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(sizeof_field(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(sizeof_field(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(sizeof_field(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(sizeof_field(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(sizeof_field(struct net_device, ifindex) != 4); 342 BUILD_BUG_ON(sizeof_field(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(sizeof_field(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(sizeof_field(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 (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 skb->tstamp = 0; 2059 2060 dev_xmit_recursion_inc(); 2061 ret = dev_queue_xmit(skb); 2062 dev_xmit_recursion_dec(); 2063 2064 return ret; 2065 } 2066 2067 static int __bpf_redirect_no_mac(struct sk_buff *skb, struct net_device *dev, 2068 u32 flags) 2069 { 2070 unsigned int mlen = skb_network_offset(skb); 2071 2072 if (mlen) { 2073 __skb_pull(skb, mlen); 2074 2075 /* At ingress, the mac header has already been pulled once. 2076 * At egress, skb_pospull_rcsum has to be done in case that 2077 * the skb is originated from ingress (i.e. a forwarded skb) 2078 * to ensure that rcsum starts at net header. 2079 */ 2080 if (!skb_at_tc_ingress(skb)) 2081 skb_postpull_rcsum(skb, skb_mac_header(skb), mlen); 2082 } 2083 skb_pop_mac_header(skb); 2084 skb_reset_mac_len(skb); 2085 return flags & BPF_F_INGRESS ? 2086 __bpf_rx_skb_no_mac(dev, skb) : __bpf_tx_skb(dev, skb); 2087 } 2088 2089 static int __bpf_redirect_common(struct sk_buff *skb, struct net_device *dev, 2090 u32 flags) 2091 { 2092 /* Verify that a link layer header is carried */ 2093 if (unlikely(skb->mac_header >= skb->network_header)) { 2094 kfree_skb(skb); 2095 return -ERANGE; 2096 } 2097 2098 bpf_push_mac_rcsum(skb); 2099 return flags & BPF_F_INGRESS ? 2100 __bpf_rx_skb(dev, skb) : __bpf_tx_skb(dev, skb); 2101 } 2102 2103 static int __bpf_redirect(struct sk_buff *skb, struct net_device *dev, 2104 u32 flags) 2105 { 2106 if (dev_is_mac_header_xmit(dev)) 2107 return __bpf_redirect_common(skb, dev, flags); 2108 else 2109 return __bpf_redirect_no_mac(skb, dev, flags); 2110 } 2111 2112 BPF_CALL_3(bpf_clone_redirect, struct sk_buff *, skb, u32, ifindex, u64, flags) 2113 { 2114 struct net_device *dev; 2115 struct sk_buff *clone; 2116 int ret; 2117 2118 if (unlikely(flags & ~(BPF_F_INGRESS))) 2119 return -EINVAL; 2120 2121 dev = dev_get_by_index_rcu(dev_net(skb->dev), ifindex); 2122 if (unlikely(!dev)) 2123 return -EINVAL; 2124 2125 clone = skb_clone(skb, GFP_ATOMIC); 2126 if (unlikely(!clone)) 2127 return -ENOMEM; 2128 2129 /* For direct write, we need to keep the invariant that the skbs 2130 * we're dealing with need to be uncloned. Should uncloning fail 2131 * here, we need to free the just generated clone to unclone once 2132 * again. 2133 */ 2134 ret = bpf_try_make_head_writable(skb); 2135 if (unlikely(ret)) { 2136 kfree_skb(clone); 2137 return -ENOMEM; 2138 } 2139 2140 return __bpf_redirect(clone, dev, flags); 2141 } 2142 2143 static const struct bpf_func_proto bpf_clone_redirect_proto = { 2144 .func = bpf_clone_redirect, 2145 .gpl_only = false, 2146 .ret_type = RET_INTEGER, 2147 .arg1_type = ARG_PTR_TO_CTX, 2148 .arg2_type = ARG_ANYTHING, 2149 .arg3_type = ARG_ANYTHING, 2150 }; 2151 2152 DEFINE_PER_CPU(struct bpf_redirect_info, bpf_redirect_info); 2153 EXPORT_PER_CPU_SYMBOL_GPL(bpf_redirect_info); 2154 2155 BPF_CALL_2(bpf_redirect, u32, ifindex, u64, flags) 2156 { 2157 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2158 2159 if (unlikely(flags & ~(BPF_F_INGRESS))) 2160 return TC_ACT_SHOT; 2161 2162 ri->flags = flags; 2163 ri->tgt_index = ifindex; 2164 2165 return TC_ACT_REDIRECT; 2166 } 2167 2168 int skb_do_redirect(struct sk_buff *skb) 2169 { 2170 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 2171 struct net_device *dev; 2172 2173 dev = dev_get_by_index_rcu(dev_net(skb->dev), ri->tgt_index); 2174 ri->tgt_index = 0; 2175 if (unlikely(!dev)) { 2176 kfree_skb(skb); 2177 return -EINVAL; 2178 } 2179 2180 return __bpf_redirect(skb, dev, ri->flags); 2181 } 2182 2183 static const struct bpf_func_proto bpf_redirect_proto = { 2184 .func = bpf_redirect, 2185 .gpl_only = false, 2186 .ret_type = RET_INTEGER, 2187 .arg1_type = ARG_ANYTHING, 2188 .arg2_type = ARG_ANYTHING, 2189 }; 2190 2191 BPF_CALL_2(bpf_msg_apply_bytes, struct sk_msg *, msg, u32, bytes) 2192 { 2193 msg->apply_bytes = bytes; 2194 return 0; 2195 } 2196 2197 static const struct bpf_func_proto bpf_msg_apply_bytes_proto = { 2198 .func = bpf_msg_apply_bytes, 2199 .gpl_only = false, 2200 .ret_type = RET_INTEGER, 2201 .arg1_type = ARG_PTR_TO_CTX, 2202 .arg2_type = ARG_ANYTHING, 2203 }; 2204 2205 BPF_CALL_2(bpf_msg_cork_bytes, struct sk_msg *, msg, u32, bytes) 2206 { 2207 msg->cork_bytes = bytes; 2208 return 0; 2209 } 2210 2211 static const struct bpf_func_proto bpf_msg_cork_bytes_proto = { 2212 .func = bpf_msg_cork_bytes, 2213 .gpl_only = false, 2214 .ret_type = RET_INTEGER, 2215 .arg1_type = ARG_PTR_TO_CTX, 2216 .arg2_type = ARG_ANYTHING, 2217 }; 2218 2219 BPF_CALL_4(bpf_msg_pull_data, struct sk_msg *, msg, u32, start, 2220 u32, end, u64, flags) 2221 { 2222 u32 len = 0, offset = 0, copy = 0, poffset = 0, bytes = end - start; 2223 u32 first_sge, last_sge, i, shift, bytes_sg_total; 2224 struct scatterlist *sge; 2225 u8 *raw, *to, *from; 2226 struct page *page; 2227 2228 if (unlikely(flags || end <= start)) 2229 return -EINVAL; 2230 2231 /* First find the starting scatterlist element */ 2232 i = msg->sg.start; 2233 do { 2234 offset += len; 2235 len = sk_msg_elem(msg, i)->length; 2236 if (start < offset + len) 2237 break; 2238 sk_msg_iter_var_next(i); 2239 } while (i != msg->sg.end); 2240 2241 if (unlikely(start >= offset + len)) 2242 return -EINVAL; 2243 2244 first_sge = i; 2245 /* The start may point into the sg element so we need to also 2246 * account for the headroom. 2247 */ 2248 bytes_sg_total = start - offset + bytes; 2249 if (!test_bit(i, &msg->sg.copy) && bytes_sg_total <= len) 2250 goto out; 2251 2252 /* At this point we need to linearize multiple scatterlist 2253 * elements or a single shared page. Either way we need to 2254 * copy into a linear buffer exclusively owned by BPF. Then 2255 * place the buffer in the scatterlist and fixup the original 2256 * entries by removing the entries now in the linear buffer 2257 * and shifting the remaining entries. For now we do not try 2258 * to copy partial entries to avoid complexity of running out 2259 * of sg_entry slots. The downside is reading a single byte 2260 * will copy the entire sg entry. 2261 */ 2262 do { 2263 copy += sk_msg_elem(msg, i)->length; 2264 sk_msg_iter_var_next(i); 2265 if (bytes_sg_total <= copy) 2266 break; 2267 } while (i != msg->sg.end); 2268 last_sge = i; 2269 2270 if (unlikely(bytes_sg_total > copy)) 2271 return -EINVAL; 2272 2273 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2274 get_order(copy)); 2275 if (unlikely(!page)) 2276 return -ENOMEM; 2277 2278 raw = page_address(page); 2279 i = first_sge; 2280 do { 2281 sge = sk_msg_elem(msg, i); 2282 from = sg_virt(sge); 2283 len = sge->length; 2284 to = raw + poffset; 2285 2286 memcpy(to, from, len); 2287 poffset += len; 2288 sge->length = 0; 2289 put_page(sg_page(sge)); 2290 2291 sk_msg_iter_var_next(i); 2292 } while (i != last_sge); 2293 2294 sg_set_page(&msg->sg.data[first_sge], page, copy, 0); 2295 2296 /* To repair sg ring we need to shift entries. If we only 2297 * had a single entry though we can just replace it and 2298 * be done. Otherwise walk the ring and shift the entries. 2299 */ 2300 WARN_ON_ONCE(last_sge == first_sge); 2301 shift = last_sge > first_sge ? 2302 last_sge - first_sge - 1 : 2303 NR_MSG_FRAG_IDS - first_sge + last_sge - 1; 2304 if (!shift) 2305 goto out; 2306 2307 i = first_sge; 2308 sk_msg_iter_var_next(i); 2309 do { 2310 u32 move_from; 2311 2312 if (i + shift >= NR_MSG_FRAG_IDS) 2313 move_from = i + shift - NR_MSG_FRAG_IDS; 2314 else 2315 move_from = i + shift; 2316 if (move_from == msg->sg.end) 2317 break; 2318 2319 msg->sg.data[i] = msg->sg.data[move_from]; 2320 msg->sg.data[move_from].length = 0; 2321 msg->sg.data[move_from].page_link = 0; 2322 msg->sg.data[move_from].offset = 0; 2323 sk_msg_iter_var_next(i); 2324 } while (1); 2325 2326 msg->sg.end = msg->sg.end - shift > msg->sg.end ? 2327 msg->sg.end - shift + NR_MSG_FRAG_IDS : 2328 msg->sg.end - shift; 2329 out: 2330 msg->data = sg_virt(&msg->sg.data[first_sge]) + start - offset; 2331 msg->data_end = msg->data + bytes; 2332 return 0; 2333 } 2334 2335 static const struct bpf_func_proto bpf_msg_pull_data_proto = { 2336 .func = bpf_msg_pull_data, 2337 .gpl_only = false, 2338 .ret_type = RET_INTEGER, 2339 .arg1_type = ARG_PTR_TO_CTX, 2340 .arg2_type = ARG_ANYTHING, 2341 .arg3_type = ARG_ANYTHING, 2342 .arg4_type = ARG_ANYTHING, 2343 }; 2344 2345 BPF_CALL_4(bpf_msg_push_data, struct sk_msg *, msg, u32, start, 2346 u32, len, u64, flags) 2347 { 2348 struct scatterlist sge, nsge, nnsge, rsge = {0}, *psge; 2349 u32 new, i = 0, l = 0, space, copy = 0, offset = 0; 2350 u8 *raw, *to, *from; 2351 struct page *page; 2352 2353 if (unlikely(flags)) 2354 return -EINVAL; 2355 2356 /* First find the starting scatterlist element */ 2357 i = msg->sg.start; 2358 do { 2359 offset += l; 2360 l = sk_msg_elem(msg, i)->length; 2361 2362 if (start < offset + l) 2363 break; 2364 sk_msg_iter_var_next(i); 2365 } while (i != msg->sg.end); 2366 2367 if (start >= offset + l) 2368 return -EINVAL; 2369 2370 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2371 2372 /* If no space available will fallback to copy, we need at 2373 * least one scatterlist elem available to push data into 2374 * when start aligns to the beginning of an element or two 2375 * when it falls inside an element. We handle the start equals 2376 * offset case because its the common case for inserting a 2377 * header. 2378 */ 2379 if (!space || (space == 1 && start != offset)) 2380 copy = msg->sg.data[i].length; 2381 2382 page = alloc_pages(__GFP_NOWARN | GFP_ATOMIC | __GFP_COMP, 2383 get_order(copy + len)); 2384 if (unlikely(!page)) 2385 return -ENOMEM; 2386 2387 if (copy) { 2388 int front, back; 2389 2390 raw = page_address(page); 2391 2392 psge = sk_msg_elem(msg, i); 2393 front = start - offset; 2394 back = psge->length - front; 2395 from = sg_virt(psge); 2396 2397 if (front) 2398 memcpy(raw, from, front); 2399 2400 if (back) { 2401 from += front; 2402 to = raw + front + len; 2403 2404 memcpy(to, from, back); 2405 } 2406 2407 put_page(sg_page(psge)); 2408 } else if (start - offset) { 2409 psge = sk_msg_elem(msg, i); 2410 rsge = sk_msg_elem_cpy(msg, i); 2411 2412 psge->length = start - offset; 2413 rsge.length -= psge->length; 2414 rsge.offset += start; 2415 2416 sk_msg_iter_var_next(i); 2417 sg_unmark_end(psge); 2418 sg_unmark_end(&rsge); 2419 sk_msg_iter_next(msg, end); 2420 } 2421 2422 /* Slot(s) to place newly allocated data */ 2423 new = i; 2424 2425 /* Shift one or two slots as needed */ 2426 if (!copy) { 2427 sge = sk_msg_elem_cpy(msg, i); 2428 2429 sk_msg_iter_var_next(i); 2430 sg_unmark_end(&sge); 2431 sk_msg_iter_next(msg, end); 2432 2433 nsge = sk_msg_elem_cpy(msg, i); 2434 if (rsge.length) { 2435 sk_msg_iter_var_next(i); 2436 nnsge = sk_msg_elem_cpy(msg, i); 2437 } 2438 2439 while (i != msg->sg.end) { 2440 msg->sg.data[i] = sge; 2441 sge = nsge; 2442 sk_msg_iter_var_next(i); 2443 if (rsge.length) { 2444 nsge = nnsge; 2445 nnsge = sk_msg_elem_cpy(msg, i); 2446 } else { 2447 nsge = sk_msg_elem_cpy(msg, i); 2448 } 2449 } 2450 } 2451 2452 /* Place newly allocated data buffer */ 2453 sk_mem_charge(msg->sk, len); 2454 msg->sg.size += len; 2455 __clear_bit(new, &msg->sg.copy); 2456 sg_set_page(&msg->sg.data[new], page, len + copy, 0); 2457 if (rsge.length) { 2458 get_page(sg_page(&rsge)); 2459 sk_msg_iter_var_next(new); 2460 msg->sg.data[new] = rsge; 2461 } 2462 2463 sk_msg_compute_data_pointers(msg); 2464 return 0; 2465 } 2466 2467 static const struct bpf_func_proto bpf_msg_push_data_proto = { 2468 .func = bpf_msg_push_data, 2469 .gpl_only = false, 2470 .ret_type = RET_INTEGER, 2471 .arg1_type = ARG_PTR_TO_CTX, 2472 .arg2_type = ARG_ANYTHING, 2473 .arg3_type = ARG_ANYTHING, 2474 .arg4_type = ARG_ANYTHING, 2475 }; 2476 2477 static void sk_msg_shift_left(struct sk_msg *msg, int i) 2478 { 2479 int prev; 2480 2481 do { 2482 prev = i; 2483 sk_msg_iter_var_next(i); 2484 msg->sg.data[prev] = msg->sg.data[i]; 2485 } while (i != msg->sg.end); 2486 2487 sk_msg_iter_prev(msg, end); 2488 } 2489 2490 static void sk_msg_shift_right(struct sk_msg *msg, int i) 2491 { 2492 struct scatterlist tmp, sge; 2493 2494 sk_msg_iter_next(msg, end); 2495 sge = sk_msg_elem_cpy(msg, i); 2496 sk_msg_iter_var_next(i); 2497 tmp = sk_msg_elem_cpy(msg, i); 2498 2499 while (i != msg->sg.end) { 2500 msg->sg.data[i] = sge; 2501 sk_msg_iter_var_next(i); 2502 sge = tmp; 2503 tmp = sk_msg_elem_cpy(msg, i); 2504 } 2505 } 2506 2507 BPF_CALL_4(bpf_msg_pop_data, struct sk_msg *, msg, u32, start, 2508 u32, len, u64, flags) 2509 { 2510 u32 i = 0, l = 0, space, offset = 0; 2511 u64 last = start + len; 2512 int pop; 2513 2514 if (unlikely(flags)) 2515 return -EINVAL; 2516 2517 /* First find the starting scatterlist element */ 2518 i = msg->sg.start; 2519 do { 2520 offset += l; 2521 l = sk_msg_elem(msg, i)->length; 2522 2523 if (start < offset + l) 2524 break; 2525 sk_msg_iter_var_next(i); 2526 } while (i != msg->sg.end); 2527 2528 /* Bounds checks: start and pop must be inside message */ 2529 if (start >= offset + l || last >= msg->sg.size) 2530 return -EINVAL; 2531 2532 space = MAX_MSG_FRAGS - sk_msg_elem_used(msg); 2533 2534 pop = len; 2535 /* --------------| offset 2536 * -| start |-------- len -------| 2537 * 2538 * |----- a ----|-------- pop -------|----- b ----| 2539 * |______________________________________________| length 2540 * 2541 * 2542 * a: region at front of scatter element to save 2543 * b: region at back of scatter element to save when length > A + pop 2544 * pop: region to pop from element, same as input 'pop' here will be 2545 * decremented below per iteration. 2546 * 2547 * Two top-level cases to handle when start != offset, first B is non 2548 * zero and second B is zero corresponding to when a pop includes more 2549 * than one element. 2550 * 2551 * Then if B is non-zero AND there is no space allocate space and 2552 * compact A, B regions into page. If there is space shift ring to 2553 * the rigth free'ing the next element in ring to place B, leaving 2554 * A untouched except to reduce length. 2555 */ 2556 if (start != offset) { 2557 struct scatterlist *nsge, *sge = sk_msg_elem(msg, i); 2558 int a = start; 2559 int b = sge->length - pop - a; 2560 2561 sk_msg_iter_var_next(i); 2562 2563 if (pop < sge->length - a) { 2564 if (space) { 2565 sge->length = a; 2566 sk_msg_shift_right(msg, i); 2567 nsge = sk_msg_elem(msg, i); 2568 get_page(sg_page(sge)); 2569 sg_set_page(nsge, 2570 sg_page(sge), 2571 b, sge->offset + pop + a); 2572 } else { 2573 struct page *page, *orig; 2574 u8 *to, *from; 2575 2576 page = alloc_pages(__GFP_NOWARN | 2577 __GFP_COMP | GFP_ATOMIC, 2578 get_order(a + b)); 2579 if (unlikely(!page)) 2580 return -ENOMEM; 2581 2582 sge->length = a; 2583 orig = sg_page(sge); 2584 from = sg_virt(sge); 2585 to = page_address(page); 2586 memcpy(to, from, a); 2587 memcpy(to + a, from + a + pop, b); 2588 sg_set_page(sge, page, a + b, 0); 2589 put_page(orig); 2590 } 2591 pop = 0; 2592 } else if (pop >= sge->length - a) { 2593 sge->length = a; 2594 pop -= (sge->length - a); 2595 } 2596 } 2597 2598 /* From above the current layout _must_ be as follows, 2599 * 2600 * -| offset 2601 * -| start 2602 * 2603 * |---- pop ---|---------------- b ------------| 2604 * |____________________________________________| length 2605 * 2606 * Offset and start of the current msg elem are equal because in the 2607 * previous case we handled offset != start and either consumed the 2608 * entire element and advanced to the next element OR pop == 0. 2609 * 2610 * Two cases to handle here are first pop is less than the length 2611 * leaving some remainder b above. Simply adjust the element's layout 2612 * in this case. Or pop >= length of the element so that b = 0. In this 2613 * case advance to next element decrementing pop. 2614 */ 2615 while (pop) { 2616 struct scatterlist *sge = sk_msg_elem(msg, i); 2617 2618 if (pop < sge->length) { 2619 sge->length -= pop; 2620 sge->offset += pop; 2621 pop = 0; 2622 } else { 2623 pop -= sge->length; 2624 sk_msg_shift_left(msg, i); 2625 } 2626 sk_msg_iter_var_next(i); 2627 } 2628 2629 sk_mem_uncharge(msg->sk, len - pop); 2630 msg->sg.size -= (len - pop); 2631 sk_msg_compute_data_pointers(msg); 2632 return 0; 2633 } 2634 2635 static const struct bpf_func_proto bpf_msg_pop_data_proto = { 2636 .func = bpf_msg_pop_data, 2637 .gpl_only = false, 2638 .ret_type = RET_INTEGER, 2639 .arg1_type = ARG_PTR_TO_CTX, 2640 .arg2_type = ARG_ANYTHING, 2641 .arg3_type = ARG_ANYTHING, 2642 .arg4_type = ARG_ANYTHING, 2643 }; 2644 2645 #ifdef CONFIG_CGROUP_NET_CLASSID 2646 BPF_CALL_0(bpf_get_cgroup_classid_curr) 2647 { 2648 return __task_get_classid(current); 2649 } 2650 2651 static const struct bpf_func_proto bpf_get_cgroup_classid_curr_proto = { 2652 .func = bpf_get_cgroup_classid_curr, 2653 .gpl_only = false, 2654 .ret_type = RET_INTEGER, 2655 }; 2656 #endif 2657 2658 BPF_CALL_1(bpf_get_cgroup_classid, const struct sk_buff *, skb) 2659 { 2660 return task_get_classid(skb); 2661 } 2662 2663 static const struct bpf_func_proto bpf_get_cgroup_classid_proto = { 2664 .func = bpf_get_cgroup_classid, 2665 .gpl_only = false, 2666 .ret_type = RET_INTEGER, 2667 .arg1_type = ARG_PTR_TO_CTX, 2668 }; 2669 2670 BPF_CALL_1(bpf_get_route_realm, const struct sk_buff *, skb) 2671 { 2672 return dst_tclassid(skb); 2673 } 2674 2675 static const struct bpf_func_proto bpf_get_route_realm_proto = { 2676 .func = bpf_get_route_realm, 2677 .gpl_only = false, 2678 .ret_type = RET_INTEGER, 2679 .arg1_type = ARG_PTR_TO_CTX, 2680 }; 2681 2682 BPF_CALL_1(bpf_get_hash_recalc, struct sk_buff *, skb) 2683 { 2684 /* If skb_clear_hash() was called due to mangling, we can 2685 * trigger SW recalculation here. Later access to hash 2686 * can then use the inline skb->hash via context directly 2687 * instead of calling this helper again. 2688 */ 2689 return skb_get_hash(skb); 2690 } 2691 2692 static const struct bpf_func_proto bpf_get_hash_recalc_proto = { 2693 .func = bpf_get_hash_recalc, 2694 .gpl_only = false, 2695 .ret_type = RET_INTEGER, 2696 .arg1_type = ARG_PTR_TO_CTX, 2697 }; 2698 2699 BPF_CALL_1(bpf_set_hash_invalid, struct sk_buff *, skb) 2700 { 2701 /* After all direct packet write, this can be used once for 2702 * triggering a lazy recalc on next skb_get_hash() invocation. 2703 */ 2704 skb_clear_hash(skb); 2705 return 0; 2706 } 2707 2708 static const struct bpf_func_proto bpf_set_hash_invalid_proto = { 2709 .func = bpf_set_hash_invalid, 2710 .gpl_only = false, 2711 .ret_type = RET_INTEGER, 2712 .arg1_type = ARG_PTR_TO_CTX, 2713 }; 2714 2715 BPF_CALL_2(bpf_set_hash, struct sk_buff *, skb, u32, hash) 2716 { 2717 /* Set user specified hash as L4(+), so that it gets returned 2718 * on skb_get_hash() call unless BPF prog later on triggers a 2719 * skb_clear_hash(). 2720 */ 2721 __skb_set_sw_hash(skb, hash, true); 2722 return 0; 2723 } 2724 2725 static const struct bpf_func_proto bpf_set_hash_proto = { 2726 .func = bpf_set_hash, 2727 .gpl_only = false, 2728 .ret_type = RET_INTEGER, 2729 .arg1_type = ARG_PTR_TO_CTX, 2730 .arg2_type = ARG_ANYTHING, 2731 }; 2732 2733 BPF_CALL_3(bpf_skb_vlan_push, struct sk_buff *, skb, __be16, vlan_proto, 2734 u16, vlan_tci) 2735 { 2736 int ret; 2737 2738 if (unlikely(vlan_proto != htons(ETH_P_8021Q) && 2739 vlan_proto != htons(ETH_P_8021AD))) 2740 vlan_proto = htons(ETH_P_8021Q); 2741 2742 bpf_push_mac_rcsum(skb); 2743 ret = skb_vlan_push(skb, vlan_proto, vlan_tci); 2744 bpf_pull_mac_rcsum(skb); 2745 2746 bpf_compute_data_pointers(skb); 2747 return ret; 2748 } 2749 2750 static const struct bpf_func_proto bpf_skb_vlan_push_proto = { 2751 .func = bpf_skb_vlan_push, 2752 .gpl_only = false, 2753 .ret_type = RET_INTEGER, 2754 .arg1_type = ARG_PTR_TO_CTX, 2755 .arg2_type = ARG_ANYTHING, 2756 .arg3_type = ARG_ANYTHING, 2757 }; 2758 2759 BPF_CALL_1(bpf_skb_vlan_pop, struct sk_buff *, skb) 2760 { 2761 int ret; 2762 2763 bpf_push_mac_rcsum(skb); 2764 ret = skb_vlan_pop(skb); 2765 bpf_pull_mac_rcsum(skb); 2766 2767 bpf_compute_data_pointers(skb); 2768 return ret; 2769 } 2770 2771 static const struct bpf_func_proto bpf_skb_vlan_pop_proto = { 2772 .func = bpf_skb_vlan_pop, 2773 .gpl_only = false, 2774 .ret_type = RET_INTEGER, 2775 .arg1_type = ARG_PTR_TO_CTX, 2776 }; 2777 2778 static int bpf_skb_generic_push(struct sk_buff *skb, u32 off, u32 len) 2779 { 2780 /* Caller already did skb_cow() with len as headroom, 2781 * so no need to do it here. 2782 */ 2783 skb_push(skb, len); 2784 memmove(skb->data, skb->data + len, off); 2785 memset(skb->data + off, 0, len); 2786 2787 /* No skb_postpush_rcsum(skb, skb->data + off, len) 2788 * needed here as it does not change the skb->csum 2789 * result for checksum complete when summing over 2790 * zeroed blocks. 2791 */ 2792 return 0; 2793 } 2794 2795 static int bpf_skb_generic_pop(struct sk_buff *skb, u32 off, u32 len) 2796 { 2797 /* skb_ensure_writable() is not needed here, as we're 2798 * already working on an uncloned skb. 2799 */ 2800 if (unlikely(!pskb_may_pull(skb, off + len))) 2801 return -ENOMEM; 2802 2803 skb_postpull_rcsum(skb, skb->data + off, len); 2804 memmove(skb->data + len, skb->data, off); 2805 __skb_pull(skb, len); 2806 2807 return 0; 2808 } 2809 2810 static int bpf_skb_net_hdr_push(struct sk_buff *skb, u32 off, u32 len) 2811 { 2812 bool trans_same = skb->transport_header == skb->network_header; 2813 int ret; 2814 2815 /* There's no need for __skb_push()/__skb_pull() pair to 2816 * get to the start of the mac header as we're guaranteed 2817 * to always start from here under eBPF. 2818 */ 2819 ret = bpf_skb_generic_push(skb, off, len); 2820 if (likely(!ret)) { 2821 skb->mac_header -= len; 2822 skb->network_header -= len; 2823 if (trans_same) 2824 skb->transport_header = skb->network_header; 2825 } 2826 2827 return ret; 2828 } 2829 2830 static int bpf_skb_net_hdr_pop(struct sk_buff *skb, u32 off, u32 len) 2831 { 2832 bool trans_same = skb->transport_header == skb->network_header; 2833 int ret; 2834 2835 /* Same here, __skb_push()/__skb_pull() pair not needed. */ 2836 ret = bpf_skb_generic_pop(skb, off, len); 2837 if (likely(!ret)) { 2838 skb->mac_header += len; 2839 skb->network_header += len; 2840 if (trans_same) 2841 skb->transport_header = skb->network_header; 2842 } 2843 2844 return ret; 2845 } 2846 2847 static int bpf_skb_proto_4_to_6(struct sk_buff *skb) 2848 { 2849 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 2850 u32 off = skb_mac_header_len(skb); 2851 int ret; 2852 2853 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) 2854 return -ENOTSUPP; 2855 2856 ret = skb_cow(skb, len_diff); 2857 if (unlikely(ret < 0)) 2858 return ret; 2859 2860 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 2861 if (unlikely(ret < 0)) 2862 return ret; 2863 2864 if (skb_is_gso(skb)) { 2865 struct skb_shared_info *shinfo = skb_shinfo(skb); 2866 2867 /* SKB_GSO_TCPV4 needs to be changed into 2868 * SKB_GSO_TCPV6. 2869 */ 2870 if (shinfo->gso_type & SKB_GSO_TCPV4) { 2871 shinfo->gso_type &= ~SKB_GSO_TCPV4; 2872 shinfo->gso_type |= SKB_GSO_TCPV6; 2873 } 2874 2875 /* Due to IPv6 header, MSS needs to be downgraded. */ 2876 skb_decrease_gso_size(shinfo, len_diff); 2877 /* Header must be checked, and gso_segs recomputed. */ 2878 shinfo->gso_type |= SKB_GSO_DODGY; 2879 shinfo->gso_segs = 0; 2880 } 2881 2882 skb->protocol = htons(ETH_P_IPV6); 2883 skb_clear_hash(skb); 2884 2885 return 0; 2886 } 2887 2888 static int bpf_skb_proto_6_to_4(struct sk_buff *skb) 2889 { 2890 const u32 len_diff = sizeof(struct ipv6hdr) - sizeof(struct iphdr); 2891 u32 off = skb_mac_header_len(skb); 2892 int ret; 2893 2894 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) 2895 return -ENOTSUPP; 2896 2897 ret = skb_unclone(skb, GFP_ATOMIC); 2898 if (unlikely(ret < 0)) 2899 return ret; 2900 2901 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 2902 if (unlikely(ret < 0)) 2903 return ret; 2904 2905 if (skb_is_gso(skb)) { 2906 struct skb_shared_info *shinfo = skb_shinfo(skb); 2907 2908 /* SKB_GSO_TCPV6 needs to be changed into 2909 * SKB_GSO_TCPV4. 2910 */ 2911 if (shinfo->gso_type & SKB_GSO_TCPV6) { 2912 shinfo->gso_type &= ~SKB_GSO_TCPV6; 2913 shinfo->gso_type |= SKB_GSO_TCPV4; 2914 } 2915 2916 /* Due to IPv4 header, MSS can be upgraded. */ 2917 skb_increase_gso_size(shinfo, len_diff); 2918 /* Header must be checked, and gso_segs recomputed. */ 2919 shinfo->gso_type |= SKB_GSO_DODGY; 2920 shinfo->gso_segs = 0; 2921 } 2922 2923 skb->protocol = htons(ETH_P_IP); 2924 skb_clear_hash(skb); 2925 2926 return 0; 2927 } 2928 2929 static int bpf_skb_proto_xlat(struct sk_buff *skb, __be16 to_proto) 2930 { 2931 __be16 from_proto = skb->protocol; 2932 2933 if (from_proto == htons(ETH_P_IP) && 2934 to_proto == htons(ETH_P_IPV6)) 2935 return bpf_skb_proto_4_to_6(skb); 2936 2937 if (from_proto == htons(ETH_P_IPV6) && 2938 to_proto == htons(ETH_P_IP)) 2939 return bpf_skb_proto_6_to_4(skb); 2940 2941 return -ENOTSUPP; 2942 } 2943 2944 BPF_CALL_3(bpf_skb_change_proto, struct sk_buff *, skb, __be16, proto, 2945 u64, flags) 2946 { 2947 int ret; 2948 2949 if (unlikely(flags)) 2950 return -EINVAL; 2951 2952 /* General idea is that this helper does the basic groundwork 2953 * needed for changing the protocol, and eBPF program fills the 2954 * rest through bpf_skb_store_bytes(), bpf_lX_csum_replace() 2955 * and other helpers, rather than passing a raw buffer here. 2956 * 2957 * The rationale is to keep this minimal and without a need to 2958 * deal with raw packet data. F.e. even if we would pass buffers 2959 * here, the program still needs to call the bpf_lX_csum_replace() 2960 * helpers anyway. Plus, this way we keep also separation of 2961 * concerns, since f.e. bpf_skb_store_bytes() should only take 2962 * care of stores. 2963 * 2964 * Currently, additional options and extension header space are 2965 * not supported, but flags register is reserved so we can adapt 2966 * that. For offloads, we mark packet as dodgy, so that headers 2967 * need to be verified first. 