1 // SPDX-License-Identifier: GPL-2.0-or-later 2 /* 3 * Linux Socket Filter - Kernel level socket filtering 4 * 5 * Based on the design of the Berkeley Packet Filter. The new 6 * internal format has been designed by PLUMgrid: 7 * 8 * Copyright (c) 2011 - 2014 PLUMgrid, http://plumgrid.com 9 * 10 * Authors: 11 * 12 * Jay Schulist <jschlst@samba.org> 13 * Alexei Starovoitov <ast@plumgrid.com> 14 * Daniel Borkmann <dborkman@redhat.com> 15 * 16 * Andi Kleen - Fix a few bad bugs and races. 17 * Kris Katterjohn - Added many additional checks in bpf_check_classic() 18 */ 19 20 #include <linux/module.h> 21 #include <linux/types.h> 22 #include <linux/mm.h> 23 #include <linux/fcntl.h> 24 #include <linux/socket.h> 25 #include <linux/sock_diag.h> 26 #include <linux/in.h> 27 #include <linux/inet.h> 28 #include <linux/netdevice.h> 29 #include <linux/if_packet.h> 30 #include <linux/if_arp.h> 31 #include <linux/gfp.h> 32 #include <net/inet_common.h> 33 #include <net/ip.h> 34 #include <net/protocol.h> 35 #include <net/netlink.h> 36 #include <linux/skbuff.h> 37 #include <linux/skmsg.h> 38 #include <net/sock.h> 39 #include <net/flow_dissector.h> 40 #include <linux/errno.h> 41 #include <linux/timer.h> 42 #include <linux/uaccess.h> 43 #include <asm/unaligned.h> 44 #include <asm/cmpxchg.h> 45 #include <linux/filter.h> 46 #include <linux/ratelimit.h> 47 #include <linux/seccomp.h> 48 #include <linux/if_vlan.h> 49 #include <linux/bpf.h> 50 #include <net/sch_generic.h> 51 #include <net/cls_cgroup.h> 52 #include <net/dst_metadata.h> 53 #include <net/dst.h> 54 #include <net/sock_reuseport.h> 55 #include <net/busy_poll.h> 56 #include <net/tcp.h> 57 #include <net/xfrm.h> 58 #include <net/udp.h> 59 #include <linux/bpf_trace.h> 60 #include <net/xdp_sock.h> 61 #include <linux/inetdevice.h> 62 #include <net/inet_hashtables.h> 63 #include <net/inet6_hashtables.h> 64 #include <net/ip_fib.h> 65 #include <net/nexthop.h> 66 #include <net/flow.h> 67 #include <net/arp.h> 68 #include <net/ipv6.h> 69 #include <net/net_namespace.h> 70 #include <linux/seg6_local.h> 71 #include <net/seg6.h> 72 #include <net/seg6_local.h> 73 #include <net/lwtunnel.h> 74 #include <net/ipv6_stubs.h> 75 #include <net/bpf_sk_storage.h> 76 77 /** 78 * sk_filter_trim_cap - run a packet through a socket filter 79 * @sk: sock associated with &sk_buff 80 * @skb: buffer to filter 81 * @cap: limit on how short the eBPF program may trim the packet 82 * 83 * Run the eBPF program and then cut skb->data to correct size returned by 84 * the program. If pkt_len is 0 we toss packet. If skb->len is smaller 85 * than pkt_len we keep whole skb->data. This is the socket level 86 * wrapper to BPF_PROG_RUN. It returns 0 if the packet should 87 * be accepted or -EPERM if the packet should be tossed. 88 * 89 */ 90 int sk_filter_trim_cap(struct sock *sk, struct sk_buff *skb, unsigned int cap) 91 { 92 int err; 93 struct sk_filter *filter; 94 95 /* 96 * If the skb was allocated from pfmemalloc reserves, only 97 * allow SOCK_MEMALLOC sockets to use it as this socket is 98 * helping free memory 99 */ 100 if (skb_pfmemalloc(skb) && !sock_flag(sk, SOCK_MEMALLOC)) { 101 NET_INC_STATS(sock_net(sk), LINUX_MIB_PFMEMALLOCDROP); 102 return -ENOMEM; 103 } 104 err = BPF_CGROUP_RUN_PROG_INET_INGRESS(sk, skb); 105 if (err) 106 return err; 107 108 err = security_sock_rcv_skb(sk, skb); 109 if (err) 110 return err; 111 112 rcu_read_lock(); 113 filter = rcu_dereference(sk->sk_filter); 114 if (filter) { 115 struct sock *save_sk = skb->sk; 116 unsigned int pkt_len; 117 118 skb->sk = sk; 119 pkt_len = bpf_prog_run_save_cb(filter->prog, skb); 120 skb->sk = save_sk; 121 err = pkt_len ? pskb_trim(skb, max(cap, pkt_len)) : -EPERM; 122 } 123 rcu_read_unlock(); 124 125 return err; 126 } 127 EXPORT_SYMBOL(sk_filter_trim_cap); 128 129 BPF_CALL_1(bpf_skb_get_pay_offset, struct sk_buff *, skb) 130 { 131 return skb_get_poff(skb); 132 } 133 134 BPF_CALL_3(bpf_skb_get_nlattr, struct sk_buff *, skb, u32, a, u32, x) 135 { 136 struct nlattr *nla; 137 138 if (skb_is_nonlinear(skb)) 139 return 0; 140 141 if (skb->len < sizeof(struct nlattr)) 142 return 0; 143 144 if (a > skb->len - sizeof(struct nlattr)) 145 return 0; 146 147 nla = nla_find((struct nlattr *) &skb->data[a], skb->len - a, x); 148 if (nla) 149 return (void *) nla - (void *) skb->data; 150 151 return 0; 152 } 153 154 BPF_CALL_3(bpf_skb_get_nlattr_nest, struct sk_buff *, skb, u32, a, u32, x) 155 { 156 struct nlattr *nla; 157 158 if (skb_is_nonlinear(skb)) 159 return 0; 160 161 if (skb->len < sizeof(struct nlattr)) 162 return 0; 163 164 if (a > skb->len - sizeof(struct nlattr)) 165 return 0; 166 167 nla = (struct nlattr *) &skb->data[a]; 168 if (nla->nla_len > skb->len - a) 169 return 0; 170 171 nla = nla_find_nested(nla, x); 172 if (nla) 173 return (void *) nla - (void *) skb->data; 174 175 return 0; 176 } 177 178 BPF_CALL_4(bpf_skb_load_helper_8, const struct sk_buff *, skb, const void *, 179 data, int, headlen, int, offset) 180 { 181 u8 tmp, *ptr; 182 const int len = sizeof(tmp); 183 184 if (offset >= 0) { 185 if (headlen - offset >= len) 186 return *(u8 *)(data + offset); 187 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 188 return tmp; 189 } else { 190 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 191 if (likely(ptr)) 192 return *(u8 *)ptr; 193 } 194 195 return -EFAULT; 196 } 197 198 BPF_CALL_2(bpf_skb_load_helper_8_no_cache, const struct sk_buff *, skb, 199 int, offset) 200 { 201 return ____bpf_skb_load_helper_8(skb, skb->data, skb->len - skb->data_len, 202 offset); 203 } 204 205 BPF_CALL_4(bpf_skb_load_helper_16, const struct sk_buff *, skb, const void *, 206 data, int, headlen, int, offset) 207 { 208 u16 tmp, *ptr; 209 const int len = sizeof(tmp); 210 211 if (offset >= 0) { 212 if (headlen - offset >= len) 213 return get_unaligned_be16(data + offset); 214 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 215 return be16_to_cpu(tmp); 216 } else { 217 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 218 if (likely(ptr)) 219 return get_unaligned_be16(ptr); 220 } 221 222 return -EFAULT; 223 } 224 225 BPF_CALL_2(bpf_skb_load_helper_16_no_cache, const struct sk_buff *, skb, 226 int, offset) 227 { 228 return ____bpf_skb_load_helper_16(skb, skb->data, skb->len - skb->data_len, 229 offset); 230 } 231 232 BPF_CALL_4(bpf_skb_load_helper_32, const struct sk_buff *, skb, const void *, 233 data, int, headlen, int, offset) 234 { 235 u32 tmp, *ptr; 236 const int len = sizeof(tmp); 237 238 if (likely(offset >= 0)) { 239 if (headlen - offset >= len) 240 return get_unaligned_be32(data + offset); 241 if (!skb_copy_bits(skb, offset, &tmp, sizeof(tmp))) 242 return be32_to_cpu(tmp); 243 } else { 244 ptr = bpf_internal_load_pointer_neg_helper(skb, offset, len); 245 if (likely(ptr)) 246 return get_unaligned_be32(ptr); 247 } 248 249 return -EFAULT; 250 } 251 252 BPF_CALL_2(bpf_skb_load_helper_32_no_cache, const struct sk_buff *, skb, 253 int, offset) 254 { 255 return ____bpf_skb_load_helper_32(skb, skb->data, skb->len - skb->data_len, 256 offset); 257 } 258 259 BPF_CALL_0(bpf_get_raw_cpu_id) 260 { 261 return raw_smp_processor_id(); 262 } 263 264 static const struct bpf_func_proto bpf_get_raw_smp_processor_id_proto = { 265 .func = bpf_get_raw_cpu_id, 266 .gpl_only = false, 267 .ret_type = RET_INTEGER, 268 }; 269 270 static u32 convert_skb_access(int skb_field, int dst_reg, int src_reg, 271 struct bpf_insn *insn_buf) 272 { 273 struct bpf_insn *insn = insn_buf; 274 275 switch (skb_field) { 276 case SKF_AD_MARK: 277 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, mark) != 4); 278 279 *insn++ = BPF_LDX_MEM(BPF_W, dst_reg, src_reg, 280 offsetof(struct sk_buff, mark)); 281 break; 282 283 case SKF_AD_PKTTYPE: 284 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_TYPE_OFFSET()); 285 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, PKT_TYPE_MAX); 286 #ifdef __BIG_ENDIAN_BITFIELD 287 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, 5); 288 #endif 289 break; 290 291 case SKF_AD_QUEUE: 292 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, queue_mapping) != 2); 293 294 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 295 offsetof(struct sk_buff, queue_mapping)); 296 break; 297 298 case SKF_AD_VLAN_TAG: 299 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_tci) != 2); 300 301 /* dst_reg = *(u16 *) (src_reg + offsetof(vlan_tci)) */ 302 *insn++ = BPF_LDX_MEM(BPF_H, dst_reg, src_reg, 303 offsetof(struct sk_buff, vlan_tci)); 304 break; 305 case SKF_AD_VLAN_TAG_PRESENT: 306 *insn++ = BPF_LDX_MEM(BPF_B, dst_reg, src_reg, PKT_VLAN_PRESENT_OFFSET()); 307 if (PKT_VLAN_PRESENT_BIT) 308 *insn++ = BPF_ALU32_IMM(BPF_RSH, dst_reg, PKT_VLAN_PRESENT_BIT); 309 if (PKT_VLAN_PRESENT_BIT < 7) 310 *insn++ = BPF_ALU32_IMM(BPF_AND, dst_reg, 1); 311 break; 312 } 313 314 return insn - insn_buf; 315 } 316 317 static bool convert_bpf_extensions(struct sock_filter *fp, 318 struct bpf_insn **insnp) 319 { 320 struct bpf_insn *insn = *insnp; 321 u32 cnt; 322 323 switch (fp->k) { 324 case SKF_AD_OFF + SKF_AD_PROTOCOL: 325 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, protocol) != 2); 326 327 /* A = *(u16 *) (CTX + offsetof(protocol)) */ 328 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 329 offsetof(struct sk_buff, protocol)); 330 /* A = ntohs(A) [emitting a nop or swap16] */ 331 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 332 break; 333 334 case SKF_AD_OFF + SKF_AD_PKTTYPE: 335 cnt = convert_skb_access(SKF_AD_PKTTYPE, BPF_REG_A, BPF_REG_CTX, insn); 336 insn += cnt - 1; 337 break; 338 339 case SKF_AD_OFF + SKF_AD_IFINDEX: 340 case SKF_AD_OFF + SKF_AD_HATYPE: 341 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, ifindex) != 4); 342 BUILD_BUG_ON(FIELD_SIZEOF(struct net_device, type) != 2); 343 344 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 345 BPF_REG_TMP, BPF_REG_CTX, 346 offsetof(struct sk_buff, dev)); 347 /* if (tmp != 0) goto pc + 1 */ 348 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_TMP, 0, 1); 349 *insn++ = BPF_EXIT_INSN(); 350 if (fp->k == SKF_AD_OFF + SKF_AD_IFINDEX) 351 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_TMP, 352 offsetof(struct net_device, ifindex)); 353 else 354 *insn = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_TMP, 355 offsetof(struct net_device, type)); 356 break; 357 358 case SKF_AD_OFF + SKF_AD_MARK: 359 cnt = convert_skb_access(SKF_AD_MARK, BPF_REG_A, BPF_REG_CTX, insn); 360 insn += cnt - 1; 361 break; 362 363 case SKF_AD_OFF + SKF_AD_RXHASH: 364 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, hash) != 4); 365 366 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, 367 offsetof(struct sk_buff, hash)); 368 break; 369 370 case SKF_AD_OFF + SKF_AD_QUEUE: 371 cnt = convert_skb_access(SKF_AD_QUEUE, BPF_REG_A, BPF_REG_CTX, insn); 372 insn += cnt - 1; 373 break; 374 375 case SKF_AD_OFF + SKF_AD_VLAN_TAG: 376 cnt = convert_skb_access(SKF_AD_VLAN_TAG, 377 BPF_REG_A, BPF_REG_CTX, insn); 378 insn += cnt - 1; 379 break; 380 381 case SKF_AD_OFF + SKF_AD_VLAN_TAG_PRESENT: 382 cnt = convert_skb_access(SKF_AD_VLAN_TAG_PRESENT, 383 BPF_REG_A, BPF_REG_CTX, insn); 384 insn += cnt - 1; 385 break; 386 387 case SKF_AD_OFF + SKF_AD_VLAN_TPID: 388 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, vlan_proto) != 2); 389 390 /* A = *(u16 *) (CTX + offsetof(vlan_proto)) */ 391 *insn++ = BPF_LDX_MEM(BPF_H, BPF_REG_A, BPF_REG_CTX, 392 offsetof(struct sk_buff, vlan_proto)); 393 /* A = ntohs(A) [emitting a nop or swap16] */ 394 *insn = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, 16); 395 break; 396 397 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 398 case SKF_AD_OFF + SKF_AD_NLATTR: 399 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 400 case SKF_AD_OFF + SKF_AD_CPU: 401 case SKF_AD_OFF + SKF_AD_RANDOM: 402 /* arg1 = CTX */ 403 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 404 /* arg2 = A */ 405 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_A); 406 /* arg3 = X */ 407 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_X); 408 /* Emit call(arg1=CTX, arg2=A, arg3=X) */ 409 switch (fp->k) { 410 case SKF_AD_OFF + SKF_AD_PAY_OFFSET: 411 *insn = BPF_EMIT_CALL(bpf_skb_get_pay_offset); 412 break; 413 case SKF_AD_OFF + SKF_AD_NLATTR: 414 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr); 415 break; 416 case SKF_AD_OFF + SKF_AD_NLATTR_NEST: 417 *insn = BPF_EMIT_CALL(bpf_skb_get_nlattr_nest); 418 break; 419 case SKF_AD_OFF + SKF_AD_CPU: 420 *insn = BPF_EMIT_CALL(bpf_get_raw_cpu_id); 421 break; 422 case SKF_AD_OFF + SKF_AD_RANDOM: 423 *insn = BPF_EMIT_CALL(bpf_user_rnd_u32); 424 bpf_user_rnd_init_once(); 425 break; 426 } 427 break; 428 429 case SKF_AD_OFF + SKF_AD_ALU_XOR_X: 430 /* A ^= X */ 431 *insn = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_X); 432 break; 433 434 default: 435 /* This is just a dummy call to avoid letting the compiler 436 * evict __bpf_call_base() as an optimization. Placed here 437 * where no-one bothers. 438 */ 439 BUG_ON(__bpf_call_base(0, 0, 0, 0, 0) != 0); 440 return false; 441 } 442 443 *insnp = insn; 444 return true; 445 } 446 447 static bool convert_bpf_ld_abs(struct sock_filter *fp, struct bpf_insn **insnp) 448 { 449 const bool unaligned_ok = IS_BUILTIN(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS); 450 int size = bpf_size_to_bytes(BPF_SIZE(fp->code)); 451 bool endian = BPF_SIZE(fp->code) == BPF_H || 452 BPF_SIZE(fp->code) == BPF_W; 453 bool indirect = BPF_MODE(fp->code) == BPF_IND; 454 const int ip_align = NET_IP_ALIGN; 455 struct bpf_insn *insn = *insnp; 456 int offset = fp->k; 457 458 if (!indirect && 459 ((unaligned_ok && offset >= 0) || 460 (!unaligned_ok && offset >= 0 && 461 offset + ip_align >= 0 && 462 offset + ip_align % size == 0))) { 463 bool ldx_off_ok = offset <= S16_MAX; 464 465 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_H); 466 if (offset) 467 *insn++ = BPF_ALU64_IMM(BPF_SUB, BPF_REG_TMP, offset); 468 *insn++ = BPF_JMP_IMM(BPF_JSLT, BPF_REG_TMP, 469 size, 2 + endian + (!ldx_off_ok * 2)); 470 if (ldx_off_ok) { 471 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 472 BPF_REG_D, offset); 473 } else { 474 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_D); 475 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_TMP, offset); 476 *insn++ = BPF_LDX_MEM(BPF_SIZE(fp->code), BPF_REG_A, 477 BPF_REG_TMP, 0); 478 } 479 if (endian) 480 *insn++ = BPF_ENDIAN(BPF_FROM_BE, BPF_REG_A, size * 8); 481 *insn++ = BPF_JMP_A(8); 482 } 483 484 *insn++ = BPF_MOV64_REG(BPF_REG_ARG1, BPF_REG_CTX); 485 *insn++ = BPF_MOV64_REG(BPF_REG_ARG2, BPF_REG_D); 486 *insn++ = BPF_MOV64_REG(BPF_REG_ARG3, BPF_REG_H); 487 if (!indirect) { 488 *insn++ = BPF_MOV64_IMM(BPF_REG_ARG4, offset); 489 } else { 490 *insn++ = BPF_MOV64_REG(BPF_REG_ARG4, BPF_REG_X); 491 if (fp->k) 492 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_ARG4, offset); 493 } 494 495 switch (BPF_SIZE(fp->code)) { 496 case BPF_B: 497 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8); 498 break; 499 case BPF_H: 500 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16); 501 break; 502 case BPF_W: 503 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32); 504 break; 505 default: 506 return false; 507 } 508 509 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_A, 0, 2); 510 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 511 *insn = BPF_EXIT_INSN(); 512 513 *insnp = insn; 514 return true; 515 } 516 517 /** 518 * bpf_convert_filter - convert filter program 519 * @prog: the user passed filter program 520 * @len: the length of the user passed filter program 521 * @new_prog: allocated 'struct bpf_prog' or NULL 522 * @new_len: pointer to store length of converted program 523 * @seen_ld_abs: bool whether we've seen ld_abs/ind 524 * 525 * Remap 'sock_filter' style classic BPF (cBPF) instruction set to 'bpf_insn' 526 * style extended BPF (eBPF). 527 * Conversion workflow: 528 * 529 * 1) First pass for calculating the new program length: 530 * bpf_convert_filter(old_prog, old_len, NULL, &new_len, &seen_ld_abs) 531 * 532 * 2) 2nd pass to remap in two passes: 1st pass finds new 533 * jump offsets, 2nd pass remapping: 534 * bpf_convert_filter(old_prog, old_len, new_prog, &new_len, &seen_ld_abs) 535 */ 536 static int bpf_convert_filter(struct sock_filter *prog, int len, 537 struct bpf_prog *new_prog, int *new_len, 538 bool *seen_ld_abs) 539 { 540 int new_flen = 0, pass = 0, target, i, stack_off; 541 struct bpf_insn *new_insn, *first_insn = NULL; 542 struct sock_filter *fp; 543 int *addrs = NULL; 544 u8 bpf_src; 545 546 BUILD_BUG_ON(BPF_MEMWORDS * sizeof(u32) > MAX_BPF_STACK); 547 BUILD_BUG_ON(BPF_REG_FP + 1 != MAX_BPF_REG); 548 549 if (len <= 0 || len > BPF_MAXINSNS) 550 return -EINVAL; 551 552 if (new_prog) { 553 first_insn = new_prog->insnsi; 554 addrs = kcalloc(len, sizeof(*addrs), 555 GFP_KERNEL | __GFP_NOWARN); 556 if (!addrs) 557 return -ENOMEM; 558 } 559 560 do_pass: 561 new_insn = first_insn; 562 fp = prog; 563 564 /* Classic BPF related prologue emission. */ 565 if (new_prog) { 566 /* Classic BPF expects A and X to be reset first. These need 567 * to be guaranteed to be the first two instructions. 568 */ 569 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 570 *new_insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_X, BPF_REG_X); 571 572 /* All programs must keep CTX in callee saved BPF_REG_CTX. 573 * In eBPF case it's done by the compiler, here we need to 574 * do this ourself. Initial CTX is present in BPF_REG_ARG1. 575 */ 576 *new_insn++ = BPF_MOV64_REG(BPF_REG_CTX, BPF_REG_ARG1); 577 if (*seen_ld_abs) { 578 /* For packet access in classic BPF, cache skb->data 579 * in callee-saved BPF R8 and skb->len - skb->data_len 580 * (headlen) in BPF R9. Since classic BPF is read-only 581 * on CTX, we only need to cache it once. 582 */ 583 *new_insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 584 BPF_REG_D, BPF_REG_CTX, 585 offsetof(struct sk_buff, data)); 586 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_H, BPF_REG_CTX, 587 offsetof(struct sk_buff, len)); 588 *new_insn++ = BPF_LDX_MEM(BPF_W, BPF_REG_TMP, BPF_REG_CTX, 589 offsetof(struct sk_buff, data_len)); 590 *new_insn++ = BPF_ALU32_REG(BPF_SUB, BPF_REG_H, BPF_REG_TMP); 591 } 592 } else { 593 new_insn += 3; 594 } 595 596 for (i = 0; i < len; fp++, i++) { 597 struct bpf_insn tmp_insns[32] = { }; 598 struct bpf_insn *insn = tmp_insns; 599 600 if (addrs) 601 addrs[i] = new_insn - first_insn; 602 603 switch (fp->code) { 604 /* All arithmetic insns and skb loads map as-is. */ 605 case BPF_ALU | BPF_ADD | BPF_X: 606 case BPF_ALU | BPF_ADD | BPF_K: 607 case BPF_ALU | BPF_SUB | BPF_X: 608 case BPF_ALU | BPF_SUB | BPF_K: 609 case BPF_ALU | BPF_AND | BPF_X: 610 case BPF_ALU | BPF_AND | BPF_K: 611 case BPF_ALU | BPF_OR | BPF_X: 612 case BPF_ALU | BPF_OR | BPF_K: 613 case BPF_ALU | BPF_LSH | BPF_X: 614 case BPF_ALU | BPF_LSH | BPF_K: 615 case BPF_ALU | BPF_RSH | BPF_X: 616 case BPF_ALU | BPF_RSH | BPF_K: 617 case BPF_ALU | BPF_XOR | BPF_X: 618 case BPF_ALU | BPF_XOR | BPF_K: 619 case BPF_ALU | BPF_MUL | BPF_X: 620 case BPF_ALU | BPF_MUL | BPF_K: 621 case BPF_ALU | BPF_DIV | BPF_X: 622 case BPF_ALU | BPF_DIV | BPF_K: 623 case BPF_ALU | BPF_MOD | BPF_X: 624 case BPF_ALU | BPF_MOD | BPF_K: 625 case BPF_ALU | BPF_NEG: 626 case BPF_LD | BPF_ABS | BPF_W: 627 case BPF_LD | BPF_ABS | BPF_H: 628 case BPF_LD | BPF_ABS | BPF_B: 629 case BPF_LD | BPF_IND | BPF_W: 630 case BPF_LD | BPF_IND | BPF_H: 631 case BPF_LD | BPF_IND | BPF_B: 632 /* Check for overloaded BPF extension and 633 * directly convert it if found, otherwise 634 * just move on with mapping. 635 */ 636 if (BPF_CLASS(fp->code) == BPF_LD && 637 BPF_MODE(fp->code) == BPF_ABS && 638 convert_bpf_extensions(fp, &insn)) 639 break; 640 if (BPF_CLASS(fp->code) == BPF_LD && 641 convert_bpf_ld_abs(fp, &insn)) { 642 *seen_ld_abs = true; 643 break; 644 } 645 646 if (fp->code == (BPF_ALU | BPF_DIV | BPF_X) || 647 fp->code == (BPF_ALU | BPF_MOD | BPF_X)) { 648 *insn++ = BPF_MOV32_REG(BPF_REG_X, BPF_REG_X); 649 /* Error with exception code on div/mod by 0. 650 * For cBPF programs, this was always return 0. 651 */ 652 *insn++ = BPF_JMP_IMM(BPF_JNE, BPF_REG_X, 0, 2); 653 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_A, BPF_REG_A); 654 *insn++ = BPF_EXIT_INSN(); 655 } 656 657 *insn = BPF_RAW_INSN(fp->code, BPF_REG_A, BPF_REG_X, 0, fp->k); 658 break; 659 660 /* Jump transformation cannot use BPF block macros 661 * everywhere as offset calculation and target updates 662 * require a bit more work than the rest, i.e. jump 663 * opcodes map as-is, but offsets need adjustment. 664 */ 665 666 #define BPF_EMIT_JMP \ 667 do { \ 668 const s32 off_min = S16_MIN, off_max = S16_MAX; \ 669 s32 off; \ 670 \ 671 if (target >= len || target < 0) \ 672 goto err; \ 673 off = addrs ? addrs[target] - addrs[i] - 1 : 0; \ 674 /* Adjust pc relative offset for 2nd or 3rd insn. */ \ 675 off -= insn - tmp_insns; \ 676 /* Reject anything not fitting into insn->off. */ \ 677 if (off < off_min || off > off_max) \ 678 goto err; \ 679 insn->off = off; \ 680 } while (0) 681 682 case BPF_JMP | BPF_JA: 683 target = i + fp->k + 1; 684 insn->code = fp->code; 685 BPF_EMIT_JMP; 686 break; 687 688 case BPF_JMP | BPF_JEQ | BPF_K: 689 case BPF_JMP | BPF_JEQ | BPF_X: 690 case BPF_JMP | BPF_JSET | BPF_K: 691 case BPF_JMP | BPF_JSET | BPF_X: 692 case BPF_JMP | BPF_JGT | BPF_K: 693 case BPF_JMP | BPF_JGT | BPF_X: 694 case BPF_JMP | BPF_JGE | BPF_K: 695 case BPF_JMP | BPF_JGE | BPF_X: 696 if (BPF_SRC(fp->code) == BPF_K && (int) fp->k < 0) { 697 /* BPF immediates are signed, zero extend 698 * immediate into tmp register and use it 699 * in compare insn. 700 */ 701 *insn++ = BPF_MOV32_IMM(BPF_REG_TMP, fp->k); 702 703 insn->dst_reg = BPF_REG_A; 704 insn->src_reg = BPF_REG_TMP; 705 bpf_src = BPF_X; 706 } else { 707 insn->dst_reg = BPF_REG_A; 708 insn->imm = fp->k; 709 bpf_src = BPF_SRC(fp->code); 710 insn->src_reg = bpf_src == BPF_X ? BPF_REG_X : 0; 711 } 712 713 /* Common case where 'jump_false' is next insn. */ 714 if (fp->jf == 0) { 715 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 716 target = i + fp->jt + 1; 717 BPF_EMIT_JMP; 718 break; 719 } 720 721 /* Convert some jumps when 'jump_true' is next insn. */ 722 if (fp->jt == 0) { 723 switch (BPF_OP(fp->code)) { 724 case BPF_JEQ: 725 insn->code = BPF_JMP | BPF_JNE | bpf_src; 726 break; 727 case BPF_JGT: 728 insn->code = BPF_JMP | BPF_JLE | bpf_src; 729 break; 730 case BPF_JGE: 731 insn->code = BPF_JMP | BPF_JLT | bpf_src; 732 break; 733 default: 734 goto jmp_rest; 735 } 736 737 target = i + fp->jf + 1; 738 BPF_EMIT_JMP; 739 break; 740 } 741 jmp_rest: 742 /* Other jumps are mapped into two insns: Jxx and JA. */ 743 target = i + fp->jt + 1; 744 insn->code = BPF_JMP | BPF_OP(fp->code) | bpf_src; 745 BPF_EMIT_JMP; 746 insn++; 747 748 insn->code = BPF_JMP | BPF_JA; 749 target = i + fp->jf + 1; 750 BPF_EMIT_JMP; 751 break; 752 753 /* ldxb 4 * ([14] & 0xf) is remaped into 6 insns. */ 754 case BPF_LDX | BPF_MSH | BPF_B: { 755 struct sock_filter tmp = { 756 .code = BPF_LD | BPF_ABS | BPF_B, 757 .k = fp->k, 758 }; 759 760 *seen_ld_abs = true; 761 762 /* X = A */ 763 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 764 /* A = BPF_R0 = *(u8 *) (skb->data + K) */ 765 convert_bpf_ld_abs(&tmp, &insn); 766 insn++; 767 /* A &= 0xf */ 768 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_A, 0xf); 769 /* A <<= 2 */ 770 *insn++ = BPF_ALU32_IMM(BPF_LSH, BPF_REG_A, 2); 771 /* tmp = X */ 772 *insn++ = BPF_MOV64_REG(BPF_REG_TMP, BPF_REG_X); 773 /* X = A */ 774 *insn++ = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 775 /* A = tmp */ 776 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_TMP); 777 break; 778 } 779 /* RET_K is remaped into 2 insns. RET_A case doesn't need an 780 * extra mov as BPF_REG_0 is already mapped into BPF_REG_A. 781 */ 782 case BPF_RET | BPF_A: 783 case BPF_RET | BPF_K: 784 if (BPF_RVAL(fp->code) == BPF_K) 785 *insn++ = BPF_MOV32_RAW(BPF_K, BPF_REG_0, 786 0, fp->k); 787 *insn = BPF_EXIT_INSN(); 788 break; 789 790 /* Store to stack. */ 791 case BPF_ST: 792 case BPF_STX: 793 stack_off = fp->k * 4 + 4; 794 *insn = BPF_STX_MEM(BPF_W, BPF_REG_FP, BPF_CLASS(fp->code) == 795 BPF_ST ? BPF_REG_A : BPF_REG_X, 796 -stack_off); 797 /* check_load_and_stores() verifies that classic BPF can 798 * load from stack only after write, so tracking 799 * stack_depth for ST|STX insns is enough 800 */ 801 if (new_prog && new_prog->aux->stack_depth < stack_off) 802 new_prog->aux->stack_depth = stack_off; 803 break; 804 805 /* Load from stack. */ 806 case BPF_LD | BPF_MEM: 807 case BPF_LDX | BPF_MEM: 808 stack_off = fp->k * 4 + 4; 809 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 810 BPF_REG_A : BPF_REG_X, BPF_REG_FP, 811 -stack_off); 812 break; 813 814 /* A = K or X = K */ 815 case BPF_LD | BPF_IMM: 816 case BPF_LDX | BPF_IMM: 817 *insn = BPF_MOV32_IMM(BPF_CLASS(fp->code) == BPF_LD ? 818 BPF_REG_A : BPF_REG_X, fp->k); 819 break; 820 821 /* X = A */ 822 case BPF_MISC | BPF_TAX: 823 *insn = BPF_MOV64_REG(BPF_REG_X, BPF_REG_A); 824 break; 825 826 /* A = X */ 827 case BPF_MISC | BPF_TXA: 828 *insn = BPF_MOV64_REG(BPF_REG_A, BPF_REG_X); 829 break; 830 831 /* A = skb->len or X = skb->len */ 832 case BPF_LD | BPF_W | BPF_LEN: 833 case BPF_LDX | BPF_W | BPF_LEN: 834 *insn = BPF_LDX_MEM(BPF_W, BPF_CLASS(fp->code) == BPF_LD ? 835 BPF_REG_A : BPF_REG_X, BPF_REG_CTX, 836 offsetof(struct sk_buff, len)); 837 break; 838 839 /* Access seccomp_data fields. */ 840 case BPF_LDX | BPF_ABS | BPF_W: 841 /* A = *(u32 *) (ctx + K) */ 842 *insn = BPF_LDX_MEM(BPF_W, BPF_REG_A, BPF_REG_CTX, fp->k); 843 break; 844 845 /* Unknown instruction. */ 846 default: 847 goto err; 848 } 849 850 insn++; 851 if (new_prog) 852 memcpy(new_insn, tmp_insns, 853 sizeof(*insn) * (insn - tmp_insns)); 854 new_insn += insn - tmp_insns; 855 } 856 857 if (!new_prog) { 858 /* Only calculating new length. */ 859 *new_len = new_insn - first_insn; 860 if (*seen_ld_abs) 861 *new_len += 4; /* Prologue bits. */ 862 return 0; 863 } 864 865 pass++; 866 if (new_flen != new_insn - first_insn) { 867 new_flen = new_insn - first_insn; 868 if (pass > 2) 869 goto err; 870 goto do_pass; 871 } 872 873 kfree(addrs); 874 BUG_ON(*new_len != new_flen); 875 return 0; 876 err: 877 kfree(addrs); 878 return -EINVAL; 879 } 880 881 /* Security: 882 * 883 * As we dont want to clear mem[] array for each packet going through 884 * __bpf_prog_run(), we check that filter loaded by user never try to read 885 * a cell if not previously written, and we check all branches to be sure 886 * a malicious user doesn't try to abuse us. 887 */ 888 static int check_load_and_stores(const struct sock_filter *filter, int flen) 889 { 890 u16 *masks, memvalid = 0; /* One bit per cell, 16 cells */ 891 int pc, ret = 0; 892 893 BUILD_BUG_ON(BPF_MEMWORDS > 16); 894 895 masks = kmalloc_array(flen, sizeof(*masks), GFP_KERNEL); 896 if (!masks) 897 return -ENOMEM; 898 899 memset(masks, 0xff, flen * sizeof(*masks)); 900 901 for (pc = 0; pc < flen; pc++) { 902 memvalid &= masks[pc]; 903 904 switch (filter[pc].code) { 905 case BPF_ST: 906 case BPF_STX: 907 memvalid |= (1 << filter[pc].k); 908 break; 909 case BPF_LD | BPF_MEM: 910 case BPF_LDX | BPF_MEM: 911 if (!(memvalid & (1 << filter[pc].k))) { 912 ret = -EINVAL; 913 goto error; 914 } 915 break; 916 case BPF_JMP | BPF_JA: 917 /* A jump must set masks on target */ 918 masks[pc + 1 + filter[pc].k] &= memvalid; 919 memvalid = ~0; 920 break; 921 case BPF_JMP | BPF_JEQ | BPF_K: 922 case BPF_JMP | BPF_JEQ | BPF_X: 923 case BPF_JMP | BPF_JGE | BPF_K: 924 case BPF_JMP | BPF_JGE | BPF_X: 925 case BPF_JMP | BPF_JGT | BPF_K: 926 case BPF_JMP | BPF_JGT | BPF_X: 927 case BPF_JMP | BPF_JSET | BPF_K: 928 case BPF_JMP | BPF_JSET | BPF_X: 929 /* A jump must set masks on targets */ 930 masks[pc + 1 + filter[pc].jt] &= memvalid; 931 masks[pc + 1 + filter[pc].jf] &= memvalid; 932 memvalid = ~0; 933 break; 934 } 935 } 936 error: 937 kfree(masks); 938 return ret; 939 } 940 941 static bool chk_code_allowed(u16 code_to_probe) 942 { 943 static const bool codes[] = { 944 /* 32 bit ALU operations */ 945 [BPF_ALU | BPF_ADD | BPF_K] = true, 946 [BPF_ALU | BPF_ADD | BPF_X] = true, 947 [BPF_ALU | BPF_SUB | BPF_K] = true, 948 [BPF_ALU | BPF_SUB | BPF_X] = true, 949 [BPF_ALU | BPF_MUL | BPF_K] = true, 950 [BPF_ALU | BPF_MUL | BPF_X] = true, 951 [BPF_ALU | BPF_DIV | BPF_K] = true, 952 [BPF_ALU | BPF_DIV | BPF_X] = true, 953 [BPF_ALU | BPF_MOD | BPF_K] = true, 954 [BPF_ALU | BPF_MOD | BPF_X] = true, 955 [BPF_ALU | BPF_AND | BPF_K] = true, 956 [BPF_ALU | BPF_AND | BPF_X] = true, 957 [BPF_ALU | BPF_OR | BPF_K] = true, 958 [BPF_ALU | BPF_OR | BPF_X] = true, 959 [BPF_ALU | BPF_XOR | BPF_K] = true, 960 [BPF_ALU | BPF_XOR | BPF_X] = true, 961 [BPF_ALU | BPF_LSH | BPF_K] = true, 962 [BPF_ALU | BPF_LSH | BPF_X] = true, 963 [BPF_ALU | BPF_RSH | BPF_K] = true, 964 [BPF_ALU | BPF_RSH | BPF_X] = true, 965 [BPF_ALU | BPF_NEG] = true, 966 /* Load instructions */ 967 [BPF_LD | BPF_W | BPF_ABS] = true, 968 [BPF_LD | BPF_H | BPF_ABS] = true, 969 [BPF_LD | BPF_B | BPF_ABS] = true, 970 [BPF_LD | BPF_W | BPF_LEN] = true, 971 [BPF_LD | BPF_W | BPF_IND] = true, 972 [BPF_LD | BPF_H | BPF_IND] = true, 973 [BPF_LD | BPF_B | BPF_IND] = true, 974 [BPF_LD | BPF_IMM] = true, 975 [BPF_LD | BPF_MEM] = true, 976 [BPF_LDX | BPF_W | BPF_LEN] = true, 977 [BPF_LDX | BPF_B | BPF_MSH] = true, 978 [BPF_LDX | BPF_IMM] = true, 979 [BPF_LDX | BPF_MEM] = true, 980 /* Store instructions */ 981 [BPF_ST] = true, 982 [BPF_STX] = true, 983 /* Misc instructions */ 984 [BPF_MISC | BPF_TAX] = true, 985 [BPF_MISC | BPF_TXA] = true, 986 /* Return instructions */ 987 [BPF_RET | BPF_K] = true, 988 [BPF_RET | BPF_A] = true, 989 /* Jump instructions */ 990 [BPF_JMP | BPF_JA] = true, 991 [BPF_JMP | BPF_JEQ | BPF_K] = true, 992 [BPF_JMP | BPF_JEQ | BPF_X] = true, 993 [BPF_JMP | BPF_JGE | BPF_K] = true, 994 [BPF_JMP | BPF_JGE | BPF_X] = true, 995 [BPF_JMP | BPF_JGT | BPF_K] = true, 996 [BPF_JMP | BPF_JGT | BPF_X] = true, 997 [BPF_JMP | BPF_JSET | BPF_K] = true, 998 [BPF_JMP | BPF_JSET | BPF_X] = true, 999 }; 1000 1001 if (code_to_probe >= ARRAY_SIZE(codes)) 1002 return false; 1003 1004 return codes[code_to_probe]; 1005 } 1006 1007 static bool bpf_check_basics_ok(const struct sock_filter *filter, 1008 unsigned int flen) 1009 { 1010 if (filter == NULL) 1011 return false; 1012 if (flen == 0 || flen > BPF_MAXINSNS) 1013 return false; 1014 1015 return true; 1016 } 1017 1018 /** 1019 * bpf_check_classic - verify socket filter code 1020 * @filter: filter to verify 1021 * @flen: length of filter 1022 * 1023 * Check the user's filter code. If we let some ugly 1024 * filter code slip through kaboom! The filter must contain 1025 * no references or jumps that are out of range, no illegal 1026 * instructions, and must end with a RET instruction. 1027 * 1028 * All jumps are forward as they are not signed. 1029 * 1030 * Returns 0 if the rule set is legal or -EINVAL if not. 1031 */ 1032 static int bpf_check_classic(const struct sock_filter *filter, 1033 unsigned int flen) 1034 { 1035 bool anc_found; 1036 int pc; 1037 1038 /* Check the filter code now */ 1039 for (pc = 0; pc < flen; pc++) { 1040 const struct sock_filter *ftest = &filter[pc]; 1041 1042 /* May we actually operate on this code? */ 1043 if (!chk_code_allowed(ftest->code)) 1044 return -EINVAL; 1045 1046 /* Some instructions need special checks */ 1047 switch (ftest->code) { 1048 case BPF_ALU | BPF_DIV | BPF_K: 1049 case BPF_ALU | BPF_MOD | BPF_K: 1050 /* Check for division by zero */ 1051 if (ftest->k == 0) 1052 return -EINVAL; 1053 break; 1054 case BPF_ALU | BPF_LSH | BPF_K: 1055 case BPF_ALU | BPF_RSH | BPF_K: 1056 if (ftest->k >= 32) 1057 return -EINVAL; 1058 break; 1059 case BPF_LD | BPF_MEM: 1060 case BPF_LDX | BPF_MEM: 1061 case BPF_ST: 1062 case BPF_STX: 1063 /* Check for invalid memory addresses */ 1064 if (ftest->k >= BPF_MEMWORDS) 1065 return -EINVAL; 1066 break; 1067 case BPF_JMP | BPF_JA: 1068 /* Note, the large ftest->k might cause loops. 1069 * Compare this with conditional jumps below, 1070 * where offsets are limited. --ANK (981016) 1071 */ 1072 if (ftest->k >= (unsigned int)(flen - pc - 1)) 1073 return -EINVAL; 1074 break; 1075 case BPF_JMP | BPF_JEQ | BPF_K: 1076 case BPF_JMP | BPF_JEQ | BPF_X: 1077 case BPF_JMP | BPF_JGE | BPF_K: 1078 case BPF_JMP | BPF_JGE | BPF_X: 1079 case BPF_JMP | BPF_JGT | BPF_K: 1080 case BPF_JMP | BPF_JGT | BPF_X: 1081 case BPF_JMP | BPF_JSET | BPF_K: 1082 case BPF_JMP | BPF_JSET | BPF_X: 1083 /* Both conditionals must be safe */ 1084 if (pc + ftest->jt + 1 >= flen || 1085 pc + ftest->jf + 1 >= flen) 1086 return -EINVAL; 1087 break; 1088 case BPF_LD | BPF_W | BPF_ABS: 1089 case BPF_LD | BPF_H | BPF_ABS: 1090 case BPF_LD | BPF_B | BPF_ABS: 1091 anc_found = false; 1092 if (bpf_anc_helper(ftest) & BPF_ANC) 1093 anc_found = true; 1094 /* Ancillary operation unknown or unsupported */ 1095 if (anc_found == false && ftest->k >= SKF_AD_OFF) 1096 return -EINVAL; 1097 } 1098 } 1099 1100 /* Last instruction must be a RET code */ 1101 switch (filter[flen - 1].code) { 1102 case BPF_RET | BPF_K: 1103 case BPF_RET | BPF_A: 1104 return check_load_and_stores(filter, flen); 1105 } 1106 1107 return -EINVAL; 1108 } 1109 1110 static int bpf_prog_store_orig_filter(struct bpf_prog *fp, 1111 const struct sock_fprog *fprog) 1112 { 1113 unsigned int fsize = bpf_classic_proglen(fprog); 1114 struct sock_fprog_kern *fkprog; 1115 1116 fp->orig_prog = kmalloc(sizeof(*fkprog), GFP_KERNEL); 1117 if (!fp->orig_prog) 1118 return -ENOMEM; 1119 1120 fkprog = fp->orig_prog; 1121 fkprog->len = fprog->len; 1122 1123 fkprog->filter = kmemdup(fp->insns, fsize, 1124 GFP_KERNEL | __GFP_NOWARN); 1125 if (!fkprog->filter) { 1126 kfree(fp->orig_prog); 1127 return -ENOMEM; 1128 } 1129 1130 return 0; 1131 } 1132 1133 static void bpf_release_orig_filter(struct bpf_prog *fp) 1134 { 1135 struct sock_fprog_kern *fprog = fp->orig_prog; 1136 1137 if (fprog) { 1138 kfree(fprog->filter); 1139 kfree(fprog); 1140 } 1141 } 1142 1143 static void __bpf_prog_release(struct bpf_prog *prog) 1144 { 1145 if (prog->type == BPF_PROG_TYPE_SOCKET_FILTER) { 1146 bpf_prog_put(prog); 1147 } else { 1148 bpf_release_orig_filter(prog); 1149 bpf_prog_free(prog); 1150 } 1151 } 1152 1153 static void __sk_filter_release(struct sk_filter *fp) 1154 { 1155 __bpf_prog_release(fp->prog); 1156 kfree(fp); 1157 } 1158 1159 /** 1160 * sk_filter_release_rcu - Release a socket filter by rcu_head 1161 * @rcu: rcu_head that contains the sk_filter to free 1162 */ 1163 static void sk_filter_release_rcu(struct rcu_head *rcu) 1164 { 1165 struct sk_filter *fp = container_of(rcu, struct sk_filter, rcu); 1166 1167 __sk_filter_release(fp); 1168 } 1169 1170 /** 1171 * sk_filter_release - release a socket filter 1172 * @fp: filter to remove 1173 * 1174 * Remove a filter from a socket and release its resources. 