *insn++ = BPF_LDX_MEM( 9477 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9478 bpf_target_off( 9479 struct sock_common, 9480 skc_v6_rcv_saddr.s6_addr32[0], 9481 sizeof_field(struct sock_common, 9482 skc_v6_rcv_saddr.s6_addr32[0]), 9483 target_size) + off); 9484 #else 9485 (void)off; 9486 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9487 #endif 9488 break; 9489 9490 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 9491 #if IS_ENABLED(CONFIG_IPV6) 9492 off = si->off; 9493 off -= offsetof(struct bpf_sock, dst_ip6[0]); 9494 *insn++ = BPF_LDX_MEM( 9495 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 9496 bpf_target_off(struct sock_common, 9497 skc_v6_daddr.s6_addr32[0], 9498 sizeof_field(struct sock_common, 9499 skc_v6_daddr.s6_addr32[0]), 9500 target_size) + off); 9501 #else 9502 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 9503 *target_size = 4; 9504 #endif 9505 break; 9506 9507 case offsetof(struct bpf_sock, src_port): 9508 *insn++ = BPF_LDX_MEM( 9509 BPF_FIELD_SIZEOF(struct sock_common, skc_num), 9510 si->dst_reg, si->src_reg, 9511 bpf_target_off(struct sock_common, skc_num, 9512 sizeof_field(struct sock_common, 9513 skc_num), 9514 target_size)); 9515 break; 9516 9517 case offsetof(struct bpf_sock, dst_port): 9518 *insn++ = BPF_LDX_MEM( 9519 BPF_FIELD_SIZEOF(struct sock_common, skc_dport), 9520 si->dst_reg, si->src_reg, 9521 bpf_target_off(struct sock_common, skc_dport, 9522 sizeof_field(struct sock_common, 9523 skc_dport), 9524 target_size)); 9525 break; 9526 9527 case offsetof(struct bpf_sock, state): 9528 *insn++ = BPF_LDX_MEM( 9529 BPF_FIELD_SIZEOF(struct sock_common, skc_state), 9530 si->dst_reg, si->src_reg, 9531 bpf_target_off(struct sock_common, skc_state, 9532 sizeof_field(struct sock_common, 9533 skc_state), 9534 target_size)); 9535 break; 9536 case offsetof(struct bpf_sock, rx_queue_mapping): 9537 #ifdef CONFIG_SOCK_RX_QUEUE_MAPPING 9538 *insn++ = BPF_LDX_MEM( 9539 BPF_FIELD_SIZEOF(struct sock, sk_rx_queue_mapping), 9540 si->dst_reg, si->src_reg, 9541 bpf_target_off(struct sock, sk_rx_queue_mapping, 9542 sizeof_field(struct sock, 9543 sk_rx_queue_mapping), 9544 target_size)); 9545 *insn++ = BPF_JMP_IMM(BPF_JNE, si->dst_reg, NO_QUEUE_MAPPING, 9546 1); 9547 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 9548 #else 9549 *insn++ = BPF_MOV64_IMM(si->dst_reg, -1); 9550 *target_size = 2; 9551 #endif 9552 break; 9553 } 9554 9555 return insn - insn_buf; 9556 } 9557 9558 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, 9559 const struct bpf_insn *si, 9560 struct bpf_insn *insn_buf, 9561 struct bpf_prog *prog, u32 *target_size) 9562 { 9563 struct bpf_insn *insn = insn_buf; 9564 9565 switch (si->off) { 9566 case offsetof(struct __sk_buff, ifindex): 9567 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 9568 si->dst_reg, si->src_reg, 9569 offsetof(struct sk_buff, dev)); 9570 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9571 bpf_target_off(struct net_device, ifindex, 4, 9572 target_size)); 9573 break; 9574 default: 9575 return bpf_convert_ctx_access(type, si, insn_buf, prog, 9576 target_size); 9577 } 9578 9579 return insn - insn_buf; 9580 } 9581 9582 static u32 xdp_convert_ctx_access(enum bpf_access_type type, 9583 const struct bpf_insn *si, 9584 struct bpf_insn *insn_buf, 9585 struct bpf_prog *prog, u32 *target_size) 9586 { 9587 struct bpf_insn *insn = insn_buf; 9588 9589 switch (si->off) { 9590 case offsetof(struct xdp_md, data): 9591 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data), 9592 si->dst_reg, si->src_reg, 9593 offsetof(struct xdp_buff, data)); 9594 break; 9595 case offsetof(struct xdp_md, data_meta): 9596 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta), 9597 si->dst_reg, si->src_reg, 9598 offsetof(struct xdp_buff, data_meta)); 9599 break; 9600 case offsetof(struct xdp_md, data_end): 9601 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end), 9602 si->dst_reg, si->src_reg, 9603 offsetof(struct xdp_buff, data_end)); 9604 break; 9605 case offsetof(struct xdp_md, ingress_ifindex): 9606 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 9607 si->dst_reg, si->src_reg, 9608 offsetof(struct xdp_buff, rxq)); 9609 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev), 9610 si->dst_reg, si->dst_reg, 9611 offsetof(struct xdp_rxq_info, dev)); 9612 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9613 offsetof(struct net_device, ifindex)); 9614 break; 9615 case offsetof(struct xdp_md, rx_queue_index): 9616 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 9617 si->dst_reg, si->src_reg, 9618 offsetof(struct xdp_buff, rxq)); 9619 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9620 offsetof(struct xdp_rxq_info, 9621 queue_index)); 9622 break; 9623 case offsetof(struct xdp_md, egress_ifindex): 9624 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, txq), 9625 si->dst_reg, si->src_reg, 9626 offsetof(struct xdp_buff, txq)); 9627 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_txq_info, dev), 9628 si->dst_reg, si->dst_reg, 9629 offsetof(struct xdp_txq_info, dev)); 9630 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9631 offsetof(struct net_device, ifindex)); 9632 break; 9633 } 9634 9635 return insn - insn_buf; 9636 } 9637 9638 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of 9639 * context Structure, F is Field in context structure that contains a pointer 9640 * to Nested Structure of type NS that has the field NF. 9641 * 9642 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make 9643 * sure that SIZE is not greater than actual size of S.F.NF. 9644 * 9645 * If offset OFF is provided, the load happens from that offset relative to 9646 * offset of NF. 9647 */ 9648 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \ 9649 do { \ 9650 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \ 9651 si->src_reg, offsetof(S, F)); \ 9652 *insn++ = BPF_LDX_MEM( \ 9653 SIZE, si->dst_reg, si->dst_reg, \ 9654 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 9655 target_size) \ 9656 + OFF); \ 9657 } while (0) 9658 9659 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \ 9660 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \ 9661 BPF_FIELD_SIZEOF(NS, NF), 0) 9662 9663 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to 9664 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation. 9665 * 9666 * In addition it uses Temporary Field TF (member of struct S) as the 3rd 9667 * "register" since two registers available in convert_ctx_access are not 9668 * enough: we can't override neither SRC, since it contains value to store, nor 9669 * DST since it contains pointer to context that may be used by later 9670 * instructions. But we need a temporary place to save pointer to nested 9671 * structure whose field we want to store to. 9672 */ 9673 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \ 9674 do { \ 9675 int tmp_reg = BPF_REG_9; \ 9676 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 9677 --tmp_reg; \ 9678 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 9679 --tmp_reg; \ 9680 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \ 9681 offsetof(S, TF)); \ 9682 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \ 9683 si->dst_reg, offsetof(S, F)); \ 9684 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \ 9685 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 9686 target_size) \ 9687 + OFF); \ 9688 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \ 9689 offsetof(S, TF)); \ 9690 } while (0) 9691 9692 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \ 9693 TF) \ 9694 do { \ 9695 if (type == BPF_WRITE) { \ 9696 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \ 9697 OFF, TF); \ 9698 } else { \ 9699 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \ 9700 S, NS, F, NF, SIZE, OFF); \ 9701 } \ 9702 } while (0) 9703 9704 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \ 9705 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \ 9706 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF) 9707 9708 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type, 9709 const struct bpf_insn *si, 9710 struct bpf_insn *insn_buf, 9711 struct bpf_prog *prog, u32 *target_size) 9712 { 9713 int off, port_size = sizeof_field(struct sockaddr_in6, sin6_port); 9714 struct bpf_insn *insn = insn_buf; 9715 9716 switch (si->off) { 9717 case offsetof(struct bpf_sock_addr, user_family): 9718 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9719 struct sockaddr, uaddr, sa_family); 9720 break; 9721 9722 case offsetof(struct bpf_sock_addr, user_ip4): 9723 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9724 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr, 9725 sin_addr, BPF_SIZE(si->code), 0, tmp_reg); 9726 break; 9727 9728 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 9729 off = si->off; 9730 off -= offsetof(struct bpf_sock_addr, user_ip6[0]); 9731 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9732 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 9733 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off, 9734 tmp_reg); 9735 break; 9736 9737 case offsetof(struct bpf_sock_addr, user_port): 9738 /* To get port we need to know sa_family first and then treat 9739 * sockaddr as either sockaddr_in or sockaddr_in6. 