2968 */ 2969 ret = bpf_skb_proto_xlat(skb, proto); 2970 bpf_compute_data_pointers(skb); 2971 return ret; 2972 } 2973 2974 static const struct bpf_func_proto bpf_skb_change_proto_proto = { 2975 .func = bpf_skb_change_proto, 2976 .gpl_only = false, 2977 .ret_type = RET_INTEGER, 2978 .arg1_type = ARG_PTR_TO_CTX, 2979 .arg2_type = ARG_ANYTHING, 2980 .arg3_type = ARG_ANYTHING, 2981 }; 2982 2983 BPF_CALL_2(bpf_skb_change_type, struct sk_buff *, skb, u32, pkt_type) 2984 { 2985 /* We only allow a restricted subset to be changed for now. */ 2986 if (unlikely(!skb_pkt_type_ok(skb->pkt_type) || 2987 !skb_pkt_type_ok(pkt_type))) 2988 return -EINVAL; 2989 2990 skb->pkt_type = pkt_type; 2991 return 0; 2992 } 2993 2994 static const struct bpf_func_proto bpf_skb_change_type_proto = { 2995 .func = bpf_skb_change_type, 2996 .gpl_only = false, 2997 .ret_type = RET_INTEGER, 2998 .arg1_type = ARG_PTR_TO_CTX, 2999 .arg2_type = ARG_ANYTHING, 3000 }; 3001 3002 static u32 bpf_skb_net_base_len(const struct sk_buff *skb) 3003 { 3004 switch (skb->protocol) { 3005 case htons(ETH_P_IP): 3006 return sizeof(struct iphdr); 3007 case htons(ETH_P_IPV6): 3008 return sizeof(struct ipv6hdr); 3009 default: 3010 return ~0U; 3011 } 3012 } 3013 3014 #define BPF_F_ADJ_ROOM_ENCAP_L3_MASK (BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 | \ 3015 BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3016 3017 #define BPF_F_ADJ_ROOM_MASK (BPF_F_ADJ_ROOM_FIXED_GSO | \ 3018 BPF_F_ADJ_ROOM_ENCAP_L3_MASK | \ 3019 BPF_F_ADJ_ROOM_ENCAP_L4_GRE | \ 3020 BPF_F_ADJ_ROOM_ENCAP_L4_UDP | \ 3021 BPF_F_ADJ_ROOM_ENCAP_L2( \ 3022 BPF_ADJ_ROOM_ENCAP_L2_MASK)) 3023 3024 static int bpf_skb_net_grow(struct sk_buff *skb, u32 off, u32 len_diff, 3025 u64 flags) 3026 { 3027 u8 inner_mac_len = flags >> BPF_ADJ_ROOM_ENCAP_L2_SHIFT; 3028 bool encap = flags & BPF_F_ADJ_ROOM_ENCAP_L3_MASK; 3029 u16 mac_len = 0, inner_net = 0, inner_trans = 0; 3030 unsigned int gso_type = SKB_GSO_DODGY; 3031 int ret; 3032 3033 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3034 /* udp gso_size delineates datagrams, only allow if fixed */ 3035 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3036 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3037 return -ENOTSUPP; 3038 } 3039 3040 ret = skb_cow_head(skb, len_diff); 3041 if (unlikely(ret < 0)) 3042 return ret; 3043 3044 if (encap) { 3045 if (skb->protocol != htons(ETH_P_IP) && 3046 skb->protocol != htons(ETH_P_IPV6)) 3047 return -ENOTSUPP; 3048 3049 if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4 && 3050 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3051 return -EINVAL; 3052 3053 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE && 3054 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3055 return -EINVAL; 3056 3057 if (skb->encapsulation) 3058 return -EALREADY; 3059 3060 mac_len = skb->network_header - skb->mac_header; 3061 inner_net = skb->network_header; 3062 if (inner_mac_len > len_diff) 3063 return -EINVAL; 3064 inner_trans = skb->transport_header; 3065 } 3066 3067 ret = bpf_skb_net_hdr_push(skb, off, len_diff); 3068 if (unlikely(ret < 0)) 3069 return ret; 3070 3071 if (encap) { 3072 skb->inner_mac_header = inner_net - inner_mac_len; 3073 skb->inner_network_header = inner_net; 3074 skb->inner_transport_header = inner_trans; 3075 skb_set_inner_protocol(skb, skb->protocol); 3076 3077 skb->encapsulation = 1; 3078 skb_set_network_header(skb, mac_len); 3079 3080 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) 3081 gso_type |= SKB_GSO_UDP_TUNNEL; 3082 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE) 3083 gso_type |= SKB_GSO_GRE; 3084 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3085 gso_type |= SKB_GSO_IPXIP6; 3086 else if (flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3087 gso_type |= SKB_GSO_IPXIP4; 3088 3089 if (flags & BPF_F_ADJ_ROOM_ENCAP_L4_GRE || 3090 flags & BPF_F_ADJ_ROOM_ENCAP_L4_UDP) { 3091 int nh_len = flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6 ? 3092 sizeof(struct ipv6hdr) : 3093 sizeof(struct iphdr); 3094 3095 skb_set_transport_header(skb, mac_len + nh_len); 3096 } 3097 3098 /* Match skb->protocol to new outer l3 protocol */ 3099 if (skb->protocol == htons(ETH_P_IP) && 3100 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV6) 3101 skb->protocol = htons(ETH_P_IPV6); 3102 else if (skb->protocol == htons(ETH_P_IPV6) && 3103 flags & BPF_F_ADJ_ROOM_ENCAP_L3_IPV4) 3104 skb->protocol = htons(ETH_P_IP); 3105 } 3106 3107 if (skb_is_gso(skb)) { 3108 struct skb_shared_info *shinfo = skb_shinfo(skb); 3109 3110 /* Due to header grow, MSS needs to be downgraded. */ 3111 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3112 skb_decrease_gso_size(shinfo, len_diff); 3113 3114 /* Header must be checked, and gso_segs recomputed. */ 3115 shinfo->gso_type |= gso_type; 3116 shinfo->gso_segs = 0; 3117 } 3118 3119 return 0; 3120 } 3121 3122 static int bpf_skb_net_shrink(struct sk_buff *skb, u32 off, u32 len_diff, 3123 u64 flags) 3124 { 3125 int ret; 3126 3127 if (flags & ~BPF_F_ADJ_ROOM_FIXED_GSO) 3128 return -EINVAL; 3129 3130 if (skb_is_gso(skb) && !skb_is_gso_tcp(skb)) { 3131 /* udp gso_size delineates datagrams, only allow if fixed */ 3132 if (!(skb_shinfo(skb)->gso_type & SKB_GSO_UDP_L4) || 3133 !(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3134 return -ENOTSUPP; 3135 } 3136 3137 ret = skb_unclone(skb, GFP_ATOMIC); 3138 if (unlikely(ret < 0)) 3139 return ret; 3140 3141 ret = bpf_skb_net_hdr_pop(skb, off, len_diff); 3142 if (unlikely(ret < 0)) 3143 return ret; 3144 3145 if (skb_is_gso(skb)) { 3146 struct skb_shared_info *shinfo = skb_shinfo(skb); 3147 3148 /* Due to header shrink, MSS can be upgraded. */ 3149 if (!(flags & BPF_F_ADJ_ROOM_FIXED_GSO)) 3150 skb_increase_gso_size(shinfo, len_diff); 3151 3152 /* Header must be checked, and gso_segs recomputed. */ 3153 shinfo->gso_type |= SKB_GSO_DODGY; 3154 shinfo->gso_segs = 0; 3155 } 3156 3157 return 0; 3158 } 3159 3160 static u32 __bpf_skb_max_len(const struct sk_buff *skb) 3161 { 3162 return skb->dev ? skb->dev->mtu + skb->dev->hard_header_len : 3163 SKB_MAX_ALLOC; 3164 } 3165 3166 BPF_CALL_4(bpf_skb_adjust_room, struct sk_buff *, skb, s32, len_diff, 3167 u32, mode, u64, flags) 3168 { 3169 u32 len_cur, len_diff_abs = abs(len_diff); 3170 u32 len_min = bpf_skb_net_base_len(skb); 3171 u32 len_max = __bpf_skb_max_len(skb); 3172 __be16 proto = skb->protocol; 3173 bool shrink = len_diff < 0; 3174 u32 off; 3175 int ret; 3176 3177 if (unlikely(flags & ~BPF_F_ADJ_ROOM_MASK)) 3178 return -EINVAL; 3179 if (unlikely(len_diff_abs > 0xfffU)) 3180 return -EFAULT; 3181 if (unlikely(proto != htons(ETH_P_IP) && 3182 proto != htons(ETH_P_IPV6))) 3183 return -ENOTSUPP; 3184 3185 off = skb_mac_header_len(skb); 3186 switch (mode) { 3187 case BPF_ADJ_ROOM_NET: 3188 off += bpf_skb_net_base_len(skb); 3189 break; 3190 case BPF_ADJ_ROOM_MAC: 3191 break; 3192 default: 3193 return -ENOTSUPP; 3194 } 3195 3196 len_cur = skb->len - skb_network_offset(skb); 3197 if ((shrink && (len_diff_abs >= len_cur || 3198 len_cur - len_diff_abs < len_min)) || 3199 (!shrink && (skb->len + len_diff_abs > len_max && 3200 !skb_is_gso(skb)))) 3201 return -ENOTSUPP; 3202 3203 ret = shrink ? bpf_skb_net_shrink(skb, off, len_diff_abs, flags) : 3204 bpf_skb_net_grow(skb, off, len_diff_abs, flags); 3205 3206 bpf_compute_data_pointers(skb); 3207 return ret; 3208 } 3209 3210 static const struct bpf_func_proto bpf_skb_adjust_room_proto = { 3211 .func = bpf_skb_adjust_room, 3212 .gpl_only = false, 3213 .ret_type = RET_INTEGER, 3214 .arg1_type = ARG_PTR_TO_CTX, 3215 .arg2_type = ARG_ANYTHING, 3216 .arg3_type = ARG_ANYTHING, 3217 .arg4_type = ARG_ANYTHING, 3218 }; 3219 3220 static u32 __bpf_skb_min_len(const struct sk_buff *skb) 3221 { 3222 u32 min_len = skb_network_offset(skb); 3223 3224 if (skb_transport_header_was_set(skb)) 3225 min_len = skb_transport_offset(skb); 3226 if (skb->ip_summed == CHECKSUM_PARTIAL) 3227 min_len = skb_checksum_start_offset(skb) + 3228 skb->csum_offset + sizeof(__sum16); 3229 return min_len; 3230 } 3231 3232 static int bpf_skb_grow_rcsum(struct sk_buff *skb, unsigned int new_len) 3233 { 3234 unsigned int old_len = skb->len; 3235 int ret; 3236 3237 ret = __skb_grow_rcsum(skb, new_len); 3238 if (!ret) 3239 memset(skb->data + old_len, 0, new_len - old_len); 3240 return ret; 3241 } 3242 3243 static int bpf_skb_trim_rcsum(struct sk_buff *skb, unsigned int new_len) 3244 { 3245 return __skb_trim_rcsum(skb, new_len); 3246 } 3247 3248 static inline int __bpf_skb_change_tail(struct sk_buff *skb, u32 new_len, 3249 u64 flags) 3250 { 3251 u32 max_len = __bpf_skb_max_len(skb); 3252 u32 min_len = __bpf_skb_min_len(skb); 3253 int ret; 3254 3255 if (unlikely(flags || new_len > max_len || new_len < min_len)) 3256 return -EINVAL; 3257 if (skb->encapsulation) 3258 return -ENOTSUPP; 3259 3260 /* The basic idea of this helper is that it's performing the 3261 * needed work to either grow or trim an skb, and eBPF program 3262 * rewrites the rest via helpers like bpf_skb_store_bytes(), 3263 * bpf_lX_csum_replace() and others rather than passing a raw 3264 * buffer here. This one is a slow path helper and intended 3265 * for replies with control messages. 3266 * 3267 * Like in bpf_skb_change_proto(), we want to keep this rather 3268 * minimal and without protocol specifics so that we are able 3269 * to separate concerns as in bpf_skb_store_bytes() should only 3270 * be the one responsible for writing buffers. 3271 * 3272 * It's really expected to be a slow path operation here for 3273 * control message replies, so we're implicitly linearizing, 3274 * uncloning and drop offloads from the skb by this. 3275 */ 3276 ret = __bpf_try_make_writable(skb, skb->len); 3277 if (!ret) { 3278 if (new_len > skb->len) 3279 ret = bpf_skb_grow_rcsum(skb, new_len); 3280 else if (new_len < skb->len) 3281 ret = bpf_skb_trim_rcsum(skb, new_len); 3282 if (!ret && skb_is_gso(skb)) 3283 skb_gso_reset(skb); 3284 } 3285 return ret; 3286 } 3287 3288 BPF_CALL_3(bpf_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3289 u64, flags) 3290 { 3291 int ret = __bpf_skb_change_tail(skb, new_len, flags); 3292 3293 bpf_compute_data_pointers(skb); 3294 return ret; 3295 } 3296 3297 static const struct bpf_func_proto bpf_skb_change_tail_proto = { 3298 .func = bpf_skb_change_tail, 3299 .gpl_only = false, 3300 .ret_type = RET_INTEGER, 3301 .arg1_type = ARG_PTR_TO_CTX, 3302 .arg2_type = ARG_ANYTHING, 3303 .arg3_type = ARG_ANYTHING, 3304 }; 3305 3306 BPF_CALL_3(sk_skb_change_tail, struct sk_buff *, skb, u32, new_len, 3307 u64, flags) 3308 { 3309 int ret = __bpf_skb_change_tail(skb, new_len, flags); 3310 3311 bpf_compute_data_end_sk_skb(skb); 3312 return ret; 3313 } 3314 3315 static const struct bpf_func_proto sk_skb_change_tail_proto = { 3316 .func = sk_skb_change_tail, 3317 .gpl_only = false, 3318 .ret_type = RET_INTEGER, 3319 .arg1_type = ARG_PTR_TO_CTX, 3320 .arg2_type = ARG_ANYTHING, 3321 .arg3_type = ARG_ANYTHING, 3322 }; 3323 3324 static inline int __bpf_skb_change_head(struct sk_buff *skb, u32 head_room, 3325 u64 flags) 3326 { 3327 u32 max_len = __bpf_skb_max_len(skb); 3328 u32 new_len = skb->len + head_room; 3329 int ret; 3330 3331 if (unlikely(flags || (!skb_is_gso(skb) && new_len > max_len) || 3332 new_len < skb->len)) 3333 return -EINVAL; 3334 3335 ret = skb_cow(skb, head_room); 3336 if (likely(!ret)) { 3337 /* Idea for this helper is that we currently only 3338 * allow to expand on mac header. This means that 3339 * skb->protocol network header, etc, stay as is. 3340 * Compared to bpf_skb_change_tail(), we're more 3341 * flexible due to not needing to linearize or 3342 * reset GSO. Intention for this helper is to be 3343 * used by an L3 skb that needs to push mac header 3344 * for redirection into L2 device. 3345 */ 3346 __skb_push(skb, head_room); 3347 memset(skb->data, 0, head_room); 3348 skb_reset_mac_header(skb); 3349 } 3350 3351 return ret; 3352 } 3353 3354 BPF_CALL_3(bpf_skb_change_head, struct sk_buff *, skb, u32, head_room, 3355 u64, flags) 3356 { 3357 int ret = __bpf_skb_change_head(skb, head_room, flags); 3358 3359 bpf_compute_data_pointers(skb); 3360 return ret; 3361 } 3362 3363 static const struct bpf_func_proto bpf_skb_change_head_proto = { 3364 .func = bpf_skb_change_head, 3365 .gpl_only = false, 3366 .ret_type = RET_INTEGER, 3367 .arg1_type = ARG_PTR_TO_CTX, 3368 .arg2_type = ARG_ANYTHING, 3369 .arg3_type = ARG_ANYTHING, 3370 }; 3371 3372 BPF_CALL_3(sk_skb_change_head, struct sk_buff *, skb, u32, head_room, 3373 u64, flags) 3374 { 3375 int ret = __bpf_skb_change_head(skb, head_room, flags); 3376 3377 bpf_compute_data_end_sk_skb(skb); 3378 return ret; 3379 } 3380 3381 static const struct bpf_func_proto sk_skb_change_head_proto = { 3382 .func = sk_skb_change_head, 3383 .gpl_only = false, 3384 .ret_type = RET_INTEGER, 3385 .arg1_type = ARG_PTR_TO_CTX, 3386 .arg2_type = ARG_ANYTHING, 3387 .arg3_type = ARG_ANYTHING, 3388 }; 3389 static unsigned long xdp_get_metalen(const struct xdp_buff *xdp) 3390 { 3391 return xdp_data_meta_unsupported(xdp) ? 0 : 3392 xdp->data - xdp->data_meta; 3393 } 3394 3395 BPF_CALL_2(bpf_xdp_adjust_head, struct xdp_buff *, xdp, int, offset) 3396 { 3397 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 3398 unsigned long metalen = xdp_get_metalen(xdp); 3399 void *data_start = xdp_frame_end + metalen; 3400 void *data = xdp->data + offset; 3401 3402 if (unlikely(data < data_start || 3403 data > xdp->data_end - ETH_HLEN)) 3404 return -EINVAL; 3405 3406 if (metalen) 3407 memmove(xdp->data_meta + offset, 3408 xdp->data_meta, metalen); 3409 xdp->data_meta += offset; 3410 xdp->data = data; 3411 3412 return 0; 3413 } 3414 3415 static const struct bpf_func_proto bpf_xdp_adjust_head_proto = { 3416 .func = bpf_xdp_adjust_head, 3417 .gpl_only = false, 3418 .ret_type = RET_INTEGER, 3419 .arg1_type = ARG_PTR_TO_CTX, 3420 .arg2_type = ARG_ANYTHING, 3421 }; 3422 3423 BPF_CALL_2(bpf_xdp_adjust_tail, struct xdp_buff *, xdp, int, offset) 3424 { 3425 void *data_end = xdp->data_end + offset; 3426 3427 /* only shrinking is allowed for now. */ 3428 if (unlikely(offset >= 0)) 3429 return -EINVAL; 3430 3431 if (unlikely(data_end < xdp->data + ETH_HLEN)) 3432 return -EINVAL; 3433 3434 xdp->data_end = data_end; 3435 3436 return 0; 3437 } 3438 3439 static const struct bpf_func_proto bpf_xdp_adjust_tail_proto = { 3440 .func = bpf_xdp_adjust_tail, 3441 .gpl_only = false, 3442 .ret_type = RET_INTEGER, 3443 .arg1_type = ARG_PTR_TO_CTX, 3444 .arg2_type = ARG_ANYTHING, 3445 }; 3446 3447 BPF_CALL_2(bpf_xdp_adjust_meta, struct xdp_buff *, xdp, int, offset) 3448 { 3449 void *xdp_frame_end = xdp->data_hard_start + sizeof(struct xdp_frame); 3450 void *meta = xdp->data_meta + offset; 3451 unsigned long metalen = xdp->data - meta; 3452 3453 if (xdp_data_meta_unsupported(xdp)) 3454 return -ENOTSUPP; 3455 if (unlikely(meta < xdp_frame_end || 3456 meta > xdp->data)) 3457 return -EINVAL; 3458 if (unlikely((metalen & (sizeof(__u32) - 1)) || 3459 (metalen > 32))) 3460 return -EACCES; 3461 3462 xdp->data_meta = meta; 3463 3464 return 0; 3465 } 3466 3467 static const struct bpf_func_proto bpf_xdp_adjust_meta_proto = { 3468 .func = bpf_xdp_adjust_meta, 3469 .gpl_only = false, 3470 .ret_type = RET_INTEGER, 3471 .arg1_type = ARG_PTR_TO_CTX, 3472 .arg2_type = ARG_ANYTHING, 3473 }; 3474 3475 static int __bpf_tx_xdp_map(struct net_device *dev_rx, void *fwd, 3476 struct bpf_map *map, struct xdp_buff *xdp) 3477 { 3478 switch (map->map_type) { 3479 case BPF_MAP_TYPE_DEVMAP: 3480 case BPF_MAP_TYPE_DEVMAP_HASH: 3481 return dev_map_enqueue(fwd, xdp, dev_rx); 3482 case BPF_MAP_TYPE_CPUMAP: 3483 return cpu_map_enqueue(fwd, xdp, dev_rx); 3484 case BPF_MAP_TYPE_XSKMAP: 3485 return __xsk_map_redirect(fwd, xdp); 3486 default: 3487 return -EBADRQC; 3488 } 3489 return 0; 3490 } 3491 3492 void xdp_do_flush(void) 3493 { 3494 __dev_flush(); 3495 __cpu_map_flush(); 3496 __xsk_map_flush(); 3497 } 3498 EXPORT_SYMBOL_GPL(xdp_do_flush); 3499 3500 static inline void *__xdp_map_lookup_elem(struct bpf_map *map, u32 index) 3501 { 3502 switch (map->map_type) { 3503 case BPF_MAP_TYPE_DEVMAP: 3504 return __dev_map_lookup_elem(map, index); 3505 case BPF_MAP_TYPE_DEVMAP_HASH: 3506 return __dev_map_hash_lookup_elem(map, index); 3507 case BPF_MAP_TYPE_CPUMAP: 3508 return __cpu_map_lookup_elem(map, index); 3509 case BPF_MAP_TYPE_XSKMAP: 3510 return __xsk_map_lookup_elem(map, index); 3511 default: 3512 return NULL; 3513 } 3514 } 3515 3516 void bpf_clear_redirect_map(struct bpf_map *map) 3517 { 3518 struct bpf_redirect_info *ri; 3519 int cpu; 3520 3521 for_each_possible_cpu(cpu) { 3522 ri = per_cpu_ptr(&bpf_redirect_info, cpu); 3523 /* Avoid polluting remote cacheline due to writes if 3524 * not needed. Once we pass this test, we need the 3525 * cmpxchg() to make sure it hasn't been changed in 3526 * the meantime by remote CPU. 3527 */ 3528 if (unlikely(READ_ONCE(ri->map) == map)) 3529 cmpxchg(&ri->map, map, NULL); 3530 } 3531 } 3532 3533 int xdp_do_redirect(struct net_device *dev, struct xdp_buff *xdp, 3534 struct bpf_prog *xdp_prog) 3535 { 3536 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3537 struct bpf_map *map = READ_ONCE(ri->map); 3538 u32 index = ri->tgt_index; 3539 void *fwd = ri->tgt_value; 3540 int err; 3541 3542 ri->tgt_index = 0; 3543 ri->tgt_value = NULL; 3544 WRITE_ONCE(ri->map, NULL); 3545 3546 if (unlikely(!map)) { 3547 fwd = dev_get_by_index_rcu(dev_net(dev), index); 3548 if (unlikely(!fwd)) { 3549 err = -EINVAL; 3550 goto err; 3551 } 3552 3553 err = dev_xdp_enqueue(fwd, xdp, dev); 3554 } else { 3555 err = __bpf_tx_xdp_map(dev, fwd, map, xdp); 3556 } 3557 3558 if (unlikely(err)) 3559 goto err; 3560 3561 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index); 3562 return 0; 3563 err: 3564 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err); 3565 return err; 3566 } 3567 EXPORT_SYMBOL_GPL(xdp_do_redirect); 3568 3569 static int xdp_do_generic_redirect_map(struct net_device *dev, 3570 struct sk_buff *skb, 3571 struct xdp_buff *xdp, 3572 struct bpf_prog *xdp_prog, 3573 struct bpf_map *map) 3574 { 3575 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3576 u32 index = ri->tgt_index; 3577 void *fwd = ri->tgt_value; 3578 int err = 0; 3579 3580 ri->tgt_index = 0; 3581 ri->tgt_value = NULL; 3582 WRITE_ONCE(ri->map, NULL); 3583 3584 if (map->map_type == BPF_MAP_TYPE_DEVMAP || 3585 map->map_type == BPF_MAP_TYPE_DEVMAP_HASH) { 3586 struct bpf_dtab_netdev *dst = fwd; 3587 3588 err = dev_map_generic_redirect(dst, skb, xdp_prog); 3589 if (unlikely(err)) 3590 goto err; 3591 } else if (map->map_type == BPF_MAP_TYPE_XSKMAP) { 3592 struct xdp_sock *xs = fwd; 3593 3594 err = xsk_generic_rcv(xs, xdp); 3595 if (err) 3596 goto err; 3597 consume_skb(skb); 3598 } else { 3599 /* TODO: Handle BPF_MAP_TYPE_CPUMAP */ 3600 err = -EBADRQC; 3601 goto err; 3602 } 3603 3604 _trace_xdp_redirect_map(dev, xdp_prog, fwd, map, index); 3605 return 0; 3606 err: 3607 _trace_xdp_redirect_map_err(dev, xdp_prog, fwd, map, index, err); 3608 return err; 3609 } 3610 3611 int xdp_do_generic_redirect(struct net_device *dev, struct sk_buff *skb, 3612 struct xdp_buff *xdp, struct bpf_prog *xdp_prog) 3613 { 3614 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3615 struct bpf_map *map = READ_ONCE(ri->map); 3616 u32 index = ri->tgt_index; 3617 struct net_device *fwd; 3618 int err = 0; 3619 3620 if (map) 3621 return xdp_do_generic_redirect_map(dev, skb, xdp, xdp_prog, 3622 map); 3623 ri->tgt_index = 0; 3624 fwd = dev_get_by_index_rcu(dev_net(dev), index); 3625 if (unlikely(!fwd)) { 3626 err = -EINVAL; 3627 goto err; 3628 } 3629 3630 err = xdp_ok_fwd_dev(fwd, skb->len); 3631 if (unlikely(err)) 3632 goto err; 3633 3634 skb->dev = fwd; 3635 _trace_xdp_redirect(dev, xdp_prog, index); 3636 generic_xdp_tx(skb, xdp_prog); 3637 return 0; 3638 err: 3639 _trace_xdp_redirect_err(dev, xdp_prog, index, err); 3640 return err; 3641 } 3642 3643 BPF_CALL_2(bpf_xdp_redirect, u32, ifindex, u64, flags) 3644 { 3645 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3646 3647 if (unlikely(flags)) 3648 return XDP_ABORTED; 3649 3650 ri->flags = flags; 3651 ri->tgt_index = ifindex; 3652 ri->tgt_value = NULL; 3653 WRITE_ONCE(ri->map, NULL); 3654 3655 return XDP_REDIRECT; 3656 } 3657 3658 static const struct bpf_func_proto bpf_xdp_redirect_proto = { 3659 .func = bpf_xdp_redirect, 3660 .gpl_only = false, 3661 .ret_type = RET_INTEGER, 3662 .arg1_type = ARG_ANYTHING, 3663 .arg2_type = ARG_ANYTHING, 3664 }; 3665 3666 BPF_CALL_3(bpf_xdp_redirect_map, struct bpf_map *, map, u32, ifindex, 3667 u64, flags) 3668 { 3669 struct bpf_redirect_info *ri = this_cpu_ptr(&bpf_redirect_info); 3670 3671 /* Lower bits of the flags are used as return code on lookup failure */ 3672 if (unlikely(flags > XDP_TX)) 3673 return XDP_ABORTED; 3674 3675 ri->tgt_value = __xdp_map_lookup_elem(map, ifindex); 3676 if (unlikely(!ri->tgt_value)) { 3677 /* If the lookup fails we want to clear out the state in the 3678 * redirect_info struct completely, so that if an eBPF program 3679 * performs multiple lookups, the last one always takes 3680 * precedence. 3681 */ 3682 WRITE_ONCE(ri->map, NULL); 3683 return flags; 3684 } 3685 3686 ri->flags = flags; 3687 ri->tgt_index = ifindex; 3688 WRITE_ONCE(ri->map, map); 3689 3690 return XDP_REDIRECT; 3691 } 3692 3693 static const struct bpf_func_proto bpf_xdp_redirect_map_proto = { 3694 .func = bpf_xdp_redirect_map, 3695 .gpl_only = false, 3696 .ret_type = RET_INTEGER, 3697 .arg1_type = ARG_CONST_MAP_PTR, 3698 .arg2_type = ARG_ANYTHING, 3699 .arg3_type = ARG_ANYTHING, 3700 }; 3701 3702 static unsigned long bpf_skb_copy(void *dst_buff, const void *skb, 3703 unsigned long off, unsigned long len) 3704 { 3705 void *ptr = skb_header_pointer(skb, off, len, dst_buff); 3706 3707 if (unlikely(!ptr)) 3708 return len; 3709 if (ptr != dst_buff) 3710 memcpy(dst_buff, ptr, len); 3711 3712 return 0; 3713 } 3714 3715 BPF_CALL_5(bpf_skb_event_output, struct sk_buff *, skb, struct bpf_map *, map, 3716 u64, flags, void *, meta, u64, meta_size) 3717 { 3718 u64 skb_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 3719 3720 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 3721 return -EINVAL; 3722 if (unlikely(!skb || skb_size > skb->len)) 3723 return -EFAULT; 3724 3725 return bpf_event_output(map, flags, meta, meta_size, skb, skb_size, 3726 bpf_skb_copy); 3727 } 3728 3729 static const struct bpf_func_proto bpf_skb_event_output_proto = { 3730 .func = bpf_skb_event_output, 3731 .gpl_only = true, 3732 .ret_type = RET_INTEGER, 3733 .arg1_type = ARG_PTR_TO_CTX, 3734 .arg2_type = ARG_CONST_MAP_PTR, 3735 .arg3_type = ARG_ANYTHING, 3736 .arg4_type = ARG_PTR_TO_MEM, 3737 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 3738 }; 3739 3740 static int bpf_skb_output_btf_ids[5]; 3741 const struct bpf_func_proto bpf_skb_output_proto = { 3742 .func = bpf_skb_event_output, 3743 .gpl_only = true, 3744 .ret_type = RET_INTEGER, 3745 .arg1_type = ARG_PTR_TO_BTF_ID, 3746 .arg2_type = ARG_CONST_MAP_PTR, 3747 .arg3_type = ARG_ANYTHING, 3748 .arg4_type = ARG_PTR_TO_MEM, 3749 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 3750 .btf_id = bpf_skb_output_btf_ids, 3751 }; 3752 3753 static unsigned short bpf_tunnel_key_af(u64 flags) 3754 { 3755 return flags & BPF_F_TUNINFO_IPV6 ? AF_INET6 : AF_INET; 3756 } 3757 3758 BPF_CALL_4(bpf_skb_get_tunnel_key, struct sk_buff *, skb, struct bpf_tunnel_key *, to, 3759 u32, size, u64, flags) 3760 { 3761 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 3762 u8 compat[sizeof(struct bpf_tunnel_key)]; 3763 void *to_orig = to; 3764 int err; 3765 3766 if (unlikely(!info || (flags & ~(BPF_F_TUNINFO_IPV6)))) { 3767 err = -EINVAL; 3768 goto err_clear; 3769 } 3770 if (ip_tunnel_info_af(info) != bpf_tunnel_key_af(flags)) { 3771 err = -EPROTO; 3772 goto err_clear; 3773 } 3774 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 3775 err = -EINVAL; 3776 switch (size) { 3777 case offsetof(struct bpf_tunnel_key, tunnel_label): 3778 case offsetof(struct bpf_tunnel_key, tunnel_ext): 3779 goto set_compat; 3780 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 3781 /* Fixup deprecated structure layouts here, so we have 3782 * a common path later on. 3783 */ 3784 if (ip_tunnel_info_af(info) != AF_INET) 3785 goto err_clear; 3786 set_compat: 3787 to = (struct bpf_tunnel_key *)compat; 3788 break; 3789 default: 3790 goto err_clear; 3791 } 3792 } 3793 3794 to->tunnel_id = be64_to_cpu(info->key.tun_id); 3795 to->tunnel_tos = info->key.tos; 3796 to->tunnel_ttl = info->key.ttl; 3797 to->tunnel_ext = 0; 3798 3799 if (flags & BPF_F_TUNINFO_IPV6) { 3800 memcpy(to->remote_ipv6, &info->key.u.ipv6.src, 3801 sizeof(to->remote_ipv6)); 3802 to->tunnel_label = be32_to_cpu(info->key.label); 3803 } else { 3804 to->remote_ipv4 = be32_to_cpu(info->key.u.ipv4.src); 3805 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 3806 to->tunnel_label = 0; 3807 } 3808 3809 if (unlikely(size != sizeof(struct bpf_tunnel_key))) 3810 memcpy(to_orig, to, size); 3811 3812 return 0; 3813 err_clear: 3814 memset(to_orig, 0, size); 3815 return err; 3816 } 3817 3818 static const struct bpf_func_proto bpf_skb_get_tunnel_key_proto = { 3819 .func = bpf_skb_get_tunnel_key, 3820 .gpl_only = false, 3821 .ret_type = RET_INTEGER, 3822 .arg1_type = ARG_PTR_TO_CTX, 3823 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 3824 .arg3_type = ARG_CONST_SIZE, 3825 .arg4_type = ARG_ANYTHING, 3826 }; 3827 3828 BPF_CALL_3(bpf_skb_get_tunnel_opt, struct sk_buff *, skb, u8 *, to, u32, size) 3829 { 3830 const struct ip_tunnel_info *info = skb_tunnel_info(skb); 3831 int err; 3832 3833 if (unlikely(!info || 3834 !(info->key.tun_flags & TUNNEL_OPTIONS_PRESENT))) { 3835 err = -ENOENT; 3836 goto err_clear; 3837 } 3838 if (unlikely(size < info->options_len)) { 3839 err = -ENOMEM; 3840 goto err_clear; 3841 } 3842 3843 ip_tunnel_info_opts_get(to, info); 3844 if (size > info->options_len) 3845 memset(to + info->options_len, 0, size - info->options_len); 3846 3847 return info->options_len; 3848 err_clear: 3849 memset(to, 0, size); 3850 return err; 3851 } 3852 3853 static const struct bpf_func_proto bpf_skb_get_tunnel_opt_proto = { 3854 .func = bpf_skb_get_tunnel_opt, 3855 .gpl_only = false, 3856 .ret_type = RET_INTEGER, 3857 .arg1_type = ARG_PTR_TO_CTX, 3858 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 3859 .arg3_type = ARG_CONST_SIZE, 3860 }; 3861 3862 static struct metadata_dst __percpu *md_dst; 3863 3864 BPF_CALL_4(bpf_skb_set_tunnel_key, struct sk_buff *, skb, 3865 const struct bpf_tunnel_key *, from, u32, size, u64, flags) 3866 { 3867 struct metadata_dst *md = this_cpu_ptr(md_dst); 3868 u8 compat[sizeof(struct bpf_tunnel_key)]; 3869 struct ip_tunnel_info *info; 3870 3871 if (unlikely(flags & ~(BPF_F_TUNINFO_IPV6 | BPF_F_ZERO_CSUM_TX | 3872 BPF_F_DONT_FRAGMENT | BPF_F_SEQ_NUMBER))) 3873 return -EINVAL; 3874 if (unlikely(size != sizeof(struct bpf_tunnel_key))) { 3875 switch (size) { 3876 case offsetof(struct bpf_tunnel_key, tunnel_label): 3877 case offsetof(struct bpf_tunnel_key, tunnel_ext): 3878 case offsetof(struct bpf_tunnel_key, remote_ipv6[1]): 3879 /* Fixup deprecated structure layouts here, so we have 3880 * a common path later on. 3881 */ 3882 memcpy(compat, from, size); 3883 memset(compat + size, 0, sizeof(compat) - size); 3884 from = (const struct bpf_tunnel_key *) compat; 3885 break; 3886 default: 3887 return -EINVAL; 3888 } 3889 } 3890 if (unlikely((!