1175 */ 1176 static void sk_filter_release(struct sk_filter *fp) 1177 { 1178 if (refcount_dec_and_test(&fp->refcnt)) 1179 call_rcu(&fp->rcu, sk_filter_release_rcu); 1180 } 1181 1182 void sk_filter_uncharge(struct sock *sk, struct sk_filter *fp) 1183 { 1184 u32 filter_size = bpf_prog_size(fp->prog->len); 1185 1186 atomic_sub(filter_size, &sk->sk_omem_alloc); 1187 sk_filter_release(fp); 1188 } 1189 1190 /* try to charge the socket memory if there is space available 1191 * return true on success 1192 */ 1193 static bool __sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1194 { 1195 u32 filter_size = bpf_prog_size(fp->prog->len); 1196 1197 /* same check as in sock_kmalloc() */ 1198 if (filter_size <= sysctl_optmem_max && 1199 atomic_read(&sk->sk_omem_alloc) + filter_size < sysctl_optmem_max) { 1200 atomic_add(filter_size, &sk->sk_omem_alloc); 1201 return true; 1202 } 1203 return false; 1204 } 1205 1206 bool sk_filter_charge(struct sock *sk, struct sk_filter *fp) 1207 { 1208 if (!refcount_inc_not_zero(&fp->refcnt)) 1209 return false; 1210 1211 if (!__sk_filter_charge(sk, fp)) { 1212 sk_filter_release(fp); 1213 return false; 1214 } 1215 return true; 1216 } 1217 1218 static struct bpf_prog *bpf_migrate_filter(struct bpf_prog *fp) 1219 { 1220 struct sock_filter *old_prog; 1221 struct bpf_prog *old_fp; 1222 int err, new_len, old_len = fp->len; 1223 bool seen_ld_abs = false; 1224 1225 /* We are free to overwrite insns et al right here as it 1226 * won't be used at this point in time anymore internally 1227 * after the migration to the internal BPF instruction 1228 * representation. 1229 */ 1230 BUILD_BUG_ON(sizeof(struct sock_filter) != 1231 sizeof(struct bpf_insn)); 1232 1233 /* Conversion cannot happen on overlapping memory areas, 1234 * so we need to keep the user BPF around until the 2nd 1235 * pass. At this time, the user BPF is stored in fp->insns. 1236 */ 1237 old_prog = kmemdup(fp->insns, old_len * sizeof(struct sock_filter), 1238 GFP_KERNEL | __GFP_NOWARN); 1239 if (!old_prog) { 1240 err = -ENOMEM; 1241 goto out_err; 1242 } 1243 1244 /* 1st pass: calculate the new program length. */ 1245 err = bpf_convert_filter(old_prog, old_len, NULL, &new_len, 1246 &seen_ld_abs); 1247 if (err) 1248 goto out_err_free; 1249 1250 /* Expand fp for appending the new filter representation. */ 1251 old_fp = fp; 1252 fp = bpf_prog_realloc(old_fp, bpf_prog_size(new_len), 0); 1253 if (!fp) { 1254 /* The old_fp is still around in case we couldn't 1255 * allocate new memory, so uncharge on that one. 1256 */ 1257 fp = old_fp; 1258 err = -ENOMEM; 1259 goto out_err_free; 1260 } 1261 1262 fp->len = new_len; 1263 1264 /* 2nd pass: remap sock_filter insns into bpf_insn insns. */ 1265 err = bpf_convert_filter(old_prog, old_len, fp, &new_len, 1266 &seen_ld_abs); 1267 if (err) 1268 /* 2nd bpf_convert_filter() can fail only if it fails 1269 * to allocate memory, remapping must succeed. Note, 1270 * that at this time old_fp has already been released 1271 * by krealloc(). 1272 */ 1273 goto out_err_free; 1274 1275 fp = bpf_prog_select_runtime(fp, &err); 1276 if (err) 1277 goto out_err_free; 1278 1279 kfree(old_prog); 1280 return fp; 1281 1282 out_err_free: 1283 kfree(old_prog); 1284 out_err: 1285 __bpf_prog_release(fp); 1286 return ERR_PTR(err); 1287 } 1288 1289 static struct bpf_prog *bpf_prepare_filter(struct bpf_prog *fp, 1290 bpf_aux_classic_check_t trans) 1291 { 1292 int err; 1293 1294 fp->bpf_func = NULL; 1295 fp->jited = 0; 1296 1297 err = bpf_check_classic(fp->insns, fp->len); 1298 if (err) { 1299 __bpf_prog_release(fp); 1300 return ERR_PTR(err); 1301 } 1302 1303 /* There might be additional checks and transformations 1304 * needed on classic filters, f.e. in case of seccomp. 1305 */ 1306 if (trans) { 1307 err = trans(fp->insns, fp->len); 1308 if (err) { 1309 __bpf_prog_release(fp); 1310 return ERR_PTR(err); 1311 } 1312 } 1313 1314 /* Probe if we can JIT compile the filter and if so, do 1315 * the compilation of the filter. 1316 */ 1317 bpf_jit_compile(fp); 1318 1319 /* JIT compiler couldn't process this filter, so do the 1320 * internal BPF translation for the optimized interpreter. 1321 */ 1322 if (!fp->jited) 1323 fp = bpf_migrate_filter(fp); 1324 1325 return fp; 1326 } 1327 1328 /** 1329 * bpf_prog_create - create an unattached filter 1330 * @pfp: the unattached filter that is created 1331 * @fprog: the filter program 1332 * 1333 * Create a filter independent of any socket. We first run some 1334 * sanity checks on it to make sure it does not explode on us later. 1335 * If an error occurs or there is insufficient memory for the filter 1336 * a negative errno code is returned. On success the return is zero. 1337 */ 1338 int bpf_prog_create(struct bpf_prog **pfp, struct sock_fprog_kern *fprog) 1339 { 1340 unsigned int fsize = bpf_classic_proglen(fprog); 1341 struct bpf_prog *fp; 1342 1343 /* Make sure new filter is there and in the right amounts. */ 1344 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1345 return -EINVAL; 1346 1347 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1348 if (!fp) 1349 return -ENOMEM; 1350 1351 memcpy(fp->insns, fprog->filter, fsize); 1352 1353 fp->len = fprog->len; 1354 /* Since unattached filters are not copied back to user 1355 * space through sk_get_filter(), we do not need to hold 1356 * a copy here, and can spare us the work. 1357 */ 1358 fp->orig_prog = NULL; 1359 1360 /* bpf_prepare_filter() already takes care of freeing 1361 * memory in case something goes wrong. 1362 */ 1363 fp = bpf_prepare_filter(fp, NULL); 1364 if (IS_ERR(fp)) 1365 return PTR_ERR(fp); 1366 1367 *pfp = fp; 1368 return 0; 1369 } 1370 EXPORT_SYMBOL_GPL(bpf_prog_create); 1371 1372 /** 1373 * bpf_prog_create_from_user - create an unattached filter from user buffer 1374 * @pfp: the unattached filter that is created 1375 * @fprog: the filter program 1376 * @trans: post-classic verifier transformation handler 1377 * @save_orig: save classic BPF program 1378 * 1379 * This function effectively does the same as bpf_prog_create(), only 1380 * that it builds up its insns buffer from user space provided buffer. 1381 * It also allows for passing a bpf_aux_classic_check_t handler. 1382 */ 1383 int bpf_prog_create_from_user(struct bpf_prog **pfp, struct sock_fprog *fprog, 1384 bpf_aux_classic_check_t trans, bool save_orig) 1385 { 1386 unsigned int fsize = bpf_classic_proglen(fprog); 1387 struct bpf_prog *fp; 1388 int err; 1389 1390 /* Make sure new filter is there and in the right amounts. */ 1391 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1392 return -EINVAL; 1393 1394 fp = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1395 if (!fp) 1396 return -ENOMEM; 1397 1398 if (copy_from_user(fp->insns, fprog->filter, fsize)) { 1399 __bpf_prog_free(fp); 1400 return -EFAULT; 1401 } 1402 1403 fp->len = fprog->len; 1404 fp->orig_prog = NULL; 1405 1406 if (save_orig) { 1407 err = bpf_prog_store_orig_filter(fp, fprog); 1408 if (err) { 1409 __bpf_prog_free(fp); 1410 return -ENOMEM; 1411 } 1412 } 1413 1414 /* bpf_prepare_filter() already takes care of freeing 1415 * memory in case something goes wrong. 1416 */ 1417 fp = bpf_prepare_filter(fp, trans); 1418 if (IS_ERR(fp)) 1419 return PTR_ERR(fp); 1420 1421 *pfp = fp; 1422 return 0; 1423 } 1424 EXPORT_SYMBOL_GPL(bpf_prog_create_from_user); 1425 1426 void bpf_prog_destroy(struct bpf_prog *fp) 1427 { 1428 __bpf_prog_release(fp); 1429 } 1430 EXPORT_SYMBOL_GPL(bpf_prog_destroy); 1431 1432 static int __sk_attach_prog(struct bpf_prog *prog, struct sock *sk) 1433 { 1434 struct sk_filter *fp, *old_fp; 1435 1436 fp = kmalloc(sizeof(*fp), GFP_KERNEL); 1437 if (!fp) 1438 return -ENOMEM; 1439 1440 fp->prog = prog; 1441 1442 if (!__sk_filter_charge(sk, fp)) { 1443 kfree(fp); 1444 return -ENOMEM; 1445 } 1446 refcount_set(&fp->refcnt, 1); 1447 1448 old_fp = rcu_dereference_protected(sk->sk_filter, 1449 lockdep_sock_is_held(sk)); 1450 rcu_assign_pointer(sk->sk_filter, fp); 1451 1452 if (old_fp) 1453 sk_filter_uncharge(sk, old_fp); 1454 1455 return 0; 1456 } 1457 1458 static 1459 struct bpf_prog *__get_filter(struct sock_fprog *fprog, struct sock *sk) 1460 { 1461 unsigned int fsize = bpf_classic_proglen(fprog); 1462 struct bpf_prog *prog; 1463 int err; 1464 1465 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1466 return ERR_PTR(-EPERM); 1467 1468 /* Make sure new filter is there and in the right amounts. */ 1469 if (!bpf_check_basics_ok(fprog->filter, fprog->len)) 1470 return ERR_PTR(-EINVAL); 1471 1472 prog = bpf_prog_alloc(bpf_prog_size(fprog->len), 0); 1473 if (!prog) 1474 return ERR_PTR(-ENOMEM); 1475 1476 if (copy_from_user(prog->insns, fprog->filter, fsize)) { 1477 __bpf_prog_free(prog); 1478 return ERR_PTR(-EFAULT); 1479 } 1480 1481 prog->len = fprog->len; 1482 1483 err = bpf_prog_store_orig_filter(prog, fprog); 1484 if (err) { 1485 __bpf_prog_free(prog); 1486 return ERR_PTR(-ENOMEM); 1487 } 1488 1489 /* bpf_prepare_filter() already takes care of freeing 1490 * memory in case something goes wrong. 1491 */ 1492 return bpf_prepare_filter(prog, NULL); 1493 } 1494 1495 /** 1496 * sk_attach_filter - attach a socket filter 1497 * @fprog: the filter program 1498 * @sk: the socket to use 1499 * 1500 * Attach the user's filter code. We first run some sanity checks on 1501 * it to make sure it does not explode on us later. If an error 1502 * occurs or there is insufficient memory for the filter a negative 1503 * errno code is returned. On success the return is zero. 1504 */ 1505 int sk_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1506 { 1507 struct bpf_prog *prog = __get_filter(fprog, sk); 1508 int err; 1509 1510 if (IS_ERR(prog)) 1511 return PTR_ERR(prog); 1512 1513 err = __sk_attach_prog(prog, sk); 1514 if (err < 0) { 1515 __bpf_prog_release(prog); 1516 return err; 1517 } 1518 1519 return 0; 1520 } 1521 EXPORT_SYMBOL_GPL(sk_attach_filter); 1522 1523 int sk_reuseport_attach_filter(struct sock_fprog *fprog, struct sock *sk) 1524 { 1525 struct bpf_prog *prog = __get_filter(fprog, sk); 1526 int err; 1527 1528 if (IS_ERR(prog)) 1529 return PTR_ERR(prog); 1530 1531 if (bpf_prog_size(prog->len) > sysctl_optmem_max) 1532 err = -ENOMEM; 1533 else 1534 err = reuseport_attach_prog(sk, prog); 1535 1536 if (err) 1537 __bpf_prog_release(prog); 1538 1539 return err; 1540 } 1541 1542 static struct bpf_prog *__get_bpf(u32 ufd, struct sock *sk) 1543 { 1544 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1545 return ERR_PTR(-EPERM); 1546 1547 return bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1548 } 1549 1550 int sk_attach_bpf(u32 ufd, struct sock *sk) 1551 { 1552 struct bpf_prog *prog = __get_bpf(ufd, sk); 1553 int err; 1554 1555 if (IS_ERR(prog)) 1556 return PTR_ERR(prog); 1557 1558 err = __sk_attach_prog(prog, sk); 1559 if (err < 0) { 1560 bpf_prog_put(prog); 1561 return err; 1562 } 1563 1564 return 0; 1565 } 1566 1567 int sk_reuseport_attach_bpf(u32 ufd, struct sock *sk) 1568 { 1569 struct bpf_prog *prog; 1570 int err; 1571 1572 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 1573 return -EPERM; 1574 1575 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SOCKET_FILTER); 1576 if (IS_ERR(prog) && PTR_ERR(prog) == -EINVAL) 1577 prog = bpf_prog_get_type(ufd, BPF_PROG_TYPE_SK_REUSEPORT); 1578 if (IS_ERR(prog)) 1579 return PTR_ERR(prog); 1580 1581 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) { 1582 /* Like other non BPF_PROG_TYPE_SOCKET_FILTER 1583 * bpf prog (e.g. sockmap). It depends on the 1584 * limitation imposed by bpf_prog_load(). 1585 * Hence, sysctl_optmem_max is not checked. 1586 */ 1587 if ((sk->sk_type != SOCK_STREAM && 1588 sk->sk_type != SOCK_DGRAM) || 1589 (sk->sk_protocol != IPPROTO_UDP && 1590 sk->sk_protocol != IPPROTO_TCP) || 1591 (sk->sk_family != AF_INET && 1592 sk->sk_family != AF_INET6)) { 1593 err = -ENOTSUPP; 1594 goto err_prog_put; 1595 } 1596 } else { 1597 /* BPF_PROG_TYPE_SOCKET_FILTER */ 1598 if (bpf_prog_size(prog->len) > sysctl_optmem_max) { 1599 err = -ENOMEM; 1600 goto err_prog_put; 1601 } 1602 } 1603 1604 err = reuseport_attach_prog(sk, prog); 1605 err_prog_put: 1606 if (err) 1607 bpf_prog_put(prog); 1608 1609 return err; 1610 } 1611 1612 void sk_reuseport_prog_free(struct bpf_prog *prog) 1613 { 1614 if (!prog) 1615 return; 1616 1617 if (prog->type == BPF_PROG_TYPE_SK_REUSEPORT) 1618 bpf_prog_put(prog); 1619 else 1620 bpf_prog_destroy(prog); 1621 } 1622 1623 struct bpf_scratchpad { 1624 union { 1625 __be32 diff[MAX_BPF_STACK / sizeof(__be32)]; 1626 u8 buff[MAX_BPF_STACK]; 1627 }; 1628 }; 1629 1630 static DEFINE_PER_CPU(struct bpf_scratchpad, bpf_sp); 1631 1632 static inline int __bpf_try_make_writable(struct sk_buff *skb, 1633 unsigned int write_len) 1634 { 1635 return skb_ensure_writable(skb, write_len); 1636 } 1637 1638 static inline int bpf_try_make_writable(struct sk_buff *skb, 1639 unsigned int write_len) 1640 { 1641 int err = __bpf_try_make_writable(skb, write_len); 1642 1643 bpf_compute_data_pointers(skb); 1644 return err; 1645 } 1646 1647 static int bpf_try_make_head_writable(struct sk_buff *skb) 1648 { 1649 return bpf_try_make_writable(skb, skb_headlen(skb)); 1650 } 1651 1652 static inline void bpf_push_mac_rcsum(struct sk_buff *skb) 1653 { 1654 if (skb_at_tc_ingress(skb)) 1655 skb_postpush_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1656 } 1657 1658 static inline void bpf_pull_mac_rcsum(struct sk_buff *skb) 1659 { 1660 if (skb_at_tc_ingress(skb)) 1661 skb_postpull_rcsum(skb, skb_mac_header(skb), skb->mac_len); 1662 } 1663 1664 BPF_CALL_5(bpf_skb_store_bytes, struct sk_buff *, skb, u32, offset, 1665 const void *, from, u32, len, u64, flags) 1666 { 1667 void *ptr; 1668 1669 if (unlikely(flags & ~(BPF_F_RECOMPUTE_CSUM | BPF_F_INVALIDATE_HASH))) 1670 return -EINVAL; 1671 if (unlikely(offset > 0xffff)) 1672 return -EFAULT; 1673 if (unlikely(bpf_try_make_writable(skb, offset + len))) 1674 return -EFAULT; 1675 1676 ptr = skb->data + offset; 1677 if (flags & BPF_F_RECOMPUTE_CSUM) 1678 __skb_postpull_rcsum(skb, ptr, len, offset); 1679 1680 memcpy(ptr, from, len); 1681 1682 if (flags & BPF_F_RECOMPUTE_CSUM) 1683 __skb_postpush_rcsum(skb, ptr, len, offset); 1684 if (flags & BPF_F_INVALIDATE_HASH) 1685 skb_clear_hash(skb); 1686 1687 return 0; 1688 } 1689 1690 static const struct bpf_func_proto bpf_skb_store_bytes_proto = { 1691 .func = bpf_skb_store_bytes, 1692 .gpl_only = false, 1693 .ret_type = RET_INTEGER, 1694 .arg1_type = ARG_PTR_TO_CTX, 1695 .arg2_type = ARG_ANYTHING, 1696 .arg3_type = ARG_PTR_TO_MEM, 1697 .arg4_type = ARG_CONST_SIZE, 1698 .arg5_type = ARG_ANYTHING, 1699 }; 1700 1701 BPF_CALL_4(bpf_skb_load_bytes, const struct sk_buff *, skb, u32, offset, 1702 void *, to, u32, len) 1703 { 1704 void *ptr; 1705 1706 if (unlikely(offset > 0xffff)) 1707 goto err_clear; 1708 1709 ptr = skb_header_pointer(skb, offset, len, to); 1710 if (unlikely(!ptr)) 1711 goto err_clear; 1712 if (ptr != to) 1713 memcpy(to, ptr, len); 1714 1715 return 0; 1716 err_clear: 1717 memset(to, 0, len); 1718 return -EFAULT; 1719 } 1720 1721 static const struct bpf_func_proto bpf_skb_load_bytes_proto = { 1722 .func = bpf_skb_load_bytes, 1723 .gpl_only = false, 1724 .ret_type = RET_INTEGER, 1725 .arg1_type = ARG_PTR_TO_CTX, 1726 .arg2_type = ARG_ANYTHING, 1727 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1728 .arg4_type = ARG_CONST_SIZE, 1729 }; 1730 1731 BPF_CALL_4(bpf_flow_dissector_load_bytes, 1732 const struct bpf_flow_dissector *, ctx, u32, offset, 1733 void *, to, u32, len) 1734 { 1735 void *ptr; 1736 1737 if (unlikely(offset > 0xffff)) 1738 goto err_clear; 1739 1740 if (unlikely(!ctx->skb)) 1741 goto err_clear; 1742 1743 ptr = skb_header_pointer(ctx->skb, offset, len, to); 1744 if (unlikely(!ptr)) 1745 goto err_clear; 1746 if (ptr != to) 1747 memcpy(to, ptr, len); 1748 1749 return 0; 1750 err_clear: 1751 memset(to, 0, len); 1752 return -EFAULT; 1753 } 1754 1755 static const struct bpf_func_proto bpf_flow_dissector_load_bytes_proto = { 1756 .func = bpf_flow_dissector_load_bytes, 1757 .gpl_only = false, 1758 .ret_type = RET_INTEGER, 1759 .arg1_type = ARG_PTR_TO_CTX, 1760 .arg2_type = ARG_ANYTHING, 1761 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1762 .arg4_type = ARG_CONST_SIZE, 1763 }; 1764 1765 BPF_CALL_5(bpf_skb_load_bytes_relative, const struct sk_buff *, skb, 1766 u32, offset, void *, to, u32, len, u32, start_header) 1767 { 1768 u8 *end = skb_tail_pointer(skb); 1769 u8 *net = skb_network_header(skb); 1770 u8 *mac = skb_mac_header(skb); 1771 u8 *ptr; 1772 1773 if (unlikely(offset > 0xffff || len > (end - mac))) 1774 goto err_clear; 1775 1776 switch (start_header) { 1777 case BPF_HDR_START_MAC: 1778 ptr = mac + offset; 1779 break; 1780 case BPF_HDR_START_NET: 1781 ptr = net + offset; 1782 break; 1783 default: 1784 goto err_clear; 1785 } 1786 1787 if (likely(ptr >= mac && ptr + len <= end)) { 1788 memcpy(to, ptr, len); 1789 return 0; 1790 } 1791 1792 err_clear: 1793 memset(to, 0, len); 1794 return -EFAULT; 1795 } 1796 1797 static const struct bpf_func_proto bpf_skb_load_bytes_relative_proto = { 1798 .func = bpf_skb_load_bytes_relative, 1799 .gpl_only = false, 1800 .ret_type = RET_INTEGER, 1801 .arg1_type = ARG_PTR_TO_CTX, 1802 .arg2_type = ARG_ANYTHING, 1803 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 1804 .arg4_type = ARG_CONST_SIZE, 1805 .arg5_type = ARG_ANYTHING, 1806 }; 1807 1808 BPF_CALL_2(bpf_skb_pull_data, struct sk_buff *, skb, u32, len) 1809 { 1810 /* Idea is the following: should the needed direct read/write 1811 * test fail during runtime, we can pull in more data and redo 1812 * again, since implicitly, we invalidate previous checks here. 1813 * 1814 * Or, since we know how much we need to make read/writeable, 1815 * this can be done once at the program beginning for direct 1816 * access case. By this we overcome limitations of only current 1817 * headroom being accessible. 1818 */ 1819 return bpf_try_make_writable(skb, len ? : skb_headlen(skb)); 1820 } 1821 1822 static const struct bpf_func_proto bpf_skb_pull_data_proto = { 1823 .func = bpf_skb_pull_data, 1824 .gpl_only = false, 1825 .ret_type = RET_INTEGER, 1826 .arg1_type = ARG_PTR_TO_CTX, 1827 .arg2_type = ARG_ANYTHING, 1828 }; 1829 1830 BPF_CALL_1(bpf_sk_fullsock, struct sock *, sk) 1831 { 1832 return sk_fullsock(sk) ? (unsigned long)sk : (unsigned long)NULL; 1833 } 1834 1835 static const struct bpf_func_proto bpf_sk_fullsock_proto = { 1836 .func = bpf_sk_fullsock, 1837 .gpl_only = false, 1838 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 1839 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 1840 }; 1841 1842 static inline int sk_skb_try_make_writable(struct sk_buff *skb, 1843 unsigned int write_len) 1844 { 1845 int err = __bpf_try_make_writable(skb, write_len); 1846 1847 bpf_compute_data_end_sk_skb(skb); 1848 return err; 1849 } 1850 1851 BPF_CALL_2(sk_skb_pull_data, struct sk_buff *, skb, u32, len) 1852 { 1853 /* Idea is the following: should the needed direct read/write 1854 * test fail during runtime, we can pull in more data and redo 1855 * again, since implicitly, we invalidate previous checks here. 1856 * 1857 * Or, since we know how much we need to make read/writeable, 1858 * this can be done once at the program beginning for direct 1859 * access case. By this we overcome limitations of only current 1860 * headroom being accessible. 1861 */ 1862 return sk_skb_try_make_writable(skb, len ? : skb_headlen(skb)); 1863 } 1864 1865 static const struct bpf_func_proto sk_skb_pull_data_proto = { 1866 .func = sk_skb_pull_data, 1867 .gpl_only = false, 1868 .ret_type = RET_INTEGER, 1869 .arg1_type = ARG_PTR_TO_CTX, 1870 .arg2_type = ARG_ANYTHING, 1871 }; 1872 1873 BPF_CALL_5(bpf_l3_csum_replace, struct sk_buff *, skb, u32, offset, 1874 u64, from, u64, to, u64, flags) 1875 { 1876 __sum16 *ptr; 1877 1878 if (unlikely(flags & ~(BPF_F_HDR_FIELD_MASK))) 1879 return -EINVAL; 1880 if (unlikely(offset > 0xffff || offset & 1)) 1881 return -EFAULT; 1882 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1883 return -EFAULT; 1884 1885 ptr = (__sum16 *)(skb->data + offset); 1886 switch (flags & BPF_F_HDR_FIELD_MASK) { 1887 case 0: 1888 if (unlikely(from != 0)) 1889 return -EINVAL; 1890 1891 csum_replace_by_diff(ptr, to); 1892 break; 1893 case 2: 1894 csum_replace2(ptr, from, to); 1895 break; 1896 case 4: 1897 csum_replace4(ptr, from, to); 1898 break; 1899 default: 1900 return -EINVAL; 1901 } 1902 1903 return 0; 1904 } 1905 1906 static const struct bpf_func_proto bpf_l3_csum_replace_proto = { 1907 .func = bpf_l3_csum_replace, 1908 .gpl_only = false, 1909 .ret_type = RET_INTEGER, 1910 .arg1_type = ARG_PTR_TO_CTX, 1911 .arg2_type = ARG_ANYTHING, 1912 .arg3_type = ARG_ANYTHING, 1913 .arg4_type = ARG_ANYTHING, 1914 .arg5_type = ARG_ANYTHING, 1915 }; 1916 1917 BPF_CALL_5(bpf_l4_csum_replace, struct sk_buff *, skb, u32, offset, 1918 u64, from, u64, to, u64, flags) 1919 { 1920 bool is_pseudo = flags & BPF_F_PSEUDO_HDR; 1921 bool is_mmzero = flags & BPF_F_MARK_MANGLED_0; 1922 bool do_mforce = flags & BPF_F_MARK_ENFORCE; 1923 __sum16 *ptr; 1924 1925 if (unlikely(flags & ~(BPF_F_MARK_MANGLED_0 | BPF_F_MARK_ENFORCE | 1926 BPF_F_PSEUDO_HDR | BPF_F_HDR_FIELD_MASK))) 1927 return -EINVAL; 1928 if (unlikely(offset > 0xffff || offset & 1)) 1929 return -EFAULT; 1930 if (unlikely(bpf_try_make_writable(skb, offset + sizeof(*ptr)))) 1931 return -EFAULT; 1932 1933 ptr = (__sum16 *)(skb->data + offset); 1934 if (is_mmzero && !do_mforce && !*ptr) 1935 return 0; 1936 1937 switch (flags & BPF_F_HDR_FIELD_MASK) { 1938 case 0: 1939 if (unlikely(from != 0)) 1940 return -EINVAL; 1941 1942 inet_proto_csum_replace_by_diff(ptr, skb, to, is_pseudo); 1943 break; 1944 case 2: 1945 inet_proto_csum_replace2(ptr, skb, from, to, is_pseudo); 1946 break; 1947 case 4: 1948 inet_proto_csum_replace4(ptr, skb, from, to, is_pseudo); 1949 break; 1950 default: 1951 return -EINVAL; 1952 } 1953 1954 if (is_mmzero && !*ptr) 1955 *ptr = CSUM_MANGLED_0; 1956 return 0; 1957 } 1958 1959 static const struct bpf_func_proto bpf_l4_csum_replace_proto = { 1960 .func = bpf_l4_csum_replace, 1961 .gpl_only = false, 1962 .ret_type = RET_INTEGER, 1963 .arg1_type = ARG_PTR_TO_CTX, 1964 .arg2_type = ARG_ANYTHING, 1965 .arg3_type = ARG_ANYTHING, 1966 .arg4_type = ARG_ANYTHING, 1967 .arg5_type = ARG_ANYTHING, 1968 }; 1969 1970 BPF_CALL_5(bpf_csum_diff, __be32 *, from, u32, from_size, 1971 __be32 *, to, u32, to_size, __wsum, seed) 1972 { 1973 struct bpf_scratchpad *sp = this_cpu_ptr(&bpf_sp); 1974 u32 diff_size = from_size + to_size; 1975 int i, j = 0; 1976 1977 /* This is quite flexible, some examples: 1978 * 1979 * from_size == 0, to_size > 0, seed := csum --> pushing data 1980 * from_size > 0, to_size == 0, seed := csum --> pulling data 1981 * from_size > 0, to_size > 0, seed := 0 --> diffing data 1982 * 1983 * Even for diffing, from_size and to_size don't need to be equal. 1984 */ 1985 if (unlikely(((from_size | to_size) & (sizeof(__be32) - 1)) || 1986 diff_size > sizeof(sp->diff))) 1987 return -EINVAL; 1988 1989 for (i = 0; i < from_size / sizeof(__be32); i++, j++) 1990 sp->diff[j] = ~from[i]; 1991 for (i = 0; i < to_size / sizeof(__be32); i++, j++) 1992 sp->diff[j] = to[i]; 1993 1994 return csum_partial(sp->diff, diff_size, seed); 1995 } 1996 1997 static const struct bpf_func_proto bpf_csum_diff_proto = { 1998 .func = bpf_csum_diff, 1999 .gpl_only = false, 2000 .pkt_access = true, 2001 .ret_type = RET_INTEGER, 2002 .arg1_type = ARG_PTR_TO_MEM_OR_NULL, 2003 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 2004 .arg3_type = ARG_PTR_TO_MEM_OR_NULL, 2005 .arg4_type = ARG_CONST_SIZE_OR_ZERO, 2006 .arg5_type = ARG_ANYTHING, 2007 }; 2008 2009 BPF_CALL_2(bpf_csum_update, struct sk_buff *, skb, __wsum, csum) 2010 { 2011 /* The interface is to be used in combination with bpf_csum_diff() 2012 * for direct packet writes. csum rotation for alignment as well 2013 * as emulating csum_sub() can be done from the eBPF program. 2014 */ 2015 if (skb->ip_summed == CHECKSUM_COMPLETE) 2016 return (skb->csum = csum_add(skb->csum, csum)); 2017 2018 return -ENOTSUPP; 2019 } 2020 2021 static const struct bpf_func_proto bpf_csum_update_proto = { 2022 .func = bpf_csum_update, 2023 .gpl_only = false, 2024 .ret_type = RET_INTEGER, 2025 .arg1_type = ARG_PTR_TO_CTX, 2026 .arg2_type = ARG_ANYTHING, 2027 }; 2028 2029 static inline int __bpf_rx_skb(struct net_device *dev, struct sk_buff *skb) 2030 { 2031 return dev_forward_skb(dev, skb); 2032 } 2033 2034 static inline int __bpf_rx_skb_no_mac(struct net_device *dev, 2035 struct sk_buff *skb) 2036 { 2037 int ret = ____dev_forward_skb(dev, skb); 2038 2039 if (likely(!ret)) { 2040 skb->dev = dev; 2041 ret = netif_rx(skb); 2042 } 2043 2044 return ret; 2045 } 2046 2047 static inline int __bpf_tx_skb(struct net_device *dev, struct sk_buff *skb) 2048 { 2049 int ret; 2050 2051 if (dev_xmit_recursion()) { 2052 net_crit_ratelimited("bpf: recursion limit reached on datapath, buggy bpf program?\n"); 2053 kfree_skb(skb); 2054 return -ENETDOWN; 2055 } 2056 2057 skb->dev = dev; 2058 2059 dev_xmit_recursion_inc(); 2060 ret = dev_queue_xmit(skb); 2061 dev_xmit_recursion_dec(); 2062 2063 return ret; 2064 } 2065 2066 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev, 2067 u32 flags) 2068 { 2069 unsigned int mlen = skb_network_offset(skb); 2070 2071 if (mlen) { 2072 __skb_pull(skb, mlen); 2073 2074 /* At ingress, the mac header has already been pulled once. 2075 * At egress, skb_pospull_rcsum has to be done in case that 2076 * the skb is originated from ingress (i.e. a forwarded skb) 2077 * to ensure that rcsum starts at net header. 2078 */ 2079 if (!skb_at_tc_ingress(skb)) 2080 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen); 2081 } 2082 skb_pop_mac_header(skb); 2083 skb_reset_mac_len(skb); 2084 return flags & BPF_F_INGRESS ? 2085 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb); 2086 } 2087 2088 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev, 2089 u32 flags) 2090 { 2091 /* Verify that a link layer header is carried */ 2092 if (unlikely(skb->mac_header >= skb->network_header)) { 2093 kfree_skb(skb); 2094 return -ERANGE; 2095 } 2096 2097 bpf_push_mac_rcsum(skb); 2098 return flags & BPF_F_INGRESS ? 2099 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb); 2100 } 2101 2102 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev, 2103 u32 flags) 2104 { 2105 if (dev_is_mac_header_xmit(dev)) 2106 return __bpf_redirect_common(skb, dev, flags); 2107 else 2108 return __bpf_redirect_no_mac(skb, dev, flags); 2109 } 2110 2111 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags) 2112 { 2113 struct net_device *dev; 2114 struct sk_buff *clone; 2115 int ret; 2116 2117 if (unlikely(flags & ~(BPF_F_INGRESS))) 2118 return -EINVAL; 2119 2120 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex); 2121 if (unlikely(!dev)) 2122 return -EINVAL; 2123 2124 clone = skb_clone(skb, GFP_ATOMIC); 2125 if (unlikely(!clone)) 2126 return -ENOMEM; 2127 2128 /* For direct write, we need to keep the invariant that the skbs 2129 * we're dealing with need to be uncloned. Should uncloning fail 2130 * here, we need to free the just generated clone to unclone once 2131 * again. 2132 */ 2133 ret = bpf_try_make_head_writable(skb); 2134 if (unlikely(ret)) { 2135 kfree_skb(clone); 2136 return -ENOMEM; 2137 } 2138 2139 return __bpf_redirect(clone, dev, flags); 2140 } 2141 2142 static const struct bpf_func_proto bpf_clone_redirect_proto = { 2143 .func = bpf_clone_redirect, 2144 .gpl_only = false, 2145 .ret_type = RET_INTEGER, 2146 .arg1_type = ARG_PTR_TO_CTX, 2147 .arg2_type = ARG_ANYTHING, 2148 .arg3_type = ARG_ANYTHING, 2149 }; 2150 2151 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 2152 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info); 2153 2154 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags) 2155 { 2156 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2157 2158 if (unlikely(flags & ~(BPF_F_INGRESS))) 2159 return TC_ACT_SHOT; 2160 2161 ri->flags = flags; 2162 ri->tgt_index = ifindex; 2163 2164 return TC_ACT_REDIRECT; 2165 } 2166 2167 int skb_do_redirect(struct sk_buff *skb) 2168 { 2169 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2170 struct net_device *dev; 2171 2172 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->tgt_index); 2173 ri->tgt_index = 0; 2174 if (unlikely(!dev)) { 2175 kfree_skb(skb); 2176 return -EINVAL; 2177 } 2178 2179 return __bpf_redirect(skb, dev, ri->flags); 2180 } 2181 2182 static const struct bpf_func_proto bpf_redirect_proto = { 2183 .func = bpf_redirect, 2184 .gpl_only = false, 2185 .ret_type = RET_INTEGER, 2186 .arg1_type = ARG_ANYTHING, 2187 .arg2_type = ARG_ANYTHING, 2188 }; 2189 2190 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes) 2191 { 2192 msg->apply_bytes = bytes; 2193 return 0; 2194 } 2195 2196 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = { 2197 .func = bpf_msg_apply_bytes, 2198 .gpl_only = false, 2199 .ret_type = RET_INTEGER, 2200 .arg1_type = ARG_PTR_TO_CTX, 2201 .arg2_type = ARG_ANYTHING, 2202 }; 2203 2204 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes) 2205 { 2206 msg->cork_bytes = bytes; 2207 return 0; 2208 } 2209 2210 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = { 2211 .func = bpf_msg_cork_bytes, 2212 .gpl_only = false, 2213 .ret_type = RET_INTEGER, 2214 .arg1_type = ARG_PTR_TO_CTX, 2215 .arg2_type = ARG_ANYTHING, 2216 }; 2217 2218 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start, 2219 u32, end, u64, flags) 2220 { 2221 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start; 2222 u32 first_sge, last_sge, i, shift, bytes_sg_total; 2223 struct scatterlist *sge; 2224 u8 *raw, *to, *from; 2225 struct page *page; 2226 2227 if (unlikely(flags || end <= start)) 2228 return -EINVAL; 2229 2230 /* First find the starting scatterlist element */ 2231 i = msg->sg.start; 2232 do { 2233 len = sk_msg_elem(msg, i)->length; 2234 if (start < offset + len) 2235 break; 2236 offset += len; 2237 sk_msg_iter_var_next(i); 2238 } while (i != msg->sg.end); 2239 2240 if (unlikely(start >= offset + len)) 2241 return -EINVAL; 2242 2243 first_sge = i; 2244 /* The start may point into the sg element so we need to also 2245 * account for the headroom. 2246 */ 2247 bytes_sg_total = start - offset + bytes; 2248 if (!msg->sg.copy[i] && bytes_sg_total <= len) 2249 goto out; 2250 2251 /* At this point we need to linearize multiple scatterlist 2252 * elements or a single shared page. Either way we need to 2253 * copy into a linear buffer exclusively owned by BPF. Then 2254 * place the buffer in the scatterlist and fixup the original 2255 * entries by removing the entries now in the linear buffer 2256 * and shifting the remaining entries. For now we do not try 2257 * to copy partial entries to avoid complexity of running out 2258 * of sg_entry slots. The downside is reading a single byte 2259 * will copy the entire sg entry. 2260 */ 2261 do { 2262 copy += sk_msg_elem(msg, i)->length; 2263 sk_msg_iter_var_next(i); 2264 if (bytes_sg_total <= copy) 2265 break; 2266 } while (i != msg->sg.end); 2267 last_sge = i; 2268 2269 if (unlikely(bytes_sg_total > copy)) 2270 return -EINVAL; 2271 2272 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2273 get_order(copy)); 2274 if (unlikely(!page)) 2275 return -ENOMEM; 2276 2277 raw = page_address(page); 2278 i = first_sge; 2279 do { 2280 sge = sk_msg_elem(msg, i); 2281 from = sg_virt(sge); 2282 len = sge->length; 2283 to = raw + poffset; 2284 2285 memcpy(to, from, len); 2286 poffset += len; 2287 sge->length = 0; 2288 put_page(sg_page(sge)); 2289 2290 sk_msg_iter_var_next(i); 2291 } while (i != last_sge); 2292 2293 sg_set_page(&msg->sg.data[first_sge], page, copy, 0); 2294 2295 /* To repair sg ring we need to shift entries. If we only 2296 * had a single entry though we can just replace it and 2297 * be done. Otherwise walk the ring and shift the entries. 2298 */ 2299 WARN_ON_ONCE(last_sge == first_sge); 2300 shift = last_sge > first_sge ? 