9740 * Though we can simplify since port field has same offset and 9741 * size in both structures. 9742 * Here we check this invariant and use just one of the 9743 * structures if it's true. 9744 */ 9745 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) != 9746 offsetof(struct sockaddr_in6, sin6_port)); 9747 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) != 9748 sizeof_field(struct sockaddr_in6, sin6_port)); 9749 /* Account for sin6_port being smaller than user_port. */ 9750 port_size = min(port_size, BPF_LDST_BYTES(si)); 9751 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9752 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 9753 sin6_port, bytes_to_bpf_size(port_size), 0, tmp_reg); 9754 break; 9755 9756 case offsetof(struct bpf_sock_addr, family): 9757 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9758 struct sock, sk, sk_family); 9759 break; 9760 9761 case offsetof(struct bpf_sock_addr, type): 9762 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9763 struct sock, sk, sk_type); 9764 break; 9765 9766 case offsetof(struct bpf_sock_addr, protocol): 9767 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 9768 struct sock, sk, sk_protocol); 9769 break; 9770 9771 case offsetof(struct bpf_sock_addr, msg_src_ip4): 9772 /* Treat t_ctx as struct in_addr for msg_src_ip4. */ 9773 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9774 struct bpf_sock_addr_kern, struct in_addr, t_ctx, 9775 s_addr, BPF_SIZE(si->code), 0, tmp_reg); 9776 break; 9777 9778 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 9779 msg_src_ip6[3]): 9780 off = si->off; 9781 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]); 9782 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */ 9783 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 9784 struct bpf_sock_addr_kern, struct in6_addr, t_ctx, 9785 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg); 9786 break; 9787 case offsetof(struct bpf_sock_addr, sk): 9788 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk), 9789 si->dst_reg, si->src_reg, 9790 offsetof(struct bpf_sock_addr_kern, sk)); 9791 break; 9792 } 9793 9794 return insn - insn_buf; 9795 } 9796 9797 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type, 9798 const struct bpf_insn *si, 9799 struct bpf_insn *insn_buf, 9800 struct bpf_prog *prog, 9801 u32 *target_size) 9802 { 9803 struct bpf_insn *insn = insn_buf; 9804 int off; 9805 9806 /* Helper macro for adding read access to tcp_sock or sock fields. */ 9807 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 9808 do { \ 9809 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 2; \ 9810 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 9811 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 9812 if (si->dst_reg == reg || si->src_reg == reg) \ 9813 reg--; \ 9814 if (si->dst_reg == reg || si->src_reg == reg) \ 9815 reg--; \ 9816 if (si->dst_reg == si->src_reg) { \ 9817 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 9818 offsetof(struct bpf_sock_ops_kern, \ 9819 temp)); \ 9820 fullsock_reg = reg; \ 9821 jmp += 2; \ 9822 } \ 9823 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9824 struct bpf_sock_ops_kern, \ 9825 is_fullsock), \ 9826 fullsock_reg, si->src_reg, \ 9827 offsetof(struct bpf_sock_ops_kern, \ 9828 is_fullsock)); \ 9829 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 9830 if (si->dst_reg == si->src_reg) \ 9831 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9832 offsetof(struct bpf_sock_ops_kern, \ 9833 temp)); \ 9834 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9835 struct bpf_sock_ops_kern, sk),\ 9836 si->dst_reg, si->src_reg, \ 9837 offsetof(struct bpf_sock_ops_kern, sk));\ 9838 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \ 9839 OBJ_FIELD), \ 9840 si->dst_reg, si->dst_reg, \ 9841 offsetof(OBJ, OBJ_FIELD)); \ 9842 if (si->dst_reg == si->src_reg) { \ 9843 *insn++ = BPF_JMP_A(1); \ 9844 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9845 offsetof(struct bpf_sock_ops_kern, \ 9846 temp)); \ 9847 } \ 9848 } while (0) 9849 9850 #define SOCK_OPS_GET_SK() \ 9851 do { \ 9852 int fullsock_reg = si->dst_reg, reg = BPF_REG_9, jmp = 1; \ 9853 if (si->dst_reg == reg || si->src_reg == reg) \ 9854 reg--; \ 9855 if (si->dst_reg == reg || si->src_reg == reg) \ 9856 reg--; \ 9857 if (si->dst_reg == si->src_reg) { \ 9858 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, \ 9859 offsetof(struct bpf_sock_ops_kern, \ 9860 temp)); \ 9861 fullsock_reg = reg; \ 9862 jmp += 2; \ 9863 } \ 9864 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9865 struct bpf_sock_ops_kern, \ 9866 is_fullsock), \ 9867 fullsock_reg, si->src_reg, \ 9868 offsetof(struct bpf_sock_ops_kern, \ 9869 is_fullsock)); \ 9870 *insn++ = BPF_JMP_IMM(BPF_JEQ, fullsock_reg, 0, jmp); \ 9871 if (si->dst_reg == si->src_reg) \ 9872 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9873 offsetof(struct bpf_sock_ops_kern, \ 9874 temp)); \ 9875 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9876 struct bpf_sock_ops_kern, sk),\ 9877 si->dst_reg, si->src_reg, \ 9878 offsetof(struct bpf_sock_ops_kern, sk));\ 9879 if (si->dst_reg == si->src_reg) { \ 9880 *insn++ = BPF_JMP_A(1); \ 9881 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->src_reg, \ 9882 offsetof(struct bpf_sock_ops_kern, \ 9883 temp)); \ 9884 } \ 9885 } while (0) 9886 9887 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \ 9888 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock) 9889 9890 /* Helper macro for adding write access to tcp_sock or sock fields. 9891 * The macro is called with two registers, dst_reg which contains a pointer 9892 * to ctx (context) and src_reg which contains the value that should be 9893 * stored. However, we need an additional register since we cannot overwrite 9894 * dst_reg because it may be used later in the program. 9895 * Instead we "borrow" one of the other register. We first save its value 9896 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore 9897 * it at the end of the macro. 9898 */ 9899 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 9900 do { \ 9901 int reg = BPF_REG_9; \ 9902 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 9903 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 9904 if (si->dst_reg == reg || si->src_reg == reg) \ 9905 reg--; \ 9906 if (si->dst_reg == reg || si->src_reg == reg) \ 9907 reg--; \ 9908 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \ 9909 offsetof(struct bpf_sock_ops_kern, \ 9910 temp)); \ 9911 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9912 struct bpf_sock_ops_kern, \ 9913 is_fullsock), \ 9914 reg, si->dst_reg, \ 9915 offsetof(struct bpf_sock_ops_kern, \ 9916 is_fullsock)); \ 9917 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \ 9918 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 9919 struct bpf_sock_ops_kern, sk),\ 9920 reg, si->dst_reg, \ 9921 offsetof(struct bpf_sock_ops_kern, sk));\ 9922 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \ 9923 reg, si->src_reg, \ 9924 offsetof(OBJ, OBJ_FIELD)); \ 9925 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \ 9926 offsetof(struct bpf_sock_ops_kern, \ 9927 temp)); \ 9928 } while (0) 9929 9930 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \ 9931 do { \ 9932 if (TYPE == BPF_WRITE) \ 9933 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 9934 else \ 9935 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 9936 } while (0) 9937 9938 if (insn > insn_buf) 9939 return insn - insn_buf; 9940 9941 switch (si->off) { 9942 case offsetof(struct bpf_sock_ops, op): 9943 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 9944 op), 9945 si->dst_reg, si->src_reg, 9946 offsetof(struct bpf_sock_ops_kern, op)); 9947 break; 9948 9949 case offsetof(struct bpf_sock_ops, replylong[0]) ... 