(flags & BPF_F_TUNINFO_IPV6) && from->tunnel_label) || 3891 from->tunnel_ext)) 3892 return -EINVAL; 3893 3894 skb_dst_drop(skb); 3895 dst_hold((struct dst_entry *) md); 3896 skb_dst_set(skb, (struct dst_entry *) md); 3897 3898 info = &md->u.tun_info; 3899 memset(info, 0, sizeof(*info)); 3900 info->mode = IP_TUNNEL_INFO_TX; 3901 3902 info->key.tun_flags = TUNNEL_KEY | TUNNEL_CSUM | TUNNEL_NOCACHE; 3903 if (flags & BPF_F_DONT_FRAGMENT) 3904 info->key.tun_flags |= TUNNEL_DONT_FRAGMENT; 3905 if (flags & BPF_F_ZERO_CSUM_TX) 3906 info->key.tun_flags &= ~TUNNEL_CSUM; 3907 if (flags & BPF_F_SEQ_NUMBER) 3908 info->key.tun_flags |= TUNNEL_SEQ; 3909 3910 info->key.tun_id = cpu_to_be64(from->tunnel_id); 3911 info->key.tos = from->tunnel_tos; 3912 info->key.ttl = from->tunnel_ttl; 3913 3914 if (flags & BPF_F_TUNINFO_IPV6) { 3915 info->mode |= IP_TUNNEL_INFO_IPV6; 3916 memcpy(&info->key.u.ipv6.dst, from->remote_ipv6, 3917 sizeof(from->remote_ipv6)); 3918 info->key.label = cpu_to_be32(from->tunnel_label) & 3919 IPV6_FLOWLABEL_MASK; 3920 } else { 3921 info->key.u.ipv4.dst = cpu_to_be32(from->remote_ipv4); 3922 } 3923 3924 return 0; 3925 } 3926 3927 static const struct bpf_func_proto bpf_skb_set_tunnel_key_proto = { 3928 .func = bpf_skb_set_tunnel_key, 3929 .gpl_only = false, 3930 .ret_type = RET_INTEGER, 3931 .arg1_type = ARG_PTR_TO_CTX, 3932 .arg2_type = ARG_PTR_TO_MEM, 3933 .arg3_type = ARG_CONST_SIZE, 3934 .arg4_type = ARG_ANYTHING, 3935 }; 3936 3937 BPF_CALL_3(bpf_skb_set_tunnel_opt, struct sk_buff *, skb, 3938 const u8 *, from, u32, size) 3939 { 3940 struct ip_tunnel_info *info = skb_tunnel_info(skb); 3941 const struct metadata_dst *md = this_cpu_ptr(md_dst); 3942 3943 if (unlikely(info != &md->u.tun_info || (size & (sizeof(u32) - 1)))) 3944 return -EINVAL; 3945 if (unlikely(size > IP_TUNNEL_OPTS_MAX)) 3946 return -ENOMEM; 3947 3948 ip_tunnel_info_opts_set(info, from, size, TUNNEL_OPTIONS_PRESENT); 3949 3950 return 0; 3951 } 3952 3953 static const struct bpf_func_proto bpf_skb_set_tunnel_opt_proto = { 3954 .func = bpf_skb_set_tunnel_opt, 3955 .gpl_only = false, 3956 .ret_type = RET_INTEGER, 3957 .arg1_type = ARG_PTR_TO_CTX, 3958 .arg2_type = ARG_PTR_TO_MEM, 3959 .arg3_type = ARG_CONST_SIZE, 3960 }; 3961 3962 static const struct bpf_func_proto * 3963 bpf_get_skb_set_tunnel_proto(enum bpf_func_id which) 3964 { 3965 if (!md_dst) { 3966 struct metadata_dst __percpu *tmp; 3967 3968 tmp = metadata_dst_alloc_percpu(IP_TUNNEL_OPTS_MAX, 3969 METADATA_IP_TUNNEL, 3970 GFP_KERNEL); 3971 if (!tmp) 3972 return NULL; 3973 if (cmpxchg(&md_dst, NULL, tmp)) 3974 metadata_dst_free_percpu(tmp); 3975 } 3976 3977 switch (which) { 3978 case BPF_FUNC_skb_set_tunnel_key: 3979 return &bpf_skb_set_tunnel_key_proto; 3980 case BPF_FUNC_skb_set_tunnel_opt: 3981 return &bpf_skb_set_tunnel_opt_proto; 3982 default: 3983 return NULL; 3984 } 3985 } 3986 3987 BPF_CALL_3(bpf_skb_under_cgroup, struct sk_buff *, skb, struct bpf_map *, map, 3988 u32, idx) 3989 { 3990 struct bpf_array *array = container_of(map, struct bpf_array, map); 3991 struct cgroup *cgrp; 3992 struct sock *sk; 3993 3994 sk = skb_to_full_sk(skb); 3995 if (!sk || !sk_fullsock(sk)) 3996 return -ENOENT; 3997 if (unlikely(idx >= array->map.max_entries)) 3998 return -E2BIG; 3999 4000 cgrp = READ_ONCE(array->ptrs[idx]); 4001 if (unlikely(!cgrp)) 4002 return -EAGAIN; 4003 4004 return sk_under_cgroup_hierarchy(sk, cgrp); 4005 } 4006 4007 static const struct bpf_func_proto bpf_skb_under_cgroup_proto = { 4008 .func = bpf_skb_under_cgroup, 4009 .gpl_only = false, 4010 .ret_type = RET_INTEGER, 4011 .arg1_type = ARG_PTR_TO_CTX, 4012 .arg2_type = ARG_CONST_MAP_PTR, 4013 .arg3_type = ARG_ANYTHING, 4014 }; 4015 4016 #ifdef CONFIG_SOCK_CGROUP_DATA 4017 BPF_CALL_1(bpf_skb_cgroup_id, const struct sk_buff *, skb) 4018 { 4019 struct sock *sk = skb_to_full_sk(skb); 4020 struct cgroup *cgrp; 4021 4022 if (!sk || !sk_fullsock(sk)) 4023 return 0; 4024 4025 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4026 return cgroup_id(cgrp); 4027 } 4028 4029 static const struct bpf_func_proto bpf_skb_cgroup_id_proto = { 4030 .func = bpf_skb_cgroup_id, 4031 .gpl_only = false, 4032 .ret_type = RET_INTEGER, 4033 .arg1_type = ARG_PTR_TO_CTX, 4034 }; 4035 4036 BPF_CALL_2(bpf_skb_ancestor_cgroup_id, const struct sk_buff *, skb, int, 4037 ancestor_level) 4038 { 4039 struct sock *sk = skb_to_full_sk(skb); 4040 struct cgroup *ancestor; 4041 struct cgroup *cgrp; 4042 4043 if (!sk || !sk_fullsock(sk)) 4044 return 0; 4045 4046 cgrp = sock_cgroup_ptr(&sk->sk_cgrp_data); 4047 ancestor = cgroup_ancestor(cgrp, ancestor_level); 4048 if (!ancestor) 4049 return 0; 4050 4051 return cgroup_id(ancestor); 4052 } 4053 4054 static const struct bpf_func_proto bpf_skb_ancestor_cgroup_id_proto = { 4055 .func = bpf_skb_ancestor_cgroup_id, 4056 .gpl_only = false, 4057 .ret_type = RET_INTEGER, 4058 .arg1_type = ARG_PTR_TO_CTX, 4059 .arg2_type = ARG_ANYTHING, 4060 }; 4061 #endif 4062 4063 static unsigned long bpf_xdp_copy(void *dst_buff, const void *src_buff, 4064 unsigned long off, unsigned long len) 4065 { 4066 memcpy(dst_buff, src_buff + off, len); 4067 return 0; 4068 } 4069 4070 BPF_CALL_5(bpf_xdp_event_output, struct xdp_buff *, xdp, struct bpf_map *, map, 4071 u64, flags, void *, meta, u64, meta_size) 4072 { 4073 u64 xdp_size = (flags & BPF_F_CTXLEN_MASK) >> 32; 4074 4075 if (unlikely(flags & ~(BPF_F_CTXLEN_MASK | BPF_F_INDEX_MASK))) 4076 return -EINVAL; 4077 if (unlikely(!xdp || 4078 xdp_size > (unsigned long)(xdp->data_end - xdp->data))) 4079 return -EFAULT; 4080 4081 return bpf_event_output(map, flags, meta, meta_size, xdp->data, 4082 xdp_size, bpf_xdp_copy); 4083 } 4084 4085 static const struct bpf_func_proto bpf_xdp_event_output_proto = { 4086 .func = bpf_xdp_event_output, 4087 .gpl_only = true, 4088 .ret_type = RET_INTEGER, 4089 .arg1_type = ARG_PTR_TO_CTX, 4090 .arg2_type = ARG_CONST_MAP_PTR, 4091 .arg3_type = ARG_ANYTHING, 4092 .arg4_type = ARG_PTR_TO_MEM, 4093 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4094 }; 4095 4096 static int bpf_xdp_output_btf_ids[5]; 4097 const struct bpf_func_proto bpf_xdp_output_proto = { 4098 .func = bpf_xdp_event_output, 4099 .gpl_only = true, 4100 .ret_type = RET_INTEGER, 4101 .arg1_type = ARG_PTR_TO_BTF_ID, 4102 .arg2_type = ARG_CONST_MAP_PTR, 4103 .arg3_type = ARG_ANYTHING, 4104 .arg4_type = ARG_PTR_TO_MEM, 4105 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4106 .btf_id = bpf_xdp_output_btf_ids, 4107 }; 4108 4109 BPF_CALL_1(bpf_get_socket_cookie, struct sk_buff *, skb) 4110 { 4111 return skb->sk ? sock_gen_cookie(skb->sk) : 0; 4112 } 4113 4114 static const struct bpf_func_proto bpf_get_socket_cookie_proto = { 4115 .func = bpf_get_socket_cookie, 4116 .gpl_only = false, 4117 .ret_type = RET_INTEGER, 4118 .arg1_type = ARG_PTR_TO_CTX, 4119 }; 4120 4121 BPF_CALL_1(bpf_get_socket_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4122 { 4123 return sock_gen_cookie(ctx->sk); 4124 } 4125 4126 static const struct bpf_func_proto bpf_get_socket_cookie_sock_addr_proto = { 4127 .func = bpf_get_socket_cookie_sock_addr, 4128 .gpl_only = false, 4129 .ret_type = RET_INTEGER, 4130 .arg1_type = ARG_PTR_TO_CTX, 4131 }; 4132 4133 BPF_CALL_1(bpf_get_socket_cookie_sock, struct sock *, ctx) 4134 { 4135 return sock_gen_cookie(ctx); 4136 } 4137 4138 static const struct bpf_func_proto bpf_get_socket_cookie_sock_proto = { 4139 .func = bpf_get_socket_cookie_sock, 4140 .gpl_only = false, 4141 .ret_type = RET_INTEGER, 4142 .arg1_type = ARG_PTR_TO_CTX, 4143 }; 4144 4145 BPF_CALL_1(bpf_get_socket_cookie_sock_ops, struct bpf_sock_ops_kern *, ctx) 4146 { 4147 return sock_gen_cookie(ctx->sk); 4148 } 4149 4150 static const struct bpf_func_proto bpf_get_socket_cookie_sock_ops_proto = { 4151 .func = bpf_get_socket_cookie_sock_ops, 4152 .gpl_only = false, 4153 .ret_type = RET_INTEGER, 4154 .arg1_type = ARG_PTR_TO_CTX, 4155 }; 4156 4157 static u64 __bpf_get_netns_cookie(struct sock *sk) 4158 { 4159 #ifdef CONFIG_NET_NS 4160 return net_gen_cookie(sk ? sk->sk_net.net : &init_net); 4161 #else 4162 return 0; 4163 #endif 4164 } 4165 4166 BPF_CALL_1(bpf_get_netns_cookie_sock, struct sock *, ctx) 4167 { 4168 return __bpf_get_netns_cookie(ctx); 4169 } 4170 4171 static const struct bpf_func_proto bpf_get_netns_cookie_sock_proto = { 4172 .func = bpf_get_netns_cookie_sock, 4173 .gpl_only = false, 4174 .ret_type = RET_INTEGER, 4175 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4176 }; 4177 4178 BPF_CALL_1(bpf_get_netns_cookie_sock_addr, struct bpf_sock_addr_kern *, ctx) 4179 { 4180 return __bpf_get_netns_cookie(ctx ? ctx->sk : NULL); 4181 } 4182 4183 static const struct bpf_func_proto bpf_get_netns_cookie_sock_addr_proto = { 4184 .func = bpf_get_netns_cookie_sock_addr, 4185 .gpl_only = false, 4186 .ret_type = RET_INTEGER, 4187 .arg1_type = ARG_PTR_TO_CTX_OR_NULL, 4188 }; 4189 4190 BPF_CALL_1(bpf_get_socket_uid, struct sk_buff *, skb) 4191 { 4192 struct sock *sk = sk_to_full_sk(skb->sk); 4193 kuid_t kuid; 4194 4195 if (!sk || !sk_fullsock(sk)) 4196 return overflowuid; 4197 kuid = sock_net_uid(sock_net(sk), sk); 4198 return from_kuid_munged(sock_net(sk)->user_ns, kuid); 4199 } 4200 4201 static const struct bpf_func_proto bpf_get_socket_uid_proto = { 4202 .func = bpf_get_socket_uid, 4203 .gpl_only = false, 4204 .ret_type = RET_INTEGER, 4205 .arg1_type = ARG_PTR_TO_CTX, 4206 }; 4207 4208 BPF_CALL_5(bpf_event_output_data, void *, ctx, struct bpf_map *, map, u64, flags, 4209 void *, data, u64, size) 4210 { 4211 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) 4212 return -EINVAL; 4213 4214 return bpf_event_output(map, flags, data, size, NULL, 0, NULL); 4215 } 4216 4217 static const struct bpf_func_proto bpf_event_output_data_proto = { 4218 .func = bpf_event_output_data, 4219 .gpl_only = true, 4220 .ret_type = RET_INTEGER, 4221 .arg1_type = ARG_PTR_TO_CTX, 4222 .arg2_type = ARG_CONST_MAP_PTR, 4223 .arg3_type = ARG_ANYTHING, 4224 .arg4_type = ARG_PTR_TO_MEM, 4225 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 4226 }; 4227 4228 BPF_CALL_5(bpf_setsockopt, struct bpf_sock_ops_kern *, bpf_sock, 4229 int, level, int, optname, char *, optval, int, optlen) 4230 { 4231 struct sock *sk = bpf_sock->sk; 4232 int ret = 0; 4233 int val; 4234 4235 if (!sk_fullsock(sk)) 4236 return -EINVAL; 4237 4238 if (level == SOL_SOCKET) { 4239 if (optlen != sizeof(int)) 4240 return -EINVAL; 4241 val = *((int *)optval); 4242 4243 /* Only some socketops are supported */ 4244 switch (optname) { 4245 case SO_RCVBUF: 4246 val = min_t(u32, val, sysctl_rmem_max); 4247 sk->sk_userlocks |= SOCK_RCVBUF_LOCK; 4248 WRITE_ONCE(sk->sk_rcvbuf, 4249 max_t(int, val * 2, SOCK_MIN_RCVBUF)); 4250 break; 4251 case SO_SNDBUF: 4252 val = min_t(u32, val, sysctl_wmem_max); 4253 sk->sk_userlocks |= SOCK_SNDBUF_LOCK; 4254 WRITE_ONCE(sk->sk_sndbuf, 4255 max_t(int, val * 2, SOCK_MIN_SNDBUF)); 4256 break; 4257 case SO_MAX_PACING_RATE: /* 32bit version */ 4258 if (val != ~0U) 4259 cmpxchg(&sk->sk_pacing_status, 4260 SK_PACING_NONE, 4261 SK_PACING_NEEDED); 4262 sk->sk_max_pacing_rate = (val == ~0U) ? ~0UL : val; 4263 sk->sk_pacing_rate = min(sk->sk_pacing_rate, 4264 sk->sk_max_pacing_rate); 4265 break; 4266 case SO_PRIORITY: 4267 sk->sk_priority = val; 4268 break; 4269 case SO_RCVLOWAT: 4270 if (val < 0) 4271 val = INT_MAX; 4272 WRITE_ONCE(sk->sk_rcvlowat, val ? : 1); 4273 break; 4274 case SO_MARK: 4275 if (sk->sk_mark != val) { 4276 sk->sk_mark = val; 4277 sk_dst_reset(sk); 4278 } 4279 break; 4280 default: 4281 ret = -EINVAL; 4282 } 4283 #ifdef CONFIG_INET 4284 } else if (level == SOL_IP) { 4285 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 4286 return -EINVAL; 4287 4288 val = *((int *)optval); 4289 /* Only some options are supported */ 4290 switch (optname) { 4291 case IP_TOS: 4292 if (val < -1 || val > 0xff) { 4293 ret = -EINVAL; 4294 } else { 4295 struct inet_sock *inet = inet_sk(sk); 4296 4297 if (val == -1) 4298 val = 0; 4299 inet->tos = val; 4300 } 4301 break; 4302 default: 4303 ret = -EINVAL; 4304 } 4305 #if IS_ENABLED(CONFIG_IPV6) 4306 } else if (level == SOL_IPV6) { 4307 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 4308 return -EINVAL; 4309 4310 val = *((int *)optval); 4311 /* Only some options are supported */ 4312 switch (optname) { 4313 case IPV6_TCLASS: 4314 if (val < -1 || val > 0xff) { 4315 ret = -EINVAL; 4316 } else { 4317 struct ipv6_pinfo *np = inet6_sk(sk); 4318 4319 if (val == -1) 4320 val = 0; 4321 np->tclass = val; 4322 } 4323 break; 4324 default: 4325 ret = -EINVAL; 4326 } 4327 #endif 4328 } else if (level == SOL_TCP && 4329 sk->sk_prot->setsockopt == tcp_setsockopt) { 4330 if (optname == TCP_CONGESTION) { 4331 char name[TCP_CA_NAME_MAX]; 4332 bool reinit = bpf_sock->op > BPF_SOCK_OPS_NEEDS_ECN; 4333 4334 strncpy(name, optval, min_t(long, optlen, 4335 TCP_CA_NAME_MAX-1)); 4336 name[TCP_CA_NAME_MAX-1] = 0; 4337 ret = tcp_set_congestion_control(sk, name, false, 4338 reinit, true); 4339 } else { 4340 struct tcp_sock *tp = tcp_sk(sk); 4341 4342 if (optlen != sizeof(int)) 4343 return -EINVAL; 4344 4345 val = *((int *)optval); 4346 /* Only some options are supported */ 4347 switch (optname) { 4348 case TCP_BPF_IW: 4349 if (val <= 0 || tp->data_segs_out > tp->syn_data) 4350 ret = -EINVAL; 4351 else 4352 tp->snd_cwnd = val; 4353 break; 4354 case TCP_BPF_SNDCWND_CLAMP: 4355 if (val <= 0) { 4356 ret = -EINVAL; 4357 } else { 4358 tp->snd_cwnd_clamp = val; 4359 tp->snd_ssthresh = val; 4360 } 4361 break; 4362 case TCP_SAVE_SYN: 4363 if (val < 0 || val > 1) 4364 ret = -EINVAL; 4365 else 4366 tp->save_syn = val; 4367 break; 4368 default: 4369 ret = -EINVAL; 4370 } 4371 } 4372 #endif 4373 } else { 4374 ret = -EINVAL; 4375 } 4376 return ret; 4377 } 4378 4379 static const struct bpf_func_proto bpf_setsockopt_proto = { 4380 .func = bpf_setsockopt, 4381 .gpl_only = false, 4382 .ret_type = RET_INTEGER, 4383 .arg1_type = ARG_PTR_TO_CTX, 4384 .arg2_type = ARG_ANYTHING, 4385 .arg3_type = ARG_ANYTHING, 4386 .arg4_type = ARG_PTR_TO_MEM, 4387 .arg5_type = ARG_CONST_SIZE, 4388 }; 4389 4390 BPF_CALL_5(bpf_getsockopt, struct bpf_sock_ops_kern *, bpf_sock, 4391 int, level, int, optname, char *, optval, int, optlen) 4392 { 4393 struct sock *sk = bpf_sock->sk; 4394 4395 if (!sk_fullsock(sk)) 4396 goto err_clear; 4397 #ifdef CONFIG_INET 4398 if (level == SOL_TCP && sk->sk_prot->getsockopt == tcp_getsockopt) { 4399 struct inet_connection_sock *icsk; 4400 struct tcp_sock *tp; 4401 4402 switch (optname) { 4403 case TCP_CONGESTION: 4404 icsk = inet_csk(sk); 4405 4406 if (!icsk->icsk_ca_ops || optlen <= 1) 4407 goto err_clear; 4408 strncpy(optval, icsk->icsk_ca_ops->name, optlen); 4409 optval[optlen - 1] = 0; 4410 break; 4411 case TCP_SAVED_SYN: 4412 tp = tcp_sk(sk); 4413 4414 if (optlen <= 0 || !tp->saved_syn || 4415 optlen > tp->saved_syn[0]) 4416 goto err_clear; 4417 memcpy(optval, tp->saved_syn + 1, optlen); 4418 break; 4419 default: 4420 goto err_clear; 4421 } 4422 } else if (level == SOL_IP) { 4423 struct inet_sock *inet = inet_sk(sk); 4424 4425 if (optlen != sizeof(int) || sk->sk_family != AF_INET) 4426 goto err_clear; 4427 4428 /* Only some options are supported */ 4429 switch (optname) { 4430 case IP_TOS: 4431 *((int *)optval) = (int)inet->tos; 4432 break; 4433 default: 4434 goto err_clear; 4435 } 4436 #if IS_ENABLED(CONFIG_IPV6) 4437 } else if (level == SOL_IPV6) { 4438 struct ipv6_pinfo *np = inet6_sk(sk); 4439 4440 if (optlen != sizeof(int) || sk->sk_family != AF_INET6) 4441 goto err_clear; 4442 4443 /* Only some options are supported */ 4444 switch (optname) { 4445 case IPV6_TCLASS: 4446 *((int *)optval) = (int)np->tclass; 4447 break; 4448 default: 4449 goto err_clear; 4450 } 4451 #endif 4452 } else { 4453 goto err_clear; 4454 } 4455 return 0; 4456 #endif 4457 err_clear: 4458 memset(optval, 0, optlen); 4459 return -EINVAL; 4460 } 4461 4462 static const struct bpf_func_proto bpf_getsockopt_proto = { 4463 .func = bpf_getsockopt, 4464 .gpl_only = false, 4465 .ret_type = RET_INTEGER, 4466 .arg1_type = ARG_PTR_TO_CTX, 4467 .arg2_type = ARG_ANYTHING, 4468 .arg3_type = ARG_ANYTHING, 4469 .arg4_type = ARG_PTR_TO_UNINIT_MEM, 4470 .arg5_type = ARG_CONST_SIZE, 4471 }; 4472 4473 BPF_CALL_2(bpf_sock_ops_cb_flags_set, struct bpf_sock_ops_kern *, bpf_sock, 4474 int, argval) 4475 { 4476 struct sock *sk = bpf_sock->sk; 4477 int val = argval & BPF_SOCK_OPS_ALL_CB_FLAGS; 4478 4479 if (!IS_ENABLED(CONFIG_INET) || !sk_fullsock(sk)) 4480 return -EINVAL; 4481 4482 tcp_sk(sk)->bpf_sock_ops_cb_flags = val; 4483 4484 return argval & (~BPF_SOCK_OPS_ALL_CB_FLAGS); 4485 } 4486 4487 static const struct bpf_func_proto bpf_sock_ops_cb_flags_set_proto = { 4488 .func = bpf_sock_ops_cb_flags_set, 4489 .gpl_only = false, 4490 .ret_type = RET_INTEGER, 4491 .arg1_type = ARG_PTR_TO_CTX, 4492 .arg2_type = ARG_ANYTHING, 4493 }; 4494 4495 const struct ipv6_bpf_stub *ipv6_bpf_stub __read_mostly; 4496 EXPORT_SYMBOL_GPL(ipv6_bpf_stub); 4497 4498 BPF_CALL_3(bpf_bind, struct bpf_sock_addr_kern *, ctx, struct sockaddr *, addr, 4499 int, addr_len) 4500 { 4501 #ifdef CONFIG_INET 4502 struct sock *sk = ctx->sk; 4503 int err; 4504 4505 /* Binding to port can be expensive so it's prohibited in the helper. 4506 * Only binding to IP is supported. 4507 */ 4508 err = -EINVAL; 4509 if (addr_len < offsetofend(struct sockaddr, sa_family)) 4510 return err; 4511 if (addr->sa_family == AF_INET) { 4512 if (addr_len < sizeof(struct sockaddr_in)) 4513 return err; 4514 if (((struct sockaddr_in *)addr)->sin_port != htons(0)) 4515 return err; 4516 return __inet_bind(sk, addr, addr_len, true, false); 4517 #if IS_ENABLED(CONFIG_IPV6) 4518 } else if (addr->sa_family == AF_INET6) { 4519 if (addr_len < SIN6_LEN_RFC2133) 4520 return err; 4521 if (((struct sockaddr_in6 *)addr)->sin6_port != htons(0)) 4522 return err; 4523 /* ipv6_bpf_stub cannot be NULL, since it's called from 4524 * bpf_cgroup_inet6_connect hook and ipv6 is already loaded 4525 */ 4526 return ipv6_bpf_stub->inet6_bind(sk, addr, addr_len, true, false); 4527 #endif /* CONFIG_IPV6 */ 4528 } 4529 #endif /* CONFIG_INET */ 4530 4531 return -EAFNOSUPPORT; 4532 } 4533 4534 static const struct bpf_func_proto bpf_bind_proto = { 4535 .func = bpf_bind, 4536 .gpl_only = false, 4537 .ret_type = RET_INTEGER, 4538 .arg1_type = ARG_PTR_TO_CTX, 4539 .arg2_type = ARG_PTR_TO_MEM, 4540 .arg3_type = ARG_CONST_SIZE, 4541 }; 4542 4543 #ifdef CONFIG_XFRM 4544 BPF_CALL_5(bpf_skb_get_xfrm_state, struct sk_buff *, skb, u32, index, 4545 struct bpf_xfrm_state *, to, u32, size, u64, flags) 4546 { 4547 const struct sec_path *sp = skb_sec_path(skb); 4548 const struct xfrm_state *x; 4549 4550 if (!sp || unlikely(index >= sp->len || flags)) 4551 goto err_clear; 4552 4553 x = sp->xvec[index]; 4554 4555 if (unlikely(size != sizeof(struct bpf_xfrm_state))) 4556 goto err_clear; 4557 4558 to->reqid = x->props.reqid; 4559 to->spi = x->id.spi; 4560 to->family = x->props.family; 4561 to->ext = 0; 4562 4563 if (to->family == AF_INET6) { 4564 memcpy(to->remote_ipv6, x->props.saddr.a6, 4565 sizeof(to->remote_ipv6)); 4566 } else { 4567 to->remote_ipv4 = x->props.saddr.a4; 4568 memset(&to->remote_ipv6[1], 0, sizeof(__u32) * 3); 4569 } 4570 4571 return 0; 4572 err_clear: 4573 memset(to, 0, size); 4574 return -EINVAL; 4575 } 4576 4577 static const struct bpf_func_proto bpf_skb_get_xfrm_state_proto = { 4578 .func = bpf_skb_get_xfrm_state, 4579 .gpl_only = false, 4580 .ret_type = RET_INTEGER, 4581 .arg1_type = ARG_PTR_TO_CTX, 4582 .arg2_type = ARG_ANYTHING, 4583 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 4584 .arg4_type = ARG_CONST_SIZE, 4585 .arg5_type = ARG_ANYTHING, 4586 }; 4587 #endif 4588 4589 #if IS_ENABLED(CONFIG_INET) || IS_ENABLED(CONFIG_IPV6) 4590 static int bpf_fib_set_fwd_params(struct bpf_fib_lookup *params, 4591 const struct neighbour *neigh, 4592 const struct net_device *dev) 4593 { 4594 memcpy(params->dmac, neigh->ha, ETH_ALEN); 4595 memcpy(params->smac, dev->dev_addr, ETH_ALEN); 4596 params->h_vlan_TCI = 0; 4597 params->h_vlan_proto = 0; 4598 params->ifindex = dev->ifindex; 4599 4600 return 0; 4601 } 4602 #endif 4603 4604 #if IS_ENABLED(CONFIG_INET) 4605 static int bpf_ipv4_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 4606 u32 flags, bool check_mtu) 4607 { 4608 struct fib_nh_common *nhc; 4609 struct in_device *in_dev; 4610 struct neighbour *neigh; 4611 struct net_device *dev; 4612 struct fib_result res; 4613 struct flowi4 fl4; 4614 int err; 4615 u32 mtu; 4616 4617 dev = dev_get_by_index_rcu(net, params->ifindex); 4618 if (unlikely(!dev)) 4619 return -ENODEV; 4620 4621 /* verify forwarding is enabled on this interface */ 4622 in_dev = __in_dev_get_rcu(dev); 4623 if (unlikely(!in_dev || !IN_DEV_FORWARD(in_dev))) 4624 return BPF_FIB_LKUP_RET_FWD_DISABLED; 4625 4626 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 4627 fl4.flowi4_iif = 1; 4628 fl4.flowi4_oif = params->ifindex; 4629 } else { 4630 fl4.flowi4_iif = params->ifindex; 4631 fl4.flowi4_oif = 0; 4632 } 4633 fl4.flowi4_tos = params->tos & IPTOS_RT_MASK; 4634 fl4.flowi4_scope = RT_SCOPE_UNIVERSE; 4635 fl4.flowi4_flags = 0; 4636 4637 fl4.flowi4_proto = params->l4_protocol; 4638 fl4.daddr = params->ipv4_dst; 4639 fl4.saddr = params->ipv4_src; 4640 fl4.fl4_sport = params->sport; 4641 fl4.fl4_dport = params->dport; 4642 4643 if (flags & BPF_FIB_LOOKUP_DIRECT) { 4644 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 4645 struct fib_table *tb; 4646 4647 tb = fib_get_table(net, tbid); 4648 if (unlikely(!tb)) 4649 return BPF_FIB_LKUP_RET_NOT_FWDED; 4650 4651 err = fib_table_lookup(tb, &fl4, &res, FIB_LOOKUP_NOREF); 4652 } else { 4653 fl4.flowi4_mark = 0; 4654 fl4.flowi4_secid = 0; 4655 fl4.flowi4_tun_key.tun_id = 0; 4656 fl4.flowi4_uid = sock_net_uid(net, NULL); 4657 4658 err = fib_lookup(net, &fl4, &res, FIB_LOOKUP_NOREF); 4659 } 4660 4661 if (err) { 4662 /* map fib lookup errors to RTN_ type */ 4663 if (err == -EINVAL) 4664 return BPF_FIB_LKUP_RET_BLACKHOLE; 4665 if (err == -EHOSTUNREACH) 4666 return BPF_FIB_LKUP_RET_UNREACHABLE; 4667 if (err == -EACCES) 4668 return BPF_FIB_LKUP_RET_PROHIBIT; 4669 4670 return BPF_FIB_LKUP_RET_NOT_FWDED; 4671 } 4672 4673 if (res.type != RTN_UNICAST) 4674 return BPF_FIB_LKUP_RET_NOT_FWDED; 4675 4676 if (fib_info_num_path(res.fi) > 1) 4677 fib_select_path(net, &res, &fl4, NULL); 4678 4679 if (check_mtu) { 4680 mtu = ip_mtu_from_fib_result(&res, params->ipv4_dst); 4681 if (params->tot_len > mtu) 4682 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 4683 } 4684 4685 nhc = res.nhc; 4686 4687 /* do not handle lwt encaps right now */ 4688 if (nhc->nhc_lwtstate) 4689 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 4690 4691 dev = nhc->nhc_dev; 4692 4693 params->rt_metric = res.fi->fib_priority; 4694 4695 /* xdp and cls_bpf programs are run in RCU-bh so 4696 * rcu_read_lock_bh is not needed here 4697 */ 4698 if (likely(nhc->nhc_gw_family != AF_INET6)) { 4699 if (nhc->nhc_gw_family) 4700 params->ipv4_dst = nhc->nhc_gw.ipv4; 4701 4702 neigh = __ipv4_neigh_lookup_noref(dev, 4703 (__force u32)params->ipv4_dst); 4704 } else { 4705 struct in6_addr *dst = (struct in6_addr *)params->ipv6_dst; 4706 4707 params->family = AF_INET6; 4708 *dst = nhc->nhc_gw.ipv6; 4709 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 4710 } 4711 4712 if (!neigh) 4713 return BPF_FIB_LKUP_RET_NO_NEIGH; 4714 4715 return bpf_fib_set_fwd_params(params, neigh, dev); 4716 } 4717 #endif 4718 4719 #if IS_ENABLED(CONFIG_IPV6) 4720 static int bpf_ipv6_fib_lookup(struct net *net, struct bpf_fib_lookup *params, 4721 u32 flags, bool check_mtu) 4722 { 4723 struct in6_addr *src = (struct in6_addr *) params->ipv6_src; 4724 struct in6_addr *dst = (struct in6_addr *) params->ipv6_dst; 4725 struct fib6_result res = {}; 4726 struct neighbour *neigh; 4727 struct net_device *dev; 4728 struct inet6_dev *idev; 4729 struct flowi6 fl6; 4730 int strict = 0; 4731 int oif, err; 4732 u32 mtu; 4733 4734 /* link local addresses are never forwarded */ 4735 if (rt6_need_strict(dst) || rt6_need_strict(src)) 4736 return BPF_FIB_LKUP_RET_NOT_FWDED; 4737 4738 dev = dev_get_by_index_rcu(net, params->ifindex); 4739 if (unlikely(!dev)) 4740 return -ENODEV; 4741 4742 idev = __in6_dev_get_safely(dev); 4743 if (unlikely(!idev || !idev->cnf.forwarding)) 4744 return BPF_FIB_LKUP_RET_FWD_DISABLED; 4745 4746 if (flags & BPF_FIB_LOOKUP_OUTPUT) { 4747 fl6.flowi6_iif = 1; 4748 oif = fl6.flowi6_oif = params->ifindex; 4749 } else { 4750 oif = fl6.flowi6_iif = params->ifindex; 4751 fl6.flowi6_oif = 0; 4752 strict = RT6_LOOKUP_F_HAS_SADDR; 4753 } 4754 fl6.flowlabel = params->flowinfo; 4755 fl6.flowi6_scope = 0; 4756 fl6.flowi6_flags = 0; 4757 fl6.mp_hash = 0; 4758 4759 fl6.flowi6_proto = params->l4_protocol; 4760 fl6.daddr = *dst; 4761 fl6.saddr = *src; 4762 fl6.fl6_sport = params->sport; 4763 fl6.fl6_dport = params->dport; 4764 4765 if (flags & BPF_FIB_LOOKUP_DIRECT) { 4766 u32 tbid = l3mdev_fib_table_rcu(dev) ? : RT_TABLE_MAIN; 4767 struct fib6_table *tb; 4768 4769 tb = ipv6_stub->fib6_get_table(net, tbid); 4770 if (unlikely(!tb)) 4771 return BPF_FIB_LKUP_RET_NOT_FWDED; 4772 4773 err = ipv6_stub->fib6_table_lookup(net, tb, oif, &fl6, &res, 4774 strict); 4775 } else { 4776 fl6.flowi6_mark = 0; 4777 fl6.flowi6_secid = 0; 4778 fl6.flowi6_tun_key.tun_id = 0; 4779 fl6.flowi6_uid = sock_net_uid(net, NULL); 4780 4781 err = ipv6_stub->fib6_lookup(net, oif, &fl6, &res, strict); 4782 } 4783 4784 if (unlikely(err || IS_ERR_OR_NULL(res.f6i) || 4785 res.f6i == net->ipv6.fib6_null_entry)) 4786 return BPF_FIB_LKUP_RET_NOT_FWDED; 4787 4788 switch (res.fib6_type) { 4789 /* only unicast is forwarded */ 4790 case RTN_UNICAST: 4791 break; 4792 case RTN_BLACKHOLE: 4793 return BPF_FIB_LKUP_RET_BLACKHOLE; 4794 case RTN_UNREACHABLE: 4795 return BPF_FIB_LKUP_RET_UNREACHABLE; 4796 case RTN_PROHIBIT: 4797 return BPF_FIB_LKUP_RET_PROHIBIT; 4798 default: 4799 return BPF_FIB_LKUP_RET_NOT_FWDED; 4800 } 4801 4802 ipv6_stub->fib6_select_path(net, &res, &fl6, fl6.flowi6_oif, 4803 fl6.flowi6_oif != 0, NULL, strict); 4804 4805 if (check_mtu) { 4806 mtu = ipv6_stub->ip6_mtu_from_fib6(&res, dst, src); 4807 if (params->tot_len > mtu) 4808 return BPF_FIB_LKUP_RET_FRAG_NEEDED; 4809 } 4810 4811 if (res.nh->fib_nh_lws) 4812 return BPF_FIB_LKUP_RET_UNSUPP_LWT; 4813 4814 if (res.nh->fib_nh_gw_family) 4815 *dst = res.nh->fib_nh_gw6; 4816 4817 dev = res.nh->fib_nh_dev; 4818 params->rt_metric = res.f6i->fib6_metric; 4819 4820 /* xdp and cls_bpf programs are run in RCU-bh so rcu_read_lock_bh is 4821 * not needed here. 4822 */ 4823 neigh = __ipv6_neigh_lookup_noref_stub(dev, dst); 4824 if (!neigh) 4825 return BPF_FIB_LKUP_RET_NO_NEIGH; 4826 4827 return bpf_fib_set_fwd_params(params, neigh, dev); 4828 } 4829 #endif 4830 4831 BPF_CALL_4(bpf_xdp_fib_lookup, struct xdp_buff *, ctx, 4832 struct bpf_fib_lookup *, params, int, plen, u32, flags) 4833 { 4834 if (plen < sizeof(*params)) 4835 return -EINVAL; 4836 4837 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 4838 return -EINVAL; 4839 4840 switch (params->family) { 4841 #if IS_ENABLED(CONFIG_INET) 4842 case AF_INET: 4843 return bpf_ipv4_fib_lookup(dev_net(ctx->rxq->dev), params, 4844 flags, true); 4845 #endif 4846 #if IS_ENABLED(CONFIG_IPV6) 4847 case AF_INET6: 4848 return bpf_ipv6_fib_lookup(dev_net(ctx->rxq->dev), params, 4849 flags, true); 4850 #endif 4851 } 4852 return -EAFNOSUPPORT; 4853 } 4854 4855 static const struct bpf_func_proto bpf_xdp_fib_lookup_proto = { 4856 .func = bpf_xdp_fib_lookup, 4857 .gpl_only = true, 4858 .ret_type = RET_INTEGER, 4859 .arg1_type = ARG_PTR_TO_CTX, 4860 .arg2_type = ARG_PTR_TO_MEM, 4861 .arg3_type = ARG_CONST_SIZE, 4862 .arg4_type = ARG_ANYTHING, 4863 }; 4864 4865 BPF_CALL_4(bpf_skb_fib_lookup, struct sk_buff *, skb, 4866 struct bpf_fib_lookup *, params, int, plen, u32, flags) 4867 { 4868 struct net *net = dev_net(skb->dev); 4869 int rc = -EAFNOSUPPORT; 4870 4871 if (plen < sizeof(*params)) 4872 return -EINVAL; 4873 4874 if (flags & ~(BPF_FIB_LOOKUP_DIRECT | BPF_FIB_LOOKUP_OUTPUT)) 4875 return -EINVAL; 4876 4877 switch (params->family) { 4878 #if IS_ENABLED(CONFIG_INET) 4879 case AF_INET: 4880 rc = bpf_ipv4_fib_lookup(net, params, flags, false); 4881 break; 4882 #endif 4883 #if IS_ENABLED(CONFIG_IPV6) 4884 case AF_INET6: 4885 rc = bpf_ipv6_fib_lookup(net, params, flags, false); 4886 break; 4887 #endif 4888 } 4889 4890 if (!rc) { 4891 struct net_device *dev; 4892 4893 dev = dev_get_by_index_rcu(net, params->ifindex); 4894 if (!is_skb_forwardable(dev, skb)) 4895 rc = BPF_FIB_LKUP_RET_FRAG_NEEDED; 4896 } 4897 4898 return rc; 4899 } 4900 4901 static const struct bpf_func_proto bpf_skb_fib_lookup_proto = { 4902 .func = bpf_skb_fib_lookup, 4903 .gpl_only = true, 4904 .ret_type = RET_INTEGER, 4905 .arg1_type = ARG_PTR_TO_CTX, 4906 .arg2_type = ARG_PTR_TO_MEM, 4907 .arg3_type = ARG_CONST_SIZE, 4908 .arg4_type = ARG_ANYTHING, 4909 }; 4910 4911 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 4912 static int bpf_push_seg6_encap(struct sk_buff *skb, u32 type, void *hdr, u32 len) 4913 { 4914 int err; 4915 struct ipv6_sr_hdr *srh = (struct ipv6_sr_hdr *)hdr; 4916 4917 if (!seg6_validate_srh(srh, len)) 4918 return -EINVAL; 4919 4920 switch (type) { 4921 case BPF_LWT_ENCAP_SEG6_INLINE: 4922 if (skb->protocol != htons(ETH_P_IPV6)) 4923 return -EBADMSG; 4924 4925 err = seg6_do_srh_inline(skb, srh); 4926 break; 4927 case BPF_LWT_ENCAP_SEG6: 4928 skb_reset_inner_headers(skb); 4929 skb->encapsulation = 1; 4930 err = seg6_do_srh_encap(skb, srh, IPPROTO_IPV6); 4931 break; 4932 default: 4933 return -EINVAL; 4934 } 4935 4936 bpf_compute_data_pointers(skb); 4937 if (err) 4938 return err; 4939 4940 ipv6_hdr(skb)->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 4941 skb_set_transport_header(skb, sizeof(struct ipv6hdr)); 4942 4943 return seg6_lookup_nexthop(skb, NULL, 0); 4944 } 4945 #endif /* CONFIG_IPV6_SEG6_BPF */ 4946 4947 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 4948 static int bpf_push_ip_encap(struct sk_buff *skb, void *hdr, u32 len, 4949 bool ingress) 4950 { 4951 return bpf_lwt_push_ip_encap(skb, hdr, len, ingress); 4952 } 4953 #endif 4954 4955 BPF_CALL_4(bpf_lwt_in_push_encap, struct sk_buff *, skb, u32, type, void *, hdr, 4956 u32, len) 4957 { 4958 switch (type) { 4959 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 4960 case BPF_LWT_ENCAP_SEG6: 4961 case BPF_LWT_ENCAP_SEG6_INLINE: 4962 return bpf_push_seg6_encap(skb, type, hdr, len); 4963 #endif 4964 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 4965 case BPF_LWT_ENCAP_IP: 4966 return bpf_push_ip_encap(skb, hdr, len, true /* ingress */); 4967 #endif 4968 default: 4969 return -EINVAL; 4970 } 4971 } 4972 4973 BPF_CALL_4(bpf_lwt_xmit_push_encap, struct sk_buff *, skb, u32, type, 4974 void *, hdr, u32, len) 4975 { 4976 switch (type) { 4977 #if IS_ENABLED(CONFIG_LWTUNNEL_BPF) 4978 case BPF_LWT_ENCAP_IP: 4979 return bpf_push_ip_encap(skb, hdr, len, false /* egress */); 4980 #endif 4981 default: 4982 return -EINVAL; 4983 } 4984 } 4985 4986 static const struct bpf_func_proto bpf_lwt_in_push_encap_proto = { 4987 .func = bpf_lwt_in_push_encap, 4988 .gpl_only = false, 4989 .ret_type = RET_INTEGER, 4990 .arg1_type = ARG_PTR_TO_CTX, 4991 .arg2_type = ARG_ANYTHING, 4992 .arg3_type = ARG_PTR_TO_MEM, 4993 .arg4_type = ARG_CONST_SIZE 4994 }; 4995 4996 static const struct bpf_func_proto bpf_lwt_xmit_push_encap_proto = { 4997 .func = bpf_lwt_xmit_push_encap, 4998 .gpl_only = false, 4999 .ret_type = RET_INTEGER, 5000 .arg1_type = ARG_PTR_TO_CTX, 5001 .arg2_type = ARG_ANYTHING, 5002 .arg3_type = ARG_PTR_TO_MEM, 5003 .arg4_type = ARG_CONST_SIZE 5004 }; 5005 5006 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5007 BPF_CALL_4(bpf_lwt_seg6_store_bytes, struct sk_buff *, skb, u32, offset, 5008 const void *, from, u32, len) 5009 { 5010 struct seg6_bpf_srh_state *srh_state = 5011 this_cpu_ptr(&seg6_bpf_srh_states); 5012 struct ipv6_sr_hdr *srh = srh_state->srh; 5013 void *srh_tlvs, *srh_end, *ptr; 5014 int srhoff = 0; 5015 5016 if (srh == NULL) 5017 return -EINVAL; 5018 5019 srh_tlvs = (void *)((char *)srh + ((srh->first_segment + 1) << 4)); 5020 srh_end = (void *)((char *)srh + sizeof(*srh) + srh_state->hdrlen); 5021 5022 ptr = skb->data + offset; 5023 if (ptr >= srh_tlvs && ptr + len <= srh_end) 5024 srh_state->valid = false; 5025 else if (ptr < (void *)&srh->flags || 5026 ptr + len > (void *)&srh->segments) 5027 return -EFAULT; 5028 5029 if (unlikely(bpf_try_make_writable(skb, offset + len))) 5030 return -EFAULT; 5031 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 5032 return -EINVAL; 5033 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5034 5035 memcpy(skb->data + offset, from, len); 5036 return 0; 5037 } 5038 5039 static const struct bpf_func_proto bpf_lwt_seg6_store_bytes_proto = { 5040 .func = bpf_lwt_seg6_store_bytes, 5041 .gpl_only = false, 5042 .ret_type = RET_INTEGER, 5043 .arg1_type = ARG_PTR_TO_CTX, 5044 .arg2_type = ARG_ANYTHING, 5045 .arg3_type = ARG_PTR_TO_MEM, 5046 .arg4_type = ARG_CONST_SIZE 5047 }; 5048 5049 static void bpf_update_srh_state(struct sk_buff *skb) 5050 { 5051 struct seg6_bpf_srh_state *srh_state = 5052 this_cpu_ptr(&seg6_bpf_srh_states); 5053 int srhoff = 0; 5054 5055 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) { 5056 srh_state->srh = NULL; 5057 } else { 5058 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5059 srh_state->hdrlen = srh_state->srh->hdrlen << 3; 5060 srh_state->valid = true; 5061 } 5062 } 5063 5064 BPF_CALL_4(bpf_lwt_seg6_action, struct sk_buff *, skb, 5065 u32, action, void *, param, u32, param_len) 5066 { 5067 struct seg6_bpf_srh_state *srh_state = 5068 this_cpu_ptr(&seg6_bpf_srh_states); 5069 int hdroff = 0; 5070 int err; 5071 5072 switch (action) { 5073 case SEG6_LOCAL_ACTION_END_X: 5074 if (!seg6_bpf_has_valid_srh(skb)) 5075 return -EBADMSG; 5076 if (param_len != sizeof(struct in6_addr)) 5077 return -EINVAL; 5078 return seg6_lookup_nexthop(skb, (struct in6_addr *)param, 0); 5079 case SEG6_LOCAL_ACTION_END_T: 5080 if (!seg6_bpf_has_valid_srh(skb)) 5081 return -EBADMSG; 5082 if (param_len != sizeof(int)) 5083 return -EINVAL; 5084 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 5085 case SEG6_LOCAL_ACTION_END_DT6: 5086 if (!seg6_bpf_has_valid_srh(skb)) 5087 return -EBADMSG; 5088 if (param_len != sizeof(int)) 5089 return -EINVAL; 5090 5091 if (ipv6_find_hdr(skb, &hdroff, IPPROTO_IPV6, NULL, NULL) < 0) 5092 return -EBADMSG; 5093 if (!pskb_pull(skb, hdroff)) 5094 return -EBADMSG; 5095 5096 skb_postpull_rcsum(skb, skb_network_header(skb), hdroff); 5097 skb_reset_network_header(skb); 5098 skb_reset_transport_header(skb); 5099 skb->encapsulation = 0; 5100 5101 bpf_compute_data_pointers(skb); 5102 bpf_update_srh_state(skb); 5103 return seg6_lookup_nexthop(skb, NULL, *(int *)param); 5104 case SEG6_LOCAL_ACTION_END_B6: 5105 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 5106 return -EBADMSG; 5107 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6_INLINE, 5108 param, param_len); 5109 if (!err) 5110 bpf_update_srh_state(skb); 5111 5112 return err; 5113 case SEG6_LOCAL_ACTION_END_B6_ENCAP: 5114 if (srh_state->srh && !seg6_bpf_has_valid_srh(skb)) 5115 return -EBADMSG; 5116 err = bpf_push_seg6_encap(skb, BPF_LWT_ENCAP_SEG6, 5117 param, param_len); 5118 if (!err) 5119 bpf_update_srh_state(skb); 5120 5121 return err; 5122 default: 5123 return -EINVAL; 5124 } 5125 } 5126 5127 static const struct bpf_func_proto bpf_lwt_seg6_action_proto = { 5128 .func = bpf_lwt_seg6_action, 5129 .gpl_only = false, 5130 .ret_type = RET_INTEGER, 5131 .arg1_type = ARG_PTR_TO_CTX, 5132 .arg2_type = ARG_ANYTHING, 5133 .arg3_type = ARG_PTR_TO_MEM, 5134 .arg4_type = ARG_CONST_SIZE 5135 }; 5136 5137 BPF_CALL_3(bpf_lwt_seg6_adjust_srh, struct sk_buff *, skb, u32, offset, 5138 s32, len) 5139 { 5140 struct seg6_bpf_srh_state *srh_state = 5141 this_cpu_ptr(&seg6_bpf_srh_states); 5142 struct ipv6_sr_hdr *srh = srh_state->srh; 5143 void *srh_end, *srh_tlvs, *ptr; 5144 struct ipv6hdr *hdr; 5145 int srhoff = 0; 5146 int ret; 5147 5148 if (unlikely(srh == NULL)) 5149 return -EINVAL; 5150 5151 srh_tlvs = (void *)((unsigned char *)srh + sizeof(*srh) + 5152 ((srh->first_segment + 1) << 4)); 5153 srh_end = (void *)((unsigned char *)srh + sizeof(*srh) + 5154 srh_state->hdrlen); 5155 ptr = skb->data + offset; 5156 5157 if (unlikely(ptr < srh_tlvs || ptr > srh_end)) 5158 return -EFAULT; 5159 if (unlikely(len < 0 && (void *)((char *)ptr - len) > srh_end)) 5160 return -EFAULT; 5161 5162 if (len > 0) { 5163 ret = skb_cow_head(skb, len); 5164 if (unlikely(ret < 0)) 5165 return ret; 5166 5167 ret = bpf_skb_net_hdr_push(skb, offset, len); 5168 } else { 5169 ret = bpf_skb_net_hdr_pop(skb, offset, -1 * len); 5170 } 5171 5172 bpf_compute_data_pointers(skb); 5173 if (unlikely(ret < 0)) 5174 return ret; 5175 5176 hdr = (struct ipv6hdr *)skb->data; 5177 hdr->payload_len = htons(skb->len - sizeof(struct ipv6hdr)); 5178 5179 if (ipv6_find_hdr(skb, &srhoff, IPPROTO_ROUTING, NULL, NULL) < 0) 5180 return -EINVAL; 5181 srh_state->srh = (struct ipv6_sr_hdr *)(skb->data + srhoff); 5182 srh_state->hdrlen += len; 5183 srh_state->valid = false; 5184 return 0; 5185 } 5186 5187 static const struct bpf_func_proto bpf_lwt_seg6_adjust_srh_proto = { 5188 .func = bpf_lwt_seg6_adjust_srh, 5189 .gpl_only = false, 5190 .ret_type = RET_INTEGER, 5191 .arg1_type = ARG_PTR_TO_CTX, 5192 .arg2_type = ARG_ANYTHING, 5193 .arg3_type = ARG_ANYTHING, 5194 }; 5195 #endif /* CONFIG_IPV6_SEG6_BPF */ 5196 5197 #ifdef CONFIG_INET 5198 static struct sock *sk_lookup(struct net *net, struct bpf_sock_tuple *tuple, 5199 int dif, int sdif, u8 family, u8 proto) 5200 { 5201 bool refcounted = false; 5202 struct sock *sk = NULL; 5203 5204 if (family == AF_INET) { 5205 __be32 src4 = tuple->ipv4.saddr; 5206 __be32 dst4 = tuple->ipv4.daddr; 5207 5208 if (proto == IPPROTO_TCP) 5209 sk = __inet_lookup(net, &tcp_hashinfo, NULL, 0, 5210 src4, tuple->ipv4.sport, 5211 dst4, tuple->ipv4.dport, 5212 dif, sdif, &refcounted); 5213 else 5214 sk = __udp4_lib_lookup(net, src4, tuple->ipv4.sport, 5215 dst4, tuple->ipv4.dport, 5216 dif, sdif, &udp_table, NULL); 5217 #if IS_ENABLED(CONFIG_IPV6) 5218 } else { 5219 struct in6_addr *src6 = (struct in6_addr *)&tuple->ipv6.saddr; 5220 struct in6_addr *dst6 = (struct in6_addr *)&tuple->ipv6.daddr; 5221 5222 if (proto == IPPROTO_TCP) 5223 sk = __inet6_lookup(net, &tcp_hashinfo, NULL, 0, 5224 src6, tuple->ipv6.sport, 5225 dst6, ntohs(tuple->ipv6.dport), 5226 dif, sdif, &refcounted); 5227 else if (likely(ipv6_bpf_stub)) 5228 sk = ipv6_bpf_stub->udp6_lib_lookup(net, 5229 src6, tuple->ipv6.sport, 5230 dst6, tuple->ipv6.dport, 5231 dif, sdif, 5232 &udp_table, NULL); 5233 #endif 5234 } 5235 5236 if (unlikely(sk && !refcounted && !sock_flag(sk, SOCK_RCU_FREE))) { 5237 WARN_ONCE(1, "Found non-RCU, unreferenced socket!"); 5238 sk = NULL; 5239 } 5240 return sk; 5241 } 5242 5243 /* bpf_skc_lookup performs the core lookup for different types of sockets, 5244 * taking a reference on the socket if it doesn't have the flag SOCK_RCU_FREE. 5245 * Returns the socket as an 'unsigned long' to simplify the casting in the 5246 * callers to satisfy BPF_CALL declarations. 5247 */ 5248 static struct sock * 5249 __bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5250 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 5251 u64 flags) 5252 { 5253 struct sock *sk = NULL; 5254 u8 family = AF_UNSPEC; 5255 struct net *net; 5256 int sdif; 5257 5258 if (len == sizeof(tuple->ipv4)) 5259 family = AF_INET; 5260 else if (len == sizeof(tuple->ipv6)) 5261 family = AF_INET6; 5262 else 5263 return NULL; 5264 5265 if (unlikely(family == AF_UNSPEC || flags || 5266 !((s32)netns_id < 0 || netns_id <= S32_MAX))) 5267 goto out; 5268 5269 if (family == AF_INET) 5270 sdif = inet_sdif(skb); 5271 else 5272 sdif = inet6_sdif(skb); 5273 5274 if ((s32)netns_id < 0) { 5275 net = caller_net; 5276 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 5277 } else { 5278 net = get_net_ns_by_id(caller_net, netns_id); 5279 if (unlikely(!net)) 5280 goto out; 5281 sk = sk_lookup(net, tuple, ifindex, sdif, family, proto); 5282 put_net(net); 5283 } 5284 5285 out: 5286 return sk; 5287 } 5288 5289 static struct sock * 5290 __bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5291 struct net *caller_net, u32 ifindex, u8 proto, u64 netns_id, 5292 u64 flags) 5293 { 5294 struct sock *sk = __bpf_skc_lookup(skb, tuple, len, caller_net, 5295 ifindex, proto, netns_id, flags); 5296 5297 if (sk) { 5298 sk = sk_to_full_sk(sk); 5299 if (!sk_fullsock(sk)) { 5300 sock_gen_put(sk); 5301 return NULL; 5302 } 5303 } 5304 5305 return sk; 5306 } 5307 5308 static struct sock * 5309 bpf_skc_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5310 u8 proto, u64 netns_id, u64 flags) 5311 { 5312 struct net *caller_net; 5313 int ifindex; 5314 5315 if (skb->dev) { 5316 caller_net = dev_net(skb->dev); 5317 ifindex = skb->dev->ifindex; 5318 } else { 5319 caller_net = sock_net(skb->sk); 5320 ifindex = 0; 5321 } 5322 5323 return __bpf_skc_lookup(skb, tuple, len, caller_net, ifindex, proto, 5324 netns_id, flags); 5325 } 5326 5327 static struct sock * 5328 bpf_sk_lookup(struct sk_buff *skb, struct bpf_sock_tuple *tuple, u32 len, 5329 u8 proto, u64 netns_id, u64 flags) 5330 { 5331 struct sock *sk = bpf_skc_lookup(skb, tuple, len, proto, netns_id, 5332 flags); 5333 5334 if (sk) { 5335 sk = sk_to_full_sk(sk); 5336 if (!sk_fullsock(sk)) { 5337 sock_gen_put(sk); 5338 return NULL; 5339 } 5340 } 5341 5342 return sk; 5343 } 5344 5345 BPF_CALL_5(bpf_skc_lookup_tcp, struct sk_buff *, skb, 5346 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 5347 { 5348 return (unsigned long)bpf_skc_lookup(skb, tuple, len, IPPROTO_TCP, 5349 netns_id, flags); 5350 } 5351 5352 static const struct bpf_func_proto bpf_skc_lookup_tcp_proto = { 5353 .func = bpf_skc_lookup_tcp, 5354 .gpl_only = false, 5355 .pkt_access = true, 5356 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 5357 .arg1_type = ARG_PTR_TO_CTX, 5358 .arg2_type = ARG_PTR_TO_MEM, 5359 .arg3_type = ARG_CONST_SIZE, 5360 .arg4_type = ARG_ANYTHING, 5361 .arg5_type = ARG_ANYTHING, 5362 }; 5363 5364 BPF_CALL_5(bpf_sk_lookup_tcp, struct sk_buff *, skb, 5365 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 5366 { 5367 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_TCP, 5368 netns_id, flags); 5369 } 5370 5371 static const struct bpf_func_proto bpf_sk_lookup_tcp_proto = { 5372 .func = bpf_sk_lookup_tcp, 5373 .gpl_only = false, 5374 .pkt_access = true, 5375 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5376 .arg1_type = ARG_PTR_TO_CTX, 5377 .arg2_type = ARG_PTR_TO_MEM, 5378 .arg3_type = ARG_CONST_SIZE, 5379 .arg4_type = ARG_ANYTHING, 5380 .arg5_type = ARG_ANYTHING, 5381 }; 5382 5383 BPF_CALL_5(bpf_sk_lookup_udp, struct sk_buff *, skb, 5384 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 5385 { 5386 return (unsigned long)bpf_sk_lookup(skb, tuple, len, IPPROTO_UDP, 5387 netns_id, flags); 5388 } 5389 5390 static const struct bpf_func_proto bpf_sk_lookup_udp_proto = { 5391 .func = bpf_sk_lookup_udp, 5392 .gpl_only = false, 5393 .pkt_access = true, 5394 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5395 .arg1_type = ARG_PTR_TO_CTX, 5396 .arg2_type = ARG_PTR_TO_MEM, 5397 .arg3_type = ARG_CONST_SIZE, 5398 .arg4_type = ARG_ANYTHING, 5399 .arg5_type = ARG_ANYTHING, 5400 }; 5401 5402 BPF_CALL_1(bpf_sk_release, struct sock *, sk) 5403 { 5404 if (sk_is_refcounted(sk)) 5405 sock_gen_put(sk); 5406 return 0; 5407 } 5408 5409 static const struct bpf_func_proto bpf_sk_release_proto = { 5410 .func = bpf_sk_release, 5411 .gpl_only = false, 5412 .ret_type = RET_INTEGER, 5413 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 5414 }; 5415 5416 BPF_CALL_5(bpf_xdp_sk_lookup_udp, struct xdp_buff *, ctx, 5417 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 5418 { 5419 struct net *caller_net = dev_net(ctx->rxq->dev); 5420 int ifindex = ctx->rxq->dev->ifindex; 5421 5422 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 5423 ifindex, IPPROTO_UDP, netns_id, 5424 flags); 5425 } 5426 5427 static const struct bpf_func_proto bpf_xdp_sk_lookup_udp_proto = { 5428 .func = bpf_xdp_sk_lookup_udp, 5429 .gpl_only = false, 5430 .pkt_access = true, 5431 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5432 .arg1_type = ARG_PTR_TO_CTX, 5433 .arg2_type = ARG_PTR_TO_MEM, 5434 .arg3_type = ARG_CONST_SIZE, 5435 .arg4_type = ARG_ANYTHING, 5436 .arg5_type = ARG_ANYTHING, 5437 }; 5438 5439 BPF_CALL_5(bpf_xdp_skc_lookup_tcp, struct xdp_buff *, ctx, 5440 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 5441 { 5442 struct net *caller_net = dev_net(ctx->rxq->dev); 5443 int ifindex = ctx->rxq->dev->ifindex; 5444 5445 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, caller_net, 5446 ifindex, IPPROTO_TCP, netns_id, 5447 flags); 5448 } 5449 5450 static const struct bpf_func_proto bpf_xdp_skc_lookup_tcp_proto = { 5451 .func = bpf_xdp_skc_lookup_tcp, 5452 .gpl_only = false, 5453 .pkt_access = true, 5454 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 5455 .arg1_type = ARG_PTR_TO_CTX, 5456 .arg2_type = ARG_PTR_TO_MEM, 5457 .arg3_type = ARG_CONST_SIZE, 5458 .arg4_type = ARG_ANYTHING, 5459 .arg5_type = ARG_ANYTHING, 5460 }; 5461 5462 BPF_CALL_5(bpf_xdp_sk_lookup_tcp, struct xdp_buff *, ctx, 5463 struct bpf_sock_tuple *, tuple, u32, len, u32, netns_id, u64, flags) 5464 { 5465 struct net *caller_net = dev_net(ctx->rxq->dev); 5466 int ifindex = ctx->rxq->dev->ifindex; 5467 5468 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, caller_net, 5469 ifindex, IPPROTO_TCP, netns_id, 5470 flags); 5471 } 5472 5473 static const struct bpf_func_proto bpf_xdp_sk_lookup_tcp_proto = { 5474 .func = bpf_xdp_sk_lookup_tcp, 5475 .gpl_only = false, 5476 .pkt_access = true, 5477 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5478 .arg1_type = ARG_PTR_TO_CTX, 5479 .arg2_type = ARG_PTR_TO_MEM, 5480 .arg3_type = ARG_CONST_SIZE, 5481 .arg4_type = ARG_ANYTHING, 5482 .arg5_type = ARG_ANYTHING, 5483 }; 5484 5485 BPF_CALL_5(bpf_sock_addr_skc_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 5486 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 5487 { 5488 return (unsigned long)__bpf_skc_lookup(NULL, tuple, len, 5489 sock_net(ctx->sk), 0, 5490 IPPROTO_TCP, netns_id, flags); 5491 } 5492 5493 static const struct bpf_func_proto bpf_sock_addr_skc_lookup_tcp_proto = { 5494 .func = bpf_sock_addr_skc_lookup_tcp, 5495 .gpl_only = false, 5496 .ret_type = RET_PTR_TO_SOCK_COMMON_OR_NULL, 5497 .arg1_type = ARG_PTR_TO_CTX, 5498 .arg2_type = ARG_PTR_TO_MEM, 5499 .arg3_type = ARG_CONST_SIZE, 5500 .arg4_type = ARG_ANYTHING, 5501 .arg5_type = ARG_ANYTHING, 5502 }; 5503 5504 BPF_CALL_5(bpf_sock_addr_sk_lookup_tcp, struct bpf_sock_addr_kern *, ctx, 5505 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 5506 { 5507 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 5508 sock_net(ctx->sk), 0, IPPROTO_TCP, 5509 netns_id, flags); 5510 } 5511 5512 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_tcp_proto = { 5513 .func = bpf_sock_addr_sk_lookup_tcp, 5514 .gpl_only = false, 5515 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5516 .arg1_type = ARG_PTR_TO_CTX, 5517 .arg2_type = ARG_PTR_TO_MEM, 5518 .arg3_type = ARG_CONST_SIZE, 5519 .arg4_type = ARG_ANYTHING, 5520 .arg5_type = ARG_ANYTHING, 5521 }; 5522 5523 BPF_CALL_5(bpf_sock_addr_sk_lookup_udp, struct bpf_sock_addr_kern *, ctx, 5524 struct bpf_sock_tuple *, tuple, u32, len, u64, netns_id, u64, flags) 5525 { 5526 return (unsigned long)__bpf_sk_lookup(NULL, tuple, len, 5527 sock_net(ctx->sk), 0, IPPROTO_UDP, 5528 netns_id, flags); 5529 } 5530 5531 static const struct bpf_func_proto bpf_sock_addr_sk_lookup_udp_proto = { 5532 .func = bpf_sock_addr_sk_lookup_udp, 5533 .gpl_only = false, 5534 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5535 .arg1_type = ARG_PTR_TO_CTX, 5536 .arg2_type = ARG_PTR_TO_MEM, 5537 .arg3_type = ARG_CONST_SIZE, 5538 .arg4_type = ARG_ANYTHING, 5539 .arg5_type = ARG_ANYTHING, 5540 }; 5541 5542 bool bpf_tcp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 5543 struct bpf_insn_access_aux *info) 5544 { 5545 if (off < 0 || off >= offsetofend(struct bpf_tcp_sock, 5546 icsk_retransmits)) 5547 return false; 5548 5549 if (off % size != 0) 5550 return false; 5551 5552 switch (off) { 5553 case offsetof(struct bpf_tcp_sock, bytes_received): 5554 case offsetof(struct bpf_tcp_sock, bytes_acked): 5555 return size == sizeof(__u64); 5556 default: 5557 return size == sizeof(__u32); 5558 } 5559 } 5560 5561 u32 bpf_tcp_sock_convert_ctx_access(enum bpf_access_type type, 5562 const struct bpf_insn *si, 5563 struct bpf_insn *insn_buf, 5564 struct bpf_prog *prog, u32 *target_size) 5565 { 5566 struct bpf_insn *insn = insn_buf; 5567 5568 #define BPF_TCP_SOCK_GET_COMMON(FIELD) \ 5569 do { \ 5570 BUILD_BUG_ON(sizeof_field(struct tcp_sock, FIELD) > \ 5571 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 5572 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct tcp_sock, FIELD),\ 5573 si->dst_reg, si->src_reg, \ 5574 offsetof(struct tcp_sock, FIELD)); \ 5575 } while (0) 5576 5577 #define BPF_INET_SOCK_GET_COMMON(FIELD) \ 5578 do { \ 5579 BUILD_BUG_ON(sizeof_field(struct inet_connection_sock, \ 5580 FIELD) > \ 5581 sizeof_field(struct bpf_tcp_sock, FIELD)); \ 5582 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 5583 struct inet_connection_sock, \ 5584 FIELD), \ 5585 si->dst_reg, si->src_reg, \ 5586 offsetof( \ 5587 struct inet_connection_sock, \ 5588 FIELD)); \ 5589 } while (0) 5590 5591 if (insn > insn_buf) 5592 return insn - insn_buf; 5593 5594 switch (si->off) { 5595 case offsetof(struct bpf_tcp_sock, rtt_min): 5596 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 5597 sizeof(struct minmax)); 5598 BUILD_BUG_ON(sizeof(struct minmax) < 5599 sizeof(struct minmax_sample)); 5600 5601 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 5602 offsetof(struct tcp_sock, rtt_min) + 5603 offsetof(struct minmax_sample, v)); 5604 break; 5605 case offsetof(struct bpf_tcp_sock, snd_cwnd): 5606 BPF_TCP_SOCK_GET_COMMON(snd_cwnd); 5607 break; 5608 case offsetof(struct bpf_tcp_sock, srtt_us): 5609 BPF_TCP_SOCK_GET_COMMON(srtt_us); 5610 break; 5611 case offsetof(struct bpf_tcp_sock, snd_ssthresh): 5612 BPF_TCP_SOCK_GET_COMMON(snd_ssthresh); 5613 break; 5614 case offsetof(struct bpf_tcp_sock, rcv_nxt): 5615 BPF_TCP_SOCK_GET_COMMON(rcv_nxt); 5616 break; 5617 case offsetof(struct bpf_tcp_sock, snd_nxt): 5618 BPF_TCP_SOCK_GET_COMMON(snd_nxt); 5619 break; 5620 case offsetof(struct bpf_tcp_sock, snd_una): 5621 BPF_TCP_SOCK_GET_COMMON(snd_una); 5622 break; 5623 case offsetof(struct bpf_tcp_sock, mss_cache): 5624 BPF_TCP_SOCK_GET_COMMON(mss_cache); 5625 break; 5626 case offsetof(struct bpf_tcp_sock, ecn_flags): 5627 BPF_TCP_SOCK_GET_COMMON(ecn_flags); 5628 break; 5629 case offsetof(struct bpf_tcp_sock, rate_delivered): 5630 BPF_TCP_SOCK_GET_COMMON(rate_delivered); 5631 break; 5632 case offsetof(struct bpf_tcp_sock, rate_interval_us): 5633 BPF_TCP_SOCK_GET_COMMON(rate_interval_us); 5634 break; 5635 case offsetof(struct bpf_tcp_sock, packets_out): 5636 BPF_TCP_SOCK_GET_COMMON(packets_out); 5637 break; 5638 case offsetof(struct bpf_tcp_sock, retrans_out): 5639 BPF_TCP_SOCK_GET_COMMON(retrans_out); 5640 break; 5641 case offsetof(struct bpf_tcp_sock, total_retrans): 5642 BPF_TCP_SOCK_GET_COMMON(total_retrans); 5643 break; 5644 case offsetof(struct bpf_tcp_sock, segs_in): 5645 BPF_TCP_SOCK_GET_COMMON(segs_in); 5646 break; 5647 case offsetof(struct bpf_tcp_sock, data_segs_in): 5648 BPF_TCP_SOCK_GET_COMMON(data_segs_in); 5649 break; 5650 case offsetof(struct bpf_tcp_sock, segs_out): 5651 BPF_TCP_SOCK_GET_COMMON(segs_out); 5652 break; 5653 case offsetof(struct bpf_tcp_sock, data_segs_out): 5654 BPF_TCP_SOCK_GET_COMMON(data_segs_out); 5655 break; 5656 case offsetof(struct bpf_tcp_sock, lost_out): 5657 BPF_TCP_SOCK_GET_COMMON(lost_out); 5658 break; 5659 case offsetof(struct bpf_tcp_sock, sacked_out): 5660 BPF_TCP_SOCK_GET_COMMON(sacked_out); 5661 break; 5662 case offsetof(struct bpf_tcp_sock, bytes_received): 5663 BPF_TCP_SOCK_GET_COMMON(bytes_received); 5664 break; 5665 case offsetof(struct bpf_tcp_sock, bytes_acked): 5666 BPF_TCP_SOCK_GET_COMMON(bytes_acked); 5667 break; 5668 case offsetof(struct bpf_tcp_sock, dsack_dups): 5669 BPF_TCP_SOCK_GET_COMMON(dsack_dups); 5670 break; 5671 case offsetof(struct bpf_tcp_sock, delivered): 5672 BPF_TCP_SOCK_GET_COMMON(delivered); 5673 break; 5674 case offsetof(struct bpf_tcp_sock, delivered_ce): 5675 BPF_TCP_SOCK_GET_COMMON(delivered_ce); 5676 break; 5677 case offsetof(struct bpf_tcp_sock, icsk_retransmits): 5678 BPF_INET_SOCK_GET_COMMON(icsk_retransmits); 