2301 last_sge - first_sge - 1 : 2302 MAX_SKB_FRAGS - first_sge + last_sge - 1; 2303 if (!shift) 2304 goto out; 2305 2306 i = first_sge; 2307 sk_msg_iter_var_next(i); 2308 do { 2309 u32 move_from; 2310 2311 if (i + shift >= MAX_MSG_FRAGS) 2312 move_from = i + shift - MAX_MSG_FRAGS; 2313 else 2314 move_from = i + shift; 2315 if (move_from == msg->sg.end) 2316 break; 2317 2318 msg->sg.data[i] = msg->sg.data[move_from]; 2319 msg->sg.data[move_from].length = 0; 2320 msg->sg.data[move_from].page_link = 0; 2321 msg->sg.data[move_from].offset = 0; 2322 sk_msg_iter_var_next(i); 2323 } while (1); 2324 2325 msg->sg.end = msg->sg.end - shift > msg->sg.end ? 2326 msg->sg.end - shift + MAX_MSG_FRAGS : 2327 msg->sg.end - shift; 2328 out: 2329 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset; 2330 msg->data_end = msg->data + bytes; 2331 return 0; 2332 } 2333 2334 static const struct bpf_func_proto bpf_msg_pull_data_proto = { 2335 .func = bpf_msg_pull_data, 2336 .gpl_only = false, 2337 .ret_type = RET_INTEGER, 2338 .arg1_type = ARG_PTR_TO_CTX, 2339 .arg2_type = ARG_ANYTHING, 2340 .arg3_type = ARG_ANYTHING, 2341 .arg4_type = ARG_ANYTHING, 2342 }; 2343 2344 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start, 2345 u32, len, u64, flags) 2346 { 2347 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge; 2348 u32 new, i = 0, l, space, copy = 0, offset = 0; 2349 u8 *raw, *to, *from; 2350 struct page *page; 2351 2352 if (unlikely(flags)) 2353 return -EINVAL; 2354 2355 /* First find the starting scatterlist element */ 2356 i = msg->sg.start; 2357 do { 2358 l = sk_msg_elem(msg, i)->length; 2359 2360 if (start < offset + l) 2361 break; 2362 offset += l; 2363 sk_msg_iter_var_next(i); 2364 } while (i != msg->sg.end); 2365 2366 if (start >= offset + l) 2367 return -EINVAL; 2368 2369 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2370 2371 /* If no space available will fallback to copy, we need at 2372 * least one scatterlist elem available to push data into 2373 * when start aligns to the beginning of an element or two 2374 * when it falls inside an element. We handle the start equals 2375 * offset case because its the common case for inserting a 2376 * header. 2377 */ 2378 if (!space || (space == 1 && start != offset)) 2379 copy = msg->sg.data[i].length; 2380 2381 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2382 get_order(copy + len)); 2383 if (unlikely(!page)) 2384 return -ENOMEM; 2385 2386 if (copy) { 2387 int front, back; 2388 2389 raw = page_address(page); 2390 2391 psge = sk_msg_elem(msg, i); 2392 front = start - offset; 2393 back = psge->length - front; 2394 from = sg_virt(psge); 2395 2396 if (front) 2397 memcpy(raw, from, front); 2398 2399 if (back) { 2400 from += front; 2401 to = raw + front + len; 2402 2403 memcpy(to, from, back); 2404 } 2405 2406 put_page(sg_page(psge)); 2407 } else if (start - offset) { 2408 psge = sk_msg_elem(msg, i); 2409 rsge = sk_msg_elem_cpy(msg, i); 2410 2411 psge->length = start - offset; 2412 rsge.length -= psge->length; 2413 rsge.offset += start; 2414 2415 sk_msg_iter_var_next(i); 2416 sg_unmark_end(psge); 2417 sk_msg_iter_next(msg, end); 2418 } 2419 2420 /* Slot(s) to place newly allocated data */ 2421 new = i; 2422 2423 /* Shift one or two slots as needed */ 2424 if (!copy) { 2425 sge = sk_msg_elem_cpy(msg, i); 2426 2427 sk_msg_iter_var_next(i); 2428 sg_unmark_end(&sge); 2429 sk_msg_iter_next(msg, end); 2430 2431 nsge = sk_msg_elem_cpy(msg, i); 2432 if (rsge.length) { 2433 sk_msg_iter_var_next(i); 2434 nnsge = sk_msg_elem_cpy(msg, i); 2435 } 2436 2437 while (i != msg->sg.end) { 2438 msg->sg.data[i] = sge; 2439 sge = nsge; 2440 sk_msg_iter_var_next(i); 2441 if (rsge.length) { 2442 nsge = nnsge; 2443 nnsge = sk_msg_elem_cpy(msg, i); 2444 } else { 2445 nsge = sk_msg_elem_cpy(msg, i); 2446 } 2447 } 2448 } 2449 2450 /* Place newly allocated data buffer */ 2451 sk_mem_charge(msg->sk, len); 2452 msg->sg.size += len; 2453 msg->sg.copy[new] = false; 2454 sg_set_page(&msg->sg.data[new], page, len + copy, 0); 2455 if (rsge.length) { 2456 get_page(sg_page(&rsge)); 2457 sk_msg_iter_var_next(new); 2458 msg->sg.data[new] = rsge; 2459 } 2460 2461 sk_msg_compute_data_pointers(msg); 2462 return 0; 2463 } 2464 2465 static const struct bpf_func_proto bpf_msg_push_data_proto = { 2466 .func = bpf_msg_push_data, 2467 .gpl_only = false, 2468 .ret_type = RET_INTEGER, 2469 .arg1_type = ARG_PTR_TO_CTX, 2470 .arg2_type = ARG_ANYTHING, 2471 .arg3_type = ARG_ANYTHING, 2472 .arg4_type = ARG_ANYTHING, 2473 }; 2474 2475 static void sk_msg_shift_left(struct sk_msg *msg, int i) 2476 { 2477 int prev; 2478 2479 do { 2480 prev = i; 2481 sk_msg_iter_var_next(i); 2482 msg->sg.data[prev] = msg->sg.data[i]; 2483 } while (i != msg->sg.end); 2484 2485 sk_msg_iter_prev(msg, end); 2486 } 2487 2488 static void sk_msg_shift_right(struct sk_msg *msg, int i) 2489 { 2490 struct scatterlist tmp, sge; 2491 2492 sk_msg_iter_next(msg, end); 2493 sge = sk_msg_elem_cpy(msg, i); 2494 sk_msg_iter_var_next(i); 2495 tmp = sk_msg_elem_cpy(msg, i); 2496 2497 while (i != msg->sg.end) { 2498 msg->sg.data[i] = sge; 2499 sk_msg_iter_var_next(i); 2500 sge = tmp; 2501 tmp = sk_msg_elem_cpy(msg, i); 2502 } 2503 } 2504 2505 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start, 2506 u32, len, u64, flags) 2507 { 2508 u32 i = 0, l, space, offset = 0; 2509 u64 last = start + len; 2510 int pop; 2511 2512 if (unlikely(flags)) 2513 return -EINVAL; 2514 2515 /* First find the starting scatterlist element */ 2516 i = msg->sg.start; 2517 do { 2518 l = sk_msg_elem(msg, i)->length; 2519 2520 if (start < offset + l) 2521 break; 2522 offset += l; 2523 sk_msg_iter_var_next(i); 2524 } while (i != msg->sg.end); 2525 2526 /* Bounds checks: start and pop must be inside message */ 2527 if (start >= offset + l || last >= msg->sg.size) 2528 return -EINVAL; 2529 2530 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2531 2532 pop = len; 2533 /* --------------| offset 2534 * -| start |-------- len -------| 2535 * 2536 * |----- a ----|-------- pop -------|----- b ----| 2537 * |______________________________________________| length 2538 * 2539 * 2540 * a: region at front of scatter element to save 2541 * b: region at back of scatter element to save when length > A + pop 2542 * pop: region to pop from element, same as input 'pop' here will be 2543 * decremented below per iteration. 2544 * 2545 * Two top-level cases to handle when start != offset, first B is non 2546 * zero and second B is zero corresponding to when a pop includes more 2547 * than one element. 2548 * 2549 * Then if B is non-zero AND there is no space allocate space and 2550 * compact A, B regions into page. If there is space shift ring to 2551 * the rigth free'ing the next element in ring to place B, leaving 2552 * A untouched except to reduce length. 2553 */ 2554 if (start != offset) { 2555 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i); 2556 int a = start; 2557 int b = sge->length - pop - a; 2558 2559 sk_msg_iter_var_next(i); 2560 2561 if (pop < sge->length - a) { 2562 if (space) { 2563 sge->length = a; 2564 sk_msg_shift_right(msg, i); 2565 nsge = sk_msg_elem(msg, i); 2566 get_page(sg_page(sge)); 2567 sg_set_page(nsge, 2568 sg_page(sge), 2569 b, sge->offset + pop + a); 2570 } else { 2571 struct page *page, *orig; 2572 u8 *to, *from; 2573 2574 page = alloc_pages(__GFP_NOWARN | 2575 __GFP_COMP | GFP_ATOMIC, 2576 get_order(a + b)); 2577 if (unlikely(!page)) 2578 return -ENOMEM; 2579 2580 sge->length = a; 2581 orig = sg_page(sge); 2582 from = sg_virt(sge); 2583 to = page_address(page); 2584 memcpy(to, from, a); 2585 memcpy(to + a, from + a + pop, b); 2586 sg_set_page(sge, page, a + b, 0); 2587 put_page(orig); 2588 } 2589 pop = 0; 2590 } else if (pop >= sge->length - a) { 2591 sge->length = a; 2592 pop -= (sge->length - a); 2593 } 2594 } 2595 2596 /* From above the current layout _must_ be as follows, 2597 * 2598 * -| offset 2599 * -| start 2600 * 2601 * |---- pop ---|---------------- b ------------| 2602 * |____________________________________________| length 2603 * 2604 * Offset and start of the current msg elem are equal because in the 2605 * previous case we handled offset != start and either consumed the 2606 * entire element and advanced to the next element OR pop == 0. 2607 * 2608 * Two cases to handle here are first pop is less than the length 2609 * leaving some remainder b above. Simply adjust the element's layout 2610 * in this case. Or pop >= length of the element so that b = 0. In this 2611 * case advance to next element decrementing pop. 2612 */ 2613 while (pop) { 2614 struct scatterlist *sge = sk_msg_elem(msg, i); 2615 2616 if (pop < sge->length) { 2617 sge->length -= pop; 2618 sge->offset += pop; 2619 pop = 0; 2620 } else { 2621 pop -= sge->length; 2622 sk_msg_shift_left(msg, i); 2623 } 2624 sk_msg_iter_var_next(i); 2625 } 2626 2627 sk_mem_uncharge(msg->sk, len - pop); 2628 msg->sg.size -= (len - pop); 2629 sk_msg_compute_data_pointers(msg); 2630 return 0; 2631 } 2632 2633 static const struct bpf_func_proto bpf_msg_pop_data_proto = { 2634 .func = bpf_msg_pop_data, 2635 .gpl_only = false, 2636 .ret_type = RET_INTEGER, 2637 .arg1_type = ARG_PTR_TO_CTX, 2638 .arg2_type = ARG_ANYTHING, 2639 .arg3_type = ARG_ANYTHING, 2640 .arg4_type = ARG_ANYTHING, 2641 }; 2642 2643 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb) 2644 { 2645 return task_get_classid(skb); 2646 } 2647 2648 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = { 2649 .func = bpf_get_cgroup_classid, 2650 .gpl_only = false, 2651 .ret_type = RET_INTEGER, 2652 .arg1_type = ARG_PTR_TO_CTX, 2653 }; 2654 2655 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb) 2656 { 2657 return dst_tclassid(skb); 2658 } 2659 2660 static const struct bpf_func_proto bpf_get_route_realm_proto = { 2661 .func = bpf_get_route_realm, 2662 .gpl_only = false, 2663 .ret_type = RET_INTEGER, 2664 .arg1_type = ARG_PTR_TO_CTX, 2665 }; 2666 2667 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb) 2668 { 2669 /* If skb_clear_hash() was called due to mangling, we can 2670 * trigger SW recalculation here. Later access to hash 2671 * can then use the inline skb->hash via context directly 2672 * instead of calling this helper again. 2673 */ 2674 return skb_get_hash(skb); 2675 } 2676 2677 static const struct bpf_func_proto bpf_get_hash_recalc_proto = { 2678 .func = bpf_get_hash_recalc, 2679 .gpl_only = false, 2680 .ret_type = RET_INTEGER, 2681 .arg1_type = ARG_PTR_TO_CTX, 2682 }; 2683 2684 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb) 2685 { 2686 /* After all direct packet write, this can be used once for 2687 * triggering a lazy recalc on next skb_get_hash() invocation. 2688 */ 2689 skb_clear_hash(skb); 2690 return 0; 2691 } 2692 2693 static const struct bpf_func_proto bpf_set_hash_invalid_proto = { 2694 .func = bpf_set_hash_invalid, 2695 .gpl_only = false, 2696 .ret_type = RET_INTEGER, 2697 .arg1_type = ARG_PTR_TO_CTX, 2698 }; 2699 2700 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash) 2701 { 2702 /* Set user specified hash as L4(+), so that it gets returned 2703 * on skb_get_hash() call unless BPF prog later on triggers a 2704 * skb_clear_hash(). 2705 */ 2706 __skb_set_sw_hash(skb, hash, true); 2707 return 0; 2708 } 2709 2710 static const struct bpf_func_proto bpf_set_hash_proto = { 2711 .func = bpf_set_hash, 2712 .gpl_only = false, 2713 .ret_type = RET_INTEGER, 2714 .arg1_type = ARG_PTR_TO_CTX, 2715 .arg2_type = ARG_ANYTHING, 2716 }; 2717 2718 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto, 2719 u16, vlan_tci) 2720 { 2721 int ret; 2722 2723 if (unlikely(vlan_proto != htons(ETH_P_8021Q) && 2724 vlan_proto != htons(ETH_P_8021AD))) 2725 vlan_proto = htons(ETH_P_8021Q); 2726 2727 bpf_push_mac_rcsum(skb); 2728 ret = skb_vlan_push(skb, vlan_proto, vlan_tci); 2729 bpf_pull_mac_rcsum(skb); 2730 2731 bpf_compute_data_pointers(skb); 2732 return ret; 2733 } 2734 2735 static const struct bpf_func_proto bpf_skb_vlan_push_proto = { 2736 .func = bpf_skb_vlan_push, 2737 .gpl_only = false, 2738 .ret_type = RET_INTEGER, 2739 .arg1_type = ARG_PTR_TO_CTX, 2740 .arg2_type = ARG_ANYTHING, 2741 .arg3_type = ARG_ANYTHING, 2742 }; 2743 2744 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb) 2745 { 2746 int ret; 2747 2748 bpf_push_mac_rcsum(skb); 2749 ret = skb_vlan_pop(skb); 2750 bpf_pull_mac_rcsum(skb); 2751 2752 bpf_compute_data_pointers(skb); 2753 return ret; 2754 } 2755 2756 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = { 2757 .func = bpf_skb_vlan_pop, 2758 .gpl_only = false, 2759 .ret_type = RET_INTEGER, 2760 .arg1_type = ARG_PTR_TO_CTX, 2761 }; 2762 2763 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len) 2764 { 2765 /* Caller already did skb_cow() with len as headroom, 2766 * so no need to do it here. 2767 */ 2768 skb_push(skb, len); 2769 memmove(skb->data, skb->data + len, off); 2770 memset(skb->data + off, 0, len); 2771 2772 /* No skb_postpush_rcsum(skb, skb->data + off, len) 2773 * needed here as it does not change the skb->csum 2774 * result for checksum complete when summing over 2775 * zeroed blocks. 2776 */ 2777 return 0; 2778 } 2779 2780 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len) 2781 { 2782 /* skb_ensure_writable() is not needed here, as we're 2783 * already working on an uncloned skb. 2784 */ 2785 if (unlikely(!pskb_may_pull(skb, off + len))) 2786 return -ENOMEM; 2787 2788 skb_postpull_rcsum(skb, skb->data + off, len); 2789 memmove(skb->data + len, skb->data, off); 2790 __skb_pull(skb, len); 2791 2792 return 0; 2793 } 2794 2795 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len) 2796 { 2797 bool trans_same = skb->transport_header == skb->network_header; 2798 int ret; 2799 2800 /* There's no need for __skb_push()/__skb_pull() pair to 2801 * get to the start of the mac header as we're guaranteed 2802 * to always start from here under eBPF. 2803 */ 2804 ret = bpf_skb_generic_push(skb, off, len); 2805 if (likely(!ret)) { 2806 skb->mac_header -= len; 2807 skb->network_header -= len; 2808 if (trans_same) 2809 skb->transport_header = skb->network_header; 2810 } 2811 2812 return ret; 2813 } 2814 2815 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len) 2816 { 2817 bool trans_same = skb->transport_header == skb->network_header; 2818 int ret; 2819 2820 /* Same here, __skb_push()/__skb_pull() pair not needed. */ 2821 ret = bpf_skb_generic_pop(skb, off, len); 2822 if (likely(!ret)) { 2823 skb->mac_header += len; 2824 skb->network_header += len; 2825 if (trans_same) 2826 skb->transport_header = skb->network_header; 2827 } 2828 2829 return ret; 2830 } 2831 2832 static int bpf_skb_proto_4_to_6(struct sk_buff *skb) 2833 { 2834 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 2835 u32 off = skb_mac_header_len(skb); 2836 int ret; 2837 2838 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) 2839 return -ENOTSUPP; 2840 2841 ret = skb_cow(skb, len_diff); 2842 if (unlikely(ret < 0)) 2843 return ret; 2844 2845 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 2846 if (unlikely(ret < 0)) 2847 return ret; 2848 2849 if (skb_is_gso(skb)) { 2850 struct skb_shared_info *shinfo = skb_shinfo(skb); 2851 2852 /* SKB_GSO_TCPV4 needs to be changed into 2853 * SKB_GSO_TCPV6. 2854 */ 2855 if (shinfo->gso_type & SKB_GSO_TCPV4) { 2856 shinfo->gso_type &= ~SKB_GSO_TCPV4; 2857 shinfo->gso_type |= SKB_GSO_TCPV6; 2858 } 2859 2860 /* Due to IPv6 header, MSS needs to be downgraded. */ 2861 skb_decrease_gso_size(shinfo, len_diff); 2862 /* Header must be checked, and gso_segs recomputed. */ 2863 shinfo->gso_type |= SKB_GSO_DODGY; 2864 shinfo->gso_segs = 0; 2865 } 2866 2867 skb->protocol = htons(ETH_P_IPV6); 2868 skb_clear_hash(skb); 2869 2870 return 0; 2871 } 2872 2873 static int bpf_skb_proto_6_to_4(struct sk_buff *skb) 2874 { 2875 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 2876 u32 off = skb_mac_header_len(skb); 2877 int ret; 2878 2879 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) 2880 return -ENOTSUPP; 2881 2882 ret = skb_unclone(skb, GFP_ATOMIC); 2883 if (unlikely(ret < 0)) 2884 return ret; 2885 2886 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 2887 if (unlikely(ret < 0)) 2888 return ret; 2889 2890 if (skb_is_gso(skb)) { 2891 struct skb_shared_info *shinfo = skb_shinfo(skb); 2892 2893 /* SKB_GSO_TCPV6 needs to be changed into 2894 * SKB_GSO_TCPV4. 2895 */ 2896 if (shinfo->gso_type & SKB_GSO_TCPV6) { 2897 shinfo->gso_type &= ~SKB_GSO_TCPV6; 2898 shinfo->gso_type |= SKB_GSO_TCPV4; 2899 } 2900 2901 /* Due to IPv4 header, MSS can be upgraded. */ 2902 skb_increase_gso_size(shinfo, len_diff); 2903 /* Header must be checked, and gso_segs recomputed. */ 2904 shinfo->gso_type |= SKB_GSO_DODGY; 2905 shinfo->gso_segs = 0; 2906 } 2907 2908 skb->protocol = htons(ETH_P_IP); 2909 skb_clear_hash(skb); 2910 2911 return 0; 2912 } 2913 2914 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto) 2915 { 2916 __be16 from_proto = skb->protocol; 2917 2918 if (from_proto == htons(ETH_P_IP) && 2919 to_proto == htons(ETH_P_IPV6)) 2920 return bpf_skb_proto_4_to_6(skb); 2921 2922 if (from_proto == htons(ETH_P_IPV6) && 2923 to_proto == htons(ETH_P_IP)) 2924 return bpf_skb_proto_6_to_4(skb); 2925 2926 return -ENOTSUPP; 2927 } 2928 2929 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto, 2930 u64, flags) 2931 { 2932 int ret; 2933 2934 if (unlikely(flags)) 2935 return -EINVAL; 2936 2937 /* General idea is that this helper does the basic groundwork 2938 * needed for changing the protocol, and eBPF program fills the 2939 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace() 2940 * and other helpers, rather than passing a raw buffer here. 2941 * 2942 * The rationale is to keep this minimal and without a need to 2943 * deal with raw packet data. F.e. even if we would pass buffers 2944 * here, the program still needs to call the bpf_lX_csum_replace() 2945 * helpers anyway. Plus, this way we keep also separation of 2946 * concerns, since f.e. bpf_skb_store_bytes() should only take 2947 * care of stores. 2948 * 2949 * Currently, additional options and extension header space are 2950 * not supported, but flags register is reserved so we can adapt 2951 * that. For offloads, we mark packet as dodgy, so that headers 2952 * need to be verified first. 2953 */ 2954 ret = bpf_skb_proto_xlat(skb, proto); 2955 bpf_compute_data_pointers(skb); 2956 return ret; 2957 } 2958 2959 static const struct bpf_func_proto bpf_skb_change_proto_proto = { 2960 .func = bpf_skb_change_proto, 2961 .gpl_only = false, 2962 .ret_type = RET_INTEGER, 2963 .arg1_type = ARG_PTR_TO_CTX, 2964 .arg2_type = ARG_ANYTHING, 2965 .arg3_type = ARG_ANYTHING, 2966 }; 2967 2968 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type) 2969 { 2970 /* We only allow a restricted subset to be changed for now. */ 2971 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) || 2972 !skb_pkt_type_ok(pkt_type))) 2973 return -EINVAL; 2974 2975 skb->pkt_type = pkt_type; 2976 return 0; 2977 } 2978 2979 static const struct bpf_func_proto bpf_skb_change_type_proto = { 2980 .func = bpf_skb_change_type, 2981 .gpl_only = false, 2982 .ret_type = RET_INTEGER, 2983 .arg1_type = ARG_PTR_TO_CTX, 2984 .arg2_type = ARG_ANYTHING, 2985 }; 2986 2987 static u32 bpf_skb_net_base_len(const struct sk_buff *skb) 2988 { 2989 switch (skb->protocol) { 2990 case htons(ETH_P_IP): 2991 return sizeof(struct iphdr); 2992 case htons(ETH_P_IPV6): 2993 return sizeof(struct ipv6hdr); 2994 default: 2995 return ~0U; 2996 } 2997 } 2998 2999 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \ 3000 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3001 3002 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \ 3003 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \ 3004 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \ 3005 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \ 3006 BPF_F_ADJ_ROOM_ENCAP_L2( \ 3007 BPF_ADJ_ROOM_ENCAP_L2_MASK)) 3008 3009 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff, 3010 u64 flags) 3011 { 3012 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT; 3013 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK; 3014 u16 mac_len = 0, inner_net = 0, inner_trans = 0; 3015 unsigned int gso_type = SKB_GSO_DODGY; 3016 int ret; 3017 3018 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3019 /* udp gso_size delineates datagrams, only allow if fixed */ 3020 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3021 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3022 return -ENOTSUPP; 3023 } 3024 3025 ret = skb_cow_head(skb, len_diff); 3026 if (unlikely(ret < 0)) 3027 return ret; 3028 3029 if (encap) { 3030 if (skb->protocol != htons(ETH_P_IP) && 3031 skb->protocol != htons(ETH_P_IPV6)) 3032 return -ENOTSUPP; 3033 3034 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 && 3035 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3036 return -EINVAL; 3037 3038 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE && 3039 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3040 return -EINVAL; 3041 3042 if (skb->encapsulation) 3043 return -EALREADY; 3044 3045 mac_len = skb->network_header - skb->mac_header; 3046 inner_net = skb->network_header; 3047 if (inner_mac_len > len_diff) 3048 return -EINVAL; 3049 inner_trans = skb->transport_header; 3050 } 3051 3052 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3053 if (unlikely(ret < 0)) 3054 return ret; 3055 3056 if (encap) { 3057 skb->inner_mac_header = inner_net - inner_mac_len; 3058 skb->inner_network_header = inner_net; 3059 skb->inner_transport_header = inner_trans; 3060 skb_set_inner_protocol(skb, skb->protocol); 3061 3062 skb->encapsulation = 1; 3063 skb_set_network_header(skb, mac_len); 3064 3065 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3066 gso_type |= SKB_GSO_UDP_TUNNEL; 3067 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE) 3068 gso_type |= SKB_GSO_GRE; 3069 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3070 gso_type |= SKB_GSO_IPXIP6; 3071 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3072 gso_type |= SKB_GSO_IPXIP4; 3073 3074 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE || 3075 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) { 3076 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ? 3077 sizeof(struct ipv6hdr) : 3078 sizeof(struct iphdr); 3079 3080 skb_set_transport_header(skb, mac_len + nh_len); 3081 } 3082 3083 /* Match skb->protocol to new outer l3 protocol */ 3084 if (skb->protocol == htons(ETH_P_IP) && 3085 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3086 skb->protocol = htons(ETH_P_IPV6); 3087 else if (skb->protocol == htons(ETH_P_IPV6) && 3088 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3089 skb->protocol = htons(ETH_P_IP); 3090 } 3091 3092 if (skb_is_gso(skb)) { 3093 struct skb_shared_info *shinfo = skb_shinfo(skb); 3094 3095 /* Due to header grow, MSS needs to be downgraded. */ 3096 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3097 skb_decrease_gso_size(shinfo, len_diff); 3098 3099 /* Header must be checked, and gso_segs recomputed. */ 3100 shinfo->gso_type |= gso_type; 3101 shinfo->gso_segs = 0; 3102 } 3103 3104 return 0; 3105 } 3106 3107 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff, 3108 u64 flags) 3109 { 3110 int ret; 3111 3112 if (flags & ~BPF_F_ADJ_ROOM_FIXED_GSO) 3113 return -EINVAL; 3114 3115 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3116 /* udp gso_size delineates datagrams, only allow if fixed */ 3117 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3118 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3119 return -ENOTSUPP; 3120 } 3121 3122 ret = skb_unclone(skb, GFP_ATOMIC); 3123 if (unlikely(ret < 0)) 3124 return ret; 3125 3126 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3127 if (unlikely(ret < 0)) 3128 return ret; 3129 3130 if (skb_is_gso(skb)) { 3131 struct skb_shared_info *shinfo = skb_shinfo(skb); 3132 3133 /* Due to header shrink, MSS can be upgraded. */ 3134 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3135 skb_increase_gso_size(shinfo, len_diff); 3136 3137 /* Header must be checked, and gso_segs recomputed. */ 3138 shinfo->gso_type |= SKB_GSO_DODGY; 3139 shinfo->gso_segs = 0; 3140 } 3141 3142 return 0; 3143 } 3144 3145 static u32 __bpf_skb_max_len(const struct sk_buff *skb) 3146 { 3147 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len : 3148 SKB_MAX_ALLOC; 3149 } 3150 3151 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3152 u32, mode, u64, flags) 3153 { 3154 u32 len_cur, len_diff_abs = abs(len_diff); 3155 u32 len_min = bpf_skb_net_base_len(skb); 3156 u32 len_max = __bpf_skb_max_len(skb); 3157 __be16 proto = skb->protocol; 3158 bool shrink = len_diff < 0; 3159 u32 off; 3160 int ret; 3161 3162 if (unlikely(flags & ~BPF_F_ADJ_ROOM_MASK)) 3163 return -EINVAL; 3164 if (unlikely(len_diff_abs > 0xfffU)) 3165 return -EFAULT; 3166 if (unlikely(proto != htons(ETH_P_IP) && 3167 proto != htons(ETH_P_IPV6))) 3168 return -ENOTSUPP; 3169 3170 off = skb_mac_header_len(skb); 3171 switch (mode) { 3172 case BPF_ADJ_ROOM_NET: 3173 off += bpf_skb_net_base_len(skb); 3174 break; 3175 case BPF_ADJ_ROOM_MAC: 3176 break; 3177 default: 3178 return -ENOTSUPP; 3179 } 3180 3181 len_cur = skb->len - skb_network_offset(skb); 3182 if ((shrink && (len_diff_abs >= len_cur || 3183 len_cur - len_diff_abs < len_min)) || 3184 (!shrink && (skb->len + len_diff_abs > len_max && 3185 !skb_is_gso(skb)))) 3186 return -ENOTSUPP; 3187 3188 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) : 3189 bpf_skb_net_grow(skb, off, len_diff_abs, flags); 3190 3191 bpf_compute_data_pointers(skb); 3192 return ret; 3193 } 3194 3195 static const struct bpf_func_proto bpf_skb_adjust_room_proto = { 3196 .func = bpf_skb_adjust_room, 3197 .gpl_only = false, 3198 .ret_type = RET_INTEGER, 3199 .arg1_type = ARG_PTR_TO_CTX, 3200 .arg2_type = ARG_ANYTHING, 3201 .arg3_type = ARG_ANYTHING, 3202 .arg4_type = ARG_ANYTHING, 3203 }; 3204 3205 static u32 __bpf_skb_min_len(const struct sk_buff *skb) 3206 { 3207 u32 min_len = skb_network_offset(skb); 3208 3209 if (skb_transport_header_was_set(skb)) 3210 min_len = skb_transport_offset(skb); 3211 if (skb->ip_summed == CHECKSUM_PARTIAL) 3212 min_len = skb_checksum_start_offset(skb) + 3213 skb->csum_offset + sizeof(__sum16); 3214 return min_len; 3215 } 3216 3217 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len) 3218 { 3219 unsigned int old_len = skb->len; 3220 int ret; 3221 3222 ret = __skb_grow_rcsum(skb, new_len); 3223 if (!ret) 3224 memset(skb->data + old_len, 0, new_len - old_len); 3225 return ret; 3226 } 3227 3228 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len) 3229 { 3230 return __skb_trim_rcsum(skb, new_len); 3231 } 3232 3233 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len, 3234 u64 flags) 3235 { 3236 u32 max_len = __bpf_skb_max_len(skb); 3237 u32 min_len = __bpf_skb_min_len(skb); 3238 int ret; 3239 3240 if (unlikely(flags || new_len > max_len || new_len < min_len)) 3241 return -EINVAL; 3242 if (skb->encapsulation) 3243 return -ENOTSUPP; 3244 3245 /* The basic idea of this helper is that it's performing the 3246 * needed work to either grow or trim an skb, and eBPF program 3247 * rewrites the rest via helpers like bpf_skb_store_bytes(), 3248 * bpf_lX_csum_replace() and others rather than passing a raw 3249 * buffer here. This one is a slow path helper and intended 3250 * for replies with control messages. 3251 * 3252 * Like in bpf_skb_change_proto(), we want to keep this rather 3253 * minimal and without protocol specifics so that we are able 3254 * to separate concerns as in bpf_skb_store_bytes() should only 3255 * be the one responsible for writing buffers. 3256 * 3257 * It's really expected to be a slow path operation here for 3258 * control message replies, so we're implicitly linearizing, 3259 * uncloning and drop offloads from the skb by this. 3260 */ 3261 ret = __bpf_try_make_writable(skb, skb->len); 3262 if (!ret) { 3263 if (new_len > skb->len) 3264 ret = bpf_skb_grow_rcsum(skb, new_len); 3265 else if (new_len < skb->len) 3266 ret = bpf_skb_trim_rcsum(skb, new_len); 3267 if (!ret && skb_is_gso(skb)) 3268 skb_gso_reset(skb); 3269 } 3270 return ret; 3271 } 3272 3273 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3274 u64, flags) 3275 { 3276 int ret = __bpf_skb_change_tail(skb, new_len, flags); 3277 3278 bpf_compute_data_pointers(skb); 3279 return ret; 3280 } 3281 3282 static const struct bpf_func_proto bpf_skb_change_tail_proto = { 3283 .func = bpf_skb_change_tail, 3284 .gpl_only = false, 3285 .ret_type = RET_INTEGER, 3286 .arg1_type = ARG_PTR_TO_CTX, 3287 .arg2_type = ARG_ANYTHING, 3288 .arg3_type = ARG_ANYTHING, 3289 }; 3290 3291 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3292 u64, flags) 3293 { 3294 int ret = __bpf_skb_change_tail(skb, new_len, flags); 3295 3296 bpf_compute_data_end_sk_skb(skb); 3297 return ret; 3298 } 3299 3300 static const struct bpf_func_proto sk_skb_change_tail_proto = { 3301 .func = sk_skb_change_tail, 3302 .gpl_only = false, 3303 .ret_type = RET_INTEGER, 3304 .arg1_type = ARG_PTR_TO_CTX, 3305 .arg2_type = ARG_ANYTHING, 3306 .arg3_type = ARG_ANYTHING, 3307 }; 3308 3309 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room, 3310 u64 flags) 3311 { 3312 u32 max_len = __bpf_skb_max_len(skb); 3313 u32 new_len = skb->len + head_room; 3314 int ret; 3315 3316 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) || 3317 new_len < skb->len)) 3318 return -EINVAL; 3319 3320 ret = skb_cow(skb, head_room); 3321 if (likely(!ret)) { 3322 /* Idea for this helper is that we currently only 3323 * allow to expand on mac header. This means that 3324 * skb->protocol network header, etc, stay as is. 3325 * Compared to bpf_skb_change_tail(), we're more 3326 * flexible due to not needing to linearize or 3327 * reset GSO. Intention for this helper is to be 3328 * used by an L3 skb that needs to push mac header 3329 * for redirection into L2 device. 3330 */ 3331 __skb_push(skb, head_room); 3332 memset(skb->data, 0, head_room); 3333 skb_reset_mac_header(skb); 3334 } 3335 3336 return ret; 3337 } 3338 3339 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room, 3340 u64, flags) 3341 { 3342 int ret = __bpf_skb_change_head(skb, head_room, flags); 3343 3344 bpf_compute_data_pointers(skb); 3345 return ret; 3346 } 3347 3348 static const struct bpf_func_proto bpf_skb_change_head_proto = { 3349 .func = bpf_skb_change_head, 3350 .gpl_only = false, 3351 .ret_type = RET_INTEGER, 3352 .arg1_type = ARG_PTR_TO_CTX, 3353 .arg2_type = ARG_ANYTHING, 3354 .arg3_type = ARG_ANYTHING, 3355 }; 3356 3357 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room, 3358 u64, flags) 3359 { 3360 int ret = __bpf_skb_change_head(skb, head_room, flags); 3361 3362 bpf_compute_data_end_sk_skb(skb); 3363 return ret; 3364 } 3365 3366 static const struct bpf_func_proto sk_skb_change_head_proto = { 3367 .func = sk_skb_change_head, 3368 .gpl_only = false, 3369 .ret_type = RET_INTEGER, 3370 .arg1_type = ARG_PTR_TO_CTX, 3371 .arg2_type = ARG_ANYTHING, 3372 .arg3_type = ARG_ANYTHING, 3373 }; 3374 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp) 3375 { 3376 return xdp_data_meta_unsupported(xdp) ? 