9950 offsetof(struct bpf_sock_ops, replylong[3]): 9951 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) != 9952 sizeof_field(struct bpf_sock_ops_kern, reply)); 9953 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) != 9954 sizeof_field(struct bpf_sock_ops_kern, replylong)); 9955 off = si->off; 9956 off -= offsetof(struct bpf_sock_ops, replylong[0]); 9957 off += offsetof(struct bpf_sock_ops_kern, replylong[0]); 9958 if (type == BPF_WRITE) 9959 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 9960 off); 9961 else 9962 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 9963 off); 9964 break; 9965 9966 case offsetof(struct bpf_sock_ops, family): 9967 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 9968 9969 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9970 struct bpf_sock_ops_kern, sk), 9971 si->dst_reg, si->src_reg, 9972 offsetof(struct bpf_sock_ops_kern, sk)); 9973 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 9974 offsetof(struct sock_common, skc_family)); 9975 break; 9976 9977 case offsetof(struct bpf_sock_ops, remote_ip4): 9978 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 9979 9980 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9981 struct bpf_sock_ops_kern, sk), 9982 si->dst_reg, si->src_reg, 9983 offsetof(struct bpf_sock_ops_kern, sk)); 9984 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9985 offsetof(struct sock_common, skc_daddr)); 9986 break; 9987 9988 case offsetof(struct bpf_sock_ops, local_ip4): 9989 BUILD_BUG_ON(sizeof_field(struct sock_common, 9990 skc_rcv_saddr) != 4); 9991 9992 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 9993 struct bpf_sock_ops_kern, sk), 9994 si->dst_reg, si->src_reg, 9995 offsetof(struct bpf_sock_ops_kern, sk)); 9996 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 9997 offsetof(struct sock_common, 9998 skc_rcv_saddr)); 9999 break; 10000 10001 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ... 10002 offsetof(struct bpf_sock_ops, remote_ip6[3]): 10003 #if IS_ENABLED(CONFIG_IPV6) 10004 BUILD_BUG_ON(sizeof_field(struct sock_common, 10005 skc_v6_daddr.s6_addr32[0]) != 4); 10006 10007 off = si->off; 10008 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]); 10009 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10010 struct bpf_sock_ops_kern, sk), 10011 si->dst_reg, si->src_reg, 10012 offsetof(struct bpf_sock_ops_kern, sk)); 10013 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10014 offsetof(struct sock_common, 10015 skc_v6_daddr.s6_addr32[0]) + 10016 off); 10017 #else 10018 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10019 #endif 10020 break; 10021 10022 case offsetof(struct bpf_sock_ops, local_ip6[0]) ... 10023 offsetof(struct bpf_sock_ops, local_ip6[3]): 10024 #if IS_ENABLED(CONFIG_IPV6) 10025 BUILD_BUG_ON(sizeof_field(struct sock_common, 10026 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 10027 10028 off = si->off; 10029 off -= offsetof(struct bpf_sock_ops, local_ip6[0]); 10030 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10031 struct bpf_sock_ops_kern, sk), 10032 si->dst_reg, si->src_reg, 10033 offsetof(struct bpf_sock_ops_kern, sk)); 10034 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10035 offsetof(struct sock_common, 10036 skc_v6_rcv_saddr.s6_addr32[0]) + 10037 off); 10038 #else 10039 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10040 #endif 10041 break; 10042 10043 case offsetof(struct bpf_sock_ops, remote_port): 10044 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 10045 10046 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10047 struct bpf_sock_ops_kern, sk), 10048 si->dst_reg, si->src_reg, 10049 offsetof(struct bpf_sock_ops_kern, sk)); 10050 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10051 offsetof(struct sock_common, skc_dport)); 10052 #ifndef __BIG_ENDIAN_BITFIELD 10053 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 10054 #endif 10055 break; 10056 10057 case offsetof(struct bpf_sock_ops, local_port): 10058 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 10059 10060 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10061 struct bpf_sock_ops_kern, sk), 10062 si->dst_reg, si->src_reg, 10063 offsetof(struct bpf_sock_ops_kern, sk)); 10064 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10065 offsetof(struct sock_common, skc_num)); 10066 break; 10067 10068 case offsetof(struct bpf_sock_ops, is_fullsock): 10069 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10070 struct bpf_sock_ops_kern, 10071 is_fullsock), 10072 si->dst_reg, si->src_reg, 10073 offsetof(struct bpf_sock_ops_kern, 10074 is_fullsock)); 10075 break; 10076 10077 case offsetof(struct bpf_sock_ops, state): 10078 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1); 10079 10080 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10081 struct bpf_sock_ops_kern, sk), 10082 si->dst_reg, si->src_reg, 10083 offsetof(struct bpf_sock_ops_kern, sk)); 10084 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg, 10085 offsetof(struct sock_common, skc_state)); 10086 break; 10087 10088 case offsetof(struct bpf_sock_ops, rtt_min): 10089 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 10090 sizeof(struct minmax)); 10091 BUILD_BUG_ON(sizeof(struct minmax) < 10092 sizeof(struct minmax_sample)); 10093 10094 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10095 struct bpf_sock_ops_kern, sk), 10096 si->dst_reg, si->src_reg, 10097 offsetof(struct bpf_sock_ops_kern, sk)); 10098 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10099 offsetof(struct tcp_sock, rtt_min) + 10100 sizeof_field(struct minmax_sample, t)); 10101 break; 10102 10103 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags): 10104 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags, 10105 struct tcp_sock); 10106 break; 10107 10108 case offsetof(struct bpf_sock_ops, sk_txhash): 10109 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash, 10110 struct sock, type); 10111 break; 10112 case offsetof(struct bpf_sock_ops, snd_cwnd): 10113 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd); 10114 break; 10115 case offsetof(struct bpf_sock_ops, srtt_us): 10116 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us); 10117 break; 10118 case offsetof(struct bpf_sock_ops, snd_ssthresh): 10119 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh); 10120 break; 10121 case offsetof(struct bpf_sock_ops, rcv_nxt): 10122 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt); 10123 break; 10124 case offsetof(struct bpf_sock_ops, snd_nxt): 10125 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt); 10126 break; 10127 case offsetof(struct bpf_sock_ops, snd_una): 10128 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una); 10129 break; 10130 case offsetof(struct bpf_sock_ops, mss_cache): 10131 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache); 10132 break; 10133 case offsetof(struct bpf_sock_ops, ecn_flags): 10134 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags); 10135 break; 10136 case offsetof(struct bpf_sock_ops, rate_delivered): 10137 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered); 10138 break; 10139 case offsetof(struct bpf_sock_ops, rate_interval_us): 10140 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us); 10141 break; 10142 case offsetof(struct bpf_sock_ops, packets_out): 10143 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out); 10144 break; 10145 case offsetof(struct bpf_sock_ops, retrans_out): 