5679 break; 5680 } 5681 5682 return insn - insn_buf; 5683 } 5684 5685 BPF_CALL_1(bpf_tcp_sock, struct sock *, sk) 5686 { 5687 if (sk_fullsock(sk) && sk->sk_protocol == IPPROTO_TCP) 5688 return (unsigned long)sk; 5689 5690 return (unsigned long)NULL; 5691 } 5692 5693 const struct bpf_func_proto bpf_tcp_sock_proto = { 5694 .func = bpf_tcp_sock, 5695 .gpl_only = false, 5696 .ret_type = RET_PTR_TO_TCP_SOCK_OR_NULL, 5697 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 5698 }; 5699 5700 BPF_CALL_1(bpf_get_listener_sock, struct sock *, sk) 5701 { 5702 sk = sk_to_full_sk(sk); 5703 5704 if (sk->sk_state == TCP_LISTEN && sock_flag(sk, SOCK_RCU_FREE)) 5705 return (unsigned long)sk; 5706 5707 return (unsigned long)NULL; 5708 } 5709 5710 static const struct bpf_func_proto bpf_get_listener_sock_proto = { 5711 .func = bpf_get_listener_sock, 5712 .gpl_only = false, 5713 .ret_type = RET_PTR_TO_SOCKET_OR_NULL, 5714 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 5715 }; 5716 5717 BPF_CALL_1(bpf_skb_ecn_set_ce, struct sk_buff *, skb) 5718 { 5719 unsigned int iphdr_len; 5720 5721 if (skb->protocol == cpu_to_be16(ETH_P_IP)) 5722 iphdr_len = sizeof(struct iphdr); 5723 else if (skb->protocol == cpu_to_be16(ETH_P_IPV6)) 5724 iphdr_len = sizeof(struct ipv6hdr); 5725 else 5726 return 0; 5727 5728 if (skb_headlen(skb) < iphdr_len) 5729 return 0; 5730 5731 if (skb_cloned(skb) && !skb_clone_writable(skb, iphdr_len)) 5732 return 0; 5733 5734 return INET_ECN_set_ce(skb); 5735 } 5736 5737 bool bpf_xdp_sock_is_valid_access(int off, int size, enum bpf_access_type type, 5738 struct bpf_insn_access_aux *info) 5739 { 5740 if (off < 0 || off >= offsetofend(struct bpf_xdp_sock, queue_id)) 5741 return false; 5742 5743 if (off % size != 0) 5744 return false; 5745 5746 switch (off) { 5747 default: 5748 return size == sizeof(__u32); 5749 } 5750 } 5751 5752 u32 bpf_xdp_sock_convert_ctx_access(enum bpf_access_type type, 5753 const struct bpf_insn *si, 5754 struct bpf_insn *insn_buf, 5755 struct bpf_prog *prog, u32 *target_size) 5756 { 5757 struct bpf_insn *insn = insn_buf; 5758 5759 #define BPF_XDP_SOCK_GET(FIELD) \ 5760 do { \ 5761 BUILD_BUG_ON(sizeof_field(struct xdp_sock, FIELD) > \ 5762 sizeof_field(struct bpf_xdp_sock, FIELD)); \ 5763 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_sock, FIELD),\ 5764 si->dst_reg, si->src_reg, \ 5765 offsetof(struct xdp_sock, FIELD)); \ 5766 } while (0) 5767 5768 switch (si->off) { 5769 case offsetof(struct bpf_xdp_sock, queue_id): 5770 BPF_XDP_SOCK_GET(queue_id); 5771 break; 5772 } 5773 5774 return insn - insn_buf; 5775 } 5776 5777 static const struct bpf_func_proto bpf_skb_ecn_set_ce_proto = { 5778 .func = bpf_skb_ecn_set_ce, 5779 .gpl_only = false, 5780 .ret_type = RET_INTEGER, 5781 .arg1_type = ARG_PTR_TO_CTX, 5782 }; 5783 5784 BPF_CALL_5(bpf_tcp_check_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 5785 struct tcphdr *, th, u32, th_len) 5786 { 5787 #ifdef CONFIG_SYN_COOKIES 5788 u32 cookie; 5789 int ret; 5790 5791 if (unlikely(th_len < sizeof(*th))) 5792 return -EINVAL; 5793 5794 /* sk_listener() allows TCP_NEW_SYN_RECV, which makes no sense here. */ 5795 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 5796 return -EINVAL; 5797 5798 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies) 5799 return -EINVAL; 5800 5801 if (!th->ack || th->rst || th->syn) 5802 return -ENOENT; 5803 5804 if (tcp_synq_no_recent_overflow(sk)) 5805 return -ENOENT; 5806 5807 cookie = ntohl(th->ack_seq) - 1; 5808 5809 switch (sk->sk_family) { 5810 case AF_INET: 5811 if (unlikely(iph_len < sizeof(struct iphdr))) 5812 return -EINVAL; 5813 5814 ret = __cookie_v4_check((struct iphdr *)iph, th, cookie); 5815 break; 5816 5817 #if IS_BUILTIN(CONFIG_IPV6) 5818 case AF_INET6: 5819 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 5820 return -EINVAL; 5821 5822 ret = __cookie_v6_check((struct ipv6hdr *)iph, th, cookie); 5823 break; 5824 #endif /* CONFIG_IPV6 */ 5825 5826 default: 5827 return -EPROTONOSUPPORT; 5828 } 5829 5830 if (ret > 0) 5831 return 0; 5832 5833 return -ENOENT; 5834 #else 5835 return -ENOTSUPP; 5836 #endif 5837 } 5838 5839 static const struct bpf_func_proto bpf_tcp_check_syncookie_proto = { 5840 .func = bpf_tcp_check_syncookie, 5841 .gpl_only = true, 5842 .pkt_access = true, 5843 .ret_type = RET_INTEGER, 5844 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 5845 .arg2_type = ARG_PTR_TO_MEM, 5846 .arg3_type = ARG_CONST_SIZE, 5847 .arg4_type = ARG_PTR_TO_MEM, 5848 .arg5_type = ARG_CONST_SIZE, 5849 }; 5850 5851 BPF_CALL_5(bpf_tcp_gen_syncookie, struct sock *, sk, void *, iph, u32, iph_len, 5852 struct tcphdr *, th, u32, th_len) 5853 { 5854 #ifdef CONFIG_SYN_COOKIES 5855 u32 cookie; 5856 u16 mss; 5857 5858 if (unlikely(th_len < sizeof(*th) || th_len != th->doff * 4)) 5859 return -EINVAL; 5860 5861 if (sk->sk_protocol != IPPROTO_TCP || sk->sk_state != TCP_LISTEN) 5862 return -EINVAL; 5863 5864 if (!sock_net(sk)->ipv4.sysctl_tcp_syncookies) 5865 return -ENOENT; 5866 5867 if (!th->syn || th->ack || th->fin || th->rst) 5868 return -EINVAL; 5869 5870 if (unlikely(iph_len < sizeof(struct iphdr))) 5871 return -EINVAL; 5872 5873 /* Both struct iphdr and struct ipv6hdr have the version field at the 5874 * same offset so we can cast to the shorter header (struct iphdr). 5875 */ 5876 switch (((struct iphdr *)iph)->version) { 5877 case 4: 5878 if (sk->sk_family == AF_INET6 && sk->sk_ipv6only) 5879 return -EINVAL; 5880 5881 mss = tcp_v4_get_syncookie(sk, iph, th, &cookie); 5882 break; 5883 5884 #if IS_BUILTIN(CONFIG_IPV6) 5885 case 6: 5886 if (unlikely(iph_len < sizeof(struct ipv6hdr))) 5887 return -EINVAL; 5888 5889 if (sk->sk_family != AF_INET6) 5890 return -EINVAL; 5891 5892 mss = tcp_v6_get_syncookie(sk, iph, th, &cookie); 5893 break; 5894 #endif /* CONFIG_IPV6 */ 5895 5896 default: 5897 return -EPROTONOSUPPORT; 5898 } 5899 if (mss == 0) 5900 return -ENOENT; 5901 5902 return cookie | ((u64)mss << 32); 5903 #else 5904 return -EOPNOTSUPP; 5905 #endif /* CONFIG_SYN_COOKIES */ 5906 } 5907 5908 static const struct bpf_func_proto bpf_tcp_gen_syncookie_proto = { 5909 .func = bpf_tcp_gen_syncookie, 5910 .gpl_only = true, /* __cookie_v*_init_sequence() is GPL */ 5911 .pkt_access = true, 5912 .ret_type = RET_INTEGER, 5913 .arg1_type = ARG_PTR_TO_SOCK_COMMON, 5914 .arg2_type = ARG_PTR_TO_MEM, 5915 .arg3_type = ARG_CONST_SIZE, 5916 .arg4_type = ARG_PTR_TO_MEM, 5917 .arg5_type = ARG_CONST_SIZE, 5918 }; 5919 5920 BPF_CALL_3(bpf_sk_assign, struct sk_buff *, skb, struct sock *, sk, u64, flags) 5921 { 5922 if (flags != 0) 5923 return -EINVAL; 5924 if (!skb_at_tc_ingress(skb)) 5925 return -EOPNOTSUPP; 5926 if (unlikely(dev_net(skb->dev) != sock_net(sk))) 5927 return -ENETUNREACH; 5928 if (unlikely(sk->sk_reuseport)) 5929 return -ESOCKTNOSUPPORT; 5930 if (sk_is_refcounted(sk) && 5931 unlikely(!refcount_inc_not_zero(&sk->sk_refcnt))) 5932 return -ENOENT; 5933 5934 skb_orphan(skb); 5935 skb->sk = sk; 5936 skb->destructor = sock_pfree; 5937 5938 return 0; 5939 } 5940 5941 static const struct bpf_func_proto bpf_sk_assign_proto = { 5942 .func = bpf_sk_assign, 5943 .gpl_only = false, 5944 .ret_type = RET_INTEGER, 5945 .arg1_type = ARG_PTR_TO_CTX, 5946 .arg2_type = ARG_PTR_TO_SOCK_COMMON, 5947 .arg3_type = ARG_ANYTHING, 5948 }; 5949 5950 #endif /* CONFIG_INET */ 5951 5952 bool bpf_helper_changes_pkt_data(void *func) 5953 { 5954 if (func == bpf_skb_vlan_push || 5955 func == bpf_skb_vlan_pop || 5956 func == bpf_skb_store_bytes || 5957 func == bpf_skb_change_proto || 5958 func == bpf_skb_change_head || 5959 func == sk_skb_change_head || 5960 func == bpf_skb_change_tail || 5961 func == sk_skb_change_tail || 5962 func == bpf_skb_adjust_room || 5963 func == bpf_skb_pull_data || 5964 func == sk_skb_pull_data || 5965 func == bpf_clone_redirect || 5966 func == bpf_l3_csum_replace || 5967 func == bpf_l4_csum_replace || 5968 func == bpf_xdp_adjust_head || 5969 func == bpf_xdp_adjust_meta || 5970 func == bpf_msg_pull_data || 5971 func == bpf_msg_push_data || 5972 func == bpf_msg_pop_data || 5973 func == bpf_xdp_adjust_tail || 5974 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 5975 func == bpf_lwt_seg6_store_bytes || 5976 func == bpf_lwt_seg6_adjust_srh || 5977 func == bpf_lwt_seg6_action || 5978 #endif 5979 func == bpf_lwt_in_push_encap || 5980 func == bpf_lwt_xmit_push_encap) 5981 return true; 5982 5983 return false; 5984 } 5985 5986 const struct bpf_func_proto * 5987 bpf_base_func_proto(enum bpf_func_id func_id) 5988 { 5989 switch (func_id) { 5990 case BPF_FUNC_map_lookup_elem: 5991 return &bpf_map_lookup_elem_proto; 5992 case BPF_FUNC_map_update_elem: 5993 return &bpf_map_update_elem_proto; 5994 case BPF_FUNC_map_delete_elem: 5995 return &bpf_map_delete_elem_proto; 5996 case BPF_FUNC_map_push_elem: 5997 return &bpf_map_push_elem_proto; 5998 case BPF_FUNC_map_pop_elem: 5999 return &bpf_map_pop_elem_proto; 6000 case BPF_FUNC_map_peek_elem: 6001 return &bpf_map_peek_elem_proto; 6002 case BPF_FUNC_get_prandom_u32: 6003 return &bpf_get_prandom_u32_proto; 6004 case BPF_FUNC_get_smp_processor_id: 6005 return &bpf_get_raw_smp_processor_id_proto; 6006 case BPF_FUNC_get_numa_node_id: 6007 return &bpf_get_numa_node_id_proto; 6008 case BPF_FUNC_tail_call: 6009 return &bpf_tail_call_proto; 6010 case BPF_FUNC_ktime_get_ns: 6011 return &bpf_ktime_get_ns_proto; 6012 default: 6013 break; 6014 } 6015 6016 if (!capable(CAP_SYS_ADMIN)) 6017 return NULL; 6018 6019 switch (func_id) { 6020 case BPF_FUNC_spin_lock: 6021 return &bpf_spin_lock_proto; 6022 case BPF_FUNC_spin_unlock: 6023 return &bpf_spin_unlock_proto; 6024 case BPF_FUNC_trace_printk: 6025 return bpf_get_trace_printk_proto(); 6026 case BPF_FUNC_jiffies64: 6027 return &bpf_jiffies64_proto; 6028 default: 6029 return NULL; 6030 } 6031 } 6032 6033 static const struct bpf_func_proto * 6034 sock_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6035 { 6036 switch (func_id) { 6037 /* inet and inet6 sockets are created in a process 6038 * context so there is always a valid uid/gid 6039 */ 6040 case BPF_FUNC_get_current_uid_gid: 6041 return &bpf_get_current_uid_gid_proto; 6042 case BPF_FUNC_get_local_storage: 6043 return &bpf_get_local_storage_proto; 6044 case BPF_FUNC_get_socket_cookie: 6045 return &bpf_get_socket_cookie_sock_proto; 6046 case BPF_FUNC_get_netns_cookie: 6047 return &bpf_get_netns_cookie_sock_proto; 6048 case BPF_FUNC_perf_event_output: 6049 return &bpf_event_output_data_proto; 6050 case BPF_FUNC_get_current_pid_tgid: 6051 return &bpf_get_current_pid_tgid_proto; 6052 case BPF_FUNC_get_current_comm: 6053 return &bpf_get_current_comm_proto; 6054 #ifdef CONFIG_CGROUPS 6055 case BPF_FUNC_get_current_cgroup_id: 6056 return &bpf_get_current_cgroup_id_proto; 6057 case BPF_FUNC_get_current_ancestor_cgroup_id: 6058 return &bpf_get_current_ancestor_cgroup_id_proto; 6059 #endif 6060 #ifdef CONFIG_CGROUP_NET_CLASSID 6061 case BPF_FUNC_get_cgroup_classid: 6062 return &bpf_get_cgroup_classid_curr_proto; 6063 #endif 6064 default: 6065 return bpf_base_func_proto(func_id); 6066 } 6067 } 6068 6069 static const struct bpf_func_proto * 6070 sock_addr_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6071 { 6072 switch (func_id) { 6073 /* inet and inet6 sockets are created in a process 6074 * context so there is always a valid uid/gid 6075 */ 6076 case BPF_FUNC_get_current_uid_gid: 6077 return &bpf_get_current_uid_gid_proto; 6078 case BPF_FUNC_bind: 6079 switch (prog->expected_attach_type) { 6080 case BPF_CGROUP_INET4_CONNECT: 6081 case BPF_CGROUP_INET6_CONNECT: 6082 return &bpf_bind_proto; 6083 default: 6084 return NULL; 6085 } 6086 case BPF_FUNC_get_socket_cookie: 6087 return &bpf_get_socket_cookie_sock_addr_proto; 6088 case BPF_FUNC_get_netns_cookie: 6089 return &bpf_get_netns_cookie_sock_addr_proto; 6090 case BPF_FUNC_get_local_storage: 6091 return &bpf_get_local_storage_proto; 6092 case BPF_FUNC_perf_event_output: 6093 return &bpf_event_output_data_proto; 6094 case BPF_FUNC_get_current_pid_tgid: 6095 return &bpf_get_current_pid_tgid_proto; 6096 case BPF_FUNC_get_current_comm: 6097 return &bpf_get_current_comm_proto; 6098 #ifdef CONFIG_CGROUPS 6099 case BPF_FUNC_get_current_cgroup_id: 6100 return &bpf_get_current_cgroup_id_proto; 6101 case BPF_FUNC_get_current_ancestor_cgroup_id: 6102 return &bpf_get_current_ancestor_cgroup_id_proto; 6103 #endif 6104 #ifdef CONFIG_CGROUP_NET_CLASSID 6105 case BPF_FUNC_get_cgroup_classid: 6106 return &bpf_get_cgroup_classid_curr_proto; 6107 #endif 6108 #ifdef CONFIG_INET 6109 case BPF_FUNC_sk_lookup_tcp: 6110 return &bpf_sock_addr_sk_lookup_tcp_proto; 6111 case BPF_FUNC_sk_lookup_udp: 6112 return &bpf_sock_addr_sk_lookup_udp_proto; 6113 case BPF_FUNC_sk_release: 6114 return &bpf_sk_release_proto; 6115 case BPF_FUNC_skc_lookup_tcp: 6116 return &bpf_sock_addr_skc_lookup_tcp_proto; 6117 #endif /* CONFIG_INET */ 6118 case BPF_FUNC_sk_storage_get: 6119 return &bpf_sk_storage_get_proto; 6120 case BPF_FUNC_sk_storage_delete: 6121 return &bpf_sk_storage_delete_proto; 6122 default: 6123 return bpf_base_func_proto(func_id); 6124 } 6125 } 6126 6127 static const struct bpf_func_proto * 6128 sk_filter_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6129 { 6130 switch (func_id) { 6131 case BPF_FUNC_skb_load_bytes: 6132 return &bpf_skb_load_bytes_proto; 6133 case BPF_FUNC_skb_load_bytes_relative: 6134 return &bpf_skb_load_bytes_relative_proto; 6135 case BPF_FUNC_get_socket_cookie: 6136 return &bpf_get_socket_cookie_proto; 6137 case BPF_FUNC_get_socket_uid: 6138 return &bpf_get_socket_uid_proto; 6139 case BPF_FUNC_perf_event_output: 6140 return &bpf_skb_event_output_proto; 6141 default: 6142 return bpf_base_func_proto(func_id); 6143 } 6144 } 6145 6146 const struct bpf_func_proto bpf_sk_storage_get_proto __weak; 6147 const struct bpf_func_proto bpf_sk_storage_delete_proto __weak; 6148 6149 static const struct bpf_func_proto * 6150 cg_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6151 { 6152 switch (func_id) { 6153 case BPF_FUNC_get_local_storage: 6154 return &bpf_get_local_storage_proto; 6155 case BPF_FUNC_sk_fullsock: 6156 return &bpf_sk_fullsock_proto; 6157 case BPF_FUNC_sk_storage_get: 6158 return &bpf_sk_storage_get_proto; 6159 case BPF_FUNC_sk_storage_delete: 6160 return &bpf_sk_storage_delete_proto; 6161 case BPF_FUNC_perf_event_output: 6162 return &bpf_skb_event_output_proto; 6163 #ifdef CONFIG_SOCK_CGROUP_DATA 6164 case BPF_FUNC_skb_cgroup_id: 6165 return &bpf_skb_cgroup_id_proto; 6166 #endif 6167 #ifdef CONFIG_INET 6168 case BPF_FUNC_tcp_sock: 6169 return &bpf_tcp_sock_proto; 6170 case BPF_FUNC_get_listener_sock: 6171 return &bpf_get_listener_sock_proto; 6172 case BPF_FUNC_skb_ecn_set_ce: 6173 return &bpf_skb_ecn_set_ce_proto; 6174 #endif 6175 default: 6176 return sk_filter_func_proto(func_id, prog); 6177 } 6178 } 6179 6180 static const struct bpf_func_proto * 6181 tc_cls_act_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6182 { 6183 switch (func_id) { 6184 case BPF_FUNC_skb_store_bytes: 6185 return &bpf_skb_store_bytes_proto; 6186 case BPF_FUNC_skb_load_bytes: 6187 return &bpf_skb_load_bytes_proto; 6188 case BPF_FUNC_skb_load_bytes_relative: 6189 return &bpf_skb_load_bytes_relative_proto; 6190 case BPF_FUNC_skb_pull_data: 6191 return &bpf_skb_pull_data_proto; 6192 case BPF_FUNC_csum_diff: 6193 return &bpf_csum_diff_proto; 6194 case BPF_FUNC_csum_update: 6195 return &bpf_csum_update_proto; 6196 case BPF_FUNC_l3_csum_replace: 6197 return &bpf_l3_csum_replace_proto; 6198 case BPF_FUNC_l4_csum_replace: 6199 return &bpf_l4_csum_replace_proto; 6200 case BPF_FUNC_clone_redirect: 6201 return &bpf_clone_redirect_proto; 6202 case BPF_FUNC_get_cgroup_classid: 6203 return &bpf_get_cgroup_classid_proto; 6204 case BPF_FUNC_skb_vlan_push: 6205 return &bpf_skb_vlan_push_proto; 6206 case BPF_FUNC_skb_vlan_pop: 6207 return &bpf_skb_vlan_pop_proto; 6208 case BPF_FUNC_skb_change_proto: 6209 return &bpf_skb_change_proto_proto; 6210 case BPF_FUNC_skb_change_type: 6211 return &bpf_skb_change_type_proto; 6212 case BPF_FUNC_skb_adjust_room: 6213 return &bpf_skb_adjust_room_proto; 6214 case BPF_FUNC_skb_change_tail: 6215 return &bpf_skb_change_tail_proto; 6216 case BPF_FUNC_skb_get_tunnel_key: 6217 return &bpf_skb_get_tunnel_key_proto; 6218 case BPF_FUNC_skb_set_tunnel_key: 6219 return bpf_get_skb_set_tunnel_proto(func_id); 6220 case BPF_FUNC_skb_get_tunnel_opt: 6221 return &bpf_skb_get_tunnel_opt_proto; 6222 case BPF_FUNC_skb_set_tunnel_opt: 6223 return bpf_get_skb_set_tunnel_proto(func_id); 6224 case BPF_FUNC_redirect: 6225 return &bpf_redirect_proto; 6226 case BPF_FUNC_get_route_realm: 6227 return &bpf_get_route_realm_proto; 6228 case BPF_FUNC_get_hash_recalc: 6229 return &bpf_get_hash_recalc_proto; 6230 case BPF_FUNC_set_hash_invalid: 6231 return &bpf_set_hash_invalid_proto; 6232 case BPF_FUNC_set_hash: 6233 return &bpf_set_hash_proto; 6234 case BPF_FUNC_perf_event_output: 6235 return &bpf_skb_event_output_proto; 6236 case BPF_FUNC_get_smp_processor_id: 6237 return &bpf_get_smp_processor_id_proto; 6238 case BPF_FUNC_skb_under_cgroup: 6239 return &bpf_skb_under_cgroup_proto; 6240 case BPF_FUNC_get_socket_cookie: 6241 return &bpf_get_socket_cookie_proto; 6242 case BPF_FUNC_get_socket_uid: 6243 return &bpf_get_socket_uid_proto; 6244 case BPF_FUNC_fib_lookup: 6245 return &bpf_skb_fib_lookup_proto; 6246 case BPF_FUNC_sk_fullsock: 6247 return &bpf_sk_fullsock_proto; 6248 case BPF_FUNC_sk_storage_get: 6249 return &bpf_sk_storage_get_proto; 6250 case BPF_FUNC_sk_storage_delete: 6251 return &bpf_sk_storage_delete_proto; 6252 #ifdef CONFIG_XFRM 6253 case BPF_FUNC_skb_get_xfrm_state: 6254 return &bpf_skb_get_xfrm_state_proto; 6255 #endif 6256 #ifdef CONFIG_SOCK_CGROUP_DATA 6257 case BPF_FUNC_skb_cgroup_id: 6258 return &bpf_skb_cgroup_id_proto; 6259 case BPF_FUNC_skb_ancestor_cgroup_id: 6260 return &bpf_skb_ancestor_cgroup_id_proto; 6261 #endif 6262 #ifdef CONFIG_INET 6263 case BPF_FUNC_sk_lookup_tcp: 6264 return &bpf_sk_lookup_tcp_proto; 6265 case BPF_FUNC_sk_lookup_udp: 6266 return &bpf_sk_lookup_udp_proto; 6267 case BPF_FUNC_sk_release: 6268 return &bpf_sk_release_proto; 6269 case BPF_FUNC_tcp_sock: 6270 return &bpf_tcp_sock_proto; 6271 case BPF_FUNC_get_listener_sock: 6272 return &bpf_get_listener_sock_proto; 6273 case BPF_FUNC_skc_lookup_tcp: 6274 return &bpf_skc_lookup_tcp_proto; 6275 case BPF_FUNC_tcp_check_syncookie: 6276 return &bpf_tcp_check_syncookie_proto; 6277 case BPF_FUNC_skb_ecn_set_ce: 6278 return &bpf_skb_ecn_set_ce_proto; 6279 case BPF_FUNC_tcp_gen_syncookie: 6280 return &bpf_tcp_gen_syncookie_proto; 6281 case BPF_FUNC_sk_assign: 6282 return &bpf_sk_assign_proto; 6283 #endif 6284 default: 6285 return bpf_base_func_proto(func_id); 6286 } 6287 } 6288 6289 static const struct bpf_func_proto * 6290 xdp_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6291 { 6292 switch (func_id) { 6293 case BPF_FUNC_perf_event_output: 6294 return &bpf_xdp_event_output_proto; 6295 case BPF_FUNC_get_smp_processor_id: 6296 return &bpf_get_smp_processor_id_proto; 6297 case BPF_FUNC_csum_diff: 6298 return &bpf_csum_diff_proto; 6299 case BPF_FUNC_xdp_adjust_head: 6300 return &bpf_xdp_adjust_head_proto; 6301 case BPF_FUNC_xdp_adjust_meta: 6302 return &bpf_xdp_adjust_meta_proto; 6303 case BPF_FUNC_redirect: 6304 return &bpf_xdp_redirect_proto; 6305 case BPF_FUNC_redirect_map: 6306 return &bpf_xdp_redirect_map_proto; 6307 case BPF_FUNC_xdp_adjust_tail: 6308 return &bpf_xdp_adjust_tail_proto; 6309 case BPF_FUNC_fib_lookup: 6310 return &bpf_xdp_fib_lookup_proto; 6311 #ifdef CONFIG_INET 6312 case BPF_FUNC_sk_lookup_udp: 6313 return &bpf_xdp_sk_lookup_udp_proto; 6314 case BPF_FUNC_sk_lookup_tcp: 6315 return &bpf_xdp_sk_lookup_tcp_proto; 6316 case BPF_FUNC_sk_release: 6317 return &bpf_sk_release_proto; 6318 case BPF_FUNC_skc_lookup_tcp: 6319 return &bpf_xdp_skc_lookup_tcp_proto; 6320 case BPF_FUNC_tcp_check_syncookie: 6321 return &bpf_tcp_check_syncookie_proto; 6322 case BPF_FUNC_tcp_gen_syncookie: 6323 return &bpf_tcp_gen_syncookie_proto; 6324 #endif 6325 default: 6326 return bpf_base_func_proto(func_id); 6327 } 6328 } 6329 6330 const struct bpf_func_proto bpf_sock_map_update_proto __weak; 6331 const struct bpf_func_proto bpf_sock_hash_update_proto __weak; 6332 6333 static const struct bpf_func_proto * 6334 sock_ops_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6335 { 6336 switch (func_id) { 6337 case BPF_FUNC_setsockopt: 6338 return &bpf_setsockopt_proto; 6339 case BPF_FUNC_getsockopt: 6340 return &bpf_getsockopt_proto; 6341 case BPF_FUNC_sock_ops_cb_flags_set: 6342 return &bpf_sock_ops_cb_flags_set_proto; 6343 case BPF_FUNC_sock_map_update: 6344 return &bpf_sock_map_update_proto; 6345 case BPF_FUNC_sock_hash_update: 6346 return &bpf_sock_hash_update_proto; 6347 case BPF_FUNC_get_socket_cookie: 6348 return &bpf_get_socket_cookie_sock_ops_proto; 6349 case BPF_FUNC_get_local_storage: 6350 return &bpf_get_local_storage_proto; 6351 case BPF_FUNC_perf_event_output: 6352 return &bpf_event_output_data_proto; 6353 case BPF_FUNC_sk_storage_get: 6354 return &bpf_sk_storage_get_proto; 6355 case BPF_FUNC_sk_storage_delete: 6356 return &bpf_sk_storage_delete_proto; 6357 #ifdef CONFIG_INET 6358 case BPF_FUNC_tcp_sock: 6359 return &bpf_tcp_sock_proto; 6360 #endif /* CONFIG_INET */ 6361 default: 6362 return bpf_base_func_proto(func_id); 6363 } 6364 } 6365 6366 const struct bpf_func_proto bpf_msg_redirect_map_proto __weak; 6367 const struct bpf_func_proto bpf_msg_redirect_hash_proto __weak; 6368 6369 static const struct bpf_func_proto * 6370 sk_msg_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6371 { 6372 switch (func_id) { 6373 case BPF_FUNC_msg_redirect_map: 6374 return &bpf_msg_redirect_map_proto; 6375 case BPF_FUNC_msg_redirect_hash: 6376 return &bpf_msg_redirect_hash_proto; 6377 case BPF_FUNC_msg_apply_bytes: 6378 return &bpf_msg_apply_bytes_proto; 6379 case BPF_FUNC_msg_cork_bytes: 6380 return &bpf_msg_cork_bytes_proto; 6381 case BPF_FUNC_msg_pull_data: 6382 return &bpf_msg_pull_data_proto; 6383 case BPF_FUNC_msg_push_data: 6384 return &bpf_msg_push_data_proto; 6385 case BPF_FUNC_msg_pop_data: 6386 return &bpf_msg_pop_data_proto; 6387 default: 6388 return bpf_base_func_proto(func_id); 6389 } 6390 } 6391 6392 const struct bpf_func_proto bpf_sk_redirect_map_proto __weak; 6393 const struct bpf_func_proto bpf_sk_redirect_hash_proto __weak; 6394 6395 static const struct bpf_func_proto * 6396 sk_skb_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6397 { 6398 switch (func_id) { 6399 case BPF_FUNC_skb_store_bytes: 6400 return &bpf_skb_store_bytes_proto; 6401 case BPF_FUNC_skb_load_bytes: 6402 return &bpf_skb_load_bytes_proto; 6403 case BPF_FUNC_skb_pull_data: 6404 return &sk_skb_pull_data_proto; 6405 case BPF_FUNC_skb_change_tail: 6406 return &sk_skb_change_tail_proto; 6407 case BPF_FUNC_skb_change_head: 6408 return &sk_skb_change_head_proto; 6409 case BPF_FUNC_get_socket_cookie: 6410 return &bpf_get_socket_cookie_proto; 6411 case BPF_FUNC_get_socket_uid: 6412 return &bpf_get_socket_uid_proto; 6413 case BPF_FUNC_sk_redirect_map: 6414 return &bpf_sk_redirect_map_proto; 6415 case BPF_FUNC_sk_redirect_hash: 6416 return &bpf_sk_redirect_hash_proto; 6417 case BPF_FUNC_perf_event_output: 6418 return &bpf_skb_event_output_proto; 6419 #ifdef CONFIG_INET 6420 case BPF_FUNC_sk_lookup_tcp: 6421 return &bpf_sk_lookup_tcp_proto; 6422 case BPF_FUNC_sk_lookup_udp: 6423 return &bpf_sk_lookup_udp_proto; 6424 case BPF_FUNC_sk_release: 6425 return &bpf_sk_release_proto; 6426 case BPF_FUNC_skc_lookup_tcp: 6427 return &bpf_skc_lookup_tcp_proto; 6428 #endif 6429 default: 6430 return bpf_base_func_proto(func_id); 6431 } 6432 } 6433 6434 static const struct bpf_func_proto * 6435 flow_dissector_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6436 { 6437 switch (func_id) { 6438 case BPF_FUNC_skb_load_bytes: 6439 return &bpf_flow_dissector_load_bytes_proto; 6440 default: 6441 return bpf_base_func_proto(func_id); 6442 } 6443 } 6444 6445 static const struct bpf_func_proto * 6446 lwt_out_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6447 { 6448 switch (func_id) { 6449 case BPF_FUNC_skb_load_bytes: 6450 return &bpf_skb_load_bytes_proto; 6451 case BPF_FUNC_skb_pull_data: 6452 return &bpf_skb_pull_data_proto; 6453 case BPF_FUNC_csum_diff: 6454 return &bpf_csum_diff_proto; 6455 case BPF_FUNC_get_cgroup_classid: 6456 return &bpf_get_cgroup_classid_proto; 6457 case BPF_FUNC_get_route_realm: 6458 return &bpf_get_route_realm_proto; 6459 case BPF_FUNC_get_hash_recalc: 6460 return &bpf_get_hash_recalc_proto; 6461 case BPF_FUNC_perf_event_output: 6462 return &bpf_skb_event_output_proto; 6463 case BPF_FUNC_get_smp_processor_id: 6464 return &bpf_get_smp_processor_id_proto; 6465 case BPF_FUNC_skb_under_cgroup: 6466 return &bpf_skb_under_cgroup_proto; 6467 default: 6468 return bpf_base_func_proto(func_id); 6469 } 6470 } 6471 6472 static const struct bpf_func_proto * 6473 lwt_in_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6474 { 6475 switch (func_id) { 6476 case BPF_FUNC_lwt_push_encap: 6477 return &bpf_lwt_in_push_encap_proto; 6478 default: 6479 return lwt_out_func_proto(func_id, prog); 6480 } 6481 } 6482 6483 static const struct bpf_func_proto * 6484 lwt_xmit_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6485 { 6486 switch (func_id) { 6487 case BPF_FUNC_skb_get_tunnel_key: 6488 return &bpf_skb_get_tunnel_key_proto; 6489 case BPF_FUNC_skb_set_tunnel_key: 6490 return bpf_get_skb_set_tunnel_proto(func_id); 6491 case BPF_FUNC_skb_get_tunnel_opt: 6492 return &bpf_skb_get_tunnel_opt_proto; 6493 case BPF_FUNC_skb_set_tunnel_opt: 6494 return bpf_get_skb_set_tunnel_proto(func_id); 6495 case BPF_FUNC_redirect: 6496 return &bpf_redirect_proto; 6497 case BPF_FUNC_clone_redirect: 6498 return &bpf_clone_redirect_proto; 6499 case BPF_FUNC_skb_change_tail: 6500 return &bpf_skb_change_tail_proto; 6501 case BPF_FUNC_skb_change_head: 6502 return &bpf_skb_change_head_proto; 6503 case BPF_FUNC_skb_store_bytes: 6504 return &bpf_skb_store_bytes_proto; 6505 case BPF_FUNC_csum_update: 6506 return &bpf_csum_update_proto; 6507 case BPF_FUNC_l3_csum_replace: 6508 return &bpf_l3_csum_replace_proto; 6509 case BPF_FUNC_l4_csum_replace: 6510 return &bpf_l4_csum_replace_proto; 6511 case BPF_FUNC_set_hash_invalid: 6512 return &bpf_set_hash_invalid_proto; 6513 case BPF_FUNC_lwt_push_encap: 6514 return &bpf_lwt_xmit_push_encap_proto; 6515 default: 6516 return lwt_out_func_proto(func_id, prog); 6517 } 6518 } 6519 6520 static const struct bpf_func_proto * 6521 lwt_seg6local_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 6522 { 6523 switch (func_id) { 6524 #if IS_ENABLED(CONFIG_IPV6_SEG6_BPF) 