0 : 3377 xdp->data - xdp->data_meta; 3378 } 3379 3380 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset) 3381 { 3382 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 3383 unsigned long metalen = xdp_get_metalen(xdp); 3384 void *data_start = xdp_frame_end + metalen; 3385 void *data = xdp->data + offset; 3386 3387 if (unlikely(data < data_start || 3388 data > xdp->data_end - ETH_HLEN)) 3389 return -EINVAL; 3390 3391 if (metalen) 3392 memmove(xdp->data_meta + offset, 3393 xdp->data_meta, metalen); 3394 xdp->data_meta += offset; 3395 xdp->data = data; 3396 3397 return 0; 3398 } 3399 3400 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = { 3401 .func = bpf_xdp_adjust_head, 3402 .gpl_only = false, 3403 .ret_type = RET_INTEGER, 3404 .arg1_type = ARG_PTR_TO_CTX, 3405 .arg2_type = ARG_ANYTHING, 3406 }; 3407 3408 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset) 3409 { 3410 void *data_end = xdp->data_end + offset; 3411 3412 /* only shrinking is allowed for now. */ 3413 if (unlikely(offset >= 0)) 3414 return -EINVAL; 3415 3416 if (unlikely(data_end < xdp->data + ETH_HLEN)) 3417 return -EINVAL; 3418 3419 xdp->data_end = data_end; 3420 3421 return 0; 3422 } 3423 3424 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = { 3425 .func = bpf_xdp_adjust_tail, 3426 .gpl_only = false, 3427 .ret_type = RET_INTEGER, 3428 .arg1_type = ARG_PTR_TO_CTX, 3429 .arg2_type = ARG_ANYTHING, 3430 }; 3431 3432 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset) 3433 { 3434 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 3435 void *meta = xdp->data_meta + offset; 3436 unsigned long metalen = xdp->data - meta; 3437 3438 if (xdp_data_meta_unsupported(xdp)) 3439 return -ENOTSUPP; 3440 if (unlikely(meta < xdp_frame_end || 3441 meta > xdp->data)) 3442 return -EINVAL; 3443 if (unlikely((metalen & (sizeof(__u32) - 1)) || 3444 (metalen > 32))) 3445 return -EACCES; 3446 3447 xdp->data_meta = meta; 3448 3449 return 0; 3450 } 3451 3452 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = { 3453 .func = bpf_xdp_adjust_meta, 3454 .gpl_only = false, 3455 .ret_type = RET_INTEGER, 3456 .arg1_type = ARG_PTR_TO_CTX, 3457 .arg2_type = ARG_ANYTHING, 3458 }; 3459 3460 static int __bpf_tx_xdp(struct net_device *dev, 3461 struct bpf_map *map, 3462 struct xdp_buff *xdp, 3463 u32 index) 3464 { 3465 struct xdp_frame *xdpf; 3466 int err, sent; 3467 3468 if (!dev->netdev_ops->ndo_xdp_xmit) { 3469 return -EOPNOTSUPP; 3470 } 3471 3472 err = xdp_ok_fwd_dev(dev, xdp->data_end - xdp->data); 3473 if (unlikely(err)) 3474 return err; 3475 3476 xdpf = convert_to_xdp_frame(xdp); 3477 if (unlikely(!xdpf)) 3478 return -EOVERFLOW; 3479 3480 sent = dev->netdev_ops->ndo_xdp_xmit(dev, 1, &xdpf, XDP_XMIT_FLUSH); 3481 if (sent <= 0) 3482 return sent; 3483 return 0; 3484 } 3485 3486 static noinline int 3487 xdp_do_redirect_slow(struct net_device *dev, struct xdp_buff *xdp, 3488 struct bpf_prog *xdp_prog, struct bpf_redirect_info *ri) 3489 { 3490 struct net_device *fwd; 3491 u32 index = ri->tgt_index; 3492 int err; 3493 3494 fwd = dev_get_by_index_rcu(dev_net(dev), index); 3495 ri->tgt_index = 0; 3496 if (unlikely(!fwd)) { 3497 err = -EINVAL; 3498 goto err; 3499 } 3500 3501 err = __bpf_tx_xdp(fwd, NULL, xdp, 0); 3502 if (unlikely(err)) 3503 goto err; 3504 3505 _trace_xdp_redirect(dev, xdp_prog, index); 3506 return 0; 3507 err: 3508 _trace_xdp_redirect_err(dev, xdp_prog, index, err); 3509 return err; 3510 } 3511 3512 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd, 3513 struct bpf_map *map, 3514 struct xdp_buff *xdp, 3515 u32 index) 3516 { 3517 int err; 3518 3519 switch (map->map_type) { 3520 case BPF_MAP_TYPE_DEVMAP: 3521 case BPF_MAP_TYPE_DEVMAP_HASH: { 3522 struct bpf_dtab_netdev *dst = fwd; 3523 3524 err = dev_map_enqueue(dst, xdp, dev_rx); 3525 if (unlikely(err)) 3526 return err; 3527 break; 3528 } 3529 case BPF_MAP_TYPE_CPUMAP: { 3530 struct bpf_cpu_map_entry *rcpu = fwd; 3531 3532 err = cpu_map_enqueue(rcpu, xdp, dev_rx); 3533 if (unlikely(err)) 3534 return err; 3535 break; 3536 } 3537 case BPF_MAP_TYPE_XSKMAP: { 3538 struct xdp_sock *xs = fwd; 3539 3540 err = __xsk_map_redirect(map, xdp, xs); 3541 return err; 3542 } 3543 default: 3544 break; 3545 } 3546 return 0; 3547 } 3548 3549 void xdp_do_flush_map(void) 3550 { 3551 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3552 struct bpf_map *map = ri->map_to_flush; 3553 3554 ri->map_to_flush = NULL; 3555 if (map) { 3556 switch (map->map_type) { 3557 case BPF_MAP_TYPE_DEVMAP: 3558 case BPF_MAP_TYPE_DEVMAP_HASH: 3559 __dev_map_flush(map); 3560 break; 3561 case BPF_MAP_TYPE_CPUMAP: 3562 __cpu_map_flush(map); 3563 break; 3564 case BPF_MAP_TYPE_XSKMAP: 3565 __xsk_map_flush(map); 3566 break; 3567 default: 3568 break; 3569 } 3570 } 3571 } 3572 EXPORT_SYMBOL_GPL(xdp_do_flush_map); 3573 3574 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index) 3575 { 3576 switch (map->map_type) { 3577 case BPF_MAP_TYPE_DEVMAP: 3578 return __dev_map_lookup_elem(map, index); 3579 case BPF_MAP_TYPE_DEVMAP_HASH: 3580 return __dev_map_hash_lookup_elem(map, index); 3581 case BPF_MAP_TYPE_CPUMAP: 3582 return __cpu_map_lookup_elem(map, index); 3583 case BPF_MAP_TYPE_XSKMAP: 3584 return __xsk_map_lookup_elem(map, index); 3585 default: 3586 return NULL; 3587 } 3588 } 3589 3590 void bpf_clear_redirect_map(struct bpf_map *map) 3591 { 3592 struct bpf_redirect_info *ri; 3593 int cpu; 3594 3595 for_each_possible_cpu(cpu) { 3596 ri = per_cpu_ptr(&bpf_redirect_info, cpu); 3597 /* Avoid polluting remote cacheline due to writes if 3598 * not needed. Once we pass this test, we need the 3599 * cmpxchg() to make sure it hasn't been changed in 3600 * the meantime by remote CPU. 3601 */ 3602 if (unlikely(READ_ONCE(ri->map) == map)) 3603 cmpxchg(&ri->map, map, NULL); 3604 } 3605 } 3606 3607 static int xdp_do_redirect_map(struct net_device *dev, struct xdp_buff *xdp, 3608 struct bpf_prog *xdp_prog, struct bpf_map *map, 3609 struct bpf_redirect_info *ri) 3610 { 3611 u32 index = ri->tgt_index; 3612 void *fwd = ri->tgt_value; 3613 int err; 3614 3615 ri->tgt_index = 0; 3616 ri->tgt_value = NULL; 3617 WRITE_ONCE(ri->map, NULL); 3618 3619 if (ri->map_to_flush && unlikely(ri->map_to_flush != map)) 3620 xdp_do_flush_map(); 3621 3622 err = __bpf_tx_xdp_map(dev, fwd, map, xdp, index); 3623 if (unlikely(err)) 3624 goto err; 3625 3626 ri->map_to_flush = map; 3627 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index); 3628 return 0; 3629 err: 3630 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err); 3631 return err; 3632 } 3633 3634 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp, 3635 struct bpf_prog *xdp_prog) 3636 { 3637 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3638 struct bpf_map *map = READ_ONCE(ri->map); 3639 3640 if (likely(map)) 3641 return xdp_do_redirect_map(dev, xdp, xdp_prog, map, ri); 3642 3643 return xdp_do_redirect_slow(dev, xdp, xdp_prog, ri); 3644 } 3645 EXPORT_SYMBOL_GPL(xdp_do_redirect); 3646 3647 static int xdp_do_generic_redirect_map(struct net_device *dev, 3648 struct sk_buff *skb, 3649 struct xdp_buff *xdp, 3650 struct bpf_prog *xdp_prog, 3651 struct bpf_map *map) 3652 { 3653 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3654 u32 index = ri->tgt_index; 3655 void *fwd = ri->tgt_value; 3656 int err = 0; 3657 3658 ri->tgt_index = 0; 3659 ri->tgt_value = NULL; 3660 WRITE_ONCE(ri->map, NULL); 3661 3662 if (map->map_type == BPF_MAP_TYPE_DEVMAP || 3663 map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) { 3664 struct bpf_dtab_netdev *dst = fwd; 3665 3666 err = dev_map_generic_redirect(dst, skb, xdp_prog); 3667 if (unlikely(err)) 3668 goto err; 3669 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) { 3670 struct xdp_sock *xs = fwd; 3671 3672 err = xsk_generic_rcv(xs, xdp); 3673 if (err) 3674 goto err; 3675 consume_skb(skb); 3676 } else { 3677 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */ 3678 err = -EBADRQC; 3679 goto err; 3680 } 3681 3682 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index); 3683 return 0; 3684 err: 3685 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err); 3686 return err; 3687 } 3688 3689 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 3690 struct xdp_buff *xdp, struct bpf_prog *xdp_prog) 3691 { 3692 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3693 struct bpf_map *map = READ_ONCE(ri->map); 3694 u32 index = ri->tgt_index; 3695 struct net_device *fwd; 3696 int err = 0; 3697 3698 if (map) 3699 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, 3700 map); 3701 ri->tgt_index = 0; 3702 fwd = dev_get_by_index_rcu(dev_net(dev), index); 3703 if (unlikely(!fwd)) { 3704 err = -EINVAL; 3705 goto err; 3706 } 3707 3708 err = xdp_ok_fwd_dev(fwd, skb->len); 3709 if (unlikely(err)) 3710 goto err; 3711 3712 skb->dev = fwd; 3713 _trace_xdp_redirect(dev, xdp_prog, index); 3714 generic_xdp_tx(skb, xdp_prog); 3715 return 0; 3716 err: 3717 _trace_xdp_redirect_err(dev, xdp_prog, index, err); 3718 return err; 3719 } 3720 EXPORT_SYMBOL_GPL(xdp_do_generic_redirect); 3721 3722 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags) 3723 { 3724 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3725 3726 if (unlikely(flags)) 3727 return XDP_ABORTED; 3728 3729 ri->flags = flags; 3730 ri->tgt_index = ifindex; 3731 ri->tgt_value = NULL; 3732 WRITE_ONCE(ri->map, NULL); 3733 3734 return XDP_REDIRECT; 3735 } 3736 3737 static const struct bpf_func_proto bpf_xdp_redirect_proto = { 3738 .func = bpf_xdp_redirect, 3739 .gpl_only = false, 3740 .ret_type = RET_INTEGER, 3741 .arg1_type = ARG_ANYTHING, 3742 .arg2_type = ARG_ANYTHING, 3743 }; 3744 3745 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, 3746 u64, flags) 3747 { 3748 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3749 3750 /* Lower bits of the flags are used as return code on lookup failure */ 3751 if (unlikely(flags > XDP_TX)) 3752 return XDP_ABORTED; 3753 3754 ri->tgt_value = __xdp_map_lookup_elem(map, ifindex); 3755 if (unlikely(!ri->tgt_value)) { 3756 /* If the lookup fails we want to clear out the state in the 3757 * redirect_info struct completely, so that if an eBPF program 3758 * performs multiple lookups, the last one always takes 3759 * precedence. 3760 */ 3761 WRITE_ONCE(ri->map, NULL); 3762 return flags; 3763 } 3764 3765 ri->flags = flags; 3766 ri->tgt_index = ifindex; 3767 WRITE_ONCE(ri->map, map); 3768 3769 return XDP_REDIRECT; 3770 } 3771 3772 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = { 3773 .func = bpf_xdp_redirect_map, 3774 .gpl_only = false, 3775 .ret_type = RET_INTEGER, 3776 .arg1_type = ARG_CONST_MAP_PTR, 3777 .arg2_type = ARG_ANYTHING, 3778 .arg3_type = ARG_ANYTHING, 3779 }; 3780 3781 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb, 3782 unsigned long off, unsigned long len) 3783 { 3784 void *ptr = skb_header_pointer(skb, off, len, dst_buff); 3785 3786 if (unlikely(!ptr)) 3787 return len; 3788 if (ptr != dst_buff) 3789 memcpy(dst_buff, ptr, len); 3790 3791 return 0; 3792 } 3793 3794 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map, 3795 u64, flags, void *, meta, u64, meta_size) 3796 { 3797 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 3798 3799 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 3800 return -EINVAL; 3801 if (unlikely(skb_size > skb->len)) 3802 return -EFAULT; 3803 3804 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size, 3805 bpf_skb_copy); 3806 } 3807 3808 static const struct bpf_func_proto bpf_skb_event_output_proto = { 3809 .func = bpf_skb_event_output, 3810 .gpl_only = true, 3811 .ret_type = RET_INTEGER, 3812 .arg1_type = ARG_PTR_TO_CTX, 3813 .arg2_type = ARG_CONST_MAP_PTR, 3814 .arg3_type = ARG_ANYTHING, 3815 .arg4_type = ARG_PTR_TO_MEM, 3816 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 3817 }; 3818 3819 static unsigned short bpf_tunnel_key_af(u64 flags) 3820 { 3821 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET; 3822 } 3823 3824 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to, 3825 u32, size, u64, flags) 3826 { 3827 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 3828 u8 compat[sizeof(struct bpf_tunnel_key)]; 3829 void *to_orig = to; 3830 int err; 3831 3832 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) { 3833 err = -EINVAL; 3834 goto err_clear; 3835 } 3836 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { 3837 err = -EPROTO; 3838 goto err_clear; 3839 } 3840 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 3841 err = -EINVAL; 3842 switch (size) { 3843 case offsetof(struct bpf_tunnel_key, tunnel_label): 3844 case offsetof(struct bpf_tunnel_key, tunnel_ext): 3845 goto set_compat; 3846 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 3847 /* Fixup deprecated structure layouts here, so we have 3848 * a common path later on. 3849 */ 3850 if (ip_tunnel_info_af(info) != AF_INET) 3851 goto err_clear; 3852 set_compat: 3853 to = (struct bpf_tunnel_key *)compat; 3854 break; 3855 default: 3856 goto err_clear; 3857 } 3858 } 3859 3860 to->tunnel_id = be64_to_cpu(info->key.tun_id); 3861 to->tunnel_tos = info->key.tos; 3862 to->tunnel_ttl = info->key.ttl; 3863 to->tunnel_ext = 0; 3864 3865 if (flags & BPF_F_TUNINFO_IPV6) { 3866 memcpy(to->remote_ipv6, &info->key.u.ipv6.src, 3867 sizeof(to->remote_ipv6)); 3868 to->tunnel_label = be32_to_cpu(info->key.label); 3869 } else { 3870 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); 3871 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 3872 to->tunnel_label = 0; 3873 } 3874 3875 if (unlikely(size != sizeof(struct bpf_tunnel_key))) 3876 memcpy(to_orig, to, size); 3877 3878 return 0; 3879 err_clear: 3880 memset(to_orig, 0, size); 3881 return err; 3882 } 3883 3884 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { 3885 .func = bpf_skb_get_tunnel_key, 3886 .gpl_only = false, 3887 .ret_type = RET_INTEGER, 3888 .arg1_type = ARG_PTR_TO_CTX, 3889 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 3890 .arg3_type = ARG_CONST_SIZE, 3891 .arg4_type = ARG_ANYTHING, 3892 }; 3893 3894 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size) 3895 { 3896 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 3897 int err; 3898 3899 if (unlikely(!info || 3900 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) { 3901 err = -ENOENT; 3902 goto err_clear; 3903 } 3904 if (unlikely(size < info->options_len)) { 3905 err = -ENOMEM; 3906 goto err_clear; 3907 } 3908 3909 ip_tunnel_info_opts_get(to, info); 3910 if (size > info->options_len) 3911 memset(to + info->options_len, 0, size - info->options_len); 3912 3913 return info->options_len; 3914 err_clear: 3915 memset(to, 0, size); 3916 return err; 3917 } 3918 3919 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { 3920 .func = bpf_skb_get_tunnel_opt, 3921 .gpl_only = false, 3922 .ret_type = RET_INTEGER, 3923 .arg1_type = ARG_PTR_TO_CTX, 3924 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 3925 .arg3_type = ARG_CONST_SIZE, 3926 }; 3927 3928 static struct metadata_dst __percpu *md_dst; 3929 3930 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb, 3931 const struct bpf_tunnel_key *, from, u32, size, u64, flags) 3932 { 3933 struct metadata_dst *md = this_cpu_ptr(md_dst); 3934 u8 compat[sizeof(struct bpf_tunnel_key)]; 3935 struct ip_tunnel_info *info; 3936 3937 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | 3938 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER))) 3939 return -EINVAL; 3940 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 3941 switch (size) { 3942 case offsetof(struct bpf_tunnel_key, tunnel_label): 3943 case offsetof(struct bpf_tunnel_key, tunnel_ext): 3944 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 3945 /* Fixup deprecated structure layouts here, so we have 3946 * a common path later on. 3947 */ 3948 memcpy(compat, from, size); 3949 memset(compat + size, 0, sizeof(compat) - size); 3950 from = (const struct bpf_tunnel_key *) compat; 3951 break; 3952 default: 3953 return -EINVAL; 3954 } 3955 } 3956 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || 3957 from->tunnel_ext)) 3958 return -EINVAL; 3959 3960 skb_dst_drop(skb); 3961 dst_hold((struct dst_entry *) md); 3962 skb_dst_set(skb, (struct dst_entry *) md); 3963 3964 info = &md->u.tun_info; 3965 memset(info, 0, sizeof(*info)); 3966 info->mode = IP_TUNNEL_INFO_TX; 3967 3968 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE; 3969 if (flags & BPF_F_DONT_FRAGMENT) 3970 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT; 3971 if (flags & BPF_F_ZERO_CSUM_TX) 3972 info->key.tun_flags &= ~TUNNEL_CSUM; 3973 if (flags & BPF_F_SEQ_NUMBER) 3974 info->key.tun_flags |= TUNNEL_SEQ; 3975 3976 info->key.tun_id = cpu_to_be64(from->tunnel_id); 3977 info->key.tos = from->tunnel_tos; 3978 info->key.ttl = from->tunnel_ttl; 3979 3980 if (flags & BPF_F_TUNINFO_IPV6) { 3981 info->mode |= IP_TUNNEL_INFO_IPV6; 3982 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, 3983 sizeof(from->remote_ipv6)); 3984 info->key.label = cpu_to_be32(from->tunnel_label) & 3985 IPV6_FLOWLABEL_MASK; 3986 } else { 3987 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); 3988 } 3989 3990 return 0; 3991 } 3992 3993 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { 3994 .func = bpf_skb_set_tunnel_key, 3995 .gpl_only = false, 3996 .ret_type = RET_INTEGER, 3997 .arg1_type = ARG_PTR_TO_CTX, 3998 .arg2_type = ARG_PTR_TO_MEM, 3999 .arg3_type = ARG_CONST_SIZE, 4000 .arg4_type = ARG_ANYTHING, 4001 }; 4002 4003 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb, 4004 const u8 *, from, u32, size) 4005 { 4006 struct ip_tunnel_info *info = skb_tunnel_info(skb); 4007 const struct metadata_dst *md = this_cpu_ptr(md_dst); 4008 4009 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) 4010 return -EINVAL; 4011 if (unlikely(size > IP_TUNNEL_OPTS_MAX)) 4012 return -ENOMEM; 4013 4014 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT); 4015 4016 return 0; 4017 } 4018 4019 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { 4020 .func = bpf_skb_set_tunnel_opt, 4021 .gpl_only = false, 4022 .ret_type = RET_INTEGER, 4023 .arg1_type = ARG_PTR_TO_CTX, 4024 .arg2_type = ARG_PTR_TO_MEM, 4025 .arg3_type = ARG_CONST_SIZE, 4026 }; 4027 4028 static const struct bpf_func_proto * 4029 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) 4030 { 4031 if (!md_dst) { 4032 struct metadata_dst __percpu *tmp; 4033 4034 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, 4035 METADATA_IP_TUNNEL, 4036 GFP_KERNEL); 4037 if (!tmp) 4038 return NULL; 4039 if (cmpxchg(&md_dst, NULL, tmp)) 4040 metadata_dst_free_percpu(tmp); 4041 } 4042 4043 switch (which) { 4044 case BPF_FUNC_skb_set_tunnel_key: 4045 return &bpf_skb_set_tunnel_key_proto; 4046 case BPF_FUNC_skb_set_tunnel_opt: 4047 return &bpf_skb_set_tunnel_opt_proto; 4048 default: 4049 return NULL; 4050 } 4051 } 4052 4053 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map, 4054 u32, idx) 4055 { 4056 struct bpf_array *array = container_of(map, struct bpf_array, map); 4057 struct cgroup *cgrp; 4058 struct sock *sk; 4059 4060 sk = skb_to_full_sk(skb); 4061 if (!sk || !sk_fullsock(sk)) 4062 return -ENOENT; 4063 if (unlikely(idx >= array->map.max_entries)) 4064 return -E2BIG; 4065 4066 cgrp = READ_ONCE(array->ptrs[idx]); 4067 if (unlikely(!cgrp)) 4068 return -EAGAIN; 4069 4070 return sk_under_cgroup_hierarchy(sk, cgrp); 4071 } 4072 4073 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = { 4074 .func = bpf_skb_under_cgroup, 4075 .gpl_only = false, 4076 .ret_type = RET_INTEGER, 4077 .arg1_type = ARG_PTR_TO_CTX, 4078 .arg2_type = ARG_CONST_MAP_PTR, 4079 .arg3_type = ARG_ANYTHING, 4080 }; 4081 4082 #ifdef CONFIG_SOCK_CGROUP_DATA 4083 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb) 4084 { 4085 struct sock *sk = skb_to_full_sk(skb); 4086 struct cgroup *cgrp; 4087 4088 if (!sk || !sk_fullsock(sk)) 4089 return 0; 4090 4091 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4092 return cgrp->kn->id.id; 4093 } 4094 4095 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = { 4096 .func = bpf_skb_cgroup_id, 4097 .gpl_only = false, 4098 .ret_type = RET_INTEGER, 4099 .arg1_type = ARG_PTR_TO_CTX, 4100 }; 4101 4102 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int, 4103 ancestor_level) 4104 { 4105 struct sock *sk = skb_to_full_sk(skb); 4106 struct cgroup *ancestor; 4107 struct cgroup *cgrp; 4108 4109 if (!sk || !sk_fullsock(sk)) 4110 return 0; 4111 4112 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4113 ancestor = cgroup_ancestor(cgrp, ancestor_level); 4114 if (!ancestor) 4115 return 0; 4116 4117 return ancestor->kn->id.id; 4118 } 4119 4120 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = { 4121 .func = bpf_skb_ancestor_cgroup_id, 4122 .gpl_only = false, 4123 .ret_type = RET_INTEGER, 4124 .arg1_type = ARG_PTR_TO_CTX, 4125 .arg2_type = ARG_ANYTHING, 4126 }; 4127 #endif 4128 4129 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff, 4130 unsigned long off, unsigned long len) 4131 { 4132 memcpy(dst_buff, src_buff + off, len); 4133 return 0; 4134 } 4135 4136 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map, 4137 u64, flags, void *, meta, u64, meta_size) 4138 { 4139 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4140 4141 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4142 return -EINVAL; 4143 if (unlikely(xdp_size > (unsigned long)(xdp->data_end - xdp->data))) 4144 return -EFAULT; 4145 4146 return bpf_event_output(map, flags, meta, meta_size, xdp->data, 4147 xdp_size, bpf_xdp_copy); 4148 } 4149 4150 static const struct bpf_func_proto bpf_xdp_event_output_proto = { 4151 .func = bpf_xdp_event_output, 4152 .gpl_only = true, 4153 .ret_type = RET_INTEGER, 4154 .arg1_type = ARG_PTR_TO_CTX, 4155 .arg2_type = ARG_CONST_MAP_PTR, 4156 .arg3_type = ARG_ANYTHING, 4157 .arg4_type = ARG_PTR_TO_MEM, 4158 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4159 }; 4160 4161 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb) 4162 { 4163 return skb->sk ? sock_gen_cookie(skb->sk) : 0; 4164 } 4165 4166 static const struct bpf_func_proto bpf_get_socket_cookie_proto = { 4167 .func = bpf_get_socket_cookie, 4168 .gpl_only = false, 4169 .ret_type = RET_INTEGER, 4170 .arg1_type = ARG_PTR_TO_CTX, 4171 }; 4172 4173 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4174 { 4175 return sock_gen_cookie(ctx->sk); 4176 } 4177 4178 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = { 4179 .func = bpf_get_socket_cookie_sock_addr, 4180 .gpl_only = false, 4181 .ret_type = RET_INTEGER, 4182 .arg1_type = ARG_PTR_TO_CTX, 4183 }; 4184 4185 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 4186 { 4187 return sock_gen_cookie(ctx->sk); 4188 } 4189 4190 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = { 4191 .func = bpf_get_socket_cookie_sock_ops, 4192 .gpl_only = false, 4193 .ret_type = RET_INTEGER, 4194 .arg1_type = ARG_PTR_TO_CTX, 4195 }; 4196 4197 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb) 4198 { 4199 struct sock *sk = sk_to_full_sk(skb->sk); 4200 kuid_t kuid; 4201 4202 if (!sk || !sk_fullsock(sk)) 4203 return overflowuid; 4204 kuid = sock_net_uid(sock_net(sk), sk); 4205 return from_kuid_munged(sock_net(sk)->user_ns, kuid); 4206 } 4207 4208 static const struct bpf_func_proto bpf_get_socket_uid_proto = { 4209 .func = bpf_get_socket_uid, 4210 .gpl_only = false, 4211 .ret_type = RET_INTEGER, 4212 .arg1_type = ARG_PTR_TO_CTX, 4213 }; 4214 4215 BPF_CALL_5(bpf_sockopt_event_output, struct bpf_sock_ops_kern *, bpf_sock, 4216 struct bpf_map *, map, u64, flags, void *, data, u64, size) 4217 { 4218 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) 4219 return -EINVAL; 4220 4221 return bpf_event_output(map, flags, data, size, NULL, 0, NULL); 4222 } 4223 4224 static const struct bpf_func_proto bpf_sockopt_event_output_proto = { 4225 .func = bpf_sockopt_event_output, 4226 .gpl_only = true, 4227 .ret_type = RET_INTEGER, 4228 .arg1_type = ARG_PTR_TO_CTX, 4229 .arg2_type = ARG_CONST_MAP_PTR, 4230 .arg3_type = ARG_ANYTHING, 4231 .arg4_type = ARG_PTR_TO_MEM, 4232 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4233 }; 4234 4235 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock, 4236 int, level, int, optname, char *, optval, int, optlen) 4237 { 4238 struct sock *sk = bpf_sock->sk; 4239 int ret = 0; 4240 int val; 4241 4242 if (!sk_fullsock(sk)) 4243 return -EINVAL; 4244 4245 if (level == SOL_SOCKET) { 4246 if (optlen != sizeof(int)) 4247 return -EINVAL; 4248 val = *((int *)optval); 4249 4250 /* Only some socketops are supported */ 4251 switch (optname) { 4252 case SO_RCVBUF: 4253 val = min_t(u32, val, sysctl_rmem_max); 4254 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 4255 sk->sk_rcvbuf = max_t(int, val * 2, SOCK_MIN_RCVBUF); 4256 break; 4257 case SO_SNDBUF: 4258 val = min_t(u32, val, sysctl_wmem_max); 4259 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 4260 sk->sk_sndbuf = max_t(int, val * 2, SOCK_MIN_SNDBUF); 4261 break; 4262 case SO_MAX_PACING_RATE: /* 32bit version */ 4263 if (val != ~0U) 4264 cmpxchg(&sk->sk_pacing_status, 4265 SK_PACING_NONE, 4266 SK_PACING_NEEDED); 4267 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val; 4268 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 4269 sk->sk_max_pacing_rate); 4270 break; 4271 case SO_PRIORITY: 4272 sk->sk_priority = val; 4273 break; 4274 case SO_RCVLOWAT: 4275 if (val < 0) 4276 val = INT_MAX; 4277 sk->sk_rcvlowat = val ? : 1; 4278 break; 4279 case SO_MARK: 4280 if (sk->sk_mark != val) { 4281 sk->sk_mark = val; 4282 sk_dst_reset(sk); 4283 } 4284 break; 4285 default: 4286 ret = -EINVAL; 4287 } 4288 #ifdef CONFIG_INET 4289 } else if (level == SOL_IP) { 4290 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 4291 return -EINVAL; 4292 4293 val = *((int *)optval); 4294 /* Only some options are supported */ 4295 switch (optname) { 4296 case IP_TOS: 4297 if (val < -1 || val > 0xff) { 4298 ret = -EINVAL; 4299 } else { 4300 struct inet_sock *inet = inet_sk(sk); 4301 4302 if (val == -1) 4303 val = 0; 4304 inet->tos = val; 4305 } 4306 break; 4307 default: 4308 ret = -EINVAL; 4309 } 4310 #if IS_ENABLED(CONFIG_IPV6) 4311 } else if (level == SOL_IPV6) { 4312 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 4313 return -EINVAL; 4314 4315 val = *((int *)optval); 4316 /* Only some options are supported */ 4317 switch (optname) { 4318 case IPV6_TCLASS: 4319 if (val < -1 || val > 0xff) { 4320 ret = -EINVAL; 4321 } else { 4322 struct ipv6_pinfo *np = inet6_sk(sk); 4323 4324 if (val == -1) 4325 val = 0; 4326 np->tclass = val; 4327 } 4328 break; 4329 default: 4330 ret = -EINVAL; 4331 } 4332 #endif 4333 } else if (level == SOL_TCP && 4334 sk->sk_prot->setsockopt == tcp_setsockopt) { 4335 if (optname == TCP_CONGESTION) { 4336 char name[TCP_CA_NAME_MAX]; 4337 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN; 4338 4339 strncpy(name, optval, min_t(long, optlen, 4340 TCP_CA_NAME_MAX-1)); 4341 name[TCP_CA_NAME_MAX-1] = 0; 4342 ret = tcp_set_congestion_control(sk, name, false, 4343 reinit, true); 4344 } else { 4345 struct tcp_sock *tp = tcp_sk(sk); 4346 4347 if (optlen != sizeof(int)) 4348 return -EINVAL; 4349 4350 val = *((int *)optval); 4351 /* Only some options are supported */ 4352 switch (optname) { 4353 case TCP_BPF_IW: 4354 if (val <= 0 || tp->data_segs_out > tp->syn_data) 4355 ret = -EINVAL; 4356 else 4357 tp->snd_cwnd = val; 4358 break; 4359 case TCP_BPF_SNDCWND_CLAMP: 4360 if (val <= 0) { 4361 ret = -EINVAL; 4362 } else { 4363 tp->snd_cwnd_clamp = val; 4364 tp->snd_ssthresh = val; 4365 } 4366 break; 4367 case TCP_SAVE_SYN: 4368 if (val < 0 || val > 1) 4369 ret = -EINVAL; 4370 else 4371 tp->save_syn = val; 4372 break; 4373 default: 4374 ret = -EINVAL; 4375 } 4376 } 4377 #endif 4378 } else { 4379 ret = -EINVAL; 4380 } 4381 return ret; 4382 } 4383 4384 static const struct bpf_func_proto bpf_setsockopt_proto = { 4385 .func = bpf_setsockopt, 4386 .gpl_only = false, 4387 .ret_type = RET_INTEGER, 4388 .arg1_type = ARG_PTR_TO_CTX, 4389 .arg2_type = ARG_ANYTHING, 4390 .arg3_type = ARG_ANYTHING, 4391 .arg4_type = ARG_PTR_TO_MEM, 4392 .arg5_type = ARG_CONST_SIZE, 4393 }; 4394 4395 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock, 4396 int, level, int, optname, char *, optval, int, optlen) 4397 { 4398 struct sock *sk = bpf_sock->sk; 4399 4400 if (!sk_fullsock(sk)) 4401 goto err_clear; 4402 #ifdef CONFIG_INET 4403 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) { 4404 struct inet_connection_sock *icsk; 4405 struct tcp_sock *tp; 4406 4407 switch (optname) { 4408 case TCP_CONGESTION: 4409 icsk = inet_csk(sk); 4410 4411 if (!icsk->icsk_ca_ops || optlen <= 1) 4412 goto err_clear; 4413 strncpy(optval, icsk->icsk_ca_ops->name, optlen); 4414 optval[optlen - 1] = 0; 4415 break; 4416 case TCP_SAVED_SYN: 4417 tp = tcp_sk(sk); 4418 4419 if (optlen <= 0 || !tp->saved_syn || 4420 optlen > tp->saved_syn[0]) 4421 goto err_clear; 4422 memcpy(optval, tp->saved_syn + 1, optlen); 4423 break; 4424 default: 4425 goto err_clear; 4426 } 4427 } else if (level == SOL_IP) { 4428 struct inet_sock *inet = inet_sk(sk); 4429 4430 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 4431 goto err_clear; 4432 4433 /* Only some options are supported */ 4434 switch (optname) { 4435 case IP_TOS: 4436 *((int *)optval) = (int)inet->tos; 4437 break; 4438 default: 4439 goto err_clear; 4440 } 4441 #if IS_ENABLED(CONFIG_IPV6) 4442 } else if (level == SOL_IPV6) { 4443 struct ipv6_pinfo *np = inet6_sk(sk); 4444 4445 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 4446 goto err_clear; 4447 4448 /* Only some options are supported */ 4449 switch (optname) { 4450 case IPV6_TCLASS: 4451 *((int *)optval) = (int)np->tclass; 4452 break; 4453 default: 4454 goto err_clear; 4455 } 4456 #endif 4457 } else { 4458 goto err_clear; 4459 } 4460 return 0; 4461 #endif 4462 err_clear: 4463 memset(optval, 0, optlen); 4464 return -EINVAL; 4465 } 4466 4467 static const struct bpf_func_proto bpf_getsockopt_proto = { 4468 .func = bpf_getsockopt, 4469 .gpl_only = false, 4470 .ret_type = RET_INTEGER, 4471 .arg1_type = ARG_PTR_TO_CTX, 4472 .arg2_type = ARG_ANYTHING, 4473 .arg3_type = ARG_ANYTHING, 4474 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 4475 .arg5_type = ARG_CONST_SIZE, 4476 }; 4477 4478 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock, 4479 int, argval) 4480 { 4481 struct sock *sk = bpf_sock->sk; 4482 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS; 4483 4484 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk)) 4485 return -EINVAL; 4486 4487 tcp_sk(sk)->bpf_sock_ops_cb_flags = val; 4488 4489 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS); 4490 } 4491 4492 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = { 4493 .func = bpf_sock_ops_cb_flags_set, 4494 .gpl_only = false, 4495 .ret_type = RET_INTEGER, 4496 .arg1_type = ARG_PTR_TO_CTX, 4497 .arg2_type = ARG_ANYTHING, 4498 }; 4499 4500 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly; 4501 EXPORT_SYMBOL_GPL(ipv6_bpf_stub); 4502 4503 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr, 4504 int, addr_len) 4505 { 4506 #ifdef CONFIG_INET 4507 struct sock *sk = ctx->sk; 4508 int err; 4509 4510 /* Binding to port can be expensive so it's prohibited in the helper. 4511 * Only binding to IP is supported. 