10146 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out); 10147 break; 10148 case offsetof(struct bpf_sock_ops, total_retrans): 10149 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans); 10150 break; 10151 case offsetof(struct bpf_sock_ops, segs_in): 10152 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in); 10153 break; 10154 case offsetof(struct bpf_sock_ops, data_segs_in): 10155 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in); 10156 break; 10157 case offsetof(struct bpf_sock_ops, segs_out): 10158 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out); 10159 break; 10160 case offsetof(struct bpf_sock_ops, data_segs_out): 10161 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out); 10162 break; 10163 case offsetof(struct bpf_sock_ops, lost_out): 10164 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out); 10165 break; 10166 case offsetof(struct bpf_sock_ops, sacked_out): 10167 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out); 10168 break; 10169 case offsetof(struct bpf_sock_ops, bytes_received): 10170 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received); 10171 break; 10172 case offsetof(struct bpf_sock_ops, bytes_acked): 10173 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked); 10174 break; 10175 case offsetof(struct bpf_sock_ops, sk): 10176 SOCK_OPS_GET_SK(); 10177 break; 10178 case offsetof(struct bpf_sock_ops, skb_data_end): 10179 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10180 skb_data_end), 10181 si->dst_reg, si->src_reg, 10182 offsetof(struct bpf_sock_ops_kern, 10183 skb_data_end)); 10184 break; 10185 case offsetof(struct bpf_sock_ops, skb_data): 10186 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10187 skb), 10188 si->dst_reg, si->src_reg, 10189 offsetof(struct bpf_sock_ops_kern, 10190 skb)); 10191 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10192 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 10193 si->dst_reg, si->dst_reg, 10194 offsetof(struct sk_buff, data)); 10195 break; 10196 case offsetof(struct bpf_sock_ops, skb_len): 10197 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10198 skb), 10199 si->dst_reg, si->src_reg, 10200 offsetof(struct bpf_sock_ops_kern, 10201 skb)); 10202 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10203 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 10204 si->dst_reg, si->dst_reg, 10205 offsetof(struct sk_buff, len)); 10206 break; 10207 case offsetof(struct bpf_sock_ops, skb_tcp_flags): 10208 off = offsetof(struct sk_buff, cb); 10209 off += offsetof(struct tcp_skb_cb, tcp_flags); 10210 *target_size = sizeof_field(struct tcp_skb_cb, tcp_flags); 10211 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_ops_kern, 10212 skb), 10213 si->dst_reg, si->src_reg, 10214 offsetof(struct bpf_sock_ops_kern, 10215 skb)); 10216 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 10217 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_skb_cb, 10218 tcp_flags), 10219 si->dst_reg, si->dst_reg, off); 10220 break; 10221 } 10222 return insn - insn_buf; 10223 } 10224 10225 /* data_end = skb->data + skb_headlen() */ 10226 static struct bpf_insn *bpf_convert_data_end_access(const struct bpf_insn *si, 10227 struct bpf_insn *insn) 10228 { 10229 int reg; 10230 int temp_reg_off = offsetof(struct sk_buff, cb) + 10231 offsetof(struct sk_skb_cb, temp_reg); 10232 10233 if (si->src_reg == si->dst_reg) { 10234 /* We need an extra register, choose and save a register. */ 10235 reg = BPF_REG_9; 10236 if (si->src_reg == reg || si->dst_reg == reg) 10237 reg--; 10238 if (si->src_reg == reg || si->dst_reg == reg) 10239 reg--; 10240 *insn++ = BPF_STX_MEM(BPF_DW, si->src_reg, reg, temp_reg_off); 10241 } else { 10242 reg = si->dst_reg; 10243 } 10244 10245 /* reg = skb->data */ 10246 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 10247 reg, si->src_reg, 10248 offsetof(struct sk_buff, data)); 10249 /* AX = skb->len */ 10250 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, len), 10251 BPF_REG_AX, si->src_reg, 10252 offsetof(struct sk_buff, len)); 10253 /* reg = skb->data + skb->len */ 10254 *insn++ = BPF_ALU64_REG(BPF_ADD, reg, BPF_REG_AX); 10255 /* AX = skb->data_len */ 10256 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data_len), 10257 BPF_REG_AX, si->src_reg, 10258 offsetof(struct sk_buff, data_len)); 10259 10260 /* reg = skb->data + skb->len - skb->data_len */ 10261 *insn++ = BPF_ALU64_REG(BPF_SUB, reg, BPF_REG_AX); 10262 10263 if (si->src_reg == si->dst_reg) { 10264 /* Restore the saved register */ 10265 *insn++ = BPF_MOV64_REG(BPF_REG_AX, si->src_reg); 10266 *insn++ = BPF_MOV64_REG(si->dst_reg, reg); 10267 *insn++ = BPF_LDX_MEM(BPF_DW, reg, BPF_REG_AX, temp_reg_off); 10268 } 10269 10270 return insn; 10271 } 10272 10273 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type, 10274 const struct bpf_insn *si, 10275 struct bpf_insn *insn_buf, 10276 struct bpf_prog *prog, u32 *target_size) 10277 { 10278 struct bpf_insn *insn = insn_buf; 10279 int off; 10280 10281 switch (si->off) { 10282 case offsetof(struct __sk_buff, data_end): 10283 insn = bpf_convert_data_end_access(si, insn); 10284 break; 10285 case offsetof(struct __sk_buff, cb[0]) ... 10286 offsetofend(struct __sk_buff, cb[4]) - 1: 10287 BUILD_BUG_ON(sizeof_field(struct sk_skb_cb, data) < 20); 10288 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 10289 offsetof(struct sk_skb_cb, data)) % 10290 sizeof(__u64)); 10291 10292 prog->cb_access = 1; 10293 off = si->off; 10294 off -= offsetof(struct __sk_buff, cb[0]); 10295 off += offsetof(struct sk_buff, cb); 10296 off += offsetof(struct sk_skb_cb, data); 10297 if (type == BPF_WRITE) 10298 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 10299 si->src_reg, off); 10300 else 10301 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 10302 si->src_reg, off); 10303 break; 10304 10305 10306 default: 10307 return bpf_convert_ctx_access(type, si, insn_buf, prog, 10308 target_size); 10309 } 10310 10311 return insn - insn_buf; 10312 } 10313 10314 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type, 10315 const struct bpf_insn *si, 10316 struct bpf_insn *insn_buf, 10317 struct bpf_prog *prog, u32 *target_size) 10318 { 10319 struct bpf_insn *insn = insn_buf; 10320 #if IS_ENABLED(CONFIG_IPV6) 10321 int off; 10322 #endif 10323 10324 /* convert ctx uses the fact sg element is first in struct */ 10325 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0); 10326 10327 switch (si->off) { 10328 case offsetof(struct sk_msg_md, data): 10329 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data), 10330 si->dst_reg, si->src_reg, 10331 offsetof(struct sk_msg, data)); 10332 break; 10333 case offsetof(struct sk_msg_md, data_end): 10334 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end), 10335 si->dst_reg, si->src_reg, 10336 offsetof(struct sk_msg, data_end)); 10337 break; 10338 case offsetof(struct sk_msg_md, family): 10339 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 10340 10341 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10342 struct sk_msg, sk), 10343 si->dst_reg, si->src_reg, 10344 offsetof(struct sk_msg, sk)); 10345 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10346 offsetof(struct sock_common, skc_family)); 10347 break; 10348 10349 case offsetof(struct sk_msg_md, remote_ip4): 10350 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 10351 10352 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10353 struct sk_msg, sk), 10354 si->dst_reg, si->src_reg, 10355 offsetof(struct sk_msg, sk)); 10356 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10357 offsetof(struct sock_common, skc_daddr)); 10358 break; 10359 10360 case offsetof(struct sk_msg_md, local_ip4): 10361 BUILD_BUG_ON(sizeof_field(struct sock_common, 10362 skc_rcv_saddr) != 4); 10363 10364 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10365 struct sk_msg, sk), 10366 si->dst_reg, si->src_reg, 10367 offsetof(struct sk_msg, sk)); 10368 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10369 offsetof(struct sock_common, 10370 skc_rcv_saddr)); 10371 break; 10372 10373 case offsetof(struct sk_msg_md, remote_ip6[0]) ... 