6525 case BPF_FUNC_lwt_seg6_store_bytes: 6526 return &bpf_lwt_seg6_store_bytes_proto; 6527 case BPF_FUNC_lwt_seg6_action: 6528 return &bpf_lwt_seg6_action_proto; 6529 case BPF_FUNC_lwt_seg6_adjust_srh: 6530 return &bpf_lwt_seg6_adjust_srh_proto; 6531 #endif 6532 default: 6533 return lwt_out_func_proto(func_id, prog); 6534 } 6535 } 6536 6537 static bool bpf_skb_is_valid_access(int off, int size, enum bpf_access_type type, 6538 const struct bpf_prog *prog, 6539 struct bpf_insn_access_aux *info) 6540 { 6541 const int size_default = sizeof(__u32); 6542 6543 if (off < 0 || off >= sizeof(struct __sk_buff)) 6544 return false; 6545 6546 /* The verifier guarantees that size > 0. */ 6547 if (off % size != 0) 6548 return false; 6549 6550 switch (off) { 6551 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 6552 if (off + size > offsetofend(struct __sk_buff, cb[4])) 6553 return false; 6554 break; 6555 case bpf_ctx_range_till(struct __sk_buff, remote_ip6[0], remote_ip6[3]): 6556 case bpf_ctx_range_till(struct __sk_buff, local_ip6[0], local_ip6[3]): 6557 case bpf_ctx_range_till(struct __sk_buff, remote_ip4, remote_ip4): 6558 case bpf_ctx_range_till(struct __sk_buff, local_ip4, local_ip4): 6559 case bpf_ctx_range(struct __sk_buff, data): 6560 case bpf_ctx_range(struct __sk_buff, data_meta): 6561 case bpf_ctx_range(struct __sk_buff, data_end): 6562 if (size != size_default) 6563 return false; 6564 break; 6565 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 6566 return false; 6567 case bpf_ctx_range(struct __sk_buff, tstamp): 6568 if (size != sizeof(__u64)) 6569 return false; 6570 break; 6571 case offsetof(struct __sk_buff, sk): 6572 if (type == BPF_WRITE || size != sizeof(__u64)) 6573 return false; 6574 info->reg_type = PTR_TO_SOCK_COMMON_OR_NULL; 6575 break; 6576 default: 6577 /* Only narrow read access allowed for now. */ 6578 if (type == BPF_WRITE) { 6579 if (size != size_default) 6580 return false; 6581 } else { 6582 bpf_ctx_record_field_size(info, size_default); 6583 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 6584 return false; 6585 } 6586 } 6587 6588 return true; 6589 } 6590 6591 static bool sk_filter_is_valid_access(int off, int size, 6592 enum bpf_access_type type, 6593 const struct bpf_prog *prog, 6594 struct bpf_insn_access_aux *info) 6595 { 6596 switch (off) { 6597 case bpf_ctx_range(struct __sk_buff, tc_classid): 6598 case bpf_ctx_range(struct __sk_buff, data): 6599 case bpf_ctx_range(struct __sk_buff, data_meta): 6600 case bpf_ctx_range(struct __sk_buff, data_end): 6601 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 6602 case bpf_ctx_range(struct __sk_buff, tstamp): 6603 case bpf_ctx_range(struct __sk_buff, wire_len): 6604 return false; 6605 } 6606 6607 if (type == BPF_WRITE) { 6608 switch (off) { 6609 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 6610 break; 6611 default: 6612 return false; 6613 } 6614 } 6615 6616 return bpf_skb_is_valid_access(off, size, type, prog, info); 6617 } 6618 6619 static bool cg_skb_is_valid_access(int off, int size, 6620 enum bpf_access_type type, 6621 const struct bpf_prog *prog, 6622 struct bpf_insn_access_aux *info) 6623 { 6624 switch (off) { 6625 case bpf_ctx_range(struct __sk_buff, tc_classid): 6626 case bpf_ctx_range(struct __sk_buff, data_meta): 6627 case bpf_ctx_range(struct __sk_buff, wire_len): 6628 return false; 6629 case bpf_ctx_range(struct __sk_buff, data): 6630 case bpf_ctx_range(struct __sk_buff, data_end): 6631 if (!capable(CAP_SYS_ADMIN)) 6632 return false; 6633 break; 6634 } 6635 6636 if (type == BPF_WRITE) { 6637 switch (off) { 6638 case bpf_ctx_range(struct __sk_buff, mark): 6639 case bpf_ctx_range(struct __sk_buff, priority): 6640 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 6641 break; 6642 case bpf_ctx_range(struct __sk_buff, tstamp): 6643 if (!capable(CAP_SYS_ADMIN)) 6644 return false; 6645 break; 6646 default: 6647 return false; 6648 } 6649 } 6650 6651 switch (off) { 6652 case bpf_ctx_range(struct __sk_buff, data): 6653 info->reg_type = PTR_TO_PACKET; 6654 break; 6655 case bpf_ctx_range(struct __sk_buff, data_end): 6656 info->reg_type = PTR_TO_PACKET_END; 6657 break; 6658 } 6659 6660 return bpf_skb_is_valid_access(off, size, type, prog, info); 6661 } 6662 6663 static bool lwt_is_valid_access(int off, int size, 6664 enum bpf_access_type type, 6665 const struct bpf_prog *prog, 6666 struct bpf_insn_access_aux *info) 6667 { 6668 switch (off) { 6669 case bpf_ctx_range(struct __sk_buff, tc_classid): 6670 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 6671 case bpf_ctx_range(struct __sk_buff, data_meta): 6672 case bpf_ctx_range(struct __sk_buff, tstamp): 6673 case bpf_ctx_range(struct __sk_buff, wire_len): 6674 return false; 6675 } 6676 6677 if (type == BPF_WRITE) { 6678 switch (off) { 6679 case bpf_ctx_range(struct __sk_buff, mark): 6680 case bpf_ctx_range(struct __sk_buff, priority): 6681 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 6682 break; 6683 default: 6684 return false; 6685 } 6686 } 6687 6688 switch (off) { 6689 case bpf_ctx_range(struct __sk_buff, data): 6690 info->reg_type = PTR_TO_PACKET; 6691 break; 6692 case bpf_ctx_range(struct __sk_buff, data_end): 6693 info->reg_type = PTR_TO_PACKET_END; 6694 break; 6695 } 6696 6697 return bpf_skb_is_valid_access(off, size, type, prog, info); 6698 } 6699 6700 /* Attach type specific accesses */ 6701 static bool __sock_filter_check_attach_type(int off, 6702 enum bpf_access_type access_type, 6703 enum bpf_attach_type attach_type) 6704 { 6705 switch (off) { 6706 case offsetof(struct bpf_sock, bound_dev_if): 6707 case offsetof(struct bpf_sock, mark): 6708 case offsetof(struct bpf_sock, priority): 6709 switch (attach_type) { 6710 case BPF_CGROUP_INET_SOCK_CREATE: 6711 goto full_access; 6712 default: 6713 return false; 6714 } 6715 case bpf_ctx_range(struct bpf_sock, src_ip4): 6716 switch (attach_type) { 6717 case BPF_CGROUP_INET4_POST_BIND: 6718 goto read_only; 6719 default: 6720 return false; 6721 } 6722 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 6723 switch (attach_type) { 6724 case BPF_CGROUP_INET6_POST_BIND: 6725 goto read_only; 6726 default: 6727 return false; 6728 } 6729 case bpf_ctx_range(struct bpf_sock, src_port): 6730 switch (attach_type) { 6731 case BPF_CGROUP_INET4_POST_BIND: 6732 case BPF_CGROUP_INET6_POST_BIND: 6733 goto read_only; 6734 default: 6735 return false; 6736 } 6737 } 6738 read_only: 6739 return access_type == BPF_READ; 6740 full_access: 6741 return true; 6742 } 6743 6744 bool bpf_sock_common_is_valid_access(int off, int size, 6745 enum bpf_access_type type, 6746 struct bpf_insn_access_aux *info) 6747 { 6748 switch (off) { 6749 case bpf_ctx_range_till(struct bpf_sock, type, priority): 6750 return false; 6751 default: 6752 return bpf_sock_is_valid_access(off, size, type, info); 6753 } 6754 } 6755 6756 bool bpf_sock_is_valid_access(int off, int size, enum bpf_access_type type, 6757 struct bpf_insn_access_aux *info) 6758 { 6759 const int size_default = sizeof(__u32); 6760 6761 if (off < 0 || off >= sizeof(struct bpf_sock)) 6762 return false; 6763 if (off % size != 0) 6764 return false; 6765 6766 switch (off) { 6767 case offsetof(struct bpf_sock, state): 6768 case offsetof(struct bpf_sock, family): 6769 case offsetof(struct bpf_sock, type): 6770 case offsetof(struct bpf_sock, protocol): 6771 case offsetof(struct bpf_sock, dst_port): 6772 case offsetof(struct bpf_sock, src_port): 6773 case bpf_ctx_range(struct bpf_sock, src_ip4): 6774 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 6775 case bpf_ctx_range(struct bpf_sock, dst_ip4): 6776 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 6777 bpf_ctx_record_field_size(info, size_default); 6778 return bpf_ctx_narrow_access_ok(off, size, size_default); 6779 } 6780 6781 return size == size_default; 6782 } 6783 6784 static bool sock_filter_is_valid_access(int off, int size, 6785 enum bpf_access_type type, 6786 const struct bpf_prog *prog, 6787 struct bpf_insn_access_aux *info) 6788 { 6789 if (!bpf_sock_is_valid_access(off, size, type, info)) 6790 return false; 6791 return __sock_filter_check_attach_type(off, type, 6792 prog->expected_attach_type); 6793 } 6794 6795 static int bpf_noop_prologue(struct bpf_insn *insn_buf, bool direct_write, 6796 const struct bpf_prog *prog) 6797 { 6798 /* Neither direct read nor direct write requires any preliminary 6799 * action. 6800 */ 6801 return 0; 6802 } 6803 6804 static int bpf_unclone_prologue(struct bpf_insn *insn_buf, bool direct_write, 6805 const struct bpf_prog *prog, int drop_verdict) 6806 { 6807 struct bpf_insn *insn = insn_buf; 6808 6809 if (!direct_write) 6810 return 0; 6811 6812 /* if (!skb->cloned) 6813 * goto start; 6814 * 6815 * (Fast-path, otherwise approximation that we might be 6816 * a clone, do the rest in helper.) 6817 */ 6818 *insn++ = BPF_LDX_MEM(BPF_B, BPF_REG_6, BPF_REG_1, CLONED_OFFSET()); 6819 *insn++ = BPF_ALU32_IMM(BPF_AND, BPF_REG_6, CLONED_MASK); 6820 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_6, 0, 7); 6821 6822 /* ret = bpf_skb_pull_data(skb, 0); */ 6823 *insn++ = BPF_MOV64_REG(BPF_REG_6, BPF_REG_1); 6824 *insn++ = BPF_ALU64_REG(BPF_XOR, BPF_REG_2, BPF_REG_2); 6825 *insn++ = BPF_RAW_INSN(BPF_JMP | BPF_CALL, 0, 0, 0, 6826 BPF_FUNC_skb_pull_data); 6827 /* if (!ret) 6828 * goto restore; 6829 * return TC_ACT_SHOT; 6830 */ 6831 *insn++ = BPF_JMP_IMM(BPF_JEQ, BPF_REG_0, 0, 2); 6832 *insn++ = BPF_ALU32_IMM(BPF_MOV, BPF_REG_0, drop_verdict); 6833 *insn++ = BPF_EXIT_INSN(); 6834 6835 /* restore: */ 6836 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_6); 6837 /* start: */ 6838 *insn++ = prog->insnsi[0]; 6839 6840 return insn - insn_buf; 6841 } 6842 6843 static int bpf_gen_ld_abs(const struct bpf_insn *orig, 6844 struct bpf_insn *insn_buf) 6845 { 6846 bool indirect = BPF_MODE(orig->code) == BPF_IND; 6847 struct bpf_insn *insn = insn_buf; 6848 6849 /* We're guaranteed here that CTX is in R6. */ 6850 *insn++ = BPF_MOV64_REG(BPF_REG_1, BPF_REG_CTX); 6851 if (!indirect) { 6852 *insn++ = BPF_MOV64_IMM(BPF_REG_2, orig->imm); 6853 } else { 6854 *insn++ = BPF_MOV64_REG(BPF_REG_2, orig->src_reg); 6855 if (orig->imm) 6856 *insn++ = BPF_ALU64_IMM(BPF_ADD, BPF_REG_2, orig->imm); 6857 } 6858 6859 switch (BPF_SIZE(orig->code)) { 6860 case BPF_B: 6861 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_8_no_cache); 6862 break; 6863 case BPF_H: 6864 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_16_no_cache); 6865 break; 6866 case BPF_W: 6867 *insn++ = BPF_EMIT_CALL(bpf_skb_load_helper_32_no_cache); 6868 break; 6869 } 6870 6871 *insn++ = BPF_JMP_IMM(BPF_JSGE, BPF_REG_0, 0, 2); 6872 *insn++ = BPF_ALU32_REG(BPF_XOR, BPF_REG_0, BPF_REG_0); 6873 *insn++ = BPF_EXIT_INSN(); 6874 6875 return insn - insn_buf; 6876 } 6877 6878 static int tc_cls_act_prologue(struct bpf_insn *insn_buf, bool direct_write, 6879 const struct bpf_prog *prog) 6880 { 6881 return bpf_unclone_prologue(insn_buf, direct_write, prog, TC_ACT_SHOT); 6882 } 6883 6884 static bool tc_cls_act_is_valid_access(int off, int size, 6885 enum bpf_access_type type, 6886 const struct bpf_prog *prog, 6887 struct bpf_insn_access_aux *info) 6888 { 6889 if (type == BPF_WRITE) { 6890 switch (off) { 6891 case bpf_ctx_range(struct __sk_buff, mark): 6892 case bpf_ctx_range(struct __sk_buff, tc_index): 6893 case bpf_ctx_range(struct __sk_buff, priority): 6894 case bpf_ctx_range(struct __sk_buff, tc_classid): 6895 case bpf_ctx_range_till(struct __sk_buff, cb[0], cb[4]): 6896 case bpf_ctx_range(struct __sk_buff, tstamp): 6897 case bpf_ctx_range(struct __sk_buff, queue_mapping): 6898 break; 6899 default: 6900 return false; 6901 } 6902 } 6903 6904 switch (off) { 6905 case bpf_ctx_range(struct __sk_buff, data): 6906 info->reg_type = PTR_TO_PACKET; 6907 break; 6908 case bpf_ctx_range(struct __sk_buff, data_meta): 6909 info->reg_type = PTR_TO_PACKET_META; 6910 break; 6911 case bpf_ctx_range(struct __sk_buff, data_end): 6912 info->reg_type = PTR_TO_PACKET_END; 6913 break; 6914 case bpf_ctx_range_till(struct __sk_buff, family, local_port): 6915 return false; 6916 } 6917 6918 return bpf_skb_is_valid_access(off, size, type, prog, info); 6919 } 6920 6921 static bool __is_valid_xdp_access(int off, int size) 6922 { 6923 if (off < 0 || off >= sizeof(struct xdp_md)) 6924 return false; 6925 if (off % size != 0) 6926 return false; 6927 if (size != sizeof(__u32)) 6928 return false; 6929 6930 return true; 6931 } 6932 6933 static bool xdp_is_valid_access(int off, int size, 6934 enum bpf_access_type type, 6935 const struct bpf_prog *prog, 6936 struct bpf_insn_access_aux *info) 6937 { 6938 if (type == BPF_WRITE) { 6939 if (bpf_prog_is_dev_bound(prog->aux)) { 6940 switch (off) { 6941 case offsetof(struct xdp_md, rx_queue_index): 6942 return __is_valid_xdp_access(off, size); 6943 } 6944 } 6945 return false; 6946 } 6947 6948 switch (off) { 6949 case offsetof(struct xdp_md, data): 6950 info->reg_type = PTR_TO_PACKET; 6951 break; 6952 case offsetof(struct xdp_md, data_meta): 6953 info->reg_type = PTR_TO_PACKET_META; 6954 break; 6955 case offsetof(struct xdp_md, data_end): 6956 info->reg_type = PTR_TO_PACKET_END; 6957 break; 6958 } 6959 6960 return __is_valid_xdp_access(off, size); 6961 } 6962 6963 void bpf_warn_invalid_xdp_action(u32 act) 6964 { 6965 const u32 act_max = XDP_REDIRECT; 6966 6967 WARN_ONCE(1, "%s XDP return value %u, expect packet loss!\n", 6968 act > act_max ? "Illegal" : "Driver unsupported", 6969 act); 6970 } 6971 EXPORT_SYMBOL_GPL(bpf_warn_invalid_xdp_action); 6972 6973 static bool sock_addr_is_valid_access(int off, int size, 6974 enum bpf_access_type type, 6975 const struct bpf_prog *prog, 6976 struct bpf_insn_access_aux *info) 6977 { 6978 const int size_default = sizeof(__u32); 6979 6980 if (off < 0 || off >= sizeof(struct bpf_sock_addr)) 6981 return false; 6982 if (off % size != 0) 6983 return false; 6984 6985 /* Disallow access to IPv6 fields from IPv4 contex and vise 6986 * versa. 6987 */ 6988 switch (off) { 6989 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 6990 switch (prog->expected_attach_type) { 6991 case BPF_CGROUP_INET4_BIND: 6992 case BPF_CGROUP_INET4_CONNECT: 6993 case BPF_CGROUP_UDP4_SENDMSG: 6994 case BPF_CGROUP_UDP4_RECVMSG: 6995 break; 6996 default: 6997 return false; 6998 } 6999 break; 7000 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 7001 switch (prog->expected_attach_type) { 7002 case BPF_CGROUP_INET6_BIND: 7003 case BPF_CGROUP_INET6_CONNECT: 7004 case BPF_CGROUP_UDP6_SENDMSG: 7005 case BPF_CGROUP_UDP6_RECVMSG: 7006 break; 7007 default: 7008 return false; 7009 } 7010 break; 7011 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 7012 switch (prog->expected_attach_type) { 7013 case BPF_CGROUP_UDP4_SENDMSG: 7014 break; 7015 default: 7016 return false; 7017 } 7018 break; 7019 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 7020 msg_src_ip6[3]): 7021 switch (prog->expected_attach_type) { 7022 case BPF_CGROUP_UDP6_SENDMSG: 7023 break; 7024 default: 7025 return false; 7026 } 7027 break; 7028 } 7029 7030 switch (off) { 7031 case bpf_ctx_range(struct bpf_sock_addr, user_ip4): 7032 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 7033 case bpf_ctx_range(struct bpf_sock_addr, msg_src_ip4): 7034 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 7035 msg_src_ip6[3]): 7036 if (type == BPF_READ) { 7037 bpf_ctx_record_field_size(info, size_default); 7038 7039 if (bpf_ctx_wide_access_ok(off, size, 7040 struct bpf_sock_addr, 7041 user_ip6)) 7042 return true; 7043 7044 if (bpf_ctx_wide_access_ok(off, size, 7045 struct bpf_sock_addr, 7046 msg_src_ip6)) 7047 return true; 7048 7049 if (!bpf_ctx_narrow_access_ok(off, size, size_default)) 7050 return false; 7051 } else { 7052 if (bpf_ctx_wide_access_ok(off, size, 7053 struct bpf_sock_addr, 7054 user_ip6)) 7055 return true; 7056 7057 if (bpf_ctx_wide_access_ok(off, size, 7058 struct bpf_sock_addr, 7059 msg_src_ip6)) 7060 return true; 7061 7062 if (size != size_default) 7063 return false; 7064 } 7065 break; 7066 case bpf_ctx_range(struct bpf_sock_addr, user_port): 7067 if (size != size_default) 7068 return false; 7069 break; 7070 case offsetof(struct bpf_sock_addr, sk): 7071 if (type != BPF_READ) 7072 return false; 7073 if (size != sizeof(__u64)) 7074 return false; 7075 info->reg_type = PTR_TO_SOCKET; 7076 break; 7077 default: 7078 if (type == BPF_READ) { 7079 if (size != size_default) 7080 return false; 7081 } else { 7082 return false; 7083 } 7084 } 7085 7086 return true; 7087 } 7088 7089 static bool sock_ops_is_valid_access(int off, int size, 7090 enum bpf_access_type type, 7091 const struct bpf_prog *prog, 7092 struct bpf_insn_access_aux *info) 7093 { 7094 const int size_default = sizeof(__u32); 7095 7096 if (off < 0 || off >= sizeof(struct bpf_sock_ops)) 7097 return false; 7098 7099 /* The verifier guarantees that size > 0. */ 7100 if (off % size != 0) 7101 return false; 7102 7103 if (type == BPF_WRITE) { 7104 switch (off) { 7105 case offsetof(struct bpf_sock_ops, reply): 7106 case offsetof(struct bpf_sock_ops, sk_txhash): 7107 if (size != size_default) 7108 return false; 7109 break; 7110 default: 7111 return false; 7112 } 7113 } else { 7114 switch (off) { 7115 case bpf_ctx_range_till(struct bpf_sock_ops, bytes_received, 7116 bytes_acked): 7117 if (size != sizeof(__u64)) 7118 return false; 7119 break; 7120 case offsetof(struct bpf_sock_ops, sk): 7121 if (size != sizeof(__u64)) 7122 return false; 7123 info->reg_type = PTR_TO_SOCKET_OR_NULL; 7124 break; 7125 default: 7126 if (size != size_default) 7127 return false; 7128 break; 7129 } 7130 } 7131 7132 return true; 7133 } 7134 7135 static int sk_skb_prologue(struct bpf_insn *insn_buf, bool direct_write, 7136 const struct bpf_prog *prog) 7137 { 7138 return bpf_unclone_prologue(insn_buf, direct_write, prog, SK_DROP); 7139 } 7140 7141 static bool sk_skb_is_valid_access(int off, int size, 7142 enum bpf_access_type type, 7143 const struct bpf_prog *prog, 7144 struct bpf_insn_access_aux *info) 7145 { 7146 switch (off) { 7147 case bpf_ctx_range(struct __sk_buff, tc_classid): 7148 case bpf_ctx_range(struct __sk_buff, data_meta): 7149 case bpf_ctx_range(struct __sk_buff, tstamp): 7150 case bpf_ctx_range(struct __sk_buff, wire_len): 7151 return false; 7152 } 7153 7154 if (type == BPF_WRITE) { 7155 switch (off) { 7156 case bpf_ctx_range(struct __sk_buff, tc_index): 7157 case bpf_ctx_range(struct __sk_buff, priority): 7158 break; 7159 default: 7160 return false; 7161 } 7162 } 7163 7164 switch (off) { 7165 case bpf_ctx_range(struct __sk_buff, mark): 7166 return false; 7167 case bpf_ctx_range(struct __sk_buff, data): 7168 info->reg_type = PTR_TO_PACKET; 7169 break; 7170 case bpf_ctx_range(struct __sk_buff, data_end): 7171 info->reg_type = PTR_TO_PACKET_END; 7172 break; 7173 } 7174 7175 return bpf_skb_is_valid_access(off, size, type, prog, info); 7176 } 7177 7178 static bool sk_msg_is_valid_access(int off, int size, 7179 enum bpf_access_type type, 7180 const struct bpf_prog *prog, 7181 struct bpf_insn_access_aux *info) 7182 { 7183 if (type == BPF_WRITE) 7184 return false; 7185 7186 if (off % size != 0) 7187 return false; 7188 7189 switch (off) { 7190 case offsetof(struct sk_msg_md, data): 7191 info->reg_type = PTR_TO_PACKET; 7192 if (size != sizeof(__u64)) 7193 return false; 7194 break; 7195 case offsetof(struct sk_msg_md, data_end): 7196 info->reg_type = PTR_TO_PACKET_END; 7197 if (size != sizeof(__u64)) 7198 return false; 7199 break; 7200 case bpf_ctx_range(struct sk_msg_md, family): 7201 case bpf_ctx_range(struct sk_msg_md, remote_ip4): 7202 case bpf_ctx_range(struct sk_msg_md, local_ip4): 7203 case bpf_ctx_range_till(struct sk_msg_md, remote_ip6[0], remote_ip6[3]): 7204 case bpf_ctx_range_till(struct sk_msg_md, local_ip6[0], local_ip6[3]): 7205 case bpf_ctx_range(struct sk_msg_md, remote_port): 7206 case bpf_ctx_range(struct sk_msg_md, local_port): 7207 case bpf_ctx_range(struct sk_msg_md, size): 7208 if (size != sizeof(__u32)) 7209 return false; 7210 break; 7211 default: 7212 return false; 7213 } 7214 return true; 7215 } 7216 7217 static bool flow_dissector_is_valid_access(int off, int size, 7218 enum bpf_access_type type, 7219 const struct bpf_prog *prog, 7220 struct bpf_insn_access_aux *info) 7221 { 7222 const int size_default = sizeof(__u32); 7223 7224 if (off < 0 || off >= sizeof(struct __sk_buff)) 7225 return false; 7226 7227 if (type == BPF_WRITE) 7228 return false; 7229 7230 switch (off) { 7231 case bpf_ctx_range(struct __sk_buff, data): 7232 if (size != size_default) 7233 return false; 7234 info->reg_type = PTR_TO_PACKET; 7235 return true; 7236 case bpf_ctx_range(struct __sk_buff, data_end): 7237 if (size != size_default) 7238 return false; 7239 info->reg_type = PTR_TO_PACKET_END; 7240 return true; 7241 case bpf_ctx_range_ptr(struct __sk_buff, flow_keys): 7242 if (size != sizeof(__u64)) 7243 return false; 7244 info->reg_type = PTR_TO_FLOW_KEYS; 7245 return true; 7246 default: 7247 return false; 7248 } 7249 } 7250 7251 static u32 flow_dissector_convert_ctx_access(enum bpf_access_type type, 7252 const struct bpf_insn *si, 7253 struct bpf_insn *insn_buf, 7254 struct bpf_prog *prog, 7255 u32 *target_size) 7256 7257 { 7258 struct bpf_insn *insn = insn_buf; 7259 7260 switch (si->off) { 7261 case offsetof(struct __sk_buff, data): 7262 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data), 7263 si->dst_reg, si->src_reg, 7264 offsetof(struct bpf_flow_dissector, data)); 7265 break; 7266 7267 case offsetof(struct __sk_buff, data_end): 7268 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, data_end), 7269 si->dst_reg, si->src_reg, 7270 offsetof(struct bpf_flow_dissector, data_end)); 7271 break; 7272 7273 case offsetof(struct __sk_buff, flow_keys): 7274 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_flow_dissector, flow_keys), 7275 si->dst_reg, si->src_reg, 7276 offsetof(struct bpf_flow_dissector, flow_keys)); 7277 break; 7278 } 7279 7280 return insn - insn_buf; 7281 } 7282 7283 static struct bpf_insn *bpf_convert_shinfo_access(const struct bpf_insn *si, 7284 struct bpf_insn *insn) 7285 { 7286 /* si->dst_reg = skb_shinfo(SKB); */ 7287 #ifdef NET_SKBUFF_DATA_USES_OFFSET 7288 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 7289 BPF_REG_AX, si->src_reg, 7290 offsetof(struct sk_buff, end)); 7291 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, head), 7292 si->dst_reg, si->src_reg, 7293 offsetof(struct sk_buff, head)); 7294 *insn++ = BPF_ALU64_REG(BPF_ADD, si->dst_reg, BPF_REG_AX); 7295 #else 7296 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, end), 7297 si->dst_reg, si->src_reg, 7298 offsetof(struct sk_buff, end)); 7299 #endif 7300 7301 return insn; 7302 } 7303 7304 static u32 bpf_convert_ctx_access(enum bpf_access_type type, 7305 const struct bpf_insn *si, 7306 struct bpf_insn *insn_buf, 7307 struct bpf_prog *prog, u32 *target_size) 7308 { 7309 struct bpf_insn *insn = insn_buf; 7310 int off; 7311 7312 switch (si->off) { 7313 case offsetof(struct __sk_buff, len): 7314 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7315 bpf_target_off(struct sk_buff, len, 4, 7316 target_size)); 7317 break; 7318 7319 case offsetof(struct __sk_buff, protocol): 7320 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 7321 bpf_target_off(struct sk_buff, protocol, 2, 7322 target_size)); 7323 break; 7324 7325 case offsetof(struct __sk_buff, vlan_proto): 7326 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 7327 bpf_target_off(struct sk_buff, vlan_proto, 2, 7328 target_size)); 7329 break; 7330 7331 case offsetof(struct __sk_buff, priority): 7332 if (type == BPF_WRITE) 7333 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 7334 bpf_target_off(struct sk_buff, priority, 4, 7335 target_size)); 7336 else 7337 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7338 bpf_target_off(struct sk_buff, priority, 4, 7339 target_size)); 7340 break; 7341 7342 case offsetof(struct __sk_buff, ingress_ifindex): 7343 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7344 bpf_target_off(struct sk_buff, skb_iif, 4, 7345 target_size)); 7346 break; 7347 7348 case offsetof(struct __sk_buff, ifindex): 7349 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 7350 si->dst_reg, si->src_reg, 7351 offsetof(struct sk_buff, dev)); 7352 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 7353 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7354 bpf_target_off(struct net_device, ifindex, 4, 7355 target_size)); 7356 break; 7357 7358 case offsetof(struct __sk_buff, hash): 7359 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7360 bpf_target_off(struct sk_buff, hash, 4, 7361 target_size)); 7362 break; 7363 7364 case offsetof(struct __sk_buff, mark): 7365 if (type == BPF_WRITE) 7366 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 7367 bpf_target_off(struct sk_buff, mark, 4, 7368 target_size)); 7369 else 7370 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7371 bpf_target_off(struct sk_buff, mark, 4, 7372 target_size)); 7373 break; 7374 7375 case offsetof(struct __sk_buff, pkt_type): 7376 *target_size = 1; 7377 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 7378 PKT_TYPE_OFFSET()); 7379 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, PKT_TYPE_MAX); 7380 #ifdef __BIG_ENDIAN_BITFIELD 7381 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, 5); 7382 #endif 7383 break; 7384 7385 case offsetof(struct __sk_buff, queue_mapping): 7386 if (type == BPF_WRITE) { 7387 *insn++ = BPF_JMP_IMM(BPF_JGE, si->src_reg, NO_QUEUE_MAPPING, 1); 7388 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 7389 bpf_target_off(struct sk_buff, 7390 queue_mapping, 7391 2, target_size)); 7392 } else { 7393 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 7394 bpf_target_off(struct sk_buff, 7395 queue_mapping, 7396 2, target_size)); 7397 } 7398 break; 7399 7400 case offsetof(struct __sk_buff, vlan_present): 7401 *target_size = 1; 7402 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->src_reg, 7403 PKT_VLAN_PRESENT_OFFSET()); 7404 if (PKT_VLAN_PRESENT_BIT) 7405 *insn++ = BPF_ALU32_IMM(BPF_RSH, si->dst_reg, PKT_VLAN_PRESENT_BIT); 7406 if (PKT_VLAN_PRESENT_BIT < 7) 7407 *insn++ = BPF_ALU32_IMM(BPF_AND, si->dst_reg, 1); 7408 break; 7409 7410 case offsetof(struct __sk_buff, vlan_tci): 7411 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 7412 bpf_target_off(struct sk_buff, vlan_tci, 2, 7413 target_size)); 7414 break; 7415 7416 case offsetof(struct __sk_buff, cb[0]) ... 