4512 */ 4513 err = -EINVAL; 4514 if (addr_len < offsetofend(struct sockaddr, sa_family)) 4515 return err; 4516 if (addr->sa_family == AF_INET) { 4517 if (addr_len < sizeof(struct sockaddr_in)) 4518 return err; 4519 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 4520 return err; 4521 return __inet_bind(sk, addr, addr_len, true, false); 4522 #if IS_ENABLED(CONFIG_IPV6) 4523 } else if (addr->sa_family == AF_INET6) { 4524 if (addr_len < SIN6_LEN_RFC2133) 4525 return err; 4526 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 4527 return err; 4528 /* ipv6_bpf_stub cannot be NULL, since it's called from 4529 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded 4530 */ 4531 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false); 4532 #endif /* CONFIG_IPV6 */ 4533 } 4534 #endif /* CONFIG_INET */ 4535 4536 return -EAFNOSUPPORT; 4537 } 4538 4539 static const struct bpf_func_proto bpf_bind_proto = { 4540 .func = bpf_bind, 4541 .gpl_only = false, 4542 .ret_type = RET_INTEGER, 4543 .arg1_type = ARG_PTR_TO_CTX, 4544 .arg2_type = ARG_PTR_TO_MEM, 4545 .arg3_type = ARG_CONST_SIZE, 4546 }; 4547 4548 #ifdef CONFIG_XFRM 4549 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index, 4550 struct bpf_xfrm_state *, to, u32, size, u64, flags) 4551 { 4552 const struct sec_path *sp = skb_sec_path(skb); 4553 const struct xfrm_state *x; 4554 4555 if (!sp || unlikely(index >= sp->len || flags)) 4556 goto err_clear; 4557 4558 x = sp->xvec[index]; 4559 4560 if (unlikely(size != sizeof(struct bpf_xfrm_state))) 4561 goto err_clear; 4562 4563 to->reqid = x->props.reqid; 4564 to->spi = x->id.spi; 4565 to->family = x->props.family; 4566 to->ext = 0; 4567 4568 if (to->family == AF_INET6) { 4569 memcpy(to->remote_ipv6, x->props.saddr.a6, 4570 sizeof(to->remote_ipv6)); 4571 } else { 4572 to->remote_ipv4 = x->props.saddr.a4; 4573 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 4574 } 4575 4576 return 0; 4577 err_clear: 4578 memset(to, 0, size); 4579 return -EINVAL; 4580 } 4581 4582 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = { 4583 .func = bpf_skb_get_xfrm_state, 4584 .gpl_only = false, 4585 .ret_type = RET_INTEGER, 4586 .arg1_type = ARG_PTR_TO_CTX, 4587 .arg2_type = ARG_ANYTHING, 4588 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 4589 .arg4_type = ARG_CONST_SIZE, 4590 .arg5_type = ARG_ANYTHING, 4591 }; 4592 #endif 4593 4594 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6) 4595 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, 4596 const struct neighbour *neigh, 4597 const struct net_device *dev) 4598 { 4599 memcpy(params->dmac, neigh->ha, ETH_ALEN); 4600 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 4601 params->h_vlan_TCI = 0; 4602 params->h_vlan_proto = 0; 4603 params->ifindex = dev->ifindex; 4604 4605 return 0; 4606 } 4607 #endif 4608 4609 #if IS_ENABLED(CONFIG_INET) 4610 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 4611 u32 flags, bool check_mtu) 4612 { 4613 struct fib_nh_common *nhc; 4614 struct in_device *in_dev; 4615 struct neighbour *neigh; 4616 struct net_device *dev; 4617 struct fib_result res; 4618 struct flowi4 fl4; 4619 int err; 4620 u32 mtu; 4621 4622 dev = dev_get_by_index_rcu(net, params->ifindex); 4623 if (unlikely(!dev)) 4624 return -ENODEV; 4625 4626 /* verify forwarding is enabled on this interface */ 4627 in_dev = __in_dev_get_rcu(dev); 4628 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev))) 4629 return BPF_FIB_LKUP_RET_FWD_DISABLED; 4630 4631 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 4632 fl4.flowi4_iif = 1; 4633 fl4.flowi4_oif = params->ifindex; 4634 } else { 4635 fl4.flowi4_iif = params->ifindex; 4636 fl4.flowi4_oif = 0; 4637 } 4638 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK; 4639 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 4640 fl4.flowi4_flags = 0; 4641 4642 fl4.flowi4_proto = params->l4_protocol; 4643 fl4.daddr = params->ipv4_dst; 4644 fl4.saddr = params->ipv4_src; 4645 fl4.fl4_sport = params->sport; 4646 fl4.fl4_dport = params->dport; 4647 4648 if (flags & BPF_FIB_LOOKUP_DIRECT) { 4649 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 4650 struct fib_table *tb; 4651 4652 tb = fib_get_table(net, tbid); 4653 if (unlikely(!tb)) 4654 return BPF_FIB_LKUP_RET_NOT_FWDED; 4655 4656 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 4657 } else { 4658 fl4.flowi4_mark = 0; 4659 fl4.flowi4_secid = 0; 4660 fl4.flowi4_tun_key.tun_id = 0; 4661 fl4.flowi4_uid = sock_net_uid(net, NULL); 4662 4663 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF); 4664 } 4665 4666 if (err) { 4667 /* map fib lookup errors to RTN_ type */ 4668 if (err == -EINVAL) 4669 return BPF_FIB_LKUP_RET_BLACKHOLE; 4670 if (err == -EHOSTUNREACH) 4671 return BPF_FIB_LKUP_RET_UNREACHABLE; 4672 if (err == -EACCES) 4673 return BPF_FIB_LKUP_RET_PROHIBIT; 4674 4675 return BPF_FIB_LKUP_RET_NOT_FWDED; 4676 } 4677 4678 if (res.type != RTN_UNICAST) 4679 return BPF_FIB_LKUP_RET_NOT_FWDED; 4680 4681 if (fib_info_num_path(res.fi) > 1) 4682 fib_select_path(net, &res, &fl4, NULL); 4683 4684 if (check_mtu) { 4685 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst); 4686 if (params->tot_len > mtu) 4687 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 4688 } 4689 4690 nhc = res.nhc; 4691 4692 /* do not handle lwt encaps right now */ 4693 if (nhc->nhc_lwtstate) 4694 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 4695 4696 dev = nhc->nhc_dev; 4697 4698 params->rt_metric = res.fi->fib_priority; 4699 4700 /* xdp and cls_bpf programs are run in RCU-bh so 4701 * rcu_read_lock_bh is not needed here 4702 */ 4703 if (likely(nhc->nhc_gw_family != AF_INET6)) { 4704 if (nhc->nhc_gw_family) 4705 params->ipv4_dst = nhc->nhc_gw.ipv4; 4706 4707 neigh = __ipv4_neigh_lookup_noref(dev, 4708 (__force u32)params->ipv4_dst); 4709 } else { 4710 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst; 4711 4712 params->family = AF_INET6; 4713 *dst = nhc->nhc_gw.ipv6; 4714 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 4715 } 4716 4717 if (!neigh) 4718 return BPF_FIB_LKUP_RET_NO_NEIGH; 4719 4720 return bpf_fib_set_fwd_params(params, neigh, dev); 4721 } 4722 #endif 4723 4724 #if IS_ENABLED(CONFIG_IPV6) 4725 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 4726 u32 flags, bool check_mtu) 4727 { 4728 struct in6_addr *src = (struct in6_addr *) params->ipv6_src; 4729 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst; 4730 struct fib6_result res = {}; 4731 struct neighbour *neigh; 4732 struct net_device *dev; 4733 struct inet6_dev *idev; 4734 struct flowi6 fl6; 4735 int strict = 0; 4736 int oif, err; 4737 u32 mtu; 4738 4739 /* link local addresses are never forwarded */ 4740 if (rt6_need_strict(dst) || rt6_need_strict(src)) 4741 return BPF_FIB_LKUP_RET_NOT_FWDED; 4742 4743 dev = dev_get_by_index_rcu(net, params->ifindex); 4744 if (unlikely(!dev)) 4745 return -ENODEV; 4746 4747 idev = __in6_dev_get_safely(dev); 4748 if (unlikely(!idev || !idev->cnf.forwarding)) 4749 return BPF_FIB_LKUP_RET_FWD_DISABLED; 4750 4751 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 4752 fl6.flowi6_iif = 1; 4753 oif = fl6.flowi6_oif = params->ifindex; 4754 } else { 4755 oif = fl6.flowi6_iif = params->ifindex; 4756 fl6.flowi6_oif = 0; 4757 strict = RT6_LOOKUP_F_HAS_SADDR; 4758 } 4759 fl6.flowlabel = params->flowinfo; 4760 fl6.flowi6_scope = 0; 4761 fl6.flowi6_flags = 0; 4762 fl6.mp_hash = 0; 4763 4764 fl6.flowi6_proto = params->l4_protocol; 4765 fl6.daddr = *dst; 4766 fl6.saddr = *src; 4767 fl6.fl6_sport = params->sport; 4768 fl6.fl6_dport = params->dport; 4769 4770 if (flags & BPF_FIB_LOOKUP_DIRECT) { 4771 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 4772 struct fib6_table *tb; 4773 4774 tb = ipv6_stub->fib6_get_table(net, tbid); 4775 if (unlikely(!tb)) 4776 return BPF_FIB_LKUP_RET_NOT_FWDED; 4777 4778 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res, 4779 strict); 4780 } else { 4781 fl6.flowi6_mark = 0; 4782 fl6.flowi6_secid = 0; 4783 fl6.flowi6_tun_key.tun_id = 0; 4784 fl6.flowi6_uid = sock_net_uid(net, NULL); 4785 4786 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict); 4787 } 4788 4789 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) || 4790 res.f6i == net->ipv6.fib6_null_entry)) 4791 return BPF_FIB_LKUP_RET_NOT_FWDED; 4792 4793 switch (res.fib6_type) { 4794 /* only unicast is forwarded */ 4795 case RTN_UNICAST: 4796 break; 4797 case RTN_BLACKHOLE: 4798 return BPF_FIB_LKUP_RET_BLACKHOLE; 4799 case RTN_UNREACHABLE: 4800 return BPF_FIB_LKUP_RET_UNREACHABLE; 4801 case RTN_PROHIBIT: 4802 return BPF_FIB_LKUP_RET_PROHIBIT; 4803 default: 4804 return BPF_FIB_LKUP_RET_NOT_FWDED; 4805 } 4806 4807 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif, 4808 fl6.flowi6_oif != 0, NULL, strict); 4809 4810 if (check_mtu) { 4811 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src); 4812 if (params->tot_len > mtu) 4813 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 4814 } 4815 4816 if (res.nh->fib_nh_lws) 4817 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 4818 4819 if (res.nh->fib_nh_gw_family) 4820 *dst = res.nh->fib_nh_gw6; 4821 4822 dev = res.nh->fib_nh_dev; 4823 params->rt_metric = res.f6i->fib6_metric; 4824 4825 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is 4826 * not needed here. 4827 */ 4828 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 4829 if (!neigh) 4830 return BPF_FIB_LKUP_RET_NO_NEIGH; 4831 4832 return bpf_fib_set_fwd_params(params, neigh, dev); 4833 } 4834 #endif 4835 4836 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx, 4837 struct bpf_fib_lookup *, params, int, plen, u32, flags) 4838 { 4839 if (plen < sizeof(*params)) 4840 return -EINVAL; 4841 4842 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 4843 return -EINVAL; 4844 4845 switch (params->family) { 4846 #if IS_ENABLED(CONFIG_INET) 4847 case AF_INET: 4848 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params, 4849 flags, true); 4850 #endif 4851 #if IS_ENABLED(CONFIG_IPV6) 4852 case AF_INET6: 4853 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params, 4854 flags, true); 4855 #endif 4856 } 4857 return -EAFNOSUPPORT; 4858 } 4859 4860 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = { 4861 .func = bpf_xdp_fib_lookup, 4862 .gpl_only = true, 4863 .ret_type = RET_INTEGER, 4864 .arg1_type = ARG_PTR_TO_CTX, 4865 .arg2_type = ARG_PTR_TO_MEM, 4866 .arg3_type = ARG_CONST_SIZE, 4867 .arg4_type = ARG_ANYTHING, 4868 }; 4869 4870 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb, 4871 struct bpf_fib_lookup *, params, int, plen, u32, flags) 4872 { 4873 struct net *net = dev_net(skb->dev); 4874 int rc = -EAFNOSUPPORT; 4875 4876 if (plen < sizeof(*params)) 4877 return -EINVAL; 4878 4879 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 4880 return -EINVAL; 4881 4882 switch (params->family) { 4883 #if IS_ENABLED(CONFIG_INET) 4884 case AF_INET: 4885 rc = bpf_ipv4_fib_lookup(net, params, flags, false); 4886 break; 4887 #endif 4888 #if IS_ENABLED(CONFIG_IPV6) 4889 case AF_INET6: 4890 rc = bpf_ipv6_fib_lookup(net, params, flags, false); 4891 break; 4892 #endif 4893 } 4894 4895 if (!rc) { 4896 struct net_device *dev; 4897 4898 dev = dev_get_by_index_rcu(net, params->ifindex); 4899 if (!is_skb_forwardable(dev, skb)) 4900 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED; 4901 } 4902 4903 return rc; 4904 } 4905 4906 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = { 4907 .func = bpf_skb_fib_lookup, 4908 .gpl_only = true, 4909 .ret_type = RET_INTEGER, 4910 .arg1_type = ARG_PTR_TO_CTX, 4911 .arg2_type = ARG_PTR_TO_MEM, 4912 .arg3_type = ARG_CONST_SIZE, 4913 .arg4_type = ARG_ANYTHING, 4914 }; 4915 4916 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 4917 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len) 4918 { 4919 int err; 4920 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr; 4921 4922 if (!seg6_validate_srh(srh, len)) 4923 return -EINVAL; 4924 4925 switch (type) { 4926 case BPF_LWT_ENCAP_SEG6_INLINE: 4927 if (skb->protocol != htons(ETH_P_IPV6)) 4928 return -EBADMSG; 4929 4930 err = seg6_do_srh_inline(skb, srh); 4931 break; 4932 case BPF_LWT_ENCAP_SEG6: 4933 skb_reset_inner_headers(skb); 4934 skb->encapsulation = 1; 4935 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6); 4936 break; 4937 default: 4938 return -EINVAL; 4939 } 4940 4941 bpf_compute_data_pointers(skb); 4942 if (err) 4943 return err; 4944 4945 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 4946 skb_set_transport_header(skb, sizeof(struct ipv6hdr)); 4947 4948 return seg6_lookup_nexthop(skb, NULL, 0); 4949 } 4950 #endif /* CONFIG_IPV6_SEG6_BPF */ 4951 4952 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 4953 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, 4954 bool ingress) 4955 { 4956 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress); 4957 } 4958 #endif 4959 4960 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr, 4961 u32, len) 4962 { 4963 switch (type) { 4964 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 4965 case BPF_LWT_ENCAP_SEG6: 4966 case BPF_LWT_ENCAP_SEG6_INLINE: 4967 return bpf_push_seg6_encap(skb, type, hdr, len); 4968 #endif 4969 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 4970 case BPF_LWT_ENCAP_IP: 4971 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */); 4972 #endif 4973 default: 4974 return -EINVAL; 4975 } 4976 } 4977 4978 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type, 4979 void *, hdr, u32, len) 4980 { 4981 switch (type) { 4982 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 4983 case BPF_LWT_ENCAP_IP: 4984 return bpf_push_ip_encap(skb, hdr, len, false /* egress */); 4985 #endif 4986 default: 4987 return -EINVAL; 4988 } 4989 } 4990 4991 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = { 4992 .func = bpf_lwt_in_push_encap, 4993 .gpl_only = false, 4994 .ret_type = RET_INTEGER, 4995 .arg1_type = ARG_PTR_TO_CTX, 4996 .arg2_type = ARG_ANYTHING, 4997 .arg3_type = ARG_PTR_TO_MEM, 4998 .arg4_type = ARG_CONST_SIZE 4999 }; 5000 5001 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = { 5002 .func = bpf_lwt_xmit_push_encap, 5003 .gpl_only = false, 5004 .ret_type = RET_INTEGER, 5005 .arg1_type = ARG_PTR_TO_CTX, 5006 .arg2_type = ARG_ANYTHING, 5007 .arg3_type = ARG_PTR_TO_MEM, 5008 .arg4_type = ARG_CONST_SIZE 5009 }; 5010 5011 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5012 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset, 5013 const void *, from, u32, len) 5014 { 5015 struct seg6_bpf_srh_state *srh_state = 5016 this_cpu_ptr(&seg6_bpf_srh_states); 5017 struct ipv6_sr_hdr *srh = srh_state->srh; 5018 void *srh_tlvs, *srh_end, *ptr; 5019 int srhoff = 0; 5020 5021 if (srh == NULL) 5022 return -EINVAL; 5023 5024 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4)); 5025 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen); 5026 5027 ptr = skb->data + offset; 5028 if (ptr >= srh_tlvs && ptr + len <= srh_end) 5029 srh_state->valid = false; 5030 else if (ptr < (void *)&srh->flags || 5031 ptr + len > (void *)&srh->segments) 5032 return -EFAULT; 5033 5034 if (unlikely(bpf_try_make_writable(skb, offset + len))) 5035 return -EFAULT; 5036 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 5037 return -EINVAL; 5038 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5039 5040 memcpy(skb->data + offset, from, len); 5041 return 0; 5042 } 5043 5044 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = { 5045 .func = bpf_lwt_seg6_store_bytes, 5046 .gpl_only = false, 5047 .ret_type = RET_INTEGER, 5048 .arg1_type = ARG_PTR_TO_CTX, 5049 .arg2_type = ARG_ANYTHING, 5050 .arg3_type = ARG_PTR_TO_MEM, 5051 .arg4_type = ARG_CONST_SIZE 5052 }; 5053 5054 static void bpf_update_srh_state(struct sk_buff *skb) 5055 { 5056 struct seg6_bpf_srh_state *srh_state = 5057 this_cpu_ptr(&seg6_bpf_srh_states); 5058 int srhoff = 0; 5059 5060 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) { 5061 srh_state->srh = NULL; 5062 } else { 5063 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5064 srh_state->hdrlen = srh_state->srh->hdrlen << 3; 5065 srh_state->valid = true; 5066 } 5067 } 5068 5069 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb, 5070 u32, action, void *, param, u32, param_len) 5071 { 5072 struct seg6_bpf_srh_state *srh_state = 5073 this_cpu_ptr(&seg6_bpf_srh_states); 5074 int hdroff = 0; 5075 int err; 5076 5077 switch (action) { 5078 case SEG6_LOCAL_ACTION_END_X: 5079 if (!seg6_bpf_has_valid_srh(skb)) 5080 return -EBADMSG; 5081 if (param_len != sizeof(struct in6_addr)) 5082 return -EINVAL; 5083 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0); 5084 case SEG6_LOCAL_ACTION_END_T: 5085 if (!seg6_bpf_has_valid_srh(skb)) 5086 return -EBADMSG; 5087 if (param_len != sizeof(int)) 5088 return -EINVAL; 5089 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 5090 case SEG6_LOCAL_ACTION_END_DT6: 5091 if (!seg6_bpf_has_valid_srh(skb)) 5092 return -EBADMSG; 5093 if (param_len != sizeof(int)) 5094 return -EINVAL; 5095 5096 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0) 5097 return -EBADMSG; 5098 if (!pskb_pull(skb, hdroff)) 5099 return -EBADMSG; 5100 5101 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff); 5102 skb_reset_network_header(skb); 5103 skb_reset_transport_header(skb); 5104 skb->encapsulation = 0; 5105 5106 bpf_compute_data_pointers(skb); 5107 bpf_update_srh_state(skb); 5108 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 5109 case SEG6_LOCAL_ACTION_END_B6: 5110 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 5111 return -EBADMSG; 5112 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE, 5113 param, param_len); 5114 if (!err) 5115 bpf_update_srh_state(skb); 5116 5117 return err; 5118 case SEG6_LOCAL_ACTION_END_B6_ENCAP: 5119 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 5120 return -EBADMSG; 5121 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6, 5122 param, param_len); 5123 if (!err) 5124 bpf_update_srh_state(skb); 5125 5126 return err; 5127 default: 5128 return -EINVAL; 5129 } 5130 } 5131 5132 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = { 5133 .func = bpf_lwt_seg6_action, 5134 .gpl_only = false, 5135 .ret_type = RET_INTEGER, 5136 .arg1_type = ARG_PTR_TO_CTX, 5137 .arg2_type = ARG_ANYTHING, 5138 .arg3_type = ARG_PTR_TO_MEM, 5139 .arg4_type = ARG_CONST_SIZE 5140 }; 5141 5142 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset, 5143 s32, len) 5144 { 5145 struct seg6_bpf_srh_state *srh_state = 5146 this_cpu_ptr(&seg6_bpf_srh_states); 5147 struct ipv6_sr_hdr *srh = srh_state->srh; 5148 void *srh_end, *srh_tlvs, *ptr; 5149 struct ipv6hdr *hdr; 5150 int srhoff = 0; 5151 int ret; 5152 5153 if (unlikely(srh == NULL)) 5154 return -EINVAL; 5155 5156 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) + 5157 ((srh->first_segment + 1) << 4)); 5158 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) + 5159 srh_state->hdrlen); 5160 ptr = skb->data + offset; 5161 5162 if (unlikely(ptr < srh_tlvs || ptr > srh_end)) 5163 return -EFAULT; 5164 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end)) 5165 return -EFAULT; 5166 5167 if (len > 0) { 5168 ret = skb_cow_head(skb, len); 5169 if (unlikely(ret < 0)) 5170 return ret; 5171 5172 ret = bpf_skb_net_hdr_push(skb, offset, len); 5173 } else { 5174 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len); 5175 } 5176 5177 bpf_compute_data_pointers(skb); 5178 if (unlikely(ret < 0)) 5179 return ret; 5180 5181 hdr = (struct ipv6hdr *)skb->data; 5182 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 5183 5184 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 5185 return -EINVAL; 5186 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5187 srh_state->hdrlen += len; 5188 srh_state->valid = false; 5189 return 0; 5190 } 5191 5192 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = { 5193 .func = bpf_lwt_seg6_adjust_srh, 5194 .gpl_only = false, 5195 .ret_type = RET_INTEGER, 5196 .arg1_type = ARG_PTR_TO_CTX, 5197 .arg2_type = ARG_ANYTHING, 5198 .arg3_type = ARG_ANYTHING, 5199 }; 5200 #endif /* CONFIG_IPV6_SEG6_BPF */ 5201 5202 #ifdef CONFIG_INET 5203 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple, 5204 int dif, int sdif, u8 family, u8 proto) 5205 { 5206 bool refcounted = false; 5207 struct sock *sk = NULL; 5208 5209 if (family == AF_INET) { 5210 __be32 src4 = tuple->ipv4.saddr; 5211 __be32 dst4 = tuple->ipv4.daddr; 5212 5213 if (proto == IPPROTO_TCP) 5214 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0, 5215 src4, tuple->ipv4.sport, 5216 dst4, tuple->ipv4.dport, 5217 dif, sdif, &refcounted); 5218 else 5219 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport, 5220 dst4, tuple->ipv4.dport, 5221 dif, sdif, &udp_table, NULL); 5222 #if IS_ENABLED(CONFIG_IPV6) 5223 } else { 5224 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr; 5225 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr; 5226 5227 if (proto == IPPROTO_TCP) 5228 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0, 5229 src6, tuple->ipv6.sport, 5230 dst6, ntohs(tuple->ipv6.dport), 5231 dif, sdif, &refcounted); 5232 else if (likely(ipv6_bpf_stub)) 5233 sk = ipv6_bpf_stub->udp6_lib_lookup(net, 5234 src6, tuple->ipv6.sport, 5235 dst6, tuple->ipv6.dport, 5236 dif, sdif, 5237 &udp_table, NULL); 5238 #endif 5239 } 5240 5241 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) { 5242 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 5243 sk = NULL; 5244 } 5245 return sk; 5246 } 5247 5248 /* bpf_skc_lookup performs the core lookup for different types of sockets, 5249 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE. 5250 * Returns the socket as an 'unsigned long' to simplify the casting in the 5251 * callers to satisfy BPF_CALL declarations. 5252 */ 5253 static struct sock * 5254 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5255 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 5256 u64 flags) 5257 { 5258 struct sock *sk = NULL; 5259 u8 family = AF_UNSPEC; 5260 struct net *net; 5261 int sdif; 5262 5263 if (len == sizeof(tuple->ipv4)) 5264 family = AF_INET; 5265 else if (len == sizeof(tuple->ipv6)) 5266 family = AF_INET6; 5267 else 5268 return NULL; 5269 5270 if (unlikely(family == AF_UNSPEC || flags || 5271 !((s32)netns_id < 0 || netns_id <= S32_MAX))) 5272 goto out; 5273 5274 if (family == AF_INET) 5275 sdif = inet_sdif(skb); 5276 else 5277 sdif = inet6_sdif(skb); 5278 5279 if ((s32)netns_id < 0) { 5280 net = caller_net; 5281 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 5282 } else { 5283 net = get_net_ns_by_id(caller_net, netns_id); 5284 if (unlikely(!net)) 5285 goto out; 5286 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 5287 put_net(net); 5288 } 5289 5290 out: 5291 return sk; 5292 } 5293 5294 static struct sock * 5295 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5296 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 5297 u64 flags) 5298 { 5299 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net, 5300 ifindex, proto, netns_id, flags); 5301 5302 if (sk) { 5303 sk = sk_to_full_sk(sk); 5304 if (!sk_fullsock(sk)) { 5305 if (!sock_flag(sk, SOCK_RCU_FREE)) 5306 sock_gen_put(sk); 5307 return NULL; 5308 } 5309 } 5310 5311 return sk; 5312 } 5313 5314 static struct sock * 5315 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5316 u8 proto, u64 netns_id, u64 flags) 5317 { 5318 struct net *caller_net; 5319 int ifindex; 5320 5321 if (skb->dev) { 5322 caller_net = dev_net(skb->dev); 5323 ifindex = skb->dev->ifindex; 5324 } else { 5325 caller_net = sock_net(skb->sk); 5326 ifindex = 0; 5327 } 5328 5329 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto, 5330 netns_id, flags); 5331 } 5332 5333 static struct sock * 5334 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5335 u8 proto, u64 netns_id, u64 flags) 5336 { 5337 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id, 5338 flags); 5339 5340 if (sk) { 5341 sk = sk_to_full_sk(sk); 5342 if (!sk_fullsock(sk)) { 5343 if (!sock_flag(sk, SOCK_RCU_FREE)) 5344 sock_gen_put(sk); 5345 return NULL; 5346 } 5347 } 5348 5349 return sk; 5350 } 5351 5352 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb, 5353 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 5354 { 5355 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP, 5356 netns_id, flags); 5357 } 5358 5359 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = { 5360 .func = bpf_skc_lookup_tcp, 5361 .gpl_only = false, 5362 .pkt_access = true, 5363 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 5364 .arg1_type = ARG_PTR_TO_CTX, 5365 .arg2_type = ARG_PTR_TO_MEM, 5366 .arg3_type = ARG_CONST_SIZE, 5367 .arg4_type = ARG_ANYTHING, 5368 .arg5_type = ARG_ANYTHING, 5369 }; 5370 5371 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb, 5372 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 5373 { 5374 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, 5375 netns_id, flags); 5376 } 5377 5378 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = { 5379 .func = bpf_sk_lookup_tcp, 5380 .gpl_only = false, 5381 .pkt_access = true, 5382 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5383 .arg1_type = ARG_PTR_TO_CTX, 5384 .arg2_type = ARG_PTR_TO_MEM, 5385 .arg3_type = ARG_CONST_SIZE, 5386 .arg4_type = ARG_ANYTHING, 5387 .arg5_type = ARG_ANYTHING, 5388 }; 5389 5390 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb, 5391 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 5392 { 5393 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, 5394 netns_id, flags); 5395 } 5396 5397 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = { 5398 .func = bpf_sk_lookup_udp, 5399 .gpl_only = false, 5400 .pkt_access = true, 5401 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5402 .arg1_type = ARG_PTR_TO_CTX, 5403 .arg2_type = ARG_PTR_TO_MEM, 5404 .arg3_type = ARG_CONST_SIZE, 5405 .arg4_type = ARG_ANYTHING, 5406 .arg5_type = ARG_ANYTHING, 5407 }; 5408 5409 BPF_CALL_1(bpf_sk_release, struct sock *, sk) 5410 { 5411 if (!sock_flag(sk, SOCK_RCU_FREE)) 5412 sock_gen_put(sk); 5413 return 0; 5414 } 5415 5416 static const struct bpf_func_proto bpf_sk_release_proto = { 5417 .func = bpf_sk_release, 5418 .gpl_only = false, 5419 .ret_type = RET_INTEGER, 5420 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 5421 }; 5422 5423 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx, 5424 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 5425 { 5426 struct net *caller_net = dev_net(ctx->rxq->dev); 5427 int ifindex = ctx->rxq->dev->ifindex; 5428 5429 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 5430 ifindex, IPPROTO_UDP, netns_id, 5431 flags); 5432 } 5433 5434 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = { 5435 .func = bpf_xdp_sk_lookup_udp, 5436 .gpl_only = false, 5437 .pkt_access = true, 5438 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5439 .arg1_type = ARG_PTR_TO_CTX, 5440 .arg2_type = ARG_PTR_TO_MEM, 5441 .arg3_type = ARG_CONST_SIZE, 5442 .arg4_type = ARG_ANYTHING, 5443 .arg5_type = ARG_ANYTHING, 5444 }; 5445 5446 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx, 5447 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 5448 { 5449 struct net *caller_net = dev_net(ctx->rxq->dev); 5450 int ifindex = ctx->rxq->dev->ifindex; 5451 5452 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net, 5453 ifindex, IPPROTO_TCP, netns_id, 5454 flags); 5455 } 5456 5457 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = { 5458 .func = bpf_xdp_skc_lookup_tcp, 5459 .gpl_only = false, 5460 .pkt_access = true, 5461 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 5462 .arg1_type = ARG_PTR_TO_CTX, 5463 .arg2_type = ARG_PTR_TO_MEM, 5464 .arg3_type = ARG_CONST_SIZE, 5465 .arg4_type = ARG_ANYTHING, 5466 .arg5_type = ARG_ANYTHING, 5467 }; 5468 5469 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx, 5470 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 5471 { 5472 struct net *caller_net = dev_net(ctx->rxq->dev); 5473 int ifindex = ctx->rxq->dev->ifindex; 5474 5475 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 5476 ifindex, IPPROTO_TCP, netns_id, 5477 flags); 5478 } 5479 5480 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = { 5481 .func = bpf_xdp_sk_lookup_tcp, 5482 .gpl_only = false, 5483 .pkt_access = true, 5484 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5485 .arg1_type = ARG_PTR_TO_CTX, 5486 .arg2_type = ARG_PTR_TO_MEM, 5487 .arg3_type = ARG_CONST_SIZE, 5488 .arg4_type = ARG_ANYTHING, 5489 .arg5_type = ARG_ANYTHING, 5490 }; 5491 5492 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 5493 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 5494 { 5495 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, 5496 sock_net(ctx->sk), 0, 5497 IPPROTO_TCP, netns_id, flags); 5498 } 5499 5500 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = { 5501 .func = bpf_sock_addr_skc_lookup_tcp, 5502 .gpl_only = false, 5503 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 5504 .arg1_type = ARG_PTR_TO_CTX, 5505 .arg2_type = ARG_PTR_TO_MEM, 5506 .arg3_type = ARG_CONST_SIZE, 5507 .arg4_type = ARG_ANYTHING, 5508 .arg5_type = ARG_ANYTHING, 5509 }; 5510 5511 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 5512 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 5513 { 5514 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 5515 sock_net(ctx->sk), 0, IPPROTO_TCP, 5516 netns_id, flags); 5517 } 5518 5519 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = { 5520 .func = bpf_sock_addr_sk_lookup_tcp, 5521 .gpl_only = false, 5522 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5523 .arg1_type = ARG_PTR_TO_CTX, 5524 .arg2_type = ARG_PTR_TO_MEM, 5525 .arg3_type = ARG_CONST_SIZE, 5526 .arg4_type = ARG_ANYTHING, 5527 .arg5_type = ARG_ANYTHING, 5528 }; 5529 5530 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx, 5531 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 5532 { 5533 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 5534 sock_net(ctx->sk), 0, IPPROTO_UDP, 5535 netns_id, flags); 5536 } 5537 5538 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = { 5539 .func = bpf_sock_addr_sk_lookup_udp, 5540 .gpl_only = false, 5541 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5542 .arg1_type = ARG_PTR_TO_CTX, 5543 .arg2_type = ARG_PTR_TO_MEM, 5544 .arg3_type = ARG_CONST_SIZE, 5545 .arg4_type = ARG_ANYTHING, 5546 .arg5_type = ARG_ANYTHING, 5547 }; 5548 5549 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 5550 struct bpf_insn_access_aux *info) 5551 { 5552 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, 5553 icsk_retransmits)) 5554 return false; 5555 5556 if (off % size != 0) 5557 return false; 5558 5559 switch (off) { 5560 case offsetof(struct bpf_tcp_sock, bytes_received): 5561 case offsetof(struct bpf_tcp_sock, bytes_acked): 5562 return size == sizeof(__u64); 5563 default: 5564 return size == sizeof(__u32); 5565 } 5566 } 5567 5568 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 5569 const struct bpf_insn *si, 5570 struct bpf_insn *insn_buf, 5571 struct bpf_prog *prog, u32 *target_size) 5572 { 5573 struct bpf_insn *insn = insn_buf; 5574 5575 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \ 5576 do { \ 5577 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, FIELD) > \ 5578 FIELD_SIZEOF(struct bpf_tcp_sock, FIELD)); \ 5579 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\ 5580 si->dst_reg, si->src_reg, \ 5581 offsetof(struct tcp_sock, FIELD)); \ 5582 } while (0) 5583 5584 #define BPF_INET_SOCK_GET_COMMON(FIELD) \ 5585 do { \ 5586 BUILD_BUG_ON(FIELD_SIZEOF(struct inet_connection_sock, \ 5587 FIELD) > \ 5588 FIELD_SIZEOF(struct bpf_tcp_sock, FIELD)); \ 5589 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 5590 struct inet_connection_sock, \ 5591 FIELD), \ 5592 si->dst_reg, si->src_reg, \ 5593 offsetof( \ 5594 struct inet_connection_sock, \ 5595 FIELD)); \ 5596 } while (0) 5597 5598 if (insn > insn_buf) 5599 return insn - insn_buf; 5600 5601 switch (si->off) { 5602 case offsetof(struct bpf_tcp_sock, rtt_min): 5603 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) != 5604 sizeof(struct minmax)); 5605 BUILD_BUG_ON(sizeof(struct minmax) < 5606 sizeof(struct minmax_sample)); 5607 5608 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 5609 offsetof(struct tcp_sock, rtt_min) + 5610 offsetof(struct minmax_sample, v)); 5611 break; 5612 case offsetof(struct bpf_tcp_sock, snd_cwnd): 5613 BPF_TCP_SOCK_GET_COMMON(snd_cwnd); 5614 break; 5615 case offsetof(struct bpf_tcp_sock, srtt_us): 5616 BPF_TCP_SOCK_GET_COMMON(srtt_us); 5617 break; 5618 case offsetof(struct bpf_tcp_sock, snd_ssthresh): 5619 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh); 5620 break; 5621 case offsetof(struct bpf_tcp_sock, rcv_nxt): 5622 BPF_TCP_SOCK_GET_COMMON(rcv_nxt); 5623 break; 5624 case offsetof(struct bpf_tcp_sock, snd_nxt): 5625 BPF_TCP_SOCK_GET_COMMON(snd_nxt); 5626 break; 5627 case offsetof(struct bpf_tcp_sock, snd_una): 5628 BPF_TCP_SOCK_GET_COMMON(snd_una); 5629 break; 5630 case offsetof(struct bpf_tcp_sock, mss_cache): 5631 BPF_TCP_SOCK_GET_COMMON(mss_cache); 5632 break; 5633 case offsetof(struct bpf_tcp_sock, ecn_flags): 5634 BPF_TCP_SOCK_GET_COMMON(ecn_flags); 5635 break; 5636 case offsetof(struct bpf_tcp_sock, rate_delivered): 5637 BPF_TCP_SOCK_GET_COMMON(rate_delivered); 5638 break; 5639 case offsetof(struct bpf_tcp_sock, rate_interval_us): 5640 BPF_TCP_SOCK_GET_COMMON(rate_interval_us); 5641 break; 5642 case offsetof(struct bpf_tcp_sock, packets_out): 5643 BPF_TCP_SOCK_GET_COMMON(packets_out); 5644 break; 5645 case offsetof(struct bpf_tcp_sock, retrans_out): 5646 BPF_TCP_SOCK_GET_COMMON(retrans_out); 5647 break; 5648 case offsetof(struct bpf_tcp_sock, total_retrans): 5649 BPF_TCP_SOCK_GET_COMMON(total_retrans); 5650 break; 5651 case offsetof(struct bpf_tcp_sock, segs_in): 5652 BPF_TCP_SOCK_GET_COMMON(segs_in); 5653 break; 5654 case offsetof(struct bpf_tcp_sock, data_segs_in): 5655 BPF_TCP_SOCK_GET_COMMON(data_segs_in); 5656 break; 5657 case offsetof(struct bpf_tcp_sock, segs_out): 5658 BPF_TCP_SOCK_GET_COMMON(segs_out); 5659 break; 5660 case offsetof(struct bpf_tcp_sock, data_segs_out): 5661 BPF_TCP_SOCK_GET_COMMON(data_segs_out); 5662 break; 5663 case offsetof(struct bpf_tcp_sock, lost_out): 5664 BPF_TCP_SOCK_GET_COMMON(lost_out); 5665 break; 5666 case offsetof(struct bpf_tcp_sock, sacked_out): 5667 BPF_TCP_SOCK_GET_COMMON(sacked_out); 5668 break; 5669 case offsetof(struct bpf_tcp_sock, bytes_received): 5670 BPF_TCP_SOCK_GET_COMMON(bytes_received); 5671 break; 5672 case offsetof(struct bpf_tcp_sock, bytes_acked): 5673 BPF_TCP_SOCK_GET_COMMON(bytes_acked); 5674 break; 5675 case offsetof(struct bpf_tcp_sock, dsack_dups): 5676 BPF_TCP_SOCK_GET_COMMON(dsack_dups); 5677 break; 5678 case offsetof(struct bpf_tcp_sock, delivered): 5679 BPF_TCP_SOCK_GET_COMMON(delivered); 5680 break; 5681 case offsetof(struct bpf_tcp_sock, delivered_ce): 5682 BPF_TCP_SOCK_GET_COMMON(delivered_ce); 5683 break; 5684 case offsetof(struct bpf_tcp_sock, icsk_retransmits): 5685 BPF_INET_SOCK_GET_COMMON(icsk_retransmits); 5686 break; 5687 } 5688 5689 return insn - insn_buf; 5690 } 5691 5692 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk) 5693 { 5694 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 5695 return (unsigned long)sk; 5696 5697 return (unsigned long)NULL; 5698 } 5699 5700 const struct bpf_func_proto bpf_tcp_sock_proto = { 5701 .func = bpf_tcp_sock, 5702 .gpl_only = false, 5703 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL, 5704 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 5705 }; 5706 5707 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk) 5708 { 5709 sk = sk_to_full_sk(sk); 5710 5711 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE)) 5712 return (unsigned long)sk; 5713 5714 return (unsigned long)NULL; 5715 } 5716 5717 static const struct bpf_func_proto bpf_get_listener_sock_proto = { 5718 .func = bpf_get_listener_sock, 5719 .gpl_only = false, 5720 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5721 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 5722 }; 5723 5724 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb) 5725 { 5726 unsigned int iphdr_len; 5727 5728 if (skb->protocol == cpu_to_be16(ETH_P_IP)) 5729 iphdr_len = sizeof(struct iphdr); 5730 else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) 5731 iphdr_len = sizeof(struct ipv6hdr); 5732 else 5733 return 0; 5734 5735 if (skb_headlen(skb) < iphdr_len) 5736 return 0; 5737 5738 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len)) 5739 return 0; 5740 5741 return INET_ECN_set_ce(skb); 5742 } 5743 5744 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 5745 struct bpf_insn_access_aux *info) 5746 { 5747 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id)) 5748 return false; 5749 5750 if (off % size != 0) 5751 return false; 5752 5753 switch (off) { 5754 default: 5755 return size == sizeof(__u32); 5756 } 5757 } 5758 5759 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 5760 const struct bpf_insn *si, 5761 struct bpf_insn *insn_buf, 5762 struct bpf_prog *prog, u32 *target_size) 5763 { 5764 struct bpf_insn *insn = insn_buf; 5765 5766 #define BPF_XDP_SOCK_GET(FIELD) \ 5767 do { \ 5768 BUILD_BUG_ON(FIELD_SIZEOF(struct xdp_sock, FIELD) > \ 5769 FIELD_SIZEOF(struct bpf_xdp_sock, FIELD)); \ 5770 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\ 5771 si->dst_reg, si->src_reg, \ 5772 offsetof(struct xdp_sock, FIELD)); \ 5773 } while (0) 5774 5775 switch (si->off) { 5776 case offsetof(struct bpf_xdp_sock, queue_id): 5777 BPF_XDP_SOCK_GET(queue_id); 5778 break; 5779 } 5780 5781 return insn - insn_buf; 5782 } 5783 5784 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = { 5785 .func = bpf_skb_ecn_set_ce, 5786 .gpl_only = false, 5787 .ret_type = RET_INTEGER, 5788 .arg1_type = ARG_PTR_TO_CTX, 5789 }; 5790 5791 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 5792 struct tcphdr *, th, u32, th_len) 5793 { 5794 #ifdef CONFIG_SYN_COOKIES 5795 u32 cookie; 5796 int ret; 5797 5798 if (unlikely(th_len < sizeof(*th))) 5799 return -EINVAL; 5800 5801 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */ 5802 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 5803 return -EINVAL; 5804 5805 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies) 5806 return -EINVAL; 5807 5808 if (!th->ack || th->rst || th->syn) 5809 return -ENOENT; 5810 5811 if (tcp_synq_no_recent_overflow(sk)) 5812 return -ENOENT; 5813 5814 cookie = ntohl(th->ack_seq) - 1; 5815 5816 switch (sk->sk_family) { 5817 case AF_INET: 5818 if (unlikely(iph_len < sizeof(struct iphdr))) 5819 return -EINVAL; 5820 5821 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie); 5822 break; 5823 5824 #if IS_BUILTIN(CONFIG_IPV6) 5825 case AF_INET6: 5826 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 5827 return -EINVAL; 5828 5829 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie); 5830 break; 5831 #endif /* CONFIG_IPV6 */ 5832 5833 default: 5834 return -EPROTONOSUPPORT; 5835 } 5836 5837 if (ret > 0) 5838 return 0; 5839 5840 return -ENOENT; 5841 #else 5842 return -ENOTSUPP; 5843 #endif 5844 } 5845 5846 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = { 5847 .func = bpf_tcp_check_syncookie, 5848 .gpl_only = true, 5849 .pkt_access = true, 5850 .ret_type = RET_INTEGER, 5851 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 5852 .arg2_type = ARG_PTR_TO_MEM, 5853 .arg3_type = ARG_CONST_SIZE, 5854 .arg4_type = ARG_PTR_TO_MEM, 5855 .arg5_type = ARG_CONST_SIZE, 5856 }; 5857 5858 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 5859 struct tcphdr *, th, u32, th_len) 5860 { 5861 #ifdef CONFIG_SYN_COOKIES 5862 u32 cookie; 5863 u16 mss; 5864 5865 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4)) 5866 return -EINVAL; 5867 5868 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 5869 return -EINVAL; 5870 5871 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies) 5872 return -ENOENT; 5873 5874 if (!th->syn || th->ack || th->fin || th->rst) 5875 return -EINVAL; 5876 5877 if (unlikely(iph_len < sizeof(struct iphdr))) 5878 return -EINVAL; 5879 5880 /* Both struct iphdr and struct ipv6hdr have the version field at the 5881 * same offset so we can cast to the shorter header (struct iphdr). 