10374 offsetof(struct sk_msg_md, remote_ip6[3]): 10375 #if IS_ENABLED(CONFIG_IPV6) 10376 BUILD_BUG_ON(sizeof_field(struct sock_common, 10377 skc_v6_daddr.s6_addr32[0]) != 4); 10378 10379 off = si->off; 10380 off -= offsetof(struct sk_msg_md, remote_ip6[0]); 10381 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10382 struct sk_msg, sk), 10383 si->dst_reg, si->src_reg, 10384 offsetof(struct sk_msg, sk)); 10385 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10386 offsetof(struct sock_common, 10387 skc_v6_daddr.s6_addr32[0]) + 10388 off); 10389 #else 10390 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10391 #endif 10392 break; 10393 10394 case offsetof(struct sk_msg_md, local_ip6[0]) ... 10395 offsetof(struct sk_msg_md, local_ip6[3]): 10396 #if IS_ENABLED(CONFIG_IPV6) 10397 BUILD_BUG_ON(sizeof_field(struct sock_common, 10398 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 10399 10400 off = si->off; 10401 off -= offsetof(struct sk_msg_md, local_ip6[0]); 10402 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10403 struct sk_msg, sk), 10404 si->dst_reg, si->src_reg, 10405 offsetof(struct sk_msg, sk)); 10406 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 10407 offsetof(struct sock_common, 10408 skc_v6_rcv_saddr.s6_addr32[0]) + 10409 off); 10410 #else 10411 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 10412 #endif 10413 break; 10414 10415 case offsetof(struct sk_msg_md, remote_port): 10416 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 10417 10418 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10419 struct sk_msg, sk), 10420 si->dst_reg, si->src_reg, 10421 offsetof(struct sk_msg, sk)); 10422 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10423 offsetof(struct sock_common, skc_dport)); 10424 #ifndef __BIG_ENDIAN_BITFIELD 10425 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 10426 #endif 10427 break; 10428 10429 case offsetof(struct sk_msg_md, local_port): 10430 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 10431 10432 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 10433 struct sk_msg, sk), 10434 si->dst_reg, si->src_reg, 10435 offsetof(struct sk_msg, sk)); 10436 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 10437 offsetof(struct sock_common, skc_num)); 10438 break; 10439 10440 case offsetof(struct sk_msg_md, size): 10441 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size), 10442 si->dst_reg, si->src_reg, 10443 offsetof(struct sk_msg_sg, size)); 10444 break; 10445 10446 case offsetof(struct sk_msg_md, sk): 10447 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, sk), 10448 si->dst_reg, si->src_reg, 10449 offsetof(struct sk_msg, sk)); 10450 break; 10451 } 10452 10453 return insn - insn_buf; 10454 } 10455 10456 const struct bpf_verifier_ops sk_filter_verifier_ops = { 10457 .get_func_proto = sk_filter_func_proto, 10458 .is_valid_access = sk_filter_is_valid_access, 10459 .convert_ctx_access = bpf_convert_ctx_access, 10460 .gen_ld_abs = bpf_gen_ld_abs, 10461 }; 10462 10463 const struct bpf_prog_ops sk_filter_prog_ops = { 10464 .test_run = bpf_prog_test_run_skb, 10465 }; 10466 10467 const struct bpf_verifier_ops tc_cls_act_verifier_ops = { 10468 .get_func_proto = tc_cls_act_func_proto, 10469 .is_valid_access = tc_cls_act_is_valid_access, 10470 .convert_ctx_access = tc_cls_act_convert_ctx_access, 10471 .gen_prologue = tc_cls_act_prologue, 10472 .gen_ld_abs = bpf_gen_ld_abs, 10473 }; 10474 10475 const struct bpf_prog_ops tc_cls_act_prog_ops = { 10476 .test_run = bpf_prog_test_run_skb, 10477 }; 10478 10479 const struct bpf_verifier_ops xdp_verifier_ops = { 10480 .get_func_proto = xdp_func_proto, 10481 .is_valid_access = xdp_is_valid_access, 10482 .convert_ctx_access = xdp_convert_ctx_access, 10483 .gen_prologue = bpf_noop_prologue, 10484 }; 10485 10486 const struct bpf_prog_ops xdp_prog_ops = { 10487 .test_run = bpf_prog_test_run_xdp, 10488 }; 10489 10490 const struct bpf_verifier_ops cg_skb_verifier_ops = { 10491 .get_func_proto = cg_skb_func_proto, 10492 .is_valid_access = cg_skb_is_valid_access, 10493 .convert_ctx_access = bpf_convert_ctx_access, 10494 }; 10495 10496 const struct bpf_prog_ops cg_skb_prog_ops = { 10497 .test_run = bpf_prog_test_run_skb, 10498 }; 10499 10500 const struct bpf_verifier_ops lwt_in_verifier_ops = { 10501 .get_func_proto = lwt_in_func_proto, 10502 .is_valid_access = lwt_is_valid_access, 10503 .convert_ctx_access = bpf_convert_ctx_access, 10504 }; 10505 10506 const struct bpf_prog_ops lwt_in_prog_ops = { 10507 .test_run = bpf_prog_test_run_skb, 10508 }; 10509 10510 const struct bpf_verifier_ops lwt_out_verifier_ops = { 10511 .get_func_proto = lwt_out_func_proto, 10512 .is_valid_access = lwt_is_valid_access, 10513 .convert_ctx_access = bpf_convert_ctx_access, 10514 }; 10515 10516 const struct bpf_prog_ops lwt_out_prog_ops = { 10517 .test_run = bpf_prog_test_run_skb, 10518 }; 10519 10520 const struct bpf_verifier_ops lwt_xmit_verifier_ops = { 10521 .get_func_proto = lwt_xmit_func_proto, 10522 .is_valid_access = lwt_is_valid_access, 10523 .convert_ctx_access = bpf_convert_ctx_access, 10524 .gen_prologue = tc_cls_act_prologue, 10525 }; 10526 10527 const struct bpf_prog_ops lwt_xmit_prog_ops = { 10528 .test_run = bpf_prog_test_run_skb, 10529 }; 10530 10531 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = { 10532 .get_func_proto = lwt_seg6local_func_proto, 10533 .is_valid_access = lwt_is_valid_access, 10534 .convert_ctx_access = bpf_convert_ctx_access, 10535 }; 10536 10537 const struct bpf_prog_ops lwt_seg6local_prog_ops = { 10538 .test_run = bpf_prog_test_run_skb, 10539 }; 10540 10541 const struct bpf_verifier_ops cg_sock_verifier_ops = { 10542 .get_func_proto = sock_filter_func_proto, 10543 .is_valid_access = sock_filter_is_valid_access, 10544 .convert_ctx_access = bpf_sock_convert_ctx_access, 10545 }; 10546 10547 const struct bpf_prog_ops cg_sock_prog_ops = { 10548 }; 10549 10550 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = { 10551 .get_func_proto = sock_addr_func_proto, 10552 .is_valid_access = sock_addr_is_valid_access, 10553 .convert_ctx_access = sock_addr_convert_ctx_access, 10554 }; 10555 10556 const struct bpf_prog_ops cg_sock_addr_prog_ops = { 10557 }; 10558 10559 const struct bpf_verifier_ops sock_ops_verifier_ops = { 10560 .get_func_proto = sock_ops_func_proto, 10561 .is_valid_access = sock_ops_is_valid_access, 10562 .convert_ctx_access = sock_ops_convert_ctx_access, 10563 }; 10564 10565 const struct bpf_prog_ops sock_ops_prog_ops = { 10566 }; 10567 10568 const struct bpf_verifier_ops sk_skb_verifier_ops = { 10569 .get_func_proto = sk_skb_func_proto, 10570 .is_valid_access = sk_skb_is_valid_access, 10571 .convert_ctx_access = sk_skb_convert_ctx_access, 10572 .gen_prologue = sk_skb_prologue, 10573 }; 10574 10575 const struct bpf_prog_ops sk_skb_prog_ops = { 10576 }; 10577 10578 const struct bpf_verifier_ops sk_msg_verifier_ops = { 10579 .get_func_proto = sk_msg_func_proto, 10580 .is_valid_access = sk_msg_is_valid_access, 10581 .convert_ctx_access = sk_msg_convert_ctx_access, 10582 .gen_prologue = bpf_noop_prologue, 10583 }; 10584 10585 const struct bpf_prog_ops sk_msg_prog_ops = { 10586 }; 10587 10588 const struct bpf_verifier_ops flow_dissector_verifier_ops = { 10589 .get_func_proto = flow_dissector_func_proto, 10590 .is_valid_access = flow_dissector_is_valid_access, 10591 .convert_ctx_access = flow_dissector_convert_ctx_access, 10592 }; 10593 10594 const struct bpf_prog_ops flow_dissector_prog_ops = { 10595 .test_run = bpf_prog_test_run_flow_dissector, 10596 }; 10597 10598 int sk_detach_filter(struct sock *sk) 10599 { 10600 int ret = -ENOENT; 10601 struct sk_filter *filter; 10602 10603 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 10604 return -EPERM; 10605 10606 filter = rcu_dereference_protected(sk->sk_filter, 10607 lockdep_sock_is_held(sk)); 10608 if (filter) { 10609 RCU_INIT_POINTER(sk->sk_filter, NULL); 10610 sk_filter_uncharge(sk, filter); 10611 ret = 0; 10612 } 10613 10614 return ret; 10615 } 10616 EXPORT_SYMBOL_GPL(sk_detach_filter); 10617 10618 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, 10619 unsigned int len) 10620 { 10621 struct sock_fprog_kern *fprog; 10622 struct sk_filter *filter; 10623 int ret = 0; 10624 10625 lock_sock(sk); 10626 filter = rcu_dereference_protected(sk->sk_filter, 10627 lockdep_sock_is_held(sk)); 10628 if (!filter) 10629 goto out; 10630 10631 /* We're copying the filter that has been originally attached, 10632 * so no conversion/decode needed anymore. eBPF programs that 10633 * have no original program cannot be dumped through this. 10634 */ 10635 ret = -EACCES; 10636 fprog = filter->prog->orig_prog; 10637 if (!fprog) 10638 goto out; 10639 10640 ret = fprog->len; 10641 if (!len) 10642 /* User space only enquires number of filter blocks. */ 10643 goto out; 10644 10645 ret = -EINVAL; 10646 if (len < fprog->len) 10647 goto out; 10648 10649 ret = -EFAULT; 10650 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog))) 10651 goto out; 10652 10653 /* Instead of bytes, the API requests to return the number 10654 * of filter blocks. 