7417 offsetofend(struct __sk_buff, cb[4]) - 1: 7418 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, data) < 20); 7419 BUILD_BUG_ON((offsetof(struct sk_buff, cb) + 7420 offsetof(struct qdisc_skb_cb, data)) % 7421 sizeof(__u64)); 7422 7423 prog->cb_access = 1; 7424 off = si->off; 7425 off -= offsetof(struct __sk_buff, cb[0]); 7426 off += offsetof(struct sk_buff, cb); 7427 off += offsetof(struct qdisc_skb_cb, data); 7428 if (type == BPF_WRITE) 7429 *insn++ = BPF_STX_MEM(BPF_SIZE(si->code), si->dst_reg, 7430 si->src_reg, off); 7431 else 7432 *insn++ = BPF_LDX_MEM(BPF_SIZE(si->code), si->dst_reg, 7433 si->src_reg, off); 7434 break; 7435 7436 case offsetof(struct __sk_buff, tc_classid): 7437 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, tc_classid) != 2); 7438 7439 off = si->off; 7440 off -= offsetof(struct __sk_buff, tc_classid); 7441 off += offsetof(struct sk_buff, cb); 7442 off += offsetof(struct qdisc_skb_cb, tc_classid); 7443 *target_size = 2; 7444 if (type == BPF_WRITE) 7445 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, 7446 si->src_reg, off); 7447 else 7448 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, 7449 si->src_reg, off); 7450 break; 7451 7452 case offsetof(struct __sk_buff, data): 7453 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, data), 7454 si->dst_reg, si->src_reg, 7455 offsetof(struct sk_buff, data)); 7456 break; 7457 7458 case offsetof(struct __sk_buff, data_meta): 7459 off = si->off; 7460 off -= offsetof(struct __sk_buff, data_meta); 7461 off += offsetof(struct sk_buff, cb); 7462 off += offsetof(struct bpf_skb_data_end, data_meta); 7463 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 7464 si->src_reg, off); 7465 break; 7466 7467 case offsetof(struct __sk_buff, data_end): 7468 off = si->off; 7469 off -= offsetof(struct __sk_buff, data_end); 7470 off += offsetof(struct sk_buff, cb); 7471 off += offsetof(struct bpf_skb_data_end, data_end); 7472 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 7473 si->src_reg, off); 7474 break; 7475 7476 case offsetof(struct __sk_buff, tc_index): 7477 #ifdef CONFIG_NET_SCHED 7478 if (type == BPF_WRITE) 7479 *insn++ = BPF_STX_MEM(BPF_H, si->dst_reg, si->src_reg, 7480 bpf_target_off(struct sk_buff, tc_index, 2, 7481 target_size)); 7482 else 7483 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->src_reg, 7484 bpf_target_off(struct sk_buff, tc_index, 2, 7485 target_size)); 7486 #else 7487 *target_size = 2; 7488 if (type == BPF_WRITE) 7489 *insn++ = BPF_MOV64_REG(si->dst_reg, si->dst_reg); 7490 else 7491 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 7492 #endif 7493 break; 7494 7495 case offsetof(struct __sk_buff, napi_id): 7496 #if defined(CONFIG_NET_RX_BUSY_POLL) 7497 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7498 bpf_target_off(struct sk_buff, napi_id, 4, 7499 target_size)); 7500 *insn++ = BPF_JMP_IMM(BPF_JGE, si->dst_reg, MIN_NAPI_ID, 1); 7501 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 7502 #else 7503 *target_size = 4; 7504 *insn++ = BPF_MOV64_IMM(si->dst_reg, 0); 7505 #endif 7506 break; 7507 case offsetof(struct __sk_buff, family): 7508 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 7509 7510 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7511 si->dst_reg, si->src_reg, 7512 offsetof(struct sk_buff, sk)); 7513 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 7514 bpf_target_off(struct sock_common, 7515 skc_family, 7516 2, target_size)); 7517 break; 7518 case offsetof(struct __sk_buff, remote_ip4): 7519 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 7520 7521 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7522 si->dst_reg, si->src_reg, 7523 offsetof(struct sk_buff, sk)); 7524 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7525 bpf_target_off(struct sock_common, 7526 skc_daddr, 7527 4, target_size)); 7528 break; 7529 case offsetof(struct __sk_buff, local_ip4): 7530 BUILD_BUG_ON(sizeof_field(struct sock_common, 7531 skc_rcv_saddr) != 4); 7532 7533 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7534 si->dst_reg, si->src_reg, 7535 offsetof(struct sk_buff, sk)); 7536 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7537 bpf_target_off(struct sock_common, 7538 skc_rcv_saddr, 7539 4, target_size)); 7540 break; 7541 case offsetof(struct __sk_buff, remote_ip6[0]) ... 7542 offsetof(struct __sk_buff, remote_ip6[3]): 7543 #if IS_ENABLED(CONFIG_IPV6) 7544 BUILD_BUG_ON(sizeof_field(struct sock_common, 7545 skc_v6_daddr.s6_addr32[0]) != 4); 7546 7547 off = si->off; 7548 off -= offsetof(struct __sk_buff, remote_ip6[0]); 7549 7550 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7551 si->dst_reg, si->src_reg, 7552 offsetof(struct sk_buff, sk)); 7553 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7554 offsetof(struct sock_common, 7555 skc_v6_daddr.s6_addr32[0]) + 7556 off); 7557 #else 7558 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 7559 #endif 7560 break; 7561 case offsetof(struct __sk_buff, local_ip6[0]) ... 7562 offsetof(struct __sk_buff, local_ip6[3]): 7563 #if IS_ENABLED(CONFIG_IPV6) 7564 BUILD_BUG_ON(sizeof_field(struct sock_common, 7565 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 7566 7567 off = si->off; 7568 off -= offsetof(struct __sk_buff, local_ip6[0]); 7569 7570 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7571 si->dst_reg, si->src_reg, 7572 offsetof(struct sk_buff, sk)); 7573 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7574 offsetof(struct sock_common, 7575 skc_v6_rcv_saddr.s6_addr32[0]) + 7576 off); 7577 #else 7578 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 7579 #endif 7580 break; 7581 7582 case offsetof(struct __sk_buff, remote_port): 7583 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 7584 7585 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7586 si->dst_reg, si->src_reg, 7587 offsetof(struct sk_buff, sk)); 7588 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 7589 bpf_target_off(struct sock_common, 7590 skc_dport, 7591 2, target_size)); 7592 #ifndef __BIG_ENDIAN_BITFIELD 7593 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 7594 #endif 7595 break; 7596 7597 case offsetof(struct __sk_buff, local_port): 7598 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 7599 7600 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7601 si->dst_reg, si->src_reg, 7602 offsetof(struct sk_buff, sk)); 7603 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 7604 bpf_target_off(struct sock_common, 7605 skc_num, 2, target_size)); 7606 break; 7607 7608 case offsetof(struct __sk_buff, tstamp): 7609 BUILD_BUG_ON(sizeof_field(struct sk_buff, tstamp) != 8); 7610 7611 if (type == BPF_WRITE) 7612 *insn++ = BPF_STX_MEM(BPF_DW, 7613 si->dst_reg, si->src_reg, 7614 bpf_target_off(struct sk_buff, 7615 tstamp, 8, 7616 target_size)); 7617 else 7618 *insn++ = BPF_LDX_MEM(BPF_DW, 7619 si->dst_reg, si->src_reg, 7620 bpf_target_off(struct sk_buff, 7621 tstamp, 8, 7622 target_size)); 7623 break; 7624 7625 case offsetof(struct __sk_buff, gso_segs): 7626 insn = bpf_convert_shinfo_access(si, insn); 7627 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_segs), 7628 si->dst_reg, si->dst_reg, 7629 bpf_target_off(struct skb_shared_info, 7630 gso_segs, 2, 7631 target_size)); 7632 break; 7633 case offsetof(struct __sk_buff, gso_size): 7634 insn = bpf_convert_shinfo_access(si, insn); 7635 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct skb_shared_info, gso_size), 7636 si->dst_reg, si->dst_reg, 7637 bpf_target_off(struct skb_shared_info, 7638 gso_size, 2, 7639 target_size)); 7640 break; 7641 case offsetof(struct __sk_buff, wire_len): 7642 BUILD_BUG_ON(sizeof_field(struct qdisc_skb_cb, pkt_len) != 4); 7643 7644 off = si->off; 7645 off -= offsetof(struct __sk_buff, wire_len); 7646 off += offsetof(struct sk_buff, cb); 7647 off += offsetof(struct qdisc_skb_cb, pkt_len); 7648 *target_size = 4; 7649 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, off); 7650 break; 7651 7652 case offsetof(struct __sk_buff, sk): 7653 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, sk), 7654 si->dst_reg, si->src_reg, 7655 offsetof(struct sk_buff, sk)); 7656 break; 7657 } 7658 7659 return insn - insn_buf; 7660 } 7661 7662 u32 bpf_sock_convert_ctx_access(enum bpf_access_type type, 7663 const struct bpf_insn *si, 7664 struct bpf_insn *insn_buf, 7665 struct bpf_prog *prog, u32 *target_size) 7666 { 7667 struct bpf_insn *insn = insn_buf; 7668 int off; 7669 7670 switch (si->off) { 7671 case offsetof(struct bpf_sock, bound_dev_if): 7672 BUILD_BUG_ON(sizeof_field(struct sock, sk_bound_dev_if) != 4); 7673 7674 if (type == BPF_WRITE) 7675 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 7676 offsetof(struct sock, sk_bound_dev_if)); 7677 else 7678 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7679 offsetof(struct sock, sk_bound_dev_if)); 7680 break; 7681 7682 case offsetof(struct bpf_sock, mark): 7683 BUILD_BUG_ON(sizeof_field(struct sock, sk_mark) != 4); 7684 7685 if (type == BPF_WRITE) 7686 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 7687 offsetof(struct sock, sk_mark)); 7688 else 7689 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7690 offsetof(struct sock, sk_mark)); 7691 break; 7692 7693 case offsetof(struct bpf_sock, priority): 7694 BUILD_BUG_ON(sizeof_field(struct sock, sk_priority) != 4); 7695 7696 if (type == BPF_WRITE) 7697 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 7698 offsetof(struct sock, sk_priority)); 7699 else 7700 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 7701 offsetof(struct sock, sk_priority)); 7702 break; 7703 7704 case offsetof(struct bpf_sock, family): 7705 *insn++ = BPF_LDX_MEM( 7706 BPF_FIELD_SIZEOF(struct sock_common, skc_family), 7707 si->dst_reg, si->src_reg, 7708 bpf_target_off(struct sock_common, 7709 skc_family, 7710 sizeof_field(struct sock_common, 7711 skc_family), 7712 target_size)); 7713 break; 7714 7715 case offsetof(struct bpf_sock, type): 7716 *insn++ = BPF_LDX_MEM( 7717 BPF_FIELD_SIZEOF(struct sock, sk_type), 7718 si->dst_reg, si->src_reg, 7719 bpf_target_off(struct sock, sk_type, 7720 sizeof_field(struct sock, sk_type), 7721 target_size)); 7722 break; 7723 7724 case offsetof(struct bpf_sock, protocol): 7725 *insn++ = BPF_LDX_MEM( 7726 BPF_FIELD_SIZEOF(struct sock, sk_protocol), 7727 si->dst_reg, si->src_reg, 7728 bpf_target_off(struct sock, sk_protocol, 7729 sizeof_field(struct sock, sk_protocol), 7730 target_size)); 7731 break; 7732 7733 case offsetof(struct bpf_sock, src_ip4): 7734 *insn++ = BPF_LDX_MEM( 7735 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 7736 bpf_target_off(struct sock_common, skc_rcv_saddr, 7737 sizeof_field(struct sock_common, 7738 skc_rcv_saddr), 7739 target_size)); 7740 break; 7741 7742 case offsetof(struct bpf_sock, dst_ip4): 7743 *insn++ = BPF_LDX_MEM( 7744 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 7745 bpf_target_off(struct sock_common, skc_daddr, 7746 sizeof_field(struct sock_common, 7747 skc_daddr), 7748 target_size)); 7749 break; 7750 7751 case bpf_ctx_range_till(struct bpf_sock, src_ip6[0], src_ip6[3]): 7752 #if IS_ENABLED(CONFIG_IPV6) 7753 off = si->off; 7754 off -= offsetof(struct bpf_sock, src_ip6[0]); 7755 *insn++ = BPF_LDX_MEM( 7756 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 7757 bpf_target_off( 7758 struct sock_common, 7759 skc_v6_rcv_saddr.s6_addr32[0], 7760 sizeof_field(struct sock_common, 7761 skc_v6_rcv_saddr.s6_addr32[0]), 7762 target_size) + off); 7763 #else 7764 (void)off; 7765 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 7766 #endif 7767 break; 7768 7769 case bpf_ctx_range_till(struct bpf_sock, dst_ip6[0], dst_ip6[3]): 7770 #if IS_ENABLED(CONFIG_IPV6) 7771 off = si->off; 7772 off -= offsetof(struct bpf_sock, dst_ip6[0]); 7773 *insn++ = BPF_LDX_MEM( 7774 BPF_SIZE(si->code), si->dst_reg, si->src_reg, 7775 bpf_target_off(struct sock_common, 7776 skc_v6_daddr.s6_addr32[0], 7777 sizeof_field(struct sock_common, 7778 skc_v6_daddr.s6_addr32[0]), 7779 target_size) + off); 7780 #else 7781 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 7782 *target_size = 4; 7783 #endif 7784 break; 7785 7786 case offsetof(struct bpf_sock, src_port): 7787 *insn++ = BPF_LDX_MEM( 7788 BPF_FIELD_SIZEOF(struct sock_common, skc_num), 7789 si->dst_reg, si->src_reg, 7790 bpf_target_off(struct sock_common, skc_num, 7791 sizeof_field(struct sock_common, 7792 skc_num), 7793 target_size)); 7794 break; 7795 7796 case offsetof(struct bpf_sock, dst_port): 7797 *insn++ = BPF_LDX_MEM( 7798 BPF_FIELD_SIZEOF(struct sock_common, skc_dport), 7799 si->dst_reg, si->src_reg, 7800 bpf_target_off(struct sock_common, skc_dport, 7801 sizeof_field(struct sock_common, 7802 skc_dport), 7803 target_size)); 7804 break; 7805 7806 case offsetof(struct bpf_sock, state): 7807 *insn++ = BPF_LDX_MEM( 7808 BPF_FIELD_SIZEOF(struct sock_common, skc_state), 7809 si->dst_reg, si->src_reg, 7810 bpf_target_off(struct sock_common, skc_state, 7811 sizeof_field(struct sock_common, 7812 skc_state), 7813 target_size)); 7814 break; 7815 } 7816 7817 return insn - insn_buf; 7818 } 7819 7820 static u32 tc_cls_act_convert_ctx_access(enum bpf_access_type type, 7821 const struct bpf_insn *si, 7822 struct bpf_insn *insn_buf, 7823 struct bpf_prog *prog, u32 *target_size) 7824 { 7825 struct bpf_insn *insn = insn_buf; 7826 7827 switch (si->off) { 7828 case offsetof(struct __sk_buff, ifindex): 7829 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_buff, dev), 7830 si->dst_reg, si->src_reg, 7831 offsetof(struct sk_buff, dev)); 7832 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7833 bpf_target_off(struct net_device, ifindex, 4, 7834 target_size)); 7835 break; 7836 default: 7837 return bpf_convert_ctx_access(type, si, insn_buf, prog, 7838 target_size); 7839 } 7840 7841 return insn - insn_buf; 7842 } 7843 7844 static u32 xdp_convert_ctx_access(enum bpf_access_type type, 7845 const struct bpf_insn *si, 7846 struct bpf_insn *insn_buf, 7847 struct bpf_prog *prog, u32 *target_size) 7848 { 7849 struct bpf_insn *insn = insn_buf; 7850 7851 switch (si->off) { 7852 case offsetof(struct xdp_md, data): 7853 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data), 7854 si->dst_reg, si->src_reg, 7855 offsetof(struct xdp_buff, data)); 7856 break; 7857 case offsetof(struct xdp_md, data_meta): 7858 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_meta), 7859 si->dst_reg, si->src_reg, 7860 offsetof(struct xdp_buff, data_meta)); 7861 break; 7862 case offsetof(struct xdp_md, data_end): 7863 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, data_end), 7864 si->dst_reg, si->src_reg, 7865 offsetof(struct xdp_buff, data_end)); 7866 break; 7867 case offsetof(struct xdp_md, ingress_ifindex): 7868 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 7869 si->dst_reg, si->src_reg, 7870 offsetof(struct xdp_buff, rxq)); 7871 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_rxq_info, dev), 7872 si->dst_reg, si->dst_reg, 7873 offsetof(struct xdp_rxq_info, dev)); 7874 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7875 offsetof(struct net_device, ifindex)); 7876 break; 7877 case offsetof(struct xdp_md, rx_queue_index): 7878 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct xdp_buff, rxq), 7879 si->dst_reg, si->src_reg, 7880 offsetof(struct xdp_buff, rxq)); 7881 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 7882 offsetof(struct xdp_rxq_info, 7883 queue_index)); 7884 break; 7885 } 7886 7887 return insn - insn_buf; 7888 } 7889 7890 /* SOCK_ADDR_LOAD_NESTED_FIELD() loads Nested Field S.F.NF where S is type of 7891 * context Structure, F is Field in context structure that contains a pointer 7892 * to Nested Structure of type NS that has the field NF. 7893 * 7894 * SIZE encodes the load size (BPF_B, BPF_H, etc). It's up to caller to make 7895 * sure that SIZE is not greater than actual size of S.F.NF. 7896 * 7897 * If offset OFF is provided, the load happens from that offset relative to 7898 * offset of NF. 7899 */ 7900 #define SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF) \ 7901 do { \ 7902 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), si->dst_reg, \ 7903 si->src_reg, offsetof(S, F)); \ 7904 *insn++ = BPF_LDX_MEM( \ 7905 SIZE, si->dst_reg, si->dst_reg, \ 7906 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 7907 target_size) \ 7908 + OFF); \ 7909 } while (0) 7910 7911 #define SOCK_ADDR_LOAD_NESTED_FIELD(S, NS, F, NF) \ 7912 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, \ 7913 BPF_FIELD_SIZEOF(NS, NF), 0) 7914 7915 /* SOCK_ADDR_STORE_NESTED_FIELD_OFF() has semantic similar to 7916 * SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF() but for store operation. 7917 * 7918 * In addition it uses Temporary Field TF (member of struct S) as the 3rd 7919 * "register" since two registers available in convert_ctx_access are not 7920 * enough: we can't override neither SRC, since it contains value to store, nor 7921 * DST since it contains pointer to context that may be used by later 7922 * instructions. But we need a temporary place to save pointer to nested 7923 * structure whose field we want to store to. 7924 */ 7925 #define SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, OFF, TF) \ 7926 do { \ 7927 int tmp_reg = BPF_REG_9; \ 7928 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 7929 --tmp_reg; \ 7930 if (si->src_reg == tmp_reg || si->dst_reg == tmp_reg) \ 7931 --tmp_reg; \ 7932 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, tmp_reg, \ 7933 offsetof(S, TF)); \ 7934 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(S, F), tmp_reg, \ 7935 si->dst_reg, offsetof(S, F)); \ 7936 *insn++ = BPF_STX_MEM(SIZE, tmp_reg, si->src_reg, \ 7937 bpf_target_off(NS, NF, sizeof_field(NS, NF), \ 7938 target_size) \ 7939 + OFF); \ 7940 *insn++ = BPF_LDX_MEM(BPF_DW, tmp_reg, si->dst_reg, \ 7941 offsetof(S, TF)); \ 7942 } while (0) 7943 7944 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF(S, NS, F, NF, SIZE, OFF, \ 7945 TF) \ 7946 do { \ 7947 if (type == BPF_WRITE) { \ 7948 SOCK_ADDR_STORE_NESTED_FIELD_OFF(S, NS, F, NF, SIZE, \ 7949 OFF, TF); \ 7950 } else { \ 7951 SOCK_ADDR_LOAD_NESTED_FIELD_SIZE_OFF( \ 7952 S, NS, F, NF, SIZE, OFF); \ 7953 } \ 7954 } while (0) 7955 7956 #define SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(S, NS, F, NF, TF) \ 7957 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( \ 7958 S, NS, F, NF, BPF_FIELD_SIZEOF(NS, NF), 0, TF) 7959 7960 static u32 sock_addr_convert_ctx_access(enum bpf_access_type type, 7961 const struct bpf_insn *si, 7962 struct bpf_insn *insn_buf, 7963 struct bpf_prog *prog, u32 *target_size) 7964 { 7965 struct bpf_insn *insn = insn_buf; 7966 int off; 7967 7968 switch (si->off) { 7969 case offsetof(struct bpf_sock_addr, user_family): 7970 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 7971 struct sockaddr, uaddr, sa_family); 7972 break; 7973 7974 case offsetof(struct bpf_sock_addr, user_ip4): 7975 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 7976 struct bpf_sock_addr_kern, struct sockaddr_in, uaddr, 7977 sin_addr, BPF_SIZE(si->code), 0, tmp_reg); 7978 break; 7979 7980 case bpf_ctx_range_till(struct bpf_sock_addr, user_ip6[0], user_ip6[3]): 7981 off = si->off; 7982 off -= offsetof(struct bpf_sock_addr, user_ip6[0]); 7983 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 7984 struct bpf_sock_addr_kern, struct sockaddr_in6, uaddr, 7985 sin6_addr.s6_addr32[0], BPF_SIZE(si->code), off, 7986 tmp_reg); 7987 break; 7988 7989 case offsetof(struct bpf_sock_addr, user_port): 7990 /* To get port we need to know sa_family first and then treat 7991 * sockaddr as either sockaddr_in or sockaddr_in6. 7992 * Though we can simplify since port field has same offset and 7993 * size in both structures. 7994 * Here we check this invariant and use just one of the 7995 * structures if it's true. 7996 */ 7997 BUILD_BUG_ON(offsetof(struct sockaddr_in, sin_port) != 7998 offsetof(struct sockaddr_in6, sin6_port)); 7999 BUILD_BUG_ON(sizeof_field(struct sockaddr_in, sin_port) != 8000 sizeof_field(struct sockaddr_in6, sin6_port)); 8001 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD(struct bpf_sock_addr_kern, 8002 struct sockaddr_in6, uaddr, 8003 sin6_port, tmp_reg); 8004 break; 8005 8006 case offsetof(struct bpf_sock_addr, family): 8007 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 8008 struct sock, sk, sk_family); 8009 break; 8010 8011 case offsetof(struct bpf_sock_addr, type): 8012 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 8013 struct sock, sk, sk_type); 8014 break; 8015 8016 case offsetof(struct bpf_sock_addr, protocol): 8017 SOCK_ADDR_LOAD_NESTED_FIELD(struct bpf_sock_addr_kern, 8018 struct sock, sk, sk_protocol); 8019 break; 8020 8021 case offsetof(struct bpf_sock_addr, msg_src_ip4): 8022 /* Treat t_ctx as struct in_addr for msg_src_ip4. */ 8023 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 8024 struct bpf_sock_addr_kern, struct in_addr, t_ctx, 8025 s_addr, BPF_SIZE(si->code), 0, tmp_reg); 8026 break; 8027 8028 case bpf_ctx_range_till(struct bpf_sock_addr, msg_src_ip6[0], 8029 msg_src_ip6[3]): 8030 off = si->off; 8031 off -= offsetof(struct bpf_sock_addr, msg_src_ip6[0]); 8032 /* Treat t_ctx as struct in6_addr for msg_src_ip6. */ 8033 SOCK_ADDR_LOAD_OR_STORE_NESTED_FIELD_SIZE_OFF( 8034 struct bpf_sock_addr_kern, struct in6_addr, t_ctx, 8035 s6_addr32[0], BPF_SIZE(si->code), off, tmp_reg); 8036 break; 8037 case offsetof(struct bpf_sock_addr, sk): 8038 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_sock_addr_kern, sk), 8039 si->dst_reg, si->src_reg, 8040 offsetof(struct bpf_sock_addr_kern, sk)); 8041 break; 8042 } 8043 8044 return insn - insn_buf; 8045 } 8046 8047 static u32 sock_ops_convert_ctx_access(enum bpf_access_type type, 8048 const struct bpf_insn *si, 8049 struct bpf_insn *insn_buf, 8050 struct bpf_prog *prog, 8051 u32 *target_size) 8052 { 8053 struct bpf_insn *insn = insn_buf; 8054 int off; 8055 8056 /* Helper macro for adding read access to tcp_sock or sock fields. */ 8057 #define SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 8058 do { \ 8059 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 8060 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 8061 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 8062 struct bpf_sock_ops_kern, \ 8063 is_fullsock), \ 8064 si->dst_reg, si->src_reg, \ 8065 offsetof(struct bpf_sock_ops_kern, \ 8066 is_fullsock)); \ 8067 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 2); \ 8068 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 8069 struct bpf_sock_ops_kern, sk),\ 8070 si->dst_reg, si->src_reg, \ 8071 offsetof(struct bpf_sock_ops_kern, sk));\ 8072 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(OBJ, \ 8073 OBJ_FIELD), \ 8074 si->dst_reg, si->dst_reg, \ 8075 offsetof(OBJ, OBJ_FIELD)); \ 8076 } while (0) 8077 8078 #define SOCK_OPS_GET_TCP_SOCK_FIELD(FIELD) \ 8079 SOCK_OPS_GET_FIELD(FIELD, FIELD, struct tcp_sock) 8080 8081 /* Helper macro for adding write access to tcp_sock or sock fields. 8082 * The macro is called with two registers, dst_reg which contains a pointer 8083 * to ctx (context) and src_reg which contains the value that should be 8084 * stored. However, we need an additional register since we cannot overwrite 8085 * dst_reg because it may be used later in the program. 8086 * Instead we "borrow" one of the other register. We first save its value 8087 * into a new (temp) field in bpf_sock_ops_kern, use it, and then restore 8088 * it at the end of the macro. 8089 */ 8090 #define SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ) \ 8091 do { \ 8092 int reg = BPF_REG_9; \ 8093 BUILD_BUG_ON(sizeof_field(OBJ, OBJ_FIELD) > \ 8094 sizeof_field(struct bpf_sock_ops, BPF_FIELD)); \ 8095 if (si->dst_reg == reg || si->src_reg == reg) \ 8096 reg--; \ 8097 if (si->dst_reg == reg || si->src_reg == reg) \ 8098 reg--; \ 8099 *insn++ = BPF_STX_MEM(BPF_DW, si->dst_reg, reg, \ 8100 offsetof(struct bpf_sock_ops_kern, \ 8101 temp)); \ 8102 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 8103 struct bpf_sock_ops_kern, \ 8104 is_fullsock), \ 8105 reg, si->dst_reg, \ 8106 offsetof(struct bpf_sock_ops_kern, \ 8107 is_fullsock)); \ 8108 *insn++ = BPF_JMP_IMM(BPF_JEQ, reg, 0, 2); \ 8109 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( \ 8110 struct bpf_sock_ops_kern, sk),\ 8111 reg, si->dst_reg, \ 8112 offsetof(struct bpf_sock_ops_kern, sk));\ 8113 *insn++ = BPF_STX_MEM(BPF_FIELD_SIZEOF(OBJ, OBJ_FIELD), \ 8114 reg, si->src_reg, \ 8115 offsetof(OBJ, OBJ_FIELD)); \ 8116 *insn++ = BPF_LDX_MEM(BPF_DW, reg, si->dst_reg, \ 8117 offsetof(struct bpf_sock_ops_kern, \ 8118 temp)); \ 8119 } while (0) 8120 8121 #define SOCK_OPS_GET_OR_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ, TYPE) \ 8122 do { \ 8123 if (TYPE == BPF_WRITE) \ 8124 SOCK_OPS_SET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 8125 else \ 8126 SOCK_OPS_GET_FIELD(BPF_FIELD, OBJ_FIELD, OBJ); \ 8127 } while (0) 8128 8129 if (insn > insn_buf) 8130 return insn - insn_buf; 8131 8132 switch (si->off) { 8133 case offsetof(struct bpf_sock_ops, op) ... 8134 offsetof(struct bpf_sock_ops, replylong[3]): 8135 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, op) != 8136 sizeof_field(struct bpf_sock_ops_kern, op)); 8137 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, reply) != 8138 sizeof_field(struct bpf_sock_ops_kern, reply)); 8139 BUILD_BUG_ON(sizeof_field(struct bpf_sock_ops, replylong) != 8140 sizeof_field(struct bpf_sock_ops_kern, replylong)); 8141 off = si->off; 8142 off -= offsetof(struct bpf_sock_ops, op); 8143 off += offsetof(struct bpf_sock_ops_kern, op); 8144 if (type == BPF_WRITE) 8145 *insn++ = BPF_STX_MEM(BPF_W, si->dst_reg, si->src_reg, 8146 off); 8147 else 8148 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->src_reg, 8149 off); 8150 break; 8151 8152 case offsetof(struct bpf_sock_ops, family): 8153 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 8154 8155 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8156 struct bpf_sock_ops_kern, sk), 8157 si->dst_reg, si->src_reg, 8158 offsetof(struct bpf_sock_ops_kern, sk)); 8159 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8160 offsetof(struct sock_common, skc_family)); 8161 break; 8162 8163 case offsetof(struct bpf_sock_ops, remote_ip4): 8164 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 8165 8166 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8167 struct bpf_sock_ops_kern, sk), 8168 si->dst_reg, si->src_reg, 8169 offsetof(struct bpf_sock_ops_kern, sk)); 8170 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8171 offsetof(struct sock_common, skc_daddr)); 8172 break; 8173 8174 case offsetof(struct bpf_sock_ops, local_ip4): 8175 BUILD_BUG_ON(sizeof_field(struct sock_common, 8176 skc_rcv_saddr) != 4); 8177 8178 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8179 struct bpf_sock_ops_kern, sk), 8180 si->dst_reg, si->src_reg, 8181 offsetof(struct bpf_sock_ops_kern, sk)); 8182 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8183 offsetof(struct sock_common, 8184 skc_rcv_saddr)); 8185 break; 8186 8187 case offsetof(struct bpf_sock_ops, remote_ip6[0]) ... 