5882 */ 5883 switch (((struct iphdr *)iph)->version) { 5884 case 4: 5885 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only) 5886 return -EINVAL; 5887 5888 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie); 5889 break; 5890 5891 #if IS_BUILTIN(CONFIG_IPV6) 5892 case 6: 5893 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 5894 return -EINVAL; 5895 5896 if (sk->sk_family != AF_INET6) 5897 return -EINVAL; 5898 5899 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie); 5900 break; 5901 #endif /* CONFIG_IPV6 */ 5902 5903 default: 5904 return -EPROTONOSUPPORT; 5905 } 5906 if (mss == 0) 5907 return -ENOENT; 5908 5909 return cookie | ((u64)mss << 32); 5910 #else 5911 return -EOPNOTSUPP; 5912 #endif /* CONFIG_SYN_COOKIES */ 5913 } 5914 5915 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = { 5916 .func = bpf_tcp_gen_syncookie, 5917 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */ 5918 .pkt_access = true, 5919 .ret_type = RET_INTEGER, 5920 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 5921 .arg2_type = ARG_PTR_TO_MEM, 5922 .arg3_type = ARG_CONST_SIZE, 5923 .arg4_type = ARG_PTR_TO_MEM, 5924 .arg5_type = ARG_CONST_SIZE, 5925 }; 5926 5927 #endif /* CONFIG_INET */ 5928 5929 bool bpf_helper_changes_pkt_data(void *func) 5930 { 5931 if (func == bpf_skb_vlan_push || 5932 func == bpf_skb_vlan_pop || 5933 func == bpf_skb_store_bytes || 5934 func == bpf_skb_change_proto || 5935 func == bpf_skb_change_head || 5936 func == sk_skb_change_head || 5937 func == bpf_skb_change_tail || 5938 func == sk_skb_change_tail || 5939 func == bpf_skb_adjust_room || 5940 func == bpf_skb_pull_data || 5941 func == sk_skb_pull_data || 5942 func == bpf_clone_redirect || 5943 func == bpf_l3_csum_replace || 5944 func == bpf_l4_csum_replace || 5945 func == bpf_xdp_adjust_head || 5946 func == bpf_xdp_adjust_meta || 5947 func == bpf_msg_pull_data || 5948 func == bpf_msg_push_data || 5949 func == bpf_msg_pop_data || 5950 func == bpf_xdp_adjust_tail || 5951 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5952 func == bpf_lwt_seg6_store_bytes || 5953 func == bpf_lwt_seg6_adjust_srh || 5954 func == bpf_lwt_seg6_action || 5955 #endif 5956 func == bpf_lwt_in_push_encap || 5957 func == bpf_lwt_xmit_push_encap) 5958 return true; 5959 5960 return false; 5961 } 5962 5963 static const struct bpf_func_proto * 5964 bpf_base_func_proto(enum bpf_func_id func_id) 5965 { 5966 switch (func_id) { 5967 case BPF_FUNC_map_lookup_elem: 5968 return &bpf_map_lookup_elem_proto; 5969 case BPF_FUNC_map_update_elem: 5970 return &bpf_map_update_elem_proto; 5971 case BPF_FUNC_map_delete_elem: 5972 return &bpf_map_delete_elem_proto; 5973 case BPF_FUNC_map_push_elem: 5974 return &bpf_map_push_elem_proto; 5975 case BPF_FUNC_map_pop_elem: 5976 return &bpf_map_pop_elem_proto; 5977 case BPF_FUNC_map_peek_elem: 5978 return &bpf_map_peek_elem_proto; 5979 case BPF_FUNC_get_prandom_u32: 5980 return &bpf_get_prandom_u32_proto; 5981 case BPF_FUNC_get_smp_processor_id: 5982 return &bpf_get_raw_smp_processor_id_proto; 5983 case BPF_FUNC_get_numa_node_id: 5984 return &bpf_get_numa_node_id_proto; 5985 case BPF_FUNC_tail_call: 5986 return &bpf_tail_call_proto; 5987 case BPF_FUNC_ktime_get_ns: 5988 return &bpf_ktime_get_ns_proto; 5989 default: 5990 break; 5991 } 5992 5993 if (!capable(CAP_SYS_ADMIN)) 5994 return NULL; 5995 5996 switch (func_id) { 5997 case BPF_FUNC_spin_lock: 5998 return &bpf_spin_lock_proto; 5999 case BPF_FUNC_spin_unlock: 6000 return &bpf_spin_unlock_proto; 6001 case BPF_FUNC_trace_printk: 6002 return bpf_get_trace_printk_proto(); 6003 default: 6004 return NULL; 6005 } 6006 } 6007 6008 static const struct bpf_func_proto * 6009 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6010 { 6011 switch (func_id) { 6012 /* inet and inet6 sockets are created in a process 6013 * context so there is always a valid uid/gid 6014 */ 6015 case BPF_FUNC_get_current_uid_gid: 6016 return &bpf_get_current_uid_gid_proto; 6017 case BPF_FUNC_get_local_storage: 6018 return &bpf_get_local_storage_proto; 6019 default: 6020 return bpf_base_func_proto(func_id); 6021 } 6022 } 6023 6024 static const struct bpf_func_proto * 6025 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6026 { 6027 switch (func_id) { 6028 /* inet and inet6 sockets are created in a process 6029 * context so there is always a valid uid/gid 6030 */ 6031 case BPF_FUNC_get_current_uid_gid: 6032 return &bpf_get_current_uid_gid_proto; 6033 case BPF_FUNC_bind: 6034 switch (prog->expected_attach_type) { 6035 case BPF_CGROUP_INET4_CONNECT: 6036 case BPF_CGROUP_INET6_CONNECT: 6037 return &bpf_bind_proto; 6038 default: 6039 return NULL; 6040 } 6041 case BPF_FUNC_get_socket_cookie: 6042 return &bpf_get_socket_cookie_sock_addr_proto; 6043 case BPF_FUNC_get_local_storage: 6044 return &bpf_get_local_storage_proto; 6045 #ifdef CONFIG_INET 6046 case BPF_FUNC_sk_lookup_tcp: 6047 return &bpf_sock_addr_sk_lookup_tcp_proto; 6048 case BPF_FUNC_sk_lookup_udp: 6049 return &bpf_sock_addr_sk_lookup_udp_proto; 6050 case BPF_FUNC_sk_release: 6051 return &bpf_sk_release_proto; 6052 case BPF_FUNC_skc_lookup_tcp: 6053 return &bpf_sock_addr_skc_lookup_tcp_proto; 6054 #endif /* CONFIG_INET */ 6055 case BPF_FUNC_sk_storage_get: 6056 return &bpf_sk_storage_get_proto; 6057 case BPF_FUNC_sk_storage_delete: 6058 return &bpf_sk_storage_delete_proto; 6059 default: 6060 return bpf_base_func_proto(func_id); 6061 } 6062 } 6063 6064 static const struct bpf_func_proto * 6065 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6066 { 6067 switch (func_id) { 6068 case BPF_FUNC_skb_load_bytes: 6069 return &bpf_skb_load_bytes_proto; 6070 case BPF_FUNC_skb_load_bytes_relative: 6071 return &bpf_skb_load_bytes_relative_proto; 6072 case BPF_FUNC_get_socket_cookie: 6073 return &bpf_get_socket_cookie_proto; 6074 case BPF_FUNC_get_socket_uid: 6075 return &bpf_get_socket_uid_proto; 6076 case BPF_FUNC_perf_event_output: 6077 return &bpf_skb_event_output_proto; 6078 default: 6079 return bpf_base_func_proto(func_id); 6080 } 6081 } 6082 6083 const struct bpf_func_proto bpf_sk_storage_get_proto __weak; 6084 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak; 6085 6086 static const struct bpf_func_proto * 6087 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6088 { 6089 switch (func_id) { 6090 case BPF_FUNC_get_local_storage: 6091 return &bpf_get_local_storage_proto; 6092 case BPF_FUNC_sk_fullsock: 6093 return &bpf_sk_fullsock_proto; 6094 case BPF_FUNC_sk_storage_get: 6095 return &bpf_sk_storage_get_proto; 6096 case BPF_FUNC_sk_storage_delete: 6097 return &bpf_sk_storage_delete_proto; 6098 case BPF_FUNC_perf_event_output: 6099 return &bpf_skb_event_output_proto; 6100 #ifdef CONFIG_SOCK_CGROUP_DATA 6101 case BPF_FUNC_skb_cgroup_id: 6102 return &bpf_skb_cgroup_id_proto; 6103 #endif 6104 #ifdef CONFIG_INET 6105 case BPF_FUNC_tcp_sock: 6106 return &bpf_tcp_sock_proto; 6107 case BPF_FUNC_get_listener_sock: 6108 return &bpf_get_listener_sock_proto; 6109 case BPF_FUNC_skb_ecn_set_ce: 6110 return &bpf_skb_ecn_set_ce_proto; 6111 #endif 6112 default: 6113 return sk_filter_func_proto(func_id, prog); 6114 } 6115 } 6116 6117 static const struct bpf_func_proto * 6118 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6119 { 6120 switch (func_id) { 6121 case BPF_FUNC_skb_store_bytes: 6122 return &bpf_skb_store_bytes_proto; 6123 case BPF_FUNC_skb_load_bytes: 6124 return &bpf_skb_load_bytes_proto; 6125 case BPF_FUNC_skb_load_bytes_relative: 6126 return &bpf_skb_load_bytes_relative_proto; 6127 case BPF_FUNC_skb_pull_data: 6128 return &bpf_skb_pull_data_proto; 6129 case BPF_FUNC_csum_diff: 6130 return &bpf_csum_diff_proto; 6131 case BPF_FUNC_csum_update: 6132 return &bpf_csum_update_proto; 6133 case BPF_FUNC_l3_csum_replace: 6134 return &bpf_l3_csum_replace_proto; 6135 case BPF_FUNC_l4_csum_replace: 6136 return &bpf_l4_csum_replace_proto; 6137 case BPF_FUNC_clone_redirect: 6138 return &bpf_clone_redirect_proto; 6139 case BPF_FUNC_get_cgroup_classid: 6140 return &bpf_get_cgroup_classid_proto; 6141 case BPF_FUNC_skb_vlan_push: 6142 return &bpf_skb_vlan_push_proto; 6143 case BPF_FUNC_skb_vlan_pop: 6144 return &bpf_skb_vlan_pop_proto; 6145 case BPF_FUNC_skb_change_proto: 6146 return &bpf_skb_change_proto_proto; 6147 case BPF_FUNC_skb_change_type: 6148 return &bpf_skb_change_type_proto; 6149 case BPF_FUNC_skb_adjust_room: 6150 return &bpf_skb_adjust_room_proto; 6151 case BPF_FUNC_skb_change_tail: 6152 return &bpf_skb_change_tail_proto; 6153 case BPF_FUNC_skb_get_tunnel_key: 6154 return &bpf_skb_get_tunnel_key_proto; 6155 case BPF_FUNC_skb_set_tunnel_key: 6156 return bpf_get_skb_set_tunnel_proto(func_id); 6157 case BPF_FUNC_skb_get_tunnel_opt: 6158 return &bpf_skb_get_tunnel_opt_proto; 6159 case BPF_FUNC_skb_set_tunnel_opt: 6160 return bpf_get_skb_set_tunnel_proto(func_id); 6161 case BPF_FUNC_redirect: 6162 return &bpf_redirect_proto; 6163 case BPF_FUNC_get_route_realm: 6164 return &bpf_get_route_realm_proto; 6165 case BPF_FUNC_get_hash_recalc: 6166 return &bpf_get_hash_recalc_proto; 6167 case BPF_FUNC_set_hash_invalid: 6168 return &bpf_set_hash_invalid_proto; 6169 case BPF_FUNC_set_hash: 6170 return &bpf_set_hash_proto; 6171 case BPF_FUNC_perf_event_output: 6172 return &bpf_skb_event_output_proto; 6173 case BPF_FUNC_get_smp_processor_id: 6174 return &bpf_get_smp_processor_id_proto; 6175 case BPF_FUNC_skb_under_cgroup: 6176 return &bpf_skb_under_cgroup_proto; 6177 case BPF_FUNC_get_socket_cookie: 6178 return &bpf_get_socket_cookie_proto; 6179 case BPF_FUNC_get_socket_uid: 6180 return &bpf_get_socket_uid_proto; 6181 case BPF_FUNC_fib_lookup: 6182 return &bpf_skb_fib_lookup_proto; 6183 case BPF_FUNC_sk_fullsock: 6184 return &bpf_sk_fullsock_proto; 6185 case BPF_FUNC_sk_storage_get: 6186 return &bpf_sk_storage_get_proto; 6187 case BPF_FUNC_sk_storage_delete: 6188 return &bpf_sk_storage_delete_proto; 6189 #ifdef CONFIG_XFRM 6190 case BPF_FUNC_skb_get_xfrm_state: 6191 return &bpf_skb_get_xfrm_state_proto; 6192 #endif 6193 #ifdef CONFIG_SOCK_CGROUP_DATA 6194 case BPF_FUNC_skb_cgroup_id: 6195 return &bpf_skb_cgroup_id_proto; 6196 case BPF_FUNC_skb_ancestor_cgroup_id: 6197 return &bpf_skb_ancestor_cgroup_id_proto; 6198 #endif 6199 #ifdef CONFIG_INET 6200 case BPF_FUNC_sk_lookup_tcp: 6201 return &bpf_sk_lookup_tcp_proto; 6202 case BPF_FUNC_sk_lookup_udp: 6203 return &bpf_sk_lookup_udp_proto; 6204 case BPF_FUNC_sk_release: 6205 return &bpf_sk_release_proto; 6206 case BPF_FUNC_tcp_sock: 6207 return &bpf_tcp_sock_proto; 6208 case BPF_FUNC_get_listener_sock: 6209 return &bpf_get_listener_sock_proto; 6210 case BPF_FUNC_skc_lookup_tcp: 6211 return &bpf_skc_lookup_tcp_proto; 6212 case BPF_FUNC_tcp_check_syncookie: 6213 return &bpf_tcp_check_syncookie_proto; 6214 case BPF_FUNC_skb_ecn_set_ce: 6215 return &bpf_skb_ecn_set_ce_proto; 6216 case BPF_FUNC_tcp_gen_syncookie: 6217 return &bpf_tcp_gen_syncookie_proto; 6218 #endif 6219 default: 6220 return bpf_base_func_proto(func_id); 6221 } 6222 } 6223 6224 static const struct bpf_func_proto * 6225 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6226 { 6227 switch (func_id) { 6228 case BPF_FUNC_perf_event_output: 6229 return &bpf_xdp_event_output_proto; 6230 case BPF_FUNC_get_smp_processor_id: 6231 return &bpf_get_smp_processor_id_proto; 6232 case BPF_FUNC_csum_diff: 6233 return &bpf_csum_diff_proto; 6234 case BPF_FUNC_xdp_adjust_head: 6235 return &bpf_xdp_adjust_head_proto; 6236 case BPF_FUNC_xdp_adjust_meta: 6237 return &bpf_xdp_adjust_meta_proto; 6238 case BPF_FUNC_redirect: 6239 return &bpf_xdp_redirect_proto; 6240 case BPF_FUNC_redirect_map: 6241 return &bpf_xdp_redirect_map_proto; 6242 case BPF_FUNC_xdp_adjust_tail: 6243 return &bpf_xdp_adjust_tail_proto; 6244 case BPF_FUNC_fib_lookup: 6245 return &bpf_xdp_fib_lookup_proto; 6246 #ifdef CONFIG_INET 6247 case BPF_FUNC_sk_lookup_udp: 6248 return &bpf_xdp_sk_lookup_udp_proto; 6249 case BPF_FUNC_sk_lookup_tcp: 6250 return &bpf_xdp_sk_lookup_tcp_proto; 6251 case BPF_FUNC_sk_release: 6252 return &bpf_sk_release_proto; 6253 case BPF_FUNC_skc_lookup_tcp: 6254 return &bpf_xdp_skc_lookup_tcp_proto; 6255 case BPF_FUNC_tcp_check_syncookie: 6256 return &bpf_tcp_check_syncookie_proto; 6257 case BPF_FUNC_tcp_gen_syncookie: 6258 return &bpf_tcp_gen_syncookie_proto; 6259 #endif 6260 default: 6261 return bpf_base_func_proto(func_id); 6262 } 6263 } 6264 6265 const struct bpf_func_proto bpf_sock_map_update_proto __weak; 6266 const struct bpf_func_proto bpf_sock_hash_update_proto __weak; 6267 6268 static const struct bpf_func_proto * 6269 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6270 { 6271 switch (func_id) { 6272 case BPF_FUNC_setsockopt: 6273 return &bpf_setsockopt_proto; 6274 case BPF_FUNC_getsockopt: 6275 return &bpf_getsockopt_proto; 6276 case BPF_FUNC_sock_ops_cb_flags_set: 6277 return &bpf_sock_ops_cb_flags_set_proto; 6278 case BPF_FUNC_sock_map_update: 6279 return &bpf_sock_map_update_proto; 6280 case BPF_FUNC_sock_hash_update: 6281 return &bpf_sock_hash_update_proto; 6282 case BPF_FUNC_get_socket_cookie: 6283 return &bpf_get_socket_cookie_sock_ops_proto; 6284 case BPF_FUNC_get_local_storage: 6285 return &bpf_get_local_storage_proto; 6286 case BPF_FUNC_perf_event_output: 6287 return &bpf_sockopt_event_output_proto; 6288 case BPF_FUNC_sk_storage_get: 6289 return &bpf_sk_storage_get_proto; 6290 case BPF_FUNC_sk_storage_delete: 6291 return &bpf_sk_storage_delete_proto; 6292 #ifdef CONFIG_INET 6293 case BPF_FUNC_tcp_sock: 6294 return &bpf_tcp_sock_proto; 6295 #endif /* CONFIG_INET */ 6296 default: 6297 return bpf_base_func_proto(func_id); 6298 } 6299 } 6300 6301 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak; 6302 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak; 6303 6304 static const struct bpf_func_proto * 6305 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6306 { 6307 switch (func_id) { 6308 case BPF_FUNC_msg_redirect_map: 6309 return &bpf_msg_redirect_map_proto; 6310 case BPF_FUNC_msg_redirect_hash: 6311 return &bpf_msg_redirect_hash_proto; 6312 case BPF_FUNC_msg_apply_bytes: 6313 return &bpf_msg_apply_bytes_proto; 6314 case BPF_FUNC_msg_cork_bytes: 6315 return &bpf_msg_cork_bytes_proto; 6316 case BPF_FUNC_msg_pull_data: 6317 return &bpf_msg_pull_data_proto; 6318 case BPF_FUNC_msg_push_data: 6319 return &bpf_msg_push_data_proto; 6320 case BPF_FUNC_msg_pop_data: 6321 return &bpf_msg_pop_data_proto; 6322 default: 6323 return bpf_base_func_proto(func_id); 6324 } 6325 } 6326 6327 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak; 6328 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak; 6329 6330 static const struct bpf_func_proto * 6331 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6332 { 6333 switch (func_id) { 6334 case BPF_FUNC_skb_store_bytes: 6335 return &bpf_skb_store_bytes_proto; 6336 case BPF_FUNC_skb_load_bytes: 6337 return &bpf_skb_load_bytes_proto; 6338 case BPF_FUNC_skb_pull_data: 6339 return &sk_skb_pull_data_proto; 6340 case BPF_FUNC_skb_change_tail: 6341 return &sk_skb_change_tail_proto; 6342 case BPF_FUNC_skb_change_head: 6343 return &sk_skb_change_head_proto; 6344 case BPF_FUNC_get_socket_cookie: 6345 return &bpf_get_socket_cookie_proto; 6346 case BPF_FUNC_get_socket_uid: 6347 return &bpf_get_socket_uid_proto; 6348 case BPF_FUNC_sk_redirect_map: 6349 return &bpf_sk_redirect_map_proto; 6350 case BPF_FUNC_sk_redirect_hash: 6351 return &bpf_sk_redirect_hash_proto; 6352 case BPF_FUNC_perf_event_output: 6353 return &bpf_skb_event_output_proto; 6354 #ifdef CONFIG_INET 6355 case BPF_FUNC_sk_lookup_tcp: 6356 return &bpf_sk_lookup_tcp_proto; 6357 case BPF_FUNC_sk_lookup_udp: 6358 return &bpf_sk_lookup_udp_proto; 6359 case BPF_FUNC_sk_release: 6360 return &bpf_sk_release_proto; 6361 case BPF_FUNC_skc_lookup_tcp: 6362 return &bpf_skc_lookup_tcp_proto; 6363 #endif 6364 default: 6365 return bpf_base_func_proto(func_id); 6366 } 6367 } 6368 6369 static const struct bpf_func_proto * 6370 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6371 { 6372 switch (func_id) { 6373 case BPF_FUNC_skb_load_bytes: 6374 return &bpf_flow_dissector_load_bytes_proto; 6375 default: 6376 return bpf_base_func_proto(func_id); 6377 } 6378 } 6379 6380 static const struct bpf_func_proto * 6381 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6382 { 6383 switch (func_id) { 6384 case BPF_FUNC_skb_load_bytes: 6385 return &bpf_skb_load_bytes_proto; 6386 case BPF_FUNC_skb_pull_data: 6387 return &bpf_skb_pull_data_proto; 6388 case BPF_FUNC_csum_diff: 6389 return &bpf_csum_diff_proto; 6390 case BPF_FUNC_get_cgroup_classid: 6391 return &bpf_get_cgroup_classid_proto; 6392 case BPF_FUNC_get_route_realm: 6393 return &bpf_get_route_realm_proto; 6394 case BPF_FUNC_get_hash_recalc: 6395 return &bpf_get_hash_recalc_proto; 6396 case BPF_FUNC_perf_event_output: 6397 return &bpf_skb_event_output_proto; 6398 case BPF_FUNC_get_smp_processor_id: 6399 return &bpf_get_smp_processor_id_proto; 6400 case BPF_FUNC_skb_under_cgroup: 6401 return &bpf_skb_under_cgroup_proto; 6402 default: 6403 return bpf_base_func_proto(func_id); 6404 } 6405 } 6406 6407 static const struct bpf_func_proto * 6408 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6409 { 6410 switch (func_id) { 6411 case BPF_FUNC_lwt_push_encap: 6412 return &bpf_lwt_in_push_encap_proto; 6413 default: 6414 return lwt_out_func_proto(func_id, prog); 6415 } 6416 } 6417 6418 static const struct bpf_func_proto * 6419 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6420 { 6421 switch (func_id) { 6422 case BPF_FUNC_skb_get_tunnel_key: 6423 return &bpf_skb_get_tunnel_key_proto; 6424 case BPF_FUNC_skb_set_tunnel_key: 6425 return bpf_get_skb_set_tunnel_proto(func_id); 6426 case BPF_FUNC_skb_get_tunnel_opt: 6427 return &bpf_skb_get_tunnel_opt_proto; 6428 case BPF_FUNC_skb_set_tunnel_opt: 6429 return bpf_get_skb_set_tunnel_proto(func_id); 6430 case BPF_FUNC_redirect: 6431 return &bpf_redirect_proto; 6432 case BPF_FUNC_clone_redirect: 6433 return &bpf_clone_redirect_proto; 6434 case BPF_FUNC_skb_change_tail: 6435 return &bpf_skb_change_tail_proto; 6436 case BPF_FUNC_skb_change_head: 6437 return &bpf_skb_change_head_proto; 6438 case BPF_FUNC_skb_store_bytes: 6439 return &bpf_skb_store_bytes_proto; 6440 case BPF_FUNC_csum_update: 6441 return &bpf_csum_update_proto; 6442 case BPF_FUNC_l3_csum_replace: 6443 return &bpf_l3_csum_replace_proto; 6444 case BPF_FUNC_l4_csum_replace: 6445 return &bpf_l4_csum_replace_proto; 6446 case BPF_FUNC_set_hash_invalid: 6447 return &bpf_set_hash_invalid_proto; 6448 case BPF_FUNC_lwt_push_encap: 6449 return &bpf_lwt_xmit_push_encap_proto; 6450 default: 6451 return lwt_out_func_proto(func_id, prog); 6452 } 6453 } 6454 6455 static const struct bpf_func_proto * 6456 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6457 { 6458 switch (func_id) { 6459 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6460 case BPF_FUNC_lwt_seg6_store_bytes: 6461 return &bpf_lwt_seg6_store_bytes_proto; 6462 case BPF_FUNC_lwt_seg6_action: 6463 return &bpf_lwt_seg6_action_proto; 6464 case BPF_FUNC_lwt_seg6_adjust_srh: 6465 return &bpf_lwt_seg6_adjust_srh_proto; 6466 #endif 6467 default: 6468 return lwt_out_func_proto(func_id, prog); 6469 } 6470 } 6471 6472 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type, 6473 const struct bpf_prog *prog, 6474 struct bpf_insn_access_aux *info) 6475 { 6476 const int size_default = sizeof(__u32); 6477 6478 if (off < 0 || off >= sizeof(struct __sk_buff)) 6479 return false; 6480 6481 /* The verifier guarantees that size > 0. */ 6482 if (off % size != 0) 6483 return false; 6484 6485 switch (off) { 6486 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 6487 if (off + size > offsetofend(struct __sk_buff, cb[4])) 6488 return false; 6489 break; 6490 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]): 6491 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]): 6492 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4): 6493 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4): 6494 case bpf_ctx_range(struct __sk_buff, data): 6495 case bpf_ctx_range(struct __sk_buff, data_meta): 6496 case bpf_ctx_range(struct __sk_buff, data_end): 6497 if (size != size_default) 6498 return false; 6499 break; 6500 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 6501 return false; 6502 case bpf_ctx_range(struct __sk_buff, tstamp): 6503 if (size != sizeof(__u64)) 6504 return false; 6505 break; 6506 case offsetof(struct __sk_buff, sk): 6507 if (type == BPF_WRITE || size != sizeof(__u64)) 6508 return false; 6509 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 6510 break; 6511 default: 6512 /* Only narrow read access allowed for now. */ 6513 if (type == BPF_WRITE) { 6514 if (size != size_default) 6515 return false; 6516 } else { 6517 bpf_ctx_record_field_size(info, size_default); 6518 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 6519 return false; 6520 } 6521 } 6522 6523 return true; 6524 } 6525 6526 static bool sk_filter_is_valid_access(int off, int size, 6527 enum bpf_access_type type, 6528 const struct bpf_prog *prog, 6529 struct bpf_insn_access_aux *info) 6530 { 6531 switch (off) { 6532 case bpf_ctx_range(struct __sk_buff, tc_classid): 6533 case bpf_ctx_range(struct __sk_buff, data): 6534 case bpf_ctx_range(struct __sk_buff, data_meta): 6535 case bpf_ctx_range(struct __sk_buff, data_end): 6536 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 6537 case bpf_ctx_range(struct __sk_buff, tstamp): 6538 case bpf_ctx_range(struct __sk_buff, wire_len): 6539 return false; 6540 } 6541 6542 if (type == BPF_WRITE) { 6543 switch (off) { 6544 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 6545 break; 6546 default: 6547 return false; 6548 } 6549 } 6550 6551 return bpf_skb_is_valid_access(off, size, type, prog, info); 6552 } 6553 6554 static bool cg_skb_is_valid_access(int off, int size, 6555 enum bpf_access_type type, 6556 const struct bpf_prog *prog, 6557 struct bpf_insn_access_aux *info) 6558 { 6559 switch (off) { 6560 case bpf_ctx_range(struct __sk_buff, tc_classid): 6561 case bpf_ctx_range(struct __sk_buff, data_meta): 6562 case bpf_ctx_range(struct __sk_buff, wire_len): 6563 return false; 6564 case bpf_ctx_range(struct __sk_buff, data): 6565 case bpf_ctx_range(struct __sk_buff, data_end): 6566 if (!capable(CAP_SYS_ADMIN)) 6567 return false; 6568 break; 6569 } 6570 6571 if (type == BPF_WRITE) { 6572 switch (off) { 6573 case bpf_ctx_range(struct __sk_buff, mark): 6574 case bpf_ctx_range(struct __sk_buff, priority): 6575 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 6576 break; 6577 case bpf_ctx_range(struct __sk_buff, tstamp): 6578 if (!capable(CAP_SYS_ADMIN)) 6579 return false; 6580 break; 6581 default: 6582 return false; 6583 } 6584 } 6585 6586 switch (off) { 6587 case bpf_ctx_range(struct __sk_buff, data): 6588 info->reg_type = PTR_TO_PACKET; 6589 break; 6590 case bpf_ctx_range(struct __sk_buff, data_end): 6591 info->reg_type = PTR_TO_PACKET_END; 6592 break; 6593 } 6594 6595 return bpf_skb_is_valid_access(off, size, type, prog, info); 6596 } 6597 6598 static bool lwt_is_valid_access(int off, int size, 6599 enum bpf_access_type type, 6600 const struct bpf_prog *prog, 6601 struct bpf_insn_access_aux *info) 6602 { 6603 switch (off) { 6604 case bpf_ctx_range(struct __sk_buff, tc_classid): 6605 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 6606 case bpf_ctx_range(struct __sk_buff, data_meta): 6607 case bpf_ctx_range(struct __sk_buff, tstamp): 6608 case bpf_ctx_range(struct __sk_buff, wire_len): 6609 return false; 6610 } 6611 6612 if (type == BPF_WRITE) { 6613 switch (off) { 6614 case bpf_ctx_range(struct __sk_buff, mark): 6615 case bpf_ctx_range(struct __sk_buff, priority): 6616 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 6617 break; 6618 default: 6619 return false; 6620 } 6621 } 6622 6623 switch (off) { 6624 case bpf_ctx_range(struct __sk_buff, data): 6625 info->reg_type = PTR_TO_PACKET; 6626 break; 6627 case bpf_ctx_range(struct __sk_buff, data_end): 6628 info->reg_type = PTR_TO_PACKET_END; 6629 break; 6630 } 6631 6632 return bpf_skb_is_valid_access(off, size, type, prog, info); 6633 } 6634 6635 /* Attach type specific accesses */ 6636 static bool __sock_filter_check_attach_type(int off, 6637 enum bpf_access_type access_type, 6638 enum bpf_attach_type attach_type) 6639 { 6640 switch (off) { 6641 case offsetof(struct bpf_sock, bound_dev_if): 6642 case offsetof(struct bpf_sock, mark): 6643 case offsetof(struct bpf_sock, priority): 6644 switch (attach_type) { 6645 case BPF_CGROUP_INET_SOCK_CREATE: 6646 goto full_access; 6647 default: 6648 return false; 6649 } 6650 case bpf_ctx_range(struct bpf_sock, src_ip4): 6651 switch (attach_type) { 6652 case BPF_CGROUP_INET4_POST_BIND: 6653 goto read_only; 6654 default: 6655 return false; 6656 } 6657 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 6658 switch (attach_type) { 6659 case BPF_CGROUP_INET6_POST_BIND: 6660 goto read_only; 6661 default: 6662 return false; 6663 } 6664 case bpf_ctx_range(struct bpf_sock, src_port): 6665 switch (attach_type) { 6666 case BPF_CGROUP_INET4_POST_BIND: 6667 case BPF_CGROUP_INET6_POST_BIND: 6668 goto read_only; 6669 default: 6670 return false; 6671 } 6672 } 6673 read_only: 6674 return access_type == BPF_READ; 6675 full_access: 6676 return true; 6677 } 6678 6679 bool bpf_sock_common_is_valid_access(int off, int size, 6680 enum bpf_access_type type, 6681 struct bpf_insn_access_aux *info) 6682 { 6683 switch (off) { 6684 case bpf_ctx_range_till(struct bpf_sock, type, priority): 6685 return false; 6686 default: 6687 return bpf_sock_is_valid_access(off, size, type, info); 6688 } 6689 } 6690 6691 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6692 struct bpf_insn_access_aux *info) 6693 { 6694 const int size_default = sizeof(__u32); 6695 6696 if (off < 0 || off >= sizeof(struct bpf_sock)) 6697 return false; 6698 if (off % size != 0) 6699 return false; 6700 6701 switch (off) { 6702 case offsetof(struct bpf_sock, state): 6703 case offsetof(struct bpf_sock, family): 6704 case offsetof(struct bpf_sock, type): 6705 case offsetof(struct bpf_sock, protocol): 6706 case offsetof(struct bpf_sock, dst_port): 6707 case offsetof(struct bpf_sock, src_port): 6708 case bpf_ctx_range(struct bpf_sock, src_ip4): 6709 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 6710 case bpf_ctx_range(struct bpf_sock, dst_ip4): 6711 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 6712 bpf_ctx_record_field_size(info, size_default); 6713 return bpf_ctx_narrow_access_ok(off, size, size_default); 6714 } 6715 6716 return size == size_default; 6717 } 6718 6719 static bool sock_filter_is_valid_access(int off, int size, 6720 enum bpf_access_type type, 6721 const struct bpf_prog *prog, 6722 struct bpf_insn_access_aux *info) 6723 { 6724 if (!bpf_sock_is_valid_access(off, size, type, info)) 6725 return false; 6726 return __sock_filter_check_attach_type(off, type, 6727 prog->expected_attach_type); 6728 } 6729 6730 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write, 6731 const struct bpf_prog *prog) 6732 { 6733 /* Neither direct read nor direct write requires any preliminary 6734 * action. 6735 */ 6736 return 0; 6737 } 6738 6739 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write, 6740 const struct bpf_prog *prog, int drop_verdict) 6741 { 6742 struct bpf_insn *insn = insn_buf; 6743 6744 if (!direct_write) 6745 return 0; 6746 6747 /* if (!skb->cloned) 6748 * goto start; 6749 * 6750 * (Fast-path, otherwise approximation that we might be 6751 * a clone, do the rest in helper.) 6752 */ 6753 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET()); 6754 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK); 6755 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7); 6756 6757 /* ret = bpf_skb_pull_data(skb, 0); */ 6758 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1); 6759 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2); 6760 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, 6761 BPF_FUNC_skb_pull_data); 6762 /* if (!ret) 6763 * goto restore; 6764 * return TC_ACT_SHOT; 6765 */ 6766 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2); 6767 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict); 6768 *insn++ = BPF_EXIT_INSN(); 6769 6770 /* restore: */ 6771 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6); 6772 /* start: */ 6773 *insn++ = prog->insnsi[0]; 6774 6775 return insn - insn_buf; 6776 } 6777 6778 static int bpf_gen_ld_abs(const struct bpf_insn *orig, 6779 struct bpf_insn *insn_buf) 6780 { 6781 bool indirect = BPF_MODE(orig->code) == BPF_IND; 6782 struct bpf_insn *insn = insn_buf; 6783 6784 /* We're guaranteed here that CTX is in R6. */ 6785 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX); 6786 if (!indirect) { 6787 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm); 6788 } else { 6789 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg); 6790 if (orig->imm) 6791 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm); 6792 } 6793 6794 switch (BPF_SIZE(orig->code)) { 6795 case BPF_B: 6796 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache); 6797 break; 6798 case BPF_H: 6799 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache); 6800 break; 6801 case BPF_W: 6802 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache); 6803 break; 6804 } 6805 6806 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2); 6807 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0); 6808 *insn++ = BPF_EXIT_INSN(); 6809 6810 return insn - insn_buf; 6811 } 6812 6813 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write, 6814 const struct bpf_prog *prog) 6815 { 6816 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT); 6817 } 6818 6819 static bool tc_cls_act_is_valid_access(int off, int size, 6820 enum bpf_access_type type, 6821 const struct bpf_prog *prog, 6822 struct bpf_insn_access_aux *info) 6823 { 6824 if (type == BPF_WRITE) { 6825 switch (off) { 6826 case bpf_ctx_range(struct __sk_buff, mark): 6827 case bpf_ctx_range(struct __sk_buff, tc_index): 6828 case bpf_ctx_range(struct __sk_buff, priority): 6829 case bpf_ctx_range(struct __sk_buff, tc_classid): 6830 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 6831 case bpf_ctx_range(struct __sk_buff, tstamp): 6832 case bpf_ctx_range(struct __sk_buff, queue_mapping): 6833 break; 6834 default: 6835 return false; 6836 } 6837 } 6838 6839 switch (off) { 6840 case bpf_ctx_range(struct __sk_buff, data): 6841 info->reg_type = PTR_TO_PACKET; 6842 break; 6843 case bpf_ctx_range(struct __sk_buff, data_meta): 6844 info->reg_type = PTR_TO_PACKET_META; 6845 break; 6846 case bpf_ctx_range(struct __sk_buff, data_end): 6847 info->reg_type = PTR_TO_PACKET_END; 6848 break; 6849 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 6850 return false; 6851 } 6852 6853 return bpf_skb_is_valid_access(off, size, type, prog, info); 6854 } 6855 6856 static bool __is_valid_xdp_access(int off, int size) 6857 { 6858 if (off < 0 || off >= sizeof(struct xdp_md)) 6859 return false; 6860 if (off % size != 0) 6861 return false; 6862 if (size != sizeof(__u32)) 6863 return false; 6864 6865 return true; 6866 } 6867 6868 static bool xdp_is_valid_access(int off, int size, 6869 enum bpf_access_type type, 6870 const struct bpf_prog *prog, 6871 struct bpf_insn_access_aux *info) 6872 { 6873 if (type == BPF_WRITE) { 6874 if (bpf_prog_is_dev_bound(prog->aux)) { 6875 switch (off) { 6876 case offsetof(struct xdp_md, rx_queue_index): 6877 return __is_valid_xdp_access(off, size); 6878 } 6879 } 6880 return false; 6881 } 6882 6883 switch (off) { 6884 case offsetof(struct xdp_md, data): 6885 info->reg_type = PTR_TO_PACKET; 6886 break; 6887 case offsetof(struct xdp_md, data_meta): 6888 info->reg_type = PTR_TO_PACKET_META; 6889 break; 6890 case offsetof(struct xdp_md, data_end): 6891 info->reg_type = PTR_TO_PACKET_END; 6892 break; 6893 } 6894 6895 return __is_valid_xdp_access(off, size); 6896 } 6897 6898 void bpf_warn_invalid_xdp_action(u32 act) 6899 { 6900 const u32 act_max = XDP_REDIRECT; 6901 6902 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n", 6903 act > act_max ? "Illegal" : "Driver unsupported", 6904 act); 6905 } 6906 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); 6907 6908 static bool sock_addr_is_valid_access(int off, int size, 6909 enum bpf_access_type type, 6910 const struct bpf_prog *prog, 6911 struct bpf_insn_access_aux *info) 6912 { 6913 const int size_default = sizeof(__u32); 6914 6915 if (off < 0 || off >= sizeof(struct bpf_sock_addr)) 6916 return false; 6917 if (off % size != 0) 6918 return false; 6919 6920 /* Disallow access to IPv6 fields from IPv4 contex and vise 6921 * versa. 