10655 */ 10656 ret = fprog->len; 10657 out: 10658 release_sock(sk); 10659 return ret; 10660 } 10661 10662 #ifdef CONFIG_INET 10663 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern, 10664 struct sock_reuseport *reuse, 10665 struct sock *sk, struct sk_buff *skb, 10666 struct sock *migrating_sk, 10667 u32 hash) 10668 { 10669 reuse_kern->skb = skb; 10670 reuse_kern->sk = sk; 10671 reuse_kern->selected_sk = NULL; 10672 reuse_kern->migrating_sk = migrating_sk; 10673 reuse_kern->data_end = skb->data + skb_headlen(skb); 10674 reuse_kern->hash = hash; 10675 reuse_kern->reuseport_id = reuse->reuseport_id; 10676 reuse_kern->bind_inany = reuse->bind_inany; 10677 } 10678 10679 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 10680 struct bpf_prog *prog, struct sk_buff *skb, 10681 struct sock *migrating_sk, 10682 u32 hash) 10683 { 10684 struct sk_reuseport_kern reuse_kern; 10685 enum sk_action action; 10686 10687 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, migrating_sk, hash); 10688 action = bpf_prog_run(prog, &reuse_kern); 10689 10690 if (action == SK_PASS) 10691 return reuse_kern.selected_sk; 10692 else 10693 return ERR_PTR(-ECONNREFUSED); 10694 } 10695 10696 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern, 10697 struct bpf_map *, map, void *, key, u32, flags) 10698 { 10699 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY; 10700 struct sock_reuseport *reuse; 10701 struct sock *selected_sk; 10702 10703 selected_sk = map->ops->map_lookup_elem(map, key); 10704 if (!selected_sk) 10705 return -ENOENT; 10706 10707 reuse = rcu_dereference(selected_sk->sk_reuseport_cb); 10708 if (!reuse) { 10709 /* Lookup in sock_map can return TCP ESTABLISHED sockets. */ 10710 if (sk_is_refcounted(selected_sk)) 10711 sock_put(selected_sk); 10712 10713 /* reuseport_array has only sk with non NULL sk_reuseport_cb. 10714 * The only (!reuse) case here is - the sk has already been 10715 * unhashed (e.g. by close()), so treat it as -ENOENT. 10716 * 10717 * Other maps (e.g. sock_map) do not provide this guarantee and 10718 * the sk may never be in the reuseport group to begin with. 10719 */ 10720 return is_sockarray ? -ENOENT : -EINVAL; 10721 } 10722 10723 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) { 10724 struct sock *sk = reuse_kern->sk; 10725 10726 if (sk->sk_protocol != selected_sk->sk_protocol) 10727 return -EPROTOTYPE; 10728 else if (sk->sk_family != selected_sk->sk_family) 10729 return -EAFNOSUPPORT; 10730 10731 /* Catch all. Likely bound to a different sockaddr. */ 10732 return -EBADFD; 10733 } 10734 10735 reuse_kern->selected_sk = selected_sk; 10736 10737 return 0; 10738 } 10739 10740 static const struct bpf_func_proto sk_select_reuseport_proto = { 10741 .func = sk_select_reuseport, 10742 .gpl_only = false, 10743 .ret_type = RET_INTEGER, 10744 .arg1_type = ARG_PTR_TO_CTX, 10745 .arg2_type = ARG_CONST_MAP_PTR, 10746 .arg3_type = ARG_PTR_TO_MAP_KEY, 10747 .arg4_type = ARG_ANYTHING, 10748 }; 10749 10750 BPF_CALL_4(sk_reuseport_load_bytes, 10751 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 10752 void *, to, u32, len) 10753 { 10754 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len); 10755 } 10756 10757 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = { 10758 .func = sk_reuseport_load_bytes, 10759 .gpl_only = false, 10760 .ret_type = RET_INTEGER, 10761 .arg1_type = ARG_PTR_TO_CTX, 10762 .arg2_type = ARG_ANYTHING, 10763 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 10764 .arg4_type = ARG_CONST_SIZE, 10765 }; 10766 10767 BPF_CALL_5(sk_reuseport_load_bytes_relative, 10768 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 10769 void *, to, u32, len, u32, start_header) 10770 { 10771 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to, 10772 len, start_header); 10773 } 10774 10775 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = { 10776 .func = sk_reuseport_load_bytes_relative, 10777 .gpl_only = false, 10778 .ret_type = RET_INTEGER, 10779 .arg1_type = ARG_PTR_TO_CTX, 10780 .arg2_type = ARG_ANYTHING, 10781 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 10782 .arg4_type = ARG_CONST_SIZE, 10783 .arg5_type = ARG_ANYTHING, 10784 }; 10785 10786 static const struct bpf_func_proto * 10787 sk_reuseport_func_proto(enum bpf_func_id func_id, 10788 const struct bpf_prog *prog) 10789 { 10790 switch (func_id) { 10791 case BPF_FUNC_sk_select_reuseport: 10792 return &sk_select_reuseport_proto; 10793 case BPF_FUNC_skb_load_bytes: 10794 return &sk_reuseport_load_bytes_proto; 10795 case BPF_FUNC_skb_load_bytes_relative: 10796 return &sk_reuseport_load_bytes_relative_proto; 10797 case BPF_FUNC_get_socket_cookie: 10798 return &bpf_get_socket_ptr_cookie_proto; 10799 case BPF_FUNC_ktime_get_coarse_ns: 10800 return &bpf_ktime_get_coarse_ns_proto; 10801 default: 10802 return bpf_base_func_proto(func_id); 10803 } 10804 } 10805 10806 static bool 10807 sk_reuseport_is_valid_access(int off, int size, 10808 enum bpf_access_type type, 10809 const struct bpf_prog *prog, 10810 struct bpf_insn_access_aux *info) 10811 { 10812 const u32 size_default = sizeof(__u32); 10813 10814 if (off < 0 || off >= sizeof(struct sk_reuseport_md) || 10815 off % size || type != BPF_READ) 10816 return false; 10817 10818 switch (off) { 10819 case offsetof(struct sk_reuseport_md, data): 10820 info->reg_type = PTR_TO_PACKET; 10821 return size == sizeof(__u64); 10822 10823 case offsetof(struct sk_reuseport_md, data_end): 10824 info->reg_type = PTR_TO_PACKET_END; 10825 return size == sizeof(__u64); 10826 10827 case offsetof(struct sk_reuseport_md, hash): 10828 return size == size_default; 10829 10830 case offsetof(struct sk_reuseport_md, sk): 10831 info->reg_type = PTR_TO_SOCKET; 10832 return size == sizeof(__u64); 10833 10834 case offsetof(struct sk_reuseport_md, migrating_sk): 10835 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 10836 return size == sizeof(__u64); 10837 10838 /* Fields that allow narrowing */ 10839 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol): 10840 if (size < sizeof_field(struct sk_buff, protocol)) 10841 return false; 10842 fallthrough; 10843 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol): 10844 case bpf_ctx_range(struct sk_reuseport_md, bind_inany): 10845 case bpf_ctx_range(struct sk_reuseport_md, len): 10846 bpf_ctx_record_field_size(info, size_default); 10847 return bpf_ctx_narrow_access_ok(off, size, size_default); 10848 10849 default: 10850 return false; 10851 } 10852 } 10853 10854 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \ 10855 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \ 10856 si->dst_reg, si->src_reg, \ 10857 bpf_target_off(struct sk_reuseport_kern, F, \ 10858 sizeof_field(struct sk_reuseport_kern, F), \ 10859 target_size)); \ 10860 }) 10861 10862 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \ 10863 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 10864 struct sk_buff, \ 10865 skb, \ 10866 SKB_FIELD) 10867 10868 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \ 10869 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 10870 struct sock, \ 10871 sk, \ 10872 SK_FIELD) 10873 10874 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type, 10875 const struct bpf_insn *si, 10876 struct bpf_insn *insn_buf, 10877 