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