8188 offsetof(struct bpf_sock_ops, remote_ip6[3]): 8189 #if IS_ENABLED(CONFIG_IPV6) 8190 BUILD_BUG_ON(sizeof_field(struct sock_common, 8191 skc_v6_daddr.s6_addr32[0]) != 4); 8192 8193 off = si->off; 8194 off -= offsetof(struct bpf_sock_ops, remote_ip6[0]); 8195 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8196 struct bpf_sock_ops_kern, sk), 8197 si->dst_reg, si->src_reg, 8198 offsetof(struct bpf_sock_ops_kern, sk)); 8199 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8200 offsetof(struct sock_common, 8201 skc_v6_daddr.s6_addr32[0]) + 8202 off); 8203 #else 8204 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8205 #endif 8206 break; 8207 8208 case offsetof(struct bpf_sock_ops, local_ip6[0]) ... 8209 offsetof(struct bpf_sock_ops, local_ip6[3]): 8210 #if IS_ENABLED(CONFIG_IPV6) 8211 BUILD_BUG_ON(sizeof_field(struct sock_common, 8212 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 8213 8214 off = si->off; 8215 off -= offsetof(struct bpf_sock_ops, local_ip6[0]); 8216 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8217 struct bpf_sock_ops_kern, sk), 8218 si->dst_reg, si->src_reg, 8219 offsetof(struct bpf_sock_ops_kern, sk)); 8220 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8221 offsetof(struct sock_common, 8222 skc_v6_rcv_saddr.s6_addr32[0]) + 8223 off); 8224 #else 8225 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8226 #endif 8227 break; 8228 8229 case offsetof(struct bpf_sock_ops, remote_port): 8230 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 8231 8232 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8233 struct bpf_sock_ops_kern, sk), 8234 si->dst_reg, si->src_reg, 8235 offsetof(struct bpf_sock_ops_kern, sk)); 8236 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8237 offsetof(struct sock_common, skc_dport)); 8238 #ifndef __BIG_ENDIAN_BITFIELD 8239 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 8240 #endif 8241 break; 8242 8243 case offsetof(struct bpf_sock_ops, local_port): 8244 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 8245 8246 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8247 struct bpf_sock_ops_kern, sk), 8248 si->dst_reg, si->src_reg, 8249 offsetof(struct bpf_sock_ops_kern, sk)); 8250 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8251 offsetof(struct sock_common, skc_num)); 8252 break; 8253 8254 case offsetof(struct bpf_sock_ops, is_fullsock): 8255 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8256 struct bpf_sock_ops_kern, 8257 is_fullsock), 8258 si->dst_reg, si->src_reg, 8259 offsetof(struct bpf_sock_ops_kern, 8260 is_fullsock)); 8261 break; 8262 8263 case offsetof(struct bpf_sock_ops, state): 8264 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_state) != 1); 8265 8266 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8267 struct bpf_sock_ops_kern, sk), 8268 si->dst_reg, si->src_reg, 8269 offsetof(struct bpf_sock_ops_kern, sk)); 8270 *insn++ = BPF_LDX_MEM(BPF_B, si->dst_reg, si->dst_reg, 8271 offsetof(struct sock_common, skc_state)); 8272 break; 8273 8274 case offsetof(struct bpf_sock_ops, rtt_min): 8275 BUILD_BUG_ON(sizeof_field(struct tcp_sock, rtt_min) != 8276 sizeof(struct minmax)); 8277 BUILD_BUG_ON(sizeof(struct minmax) < 8278 sizeof(struct minmax_sample)); 8279 8280 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8281 struct bpf_sock_ops_kern, sk), 8282 si->dst_reg, si->src_reg, 8283 offsetof(struct bpf_sock_ops_kern, sk)); 8284 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8285 offsetof(struct tcp_sock, rtt_min) + 8286 sizeof_field(struct minmax_sample, t)); 8287 break; 8288 8289 case offsetof(struct bpf_sock_ops, bpf_sock_ops_cb_flags): 8290 SOCK_OPS_GET_FIELD(bpf_sock_ops_cb_flags, bpf_sock_ops_cb_flags, 8291 struct tcp_sock); 8292 break; 8293 8294 case offsetof(struct bpf_sock_ops, sk_txhash): 8295 SOCK_OPS_GET_OR_SET_FIELD(sk_txhash, sk_txhash, 8296 struct sock, type); 8297 break; 8298 case offsetof(struct bpf_sock_ops, snd_cwnd): 8299 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_cwnd); 8300 break; 8301 case offsetof(struct bpf_sock_ops, srtt_us): 8302 SOCK_OPS_GET_TCP_SOCK_FIELD(srtt_us); 8303 break; 8304 case offsetof(struct bpf_sock_ops, snd_ssthresh): 8305 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_ssthresh); 8306 break; 8307 case offsetof(struct bpf_sock_ops, rcv_nxt): 8308 SOCK_OPS_GET_TCP_SOCK_FIELD(rcv_nxt); 8309 break; 8310 case offsetof(struct bpf_sock_ops, snd_nxt): 8311 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_nxt); 8312 break; 8313 case offsetof(struct bpf_sock_ops, snd_una): 8314 SOCK_OPS_GET_TCP_SOCK_FIELD(snd_una); 8315 break; 8316 case offsetof(struct bpf_sock_ops, mss_cache): 8317 SOCK_OPS_GET_TCP_SOCK_FIELD(mss_cache); 8318 break; 8319 case offsetof(struct bpf_sock_ops, ecn_flags): 8320 SOCK_OPS_GET_TCP_SOCK_FIELD(ecn_flags); 8321 break; 8322 case offsetof(struct bpf_sock_ops, rate_delivered): 8323 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_delivered); 8324 break; 8325 case offsetof(struct bpf_sock_ops, rate_interval_us): 8326 SOCK_OPS_GET_TCP_SOCK_FIELD(rate_interval_us); 8327 break; 8328 case offsetof(struct bpf_sock_ops, packets_out): 8329 SOCK_OPS_GET_TCP_SOCK_FIELD(packets_out); 8330 break; 8331 case offsetof(struct bpf_sock_ops, retrans_out): 8332 SOCK_OPS_GET_TCP_SOCK_FIELD(retrans_out); 8333 break; 8334 case offsetof(struct bpf_sock_ops, total_retrans): 8335 SOCK_OPS_GET_TCP_SOCK_FIELD(total_retrans); 8336 break; 8337 case offsetof(struct bpf_sock_ops, segs_in): 8338 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_in); 8339 break; 8340 case offsetof(struct bpf_sock_ops, data_segs_in): 8341 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_in); 8342 break; 8343 case offsetof(struct bpf_sock_ops, segs_out): 8344 SOCK_OPS_GET_TCP_SOCK_FIELD(segs_out); 8345 break; 8346 case offsetof(struct bpf_sock_ops, data_segs_out): 8347 SOCK_OPS_GET_TCP_SOCK_FIELD(data_segs_out); 8348 break; 8349 case offsetof(struct bpf_sock_ops, lost_out): 8350 SOCK_OPS_GET_TCP_SOCK_FIELD(lost_out); 8351 break; 8352 case offsetof(struct bpf_sock_ops, sacked_out): 8353 SOCK_OPS_GET_TCP_SOCK_FIELD(sacked_out); 8354 break; 8355 case offsetof(struct bpf_sock_ops, bytes_received): 8356 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_received); 8357 break; 8358 case offsetof(struct bpf_sock_ops, bytes_acked): 8359 SOCK_OPS_GET_TCP_SOCK_FIELD(bytes_acked); 8360 break; 8361 case offsetof(struct bpf_sock_ops, sk): 8362 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8363 struct bpf_sock_ops_kern, 8364 is_fullsock), 8365 si->dst_reg, si->src_reg, 8366 offsetof(struct bpf_sock_ops_kern, 8367 is_fullsock)); 8368 *insn++ = BPF_JMP_IMM(BPF_JEQ, si->dst_reg, 0, 1); 8369 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8370 struct bpf_sock_ops_kern, sk), 8371 si->dst_reg, si->src_reg, 8372 offsetof(struct bpf_sock_ops_kern, sk)); 8373 break; 8374 } 8375 return insn - insn_buf; 8376 } 8377 8378 static u32 sk_skb_convert_ctx_access(enum bpf_access_type type, 8379 const struct bpf_insn *si, 8380 struct bpf_insn *insn_buf, 8381 struct bpf_prog *prog, u32 *target_size) 8382 { 8383 struct bpf_insn *insn = insn_buf; 8384 int off; 8385 8386 switch (si->off) { 8387 case offsetof(struct __sk_buff, data_end): 8388 off = si->off; 8389 off -= offsetof(struct __sk_buff, data_end); 8390 off += offsetof(struct sk_buff, cb); 8391 off += offsetof(struct tcp_skb_cb, bpf.data_end); 8392 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(void *), si->dst_reg, 8393 si->src_reg, off); 8394 break; 8395 default: 8396 return bpf_convert_ctx_access(type, si, insn_buf, prog, 8397 target_size); 8398 } 8399 8400 return insn - insn_buf; 8401 } 8402 8403 static u32 sk_msg_convert_ctx_access(enum bpf_access_type type, 8404 const struct bpf_insn *si, 8405 struct bpf_insn *insn_buf, 8406 struct bpf_prog *prog, u32 *target_size) 8407 { 8408 struct bpf_insn *insn = insn_buf; 8409 #if IS_ENABLED(CONFIG_IPV6) 8410 int off; 8411 #endif 8412 8413 /* convert ctx uses the fact sg element is first in struct */ 8414 BUILD_BUG_ON(offsetof(struct sk_msg, sg) != 0); 8415 8416 switch (si->off) { 8417 case offsetof(struct sk_msg_md, data): 8418 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data), 8419 si->dst_reg, si->src_reg, 8420 offsetof(struct sk_msg, data)); 8421 break; 8422 case offsetof(struct sk_msg_md, data_end): 8423 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg, data_end), 8424 si->dst_reg, si->src_reg, 8425 offsetof(struct sk_msg, data_end)); 8426 break; 8427 case offsetof(struct sk_msg_md, family): 8428 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_family) != 2); 8429 8430 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8431 struct sk_msg, sk), 8432 si->dst_reg, si->src_reg, 8433 offsetof(struct sk_msg, sk)); 8434 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8435 offsetof(struct sock_common, skc_family)); 8436 break; 8437 8438 case offsetof(struct sk_msg_md, remote_ip4): 8439 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_daddr) != 4); 8440 8441 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8442 struct sk_msg, sk), 8443 si->dst_reg, si->src_reg, 8444 offsetof(struct sk_msg, sk)); 8445 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8446 offsetof(struct sock_common, skc_daddr)); 8447 break; 8448 8449 case offsetof(struct sk_msg_md, local_ip4): 8450 BUILD_BUG_ON(sizeof_field(struct sock_common, 8451 skc_rcv_saddr) != 4); 8452 8453 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8454 struct sk_msg, sk), 8455 si->dst_reg, si->src_reg, 8456 offsetof(struct sk_msg, sk)); 8457 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8458 offsetof(struct sock_common, 8459 skc_rcv_saddr)); 8460 break; 8461 8462 case offsetof(struct sk_msg_md, remote_ip6[0]) ... 8463 offsetof(struct sk_msg_md, remote_ip6[3]): 8464 #if IS_ENABLED(CONFIG_IPV6) 8465 BUILD_BUG_ON(sizeof_field(struct sock_common, 8466 skc_v6_daddr.s6_addr32[0]) != 4); 8467 8468 off = si->off; 8469 off -= offsetof(struct sk_msg_md, remote_ip6[0]); 8470 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8471 struct sk_msg, sk), 8472 si->dst_reg, si->src_reg, 8473 offsetof(struct sk_msg, sk)); 8474 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8475 offsetof(struct sock_common, 8476 skc_v6_daddr.s6_addr32[0]) + 8477 off); 8478 #else 8479 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8480 #endif 8481 break; 8482 8483 case offsetof(struct sk_msg_md, local_ip6[0]) ... 8484 offsetof(struct sk_msg_md, local_ip6[3]): 8485 #if IS_ENABLED(CONFIG_IPV6) 8486 BUILD_BUG_ON(sizeof_field(struct sock_common, 8487 skc_v6_rcv_saddr.s6_addr32[0]) != 4); 8488 8489 off = si->off; 8490 off -= offsetof(struct sk_msg_md, local_ip6[0]); 8491 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8492 struct sk_msg, sk), 8493 si->dst_reg, si->src_reg, 8494 offsetof(struct sk_msg, sk)); 8495 *insn++ = BPF_LDX_MEM(BPF_W, si->dst_reg, si->dst_reg, 8496 offsetof(struct sock_common, 8497 skc_v6_rcv_saddr.s6_addr32[0]) + 8498 off); 8499 #else 8500 *insn++ = BPF_MOV32_IMM(si->dst_reg, 0); 8501 #endif 8502 break; 8503 8504 case offsetof(struct sk_msg_md, remote_port): 8505 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_dport) != 2); 8506 8507 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8508 struct sk_msg, sk), 8509 si->dst_reg, si->src_reg, 8510 offsetof(struct sk_msg, sk)); 8511 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8512 offsetof(struct sock_common, skc_dport)); 8513 #ifndef __BIG_ENDIAN_BITFIELD 8514 *insn++ = BPF_ALU32_IMM(BPF_LSH, si->dst_reg, 16); 8515 #endif 8516 break; 8517 8518 case offsetof(struct sk_msg_md, local_port): 8519 BUILD_BUG_ON(sizeof_field(struct sock_common, skc_num) != 2); 8520 8521 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF( 8522 struct sk_msg, sk), 8523 si->dst_reg, si->src_reg, 8524 offsetof(struct sk_msg, sk)); 8525 *insn++ = BPF_LDX_MEM(BPF_H, si->dst_reg, si->dst_reg, 8526 offsetof(struct sock_common, skc_num)); 8527 break; 8528 8529 case offsetof(struct sk_msg_md, size): 8530 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_msg_sg, size), 8531 si->dst_reg, si->src_reg, 8532 offsetof(struct sk_msg_sg, size)); 8533 break; 8534 } 8535 8536 return insn - insn_buf; 8537 } 8538 8539 const struct bpf_verifier_ops sk_filter_verifier_ops = { 8540 .get_func_proto = sk_filter_func_proto, 8541 .is_valid_access = sk_filter_is_valid_access, 8542 .convert_ctx_access = bpf_convert_ctx_access, 8543 .gen_ld_abs = bpf_gen_ld_abs, 8544 }; 8545 8546 const struct bpf_prog_ops sk_filter_prog_ops = { 8547 .test_run = bpf_prog_test_run_skb, 8548 }; 8549 8550 const struct bpf_verifier_ops tc_cls_act_verifier_ops = { 8551 .get_func_proto = tc_cls_act_func_proto, 8552 .is_valid_access = tc_cls_act_is_valid_access, 8553 .convert_ctx_access = tc_cls_act_convert_ctx_access, 8554 .gen_prologue = tc_cls_act_prologue, 8555 .gen_ld_abs = bpf_gen_ld_abs, 8556 }; 8557 8558 const struct bpf_prog_ops tc_cls_act_prog_ops = { 8559 .test_run = bpf_prog_test_run_skb, 8560 }; 8561 8562 const struct bpf_verifier_ops xdp_verifier_ops = { 8563 .get_func_proto = xdp_func_proto, 8564 .is_valid_access = xdp_is_valid_access, 8565 .convert_ctx_access = xdp_convert_ctx_access, 8566 .gen_prologue = bpf_noop_prologue, 8567 }; 8568 8569 const struct bpf_prog_ops xdp_prog_ops = { 8570 .test_run = bpf_prog_test_run_xdp, 8571 }; 8572 8573 const struct bpf_verifier_ops cg_skb_verifier_ops = { 8574 .get_func_proto = cg_skb_func_proto, 8575 .is_valid_access = cg_skb_is_valid_access, 8576 .convert_ctx_access = bpf_convert_ctx_access, 8577 }; 8578 8579 const struct bpf_prog_ops cg_skb_prog_ops = { 8580 .test_run = bpf_prog_test_run_skb, 8581 }; 8582 8583 const struct bpf_verifier_ops lwt_in_verifier_ops = { 8584 .get_func_proto = lwt_in_func_proto, 8585 .is_valid_access = lwt_is_valid_access, 8586 .convert_ctx_access = bpf_convert_ctx_access, 8587 }; 8588 8589 const struct bpf_prog_ops lwt_in_prog_ops = { 8590 .test_run = bpf_prog_test_run_skb, 8591 }; 8592 8593 const struct bpf_verifier_ops lwt_out_verifier_ops = { 8594 .get_func_proto = lwt_out_func_proto, 8595 .is_valid_access = lwt_is_valid_access, 8596 .convert_ctx_access = bpf_convert_ctx_access, 8597 }; 8598 8599 const struct bpf_prog_ops lwt_out_prog_ops = { 8600 .test_run = bpf_prog_test_run_skb, 8601 }; 8602 8603 const struct bpf_verifier_ops lwt_xmit_verifier_ops = { 8604 .get_func_proto = lwt_xmit_func_proto, 8605 .is_valid_access = lwt_is_valid_access, 8606 .convert_ctx_access = bpf_convert_ctx_access, 8607 .gen_prologue = tc_cls_act_prologue, 8608 }; 8609 8610 const struct bpf_prog_ops lwt_xmit_prog_ops = { 8611 .test_run = bpf_prog_test_run_skb, 8612 }; 8613 8614 const struct bpf_verifier_ops lwt_seg6local_verifier_ops = { 8615 .get_func_proto = lwt_seg6local_func_proto, 8616 .is_valid_access = lwt_is_valid_access, 8617 .convert_ctx_access = bpf_convert_ctx_access, 8618 }; 8619 8620 const struct bpf_prog_ops lwt_seg6local_prog_ops = { 8621 .test_run = bpf_prog_test_run_skb, 8622 }; 8623 8624 const struct bpf_verifier_ops cg_sock_verifier_ops = { 8625 .get_func_proto = sock_filter_func_proto, 8626 .is_valid_access = sock_filter_is_valid_access, 8627 .convert_ctx_access = bpf_sock_convert_ctx_access, 8628 }; 8629 8630 const struct bpf_prog_ops cg_sock_prog_ops = { 8631 }; 8632 8633 const struct bpf_verifier_ops cg_sock_addr_verifier_ops = { 8634 .get_func_proto = sock_addr_func_proto, 8635 .is_valid_access = sock_addr_is_valid_access, 8636 .convert_ctx_access = sock_addr_convert_ctx_access, 8637 }; 8638 8639 const struct bpf_prog_ops cg_sock_addr_prog_ops = { 8640 }; 8641 8642 const struct bpf_verifier_ops sock_ops_verifier_ops = { 8643 .get_func_proto = sock_ops_func_proto, 8644 .is_valid_access = sock_ops_is_valid_access, 8645 .convert_ctx_access = sock_ops_convert_ctx_access, 8646 }; 8647 8648 const struct bpf_prog_ops sock_ops_prog_ops = { 8649 }; 8650 8651 const struct bpf_verifier_ops sk_skb_verifier_ops = { 8652 .get_func_proto = sk_skb_func_proto, 8653 .is_valid_access = sk_skb_is_valid_access, 8654 .convert_ctx_access = sk_skb_convert_ctx_access, 8655 .gen_prologue = sk_skb_prologue, 8656 }; 8657 8658 const struct bpf_prog_ops sk_skb_prog_ops = { 8659 }; 8660 8661 const struct bpf_verifier_ops sk_msg_verifier_ops = { 8662 .get_func_proto = sk_msg_func_proto, 8663 .is_valid_access = sk_msg_is_valid_access, 8664 .convert_ctx_access = sk_msg_convert_ctx_access, 8665 .gen_prologue = bpf_noop_prologue, 8666 }; 8667 8668 const struct bpf_prog_ops sk_msg_prog_ops = { 8669 }; 8670 8671 const struct bpf_verifier_ops flow_dissector_verifier_ops = { 8672 .get_func_proto = flow_dissector_func_proto, 8673 .is_valid_access = flow_dissector_is_valid_access, 8674 .convert_ctx_access = flow_dissector_convert_ctx_access, 8675 }; 8676 8677 const struct bpf_prog_ops flow_dissector_prog_ops = { 8678 .test_run = bpf_prog_test_run_flow_dissector, 8679 }; 8680 8681 int sk_detach_filter(struct sock *sk) 8682 { 8683 int ret = -ENOENT; 8684 struct sk_filter *filter; 8685 8686 if (sock_flag(sk, SOCK_FILTER_LOCKED)) 8687 return -EPERM; 8688 8689 filter = rcu_dereference_protected(sk->sk_filter, 8690 lockdep_sock_is_held(sk)); 8691 if (filter) { 8692 RCU_INIT_POINTER(sk->sk_filter, NULL); 8693 sk_filter_uncharge(sk, filter); 8694 ret = 0; 8695 } 8696 8697 return ret; 8698 } 8699 EXPORT_SYMBOL_GPL(sk_detach_filter); 8700 8701 int sk_get_filter(struct sock *sk, struct sock_filter __user *ubuf, 8702 unsigned int len) 8703 { 8704 struct sock_fprog_kern *fprog; 8705 struct sk_filter *filter; 8706 int ret = 0; 8707 8708 lock_sock(sk); 8709 filter = rcu_dereference_protected(sk->sk_filter, 8710 lockdep_sock_is_held(sk)); 8711 if (!filter) 8712 goto out; 8713 8714 /* We're copying the filter that has been originally attached, 8715 * so no conversion/decode needed anymore. eBPF programs that 8716 * have no original program cannot be dumped through this. 8717 */ 8718 ret = -EACCES; 8719 fprog = filter->prog->orig_prog; 8720 if (!fprog) 8721 goto out; 8722 8723 ret = fprog->len; 8724 if (!len) 8725 /* User space only enquires number of filter blocks. */ 8726 goto out; 8727 8728 ret = -EINVAL; 8729 if (len < fprog->len) 8730 goto out; 8731 8732 ret = -EFAULT; 8733 if (copy_to_user(ubuf, fprog->filter, bpf_classic_proglen(fprog))) 8734 goto out; 8735 8736 /* Instead of bytes, the API requests to return the number 8737 * of filter blocks. 8738 */ 8739 ret = fprog->len; 8740 out: 8741 release_sock(sk); 8742 return ret; 8743 } 8744 8745 #ifdef CONFIG_INET 8746 static void bpf_init_reuseport_kern(struct sk_reuseport_kern *reuse_kern, 8747 struct sock_reuseport *reuse, 8748 struct sock *sk, struct sk_buff *skb, 8749 u32 hash) 8750 { 8751 reuse_kern->skb = skb; 8752 reuse_kern->sk = sk; 8753 reuse_kern->selected_sk = NULL; 8754 reuse_kern->data_end = skb->data + skb_headlen(skb); 8755 reuse_kern->hash = hash; 8756 reuse_kern->reuseport_id = reuse->reuseport_id; 8757 reuse_kern->bind_inany = reuse->bind_inany; 8758 } 8759 8760 struct sock *bpf_run_sk_reuseport(struct sock_reuseport *reuse, struct sock *sk, 8761 struct bpf_prog *prog, struct sk_buff *skb, 8762 u32 hash) 8763 { 8764 struct sk_reuseport_kern reuse_kern; 8765 enum sk_action action; 8766 8767 bpf_init_reuseport_kern(&reuse_kern, reuse, sk, skb, hash); 8768 action = BPF_PROG_RUN(prog, &reuse_kern); 8769 8770 if (action == SK_PASS) 8771 return reuse_kern.selected_sk; 8772 else 8773 return ERR_PTR(-ECONNREFUSED); 8774 } 8775 8776 BPF_CALL_4(sk_select_reuseport, struct sk_reuseport_kern *, reuse_kern, 8777 struct bpf_map *, map, void *, key, u32, flags) 8778 { 8779 bool is_sockarray = map->map_type == BPF_MAP_TYPE_REUSEPORT_SOCKARRAY; 8780 struct sock_reuseport *reuse; 8781 struct sock *selected_sk; 8782 8783 selected_sk = map->ops->map_lookup_elem(map, key); 8784 if (!selected_sk) 8785 return -ENOENT; 8786 8787 reuse = rcu_dereference(selected_sk->sk_reuseport_cb); 8788 if (!reuse) { 8789 /* reuseport_array has only sk with non NULL sk_reuseport_cb. 8790 * The only (!reuse) case here is - the sk has already been 8791 * unhashed (e.g. by close()), so treat it as -ENOENT. 8792 * 8793 * Other maps (e.g. sock_map) do not provide this guarantee and 8794 * the sk may never be in the reuseport group to begin with. 8795 */ 8796 return is_sockarray ? -ENOENT : -EINVAL; 8797 } 8798 8799 if (unlikely(reuse->reuseport_id != reuse_kern->reuseport_id)) { 8800 struct sock *sk = reuse_kern->sk; 8801 8802 if (sk->sk_protocol != selected_sk->sk_protocol) 8803 return -EPROTOTYPE; 8804 else if (sk->sk_family != selected_sk->sk_family) 8805 return -EAFNOSUPPORT; 8806 8807 /* Catch all. Likely bound to a different sockaddr. */ 8808 return -EBADFD; 8809 } 8810 8811 reuse_kern->selected_sk = selected_sk; 8812 8813 return 0; 8814 } 8815 8816 static const struct bpf_func_proto sk_select_reuseport_proto = { 8817 .func = sk_select_reuseport, 8818 .gpl_only = false, 8819 .ret_type = RET_INTEGER, 8820 .arg1_type = ARG_PTR_TO_CTX, 8821 .arg2_type = ARG_CONST_MAP_PTR, 8822 .arg3_type = ARG_PTR_TO_MAP_KEY, 8823 .arg4_type = ARG_ANYTHING, 8824 }; 8825 8826 BPF_CALL_4(sk_reuseport_load_bytes, 8827 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 8828 void *, to, u32, len) 8829 { 8830 return ____bpf_skb_load_bytes(reuse_kern->skb, offset, to, len); 8831 } 8832 8833 static const struct bpf_func_proto sk_reuseport_load_bytes_proto = { 8834 .func = sk_reuseport_load_bytes, 8835 .gpl_only = false, 8836 .ret_type = RET_INTEGER, 8837 .arg1_type = ARG_PTR_TO_CTX, 8838 .arg2_type = ARG_ANYTHING, 8839 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 8840 .arg4_type = ARG_CONST_SIZE, 8841 }; 8842 8843 BPF_CALL_5(sk_reuseport_load_bytes_relative, 8844 const struct sk_reuseport_kern *, reuse_kern, u32, offset, 8845 void *, to, u32, len, u32, start_header) 8846 { 8847 return ____bpf_skb_load_bytes_relative(reuse_kern->skb, offset, to, 8848 len, start_header); 8849 } 8850 8851 static const struct bpf_func_proto sk_reuseport_load_bytes_relative_proto = { 8852 .func = sk_reuseport_load_bytes_relative, 8853 .gpl_only = false, 8854 .ret_type = RET_INTEGER, 8855 .arg1_type = ARG_PTR_TO_CTX, 8856 .arg2_type = ARG_ANYTHING, 8857 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 8858 .arg4_type = ARG_CONST_SIZE, 8859 .arg5_type = ARG_ANYTHING, 8860 }; 8861 8862 static const struct bpf_func_proto * 8863 sk_reuseport_func_proto(enum bpf_func_id func_id, 8864 const struct bpf_prog *prog) 8865 { 8866 switch (func_id) { 8867 case BPF_FUNC_sk_select_reuseport: 8868 return &sk_select_reuseport_proto; 8869 case BPF_FUNC_skb_load_bytes: 8870 return &sk_reuseport_load_bytes_proto; 8871 case BPF_FUNC_skb_load_bytes_relative: 8872 return &sk_reuseport_load_bytes_relative_proto; 8873 default: 8874 return bpf_base_func_proto(func_id); 8875 } 8876 } 8877 8878 static bool 8879 sk_reuseport_is_valid_access(int off, int size, 8880 enum bpf_access_type type, 8881 const struct bpf_prog *prog, 8882 struct bpf_insn_access_aux *info) 8883 { 8884 const u32 size_default = sizeof(__u32); 8885 8886 if (off < 0 || off >= sizeof(struct sk_reuseport_md) || 8887 off % size || type != BPF_READ) 8888 return false; 8889 8890 switch (off) { 8891 case offsetof(struct sk_reuseport_md, data): 8892 info->reg_type = PTR_TO_PACKET; 8893 return size == sizeof(__u64); 8894 8895 case offsetof(struct sk_reuseport_md, data_end): 8896 info->reg_type = PTR_TO_PACKET_END; 8897 return size == sizeof(__u64); 8898 8899 case offsetof(struct sk_reuseport_md, hash): 8900 return size == size_default; 8901 8902 /* Fields that allow narrowing */ 8903 case bpf_ctx_range(struct sk_reuseport_md, eth_protocol): 8904 if (size < sizeof_field(struct sk_buff, protocol)) 8905 return false; 8906 /* fall through */ 8907 case bpf_ctx_range(struct sk_reuseport_md, ip_protocol): 8908 case bpf_ctx_range(struct sk_reuseport_md, bind_inany): 8909 case bpf_ctx_range(struct sk_reuseport_md, len): 8910 bpf_ctx_record_field_size(info, size_default); 8911 return bpf_ctx_narrow_access_ok(off, size, size_default); 8912 8913 default: 8914 return false; 8915 } 8916 } 8917 8918 #define SK_REUSEPORT_LOAD_FIELD(F) ({ \ 8919 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct sk_reuseport_kern, F), \ 8920 si->dst_reg, si->src_reg, \ 8921 bpf_target_off(struct sk_reuseport_kern, F, \ 8922 sizeof_field(struct sk_reuseport_kern, F), \ 8923 target_size)); \ 8924 }) 8925 8926 #define SK_REUSEPORT_LOAD_SKB_FIELD(SKB_FIELD) \ 8927 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 8928 struct sk_buff, \ 8929 skb, \ 8930 SKB_FIELD) 8931 8932 #define SK_REUSEPORT_LOAD_SK_FIELD(SK_FIELD) \ 8933 SOCK_ADDR_LOAD_NESTED_FIELD(struct sk_reuseport_kern, \ 8934 struct sock, \ 8935 sk, \ 8936 SK_FIELD) 8937 8938 static u32 sk_reuseport_convert_ctx_access(enum bpf_access_type type, 8939 const struct bpf_insn *si, 8940 struct bpf_insn *insn_buf, 8941 struct bpf_prog *prog, 8942 u32 *target_size) 8943 { 8944 struct bpf_insn *insn = insn_buf; 8945 8946 switch (si->off) { 8947 case offsetof(struct sk_reuseport_md, data): 8948 SK_REUSEPORT_LOAD_SKB_FIELD(data); 8949 break; 8950 8951 case offsetof(struct sk_reuseport_md, len): 8952 SK_REUSEPORT_LOAD_SKB_FIELD(len); 8953 break; 8954 8955 case offsetof(struct sk_reuseport_md, eth_protocol): 8956 SK_REUSEPORT_LOAD_SKB_FIELD(protocol); 8957 break; 8958 8959 case offsetof(struct sk_reuseport_md, ip_protocol): 8960 SK_REUSEPORT_LOAD_SK_FIELD(sk_protocol); 8961 break; 8962 8963 case offsetof(struct sk_reuseport_md, data_end): 8964 SK_REUSEPORT_LOAD_FIELD(data_end); 8965 break; 8966 8967 case offsetof(struct sk_reuseport_md, hash): 8968 SK_REUSEPORT_LOAD_FIELD(hash); 8969 break; 8970 8971 case offsetof(struct sk_reuseport_md, bind_inany): 8972 SK_REUSEPORT_LOAD_FIELD(bind_inany); 8973 break; 8974 } 8975 8976 return insn - insn_buf; 8977 } 8978 8979 const struct bpf_verifier_ops sk_reuseport_verifier_ops = { 8980 .get_func_proto = sk_reuseport_func_proto, 8981 .is_valid_access = sk_reuseport_is_valid_access, 8982 .convert_ctx_access = sk_reuseport_convert_ctx_access, 8983 }; 8984 8985 const struct bpf_prog_ops sk_reuseport_prog_ops = { 8986 }; 8987 #endif /* CONFIG_INET */ 8988 8989 DEFINE_BPF_DISPATCHER(xdp) 8990 8991 void bpf_prog_change_xdp(struct bpf_prog *prev_prog, struct bpf_prog *prog) 8992 { 8993 bpf_dispatcher_change_prog(BPF_DISPATCHER_PTR(xdp), prev_prog, prog); 8994 } 8995