struct bpf_prog *prog, 10878 u32 *target_size) 10879 { 10880 struct bpf_insn *insn = insn_buf; 10881 10882 switch (si->off) { 10883 case offsetof(struct sk_reuseport_md, data): 10884 SK_REUSEPORT_LOAD_SKB_FIELD(data); 10885 break; 10886 10887 case offsetof(struct sk_reuseport_md, len): 10888 SK_REUSEPORT_LOAD_SKB_FIELD(len); 10889 break; 10890 10891 case offsetof(struct sk_reuseport_md, eth_protocol): 10892 SK_REUSEPORT_LOAD_SKB_FIELD(protocol); 10893 break; 10894 10895 case offsetof(struct sk_reuseport_md, ip_protocol): 10896 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol); 10897 break; 10898 10899 case offsetof(struct sk_reuseport_md, data_end): 10900 SK_REUSEPORT_LOAD_FIELD(data_end); 10901 break; 10902 10903 case offsetof(struct sk_reuseport_md, hash): 10904 SK_REUSEPORT_LOAD_FIELD(hash); 10905 break; 10906 10907 case offsetof(struct sk_reuseport_md, bind_inany): 10908 SK_REUSEPORT_LOAD_FIELD(bind_inany); 10909 break; 10910 10911 case offsetof(struct sk_reuseport_md, sk): 10912 SK_REUSEPORT_LOAD_FIELD(sk); 10913 break; 10914 10915 case offsetof(struct sk_reuseport_md, migrating_sk): 10916 SK_REUSEPORT_LOAD_FIELD(migrating_sk); 10917 break; 10918 } 10919 10920 return insn - insn_buf; 10921 } 10922 10923 const struct bpf_verifier_ops sk_reuseport_verifier_ops = { 10924 .get_func_proto = sk_reuseport_func_proto, 10925 .is_valid_access = sk_reuseport_is_valid_access, 10926 .convert_ctx_access = sk_reuseport_convert_ctx_access, 10927 }; 10928 10929 const struct bpf_prog_ops sk_reuseport_prog_ops = { 10930 }; 10931 10932 DEFINE_STATIC_KEY_FALSE(bpf_sk_lookup_enabled); 10933 EXPORT_SYMBOL(bpf_sk_lookup_enabled); 10934 10935 BPF_CALL_3(bpf_sk_lookup_assign, struct bpf_sk_lookup_kern *, ctx, 10936 struct sock *, sk, u64, flags) 10937 { 10938 if (unlikely(flags & ~(BPF_SK_LOOKUP_F_REPLACE | 10939 BPF_SK_LOOKUP_F_NO_REUSEPORT))) 10940 return -EINVAL; 10941 if (unlikely(sk && sk_is_refcounted(sk))) 10942 return -ESOCKTNOSUPPORT; /* reject non-RCU freed sockets */ 10943 if (unlikely(sk && sk_is_tcp(sk) && sk->sk_state != TCP_LISTEN)) 10944 return -ESOCKTNOSUPPORT; /* only accept TCP socket in LISTEN */ 10945 if (unlikely(sk && sk_is_udp(sk) && sk->sk_state != TCP_CLOSE)) 10946 return -ESOCKTNOSUPPORT; /* only accept UDP socket in CLOSE */ 10947 10948 /* Check if socket is suitable for packet L3/L4 protocol */ 10949 if (sk && sk->sk_protocol != ctx->protocol) 10950 return -EPROTOTYPE; 10951 if (sk && sk->sk_family != ctx->family && 10952 (sk->sk_family == AF_INET || ipv6_only_sock(sk))) 10953 return -EAFNOSUPPORT; 10954 10955 if (ctx->selected_sk && !(flags & BPF_SK_LOOKUP_F_REPLACE)) 10956 return -EEXIST; 10957 10958 /* Select socket as lookup result */ 10959 ctx->selected_sk = sk; 10960 ctx->no_reuseport = flags & BPF_SK_LOOKUP_F_NO_REUSEPORT; 10961 return 0; 10962 } 10963 10964 static const struct bpf_func_proto bpf_sk_lookup_assign_proto = { 10965 .func = bpf_sk_lookup_assign, 10966 .gpl_only = false, 10967 .ret_type = RET_INTEGER, 10968 .arg1_type = ARG_PTR_TO_CTX, 10969 .arg2_type = ARG_PTR_TO_SOCKET_OR_NULL, 10970 .arg3_type = ARG_ANYTHING, 10971 }; 10972 10973 static const struct bpf_func_proto * 10974 sk_lookup_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 10975 { 10976 switch (func_id) { 10977 case BPF_FUNC_perf_event_output: 10978 return &bpf_event_output_data_proto; 10979 case BPF_FUNC_sk_assign: 10980 return &bpf_sk_lookup_assign_proto; 10981 case BPF_FUNC_sk_release: 10982 return &bpf_sk_release_proto; 10983 default: 10984 return bpf_sk_base_func_proto(func_id); 10985 } 10986 } 10987 10988 static bool sk_lookup_is_valid_access(int off, int size, 10989 enum bpf_access_type type, 10990 const struct bpf_prog *prog, 10991 struct bpf_insn_access_aux *info) 10992 { 10993 if (off < 0 || off >= sizeof(struct bpf_sk_lookup)) 10994 return false; 10995 if (off % size != 0) 10996 return false; 10997 if (type != BPF_READ) 10998 return false; 10999 11000 switch (off) { 11001 case offsetof(struct bpf_sk_lookup, sk): 11002 info->reg_type = PTR_TO_SOCKET_OR_NULL; 11003 return size == sizeof(__u64); 11004 11005 case bpf_ctx_range(struct bpf_sk_lookup, family): 11006 case bpf_ctx_range(struct bpf_sk_lookup, protocol): 11007 case bpf_ctx_range(struct bpf_sk_lookup, remote_ip4): 11008 case bpf_ctx_range(struct bpf_sk_lookup, local_ip4): 11009 case bpf_ctx_range_till(struct bpf_sk_lookup, remote_ip6[0], remote_ip6[3]): 11010 case bpf_ctx_range_till(struct bpf_sk_lookup, local_ip6[0], local_ip6[3]): 11011 case bpf_ctx_range(struct bpf_sk_lookup, local_port): 11012 case bpf_ctx_range(struct bpf_sk_lookup, ingress_ifindex): 11013 bpf_ctx_record_field_size(info, sizeof(__u32)); 11014 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u32)); 11015 11016 case bpf_ctx_range(struct bpf_sk_lookup, remote_port): 11017 /* Allow 4-byte access to 2-byte field for backward compatibility */ 11018 if (size == sizeof(__u32)) 11019 return true; 11020 bpf_ctx_record_field_size(info, sizeof(__be16)); 11021 return bpf_ctx_narrow_access_ok(off, size, sizeof(__be16)); 11022 11023 case offsetofend(struct bpf_sk_lookup, remote_port) ... 11024 offsetof(struct bpf_sk_lookup, local_ip4) - 1: 11025 /* Allow access to zero padding for backward compatibility */ 11026 bpf_ctx_record_field_size(info, sizeof(__u16)); 11027 return bpf_ctx_narrow_access_ok(off, size, sizeof(__u16)); 11028 11029 default: 11030 return false; 11031 } 11032 } 11033 11034 static u32 sk_lookup_convert_ctx_access(enum bpf_access_type type, 11035 const struct bpf_insn *si, 11036 struct bpf_insn *insn_buf, 11037 struct bpf_prog *prog, 11038 u32 *target_size) 11039 { 11040 struct bpf_insn *insn = insn_buf; 11041 11042 switch (si->off) { 11043 case offsetof(struct bpf_sk_lookup, sk): 11044 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11045 offsetof(struct bpf_sk_lookup_kern, selected_sk)); 11046 break; 11047 11048 case offsetof(struct bpf_sk_lookup, family): 11049 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11050 bpf_target_off(struct bpf_sk_lookup_kern, 11051 family, 2, target_size)); 11052 break; 11053 11054 case offsetof(struct bpf_sk_lookup, protocol): 11055 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11056 bpf_target_off(struct bpf_sk_lookup_kern, 11057 protocol, 2, target_size)); 11058 break; 11059 11060 case offsetof(struct bpf_sk_lookup, remote_ip4): 11061 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11062 bpf_target_off(struct bpf_sk_lookup_kern, 11063 v4.saddr, 4, target_size)); 11064 break; 11065 11066 case offsetof(struct bpf_sk_lookup, local_ip4): 11067 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11068 bpf_target_off(struct bpf_sk_lookup_kern, 11069 v4.daddr, 4, target_size)); 11070 break; 11071 11072 case bpf_ctx_range_till(struct bpf_sk_lookup, 11073 remote_ip6[0], remote_ip6[3]): { 11074 #if IS_ENABLED(CONFIG_IPV6) 11075 int off = si->off; 11076 11077 off -= offsetof(struct bpf_sk_lookup, remote_ip6[0]); 11078 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 11079 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11080 offsetof(struct bpf_sk_lookup_kern, v6.saddr)); 11081 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11082 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 11083 #else 11084 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11085 #endif 11086 break; 11087 } 11088 case bpf_ctx_range_till(struct bpf_sk_lookup, 11089 local_ip6[0], local_ip6[3]): { 11090 #if IS_ENABLED(CONFIG_IPV6) 11091 int off = si->off; 11092 11093 off -= offsetof(struct bpf_sk_lookup, local_ip6[0]); 11094 off += bpf_target_off(struct in6_addr, s6_addr32[0], 4, target_size); 11095 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, si->src_reg, 11096 offsetof(struct bpf_sk_lookup_kern, v6.daddr)); 11097 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 11098 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, off); 11099 #else 11100 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11101 #endif 11102 break; 11103 } 11104 case offsetof(struct bpf_sk_lookup, remote_port): 11105 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11106 bpf_target_off(struct bpf_sk_lookup_kern, 11107 