6922 */ 6923 switch (off) { 6924 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 6925 switch (prog->expected_attach_type) { 6926 case BPF_CGROUP_INET4_BIND: 6927 case BPF_CGROUP_INET4_CONNECT: 6928 case BPF_CGROUP_UDP4_SENDMSG: 6929 case BPF_CGROUP_UDP4_RECVMSG: 6930 break; 6931 default: 6932 return false; 6933 } 6934 break; 6935 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 6936 switch (prog->expected_attach_type) { 6937 case BPF_CGROUP_INET6_BIND: 6938 case BPF_CGROUP_INET6_CONNECT: 6939 case BPF_CGROUP_UDP6_SENDMSG: 6940 case BPF_CGROUP_UDP6_RECVMSG: 6941 break; 6942 default: 6943 return false; 6944 } 6945 break; 6946 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 6947 switch (prog->expected_attach_type) { 6948 case BPF_CGROUP_UDP4_SENDMSG: 6949 break; 6950 default: 6951 return false; 6952 } 6953 break; 6954 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 6955 msg_src_ip6[3]): 6956 switch (prog->expected_attach_type) { 6957 case BPF_CGROUP_UDP6_SENDMSG: 6958 break; 6959 default: 6960 return false; 6961 } 6962 break; 6963 } 6964 6965 switch (off) { 6966 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 6967 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 6968 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 6969 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 6970 msg_src_ip6[3]): 6971 if (type == BPF_READ) { 6972 bpf_ctx_record_field_size(info, size_default); 6973 6974 if (bpf_ctx_wide_access_ok(off, size, 6975 struct bpf_sock_addr, 6976 user_ip6)) 6977 return true; 6978 6979 if (bpf_ctx_wide_access_ok(off, size, 6980 struct bpf_sock_addr, 6981 msg_src_ip6)) 6982 return true; 6983 6984 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 6985 return false; 6986 } else { 6987 if (bpf_ctx_wide_access_ok(off, size, 6988 struct bpf_sock_addr, 6989 user_ip6)) 6990 return true; 6991 6992 if (bpf_ctx_wide_access_ok(off, size, 6993 struct bpf_sock_addr, 6994 msg_src_ip6)) 6995 return true; 6996 6997 if (size != size_default) 6998 return false; 6999 } 7000 break; 7001 case bpf_ctx_range(struct bpf_sock_addr, user_port): 7002 if (size != size_default) 7003 return false; 7004 break; 7005 case offsetof(struct bpf_sock_addr, sk): 7006 if (type != BPF_READ) 7007 return false; 7008 if (size != sizeof(__u64)) 7009 return false; 7010 info->reg_type = PTR_TO_SOCKET; 7011 break; 7012 default: 7013 if (type == BPF_READ) { 7014 if (size != size_default) 7015 return false; 7016 } else { 7017 return false; 7018 } 7019 } 7020 7021 return true; 7022 } 7023 7024 static bool sock_ops_is_valid_access(int off, int size, 7025 enum bpf_access_type type, 7026 const struct bpf_prog *prog, 7027 struct bpf_insn_access_aux *info) 7028 { 7029 const int size_default = sizeof(__u32); 7030 7031 if (off < 0 || off >= sizeof(struct bpf_sock_ops)) 7032 return false; 7033 7034 /* The verifier guarantees that size > 0. */ 7035 if (off % size != 0) 7036 return false; 7037 7038 if (type == BPF_WRITE) { 7039 switch (off) { 7040 case offsetof(struct bpf_sock_ops, reply): 7041 case offsetof(struct bpf_sock_ops, sk_txhash): 7042 if (size != size_default) 7043 return false; 7044 break; 7045 default: 7046 return false; 7047 } 7048 } else { 7049 switch (off) { 7050 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received, 7051 bytes_acked): 7052 if (size != sizeof(__u64)) 7053 return false; 7054 break; 7055 case offsetof(struct bpf_sock_ops, sk): 7056 if (size != sizeof(__u64)) 7057 return false; 7058 info->reg_type = PTR_TO_SOCKET_OR_NULL; 7059 break; 7060 default: 7061 if (size != size_default) 7062 return false; 7063 break; 7064 } 7065 } 7066 7067 return true; 7068 } 7069 7070 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write, 7071 const struct bpf_prog *prog) 7072 { 7073 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP); 7074 } 7075 7076 static bool sk_skb_is_valid_access(int off, int size, 7077 enum bpf_access_type type, 7078 const struct bpf_prog *prog, 7079 struct bpf_insn_access_aux *info) 7080 { 7081 switch (off) { 7082 case bpf_ctx_range(struct __sk_buff, tc_classid): 7083 case bpf_ctx_range(struct __sk_buff, data_meta): 7084 case bpf_ctx_range(struct __sk_buff, tstamp): 7085 case bpf_ctx_range(struct __sk_buff, wire_len): 7086 return false; 7087 } 7088 7089 if (type == BPF_WRITE) { 7090 switch (off) { 7091 case bpf_ctx_range(struct __sk_buff, tc_index): 7092 case bpf_ctx_range(struct __sk_buff, priority): 7093 break; 7094 default: 7095 return false; 7096 } 7097 } 7098 7099 switch (off) { 7100 case bpf_ctx_range(struct __sk_buff, mark): 7101 return false; 7102 case bpf_ctx_range(struct __sk_buff, data): 7103 info->reg_type = PTR_TO_PACKET; 7104 break; 7105 case bpf_ctx_range(struct __sk_buff, data_end): 7106 info->reg_type = PTR_TO_PACKET_END; 7107 break; 7108 } 7109 7110 return bpf_skb_is_valid_access(off, size, type, prog, info); 7111 } 7112 7113 static bool sk_msg_is_valid_access(int off, int size, 7114 enum bpf_access_type type, 7115 const struct bpf_prog *prog, 7116 struct bpf_insn_access_aux *info) 7117 { 7118 if (type == BPF_WRITE) 7119 return false; 7120 7121 if (off % size != 0) 7122 return false; 7123 7124 switch (off) { 7125 case offsetof(struct sk_msg_md, data): 7126 info->reg_type = PTR_TO_PACKET; 7127 if (size != sizeof(__u64)) 7128 return false; 7129 break; 7130 case offsetof(struct sk_msg_md, data_end): 7131 info->reg_type = PTR_TO_PACKET_END; 7132 if (size != sizeof(__u64)) 7133 return false; 7134 break; 7135 case bpf_ctx_range(struct sk_msg_md, family): 7136 case bpf_ctx_range(struct sk_msg_md, remote_ip4): 7137 case bpf_ctx_range(struct sk_msg_md, local_ip4): 7138 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]): 7139 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]): 7140 case bpf_ctx_range(struct sk_msg_md, remote_port): 7141 case bpf_ctx_range(struct sk_msg_md, local_port): 7142 case bpf_ctx_range(struct sk_msg_md, size): 7143 if (size != sizeof(__u32)) 7144 return false; 7145 break; 7146 default: 7147 return false; 7148 } 7149 return true; 7150 } 7151 7152 static bool flow_dissector_is_valid_access(int off, int size, 7153 enum bpf_access_type type, 7154 const struct bpf_prog *prog, 7155 struct bpf_insn_access_aux *info) 7156 { 7157 const int size_default = sizeof(__u32); 7158 7159 if (off < 0 || off >= sizeof(struct __sk_buff)) 7160 return false; 7161 7162 if (type == BPF_WRITE) 7163 return false; 7164 7165 switch (off) { 7166 case bpf_ctx_range(struct __sk_buff, data): 7167 if (size != size_default) 7168 return false; 7169 info->reg_type = PTR_TO_PACKET; 7170 return true; 7171 case bpf_ctx_range(struct __sk_buff, data_end): 7172 if (size != size_default) 7173 return false; 7174 info->reg_type = PTR_TO_PACKET_END; 7175 return true; 7176 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 7177 if (size != sizeof(__u64)) 7178 return false; 7179 info->reg_type = PTR_TO_FLOW_KEYS; 7180 return true; 7181 default: 7182 return false; 7183 } 7184 } 7185 7186 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type, 7187 const struct bpf_insn *si, 7188 struct bpf_insn *insn_buf, 7189 struct bpf_prog *prog, 7190 u32 *target_size) 7191 7192 { 7193 struct bpf_insn *insn = insn_buf; 7194 7195 switch (si->off) { 7196 case offsetof(struct __sk_buff, data): 7197 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data), 7198 si->dst_reg, si->src_reg, 7199 offsetof(struct bpf_flow_dissector, data)); 7200 break; 7201 7202 case offsetof(struct __sk_buff, data_end): 7203 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end), 7204 si->dst_reg, si->src_reg, 7205 offsetof(struct bpf_flow_dissector, data_end)); 7206 break; 7207 7208 case offsetof(struct __sk_buff, flow_keys): 7209 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys), 7210 si->dst_reg, si->src_reg, 7211 offsetof(struct bpf_flow_dissector, flow_keys)); 7212 break; 7213 } 7214 7215 return insn - insn_buf; 7216 } 7217 7218 static u32 bpf_convert_ctx_access(enum bpf_access_type type, 7219 const struct bpf_insn *si, 7220 struct bpf_insn *insn_buf, 7221 struct bpf_prog *prog, u32 *target_size) 7222 { 7223 struct bpf_insn *insn = insn_buf; 7224 int off; 7225 7226 switch (si->off) { 7227 case offsetof(struct __sk_buff, len): 7228 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7229 bpf_target_off(struct sk_buff, len, 4, 7230 target_size)); 7231 break; 7232 7233 case offsetof(struct __sk_buff, protocol): 7234 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 7235 bpf_target_off(struct sk_buff, protocol, 2, 7236 target_size)); 7237 break; 7238 7239 case offsetof(struct __sk_buff, vlan_proto): 7240 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 7241 bpf_target_off(struct sk_buff, vlan_proto, 2, 7242 target_size)); 7243 break; 7244 7245 case offsetof(struct __sk_buff, priority): 7246 if (type == BPF_WRITE) 7247 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 7248 bpf_target_off(struct sk_buff, priority, 4, 7249 target_size)); 7250 else 7251 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7252 bpf_target_off(struct sk_buff, priority, 4, 7253 target_size)); 7254 break; 7255 7256 case offsetof(struct __sk_buff, ingress_ifindex): 7257 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7258 bpf_target_off(struct sk_buff, skb_iif, 4, 7259 target_size)); 7260 break; 7261 7262 case offsetof(struct __sk_buff, ifindex): 7263 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 7264 si->dst_reg, si->src_reg, 7265 offsetof(struct sk_buff, dev)); 7266 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 7267 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7268 bpf_target_off(struct net_device, ifindex, 4, 7269 target_size)); 7270 break; 7271 7272 case offsetof(struct __sk_buff, hash): 7273 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7274 bpf_target_off(struct sk_buff, hash, 4, 7275 target_size)); 7276 break; 7277 7278 case offsetof(struct __sk_buff, mark): 7279 if (type == BPF_WRITE) 7280 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 7281 bpf_target_off(struct sk_buff, mark, 4, 7282 target_size)); 7283 else 7284 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7285 bpf_target_off(struct sk_buff, mark, 4, 7286 target_size)); 7287 break; 7288 7289 case offsetof(struct __sk_buff, pkt_type): 7290 *target_size = 1; 7291 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 7292 PKT_TYPE_OFFSET()); 7293 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX); 7294 #ifdef __BIG_ENDIAN_BITFIELD 7295 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5); 7296 #endif 7297 break; 7298 7299 case offsetof(struct __sk_buff, queue_mapping): 7300 if (type == BPF_WRITE) { 7301 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1); 7302 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 7303 bpf_target_off(struct sk_buff, 7304 queue_mapping, 7305 2, target_size)); 7306 } else { 7307 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 7308 bpf_target_off(struct sk_buff, 7309 queue_mapping, 7310 2, target_size)); 7311 } 7312 break; 7313 7314 case offsetof(struct __sk_buff, vlan_present): 7315 *target_size = 1; 7316 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 7317 PKT_VLAN_PRESENT_OFFSET()); 7318 if (PKT_VLAN_PRESENT_BIT) 7319 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT); 7320 if (PKT_VLAN_PRESENT_BIT < 7) 7321 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1); 7322 break; 7323 7324 case offsetof(struct __sk_buff, vlan_tci): 7325 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 7326 bpf_target_off(struct sk_buff, vlan_tci, 2, 7327 target_size)); 7328 break; 7329 7330 case offsetof(struct __sk_buff, cb[0]) ... 7331 offsetofend(struct __sk_buff, cb[4]) - 1: 7332 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, data) < 20); 7333 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 7334 offsetof(struct qdisc_skb_cb, data)) % 7335 sizeof(__u64)); 7336 7337 prog->cb_access = 1; 7338 off = si->off; 7339 off -= offsetof(struct __sk_buff, cb[0]); 7340 off += offsetof(struct sk_buff, cb); 7341 off += offsetof(struct qdisc_skb_cb, data); 7342 if (type == BPF_WRITE) 7343 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 7344 si->src_reg, off); 7345 else 7346 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 7347 si->src_reg, off); 7348 break; 7349 7350 case offsetof(struct __sk_buff, tc_classid): 7351 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, tc_classid) != 2); 7352 7353 off = si->off; 7354 off -= offsetof(struct __sk_buff, tc_classid); 7355 off += offsetof(struct sk_buff, cb); 7356 off += offsetof(struct qdisc_skb_cb, tc_classid); 7357 *target_size = 2; 7358 if (type == BPF_WRITE) 7359 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, 7360 si->src_reg, off); 7361 else 7362 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, 7363 si->src_reg, off); 7364 break; 7365 7366 case offsetof(struct __sk_buff, data): 7367 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 7368 si->dst_reg, si->src_reg, 7369 offsetof(struct sk_buff, data)); 7370 break; 7371 7372 case offsetof(struct __sk_buff, data_meta): 7373 off = si->off; 7374 off -= offsetof(struct __sk_buff, data_meta); 7375 off += offsetof(struct sk_buff, cb); 7376 off += offsetof(struct bpf_skb_data_end, data_meta); 7377 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 7378 si->src_reg, off); 7379 break; 7380 7381 case offsetof(struct __sk_buff, data_end): 7382 off = si->off; 7383 off -= offsetof(struct __sk_buff, data_end); 7384 off += offsetof(struct sk_buff, cb); 7385 off += offsetof(struct bpf_skb_data_end, data_end); 7386 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 7387 si->src_reg, off); 7388 break; 7389 7390 case offsetof(struct __sk_buff, tc_index): 7391 #ifdef CONFIG_NET_SCHED 7392 if (type == BPF_WRITE) 7393 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 7394 bpf_target_off(struct sk_buff, tc_index, 2, 7395 target_size)); 7396 else 7397 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 7398 bpf_target_off(struct sk_buff, tc_index, 2, 7399 target_size)); 7400 #else 7401 *target_size = 2; 7402 if (type == BPF_WRITE) 7403 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg); 7404 else 7405 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 7406 #endif 7407 break; 7408 7409 case offsetof(struct __sk_buff, napi_id): 7410 #if defined(CONFIG_NET_RX_BUSY_POLL) 7411 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7412 bpf_target_off(struct sk_buff, napi_id, 4, 7413 target_size)); 7414 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1); 7415 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 7416 #else 7417 *target_size = 4; 7418 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 7419 #endif 7420 break; 7421 case offsetof(struct __sk_buff, family): 7422 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2); 7423 7424 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7425 si->dst_reg, si->src_reg, 7426 offsetof(struct sk_buff, sk)); 7427 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 7428 bpf_target_off(struct sock_common, 7429 skc_family, 7430 2, target_size)); 7431 break; 7432 case offsetof(struct __sk_buff, remote_ip4): 7433 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4); 7434 7435 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7436 si->dst_reg, si->src_reg, 7437 offsetof(struct sk_buff, sk)); 7438 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7439 bpf_target_off(struct sock_common, 7440 skc_daddr, 7441 4, target_size)); 7442 break; 7443 case offsetof(struct __sk_buff, local_ip4): 7444 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, 7445 skc_rcv_saddr) != 4); 7446 7447 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7448 si->dst_reg, si->src_reg, 7449 offsetof(struct sk_buff, sk)); 7450 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7451 bpf_target_off(struct sock_common, 7452 skc_rcv_saddr, 7453 4, target_size)); 7454 break; 7455 case offsetof(struct __sk_buff, remote_ip6[0]) ... 7456 offsetof(struct __sk_buff, remote_ip6[3]): 7457 #if IS_ENABLED(CONFIG_IPV6) 7458 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, 7459 skc_v6_daddr.s6_addr32[0]) != 4); 7460 7461 off = si->off; 7462 off -= offsetof(struct __sk_buff, remote_ip6[0]); 7463 7464 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7465 si->dst_reg, si->src_reg, 7466 offsetof(struct sk_buff, sk)); 7467 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7468 offsetof(struct sock_common, 7469 skc_v6_daddr.s6_addr32[0]) + 7470 off); 7471 #else 7472 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 7473 #endif 7474 break; 7475 case offsetof(struct __sk_buff, local_ip6[0]) ... 7476 offsetof(struct __sk_buff, local_ip6[3]): 7477 #if IS_ENABLED(CONFIG_IPV6) 7478 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, 7479 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 7480 7481 off = si->off; 7482 off -= offsetof(struct __sk_buff, local_ip6[0]); 7483 7484 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7485 si->dst_reg, si->src_reg, 7486 offsetof(struct sk_buff, sk)); 7487 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7488 offsetof(struct sock_common, 7489 skc_v6_rcv_saddr.s6_addr32[0]) + 7490 off); 7491 #else 7492 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 7493 #endif 7494 break; 7495 7496 case offsetof(struct __sk_buff, remote_port): 7497 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2); 7498 7499 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7500 si->dst_reg, si->src_reg, 7501 offsetof(struct sk_buff, sk)); 7502 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 7503 bpf_target_off(struct sock_common, 7504 skc_dport, 7505 2, target_size)); 7506 #ifndef __BIG_ENDIAN_BITFIELD 7507 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 7508 #endif 7509 break; 7510 7511 case offsetof(struct __sk_buff, local_port): 7512 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2); 7513 7514 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7515 si->dst_reg, si->src_reg, 7516 offsetof(struct sk_buff, sk)); 7517 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 7518 bpf_target_off(struct sock_common, 7519 skc_num, 2, target_size)); 7520 break; 7521 7522 case offsetof(struct __sk_buff, tstamp): 7523 BUILD_BUG_ON(FIELD_SIZEOF(struct sk_buff, tstamp) != 8); 7524 7525 if (type == BPF_WRITE) 7526 *insn++ = BPF_STX_MEM(BPF_DW, 7527 si->dst_reg, si->src_reg, 7528 bpf_target_off(struct sk_buff, 7529 tstamp, 8, 7530 target_size)); 7531 else 7532 *insn++ = BPF_LDX_MEM(BPF_DW, 7533 si->dst_reg, si->src_reg, 7534 bpf_target_off(struct sk_buff, 7535 tstamp, 8, 7536 target_size)); 7537 break; 7538 7539 case offsetof(struct __sk_buff, gso_segs): 7540 /* si->dst_reg = skb_shinfo(SKB); */ 7541 #ifdef NET_SKBUFF_DATA_USES_OFFSET 7542 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 7543 BPF_REG_AX, si->src_reg, 7544 offsetof(struct sk_buff, end)); 7545 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head), 7546 si->dst_reg, si->src_reg, 7547 offsetof(struct sk_buff, head)); 7548 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX); 7549 #else 7550 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 7551 si->dst_reg, si->src_reg, 7552 offsetof(struct sk_buff, end)); 7553 #endif 7554 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs), 7555 si->dst_reg, si->dst_reg, 7556 bpf_target_off(struct skb_shared_info, 7557 gso_segs, 2, 7558 target_size)); 7559 break; 7560 case offsetof(struct __sk_buff, wire_len): 7561 BUILD_BUG_ON(FIELD_SIZEOF(struct qdisc_skb_cb, pkt_len) != 4); 7562 7563 off = si->off; 7564 off -= offsetof(struct __sk_buff, wire_len); 7565 off += offsetof(struct sk_buff, cb); 7566 off += offsetof(struct qdisc_skb_cb, pkt_len); 7567 *target_size = 4; 7568 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off); 7569 break; 7570 7571 case offsetof(struct __sk_buff, sk): 7572 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7573 si->dst_reg, si->src_reg, 7574 offsetof(struct sk_buff, sk)); 7575 break; 7576 } 7577 7578 return insn - insn_buf; 7579 } 7580 7581 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 7582 const struct bpf_insn *si, 7583 struct bpf_insn *insn_buf, 7584 struct bpf_prog *prog, u32 *target_size) 7585 { 7586 struct bpf_insn *insn = insn_buf; 7587 int off; 7588 7589 switch (si->off) { 7590 case offsetof(struct bpf_sock, bound_dev_if): 7591 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_bound_dev_if) != 4); 7592 7593 if (type == BPF_WRITE) 7594 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 7595 offsetof(struct sock, sk_bound_dev_if)); 7596 else 7597 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7598 offsetof(struct sock, sk_bound_dev_if)); 7599 break; 7600 7601 case offsetof(struct bpf_sock, mark): 7602 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_mark) != 4); 7603 7604 if (type == BPF_WRITE) 7605 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 7606 offsetof(struct sock, sk_mark)); 7607 else 7608 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7609 offsetof(struct sock, sk_mark)); 7610 break; 7611 7612 case offsetof(struct bpf_sock, priority): 7613 BUILD_BUG_ON(FIELD_SIZEOF(struct sock, sk_priority) != 4); 7614 7615 if (type == BPF_WRITE) 7616 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 7617 offsetof(struct sock, sk_priority)); 7618 else 7619 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7620 offsetof(struct sock, sk_priority)); 7621 break; 7622 7623 case offsetof(struct bpf_sock, family): 7624 *insn++ = BPF_LDX_MEM( 7625 BPF_FIELD_SIZEOF(struct sock_common, skc_family), 7626 si->dst_reg, si->src_reg, 7627 bpf_target_off(struct sock_common, 7628 skc_family, 7629 FIELD_SIZEOF(struct sock_common, 7630 skc_family), 7631 target_size)); 7632 break; 7633 7634 case offsetof(struct bpf_sock, type): 7635 BUILD_BUG_ON(HWEIGHT32(SK_FL_TYPE_MASK) != BITS_PER_BYTE * 2); 7636 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7637 offsetof(struct sock, __sk_flags_offset)); 7638 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK); 7639 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT); 7640 *target_size = 2; 7641 break; 7642 7643 case offsetof(struct bpf_sock, protocol): 7644 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE); 7645 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7646 offsetof(struct sock, __sk_flags_offset)); 7647 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK); 7648 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_PROTO_SHIFT); 7649 *target_size = 1; 7650 break; 7651 7652 case offsetof(struct bpf_sock, src_ip4): 7653 *insn++ = BPF_LDX_MEM( 7654 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 7655 bpf_target_off(struct sock_common, skc_rcv_saddr, 7656 FIELD_SIZEOF(struct sock_common, 7657 skc_rcv_saddr), 7658 target_size)); 7659 break; 7660 7661 case offsetof(struct bpf_sock, dst_ip4): 7662 *insn++ = BPF_LDX_MEM( 7663 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 7664 bpf_target_off(struct sock_common, skc_daddr, 7665 FIELD_SIZEOF(struct sock_common, 7666 skc_daddr), 7667 target_size)); 7668 break; 7669 7670 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 7671 #if IS_ENABLED(CONFIG_IPV6) 7672 off = si->off; 7673 off -= offsetof(struct bpf_sock, src_ip6[0]); 7674 *insn++ = BPF_LDX_MEM( 7675 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 7676 bpf_target_off( 7677 struct sock_common, 7678 skc_v6_rcv_saddr.s6_addr32[0], 7679 FIELD_SIZEOF(struct sock_common, 7680 skc_v6_rcv_saddr.s6_addr32[0]), 7681 target_size) + off); 7682 #else 7683 (void)off; 7684 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 7685 #endif 7686 break; 7687 7688 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 7689 #if IS_ENABLED(CONFIG_IPV6) 7690 off = si->off; 7691 off -= offsetof(struct bpf_sock, dst_ip6[0]); 7692 *insn++ = BPF_LDX_MEM( 7693 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 7694 bpf_target_off(struct sock_common, 7695 skc_v6_daddr.s6_addr32[0], 7696 FIELD_SIZEOF(struct sock_common, 7697 skc_v6_daddr.s6_addr32[0]), 7698 target_size) + off); 7699 #else 7700 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 7701 *target_size = 4; 7702 #endif 7703 break; 7704 7705 case offsetof(struct bpf_sock, src_port): 7706 *insn++ = BPF_LDX_MEM( 7707 BPF_FIELD_SIZEOF(struct sock_common, skc_num), 7708 si->dst_reg, si->src_reg, 7709 bpf_target_off(struct sock_common, skc_num, 7710 FIELD_SIZEOF(struct sock_common, 7711 skc_num), 7712 target_size)); 7713 break; 7714 7715 case offsetof(struct bpf_sock, dst_port): 7716 *insn++ = BPF_LDX_MEM( 7717 BPF_FIELD_SIZEOF(struct sock_common, skc_dport), 7718 si->dst_reg, si->src_reg, 7719 bpf_target_off(struct sock_common, skc_dport, 7720 FIELD_SIZEOF(struct sock_common, 7721 skc_dport), 7722 target_size)); 7723 break; 7724 7725 case offsetof(struct bpf_sock, state): 7726 *insn++ = BPF_LDX_MEM( 7727 BPF_FIELD_SIZEOF(struct sock_common, skc_state), 7728 si->dst_reg, si->src_reg, 7729 bpf_target_off(struct sock_common, skc_state, 7730 FIELD_SIZEOF(struct sock_common, 7731 skc_state), 7732 target_size)); 7733 break; 7734 } 7735 7736 return insn - insn_buf; 7737 } 7738 7739 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, 7740 const struct bpf_insn *si, 7741 struct bpf_insn *insn_buf, 7742 struct bpf_prog *prog, u32 *target_size) 7743 { 7744 struct bpf_insn *insn = insn_buf; 7745 7746 switch (si->off) { 7747 case offsetof(struct __sk_buff, ifindex): 7748 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 7749 si->dst_reg, si->src_reg, 7750 offsetof(struct sk_buff, dev)); 7751 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7752 bpf_target_off(struct net_device, ifindex, 4, 7753 target_size)); 7754 break; 7755 default: 7756 return bpf_convert_ctx_access(type, si, insn_buf, prog, 7757 target_size); 7758 } 7759 7760 return insn - insn_buf; 7761 } 7762 7763 static u32 xdp_convert_ctx_access(enum bpf_access_type type, 7764 const struct bpf_insn *si, 7765 struct bpf_insn *insn_buf, 7766 struct bpf_prog *prog, u32 *target_size) 7767 { 7768 struct bpf_insn *insn = insn_buf; 7769 7770 switch (si->off) { 7771 case offsetof(struct xdp_md, data): 7772 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data), 7773 si->dst_reg, si->src_reg, 7774 offsetof(struct xdp_buff, data)); 7775 break; 7776 case offsetof(struct xdp_md, data_meta): 7777 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta), 7778 si->dst_reg, si->src_reg, 7779 offsetof(struct xdp_buff, data_meta)); 7780 break; 7781 case offsetof(struct xdp_md, data_end): 7782 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end), 7783 si->dst_reg, si->src_reg, 7784 offsetof(struct xdp_buff, data_end)); 7785 break; 7786 case offsetof(struct xdp_md, ingress_ifindex): 7787 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 7788 si->dst_reg, si->src_reg, 7789 offsetof(struct xdp_buff, rxq)); 7790 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev), 7791 si->dst_reg, si->dst_reg, 7792 offsetof(struct xdp_rxq_info, dev)); 7793 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7794 offsetof(struct net_device, ifindex)); 7795 break; 7796 case offsetof(struct xdp_md, rx_queue_index): 7797 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 7798 si->dst_reg, si->src_reg, 7799 offsetof(struct xdp_buff, rxq)); 7800 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7801 offsetof(struct xdp_rxq_info, 7802 queue_index)); 7803 break; 7804 } 7805 7806 return insn - insn_buf; 7807 } 7808 7809 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of 7810 * context Structure, F is Field in context structure that contains a pointer 7811 * to Nested Structure of type NS that has the field NF. 7812 * 7813 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make 7814 * sure that SIZE is not greater than actual size of S.F.NF. 7815 * 7816 * If offset OFF is provided, the load happens from that offset relative to 7817 * offset of NF. 7818 */ 7819 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \ 7820 do { \ 7821 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \ 7822 si->src_reg, offsetof(S, F)); \ 7823 *insn++ = BPF_LDX_MEM( \ 7824 SIZE, si->dst_reg, si->dst_reg, \ 7825 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \ 7826 target_size) \ 7827 + OFF); \ 7828 } while (0) 7829 7830 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \ 7831 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \ 7832 BPF_FIELD_SIZEOF(NS, NF), 0) 7833 7834 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to 7835 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation. 7836 * 7837 * In addition it uses Temporary Field TF (member of struct S) as the 3rd 7838 * "register" since two registers available in convert_ctx_access are not 7839 * enough: we can't override neither SRC, since it contains value to store, nor 7840 * DST since it contains pointer to context that may be used by later 7841 * instructions. But we need a temporary place to save pointer to nested 7842 * structure whose field we want to store to. 7843 */ 7844 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \ 7845 do { \ 7846 int tmp_reg = BPF_REG_9; \ 7847 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 7848 --tmp_reg; \ 7849 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 7850 --tmp_reg; \ 7851 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \ 7852 offsetof(S, TF)); \ 7853 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \ 7854 si->dst_reg, offsetof(S, F)); \ 7855 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \ 7856 bpf_target_off(NS, NF, FIELD_SIZEOF(NS, NF), \ 7857 target_size) \ 7858 + OFF); \ 7859 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \ 7860 offsetof(S, TF)); \ 7861 } while (0) 7862 7863 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \ 7864 TF) \ 7865 do { \ 7866 if (type == BPF_WRITE) { \ 7867 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \ 7868 OFF, TF); \ 7869 } else { \ 7870 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \ 7871 S, NS, F, NF, SIZE, OFF); \ 7872 } \ 7873 } while (0) 7874 7875 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \ 7876 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \ 7877 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF) 7878 7879 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type, 7880 const struct bpf_insn *si, 7881 struct bpf_insn *insn_buf, 7882 struct bpf_prog *prog, u32 *target_size) 7883 { 7884 struct bpf_insn *insn = insn_buf; 7885 int off; 7886 7887 switch (si->off) { 7888 case offsetof(struct bpf_sock_addr, user_family): 7889 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 7890 struct sockaddr, uaddr, sa_family); 7891 break; 7892 7893 case offsetof(struct bpf_sock_addr, user_ip4): 7894 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 7895 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr, 7896 sin_addr, BPF_SIZE(si->code), 0, tmp_reg); 7897 break; 7898 7899 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 7900 off = si->off; 7901 off -= offsetof(struct bpf_sock_addr, user_ip6[0]); 7902 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 7903 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 7904 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off, 7905 tmp_reg); 7906 break; 7907 7908 case offsetof(struct bpf_sock_addr, user_port): 7909 /* To get port we need to know sa_family first and then treat 7910 * sockaddr as either sockaddr_in or sockaddr_in6. 7911 * Though we can simplify since port field has same offset and 7912 * size in both structures. 