sport, 2, target_size)); 11108 break; 11109 11110 case offsetofend(struct bpf_sk_lookup, remote_port): 11111 *target_size = 2; 11112 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 11113 break; 11114 11115 case offsetof(struct bpf_sk_lookup, local_port): 11116 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 11117 bpf_target_off(struct bpf_sk_lookup_kern, 11118 dport, 2, target_size)); 11119 break; 11120 11121 case offsetof(struct bpf_sk_lookup, ingress_ifindex): 11122 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 11123 bpf_target_off(struct bpf_sk_lookup_kern, 11124 ingress_ifindex, 4, target_size)); 11125 break; 11126 } 11127 11128 return insn - insn_buf; 11129 } 11130 11131 const struct bpf_prog_ops sk_lookup_prog_ops = { 11132 .test_run = bpf_prog_test_run_sk_lookup, 11133 }; 11134 11135 const struct bpf_verifier_ops sk_lookup_verifier_ops = { 11136 .get_func_proto = sk_lookup_func_proto, 11137 .is_valid_access = sk_lookup_is_valid_access, 11138 .convert_ctx_access = sk_lookup_convert_ctx_access, 11139 }; 11140 11141 #endif /* CONFIG_INET */ 11142 11143 DEFINE_BPF_DISPATCHER(xdp) 11144 11145 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog) 11146 { 11147 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog); 11148 } 11149 11150 BTF_ID_LIST_GLOBAL(btf_sock_ids, MAX_BTF_SOCK_TYPE) 11151 #define BTF_SOCK_TYPE(name, type) BTF_ID(struct, type) 11152 BTF_SOCK_TYPE_xxx 11153 #undef BTF_SOCK_TYPE 11154 11155 BPF_CALL_1(bpf_skc_to_tcp6_sock, struct sock *, sk) 11156 { 11157 /* tcp6_sock type is not generated in dwarf and hence btf, 11158 * trigger an explicit type generation here. 11159 */ 11160 BTF_TYPE_EMIT(struct tcp6_sock); 11161 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP && 11162 sk->sk_family == AF_INET6) 11163 return (unsigned long)sk; 11164 11165 return (unsigned long)NULL; 11166 } 11167 11168 const struct bpf_func_proto bpf_skc_to_tcp6_sock_proto = { 11169 .func = bpf_skc_to_tcp6_sock, 11170 .gpl_only = false, 11171 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11172 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11173 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP6], 11174 }; 11175 11176 BPF_CALL_1(bpf_skc_to_tcp_sock, struct sock *, sk) 11177 { 11178 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 11179 return (unsigned long)sk; 11180 11181 return (unsigned long)NULL; 11182 } 11183 11184 const struct bpf_func_proto bpf_skc_to_tcp_sock_proto = { 11185 .func = bpf_skc_to_tcp_sock, 11186 .gpl_only = false, 11187 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11188 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11189 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP], 11190 }; 11191 11192 BPF_CALL_1(bpf_skc_to_tcp_timewait_sock, struct sock *, sk) 11193 { 11194 /* BTF types for tcp_timewait_sock and inet_timewait_sock are not 11195 * generated if CONFIG_INET=n. Trigger an explicit generation here. 11196 */ 11197 BTF_TYPE_EMIT(struct inet_timewait_sock); 11198 BTF_TYPE_EMIT(struct tcp_timewait_sock); 11199 11200 #ifdef CONFIG_INET 11201 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_TIME_WAIT) 11202 return (unsigned long)sk; 11203 #endif 11204 11205 #if IS_BUILTIN(CONFIG_IPV6) 11206 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_TIME_WAIT) 11207 return (unsigned long)sk; 11208 #endif 11209 11210 return (unsigned long)NULL; 11211 } 11212 11213 const struct bpf_func_proto bpf_skc_to_tcp_timewait_sock_proto = { 11214 .func = bpf_skc_to_tcp_timewait_sock, 11215 .gpl_only = false, 11216 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11217 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11218 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_TW], 11219 }; 11220 11221 BPF_CALL_1(bpf_skc_to_tcp_request_sock, struct sock *, sk) 11222 { 11223 #ifdef CONFIG_INET 11224 if (sk && sk->sk_prot == &tcp_prot && sk->sk_state == TCP_NEW_SYN_RECV) 11225 return (unsigned long)sk; 11226 #endif 11227 11228 #if IS_BUILTIN(CONFIG_IPV6) 11229 if (sk && sk->sk_prot == &tcpv6_prot && sk->sk_state == TCP_NEW_SYN_RECV) 11230 return (unsigned long)sk; 11231 #endif 11232 11233 return (unsigned long)NULL; 11234 } 11235 11236 const struct bpf_func_proto bpf_skc_to_tcp_request_sock_proto = { 11237 .func = bpf_skc_to_tcp_request_sock, 11238 .gpl_only = false, 11239 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11240 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11241 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_TCP_REQ], 11242 }; 11243 11244 BPF_CALL_1(bpf_skc_to_udp6_sock, struct sock *, sk) 11245 { 11246 /* udp6_sock type is not generated in dwarf and hence btf, 11247 * trigger an explicit type generation here. 11248 */ 11249 BTF_TYPE_EMIT(struct udp6_sock); 11250 if (sk && sk_fullsock(sk) && sk->sk_protocol == IPPROTO_UDP && 11251 sk->sk_type == SOCK_DGRAM && sk->sk_family == AF_INET6) 11252 return (unsigned long)sk; 11253 11254 return (unsigned long)NULL; 11255 } 11256 11257 const struct bpf_func_proto bpf_skc_to_udp6_sock_proto = { 11258 .func = bpf_skc_to_udp6_sock, 11259 .gpl_only = false, 11260 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11261 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11262 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UDP6], 11263 }; 11264 11265 BPF_CALL_1(bpf_skc_to_unix_sock, struct sock *, sk) 11266 { 11267 /* unix_sock type is not generated in dwarf and hence btf, 11268 * trigger an explicit type generation here. 11269 */ 11270 BTF_TYPE_EMIT(struct unix_sock); 11271 if (sk && sk_fullsock(sk) && sk->sk_family == AF_UNIX) 11272 return (unsigned long)sk; 11273 11274 return (unsigned long)NULL; 11275 } 11276 11277 const struct bpf_func_proto bpf_skc_to_unix_sock_proto = { 11278 .func = bpf_skc_to_unix_sock, 11279 .gpl_only = false, 11280 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11281 .arg1_type = ARG_PTR_TO_BTF_ID_SOCK_COMMON, 11282 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_UNIX], 11283 }; 11284 11285 BPF_CALL_1(bpf_skc_to_mptcp_sock, struct sock *, sk) 11286 { 11287 BTF_TYPE_EMIT(struct mptcp_sock); 11288 return (unsigned long)bpf_mptcp_sock_from_subflow(sk); 11289 } 11290 11291 const struct bpf_func_proto bpf_skc_to_mptcp_sock_proto = { 11292 .func = bpf_skc_to_mptcp_sock, 11293 .gpl_only = false, 11294 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11295 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 11296 .ret_btf_id = &btf_sock_ids[BTF_SOCK_TYPE_MPTCP], 11297 }; 11298 11299 BPF_CALL_1(bpf_sock_from_file, struct file *, file) 11300 { 11301 return (unsigned long)sock_from_file(file); 11302 } 11303 11304 BTF_ID_LIST(bpf_sock_from_file_btf_ids) 11305 BTF_ID(struct, socket) 11306 BTF_ID(struct, file) 11307 11308 const struct bpf_func_proto bpf_sock_from_file_proto = { 11309 .func = bpf_sock_from_file, 11310 .gpl_only = false, 11311 .ret_type = RET_PTR_TO_BTF_ID_OR_NULL, 11312 .ret_btf_id = &bpf_sock_from_file_btf_ids[0], 11313 .arg1_type = ARG_PTR_TO_BTF_ID, 11314 .arg1_btf_id = &bpf_sock_from_file_btf_ids[1], 11315 }; 11316 11317 static const struct bpf_func_proto * 11318 bpf_sk_base_func_proto(enum bpf_func_id func_id) 11319 { 11320 const struct bpf_func_proto *func; 11321 11322 switch (func_id) { 11323 case BPF_FUNC_skc_to_tcp6_sock: 11324 func = &bpf_skc_to_tcp6_sock_proto; 11325 break; 11326 case BPF_FUNC_skc_to_tcp_sock: 11327 func = &bpf_skc_to_tcp_sock_proto; 11328 break; 11329 case BPF_FUNC_skc_to_tcp_timewait_sock: 11330 func = &bpf_skc_to_tcp_timewait_sock_proto; 11331 break; 11332 case BPF_FUNC_skc_to_tcp_request_sock: 11333 func = &bpf_skc_to_tcp_request_sock_proto; 11334 break; 11335 case BPF_FUNC_skc_to_udp6_sock: 11336 func = &bpf_skc_to_udp6_sock_proto; 11337 break; 11338 case BPF_FUNC_skc_to_unix_sock: 11339 func = &bpf_skc_to_unix_sock_proto; 11340 break; 11341 case BPF_FUNC_skc_to_mptcp_sock: 11342 func = &bpf_skc_to_mptcp_sock_proto; 11343 break; 11344 case BPF_FUNC_ktime_get_coarse_ns: 11345 return &bpf_ktime_get_coarse_ns_proto; 11346 default: 11347 return bpf_base_func_proto(func_id); 11348 } 11349 11350 if (!perfmon_capable()) 11351 return NULL; 11352 11353 return func; 11354 } 11355