7913 * Here we check this invariant and use just one of the 7914 * structures if it's true. 7915 */ 7916 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) != 7917 offsetof(struct sockaddr_in6, sin6_port)); 7918 BUILD_BUG_ON(FIELD_SIZEOF(struct sockaddr_in, sin_port) != 7919 FIELD_SIZEOF(struct sockaddr_in6, sin6_port)); 7920 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern, 7921 struct sockaddr_in6, uaddr, 7922 sin6_port, tmp_reg); 7923 break; 7924 7925 case offsetof(struct bpf_sock_addr, family): 7926 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 7927 struct sock, sk, sk_family); 7928 break; 7929 7930 case offsetof(struct bpf_sock_addr, type): 7931 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( 7932 struct bpf_sock_addr_kern, struct sock, sk, 7933 __sk_flags_offset, BPF_W, 0); 7934 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_TYPE_MASK); 7935 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, SK_FL_TYPE_SHIFT); 7936 break; 7937 7938 case offsetof(struct bpf_sock_addr, protocol): 7939 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( 7940 struct bpf_sock_addr_kern, struct sock, sk, 7941 __sk_flags_offset, BPF_W, 0); 7942 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK); 7943 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 7944 SK_FL_PROTO_SHIFT); 7945 break; 7946 7947 case offsetof(struct bpf_sock_addr, msg_src_ip4): 7948 /* Treat t_ctx as struct in_addr for msg_src_ip4. */ 7949 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 7950 struct bpf_sock_addr_kern, struct in_addr, t_ctx, 7951 s_addr, BPF_SIZE(si->code), 0, tmp_reg); 7952 break; 7953 7954 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 7955 msg_src_ip6[3]): 7956 off = si->off; 7957 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]); 7958 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */ 7959 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 7960 struct bpf_sock_addr_kern, struct in6_addr, t_ctx, 7961 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg); 7962 break; 7963 case offsetof(struct bpf_sock_addr, sk): 7964 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk), 7965 si->dst_reg, si->src_reg, 7966 offsetof(struct bpf_sock_addr_kern, sk)); 7967 break; 7968 } 7969 7970 return insn - insn_buf; 7971 } 7972 7973 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type, 7974 const struct bpf_insn *si, 7975 struct bpf_insn *insn_buf, 7976 struct bpf_prog *prog, 7977 u32 *target_size) 7978 { 7979 struct bpf_insn *insn = insn_buf; 7980 int off; 7981 7982 /* Helper macro for adding read access to tcp_sock or sock fields. */ 7983 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 7984 do { \ 7985 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \ 7986 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \ 7987 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 7988 struct bpf_sock_ops_kern, \ 7989 is_fullsock), \ 7990 si->dst_reg, si->src_reg, \ 7991 offsetof(struct bpf_sock_ops_kern, \ 7992 is_fullsock)); \ 7993 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2); \ 7994 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 7995 struct bpf_sock_ops_kern, sk),\ 7996 si->dst_reg, si->src_reg, \ 7997 offsetof(struct bpf_sock_ops_kern, sk));\ 7998 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \ 7999 OBJ_FIELD), \ 8000 si->dst_reg, si->dst_reg, \ 8001 offsetof(OBJ, OBJ_FIELD)); \ 8002 } while (0) 8003 8004 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \ 8005 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock) 8006 8007 /* Helper macro for adding write access to tcp_sock or sock fields. 8008 * The macro is called with two registers, dst_reg which contains a pointer 8009 * to ctx (context) and src_reg which contains the value that should be 8010 * stored. However, we need an additional register since we cannot overwrite 8011 * dst_reg because it may be used later in the program. 8012 * Instead we "borrow" one of the other register. We first save its value 8013 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore 8014 * it at the end of the macro. 8015 */ 8016 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 8017 do { \ 8018 int reg = BPF_REG_9; \ 8019 BUILD_BUG_ON(FIELD_SIZEOF(OBJ, OBJ_FIELD) > \ 8020 FIELD_SIZEOF(struct bpf_sock_ops, BPF_FIELD)); \ 8021 if (si->dst_reg == reg || si->src_reg == reg) \ 8022 reg--; \ 8023 if (si->dst_reg == reg || si->src_reg == reg) \ 8024 reg--; \ 8025 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \ 8026 offsetof(struct bpf_sock_ops_kern, \ 8027 temp)); \ 8028 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 8029 struct bpf_sock_ops_kern, \ 8030 is_fullsock), \ 8031 reg, si->dst_reg, \ 8032 offsetof(struct bpf_sock_ops_kern, \ 8033 is_fullsock)); \ 8034 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \ 8035 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 8036 struct bpf_sock_ops_kern, sk),\ 8037 reg, si->dst_reg, \ 8038 offsetof(struct bpf_sock_ops_kern, sk));\ 8039 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \ 8040 reg, si->src_reg, \ 8041 offsetof(OBJ, OBJ_FIELD)); \ 8042 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \ 8043 offsetof(struct bpf_sock_ops_kern, \ 8044 temp)); \ 8045 } while (0) 8046 8047 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \ 8048 do { \ 8049 if (TYPE == BPF_WRITE) \ 8050 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 8051 else \ 8052 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 8053 } while (0) 8054 8055 if (insn > insn_buf) 8056 return insn - insn_buf; 8057 8058 switch (si->off) { 8059 case offsetof(struct bpf_sock_ops, op) ... 8060 offsetof(struct bpf_sock_ops, replylong[3]): 8061 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, op) != 8062 FIELD_SIZEOF(struct bpf_sock_ops_kern, op)); 8063 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, reply) != 8064 FIELD_SIZEOF(struct bpf_sock_ops_kern, reply)); 8065 BUILD_BUG_ON(FIELD_SIZEOF(struct bpf_sock_ops, replylong) != 8066 FIELD_SIZEOF(struct bpf_sock_ops_kern, replylong)); 8067 off = si->off; 8068 off -= offsetof(struct bpf_sock_ops, op); 8069 off += offsetof(struct bpf_sock_ops_kern, op); 8070 if (type == BPF_WRITE) 8071 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8072 off); 8073 else 8074 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8075 off); 8076 break; 8077 8078 case offsetof(struct bpf_sock_ops, family): 8079 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2); 8080 8081 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8082 struct bpf_sock_ops_kern, sk), 8083 si->dst_reg, si->src_reg, 8084 offsetof(struct bpf_sock_ops_kern, sk)); 8085 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8086 offsetof(struct sock_common, skc_family)); 8087 break; 8088 8089 case offsetof(struct bpf_sock_ops, remote_ip4): 8090 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4); 8091 8092 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8093 struct bpf_sock_ops_kern, sk), 8094 si->dst_reg, si->src_reg, 8095 offsetof(struct bpf_sock_ops_kern, sk)); 8096 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8097 offsetof(struct sock_common, skc_daddr)); 8098 break; 8099 8100 case offsetof(struct bpf_sock_ops, local_ip4): 8101 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, 8102 skc_rcv_saddr) != 4); 8103 8104 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8105 struct bpf_sock_ops_kern, sk), 8106 si->dst_reg, si->src_reg, 8107 offsetof(struct bpf_sock_ops_kern, sk)); 8108 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8109 offsetof(struct sock_common, 8110 skc_rcv_saddr)); 8111 break; 8112 8113 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ... 8114 offsetof(struct bpf_sock_ops, remote_ip6[3]): 8115 #if IS_ENABLED(CONFIG_IPV6) 8116 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, 8117 skc_v6_daddr.s6_addr32[0]) != 4); 8118 8119 off = si->off; 8120 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]); 8121 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8122 struct bpf_sock_ops_kern, sk), 8123 si->dst_reg, si->src_reg, 8124 offsetof(struct bpf_sock_ops_kern, sk)); 8125 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8126 offsetof(struct sock_common, 8127 skc_v6_daddr.s6_addr32[0]) + 8128 off); 8129 #else 8130 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8131 #endif 8132 break; 8133 8134 case offsetof(struct bpf_sock_ops, local_ip6[0]) ... 8135 offsetof(struct bpf_sock_ops, local_ip6[3]): 8136 #if IS_ENABLED(CONFIG_IPV6) 8137 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, 8138 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 8139 8140 off = si->off; 8141 off -= offsetof(struct bpf_sock_ops, local_ip6[0]); 8142 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8143 struct bpf_sock_ops_kern, sk), 8144 si->dst_reg, si->src_reg, 8145 offsetof(struct bpf_sock_ops_kern, sk)); 8146 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8147 offsetof(struct sock_common, 8148 skc_v6_rcv_saddr.s6_addr32[0]) + 8149 off); 8150 #else 8151 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8152 #endif 8153 break; 8154 8155 case offsetof(struct bpf_sock_ops, remote_port): 8156 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2); 8157 8158 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8159 struct bpf_sock_ops_kern, sk), 8160 si->dst_reg, si->src_reg, 8161 offsetof(struct bpf_sock_ops_kern, sk)); 8162 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8163 offsetof(struct sock_common, skc_dport)); 8164 #ifndef __BIG_ENDIAN_BITFIELD 8165 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 8166 #endif 8167 break; 8168 8169 case offsetof(struct bpf_sock_ops, local_port): 8170 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2); 8171 8172 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8173 struct bpf_sock_ops_kern, sk), 8174 si->dst_reg, si->src_reg, 8175 offsetof(struct bpf_sock_ops_kern, sk)); 8176 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8177 offsetof(struct sock_common, skc_num)); 8178 break; 8179 8180 case offsetof(struct bpf_sock_ops, is_fullsock): 8181 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8182 struct bpf_sock_ops_kern, 8183 is_fullsock), 8184 si->dst_reg, si->src_reg, 8185 offsetof(struct bpf_sock_ops_kern, 8186 is_fullsock)); 8187 break; 8188 8189 case offsetof(struct bpf_sock_ops, state): 8190 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_state) != 1); 8191 8192 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8193 struct bpf_sock_ops_kern, sk), 8194 si->dst_reg, si->src_reg, 8195 offsetof(struct bpf_sock_ops_kern, sk)); 8196 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg, 8197 offsetof(struct sock_common, skc_state)); 8198 break; 8199 8200 case offsetof(struct bpf_sock_ops, rtt_min): 8201 BUILD_BUG_ON(FIELD_SIZEOF(struct tcp_sock, rtt_min) != 8202 sizeof(struct minmax)); 8203 BUILD_BUG_ON(sizeof(struct minmax) < 8204 sizeof(struct minmax_sample)); 8205 8206 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8207 struct bpf_sock_ops_kern, sk), 8208 si->dst_reg, si->src_reg, 8209 offsetof(struct bpf_sock_ops_kern, sk)); 8210 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8211 offsetof(struct tcp_sock, rtt_min) + 8212 FIELD_SIZEOF(struct minmax_sample, t)); 8213 break; 8214 8215 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags): 8216 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags, 8217 struct tcp_sock); 8218 break; 8219 8220 case offsetof(struct bpf_sock_ops, sk_txhash): 8221 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash, 8222 struct sock, type); 8223 break; 8224 case offsetof(struct bpf_sock_ops, snd_cwnd): 8225 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd); 8226 break; 8227 case offsetof(struct bpf_sock_ops, srtt_us): 8228 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us); 8229 break; 8230 case offsetof(struct bpf_sock_ops, snd_ssthresh): 8231 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh); 8232 break; 8233 case offsetof(struct bpf_sock_ops, rcv_nxt): 8234 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt); 8235 break; 8236 case offsetof(struct bpf_sock_ops, snd_nxt): 8237 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt); 8238 break; 8239 case offsetof(struct bpf_sock_ops, snd_una): 8240 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una); 8241 break; 8242 case offsetof(struct bpf_sock_ops, mss_cache): 8243 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache); 8244 break; 8245 case offsetof(struct bpf_sock_ops, ecn_flags): 8246 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags); 8247 break; 8248 case offsetof(struct bpf_sock_ops, rate_delivered): 8249 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered); 8250 break; 8251 case offsetof(struct bpf_sock_ops, rate_interval_us): 8252 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us); 8253 break; 8254 case offsetof(struct bpf_sock_ops, packets_out): 8255 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out); 8256 break; 8257 case offsetof(struct bpf_sock_ops, retrans_out): 8258 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out); 8259 break; 8260 case offsetof(struct bpf_sock_ops, total_retrans): 8261 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans); 8262 break; 8263 case offsetof(struct bpf_sock_ops, segs_in): 8264 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in); 8265 break; 8266 case offsetof(struct bpf_sock_ops, data_segs_in): 8267 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in); 8268 break; 8269 case offsetof(struct bpf_sock_ops, segs_out): 8270 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out); 8271 break; 8272 case offsetof(struct bpf_sock_ops, data_segs_out): 8273 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out); 8274 break; 8275 case offsetof(struct bpf_sock_ops, lost_out): 8276 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out); 8277 break; 8278 case offsetof(struct bpf_sock_ops, sacked_out): 8279 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out); 8280 break; 8281 case offsetof(struct bpf_sock_ops, bytes_received): 8282 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received); 8283 break; 8284 case offsetof(struct bpf_sock_ops, bytes_acked): 8285 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked); 8286 break; 8287 case offsetof(struct bpf_sock_ops, sk): 8288 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8289 struct bpf_sock_ops_kern, 8290 is_fullsock), 8291 si->dst_reg, si->src_reg, 8292 offsetof(struct bpf_sock_ops_kern, 8293 is_fullsock)); 8294 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 8295 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8296 struct bpf_sock_ops_kern, sk), 8297 si->dst_reg, si->src_reg, 8298 offsetof(struct bpf_sock_ops_kern, sk)); 8299 break; 8300 } 8301 return insn - insn_buf; 8302 } 8303 8304 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type, 8305 const struct bpf_insn *si, 8306 struct bpf_insn *insn_buf, 8307 struct bpf_prog *prog, u32 *target_size) 8308 { 8309 struct bpf_insn *insn = insn_buf; 8310 int off; 8311 8312 switch (si->off) { 8313 case offsetof(struct __sk_buff, data_end): 8314 off = si->off; 8315 off -= offsetof(struct __sk_buff, data_end); 8316 off += offsetof(struct sk_buff, cb); 8317 off += offsetof(struct tcp_skb_cb, bpf.data_end); 8318 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 8319 si->src_reg, off); 8320 break; 8321 default: 8322 return bpf_convert_ctx_access(type, si, insn_buf, prog, 8323 target_size); 8324 } 8325 8326 return insn - insn_buf; 8327 } 8328 8329 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type, 8330 const struct bpf_insn *si, 8331 struct bpf_insn *insn_buf, 8332 struct bpf_prog *prog, u32 *target_size) 8333 { 8334 struct bpf_insn *insn = insn_buf; 8335 #if IS_ENABLED(CONFIG_IPV6) 8336 int off; 8337 #endif 8338 8339 /* convert ctx uses the fact sg element is first in struct */ 8340 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0); 8341 8342 switch (si->off) { 8343 case offsetof(struct sk_msg_md, data): 8344 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data), 8345 si->dst_reg, si->src_reg, 8346 offsetof(struct sk_msg, data)); 8347 break; 8348 case offsetof(struct sk_msg_md, data_end): 8349 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end), 8350 si->dst_reg, si->src_reg, 8351 offsetof(struct sk_msg, data_end)); 8352 break; 8353 case offsetof(struct sk_msg_md, family): 8354 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_family) != 2); 8355 8356 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8357 struct sk_msg, sk), 8358 si->dst_reg, si->src_reg, 8359 offsetof(struct sk_msg, sk)); 8360 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8361 offsetof(struct sock_common, skc_family)); 8362 break; 8363 8364 case offsetof(struct sk_msg_md, remote_ip4): 8365 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_daddr) != 4); 8366 8367 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8368 struct sk_msg, sk), 8369 si->dst_reg, si->src_reg, 8370 offsetof(struct sk_msg, sk)); 8371 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8372 offsetof(struct sock_common, skc_daddr)); 8373 break; 8374 8375 case offsetof(struct sk_msg_md, local_ip4): 8376 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, 8377 skc_rcv_saddr) != 4); 8378 8379 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8380 struct sk_msg, sk), 8381 si->dst_reg, si->src_reg, 8382 offsetof(struct sk_msg, sk)); 8383 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8384 offsetof(struct sock_common, 8385 skc_rcv_saddr)); 8386 break; 8387 8388 case offsetof(struct sk_msg_md, remote_ip6[0]) ... 8389 offsetof(struct sk_msg_md, remote_ip6[3]): 8390 #if IS_ENABLED(CONFIG_IPV6) 8391 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, 8392 skc_v6_daddr.s6_addr32[0]) != 4); 8393 8394 off = si->off; 8395 off -= offsetof(struct sk_msg_md, remote_ip6[0]); 8396 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8397 struct sk_msg, sk), 8398 si->dst_reg, si->src_reg, 8399 offsetof(struct sk_msg, sk)); 8400 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8401 offsetof(struct sock_common, 8402 skc_v6_daddr.s6_addr32[0]) + 8403 off); 8404 #else 8405 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8406 #endif 8407 break; 8408 8409 case offsetof(struct sk_msg_md, local_ip6[0]) ... 8410 offsetof(struct sk_msg_md, local_ip6[3]): 8411 #if IS_ENABLED(CONFIG_IPV6) 8412 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, 8413 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 8414 8415 off = si->off; 8416 off -= offsetof(struct sk_msg_md, local_ip6[0]); 8417 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8418 struct sk_msg, sk), 8419 si->dst_reg, si->src_reg, 8420 offsetof(struct sk_msg, sk)); 8421 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8422 offsetof(struct sock_common, 8423 skc_v6_rcv_saddr.s6_addr32[0]) + 8424 off); 8425 #else 8426 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8427 #endif 8428 break; 8429 8430 case offsetof(struct sk_msg_md, remote_port): 8431 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_dport) != 2); 8432 8433 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8434 struct sk_msg, sk), 8435 si->dst_reg, si->src_reg, 8436 offsetof(struct sk_msg, sk)); 8437 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8438 offsetof(struct sock_common, skc_dport)); 8439 #ifndef __BIG_ENDIAN_BITFIELD 8440 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 8441 #endif 8442 break; 8443 8444 case offsetof(struct sk_msg_md, local_port): 8445 BUILD_BUG_ON(FIELD_SIZEOF(struct sock_common, skc_num) != 2); 8446 8447 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8448 struct sk_msg, sk), 8449 si->dst_reg, si->src_reg, 8450 offsetof(struct sk_msg, sk)); 8451 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8452 offsetof(struct sock_common, skc_num)); 8453 break; 8454 8455 case offsetof(struct sk_msg_md, size): 8456 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size), 8457 si->dst_reg, si->src_reg, 8458 offsetof(struct sk_msg_sg, size)); 8459 break; 8460 } 8461 8462 return insn - insn_buf; 8463 } 8464 8465 const struct bpf_verifier_ops sk_filter_verifier_ops = { 8466 .get_func_proto = sk_filter_func_proto, 8467 .is_valid_access = sk_filter_is_valid_access, 8468 .convert_ctx_access = bpf_convert_ctx_access, 8469 .gen_ld_abs = bpf_gen_ld_abs, 8470 }; 8471 8472 const struct bpf_prog_ops sk_filter_prog_ops = { 8473 .test_run = bpf_prog_test_run_skb, 8474 }; 8475 8476 const struct bpf_verifier_ops tc_cls_act_verifier_ops = { 8477 .get_func_proto = tc_cls_act_func_proto, 8478 .is_valid_access = tc_cls_act_is_valid_access, 8479 .convert_ctx_access = tc_cls_act_convert_ctx_access, 8480 .gen_prologue = tc_cls_act_prologue, 8481 .gen_ld_abs = bpf_gen_ld_abs, 8482 }; 8483 8484 const struct bpf_prog_ops tc_cls_act_prog_ops = { 8485 .test_run = bpf_prog_test_run_skb, 8486 }; 8487 8488 const struct bpf_verifier_ops xdp_verifier_ops = { 8489 .get_func_proto = xdp_func_proto, 8490 .is_valid_access = xdp_is_valid_access, 8491 .convert_ctx_access = xdp_convert_ctx_access, 8492 .gen_prologue = bpf_noop_prologue, 8493 }; 8494 8495 const struct bpf_prog_ops xdp_prog_ops = { 8496 .test_run = bpf_prog_test_run_xdp, 8497 }; 8498 8499 const struct bpf_verifier_ops cg_skb_verifier_ops = { 8500 .get_func_proto = cg_skb_func_proto, 8501 .is_valid_access = cg_skb_is_valid_access, 8502 .convert_ctx_access = bpf_convert_ctx_access, 8503 }; 8504 8505 const struct bpf_prog_ops cg_skb_prog_ops = { 8506 .test_run = bpf_prog_test_run_skb, 8507 }; 8508 8509 const struct bpf_verifier_ops lwt_in_verifier_ops = { 8510 .get_func_proto = lwt_in_func_proto, 8511 .is_valid_access = lwt_is_valid_access, 8512 .convert_ctx_access = bpf_convert_ctx_access, 8513 }; 8514 8515 const struct bpf_prog_ops lwt_in_prog_ops = { 8516 .test_run = bpf_prog_test_run_skb, 8517 }; 8518 8519 const struct bpf_verifier_ops lwt_out_verifier_ops = { 8520 .get_func_proto = lwt_out_func_proto, 8521 .is_valid_access = lwt_is_valid_access, 8522 .convert_ctx_access = bpf_convert_ctx_access, 8523 }; 8524 8525 const struct bpf_prog_ops lwt_out_prog_ops = { 8526 .test_run = bpf_prog_test_run_skb, 8527 }; 8528 8529 const struct bpf_verifier_ops lwt_xmit_verifier_ops = { 8530 .get_func_proto = lwt_xmit_func_proto, 8531 .is_valid_access = lwt_is_valid_access, 8532 .convert_ctx_access = bpf_convert_ctx_access, 8533 .gen_prologue = tc_cls_act_prologue, 8534 }; 8535 8536 const struct bpf_prog_ops lwt_xmit_prog_ops = { 8537 .test_run = bpf_prog_test_run_skb, 8538 }; 8539 8540 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = { 8541 .get_func_proto = lwt_seg6local_func_proto, 8542 .is_valid_access = lwt_is_valid_access, 8543 .convert_ctx_access = bpf_convert_ctx_access, 8544 }; 8545 8546 const struct bpf_prog_ops lwt_seg6local_prog_ops = { 8547 .test_run = bpf_prog_test_run_skb, 8548 }; 8549 8550 const struct bpf_verifier_ops cg_sock_verifier_ops = { 8551 .get_func_proto = sock_filter_func_proto, 8552 .is_valid_access = sock_filter_is_valid_access, 8553 .convert_ctx_access = bpf_sock_convert_ctx_access, 8554 }; 8555 8556 const struct bpf_prog_ops cg_sock_prog_ops = { 8557 }; 8558 8559 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = { 8560 .get_func_proto = sock_addr_func_proto, 8561 .is_valid_access = sock_addr_is_valid_access, 8562 .convert_ctx_access = sock_addr_convert_ctx_access, 8563 }; 8564 8565 const struct bpf_prog_ops cg_sock_addr_prog_ops = { 8566 }; 8567 8568 const struct bpf_verifier_ops sock_ops_verifier_ops = { 8569 .get_func_proto = sock_ops_func_proto, 8570 .is_valid_access = sock_ops_is_valid_access, 8571 .convert_ctx_access = sock_ops_convert_ctx_access, 8572 }; 8573 8574 const struct bpf_prog_ops sock_ops_prog_ops = { 8575 }; 8576 8577 const struct bpf_verifier_ops sk_skb_verifier_ops = { 8578 .get_func_proto = sk_skb_func_proto, 8579 .is_valid_access = sk_skb_is_valid_access, 8580 .convert_ctx_access = sk_skb_convert_ctx_access, 8581 .gen_prologue = sk_skb_prologue, 8582 }; 8583 8584 const struct bpf_prog_ops sk_skb_prog_ops = { 8585 }; 8586 8587 const struct bpf_verifier_ops sk_msg_verifier_ops = { 8588 .get_func_proto = sk_msg_func_proto, 8589 .is_valid_access = sk_msg_is_valid_access, 8590 .convert_ctx_access = sk_msg_convert_ctx_access, 8591 .gen_prologue = bpf_noop_prologue, 8592 }; 8593 8594 const struct bpf_prog_ops sk_msg_prog_ops = { 8595 }; 8596 8597 const struct bpf_verifier_ops flow_dissector_verifier_ops = { 8598 .get_func_proto = flow_dissector_func_proto, 8599 .is_valid_access = flow_dissector_is_valid_access, 8600 .convert_ctx_access = flow_dissector_convert_ctx_access, 8601 }; 8602 8603 const struct bpf_prog_ops flow_dissector_prog_ops = { 8604 .test_run = bpf_prog_test_run_flow_dissector, 8605 }; 8606 8607 int sk_detach_filter(struct sock *sk) 8608 { 8609 int ret = -ENOENT; 8610 struct sk_filter *filter; 8611 8612 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 8613 return -EPERM; 8614 8615 filter = rcu_dereference_protected(sk->sk_filter, 8616 lockdep_sock_is_held(sk)); 8617 if (filter) { 8618 RCU_INIT_POINTER(sk->sk_filter, NULL); 8619 sk_filter_uncharge(sk, filter); 8620 ret = 0; 8621 } 8622 8623 return ret; 8624 } 8625 EXPORT_SYMBOL_GPL(sk_detach_filter); 8626 8627 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, 8628 unsigned int len) 8629 { 8630 struct sock_fprog_kern *fprog; 8631 struct sk_filter *filter; 8632 int ret = 0; 8633 8634 lock_sock(sk); 8635 filter = rcu_dereference_protected(sk->sk_filter, 8636 lockdep_sock_is_held(sk)); 8637 if (!filter) 8638 goto out; 8639 8640 /* We're copying the filter that has been originally attached, 8641 * so no conversion/decode needed anymore. eBPF programs that 8642 * have no original program cannot be dumped through this. 8643 */ 8644 ret = -EACCES; 8645 fprog = filter->prog->orig_prog; 8646 if (!fprog) 8647 goto out; 8648 8649 ret = fprog->len; 8650 if (!len) 8651 /* User space only enquires number of filter blocks. */ 8652 goto out; 8653 8654 ret = -EINVAL; 8655 if (len < fprog->len) 8656 goto out; 8657 8658 ret = -EFAULT; 8659 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog))) 8660 goto out; 8661 8662 /* Instead of bytes, the API requests to return the number 8663 * of filter blocks. 8664 */ 8665 ret = fprog->len; 8666 out: 8667 release_sock(sk); 8668 return ret; 8669 } 8670 8671 #ifdef CONFIG_INET 8672 struct sk_reuseport_kern { 8673 struct sk_buff *skb; 8674 struct sock *sk; 8675 struct sock *selected_sk; 8676 void *data_end; 8677 u32 hash; 8678 u32 reuseport_id; 8679 bool bind_inany; 8680 }; 8681 8682 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern, 8683 struct sock_reuseport *reuse, 8684 struct sock *sk, struct sk_buff *skb, 8685 u32 hash) 8686 { 8687 reuse_kern->skb = skb; 8688 reuse_kern->sk = sk; 8689 reuse_kern->selected_sk = NULL; 8690 reuse_kern->data_end = skb->data + skb_headlen(skb); 8691 reuse_kern->hash = hash; 8692 reuse_kern->reuseport_id = reuse->reuseport_id; 8693 reuse_kern->bind_inany = reuse->bind_inany; 8694 } 8695 8696 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 8697 struct bpf_prog *prog, struct sk_buff *skb, 8698 u32 hash) 8699 { 8700 struct sk_reuseport_kern reuse_kern; 8701 enum sk_action action; 8702 8703 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash); 8704 action = BPF_PROG_RUN(prog, &reuse_kern); 8705 8706 if (action == SK_PASS) 8707 return reuse_kern.selected_sk; 8708 else 8709 return ERR_PTR(-ECONNREFUSED); 8710 } 8711 8712 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern, 8713 struct bpf_map *, map, void *, key, u32, flags) 8714 { 8715 struct sock_reuseport *reuse; 8716 struct sock *selected_sk; 8717 8718 selected_sk = map->ops->map_lookup_elem(map, key); 8719 if (!selected_sk) 8720 return -ENOENT; 8721 8722 reuse = rcu_dereference(selected_sk->sk_reuseport_cb); 8723 if (!reuse) 8724 /* selected_sk is unhashed (e.g. by close()) after the 8725 * above map_lookup_elem(). Treat selected_sk has already 8726 * been removed from the map. 8727 */ 8728 return -ENOENT; 8729 8730 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) { 8731 struct sock *sk; 8732 8733 if (unlikely(!reuse_kern->reuseport_id)) 8734 /* There is a small race between adding the 8735 * sk to the map and setting the 8736 * reuse_kern->reuseport_id. 8737 * Treat it as the sk has not been added to 8738 * the bpf map yet. 8739 */ 8740 return -ENOENT; 8741 8742 sk = reuse_kern->sk; 8743 if (sk->sk_protocol != selected_sk->sk_protocol) 8744 return -EPROTOTYPE; 8745 else if (sk->sk_family != selected_sk->sk_family) 8746 return -EAFNOSUPPORT; 8747 8748 /* Catch all. Likely bound to a different sockaddr. */ 8749 return -EBADFD; 8750 } 8751 8752 reuse_kern->selected_sk = selected_sk; 8753 8754 return 0; 8755 } 8756 8757 static const struct bpf_func_proto sk_select_reuseport_proto = { 8758 .func = sk_select_reuseport, 8759 .gpl_only = false, 8760 .ret_type = RET_INTEGER, 8761 .arg1_type = ARG_PTR_TO_CTX, 8762 .arg2_type = ARG_CONST_MAP_PTR, 8763 .arg3_type = ARG_PTR_TO_MAP_KEY, 8764 .arg4_type = ARG_ANYTHING, 8765 }; 8766 8767 BPF_CALL_4(sk_reuseport_load_bytes, 8768 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 8769 void *, to, u32, len) 8770 { 8771 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len); 8772 } 8773 8774 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = { 8775 .func = sk_reuseport_load_bytes, 8776 .gpl_only = false, 8777 .ret_type = RET_INTEGER, 8778 .arg1_type = ARG_PTR_TO_CTX, 8779 .arg2_type = ARG_ANYTHING, 8780 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 8781 .arg4_type = ARG_CONST_SIZE, 8782 }; 8783 8784 BPF_CALL_5(sk_reuseport_load_bytes_relative, 8785 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 8786 void *, to, u32, len, u32, start_header) 8787 { 8788 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to, 8789 len, start_header); 8790 } 8791 8792 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = { 8793 .func = sk_reuseport_load_bytes_relative, 8794 .gpl_only = false, 8795 .ret_type = RET_INTEGER, 8796 .arg1_type = ARG_PTR_TO_CTX, 8797 .arg2_type = ARG_ANYTHING, 8798 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 8799 .arg4_type = ARG_CONST_SIZE, 8800 .arg5_type = ARG_ANYTHING, 8801 }; 8802 8803 static const struct bpf_func_proto * 8804 sk_reuseport_func_proto(enum bpf_func_id func_id, 8805 const struct bpf_prog *prog) 8806 { 8807 switch (func_id) { 8808 case BPF_FUNC_sk_select_reuseport: 8809 return &sk_select_reuseport_proto; 8810 case BPF_FUNC_skb_load_bytes: 8811 return &sk_reuseport_load_bytes_proto; 8812 case BPF_FUNC_skb_load_bytes_relative: 8813 return &sk_reuseport_load_bytes_relative_proto; 8814 default: 8815 return bpf_base_func_proto(func_id); 8816 } 8817 } 8818 8819 static bool 8820 sk_reuseport_is_valid_access(int off, int size, 8821 enum bpf_access_type type, 8822 const struct bpf_prog *prog, 8823 struct bpf_insn_access_aux *info) 8824 { 8825 const u32 size_default = sizeof(__u32); 8826 8827 if (off < 0 || off >= sizeof(struct sk_reuseport_md) || 8828 off % size || type != BPF_READ) 8829 return false; 8830 8831 switch (off) { 8832 case offsetof(struct sk_reuseport_md, data): 8833 info->reg_type = PTR_TO_PACKET; 8834 return size == sizeof(__u64); 8835 8836 case offsetof(struct sk_reuseport_md, data_end): 8837 info->reg_type = PTR_TO_PACKET_END; 8838 return size == sizeof(__u64); 8839 8840 case offsetof(struct sk_reuseport_md, hash): 8841 return size == size_default; 8842 8843 /* Fields that allow narrowing */ 8844 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol): 8845 if (size < FIELD_SIZEOF(struct sk_buff, protocol)) 8846 return false; 8847 /* fall through */ 8848 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol): 8849 case bpf_ctx_range(struct sk_reuseport_md, bind_inany): 8850 case bpf_ctx_range(struct sk_reuseport_md, len): 8851 bpf_ctx_record_field_size(info, size_default); 8852 return bpf_ctx_narrow_access_ok(off, size, size_default); 8853 8854 default: 8855 return false; 8856 } 8857 } 8858 8859 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \ 8860 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \ 8861 si->dst_reg, si->src_reg, \ 8862 bpf_target_off(struct sk_reuseport_kern, F, \ 8863 FIELD_SIZEOF(struct sk_reuseport_kern, F), \ 8864 target_size)); \ 8865 }) 8866 8867 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \ 8868 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 8869 struct sk_buff, \ 8870 skb, \ 8871 SKB_FIELD) 8872 8873 #define SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(SK_FIELD, BPF_SIZE, EXTRA_OFF) \ 8874 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(struct sk_reuseport_kern, \ 8875 struct sock, \ 8876 sk, \ 8877 SK_FIELD, BPF_SIZE, EXTRA_OFF) 8878 8879 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type, 8880 const struct bpf_insn *si, 8881 struct bpf_insn *insn_buf, 8882 struct bpf_prog *prog, 8883 u32 *target_size) 8884 { 8885 struct bpf_insn *insn = insn_buf; 8886 8887 switch (si->off) { 8888 case offsetof(struct sk_reuseport_md, data): 8889 SK_REUSEPORT_LOAD_SKB_FIELD(data); 8890 break; 8891 8892 case offsetof(struct sk_reuseport_md, len): 8893 SK_REUSEPORT_LOAD_SKB_FIELD(len); 8894 break; 8895 8896 case offsetof(struct sk_reuseport_md, eth_protocol): 8897 SK_REUSEPORT_LOAD_SKB_FIELD(protocol); 8898 break; 8899 8900 case offsetof(struct sk_reuseport_md, ip_protocol): 8901 BUILD_BUG_ON(HWEIGHT32(SK_FL_PROTO_MASK) != BITS_PER_BYTE); 8902 SK_REUSEPORT_LOAD_SK_FIELD_SIZE_OFF(__sk_flags_offset, 8903 BPF_W, 0); 8904 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, SK_FL_PROTO_MASK); 8905 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 8906 SK_FL_PROTO_SHIFT); 8907 /* SK_FL_PROTO_MASK and SK_FL_PROTO_SHIFT are endian 8908 * aware. No further narrowing or masking is needed. 8909 */ 8910 *target_size = 1; 8911 break; 8912 8913 case offsetof(struct sk_reuseport_md, data_end): 8914 SK_REUSEPORT_LOAD_FIELD(data_end); 8915 break; 8916 8917 case offsetof(struct sk_reuseport_md, hash): 8918 SK_REUSEPORT_LOAD_FIELD(hash); 8919 break; 8920 8921 case offsetof(struct sk_reuseport_md, bind_inany): 8922 SK_REUSEPORT_LOAD_FIELD(bind_inany); 8923 break; 8924 } 8925 8926 return insn - insn_buf; 8927 } 8928 8929 const struct bpf_verifier_ops sk_reuseport_verifier_ops = { 8930 .get_func_proto = sk_reuseport_func_proto, 8931 .is_valid_access = sk_reuseport_is_valid_access, 8932 .convert_ctx_access = sk_reuseport_convert_ctx_access, 8933 }; 8934 8935 const struct bpf_prog_ops sk_reuseport_prog_ops = { 8936 }; 8937 #endif /* CONFIG_INET */ 8938